Hopefully this doesn't cause any failures in wdspec tests:

try: https://treeherder.mozilla.org/#/jobs?repo=try&revision=1034e3a0e81589a0f9619051269ca5d15bedc1d9

wdspec launches Firefox using geckodriver. This was going through a
different codepath to the other wpt tests and in particular it wasn't
setting the correct environment variables.

This patch factors the environment variable setting out into a
function that can be called both when starting Firefox normally and
before invoking pytest. This isn't perfect, but does allow more code
sharing between the two paths.

I'm still seeing the issue with wdspec tests:

https://treeherder.mozilla.org/jobs?repo=try&revision=5fcbc4c7121894133f09f9945e5af8c6745046e1&selectedTaskRun=KDrcrnd8Sd-MbDDHOBuXzQ.0&searchStr=Linux%2C18.04%2Cx64%2CWebRender%2Copt%2CWeb%2Cplatform%2Ctests%2Ctest-linux1804-64-qr%2Fopt-web-platform-tests-wdspec-e10s%2CWd1

I added logs:

[task 2020-11-19T16:15:41.680Z] 16:15:41 INFO - PID 911 | [AO] wr=1 sw_wr=1; ee=0 ed=0 ps=0

which show MOZ_WEBRENDER=1 wasn't set (ee=0).

The previous patch set the variables too late, after the webdriver server was already
started. Pass them down into the server startup code instead

https://hg.mozilla.org/mozilla-central/rev/c4756cf7f5f1

This ensures that we always pass a profile in the capabilitites. When
we're running with a single process it simply unpacks the profile to a
temporary folder and passes in that path as a firefox argument. When
we're running with multiple processes, that won't work. We could
invent a mechanism to get one profile per process, but for simplicity
we instead pass in the profile as a base64-encoded string.

This bug is fixed. James, should all these revisions actually be part of bug 1680963?

As mentioned by James this bug isn't fixed. So lets reopen. It will also block bug 1680963.

try: https://treeherder.mozilla.org/jobs?repo=try&selectedTaskRun=K57ozD7uSdKjbm1LB_mynA.0&revision=0d9368f856f1f0a709e8560156891b6baffa19f1

The debug prints show we aren't getting the MOZ_WEBRENDER envvar set in the qr and non-qr variations.

https://hg.mozilla.org/mozilla-central/rev/0e5887add977

This requires passing the test_environment into the get_executor_kwargs function
so that in the firefox wdspec case we can add a cleanup function to the environment
when running wdspec tests. That seems reasonable since we were previously using
a variety of data in the environment to setup the kwargs anyway

Otherwise it seems to interfere with the rest of the test.

Setting Fission Milestone to M8 because this bug blocks Fission M8 bug 1680963.

Backout changeset for causing marionette failures at test_fission_autostart.py .

Push with failures

Failure log

also failes here python/test_mochitest_integration.py
Failure log

Backout link

Backed out 14 changesets (bug 1686707, bug 1707876, bug 1678044) for Spidermonkey failure. CLOSED TREE

Log:
https://treeherder.mozilla.org/logviewer?job_id=340215481&repo=autoland&lineNumber=45654

Backout:
https://hg.mozilla.org/integration/autoland/rev/532f607c9ca037bdfc9d671a7447b28b19c05c77

https://hg.mozilla.org/mozilla-central/rev/222c0fda8126
https://hg.mozilla.org/mozilla-central/rev/6dc4d87d1b65
https://hg.mozilla.org/mozilla-central/rev/1f77f1f4241f
https://hg.mozilla.org/mozilla-central/rev/0a5ac7d40966
https://hg.mozilla.org/mozilla-central/rev/98347e378726

James, I think that we can close this bug now, right? If not please reopen.

https://hg.mozilla.org/mozilla-central/rev/2b2ea8650148