Bug 200901 – Pass conformance/attribs WebGL conformance tests

• Status: RESOLVED
• Resolution: FIXED
• Component: WebGL

[Kenneth Russell]

Currently WebKit's WebGL implementation fails a few of the conformance/attribs WebGL conformance tests, specifically:

https://www.khronos.org/registry/webgl/sdk/tests/conformance/attribs/gl-disabled-vertex-attrib-update.html?webglVersion=1&quiet=0&quick=1
https://www.khronos.org/registry/webgl/sdk/tests/conformance/attribs/gl-vertexattribpointer.html?webglVersion=1&quiet=0&quick=1
https://www.khronos.org/registry/webgl/sdk/tests/conformance/attribs/gl-vertex-attrib-unconsumed-out-of-bounds.html?webglVersion=1&quiet=0&quick=1

The reason appears to be that the spec was changed recently to allow vertexAttribPointer to be called with a null WebGLBuffer bound, but only if the associated offset is 0, in order to reset the vertex attribute's state. This one fix will likely fix all of these tests.

[Kenneth Russell]

Kai, may I ask you to take this bug?

[Kai Ninomiya]

Created attachment 376829 [details]
Patch

[Alex Christensen]

Please update the LayoutTests/...-expected.txt of the newly passing tests.  You can do this automatically by using run-webkit-tests <test name> --reset

[Kai Ninomiya]

Ah, thanks for the pointer. I was trying to find where I could update the expectations but couldn't find anything because I just grepped for the filename.

[Kai Ninomiya]

Created attachment 376914 [details]
Patch

[Kai Ninomiya]

I can't find the test logs for the first patch anymore (they seem to have vanished between this morning and reloading the page while working on the second patch?), so actually see what failed.
AFAICT without that, out of the 3 tests Ken listed, only 1 actually exists (in several places) in LayoutTests:

- webgl/2.0.0/conformance/attribs/gl-vertexattribpointer.html: updated results
- webgl/1.0.2/conformance/attribs/gl-vertexattribpointer.html: old snapshot, already passing
- fast/canvas/webgl/gl-vertexattribpointer.html: old snapshot, already passing

[Alex Christensen]

Click "Show Obsolete", click the red bubble, then "view layout test results"

[Kenneth Russell]

New patch LGTM though Alex should review and approve it.

I also only see the style and win bots having run against the now-obsolete patch.

[Kai Ninomiya]

Yup, only "style" and "win" buttons; all of the other test jobs seem to disappear gradually as each one completes.

Is there a permission necessary to view the complete results?

[Alex Christensen]

EWS indicates this patch still makes tests not match their expectations. Run this command:
Tools/Scripts/run-webkit-tests webgl/2.0.0/conformance/attribs/gl-vertexattribpointer.html webgl/1.0.2/conformance/attribs/gl-vertexattribpointer.html fast/canvas/webgl/gl-vertexattribpointer.html --reset

[Kai Ninomiya]

Found that if I click "style" then "All EWS Queues", I can see the 4 other queues I could see before. But they are all jsc-* and all of them say "Patch was not relevant", so I guess that's why they disappeared.

FWIW, all of that looks exactly the same if I'm logged out (in incognito).

[Kai Ninomiya]

Thanks for providing that command. Unfortunately those are the same files I tried to rerun earlier. Running them does not change the -expected.txt files, and if I look at the 3 "Writing new expected result" files manually, they already say everything is passing.

[Alex Christensen]

Hmm, I just verified the tests pass.  Aakash, is something wrong with EWS?

[Aakash Jain]

(In reply to Alex Christensen from comment #13)
> Hmm, I just verified the tests pass.  Aakash, is something wrong with EWS?
I don't see anything visibly wrong in EWS. EWS shows exact same failure on 4 different queues (4 different bots). Also, those failures were reported only on this patch, not on any other patch recently (so can't blame those tests as flaky).

Maybe the failures are specific to the OS/configuration. The bots are running High-Sierra.

[Kai Ninomiya]

Thanks for looking into it and for the suggestion. Is it possible to get a link to or copy of the test logs that I can view to try to figure out the issue?

[Aakash Jain]

(In reply to Kai Ninomiya from comment #15)
> Thanks for looking into it and for the suggestion. Is it possible to get a
> link to or copy of the test logs that I can view to try to figure out the issue?
Please click the red bubbles (e.g.: ios-wk2 bubble), it would take you to another (buildbot) page, in that click on "view layout test results" to see the test results.

e.g.: 
https://ews-build.webkit.org/results/iOS-12-Simulator-WK2-Tests-EWS/r376914-2216/results.html
https://ews-build.webkit.org/results/macOS-High-Sierra-Release-WK1-Tests-EWS/r376914-1304/results.html
https://ews-build.webkit.org/results/macOS-High-Sierra-Release-WK2-Tests-EWS/r376914-1627/results.html
https://ews-build.webkit.org/results/macOS-High-Sierra-Debug-WK1-Tests-EWS/r376914-157/results.html

[Kai Ninomiya]

Thank you for the links! I will take a look at the results tomorrow.

As Ken and I pointed out in #c8 #c9 though, I am still having the issue where there are no red bubbles to click on, and I can only see 6 queues. Here are some screenshots:
https://drive.google.com/drive/folders/1cdIErzConMbGdY3yJTjgeDu4spM_32mH?usp=sharing

[Kai Ninomiya]

Created attachment 377003 [details]
Patch

[Kai Ninomiya]

Oh, the problem was just that this change also fixed conformance/more/functions/vertexAttribPointerBadArgs.html and I didn't know that.

Unfortunately, the 1.0.2 snapshot of this test is not compatible with 1.0.3+ and 2.0.0+, and it is expected to fail. (It fails in Chrome and Firefox as well.)

PTAL!

[Alex Christensen]

Comment on attachment 377003 [details]
Patch

r=me

Aakash, I can also not see the bubbles from home, but I could see them from all browsers logged in or not from work.  Maybe they're only accessible from within Apple?  "curl 'https://ews.webkit.org/status-bubble/376829/' -v" returns 404 for me at home.

[WebKit Commit Bot]

Comment on attachment 377003 [details]
Patch

Clearing flags on attachment: 377003

Committed r249007: <https://trac.webkit.org/changeset/249007>

[WebKit Commit Bot]

All reviewed patches have been landed.  Closing bug.

[Radar WebKit Bug Importer]

<rdar://problem/54599744>

[Kenneth Russell]

Great work Kai getting to the bottom of this!

[Aakash Jain]

(In reply to Kai Ninomiya from comment #17)
> Thank you for the links! I will take a look at the results tomorrow.
> 
> As Ken and I pointed out in #c8 #c9 though, I am still having the issue
> where there are no red bubbles to click on, and I can only see 6 queues.
> Here are some screenshots:
> https://drive.google.com/drive/folders/
> 1cdIErzConMbGdY3yJTjgeDu4spM_32mH?usp=sharing
Sorry, I missed reading your earlier comments, and was looking into the test failures.
There were some DNS changes yesterday, and because of that EWS bubbles are not loading for some of the users. We are looking into it.

https://lists.webkit.org/pipermail/webkit-dev/2019-August/030761.html

[Kai Ninomiya]

No problem, there were 2 issues to solve, that one was just blocking the other :)

FYI, the bubbles have been working again recently!