Re: [icqlib-dev] BSD Fix, and question about two callbacks

[Michael M. <mic...@in...>]

On Thursday, 31. May 2001 02:15, Denis V. Dmitrienko wrote:
[...]
>
> > Also, I had a question about the InvalidUIN and WrongPassword
> > callbacks. In udphandle.c, the both have invoke_callback followed by
> >
> > icq_UDPAck(icqlink, seq);
> >
> > In kicq though, both callbacks lead to the icqlink being deleted
> > (both callbacks call RegWindow::stop, which calls
> > RegWindow::disconnect, which calls icq_ICQLINKDelete).
> >
> > Once the icqlink is deleted, it returns to icq_ServerResponse in
> > udphandle.c, which immediately tries to use it.
> >
> > Shouldn't the calls be reversed? The icq_UDPAck first and then
> > invoke_callback?
>
> Hmm, it is more than just 2 callbacks. Most of the time we send
> acknowledgement packet after calling callback. Which I think could cause
> problems if client try to disconnect/delete icq_Link inside the
> callback, and which is causing them in kicq's RegWindow right now.
> Probably calling callbacks after acknowledging the packet would solve
> this kind of problems in general and RegWindow crash in particular.

I agree with you, that acknowledging the packet before calling callbacks 
should be more save. But I encountered another problem with the same 
result (RegWindow crash): When registering a new UIN with correct password 
the icqlib tried to delete some tcp stuff, although it was not in use and 
therefore not allocated. In my situation testing for icq_UseTCP before 
calling icq_TCPDone helped. Perhaps this will fix Eric Warmenhoven's 
problem as well. I submitted a patch on 
http://sourceforge.net/projects/icqlib

-- 
Michael Moeller      Universitaet Oldenburg, Fachbereich Informatik, 
                     Postfach 25 03, 26111 Oldenburg 
                     Tel.:0441 798-3483              Raum: A3 2-218
        mailto:Mic...@in...