]> source.dussan.org Git - jgit.git/log
jgit.git
14 years agoConsistently fail work tree methods on bare repositories 54/954/1
Shawn O. Pearce [Fri, 25 Jun 2010 01:45:57 +0000 (18:45 -0700)]
Consistently fail work tree methods on bare repositories

If the working tree isn't available, it doesn't make any sense to
obtain the merge heads, or the buffered commit message.  The
repository shouldn't have a partial merge state to read.  Throw back
the same exception we do when invoking getWorkDir() on a bare
repository instance.

Change-Id: I762c55890b7fe272a183da583f910671d1cadf71
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years agoConsistently use getDirectory() for work tree state 53/953/1
Shawn O. Pearce [Fri, 25 Jun 2010 01:41:21 +0000 (18:41 -0700)]
Consistently use getDirectory() for work tree state

This permits us to leave the implementation of these methods here in
the Repository class, but later refactor how the directory is accessed
into a subclass.

Change-Id: I5785b2009c5b7cca0fb070a968e50814ce847076
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years agoAdd RepositoryState.BARE 52/952/1
Shawn O. Pearce [Thu, 24 Jun 2010 19:57:02 +0000 (12:57 -0700)]
Add RepositoryState.BARE

A bare repository cannot be checked out, committed to, etc. as it
doesn't have a working directory.  Define this as a state since the
state enumeration exists only to describe how a working directory
can be modified.

Change-Id: I0a299013c6e42fef6cae3f6a9446f8f6c8e0514a
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years agoRename Repository 'config' as 'repoConfig' 51/951/1
Shawn O. Pearce [Fri, 25 Jun 2010 01:18:46 +0000 (18:18 -0700)]
Rename Repository 'config' as 'repoConfig'

This better matches with the other configuration variable,
'userConfig', and helps to make it clear what config object
we are dealing with.

Change-Id: I2c585649aecc805e8e66db2f094828cd2649e549
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years agoRemove RepositoryConfig and use FileBasedConfig instead 50/950/1
Shawn O. Pearce [Wed, 23 Jun 2010 17:13:55 +0000 (10:13 -0700)]
Remove RepositoryConfig and use FileBasedConfig instead

Change the Repository API to use straight-up FileBasedConfig.
This lets us remove the subclass RepositoryConfig and stop having
a specialized configuration type for repository, letting us instead
focus the config type heirarchy on type-of-storage rather than use.

Change-Id: I7236800e8090624453a89cb0c7a9a632702691c6
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years agoDelegate repository access to refs, objects 49/949/1
Shawn O. Pearce [Thu, 24 Jun 2010 22:18:14 +0000 (15:18 -0700)]
Delegate repository access to refs, objects

Instead of using the internal field directly to access references
or objects, use the getter method to obtain the proper type of
database, and follow down from there.  This permits us to later
do a refactoring that makes those methods abstract and strips the
field out of the Repository class, moving it into a concrete base
class that is more storage implementation specific.

Change-Id: Ic21dd48800e68a04ce372965ad233485b2a84bef
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years agoCleanup Repository.create() 48/948/1
Shawn O. Pearce [Thu, 24 Jun 2010 22:19:05 +0000 (15:19 -0700)]
Cleanup Repository.create()

This method doesn't need to be synchronized, as its only a proxy to
create(boolean), which is the real worker.  While we are touching
it try to improve the Javadoc and whitespace nearby.

Change-Id: Ibdddec6e518ca6d7439cfad90fedfcdc2d6b7a2e
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years agoMove additional have enumeration to Repository 14/914/4
Shawn O. Pearce [Sat, 19 Jun 2010 03:23:06 +0000 (20:23 -0700)]
Move additional have enumeration to Repository

This permits the repository implementation to know what its
alternates concept means, and avoids needing to expose finer details
about the ObjectDatabase to network code like the RefAdvertiser.

Change-Id: Ic6d173f300cb72de34519c7607cf7b0ff3ea6882
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years agoRefactor amazon-s3:// property file loading to support no directory 47/947/1
Shawn O. Pearce [Sat, 26 Jun 2010 00:42:01 +0000 (17:42 -0700)]
Refactor amazon-s3:// property file loading to support no directory

In the future getDirectory() can return null.  Avoid an NPE here by
refactoring the code to support conditionally skipping a check for
the properties file in the repository directory, falling to only
the user's ~/ file location.

Change-Id: I76f5503d4063fdd9d24b7c1b58e1b09ddf1a5670
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years agoDownload pack-*.idx to /tmp if not on local filesystem 46/946/1
Shawn O. Pearce [Thu, 24 Jun 2010 18:30:19 +0000 (11:30 -0700)]
Download pack-*.idx to /tmp if not on local filesystem

If the destination repository doesn't use an ObjectDirectory to
store its objects, we can't download to the object directory.
Instead pull the pack-*.idx files down to temporary files in the
JVM's default temporary directory.

Change-Id: Ied16bc89be624d87110ba42ba52d698a6ea7d982
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years agoUse FileKey to resolve Git repository arguments 45/945/1
Shawn O. Pearce [Thu, 24 Jun 2010 18:39:20 +0000 (11:39 -0700)]
Use FileKey to resolve Git repository arguments

upload-pack and receive-pack take a git repository as an argument,
but its a lenient path format.  Locate the repository and open it.

Change-Id: I4b377e57b28ba3b1717c13d9ab51a602de1ad257
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years agoDirCache must use getIndexFile 44/944/1
Shawn O. Pearce [Sat, 26 Jun 2010 00:32:12 +0000 (17:32 -0700)]
DirCache must use getIndexFile

When reading or locking the index of a repository, we need to use
the index file specified by the repository, to ensure we correctly
honor what the repository was configured with.

Change-Id: I5be366ce32d7923b888dc01d19335912b01b7c4c
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years agoDisable topological sorting in PackWriter 85/885/3
Shawn O. Pearce [Tue, 15 Jun 2010 01:31:14 +0000 (18:31 -0700)]
Disable topological sorting in PackWriter

Its not strictly required that we sort topologically in order to
produce a valid pack file.  This was just something that Linus
thought would be a good idea to do.  In practice its not that
important for most repositories.  Local file IO quickly falls
out of the pattern that topological sorting provides any sort
of benefit for, so expending extra resources to enforce it when
we make a pack isn't really worth it.

