GitLab

Part-of: <!389>

This was fixed 3 weeks ago and should no longer be needed.

Part-of: <!387>

Most of the networking issues with the runner should be fixed now.

Part-of: <!385>

As gst-build can select features to be exposed by
gstreamer-full, the test will check that feature
are present or not present according to the current
configuration.

Part-of: <!381>

This reverts commit 12c1689a.

Revert ci-fairy check-commits job for now as it seems a bit
buggy. It picks up commits that are not part of the branch
and breaks the CI.

Will investigate further and reintroduce it later.

Make sure merge requests have enabled the "allow-edit"
option in the MR so we can rebase them.

Close gstreamer-project#34

Part-of: <!304>

See https://freedesktop.pages.freedesktop.org/ci-templates/ci-fairy.html#checking-commits

In this setup it will check for any leftover fixup!/squash! commits and
that messages don't have the 'Signed-off-by:' tag.

Part-of: <!304>

explicitly enable the cairo plugin build as well

Part-of: <!377>

This reverts commit c3b56937.

This broke the build and needs further investigation

https://gitlab.freedesktop.org/gstreamer/gst-ci/-/jobs/5630095

Sometimes the git repo cached on the docker image gets corrupted on
Windows, in which case there is no way to clean things and resolve
that error. For example: https://gitlab.freedesktop.org/bilboed/gst-plugins-bad/-/jobs/5517037

```
++ cp -a C:/cerbero/.git .
cp: cannot create regular file `./.git/objects/pack/pack-cbdde9dc4dd07bff6761e6715e20d7c0abe62b9e.idx': Permission denied
cp: cannot create regular file `./.git/objects/pack/pack-cbdde9dc4dd07bff6761e6715e20d7c0abe62b9e.pack': Permission denied
```

Always delete the cached git repo and copy a pristine copy from the
image. This will also avoid accumulation of garbage in the git repo.

Part-of: <!376>

This is meant to be used for containers that are going
to be periodically rebuilt. For our usecase specifying
`fdo.upstream-repo` is enough to garbage collect all
the tags from the forks.

We also need to rebuild the images, so they don't have
include the label anymore.

Part-of: <!368>

`bootstrap --build-tools=no --toolchains=no` now do nothing, and
just repeat what the bootstrap line above does which is to only
run the system bootstraper

Part-of: <!368>

With the MergeRequest rules template, its was impossible
to manually create or trigger a pipeline for a branch/commit
that was not part of a merge request.

This tweaks the workflow:rules such that there will always be
a pipeline for each commit (which will be set to a manual job trigger).

There is the downside that we will now be triggering 2 pipelines in
case of merge requests, but they will require manual action to be
started.

Part-of: <!369>

Run test-gst-full in gstreamer-full job

Part-of: <!365>

Part-of: <!364>

Gitlab was slightly bugged in previous version, and we
had to use branch-only pipelines. Let's try this again.

Part-of: <!363>

This tests that all fallbacks works to get pango plugin built with no
system dependencies available.

Part-of: <!360>

This reverts commit 19d7ff53

Revert me after merge.

Part-of: <!361>

Part-of: <!361>

Will avoid packaging errors due to the old package file still being
around from a previous job, such as in:

https://gitlab.freedesktop.org/nirbheek/cerbero/-/jobs/4804147

Part-of: <!361>

also:

Revert "citemplate: set the cerbero msvc build to allow_failure again"

This reverts commit 3dddfacf.

Part-of: <!359>

Noticed in this job, which failed because we rebuilt on top of an
existing builddir:

https://gitlab.freedesktop.org/nirbheek/cerbero/-/jobs/4772728

Part-of: <!358>

Very weird things going on..

```
UnicodeDecodeError: 'charmap' codec can't decode byte 0x9d in position 8608: character maps to <undefined>
```

https://gitlab.freedesktop.org/Hosang/gst-plugins-bad/-/jobs/4731636

Part-of: <!356>

When quoting in rules it takes the string as literal and doesn't
expand the variable

Part-of: <!353>

This reverts commit bdbec90a.

Everything seems to be working fine.

Part-of: <!352>

This reverts commit ef981baf.

Part-of: <!352>

We will use these when running bootstrap in the actual jobs.

Part-of: <!350>

Instead of putting a hammer-like symlink to ../../gstreamer/cerbero,
do recipe-specific relocation hacks. Fixes random errors like:

```
++ ln -sf /builds/nirbheek/cerbero ../../gstreamer/cerbero
ln: ../../gstreamer/cerbero/cerbero: cannot overwrite directory
```

Part-of: <!350>

We were doing a few things suboptimally:

1. Making a symlink from the current directory to
   ../../gstreamer/cerbero so that build-tools kept working, which
   should not be necessary since cerbero relocates all binaries inside
   the cerbero homedir after extracting the cache.

2. Hard-coding the cerbero builddir when building on Windows because
   `pwd` was outputting an MSYS-style path which Python can't read

3. rsync-ing all the contents of $CERBERO_HOST_DIR, which is
   bad because we should be building on top of the artifact cache.
   The only thing we want from $CERBERO_HOST_DIR is cerbero-sources
   (the network cache), and we can setup the toolchain during
   bootstrap with the new --system and --toolchains arguments which
   were split out of --system-only.

Part-of: <!350>

Revert after merging.

Part-of: <!350>

Instead of picking the 'slowest build', look at all the deps logs and
pick the commit that is mentioned in the maximum number of deps logs.

In practice, this will either be the newest commit or the previous
commit. If it's not the newest commit, we will warn and use an older
one.

Part-of: <!349>

In case python segfaults, which might happen for things with
a C-FFI like pygobject, make it so we get a proper backtrace.

https://docs.python.org/3/using/cmdline.html#envvar-PYTHONFAULTHANDLER

Part-of: <!341>

There was a missing 'f' prefix, so the {branch} was not expanded.

Part-of: <!348>

This helps detecting issue with optional steps like cerbero deps
lookup.

Part-of: <!348>

This reverts commit 555c49f9.

The job is working fine now and takes slightly more time than the iOS
build, so it can be re-enabled.

Part-of: <!347>

Sprinkle some 'time' commands so we can measure how long things take
and we can know where to spend time optimizing first.

Part-of: <!346>

Part-of: <!345>

There seems to be some issues uploading the cache from the Windows
runner.

Part-of: <!344>

Can't ln twice: it does a cp on MSYS.

Part-of: <!344>