944946 - Use accessor methods for script->code and script->length

Status: RESOLVED FIXED

(Core :: JavaScript Engine, defect)

Description

[Brian Hackett [Laid off!]]

Attachment: patch

To mark script->code and script->length accesses as threadsafe when used during Ion compilation, these members need to be accessed only via methods.  The attached patch makes this change, and also adds some helper methods to capture the great majority of uses of these members: script->codeEnd(), script->containsPC(pc), script->pcToOffset(pc) and script->offsetToPC(offset).

Comment 1

[Jan de Mooij [:jandem]]

Comment on attachment 8340707 [details] [diff] [review]
patch

Review of attachment 8340707 [details] [diff] [review]:
-----------------------------------------------------------------

Nice!

::: js/src/jit/CompileInfo.h
@@ +100,3 @@
>      }
>      jsbytecode *limitPC() const {
> +        return script_->code() + script_->length();

Nit: script_->codeEnd()

Comment 2

[Jan de Mooij [:jandem]]

Comment on attachment 8340707 [details] [diff] [review]
patch

Review of attachment 8340707 [details] [diff] [review]:
-----------------------------------------------------------------

::: js/src/jsopcode.cpp
@@ +288,2 @@
>          JSOp op = JSOp(*pc);
> +        jsbytecode *next = GetNextPc(pc);

I just ran into a problem with GetNextPc and it made me think of this line. Unlike GetBytecodeLength, GetNextPc doesn't handle variable-length ops. So change it from

    return pc + js_CodeSpec[JSOp(*pc)].length;

to

    return pc + GetBytecodeLength(pc);

So that it doesn't iloop when there's a TABLESWITCH or something (CodeSpec length == -1).

Comment 3

[Brian Hackett [Laid off!]]

https://hg.mozilla.org/integration/mozilla-inbound/rev/d0bdb5b5dbd6

Comment 4

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/mozilla-central/rev/d0bdb5b5dbd6