As we currently describe in t/README, it can happen that:
Some tests run "git" (or "test-tool" etc.) without properly checking
the exit code, or git will invoke itself and fail to ferry the
abort() exit code to the original caller.
Therefore, GIT_TEST_SANITIZE_LEAK_LOG=true is needed to be set to
capture all memory leaks triggered by our tests.
It seems unnecessary to force users to remember this option, as
forgetting it could lead to missed memory leaks.
We could solve the problem by making it "true" by default, but that
might suggest we think "false" makes sense, which isn't the case.
Therefore, the best approach is to remove the option entirely while
maintaining the capability to detect memory leaks in blind spots of our
tests.
Signed-off-by: Rubén Justo <rjusto@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
In 5ca0c455f1 (ci: fix Python dependency on Ubuntu 24.04, 2024-05-06),
we made the use of Python 2 conditional on whether or not the CI job
runs Ubuntu 20.04. There was a brown-paper-bag-style bug though, where
the condition forgot to invoke the `test` builtin. The result of it is
that the check always fails, and thus all of our jobs run with Python 3
by accident.
Fix this.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Unbreak CI jobs so that we do not attempt to use Python 2 that has
been removed from the platform.
* ps/ci-python-2-deprecation:
ci: fix Python dependency on Ubuntu 24.04
* fixes/2.45.1/2.44:
Revert "fsck: warn about symlink pointing inside a gitdir"
Revert "Add a helper function to compare file contents"
clone: drop the protections where hooks aren't run
tests: verify that `clone -c core.hooksPath=/dev/null` works again
Revert "core.hooksPath: add some protection while cloning"
init: use the correct path of the templates directory again
hook: plug a new memory leak
ci: stop installing "gcc-13" for osx-gcc
ci: avoid bare "gcc" for osx-gcc job
ci: drop mention of BREW_INSTALL_PACKAGES variable
send-email: avoid creating more than one Term::ReadLine object
send-email: drop FakeTerm hack
* fixes/2.45.1/2.43:
Revert "fsck: warn about symlink pointing inside a gitdir"
Revert "Add a helper function to compare file contents"
clone: drop the protections where hooks aren't run
tests: verify that `clone -c core.hooksPath=/dev/null` works again
Revert "core.hooksPath: add some protection while cloning"
init: use the correct path of the templates directory again
hook: plug a new memory leak
ci: stop installing "gcc-13" for osx-gcc
ci: avoid bare "gcc" for osx-gcc job
ci: drop mention of BREW_INSTALL_PACKAGES variable
send-email: avoid creating more than one Term::ReadLine object
send-email: drop FakeTerm hack
* fixes/2.45.1/2.42:
Revert "fsck: warn about symlink pointing inside a gitdir"
Revert "Add a helper function to compare file contents"
clone: drop the protections where hooks aren't run
tests: verify that `clone -c core.hooksPath=/dev/null` works again
Revert "core.hooksPath: add some protection while cloning"
init: use the correct path of the templates directory again
hook: plug a new memory leak
ci: stop installing "gcc-13" for osx-gcc
ci: avoid bare "gcc" for osx-gcc job
ci: drop mention of BREW_INSTALL_PACKAGES variable
send-email: avoid creating more than one Term::ReadLine object
send-email: drop FakeTerm hack
* fixes/2.45.1/2.41:
Revert "fsck: warn about symlink pointing inside a gitdir"
Revert "Add a helper function to compare file contents"
clone: drop the protections where hooks aren't run
tests: verify that `clone -c core.hooksPath=/dev/null` works again
Revert "core.hooksPath: add some protection while cloning"
init: use the correct path of the templates directory again
hook: plug a new memory leak
ci: stop installing "gcc-13" for osx-gcc
ci: avoid bare "gcc" for osx-gcc job
ci: drop mention of BREW_INSTALL_PACKAGES variable
send-email: avoid creating more than one Term::ReadLine object
send-email: drop FakeTerm hack
* fixes/2.45.1/2.40:
Revert "fsck: warn about symlink pointing inside a gitdir"
Revert "Add a helper function to compare file contents"
clone: drop the protections where hooks aren't run
tests: verify that `clone -c core.hooksPath=/dev/null` works again
Revert "core.hooksPath: add some protection while cloning"
init: use the correct path of the templates directory again
hook: plug a new memory leak
ci: stop installing "gcc-13" for osx-gcc
ci: avoid bare "gcc" for osx-gcc job
ci: drop mention of BREW_INSTALL_PACKAGES variable
send-email: avoid creating more than one Term::ReadLine object
send-email: drop FakeTerm hack
Revert overly aggressive "layered defence" that went into 2.45.1
and friends, which broke "git-lfs", "git-annex", and other use
cases, so that we can rebuild necessary counterparts in the open.
* jc/fix-2.45.1-and-friends-for-2.39:
Revert "fsck: warn about symlink pointing inside a gitdir"
Revert "Add a helper function to compare file contents"
clone: drop the protections where hooks aren't run
tests: verify that `clone -c core.hooksPath=/dev/null` works again
Revert "core.hooksPath: add some protection while cloning"
init: use the correct path of the templates directory again
hook: plug a new memory leak
ci: stop installing "gcc-13" for osx-gcc
ci: avoid bare "gcc" for osx-gcc job
ci: drop mention of BREW_INSTALL_PACKAGES variable
send-email: avoid creating more than one Term::ReadLine object
send-email: drop FakeTerm hack
The last user of this variable went away in 4a6e4b9602 (CI: remove
Travis CI support, 2021-11-23), so it's doing nothing except making it
more confusing to find out which packages _are_ installed.
[jc: cherry-picked from v2.45.0-1-g9d4453e8d6]
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Newer versions of Ubuntu have dropped Python 2 starting with Ubuntu
23.04. By default though, our CI setups will try to use that Python
version on all Ubuntu-based jobs except for the "linux-gcc" one.
We didn't notice this issue due to two reasons:
- The "ubuntu:latest" tag always points to the latest LTS release.
Until a few weeks ago this was Ubuntu 22.04, which still had Python
2.
- Our Docker-based CI jobs had their own script to install
dependencies until 9cdeb34b96 (ci: merge scripts which install
dependencies, 2024-04-12), where we didn't even try to install
Python at all for many of them.
Since the CI refactorings have originally been implemented, Ubuntu
24.04 was released, and it being an LTS versions means that the "latest"
tag now points to that Python-2-less version. Consequently, those jobs
that use "ubuntu:latest" broke.
Address this by using Python 2 on Ubuntu 20.04, only, whereas we use
Python 3 on all other Ubuntu jobs. Eventually, we should think about
dropping support for Python 2 completely.
Reported-by: Justin Tobler <jltobler@gmail.com>
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
* maint-2.41: (38 commits)
Git 2.41.1
Git 2.40.2
Git 2.39.4
fsck: warn about symlink pointing inside a gitdir
core.hooksPath: add some protection while cloning
init.templateDir: consider this config setting protected
clone: prevent hooks from running during a clone
Add a helper function to compare file contents
init: refactor the template directory discovery into its own function
find_hook(): refactor the `STRIP_EXTENSION` logic
clone: when symbolic links collide with directories, keep the latter
entry: report more colliding paths
t5510: verify that D/F confusion cannot lead to an RCE
submodule: require the submodule path to contain directories only
clone_submodule: avoid using `access()` on directories
submodules: submodule paths must not contain symlinks
clone: prevent clashing git dirs when cloning submodule in parallel
t7423: add tests for symlinked submodule directories
has_dir_name(): do not get confused by characters < '/'
docs: document security issues around untrusted .git dirs
...
* maint-2.40: (39 commits)
Git 2.40.2
Git 2.39.4
fsck: warn about symlink pointing inside a gitdir
core.hooksPath: add some protection while cloning
init.templateDir: consider this config setting protected
clone: prevent hooks from running during a clone
Add a helper function to compare file contents
init: refactor the template directory discovery into its own function
find_hook(): refactor the `STRIP_EXTENSION` logic
clone: when symbolic links collide with directories, keep the latter
entry: report more colliding paths
t5510: verify that D/F confusion cannot lead to an RCE
submodule: require the submodule path to contain directories only
clone_submodule: avoid using `access()` on directories
submodules: submodule paths must not contain symlinks
clone: prevent clashing git dirs when cloning submodule in parallel
t7423: add tests for symlinked submodule directories
has_dir_name(): do not get confused by characters < '/'
docs: document security issues around untrusted .git dirs
upload-pack: disable lazy-fetching by default
...
* maint-2.39: (38 commits)
Git 2.39.4
fsck: warn about symlink pointing inside a gitdir
core.hooksPath: add some protection while cloning
init.templateDir: consider this config setting protected
clone: prevent hooks from running during a clone
Add a helper function to compare file contents
init: refactor the template directory discovery into its own function
find_hook(): refactor the `STRIP_EXTENSION` logic
clone: when symbolic links collide with directories, keep the latter
entry: report more colliding paths
t5510: verify that D/F confusion cannot lead to an RCE
submodule: require the submodule path to contain directories only
clone_submodule: avoid using `access()` on directories
submodules: submodule paths must not contain symlinks
clone: prevent clashing git dirs when cloning submodule in parallel
t7423: add tests for symlinked submodule directories
has_dir_name(): do not get confused by characters < '/'
docs: document security issues around untrusted .git dirs
upload-pack: disable lazy-fetching by default
fetch/clone: detect dubious ownership of local repositories
...
Every once in a while, the `git-p4` tests flake for reasons outside of
our control. It typically fails with "Connection refused" e.g. here:
https://github.com/git/git/actions/runs/5969707156/job/16196057724
[...]
+ git p4 clone --dest=/home/runner/work/git/git/t/trash directory.t9807-git-p4-submit/git //depot
Initialized empty Git repository in /home/runner/work/git/git/t/trash directory.t9807-git-p4-submit/git/.git/
Perforce client error:
Connect to server failed; check $P4PORT.
TCP connect to localhost:9807 failed.
connect: 127.0.0.1:9807: Connection refused
failure accessing depot: could not run p4
Importing from //depot into /home/runner/work/git/git/t/trash directory.t9807-git-p4-submit/git
[...]
This happens in other jobs, too, but in the `linux-asan`/`linux-ubsan`
jobs it hurts the most because those jobs often take an _awfully_ long
time to run, therefore re-running a failed `linux-asan`/`linux-ubsan`
jobs is _very_ costly.
