Bug 145479 – Crash in GraphicsContext3D::getInternalFramebufferSize

• Status: RESOLVED
• Resolution: FIXED
• Component: New Bugs

[Dean Jackson]

Crash in GraphicsContext3D::getInternalFramebufferSize

[Dean Jackson]

<rdar://problem/16461048>

[Dean Jackson]

Created attachment 253928 [details]
Patch

[Sam Weinig]

Comment on attachment 253928 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=253928&action=review

> Source/WebCore/ChangeLog:12
> +        This can't be easily tested automatically, because it's a specific
> +        configuration of Safari that uses WebGL in this manner.

We should add the necessary infrastructure to make it testable.

[Dean Jackson]

Created attachment 254103 [details]
Patch

[Eric Carlson]

Comment on attachment 254103 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=254103&action=review

> LayoutTests/fast/canvas/webgl/useWhilePending.html:18
> +        window.internals.settings.setForcePendingWebGLPolicy(false);

If you reset the property in InternalSettings::resetToConsistentState this won't be necessary.

[Dean Jackson]

Committed r185128: <http://trac.webkit.org/changeset/185128>

[Alexey Proskuryakov]

The new test times out on some bots:

https://webkit-test-results.appspot.com/dashboards/flakiness_dashboard.html#showAllRuns=true&tests=fast%2Fcanvas%2Fwebgl%2FuseWhilePending.html

[Alexey Proskuryakov]

That was because the test is not expected to work with WebKit1.

However, WebKit2 is broken too, it hits a very scary assertion: <https://build.webkit.org/results/Apple%20Mavericks%20Debug%20WK2%20(Tests)/r185140%20(12167)/fast/canvas/webgl/viewport-unchanged-upon-resize-crash-log.txt>.

This may be a pre-existing bug, but given that:
1) it's a very bad assertion to hit, and
2) it essentially forces us to skip the test, and
3) Sam requested this fix to be tested,

I'm going to roll out.

I suspect that there is an actual issue with uninitialized WebGL contexts when navigating.

[WebKit Commit Bot]

Re-opened since this is blocked by bug 145597

[Dean Jackson]

Re-landed in https://trac.webkit.org/r185172

Alexey was right - it was doing some "bad stuff" with the unitialised context, and hadn't been tested before. This bug has been around for about a year. The fix was pretty clear - to add
renderingContext->suspendIfNeeded();