932276 - Compute static scope extents at compile-time and store them in JSScript

Status: RESOLVED FIXED

(Core :: JavaScript Engine, defect)

Description

[Andy Wingo [:wingo]]

The bytecode emitter should record the extent of block scopes at compile-time and store that in the script object.

This will allow computation of the block chain corresponding to a PC solely based on the PC, without abstract interpretation of the bytecode.  In the case of lexical blocks with unaliased locals, this will remove the need for ENTERBLOCK / LEAVEBLOCK instructions.  It also removes the need for SRC_HIDDEN annotations for nonlocal block exits -- the last use of SRC_HIDDEN after bug 932180.

Comment 1

[Andy Wingo [:wingo]]

Attachment: Reserve space in JSScript for an optional block scope array (1/3)

Comment 2

[Andy Wingo [:wingo]]

Attachment: Bytecode emitter records static scope extents (2/3)

Comment 3

[Andy Wingo [:wingo]]

Attachment: Reimplement GetBlockChainAtPC using JSScript::blockScopes() (3/3)

Comment 4

[Jason Orendorff [:jorendorff]]

Comment on attachment 823986 [details] [diff] [review]
Bytecode emitter records static scope extents (2/3)

Review of attachment 823986 [details] [diff] [review]:
-----------------------------------------------------------------

::: js/src/frontend/BytecodeEmitter.cpp
@@ +786,5 @@
>  
>  static bool
> +EmitInternedObjectOp(ExclusiveContext *cx, uint32_t index, JSOp op, BytecodeEmitter *bce)
> +{
> +    JS_ASSERT(JOF_OPTYPE(op) == JOF_OBJECT);

Can this also assert that index is in range?

@@ +1087,5 @@
> +    JS_ASSERT(stmt->isBlockScope);
> +    JS_ASSERT(stmt->blockScopeIndex == bce->blockScopeList.length() - 1);
> +    JS_ASSERT(bce->blockScopeList.list[stmt->blockScopeIndex].length == 0);
> +    uint32_t scopeObjectIndex = bce->blockScopeList.list[stmt->blockScopeIndex].index;
> +    JS_ASSERT(scopeObjectIndex == bce->objectList.length - 1);

These assertions seem to explain the timing of EmitEnterBlock relative to the initialization of all these other bits of the script, which is nice. Can we also assert that pn->pn_objbox and scopeObjectIndex refer to the same thing?

Comment 5

[Andy Wingo [:wingo]]

Attachment: Reimplement GetBlockChainAtPC using JSScript::blockScopes() (3/3) v2

Comment 6

[Andy Wingo [:wingo]]

Comment on attachment 830704 [details] [diff] [review]
Reimplement GetBlockChainAtPC using JSScript::blockScopes() (3/3) v2

Thanks for the review, nits addressed.

Comment 7

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/integration/mozilla-inbound/rev/a8884e90aace
https://hg.mozilla.org/integration/mozilla-inbound/rev/a4ba9af376f3
https://hg.mozilla.org/integration/mozilla-inbound/rev/d3091d10dd64

Comment 8

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/mozilla-central/rev/a8884e90aace
https://hg.mozilla.org/mozilla-central/rev/a4ba9af376f3
https://hg.mozilla.org/mozilla-central/rev/d3091d10dd64