You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

RefDirectory.java 41KB

Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Redo event listeners to be more generic Replace the old crude event listener system with a much more generic implementation, patterned after the event dispatch techniques used in Google Web Toolkit 1.5 and later. Each event delivers to an interface that defines a single method, and the event itself is what performs the delivery in a type-safe way through its own dispatch method. Listeners are registered in a generic listener list, indexed by the interface they implement and wish to receive an event for. Delivery of events is performed by looping through all listeners implementing the event's corresponding listener interface, and using the event's own dispatch method to deliver the event. This is the classical "double dispatch" pattern for event delivery. Listeners can be unregistered by invoking remove() on their registration handle. This change therefore requires application code to track the handle if it wishes to remove the listener at a later point in time. Event delivery is now exposed as a generic public method on the Repository class, making it easier for any type of message to be sent out to any type of listener that has registered, without needing to pre-arrange for type-safe fireFoo() methods. New event types can be added in the future simply by defining a new RepositoryEvent subclass and a corresponding RepositoryListener interface that it dispatches to. By always adding new events through a new interface, we never need to worry about defining an Adapter to provide default no-op implementations of new event methods. Change-Id: I651417b3098b9afc93d91085e9f0b2265df8fc81 Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Handle stale file handles on packed-refs file On a local filesystem the packed-refs file will be orphaned if it is replaced by another client while the current client is reading the old one. However, since NFS servers do not keep track of open files, instead of orphaning the old packed-refs file, such a replacement will cause the old file to be garbage collected instead. A stale file handle exception will be raised on NFS servers if the file is garbage collected (deleted) on the server while it is being read. Since we no longer have access to the old file in these cases, the previous code would just fail. However, in these cases, reopening the file and rereading it will succeed (since it will reopen the new replacement file). So retrying the read is a viable strategy to deal with stale file handles on the packed-refs file, implement such a strategy. Since it is possible that the packed-refs file could be replaced again while rereading it (multiple consecutive updates can easily occur with ref deletions), loop on stale file handle exceptions, up to 5 extra times, trying to read the packed-refs file again, until we either read the new file, or find that the file no longer exists. The limit of 5 is arbitrary, and provides a safe upper bounds to prevent infinite loops consuming resources in a potential unforeseen persistent error condition. Change-Id: I085c472bafa6e2f32f610a33ddc8368bb4ab1814 Signed-off-by: Martin Fick<mfick@codeaurora.org> Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
8 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Handle stale file handles on packed-refs file On a local filesystem the packed-refs file will be orphaned if it is replaced by another client while the current client is reading the old one. However, since NFS servers do not keep track of open files, instead of orphaning the old packed-refs file, such a replacement will cause the old file to be garbage collected instead. A stale file handle exception will be raised on NFS servers if the file is garbage collected (deleted) on the server while it is being read. Since we no longer have access to the old file in these cases, the previous code would just fail. However, in these cases, reopening the file and rereading it will succeed (since it will reopen the new replacement file). So retrying the read is a viable strategy to deal with stale file handles on the packed-refs file, implement such a strategy. Since it is possible that the packed-refs file could be replaced again while rereading it (multiple consecutive updates can easily occur with ref deletions), loop on stale file handle exceptions, up to 5 extra times, trying to read the packed-refs file again, until we either read the new file, or find that the file no longer exists. The limit of 5 is arbitrary, and provides a safe upper bounds to prevent infinite loops consuming resources in a potential unforeseen persistent error condition. Change-Id: I085c472bafa6e2f32f610a33ddc8368bb4ab1814 Signed-off-by: Martin Fick<mfick@codeaurora.org> Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
8 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Scan loose ref before packed in case gc about to remove the loose Before this change, jgit used to read packed-refs before scanning loose refs. That was not a problem if gc didn't run concurrently. When gc did run concurrently with such refs reading, that order sometimes broke the latter. This lead to reading an older version of a ref's tip, which meant "losing" the real tip or commit. The specific read-Vs-gc concurrency scenario which broke reading that way follows: 1. let ref R be in packed-refs and R' be in loose 2. jgit starts reading packed-refs 3. gc also starts its business around that very time 4. jgit still has the time to read R from packed-refs 5. as gc is not done yet updating packed-refs with R' 6. jgit then starts scanning loose refs (or is about to) 7. gc quickly ends up being done moving loose R' to packed-refs 8. so gc (quickly) removes loose refs 9. -while jgit is scanning loose refs, now gone 10. so jgit assumes no loose to consider => packed-refs winning 11. so jgit wrongfully returns R (from 4.) as the tip, instead of R'. This fix switches the order so loose refs are scanned (secured) before taking the time to read packed-refs. This way, knowledge of the likelier tip is guaranteed for ref reading to return the true tip - despite concurrent gc. If there is no loose ref to scan, jgit reads packed-refs and lands on R' (or S), which it then returns, as expected. The gerrit issue [1] should be solved by this fix. [1] https://code.google.com/p/gerrit/issues/detail?id=2302 Change-Id: Ibd120120a361a3a6ed565f3836afc1db706fbcdd Signed-off-by: Marco Miller <marco.miller@ericsson.com> Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
8 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Do not fake a SymbolicRef as an ObjectIdRef When doing a detaching operation, JGit fakes a SymbolicRef as an ObjectIdRef. This is because RefUpdate#updateImpl dereferences the SymbolicRef when updating it. For example, assume that HEAD is pointing to refs/heads/master. If I try to make a detached HEAD pointing to a commit c0ffee, RefUpdate dereferences HEAD as refs/heads/master first and changes refs/heads/master to c0ffee. The detach argument of RefDatabase#newUpdate avoids this dereference by faking HEAD as ObjectIdRef. This faking is problematic for the linking operation of DfsRefDatabase. It does a compare-and-swap operation on every reference change because of its distributed systems nature. If a SymbolicRef is faked as an ObjectRef, it thinks that there is a racing change in the reference and rejects the update. Because of this, DFS based repositories cannot change the link target of symbolic refs. This has not been a problem for file-based repositories because they have a file-lock based semantics instead of the CAS based one. The reference implementation, InMemoryRepository, is not affected because it only compares ObjectIds. When [1] introduced this faking code, there was no way for RefUpdate to distinguish the detaching operation. When [2] fixed the detaching operation, it introduced a detachingSymbolicRef flag. This commit uses this flag to control whether it needs to dereference the symbolic refs by calling Ref#getLeaf. The same flag is used in the reflog update operation. This commit does not affect any operation that succeeds currently. In some DFS repository implementations, this fixes a ref linking operation, which is currently failing. [1]: https://eclipse.googlesource.com/jgit/jgit/+/01b5392cdbc12ce2e21fd1d1afbd61fdf97e1c38 [2]: https://eclipse.googlesource.com/jgit/jgit/+/3a86868c0883d2a564db88bf9ae4a5fe235bb63f Change-Id: I118f85f0414dbfad02250944e28d74dddd59469b Signed-off-by: Masaya Suzuki <masayasuzuki@google.com>
7 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Handle stale file handles on packed-refs file On a local filesystem the packed-refs file will be orphaned if it is replaced by another client while the current client is reading the old one. However, since NFS servers do not keep track of open files, instead of orphaning the old packed-refs file, such a replacement will cause the old file to be garbage collected instead. A stale file handle exception will be raised on NFS servers if the file is garbage collected (deleted) on the server while it is being read. Since we no longer have access to the old file in these cases, the previous code would just fail. However, in these cases, reopening the file and rereading it will succeed (since it will reopen the new replacement file). So retrying the read is a viable strategy to deal with stale file handles on the packed-refs file, implement such a strategy. Since it is possible that the packed-refs file could be replaced again while rereading it (multiple consecutive updates can easily occur with ref deletions), loop on stale file handle exceptions, up to 5 extra times, trying to read the packed-refs file again, until we either read the new file, or find that the file no longer exists. The limit of 5 is arbitrary, and provides a safe upper bounds to prevent infinite loops consuming resources in a potential unforeseen persistent error condition. Change-Id: I085c472bafa6e2f32f610a33ddc8368bb4ab1814 Signed-off-by: Martin Fick<mfick@codeaurora.org> Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
8 years ago
Handle stale file handles on packed-refs file On a local filesystem the packed-refs file will be orphaned if it is replaced by another client while the current client is reading the old one. However, since NFS servers do not keep track of open files, instead of orphaning the old packed-refs file, such a replacement will cause the old file to be garbage collected instead. A stale file handle exception will be raised on NFS servers if the file is garbage collected (deleted) on the server while it is being read. Since we no longer have access to the old file in these cases, the previous code would just fail. However, in these cases, reopening the file and rereading it will succeed (since it will reopen the new replacement file). So retrying the read is a viable strategy to deal with stale file handles on the packed-refs file, implement such a strategy. Since it is possible that the packed-refs file could be replaced again while rereading it (multiple consecutive updates can easily occur with ref deletions), loop on stale file handle exceptions, up to 5 extra times, trying to read the packed-refs file again, until we either read the new file, or find that the file no longer exists. The limit of 5 is arbitrary, and provides a safe upper bounds to prevent infinite loops consuming resources in a potential unforeseen persistent error condition. Change-Id: I085c472bafa6e2f32f610a33ddc8368bb4ab1814 Signed-off-by: Martin Fick<mfick@codeaurora.org> Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
