If we hit a corrupt object or invalid pack remove the pack from the pack
list. Other IOException could be transient hence we should not remove
the pack from the list to avoid the problem reported on the Gerrit list
[1]. It looks like in the reported case the pack was removed from the
pack list causing MissingObjectExceptions which disappear when the
server is restarted.

[1] https://groups.google.com/forum/#!topic/repo-discuss/Qdmbl-YZ4NU

Change-Id: I331626110d54b190e46cddc2c40f29ddeb9613cd
Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>

This should help to identify the root cause of the problem discussed on
the Gerrit list [1].

[1] https://groups.google.com/forum/#!topic/repo-discuss/Qdmbl-YZ4NU

Change-Id: I871f70e4bb1227952e1544b789013583b14e2b96
Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>

Native git's "init" command allows to specify the location of the .git
folder with the option "--separate-git-dir". This allows for example to
setup repositories with a non-standard layout. E.g. .git folder under
/repos/a.git and the worktree under /home/git/a. Both directories
contain pointers to the other side: /repos/a.git/config contains
core.worktree=/home/git/a . And /home/git/a/.git is a file containing
"gitdir: /repos/a.git". This commit adds that option to InitCommand.
This feature is needed to support the new submodule layout where the
.git folder of the submodules is under .git/modules/<submodule>.

Change-Id: I0208f643808bf8f28e2c979d6e33662607775f1f

ObjectDirectory.searchPacksAgain() should always read trustFolderStat
from the config and not rely on a cached value.

Change-Id: I90edbaae3c64eea0c9894d05acde4267991575ee

JGit's ObjectDirectory implements the optimization that it remembers the
pack folders (.git/objects/pack) lastModified timestamp and doesn't
check for new packfiles in this folder if the lastModified attribute has
not changed.
In environments using NFS this can cause trouble. If multiple JGit
instances from multiple machines work on the same repository and one
instance creates a new ref and a new packfile (e.g. by doing a fetch)
then the other machines may detect the new ref but can't resolve the
referenced object because it doesn't detect that pack folder has a new
packfile. That's because NFS may cache file/folder metadata for quite a
long time and the pack folders modification time is not updated although
a new packfile is there and could be read.

The new config parameter core.trustfolderstat controls this behaviour.
The default is true and jgits behaviours is unchanged. But if this
parameter is set to false then jgit doesn't trust the pack directories
lastmodified anymore. Instead it will always iterate through the content
of that folder to detect new packfiles.

Change-Id: Ie3b4e92933286aa9916070a22422e629b3147f54
Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>

This problem was raised on the jgit list
https://dev.eclipse.org/mhonarc/lists/jgit-dev/msg02675.html

Change-Id: I8ce8a159120ce3747d31d6f012c6d2c0d0ce7917
Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>

When writing new packs it should be allowed to specify objects as "have"
(objects which should not be included in the pack) which do not exist in
the local repository.

This works with the traditional PackWriter, but when PackWriter was
working on a repository with bitmap indexes and used
PackWriterBitmapWalker then this feature was broken. Non-existing "have"
objects lead to MissingObjectExceptions. That broke push and Gerrit
replication. When the replication target had branches unknown to the
replication source then the source repository wanted to build pack files
where "have" included branch-tips which were unknown in the source 
repository.

Bug: 427107
Change-Id: I6b6598a1ec49af68aa77ea6f1f06e827982ea4ac
Also-by: Matthias Sohn <matthias.sohn@sap.com>

JGit should offer the possibility to do a garbage collection in
"aggressive" mode. In this mode garbage collection more aggressively
optimize the repository at the expense of taking much more time.
Technically a aggressive mode garbage collection differs from a
non-aggressive one by:
- not reusing packed objects found in old packs. Recompress every object
- the configuration pack.window is set to 250 (the default is 10)
- the configuration pack.depths is set to 250 (the default is 50)

The associated classes in org.eclipse.jgit.api and the command line
command in org.eclipse.jgit.pgm expose this new option.

The configuration parameters gc.aggressiveDepth and gc.aggressiveWindow
have been introduced to configure this feature.

Bug: 444332
Change-Id: I024101f2810acf6be13ce144c9893d98f5c4ae76

Change-Id: I1dd741d3e522e396950c30d2f96e9713d0439078
Signed-off-by: Marc Strapetz <marc.strapetz@syntevo.com>

It can be useful for a server operator to know how long a pack
writer spent writing out objects, even if the request aborts and
never finishes.

Record more of the stats data inside of a finally block, to
ensure these can be included into the server's monitoring.

Change-Id: I00858aa393a948f8e742e64ae4c00953eadaef95

Other .git files are not hidden with this patch

Change-Id: Idf63ca08d08f3a77c33f5848d02074f8d6a75758
Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>

Change-Id: I1803ec6d8141f07dd4085778da6461abe81c30a9
Signed-off-by: Marc Strapetz <marc.strapetz@syntevo.com>

According to http://stackoverflow.com/a/8381338, the maximum array
size is not Integer.MAX_VALUE, but Integer.MAX_VALUE - 8

Change-Id: I6ddc7470368acd20abf0885c53c89a982bb0f176
Signed-off-by: Marc Strapetz <marc.strapetz@syntevo.com>

The native implementation of inflate() can set finished to return
true at the same time as it copies the last bytes into the buffer.

Check for finished on each iteration, terminating as soon as libz
knows the stream was completely inflated.

If not finished, it is likely input is required before the next
native call could do any useful work. Most invocations are passing
in a buffer large enough to store the entire result. A partial return
from inflate() will need more input before it can continue. Checking
right away that needsInput() is true saves a native call to determine
no bytes can be inflated without more input.

This should fix a rare infinite loop condition inside of inflation
when an object ends exactly at the end of a block boundary, and
the next block contains only the 20 byte trailing SHA-1.

When the stream is finished each new attempt to inflate() returns
n == 0, as no additional bytes were output. The needsInput() test
tries to add the length of the footer block to itself, but then loops
back around an reloads the same block as the block is smaller than
a full block size. A zero length input is set to the inflater,
which triggers needsInput() condition again.

Change-Id: I95d02bfeab4bf995a254d49166b4ae62d1f21346

This reverts commit b646578d89.

openInputStream() is never used in JGit, nor is it used by any
known working DFS implementation. The method was added as a
utility for reading back from a DfsInserter, but the final
implementation of that feature does not requrire this method.

Change-Id: I075ad95e40af49c92b554480f8993ef5658f7684

In the DFS implementation, flushing an inserter writes a new pack to
the storage system and is potentially very slow, but was the only way
to ensure previously-inserted objects were available.  For some tasks,
like performing a series of three-way merges, the total size of all
inserted objects may be small enough to avoid flushing the in-memory
buffered data.

DfsOutputStream already provides a read method to read back from the
not-yet-flushed data, so use this to provide an ObjectReader in the
DFS case.

In the file-backed case, objects are written out loosely on the fly,
so the implementation can just return the existing WindowCursor.

Change-Id: I454fdfb88f4d215e31b7da2b2a069853b197b3dd

Since 2badedcbe0 in-core merges can write up to 10 MiB
into a TemporaryBuffer.Heap strategy, where the data is stored
as a chain of byte[] blocks.

Support the inserter reading up to the streamFileThreshold (default 50
MiB) from the supplied input stream and hash the content to determine
if the merged result blob is already present in the repository.  This
allows the inserter to avoid creating duplicate objects in more cases,
reducing repository pack file churn.

Change-Id: I38967e2a0cff14c0a856cdb46a2c8fedbeb21ed5

Previously an equality check was performed so an exception would
be thrown if any other options were set.

Change-Id: I36b60e2c0a8aef9fcfe663055dba520192996872

DeltaTask$Block.partitionTask was doing an infinite loop if number of
threads was greater than the totalWeight. The weightPerThread was 0
which was causing the infinite loop. Set the weightPerThread to a
minimal value of one.

