[sldev] Re: [offlist] Quirk in end_net ( net.cpp ) / VWR-1410

Able Whitman able.whitman at gmail.com
Fri Jun 29 07:43:34 PDT 2007


Honestly, my investigation into this bug has only gotten far enough to get a
fairly-reliable repro. (It was aggrivating to no end trying to figure out
what was going on when I only hit the crash occasionally.)

Like you found, I saw that the crash only happened when
LLEventPoll::Impl::error() was called with an HTTP error code other than
499. I can't repro it 100% but my most reliable (about 90% of the time)
repro is this:

1. start the viewer under the debugger (F5)
2. login to the main grid
3. once logged in, press Ctrl+M to open the map
4. while the map is still rendering, Break All into the application ( i.e.,
pause debugging)
5. wait for 2 minutes or so
- waiting less than 2 minutes doesn't reliably trigger the bug
6. continue the application
- at this point you'll probably be getting lots of "unknown circuit"
warnings in the debug console
7. press Ctrl+Q to quit the viewer

The repro above ususally results in an error in the debug log like this: "".
Which indicates that the error() method has called the stop() method before
the LLEventPoll destructor is called, causing the crash.

When you resume the viewer in step 6, pretty quickly you'll see a couple of
debug messages like this:

WARNING: sendMessage - Trying to send AgentUpdate on unknown circuit
64.129.45.197:13004
WARNING: LLEventPoll::Impl::error: <1> got 404: Not Found

Once you see the 404 error from error(), you're golden, because you will
crash. Hopefeully this is helpful.

--Able

On 6/29/07, Nicholaz Beresford <nicholaz at blueflash.cc> wrote:
>
>
>
> Able Whitman wrote:
> > Oh man, I think I've just been digging into that same crash, Nick. Is
> > this the crash that happens in a call to
> > boost::intrusive_ptr_release(LLHTTPClient::Responder * p=0xfeeefeee),
> > which originates from the LLEventPoll destructor?
> >
> > If it is, let me know if I can do anything to help you out.
>
> Yes, it is that one.  I have it already, I just don't know
> the best way to fix it yet.   Let me know if you want the
> details upfront.
>
> Were you able to reproduce it?
>
> If you want details upfront, it's a nice riddle/puzzle (I just
> had it one of 10 times with the new leak enabled in one area).
>
>
> Nick
>
>
>
> Second Life from the inside out:
> http://nicholaz-beresford.blogspot.com/
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.secondlife.com/pipermail/sldev/attachments/20070629/ae158b7d/attachment.htm


More information about the SLDev mailing list