8 years ago
Handle stale file handles on packed-refs file On a local filesystem the packed-refs file will be orphaned if it is replaced by another client while the current client is reading the old one. However, since NFS servers do not keep track of open files, instead of orphaning the old packed-refs file, such a replacement will cause the old file to be garbage collected instead. A stale file handle exception will be raised on NFS servers if the file is garbage collected (deleted) on the server while it is being read. Since we no longer have access to the old file in these cases, the previous code would just fail. However, in these cases, reopening the file and rereading it will succeed (since it will reopen the new replacement file). So retrying the read is a viable strategy to deal with stale file handles on the packed-refs file, implement such a strategy. Since it is possible that the packed-refs file could be replaced again while rereading it (multiple consecutive updates can easily occur with ref deletions), loop on stale file handle exceptions, up to 5 extra times, trying to read the packed-refs file again, until we either read the new file, or find that the file no longer exists. The limit of 5 is arbitrary, and provides a safe upper bounds to prevent infinite loops consuming resources in a potential unforeseen persistent error condition. Change-Id: I085c472bafa6e2f32f610a33ddc8368bb4ab1814 Signed-off-by: Martin Fick<mfick@codeaurora.org> Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
8 years ago
Handle stale file handles on packed-refs file On a local filesystem the packed-refs file will be orphaned if it is replaced by another client while the current client is reading the old one. However, since NFS servers do not keep track of open files, instead of orphaning the old packed-refs file, such a replacement will cause the old file to be garbage collected instead. A stale file handle exception will be raised on NFS servers if the file is garbage collected (deleted) on the server while it is being read. Since we no longer have access to the old file in these cases, the previous code would just fail. However, in these cases, reopening the file and rereading it will succeed (since it will reopen the new replacement file). So retrying the read is a viable strategy to deal with stale file handles on the packed-refs file, implement such a strategy. Since it is possible that the packed-refs file could be replaced again while rereading it (multiple consecutive updates can easily occur with ref deletions), loop on stale file handle exceptions, up to 5 extra times, trying to read the packed-refs file again, until we either read the new file, or find that the file no longer exists. The limit of 5 is arbitrary, and provides a safe upper bounds to prevent infinite loops consuming resources in a potential unforeseen persistent error condition. Change-Id: I085c472bafa6e2f32f610a33ddc8368bb4ab1814 Signed-off-by: Martin Fick<mfick@codeaurora.org> Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
8 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
RefDirectory: Throw exception if CAS of packed ref list fails The contents of the packedRefList AtomicReference should never differ from what we expect prior to writing, because this segment of the code is protected by the packed-refs lock file on disk. If it does happen, whether due to programmer error or a rogue process not respecting the locking protocol, it's better to let the caller know than to silently drop the whole commit operation on the floor. The existing concurrentOnlyOneWritesPackedRefs test is inherently nondeterministic as written, and was already about 6% flaky as measured by bazel: $ bazel test --runs_per_test=200 //org.eclipse.jgit.test:org_eclipse_jgit_internal_storage_file_GcPackRefsTest ... INFO: Elapsed time: 42.608s, Critical Path: 10.35s //org.eclipse.jgit.test:org_eclipse_jgit_internal_storage_file_GcPackRefsTest FAILED in 12 out of 200 in 1.6s Stats over 200 runs: max = 1.6s, min = 1.1s, avg = 1.3s, dev = 0.1s This flakiness was caused by the assumption that exactly one of the 2 threads would fail, when both might actually succeed in practice due to racing on the compare-and-swap. For whatever reason, this change affected the interleaving behavior in such a way that the flakiness jumped to around 50%. Making the interleaving of the test fully deterministic is beyond the scope of this change, but a simple tweak to the assertion is enough to make it pass consistently 200+ times both before and after this change. Change-Id: I5ff4dc39ee05bda88d47909acb70118f3d0c8f74
6 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Redo event listeners to be more generic Replace the old crude event listener system with a much more generic implementation, patterned after the event dispatch techniques used in Google Web Toolkit 1.5 and later. Each event delivers to an interface that defines a single method, and the event itself is what performs the delivery in a type-safe way through its own dispatch method. Listeners are registered in a generic listener list, indexed by the interface they implement and wish to receive an event for. Delivery of events is performed by looping through all listeners implementing the event's corresponding listener interface, and using the event's own dispatch method to deliver the event. This is the classical "double dispatch" pattern for event delivery. Listeners can be unregistered by invoking remove() on their registration handle. This change therefore requires application code to track the handle if it wishes to remove the listener at a later point in time. Event delivery is now exposed as a generic public method on the Repository class, making it easier for any type of message to be sent out to any type of listener that has registered, without needing to pre-arrange for type-safe fireFoo() methods. New event types can be added in the future simply by defining a new RepositoryEvent subclass and a corresponding RepositoryListener interface that it dispatches to. By always adding new events through a new interface, we never need to worry about defining an Adapter to provide default no-op implementations of new event methods. Change-Id: I651417b3098b9afc93d91085e9f0b2265df8fc81 Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
Rewrite reference handling to be abstract and accurate This commit actually does three major changes to the way references are handled within JGit. Unfortunately they were easier to do as a single massive commit than to break them up into smaller units. Disambiguate symbolic references: --------------------------------- Reporting a symbolic reference such as HEAD as though it were any other normal reference like refs/heads/master causes subtle programming errors. We have been bitten by this error on several occasions, as have some downstream applications written by myself. Instead of reporting HEAD as a reference whose name differs from its "original name", report it as an actual SymbolicRef object that the application can test the type and examine the target of. With this change, Ref is now an abstract type with different subclasses for the different types. In the classical example of "HEAD" being a symbolic reference to branch "refs/heads/master", the Repository.getAllRefs() method will now return: Map<String, Ref> all = repository.getAllRefs(); SymbolicRef HEAD = (SymbolicRef) all.get("HEAD"); ObjectIdRef master = (ObjectIdRef) all.get("refs/heads/master"); assertSame(master, HEAD.getTarget()); assertSame(master.getObjectId(), HEAD.getObjectId()); assertEquals("HEAD", HEAD.getName()); assertEquals("refs/heads/master", master.getName()); A nice side-effect of this change is the storage type of the symbolic reference is no longer ambiguous with the storge type of the underlying reference it targets. In the above example, if master was only available in the packed-refs file, then the following is also true: assertSame(Ref.Storage.LOOSE, HEAD.getStorage()); assertSame(Ref.Storage.PACKED, master.getStorage()); (Prior to this change we returned the ambiguous storage of LOOSE_PACKED for HEAD, which was confusing since it wasn't actually true on disk). Another nice side-effect of this change is all intermediate symbolic references are preserved, and are therefore visible to the application when they walk the target chain. We can now correctly inspect chains of symbolic references. As a result of this change the Ref.getOrigName() method has been removed from the API. Applications should identify a symbolic reference by testing for isSymbolic() and not by using an arcane string comparsion between properties. Abstract the RefDatabase storage: --------------------------------- RefDatabase is now abstract, similar to ObjectDatabase, and a new concrete implementation called RefDirectory is used for the traditional on-disk storage layout. In the future we plan to support additional implementations, such as a pure in-memory RefDatabase for unit testing purposes. Optimize RefDirectory: ---------------------- The implementation of the in-memory reference cache, reading, and update routines has been completely rewritten. Much of the code was heavily borrowed or cribbed from the prior implementation, so copyright notices have been left intact as much as possible. The RefDirectory cache no longer confuses symbolic references with normal references. This permits the cache to resolve the value of a symbolic reference as late as possible, ensuring it is always current, without needing to maintain reverse pointers. The cache is now 2 sorted RefLists, rather than 3 HashMaps. Using sorted lists allows the implementation to reduce the in-memory footprint when storing many refs. Using specialized types for the elements allows the code to avoid additional map lookups for auxiliary stat information. To improve scan time during getRefs(), the lists are returned via a copy-on-write contract. Most callers of getRefs() do not modify the returned collections, so the copy-on-write semantics improves access on repositories with a large number of packed references. Iterator traversals of the returned Map<String,Ref> are performed using a simple merge-join of the two cache lists, ensuring we can perform the entire traversal in linear time as a function of the number of references: O(PackedRefs + LooseRefs). Scans of the loose reference space to update the cache run in O(LooseRefs log LooseRefs) time, as the directory contents are sorted before being merged against the in-memory cache. Since the majority of stable references are kept packed, there typically are only a handful of reference names to be sorted, so the sorting cost should not be very high. Locking is reduced during getRefs() by taking advantage of the copy-on-write semantics of the improved cache data structure. This permits concurrent readers to pull back references without blocking each other. If there is contention updating the cache during a scan, one or more updates are simply skipped and will get picked up again in a future scan. Writing to the $GIT_DIR/packed-refs during reference delete is now fully atomic. The file is locked, reparsed fresh, and written back out if a change is necessary. This avoids all race conditions with concurrent external updates of the packed-refs file. The RefLogWriter class has been fully folded into RefDirectory and is therefore deleted. Maintaining the reference's log is the responsiblity of the database implementation, and not all implementations will use java.io for access. Future work still remains to be done to abstract the ReflogReader class away from local disk IO. Change-Id: I26b9287c45a4b2d2be35ba2849daa316f5eec85d Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years ago
