Layout Test webgl/2.0.0/conformance/glsl/misc/shaders-with-invariance.html is failing since ANGLE roll

[Russell Epstein]

The following layout test is failing on iOS:

webgl/2.0.0/conformance/glsl/misc/shaders-with-invariance.html

Probable cause:

Test started failing with r249823
From James Darpinian in associated bug 201156:
"Looks like the shaders-with-invariance.html test is broken. It was updated upstream. WebKit's copy of the WebGL conformance tests should be updated."

Flakiness Dashboard:

https://webkit-test-results.webkit.org/dashboards/flakiness_dashboard.html#showAllRuns=true&tests=webgl%2F2.0.0%2Fconformance%2Fglsl%2Fmisc%2Fshaders-with-invariance.html

[Radar WebKit Bug Importer]

<rdar://problem/55358159>

[Russell Epstein]

Marked test as failing in r249860.

[Dean Jackson]

It was the ANGLE update in r249823 that caused this.

The actual change in ANGLE is: https://chromium.googlesource.com/angle/angle/+/b7418a177f70ab736c5d4b1c3b42f9d387244718

although that seems to be an attempt to fix the same failure in Chrome.

Note that our version of this test is out of date.

I'm looking through https://bugs.chromium.org/p/chromium/issues/detail?id=980675 to see how Chrome addressed this.

[Dean Jackson]

The difference is in the translated fragment shader source.

Before:
invariant varying mediump vec4 webgl_a78e73e400000001;
void main(){
(gl_FragColor = vec4(0.0, 0.0, 0.0, 0.0));
(gl_FragColor = webgl_a78e73e400000001);
}

After:
varying mediump vec4 webgl_a78e73e400000001;
void main(){
(gl_FragColor = vec4(0.0, 0.0, 0.0, 0.0));
(gl_FragColor = webgl_a78e73e400000001);
}

However, my reading of the GLSL specifications says this is ok.

"Only variables output from a shader can be candidates for invariance. This includes user-defined output variables and the built-in output variables. As only outputs can be declared as invariant, an output from one shader stage will still match an input of a subsequent stage without the input being declared as invariant."

[Dean Jackson]

False alarm. I was looking at the wrong file :(

[Dean Jackson]

Created attachment 380486 [details]
Patch

[Kenneth Russell]

Ah, relieved to know that the test was out of date and that there wasn't a bug in the version of ANGLE that WebKit rolled forward!

[Dean Jackson]

Created attachment 380538 [details]
Patch

[Dean Jackson]

Committed r250917: <https://trac.webkit.org/changeset/250917>