Bug: 420915
Change-Id: Ia8e3ad956d53d8193937b7fa1bc19aafde9767ff
Signed-off-by: Hugo Arès <hugo.ares@ericsson.com>

If a client passes a multiline message as argument to ReflogWriter.log()
the Reflog gets corrupted and cannot be parsed. ReflogWriter.log() is
invoked implicitly from various commands such as StashCreate, Rebase and
many more. However the message is not always filtered for line feeds.

Such an example is the StashCreateOperation of EGit which passes
unchecked user input as commit message. If a multiline comment is pasted
to the stash create dialog, the reflog gets corrupted.

ReflogWriter now replaces line endings in log message with spaces.

Bug: 435509
Change-Id: I3010cc902e13bee4d7b6696dfd11ab51062739d3
Signed-off-by: Andreas Hermann <a.v.hermann@gmail.com>
Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>

Streaming packed deltas is so slow that it never feasibly completes
(it will take hours for it to stream a few hundred megabytes on
relatively fast systems with a large amount of storage).  This
was indicated as a "failed experiment" by Shawn in the following
mailing list post:
http://dev.eclipse.org/mhonarc/lists/jgit-dev/msg01674.html

Change-Id: Idc12f59e37b122f13856d7b533a5af9d8867a8a5
Signed-off-by: Doug Kelly <dougk.ff7@gmail.com>

Change-Id: If8522d4e02254766e3310a70e42badafa5ab9768
Signed-off-by: Robin Rosenberg <robin.rosenberg@dewire.com>

Bug: 431552
Change-Id: I469316f5645205016e1fa6b0fbd2ff3b509b14bc
Signed-off-by: Robin Stocker <robin@nibor.org>

When working on a non-bare repository with a detached HEAD jgit's GC was
packing the ref named "HEAD" into the packed-refs file and deleted the
loose ref (the file .git/HEAD!). This made the repo unusable for native
git. This is fixed by telling jgit to only pack refs starting from
"refs/"

Change-Id: I50018aa006f18b244d2cae2ff78b5ffe1b821d63

Change-Id: I3137eba00d6eba96ca9051b6687fcf62e0871bcc
Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>

PostReceiveHooks can make use of this information to, for example,
update a cached size of the Git repository.

Change-Id: I2bf1200959a50531e2155a7609c96035ba45b10d
Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>

usable"

This reverts commit 2670fd427c.

By returning an instance of File from the ReceivePack.getPackFile the
abstraction of the persistence implementation was broken.

Change-Id: I28e3ebf3a659a7cbc94be51bba9e1ad338f2b786
Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>

Some of our Windows users have reported sporadic file system access
problems related to ObjectDirectory(Inserter) file deletion code in
combination with antiviral/firewall tools. For one of these users the
problem was fairly reproducible and changing deletion to RETRY solved
his problem.

Change-Id: I1e4001d5557fca693b7bac401268599467cb0c9e
Signed-off-by: Marc Strapetz <marc.strapetz@syntevo.com>‌

Having access to the pack file that was created by the ReceivePack
may be useful for post receive hooks. For example, a hook may want
to check the size of the received pack and the created index.

Change-Id: I4d51758e4565d32c9f8892242947eb72644b847d
Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>

The change includes comparing symbolic links between disk and index,
adding symbolic links to the index, creating/modifying links on
checkout. The behavior is controlled by the core.symlinks setting, just
as C Git does. When a new repository is created core.symlinks will be
set depending on the capabilities of the operating system and Java
runtime.

If core.symlinks is set to true, the assumption is that symlinks are
supported, which may result in runtime errors if this turns out not to
be the case.

Measuring the cost of jgit status on a repository with ~70000 files,
of which ~30000 are tracked reveals a penalty of about 10% for using
the Java7 (really NIO2) support module.

Bug: 354367
Change-Id: I12f0fdd9d26212324a586896ef7eb1f6ff89c39c
Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>

Allegedly this should improve performance, but I could not see it.