The purpose of the `linux-asan`/`linux-ubsan` jobs is to exercise the C
code of Git, anyway, and any part of Git's source code that the `git-p4`
tests run and that would benefit from the attention of ASAN/UBSAN are
run better in other tests anyway, as debugging C code run via Python
scripts can get a bit hairy.
In fact, it is not even just `git-p4` that is the problem (even if it
flakes often enough to be problematic in the CI builds), but really the
part about Python scripts. So let's just skip any Python parts of the
tests from being run in that job.
For good measure, also skip the Subversion tests because debugging C
code run via Perl scripts is as much fun as debugging C code run via
Python scripts. And it will reduce the time this very expensive job
takes, which is a big benefit.
Backported to `maint-2.39` as another step to get that branch's CI
builds back to a healthy state.
Backported-from: 6ba913629f (ci(linux-asan-ubsan): let's save some time, 2023-08-29)
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
In April, GitHub announced that the `macos-13` pool is available:
https://github.blog/changelog/2023-04-24-github-actions-macos-13-is-now-available/.
It is only a matter of time until the `macos-12` pool is going away,
therefore we should switch now, without pressure of a looming deadline.
Since the `macos-13` runners no longer include Python2, we also drop
specifically testing with Python2 and switch uniformly to Python3, see
https://github.com/actions/runner-images/blob/HEAD/images/macos/macos-13-Readme.md
for details about the software available on the `macos-13` pool's
runners.
Also, on macOS 13, Homebrew seems to install a `gcc@9` package that no
longer comes with a regular `unistd.h` (there seems only to be a
`ssp/unistd.h`), and hence builds would fail with:
In file included from base85.c:1:
git-compat-util.h:223:10: fatal error: unistd.h: No such file or directory
223 | #include <unistd.h>
| ^~~~~~~~~~
compilation terminated.
The reason why we install GCC v9.x explicitly is historical, and back in
the days it was because it was the _newest_ version available via
Homebrew: 176441bfb5 (ci: build Git with GCC 9 in the 'osx-gcc' build
job, 2019-11-27).
To reinstate the spirit of that commit _and_ to fix that build failure,
let's switch to the now-newest GCC version: v13.x.
Backported-from: 682a868f67 (ci: upgrade to using macos-13, 2023-11-03)
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Add a new fuzz target that exercises the parsing of git configs.
The existing git_config_from_mem function is a perfect entry point
for fuzzing as it exercises the same code paths as the rest of the
config parsing functions and offers an easily fuzzable interface.
Config parsing is a useful thing to fuzz because it operates on user
controlled data and is a central component of many git operations.
Signed-off-by: Brian C Tracy <brian.tracy33@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The way CI testing used "prove" could lead to running the test
suite twice needlessly, which has been corrected.
* js/ci-discard-prove-state:
ci: avoid running the test suite _twice_
ci: add support for GitLab CI
ci: install test dependencies for linux-musl
ci: squelch warnings when testing with unusable Git repo
ci: unify setup of some environment variables
ci: split out logic to set up failed test artifacts
ci: group installation of Docker dependencies
ci: make grouping setup more generic
ci: reorder definitions for grouping functions
Add CI jobs for both GitHub Workflows and GitLab CI to run Git with the
new reftable backend.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
To prevent bitrot, we would like to regularly exercise the fuzz tests in
order to make sure they still link & run properly. We already compile
the fuzz test objects as part of the default `make` target, but we do
not link the executables due to the fuzz tests needing specific
compilers and compiler features. This has lead to frequent build
breakages for the fuzz tests.
To remedy this, we can add a CI step to actually link the fuzz
executables, and run them (with finite input rather than the default
infinite random input mode) to verify that they execute properly.
Since the main use of the fuzz tests is via OSS-Fuzz [1], and OSS-Fuzz
only runs tests on Linux [2], we only set up a CI test for the fuzzers
on Linux.
[1] https://github.com/google/oss-fuzz
[2] https://google.github.io/oss-fuzz/further-reading/fuzzer-environment/
Signed-off-by: Josh Steadmon <steadmon@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Add a job to GitLab CI which runs tests on macOS, which matches the
equivalent "osx-clang" job that we have for GitHub Workflows. One
significant difference though is that this new job runs on Apple M1
machines and thus uses the "arm64" architecture. As GCC does not yet
support this comparatively new architecture we cannot easily include an
equivalent for the "osx-gcc" job that exists in GitHub Workflows.
Note that one test marked as `test_must_fail` is surprisingly passing:
t7815-grep-binary.sh (Wstat: 0 Tests: 22 Failed: 0)
TODO passed: 12
This seems to boil down to an unexpected difference in how regcomp(3P)
works when matching NUL bytes. Cross-checking with the respective GitHub
job shows that this is not an issue unique to the GitLab CI job as it
passes in the same way there.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When setting up Perforce on macOS we put both `p4` and `p4d` into
"$HOME/bin". On GitHub CI this directory is indeed contained in the PATH
environment variable and thus there is no need for additional setup than
to put the binaries there. But GitLab CI does not do this, and thus our
Perforce-based tests would be skipped there even though we download the
binaries.
Refactor the setup code to become more robust by downloading binaries
into a separate directory which we then manually append to our PATH.