12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061626364656667686970717273747576777879808182838485868788899091929394959697989910010110210310410510610710810911011111211311411511611711811912012112212312412512612712812913013113213313413513613713813914014114214314414514614714814915015115215315415515615715815916016116216316416516616716816917017117217317417517617717817918018118218318418518618718818919019119219319419519619719819920020120220320420520620720820921021121221321421521621721821922022122222322422522622722822923023123223323423523623723823924024124224324424524624724824925025125225325425525625725825926026126226326426526626726826927027127227327427527627727827928028128228328428528628728828929029129229329429529629729829930030130230330430530630730830931031131231331431531631731831932032132232332432532632732832933033133233333433533633733833934034134234334434534634734834935035135235335435535635735835936036136236336436536636736836937037137237337437537637737837938038138238338438538638738838939039139239339439539639739839940040140240340440540640740840941041141241341441541641741841942042142242342442542642742842943043143243343443543643743843944044144244344444544644744844945045145245345445545645745845946046146246346446546646746846947047147247347447547647747847948048148248348448548648748848949049149249349449549649749849950050150250350450550650750850951051151251351451551651751851952052152252352452552652752852953053153253353453553653753853954054154254354454554654754854955055155255355455555655755855956056156256356456556656756856957057157257357457557657757857958058158258358458558658758858959059159259359459559659759859960060160260360460560660760860961061161261361461561661761861962062162262362462562662762862963063163263363463563663763863964064164264364464564664764864965065165265365465565665765865966066166266366466566666766866967067167267367467567667767867968068168268368468568668768868969069169269369469569669769869970070170270370470570670770870971071171271371471571671771871972072172272372472572672772872973073173273373473573673773873974074174274374474574674774874975075175275375475575675775875976076176276376476576676776876977077177277377477577677777877978078178278378478578678778878979079179279379479579679779879980080180280380480580680780880981081181281381481581681781881982082182282382482582682782882983083183283383483583683783883984084184284384484584684784884985085185285385485585685785885986086186286386486586686786886987087187287387487587687787887988088188288388488588688788888989089189289389489589689789889990090190290390490590690790890991091191291391491591691791891992092192292392492592692792892993093193293393493593693793893994094194294394494594694794894995095195295395495595695795895996096196296396496596696796896997097197297397497597697797897998098198298398498598698798898999099199299399499599699799899910001001100210031004100510061007100810091010101110121013101410151016101710181019102010211022102310241025102610271028102910301031103210331034103510361037103810391040104110421043104410451046104710481049105010511052105310541055105610571058105910601061106210631064106510661067106810691070107110721073107410751076107710781079108010811082108310841085108610871088108910901091109210931094109510961097109810991100110111021103110411051106110711081109111011111112111311141115111611171118111911201121112211231124112511261127112811291130113111321133113411351136113711381139114011411142114311441145114611471148114911501151115211531154115511561157115811591160116111621163116411651166116711681169117011711172117311741175117611771178117911801181118211831184118511861187118811891190119111921193119411951196119711981199120012011202120312041205120612071208120912101211121212131214121512161217121812191220122112221223122412251226122712281229123012311232123312341235123612371238123912401241124212431244124512461247124812491250125112521253125412551256125712581259126012611262126312641265126612671268126912701271127212731274127512761277127812791280128112821283128412851286128712881289129012911292129312941295129612971298129913001301130213031304130513061307130813091310131113121313131413151316131713181319132013211322132313241325132613271328132913301331133213331334133513361337133813391340134113421343134413451346134713481349135013511352135313541355135613571358135913601361136213631364136513661367136813691370137113721373137413751376137713781379138013811382138313841385138613871388138913901391139213931394139513961397139813991400140114021403140414051406140714081409141014111412141314141415141614171418141914201421142214231424142514261427142814291430143114321433143414351436143714381439144014411442144314441445144614471448
  1. /*
  2. * Copyright (C) 2007, Dave Watson <dwatson@mimvista.com>
  3. * Copyright (C) 2009-2010, Google Inc.
  4. * Copyright (C) 2007, Robin Rosenberg <robin.rosenberg@dewire.com>
  5. * Copyright (C) 2006, Shawn O. Pearce <spearce@spearce.org> and others
  6. *
  7. * This program and the accompanying materials are made available under the
  8. * terms of the Eclipse Distribution License v. 1.0 which is available at
  9. * https://www.eclipse.org/org/documents/edl-v10.php.
  10. *
  11. * SPDX-License-Identifier: BSD-3-Clause
  12. */
  13. package org.eclipse.jgit.internal.storage.file;
  14. import static java.nio.charset.StandardCharsets.UTF_8;
  15. import static org.eclipse.jgit.lib.Constants.HEAD;
  16. import static org.eclipse.jgit.lib.Constants.LOGS;
  17. import static org.eclipse.jgit.lib.Constants.OBJECT_ID_STRING_LENGTH;
  18. import static org.eclipse.jgit.lib.Constants.PACKED_REFS;
  19. import static org.eclipse.jgit.lib.Constants.R_HEADS;
  20. import static org.eclipse.jgit.lib.Constants.R_REFS;
  21. import static org.eclipse.jgit.lib.Constants.R_TAGS;
  22. import static org.eclipse.jgit.lib.Ref.Storage.LOOSE;
  23. import static org.eclipse.jgit.lib.Ref.Storage.NEW;
  24. import static org.eclipse.jgit.lib.Ref.Storage.PACKED;
  25. import java.io.BufferedReader;
  26. import java.io.File;
  27. import java.io.FileInputStream;
  28. import java.io.FileNotFoundException;
  29. import java.io.IOException;
  30. import java.io.InputStreamReader;
  31. import java.io.InterruptedIOException;
  32. import java.nio.file.DirectoryNotEmptyException;
  33. import java.nio.file.Files;
  34. import java.nio.file.Path;
  35. import java.security.DigestInputStream;
  36. import java.security.MessageDigest;
  37. import java.text.MessageFormat;
  38. import java.util.Arrays;
  39. import java.util.Collection;
  40. import java.util.Collections;
  41. import java.util.HashMap;
  42. import java.util.LinkedList;
  43. import java.util.List;
  44. import java.util.Map;
  45. import java.util.concurrent.atomic.AtomicInteger;
  46. import java.util.concurrent.atomic.AtomicReference;
  47. import java.util.concurrent.locks.ReentrantLock;
  48. import java.util.stream.Stream;
  49. import org.eclipse.jgit.annotations.NonNull;
  50. import org.eclipse.jgit.annotations.Nullable;
  51. import org.eclipse.jgit.errors.InvalidObjectIdException;
  52. import org.eclipse.jgit.errors.LockFailedException;
  53. import org.eclipse.jgit.errors.MissingObjectException;
  54. import org.eclipse.jgit.errors.ObjectWritingException;
  55. import org.eclipse.jgit.events.RefsChangedEvent;
  56. import org.eclipse.jgit.internal.JGitText;
  57. import org.eclipse.jgit.lib.ConfigConstants;
  58. import org.eclipse.jgit.lib.Constants;
  59. import org.eclipse.jgit.lib.ObjectId;
  60. import org.eclipse.jgit.lib.ObjectIdRef;
  61. import org.eclipse.jgit.lib.Ref;
  62. import org.eclipse.jgit.lib.RefComparator;
  63. import org.eclipse.jgit.lib.RefDatabase;
  64. import org.eclipse.jgit.lib.RefUpdate;
  65. import org.eclipse.jgit.lib.RefWriter;
  66. import org.eclipse.jgit.lib.Repository;
  67. import org.eclipse.jgit.lib.SymbolicRef;
  68. import org.eclipse.jgit.revwalk.RevObject;
  69. import org.eclipse.jgit.revwalk.RevTag;
  70. import org.eclipse.jgit.revwalk.RevWalk;
  71. import org.eclipse.jgit.util.FS;
  72. import org.eclipse.jgit.util.FileUtils;
  73. import org.eclipse.jgit.util.IO;
  74. import org.eclipse.jgit.util.RawParseUtils;
  75. import org.eclipse.jgit.util.RefList;
  76. import org.eclipse.jgit.util.RefMap;
  77. import org.slf4j.Logger;
  78. import org.slf4j.LoggerFactory;
  79. /**
  80. * Traditional file system based {@link org.eclipse.jgit.lib.RefDatabase}.
  81. * <p>
  82. * This is the classical reference database representation for a Git repository.
  83. * References are stored in two formats: loose, and packed.
  84. * <p>
  85. * Loose references are stored as individual files within the {@code refs/}
  86. * directory. The file name matches the reference name and the file contents is
  87. * the current {@link org.eclipse.jgit.lib.ObjectId} in string form.
  88. * <p>
  89. * Packed references are stored in a single text file named {@code packed-refs}.
  90. * In the packed format, each reference is stored on its own line. This file
  91. * reduces the number of files needed for large reference spaces, reducing the
  92. * overall size of a Git repository on disk.
  93. */
  94. public class RefDirectory extends RefDatabase {
  95. private static final Logger LOG = LoggerFactory
  96. .getLogger(RefDirectory.class);
  97. /** Magic string denoting the start of a symbolic reference file. */
  98. public static final String SYMREF = "ref: "; //$NON-NLS-1$
  99. /** Magic string denoting the header of a packed-refs file. */
  100. public static final String PACKED_REFS_HEADER = "# pack-refs with:"; //$NON-NLS-1$
  101. /** If in the header, denotes the file has peeled data. */
  102. public static final String PACKED_REFS_PEELED = " peeled"; //$NON-NLS-1$
  103. /** The names of the additional refs supported by this class */
  104. private static final String[] additionalRefsNames = new String[] {
  105. Constants.MERGE_HEAD, Constants.FETCH_HEAD, Constants.ORIG_HEAD,
  106. Constants.CHERRY_PICK_HEAD };
  107. @SuppressWarnings("boxing")
  108. private static final List<Integer> RETRY_SLEEP_MS =
  109. Collections.unmodifiableList(Arrays.asList(0, 100, 200, 400, 800, 1600));
  110. private final FileRepository parent;
  111. private final File gitDir;
  112. final File refsDir;
  113. final File packedRefsFile;
  114. final File logsDir;
  115. final File logsRefsDir;
  116. /**
  117. * Immutable sorted list of loose references.
