A commit A can reach a commit B only if the generation number of A is
strictly larger than the generation number of B. This condition allows
significantly short-circuiting commit-graph walks.
On a copy of the Linux repository where HEAD is contained in v6.3-rc4
but no earlier tag, the command 'git tag --contains HEAD' of
ListTagCommand#call() had the following peformance improvement:
(excluded the startup time of the repo)
Before: 2649ms (core.commitgraph=true)
11909ms (core.commitgraph=false)
After: 91ms (core.commitgraph=true)
11934ms (core.commitgraph=false)
Bug: 574368
Change-Id: Ia2efaa4e9ae598266f72e70eb7e3b27655cbf85b
Signed-off-by: kylezhao <kylezhao@tencent.com>
* stable-6.5:
Ensure parsed RevCommitCG has derived data from commit-graph
Downgrade maven-site-plugin to 3.12.1
Use wagon-ssh-external to deploy Maven site
Change-Id: Ide721fb088fa04f6276ac495968a45e732f6e139
If a RevCommitCG was newly created and called #parseCanonical(RevWalk,
byte[]) method immediately, its flag was marked as PARSED, but no
derived data was obtained from the commit-graph. This is different from
what we expected.
Change-Id: I5d417efa3c42d211f19e6acf255f761e84d84450
Signed-off-by: kylezhao <kylezhao@tencent.com>
Running
bazel test //org.eclipse.jgit.test:org_eclipse_jgit_patch_PatchApplierTest
results in a DefaultCharset error on my machine, which can be avoided
by explicitly specifying a charset when calling getBytes on a string. In
these tests, the charset used doesn't really matter, so go with UTF_8.
Signed-off-by: Jonathan Tan <jonathantanmy@google.com>
Change-Id: Id7721cc5b7ea650e77c2db47042715487983cae6
During my development of Id7721cc5b7ea650e77c2db47042715487983cae6, I
have found this test to be flaky when run by CI. As a speculative fix,
mark this test as @Ignore so it won't be run.
Signed-off-by: Jonathan Tan <jonathantanmy@google.com>
Change-Id: Idfe04d7f1fb72a772d4c8d249ca86a9c2eec0b1a
Introduce a SnapshottingRefDirectory class which allows users to get
a snapshot of the ref database and use it in a request scope (for
example a Gerrit query) instead of having to re-read packed-refs
several times in a request.
This can potentially be further improved to avoid scanning/reading a
loose ref several times in a request. This would especially help
repeated lookups of a packed ref, where we check for the existence of
a loose ref each time.
Change-Id: I634b92877f819f8bf36a3b9586bbc1815108189a
Signed-off-by: Kaushik Lingarkar <quic_kaushikl@quicinc.com>
Since the first attempt to read a ref is not expected to trigger
a RefsChangedEvent, update the test to ensure 'lastNotifiedModCnt'
is not 0 before we start the actual work. The test has been passing
luckily because createBareRepository in setUp() happens to bump
'lastNotifiedModCnt'.
Change-Id: Ibd981f677920e8c3b965aa742fe669c42b8c1c93
Signed-off-by: Kaushik Lingarkar <quic_kaushikl@quicinc.com>
By default, Git will report, to the server, commits reachable from all local refs to find common commits in an attempt to reduce the size of the to-be-received packfile. If specified with negotiation tip, Git will only report commits reachable from the given tips. This is useful to speed up fetches when the user knows which local ref is likely to have commits in common with the upstream ref being fetched.
When negotation-tip is on, use the wanted refs instead of all refs as source of the "have" list to send.
This is controlled by the `fetch.usenegotationtip` flag, false by default. This works only for programmatic fetches and there is no support for it yet in the CLI.
Change-Id: I19f8fe48889bfe0ece7cdf78019b678ede5c6a32
Support the new option index.skipHash which was introduced in git 2.40
[1]. If it is set to true skip computing the git index checksum. This
accelerates Git commands that manipulate the index, such as git add, git
commit, or git status. Instead of storing the checksum, write a trailing
set of bytes with value zero, indicating that the computation was
skipped.
Accept a skipped checksum consisting of 20 null bytes when reading the
index since the option could have been set to true at the time when the
index was written.
[1] https://git-scm.com/docs/git-config#Documentation/git-config.txt-indexskipHash
Bug: 581723
Change-Id: I28ebe44c5ca1cbcb882438665d686452a0c111b2
Some test data introduced in [1] includes code with Apache
license. This triggers warnings in license analyzers.
Remove the licensed files and the test case as a quick relief. We
should restore the test with appropiate data in a follow-up change.
[1] https://git.eclipse.org/r/c/jgit/jgit/+/197724
Change-Id: I42670dc7d994f77d2c7f2c2156bcf1e112374022
1. For general errors, throw IOException instead of wrapping them with
PatchApplyException. The wrapping was moved (back) to ApplyCommand.
2. For file specific errors, log the errors as part of
PatchApplier::Result.
3. Change applyPatch() to receive the parsed Patch object, so the caller
can decide how to handle parsing errors.
Background: this utility class was extracted from ApplyCommand on V6.4.0.
During the extraction, we left the exception wrapping by
PatchApplyException intact. This attitude made it harder for the callers to
distinguish between the actual error causes.
Change-Id: Ib0f2b5e97a13df2339d8b65f2fea1c819c161ac3
When commit-graph file already exists in the repository, a newly
created FileCommitGraph didn't scan CommitGraph until the file was
modified, resulting in wrong result.
Change-Id: Ic85676f2d3b6a88f3ae28d4065729926b6fb2f23
Signed-off-by: kylezhao <kylezhao@tencent.com>
As of today, we don't have test coverage for RefDirectory when
core.trustPackedRefsStat config is set to after_open. Thus create new
test classes which set core.trustPackedRefsStat config to after_open in
setup and extend RefDirectoryTest and FileRepositoryBuilderTest
respectively.
Change-Id: I1db6fcf414bc488106ad4c85fb934480f299c995
Signed-off-by: Prudhvi Akhil Alahari <quic_prudhvi@quicinc.com>
The native git option to preserve merge commits during rebase
has been changed from pull.rebase=preserve to pull.rebase=merges.
