A non-bare clone command with null remote produces a
NullPointerException when trying to produce a refspec to fetch against.
In a bare repository, a null remote name is accepted by mistake,
producing a configuration with items like 'remote.url' instead of
'remote.<remote>.url'. This was never meant to work.
Instead, let's make setRemote(null) undo any previous setRemote calls
and re-set the remote name to DEFAULT_REMOTE, imitating C git clone's
--no-origin option.
While we're here, add some tests for setRemote working normally.
Change-Id: I76f502da5e677df501d3ef387e7f61f42a7ca238
Signed-off-by: Jonathan Nieder <jrn@google.com>
- Consistently return structured data, such as actual ReceiveCommands,
which is more useful for callers that are doing things other than
verifying the signature, e.g. recording the set of commands.
- Store the certificate version field, as this is required to be part
of the signed payload.
- Add a toText() method to recreate the actual payload for signature
verification. This requires keeping track of the un-chomped command
strings from the original protocol stream.
- Separate the parser from the certificate itself, so the actual
PushCertificate object can be immutable. Make a fair attempt at deep
immutability, but this is not possible with the current mutable
ReceiveCommand structure.
- Use more detailed error messages that don't involve NON-NLS strings.
- Document null return values more thoroughly. Instead of having the
undocumented behavior of throwing NPE from certain methods if they
are not first guarded by enabled(), eliminate enabled() and return
null from those methods.
- Add tests for parsing a push cert from a section of pkt-line stream
using a real live stream captured with Wireshark (which, it should
be noted, uncovered several simply incorrect statements in C git's
Documentation/technical/pack-protocol.txt).
This is a slightly breaking API change to classes that were
technically public and technically released in 4.0. However, it is
highly unlikely that people were actually depending on public
behavior, since there were no public methods to create
PushCertificates with anything other than null field values, or a
PushCertificateParser that did anything other than infinite loop or
throw exceptions when reading.
Change-Id: I5382193347a8eb1811032d9b32af9651871372d0
C git's receive-pack.c strips trailing newlines in command lists when
present[1], although send-pack.c does not send them, at least in the
case of command lists[2]. Change JGit to match this behavior.
Add tests.
This also fixes parsing of commands in the push cert, which, unlike
commands sent in the non-push case, always have trailing newlines.
[1] 7974889a05/builtin/receive-pack.c (L1380)
where packet_read_line chomps newlines:
7974889a05/pkt-line.c (L202)
[2] 7974889a05/send-pack.c (L470)
Change-Id: I4bca6342a7482a53c9a5815a94b3c181a479d04b
The new submodule layout where GITDIR of a submodule is located at
<parent-repo-GITDIR>/modules/<submodule-path> was only used during
clone. Teach SubmoduleAddCommand to use the new layout.
Bug: 469666
Change-Id: Ie97dc0607b71499560444616f362bccee9cce515
Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
Most relative-URL tests for SubmoduleInitCommand carry out the following
steps:
1. add a submodule at path "sub" to the index
2. set remote.origin.url in .git/config
3. configure .gitmodules, possibly using relative URLs, and see what
happens
resolveWorkingDirectoryRelativeUrl() is meant to test the fallback when
remote.origin.url is not set, to match C git which treats the URL as
relative to the cwd in that case. To do so, in step (2) it sets
remote.origin.url to null.
However, Config.setString when taking a null value does not actually
unset that value from the configuration --- it sets it to the empty
string. This means we are testing a behavior that C git never
supported. Use Config.unset instead.
Change-Id: I7af29fbbd333a2598843d62c320093c48b2ad972
Signed-off-by: Jonathan Nieder <jrn@google.com>
Move ObjectCountCallback and WriteAbortedException to package
org.eclipse.jgit.transport, so that they'll become public API.
Change-Id: I95e3cfaa49f3f7371e794d5c253cf6981f87cae0
Signed-off-by: Yuxuan 'fishy' Wang <fishywang@google.com>
This reverts commit 96eb3ee397, which
broke Gerrit tests that set a config value to 'null', serialize the
result, deserialize, and expect 'null' from Config.getString[1].
