Open Bug 1558515 Opened 5 months ago Updated 4 months ago

Intermittent SVG object-fit reftest failures on geckoview (x86-64 Android 7.0 emulator)

Categories

(Core :: SVG, defect, P3)

Other Branch
Unspecified
Android
defect

Tracking

()

Tracking Status
firefox69 --- affected

People

(Reporter: kats, Unassigned)

References

Details

I'm turning on gecko reftests on Geckoview in bug 1501582 and as part of that annotating any failures I'm running into. There's a whole bunch of tests in layout/reftests/w3c-css/submitted/images/reftest.list that are intermittently failing and I had to mark them random-if(geckoview). Every test of the form object-fit-{contain,cover,none,scale-down}-svg-00?{e,o} and object-position-svg-00?{e,o} is affected.

Visually it looks like some of the boxes don't apply the object-fit/object-position properties and it just fills the entire box instead of being positioned where it's supposed to. Given that the png variants of these tests are not affected I'm assuming it's specific to SVG.

OS: Unspecified → Android
Whiteboard: [geckoview]

"e" and "o" are for <embed> and <object>, which use nsSubDocumentFrame to host the SVG content.

I wonder if (and hope that) this is a race condition, and maybe these tests end up rendering correctly but we take the snapshot too early, due to the subframe's pending repaint not getting picked up by the reftest harness...? That seems conceivable, particularly if we're using special reftest harness code for GeckoView.

Priority: -- → P3
Whiteboard: [geckoview]

I'd also note that a bunch of these tests fail much more often when run in the WPT harness than they do in the primary reftest harness.

You need to log in before you can comment on or make changes to this bug.