I'm removing this sort in the pipeline because later changes
would support really efficient COMMIT_TIME_DESC sorting on a
non-file storage system, but TOPO sorting would be a bit more
ugly to run, due to the in-memory delays it imposes.

Change-Id: I0121453461c2140c6917cb10c6df584eb47e5795
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years agoUploadPack: Permit flushing progress messages under smart HTTP 58/858/4
Shawn O. Pearce [Mon, 14 Jun 2010 19:49:41 +0000 (12:49 -0700)]
UploadPack: Permit flushing progress messages under smart HTTP

If UploadPack invokes flush() on the output stream we pass it, its
most likely the progress messages coming down the side band stream.
As pack generation can take a while, we want to push that down
at the client as early as we can, to keep the connection alive,
and to let the user know we are still working on their behalf.

Ensure we dump the temporary buffer whenever flush() is invoked,
otherwise the messages don't get sent in a timely fashion to the
user agent (in this case, git fetch).

We specifically don't implement flush() for ReceivePack right now,
as that protocol currently does not provide progress messages to
the user, but it does invoke flush several times, as the different
streams include '0000' type flush-pkts to denote various end points.

Change-Id: I797c90a2c562a416223dc0704785f61ac64e0220
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years agoRewrite resolve in terms of RevWalk 13/913/4
Shawn O. Pearce [Thu, 24 Jun 2010 00:26:43 +0000 (17:26 -0700)]
Rewrite resolve in terms of RevWalk

We want to eventually get rid of the mapCommit, mapTree APIs on
Repository and force everyone into the faster parsers that exist
in RevWalk.  Rewriting resolve in terms of the faster parsers is
a good first step.

It actually simplifies the code a bit, as we no longer need to
keep track of an ObjectId and an Object (the parsed form), since
all RevObjects implicitly have their ObjectId readily available.

Change-Id: I4d234630195616e2c263e7e70038b55a1be4e7a3
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years agoReplace manual peel loops with RevWalk.peel 35/935/1
Shawn O. Pearce [Thu, 24 Jun 2010 00:26:44 +0000 (17:26 -0700)]
Replace manual peel loops with RevWalk.peel

Instead of peeling things by hand in application level code, defer
the peeling logic into RevWalk's new peel utility method.

Change-Id: Idabd10dc41502e782f6a2eeb56f09566b97775a8
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years agoUse RevTag/RevCommit to sort in a PlotWalk 84/884/3
Shawn O. Pearce [Tue, 15 Jun 2010 01:51:09 +0000 (18:51 -0700)]
Use RevTag/RevCommit to sort in a PlotWalk

We already have these objects parsed and cached in our object pool.
We shouldn't be looking them up via the legacy mapObject API, but
instead can use the pool and the faster parsing routines available
through the RevWalk that we extend.

While we are here fixing the code, lets also correct the tag date
sorting to accept tags that have no tagger identity, because they
were created before Git knew how to store that field.

Change-Id: Id49a11f6d9c050c82b876e5e11058840c894b2d7
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years agoUse CoreConfig, UserConfig and TransferConfig directly 34/934/1
Shawn O. Pearce [Wed, 23 Jun 2010 17:11:34 +0000 (10:11 -0700)]
Use CoreConfig, UserConfig and TransferConfig directly

Rather than relying on the helpers in RepositoryConfig to get
these objects, obtain them directly through the Config API.
Its only slightly more verbose, but permits us to work with the
base Config class, which is more flexible than the highly file
specific RepositoryConfig.

This is what I really meant to do when I added the section parser
and caching support to Config, we just failed to finish updating
all of the call sites.

Change-Id: I481cb365aa00bfa8c21e5ad0cd367ddd9c6c0edd
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years agoUse higher level Config types when possible 33/933/1
Shawn O. Pearce [Wed, 23 Jun 2010 17:09:27 +0000 (10:09 -0700)]
Use higher level Config types when possible

We don't have to assume/depend on RepositoryConfig here, these
two tests can use higher level versions of the class and still
come up with the same test.  That frees us up to do some changes
to the RepositoryConfig API.

Change-Id: Ia7b263c8c5efa3fae1054416d39c546867288132
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years agoRemove test of the unsupported core.legacyHeaders variable 32/932/1
Shawn O. Pearce [Wed, 23 Jun 2010 17:07:44 +0000 (10:07 -0700)]
Remove test of the unsupported core.legacyHeaders variable

Long ago we stopped supporting the core.legacyHeaders variable,
as JGit (like C Git) stopped creating the new pack-style loose
objects, rendering this variable pointless.  The test is still
valid, it proves we write the standard loose object format for
a commit, but the variable assignment has no impact on the test
so drop it from the code.

Change-Id: I051336ada23033c05e86bbff73ae5d78a37b1640
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years agoUploadPack: Avoid unnecessary flush in smart HTTP 57/857/4
Shawn O. Pearce [Mon, 14 Jun 2010 19:37:17 +0000 (12:37 -0700)]
UploadPack: Avoid unnecessary flush in smart HTTP

Under smart HTTP the biDirectionalPipe flag is false, and we return
back immediately at this point in the negotiation process.  There is
no need to flush the stream to the client, the request is over and
it will be automatically flushed out by the higher level servlet
that invoked us.  Avoiding flush here allows us to only use flush
after a progress message is sent during pack generation.

Change-Id: Id0c8b7e95e3be6ca4c1b479e096bed6b0283b828
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years agoAdd MutableObjectId.copyFrom(AnyObjectId) 22/922/2
Shawn O. Pearce [Tue, 22 Jun 2010 23:42:29 +0000 (16:42 -0700)]
Add MutableObjectId.copyFrom(AnyObjectId)

This simplifies the PackIndex code, which is trying to quickly copy
an existing ObjectId into a MutableObjectId.  Rather than having
the PackIndex violate the ObjectId's internals, expose a copy from
function similar to the other ones for copying from raw byte arrays
or hex formatted strings.

Change-Id: I142635cbece54af2ab83c58477961ce925dc8255
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years agoExpose AnyObjectId compareTo(byte[]) and compareTo(int[]) 17/917/4
Shawn O. Pearce [Sat, 19 Jun 2010 00:38:22 +0000 (17:38 -0700)]
Expose AnyObjectId compareTo(byte[]) and compareTo(int[])

Storage systems can use these implementations to compare a passed
AnyObjectId with a stored representation of an ObjectId in the
canonical network byte order format.  This can be useful to do a
binary search, or just linear scan, over an encoded storage file.

