I think it's time we considered having tests running on the build
machines in HiDPI mode so that we can catch regressions that only
manifest themselves in high resolution configurations. We have at least
three platforms we're targeting where this is going to be useful:
Firefox on Retina MBPs, and Firefox for Metro and Firefox for Android on
high res devices.
I believe we can get good HiDPI test coverage just by running tests with
layout.css.devPixelsPerPx > 1 on non-high resolution hardware. There is
some work that needs to be done on the reftest framework to get that
working -- https://bugzilla.mozilla.org/show_bug.cgi?id=821210 and its
dependent -- to get high resolution snapshots happening, but that
shouldn't be difficult.
Last I tested in https://bugzilla.mozilla.org/show_bug.cgi?id=859742
there were around 300 test failures when running with
layout.css.devPixelsPerPx = 2. I haven't looked into how many of those
are real bugs or just tests that don't correctly test things in HiDPI mode.
I think the main question to ask is: what platforms would we want/need
to run tests in HiDPI mode on?
Given the capacity constraints we have in the build farm, would it be
sufficient to run HiDPI tests on Linux or Linux64, where we can easily
spin up more AWS machines? There are probably few enough differences
across platforms that we can just write individual tests to handle HiDPI
on Mac/Metro/Android when we need to. (roc pointed out to me that
reftest-zoom can be used for individual tests to get the same effect as
setting layout.css.devPixelsPerPx.)
As for getting the tree to a state where we can consider HiDPI test
failures as critical, we can start off with these tests hidden on tbpl
until we've fixed those ~300 tests, like we've done with other new
platforms/configurations.
Any reasons not to go ahead with this? If not, how do we proceed?
_______________________________________________
dev-platform mailing list
dev-platform@lists.mozilla.org
https://lists.mozilla.org/listinfo/dev-platform
