Bug 52341 – fast/canvas/canvas-large-dimensions.html (introduced in r75648) is failing on all platforms except Snow Leopard

• Status: RESOLVED
• Resolution: FIXED
• Component: Canvas

[Justin Schuh]

I don't know exactly what change, but this seems like the culprit: http://trac.webkit.org/changeset/75639

Snow Leopard appears to be the only platform that's passing.

[Hayato Ito]

We disabled this test for chromium and chromium-gpu:

https://trac.webkit.org/changeset/75666
https://trac.webkit.org/changeset/75678

[Ryosuke Niwa]

The test was added in http://trac.webkit.org/changeset/75648.  It can't possibly be failing on r75639.

[Hayato Ito]

You are right. The comments in the test_expectations files are wrong.
r57648 is correct.

(In reply to comment #2)
> The test was added in http://trac.webkit.org/changeset/75648.  It can't possibly be failing on r75639.

[Hayato Ito]

s/r57648/r75648/.

(In reply to comment #3)
> You are right. The comments in the test_expectations files are wrong.
> r57648 is correct.
> 
> (In reply to comment #2)
> > The test was added in http://trac.webkit.org/changeset/75648.  It can't possibly be failing on r75639.

[Ryosuke Niwa]

Regression is P1.  Currently, this is the only test failing on GTK 64-bit debug, GTK 32-bit release, Leopard Intel Release, and Leopard Intel Debug and this bug needs to be fixed ASAP.

[Ryosuke Niwa]

Oops, thanks for fixing the typo in the summary, Simon.

[Justin Schuh]

(In reply to comment #2)
> The test was added in http://trac.webkit.org/changeset/75648.  It can't possibly be failing on r75639.

Apologies for the confusion. I was in a hurry and must have grabbed the wrong version number.

[Stephen White]

(In reply to comment #7)
> (In reply to comment #2)
> > The test was added in http://trac.webkit.org/changeset/75648.  It can't possibly be failing on r75639.
> 
> Apologies for the confusion. I was in a hurry and must have grabbed the wrong version number.

Skia has a hard limit in each dimension right now, I think it's 32K, to avoid overflow inside skia.  I think wjmaclean implemented it.  I don't think we can pass this test as-written, since 100K is too big.

reed@ just told me that there's a #define to allow larger bitmaps inside skia, but I don't know if this applies to canvases too.

[Matthew Delaney]

The two checks inside the test for making a canvas of width 100k and 1000k are what fails on all the other platforms. I'm going to remove those from the test and add in one that is ~32k that all the platforms should be able to handle. Then we (still) have a test that stresses absurdly large sized canvases but makes sense for all platforms. As a follow up, I'll create a bugzilla to track the discussion of sizes larger than this.

[Matthew Delaney]

Created attachment 78841 [details]
Patch

[James Robinson]

Comment on attachment 78841 [details]
Patch

Looks good

[Matthew Delaney]

<http://trac.webkit.org/changeset/75732>

[Jeff Timanus]

Created attachment 88078 [details]
Addition of support for falling back to software rendering when creating a canvas that is too large for the GPU device.

[WebKit Review Bot]

Attachment 88078 [details] did not pass style-queue:

Failed to run "['Tools/Scripts/check-webkit-style']" exit_code: 1

Source/WebCore/html/canvas/CanvasRenderingContext2D.cpp:142:  One line control clauses should not use braces.  [whitespace/braces] [4]
Total errors found: 1 in 8 files


If any of these errors are false positives, please file a bug against check-webkit-style.

[James Robinson]

Please post this patch on a new, separate bug.

[James Robinson]

Comment on attachment 88078 [details]
Addition of support for falling back to software rendering when creating a canvas that is too large for the GPU device.

Not valid for this bug (please file another one for the chromium-specific issue)

[Jeff Timanus]

Created attachment 88655 [details]
Patch