803837 - Clearing object properties by setting property

Status: NEW

(Core :: JavaScript Engine, defect)

Description

[Maks V. Sidorenko]

User Agent: Mozilla/5.0 (Windows NT 6.1; WOW64; rv:16.0) Gecko/20100101 Firefox/16.0
Build ID: 20121010144125

Steps to reproduce:

Greetings!

Code like this doesn't work properly on FF16:

/************** code start **********/
function setValue(ob)
{
    ob['a' + 555] = 2;
}

function cl()
{
    this.bInited = this.Init();
}

cl.prototype.Init = function()
{
    this.a = 1;
    setValue(this);
    return true;
}

var e = new cl();
alert(e.a);

/************** code end **********/

After line "ob['a' + 555] = 2;" all object properties are cleared. If we change this on ob['a555'] = 2 everything is ok. If we remove 'this.bInited = ' - everything is also ok.

In FF15 and other browsers this code worked as it should.



Actual results:

we recieve 'undefined'


Expected results:

we should recieve '1'

Comment 1

[Nickolay_Ponomarev]

Confirming on 16.0.2. Works fine on Nightly, though.

Couldn't find the bug about this, moving to Core since I'm curious to see it, and it seems serious enough to warrant fixing ASAP. (Guessing JS engine, although I don't have a JS shell from the 16.x branch to check).

(Note: the problem cannot be reproduced from scratchpad, I ran this as a top-level <script> in an html file to reproduce).

Comment 2

[Nickolay_Ponomarev]

Attachment: testcase to run from browser, slightly simplified from the reporter's

Comment 3

[Boris Zbarsky [:bzbarsky]]

Scratchpad may well not be jitted or something...

Nickolay, can you find a fix range?

Comment 4

[Nickolay_Ponomarev]

bz, sorry, don't count on me. I only had time yesterday.

Comment 5

[Boris Zbarsky [:bzbarsky]]

Not a problem at all.

This got fixed back in July: http://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=045c11dd41a6&tochange=462106f027af

Given that, this looks like this probably got fixed by the fix for bug 646599... but only because after that change "ob['a' + 555] = 2" and "ob['a555'] = 2" are treated identically.  So the underlying bug might still be there.

Comment 6

[Boris Zbarsky [:bzbarsky]]

Tom, any idea whether this is really fixed or just masked?

Comment 7

[Maks V. Sidorenko]

FF 17.0 - fixed
Thanks!

Comment 8

[Tom S [:evilpie]]

I am pretty sure this bug should not have changed anything observable like that. cc'ing Jeff.

Comment 9

[Jeff Walden [:Waldo]]

Yeah, seems like there's an underlying issue here.  Given the this-property-setting bits here, I suspect a TI issue, albeit with no hard evidence in support of the theory.