Bug 233965 – REGRESSION(r286603) [GTK][WPE] WebGL broken after the ANGLE update.

• Component: ANGLE
• Status: RESOLVED FIXED

[Arcady Goldmints-Orlov]

Changelist r286603, which updated ANGLE to a more recent version, seems to have broken WebGL entirely on both GTK and WPE platforms.

[Lauro Moura]

Created attachment 446280 [details]
Crash log from debug build

Running a debug build, the test crashes with the following message (full log attached):

STDERR: FATAL: UnfoldShortCircuitAST.h:26 (UnfoldShortCircuitAST):      ! Unreachable reached: UnfoldShortCircuitAST(/app/webkit/Source/ThirdParty/ANGLE/src/compiler/translator/tree_ops/apple/UnfoldShortCircuitAST.h:26)
STDERR: (ProcessSwapping) Removing process with pid 51 from the origin cache set
STDERR: UNIMPLEMENTED: 
STDERR: /app/webkit/Source/WebKit/UIProcess/API/wpe/PageClientImpl.cpp(371) : virtual void WebKit::PageClientImpl::closeFullScreenManager()
STDERR: WPEWebProcess terminated (pid 51) because the process crashed
STDERR: LEAK: 1 WebPageProxy⏎

[Kimmo Kinnunen]

Could you specify if it's the OpenGL backends, where the ANGLE is used only as a shader translator, or ANGLE backends, where it's used for real?

[Lauro Moura]

(In reply to Kimmo Kinnunen from comment #2)
> Could you specify if it's the OpenGL backends, where the ANGLE is used only
> as a shader translator, or ANGLE backends, where it's used for real?

The former, WebKitGTK and WPE use the OpenGL backend. We're still working on the full ANGLE support.

[Lauro Moura]

I'm did some tests in the WPE/GTK cmake build defining ANGLE_ENABLE_APPLE_WORKAROUNDS and compiling the apple translator sources. Is this a sensible approach?

Results with WPE:

=> Results: 623/654 tests passed (95.3%)

=> Tests to be fixed (6300):
      1 crashes                  ( 0.0%)
      9 timeouts                 ( 0.1%)

=> Tests that will only be fixed if they crash (WONTFIX) (0):


Expected to fail, but passed: (1)
  webgl/1.0.x/conformance/extensions/oes-vertex-array-object.html


Regressions: Unexpected text-only failures (3)
  webgl/1.0.3/conformance/extensions/webgl-compressed-texture-s3tc.html [ Failure ]
  webgl/1.0.3/conformance/glsl/misc/shaders-with-name-conflicts.html [ Failure ]
  webgl/1.0.3/conformance/textures/tex-image-and-sub-image-2d-with-video.html [ Failure ]

Regressions: Unexpected timeouts (2)
  webgl/1.0.3/conformance/context/context-eviction-with-garbage-collection.html [ Timeout ]
  webgl/1.0.3/conformance/glsl/misc/expression-list-in-declarator-initializer.html [ Timeout ]

[Kimmo Kinnunen]

I think you should remove the use of SH_UNFOLD_SHORT_CIRCUIT.

[Lauro Moura]

(In reply to Kimmo Kinnunen from comment #5)
> I think you should remove the use of SH_UNFOLD_SHORT_CIRCUIT.

Seems to have worked like a charm, thanks!

Doing a few more tests with both ports before submitting.

[Lauro Moura]

Created attachment 446359 [details]
Patch

[EWS]

Committed r286653 (?): <https://commits.webkit.org/r286653>

All reviewed patches have been landed. Closing bug and clearing flags on attachment 446359 [details].

[Radar WebKit Bug Importer]

<rdar://problem/86210561>