tests/test_git_superproject.py::SuperprojectTestCase::test_Fetch was
failing in Python 3.6 and 3.7 due to attribute args only being
introduced in Python 3.8. Falling back on old way of accessing
the arguments.
Test: tox with Python 3.6 to 3.11
Change-Id: Iae1934a7bce8cbd6b4519e4dbc92d94e21b43435
Reviewed-on: https://gerrit-review.googlesource.com/c/git-repo/+/382818
Tested-by: Daniel Kutik <daniel.kutik@lavawerk.com>
Reviewed-by: Mike Frysinger <vapier@google.com>
Commit-Queue: Daniel Kutik <daniel.kutik@lavawerk.com>
When comparing 2 manifests, most of the time is
spent getting the relevant commit id as it relies
on _allrefs which ends up loading all git references.
However, the value from `revisionIs` (when it is valid)
could be used directly leading to a huge performance improvement
(from 180+ seconds to less than 0.01 sec which is more
than 25000 times faster for manifests with 700+ projects).
Bug: 295282548
Change-Id: I5881aa4b2326cc17bbb4ee91d23293111f76ad7e
Reviewed-on: https://gerrit-review.googlesource.com/c/git-repo/+/385834
Tested-by: Sylvain Desodt <sylvain.desodt@gmail.com>
Reviewed-by: Mike Frysinger <vapier@google.com>
Commit-Queue: Sylvain Desodt <sylvain.desodt@gmail.com>
The logic in projectsDiff performs various operations which
suggest that a set is more appropriate than a list:
- membership lookup ("in")
- removal
Also, sorting can be performed on the the remaining elements at the
end (which will usually involve a much smaller number of elements).
(The performance gain is invisible in comparison to the time being
spent performing git operations).
Cosmetic chance:
- the definition of 'fromProj' is moved to be used in more places
- the values in diff["added"] are added with a single call to extend
Change-Id: I5ed22ba73b50650ca2d3a49a1ae81f02be3b3055
Reviewed-on: https://gerrit-review.googlesource.com/c/git-repo/+/383434
Tested-by: Sylvain Desodt <sylvain.desodt@gmail.com>
Reviewed-by: Mike Frysinger <vapier@google.com>
Commit-Queue: Sylvain Desodt <sylvain.desodt@gmail.com>
If a KeyboardInterrupt is encountered before an error is aggregated then
the context surrounding the interrupt is lost. This change aggregates
errors as soon as possible for the sync command
Bug: b/293344017
Change-Id: Iac14f9d59723cc9dedbb960f14fdc1fa5b348ea3
Reviewed-on: https://gerrit-review.googlesource.com/c/git-repo/+/384974
Tested-by: Jason Chang <jasonnc@google.com>
Commit-Queue: Jason Chang <jasonnc@google.com>
Reviewed-by: Gavin Mak <gavinmak@google.com>
Black will only check .py files when given a dir and --check, so list
our few standalone programs explicitly. This causes the repo launcher
to be reformatted since it was missed in the previous mass reformat.
Bug: b/267675342
Change-Id: Ic90a7f5d84fc02e9fccb05945310fd067e2ed764
Reviewed-on: https://gerrit-review.googlesource.com/c/git-repo/+/385034
Tested-by: Mike Frysinger <vapier@google.com>
Reviewed-by: Gavin Mak <gavinmak@google.com>
Commit-Queue: Mike Frysinger <vapier@google.com>
Per discussion in go/repo-error-update updated aggregated and exit
errors for sync command.
Aggregated errors are errors that result in eventual command failure.
Exit errors are errors that result in immediate command failure.
Also updated main.py to log aggregated and exit errors to git sessions
log
Bug: b/293344017
Change-Id: I77a21f14da32fe2e68c16841feb22de72e86a251
Reviewed-on: https://gerrit-review.googlesource.com/c/git-repo/+/379614
Reviewed-by: Aravind Vasudevan <aravindvasudev@google.com>
Tested-by: Jason Chang <jasonnc@google.com>
Commit-Queue: Jason Chang <jasonnc@google.com>
In order to better analyze and track repo errors, repo command failures
need to be tied to specific errors in repo source code.
Additionally a new GitCommandError was added to differentiate between
general git related errors to failed git commands. Git commands that opt
into verification will raise a GitCommandError if the command failed.
