84250 - [Linux only] When select cancel in activation window browser would not start up and hangs

Status: VERIFIED FIXED

(SeaMonkey :: General, defect, P2)

Description

[chris hofmann]

Linux Build ID 2001050808

Steps to reproduce:
1) remove .mozilla profile
2) start N6
3) in activation window click on cancel
4) click on cancel again

Result: hangs

Expected result: N6 browser window



------- Additional Comments From [e-mail] 2001-05-08 11:45:10 ----

*** Bug 4891 has been marked as a duplicate of this bug. ***



------- Additional Comments From [e-mail] 2001-05-08 12:32:51 ----

*** Bug 4893 has been marked as a duplicate of this bug. ***



------- Additional Comments From [e-mail] 2001-05-11 10:24:01 ----

I am seeing this also.  It's 100% reproducable.  

note:  relaunching with the same profile is successful (apparently the 
cancel/cancel took the first time, although it hung that first attempt to 
launch)



------- Additional Comments From [e-mail] 2001-05-17 10:30:41 ----

Marking nsbeta1+, setting milestone 0.9.1.  Reassigning to Bhuvan.



------- Additional Comments From [e-mail] 2001-05-18 18:12:38 ----

Is this Linux only, or are we seeing it on other platforms?



------- Additional Comments From [e-mail] 2001-05-18 19:16:03 ----

I just it on windows (WinNT) with the latest build available on sweetlou (build
ID : 2001051804).

I am unable to reproduce the bug. I created a new profile and did cancel +
cancel on Activation window. I was launched into the browser. On the second run
with the same profile, I launched into the browser directly. This is the
expected behavior.

Let me check Linux...



------- Additional Comments From [e-mail] 2001-05-18 19:28:49 ----

I just it on windows -> I just tested the behavior on windows





------- Additional Comments From [e-mail] 2001-05-18 19:58:40 ----

It is reproducible on Linux. 

One thing I noticed is that it is getting a cookie from activation server and it
is being processed properly (as, on the next launch you go into the browser
directly). So, the problem seems to be with the code that is reponsible to close
windows properly and bring up the browser window. Bug 4922 reporting similar
problem (with crash) was fixed. But this one seems to be falling into platform
dependent path somewhere. Let me debug to findout that code segment.



------- Additional Comments From [e-mail] 2001-05-21 13:35:55 ----

Adding Saari & jst to the cc list. This one from the date seem to be happening
right around the XPCDOM landing...

I just got my linux build done. I am going debug further..

saari & jst,

Can you please a take look at this one? You need to a commercial debug build on
Linux for this.



------- Additional Comments From [e-mail] 2001-05-22 13:55:48 ----

I see that appshell component enumeration is failing and then it's hanging on
appShell->Run() in nsAppRunner.cpp..

Bill, what are the consequences of (appshell component) enumeration failure ?

bhuvan




------- Additional Comments From [e-mail] 2001-05-22 21:17:42 ----

Adding bryner to the cc list.



------- Additional Comments From [e-mail] 2001-05-22 23:12:52 ----

Seth volunteered to debug this with me (thanks to him). Bryner suggested that we 
look for possible leaks in the eventQueues. XPCOM_MEM_LEAK_LOG suggests that 
nsEventQueue(impl/ServiceImpl) leaks occur when the commercial builds are used 
(with/without activation in the path). Having activation window in the path of 
leaked objects is probably causing hang reported in the bug. Event queue leaks 
were found in various paths into the browser window (netscape/netscape 
-profilemanager/netscape -p <profile>). 

Mozilla does not show any leaked instances of nsEventQueue(Impl/ServiceImpl).

We will need to Saari's expertise with any possbile missing events here.

More debugging to come....



------- Additional Comments From [e-mail] 2001-05-23 09:01:26 ----

Not sure if this matters or not, but this is also reproducable on Hpux 11i using 
build #2001052209



------- Additional Comments From [e-mail] 2001-05-23 20:28:16 ----

Things still go wrong on windows with this code path, asserts out.



------- Additional Comments From [e-mail] 2001-05-24 12:53:22 ----

Adding DanM, Pavlov & DougT to the cc list. 



------- Additional Comments From [e-mail] 2001-05-24 17:11:50 ----

I think we have a breakthrough here in terms of reproducing this with much
simpler reduced case (as opposed to loading the activation window and all).

With the new test cases, it is now more apparent that modal dialog is the
culprit in this process.

Today, If we look at a regular xpcom leak log (XPCOM_MEM_LEAK_LOG=1) after
running the browser once and closing it, we will notice a leak in
nsEventQueueImpl. Looks like this is OK as we already reached the end.

