Created attachment 73964 [details] Test case See attached test case. The problem is that RenderImage::imageChanged() is called during attach() when an ancestor is attaching during recalcStyle(), and imageChanged() calls setNeedsStyleRecalc(). After recalcStyle() calls attach, it clears the childNeedsStyleRecalc state for itself, but not for all of its descendants leading down to the image. Now there is an island out of which setNeedsStyleRecalc() cannot propagate to the root.
<rdar://problem/8672000>
This bug was causing fast/block/float/015.html to fail intermittently, so I disabled it in <http://trac.webkit.org/projects/webkit/changeset/72064>.
Created attachment 74030 [details] Only defer intrinsic size compoutation if a style recalc is coming
Comment on attachment 74030 [details] Only defer intrinsic size compoutation if a style recalc is coming View in context: https://bugs.webkit.org/attachment.cgi?id=74030&action=review > WebCore/ChangeLog:8 > + https://bugs.webkit.org/show_bug.cgi?id=49579 > + > + Tests: fast/block/float/015.html I'd like to see some words here about what the problem was. I don't think the but title is quite enough.
Fixed in <http://trac.webkit.org/changeset/72135>.