Change-Id: I8c72993c4f4c6e98d599ac2c9867453752f25fd2
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years agoExpose RefWriter constructor taking RefList 25/925/2
Shawn O. Pearce [Wed, 23 Jun 2010 00:07:09 +0000 (17:07 -0700)]
Expose RefWriter constructor taking RefList

An implementation might prefer to use the RefList type here, and
RefList is part of our public API.  Expose the constructor so callers
who have a RefList can take advantage of the existing sorting.

Change-Id: I545867f85aa2c479d2d610024ebbe318144709c8
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years agoExpose RefUpdate constructor to any subclass 24/924/2
Shawn O. Pearce [Tue, 22 Jun 2010 23:48:37 +0000 (16:48 -0700)]
Expose RefUpdate constructor to any subclass

When we finally move RefDirectory to the new storage.file package,
its associated RefDirectoryUpdate will need visiblity to this
constructor in order to initialize itself.  This is true of any
other repository implementation, so make it protected rather than
package level visible.

Change-Id: If838aec9baeb80ee2f12dcbca717657c725a9242
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years agoExpose repository change event constructors 23/923/2
Shawn O. Pearce [Tue, 22 Jun 2010 23:45:26 +0000 (16:45 -0700)]
Expose repository change event constructors

Repository implementations outside of .lib need to be able to
create these events and deliver them to listening application code.

Expose and document the constructors so that they are visible when
we move FileRepository into storage.file.FileRepository.

Change-Id: I7fb6e8f4f5fdab683c5ebb5267673aa6d5b560bb
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years agoisValidRefName: Inline the forbidden ref suffix of ".lock" 20/920/2
Shawn O. Pearce [Tue, 22 Jun 2010 23:37:38 +0000 (16:37 -0700)]
isValidRefName: Inline the forbidden ref suffix of ".lock"

A Git reference name must never end with ".lock", as it would
confuse any existing C client that tries to obtain a clone of the
repository over the network.  Even if the repository isn't on a
local filesystem, it still should ban that suffix.

Because I plan to move LockFile to storage.file and make it a private
implementation detail of the local file system storage model,
we can't rely on its package level SUFFIX field here.  Making it
public probably won't work long-term either, as I also plan to
pull storage.file into its own separate project that depends on
the core library.

So, just inline the constant here.  Its as foribidden as ":" is.

Change-Id: If85076861baeacc183b82696375a13e935ba8836
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years agoRemove pack stream from PackWriterTest 19/919/2
Shawn O. Pearce [Tue, 22 Jun 2010 23:54:23 +0000 (16:54 -0700)]
Remove pack stream from PackWriterTest

This stream was used only to determine how many bytes had been
written thus far.  Except we're always dumping it into a simple
ByteArrayOutputStream, which also knows that.  Drop the dependency
on the pack stream and use ByteArrayOutputStream directly.

This lets us later move this test into the new storage.file
package without dragging along the pack stream that is an internal
implementation detail of PackWriter, which is more general than
just the file storage layer.

Change-Id: I291689c0b1ed799270c213ee73b710b2637fb238
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years agoRemove pointless setOldObjectId in test 18/918/2
Shawn O. Pearce [Tue, 22 Jun 2010 23:18:22 +0000 (16:18 -0700)]
Remove pointless setOldObjectId in test

Setting this value is pointless, because its automatically set
by the refs.newUpdate call that created the update operation.
The API is protected by default, because application level code,
including this test, should not be calling it.

Change-Id: I8867a4e8007892e2bd44a05d7dec619081081943
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years agoRemove speed tests based on mapCommit 12/912/2
Shawn O. Pearce [Sat, 19 Jun 2010 01:10:32 +0000 (18:10 -0700)]
Remove speed tests based on mapCommit

The mapCommit API is being deprecated because it doesn't run very
fast.  Leaving tests around to test how fast it is relative to C Git
isn't instructive.  Remove them, which should help aid the transition
away from the mapCommit API.

Change-Id: I27e1c844610d7da5b2c44b33a00602706973c9cc
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years agoChange default target platform for maven build to galileo 03/903/1
Matthias Sohn [Fri, 18 Jun 2010 23:06:14 +0000 (01:06 +0200)]
Change default target platform for maven build to galileo

Starting with 0.9 we do no longer support ganymede.
http://dev.eclipse.org/mhonarc/lists/egit-dev/msg01277.html

Change-Id: Ibf40342f67d9706e86336748f15d10ea47278096
Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
14 years agoMerge "Fix line endings"
Shawn Pearce [Fri, 18 Jun 2010 22:15:53 +0000 (18:15 -0400)]
Merge "Fix line endings"

14 years agoFix line endings 99/899/1
Matthias Sohn [Fri, 18 Jun 2010 21:24:56 +0000 (23:24 +0200)]
Fix line endings

Some sources had dos line endings. Also configure all projects to use
unix line endings and UTF-8 text encoding.

Change-Id: I8fc9a1dbb219ffa91d1b3011b3b11b7e48e74ca7
Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
14 years agoMerge ""Bare" Repository should not return working directory."
Shawn Pearce [Thu, 17 Jun 2010 02:34:46 +0000 (22:34 -0400)]
Merge ""Bare" Repository should not return working directory."

14 years agoMake ObjectId, RefSpec, RemoteConfig, URIish serializable 80/880/2
Andrew Bayer [Wed, 16 Jun 2010 22:49:49 +0000 (15:49 -0700)]
Make ObjectId, RefSpec, RemoteConfig, URIish serializable

Modifications to various classes in order to allow serialization
for use of JGit in Hudson's git plugin.

Change-Id: If088717d3da7483538c00a927e433a74085ae9e6

14 years ago"Bare" Repository should not return working directory. 36/836/10
Mathias Kinzler [Wed, 16 Jun 2010 06:50:26 +0000 (08:50 +0200)]
"Bare" Repository should not return working directory.

If a repository is "bare", it currently still returns a working directory.
This conflicts with the specification of "bare"-ness.

