Closed Bug 618683 Opened 14 years ago Closed 13 years ago

Flash plugin hang [@ hang | NtUserMessageCall | RealDefWindowProcW ] [@ F1546913566_____________________________________ ]

Categories

(Core Graveyard :: Plug-ins, defect)

6 Branch
x86
Windows XP
defect
Not set
critical

Tracking

(firefox6+ fixed, firefox7 fixed, firefox8 fixed)

VERIFIED FIXED
mozilla8
Tracking Status
firefox6 + fixed
firefox7 --- fixed
firefox8 --- fixed

People

(Reporter: scoobidiver, Assigned: jimm)

References

Details

(Keywords: hang, regression)

Crash Data

Attachments

(2 files)

[@ hang | KiFastSystemCallRet ] signature is #4 top crasher on 4.0b8pre for the last week.
I can not determine how many crashes are caused only by F1546913566 signature.

Signature	hang | KiFastSystemCallRet
UUID	621fe45b-4df4-4b04-8de5-77c7f2101211
Process Type	plugin Version: Filename: NPSWF32.dll
Time 	2010-12-11 13:21:05.711136
Uptime	22190
Install Age	22221 seconds (6.2 hours) since version was first installed.
Product	Firefox
Version	4.0b8pre
Build ID	20101208030336
Branch	2.0
OS	Windows NT
OS Version	5.1.2600 Service Pack 3
CPU	x86
CPU Info	GenuineIntel family 6 model 14 stepping 12
Crash Reason	EXCEPTION_BREAKPOINT
Crash Address	0x7c90e514
App Notes 	AdapterVendorID: 8086, AdapterDeviceID: 27a2
MSAFD Tcpip [TCP/IP] : 2 : 1 :
MSAFD Tcpip [UDP/IP] : 2 : 2 : %SystemRoot%\system32\mswsock.dll
MSAFD Tcpip [RAW/IP] : 2 : 3 : %SystemRoot%\system32\mswsock.dll
RSVP UDP Service Provider : 6 : 2 : %SystemRoot%\system32\rsvpsp.dll
RSVP TCP Service Provider : 6 : 1 : %SystemRoot%\system32\rsvpsp.dll
MSAFD NetBIOS [\Device\NetBT_Tcpip_{DCE1BE57-A22F-412D-8D74-9FB981C63DCF}] SEQPACKET 4 : 2 : 5 : %SystemRoot%\system32\mswsock.dll
MSAFD NetBIOS [\Device\NetBT_Tcpip_{DCE1BE57-A22F-412D-8D74-9FB981C63DCF}] DATAGRAM 4 : 2 : 2 : %SystemRoot%\system32\mswsock.dll
MSAFD NetBIOS [\Device\NetBT_Tcpip_{131D0887-D1D0-43E3-A252-6C905F46D3E0}] SEQPACKET 0 : 2 : 5 : %SystemRoot%\system32\mswsock.dll
MSAFD NetBIOS [\Device\NetBT_Tcpip_{131D0887-D1D0-43E3-A252-6C905F46D3E0}] DATAGRAM 0 : 2 : 2 : %SystemRoot%\system32\mswsock.dll
MSAFD NetBIOS [\Device\NetBT_Tcpip_{C92601FE-C4E2-4227-B174-757F7E8BAE91}] SEQPACKET 1 : 2 : 5 : %SystemR

Frame 	Module 	Signature [Expand] 	Source
0 	ntdll.dll 	KiFastSystemCallRet 	
1 	user32.dll 	NtUserMessageCall 	
2 	user32.dll 	RealDefWindowProcW 	
3 	user32.dll 	DefWindowProcW 	
4 	NPSWF32.dll 	F1546913566_____________________________________ 	F58547254____________________________________________________________________:4857
5 	NPSWF32.dll 	F1185421069_____________________________ 	F58547254____________________________________________________________________:268
6 	user32.dll 	InternalCallWinProc 	
7 	user32.dll 	UserCallWinProcCheckWow 	
8 	user32.dll 	CallWindowProcAorW 	
9 	user32.dll 	CallWindowProcW 	
10 	xul.dll 	mozilla::plugins::PluginInstanceChild::PluginWindowProc 	dom/plugins/PluginInstanceChild.cpp:1196
11 	user32.dll 	InternalCallWinProc 	
12 	user32.dll 	UserCallWinProcCheckWow 	
13 	user32.dll 	DispatchClientMessage 	
14 	user32.dll 	__fnDWORD 	
15 	ntdll.dll 	KiUserCallbackDispatcher 	
16 	xul.dll 	mozilla::plugins::PluginInstanceChild::CreateWinlessPopupSurrogate 	dom/plugins/PluginInstanceChild.cpp:1460
17 	user32.dll 	NtUserPeekMessage 	
18 	xul.dll 	base::MessagePumpForUI::ProcessNextWindowsMessage 	ipc/chromium/src/base/message_pump_win.cc:339
19 	xul.dll 	base::MessagePumpForUI::DoRunLoop 	ipc/chromium/src/base/message_pump_win.cc:209
20 	xul.dll 	base::MessagePumpWin::RunWithDispatcher 	ipc/chromium/src/base/message_pump_win.cc:52
21 	xul.dll 	base::MessagePumpWin::Run 	ipc/chromium/src/base/message_pump_win.h:78
22 	xul.dll 	MessageLoop::RunInternal 	ipc/chromium/src/base/message_loop.cc:219
23 	xul.dll 	MessageLoop::RunHandler 	
24 	xul.dll 	MessageLoop::Run 	ipc/chromium/src/base/message_loop.cc:176
25 	xul.dll 	XRE_InitChildProcess 	toolkit/xre/nsEmbedFunctions.cpp:506
26 	plugin-container.exe 	wmain 	toolkit/xre/nsWindowsWMain.cpp:128
27 	plugin-container.exe 	__tmainCRTStartup 	obj-firefox/memory/jemalloc/crtsrc/crtexe.c:591
28 	kernel32.dll 	BaseProcessStart 	

NPSWF32.dll 	10.1.102.64 	5BFFF4B3CC284B94A7DD3026F0CD791E1 	NPSWF32.pdb
Summary: Flash plugin hang [@ hang | KiFastSystemCallRet ] [@ F1546913566_____________________________________ ] → Flash plugin hang [@ hang | NtUserMessageCall | RealDefWindowProcW ] [@ F1546913566_____________________________________ ]
Reproduced this while trying to reproduce bug 632196 using 

http://mais.uol.com.br/view/fu1e5ebic3ev/anderson-silva-vs-chael-sonnen-ufc-117-full-04029C3666E09993A6?types=A

I reloaded four or five times trying to get the video to display. Flash crashed while I was scrolling during a reload attempt.

bp-6e045f81-048a-4f41-a0dd-9dfeb2110207
Mozilla/5.0 (Windows NT 5.1; rv:6.0a1) Gecko/20110524 Firefox/6.0a1

crash report:
e529ecc3-cfb0-41a2-9b85-e8a152110524
This dropped significantly along with the browser-side hang signature in bug 659924 between the trunk builds from 2011-05-27 and 2011-05-28: -05-25 through -05-27 had way over 300 such hangs, while since the -05-28 builds, we have way below 50 such hangs reported per build. (Before -05-25, it wasn't 7.0a1 yet, so the data set I'm looking at right now only starts with that date.)

Whatever landed on trunk that day - the range is http://hg.mozilla.org/mozilla-central/pushloghtml?startdate=2011-05-27+03%3A00%3A00&enddate=2011-05-28+03%3A00%3A00 - seems to have significantly reduced those hangs. This range includes a major cairo update, not sure what else could have influenced this...
Probably b8c7dd3bddbc which was causing plugin hang on trunk and aurora. See bug 658741.
(In reply to comment #5)
> Probably b8c7dd3bddbc which was causing plugin hang on trunk and aurora. See
> bug 658741.

Ah, sounds logical. Which doesn't make me less worried about 5.0 seeming to have elevated "hang" volume compared to other versions, though. :-/
Assignee: nobody → smooney
Not sure if it is related to this signature, but https://crash-stats.mozilla.com/report/list?signature=hang%20|%20NtUserMessageCall%20|%20RealDefWindowProcA shows us as a very similar signature and shows as #2 in today's report: https://crash-analysis.mozilla.com/chofmann/20110719/top-plugin-crashes-6.0.html
Crash Signature: [@ hang | NtUserMessageCall | RealDefWindowProcW ] [@ F1546913566_____________________________________ ]
yes, it seems this signature has quite an increase in FF5/FF6.  any reason for the sharp increase?  it is ranked #1 in crash-stats from our last rollup on 7/15/11?
Summary: Flash plugin hang [@ hang | NtUserMessageCall | RealDefWindowProcW ] [@ F1546913566_____________________________________ ] → [adbe 2811589] Flash plugin hang [@ hang | NtUserMessageCall | RealDefWindowProcW ] [@ F1546913566_____________________________________ ]
Here's a history of the "hang | NtUserMessageCall | RealDefWindowProcW" signature on Firefox 6* - attaching instead of posting inline it as the list is pretty long.

It looks like this was pretty low on the radar until 2011-05-14/-15, when it rose to roughly rank 5 among all reports for Firefox 6*. A few days later it regressed due to a checkin by jimm, which was backed out on 2011-06-10 on what was meanwhile Aurora. This fixed the worst part, but since then, this stayed on rank 3-4 (of course, numbers are rising as the amount of users are rising with 6 being on Beta now).

Either a checkin, a Flash release or some website (Youtube?) change on May 13th or 14th seems to be the problem that is left over.
This occurs on all Flash versions, as http://test.kairo.at/socorro/2011-07-21.firefox.6.0.flashhangdetails.html shows (same reports for 7.0a2 and 8.0a1 look similar), just that Flash 10.0.* has RealDefWindowProcA instead of RealDefWindowProcW, so we can probably close out a Flash release causing this, the other two remain. We don't have user comments for plugin hangs, so we don't know too much about what users are doing there, unfortunately.

jimm, do you know of any checkin that could have caused that back then on May 13th or 14th?
As this signature hovers around ranks in the 30s and 40s on the released 5.0 and in the top 5 in all of 6, 7, and 8, this somehow seems the most likely scenario to me.
The two day checkin window of http://hg.mozilla.org/mozilla-central/pushloghtml?startdate=2011-05-13&enddate=2011-05-15+04%3A00 seems most likely to contain something that caused Flash to be more likely to hang - probably it was a checkin from the 13th, as 14th still being lower than 15th could be a connection of it having been a Saturday and people often needing more than a day to update - and indeed, a significant portion of the hangs reported on the 15th are from a build id starting with 20110514 (next to those with 20110515).

So, it looks like the regression window is most likely http://hg.mozilla.org/mozilla-central/pushloghtml?startdate=2011-05-13&enddate=2011-05-14+04%3A00 - we probably need someone familiar with plugin code to figure out regression candidates, as I can't really make out what in this range, including the merges, could cause slowdown in plugin code. From what I see, it was mostly small things that landed there that fiddled with things related to plugins, like bug 608013 by jimm (I guess bug 633282 didn't really touch anything the plugin would care about, right?) or bug 648277 by roc, there was a tracemonkey merge as well, but I'd guess that shouldn't affect plugins. As I said, we'll probably need someone knowing about plugins to dig into that.
jim, roc, can either of you shed any light on this? It seems like this issue goes way back into May but we didn't see it explode until we migrated 6.0 to beta. Any thoughts on the candidates that ben proposed?
This doesn't make any sense to me but, it seems these crashes dropped off the radar for 6.0 between 7/13 and 7/21 - 

https://crash-stats.mozilla.com/report/list?product=Firefox&version=Firefox%3A6.0&platform=windows&query_search=signature&query_type=contains&query=RealDefWindowProcW&reason_type=contains&date=07%2F26%2F2011%2007%3A51%3A49&range_value=4&range_unit=weeks&hang_type=any&process_type=any&do_query=1&signature=hang%20|%20NtUserMessageCall%20|%20RealDefWindowProcW

Possibly Adobe released a bug fix? Or is crash stats throwing me off?

We merged 6.0 to beta on 7/5, and AFAICT nothing has really landed on the beta channel since.
(In reply to comment #10)
> Created attachment 547718 [details]
> Firefox 6 hang report history of "hang | NtUserMessageCall |
> RealDefWindowProcW"
> 
> Here's a history of the "hang | NtUserMessageCall | RealDefWindowProcW"
> signature on Firefox 6* - attaching instead of posting inline it as the list
> is pretty long.
> 
> It looks like this was pretty low on the radar until 2011-05-14/-15, when it
> rose to roughly rank 5 among all reports for Firefox 6*. A few days later it
> regressed due to a checkin by jimm, which was backed out on 2011-06-10 on
> what was meanwhile Aurora. This fixed the worst part, but since then, this
> stayed on rank 3-4 (of course, numbers are rising as the amount of users are
> rising with 6 being on Beta now).
> 
> Either a checkin, a Flash release or some website (Youtube?) change on May
> 13th or 14th seems to be the problem that is left over.
> This occurs on all Flash versions, as
> http://test.kairo.at/socorro/2011-07-21.firefox.6.0.flashhangdetails.html
> shows (same reports for 7.0a2 and 8.0a1 look similar), just that Flash
> 10.0.* has RealDefWindowProcA instead of RealDefWindowProcW, so we can
> probably close out a Flash release causing this, the other two remain. We
> don't have user comments for plugin hangs, so we don't know too much about
> what users are doing there, unfortunately.
> 
> jimm, do you know of any checkin that could have caused that back then on
> May 13th or 14th?
> As this signature hovers around ranks in the 30s and 40s on the released 5.0
> and in the top 5 in all of 6, 7, and 8, this somehow seems the most likely
> scenario to me.

http://test.kairo.at/socorro/2011-07-21.firefox.6.0.flashhangdetails.html

This doesn't show much data:

https://crash-stats.mozilla.com/report/list?signature=hang%20|%20NtUserMessageCall%20|%20RealDefWindowProcW 

I feel like I'm missing something here..
(In reply to comment #14)
> This doesn't make any sense to me but, it seems these crashes dropped off
> the radar for 6.0 between 7/13 and 7/21

That's showing Build IDs and our beta builds have those IDs of 20110705195857 for 6.0b1, 20110713171652 for 6.0b2, 20110721152715 for 6.0b3 (which was just pushed out to users yesterday), so it's clear those are the only relevant build IDs you get for a "6.0" version. See my attachment 547718 [details] for history on all 6* versions, which include the Nightly (6.0a1) and Aurora (6.0a2) phases.

(In reply to comment #15)
> http://test.kairo.at/socorro/2011-07-21.firefox.6.0.flashhangdetails.html
> 
> This doesn't show much data:
> 
> https://crash-stats.mozilla.com/report/
> list?signature=hang%20|%20NtUserMessageCall%20|%20RealDefWindowProcW 
> 
> I feel like I'm missing something here..

That's all the data we get from the crash-stats system, and yes, it's a bit hard to tell from that what's going on, but you can even make more of it than me. What we know is that we have a bad regression of this kind of hang, my report basically just tells us that this signature on the plugin side is always paired with a "hang | mozilla::plugins::PPluginInstanceParent::CallNPP_SetWindow(mozilla::plugins::NPRemoteWindow const&)" signature on the browser side, if that helps anything.
We also know that it didn't regress on 5, but only on 6 and higher, and that the regression happened on Nightlies from May 14 first, so everything points to code changes on our side from May 13.
We need to track this for FF6. Sorry, I should have nominated it yesterday.
We getting about 1400 of these hangs a day on 6.0 with 1 million users. We are getting about 400-500 on 5.0 with 60 million users. It doesn't make sense that this is entirely related to a flash version since the distribution of who is on which flash versions wouldn't be that different between 6.0 and 5.0. I am mostly worried about this exploding when we release 6.0 - we already are correlating this to feedback about crashiness (probably hangs).
As plugin process don't report any URLs to us, I've looked at the URLs from the matching browser side (no automated report, just manually looking into a couple of reports) and in the 30 reports or so I looked into, a good bunch are facebook and youtube URLs, but also a lot of other sites, many of them video-related. This clearly is not a single-site issue, though, a lot of the URLs point to video stuff, though.
(In reply to comment #18)
> We getting about 1400 of these hangs a day on 6.0 with 1 million users. We
> are getting about 400-500 on 5.0 with 60 million users.
Stats in relative value are more explicit:
It's #3 top crasher/hanger in 6.0 (5.3% crash/ADU) while only #40 in 5.0 (0.29% crash/ADU), meaning that it's 18 times more crashy.

