Filed by: ncsoregi [at] mozilla.com
Parsed log: https://treeherder.mozilla.org/logviewer.html#?job_id=261863125&repo=mozilla-central
Full log: https://queue.taskcluster.net/v1/task/KL3nctMCSIaHz7SMmlZffw/runs/0/artifacts/public/logs/live_backing.log
Reftest URL: https://hg.mozilla.org/mozilla-central/raw-file/tip/layout/tools/reftest/reftest-analyzer.xhtml#logurl=https://queue.taskcluster.net/v1/task/KL3nctMCSIaHz7SMmlZffw/runs/0/artifacts/public/logs/live_backing.log&only_show_unexpected=1

[task 2019-08-15T20:47:53.865Z] 20:47:53 INFO - TEST-START | /css/css-flexbox/auto-margins-001.html
[task 2019-08-15T20:47:53.868Z] 20:47:53 INFO - PID 7413 | 1565902073864 Marionette INFO Testing http://web-platform.test:8000/css/css-flexbox/auto-margins-001.html == http://web-platform.test:8000/css/css-flexbox/auto-margins-001-ref.html
[task 2019-08-15T20:47:54.066Z] 20:47:54 INFO - PID 7413 | 1565902074061 Marionette INFO No differences allowed
[task 2019-08-15T20:47:54.123Z] 20:47:54 INFO - TEST-UNEXPECTED-FAIL | /css/css-flexbox/auto-margins-001.html | Testing http://web-platform.test:8000/css/css-flexbox/auto-margins-001.html == http://web-platform.test:8000/css/css-flexbox/auto-margins-001-ref.html

The screenshot of the testcase shows that it's doing ink-skipping for the underline on the link, whereas we're not doing ink-skipping in the screenshot of the reference case.

The underline is just from a standard <a> element -- no fancy styling or anything.

So, it looks like we're intermittently failing to do ink-skipping (and we fail over to just drawing the full line). Charlie mentioned that this might happen if we fail to create an SkFont, for example -- that might be what happened here...

If I notice another bug where this happens, I'll probably file a general "[some reftest] intermittent reftest failure due to inconsistently honoring skip-ink" metabug and dupe this against that. Worst-case, we could disable the skip-ink pref for some subset of platforms/tests after we've got more data about where this is happening. (In this case, it was for a linux64 webrender web-platform-test.)

So far, we've hit this 6 times, all 6 of them on opt linux64-qr or linux64-shippable-qr builds:
https://treeherder.mozilla.org/intermittent-failures.html#/bugdetails?startday=2019-08-13&endday=2019-08-16&tree=all&bug=1574306

Bug 1574031 might be similar (adding to see also). That one is also opt-linux-only so far (both -qr and not-QR, in that case).

I landed a band-aid for this over in bug 1575338 comment 6 (turning off the pref for this test). Anyway, this should be FIXED by that patch (once it hits central at least)

(I meant to post that patch here, but not a big deal.)