483886 - static analysis incorrectly requires JS_REQUIRES_STACK

Status: RESOLVED WORKSFORME

(Core :: JavaScript Engine, defect)

Comment 1

[Andreas Gal :gal]

Applying this patch makes static analysis fail, but shouldn't.

http://hg.mozilla.org/tracemonkey/rev/186ae511d5f2

Comment 2

[Andreas Gal :gal]

http://tinderbox.mozilla.org/showlog.cgi?log=TraceMonkey/1237330299.1237330339.25955.gz&fulltext=1

Comment 3

[Benjamin Smedberg]

It looks like the analysis is correct: you're calling guard(), which requires a stack, from guardNotGlobalObject which doesn't have a stack.

Comment 4

[Brendan Eich [:brendan]]

(In reply to comment #3)
> It looks like the analysis is correct: you're calling guard(), which requires a
> stack, from guardNotGlobalObject which doesn't have a stack.

Why do we have to hand-annotate the transitive closure of calls-JS_REQUIRES_STACK? I recall being told we did not need to do anything that computers can infer for us. If a method uses cx->fp, we should hand-annotate. If a method calls a JS_REQUIRES_STACK method, no annotation should be required.

/be

Comment 5

[Benjamin Smedberg]

All the automatic analyses are function-local. They can only operate on the declared signature of the functions that are being called. So any method that calls a JS_REQUIRES_STACK method must either

1) be annotated JS_REQUIRES_STACK
2) somewhere in the body of the method, call a function which is JS_FORCES_STACK, so that you know you have safely bailed off trace.

Comment 6

[Luke Wagner [:luke]]

JS_REQUIRES_STACK has been removed.