> It doesn't make sense that this is entirely related to a flash version since
> the distribution of who is on which flash versions wouldn't be that different
> between 6.0 and 5.0.
Confirmed by stats:
* Distribution on 5.0:
         8% (25/304) vs.   1% (1103/104798) 10.1.102.64
          1% (2/304) vs.   0% (405/104798) 10.1.53.64
          1% (3/304) vs.   0% (264/104798) 10.1.82.76
          2% (5/304) vs.   0% (346/104798) 10.1.85.3
          3% (8/304) vs.   0% (381/104798) 10.2.152.26
          3% (10/304) vs.   1% (858/104798) 10.2.152.32
          6% (19/304) vs.   1% (808/104798) 10.2.153.1
          7% (20/304) vs.   1% (1147/104798) 10.2.159.1
         10% (31/304) vs.   1% (1344/104798) 10.3.181.14
          4% (13/304) vs.   1% (655/104798) 10.3.181.22
         33% (100/304) vs.   8% (8385/104798) 10.3.181.26
         18% (56/304) vs.   3% (3375/104798) 10.3.181.34
* Distribution on 6.0:
          7% (104/1526) vs.   2% (475/28650) 10.1.102.64
          1% (17/1526) vs.   1% (146/28650) 10.1.53.64
          1% (13/1526) vs.   0% (92/28650) 10.1.82.76
          1% (14/1526) vs.   0% (75/28650) 10.1.85.3
          2% (37/1526) vs.   1% (173/28650) 10.2.152.26
          3% (52/1526) vs.   1% (241/28650) 10.2.152.32
          2% (32/1526) vs.   1% (207/28650) 10.2.153.1
          4% (61/1526) vs.   1% (296/28650) 10.2.159.1
          1% (16/1526) vs.   0% (57/28650) 10.2.161.23
          5% (79/1526) vs.   1% (344/28650) 10.3.181.14
          5% (74/1526) vs.   1% (219/28650) 10.3.181.22
         39% (595/1526) vs.   9% (2661/28650) 10.3.181.26
         25% (385/1526) vs.   5% (1423/28650) 10.3.181.34
          1% (14/1526) vs.   1% (211/28650) 11.0.1.60
(In reply to comment #10)
> jimm, do you know of any checkin that could have caused that back then on
> May 13th or 14th?
> As this signature hovers around ranks in the 30s and 40s on the released 5.0
> and in the top 5 in all of 6, 7, and 8, this somehow seems the most likely
> scenario to me.

Ok this is making better sense. I'd say bug 608013 is the likely culprit. We can back that patch out if we want, regressing the silverlight bug it fixed. Maybe a better solution would be to keep that fix and special case it to silverlight.
(In reply to comment #21)
> Ok this is making better sense. I'd say bug 608013 is the likely culprit. We
> can back that patch out if we want, regressing the silverlight bug it fixed.
> Maybe a better solution would be to keep that fix and special case it to
> silverlight.

If we can back out on trunk fast and watch stats for a few days while you perhaps work on a patch that narrows things down (i.e. to use the change for Silverlight only), I'd hope we can come in with a good fix in time to make 6.0, which is the most pressing matter. That is, if the stats show the backout being successful in reducing this problem.

Does that sounds good?
Attached patch patchSplinter Review
Let's try this on trunk.
Assignee: smooney → jmathies
Attachment #548539 - Flags: review?(benjamin)
Blocks: 608013
Attachment #548539 - Flags: review?(benjamin) → review+
http://hg.mozilla.org/mozilla-central/rev/0a936ddb70e9

Lets give it a few days, if stats on mc don't come down, we back this out and re-open the bug. If they do, we land this on aurora and beta.
Status: NEW → RESOLVED
Closed: 13 years ago
Resolution: --- → FIXED
Component: Flash (Adobe) → Plug-ins
Product: Plugins → Core
QA Contact: adobe-flash → plugins
Target Milestone: --- → mozilla8
Version: 10.x → 6 Branch
Keywords: regression
Jim, is this patch only Silverlight specific? If so, why do we think this will fix the problem? The comments in the bugs complain about youtube and facebook.
The regressing bug was silverlight-specific, but we applied the quirk to all plugins. This patch limits the quirk to silverlight so that it doesn't apply to Flash.
Summary: [adbe 2811589] Flash plugin hang [@ hang | NtUserMessageCall | RealDefWindowProcW ] [@ F1546913566_____________________________________ ] → Flash plugin hang [@ hang | NtUserMessageCall | RealDefWindowProcW ] [@ F1546913566_____________________________________ ]
After more than 24h of a Nightly with this being out there, we have only seen a single hang with this signature with a build ID having 20110727 (or later), while we have seen 30-80 for every of the earlier build IDs in the last weeks on Nightly.

This means your patch is a full success from all we know so far! \o/

Can we get it onto Aurora and Beta ASAP, please?
(In reply to comment #27)
> After more than 24h of a Nightly with this being out there, we have only
> seen a single hang with this signature with a build ID having 20110727 (or
> later), while we have seen 30-80 for every of the earlier build IDs in the
> last weeks on Nightly.
> 
> This means your patch is a full success from all we know so far! \o/
> 
> Can we get it onto Aurora and Beta ASAP, please?