The intent of that commit was to make it possible to distinguish between
an absent and an empty config value, which we'll have to do with a new
method.
Revert the behavior change. Keep the tests from 428cb23f2de8, since
they test the behavior more precisely than the old tests did.
[1] https://gerrit-review.googlesource.com/68452
Change-Id: Ie8042f380ea0e34e3203e1991aa0feb2e6e44641
Signed-off-by: Jonathan Nieder <jrn@google.com>
Added callback in PackWriter and BundleWriter for the caller to get the
count of objects to write, and a chance to abort the write operation.
Change-Id: I1baeedcc6946b1093652de4a707fe597a577e526
Signed-off-by: Yuxuan 'fishy' Wang <fishywang@google.com>
exactRef(ref1, ref2, ref3) requests multiple specific refs in a single
lookup, which may be faster in some backends than looking them up one by
one.
firstExactRef generalizes getRef by finding the first existing ref from
the list of refs named. Its main purpose is for the default
implementation of getRef (finding the first existing ref in a search
path). Hopefully it can be useful for other operations that look for
refs in a search path (e.g., git log --notes=<name>), too.
Change-Id: I5c6fcf1d3920f6968b8b97f3d4c3a267258c4b86
Signed-off-by: Jonathan Nieder <jrn@google.com>
Unlike getRef(name), the new exactRef method does not walk the search
path. This should produce a less confusing result than getRef when the
exact ref name is known: it will not try to resolve refs/foo/bar to
refs/heads/refs/foo/bar even when refs/foo/bar does not exist.
It can be faster than both getRefs(ALL).get(name) and getRef(name)
because it only needs to examine a single ref.
A follow-up change will introduce a findRef synonym to getRef and
deprecate getRef to make the choice a caller is making more obvious
(exactRef or findRef, with the same semantics as getRefs(ALL).get and
getRefs(ALL).findRef).
Change-Id: If1bd09bcfc9919e7976a4d77f13184ea58dcda52
Signed-off-by: Jonathan Nieder <jrn@google.com>
This is a perfectly valid construction according to C git:
$ echo -en '[a]\nx =' > foo.config
$ git config -f foo.config a.x; echo $?
0
Change-Id: Icfcf8304adb43c79e2b8b998f8d651b2a94f6acb
The C git API and command line tools distinguish between a key having
the empty string as a value and no key being present in the config
file:
$ echo -e '[a]\nx =' > foo.config
$ git config -f foo.config a.x; echo $?
0
$ git config -f foo.config a.y; echo $?
1
Make JGit make the same distinction. This is in line with the current
Javadoc of getString, which claims to return "a String value from the
config, null if not found". It is more reasonable to interpret "x ="
in the above example as "found" rather than "missing".
We need to maintain the special handling of a key name with no "="
resolving to a boolean true, but "=" with an empty string is still not
a valid boolean.
Change-Id: If0dbb7470c524259de0b167148db87f81be2d04a
When CommitCommand.setOnly(path) is used a temporary index has to be
created containing only modifications for the specified pathes. The
process to fill this temporary index was broken because filling the
index was stopped when we reached the first untracked file. Instead the
process to fill the temporary index should continue until we processed
all pathes.
Bug: 451465
Change-Id: I22af50d70bd3b81e7c056358724956122b0d158d
Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
Otherwise MergeCommandTest using Sets doesn't compile in Eclipse since
adbcbc79 moved Sets from the "tst" to the "src" folder.
Change-Id: I661b987513365a8af0b568ec95b0898e5758f59f
Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
ObjectReader release method was replaced by close method but
WindowCursor was still implementing release method.
To prevent the same mistake again, make ObjectReader close method
abstract to force sub classes to implement it.
Change-Id: I50d0d1d19a26e306fd0dba77b246a95a44fd6584
Signed-off-by: Hugo Arès <hugo.ares@ericsson.com>
Commit d3348e introduced few errors in Eclipse.
This commit cleans up the new API:
- fixes API error in RepoCommand after moving IncludedFileReader type
- fixes unused imports in RepoCommand & RepoCommandTest
- fix javadoc errors in ManifestParser & RepoProject
- makes three (implicitly final) fields in ManifestParser final.