But unfortunately, in case of activation, that leak doers matter much and that
seems to the core reason for the hang described in this bug.

Here are the test cases which can be used to reproduce the problem :

1. Modal dialog in the path :
- to simulate activation window (ns build):
  * Change the property "browser.registation.url" (in file
ns/xpfe/browser/resources/locale/en-US/region.properties) to
"http://sspitzer/bhuvan/modal.html" (from activation url). build it.
  * Create a new profile and launch the app
  * Activation window will come up and you will see a confirm dialog that asks
"stay with activation process ?". Click on cancel.
  * Clicking on calcel triggers window.close(). So, the activation window
disappears and the app simply hangs

2. No modal dialog in the path
  * Change the property URL mentioned in the case 1, from
"http://sspitzer/bhuvan/modal.html" to "http://sspitzer/bhuvan/non_modal.html"
  * This one has no confirm dialog. Activaiton window is loaded and
window.close()is executed directly.
  * Activation window disappears and the browser window comes up.....

3. Leak stats (ns or mozilla)
  * Load url "http://sspitzer/bhuvan/modal.html" in the browser url bar for path
that is troubling us.
  * Get stats for non_modal dialog by loading non_modal.html file.

Upcoming attachments :

1. http://sspitzer/bhuvan/modal.html
2. http://sspitzer/bhuvan/non_modal.html
3. leak stats when modal.html is loaded
4. leak stats when non_modal.html is loaded

Reassignig this to DanM as his expertise & experience with modal dialogs, event
queues and appshell component may help us knock this one down quicker. Dan, if
needed, please do reassign if you know anyone else who worked closely with the
above concenpts and components.

thanks
bhuvan




------- Additional Comments From [e-mail] 2001-05-24 17:19:59 ----

Created an attachment (id=1557)
modal.html




------- Additional Comments From [e-mail] 2001-05-24 17:20:28 ----

Created an attachment (id=1558)
non_modal.html




------- Additional Comments From [e-mail] 2001-05-24 17:23:06 ----

Let me post the html files as text type attachments. Please ignore the above as
clicking on them executes the script directly as I posted them as html type.



------- Additional Comments From [e-mail] 2001-05-24 17:23:40 ----

Created an attachment (id=1559)
modal.html (text/plain)




------- Additional Comments From [e-mail] 2001-05-24 17:24:18 ----

Created an attachment (id=1560)
non_modal.html (text/plain)




------- Additional Comments From [e-mail] 2001-05-24 17:25:30 ----

Created an attachment (id=1561)
leak stats when modal.html is loaded




------- Additional Comments From [e-mail] 2001-05-24 17:26:11 ----

Created an attachment (id=1562)
leak stats when non_modal.html is loaded




------- Additional Comments From [e-mail] 2001-05-24 17:31:58 ----

Thanks to Seth for working with me on narrowing down the problem. Also, thanks
to bryner and saari for suggesting the possible causes for this behavior.




------- Additional Comments From [e-mail] 2001-05-24 17:32:54 ----

adding myself to the cc list.



------- Additional Comments From [e-mail] 2001-05-24 19:30:08 ----

Created an attachment (id=1563)
refcount balancer output for leaked nsEventQueueImpl object (when browser loads http://sspitzer/bhuvan/modal.html and quits after clicking on cancel)




------- Additional Comments From [e-mail] 2001-05-30 11:03:09 ----

Should this be assigned to racham? DanM, are you working on this?



------- Additional Comments From [e-mail] 2001-05-30 12:16:11 ----

Activation exposes it, but it's been narrowed down to not be an activation bug.  



------- Additional Comments From [e-mail] 2001-05-30 17:05:04 ----

It got hard, then it got assigned to me. I live for that. Yes, I'm looking at it. 
No ETA.




------- Additional Comments From [e-mail] 2001-06-01 18:33:46 ----

We need this fix for m91, noting in whiteboard.



------- Additional Comments From [e-mail] 2001-06-05 10:00:06 ----

see also bugs 1730, 2659 and 2683- app hangs when certain links are pressed in
Activation- special features/T&C



------- Additional Comments From [e-mail] 2001-06-05 17:04:01 ----

  I've been running with the patch I'm about to attach for several hours and
I've seen no problems. It fixes this bug. I haven't managed yet to find out why
the problem this patch fixes hangs the app. (I suspect it's events being
processed out of order, which is very hard to track down). And the 0.9.1
deadline is today, so it's time to get this patch reviewed.
  I suspect it won't fix the bugs mentioned just above, since I'm adjusting code
that was added fairly recently, and those are old bugs. (I'll ask offline about
testing: seems to me I have to actually activate a new account.)



------- Additional Comments From [e-mail] 2001-06-05 17:05:55 ----

Created an attachment (id=1734)
make nsCacheEntry not hold a strong reference to its event queue




------- Additional Comments From [e-mail] 2001-06-05 17:34:35 ----

what if the thread identified by mThread does not have an event queue?  it looks
like we'd crash if that happened.  we should either do nothing (and simply leak
memory) or we should just go for broke and delete the object on the current
thread (i understand that there may be deadlock issues if we were to delete the
object on the current thread).  i think this could be an issue at shutdown...
that is, if memory object cache clients ever used the cache from a thread other
than the ui thread, which i think is currently never the case, we could crash.



------- Additional Comments From [e-mail] 2001-06-05 17:36:01 ----

Created an attachment (id=1736)
modified version of patch danm made to avoid having nsCacheEntry hold on to event queues.




------- Additional Comments From [e-mail] 2001-06-05 17:51:05 ----

Is this patch ready for r and sr and approval for 0.9.1?  Thanks.



------- Additional Comments From [e-mail] 2001-06-05 18:01:11 ----

PDT met with danm and gordon.  Fix ready for approvals, expected tonight.
Risk not very high, same logic.



------- Additional Comments From [e-mail] 2001-06-05 18:03:29 ----

PDT met with danm and gordon.  Fix ready for approvals, expected tonight.
Risk not very high, same logic.
a= conditional on r= and sr=



------- Additional Comments From [e-mail] 2001-06-05 18:15:26 ----

r=beard





------- Additional Comments From [e-mail] 2001-06-05 18:32:02 ----

sr=sfraser




------- Additional Comments From [e-mail] 2001-06-05 19:30:19 ----

a=chofmann



------- Bug moved to this database by [e-mail] 2001-06-05 19:35 -------

This bug previously known as bug 4890 at http://bugscape.netscape.com/
http://bugscape.netscape.com/show_bug.cgi?id=4890
Originally filed under the Browser product and Activation component.
Bug blocks bug(s) 5720.

---
Unknown bug field "attachment" encountered while moving bug
<attachment>
  <attachid> 1557 </attachid>
  <date> 05/24/01 17:19 </date>
  <desc> modal.html </desc>
</attachment>
---
Unknown bug field "attachment" encountered while moving bug
<attachment>
  <attachid> 1558 </attachid>
  <date> 05/24/01 17:20 </date>
  <desc> non_modal.html </desc>
</attachment>
---
Unknown bug field "attachment" encountered while moving bug
<attachment>
  <attachid> 1559 </attachid>
  <date> 05/24/01 17:23 </date>
  <desc> modal.html (text/plain) </desc>
</attachment>
---
Unknown bug field "attachment" encountered while moving bug
<attachment>
  <attachid> 1560 </attachid>
  <date> 05/24/01 17:24 </date>
  <desc> non_modal.html (text/plain) </desc>
</attachment>
---
Unknown bug field "attachment" encountered while moving bug
<attachment>
  <attachid> 1561 </attachid>
  <date> 05/24/01 17:25 </date>
  <desc> leak stats when modal.html is loaded </desc>
</attachment>
---
Unknown bug field "attachment" encountered while moving bug
<attachment>
  <attachid> 1562 </attachid>
  <date> 05/24/01 17:26 </date>
  <desc> leak stats when non_modal.html is loaded </desc>
</attachment>
---
Unknown bug field "attachment" encountered while moving bug
<attachment>
  <attachid> 1563 </attachid>
  <date> 05/24/01 19:30 </date>
  <desc> refcount balancer output for leaked nsEventQueueImpl object (when browser loads http://sspitzer/bhuvan/modal.html and quits after clicking on cancel) </desc>
</attachment>
---
Unknown bug field "attachment" encountered while moving bug
<attachment>
  <attachid> 1734 </attachid>
  <date> 06/05/01 17:05 </date>
  <desc> make nsCacheEntry not hold a strong reference to its event queue </desc>
</attachment>
---
Unknown bug field "attachment" encountered while moving bug
<attachment>
  <attachid> 1736 </attachid>
  <date> 06/05/01 17:36 </date>
  <desc> modified version of patch danm made to avoid having nsCacheEntry hold on to event queues. </desc>
</attachment>

Comment 1

[Dan M]

*** Bug 84247 has been marked as a duplicate of this bug. ***

Comment 2

[Dan M]

patch is in the trunk and 0.9.1 branch (you can see the patch in dup. bug 84247)

Comment 3

[Tracy Walker [:tracy]]

verifed fixed linux branch build 2001-06-06-0.9.1