This is totally asymptomatic as far as I can tell even though it's a violation of IR semantics. It just sort of "works" because this only arises if you already had a transition watchpoint for that structure, which usually ends up being good enough since watchpoints invalidate the whole function.
Created attachment 216005 [details] the patch
Comment on attachment 216005 [details] the patch r=me The phrase "have structures" does not communicate this detail to me. I think by "have" you mean "speculated about, in a way which is invalid after an operation that clobbers the world". Maybe a better name would be "hasStructureSpeculations" or "hasStructuresToClobber" or "hasClobberableStructures".
(In reply to comment #2) > (From update of attachment 216005 [details]) > r=me > > The phrase "have structures" does not communicate this detail to me. I think by "have" you mean "speculated about, in a way which is invalid after an operation that clobbers the world". Maybe a better name would be "hasStructureSpeculations" or "hasStructuresToClobber" or "hasClobberableStructures". Fair point. I'll try to think of a better name.
Landed in http://trac.webkit.org/changeset/158680