Change-Id: I4185f451d97039d155391f62815bb9d3afe39fa6
Signed-off-by: Andrey Loskutov <loskutov@gmx.de>
The repo xml manifest parser used in RepoCommand could also be useful for
others, so refactor to make it public.
Also this breaks backward compatibility slightly.
Change-Id: I5001bd2fe77541109fe32dbe2597a065e6ad585e
Signed-off-by: Yuxuan 'fishy' Wang <fishywang@google.com>
Test that
- the default ObjectFilter is ALL
- ObjectFilter affects nextObject() and not next()
- omitting a tree implies omitting its subtrees
- a blob or tree reached by another path is still returned
- ObjectFilter can be mixed with RevFilter
Change-Id: I144a53fe677070fff8c3ddf8cba07a848773bc1b
Signed-off-by: Jonathan Nieder <jrn@google.com>
A later patch will make use of this class in a org.eclipse.jgit.lib
test.
Change-Id: I2b268e6a5dbf12174201f45259f9f007686708d2
Signed-off-by: Jonathan Nieder <jrn@google.com>
Change-Id: I700540eec06efb24eeb09bfcb40420820c32d156
Signed-off-by: Sebastian Schuberth <sschuberth@gmail.com>
Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
The LRU chain management code was broken leading to situations where
the chain was incomplete. This prevented the cache from removing
items when it exceeded its memory target, causing a leak.
One case was repeated hit on the head of the chain. moveToHead(e)
was invoked linking the head back to itself in a cycle orphaning
the rest of the table.
Add some unit tests to cover this and a few other paths.
Change-Id: Ib27486eaa1b1d2bf1c745a56d0a5832bfb029322
Avoid always calling `sh -c umask` on startup, instead deferring
the invocation until the first time a working tree file needs to
use the execute bit. This allows servers using bare repos to avoid
a costly fork+exec for a value that is never used.
Store the umask as an int instead of two Boolean. This is slightly
smaller memory (one int vs. two references) and makes it easier for
an application to force setting the umask to a value that overrides
whatever the shell told JGit.
Simplify the code to bail by returning early when canExecute is
false, which is the common case for working tree files.
Change-Id: Ie713647615bc5bdf5d71b731a6748c28ea21c900
Despite being the primary author of RevWalk and ObjectWalk I still
fail to remember to setRetainBody(false) in application code using
an ObjectWalk to examine the graph.
Document the default for RevWalk is setRetainBody(true), where the
application usually wants the commit bodies to display or inspect.
Change the default for ObjectWalk to setRetainBody(false), as nearly
all callers want only the graph shape and do not need the larger text
inside a commit body. This allows some code in JGit to be simplified.
Change-Id: I367e42209e805bd5e1f41b4072aeb2fa98ec9d99
A larger than expected number of real-world repositories found on
the Internet contain invalid author, committer and tagger lines
in their history. Many of these seem to be caused by users misusing
the user.name and user.email fields, e.g.:
[user]
name = Au Thor <author@example.com>
email = author@example.com
that some version of Git (or a reimplementation thereof) copied
directly into the object header. These headers are not valid and
are rejected by a strict fsck, making it impossible to transfer
the repository with JGit/EGit.
Another form is an invalid committer line with double negative for
the time zone, e.g.
committer Au Thor <a@b> 1288373970 --700
The real world is messy. :(
Allow callers and users to weaken the fsck settings to accept these
sorts of breakages if they really want to work on a repo that has
broken history. Most routines will still function fine, however
commit timestamp sorting in RevWalk may become confused by a corrupt
committer line and sort commits out of order. This is mostly fine if
the corrupted chain is shorter than the slop window.
Change-Id: I6d529542c765c131de590f4f7ef8e7c1c8cb9db9
The Iff2de881 tried to fix missing tree ..." but introduced severe
performance degradation (>10x in some cases) when acting as server
(git push) and as client (replication). IOW cure is worse than the
disease.
This reverts commit c4797fe986.