This matches what we do on Linux-based jobs.
Note that it may seem like we already did append "$HOME/bin" to PATH
because we're actually removing the lines that adapt PATH. But we only
ever adapted the PATH variable in "ci/install-dependencies.sh", and
didn't adapt it when running "ci/run-build-and-test.sh". Consequently,
the required binaries wouldn't be found during the test run unless the
CI platform already had the "$HOME/bin" in PATH right from the start.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The TEST_OUTPUT_DIRECTORY environment variable can be used to instruct
the test suite to write test data and test results into a different
location than into "t/". The "ci/print-test-failures.sh" script does not
know to handle this environment variable though, which means that it
will search for test results in the wrong location if it was set.
Update the script to handle TEST_OUTPUT_DIRECTORY so that we can start
to set it in our CI.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Our GitHub Workflows definitions have a static analysis job that
runs the following tasks:
- Coccinelle to check for suggested refactorings.
- `make hdr-check` to check for missing includes or forward
declarations in our header files.
- `make check-pot` to check our translations for issues.
- `./ci/check-directional-formatting.bash` to check whether our
sources contain any Unicode directional formatting code points.
Add an equivalent job to our GitLab CI definitions.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The way CI testing used "prove" could lead to running the test
suite twice needlessly, which has been corrected.
* js/ci-discard-prove-state:
ci: avoid running the test suite _twice_
Add support for GitLab CI.
* ps/ci-gitlab:
ci: add support for GitLab CI
ci: install test dependencies for linux-musl
ci: squelch warnings when testing with unusable Git repo
ci: unify setup of some environment variables
ci: split out logic to set up failed test artifacts
ci: group installation of Docker dependencies
ci: make grouping setup more generic
ci: reorder definitions for grouping functions
Process to add some form of low-level unit tests has started.
* js/doc-unit-tests:
ci: run unit tests in CI
unit tests: add TAP unit test framework
unit tests: add a project plan document
This is a late amendment of 4a6e4b9602 (CI: remove Travis CI support,
2021-11-23), whereby the `.prove` file (being written by the `prove`
command that is used to run the test suite) is no longer retained
between CI builds: This feature was only ever used in the Travis CI
builds, we tried for a while to do the same in Azure Pipelines CI runs
(but I gave up on it after a while), and we never used that feature in
GitHub Actions (nor does the new GitLab CI code use it).
Retaining the Prove cache has been fragile from the start, even though
the idea seemed good at the time, the idea being that the `.prove` file
caches information about previous `prove` runs (`save`) and uses them
(`slow`) to run the tests in the order from longer-running to shorter
ones, making optimal use of the parallelism implied by `--jobs=<N>`.
However, using a Prove cache can cause some surprising behavior: When
the `prove` caches information about a test script it has run,
subsequent `prove` runs (with `--state=slow`) will run the same test
script again even if said script is not specified on the `prove`
command-line!
So far, this bug did not matter. Right until d8f416bbb8 (ci: run unit
tests in CI, 2023-11-09) did it not matter.
But starting with that commit, we invoke `prove` _twice_ in CI, once to
run the regular test suite of regression test scripts, and once to run
the unit tests. Due to the bug, the second invocation re-runs all of the
tests that were already run as part of the first invocation. This not
only wastes build minutes, it also frequently causes the `osx-*` jobs to
fail because they already take a long time and now are likely to run
into a timeout.
The worst part about it is that there is actually no benefit to keep
running with `--state=slow,save`, ever since we decided no longer to
try to reuse the Prove cache between CI runs.
So let's just drop that Prove option and live happily ever after.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Run unit tests in both Cirrus and GitHub CI. For sharded CI instances
(currently just Windows on GitHub), run only on the first shard. This is
OK while we have only a single unit test executable, but we may wish to
distribute tests more evenly when we add new unit tests in the future.
We may also want to add more status output in our unit test framework,
so that we can do similar post-processing as in
ci/lib.sh:handle_failed_tests().
Signed-off-by: Josh Steadmon <steadmon@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
We already support Azure Pipelines and GitHub Workflows in the Git
project, but until now we do not have support for GitLab CI. While it is
arguably not in the interest of the Git project to maintain a ton of
different CI platforms, GitLab has recently ramped up its efforts and
tries to contribute to the Git project more regularly.
Part of a problem we hit at GitLab rather frequently is that our own,
custom CI setup we have is so different to the setup that the Git
project has. More esoteric jobs like "linux-TEST-vars" that also set a
couple of environment variables do not exist in GitLab's custom CI
setup, and maintaining them to keep up with what Git does feels like
wasted time. The result is that we regularly send patch series upstream
that fail to compile or pass tests in GitHub Workflows. We would thus
like to integrate the GitLab CI configuration into the Git project to
help us send better patch series upstream and thus reduce overhead for
the maintainer. Results of these pipeline runs will be made available
(at least) in GitLab's mirror of the Git project at [1].
This commit introduces the integration into our regular CI scripts so
that most of the setup continues to be shared across all of the CI
solutions. Note that as the builds on GitLab CI run as unprivileged
user, we need to pull in both sudo and shadow packages to our Alpine
based job to set this up.
[1]: https://gitlab.com/gitlab-org/git
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The linux-musl CI job executes tests on Alpine Linux, which is based on
musl libc instead of glibc. We're missing some test dependencies though,
which causes us to skip a subset of tests.
