[subixel] FrameView's converTo/FromRenderer methods should not pixel snap transforms

[Levi Weintraub]

Pixel snapping these methods can lead to repaint issues, as the resulting values are sent through repaint code that's aware of pixel snapping. Unfortunately, there doesn't seem to be a way to repeat this without a plugin redrawing itself. http://news.nana10.co.il/Article/?ArticleID=930551 demonstrates the issue (see http://code.google.com/p/chromium/issues/detail?id=156528 for details).

This is a regression introduced by me in http://trac.webkit.org/changeset/131111

[Levi Weintraub]

Created attachment 171887 [details]
Patch

[Emil A Eklund]

Comment on attachment 171887 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=171887&action=review

> Source/WebCore/page/FrameView.cpp:-3488
> -    IntRect rect = renderer->localToAbsoluteQuad(FloatRect(rendererRect), SnapOffsetForTransforms).enclosingBoundingBox();

This change restores the logic we had prior to r131111 for these methods, correct?

[Levi Weintraub]

Comment on attachment 171887 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=171887&action=review

>> Source/WebCore/page/FrameView.cpp:-3488
>> -    IntRect rect = renderer->localToAbsoluteQuad(FloatRect(rendererRect), SnapOffsetForTransforms).enclosingBoundingBox();
> 
> This change restores the logic we had prior to r131111 for these methods, correct?

That's correct.

[Darin Adler]

Comment on attachment 171887 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=171887&action=review

The comments make it sound like these fixes are specific to plug-ins, yet the code changes are in non-plug-in-specific functions.

Are these code paths really unused outside of the plug-in world?

We have test plug-ins and tests that use those plug-ins. Is it really impossible to cover this with a regression test?

If so, can you check in some test cases to be run manually?

> Source/WebCore/ChangeLog:3
> +        [subixel] FrameView's converTo/FromRenderer methods should not pixel snap transforms

Missing p in subpixel?

[WebKit Review Bot]

Comment on attachment 171887 [details]
Patch

Attachment 171887 [details] did not pass chromium-ews (chromium-xvfb):
Output: http://queues.webkit.org/results/14686591

New failing tests:
fast/repaint/repaint-during-scroll-with-zoom.html
fast/sub-pixel/sub-pixel-iframe-copy-on-scroll.html

[Levi Weintraub]

I found a better fix for this.