The first step in this process is a general error refactoring
Bug: b/293344017
Change-Id: I46944b1825ce892757c8dd3f7e2fab7e460760c0
Reviewed-on: https://gerrit-review.googlesource.com/c/git-repo/+/380994
Commit-Queue: Jason Chang <jasonnc@google.com>
Reviewed-by: Aravind Vasudevan <aravindvasudev@google.com>
Tested-by: Jason Chang <jasonnc@google.com>
Reviewed-by: Joanna Wang <jojwang@google.com>
Fixed a couple of bugs in ExitEvent logging:
- log exitcode 130 on KeyboardInterrupt
- log exitcode 1 on unhandled Exception
- log errorevent with specific reason for exit
Before this CL an exitcode of 0 would be logged, and it would be
difficult to determine the cause of non-zero exit codes
Bug: b/287105597
Change-Id: I2d34f180581f9fbd77a1c78c966ebed065223af6
Reviewed-on: https://gerrit-review.googlesource.com/c/git-repo/+/377834
Tested-by: Jason Chang <jasonnc@google.com>
Reviewed-by: Josip Sokcevic <sokcevic@google.com>
A something.xml that gets included by two different
files, that both remove and add same shared project
to two different locations, would not work
prior to this change.
Reason is that remove killed all name keys, even
though reuse of same repo in different locations
is allowed.
Solve by adding optional attrib path to
<remove-project name="foo" path="only_this_path" />
and tweak remove-project.
Behaves as before without path, and deletes
more selectively when remove path is supplied.
As secondary feature, a project can now also be removed
by only using path, assuming a matching project name
can be found.
Change-Id: I502d9f949f5d858ddc1503846b170473f76dc8e2
Reviewed-on: https://gerrit-review.googlesource.com/c/git-repo/+/375694
Tested-by: Fredrik de Groot <fredrik.de.groot@aptiv.com>
Reviewed-by: Mike Frysinger <vapier@google.com>
An investigation go/git-repo-shallow shows a number of problems
when doing a shallow git fetch/clone. This change introduces an
environment variable REPO_ALLOW_SHALLOW. When this environment variable
is set to 1 during a repo init or repo sync all shallow git fetch
commands are replaced with partial fetch commands. Any shallow
repository needing update is unshallowed. This behavior continues until
a subsequent repo sync command is run with REPO_ALLOW_SHALLOW set to 1.
Bug: b/274340522
Change-Id: I1c3188270629359e52449788897d9d4988ebf280
Reviewed-on: https://gerrit-review.googlesource.com/c/git-repo/+/374754
Reviewed-by: Josip Sokcevic <sokcevic@google.com>
Tested-by: Jason Chang <jasonnc@google.com>
If the original fetch attempt did not want tags, we should continue to
honor that when doing a retry fetch with depth set to None. This seems
to match the intent of the retry based on the inline comment and results
in a significant performance improvement when the original fetch-by-sha1
fails due to the server not allowing requests for unadvertised objects.
Change-Id: Ia26bb31ea9aecc4ba2d3e87fc0c5412472cd98c4
Reviewed-on: https://gerrit-review.googlesource.com/c/git-repo/+/374918
Reviewed-by: Mike Frysinger <vapier@google.com>
Reviewed-by: Kaushik Lingarkar <kaushik.lingarkar@linaro.org>
Tested-by: Kaushik Lingarkar <kaushik.lingarkar@linaro.org>
Attribute groups can now be added to manifest include, thus
all projects in an included manifest file can easily modify
default branch without modifying all projects in that manifest file.
For example,
the main manifest.xml has an include node contain revision attribute,
```
<include name="include.xml" revision="r1" />
```
and the include.xml has some projects,
```
<project path="project1_path" name="project1_name" revision="r2" />
<project path="project2_path" name="project2_name" />
```
With this change, the final manifest will have revision="r1" for project2.
```
<project name="project1_name" path="project1_path" revision="r2" />
<project name="project2_name" path="project2_path" revision="r1" />
```
Test: added unit tests to cover the inheritance
Change-Id: I4b8547a7198610ec3a3c6aeb2136e0c0f3557df0
Reviewed-on: https://gerrit-review.googlesource.com/c/git-repo/+/369714
Reviewed-by: Mike Frysinger <vapier@google.com>
Commit-Queue: Shuchuan Zeng <zengshuchuan@allwinnertech.com>
Tested-by: Shuchuan Zeng <zengshuchuan@allwinnertech.com>