Upstream bug: http://code.google.com/p/chromium/issues/detail?id=150624 RenderLayerBacking::paintIntoLayer passes LayoutSize() into RenderLayer::paint, which can result in off-by-one paint errors with sub-pixel layout enabled. It should use the sub-pixel fraction from the root layer. I'm not sure of how, or if, this can be tested with a LayoutTest...
My previous comment wasn't entirely accurate. The issue comes to losing the correct sub-pixel offset by virtue of the layer bounds being zero when painting a layer in a vacuum. This is because convertToLayerCoords is a no-op when we're treating ourselves as our own root.
Created attachment 165860 [details] Fixes the one known occurrence of the bug This eliminates the issue that I've seen, but I'm not convinced this is the right thing for all cases. I'm continuing to investigate...
Created attachment 166324 [details] Patch
Comment on attachment 166324 [details] Patch View in context: https://bugs.webkit.org/attachment.cgi?id=166324&action=review Seems reasonable. I normally defer to Simon or James for Layer reviews, so you might want to poke one of them in #webkit to make sure. > Source/WebCore/ChangeLog:9 > + happens, a sub-RenderLayer is called to paint as its own root. Normally, we attempt to preserve a non-root render layer?
Committed r130322: <http://trac.webkit.org/changeset/130322>
Comment on attachment 166324 [details] Patch Clearing flags.