Yeah, I can reproduce this on my Windows bot. On Mac it works fine.

Bisecting ...

Note that this is already fixed on Canary.

https://chromium.googlesource.com/chromium/src/+/11b54cadcd04ea3b5650839abff6a6060d63e978 ANGLE roll is responsible

Thanks, Mo. Any idea where the breakage was? I'll try to figure out a workaround in the meantime if I can get an idea of what exactly the issue is.

Lots of dirty bit changes in that roll. Good that it's fixed now, but would be good to know which it was.

Tarek, it would be most helpful if you could occasionally test with Chrome Beta (at least, if not also Chrome Dev/Canary). The release channels can now be installed side-by-side on Windows.

This is a bug introduced by dirty-bit optimization in ANGLE. The fix has already landed in M63 so the next push will get the bug fixed (I don't know when is the next bug fix push though).

Beta and Canary already has this bug fixed.

Good to know, Ken. Will do. Thanks for the info, Mo.

BTW, is there a way to avoid Firefox using the buggy version of ANGLE? Maybe mark it as problematic. I know they tend to use a slightly older version than Chrome does. It would be a shame for them to introduce this bug into Firefox after it's already been fixed.

Yeah, I'll email Jeff Gilbert @ Mozilla about the ANGLE bug.

tsherif: We have some WebGL tests in place, and it's on me to add additional WebGL tests to ensure other browsers use the updated ANGLE. If anyone feels like helping here I can point you to the ANGLE tests that would need to be translated to javascript.

Yep, we have a problematic version in Firefox 58.

We can't pull in a new ANGLE update this late in Beta, so we'll be looking at patching what we have.

Thanks for the heads-up.

jmadill: If you can point me to the ANGLE tests and let me know where they should go in the conformance suite, I'd be happy to help with that.

tsherif:

in angle's src/tests/gl_tests/UniformBufferTest.cpp, UniformBufferTest.SimpleBindingChange is based on a test Mo added to the WebGL CTS here: https://github.com/KhronosGroup/WebGL/pull/2559. However there are two other test cases that should be ported:

UniformBufferTest.SimpleBufferChange and UniformBufferTest.DependentBufferChange. You could probably add these to the same test file.

Additionally there were other dirty bit bug tests:

src/tests/gl_tests/VertexAttributeTest.cpp: VertexAttributeTest.DisabledAttribUpdates (related to  issue 779675 )

src/tests/gl_tests/StateChangeTest.cpp:
StateChangeTest.VertexBufferUpdatedAfterDraw and StateChangeTestES3.VertexArrayObjectAndDisabledAttributes (related to  issue 778689 )

Note that a few might have to be WebGL-2 only and some can be WebGL 1. Thanks if you can help porting these.

jmadill: Sure those look pretty straightforward to port. I can probably get something ready in a week or so.

BTW, Is there a way to trigger UBO binding dirty state through the GL in the meantime?

I think unbinding and re-binding the program should clear the dirty states in any case. Thanks for agreeing to port the tests.

PR for the UBO conformance tests: https://github.com/KhronosGroup/WebGL/pull/2573

Thanks for contributing these tests Tarek!

PR for the rest of the dirty bit conformance tests: https://github.com/KhronosGroup/WebGL/pull/2600