Bug: 311902

Change-Id: Ib54b31ddc80b9032e6e7bf013948bb83e12cfd88
Signed-off-by: Mathias Kinzler <mathias.kinzler@sap.com>
14 years agoMerge "tools/version.sh: Use backup files on Win32"
Matthias Sohn [Tue, 15 Jun 2010 23:34:52 +0000 (19:34 -0400)]
Merge "tools/version.sh: Use backup files on Win32"

14 years agoMerge "Add missing @Override tags in AlternateRepositoryDatabase"
Chris Aniszczyk [Tue, 15 Jun 2010 15:40:04 +0000 (11:40 -0400)]
Merge "Add missing @Override tags in AlternateRepositoryDatabase"

14 years agoAllow to read configured keys 35/835/3
Mathias Kinzler [Mon, 14 Jun 2010 16:03:30 +0000 (18:03 +0200)]
Allow to read configured keys

Currently, there is no way to read the content
of the Git Configuration in a  way that would
allow to list all configured values generically.
This change extends the Config class in such a
way as to being able to get a list of sections and
to get a list of names for any given section or
subsection.
This is required in able to implement proper
configuration handling in EGit (show all the
content of a given configuration similar to
"git config -l").

Change-Id: Idd4bc47be18ed0e36b11be8c23c9c707159dc830
Signed-off-by: Mathias Kinzler <mathias.kinzler@sap.com>
14 years agoMerge changes I53f71dc0,I3a899a3a,I3e8bd245,Ie7c9db83,If396326e,I6f4cf8da,I3bf96dd0...
Shawn Pearce [Mon, 14 Jun 2010 23:59:48 +0000 (19:59 -0400)]
Merge changes I53f71dc0,I3a899a3a,I3e8bd245,Ie7c9db83,If396326e,I6f4cf8da,I3bf96dd0,I3a2a43a1,I292fe88c,Ia1cf40cf

* changes:
  git-servlet: Fix comparing uploadFactory with the wrong DISABLED instance
  Prefer static inner classes
  Override equals for SwingLane since super class PlotLane defines it
  Make sure a Stream is closed upon errors in IpLogGenerator
  Make constant static in RebuildCommitGraph
  Make inner classes static in http code
  Cache filemode in GitIndex
  Remove unused parent field in PlotLane
  Removed unused repo field in WorkDirCheckout
  Extend DiffFormatter API to simplify styling

14 years agogit-servlet: Fix comparing uploadFactory with the wrong DISABLED instance 54/854/3
Robin Rosenberg [Mon, 14 Jun 2010 21:33:02 +0000 (23:33 +0200)]
git-servlet: Fix comparing uploadFactory with the wrong DISABLED instance

Change-Id: I53f71dc0e3c68839da5ff5a2e0f3eeb8340e4793
Signed-off-by: Robin Rosenberg <robin.rosenberg@dewire.com>
14 years agoAdd missing @Override tags in AlternateRepositoryDatabase 56/856/1
Shawn O. Pearce [Sun, 26 Jul 2009 00:33:30 +0000 (17:33 -0700)]
Add missing @Override tags in AlternateRepositoryDatabase

Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years agotools/version.sh: Use backup files on Win32 42/842/1
Shawn O. Pearce [Mon, 14 Jun 2010 15:18:47 +0000 (08:18 -0700)]
tools/version.sh: Use backup files on Win32

Windows doesn't permit us to edit a file in-place with Perl.
So create backup files when we perform the edit, and remove them
when we are done.  This is a tad slower on POSIX systems, but is
much more portable.

Change-Id: I429c7d698924cb32e709363f5da82f7232bbdab2
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years agoMerge branch 'stable-0.8' 41/841/1
Shawn O. Pearce [Mon, 14 Jun 2010 15:12:48 +0000 (08:12 -0700)]
Merge branch 'stable-0.8'

* stable-0.8:
  Qualify post-0.8.4 builds
  JGit 0.8.4
  JGit 0.8.3
  Include about.html in org.eclipse.jgit artifact
  Fix build.properties of the JGit feature
  Added the standard SULA for JGit
  Add "resources/" as a source folder

Change-Id: I4ecb0af41184ef84d104345fd1adcc4a240a38f6

14 years agoStart 0.9 development 40/840/1
Shawn O. Pearce [Mon, 14 Jun 2010 15:11:27 +0000 (08:11 -0700)]
Start 0.9 development

Change-Id: I84173ece5100f1fcb78168e2e102b649d9466c08
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years agoQualify post-0.8.4 builds 39/839/1 stable-0.8
Shawn O. Pearce [Mon, 14 Jun 2010 15:09:54 +0000 (08:09 -0700)]
Qualify post-0.8.4 builds

Change-Id: I21efed66921eb7e1e4010fccc9fa9af6c4150fc1
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years agoJGit 0.8.4 37/837/1 v0.8.4
Matthias Sohn [Mon, 14 Jun 2010 13:34:28 +0000 (15:34 +0200)]
JGit 0.8.4

Created wrong tags for 0.8.3 hence creating another version.

Change-Id: I4e00bbcffe1cf872e2d7e3f3d88d068701fb5330
Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
14 years agoJGit 0.8.3 30/830/1
Matthias Sohn [Sun, 13 Jun 2010 23:34:34 +0000 (01:34 +0200)]
JGit 0.8.3

Change-Id: I845da83c74475d74ec25d68f53c0a4738a898550
Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
14 years agoPrefer static inner classes 53/853/1
Robin Rosenberg [Sun, 13 Jun 2010 01:31:52 +0000 (03:31 +0200)]
Prefer static inner classes

Signed-off-by: Robin Rosenberg <robin.rosenberg@dewire.com>
14 years agoOverride equals for SwingLane since super class PlotLane defines it 52/852/1
Robin Rosenberg [Sun, 13 Jun 2010 01:30:10 +0000 (03:30 +0200)]
Override equals for SwingLane since super class PlotLane defines it

Signed-off-by: Robin Rosenberg <robin.rosenberg@dewire.com>
14 years agoMake sure a Stream is closed upon errors in IpLogGenerator 51/851/1
Robin Rosenberg [Sun, 13 Jun 2010 01:28:04 +0000 (03:28 +0200)]
Make sure a Stream is closed upon errors in IpLogGenerator

Signed-off-by: Robin Rosenberg <robin.rosenberg@dewire.com>
14 years agoMake constant static in RebuildCommitGraph 50/850/1
Robin Rosenberg [Sun, 13 Jun 2010 01:27:06 +0000 (03:27 +0200)]
Make constant static in RebuildCommitGraph

Signed-off-by: Robin Rosenberg <robin.rosenberg@dewire.com>
14 years agoMake inner classes static in http code 49/849/1
Robin Rosenberg [Sun, 13 Jun 2010 01:19:47 +0000 (03:19 +0200)]
Make inner classes static in http code

Static classes are preferrable to keep unwanted dependencies away,
and they have one less member field.

