475091 - JSCodeSpec cleanup

Status: RESOLVED INCOMPLETE

(Core :: JavaScript Engine, defect)

Description

[Mike Shaver (:shaver emeritus)]

https://bugzilla.mozilla.org/show_bug.cgi?id=465443#c28 ,
https://bugzilla.mozilla.org/show_bug.cgi?id=473282#c10 ,
passim.

Comment 1

[Jeff Walden [:Waldo]]

We should also add inlines to get operands and operand sizes, since right now (until I fix bug 473096) e.g. JSOP_NEWARRAY has a handful of different places that encode implicit dependencies on JOF_UINT24 (a 3 trailing argument to js_EmitN, js_GetVariableStackUses, the writing of the count into the bytecode stream, the reading of it from the bytecode stream in the interpreter, &c.).

Comment 2

[Jeff Walden [:Waldo]]

https://bugzilla.mozilla.org/show_bug.cgi?id=482743#c16

although I'd prefer |size_t JSCodeSpec.stackUses(jsbytecode* pc)| to what was suggested there.

Comment 3

[Nicolas B. Pierron [:nbp]]

Can you clarify what this bug means?

From what I understand, the goal sounds to be that we want to abstract the load of the operands / meta-data of any bytecode.  In which case this bug is far from being fixed.

Comment 4

[Jeff Walden [:Waldo]]

That sounds roughly right.  Properly making all the changes implied here would be a very substantial refactoring, tho.  So it seems somewhat more likely that we'll switch to AST walking than that we'll actually make any of the changes considered here.