This changeset in jgit makes the same config change. The old "preserve"
option is no longer recognized and is replaced by new option called
"merges".
This makes jgit's rebase configuration compatible with native git
versions 2.34 and newer where the old "preserve" option has been
removed.
Change-Id: Ic07ff954e258115e76465a1593ef3259f4c418a3
Display elapsed time per task if enabled via
ProgressMonitor#showDuration or if system property or environment
variable GIT_TRACE_PERFORMANCE is set to "true". If both the system
property and the environment variable are set the system property takes
precedence.
E.g. using jgit CLI:
$ GIT_TRACE_PERFORMANCE=true jgit clone https://foo.bar/foobar
Cloning into 'foobar'...
remote: Counting objects: 1 [0.002s]
remote: Finding sources: 100% (15531/15531) [0.006s]
Receiving objects: 100% (169737/169737) [13.045s]
Resolving deltas: 100% (67579/67579) [1.842s]
Change-Id: I4d624e7858b286aeddbe7d4e557589986d73659e
PackWriter callers tell the writer what do the want to include in the
pack and invoke #writePack(). Afterwards, they can invoke #writeIndex()
to write the corresponding pack index.
Mirror this for the object-size index, adding a #writeObjectSizeIndex()
method.
Change-Id: Ic319975c72c239cd6488303f7d4cced797e6fe00
* stable-6.4:
If tryLock fails to get the lock another gc has it
Fix GcConcurrentTest#testInterruptGc
Don't swallow IOException in GC.PidLock#lock
Check if FileLock is valid before using or releasing it
Change-Id: Ia2797b44a60342eb9df53f0b3d674cba92a512fc
* stable-6.3:
If tryLock fails to get the lock another gc has it
Fix GcConcurrentTest#testInterruptGc
Don't swallow IOException in GC.PidLock#lock
Check if FileLock is valid before using or releasing it
Change-Id: I5af34c92e423a651db53b4dc45ed844d5f39910d
* stable-6.2:
If tryLock fails to get the lock another gc has it
Fix GcConcurrentTest#testInterruptGc
Don't swallow IOException in GC.PidLock#lock
Check if FileLock is valid before using or releasing it
Change-Id: I5b6b10622b61fde3f0f10455a74ae159a0b69082
* stable-6.1:
If tryLock fails to get the lock another gc has it
Fix GcConcurrentTest#testInterruptGc
Don't swallow IOException in GC.PidLock#lock
Check if FileLock is valid before using or releasing it
Change-Id: I3ffe92566cc145053bb762f612dd96bc6d542c62
* stable-6.0:
If tryLock fails to get the lock another gc has it
Fix GcConcurrentTest#testInterruptGc
Don't swallow IOException in GC.PidLock#lock
Check if FileLock is valid before using or releasing it
Change-Id: Idea23e555c024557d7e39a86efe25f609400b962
* stable-5.13:
If tryLock fails to get the lock another gc has it
Fix GcConcurrentTest#testInterruptGc
Don't swallow IOException in GC.PidLock#lock
Check if FileLock is valid before using or releasing it
Change-Id: I708d0936fa86b028e4da4e7e21f332f8b48ad293
C git 2.11 supports setting the equivalent of RequestPolicy.ANY with
uploadpack.allowAnySHA1InWant[1]. Parse this into TransportConfig and
use it from UploadPack.
Add additional tests for [2] and this change.
We can execute "git clone --filter=blob:none --no-checkout" successfully
with config uploadPack.allowFilter is true. But when we checkout, the
git will fetch other missing objects required by the checkout(this is
why we need this config).
When both uploadPack.allowFilter and uploadPack.allowAnySHA1InWant are
true, jgit will support partial clone. If you are using an extremely
large monorepo, this feature can help. It allows users to work on an
incomplete repo which reduces disk usage.
[1] f8edeaa05d
[2] change Id39771a6e42d8082099acde11249306828a053c0
Bug: 573390
Change-Id: I8fe75f03bf1fea7c11e0d67c8637bd05dd1f9b89
Signed-off-by: kylezhao <kylezhao@tencent.com>
Operations like "clone --filter=blob:limit=N" or the "object-info"
command need to read the size of the objects from the storage. An
index would provide those sizes at once rather than having to seek in
the packfile.
Introduce an interface for the Object-size index. This index returns
the inflated size of an object. Not all objects could be indexed (to
limit memory usage).
This implementation indexes only blobs (no trees, nor
commits) *above* certain size threshold (configurable). Lower
threshold adds more objects to the index, consumes more memory and
provides better performance. 0 means "all blobs" and -1 "disabled".
If we don't index everything, for the filter use case is more
efficient to index the biggest objects first: the set is small and
most objects are filtered by NOT being in the index. For the
object-size, the more objects in the index the better, regardless
their size. All together, it is more helpful to index above threshold.
Change-Id: I9ed608ac240677e199b90ca40d420bcad9231489
The object size index stores positions of objects in the main
index (when ordered by sha1). These positions are per-pack and usually
a pack has <16 million objects (there are exceptions but rather
rare). It could save some memory storing these positions in three bytes
instead of four. Note that these positions are sorted and always positive.
Implement a wrapper around a byte[] to access and search "ints" while
they are stored as unsigned 3 bytes.
Change-Id: Iaa26ce8e2272e706e35fe4cdb648fb6ca7591972
The primary index returns the offset in the pack for an
objectId. Internally it keeps the object-ids in lexicographical order,
but doesn't expose an API to find the position of an object-id in that
list. This is needed for the object-size index, that we want to store
as "position-in-idx, size".
Add a #findPosition(object-id) method to the PackIndex interface to
know where an object-id sits in the ordered list of ids in the pack.
Note that this index position is over the list of ordered object-ids,
while reverse-index position is over the list of objects in packed
order.
Change-Id: I89fa146599e347a26d3012d3477d7f5bbbda7ba4
JGit knows how to read/write commit graphs but the DFS stack is not
using it yet.
