606314 - Negative screenX for main window in localstore.rdf causes Minefield to be unable to create any windows

Status: RESOLVED WORKSFORME

(Core Graveyard :: Profile: BackEnd, defect)

Description

[Steven Michaud [:smichaud] (Retired)]

Several times in the past month or two, I've had Minefield
mysteriously hang on startup.  Then I deleted my old profile and the
problem went away.

Just now it happened again, and I've managed (I think) to isolate the
problem to my localstore.rdf file.

Currently (on OS X 10.5.8) here's all it takes me to reproduce this
problem:

1) Delete (rm -rf) ~/Library/Application Support/Firefox and
   ~/Library/Caches/Firefox.

2) Start a recent Minefield nightly, and choose not to import any
   settings.

3) Quit Minefield.

4) Copy my "bad" localstore.rdf to ~/Library/Application
   Support/Firefox/Profiles/[profiledir]/

5) Restart Minefield.

   I always hang on startup.

In my next comment I'll attach my "bad" localstore.rdf.

Comment 1

[Steven Michaud [:smichaud] (Retired)]

Attachment: "Bad" localstore.rdf that causes hang on startup

Comment 2

[Steven Michaud [:smichaud] (Retired)]

I'll test on other OS versions, on other platforms, and with other FF versions as and when I have the time.

Others should test too, of course.

I've *no* idea what causes these hangs.

I'll post a gdb trace of one of them in my next comment.

Comment 3

[Dave Garrett]

I attempted using the attached localstore.rdf in a new profile on Linux and got no hang. Could be Mac specific.

The ideal fix would of course be to kill localstore.rdf -> bug 559505

Comment 4

[Steven Michaud [:smichaud] (Retired)]

I'm actually not able to reproduce this in the most recent Minefield
nightlies (on OS X 10.5.8).

The last one that I see the problem with is
firefox-2010-10-15-03-mozilla-central.  So the "fix range" is:

http://hg.mozilla.org/mozilla-central/pushloghtml?startdate=2010-10-15+03%3A00%3A00&enddate=2010-10-16+03%3A00%3A00

Dave, could you try my STR in a Linux nightly for 2010-10-15 or
2010-10-14?

I'm not sure if the problem is now "fixed", or if it went away for
some other reason.

Comment 5

[Steven Michaud [:smichaud] (Retired)]

Attachment: Gdb stack trace of hang

Comment 6

[Steven Michaud [:smichaud] (Retired)]

I'm able to reproduce the same hang on OS X 10.6.4 (with the firefox-2010-10-15-03-mozilla-central nightly).  And it has the same "fix range" there:  I don't see the problem with the 2010-10-16-03-mozilla-central nightly.

I tested in both 64-bit and 32-bit mode, with the same result.

Comment 7

[Dave Garrett]

(In reply to comment #4)
> Dave, could you try my STR in a Linux nightly for 2010-10-15 or
> 2010-10-14?

I just tried new profiles for those dates and no hangs with your given localstore.rdf file. It doesn't look like I can reproduce this here.

I could give Windows XP a shot in a VM in a bit if you'd like, but if it's fixed it's probably not worth it.

> I'm actually not able to reproduce this in the most recent Minefield
> nightlies (on OS X 10.5.8).

If you also can't reproduce this on a stable branch, then just close this as WFM. Would be nice to find the exact bug with the fix to dupe to, but that's not really needed if it stays fixed. I was going to suggest narrowing down the range using hourly builds, but for some reason the hourly archive has Windows and Linux but no Mac files:
http://hourly-archive.localgho.st/hourly-archive2/
Even the regular tinderbox Mac build folders are empty:
http://ftp.mozilla.org/pub/mozilla.org/firefox/tinderbox-builds/
(debug and 64-bit are working there, but only go back a couple days)

Comment 8

[Steven Michaud [:smichaud] (Retired)]

> I could give Windows XP a shot in a VM in a bit if you'd like, but if it's
> fixed it's probably not worth it.

It's not at all clear this has actually been fixed.

But if it doesn't happen on Linux or Windows with my "bad" localstore.rdf file, it's (probably) Mac-specific -- which tells us *something*.

Please do try to reproduce in Windows XP.  If you can't, go ahead and close this WFM.

At some point I'll try to find out which part of my localstore.rdf is causing trouble.

Even if this bug *has* been fixed, it'd be worthwhile knowing what the issue was.

Comment 9

[Steven Michaud [:smichaud] (Retired)]

And if I have time (and this bothers me enough), I can use hg bisect to find out which patch made this problem go away.

This'd presumably give us a better idea of whether or not this bug has actually been fixed.

Comment 10

[Steven Michaud [:smichaud] (Retired)]

Actually I also can test in a Windows XP VM.

Testing with the firefox-2010-10-14-04-mozilla-central nightly, I wasn't able to reproduce the hang (using my "bad" localstore.rdf).  So I'll mark this WFM.

I'll reopen it if I later discover something interesting.

Comment 11

[Steven Michaud [:smichaud] (Retired)]

Attachment: Reduced "bad" localstore.rdf

For the record, here's the most-reduced localstore.rdf with which I can still reproduce these hangs.

Comment 12

[Steven Michaud [:smichaud] (Retired)]

> screenX="-3"

Further tests show that setting this to a negative number always cause a hang (in Minefield nightlies that are susceptible to these hangs).

One way to make screenX negative without editing localstore.rdf is to move the browser window so that part of it is off-screen to the left, then quit.  On restart I hang.

Comment 13

[Steven Michaud [:smichaud] (Retired)]

I'm now beginning to think that my patch for bug 582466 may have fixed this :-)

At some point I'll hg bisect to find out.

Comment 14

[Steven Michaud [:smichaud] (Retired)]

> I'm now beginning to think that my patch for bug 582466 may have fixed this :-)

No, that can't be.

Comment 15

[Dave Garrett]

(In reply to comment #12)
> > screenX="-3"
> 
> Further tests show that setting this to a negative number always cause a hang
> (in Minefield nightlies that are susceptible to these hangs).

That was my guess for the culprit.

If you're worried about it regressing you could write a new test to cover negative position persist values. (for all OSes, just in case)

Comment 16

[Steven Michaud [:smichaud] (Retired)]

What I'm most worried about is how a negative screenX could ever have caused hangs on startup :-)

Comment 17

[Dave Garrett]

Obvious question: can you reproduce in affected builds with GPU accel off?

Comment 18

[Steven Michaud [:smichaud] (Retired)]

> Obvious question: can you reproduce in affected builds with GPU accel off?

You mean after turning layers.accelerate-all off?

If so no -- the problem still happens with accelerated layers off.

Comment 19

[Dave Garrett]

(In reply to comment #18)
> You mean after turning layers.accelerate-all off?

That disables full acceleration for everything by default. You may need to set layers.accelerate-none to true to disable everything totally.

Comment 20

[Steven Michaud [:smichaud] (Retired)]

Setting layers.accelerate-none to 'true' also makes no difference.

I'm no longer so sure what I'm seeing are really hangs.  It may just be that the browser window fails to display.

Comment 21

[Steven Michaud [:smichaud] (Retired)]

I'm pretty sure this is related to bug 597133, and that (like that bug) it was fixed by backing out the patch for bug 574690.