Install these test dependencies to increase our test coverage on this
platform. There are still some missing test dependecies, but these do
not have a corresponding package in the Alpine repositories:
- p4 and p4d, both parts of the Perforce version control system.
- cvsps, which generates patch sets for CVS.
- Subversion and the SVN::Core Perl library, the latter of which is
not available in the Alpine repositories. While the tool itself is
available, all Subversion-related tests are skipped without the
SVN::Core Perl library anyway.
The Apache2-based tests require a bit more care though. For one, the
module path is different on Alpine Linux, which requires us to add it to
the list of known module paths to detect it. But second, the WebDAV
module on Alpine Linux is broken because it does not bundle the default
database backend [1]. We thus need to skip the WebDAV-based tests on
Alpine Linux for now.
[1]: https://gitlab.alpinelinux.org/alpine/aports/-/issues/13112
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Our CI jobs that run on Docker also use mostly the same architecture to
build and test Git via the "ci/run-build-and-tests.sh" script. These
scripts also provide some functionality to massage the Git repository
we're supposedly operating in.
In our Docker-based infrastructure we may not even have a Git repository
available though, which leads to warnings when those functions execute.
Make the helpers exit gracefully in case either there is no Git in our
PATH, or when not running in a Git repository.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Both GitHub Actions and Azure Pipelines set up the environment variables
GIT_TEST_OPTS, GIT_PROVE_OPTS and MAKEFLAGS. And while most values are
actually the same, the setup is completely duplicate. With the upcoming
support for GitLab CI this duplication would only extend even further.
Unify the setup of those environment variables so that only the uncommon
parts are separated. While at it, we also perform some additional small
improvements:
- We now always pass `--state=failed,slow,save` via GIT_PROVE_OPTS.
It doesn't hurt on platforms where we don't persist the state, so
this further reduces boilerplate.
- When running on Windows systems we set `--no-chain-lint` and
`--no-bin-wrappers`. Interestingly though, we did so _after_
already having exported the respective environment variables.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
We have some logic in place to create a directory with the output from
failed tests, which will then subsequently be uploaded as CI artifacts.
We're about to add support for GitLab CI, which will want to reuse the
logic.
Split the logic into a separate function so that it is reusable.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The output of CI jobs tends to be quite long-winded and hard to digest.
To help with this, many CI systems provide the ability to group output
into collapsible sections, and we're also doing this in some of our
scripts.
One notable omission is the script to install Docker dependencies.
Address it to bring more structure to the output for Docker-based jobs.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Make the grouping setup more generic by always calling `begin_group ()`
and `end_group ()` regardless of whether we have stubbed those functions
or not. This ensures we can more readily add support for additional CI
platforms.
Furthermore, the `group ()` function is made generic so that it is the
same for both GitHub Actions and for other platforms. There is a
semantic conflict here though: GitHub Actions used to call `set +x` in
`group ()` whereas the non-GitHub case unconditionally uses `set -x`.
The latter would get overriden if we kept the `set +x` in the generic
version of `group ()`. To resolve this conflict, we simply drop the `set
+x` in the generic variant of this function. As `begin_group ()` calls
`set -x` anyway this is not much of a change though, as the only
commands that aren't printed anymore now are the ones between the
beginning of `group ()` and the end of `begin_group ()`.
Last, this commit changes `end_group ()` to also accept a parameter that
indicates _which_ group should end. This will be required by a later
commit that introduces support for GitLab CI.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
We define a set of grouping functions that are used to group together
output in our CI, where these groups then end up as collapsible sections
in the respective pipeline platform. The way these functions are defined
is not easily extensible though as we have an up front check for the CI
_not_ being GitHub Actions, where we define the non-stub logic in the
else branch.
Reorder the conditional branches such that we explicitly handle GitHub
Actions.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
In April, GitHub announced that the `macos-13` pool is available:
https://github.blog/changelog/2023-04-24-github-actions-macos-13-is-now-available/.
It is only a matter of time until the `macos-12` pool is going away,
therefore we should switch now, without pressure of a looming deadline.
Since the `macos-13` runners no longer include Python2, we also drop
specifically testing with Python2 and switch uniformly to Python3, see
https://github.com/actions/runner-images/blob/HEAD/images/macos/macos-13-Readme.md
for details about the software available on the `macos-13` pool's
runners.
Also, on macOS 13, Homebrew seems to install a `gcc@9` package that no
longer comes with a regular `unistd.h` (there seems only to be a
`ssp/unistd.h`), and hence builds would fail with:
In file included from base85.c:1:
git-compat-util.h:223:10: fatal error: unistd.h: No such file or directory
223 | #include <unistd.h>
| ^~~~~~~~~~
compilation terminated.
The reason why we install GCC v9.x explicitly is historical, and back in
the days it was because it was the _newest_ version available via
Homebrew: 176441bfb5 (ci: build Git with GCC 9 in the 'osx-gcc' build
job, 2019-11-27).
To reinstate the spirit of that commit _and_ to fix that build failure,
let's switch to the now-newest GCC version: v13.x.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Now that we have the CI_BRANCHES mechanism, there is no need for anybody
to use the ci/config/allow-ref mechanism. In the long run, we can
hopefully remove it and the whole "config" job, as it consumes CPU and
adds to the end-to-end latency of the whole workflow. But we don't want
to do that immediately, as people need time to migrate until the
CI_BRANCHES change has made it into the workflow file of every branch.