The DFS garbage collector generates a commit-graph with commits
reachable from any ref. The pack is stored as extra stream in the GC
pack. DfsPackFile mimicks how other indices are loaded storing the
reference in DFS cache.
Signed-off-by: Xing Huang <xingkhuang@google.com>
Change-Id: I3f94997377986d21a56b300d8358dd27be37f5de
The 'size' packet line is an argument, so it
must be preceeded by a 0001 delimiter. See also git's
t5701-git-serve.sh test,
https://github.com/git/git/blob/8b8d9a2/t/t5701-git-serve.sh#L329
Without this fix, the server will choke on the delimiter line, saying
PackProtocolException: unexpected <empty string>
To test, I ran Gerrit locally with this fix
$ curl -X POST -H 'git-protocol: version=2' -H 'content-type:
application/x-git-upload-pack-request' -H 'accept:
application/x-git-upload-pack-result' --data
$'0018command=object-info\n00010009size\n0031oid
d38b1b92bdb2893eb4505667375563f2d6d4086b\n0000'
http://localhost:8080/git.git/git-upload-pack
=>
0008size0032d38b1b92bdb2893eb4505667375563f2d6d4086b 268590000
The same command completes identically on Gitlab (which supports the
object-info command)
$ curl -X POST -H 'git-protocol: version=2' -H 'content-type:
application/x-git-upload-pack-request' -H 'accept:
application/x-git-upload-pack-result' --data
$'0018command=object-info\n00010009size\n0031oid
d38b1b92bdb2893eb4505667375563f2d6d4086b\n0000'
https://gitlab.com/gitlab-org/git.git/git-upload-pack
=>
0008size0032d38b1b92bdb2893eb4505667375563f2d6d4086b 268590000
In this case, the blob is for the COPYING file in the Git source tree,
which is 26859 bytes long.
Change-Id: Ief4ce1eb9303a3b2479547d7950ef01c7c28f472
This change only affects inCore repositories.
Before this change, any file that wasn't part of the patch
wasn't read, and therefore wasn't part of the output tree.
Change-Id: I246ef957088f17aaf367143f7a0b3af0f8264ffb
Bug: Google b/267270348
* stable-6.4:
Shortcut during git fetch for avoiding looping through all local refs
FetchCommand: fix fetchSubmodules to work on a Ref to a blob
Silence API warnings introduced by I466dcde6
Allow the exclusions of refs prefixes from bitmap
PackWriterBitmapPreparer: do not include annotated tags in bitmap
BatchingProgressMonitor: avoid int overflow when computing percentage
Speedup GC listing objects referenced from reflogs
FileSnapshotTest: Add more MISSING_FILE coverage
Change-Id: Id0ebfbd85eb815716383b9495eb7dd1f54cf4d74
* stable-6.3:
Shortcut during git fetch for avoiding looping through all local refs
FetchCommand: fix fetchSubmodules to work on a Ref to a blob
Silence API warnings introduced by I466dcde6
Allow the exclusions of refs prefixes from bitmap
PackWriterBitmapPreparer: do not include annotated tags in bitmap
BatchingProgressMonitor: avoid int overflow when computing percentage
Speedup GC listing objects referenced from reflogs
FileSnapshotTest: Add more MISSING_FILE coverage
Change-Id: Iefcf5d832bd0087c1027876f2200689e1150abce
* stable-6.2:
Shortcut during git fetch for avoiding looping through all local refs
FetchCommand: fix fetchSubmodules to work on a Ref to a blob
Silence API warnings introduced by I466dcde6
Allow the exclusions of refs prefixes from bitmap
PackWriterBitmapPreparer: do not include annotated tags in bitmap
BatchingProgressMonitor: avoid int overflow when computing percentage
Speedup GC listing objects referenced from reflogs
FileSnapshotTest: Add more MISSING_FILE coverage
Change-Id: I2ff386d9a096277360e6c7bd5535b49984620fb3
* stable-6.1:
Shortcut during git fetch for avoiding looping through all local refs
FetchCommand: fix fetchSubmodules to work on a Ref to a blob
Silence API warnings introduced by I466dcde6
Allow the exclusions of refs prefixes from bitmap
PackWriterBitmapPreparer: do not include annotated tags in bitmap
BatchingProgressMonitor: avoid int overflow when computing percentage
Speedup GC listing objects referenced from reflogs
FileSnapshotTest: Add more MISSING_FILE coverage
Change-Id: Iff2fba026b49463016015b2fae1a42cf76ee2dbb
* stable-6.0:
Shortcut during git fetch for avoiding looping through all local refs
FetchCommand: fix fetchSubmodules to work on a Ref to a blob
Silence API warnings introduced by I466dcde6
Allow the exclusions of refs prefixes from bitmap
PackWriterBitmapPreparer: do not include annotated tags in bitmap
BatchingProgressMonitor: avoid int overflow when computing percentage
Speedup GC listing objects referenced from reflogs
FileSnapshotTest: Add more MISSING_FILE coverage
Change-Id: Ib5055f2f3b8a313c178d6f6c7c5630285ad5a726
* stable-5.13:
Shortcut during git fetch for avoiding looping through all local refs
FetchCommand: fix fetchSubmodules to work on a Ref to a blob
Silence API warnings introduced by I466dcde6
Allow the exclusions of refs prefixes from bitmap
PackWriterBitmapPreparer: do not include annotated tags in bitmap
BatchingProgressMonitor: avoid int overflow when computing percentage
Speedup GC listing objects referenced from reflogs
FileSnapshotTest: Add more MISSING_FILE coverage
Change-Id: I58ad4c210a5e7e5a1ba6b22315b04211c8909950
When running a GC.repack() against a repository with over one
thousands of refs/heads and tens of millions of ObjectIds,
the calculation of all bitmaps associated with all the refs
would result in an unreasonable big file that would take up to
several hours to compute.
Test scenario: repo with 2500 heads / 10M obj Intel Xeon E5-2680 2.5GHz
Before this change: 20 mins
After this change and 2300 heads excluded: 10 mins (90s for bitmap)
Having such a large bitmap file is also slow in the runtime
processing and have negligible or even negative benefits, because
the time lost in reading and decompressing the bitmap in memory
would not be compensated by the time saved by using it.
