841558 - GC: Don't exactly mark through tagged protos and other tagged things

Status: RESOLVED FIXED

(Core :: JavaScript Engine, defect)

Description

[Terrence Cole [:terrence]]

Attachment: v0

I hit this when working on GGC. This is the same test we have in place on the object pre-barrier and for the same reason: NULL tagged pointers.

Comment 1

[Steve Fink [:sfink] [:s:]]

Comment on attachment 714115 [details] [diff] [review]
v0

Review of attachment 714115 [details] [diff] [review]:
-----------------------------------------------------------------

::: js/src/gc/RootMarking.cpp
@@ +45,5 @@
>  static inline void
>  MarkExactStackRoot(JSTracer *trc, Rooted<void*> *rooter, ThingRootKind kind)
>  {
>      void **addr = (void **)rooter->address();
> +    if (uintptr_t(*addr) < 32)

Ugh. No problem with the implementation, but could this check and the ones for barriers in ObjectImpl-inl.h be put into a meaningfully-named macro? It's both cryptic and yet not worth a full comment at the same time. Hopefully there's something shorter than IS_NULL_POSSIBLY_TAGGED_POINTER(p).

Comment 2

[Terrence Cole [:terrence]]

https://hg.mozilla.org/integration/mozilla-inbound/rev/4ba0c687f0b4

Comment 3

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/mozilla-central/rev/4ba0c687f0b4