Age | Commit message (Collapse) | Author |
|
Patches that just change binary files like:
https://git.linaro.org/uefi/OpenPlatformPkg.git/commit/?id=7ab4bb34b2464a2491868264bdf2931f2acd6452
will produce a None value. We need to treat this like '' to make the
rest of our code sane.
Change-Id: Id9d1b7104cd8410709e01e08e81f96a1ed80f451
|
|
We occasionally see giant patches come in, normally from libvirt
releases such as:
http://libvirt.org/git/?p=libvirt.git;a=commit;h=0086221cb377c7a1c2df58b8ee5eb8223f989765
Patchwork looks at patches line by line, and this consumes tons of
memory and runs for hours blocking everything. This detects giant
patches and safely discards them.
Change-Id: I2403941c708c0923bbe143e783f5869da88f880d
|
|
We often find upstream repos are periodically unreachable. This allows
us to fail a few times before we actually log an error. (Error logs
result in cron email so its good to make sure they are real issues)
Change-Id: I8f1a0ae19b36fbd1af4ef26b3d68fa85501577c5
|
|
Code that uses subprocess.* is always a bit weird to manage when called
via cron where any output will cause a warning email. This change allows
commands to be run in their natural non "quiet" mode, and only dump
stdout/stderr when a problem happens or log.DEBUG is enabled.
Change-Id: I8d4e83044c3132d287356cb6d3283e99c902b170
|
|
We have unnecessary complication in our code due to the fact we don't
use bare repos. Additionally bare repos are much faster for our needs
because we do less file I/O. Why was there ever non-bare? I suspect its
just historical baggage. I've converted the repos on
staging-patches.linaro.org over and things are working much more
reliably from cron now.
Change-Id: Ie96022b9dcda7ee34ec7c35d9befea6b5f03f8c1
|
|
Occasionally, the linux kernel has commits with non-utf8 characters.
This keeps us from blowing up and should still work well with the patch
compare logic we have that has "fuzzy" matching logic.
Change-Id: Ia09248f2574b47c68897699c2ec4976f032b275b
|
|
update_patches needs similar logic that we had in import_email for
marking patches as superseded. This logic is very close to what is
currently in linaro_patchmetrics but done in a more pythonic and
testable way.
This was tested on the current lng-odp repo and found over 100 patches
that this change corrects. The test logic includes some patch examples
to help start building a nice base to test our algorithms against:
v2 of a patch: http://patches.opendataplane.org/patch/1781/
v3 of a patch: http://patches.opendataplane.org/patch/1792/
Change-Id: I3c41c9ad2cab274389a327d6dac266fc0b7f8244
|
|
At times we need to debug this command to see why certain patches get
missed. This change allows us to target the new commits for a specific project,
or just parse through a specific list of commits in a project.
adapted from linaro-patchmetrics:
https://git.linaro.org/infrastructure/linaro-patchmetrics.git/commit/54555f4bbee80e79c8694f8c18f1ef02aca9fb36
Change-Id: Idb93721c5d022716282817210c1e5724ac687f8d
|
|
future changes will need to look at the author, committer, and message
of a commit.
Change-Id: Idbe898befd12df653dbed876f002f03f8a2a50a4
|
|
This was causing cron emails
Change-Id: Ib212fbb619abdd30f0c21fd4ad59b923e1471471
|
|
We need a way to enable debug logging and the current logging setup was
inadequate. This provides a common way to add a "--log" option which
includes an argparse.Action that can set the logging level if the users
specifies.
This also changes the default logging level to WARN so that cron won't
send emails unless there is a problem.
gitrepo.py was updating to make its output adhere to logging settings
and not dump debug info the stdout when not needed.
Change-Id: I9c43217c5cf93c6f9cd12555733250f5e6d25c6b
|
|
This makes doing local development and testing much easier.
Change-Id: Ic2e921d1095e59e3a023d4f9420a7e0387b2c859
|
|
Looking at update-committed-patches in the linaro-patchmetrics file and
seeing where this is headed, I think now is the time to split
functionality up so its easier to manage.
This also improves the logic of managing the last analyzed commit by
introducing a process_unchecked_commits method. The result of the change
also simplifies/improves some of the logic needed for unit-testing.
Change-Id: I1e28f3fe3b27e1afd414939d28acc5eb5f6387cf
|