It is key to preserve the bitmaps for those refs that are mostly
used in clone/fetch and give the ability to exlude some refs
prefixes that are known to be less frequently accessed, even
though they may actually be actively written.
Example: Gerrit sandbox branches may even be actively
used and selected automatically because its commits are very
recent, however, they may bloat the bitmap, making it ineffective.
A mono-repo with tens of thousands of developers may have
a relatively small number of active branches where the
CI/CD jobs are continuously fetching/cloning the code. However,
because Gerrit allows the use of sandbox branches, the
total number of refs/heads may be even tens to hundred
thousands.
Change-Id: I466dcde69fa008e7f7785735c977f6e150e3b644
Signed-off-by: Luca Milanesio <luca.milanesio@gmail.com>
The annotated tags should be excluded from the bitmap associated
with the heads-only packfile. However, this was not happening
because of the check of exclusion of the peeled object instead
of the objectId to be excluded from the bitmap.
Sample use-case:
refs/heads/main
^
|
commit1 <-- commit2 <- annotated-tag1 <- tag1
^
|
commit0
When creating a bitmap for the above commit graph, before this
change all the commits are included (3 bitmaps), which is
incorrect, because all commits reachable from annotated tags
should not be included.
The heads-only bitmap should include only commit0 and commit1
but because PackWriterBitPreparer was checking for the peeled
pointer of tag1 to be excluded (commit2) which was not found in
the list of tags to exclude (annotated-tag1), the commit2 was
included, even if it wasn't reachable only from the head.
Add an additional check for exclusion of the original objectId
for allowing the exclusion of annotated tags and their pointed
commits. Add one specific test associated with an annotated tag
for making sure that this use-case is covered also.
Example repository benchmark for measuring the improvement:
# refs: 400k (2k heads, 88k tags, 310k changes)
# objects: 11M (88k of them are annotate tags)
# packfiles: 2.7G
Before this change:
GC time: 5h
clone --bare time: 7 mins
After this change:
GC time: 20 mins
clone --bare time: 3 mins
Bug: 581267
Signed-off-by: Luca Milanesio <luca.milanesio@gmail.com>
Change-Id: Iff2bfc6587153001837220189a120ead9ac649dc
When cloning huge repositories I observed percentage of object counts
turning negative. This happened if lastWork * 100 exceeded
Integer.MAX_VALUE.
Change-Id: Ic5f5cf5a911a91338267aace4daba4b873ab3900
RevWalk#createCommit() will inspect the commit-graph file to find the
specified object's graph position and then return a new RevCommitCG
instance.
RevCommitGC is a RevCommit with an additional "pointer" (the position)
to the commit-graph, so it can load the headers and metadata from there
instead of the pack. This saves IO access in walks where the body is not
needed (i.e. #isRetainBody is false and #parseBody is not invoked).
RevWalk uses automatically the commit-graph if available, no action
needed from callers. The commit-graph is fetched on first access from
the reader (that internally can keep it loaded and reuse it between
walks).
The startup cost of reading the entire commit graph is small. After
testing, reading a commit-graph with 1 million commits takes less than
50ms. If we use RepositoryCache, it will not be initialized util the
commit-graph is rewritten.
Bug: 574368
Change-Id: I90d0f64af24f3acc3eae6da984eae302d338f5ee
Signed-off-by: kylezhao <kylezhao@tencent.com>
Add a couple tests that confirm what the docs say about isModified() and
equals(MISSING_FILE) behavior.
Change-Id: I6093040ba3594934c3270331405a44b2634b97c5
Signed-off-by: Nasser Grainawi <quic_nasserg@quicinc.com>
In shallow repos, GC writes to the commit-graph that shallow commits
do not have parents. This won't be true after a "git fetch --unshallow"
(and before another GC).
Do not write the commit-graph from shallow clones of a repo. The
commit-graph must have the real metadata of commits and that is not
available in a shallow view of the repo.
Change-Id: Ic9f2358ddaa607c74f4dbf289c9bf2a2f0af9ce0
Signed-off-by: kylezhao <kylezhao@tencent.com>
A ternary search tree is a type of tree where nodes are arranged in a
manner similar to a binary search tree, but with up to three children
rather than the binary tree's limit of two.
Each node of a ternary search tree stores a single character, a
reference to a value object and references to its three children named
equal kid, lo kid and hi kid. The lo kid pointer must point to a node
whose character value is less than the current node. The hi kid pointer
must point to a node whose character is greater than the current
node.[1] The equal kid points to the next character in the word. Each
node in a ternary search tree represents a prefix of the stored strings.
All strings in the middle subtree of a node start with that prefix.
Like other prefix trees, a ternary search tree can be used as an
associative map with the ability for incremental string search. Ternary
search trees are more space efficient compared to standard prefix trees,
at the cost of speed.
They allow efficient prefix search which is important to implement
searching refs by prefix in a RefDatabase.
Searching by prefix returns all keys if the prefix is an empty string.
Bug: 576165
Change-Id: If160df70151a8e1c1bd6716ee4968e4c45b2c7ac
FileRepository's ObjectReader#getCommitGraph will return commit-graph
when it exists and core.commitGraph is true.
DfsRepository is not supported currently.
Change-Id: I992d43d104cf542797e6949470e95e56de025107
Signed-off-by: kylezhao <kylezhao@tencent.com>
If the last line came from the patch, use the patch to determine whether
or not there should be a trailing newline. Otherwise use the old text.
Add test cases for
- no newline at end, last line not in patch hunk
- no newline at end, last line in patch hunk
- patch removing the last newline
- patch adding a newline at the end of file not having one
all for core.autocrlf false, true, and input.
Add a test case where the "no newline" indicator line is not the last
line of the last hunk. This can happen if the patch ends with removals
at the file end.
Bug: 581234
Change-Id: I09d079b51479b89400ad300d0662c1dcb50deab6
Also-by: Yuriy Mitrofanov <a2terminator@mail.ru>
Signed-off-by: Thomas Wolf <twolf@apache.org>
This change makes JGit can read .git/objects/info/commit-graph file
and then get CommitGraph.
