Right now, below where PICs are flushed there's a commented out call to "FlushMICs" which doesn't exist.

We only need to patch the shape guard for setgname and getgname, since the shapes could be changed during GC, and we may have an ABA problem. Exciting!

Same as before, minus debugging cruft. Whoops.

Comment on attachment 465429 [details] [diff] [review] Implement PurgeMICs(). Only a rare GC regenerates shapes (although we generate shapes like mad in Firefox, mainly due to XBL, currently). If nothing else in the MIC is a weak ref that might dangle unsafely, you shouldn't do this unless cx->runtime->gcRegenShapes is true. If other weak refs than shapes might dangle or replay badly, let's have a followup bug on invalidating them precisely (sorry if I missed one on file). /be

Thanks, Brendan. This is the same as the previous patch, except that we respect cx->runtime->gcRegenShapes at the PurgeMICs callsite.

Comment on attachment 465454 [details] [diff] [review] Previous + only purge MICs if necessary >+ /* MICs do not need to be purged unless shape guards are invalidated. */ >+ if (cx->runtime->gcRegenShapes) >+ mjit::ic::PurgeMICs(cx, script); This check already exists above, we don't purge PICs unless there's a shape-regenerating GC. >+ ic::MICInfo &mic = script->mics[i]; >+ switch (mic.kind) { >+ case ic::MICInfo::SET: >+ case ic::MICInfo::GET: >+ { Align { with the "c" of "case", and >+ /* Patch shape guard. */ ... this block to where the { was.