  118. * <p>
  119. * Symbolic references in this collection are stored unresolved, that is
  120. * their target appears to be a new reference with no ObjectId. These are
  121. * converted into resolved references during a get operation, ensuring the
  122. * live value is always returned.
  123. */
  124. private final AtomicReference<RefList<LooseRef>> looseRefs = new AtomicReference<>();
  125. /** Immutable sorted list of packed references. */
  126. final AtomicReference<PackedRefList> packedRefs = new AtomicReference<>();
  127. /**
  128. * Lock for coordinating operations within a single process that may contend
  129. * on the {@code packed-refs} file.
  130. * <p>
  131. * All operations that write {@code packed-refs} must still acquire a
  132. * {@link LockFile} on {@link #packedRefsFile}, even after they have acquired
  133. * this lock, since there may be multiple {@link RefDirectory} instances or
  134. * other processes operating on the same repo on disk.
  135. * <p>
  136. * This lock exists so multiple threads in the same process can wait in a fair
  137. * queue without trying, failing, and retrying to acquire the on-disk lock. If
  138. * {@code RepositoryCache} is used, this lock instance will be used by all
  139. * threads.
  140. */
  141. final ReentrantLock inProcessPackedRefsLock = new ReentrantLock(true);
  142. /**
  143. * Number of modifications made to this database.
  144. * <p>
  145. * This counter is incremented when a change is made, or detected from the
  146. * filesystem during a read operation.
  147. */
  148. private final AtomicInteger modCnt = new AtomicInteger();
  149. /**
  150. * Last {@link #modCnt} that we sent to listeners.
  151. * <p>
  152. * This value is compared to {@link #modCnt}, and a notification is sent to
  153. * the listeners only when it differs.
  154. */
  155. private final AtomicInteger lastNotifiedModCnt = new AtomicInteger();
  156. private List<Integer> retrySleepMs = RETRY_SLEEP_MS;
  157. RefDirectory(FileRepository db) {
  158. final FS fs = db.getFS();
  159. parent = db;
  160. gitDir = db.getDirectory();
  161. refsDir = fs.resolve(gitDir, R_REFS);
  162. logsDir = fs.resolve(gitDir, LOGS);
  163. logsRefsDir = fs.resolve(gitDir, LOGS + '/' + R_REFS);
  164. packedRefsFile = fs.resolve(gitDir, PACKED_REFS);
  165. looseRefs.set(RefList.<LooseRef> emptyList());
  166. packedRefs.set(NO_PACKED_REFS);
  167. }
  168. Repository getRepository() {
  169. return parent;
  170. }
  171. ReflogWriter newLogWriter(boolean force) {
  172. return new ReflogWriter(this, force);
  173. }
  174. /**
  175. * Locate the log file on disk for a single reference name.
  176. *
  177. * @param name
  178. * name of the ref, relative to the Git repository top level
  179. * directory (so typically starts with refs/).
  180. * @return the log file location.
  181. */
  182. public File logFor(String name) {
  183. if (name.startsWith(R_REFS)) {
  184. name = name.substring(R_REFS.length());
  185. return new File(logsRefsDir, name);
  186. }
  187. return new File(logsDir, name);
  188. }
  189. /** {@inheritDoc} */
  190. @Override
  191. public void create() throws IOException {
  192. FileUtils.mkdir(refsDir);
  193. FileUtils.mkdir(new File(refsDir, R_HEADS.substring(R_REFS.length())));
  194. FileUtils.mkdir(new File(refsDir, R_TAGS.substring(R_REFS.length())));
  195. newLogWriter(false).create();
  196. }
  197. /** {@inheritDoc} */
  198. @Override
  199. public void close() {
  200. clearReferences();
  201. }
  202. private void clearReferences() {
  203. looseRefs.set(RefList.<LooseRef> emptyList());
  204. packedRefs.set(NO_PACKED_REFS);
  205. }
  206. /** {@inheritDoc} */
  207. @Override
  208. public void refresh() {
  209. super.refresh();
  210. clearReferences();
  211. }
  212. /** {@inheritDoc} */
  213. @Override
  214. public boolean isNameConflicting(String name) throws IOException {
  215. // Cannot be nested within an existing reference.
  216. int lastSlash = name.lastIndexOf('/');
  217. while (0 < lastSlash) {
  218. String needle = name.substring(0, lastSlash);
  219. if (exactRef(needle) != null) {
  220. return true;
  221. }
  222. lastSlash = name.lastIndexOf('/', lastSlash - 1);
  223. }
  224. // Cannot be the container of an existing reference.
  225. return !getRefsByPrefix(name + '/').isEmpty();
  226. }
  227. @Nullable
  228. private Ref readAndResolve(String name, RefList<Ref> packed) throws IOException {
  229. try {
  230. Ref ref = readRef(name, packed);
  231. if (ref != null) {
  232. ref = resolve(ref, 0, null, null, packed);
  233. }
  234. return ref;
  235. } catch (IOException e) {
  236. if (name.contains("/") //$NON-NLS-1$
  237. || !(e.getCause() instanceof InvalidObjectIdException)) {
  238. throw e;
  239. }
  240. // While looking for a ref outside of refs/ (e.g., 'config'), we
  241. // found a non-ref file (e.g., a config file) instead. Treat this
  242. // as a ref-not-found condition.
  243. return null;
  244. }
  245. }
  246. /** {@inheritDoc} */
  247. @Override
  248. public Ref exactRef(String name) throws IOException {
  249. try {
  250. return readAndResolve(name, getPackedRefs());
  251. } finally {
  252. fireRefsChanged();
  253. }
  254. }
  255. /** {@inheritDoc} */
  256. @Override
  257. @NonNull
  258. public Map<String, Ref> exactRef(String... refs) throws IOException {
  259. try {
  260. RefList<Ref> packed = getPackedRefs();
  261. Map<String, Ref> result = new HashMap<>(refs.length);
  262. for (String name : refs) {
  263. Ref ref = readAndResolve(name, packed);
  264. if (ref != null) {
  265. result.put(name, ref);
  266. }
  267. }
  268. return result;
  269. } finally {
  270. fireRefsChanged();
  271. }
  272. }
  273. /** {@inheritDoc} */
  274. @Override
  275. @Nullable
  276. public Ref firstExactRef(String... refs) throws IOException {
  277. try {
  278. RefList<Ref> packed = getPackedRefs();
  279. for (String name : refs) {
  280. Ref ref = readAndResolve(name, packed);
  281. if (ref != null) {
  282. return ref;
  283. }
  284. }
  285. return null;
  286. } finally {
  287. fireRefsChanged();
  288. }
  289. }
  290. /** {@inheritDoc} */
  291. @Override
  292. public Map<String, Ref> getRefs(String prefix) throws IOException {
  293. final RefList<LooseRef> oldLoose = looseRefs.get();
  294. LooseScanner scan = new LooseScanner(oldLoose);
  295. scan.scan(prefix);
  296. final RefList<Ref> packed = getPackedRefs();
  297. RefList<LooseRef> loose;
  298. if (scan.newLoose != null) {
  299. scan.newLoose.sort();
  300. loose = scan.newLoose.toRefList();
  301. if (looseRefs.compareAndSet(oldLoose, loose))
  302. modCnt.incrementAndGet();
  303. } else
  304. loose = oldLoose;
  305. fireRefsChanged();
  306. RefList.Builder<Ref> symbolic = scan.symbolic;
  307. for (int idx = 0; idx < symbolic.size();) {
  308. final Ref symbolicRef = symbolic.get(idx);
  309. final Ref resolvedRef = resolve(symbolicRef, 0, prefix, loose, packed);
  310. if (resolvedRef != null && resolvedRef.getObjectId() != null) {
  311. symbolic.set(idx, resolvedRef);
  312. idx++;
  313. } else {
  314. // A broken symbolic reference, we have to drop it from the
  315. // collections the client is about to receive. Should be a
  316. // rare occurrence so pay a copy penalty.
  317. symbolic.remove(idx);
  318. final int toRemove = loose.find(symbolicRef.getName());
  319. if (0 <= toRemove)
  320. loose = loose.remove(toRemove);
  321. }
  322. }
  323. symbolic.sort();
  324. return new RefMap(prefix, packed, upcast(loose), symbolic.toRefList());
  325. }
  326. /** {@inheritDoc} */
  327. @Override
  328. public List<Ref> getAdditionalRefs() throws IOException {
  329. List<Ref> ret = new LinkedList<>();
  330. for (String name : additionalRefsNames) {
  331. Ref r = exactRef(name);
  332. if (r != null)
  333. ret.add(r);
  334. }
  335. return ret;
  336. }
  337. @SuppressWarnings("unchecked")
  338. private RefList<Ref> upcast(RefList<? extends Ref> loose) {
  339. return (RefList<Ref>) loose;
  340. }
  341. private class LooseScanner {
  342. private final RefList<LooseRef> curLoose;
  343. private int curIdx;
  344. final RefList.Builder<Ref> symbolic = new RefList.Builder<>(4);
  345. RefList.Builder<LooseRef> newLoose;
  346. LooseScanner(RefList<LooseRef> curLoose) {
  347. this.curLoose = curLoose;
  348. }
  349. void scan(String prefix) {
  350. if (ALL.equals(prefix)) {
  351. scanOne(HEAD);
  352. scanTree(R_REFS, refsDir);
  353. // If any entries remain, they are deleted, drop them.
