webkit_layout_tests fail on 10.14 |
||||||
Issue descriptionBot: https://luci-milo.appspot.com/p/chromium/builders/luci.chromium.ci/mac-osxbeta-rel Failing run: https://luci-milo.appspot.com/p/chromium/builders/luci.chromium.ci/mac-osxbeta-rel/69 Failure log: https://logs.chromium.org/logs/chromium/buildbucket/cr-buildbucket.appspot.com/8933611326168921680/+/steps/webkit_layout_tests_on_Intel_GPU_on_Mac_on_Mac-10.14/0/stdout Invalid test result
,
Nov 15
The following revision refers to this bug: https://chromium.googlesource.com/chromium/src.git/+/194ae154045ac6e880fae4ad14b3b4ca9f82e395 commit 194ae154045ac6e880fae4ad14b3b4ca9f82e395 Author: Elly Fong-Jones <ellyjones@chromium.org> Date: Thu Nov 15 13:48:13 2018 blink: use php7 for web tests on 10.14 php5 is unsupported, same as 10.13. Bug: 892187 Change-Id: I52bc19b4f6ea8408093b7fcc61f6fe5153c984fd Reviewed-on: https://chromium-review.googlesource.com/c/1335947 Reviewed-by: Dirk Pranke <dpranke@chromium.org> Commit-Queue: Elly Fong-Jones <ellyjones@chromium.org> Cr-Commit-Position: refs/heads/master@{#608349} [modify] https://crrev.com/194ae154045ac6e880fae4ad14b3b4ca9f82e395/third_party/blink/tools/blinkpy/web_tests/port/mac.py
,
Dec 3
There are now a lot of tests (~8000) that are failing, mostly with "image diff". I think that perhaps we need to create a mac-mac10.14 directory in third_party/blink/web_tests/platform/ and put the expectations for these tests in there. How do we do that? Over to tkent@: I'm a bit out of my depth here, I could use either a lot of advice or some extra hands :)
,
Dec 3
You'd actually create a mac-mac10.13 directory, and move the 10.13-specific baselines there and put the baselines for 10.14 into mac/. robertma@ and/or foolip@ can probably help here.
,
Dec 3
So I've done this for 10.13. Essentially as Dirk said, we need to create new baselines for Mac 10.14. I was hoping there would be fewer breakages this time than 10.12 -> 10.13, but looks like the number is on the same scale. The work itself isn't technical at all, but it's a lot of babysitting as we are talking about landing a CL with thousands of files (or multiple, split CLs if you want to minimize merge conflicts) and maintaining a suite to be green on a bot that is not on CQ. You'd need to modify the following stuff: 0. Set up a Blink try bot for 10.14. 1. Port: https://cs.chromium.org/chromium/src/third_party/blink/tools/blinkpy/web_tests/port/mac.py?l=53&rcl=9b69b0c6203dc822162d903db77a63225aff4125 FALLBACK_PATHS['mac10.13'] = ['mac-mac10.13'] + FALLBACK_PATHS['mac10.14'] 2. rebaseline config: https://cs.chromium.org/chromium/src/third_party/blink/tools/blinkpy/common/config/builders.json?l=74&rcl=4f51d5d775713e1b428ad8fa5ee131c411753f3c Add a new entry for the 10.14 bot (remember to set is_try_builder to true). 3. Run `blink_tools.py rebaseline-cl` and sit back. (You don't need to rename any baselines yourself. rebaseline-cl should take care of it.)
,
Dec 3
It is par for the course for Apple to make enough changes to their text rendering and form controls to break every pixel test in a major OS release. I think that's only *not* happened once or twice.
,
Dec 4
What is a blink try bot & is it different from a regular try bot? +cc jbudorick@ for #5 - I think I need a bot to do this.
,
Dec 4
These are the blink trybots: https://ci.chromium.org/search?q=blink-rel They are different from the regular trybots in that (a) they are regular release builds, not DCHECK builds, and (b) they only run the layout_tests and not all of the test suites. They are used by the blink tooling (like rebaseline-cl) to generate all of the new baselines for a change. Yes, we should set up a 10.14 trybot.
,
Dec 5
I haven't done this kind of task, and have no time for it.
,
Dec 5
Okay - I will try to work through this. jbudorick@, I'm going to need a bot per #5 and #8. |
||||||
►
Sign in to add a comment |
||||||
Comment 1 by ellyjo...@chromium.org
, Oct 19Owner: ellyjo...@chromium.org
Status: Assigned (was: Untriaged)