REGRESSION: fast/canvas/webgl/context-attributes-alpha.html fails
after bug 217212
Created attachment 411306 [details]
Comment on attachment 411306 [details]
View in context: https://bugs.webkit.org/attachment.cgi?id=411306&action=review
> + WebGLLayer contentsOpaque. Neither the EWS nor the local testing
Why is. that?
Committed r268446: <https://trac.webkit.org/changeset/268446>
All reviewed patches have been landed. Closing bug and clearing flags on attachment 411306 [details].
>> + WebGLLayer contentsOpaque. Neither the EWS nor the local testing
> Why is. that?
It is unclear at this point, I'll investigate this more.
- I developed the patch on iMac with AMD card. I thought I ran the tests as supposed.
- I noticed the regression on MacBook Pro with Intel/AMD, with automatic switching on
- EWS runs Intel opus
- Intel GPUs have bugs around IOSurfaces and their alpha channels. ANGLE tries to work around them.
- The test and the expected.html both use the WebGL canvas alpha feature, which turned out to have the bug
- The test should be run on Mac and expected to work
- The -expected.html probably should not use the exact same code as what the test is testing, but it's somewhat tricky to alter the content in GPU-agnostic way
- Did I run the tests locally?
- Does Intel alpha bugs, ANGLE workarounds and the bug in question cancel each other out in EWS?
- Does EWS run fast/canvas/webgl/context-attributes-alpha.html
- If the bugs cancel each other out, why don't they cancel out with my MacBook Pro? Does it use the AMD part or are the drivers different or is there a bug in ANGLE workaround selection.
I'll try to investigate the unknowns
To clarify, EWS is Mac Pros with AMD GPUs.
Bots that saw this failure were all newer AMD ones: