Crash Backspacing in Addressing Widget of Compose Window

VERIFIED FIXED in mozilla0.9.1


18 years ago
11 years ago


(Reporter: sol, Assigned: dmose)


(Blocks: 2 bugs, {crash})

Windows NT
Dependency tree / graph

Firefox Tracking Flags

(Not tracked)


(Whiteboard: fix in hand; have r=, sr=, a=; awaiting tree verifications)


(2 attachments)



18 years ago
I crash when I (a) send a web page via email from Navigator *and* (b) backspace 
in the addressing widget of the compose window.

Running 2001-05-17-04 (commercial) on WinNT.

I can reproduce this about 75% of the time. It seems that following a particular 
order of steps is very important in reproducing this bug.

Steps to reproduce.

1. Launch MoJo.

2. Launch Mail (Tasks | Mail). Authenticate to email server when prompted. I'm 
using IMAP.

3. Go back to Navigator. Browse to a site - say,

4. Send the web page via email (File | Send Page ...).

5. *As soon as you see the cursor* in the addressing widget in the email compose 
window, begin typing, and then - after you've typed about 5 characters - quickly 
start hitting the backspace key (like you made a mistake in your typing and want 
to correct it right away). 

Note - I have configured email addressing to run against my LDAP server at work.

Expected result: The text you have typed in the addressing widget is deleted, 
and the cursor backspaces.

Actual result: CRASH!!

Stack trace:

Trigger Type:  Program Crash 

Trigger Reason:  Access violation 

Thread ID: 203  

Call Stack:    (Signature = nsLDAPAutoCompleteSession::FinishAutoCompleteLookup 

Session.cpp, line 923]

Session.cpp, line 601]

Session.cpp, line 327]


Seems to be crashing in LDAP code.

For more information, see Talkback incidents: 30630928 and 30631144 (identical 
stack traces).
Looks like a LDAP problem, reassign to dmose
Assignee: ducarroz → dmose


18 years ago
Severity: normal → critical
Keywords: crash


18 years ago
Blocks: 17880
Keywords: nsbeta1
Priority: -- → P1
Target Milestone: --- → mozilla0.9.1

Comment 2

18 years ago
are the options 1) fix the crash  or 2) take out the feature??

Comment 3

18 years ago
chofmann: a third option would be to release note it.  My impression is that
this crash isn't all that common, and the easy workaround is to disable LDAP in
the prefs.

I've been able to reproduce this maybe twice (out of about 10 attempts) running
on Linux with the release build.  Unfortunately, I've never seen it under debug
builds at all (let alone while running under the debugger) after about 30
attempts, so it's a very timing sensitive thing.  Haven't been able to catch it
under the debugger on Mac either; Srilatha's had similar experiences on windows;
she's going to try over a slower connection once she gets home this evening.

However, from reading the code, I have a guess as to what could be happening. 
I'm gonna go grovel through the code some more, and will post my thoughts in a bit.


Comment 4

18 years ago
OK, after grovelling through the source for a while (and thanks to talkback) I
see what could cause this crash.  With luck I'll have a patch later tonight.

Comment 5

18 years ago
Created attachment 35772 [details] [diff] [review]
minimal paper-over-the-problem patch

Comment 6

18 years ago
I've just attached a minimal paper-over-the-problem fix.  With just this patch,
the race still exists, but if we lose the race, the result will just be a
confused autocomplete session rather than a crash.  Worst case, I'd suggest
shipping with this patch.

However, I expect that with a few more hours of work tommorrow (Weds), I'll have
a better patch that includes this paper-over fallback, but also eliminates the
race entirely.


18 years ago
Whiteboard: minimal fix avail now; better fix later today

Comment 7

18 years ago
OK, I've got the real fix.  A side-effect of this fix is that the race described
in 70422 is fixed also.
Keywords: patch, review
Whiteboard: minimal fix avail now; better fix later today → fix in hand

Comment 9

18 years ago
Actually, the side effect is that 77452 is fixed, NOT 70422.
Blocks: 77452

Comment 10

18 years ago
who should be the reviewer?
Whiteboard: fix in hand → fix in hand need review

Comment 11

18 years ago
ducarroz, hewitt, or chuang.  mscott would make a fine sr=.  :-)
the patch looks good. R=ducarroz


18 years ago
Whiteboard: fix in hand need review → fix in hand, have r=, need sr=

Comment 13

18 years ago
This has been built and tested on all three platforms.

Comment 14

18 years ago


18 years ago
Whiteboard: fix in hand, have r=, need sr= → fix in hand; have r=, sr=; need a=
a= for checkin to 0.9.1


18 years ago
Whiteboard: fix in hand; have r=, sr=; need a= → fix in hand; have r=, sr=, a=; awaiting tree verifications

Comment 16

18 years ago
Fix checked in.
Last Resolved: 18 years ago
Resolution: --- → FIXED

Comment 17

18 years ago
Using build 2001-05-31 on mac, linux and windows this is fixed.  Verified.
I added the LDAP directory and went back to a 5-23 build to see the crash.  I 
then ran sol's test on all three platforms and did not crash. 
Product: MailNews → Core
Product: Core → MailNews Core
You need to log in before you can comment on or make changes to this bug.