REGRESSION: [ Mac WK2 ] animations/suspend-resume-animation-events.html is a flaky failure

[Jacob Uphoff]

This is a flakey failure on Mac wk2

I did not try to reproduce, however failure goes back as far as I can see the history.

History: 

https://results.webkit.org/?suite=layout-tests&test=animations%2Fsuspend-resume-animation-events.html

Diff: 

--- /Volumes/Data/slave/catalina-release-tests-wk2/build/layout-test-results/animations/suspend-resume-animation-events-expected.txt
+++ /Volumes/Data/slave/catalina-release-tests-wk2/build/layout-test-results/animations/suspend-resume-animation-events-actual.txt
@@ -4,5 +4,4 @@
 
 start animation move
 iteration animation move
-end animation move

[Radar WebKit Bug Importer]

<rdar://problem/58814195>

[Radar WebKit Bug Importer]

<rdar://problem/58814196>

[Ryan Haddad]

This became very frequent starting on 1/7/2020, around seemingly unrelated commit r254181.

Maybe this commit adding animations logging could be related?
Add some more Animations logging
​https://bugs.webkit.org/show_bug.cgi?id=205890
https://trac.webkit.org/changeset/254177/webkit

[Darin Adler]

This test is inherently racy and timing-dependent. If the 350ms timeout fires too soon, we would move on to the next test without seeing the end of the animation. I think we can make it much less racy by having each timer set the subsequent one instead of setting all three timers at once. I’ll try that, but no easy way to test if it worked.

[Darin Adler]

Created attachment 388773 [details]
Patch

[Alexey Proskuryakov]

Comment on attachment 388773 [details]
Patch

The change looks harmless with regards to what this tests check for, and likely beneficial for test reliability. However, I do not understand what is wrong with the test as is. 

The animation starts before any setTimeouts, and is shorter. Wouldn’t it be an actual WebKit bug if there is raciness?

[Darin Adler]

(In reply to Alexey Proskuryakov from comment #6)
> The animation starts before any setTimeouts, and is shorter. Wouldn’t it be
> an actual WebKit bug if there is raciness?

As far as I know, WebKit does not attempt to have a single unified timeline between JavaScript timers and webpage animation. Unless there’s something I’m not aware of. Could be a neat feature and might help webpages work more reliably. Not sure I’d call it a "bug" exactly.

[Darin Adler]

I’m going to land this change, even though I’m not sure it will eliminate the flakiness. If this attempt fails then we might need to find another way to write the test.

And I think it’s exciting to contemplate an architecture change where we unify JavaScript timers with animation timing and possibly even other time bases that affect web pages.

[Darin Adler]

Committed r255124: <https://trac.webkit.org/changeset/255124>