Bug 80560 – [CG] Canvas lineDashOffset does not handle negative numbers correctly

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

Created attachment 130748 [details]
demonstration of problem

webkitLineDashOffset is sometimes not handled correctly -- or at least, is handled differently from Firefox's mozDashOffset, and I like how mozDashOffset is handled better. The attachment demonstrates the problem. Open it in Firefox 10 to see the desired rendering, then open it in WebKit to compare. I am currently using WebKit r110098.

I have only observed the problem when webkitLineDashOffset is negative and its absolute value is greater than the sum of all dash lengths. Uncommenting the commented-out line in the attachment works around the problem by setting webkitLineDashOffset to its modulo.

[Dirk Schulze]

Created attachment 228507 [details]
Patch

[Dean Jackson]

Comment on attachment 228507 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=228507&action=review

> LayoutTests/ChangeLog:8
> +        Test correct rendering of negative offset fot Canvas dash arrays.

fot!

[Dirk Schulze]

Created attachment 228540 [details]
Patch for laning

[WebKit Commit Bot]

Comment on attachment 228540 [details]
Patch for laning

Clearing flags on attachment: 228540

Committed r166746: <http://trac.webkit.org/changeset/166746>

[WebKit Commit Bot]

All reviewed patches have been landed.  Closing bug.

[Darin Adler]

Comment on attachment 228540 [details]
Patch for laning

View in context: https://bugs.webkit.org/attachment.cgi?id=228540&action=review

> Source/WebCore/platform/graphics/cg/GraphicsContextCG.cpp:1189
> +        for (size_t i = 0; i < dashes.size(); ++i)
> +            length += static_cast<float>(dashes[i]);

New style for loop is better here:

    for (auto& dash : dashes)
        length += static_cast<float>(dash);

Also wondering why the static_cast<float> is helpful.

> Source/WebCore/platform/graphics/cg/GraphicsContextCG.cpp:1191
> +            dashOffset = fmod(dashOffset, length) + length;

Because we use fmod here instead of fmodf, I believe this involves two conversions from float to double and one conversion from double to float.

Please change to use fmodf, unless I am mistaken about that.