357696 - Crash after using nsDocumentEncoder

Status: RESOLVED INVALID

(Core :: DOM: Core & HTML, defect)

Description

[Mike Pinkerton (not reading bugmail)]

I have a C++ component that uses nsIDocumentEncoder. It hooks in as a web progress listener and gets the DOM Document from that, which it passes to nsIDocumentEncoder::Init().

On builds after 20060614, this will crash sometime later with the following stack:

http://pastebin.mozilla.org/980

My code works fine in either case. Explicitly leaking the document doesn't change anything. If I pass NULL to Init, or skip that chunk of code in my extension, everything is fine. 

Here's the changelist between the builds:

http://bonsai.mozilla.org/cvsquery.cgi?treeid=default&module=all&branch=MOZILLA_1_8_BRANCH&branchtype=match&dir=&file=&filetype=match&who=&whotype=match&sortby=Date&hours=2&date=explicit&mindate=2006-06-14+05%3A00%3A00&maxdate=2006-06-15+10%3A00%3A00&cvsroot=%2Fcvsroot

cc'ing the people involved in key patches from that day. It would be bad if ff2 ships with this bug.

Comment 1

[Mike Pinkerton (not reading bugmail)]

forgot to mention i'm testing the 1.8branch for ff2. sorry.

Comment 2

[Boris Zbarsky [:bzbarsky]]

I asked Mike to see where after his code runs we end up in ~nsDocument...

Comment 3

[Mike Beltzner [:beltzner, not reading bugmail]]

You know that Firefox 2 ships tomorrow, right? Could you elaborate on "bad"? Like, all Mac clients will always crash, always-always bad? Or "this might cause some crashes on Mac" bad?

I ask because AFAIK, we haven't seen this as a topcrash (or have we?) and I haven't seen any significant instability on my Intel Mac (which doesn't yield talkback, sadly), so I'm curious as to how we haven't hit upon this issue yet.

Comment 4

[Mike Pinkerton (not reading bugmail)]

this would prevent an under-development component from my employer from working with ff2.

Comment 5

[Mike Pinkerton (not reading bugmail)]

i keep running out of processes to build, and i have class in 2hrs anyway. I can't get to the dtor check today.

Comment 6

[Mike Connor [:mconnor]]

This kinda sucks, but it's nowhere near severe enough to stop ship.  Hopefully we can get this fixed in 1.8.1.1 (the extension could just have a minVersion of 2.0.0.1 in that case).

Comment 7

[Mike Pinkerton (not reading bugmail)]

whoops, this really was me. Not having an SDK for mozilla1.8 I had to cobble a few things together and my extension made some assumptions about nsIDocument that were no longer valid after June 14.

How about using this as a call for a real 1.8 SDK so that developers don't have to make their own? :) :) Maybe?

Comment 8

[Alex Vincent [:WeirdAl]]

pink: reference bug 321471 for the SDK.