Bug 221011

Summary:	[ macOS wk2 ] fast/selectors/selection-window-inactive-text-shadow.html flakey image failure
Product:	WebKit	Reporter:	Robert Jenner <jenner>
Component:	Layout and Rendering	Assignee:	Carlos Garcia Campos <cgarcia>
Status:	NEW ---
Severity:	Normal	CC:	ap, bfulgham, ryanhaddad, sihui_liu, simon.fraser, thorton, tsavell, webkit-bot-watchers-bugzilla, webkit-bug-importer, zalan
Priority:	P2	Keywords:	InRadar
Version:	WebKit Nightly Build
Hardware:	Unspecified
OS:	Unspecified
See Also:	https://bugs.webkit.org/show_bug.cgi?id=221857 https://bugs.webkit.org/show_bug.cgi?id=228248 https://bugs.webkit.org/show_bug.cgi?id=228337

Description Robert Jenner 2021-01-26 14:56:45 PST

fast/selectors/selection-window-inactive-text-shadow.html

Is a flakey image failure on Mac.

History: https://results.webkit.org/?suite=layout-tests&test=fast%2Fselectors%2Fselection-window-inactive-text-shadow.html

Image Diff: https://build.webkit.org/results/Apple-BigSur-Release-WK2-Tests/r271897%20(1193)/fast/selectors/selection-window-inactive-text-shadow-diffs.html

Comment 1 Radar WebKit Bug Importer 2021-01-26 14:57:21 PST

<rdar://problem/73634124>

Comment 2 Robert Jenner 2021-01-26 16:08:18 PST

Bug reproducible in r271900

Produced bug using:

run-webkit-test fast/selectors/selection-window-inactive-text-shadow.html --iterations 1000 -f

Regressed to build r27164 where bug was not occurring.

Testing forward we found that the bug was not reproducible on r271878, and was reproducible in r271879. 

It appears the bug was introduced in r271879.

Comment 3 Carlos Garcia Campos 2021-01-27 00:38:28 PST

There isn't any cross-platform change in r271879, only soup based ports should have a behavior change. And the test failing has nothing to do with the network loads. So, I'm quite sure, it wasn't r271879 the one introducing this, I guess it's difficult to find when it's flaky

Comment 4 Alexey Proskuryakov 2021-01-27 10:09:30 PST

There have been flaky failures on test for much longer.

When this fails, all of the text is blue instead of green. The test calls testRunner.setWindowIsKey(false), and doesn't wait at all. There is a lot of potential for async updates that need to be waited for here.

But also, we have other tests that do the same, and at least some were identified to have regressed with r266634.

A few examples from LayoutTests/fast:

fast/selectors/selection-window-inactive.html
fast//selectors/selection-window-inactive-stroke-color.html
fast/selectors/selection-window-inactive-text-shadow.html

It's not good that we broke all inactive window rendering tests.

Interestingly, fast/selectors/querySelector-window-inactive.html isn't flaky, so a:window-inactive gets updated synchronously, but rendering is not.

Comment 5 Robert Jenner 2021-03-05 13:17:48 PST

Updated test expectations here:
https://trac.webkit.org/changeset/274002/webkit