712379 - JS Correctness: Different error messages with -m -a and with -m or interp

Status: RESOLVED WORKSFORME

(Core :: JavaScript Engine, defect)

Description

[Gary Kwong [:gkw] [:nth10sd] (NOT official MoCo now)]

"use strict";
for each(let z in [0, 0]) {
    try {
        for (s = 0;;) {
            t()
        }
    } catch (e) {
        print('' + e)
    }
}

On js opt shell on m-c changeset a0d32d0353e1 passing testcase as a CLI argument:

$ ./js-opt-64-mc-darwin w8511-cj-in.js 
ReferenceError: assignment to undeclared variable s
ReferenceError: assignment to undeclared variable s

$ ./js-opt-64-mc-darwin -m w8511-cj-in.js 
ReferenceError: assignment to undeclared variable s
ReferenceError: assignment to undeclared variable s

$ ./js-opt-64-mc-darwin -m -a w8511-cj-in.js 
ReferenceError: assignment to undeclared variable s
ReferenceError: t is not defined

we get different error messages with ( interp or -m ) and -m -a.

autoBisect shows this is probably related to the following changeset:

The first bad revision is:
changeset:   81356:39a0770a0541
user:        Brian Hackett
date:        Fri Nov 25 17:45:41 2011 -0500
summary:     Fix inverted test, improve test when marking objects with uncacheable prototypes, bug 703047.

Comment 1

[Jesse Ruderman]

WFM on mozilla-central. I now get "ReferenceError: assignment to undeclared variable s" in all execution modes.

The first good revision is:
changeset:   38344f96b3e3
user:        Luke Wagner
date:        Fri Oct 07 12:02:50 2011 -0700
summary:     Bug Bug 692274, part 4 - Rewrite parsing, emitting and decompiling of let to fix scoping properly (r=jorendorff)

Comment 2

[Gary Kwong [:gkw] [:nth10sd] (NOT official MoCo now)]

Bisections seem weird. How can the first good revision be *before* the first bad revision?

Comment 3

[Jesse Ruderman]

It isn't. 39a0770a0541 is an ancestor of 38344f96b3e3. The date on 38344f96b3e3 is just weird.