616618 - Assigning to a const is unsoundly optimized away

Status: RESOLVED WORKSFORME

(Core :: JavaScript Engine, defect)

Description

[Jason Orendorff [:jorendorff]]

"use strict";
const k = 0;
k = 1;

This should throw an exception, but doesn't, because the assignment to k is optimized away in the bytecode compiler.

Does any later stage depend on this optimization? It seems like it can't be for performance, since we're talking about assignment to a const here. I guess I'll find out.

Comment 1

[Jason Orendorff [:jorendorff]]

I looked into this a bit. The question, of course, is what to do once you know you're assigning to a const.

We could deoptimize the entire enclosing script/function. The emitter would then emit a JOF_NAME opcode for the assignment, inc/dec, or delete. That would work correctly. To do it during the parser's first pass, we would need to touch two places (assignExpr and BindVarOrConst). Maybe it could be done during the analyzeFunctions pass with less hassle. (?)

The other approach is to do it in the emitter, but it's not safe to emit JOF_NAME ops in those cases, is it? So, what bytecode should we emit?

k = 1;
    one
    ???  /* opcode that behaves like setname */
    pop

k++;
    ???  /* opcode that behaves like nameinc */
    pop

delete k;
    ???  /* opcode that behaves like delname */
    pop

I'm not taking this.

Comment 2

[André Bargull [:anba]]

No longer reproducible - resolving as WFM.