This came up for undefined on jsperf.com in bug 605486, but just doing all primitives is easy enough.

Comment on attachment 484715 [details] [diff] [review] Trace inc() for all primitive values, not just numbers. I used insImmD in the null case because that's what the number case used to do.... but should they both use insImmI? Or is that handled elsewhere?

Comment on attachment 484715 [details] [diff] [review] Trace inc() for all primitive values, not just numbers. insImmD is right, integers shouldn't leak into the LIR (doubles are promoted to integers in a forward pipeline).

Comment on attachment 484715 [details] [diff] [review] Trace inc() for all primitive values, not just numbers. Requesting approval. This is pretty safe, and is needed (barring site changes) for jsperf.com to trace.

Comment on attachment 484715 [details] [diff] [review] Trace inc() for all primitive values, not just numbers. >+ // XXXbz can we manage to do this for objects with an imacro? Or >+ // will that be unsafe for some of our callers? No need to ask and we avoid XXX comments now in favor of FIXME: bug NNNNNN comments. This one is completely fixable with imacrology, so please file a future bug. >+ if (!v.isPrimitive()) >+ RETURN_STOP("can only inc primitives"); Uber-nit: "can inc primitives only". Nice patch. bz, you need to do a JM patch too some time, get both decoder rings ;-). /be

(In reply to comment #5) > No need to ask and we avoid XXX comments now in favor of FIXME: bug NNNNNN Done. Bug 606071. > Uber-nit: "can inc primitives only". Done. > Nice patch. bz, you need to do a JM patch too some time I looked at the pic code some; got lost and gave up for the time being. ;)

Robert, want to either approve or deny so I can stop worrying about merging this? ;)

Pushed http://hg.mozilla.org/tracemonkey/rev/19f70f8c2b88