Change-Id: I4e6056eb352d51277867f857a0cab380eca153ac
Signed-off-by: David Ostrovsky <david@ostrovsky.org>
Bug: 390833
Change-Id: I29f7b79b241929877c93ac485c677487a91bb77b
Signed-off-by: André de Oliveira <andre.oliveira@liferay.com>
Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
If a non interactive rebase is launched, stopping after a conflict
should set the repository state to RepositoryState.REBASING_MERGE
instead of RepositoryState.REBASING_INTERACTIVE.
Bug: 452623
Change-Id: Ie885aab6d71dabd158a718af0d14fff643c9b850
Also-by: Arthur Daussy <arthur.daussy@obeo.fr>
Signed-off-by: Laurent Delaigue <laurent.delaigue@obeo.fr>
Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
When RecursiveMerger found that there are multiple base-commits for the
commits to be merged it tries to temporarily merge the base commits. But
if these base commits have no common predecessor there was a bug in JGit
leading to a NPE. This commit fixes this by enforcing that an empty tree
is used as base when merging two unrelated base commits.
This logic was already there when merging two commits which have no
common predecessor (ThreeWayMerger.mergeBase()). But the code which was
computing a new temporary base commit in case of criss-cross merges
didn't take care to pick an empty tree when no common predecessor can be
found.
Bug: 462671
Change-Id: Ibd96302f5f81383f36d3b1e3edcbf5822147b1a4
CherryPickCommand only works on a non-bare repository, as it must
modify the working tree and index in case of a merge conflict. In
tests, being able to recover from a merge conflict is less important,
as the caller should be able to control the full contents of files in
advance of the cherry-pick.
Change-Id: Ic332e44df1308b9336e884666b08c1f6db64513d
As we moved minimum Java version to 7 we don't need a separate bundle
and feature for JGit features depending on Java 7 anymore.
Change-Id: Ib5da61b0886ddbdea65298f1e8c6d65c9879ced1
Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
This reverts commit 6bc48cdc62.
Until git v1.7.10.2~29^2~1 (builtin/merge.c: reduce parents early,
2012-04-17), C git merge would make merge commits with duplicate parents
when asked to with a series of commands like the following:
git checkout origin/master
git merge --no-ff origin/master
Nowadays "git merge" removes redundant parents more aggressively
(whenever one parent is an ancestor of another and not just when
duplicates exist) but merges with duplicate parents are still permitted
and can be created with git fast-import or git commit-tree and history
viewers need to be able to cope with them.
CommitBuilder is an interface analagous to commit-tree, so it should
allow duplicate parents. (That said, an option to automatically remove
redundant parents would be useful.)
Reported-by: Dave Borowitz <dborowitz@google.com>
Change-Id: Ia682238397eb1de8541802210fa875fdd50f62f0
Signed-off-by: Jonathan Nieder <jrn@google.com>
Callers may wish to use TemporaryBuffer as an essentially unbounded
buffer by passing Integer.MAX_VALUE as the size. (This makes it
behave like ByteArrayOutputStream, only without requiring contiguous
memory.) Unfortunately, it was always allocating an array in the
backing block pointer list to hold enough blocks to MAX_VALUE--all
262,016 of them. It wasn't allocating the blocks themselves, but this
array was still extremely wasteful, using about 2MiB of memory on a
64-bit system.
Tweak the interface to specify an estimated size, and only allocate
the block pointer list enough entries to hold that size. It's an
ArrayList, so if that estimate was wrong, it'll grow. We assume the
cost of finding enough contiguous memory to grow that array is
acceptable.
While we're in there, fix an off-by-one error: due to integer division
we were undercounting the number of blocks needed to store n bytes of
data as (n / SZ).
Change-Id: I794eca3ac4472bcc605b3641e177922aca92b9c0
Jonathan Nieder
Dave Borowitz
Christian Halstrick
Yuxuan 'fishy' Wang
Matthias Sohn
Hugo Arès
Andrey Loskutov
Sebastian Schuberth
Shawn Pearce
David Ostrovsky
André de Oliveira
Laurent Delaigue