Created attachment 553340 [details] [diff] [review]
Asserting regs.sp == regs.fp->base() + js_ReconstructStackDepth in js_InternalThrow revealed a few cases where sp doesn't match pc. I suspect this is causing the slow trickle of CrashIfInvalidSlot crashes (e.g. http://crash-stats.mozilla.com/report/index/bb39c9bf-bb35-4567-bbfe-64c742110809).
Hrm, I meant for this to go out with aurora but I doubt there will be an m-i -> m-c merge in the next 2 hours. I guess I'll land on m-c and hopefully 3-way merge should avoid any merge pain. Apologies to the m-i sheriff for the bother.