Change-Id: Id2057cb2cfcb46e94ff954483ce23f9c4a7edc5e

Johannes Carlsson identified a race condition[1] that can lead to
spurious MissingObjectExceptions at read time. If two threads are
active inside of ObjectDirectory looking for a packed object and the
packList is currently the empty NO_PACKS list, thread A will find
no object and eventually consider tryAgain1(). If thread A is put
to sleep and this point and thread B also does not find the object,
loads the packs, when thread A wakes up its tryAgain1 would return
false and the thread never considers the packs.

Rework the internal API of ObjectDirectory to keep a handle on the
exact PackList that was iterated by thread A, allowing it to always
retry walking through the packs if the new PackList is different.

This had some ripple effect into the CachedObjectDirectory and
the shared FileObjectDatabase interface. The new code should be
slightly easier to follow, especially from the perspective of the
CachedObjectDirectory trying to minimize the number of open system
calls it makes to files matching "$GIT_DIR/objects/??/?x{38}".

[1] http://dev.eclipse.org/mhonarc/lists/jgit-dev/msg02401.html

Change-Id: I9a1c9d6ad6cb38404b7b9178167b714077561353

Change-Id: Iaa07bb23f163b75e222bc6a5ce3acc91ffef435e
Signed-off-by: Tomasz Zarna <tomasz.zarna@tasktop.com>

Package was renamed, so I had to update the imports. Also, I verified
bitmap serialization was still compatible.

Change-Id: I161ad3875b963b56001beab477ef8d072accee4f

Change-Id: If3cc05931683d396b5ae2ea8952adceeb9a82ab9

Otherwise switching to another locale yields wrong results when parsing
date strings in GitDateParser. Since the MockSystemReader explicitly
uses english locale the tests need to specify the locale to be used when
parsing date strings.

Bug: 420772
Change-Id: I313ef6b1e9ef3bfb43d929ce34712ebd21f2cd9c
Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>

DfsPackFile.isIndexLoaded() uses the DfsBlockCache.Ref.get() method
to check if the index loaded. However, using the get() method marks
a hot bit in the cache, which can cause the index to never be unloaded
and seem hotter than it really is. Add a has() method which only
checks if the value is not null and does not update the hot bit.

Change-Id: I7e9ed216f6e273e8f5d79ae573973197654419b4

If during an garbage collection old packfiles are deleted it could
happen that on certain platforms the index file can be deleted but the
packfile can't be deleted (because someone locked the file). This led
to repositories with packfiles without corresponding index files. Those
zombie-packfiles potentially consume a lot of space on disk and it is
never tried to delete them again. Try to avoid this situation by
deleting packfiles first and don't try to delete the other files if we
can't delete the packfile. This gives us the chance to delete the
packfile during next GC.

This commit only improves the situation - there is still the chance for
orphan files during packfile deletion. We don't have an atomic delete
of multiple files .

Change-Id: I0a19ae630186f07d0cc7fe9df246fa1cedeca8f6

Currently, Repository.getAllRefs() and Repository.getTags() silently
ignores an IOException and instead returns an empty map. Repository
is a public API and as such cannot be changed until the next major
revision change. Where possible, update the internal jgit APIs to
use the RefDatabase directly, since it propagates the error.

Change-Id: I4e4537d8bd0fa772f388262684c5c4ca1929dc4c

If `git gc` creates a new pack with the same file name, the
pack checksum may not match that in the .bitmap. Fix the PackFile
implementaion to silently ignore invalid bitmap indexes.

Fixes Issue https://code.google.com/p/gerrit/issues/detail?id=2131

Change-Id: I378673c00de32385ba90f4b639cb812f9574a216

Previously, the DfsPackCompactor exited without pruning the existing
packs, when no new packs were created.

Change-Id: I5e3b6f8c789706c7a982e6ae93cf7c3d4346797c

OpenJDK 7 does not benefit from using an inflate stride on the input
array. The implementation of java.util.zip.Inflater supplies the
entire input byte[] to libz, with no regards for the bounds supplied.
Slicing at 512 byte increments in DfsBlock no longer has any benefit.

