(Even after the infallible-malloc hack in bug 608776) callgrind shows 30% of date-format-xparb is spent below stubs::Add. 8% is spent in stubs::Add itself. With bug 608776 taking the malloc out of js_ConcatStrings (usually), we should be able to write a stub-call-free fast path which either creates a rope node or fills a short string. Because xparb uses String objects (costing, even after bug 603017, 5% in js::DefaultValue), this fast path should also include an IC for 'toString' on objects.
I think bug 609440 is going ahead instead of bug 608776. The net effect w.r.t. this bug should be the same, though.
Throwing back in the pool.
un-good-first-bugging. With IonMonkey development full steam, this should either be WONTFIX'd or re-targeted towards IonMonkey.
Too late, JM was removed.