Change from -s to -S to fix:
  argparse.ArgumentError: argument -s: conflicting option string: -s

Send the splitrepo a job belongs to for each API request. This is in
preparation for the split systemd repository. As always, add the new
parameter to all existing jobs (helper fails if env vars are missing).

Luca Weiss authored 6 months ago and Oliver Smith committed 5 months ago

Sometimes just resetting e.g. packages is desired, so add that functionality using arguments.

pmbootstrap v3 doesn't create the "aports" symlink anymore that points
to the pmaports dir. Make it possible to set the pmaports dir via an
environment variable and set it from related scripts.

Add a helper script that can be used to run just one test with pytest,
but using pmbootstrap v2.

Support setting and getting the retry_count. This is useful for
intentionally setting packages from queued to failed with max retry
count.

This is not needed anymore, pytest_logs.sh has the same purpose and is a
better implementation (as it can follow all job logs properly).

Don't complain about job logs disappearing.

* move it to the bottom, so it gets displayed last: the pytest.log file
  is most relevant, the local job logs are only filled once a job gets
  started
* touch the file, so there is no error about the file not existing if
  the user runs pytest_logs.sh before starting pytest
* fix the path, it is pytest.log, not test/pytest.log

Luca Weiss authored 1 year ago and Oliver Smith committed 1 year ago

Inspired by the recent pmbootstrap change from Hugo Osvaldo Barrera, use
ruff instead of flake8. Aside from being substantially faster, ruff
includes a lot of flake's plug-ins built-in, some of which may be useful
to enable as a follow-up.

Remove an outdated section from README (replaced by pmbootstrap ci) and
ignore the E741 warning for now since flake8 didn't complain about that.

Edit by ollieparanoid: use edge in .gitlab-ci.yml for ruff, as it is
currently only available in testing.

Add a script that runs tail -F on all logs that pytest generates.

Similar to bpo_failed_to_queued.py. Currently sometimes sourcehut
builds get stuck in 'running' state like here and can't be canceled:
https://builds.sr.ht/~postmarketos/job/920725

So this script marks them as failed.

Fix running into max upload file size errors by uploading only one file
at once instead of uploading all at once. We hit this limit currently
with some installer images and builds with multiple kernels.

Fixes: #114

Prepare to split the image upload across multiple api calls. There will
be one api call per upload, and a last one that doesn't have
BPO_PAYLOAD_FILES set, which indicates that the upload is finished.

Add a new maintenance script that simply resets all failed entries back
to queued. If we need something more sophisticated to deal with images,
we can add it later, this should cover the most important use case.

Do not set the combined read and connect timeout, as this causes
problems when uploading big image files in the testsuite.

Instead of overwriting the defaults at the end of init(), read the
defaults from a separate file. This allows running init() more than once
in the testsuite without risking to use a previous argument value if the
argument was first specified and then absent.

Example (without the patch, this returns 'test' twice):
$ python3
>>> import sys, bpo, bpo.config.args
>>> sys.argv = ["bpo.py", "-m" "test", "local"]
>>> bpo.main()
^C>> bpo.config.args.mirror
'test'
>>> sys.argv = ["bpo.py", "local"]
>>> bpo.main()
^C>> bpo.config.args.mirror
'http://mirror.postmarketos.org/postmarketos'

Prepare to submit images.

List all packages before modifying them, so it's easy to see the
previous value.

Not the most elegant solution, but let's focus on getting the v20.05
branch building.

Put submit.py in this repository. It is only useful here, as changes in
submit.py always need to be syntax with the bpo code and bpo is the only
user of the script. Also, with the upcoming release branches, we would
need to sync the changes in pmaports.git across multiple branches.

To make this work, adjust the sourcehut builds and local job services to
make submit.py from build.postmarketos.org.git available during the build.

Make it easy to change the status for all failed packages back to queued
(which was the original intention of this tool).

$ ./bpo_package_status.py -f failed -s queued
Will change status to 'queued' for the following packages: ['hello-world', 'hello-world-wrapper']
Are you sure? [y/N]

Simple script for setting and getting the package status from the bpo
sqlite database. See "bpo_package_status.py -h" for details.

usage: bpo_package_status.py [-h] [-a ARCH] [-b BRANCH] [-d DB_PATH]
                             (-g | -s {building,built,failed,published,queued})
                             [pkgname [pkgname ...]]

Closes: #44