In OpenJDK 6 the native portion of Inflater used GetByteArrayRegion
to obtain a copy of the input buffer for libz. In this use case
supplying a small stride made sense, it avoided allocating space
for and copying data past the end of the object's compressed stream.

In OpenJDK 7 the native code uses GetPrimitiveArrayCritical,
which tries to avoid copying by freezing Java garbage collection
and accessing the byte[] contents in place. On OpenJDK 7 derived
JVMs it is likely more efficient to supply the entire DfsBlock.

Since OpenJDK 5 and 6 are deprecated and replaced by OpenJDK 7
it is reasonable to suggest any consumers running JGit with DFS
support use an OpenJDK 7 derived JVM. However, JGit still targets
local filesystem support on Java 5, so it is still not reasonble to
apply this same simplification to the internal.storage.file package.

See: JDK-6751338 (http://bugs.sun.com/bugdatabase/view_bug.do?bug_id=6751338)
Change-Id: Ib248b6d383da5c8aa887d9c355a0df6f3e2247a5

Previously it took 1200ms to create a reverse index (sorted by offset).
Using a simple bucket sort algorithm, that time is reduced to 450ms.
The bucket index into the offset array is kept, in order to decrease
the binary search window.

Don't keep a copy of the offsets. Instead, use nth position
to lookup the offset in the PackIndex.

Change-Id: If51ab76752622e04a4430d9a14db95ad02f5329d

Currently, the offset can only be retrieved by ObjectId or iterating all
of the entries. Add a method to lookup the offset by position in the
index sorted by SHA1.

Change-Id: I45e9ac8b752d1dab47b202753a1dcca7122b958e

Change-Id: I0ec1f17a88bc14f22c10f9bc8d6f5b5118410e3a

A parenthesis was in the wrong place passing arguments to the wrong
format call. Also fix formatting of enclosing switch statement.

Change-Id: I4cb9642f08b58c39033c3a81dab4bd56bebf4fd2
Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>

Change-Id: I8ced7098da5b345fd9af2fdfafd1ef6a44ccee0d

Instead of counting objects processed, count number of bytes added
into the window. This should rescale the progress meter so that 30%
complete means 30% of the total uncompressed content size has been
inflated and fed into the window.

In theory the progress meter should be more accurate about its
percentage complete/remaining fraction than with objects. When
counting objects small objects move the progress meter more rapidly
than large objects, but demand a smaller amount of work than large
objects being compressed.

Change-Id: Id2848c16a2148b5ca51e0ca1e29c5be97eefeb48

Instead of assuming all objects cost the same amount of time to
delta compress, aggregate the byte size of objects in the list
and partition threads with roughly equal total bytes.

Before splitting the list select the N largest paths and assign
each one to its own thread. This allows threads to get through the
worst cases in parallel before attempting smaller paths that are
more likely to be splittable.

By running the largest path buckets first on each thread the likely
slowest part of compression is done early, while progress is still
reporting a low percentage. This gives users a better impression of
how fast the phase will run. On very complex inputs the slow part
is more likely to happen first, making a user realize its time to
go grab lunch, or even run it overnight.

If the worst sections are earlier, memory overruns may show up
earlier, giving the user a chance to correct the configuration and
try again before wasting large amounts of time. It also makes it
less likely the delta compression phase reaches 92% in 30 minutes
and then crawls for 10 hours through the remaining 8%.

Change-Id: I7621c4349b99e40098825c4966b8411079992e5f

By excluding objects the compactor can avoid storing objects that
are already well packed in the base GC packs, or any other pack
not being replaced by the current compaction operation.

For deltas the base object is still included even if the base exists
in another exclusion set.  This favors keeping deltas for recent
history, to support faster fetch operations for clients.

Change-Id: Ie822fe075fe5072fe3171450fda2f0ca507796a1