  354. if (newLoose == null && curIdx < curLoose.size())
  355. newLoose = curLoose.copy(curIdx);
  356. } else if (prefix.startsWith(R_REFS) && prefix.endsWith("/")) { //$NON-NLS-1$
  357. curIdx = -(curLoose.find(prefix) + 1);
  358. File dir = new File(refsDir, prefix.substring(R_REFS.length()));
  359. scanTree(prefix, dir);
  360. // Skip over entries still within the prefix; these have
  361. // been removed from the directory.
  362. while (curIdx < curLoose.size()) {
  363. if (!curLoose.get(curIdx).getName().startsWith(prefix))
  364. break;
  365. if (newLoose == null)
  366. newLoose = curLoose.copy(curIdx);
  367. curIdx++;
  368. }
  369. // Keep any entries outside of the prefix space, we
  370. // do not know anything about their status.
  371. if (newLoose != null) {
  372. while (curIdx < curLoose.size())
  373. newLoose.add(curLoose.get(curIdx++));
  374. }
  375. }
  376. }
  377. private boolean scanTree(String prefix, File dir) {
  378. final String[] entries = dir.list(LockFile.FILTER);
  379. if (entries == null) // not a directory or an I/O error
  380. return false;
  381. if (0 < entries.length) {
  382. for (int i = 0; i < entries.length; ++i) {
  383. String e = entries[i];
  384. File f = new File(dir, e);
  385. if (f.isDirectory())
  386. entries[i] += '/';
  387. }
  388. Arrays.sort(entries);
  389. for (String name : entries) {
  390. if (name.charAt(name.length() - 1) == '/')
  391. scanTree(prefix + name, new File(dir, name));
  392. else
  393. scanOne(prefix + name);
  394. }
  395. }
  396. return true;
  397. }
  398. private void scanOne(String name) {
  399. LooseRef cur;
  400. if (curIdx < curLoose.size()) {
  401. do {
  402. cur = curLoose.get(curIdx);
  403. int cmp = RefComparator.compareTo(cur, name);
  404. if (cmp < 0) {
  405. // Reference is not loose anymore, its been deleted.
  406. // Skip the name in the new result list.
  407. if (newLoose == null)
  408. newLoose = curLoose.copy(curIdx);
  409. curIdx++;
  410. cur = null;
  411. continue;
  412. }
  413. if (cmp > 0) // Newly discovered loose reference.
  414. cur = null;
  415. break;
  416. } while (curIdx < curLoose.size());
  417. } else
  418. cur = null; // Newly discovered loose reference.
  419. LooseRef n;
  420. try {
  421. n = scanRef(cur, name);
  422. } catch (IOException notValid) {
  423. n = null;
  424. }
  425. if (n != null) {
  426. if (cur != n && newLoose == null)
  427. newLoose = curLoose.copy(curIdx);
  428. if (newLoose != null)
  429. newLoose.add(n);
  430. if (n.isSymbolic())
  431. symbolic.add(n);
  432. } else if (cur != null) {
  433. // Tragically, this file is no longer a loose reference.
  434. // Kill our cached entry of it.
  435. if (newLoose == null)
  436. newLoose = curLoose.copy(curIdx);
  437. }
  438. if (cur != null)
  439. curIdx++;
  440. }
  441. }
  442. /** {@inheritDoc} */
  443. @Override
  444. public Ref peel(Ref ref) throws IOException {
  445. final Ref leaf = ref.getLeaf();
  446. if (leaf.isPeeled() || leaf.getObjectId() == null)
  447. return ref;
  448. ObjectIdRef newLeaf = doPeel(leaf);
  449. // Try to remember this peeling in the cache, so we don't have to do
  450. // it again in the future, but only if the reference is unchanged.
  451. if (leaf.getStorage().isLoose()) {
  452. RefList<LooseRef> curList = looseRefs.get();
  453. int idx = curList.find(leaf.getName());
  454. if (0 <= idx && curList.get(idx) == leaf) {
  455. LooseRef asPeeled = ((LooseRef) leaf).peel(newLeaf);
  456. RefList<LooseRef> newList = curList.set(idx, asPeeled);
  457. looseRefs.compareAndSet(curList, newList);
  458. }
  459. }
  460. return recreate(ref, newLeaf);
  461. }
  462. private ObjectIdRef doPeel(Ref leaf) throws MissingObjectException,
  463. IOException {
  464. try (RevWalk rw = new RevWalk(getRepository())) {
  465. RevObject obj = rw.parseAny(leaf.getObjectId());
  466. if (obj instanceof RevTag) {
  467. return new ObjectIdRef.PeeledTag(leaf.getStorage(), leaf
  468. .getName(), leaf.getObjectId(), rw.peel(obj).copy());
  469. }
  470. return new ObjectIdRef.PeeledNonTag(leaf.getStorage(),
  471. leaf.getName(), leaf.getObjectId());
  472. }
  473. }
  474. private static Ref recreate(Ref old, ObjectIdRef leaf) {
  475. if (old.isSymbolic()) {
  476. Ref dst = recreate(old.getTarget(), leaf);
  477. return new SymbolicRef(old.getName(), dst);
  478. }
  479. return leaf;
  480. }
  481. void storedSymbolicRef(RefDirectoryUpdate u, FileSnapshot snapshot,
  482. String target) {
  483. putLooseRef(newSymbolicRef(snapshot, u.getRef().getName(), target));
  484. fireRefsChanged();
  485. }
  486. /** {@inheritDoc} */
  487. @Override
  488. public RefDirectoryUpdate newUpdate(String name, boolean detach)
  489. throws IOException {
  490. boolean detachingSymbolicRef = false;
  491. final RefList<Ref> packed = getPackedRefs();
  492. Ref ref = readRef(name, packed);
  493. if (ref != null)
  494. ref = resolve(ref, 0, null, null, packed);
  495. if (ref == null)
  496. ref = new ObjectIdRef.Unpeeled(NEW, name, null);
  497. else {
  498. detachingSymbolicRef = detach && ref.isSymbolic();
  499. }
  500. RefDirectoryUpdate refDirUpdate = new RefDirectoryUpdate(this, ref);
  501. if (detachingSymbolicRef)
  502. refDirUpdate.setDetachingSymbolicRef();
  503. return refDirUpdate;
  504. }
  505. /** {@inheritDoc} */
  506. @Override
  507. public RefDirectoryRename newRename(String fromName, String toName)
  508. throws IOException {
  509. RefDirectoryUpdate from = newUpdate(fromName, false);
  510. RefDirectoryUpdate to = newUpdate(toName, false);
  511. return new RefDirectoryRename(from, to);
  512. }
  513. /** {@inheritDoc} */
  514. @Override
  515. public PackedBatchRefUpdate newBatchUpdate() {
  516. return new PackedBatchRefUpdate(this);
  517. }
  518. /** {@inheritDoc} */
  519. @Override
  520. public boolean performsAtomicTransactions() {
  521. return true;
  522. }
  523. void stored(RefDirectoryUpdate update, FileSnapshot snapshot) {
  524. final ObjectId target = update.getNewObjectId().copy();
  525. final Ref leaf = update.getRef().getLeaf();
  526. putLooseRef(new LooseUnpeeled(snapshot, leaf.getName(), target));
  527. }
  528. private void putLooseRef(LooseRef ref) {
  529. RefList<LooseRef> cList, nList;
  530. do {
  531. cList = looseRefs.get();
  532. nList = cList.put(ref);
  533. } while (!looseRefs.compareAndSet(cList, nList));
  534. modCnt.incrementAndGet();
  535. fireRefsChanged();
  536. }
  537. void delete(RefDirectoryUpdate update) throws IOException {
  538. Ref dst = update.getRef();
  539. if (!update.isDetachingSymbolicRef()) {
  540. dst = dst.getLeaf();
  541. }
  542. String name = dst.getName();
  543. // Write the packed-refs file using an atomic update. We might
  544. // wind up reading it twice, before and after the lock, to ensure
  545. // we don't miss an edit made externally.
  546. final PackedRefList packed = getPackedRefs();
  547. if (packed.contains(name)) {
  548. inProcessPackedRefsLock.lock();
  549. try {
  550. LockFile lck = lockPackedRefsOrThrow();
  551. try {
  552. PackedRefList cur = readPackedRefs();
  553. int idx = cur.find(name);
  554. if (0 <= idx) {
  555. commitPackedRefs(lck, cur.remove(idx), packed, true);
  556. }
  557. } finally {
  558. lck.unlock();
  559. }
  560. } finally {
  561. inProcessPackedRefsLock.unlock();
  562. }
  563. }
  564. RefList<LooseRef> curLoose, newLoose;
  565. do {
  566. curLoose = looseRefs.get();
  567. int idx = curLoose.find(name);
  568. if (idx < 0)
  569. break;
  570. newLoose = curLoose.remove(idx);
  571. } while (!looseRefs.compareAndSet(curLoose, newLoose));
  572. int levels = levelsIn(name) - 2;
  573. delete(logFor(name), levels);
  574. if (dst.getStorage().isLoose()) {
  575. update.unlock();
  576. delete(fileFor(name), levels);
  577. }
  578. modCnt.incrementAndGet();
  579. fireRefsChanged();
  580. }
  581. /**
  582. * Adds a set of refs to the set of packed-refs. Only non-symbolic refs are
  583. * added. If a ref with the given name already existed in packed-refs it is
  584. * updated with the new value. Each loose ref which was added to the
  585. * packed-ref file is deleted. If a given ref can't be locked it will not be
  586. * added to the pack file.