Loading a new commit-graph into memory requires additional time. After
testing, loading a copy of the Linux's commit-graph(1039139 commits)
is under 50ms.
Bug: 574368
Change-Id: Iadfdd6ed437945d3cdfdbe988cf541198140a8bf
Signed-off-by: kylezhao <kylezhao@tencent.com>
Some lines were too long, unnecessary fully qualified class names,
and an assertEquals(actual, expected) when it should have been
assertEquals(expected, actual).
Change-Id: I3b3c46c963afe2fb82a79c1e93970e73778877e5
Signed-off-by: Thomas Wolf <twolf@apache.org>
IO#readFully is often called with the intent to fill the destination
array from beginning to end. The redundant arguments for where to start
and stop filling are opportunities for bugs if specified incorrectly or
if not changed to match a changed array length.
Provide a overloaded method for filling the full destination array.
Change-Id: I964f18f4a061189cce1ca00ff0258669277ff499
Signed-off-by: Anna Papitto <annapapitto@google.com>
If 'core.commitGraph' and 'gc.writeCommitGraph' are both true, then gc
will rewrite the commit-graph file when 'git gc' is run. Defaults to
false while the commit-graph feature matures.
Bug: 574368
Change-Id: Ic94cd69034c524285c938414610f2e152198e06e
Signed-off-by: kylezhao <kylezhao@tencent.com>
Git introduced a new file storing the topology and some metadata of
the commits in the repo (commitGraph). With this data, git can browse
commit history without parsing the pack, speeding up e.g.
reachability checks.
This change teaches JGit to read commit-graph-format file, following
the upstream format([1]).
JGit can read a commit-graph file from a buffered stream, which means
that we can provide this feature for both FileRepository and
DfsRepository.
[1] https://git-scm.com/docs/commit-graph-format/2.21.0
Bug: 574368
Change-Id: Ib5c0d6678cb242870a0f5841bd413ad3885e95f6
Signed-off-by: kylezhao <kylezhao@tencent.com>
Deleting orphan files depends on .pack and .keep being reverse-sorted
to before the corresponding index files that could be orphans. The new
reverse index file extension (.rev) will break that frail dependency.
Rewrite Gc#deleteOrphans to avoid that dependence by tracking which pack
names have a .pack or .keep file and then deleting any index files that
without a corresponding one. This approach takes linear time instead of
the O(n logn) time needed for sorting.
Change-Id: If83c378ea070b8871d4b01ae008e7bf8270de763
Signed-off-by: Anna Papitto <annapapitto@google.com>
The NameConflictTreeWalk class is used in merge for iterating over
entries in commits. The class uses a separate iterator for each
commit's tree. In rare cases it can incorrectly report the same entry
twice. As a result, duplicated entries are added to the merge result
and later jgit throws an exception when it tries to process merge
result.
The problem appears only when there is a directory-file conflict for
the last item in trees. Example from the bug:
Commit 1:
* subtree - file
* subtree-0 - file
Commit 2:
* subtree - directory
* subtree-0 - file
Here the names are ordered like this:
"subtree" file <"subtree-0" file < "subtree" directory.
The NameConflictTreeWalk handles similar cases correctly if there are
other files after subtree... in commits - this is processed in the
AbstractTreeIterator.min function. Existing code has a special
optimization for the case, when all trees are pointed to the same
entry name - it skips additional checks. However, this optimization
incorrectly skips checks if one of trees reached the end.
The fix processes a situation when some trees reached the end, while
others are still point to an entry.
bug: 535919
Change-Id: I62fde3dd89779fac282479c093400448b4ac5c86
Keys used for identifying reverse indexes in the DfsBlockCache use a
custom subclass ForReverseIndex because there was no PackExt for them.
This conflates BlockCacheMetrics for reverse indexes with those for
packs, since the key falls back onto 0 when there is no extension.
Replace the custom ForReverseIndex with a DfsStreamKey usage to bring
keys for the new REVERSE_INDEX extension in line with INDEX and BITMAP
and separate reverse index and pack BlockCacheMetrics.
Change-Id: I305e2c16d2a8cb2a824855ea92e0c9a9b188fce5
Signed-off-by: Anna Papitto <annapapitto@google.com>
Make sure we always get consistent results, whether or not we have the
full data in the buffer.
Change-Id: Ieb379a0c375ad3dd352e63ac2f23bda6ef16c215
Signed-off-by: Thomas Wolf <twolf@apache.org>
* stable-6.3:
[benchmarks] Remove profiler configuration
Add SHA1 benchmark
[benchmarks] Set version of maven-compiler-plugin to 3.8.1
Fix running JMH benchmarks
Add option to allow using JDK's SHA1 implementation
Fix API breakage caused by extracting WorkTreeUpdater
Extract Exception -> HTTP status code mapping for reuse
Don't handle internal git errors as an HTTP error
Ignore IllegalStateException if JVM is already shutting down
Allow to perform PackedBatchRefUpdate without locking loose refs
Change-Id: Ib58879be292c54a2a7f4936ac0986997985c822b
* stable-6.2:
[benchmarks] Remove profiler configuration
Add SHA1 benchmark
[benchmarks] Set version of maven-compiler-plugin to 3.8.1
Fix running JMH benchmarks
Add option to allow using JDK's SHA1 implementation
Ignore IllegalStateException if JVM is already shutting down
Change-Id: I9c1576011c11b4ff8f453d18d9e786cee59860fa
* stable-6.1:
[benchmarks] Remove profiler configuration
Add SHA1 benchmark
[benchmarks] Set version of maven-compiler-plugin to 3.8.1
Fix running JMH benchmarks
Add option to allow using JDK's SHA1 implementation
Ignore IllegalStateException if JVM is already shutting down
Change-Id: Ie433c46a01a0f33848d54ecf99b30a44ca01e286
* stable-6.0:
[benchmarks] Remove profiler configuration
Add SHA1 benchmark
[benchmarks] Set version of maven-compiler-plugin to 3.8.1
Fix running JMH benchmarks
Add option to allow using JDK's SHA1 implementation
Ignore IllegalStateException if JVM is already shutting down
Change-Id: I176419026c3f4fdd8ebd34c61468c1ec3482ff45
* stable-5.13:
[benchmarks] Remove profiler configuration
Add SHA1 benchmark
[benchmarks] Set version of maven-compiler-plugin to 3.8.1
Fix running JMH benchmarks
Add option to allow using JDK's SHA1 implementation
Ignore IllegalStateException if JVM is already shutting down
Change-Id: I40105336f0b9e593a8a2c242a9557f854c274fdc
The change If6da9833 moved the computation of SHA1 from the JVM's
JCE to a pure Java implementation with collision detection.
