1130156 - Main process hang when opening a window-modal dialog

Status: REOPENED

(Firefox :: General, defect)

Description

[Mike Conley (:mconley) (:⚙️)]

Attachment: Process sample

Myself and a few others have been seeing this the past day or so. Margaret was able to get a process sample, which I've included.

STR, as far as I can tell, are to have a browser session open for a while (e10s or not, it doesn't seem to matter), and then do something to cause a window modal to open. For example, go to any page requiring HTTP Authentication, or a page that might cause the Master Password prompt to appear.

Comment 2

[Mike Conley (:mconley) (:⚙️)]

[Tracking Requested - why for this release]:

This is a particularly nasty regression, and the train merge isn't too far down the road. I don't want this to get lost in the shuffle.

Comment 3

[Lukas Blakk [:lsblakk] use ?needinfo]

Regression window will help here - Margaret, any chance you could run mozregression on recent builds to narrow this down?

Comment 4

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

For what it's worth I haven't seen this again since we were chatting about shortly before Mike filed this.

Comment 5

[:Margaret Leibovic]

(In reply to :Gavin Sharp [email: [e-mail]] from comment #4)
> For what it's worth I haven't seen this again since we were chatting about
> shortly before Mike filed this.

I also haven't seen again, so I don't have reliable STR, so it would be hard to get a regression range.

Has something changed in the modal window code recently? Could we start by looking into that?

Comment 6

[Mike Conley (:mconley) (:⚙️)]

I just tried to reproduce it with no luck. I suspect whatever caused it has been fixed or backed out.

Comment 7

[Mike Conley (:mconley) (:⚙️)]

Ok, I haven't seen this for a bunch of days now, at least with e10s enabled. I'm closing WFM unless somebody is still seeing it.

Comment 8

[Mike Conley (:mconley) (:⚙️)]

gavin said he's still seeing this, and on release builds. :(

Comment 9

[Mike Conley (:mconley) (:⚙️)]

Can you give me a list of all of your enabled add-ons, gavin?

Comment 10

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

In the 35 profile that I just saw this in I have:
Mass Password Reset 1.05 ([e-mail])
TabCloser 1.07 ([e-mail])
URL Lister 1.3 ([e-mail])
User Agent Switcher 0.7.3 ({e968fc70-8f95-4ab9-9e79-304de2a71ee1})
Developer Assistant 0.3.0.20140917 ({75739dec-72db-4020-aa9a-6afa6744759b})
Gecko Profiler 1.15.7 (jid0-edalmuivkozlouyij0lpdx548bc@jetpack)
Firefox Interest Dashboard 0.9.1 ([e-mail])

Comment 11

[Blake Winton (:bwinton) (:☕️)]

In common, I have (or have had recently):
Mass Password Reset
User Agent Switcher
Gecko Profiler,
and Firefox Interest Dashboard.

Comment 12

[Mike Conley (:mconley) (:⚙️)]

How easily can you reproduce this, gavin? Can you try to reproduce it with those add-ons disabled?

And if you're able to hit it again, can you also get a process sample? I'd be curious to see if it matched margarets.

Comment 13

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

I haven't seen this again since. Seems very sporadic...

Comment 14

[Sylvestre Ledru [:Sylvestre]]

As it seems very hard to reproduce, I am going to untrack it. Please resubmit if you have it again!

Comment 15

[Rich Gray (:rbgray)]

Mozilla/5.0 (Macintosh; Intel Mac OS X 10.6; rv:36.0) Gecko/20100101 Firefox/36.0 SeaMonkey/2.33.1

Been getting these every now and then for several months with various dialog boxes.  CPU goes up to around 100% and the buttons in the dialog box will not click.  Game over.  I've been forcing crash dumps by sending sigabrt to the SM process, latest one about an hour or two ago.  I don't know if this is the bug I'm looking for, but the timeframe is about right.   Today's hang was for an unresponsive script. Ah! Found the crash dumps:

Today: Crash ID: bp-55ca5a0c-48e9-4cb6-9df4-62eb72150716   unresponsive script
24Jun: Crash ID: bp-10777d64-c866-48d1-a47a-06cbf2150624   save message for unsent message
17Jun: Crash ID: bp-6f6ad78e-9dc6-4625-9f1e-a8f612150617   confirm certificate security exception

Comment 16

[Mike Conley (:mconley) (:⚙️)]

So window modal dialogs run their own nested event loops, and nested event loops can be pretty dangerous and awful. I know that's pretty vague and hand-wavey, but trust me that it's true.

This, to me, sounds like us running a nested event loop, and then a timer or event fires somewhere that causes us to do something that is not possible until the inner loop is exited, and so we deadlock or something.

That's all speculation, but I think it's plausible.

This would also explain the trickiness in reproducing; we'd need to make sure we were running a nested event loop at the right time.

The easiest way to fix this class of bug, imo, is by eliminating the nested event loop, and by moving as many of our modal dialogs into content as we can (like we do for alert/confirm/prompt).

I think finding a regression window for this bug is pretty unrealistic, tbh. At least, not until we get more reliable STR.

Comment 17

[Rich Gray (:rbgray)]

Hasn't recurred for me.  I'm running currently running 
Mozilla/5.0 (Macintosh; Intel Mac OS X 10.6; rv:41.0) Gecko/20100101 Firefox/41.0 SeaMonkey/2.38

Comment 18

[tino84]

Hi,
I am seeing this since months, using Firefox ESR.
Sometimes, when opening "save as" or "browse" windows, firefox hangs and I need to terminate the process.

currently using Firefox ESR 38.7.0, Windows 7 Enterprise SP1.