Signed-off-by: Robin Rosenberg <robin.rosenberg@dewire.com>
14 years agoCache filemode in GitIndex 48/848/1
Robin Rosenberg [Sun, 13 Jun 2010 01:16:32 +0000 (03:16 +0200)]
Cache filemode in GitIndex

Apparently this was the intention, but never happened

Signed-off-by: Robin Rosenberg <robin.rosenberg@dewire.com>
14 years agoRemove unused parent field in PlotLane 47/847/1
Robin Rosenberg [Sun, 13 Jun 2010 01:13:57 +0000 (03:13 +0200)]
Remove unused parent field in PlotLane

Signed-off-by: Robin Rosenberg <robin.rosenberg@dewire.com>
14 years agoRemoved unused repo field in WorkDirCheckout 46/846/1
Robin Rosenberg [Sun, 13 Jun 2010 01:12:41 +0000 (03:12 +0200)]
Removed unused repo field in WorkDirCheckout

Signed-off-by: Robin Rosenberg <robin.rosenberg@dewire.com>
14 years agoExtend DiffFormatter API to simplify styling 28/828/2
Robin Rosenberg [Sat, 12 Jun 2010 13:31:04 +0000 (15:31 +0200)]
Extend DiffFormatter API to simplify styling

Refactor and extend the internals so users can override and
intervene during formatting, e.g. to colorize output.

Change-Id: Ia1cf40cfd4a5ed7dfb6503f8dfc617237bee0659
Signed-off-by: Robin Rosenberg <robin.rosenberg@dewire.com>
14 years agoInclude about.html in org.eclipse.jgit artifact 20/820/1
Matthias Sohn [Tue, 8 Jun 2010 00:36:43 +0000 (02:36 +0200)]
Include about.html in org.eclipse.jgit artifact

This is required to enable accessing legal info for
org.eclipse.jgit from
Help > About > Installation Details > Plugins

Change-Id: I73f40dd2018112cd23102954d7647ecdbbbf0d89
Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
14 years agoFix build.properties of the JGit feature 18/818/1
Chris Aniszczyk [Mon, 7 Jun 2010 21:55:48 +0000 (16:55 -0500)]
Fix build.properties of the JGit feature

The JGit feature's build.properties was referring
to a notice.html instead of license.html

Change-Id: I642e2a05d1be58b1d47fd9701edf1a0f2bfa3660
Signed-off-by: Chris Aniszczyk <caniszczyk@gmail.com>
14 years agoAdded the standard SULA for JGit 17/817/1
Chris Aniszczyk [Mon, 7 Jun 2010 21:26:57 +0000 (16:26 -0500)]
Added the standard SULA for JGit

The Eclipse Foundation requires the standard SULA be present
in every feature. We had the license present via edl-v10.html
but we were missing the SULA via the license.html file. The
fix is to simply add the SULA.

Change-Id: I75b43ce098f544b95181755b5cc81a9b1dee6391
Signed-off-by: Chris Aniszczyk <caniszczyk@gmail.com>
14 years agoAdd "resources/" as a source folder 10/810/1
Matthias Sohn [Sat, 5 Jun 2010 12:39:27 +0000 (14:39 +0200)]
Add "resources/" as a source folder

Building jgit with pde.build was broken without resources.

Bug:315823
Change-Id: I45be510ada068b3ffab0feb30ec60f2c96a5ca32
Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
14 years agoRepository can be configured with FS 88/588/3
Marc Strapetz [Tue, 20 Apr 2010 19:01:19 +0000 (21:01 +0200)]
Repository can be configured with FS

On Windows, FS_Win32_Cygwin has been used if a Cygwin Git installation
is present in the PATH. Assuming that the user works with the Cygwin
Git installation may result in unnecessary overhead if he actually
does not.

Applications built on top of jgit may have more knowledge on the
actually used Git client (Cygwin or not) and hence should be able to
configure which FS to use accordingly.

Change-Id: Ifc4278078b298781d55cf5421e9647a21fa5db24

14 years agoMerge "Refactor ReadTreeTest to allow testing other checkout classes"
Shawn Pearce [Sat, 5 Jun 2010 01:48:54 +0000 (21:48 -0400)]
Merge "Refactor ReadTreeTest to allow testing other checkout classes"

14 years agoAdd support for computing a Change-Id à la Gerrit 23/523/7
Robin Rosenberg [Mon, 24 May 2010 19:19:59 +0000 (21:19 +0200)]
Add support for computing a Change-Id Ã  la Gerrit

A Change-Id helps tools like Gerrit Code Review to keeps different
versions of a patch together. The Change-Id is computed as a SHA-1
hash of some of the same basic information as a commit id on the first
commit intended to solve a particular problem and then reused for
updated solutions.

Change-Id: I04334f84e76e83a4185283cb72ea0308b1cb4182
Signed-off-by: Robin Rosenberg <robin.rosenberg@dewire.com>
14 years agoRefactor ReadTreeTest to allow testing other checkout classes 92/792/3
Christian Halstrick [Fri, 4 Jun 2010 22:18:17 +0000 (00:18 +0200)]
Refactor ReadTreeTest to allow testing other checkout classes

ReadTreeTest contains a lot of useful tests for "checkout"
implementations. But ReadTreeTest was hardcoded to test only
WorkDirCheckout. This change doesn't add/modify any tests semantically
but refactors ReadTreeTest so that a different implementations of
checkout can be tested. This was done to allow DirCacheCheckout to be
tested without rewriting all these tests.