The extra security for public sites comes with a cost of slower
SHA1 processing compared to the native implementation in the JDK.
When JGit is used internally and not exposed to any traffic from
external or untrusted users, the extra cost of the pure Java SHA1
implementation can be avoided, falling back to the previous
native MessageDigest implementation.
Bug: 580310
Change-Id: Ic24c0ba1cb0fb6282b8ca3025ffbffa84035565e
and update
- com.sun.jna to 5.12.1.v20221103-2317
- com.sun.jna.platform to 5.12.1.v20221103-2317
- org.bouncycastle.bcpg to 1.72.0.v20221013-1810
- org.bouncycastle.bcpkix to 1.72.0.v20221013-1810
- org.bouncycastle.bcprov to 1.72.0.v20221013-1810
- org.bouncycastle.bcutil to 1.72.0.v20221013-1810
- org.mockito.mockito-core to 4.8.1.v20221103-2317
- org.objenesis to 3.3.0.v20221103-2317
Change-Id: If00094d23e51d5f66928f83c1334aa6b18b98dfe
The NameConflictTreeWalk class is used in merge for iterating over
entries in commits. The class uses a separate iterator for each
commit's tree. In rare cases it can incorrectly report the same entry
twice. As a result, duplicated entries are added to the merge result
and later jgit throws an exception when it tries to process merge
result.
The problem appears only when there is a directory-file conflict for
the last item in trees. Example from the bug:
Commit 1:
* subtree - file
* subtree-0 - file
Commit 2:
* subtree - directory
* subtree-0 - file
Here the names are ordered like this:
"subtree" file <"subtree-0" file < "subtree" directory.
The NameConflictTreeWalk handles similar cases correctly if there are
other files after subtree... in commits - this is processed in the
AbstractTreeIterator.min function. Existing code has a special
optimization for the case, when all trees are pointed to the same
entry name - it skips additional checks. However, this optimization
incorrectly skips checks if one of trees reached the end.
The fix processes a situation when some trees reached the end, while
others are still point to an entry.
bug: 535919
Change-Id: I62fde3dd89779fac282479c093400448b4ac5c86
Before this change JGit did not support the session-id capability
implemented by native Git in UploadPack. This change implements
advertising the capability from the server and parsing the session-id
received from the client during an UploadPack operation.
Enable the transfer.advertisesid config setting to advertise the
capability from the server. The client may send a session-id capability
in response. If received, the value from this is parsed and available
via the getClientSID method on the UploadPack object.
This change does not add the capability to send a session-id from the
JGit client.
https://git-scm.com/docs/gitprotocol-capabilities#_session_idsession_id
Change-Id: Ib1b6929ff1b3a4528e767925b5e5c44b5d18182f
Signed-off-by: Josh Brown <sjoshbrown@google.com>
The config setting to enable advertising the session-id capability is
currently read in the ReceivePack class. This change moves it to a
common location in the TransferConfig class so that it can be reused
in other places like UploadPack. TransferConfig is also a more logical
place for the setting as it resides in the `transfer` config section.
Set the transfer.advertisesid setting to true to send the session-id
capability to the client.
Change-Id: If68ecb5e68b59f5c452a7992d02e3688b0a86747
Signed-off-by: Josh Brown <sjoshbrown@google.com>
In protocol V0 the client capabilities are appended to the first line.
Parsing session-id is currently only supported during a ReceivePack
operation. This change will parse the client session-id capability if
it has been sent by the client.
If the server sends the session-id capability to the client. The client
may respond with a session ID of its own. FirstWant.fromLine will now
parse the ID and make it available via the getClientSID method.
This change does not add support to send the session-id capability from
the server. The change is necessary to support session-id in UploadPack.
Change-Id: Id3fe44fdf9a72984ee3de9cf40cc4e71d434df4a
Signed-off-by: Josh Brown <sjoshbrown@google.com>
Before this change JGit did not support the session-id capability
implemented by native Git. This change implements advertising the
capability from the server and parsing the session-id received from
the client during a ReceivePack operation.
Enable the transfer.advertisesid config setting to advertise the
capability from the server. The client may send a session-id capability
in response. If received, the value from this is parsed and available
via the getClientSID method on the ReceivePack object. All capabilities
in the form `capability=value` are now split into key value pairs at the
first `=` character. This change replaces specific handling for the
agent capability.
This change does not add advertisement or parsing to UploadPack. This
change also does not add the ability to send a session ID from the JGit
client.
https://git-scm.com/docs/protocol-v2/2.33.0#_session_idsession_id
Change-Id: I56fb115e843b11b27e128c4ac427b05d5ec129d0
Signed-off-by: Josh Brown <sjoshbrown@google.com>
Reachability bitmaps are designed to speed up the "counting objects"
phase of generating a pack during a clone or fetch. They are not
optimized for Git clients sending a small topic branch via "git push".
In some cases (see [1]), using reachability bitmaps during "git push"
can cause significant performance regressions.
Add PushCommand#setUseBitmaps(boolean) to allow users to tell "git push"
not to use bitmaps.
[1]: https://lore.kernel.org/git/87zhoz8b9o.fsf@evledraar.gmail.com/
Change-Id: I7fb7d26084ec63ddfa7249cf58abb85929b30e56
Signed-off-by: kylezhao <kylezhao@tencent.com>
Fix and complete the implementation of calling the pre-push hook.
Add the missing error stream redirect, and add the missing setters
in Transport and in PushCommand. In Transport, delay setting up a
PrePushHook such that it happens only on a push. Previously, the
hook was set up also for fetches.