So let's issue a warning, which will appear in the "annotations" section
below the workflow result in GitHub's web interface. And let's remove
the sample allow-refs script, as we don't want to encourage anybody to
use it.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When we added config to skip CI for certain branches in e76eec3554 (ci:
allow per-branch config for GitHub Actions, 2020-05-07), there wasn't
any way to avoid spinning up a VM just to check the config. From the
developer's perspective this isn't too bad, as the "skipped" branches
complete successfully after running the config job (the workflow result
is "success" instead of "skipped", but that is a minor lie).
But we are still wasting time and GitHub's CPU to spin up a VM just to
check the result of a short shell script. At the time there wasn't any
way to avoid this. But they've since introduced repo-level variables
that should let us do the same thing:
https://github.blog/2023-01-10-introducing-required-workflows-and-configuration-variables-to-github-actions/#configuration-variables
This is more efficient, and as a bonus is probably less confusing to
configure (the existing system requires sticking your config on a magic
ref).
See the included docs for how to configure it.
The code itself is pretty simple: it checks the variable and skips the
config job if appropriate (and everything else depends on the config job
already). There are two slight inaccuracies here:
- we don't insist on branches, so this likewise applies to tag names
or other refs. I think in practice this is OK, and keeping the code
(and docs) short is more important than trying to be more exact. We
are targeting developers of git.git and their limited workflows.
- the match is done as a substring (so if you want to run CI for
"foobar", then branch "foo" will accidentally match). Again, this
should be OK in practice, as anybody who uses this is likely to only
specify a handful of well-known names. If we want to be more exact,
we can have the code check for adjoining spaces. Or even move to a
more general CI_CONFIG variable formatted as JSON. I went with this
scheme for the sake of simplicity.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Every once in a while, the `git-p4` tests flake for reasons outside of
our control. It typically fails with "Connection refused" e.g. here:
https://github.com/git/git/actions/runs/5969707156/job/16196057724
[...]
+ git p4 clone --dest=/home/runner/work/git/git/t/trash directory.t9807-git-p4-submit/git //depot
Initialized empty Git repository in /home/runner/work/git/git/t/trash directory.t9807-git-p4-submit/git/.git/
Perforce client error:
Connect to server failed; check $P4PORT.
TCP connect to localhost:9807 failed.
connect: 127.0.0.1:9807: Connection refused
failure accessing depot: could not run p4
Importing from //depot into /home/runner/work/git/git/t/trash directory.t9807-git-p4-submit/git
[...]
This happens in other jobs, too, but in the `linux-asan-ubsan` job it
hurts the most because that job often takes over a full hour to run,
therefore re-running a failed `linux-asan-ubsan` job is _very_ costly.
The purpose of the `linux-asan-ubsan` job is to exercise the C code of
Git, anyway, and any part of Git's source code that the `git-p4` tests
run and that would benefit from the attention of ASAN/UBSAN are run
better in other tests anyway, as debugging C code run via Python scripts
can get a bit hairy.
In fact, it is not even just `git-p4` that is the problem (even if it
flakes often enough to be problematic in the CI builds), but really the
part about Python scripts. So let's just skip any Python parts of the
tests from being run in that job.
For good measure, also skip the Subversion tests because debugging C
code run via Perl scripts is as much fun as debugging C code run via
Python scripts. And it will reduce the time this very expensive job
takes, which is a big benefit.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The object traversal using reachability bitmap done by
"pack-object" has been tweaked to take advantage of the fact that
using "boundary" commits as representative of all the uninteresting
ones can save quite a lot of object enumeration.
* tb/pack-bitmap-traversal-with-boundary:
pack-bitmap.c: use commit boundary during bitmap traversal
pack-bitmap.c: extract `fill_in_bitmap()`
object: add object_array initializer helper function
Clang's sanitizer implementation seems to work better than GCC's.
* jk/ci-use-clang-for-sanitizer-jobs:
ci: drop linux-clang job
ci: run ASan/UBSan in a single job
ci: use clang for ASan/UBSan checks
When we started running sanitizers in CI via 1c0962c0c4 (ci: add address
and undefined sanitizer tasks, 2022-10-20), we ran them as two separate
CI jobs, since as that commit notes, the combination "seems to take
forever".
And indeed, it does with gcc. However, since the previous commit
switched to using clang, the situation is different, and we can save
some CPU by using a single job for both. Comparing before/after CI runs,
this saved about 14 minutes (the single combined job took 54m, versus
44m plus 24m for ASan and UBSan jobs, respectively). That's wall-clock
and not CPU, but since our jobs are mostly CPU-bound, the two should be
closely proportional.
This does increase the end-to-end time of a CI run, though, since before
this patch the two jobs could run in parallel, and the sanitizer job is
our longest single job. It also means that we won't get a separate
result for "this passed with UBSan but not with ASan" or vice versa).
But as 1c0962c0c4 noted, that is not a very useful signal in practice.
Below are some more detailed timings of gcc vs clang that I measured by
running the test suite on my local workstation. Each measurement counts
only the time to run the test suite with each compiler (not the compile
time itself). We'll focus on the wall-clock times for simplicity, though
the CPU times follow roughly similar trends.
