We shouldn't waste time painting backgrounds for things like images, for which it's easy to determine that the foreground fully obscures that background. This helps particularly in cases where the background might contain an animated "loading" image; that currently causes continual repaints, and it should not.
This is only going to work for JPEG initially: bug 65033.
Created attachment 101774 [details] Patch
Comment on attachment 101774 [details] Patch View in context: https://bugs.webkit.org/attachment.cgi?id=101774&action=review > Source/WebCore/ChangeLog:8 > + Some pages used animated loading GIFs as the background of <img>, Confused by the use of “used” in the past tense. > Source/WebCore/rendering/RenderImage.cpp:411 > + if (!image->isBitmapImage()) I think you need to null-check image.
http://trac.webkit.org/changeset/91628
This patch seems to have caused a crash on Chromium Linux: http://build.webkit.org/results/Chromium%20Linux%20Release%20(Tests)/r91628%20(21410)/results.html http://build.webkit.org/results/Chromium%20Linux%20Release%20(Flexbox)/r91628%20(1508)/results.html Given these are two different bots, I highly doubt that this is a flake. Unfortunately, these are release builds so we can't get stack trace out of them. There's also one failing test on Chromium Mac: http://build.webkit.org/results/Chromium%20Mac%20Release%20(Tests)/r91628%20(9716)/results.html But I'm not certain if this is a real regression or just a flakiness.
Do we have any more information on whether those crashes persisted?
(In reply to comment #6) > Do we have any more information on whether those crashes persisted? They have been crashing quite consistently: http://test-results.appspot.com/dashboards/flakiness_dashboard.html#tests=fast%2Fcanvas%2Fwebgl%2Fdrawingbuffer-test.html%2Cfast%2Fcanvas%2Fcanvas-bg-multiple-removal.html%2Cfast%2Fcanvas%2Fcanvas-as-image-incremental-repaint.html
Odd that it's just Linux that's crashing.
(In reply to comment #8) > Odd that it's just Linux that's crashing. It's possible that there's a cr-linux or skia bug; or that only cr-linux hit some specific timing.
Let's continue discussing the crashes in bug 65063.