802318 - GC: remove skip-roots from jsatom.cpp

Status: RESOLVED FIXED

(Core :: JavaScript Engine, defect)

Description

[Terrence Cole [:terrence]]

At the moment, AtomizeInline accepts a jschar* which may point into an inline string.  Forcing this to take a StableCharPtr would, quite unacceptably, force all atoms to be stable strings.  Instead, we should reorganize this as following: (1) AtomizeChars should take a StableCharPtr or callers should get rewritten to use AtomizeString; (2) AtomizeString should be rewritten such that it only needs to extract characters after allocating the new Atom, allowing even short/inline strings to work here.

Comment 1

[Terrence Cole [:terrence]]

Attachment: v0

This is largely code motion to allow the Copy and TakeOwnership to specialize on the stableness and inlinedness of the string. A nice knock-on effect from this is that AtomizeString no longer needs to stabilize inline strings.

Comment 2

[Bill McCloskey [inactive unless it's an emergency] (:billm)]

Comment on attachment 698829 [details] [diff] [review]
v0

Review of attachment 698829 [details] [diff] [review]:
-----------------------------------------------------------------

::: js/src/jsatom.cpp
@@ +242,4 @@
>   */
>  JS_ALWAYS_INLINE
> +static UnrootedAtom
> +AtomizeInlineTakeOwnership(JSContext *cx, StableCharPtr tbchars, size_t length,

How about AtomizeAndTakeOwnership?

@@ +249,3 @@
>  
> +    if (UnrootedAtom s = cx->runtime->staticStrings.lookup(tbchars.get(), length)) {
> +        JS_free(cx, (void*)tbchars.get());

Please just use js_free here and for the other JS_free calls below.

@@ +255,5 @@
> +    /*
> +     * If a GC occurs at js_NewStringCopy then |p| will still have the correct
> +     * hash, allowing us to avoid rehashing it. Even though the hash is
> +     * unchanged, we need to re-lookup the table position because a last-ditch
> +     * GC will unlock the atomsCompartment and potentially free some table

Please remove the phrase "unlock the atoms compartment and" since I don't even know what it means.

@@ +288,5 @@
> +}
> +
> +JS_ALWAYS_INLINE
> +static UnrootedAtom
> +AtomizeInlineCopy(JSContext *cx, const jschar *tbchars, size_t length, InternBehavior ib)

How about AtomAndCopyStableChars?

Also, please put a comment above this function saying that tbchars should not point into an inline JSString.

@@ +297,5 @@
> +    /*
> +     * If a GC occurs at js_NewStringCopy then |p| will still have the correct
> +     * hash, allowing us to avoid rehashing it. Even though the hash is
> +     * unchanged, we need to re-lookup the table position because a last-ditch
> +     * GC will unlock the atomsCompartment and potentially free some table

Same issue with the comment.

@@ +319,4 @@
>  
>      /*
> +     * We re-generate the Lookup from |flat| here because |tbchars| may be
> +     * inline in a moved JSString and is thus not safe to use after

Please revise this comment. As we discussed, this function won't work if tbchars points into an inline JSString.

@@ +350,5 @@
>          p->setTagged(bool(ib));
>          return &atom;
>      }
>  
> +    JSLinearString *linear = str->ensureLinear(cx);

As we discussed, this needs to remain ensureStable.

Comment 3

[Terrence Cole [:terrence]]

Green try run at:
https://tbpl.mozilla.org/?tree=Try&rev=bc8efe416481

Pushed at:
https://hg.mozilla.org/integration/mozilla-inbound/rev/c420b26e7e84

Comment 4

[Ed Morley [:emorley]]

https://hg.mozilla.org/mozilla-central/rev/c420b26e7e84