[libcamera-ci] [RFC PATCH v1 2/3] Separate the building and running of unit tests

Laurent Pinchart laurent.pinchart at ideasonboard.com
Mon Dec 16 21:01:41 CET 2024


On Mon, Dec 16, 2024 at 06:26:14PM +0000, Kieran Bingham wrote:
> Quoting Barnabás Pőcze (2024-12-16 17:28:46)
> > 2024. 12. 16. 12:04 keltezéssel, Laurent Pinchart írta:
> > > On Mon, Dec 16, 2024 at 10:13:54AM +0100, Barnabás Pőcze wrote:
> > >> 2024. 12. 15. 21:43 keltezéssel, Laurent Pinchart írta:
> > >>> On Sun, Dec 15, 2024 at 09:43:20PM +0200, Laurent Pinchart wrote:
> > >>>> On Sun, Dec 15, 2024 at 09:04:08PM +0200, Laurent Pinchart wrote:
> > >>>>> Hi Barnabás,
> > >>>>>
> > >>>>> Thank you for the patch.
> > >>>>>
> > >>>>> On Thu, Dec 12, 2024 at 07:16:54PM +0100, Barnabás Pőcze wrote:
> > >>>>>> The built artifacts will be reused in a later job, so split
> > >>>>>> the "test-unit" into the "build-test" and "test-unit" jobs.
> > >>>>>>
> > >>>>>> The `libevent` development package cannot be installed in the container
> > >>>>>
> > >>>>> I've write `libevent-dev` here to avoid ambiguities.
> > >>>>>
> > >>>>>> directly because it is not multiarch compatible. It is, however, installed
> > >>>>>> in the architecture specific build jobs, right before building. To ensure
> > >>>>>> that the it is available for already built executables in different jobs,
> > >>>>>
> > >>>>> "that the it is" ?
> > >>>>>
> > >>>>>> install just the libraries in the container.
> > >>>>>
> > >>>>> And name here `libevent`.
> > >>>>>
> > >>>>>>
> > >>>>>> Signed-off-by: Barnabás Pőcze <barnabas.pocze at ideasonboard.com>
> > >>>>>> ---
> > >>>>>>    .gitlab-ci/setup-container.sh |  3 +++
> > >>>>>>    gitlab-ci.yml                 | 42 +++++++++++++++++++++++------------
> > >>>>>>    2 files changed, 31 insertions(+), 14 deletions(-)
> > >>>>>>
> > >>>>>> diff --git a/.gitlab-ci/setup-container.sh b/.gitlab-ci/setup-container.sh
> > >>>>>> index d2909c7..0658368 100755
> > >>>>>> --- a/.gitlab-ci/setup-container.sh
> > >>>>>> +++ b/.gitlab-ci/setup-container.sh
> > >>>>>> @@ -103,6 +103,9 @@ case $FDO_DISTRIBUTION_VERSION in
> > >>>>>>    'bookworm')
> > >>>>>>          # libclang-rt-dev for the clang ASan runtime.
> > >>>>>>          PKGS_LIBCAMERA_RUNTIME_MULTIARCH+=( libclang-rt-dev )
> > >>>>>> +        # For cam and lc-compliance
> > >>>>>> +        # libevent-dev cannot be used here, see build-libcamera-common.sh
> > >>>>>> +        PKGS_LIBCAMERA_RUNTIME_MULTIARCH+=( libevent-2.1-7 libevent-pthreads-2.1-7 )
> > >>>>>>          ;;
> > >>>>>>    'trixie')
> > >>>>>>          # gcc 13 to expand compilation testing coverage.
> > >>>>>> diff --git a/gitlab-ci.yml b/gitlab-ci.yml
> > >>>>>> index 8bc8bc2..c7448b8 100644
> > >>>>>> --- a/gitlab-ci.yml
> > >>>>>> +++ b/gitlab-ci.yml
> > >>>>>> @@ -64,7 +64,7 @@ include:
> > >>>>>>    .libcamera-ci.debian:12:
> > >>>>>>      variables:
> > >>>>>>        FDO_DISTRIBUTION_VERSION: 'bookworm'
> > >>>>>> -    FDO_DISTRIBUTION_TAG: '2024-12-12.1'
> > >>>>>> +    FDO_DISTRIBUTION_TAG: '2024-12-12.2'
> > >>>>>>
> > >>>>>>    .libcamera-ci.debian:13:
> > >>>>>>      variables:
> > >>>>>> @@ -363,28 +363,18 @@ test-soraka:
> > >>>>>>      script:
> > >>>>>>        - submit .gitlab-ci/lava/soraka-camera-test.yml
> > >>>>>>
> > >>>>>> -# Run the unit tests in a virtual machine. Enable only the options exercised by
> > >>>>>> -# the unit tests.
> > >>>>>> -test-unit:
> > >>>>>> +# Enable only the options exercised by the unit tests.
> > >>>>>> +build-test:debug:
> > >>>>>
> > >>>>> I'd call this build-package:amd64, as we have build-package:arm64 and
> > >>>>> build-package:cros. I think it would also make sense to use the same
> > >>>>> build options for the amd64 and arm64 packages (beside possibly the
> > >>>>> selected pipeline handlers, although the 'auto' option may work for
> > >>>>> both).
> > >>>>>
> > >>>>>>      extends:
> > >>>>>>        - .fdo.distribution-image at debian
> > >>>>>>        - .libcamera-ci.debian:12
> > >>>>>>        - .libcamera-ci.scripts
> > >>>>>> -  stage: test
> > >>>>>> +  stage: build
> > >>>>>>      needs:
> > >>>>>>        - job: container-debian:12
> > >>>>>>          artifacts: false
> > >>>>>> -  tags:
> > >>>>>> -    - kvm
> > >>>>>>      script:
> > >>>>>>        - $CI_PROJECT_DIR/.gitlab-ci/build-libcamera.sh
> > >>>>>> -    - $CI_PROJECT_DIR/.gitlab-ci/test-libcamera-qemu.sh
> > >>>>>> -  artifacts:
> > >>>>>> -    name: libcamera-unit-tests-${CI_COMMIT_SHA}
> > >>>>>> -    when: always
> > >>>>>> -    expire_in: 1 week
> > >>>>>> -    paths:
> > >>>>>> -      - build/meson-logs/
> > >>>>>>      variables:
> > >>>>>>        BUILD_TYPE: debug
> > >>>>>>        MESON_OPTIONS: >-
> > >>>>>> @@ -399,6 +389,30 @@ test-unit:
> > >>>>>>          -D qcam=disabled
> > >>>>>>          -D test=true
> > >>>>>>          -D v4l2=true
> > >>>>>> +  artifacts:
> > >>>>>> +    paths:
> > >>>>>> +      - build/
> > >>>>>
> > >>>>> The whole build directory can be very large. Can't we do the same as
> > >>>>> build-package:arm64 and run package-libcamera.sh to only package what we
> > >>>>> need ? We'll need probably need an unpackage script for the test-unit
> > >>>>> job.
> > >>>>
> > >>>> But of course the unit test binaries don't get installed... Can we fix
> > >>>> that and install them ? You can specify "install_tag : 'tests'" in
> > >>>> meson.build so they won't be installed by default (an appropriate
> > >>>> install_dir is also needed). This in turn requires bumping the minimum
> > >>>> meson version from 0.63.0 to 0.64.0, which shouldn't be an issue.
> > >>>
> > >>> I've been told on IRC that the motivation for the "tests" install tag in
> > >>> meson is https://gitlab.gnome.org/GNOME/gnome-desktop-testing. I don't
> > >>> think we should switch to a separate runner for unit tests (the pain is
> > >>> not worth the gain at this point in my opinion), but it could be useful
> > >>> to tag lc-compliance with install_tag = 'tests'.
> > >>>
> > >>>> And now that I've said this, I realize we wouldn't be able to run "meson
> > >>>> test" to run the tests :-/ I'm not sure there's an appropriate solution
> > >>>> for this. If not, given the size of the build directory, and to avoid
> > >>>> transferring a large amount of data between runners, we may need to keep
> > >>>> building libcamera within the test-unit job :-(
> > >>>>
> > >>>> A separate build-package target for lc-compliance would still make
> > >>>> sense.
> > >>
> > >> I think it would be unfortunate to give up the usage `meson test` as you
> > >> mentioned.
> > > 
> > > We could work on a replacement, but it would require a significant
> > > amount of work and I think there are better things to do.
> > > 
> > >> I have not noticed that these build artifacts would put any
> > >> appreciable strain on the infrastructure. The compressed build directory
> > >> comes out to around 167 MiB; I am not sure if I would consider that a
> > >> large amount of data. It is definitely cheaper, in terms of time, than
> > >> building libcamera twice. Clearing the object files could be another
> > >> option. With `artifacts:exclude: build/**/*.o` we can seemingly
> > >> remove more than half of the uncompressed size, and about 1/4 of
> > >> the compressed size. Does this look acceptable?
> > > 
> > > Possibly. We should probably ask the fdo sysadmins about what is
> > > acceptable.
> > > 
> > > I gave it a try locally though, and deleting all *.o files in the build
> > > directory results in "meson test" rebuilding everything.
> > 
> > As far as I can see that does not happen with the `--no-rebuild` option,
> > which is already used in `test-libcamera-qemu.sh`.