  587. *
  588. * @param refs
  589. * the refs to be added. Must be fully qualified.
  590. * @throws java.io.IOException
  591. */
  592. public void pack(List<String> refs) throws IOException {
  593. pack(refs, Collections.emptyMap());
  594. }
  595. PackedRefList pack(Map<String, LockFile> heldLocks) throws IOException {
  596. return pack(heldLocks.keySet(), heldLocks);
  597. }
  598. private PackedRefList pack(Collection<String> refs,
  599. Map<String, LockFile> heldLocks) throws IOException {
  600. for (LockFile ol : heldLocks.values()) {
  601. ol.requireLock();
  602. }
  603. if (refs.isEmpty()) {
  604. return null;
  605. }
  606. FS fs = parent.getFS();
  607. // Lock the packed refs file and read the content
  608. inProcessPackedRefsLock.lock();
  609. try {
  610. LockFile lck = lockPackedRefsOrThrow();
  611. try {
  612. final PackedRefList packed = getPackedRefs();
  613. RefList<Ref> cur = readPackedRefs();
  614. // Iterate over all refs to be packed
  615. boolean dirty = false;
  616. for (String refName : refs) {
  617. Ref oldRef = readRef(refName, cur);
  618. if (oldRef == null) {
  619. continue; // A non-existent ref is already correctly packed.
  620. }
  621. if (oldRef.isSymbolic()) {
  622. continue; // can't pack symbolic refs
  623. }
  624. // Add/Update it to packed-refs
  625. Ref newRef = peeledPackedRef(oldRef);
  626. if (newRef == oldRef) {
  627. // No-op; peeledPackedRef returns the input ref only if it's already
  628. // packed, and readRef returns a packed ref only if there is no
  629. // loose ref.
  630. continue;
  631. }
  632. dirty = true;
  633. int idx = cur.find(refName);
  634. if (idx >= 0) {
  635. cur = cur.set(idx, newRef);
  636. } else {
  637. cur = cur.add(idx, newRef);
  638. }
  639. }
  640. if (!dirty) {
  641. // All requested refs were already packed accurately
  642. return packed;
  643. }
  644. // The new content for packed-refs is collected. Persist it.
  645. PackedRefList result = commitPackedRefs(lck, cur, packed,
  646. false);
  647. // Now delete the loose refs which are now packed
  648. for (String refName : refs) {
  649. // Lock the loose ref
  650. File refFile = fileFor(refName);
  651. if (!fs.exists(refFile)) {
  652. continue;
  653. }
  654. LockFile rLck = heldLocks.get(refName);
  655. boolean shouldUnlock;
  656. if (rLck == null) {
  657. rLck = new LockFile(refFile);
  658. if (!rLck.lock()) {
  659. continue;
  660. }
  661. shouldUnlock = true;
  662. } else {
  663. shouldUnlock = false;
  664. }
  665. try {
  666. LooseRef currentLooseRef = scanRef(null, refName);
  667. if (currentLooseRef == null || currentLooseRef.isSymbolic()) {
  668. continue;
  669. }
  670. Ref packedRef = cur.get(refName);
  671. ObjectId clr_oid = currentLooseRef.getObjectId();
  672. if (clr_oid != null
  673. && clr_oid.equals(packedRef.getObjectId())) {
  674. RefList<LooseRef> curLoose, newLoose;
  675. do {
  676. curLoose = looseRefs.get();
  677. int idx = curLoose.find(refName);
  678. if (idx < 0) {
  679. break;
  680. }
  681. newLoose = curLoose.remove(idx);
  682. } while (!looseRefs.compareAndSet(curLoose, newLoose));
  683. int levels = levelsIn(refName) - 2;
  684. delete(refFile, levels, rLck);
  685. }
  686. } finally {
  687. if (shouldUnlock) {
  688. rLck.unlock();
  689. }
  690. }
  691. }
  692. // Don't fire refsChanged. The refs have not change, only their
  693. // storage.
  694. return result;
  695. } finally {
  696. lck.unlock();
  697. }
  698. } finally {
  699. inProcessPackedRefsLock.unlock();
  700. }
  701. }
  702. @Nullable
  703. LockFile lockPackedRefs() throws IOException {
  704. LockFile lck = new LockFile(packedRefsFile);
  705. for (int ms : getRetrySleepMs()) {
  706. sleep(ms);
  707. if (lck.lock()) {
  708. return lck;
  709. }
  710. }
  711. return null;
  712. }
  713. private LockFile lockPackedRefsOrThrow() throws IOException {
  714. LockFile lck = lockPackedRefs();
  715. if (lck == null) {
  716. throw new LockFailedException(packedRefsFile);
  717. }
  718. return lck;
  719. }
  720. /**
  721. * Make sure a ref is peeled and has the Storage PACKED. If the given ref
  722. * has this attributes simply return it. Otherwise create a new peeled
  723. * {@link ObjectIdRef} where Storage is set to PACKED.
  724. *
  725. * @param f
  726. * @return a ref for Storage PACKED having the same name, id, peeledId as f
  727. * @throws MissingObjectException
  728. * @throws IOException
  729. */
  730. private Ref peeledPackedRef(Ref f)
  731. throws MissingObjectException, IOException {
  732. if (f.getStorage().isPacked() && f.isPeeled()) {
  733. return f;
  734. }
  735. if (!f.isPeeled()) {
  736. f = peel(f);
  737. }
  738. ObjectId peeledObjectId = f.getPeeledObjectId();
  739. if (peeledObjectId != null) {
  740. return new ObjectIdRef.PeeledTag(PACKED, f.getName(),
  741. f.getObjectId(), peeledObjectId);
  742. }
  743. return new ObjectIdRef.PeeledNonTag(PACKED, f.getName(),
  744. f.getObjectId());
  745. }
  746. void log(boolean force, RefUpdate update, String msg, boolean deref)
  747. throws IOException {
  748. newLogWriter(force).log(update, msg, deref);
  749. }
  750. private Ref resolve(final Ref ref, int depth, String prefix,
  751. RefList<LooseRef> loose, RefList<Ref> packed) throws IOException {
  752. if (ref.isSymbolic()) {
  753. Ref dst = ref.getTarget();
  754. if (MAX_SYMBOLIC_REF_DEPTH <= depth)
  755. return null; // claim it doesn't exist
  756. // If the cached value can be assumed to be current due to a
  757. // recent scan of the loose directory, use it.
  758. if (loose != null && dst.getName().startsWith(prefix)) {
  759. int idx;
  760. if (0 <= (idx = loose.find(dst.getName())))
  761. dst = loose.get(idx);
  762. else if (0 <= (idx = packed.find(dst.getName())))
  763. dst = packed.get(idx);
  764. else
  765. return ref;
  766. } else {
  767. dst = readRef(dst.getName(), packed);
  768. if (dst == null)
  769. return ref;
  770. }
  771. dst = resolve(dst, depth + 1, prefix, loose, packed);
  772. if (dst == null)
  773. return null;
  774. return new SymbolicRef(ref.getName(), dst);
  775. }
  776. return ref;
  777. }
  778. PackedRefList getPackedRefs() throws IOException {
  779. boolean trustFolderStat = getRepository().getConfig().getBoolean(
  780. ConfigConstants.CONFIG_CORE_SECTION,
  781. ConfigConstants.CONFIG_KEY_TRUSTFOLDERSTAT, true);
  782. final PackedRefList curList = packedRefs.get();
  783. if (trustFolderStat && !curList.snapshot.isModified(packedRefsFile)) {
  784. return curList;
  785. }
  786. final PackedRefList newList = readPackedRefs();
  787. if (packedRefs.compareAndSet(curList, newList)
  788. && !curList.id.equals(newList.id)) {
  789. modCnt.incrementAndGet();
  790. }
  791. return newList;
  792. }
  793. private PackedRefList readPackedRefs() throws IOException {
  794. int maxStaleRetries = 5;
  795. int retries = 0;
  796. while (true) {
  797. final FileSnapshot snapshot = FileSnapshot.save(packedRefsFile);
  798. final MessageDigest digest = Constants.newMessageDigest();
  799. try (BufferedReader br = new BufferedReader(new InputStreamReader(
  800. new DigestInputStream(new FileInputStream(packedRefsFile),
  801. digest),
  802. UTF_8))) {
  803. try {
  804. return new PackedRefList(parsePackedRefs(br), snapshot,
  805. ObjectId.fromRaw(digest.digest()));
  806. } catch (IOException e) {
  807. if (FileUtils.isStaleFileHandleInCausalChain(e)
  808. && retries < maxStaleRetries) {
  809. if (LOG.isDebugEnabled()) {
  810. LOG.debug(MessageFormat.format(
  811. JGitText.get().packedRefsHandleIsStale,
  812. Integer.valueOf(retries)), e);
  813. }
  814. retries++;
  815. continue;
  816. }
  817. throw e;
  818. }
  819. } catch (FileNotFoundException noPackedRefs) {
  820. if (packedRefsFile.exists()) {
  821. throw noPackedRefs;
  822. }
  823. // Ignore it and leave the new list empty.
