During layout we should never access stale geometry of other boxes (containing block etc).
Created attachment 339664 [details] Patch
Created attachment 339669 [details] Patch
Comment on attachment 339669 [details] Patch View in context: https://bugs.webkit.org/attachment.cgi?id=339669&action=review > Source/WebCore/layout/displaytree/DisplayBox.h:138 > +#if !ASSERT_DISABLED > + bool m_hasValidTop : 1; > + bool m_hasValidLeft : 1; > + bool m_hasValidWidth : 1; > + bool m_hasValidHeight : 1; > + bool m_hasValidMargin : 1; > + bool m_hasValidBorder : 1; > + bool m_hasValidPadding : 1; > +#endif I would just use bools. There is no need to optimize memory use of the debug build. Do all these really need separate flags?
(In reply to Antti Koivisto from comment #3) > Comment on attachment 339669 [details] > Patch > > View in context: > https://bugs.webkit.org/attachment.cgi?id=339669&action=review > > > Source/WebCore/layout/displaytree/DisplayBox.h:138 > > +#if !ASSERT_DISABLED > > + bool m_hasValidTop : 1; > > + bool m_hasValidLeft : 1; > > + bool m_hasValidWidth : 1; > > + bool m_hasValidHeight : 1; > > + bool m_hasValidMargin : 1; > > + bool m_hasValidBorder : 1; > > + bool m_hasValidPadding : 1; > > +#endif > > I would just use bools. There is no need to optimize memory use of the debug > build. > > Do all these really need separate flags? For block context, I could definitely collapse top with left and the padding with border/margin since we compute them in one go, but other formatting contexts might differ. I guess padding/border/margin should always be collapsible.
Created attachment 339694 [details] Patch
Comment on attachment 339694 [details] Patch Clearing flags on attachment: 339694 Committed r231401: <https://trac.webkit.org/changeset/231401>
All reviewed patches have been landed. Closing bug.
<rdar://problem/40013012>