Change-Id: I36e34264482b855ed22c9dde98824f573cf8ae22
Signed-off-by: Christian Halstrick <christian.halstrick@sap.com>
Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
14 years agoProvide a public entry method to determine whether a URI protocol is supported 03/803/1
Alex Blewitt [Thu, 3 Jun 2010 23:38:50 +0000 (00:38 +0100)]
Provide a public entry method to determine whether a URI protocol is supported

14 years agoQualify post-0.8.1 builds 94/794/3
Shawn O. Pearce [Wed, 2 Jun 2010 22:20:42 +0000 (15:20 -0700)]
Qualify post-0.8.1 builds

Change-Id: Id86e5876b2f684b2a272c07061a276b054ba410d
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years agoJGit 0.8.1 97/797/1 v0.8.1
Shawn O. Pearce [Wed, 2 Jun 2010 21:47:31 +0000 (14:47 -0700)]
JGit 0.8.1

Change-Id: I3d4ac7d0617a3575019e2ed748ed2a298a988340
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years agoBump all features during release 96/796/2
Shawn O. Pearce [Wed, 2 Jun 2010 21:30:11 +0000 (14:30 -0700)]
Bump all features during release

Change-Id: I3103c54a2a525f5f190cf35b63394dad6d02cc5e
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years agoMerge "Describe how to generate iplog."
Shawn Pearce [Tue, 1 Jun 2010 15:10:38 +0000 (11:10 -0400)]
Merge "Describe how to generate iplog."

14 years agoDescribe how to generate iplog. 73/773/3
Matthias Sohn [Fri, 28 May 2010 23:03:04 +0000 (01:03 +0200)]
Describe how to generate iplog.

Change-Id: I91f249422efbb6aea0491e276ccfe00f5ae7d212
Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
14 years agoeclipse-iplog: Use contribution rather than bug element 72/772/1
Shawn O. Pearce [Fri, 28 May 2010 22:06:29 +0000 (15:06 -0700)]
eclipse-iplog: Use contribution rather than bug element

Wayne changed the schema to no longer be dependent upon the Bugzilla
notion of a contribution, but instead be more generic and better
support systems like Gerrit Code Review.  Update our output to
use the <contribution> element and include a link to the change
in Gerrit.

Change-Id: Ibc8a436918bd8e7597dc17743824201a74bce09b
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years agoeclipse-iplog: Add new consumes element to IP log 71/771/1
Shawn O. Pearce [Fri, 28 May 2010 21:58:42 +0000 (14:58 -0700)]
eclipse-iplog: Add new consumes element to IP log

This element documents other Eclipse projects that are consumed by
this project.  JGit doesn't consume any projects, but its sister
project EGit does.

Change-Id: Ie922d27c580f6742e2acb051815a381af48df7ca
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years agoUpdate cache of IPzilla CQ 3904 68/768/1
Shawn O. Pearce [Fri, 28 May 2010 16:48:38 +0000 (09:48 -0700)]
Update cache of IPzilla CQ 3904

Change-Id: I552e77e434487ece9a600de9e09bc17066579773
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years agoeclipse-ipzilla: Correctly parse result with empty last field 67/767/1
Shawn O. Pearce [Fri, 28 May 2010 21:30:27 +0000 (14:30 -0700)]
eclipse-ipzilla: Correctly parse result with empty last field

If the last field of our IPzilla query comes back empty, we were
skipping over and not including it in the result List, causing an
IndexOutOfBoundsException when it was read into our data model.

If the last field is empty, actually add the empty string.

Change-Id: Ib18b335990c73e036b185199d0004f4ffc395867
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years agoExternalize strings in Commit command 66/766/1
Shawn O. Pearce [Fri, 28 May 2010 16:44:10 +0000 (09:44 -0700)]
Externalize strings in Commit command

Change-Id: I08b2cdd147ad6f3c67795f2e75a5b5d2ce2fb20f
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years agoDon't use interruptable pread() to access pack files 59/759/1
Shawn O. Pearce [Thu, 27 May 2010 02:00:06 +0000 (19:00 -0700)]
Don't use interruptable pread() to access pack files

The J2SE NIO APIs require that FileChannel close the underlying file
descriptor if a thread is interrupted while it is inside of a read or
write operation on that channel.  This is insane, because it means we
cannot share the file descriptor between threads.  If a thread is in
the middle of the FileChannel variant of IO.readFully() and it
receives an interrupt, the pack will be automatically closed on us.
This causes the other threads trying to use that same FileChannel to
receive IOExceptions, which leads to the pack getting marked as
invalid.  Once the pack is marked invalid, JGit loses access to its
entire contents and starts to report MissingObjectExceptions.

Because PackWriter must ensure that the chosen pack file stays
available until the current object's data is fully copied to the
output, JGit cannot simply reopen the pack when its automatically
closed due to an interrupt being sent at the wrong time.  The pack may
have been deleted by a concurrent `git gc` process, and that open file
descriptor might be the last reference to the inode on disk.  Once its
closed, the PackWriter loses access to that object representation, and
it cannot complete sending the object the client.

Fortunately, RandomAccessFile's readFully method does not have this
problem.  Interrupts during readFully() are ignored.  However, it
requires us to first seek to the offset we need to read, then issue
the read call.  This requires locking around the file descriptor to
prevent concurrent threads from moving the pointer before the read.

This reduces the concurrency level, as now only one window can be
paged in at a time from each pack.  However, the WindowCache should
already be holding most of the pages required to handle the working
set for a process, and its own internal locking was already limiting
us on the number of concurrent loads possible.  Provided that most
concurrent accesses are getting hits in the WindowCache, or are for
different repositories on the same server, we shouldn't see a major
performance hit due to the more serialized loading.

I would have preferred to use a pool of RandomAccessFiles for each
pack, with threads borrowing an instance dedicated to that thread
whenever they needed to page in a window.  This would permit much
higher levels of concurrency by using multiple file descriptors (and
file pointers) for each pack.  However the code became too complex to
develop in any reasonable period of time, so I've chosen to retrofit
the existing code with more serialization instead.

Bug: 308945
Change-Id: I2e6e11c6e5a105e5aef68871b66200fd725134c9
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years agoAdd a merge command to the jgit API 23/723/6
Stefan Lay [Thu, 20 May 2010 13:09:39 +0000 (15:09 +0200)]
Add a merge command to the jgit API

Merges the current head with one other commit.
In this first iteration the merge command supports
only fast forward and already up-to-date.

Change-Id: I0db480f061e01b343570cf7da02cac13a0cbdf8f
Signed-off-by: Stefan Lay <stefan.lay@sap.com>
Signed-off-by: Christian Halstrick <christian.halstrick@sap.com>
Signed-off-by: Chris Aniszczyk <caniszczyk@gmail.com>
14 years agoAdded merge support to CommitCommand 21/721/4
Christian Halstrick [Wed, 19 May 2010 08:01:25 +0000 (10:01 +0200)]
Added merge support to CommitCommand

The CommitCommand should take care to create a merge commit if the file
$GIT_DIR/MERGE_HEAD exists. It should then read the parents for the merge
commit out of this file. It should also take care that when commiting
a merge and no commit message was specified to read the message from
$GIT_DIR/MERGE_MSG.
Finally the CommitCommand should remove these files if the commit
succeeded.

