589095 - sessionstore.js and sessionstore.bak wiped on FF4 b4 crash

Status: RESOLVED FIXED

(Firefox :: Session Restore, defect)

Description

[Jonathan Watt [:jwatt]]

FF 4 beta 4 just crashed for me on Mac, and lost both sessionstore.js and sessionstore.bak completely loosing my session. Unfortunately Crash Reporter also crashed while trying to file the crash report, so I don't even have any info on the crash.

I'm not even sure how this is possible given that I believe we use nsSafeFileOutput or some such thing, but clearly there's a problem in the code somewhere.

Comment 1

[Marcia Knous [:marcia]]

Jonathan: I tried to repro your steps on the latest trunk build, Mozilla/5.0 (Macintosh; Intel Mac OS X 10.6; rv:2.0b5pre) Gecko/20100820 Minefield/4.0b5pre since it is easy to crash because of the Gmail bug. My session was restored fine in that instance.

I will next try to reproduce using B4.

Is there any other information you can provide - are you running with any extensions?

Comment 2

[juan becerra [:juanb]]

I've been trying to reproduce this problem on the latest Fx4b4 build candidate, but I haven't been able to. To force the crash we use the crashme extension. So any information that can help us reproduce the problem as you saw it would be very helpful.

Comment 3

[Jonathan Watt [:jwatt]]

I don't expect this is very reproducible. I crash often enough, but this was probably only the second time I've seen this problem in the last 12 months or so. Really I think this needs someone who knows the code to ponder exactly under what circumstances this could happen.

Comment 4

[Jonathan Watt [:jwatt]]

Anyway, to answer your question, I had about a dozen windows with lots of tabs in them open when I crashed. The only slightly unusual thing that I can think of that was doing was that one of the tabs was paused half way through:

http://media.libsyn.com/media/donmc1/SCO0178-omnifocuspt1-free-960x540-h264-free-540p-OS.mov

Shark was also processing a sample I'd taken from Firefox, but it had finished the actual sample without problem, so I don't think that was an issue.

I also some time earlier had started a separate Firefox binary without specifying a separate profile and got the "Firefox is already open" error from it, but again that didn't seem to immediately cause a problem, and it shouldn't have touched anything in the profile other than the '.parentlock' file, so again, I don't think that's likely to be relevant.

Comment 5

[Marcia Knous [:marcia]]

FWIW, I did see crash reporter crashing once a while back - see Bug 461702. That is the only time I have really seen that happen on the Mac.

Comment 6

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

(In reply to comment #3)
> Really I think this needs someone who knows the code to ponder exactly
> under what circumstances this could happen.

Not really sure how you would lose both. ss.bak is created at startup (just a copy of whatever ss.js). We might have called _clearDisk, but that should only happen under specific circumstances.

I'm going to assume you don't (or didn't) have Firefox set to clear browsing history at quit. I'll also assume you have browser.sessionstore.resume_from_crash = true.

Let me know if you see it again. I haven't seen any other instances of this happening, but if we can reproduce, it's pretty bad.

Comment 7

[Jonathan Watt [:jwatt]]

(In reply to comment #6)
> I'm going to assume you don't (or didn't) have Firefox set to clear browsing
> history at quit. I'll also assume you have
> browser.sessionstore.resume_from_crash = true.

Correct on both counts.

So I've just had a friend report to me that Firefox 4.01 just crashed for him, then crashed again immediately on restart. The second crash lost all his tabs.

Comment 8

[Wayne Mery (:wsmwk)]

(In reply to Jonathan Watt [:jwatt] (away 1-23 June) from comment #7)
> (In reply to comment #6)
> > I'm going to assume you don't (or didn't) have Firefox set to clear browsing
> > history at quit. I'll also assume you have
> > browser.sessionstore.resume_from_crash = true.
> 
> Correct on both counts.
> 
> So I've just had a friend report to me that Firefox 4.01 just crashed for
> him, then crashed again immediately on restart. The second crash lost all
> his tabs.

This is much like bug 637148.

Other potential examples: bug 746649

Comment 9

[Wayne Mery (:wsmwk)]

Other potential examples: bug 746649, (my) bug 873656

bug 735694 may be something of a variation.
bug 558425 would help mitigate the dataloss.

It seems to me that this should be easily reproduced, but I haven't tested

Comment 10

[u279076]

We've tried testing this in the past and have been unsuccessful. Wayne, if you want to give it a shot we'd appreciate that. Until more reliable steps to reproduce are brought to our attention by users affected by this bug I don't see there's much QA can do to move this forward.

Comment 11

[Wayne Mery (:wsmwk)]

(In reply to Wayne Mery (:wsmwk) from comment #9)
> Other potential examples: bug 746649, (my) bug 873656
> 
> bug 735694 may be something of a variation.
> bug 558425 would help mitigate the dataloss.
> 
> It seems to me that this should be easily reproduced, but I haven't tested

reminder to self

Comment 12

[Wayne Mery (:wsmwk)]

At this point I'm not likely yto attempt reproducing

Comment 13

[Tim Taubert [:ttaubert] (inactive)]

Marking as fixed by bug 883609. We should recover a lot better from corrupted sessionstore files now.