  824. return NO_PACKED_REFS;
  825. }
  826. }
  827. }
  828. private RefList<Ref> parsePackedRefs(BufferedReader br)
  829. throws IOException {
  830. RefList.Builder<Ref> all = new RefList.Builder<>();
  831. Ref last = null;
  832. boolean peeled = false;
  833. boolean needSort = false;
  834. String p;
  835. while ((p = br.readLine()) != null) {
  836. if (p.charAt(0) == '#') {
  837. if (p.startsWith(PACKED_REFS_HEADER)) {
  838. p = p.substring(PACKED_REFS_HEADER.length());
  839. peeled = p.contains(PACKED_REFS_PEELED);
  840. }
  841. continue;
  842. }
  843. if (p.charAt(0) == '^') {
  844. if (last == null)
  845. throw new IOException(JGitText.get().peeledLineBeforeRef);
  846. ObjectId id = ObjectId.fromString(p.substring(1));
  847. last = new ObjectIdRef.PeeledTag(PACKED, last.getName(), last
  848. .getObjectId(), id);
  849. all.set(all.size() - 1, last);
  850. continue;
  851. }
  852. int sp = p.indexOf(' ');
  853. if (sp < 0) {
  854. throw new IOException(MessageFormat.format(
  855. JGitText.get().packedRefsCorruptionDetected,
  856. packedRefsFile.getAbsolutePath()));
  857. }
  858. ObjectId id = ObjectId.fromString(p.substring(0, sp));
  859. String name = copy(p, sp + 1, p.length());
  860. ObjectIdRef cur;
  861. if (peeled)
  862. cur = new ObjectIdRef.PeeledNonTag(PACKED, name, id);
  863. else
  864. cur = new ObjectIdRef.Unpeeled(PACKED, name, id);
  865. if (last != null && RefComparator.compareTo(last, cur) > 0)
  866. needSort = true;
  867. all.add(cur);
  868. last = cur;
  869. }
  870. if (needSort)
  871. all.sort();
  872. return all.toRefList();
  873. }
  874. private static String copy(String src, int off, int end) {
  875. // Don't use substring since it could leave a reference to the much
  876. // larger existing string. Force construction of a full new object.
  877. return new StringBuilder(end - off).append(src, off, end).toString();
  878. }
  879. PackedRefList commitPackedRefs(final LockFile lck, final RefList<Ref> refs,
  880. final PackedRefList oldPackedList, boolean changed)
  881. throws IOException {
  882. // Can't just return packedRefs.get() from this method; it might have been
  883. // updated again after writePackedRefs() returns.
  884. AtomicReference<PackedRefList> result = new AtomicReference<>();
  885. new RefWriter(refs) {
  886. @Override
  887. protected void writeFile(String name, byte[] content)
  888. throws IOException {
  889. lck.setFSync(true);
  890. lck.setNeedSnapshot(true);
  891. try {
  892. lck.write(content);
  893. } catch (IOException ioe) {
  894. throw new ObjectWritingException(MessageFormat.format(JGitText.get().unableToWrite, name), ioe);
  895. }
  896. try {
  897. lck.waitForStatChange();
  898. } catch (InterruptedException e) {
  899. lck.unlock();
  900. throw new ObjectWritingException(
  901. MessageFormat.format(
  902. JGitText.get().interruptedWriting, name),
  903. e);
  904. }
  905. if (!lck.commit())
  906. throw new ObjectWritingException(MessageFormat.format(JGitText.get().unableToWrite, name));
  907. byte[] digest = Constants.newMessageDigest().digest(content);
  908. PackedRefList newPackedList = new PackedRefList(
  909. refs, lck.getCommitSnapshot(), ObjectId.fromRaw(digest));
  910. // This thread holds the file lock, so no other thread or process should
  911. // be able to modify the packed-refs file on disk. If the list changed,
  912. // it means something is very wrong, so throw an exception.
  913. //
  914. // However, we can't use a naive compareAndSet to check whether the
  915. // update was successful, because another thread might _read_ the
  916. // packed refs file that was written out by this thread while holding
  917. // the lock, and update the packedRefs reference to point to that. So
  918. // compare the actual contents instead.
  919. PackedRefList afterUpdate = packedRefs.updateAndGet(
  920. p -> p.id.equals(oldPackedList.id) ? newPackedList : p);
  921. if (!afterUpdate.id.equals(newPackedList.id)) {
  922. throw new ObjectWritingException(
  923. MessageFormat.format(JGitText.get().unableToWrite, name));
  924. }
  925. if (changed) {
  926. modCnt.incrementAndGet();
  927. }
  928. result.set(newPackedList);
  929. }
  930. }.writePackedRefs();
  931. return result.get();
  932. }
  933. private Ref readRef(String name, RefList<Ref> packed) throws IOException {
  934. final RefList<LooseRef> curList = looseRefs.get();
  935. final int idx = curList.find(name);
  936. if (0 <= idx) {
  937. final LooseRef o = curList.get(idx);
  938. final LooseRef n = scanRef(o, name);
  939. if (n == null) {
  940. if (looseRefs.compareAndSet(curList, curList.remove(idx)))
  941. modCnt.incrementAndGet();
  942. return packed.get(name);
  943. }
  944. if (o == n)
  945. return n;
  946. if (looseRefs.compareAndSet(curList, curList.set(idx, n)))
  947. modCnt.incrementAndGet();
  948. return n;
  949. }
  950. final LooseRef n = scanRef(null, name);
  951. if (n == null)
  952. return packed.get(name);
  953. // check whether the found new ref is the an additional ref. These refs
  954. // should not go into looseRefs
  955. for (String additionalRefsName : additionalRefsNames) {
  956. if (name.equals(additionalRefsName)) {
  957. return n;
  958. }
  959. }
  960. if (looseRefs.compareAndSet(curList, curList.add(idx, n)))
  961. modCnt.incrementAndGet();
  962. return n;
  963. }
  964. LooseRef scanRef(LooseRef ref, String name) throws IOException {
  965. final File path = fileFor(name);
  966. FileSnapshot currentSnapshot = null;
  967. if (ref != null) {
  968. currentSnapshot = ref.getSnapShot();
  969. if (!currentSnapshot.isModified(path))
  970. return ref;
  971. name = ref.getName();
  972. }
  973. final int limit = 4096;
  974. final byte[] buf;
  975. FileSnapshot otherSnapshot = FileSnapshot.save(path);
  976. try {
  977. buf = IO.readSome(path, limit);
  978. } catch (FileNotFoundException noFile) {
  979. if (path.exists() && path.isFile()) {
  980. throw noFile;
  981. }
  982. return null; // doesn't exist or no file; not a reference.
  983. }
  984. int n = buf.length;
  985. if (n == 0)
  986. return null; // empty file; not a reference.
  987. if (isSymRef(buf, n)) {
  988. if (n == limit)
  989. return null; // possibly truncated ref
  990. // trim trailing whitespace
  991. while (0 < n && Character.isWhitespace(buf[n - 1]))
  992. n--;
  993. if (n < 6) {
  994. String content = RawParseUtils.decode(buf, 0, n);
  995. throw new IOException(MessageFormat.format(JGitText.get().notARef, name, content));
  996. }
  997. final String target = RawParseUtils.decode(buf, 5, n);
  998. if (ref != null && ref.isSymbolic()
  999. && ref.getTarget().getName().equals(target)) {
  1000. assert(currentSnapshot != null);
  1001. currentSnapshot.setClean(otherSnapshot);
  1002. return ref;
  1003. }
  1004. return newSymbolicRef(otherSnapshot, name, target);
  1005. }
  1006. if (n < OBJECT_ID_STRING_LENGTH)
  1007. return null; // impossibly short object identifier; not a reference.
  1008. final ObjectId id;
  1009. try {
  1010. id = ObjectId.fromString(buf, 0);
  1011. if (ref != null && !ref.isSymbolic()
  1012. && id.equals(ref.getTarget().getObjectId())) {
  1013. assert(currentSnapshot != null);
  1014. currentSnapshot.setClean(otherSnapshot);
  1015. return ref;
  1016. }
  1017. } catch (IllegalArgumentException notRef) {
  1018. while (0 < n && Character.isWhitespace(buf[n - 1]))
  1019. n--;
  1020. String content = RawParseUtils.decode(buf, 0, n);
  1021. throw new IOException(MessageFormat.format(JGitText.get().notARef,
  1022. name, content), notRef);
  1023. }
  1024. return new LooseUnpeeled(otherSnapshot, name, id);
  1025. }
  1026. private static boolean isSymRef(byte[] buf, int n) {
  1027. if (n < 6)
  1028. return false;
  1029. return /**/buf[0] == 'r' //
  1030. && buf[1] == 'e' //
  1031. && buf[2] == 'f' //
  1032. && buf[3] == ':' //
  1033. && buf[4] == ' ';
  1034. }
  1035. /**
  1036. * Detect if we are in a clone command execution
  1037. *
  1038. * @return {@code true} if we are currently cloning a repository
  1039. * @throws IOException
  1040. */
  1041. boolean isInClone() throws IOException {
  1042. return hasDanglingHead() && !packedRefsFile.exists() && !hasLooseRef();
  1043. }
  1044. private boolean hasDanglingHead() throws IOException {
  1045. Ref head = exactRef(Constants.HEAD);
  1046. if (head != null) {
  1047. ObjectId id = head.getObjectId();
  1048. return id == null || id.equals(ObjectId.zeroId());
  1049. }
  1050. return false;
  1051. }
  1052. private boolean hasLooseRef() throws IOException {
  1053. try (Stream<Path> stream = Files.walk(refsDir.toPath())) {
  1054. return stream.anyMatch(Files::isRegularFile);
  1055. }
  1056. }
  1057. /** If the parent should fire listeners, fires them. */
  1058. void fireRefsChanged() {
  1059. final int last = lastNotifiedModCnt.get();
  1060. final int curr = modCnt.get();
  1061. if (last != curr && lastNotifiedModCnt.compareAndSet(last, curr) && last != 0)
  1062. parent.fireEvent(new RefsChangedEvent());
  1063. }
  1064. /**
  1065. * Create a reference update to write a temporary reference.
