Bug 37098 – A few tests in fast/canvas/webgl failed randomly on Leopard Commit Bot

• Component: Tools / Tests
• Status: RESOLVED FIXED

[Eric Seidel (no email)]

fast/canvas/webgl/drawArraysOutOfBounds.html failed on Leopard Commit Bot

https://bugs.webkit.org/show_bug.cgi?id=37064#c3

/tmp/layout-test-results/fast/canvas/webgl/drawArraysOutOfBounds-actual.txt

1313 PASS context.drawArrays(context.TRIANGLES, -1, 1) threw exception GL error 1282 in drawArrays.
1414 
1515 Test buffer with 3 float vectors
16  PASS context.drawArrays(context.TRIANGLES, 0, 3) is undefined
 16 FAIL context.drawArrays(context.TRIANGLES, 0, 3) should be undefined. Threw exception GL error 1286 in drawArrays
1717 PASS context.drawArrays(context.TRIANGLES, 3, 2) threw exception GL error 1282 in drawArrays.
1818 PASS context.drawArrays(context.TRIANGLES, 0, 10000) threw exception GL error 1282 in drawArrays.
1919 PASS context.drawArrays(context.TRIANGLES, 0, 10000000000000) threw exception GL error 1282 in drawArrays.

Probably more/same driver bugs.  This may need to be duped against other recent webgl failures.

[Zhenyao Mo]

Created attachment 52768 [details]
patch

We turn off stencil/antialias by default for tests that create canvas dynamically.  This should tremendously reduce the probability of test failure, hopefully next to zero.

The true cause of these failures are very likely a driver bug in Mac where multi-threading competition is on.  We could turn them on again when this driver bug is fixed.

We filed another bug for this driver bug: https://bugs.webkit.org/show_bug.cgi?id=36972

[Zhenyao Mo]

Created attachment 52770 [details]
revised patch: fixed a typo

[Eric Seidel (no email)]

Comment on attachment 52770 [details]
revised patch: fixed a typo

The JS change need a FIXME comment next to it with a link to one of these bugs.

Also, why not just do if (!attrs) instead of checking for typeof == undefined?

[Zhenyao Mo]

(In reply to comment #3)
> (From update of attachment 52770 [details])
> The JS change need a FIXME comment next to it with a link to one of these bugs.
> 

I'll add it.  Thanks.

> Also, why not just do if (!attrs) instead of checking for typeof == undefined?

The reason for using typeof==undefined instead of (!attrs) is the latter could be a intended null/0 input, whereas the former always indicates a default value situation.

[Zhenyao Mo]

Created attachment 52779 [details]
revised patch: add a FIXME comment

[Kenneth Russell]

(In reply to comment #5)
> Created an attachment (id=52779) [details]
> revised patch: add a FIXME comment

Patch looks good. Please update the synopsis of this bug to match that in the ChangeLog.

[WebKit Commit Bot]

Comment on attachment 52779 [details]
revised patch: add a FIXME comment

Clearing flags on attachment: 52779

Committed r57234: <http://trac.webkit.org/changeset/57234>

[WebKit Commit Bot]

All reviewed patches have been landed.  Closing bug.