Created attachment 25021 [details] Testcase

Created attachment 25332 [details] Simple testcase with outline This bug has nothing to do with focus, or animations. There's a basic repaint bug with outlines.

The problem is that RenderBox::absoluteClippedOverflowRect() only looks at the outline size on the style of the element being laid out, but not any of its contents. We could either grovel through the child objects looking at outline size, or just use view->maximalOutlineSize().

I think it's going to have to be view->maximalOutlineSize() :-(

Created attachment 25338 [details] Patch, testcase, changelog Patch includes some new bases for repaint tests whose repaint rects alter with this change (because of focus rings)

Comment on attachment 25338 [details] Patch, testcase, changelog > + if (style()->outlineWidth() > 0 && style()->outlineSize() > maximalOutlineSize(PaintPhaseOutline)) > + static_cast<RenderView*>(document()->renderer())->setMaximalOutlineSize(style()->outlineSize()); Since you had to move this code, maybe there's a way to write a brief comment explaining the order dependency? I feel a little uncomfortable reviewing this with Hyatt on vacation. You should check with him afterwards just in case he thinks this will result in excessive painting. r=me

Yes, the over-repaint is a worry, especially since any focus ring makes setMaximalOutlineSize() non-zero. The alternatives would be: 1. Store outline dimensions on RenderBox, and explicitly account for them whem computing repaint rects (but they don't affect overflow, unlike shadows, so we can't just put them in the overflow sizes), or 2. Set a bit on RenderObject to say that it has outlined children, and only inflate by maxOutlineSize when the object or children have outline.

Committed r38678 M WebCore/ChangeLog M WebCore/rendering/RenderBox.cpp M LayoutTests/platform/mac/fast/repaint/outline-repaint-glitch-expected.checksum A LayoutTests/platform/mac/fast/repaint/outline-child-repaint-expected.checksum A LayoutTests/platform/mac/fast/repaint/outline-child-repaint-expected.png M LayoutTests/platform/mac/fast/repaint/4776765-expected.checksum A LayoutTests/platform/mac/fast/repaint/outline-child-repaint-expected.txt M LayoutTests/platform/mac/fast/repaint/delete-into-nested-block-expected.png M LayoutTests/platform/mac/fast/repaint/4776765-expected.png M LayoutTests/platform/mac/fast/repaint/delete-into-nested-block-expected.checksum M LayoutTests/platform/mac/fast/repaint/outline-repaint-glitch-expected.png M LayoutTests/ChangeLog A LayoutTests/fast/repaint/outline-child-repaint.html r38678 = 0ebe80da11f747b44e3ba9e89ae67bf891820c51 (trunk) I'll send mail to Hyatt to look at this change.

Maybe this should have used getAbsoluteRepaintRectWithOutline()?

*** Bug 14050 has been marked as a duplicate of this bug. ***