Ah, good point, I missed that.

> > > For other uses of the artifacts (in particular deployment on real
> > > devices), I would still prefer minimizing the bandwidth, creating a
> > > package similarly to what build-package:arm64 does. How about keeping
> > > test-unit as-is, at the cost of a recompilation, and creating a
> > > build-package:amd64 that will be used by the lc-compliance test job ? We
> > > can try to improve on top when/if needed.
> > 
> > Couple observations:
> > 
> > 1. The virtual pipeline handler configuration is not installed, so
> >     that needs to be addressed. (Was this omitted intentionally?)

I don't know. We have to be careful here, we don't want to virtual
pipeline handler to end up being enabled with a default valid
configuration in distributions, otherwise people will all of a sudden
see virtual cameras poluting their devices list.

> > 2. I am not a fan of the extra `tar` and `ldconfig`  calls that
> >     need to be sprinkled in. I think this would be much better
> >     if the package was not a mere tar archive but a proper deb/etc.
> >     package. I imagine that is a prerequisite of deploying on real
> >     hardware in any case, correct?
> 
> Having the server build a 'real' deb would be a real benefit IMO, and
> indeed could help with installation/set up on real targets for testing
> in defined environments.
> 
> I've wanted to tackle that for a while but never got time.

Same, I think it's probably worth it, but it will require some effort
and I didn't have enough time when I implemented build-package:arm64.

-- 
Regards,

Laurent Pinchart


More information about the libcamera-devel mailing list