613153 - GMail broken with method JIT on

Status: RESOLVED FIXED

(Core :: JavaScript Engine, defect)

Description

[Olipro]

User-Agent:       Mozilla/5.0 (Windows NT 6.1; Win64; x64; rv:2.0b8pre) Gecko/20101118 Firefox/4.0b8pre
Build Identifier: "parent: 57846:579f63017921 tip Bug 611566" (from hg summary)

Just load up GMail and it'll immediately become evident; the reading pane completely fails to load, major javascript snafu I suspect.

Reproducible: Always

Steps to Reproduce:
1. go to gmail.google.com
2. login and ensure you're using their standard view, not basic HTML
3. look on in horror.

Comment 1

[Ed Morley [:emorley]]

Broken for me too using:
Mozilla/5.0 (Windows NT 6.1; WOW64; rv:2.0b8pre) Gecko/20101118 Firefox/4.0b8pre

Comment 2

[Ed Morley [:emorley]]

Ok, this is weird, if I restart in safe mode, gmail works again.

However in standard mode, if I disable every single addon and plugin manually (one by one using the disable button in the addons manager), clear the cache and restart, gmail is still broken.

Comment 3

[Ed Morley [:emorley]]

Attachment: Screenshot of gmail partial page load

Comment 4

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

wfm, Mozilla/5.0 (Macintosh; Intel Mac OS X 10.6; rv:2.0b8pre) Gecko/20101118 Firefox/4.0b8pre

Comment 5

[Paul O'Shannessy [:zpao] (not bugmail, email directly)]

Turning off method JIT fixes it. Moving.

So this only is happening with GMail for apps (mail.google.com/a/whatever.com) but not with a normal gmail account. If access is needed to one, I have a couple domains with these set up and can set up a new account.

Comment 6

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

Limi reports that it's also broken with standard Gmail, too:

http://img.skitch.com/20101118-xjbjehm319x3kdfdu2nu3xb94m.png

Confusingly, this doesn't seem to be happening to everyone. Possible that we're also seeing different content served to different users :(

Comment 7

[Paul O'Shannessy [:zpao] (not bugmail, email directly)]

Adding ?labs=0 to the URL fixes this for me, so one of those is the likely culprit. Trying to narrow down now.

Comment 8

[Olipro]

(In reply to comment #5)
> Turning off method JIT fixes it. Moving.
> 
> So this only is happening with GMail for apps (mail.google.com/a/whatever.com)
> but not with a normal gmail account. If access is needed to one, I have a
> couple domains with these set up and can set up a new account.

incorrect, I have the problem and I'm using GMail, not the Apps service.

I suspect something in the JIT has changed that has broken GMail.

If you try an hourly build of Firefox from around 17:00 yesterday, GMail functions somewhat but replying is broken (someone actually logged a bug about this and it was reported as fixed, I wonder if perhaps the "fix" actually broke it completely)

Comment 9

[Ed Morley [:emorley]]

(In reply to comment #8)
> someone actually logged a bug about this and it was reported as fixed,
> I wonder if perhaps the "fix" actually broke it completely)
Rob replied in that bug and said he thought it was unrelated:
https://bugzilla.mozilla.org/show_bug.cgi?id=612861#c34

Comment 10

[Paul O'Shannessy [:zpao] (not bugmail, email directly)]

(In reply to comment #8)
> (In reply to comment #5)
> > Turning off method JIT fixes it. Moving.
> > 
> > So this only is happening with GMail for apps (mail.google.com/a/whatever.com)
> > but not with a normal gmail account. If access is needed to one, I have a
> > couple domains with these set up and can set up a new account.
> 
> incorrect, I have the problem and I'm using GMail, not the Apps service.

Indeed. beltzner pointed this out as well, though the bug summary hasn't been updated (I didn't want to spam too much).

For me, Gmail breaks when I enable the "quick links" item from labs. Olipro, Limi, (and anybody else who wants to try), can you confirm that you have that enabled? You can add ?labs=0 to your url to not load any labs then go to the labs section in settings and see what you have enabled.

Comment 11

[Scott Turner]

This is broken for me with ?labs=0.

Setting javascript.options.methodjit.content false fixes it for me.

Comment 12

[David Mandelin [:dmandelin]]

1. Has this been seen on 32-bit browser builds, or is everyone that gets this using a 64-bit build?

2. Can we get a regression range?

Comment 13

[Ed Morley [:emorley]]

I'm using 32bit build (see comment 1).

Comment 14

[Paul O'Shannessy [:zpao] (not bugmail, email directly)]

(In reply to comment #12)
> 1. Has this been seen on 32-bit browser builds, or is everyone that gets this
> using a 64-bit build?

I see it on my 32-bit nightly on windows

> 2. Can we get a regression range?

2 consecutive builds from give me this:
http://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=6cfd43067a07&tochange=b433764562cc

Comment 15

[Ed Morley [:emorley]]

Works: http://hg.mozilla.org/mozilla-central/rev/2cd50afa5a00
Doesn't: http://hg.mozilla.org/mozilla-central/rev/35f8ec6fa9e6

Slightly narrower range than above:
http://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=2cd50afa5a00
&tochange=35f8ec6fa9e6

Comment 16

[Ed Morley [:emorley]]

The last range included 170 merged changesets from Tracemonkey, so used the TM nightlies/hourlies to narrow down.

Works: http://hg.mozilla.org/tracemonkey/rev/7b8898c9b54c
Doesn't: http://hg.mozilla.org/tracemonkey/rev/08794e076ded

Pushlog:
http://hg.mozilla.org/tracemonkey/pushloghtml?fromchange=7b8898c9b54c&tochange=08794e076ded

Comment 17

[David Mandelin [:dmandelin]]

Great work, Ed! In that tiny range, it's most likely:

 http://hg.mozilla.org/tracemonkey/rev/08794e076ded

Comment 18

[Bill McCloskey [inactive unless it's an emergency] (:billm)]

(In reply to comment #17)
> Great work, Ed! In that tiny range, it's most likely:
> 
>  http://hg.mozilla.org/tracemonkey/rev/08794e076ded

I backed this out and it seems to fix the problem.

Comment 19

[Ed Morley [:emorley]]

Fixed for me using:
Mozilla/5.0 (Windows NT 6.1; WOW64; rv:2.0b8pre) Gecko/20101119 Firefox/4.0b8pre

Comment 20

[Scott Turner]

Works for me in the latest 4.0b8pre.

Comment 21

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

Ed; that's surprising, since I don't think we made any code changes to mozilla-central. We'd previously established that this was intermittent. Any chance that GMail changed the code on their side? Can you confirm that it's still broken with the 20101118 nightly?

Comment 22

[Mike Shaver (:shaver emeritus)]

It was backed out on m-c as well: http://hg.mozilla.org/mozilla-central/rev/0096453a93e7

It was broken for me yesterday's nightly, I just updated and it's fixed.