Bug: 549246
Change-Id: I64a576dfc6b139426f05d9ea6654027ab805734e
Signed-off-by: Thomas Wolf <twolf@apache.org>
We need the full size of the objects to populate the object-size index
of a pack. This size is not always the one encoded in the object header
in the pack (e.g. for deltas).
Populate the full size of PackedObjectInfos in the PackParser, which is
invoked when receiving a pack e.g. in a push.
Change-Id: I102c20901aefb5e85047e2e526c0d733f82ff74b
CloneCommand, when setNoCheckout(true) was set, did not set HEAD.
With C git, "git clone --no-checkout" does.
Change-Id: Ief3df7e904ce90829a6345a6c3e9ee6a68486ab0
Signed-off-by: Thomas Wolf <twolf@apache.org>
PatchApplier now routes updates through the index. This has two
results:
* we can now execute patches in-memory.
* the JGit apply command will now always update the
index to match the working tree.
Change-Id: Id60a88232f05d0367787d038d2518c670cdb543f
Co-authored-by: Han-Wen Nienhuys <hanwen@google.com>
Co-authored-by: Nitzan Gur-Furman <nitzan@google.com>
* stable-6.3:
Prepare 6.3.1-SNAPSHOT builds
JGit v6.3.0.202209071007-r
JGit v6.3.0.2022009070944-r
[merge] Fix merge conflicts with symlinks
Update DEPENDENCIES for 6.3.0
Update tycho to 2.7.5
Revert "Adds FilteredRevCommit that can overwrites its parents in the DAG."
Revert "Option to pass start RevCommit to be blamed on to the BlameGenerator."
Prepare 6.3.0-SNAPSHOT builds
JGit v6.3.0.202208161710-m3
Change-Id: Ia9430fb516dca795e25064a190704b70689af364
Add a test case to make sure that the bundle writer writes objects in
GC_REST packs as well.
Signed-off-by: Yunjie Li <yunjieli@google.com>
Change-Id: Iba4d88c573aa1cda4505afbe2b83581a09a343df
Previous code would do a content merge on symlinks, and write the merge
result to the working tree as a file. C git doesn't do this; it leaves
a symlink in the working tree unchanged, or in a delete-modify conflict
it would check out "theirs".
Moreover, previous code would write the merge result to the link target,
not to the link. This would overwrite an existing link target, or fail
if the link pointed to a directory.
In link/file conflicts or file/link conflicts, C git always puts the
file into the working tree.
Change conflict handling accordingly. Add tests for all the conflict
cases.
Bug: 580347
Change-Id: I3cffcb4bcf8e336a85186031fff23f0c4b6ee19d
Signed-off-by: Thomas Wolf <twolf@apache.org>
DAG."
This reverts commit 6297491e8a.
This is done as a quick fix for the failure of egit tests caused by the
introduction of FilteredRevCommit.
Bug: 580690
Change-Id: Ia6b651dd11b0a4b02d5e52247eb4bf13adf94e27
BlameGenerator."
This reverts commit 5747bba48b.
This is done as a quick fix for the failure of egit tests caused by the
introduction of FilteredRevCommit.
Bug: 580690
Change-Id: Ia0178bc2de4fc825a81207bbd7979bf3a386c955
An invalid path in the manifest (e.g. '.') is reported by DirCache in a
runtime exception. In server context this becomes a 500 instead of a user error.
Wrap the runtime invalid path exception into a checked ManifestErrorException that
caller can handle.
Change-Id: I61a2104922765506ae232334891057bb06141d97
On executing a copy, mark the destination as updated.
On executing a rename, mark both source and destination as updated.
Change-Id: Ied5b9b0e5a14eac59a06cdd0961e25e143f50ff0
This can allow passing a FilteredRevCommit which is the filtered list of
commit graph making it easier for Blame to work on. This can
significantly improve blame performance since blame can skip expensive
RevWalk.
Change-Id: Ie127cb710d004079e9f53a5802130afdb49a7de1
Previous code loaded the WorkingTreeOptions afresh for every single
file being checked out. This checked the git config (all three files,
repo, user and system config) for having been modified every time.
These checks can be costly, for instance on Windows, or if one of the
three config files is not on a local disk, or on an otherwise slow
storage.
Improve this by loading the options and thus checking the git config
only once before the checkout.
Bug: 579715
Change-Id: I21cd5a808f9d90b5ca2d022f91f0eeb8ca26091c
Signed-off-by: Thomas Wolf <twolf@apache.org>
With core.symlinks=false, symlinks are checked out as plain files.
When such a file is re-added to the index, and the index already
contains a symlink there, add the file as a symlink. Previous code
changed the index entry to a regular file.
Bug: 580412
Change-Id: I5497bedc3da89c8b10120b8077c56bc5b67cb791
Signed-off-by: Thomas Wolf <twolf@apache.org>
String.startsWith() is not a valid test for file path prefixes:
directory "a" is _not_ a prefix of a file "ab", only of "a/b".
Add a proper Paths.isEqualOrPrefix() method and use it in CleanCommand.
Bug: 580478
Change-Id: I6863e6ba94a8ffba6561835cc57044a0945d2770
Signed-off-by: Thomas Wolf <twolf@apache.org>
This can allow passing a FilteredRevCommit which is the filtered list of
commit graph making it easier for Blame to work on. This can
significantly improve blame performance since blame can skip expensive
RevWalk.
Change-Id: I5dab25301d6aef7df6a0bc25a4c553c730199272
This makes RevCommit extensible to allow having different structure of
child-parent relationship. This change is a pre-requsite for having a
FilteredRevCommit that overrides parents from the RevCommit. That then
provides a cheaper way to walk over a subset of RevCommits instead of
an expensive way that applies filters while walking over selected
commits. Useful with Blame which works on a single file and that can be
made performant, if we know all the commits needed by the Blame
algorithm. So Blame algorithm can avoid walking over finding what
commits to blame on.
This change makes parents field on RevCommit private and exposes it
thrrough overrideable methods such as getParents, getParent at index,
getParentCount and setParents. All other files other than RevCommit are
updating the usages of accessing RevCommits parents.
