Docker image for Linux tests ships minidump-stack-walk. The mozharness
script finds it instead of tooltool downloaded file, and when try to
give executable permission, it gets permission denied error because the
file is localted outside home directory. As minidump-stack-walk already
has executable permission, we skip the permission denied error.
MozReview-Commit-ID: 5IyUna02iPb
--HG--
extra : rebase_source : 0d60e53df9f925e799c30fadeff7cc7bfd9cb2da
This patch provides the implementation that makes it possible to run the devtools test suite with the linux64-jsdcov build to collect js code coverage.
MozReview-Commit-ID: KFmFhKsDq5s
--HG--
extra : rebase_source : 78894caa6b45a0e43fd1a4c29190788523b10e12
Currently, perfherder data containing resource usage goes into the same
bucket regardless of the execution environment. This means results for
buildbot, taskcluster, and for different AWS instance types are all
mixed together, making the data noisy.
This commit adds extraOptions values to the perfherder data to
identify:
* buildbot vs taskcluster
* the TC instance type
Data will go to separate buckets and should be much more consistent.
MozReview-Commit-ID: A0zSkdVI6ZM
--HG--
extra : rebase_source : 6047a3a0fcd9103747862fb2db7ba9b11a0cf95f
aa1e7115e8d7 removed code related to influx. Unfortunately, part of a
deleted class loaded build_resources.json and populated an attribute
holding the results. This attribute was consulted by
buildbase.BuildScript to send build metrics to perfherder. Oh, the
joy of mixins.
This commit essentially restores the build_resources.json loading
code as part of BuildScript.
MozReview-Commit-ID: IuFJd9Y6wFO
--HG--
extra : rebase_source : d2e5f1323c85207a6aaf096298c7f09d8bda87c8
This changeset adds a taskcluster specific mozharness script and posix config
for the external media tests. It also removes some of the no longer needed
config from the windows task cluster config for these tests.
MozReview-Commit-ID: 1E5fkFV5DeV
--HG--
extra : rebase_source : 268b09da0601b32e61a24195f703de0578de1317
I haven't seen a case of this happening in the wild, but I believe it is possible for a test to dump "TEST-UNEXPECTED-FAIL"
directly to the log and mozharness using the StructuredOutputParser would not pick it up. This patch is just me being extra
careful to flag potential errors like that. I'm not sure they even exist.
This patch also purposefully uses substr to avoid requiring said string to show up at the beginning, which should avoid
certain prefix issues we've run into in the past.
MozReview-Commit-ID: 99n9YizlEDH
--HG--
extra : rebase_source : 2063f2fc4ca109a4ac24ccdbc0672e4e1b4c0fb6
Bug 1306421 changed the behavior of proxxy in an attempt to make it
more consistent. As a result of that, it enabled proxxy in places
where it wasn't used before. One of them was the Jenkins QA environment
which runs update UI tests. Unfortunately, that environment
doesn't have connectivity to proxxy.
This commit simply blanks out the proxxy config in the Jenkins QA
environment, effectively disabling proxxy.
MozReview-Commit-ID: 3A37VKEGQ49
--HG--
extra : rebase_source : ac63444f66909079297defcf5e0e9ff10b66e510
This refactors the mozharness configs such that a specific flavor of an overall suite can still use the old
DesktopUnittestOutputParser. This is necessary because mochitest-jetpack still does not use the structured
logger.
MozReview-Commit-ID: E8EpSLH4xt2
--HG--
extra : rebase_source : be18e3e4aa1f693b563a82b90fdda2f93201d9ee
3fd83c9c0548 introduced the use of --trusted-host for use with pip.
Unfortunately, it unconditionally added --trusted-host, even when
easy_install is being used. This broke modules using easy_install
(which is currently limited to pywin32). Removing easy_install
and installing pywin32 is tracked in bug 1306691.
MozReview-Commit-ID: ExiO22EUr0B
--HG--
extra : rebase_source : 25dcded06a47ec8c0e61637bdd425131fe20a543