  1066. *
  1067. * @return an update for a new temporary reference.
  1068. * @throws IOException
  1069. * a temporary name cannot be allocated.
  1070. */
  1071. RefDirectoryUpdate newTemporaryUpdate() throws IOException {
  1072. File tmp = File.createTempFile("renamed_", "_ref", refsDir); //$NON-NLS-1$ //$NON-NLS-2$
  1073. String name = Constants.R_REFS + tmp.getName();
  1074. Ref ref = new ObjectIdRef.Unpeeled(NEW, name, null);
  1075. return new RefDirectoryUpdate(this, ref);
  1076. }
  1077. /**
  1078. * Locate the file on disk for a single reference name.
  1079. *
  1080. * @param name
  1081. * name of the ref, relative to the Git repository top level
  1082. * directory (so typically starts with refs/).
  1083. * @return the loose file location.
  1084. */
  1085. File fileFor(String name) {
  1086. if (name.startsWith(R_REFS)) {
  1087. name = name.substring(R_REFS.length());
  1088. return new File(refsDir, name);
  1089. }
  1090. return new File(gitDir, name);
  1091. }
  1092. static int levelsIn(String name) {
  1093. int count = 0;
  1094. for (int p = name.indexOf('/'); p >= 0; p = name.indexOf('/', p + 1))
  1095. count++;
  1096. return count;
  1097. }
  1098. static void delete(File file, int depth) throws IOException {
  1099. delete(file, depth, null);
  1100. }
  1101. private static void delete(File file, int depth, LockFile rLck)
  1102. throws IOException {
  1103. if (!file.delete() && file.isFile()) {
  1104. throw new IOException(MessageFormat.format(
  1105. JGitText.get().fileCannotBeDeleted, file));
  1106. }
  1107. if (rLck != null) {
  1108. rLck.unlock(); // otherwise cannot delete dir below
  1109. }
  1110. File dir = file.getParentFile();
  1111. for (int i = 0; i < depth; ++i) {
  1112. try {
  1113. Files.deleteIfExists(dir.toPath());
  1114. } catch (DirectoryNotEmptyException e) {
  1115. // Don't log; normal case when there are other refs with the
  1116. // same prefix
  1117. break;
  1118. } catch (IOException e) {
  1119. LOG.warn(MessageFormat.format(JGitText.get().unableToRemovePath,
  1120. dir), e);
  1121. break;
  1122. }
  1123. dir = dir.getParentFile();
  1124. }
  1125. }
  1126. /**
  1127. * Get times to sleep while retrying a possibly contentious operation.
  1128. * <p>
  1129. * For retrying an operation that might have high contention, such as locking
  1130. * the {@code packed-refs} file, the caller may implement a retry loop using
  1131. * the returned values:
  1132. *
  1133. * <pre>
  1134. * for (int toSleepMs : getRetrySleepMs()) {
  1135. * sleep(toSleepMs);
  1136. * if (isSuccessful(doSomething())) {
  1137. * return success;
  1138. * }
  1139. * }
  1140. * return failure;
  1141. * </pre>
  1142. *
  1143. * The first value in the returned iterable is 0, and the caller should treat
  1144. * a fully-consumed iterator as a timeout.
  1145. *
  1146. * @return iterable of times, in milliseconds, that the caller should sleep
  1147. * before attempting an operation.
  1148. */
  1149. Iterable<Integer> getRetrySleepMs() {
  1150. return retrySleepMs;
  1151. }
  1152. void setRetrySleepMs(List<Integer> retrySleepMs) {
  1153. if (retrySleepMs == null || retrySleepMs.isEmpty()
  1154. || retrySleepMs.get(0).intValue() != 0) {
  1155. throw new IllegalArgumentException();
  1156. }
  1157. this.retrySleepMs = retrySleepMs;
  1158. }
  1159. /**
  1160. * Sleep with {@link Thread#sleep(long)}, converting {@link
  1161. * InterruptedException} to {@link InterruptedIOException}.
  1162. *
  1163. * @param ms
  1164. * time to sleep, in milliseconds; zero or negative is a no-op.
  1165. * @throws InterruptedIOException
  1166. * if sleeping was interrupted.
  1167. */
  1168. static void sleep(long ms) throws InterruptedIOException {
  1169. if (ms <= 0) {
  1170. return;
  1171. }
  1172. try {
  1173. Thread.sleep(ms);
  1174. } catch (InterruptedException e) {
  1175. InterruptedIOException ie = new InterruptedIOException();
  1176. ie.initCause(e);
  1177. throw ie;
  1178. }
  1179. }
  1180. static class PackedRefList extends RefList<Ref> {
  1181. private final FileSnapshot snapshot;
  1182. private final ObjectId id;
  1183. private PackedRefList(RefList<Ref> src, FileSnapshot s, ObjectId i) {
  1184. super(src);
  1185. snapshot = s;
  1186. id = i;
  1187. }
  1188. }
  1189. private static final PackedRefList NO_PACKED_REFS = new PackedRefList(
  1190. RefList.emptyList(), FileSnapshot.MISSING_FILE,
  1191. ObjectId.zeroId());
  1192. private static LooseSymbolicRef newSymbolicRef(FileSnapshot snapshot,
  1193. String name, String target) {
  1194. Ref dst = new ObjectIdRef.Unpeeled(NEW, target, null);
  1195. return new LooseSymbolicRef(snapshot, name, dst);
  1196. }
  1197. private static interface LooseRef extends Ref {
  1198. FileSnapshot getSnapShot();
  1199. LooseRef peel(ObjectIdRef newLeaf);
  1200. }
  1201. private static final class LoosePeeledTag extends ObjectIdRef.PeeledTag
  1202. implements LooseRef {
  1203. private final FileSnapshot snapShot;
  1204. LoosePeeledTag(FileSnapshot snapshot, @NonNull String refName,
  1205. @NonNull ObjectId id, @NonNull ObjectId p) {
  1206. super(LOOSE, refName, id, p);
  1207. this.snapShot = snapshot;
  1208. }
  1209. @Override
  1210. public FileSnapshot getSnapShot() {
  1211. return snapShot;
  1212. }
  1213. @Override
  1214. public LooseRef peel(ObjectIdRef newLeaf) {
  1215. return this;
  1216. }
  1217. }
  1218. private static final class LooseNonTag extends ObjectIdRef.PeeledNonTag
  1219. implements LooseRef {
  1220. private final FileSnapshot snapShot;
  1221. LooseNonTag(FileSnapshot snapshot, @NonNull String refName,
  1222. @NonNull ObjectId id) {
  1223. super(LOOSE, refName, id);
  1224. this.snapShot = snapshot;
  1225. }
  1226. @Override
  1227. public FileSnapshot getSnapShot() {
  1228. return snapShot;
  1229. }
  1230. @Override
  1231. public LooseRef peel(ObjectIdRef newLeaf) {
  1232. return this;
  1233. }
  1234. }
  1235. private static final class LooseUnpeeled extends ObjectIdRef.Unpeeled
  1236. implements LooseRef {
  1237. private FileSnapshot snapShot;
  1238. LooseUnpeeled(FileSnapshot snapShot, @NonNull String refName,
  1239. @NonNull ObjectId id) {
  1240. super(LOOSE, refName, id);
  1241. this.snapShot = snapShot;
  1242. }
  1243. @Override
  1244. public FileSnapshot getSnapShot() {
  1245. return snapShot;
  1246. }
  1247. @NonNull
  1248. @Override
  1249. public ObjectId getObjectId() {
  1250. ObjectId id = super.getObjectId();
  1251. assert id != null; // checked in constructor
  1252. return id;
  1253. }
  1254. @Override
  1255. public LooseRef peel(ObjectIdRef newLeaf) {
  1256. ObjectId peeledObjectId = newLeaf.getPeeledObjectId();
  1257. ObjectId objectId = getObjectId();
  1258. if (peeledObjectId != null) {
  1259. return new LoosePeeledTag(snapShot, getName(),
  1260. objectId, peeledObjectId);
  1261. }
  1262. return new LooseNonTag(snapShot, getName(), objectId);
  1263. }
  1264. }
  1265. private static final class LooseSymbolicRef extends SymbolicRef implements
  1266. LooseRef {
  1267. private final FileSnapshot snapShot;
  1268. LooseSymbolicRef(FileSnapshot snapshot, @NonNull String refName,
  1269. @NonNull Ref target) {
  1270. super(refName, target);
  1271. this.snapShot = snapshot;
  1272. }
  1273. @Override
  1274. public FileSnapshot getSnapShot() {
  1275. return snapShot;
  1276. }
  1277. @Override
  1278. public LooseRef peel(ObjectIdRef newLeaf) {
  1279. // We should never try to peel the symbolic references.
  1280. throw new UnsupportedOperationException();
  1281. }
  1282. }
  1283. }