[Win] Layout Test fast/canvas/canvas-path-addPath.html is failing

• Component: Layout and Rendering
• Status: RESOLVED FIXED

[Brent Fulgham]

The following layout test is failing on Windows.

fast/canvas/canvas-path-addPath.html

Probable cause:

This is crashing because we are attempting to add a path to itself in Path::addPath. The bug is here:

void Path::addPath(const Path& path, const AffineTransform& transform)
{
    if (!path.platformPath())
        return;

    if (!transform.isInvertible())
        return;

    CGAffineTransform transformCG = transform;
    // CG doesn't allow adding a path to itself. Optimize for the common case
    // and copy the path for the self referencing case.
    if (ensurePlatformPath() != path.platformPath()) {
        CGPathAddPath(ensurePlatformPath(), &transformCG, path.platformPath());
        return;
    }
    CGPathRef pathCopy = CGPathCreateCopy(path.platformPath());
    CGPathAddPath(ensurePlatformPath(), &transformCG, path.platformPath());
    CGPathRelease(pathCopy);
}

Note that even though we are creating the pathCopy, we are not using it. Instead, we use the original path (violating the contract of the CG API).

I'm not sure why this code seems to work on Mac, but it crashes on Windows.

[Brent Fulgham]

<rdar://problem/19428865>

[Brent Fulgham]

Created attachment 244359 [details]
Patch

[Brent Fulgham]

Committed r178186: <http://trac.webkit.org/changeset/178186>

[Chris Dumez]

May have caused 1 test failure on the bots (may be a simple rebaseline):
https://build.webkit.org/results/Apple%20MountainLion%20Debug%20WK2%20(Tests)/r178186%20(20724)/results.html

[Brent Fulgham]

(In reply to comment #4)
> May have caused 1 test failure on the bots (may be a simple rebaseline):
> https://build.webkit.org/results/
> Apple%20MountainLion%20Debug%20WK2%20(Tests)/r178186%20(20724)/results.html

Interesting. Looking at the history for this test on our platform (<http://webkit-test-results.appspot.com/dashboards/flakiness_dashboard.html#showAllRuns=true&tests=css3%2Fbackground%2Fbackground-repeat-space-content.html>), it looks as if a change between r178181 and r178183 introduced the slight image difference. (See http://trac.webkit.org/log/?verbose=on&rev=178183&stop_rev=178181>)

Looking at the whole set of bots, it looks like r178183 is the change that triggered the image difference.

[Brent Fulgham]

(In reply to comment #4)
> May have caused 1 test failure on the bots (may be a simple rebaseline):
> https://build.webkit.org/results/
> Apple%20MountainLion%20Debug%20WK2%20(Tests)/r178186%20(20724)/results.html

I think it was Bug 140298 that introduced that small image difference. I've updated the bug with some notes.