Change-Id:  I4e292115085099d5b86546d2021680cb1454266c
Signed-off-by: Christian Halstrick <christian.halstrick@sap.com>
14 years agoTest root locale translations 29/729/1
Sasa Zivkov [Wed, 19 May 2010 21:37:33 +0000 (14:37 -0700)]
Test root locale translations

Ensures all translations exist in the root locale.

Change-Id: Ic8a8bdfd4a06c6d1ebd1e85a8082a32c82d155c7

14 years agoExternalize strings from JGit 98/598/4
Sasa Zivkov [Wed, 19 May 2010 14:59:28 +0000 (16:59 +0200)]
Externalize strings from JGit

The strings are externalized into the root resource bundles.
The resource bundles are stored under the new "resources" source
folder to get proper maven build.

Strings from tests are, in general, not externalized. Only in
cases where it was necessary to make the test pass the strings
were externalized. This was typically necessary in cases where
e.getMessage() was used in assert and the exception message was
slightly changed due to reuse of the externalized strings.

Change-Id: Ic0f29c80b9a54fcec8320d8539a3e112852a1f7b
Signed-off-by: Sasa Zivkov <sasa.zivkov@sap.com>
14 years agoFix SSH deadlock during OutOfMemoryError 08/708/3
Shawn O. Pearce [Sun, 16 May 2010 00:01:49 +0000 (17:01 -0700)]
Fix SSH deadlock during OutOfMemoryError

In close() method of SshFetchConnection and SshPushConnection
errorThread.join() can wait forever if JSch will not close the
channel's error stream.  Join with a timeout, and interrupt the
copy thread if its blocked on data that will never arrive.

Bug: 312863
Change-Id: I763081267653153eed9cd7763a015059338c2df8
Reported-by: Dmitry Neverov <dmitry.neverov@gmail.com>
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years agoFix race condition in StreamCopyThread 28/728/1
Dmitry Neverov [Wed, 19 May 2010 18:39:17 +0000 (11:39 -0700)]
Fix race condition in StreamCopyThread

If we get an interrupt during an IO operation (src.read or dst.write)
caused by the flush() method incrementing the flush counter, ensure
we restart the proper section of code.  Just ignore the interrupt
and continue running.

Bug: 313082
Change-Id: Ib2b37901af8141289bbac9807cacf42b4e2461bd
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years agoRemove unnecessary truncation of in-pack size during copy 12/712/2
Shawn O. Pearce [Sun, 16 May 2010 02:10:47 +0000 (19:10 -0700)]
Remove unnecessary truncation of in-pack size during copy

The number of bytes to copy was truncated to an int, but the
pack's copyToStream() method expected to be passed a long here.
Pass through the long so we don't truncate a giant object.

Change-Id: I0786ad60a3a33f84d8746efe51f68d64e127c332
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years agoReduce the size of PackWriter's ObjectToPack instances 10/710/1
Shawn O. Pearce [Sun, 16 May 2010 00:51:03 +0000 (17:51 -0700)]
Reduce the size of PackWriter's ObjectToPack instances

Rather than holding onto the PackedObjectLoader, only hold the
PackFile and the object offset.  During a reuse copy that is all
we should need to complete a reuse, and the other parts of the
PackedObjectLoader just waste memory.

This change reduces the per-object memory usage of a PackWriter by
32 bytes on a 32 bit JVM using only OFS_DELTA formatted objects.
The savings is even larger (by another 20 bytes) for REF_DELTAs.
This is close to a 50% reduction in the size of ObjectToPack,
making it rather worthwhile to do.

Beyond the memory reduction, this change will help to make future
refactoring work easier.  We need to redo the API used to support
copying data, and disconnecting it from the PackedObjectLoader is
a good first step.

Change-Id: I24ba4e621e101f14e79a16463aec5379f447aa9b
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years agoReduce size of PackedObjectLoader by dropping long to int 09/709/1
Shawn O. Pearce [Sun, 16 May 2010 00:37:14 +0000 (17:37 -0700)]
Reduce size of PackedObjectLoader by dropping long to int

Rather than keep track of both the position of the object, and the
position of its data, just keep track of the number of bytes used
by the object's header in the pack.  This shaves 4 bytes out of the
size of the PackedObjectLoader instances.

We also can defer the addition instruction to the materialize()
operation, avoiding it entirely if the caller never actually uses
the loader.  This may be relevant for PackWriter invocations,
where only 1 loader gets chosen for a given object, even though
the object may appear on disk in more than one pack file.

Error reporting is now simplified, as we can rely on the object
offset rather than its data offset.  This is the value displayed
by pack debugging tools like `git verify-pack -v`, so its better
to use that in our own errors.

Because nobody needs getDataOffset() now, we can drop that from
the public API.

Change-Id: Ic639c0d5a722315f4f5c8ffda6e26643d90e5f42
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years agoFactor out duplicate Inflater setup in WindowCursor 06/706/1
Shawn O. Pearce [Sat, 15 May 2010 23:18:44 +0000 (16:18 -0700)]
Factor out duplicate Inflater setup in WindowCursor

Since we use this code twice, pull it into a private method.  Let
the compiler/JIT worry about whether or not this logic should be
inlined into the call sites.

Change-Id: Ia44fb01e0328485bcdfd7af96835d62b227a0fb1
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years agoSquash OffsetCache into WindowCache 05/705/2
Shawn O. Pearce [Fri, 14 May 2010 22:02:31 +0000 (15:02 -0700)]
Squash OffsetCache into WindowCache

Originally when I wrote this code I had hoped to use OffsetCache
to also implement the UnpackedObjectCache.  But it turns out they
need rather different code, and it just wasn't worth trying to
reuse the OffsetCache base class.

Before doing any major refactoring or code cleanups here, squash the
two classes together and delete OffsetCache.  As WindowCache is our
only subclass, this is pretty simple to do.  We also get a minor
code reduction due to less duplication between the two classes,
and the JIT should be able to do a better job of optimization here
as we can define types up front rather than relying on generics
that erase back to java.lang.Object.