Change-Id: I2d13b001c599cc4ebc92d1ab6e07b07acb3b7fe5
The FilteredRenameDetectorTest was added in [1], but bazel build rules
were not updated.
[1] https://git.eclipse.org/r/c/jgit/jgit/+/194200
Change-Id: I7fc713e19b4768176eb84e9768137431b33a805e
This adds support for shallow cloning. The CloneCommand and the
FetchCommand now have the new methods setDepth, setShallowSince and
addShallowExclude to tell the server that the client doesn't want to
download the complete history.
Bug: 475615
Change-Id: Ic80fb6efb5474543ae59be590ebe385bec21cc0d
The NameConflictTreeWalk class is used in 3-way merge for iterating over
entries in 3 different commits. The class provides information about a
current entry and a state of the entry in commits (e.g entry is file,
entry is directory, entry is missing). In rare cases, the tree walker
can mix information about entries with different name.
The problem appears, because git uses unusual sorting order for
files. Example (this is a simplified real-life example):
Commit 1:
* gradle.properties - file
* gradle - directory (with nested files)
* gradle/file - file in gradle directory
Commit 2:
* gradle.properties - file
* no entry with the name gradle
Commit 3:
* gradle.properties - file
* gradle - file
Here the names are ordered like this:
"gradle" file <"gradle.properties" file < "gradle/file" file.
NameConflictTreeWalk iterator already have code for processing
git sorting order, however in the example above the code doesn't
work correctly. Before the fix, NameConflictTreeWalk returns:
#next()
"gradle - directory" | "gradle.properties" | "gradle - file" - which is
wrong. The expected result is
#next()
"gradle - directory | MISSED_FILE | "gradle - file"
#next()
"gradle.properties"|"gradle.properties"|"gradle.properties"
Ensure that the "matches" field of tree iterators (which contains the
current path) is kept in sync in the case above.
Change-Id: Ief5aa06d80b358f4080043c8694aa0fd7c60045b
Signed-off-by: Dmitrii Filippov <dmfilippov@google.com>
The RootIgnoreNode in a WorkingTreeIterator must _not_ add the rules
from .git/info/exclude or from the file designated by git config
core.excludesFile to the list of rules read from the root .gitignore.
These really must be separate nodes in a hierarchy, otherwise the
precedence rules from [1] are violated and the outcome is not the
same as in C git.
[1] https://git-scm.com/docs/gitignore
Bug: 580381
Change-Id: I57802ba7bbbe4f183504c882b6c77a78cc3a9b99
Signed-off-by: Thomas Wolf <twolf@apache.org>
* stable-6.2:
UploadPack: don't prematurely terminate timer in case of error
Do not create reflog for remote tracking branches during clone
UploadPack: do not check reachability of visible SHA1s
Add missing package import javax.management to org.eclipse.jgit
Change-Id: Ia2280b848cea7bdfa5bc48356484e040d159845d
* stable-6.1:
UploadPack: don't prematurely terminate timer in case of error
Do not create reflog for remote tracking branches during clone
UploadPack: do not check reachability of visible SHA1s
Add missing package import javax.management to org.eclipse.jgit
Change-Id: Ib3ff6a038e1b92a931a7f2624c19bbefd8cb95fd
* stable-6.0:
UploadPack: don't prematurely terminate timer in case of error
Do not create reflog for remote tracking branches during clone
UploadPack: do not check reachability of visible SHA1s
Add missing package import javax.management to org.eclipse.jgit
Change-Id: I08734ee2c8f3296d908da6a29d53ed87c4b48eb2
* stable-5.13:
UploadPack: don't prematurely terminate timer in case of error
Do not create reflog for remote tracking branches during clone
UploadPack: do not check reachability of visible SHA1s
Add missing package import javax.management to org.eclipse.jgit
Change-Id: I6db0a4d74399fde892eeec62efd2946f97547a5d
Adjusted BlameGenerator to filter rename detection with the blame path.
This reduces the running time of the blame computation significantly,
for repositories with massive commits involving renames.
The filtered rename detection is made (internally) available with:
org.eclipse.jgit.internal.diff.FilteredRenameDetector
Bug: 578900
Change-Id: I6580004e81102d685081b8180da1587a35073d36
Signed-off-by: Simeon Andreev <simeon.danailov.andreev@gmail.com>
When using JGit on a non-bare repository, the CloneCommand
it previously created local reflogs for all branches including remote
tracking ones, causing the generation of a potentially large
number of files on the local filesystem.
The creation of the remote-tracking branches (refs/remotes/*) during
clone is not an issue for the local filesystem because all of them are
stored in a single packed-refs file. However, the creation of a large
number of ref logs on a local filesystem IS an issue because it
may not be tuned or initialised in term of inodes to contain a very
large number of files.
When a user (or a CI system) performs the CloneCommand against
a potentially large repository (e.g., millions of branches), it is
interested in working or validating a single branch or tag and is
unlikely to work with all the remote-tracking branches.
The eager creation of a reflogs for all the remote-tracking branches is
not just a performance issue but may also compromise the ability to
use JGit for cloning a large repository.
The behaviour implemented in this change is also consistent with the
optimisation done in the C code-base [1].
We differentiate between clone and fetch commands using --branch
<initialBranch> option, that is only available in clone command,
and is set as HEAD per default.
[1] 58f233ce1e
Bug: 579805
Change-Id: I58d0d36a8a4ce42e0f59b8bf063747c4b81bd859
Signed-off-by: Luca Milanesio <luca.milanesio@gmail.com>
* stable-6.2:
Prepare 5.13.2-SNAPSHOT builds
JGit v5.13.1.202206130422-r
AmazonS3: Add support for AWS API signature version 4
Change-Id: I13514595f3a710b0d6e3c2de62f6508b950d22f9
* stable-6.1:
Prepare 5.13.2-SNAPSHOT builds
JGit v5.13.1.202206130422-r
AmazonS3: Add support for AWS API signature version 4
Change-Id: Id4965aacd4e2ea1e8575a2c1bd4845729db6049a