Here's a run with CC=gcc as a baseline:
real 1m12.931s
user 9m30.566s
sys 8m9.538s
Running with SANITIZE=address increases the time by a factor of ~4.7x:
real 5m40.352s
user 49m37.044s
sys 36m42.950s
Running with SANITIZE=undefined increases the time by a factor of ~1.7x:
real 2m5.956s
user 12m42.847s
sys 19m27.067s
So let's call that 6.4 time units to run them separately (where a unit
is the time it takes to run the test suite with no sanitizers). As a
simplistic model, we might imagine that running them together would take
5.4 units (we save 1 unit because we are no longer running the test
suite twice, but just paying the sanitizer overhead on top of a single
run).
But that's not what happens. Running with SANITIZE=address,undefined
results in a factor of 9.3x:
real 11m9.817s
user 77m31.284s
sys 96m40.454s
So not only did we not get faster when doing them together, we actually
spent 1.5x as much CPU as doing them separately! And while those
wall-clock numbers might not look too terrible, keep in mind that this
is on an unloaded 8-core machine. In the CI environment, wall-clock
times will be much closer to CPU times. So not only are we wasting CPU,
but we risk hitting timeouts.
Now let's try the same thing with clang. Here's our no-sanitizer
baseline run, which is almost identical to the gcc one (which is quite
convenient, because we can keep using the same "time units" to get an
apples-to-apples comparison):
real 1m11.844s
user 9m28.313s
sys 8m8.240s
And now again with SANITIZE=address, we get a 5x factor (so slightly
worse than gcc's 4.7x, though I wouldn't read too much into it; there is
a fair bit of run-to-run noise):
real 6m7.662s
user 49m24.330s
sys 44m13.846s
And with SANITIZE=undefined, we are at 1.5x, slightly outperforming gcc
(though again, that's probably mostly noise):
real 1m50.028s
user 11m0.973s
sys 16m42.731s
So running them separately, our total cost is 6.5x. But if we combine
them in a single run (SANITIZE=address,undefined), we get:
real 6m51.804s
user 52m32.049s
sys 51m46.711s
which is a factor of 5.7x. That's along the lines we'd hoped for!
Running them together saves us almost a whole time unit. And that's not
counting any time spent outside the test suite itself (starting the job,
setting up the environment, compiling) that we're no longer duplicating
by having two jobs.
So clang behaves like we'd hope: the overhead to run the sanitizers is
additive as you add more sanitizers. Whereas gcc's numbers seem very
close to multiplicative, almost as if the sanitizers were enforcing
their overheads on each other (though that is purely a guess on what is
going on; ultimately what matters to us is the amount of time it takes).
And that roughly matches the CI improvement I saw. A "time unit" there
is more like 12 minutes, and the observed time savings was 14 minutes
(with the extra presumably coming from avoiding duplicated setup, etc).
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When reachability bitmap coverage exists in a repository, Git will use a
different (and hopefully faster) traversal to compute revision walks.
Consider a set of positive and negative tips (which we'll refer to with
their standard bitmap parlance by "wants", and "haves"). In order to
figure out what objects exist between the tips, the existing traversal
in `prepare_bitmap_walk()` does something like:
1. Consider if we can even compute the set of objects with bitmaps,
and fall back to the usual traversal if we cannot. For example,
pathspec limiting traversals can't be computed using bitmaps (since
they don't know which objects are at which paths). The same is true
of certain kinds of non-trivial object filters.
2. If we can compute the traversal with bitmaps, partition the
(dereferenced) tips into two object lists, "haves", and "wants",
based on whether or not the objects have the UNINTERESTING flag,
respectively.
3. Fall back to the ordinary object traversal if either (a) there are
more than zero haves, none of which are in the bitmapped pack or
MIDX, or (b) there are no wants.
4. Construct a reachability bitmap for the "haves" side by walking
from the revision tips down to any existing bitmaps, OR-ing in any
bitmaps as they are found.
5. Then do the same for the "wants" side, stopping at any objects that
appear in the "haves" bitmap.
6. Filter the results if any object filter (that can be easily
computed with bitmaps alone) was given, and then return back to the
caller.
When there is good bitmap coverage relative to the traversal tips, this
walk is often significantly faster than an ordinary object traversal
because it can visit far fewer objects.
But in certain cases, it can be significantly *slower* than the usual
object traversal. Why? Because we need to compute complete bitmaps on
either side of the walk. If either one (or both) of the sides require
walking many (or all!) objects before they get to an existing bitmap,
the extra bitmap machinery is mostly or all overhead.
One of the benefits, however, is that even if the walk is slower, bitmap
traversals are guaranteed to provide an *exact* answer. Unlike the
traditional object traversal algorithm, which can over-count the results
by not opening trees for older commits, the bitmap walk builds an exact
reachability bitmap for either side, meaning the results are never
over-counted.
But producing non-exact results is OK for our traversal here (both in
the bitmap case and not), as long as the results are over-counted, not
under.
Relaxing the bitmap traversal to allow it to produce over-counted
results gives us the opportunity to make some significant improvements.
Instead of the above, the new algorithm only has to walk from the
*boundary* down to the nearest bitmap, instead of from each of the
UNINTERESTING tips.