I'll nominate. Should we keep this open or close it and open a new bug on residual hangs? (The stack on the 27th build look about the same.) Apparently my silverlight patch fingered an existing but rare flash hang that still exists.
Attachment #548539 - Flags: approval-mozilla-beta?
Attachment #548539 - Flags: approval-mozilla-aurora?
(In reply to comment #28)
> Should we keep this open or close it and open a new bug on residual hangs?
It already exists: bug 634892.
(In reply to comment #28)
> Apparently my silverlight patch fingered an existing but rare flash hang
> that still exists.

As my analysis in attachment 547718 [details] and also our 5.0 or earlier stats show, there always has been some smaller-volume hang before that Silverlight patch that caused this particular regression.
I'm all for hunting all those issues down, but let's do separate bugs for separate issues - this one is for the regression caused by the Silverlight fix, and your fix for that seems to have worked out pretty well. :)
awesome...  :)
Comment on attachment 548539 [details] [diff] [review]
patch

Discussed in release-drivers and approved by asa, bsmedberg and me -- please land this immediately so that we make the next beta cut off
Attachment #548539 - Flags: approval-mozilla-beta?
Attachment #548539 - Flags: approval-mozilla-beta+
Attachment #548539 - Flags: approval-mozilla-aurora?
Attachment #548539 - Flags: approval-mozilla-aurora+
(In reply to comment #32)
> Comment on attachment 548539 [details] [diff] [review] [review]
> patch
> 
> Discussed in release-drivers and approved by asa, bsmedberg and me -- please
> land this immediately so that we make the next beta cut off

http://hg.mozilla.org/releases/mozilla-aurora/rev/19d0db5cd4a0
http://hg.mozilla.org/releases/mozilla-beta/rev/c9f4bee8ee55
2011072900 	1 - 100.000%
2011072100 	7077 - 100.000%
2011071300 	2426 - 100.000%
2011070500 	305 - 100.000%

Nice looking stats.
Status: RESOLVED → VERIFIED
Maybe that was premature, all the 29th build stats look like that. I'm guessing that build might have just gone live recently, instead of on the 29th?
Status: VERIFIED → RESOLVED
Closed: 13 years ago13 years ago
jimm, it would of course help to tell where you get that data from to know what it means. ;-)

For beta, it's useless to look at data yet because we haven't shipped 6.0b4 (the one with the 20110729 build ID) to users on the beta channel yet, and those few hundred that already use it of course generate fewer crashes/hangs than the million people on 6.0b3 (the one with the 20110721 build ID).
Still, the graphs on https://crash-stats.mozilla.com/report/list?signature=hang%20|%20NtUserMessageCall%20|%20RealDefWindowProcW&version=Firefox:8.0a1&range_value=4&range_unit=weeks (Nightly) and https://crash-stats.mozilla.com/report/list?signature=hang%20|%20NtUserMessageCall%20|%20RealDefWindowProcW&version=Firefox:7.0a2&range_value=4&range_unit=weeks (Aurora) make a pretty good impression and very much look like the regression this bug is about has been fixed there. Once we have data for beta (in 2-3 days I hope), we should see the same there.

(Of course, building beta builds on a Friday and only pushing them to users on Monday causes us to not get reasonable data as early as we should, but that's a release management/process problem we need to fix, which I'm pushing for, but it's out of the scope for this bug.)
(In reply to comment #36)
> jimm, it would of course help to tell where you get that data from to know
> what it means. ;-)
> 
> For beta, it's useless to look at data yet because we haven't shipped 6.0b4
> (the one with the 20110729 build ID) to users on the beta channel yet, and
> those few hundred that already use it of course generate fewer crashes/hangs
> than the million people on 6.0b3 (the one with the 20110721 build ID).

From crash stats for version 6.0:

https://crash-stats.mozilla.com/query/query?product=Firefox&version=Firefox%3A6.0&platform=windows&range_value=1&range_unit=weeks&date=08%2F01%2F2011+08%3A48%3A57&query_search=signature&query_type=contains&query=&reason=&build_id=&process_type=plugin&hang_type=any&plugin_field=filename&plugin_query_type=exact&plugin_query=&do_query=1

I assumed it was live because the build was on the 29th. My mistake. There are so many builds flying around these days it's hard to keep track.
It was #3 top crasher in 6.0b3 and is now #19 in 6.0b5 and #16 in 7.0a2.
It is only #42 top crasher in 5.0.
So it is partially fixed. See bug 634892 for further investigations.
Status: RESOLVED → VERIFIED
Product: Core → Core Graveyard
You need to log in before you can comment on or make changes to this bug.