Change-Id: Icac8bda01260e405899efabfdd274928e98f3521
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years agoAvoid unnecessary second read on OBJ_OFS_DELTA headers 04/704/1
Shawn O. Pearce [Fri, 14 May 2010 21:03:32 +0000 (14:03 -0700)]
Avoid unnecessary second read on OBJ_OFS_DELTA headers

When we read the object header we copy 20 bytes from the pack data,
then start parsing out the type and the inflated size.  For most
objects, this is only going to require 3 bytes, which is sufficient
to represent objects with inflated sizes of up to 2^16.  The local
buffer however still has 17 bytes remaining in it, and that can be
used to satisfy the OBJ_OFS_DELTA header.

We shouldn't need to worry about walking off the end of the buffer
here, because delta offsets cannot be larger than 64 bits, and that
requires only 9 bytes in the OFS_DELTA encoding.

Assuming worst-case scenarios of 9 bytes for the OFS_DELTA encoding,
the pack file itself must be approaching 2^64 bytes, an infeasible
size to store on any current technology.  However, even if this
were the case we still have 11 bytes for the type/size header.
In that encoding we can represent an object as large as 2^74 bytes,
which is also an infeasible size to process in JGit.

So drop the second read here.

The data offsets we pass into the ObjectLoaders being constructed
need to be computed individually now.  This saves a local variable,
but pushes the addition operation into each branch of the switch.

Change-Id: I6cf64697a9878db87bbf31c7636c03392b47a062
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years agoFix hang when fetching over SSH 96/696/1
Shawn O. Pearce [Thu, 13 May 2010 17:23:33 +0000 (10:23 -0700)]
Fix hang when fetching over SSH

JSch may hang or abort with the timeout if JGit connects before
its obtained the streams.  Instead defer the connect() call until
after the streams have been configured.

Bug: 312383
Change-Id: I7c3a687ba4cb69a41a85e2b60d381d42b9090e3f
Reported-by: Dmitry Neverov <dmitry.neverov@gmail.com>
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years agoFix interrupted write in StreamCopyThread 95/695/1
Shawn O. Pearce [Thu, 13 May 2010 16:56:15 +0000 (09:56 -0700)]
Fix interrupted write in StreamCopyThread

If a flush() gets delivered at the same time that we are blocking
while writing to an interruptable stream, the copy thread will
abort assuming its a stream error.  Instead ignore the interrupt,
and retry the write.

Change-Id: Icbf62d1b8abe0fabbb532dbee088020eecf4c6c2
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years agoFix missing flush in StreamCopyThread 94/694/1
Dmitry Neverov [Thu, 13 May 2010 16:50:44 +0000 (09:50 -0700)]
Fix missing flush in StreamCopyThread

It is possible to miss flush() invocation in StreamCopyThread.
In this case some data will not be sent to remote host and we will
wait forever (or until timeout) in src.read().

Use a counter to keep track of the flush requests.

Change-Id: Ia818be9b109a1674d9e2a9c78e125ab248cfb75b
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
14 years agoAdd commit command to the pgm package 03/603/3
Christian Halstrick [Mon, 26 Apr 2010 13:59:59 +0000 (15:59 +0200)]
Add commit command to the pgm package

The commit command is added using the new Git class. Currently
this supports only the author and commit-message option.

Change-Id: I13152575b5b03f6f9e816d0747e7a8c5c6fccade
Signed-off-by: Christian Halstrick <christian.halstrick@sap.com>
14 years agoMerge "Fix Maven Javadoc generation problem"
Matthias Sohn [Tue, 11 May 2010 23:59:03 +0000 (19:59 -0400)]
Merge "Fix Maven Javadoc generation problem"

14 years agoFix Maven Javadoc generation problem 85/685/1
Jonathan Gossage [Tue, 11 May 2010 23:06:28 +0000 (18:06 -0500)]
Fix Maven Javadoc generation problem

There is a serious problem with the Maven Javadoc plugin. Please see
http://jira.codehaus.org/browse/MJAVADOC-275
for details. This problem is fixed by using maven-javadoc-plugin V2.7
instead of maven-javadoc-plugin v2.6.1.

14 years agoAdd missing pom dependency to org.eclipse.jgit.junit 84/684/1
Matthias Sohn [Tue, 11 May 2010 19:53:42 +0000 (21:53 +0200)]
Add missing pom dependency to org.eclipse.jgit.junit

Change-Id: I86f292039f1b8e499baf05238f55b1d550d098a5
Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
14 years agoExpose org.eclipse.jgit.junit via jgit p2 repository 78/678/1
Matthias Sohn [Tue, 11 May 2010 12:35:26 +0000 (14:35 +0200)]
Expose org.eclipse.jgit.junit via jgit p2 repository

EGit Tycho builds on build.eclipse.org frequently hit corrupted artifacts
which leads to broken builds. Cleaning up these corrupted files is tedious
since it requires file system access on the build server. Hence we want to
switch to use job-local m2 repositories. This requires that build artifacts
are shared between the jgit and egit build jobs via p2. Therefore the
bundle org.eclipse.jgit.junit needs to be exposed via p2 repository.

Change-Id: I0ccd7763eede117cb68240fdd25f13d6e6f6a2c1
Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
14 years agoAdd builder-style API to jgit and Commit & Log cmd 91/591/10
Christian Halstrick [Mon, 26 Apr 2010 11:35:30 +0000 (13:35 +0200)]
Add builder-style API to jgit and Commit & Log cmd

Added a new package org.eclipse.jgit.api and a builder-style API for
jgit. Added also the first implementation for two git commands: Commit
and Log.

This API is intended to be used by external components when
functionalities of the standard git commands are required. It will also
help to ease writing JGit tests.

For internal usages this API may often not be optimal because the git
commands are doing much more than required or they expect parameters of
an unappropriate type.

Change-Id: I71ac4839ab9d2f848307eba9252090c586b4146b
Signed-off-by: Christian Halstrick <christian.halstrick@sap.com>
14 years agoMerge "Added MERGING_RESOLVED repository state"
Robin Rosenberg [Sat, 8 May 2010 21:16:26 +0000 (17:16 -0400)]
Merge "Added MERGING_RESOLVED repository state"

14 years agoMerge "A stages field and getter for GitIndex entry introduced"
Robin Rosenberg [Sat, 8 May 2010 21:13:25 +0000 (17:13 -0400)]
Merge "A stages field and getter for GitIndex entry introduced"