355913 - Inappropriate "reference to undefined property" strict warnings

Status: RESOLVED DUPLICATE of bug 355145

(Core :: JavaScript Engine, defect)

Description

[Simon Bünzli]

Regression introduced between the 2006093003 and 2006100103 branch nightlies.

Steps to reproduce:
1. Install the Console² extension ( http://forums.mozillazine.org/viewtopic.php?t=318102 )
2. Tools -> Error Console
3. Make sure that Options -> Report Strict Warnings is checked
4. Click on the All button

Actual results:
Warning: reference to undefined property gModes[aType]
Source file: chrome://console2/content/console2.js
Line: 79
 ----------
Warning: reference to undefined property gModes[aType]
Source file: chrome://console2/content/console2.js
Line: 577

Note that |aType in gModes == true| and the extension works as flawless as it did with the September nightlies.

Comment 1

[Mike Beltzner [:beltzner, not reading bugmail]]

Bogus error, not gonna block Firefox 2 on this. Please renominate for 1.8.1.1 only when you've hunted down the regression range.

Comment 2

[Simon Bünzli]

(In reply to comment #1)
The regression is very recent and _might_ have other unintended consequences.

Comment #0:
> Regression introduced between the 2006093003 and 2006100103 branch nightlies.

About a dozen JavaScript fixes fall into that range, half of them non-disclosed, so it's really up to Brendan to determine the exact problem.

Comment 3

[Brendan Eich [:brendan]]

Reduced testcase that fits in the js shell would be a big help.

/be

Comment 4

[Igor Bukanov]

Given that this is the second dup of bug 355145. Given that perhaps we can apply it now?


*** This bug has been marked as a duplicate of 355145 ***

Comment 5

[:Gavin Sharp [email: [e-mail]]]

(In reply to comment #4)
> Given that this is the second dup of bug 355145. Given that perhaps we can
> apply it now?

I don't understand how this (and bug 356129) are duplicates of bug 355145. They seem like exact duplicates of bug 355075, and as I understand the patch in bug 355145 is just a new approach to fixing bug 355075 without introducing an E4X related regression.

In other words: what caused bug 355075 to regress on the branch?

Comment 6

[Igor Bukanov]

A reduced test case for js shell that produces a warning in the strict mode:

var gModes = {	a: { Errors: true } };
for (var mode in gModes.a)
  gModes.a[mode] = true;

Comment 7

[Igor Bukanov]

Or even smaller test case that the patch from bug 355145 somehow fixes (I do not know at this moment why exactly):

var gModes = {	a: { } };
for (var mode in gModes.a);

So far it looks like that second property lookup affects bytecode calculations.

Comment 8

[Igor Bukanov]

I reopn the bug until it becomes clear why the patch from bug 355145 fixes this.

Comment 9

[Igor Bukanov]

Here is an explanation why is this a dup of 355145:

Now the branch got the fix for the bug 353165. But the fix do not call JS_GetMethodById any longer and rather calls OBJ_GET_PROPERTY. But that assumed that 355145 is fixed and OBJ_GET_PROPERTY does not warn about missing __iterator__. 

*** This bug has been marked as a duplicate of 355145 ***