The boundary-based approach still has degenerate cases, but we'll show
in a moment that it is often a significant improvement.
The new algorithm works as follows:
1. Build a (partial) bitmap of the haves side by first OR-ing any
bitmap(s) that already exist for UNINTERESTING commits between the
haves and the boundary.
2. For each commit along the boundary, add it as a fill-in traversal
tip (where the traversal terminates once an existing bitmap is
found), and perform fill-in traversal.
3. Build up a complete bitmap of the wants side as usual, stopping any
time we intersect the (partial) haves side.
4. Return the results.
And is more-or-less equivalent to using the *old* algorithm with this
invocation:
$ git rev-list --objects --use-bitmap-index $WANTS --not \
$(git rev-list --objects --boundary $WANTS --not $HAVES |
perl -lne 'print $1 if /^-(.*)/')
The new result performs significantly better in many cases, particularly
when the distance from the boundary commit(s) to an existing bitmap is
shorter than the distance from (all of) the have tips to the nearest
bitmapped commit.
Note that when using the old bitmap traversal algorithm, the results can
be *slower* than without bitmaps! Under the new algorithm, the result is
computed faster with bitmaps than without (at the cost of over-counting
the true number of objects in a similar fashion as the non-bitmap
traversal):
# (Computing the number of tagged objects not on any branches
# without bitmaps).
$ time git rev-list --count --objects --tags --not --branches
20
real 0m1.388s
user 0m1.092s
sys 0m0.296s
# (Computing the same query using the old bitmap traversal).
$ time git rev-list --count --objects --tags --not --branches --use-bitmap-index
19
real 0m22.709s
user 0m21.628s
sys 0m1.076s
# (this commit)
$ time git.compile rev-list --count --objects --tags --not --branches --use-bitmap-index
19
real 0m1.518s
user 0m1.234s
sys 0m0.284s
The new algorithm is still slower than not using bitmaps at all, but it
is nearly a 15-fold improvement over the existing traversal.
In a more realistic setting (using my local copy of git.git), I can
observe a similar (if more modest) speed-up:
$ argv="--count --objects --branches --not --tags"
hyperfine \
-n 'no bitmaps' "git.compile rev-list $argv" \
-n 'existing traversal' "git.compile rev-list --use-bitmap-index $argv" \
-n 'boundary traversal' "git.compile -c pack.useBitmapBoundaryTraversal=true rev-list --use-bitmap-index $argv"
Benchmark 1: no bitmaps
Time (mean ± σ): 124.6 ms ± 2.1 ms [User: 103.7 ms, System: 20.8 ms]
Range (min … max): 122.6 ms … 133.1 ms 22 runs
Benchmark 2: existing traversal
Time (mean ± σ): 368.6 ms ± 3.0 ms [User: 325.3 ms, System: 43.1 ms]
Range (min … max): 365.1 ms … 374.8 ms 10 runs
Benchmark 3: boundary traversal
Time (mean ± σ): 167.6 ms ± 0.9 ms [User: 139.5 ms, System: 27.9 ms]
Range (min … max): 166.1 ms … 169.2 ms 17 runs
Summary
'no bitmaps' ran
1.34 ± 0.02 times faster than 'boundary traversal'
2.96 ± 0.05 times faster than 'existing traversal'
Here, the new algorithm is also still slower than not using bitmaps, but
represents a more than 2-fold improvement over the existing traversal in
a more modest example.
Since this algorithm was originally written (nearly a year and a half
ago, at the time of writing), the bitmap lookup table shipped, making
the new algorithm's result more competitive. A few other future
directions for improving bitmap traversal times beyond not using bitmaps
at all:
- Decrease the cost to decompress and OR together many bitmaps
together (particularly when enumerating the uninteresting side of
the walk). Here we could explore more efficient bitmap storage
techniques, like Roaring+Run and/or use SIMD instructions to speed
up ORing them together.
- Store pseudo-merge bitmaps, which could allow us to OR together
fewer "summary" bitmaps (which would also help with the above).
Helped-by: Jeff King <peff@peff.net>
Helped-by: Derrick Stolee <derrickstolee@github.com>
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Back in e8c58f894b (t: support GIT_TEST_WRITE_REV_INDEX, 2021-01-25), we
added a test knob to conditionally enable writing a ".rev" file when
indexing a pack. At the time, this was used to ensure that the test
suite worked even when ".rev" files were written, which served as a
stress-test for the on-disk reverse index implementation.
Now that reading from on-disk ".rev" files is enabled by default, the
test knob `GIT_TEST_WRITE_REV_INDEX` no longer has any meaning.
We could get rid of the option entirely, but there would be no
convenient way to test Git when ".rev" files *aren't* in place.
Instead of getting rid of the option, invert its meaning to instead
disable writing ".rev" files, thereby running the test suite in a mode
where the reverse index is generated from scratch.
This ensures that, when GIT_TEST_NO_WRITE_REV_INDEX is set to some
spelling of "true", we are still running and exercising Git's behavior
when forced to generate reverse indexes from scratch. Do so by setting
it in the linux-TEST-vars CI run to ensure that we are maintaining good
coverage of this now-legacy code.
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Acked-by: Derrick Stolee <derrickstolee@github.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
