1188124 - Use rootKind to find the right PersistentRooted list

Status: RESOLVED FIXED

(Core :: JavaScript: GC, defect)

Description

[Terrence Cole [:terrence]]

Attachment: use_rootKind_to_select_persistentrooted_list-v0.diff

At least in the PersistentRooted constructor. Finish and Trace both want a loop, which we will get to later. In the meantime we assert that only Traced and Finished lists are used.

Comment 1

[Terrence Cole [:terrence]]

Comment on attachment 8639534 [details] [diff] [review]
use_rootKind_to_select_persistentrooted_list-v0.diff

Steve is on PTO this week and next.

Comment 2

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

Comment on attachment 8639534 [details] [diff] [review]
use_rootKind_to_select_persistentrooted_list-v0.diff

Review of attachment 8639534 [details] [diff] [review]:
-----------------------------------------------------------------

I'm not gone quite yet. I'll be around today (though I'll be shifting my hours a little.)

::: js/public/RootingAPI.h
@@ +994,5 @@
>      void registerWithRootLists(js::RootLists& roots) {
>          MOZ_ASSERT(!initialized());
> +        js::ThingRootKind kind = js::RootKind<T>::rootKind();
> +        reinterpret_cast<mozilla::LinkedList<JS::PersistentRooted<T>>&>(roots.heapRoots_[kind])
> +            .insertBack(this);

Wow, that's an awful looking cast!

I almost wonder if it should be

  roots.typedHeapRoot<T>(kind).insertBack(this);

but if this is going to be the only use, I'd rather not hide what's going on.

Comment 3

[Terrence Cole [:terrence]]

(In reply to Steve Fink [:sfink, :s:] from comment #2)
> Comment on attachment 8639534 [details] [diff] [review]
> use_rootKind_to_select_persistentrooted_list-v0.diff
> 
> Review of attachment 8639534 [details] [diff] [review]:
> -----------------------------------------------------------------
> 
> I'm not gone quite yet. I'll be around today (though I'll be shifting my
> hours a little.)
> 
> ::: js/public/RootingAPI.h
> @@ +994,5 @@
> >      void registerWithRootLists(js::RootLists& roots) {
> >          MOZ_ASSERT(!initialized());
> > +        js::ThingRootKind kind = js::RootKind<T>::rootKind();
> > +        reinterpret_cast<mozilla::LinkedList<JS::PersistentRooted<T>>&>(roots.heapRoots_[kind])
> > +            .insertBack(this);
> 
> Wow, that's an awful looking cast!

Turns out it breaks gcc's strict alias analysis because we're casting from void* to multiple unrelated T. We actually need to cast this to PersistentRooted<void*>*, which is a ton tamer.

Comment 4

[Pulsebot]

https://hg.mozilla.org/integration/mozilla-inbound/rev/ca8eb562952d

Comment 5

[Wes Kocher (:KWierso) (Not reading bugmail; email directly if needed)]

https://hg.mozilla.org/mozilla-central/rev/ca8eb562952d