Re: [Gtk-gnutella-devel] What causes the SIGPIPE exception?

[alex <al...@be...>]

On Sat, 2002-08-31 at 04:00, Michael Core wrote:
> alex <al...@be...> wrote:
> 
> > According to the man pages a SIGPIPE signal is sent to the process when
> > it tries to write data to something that is not listening. However what
> > I can't figure out is what "pipe" is this SIGPIPE referring to.
> 
> The sockets. This happens when you're disconnected unexpectedly. You
> cannot really prevent to trigger SIGPIPE because there's a race condition
> between write() and receiving a RST/FIN-packet. You cannot predict it.

Ahh, I understand now. I assume a lot of network applications must do
the same thing, catch SIGPIPEs as they occur whenever a connection goes
down suddenly.

> > The only clue I have so far is that a null signal handler is set up in
> > main.c to handle SIGPIPE's and that a hack is employed as occasionaly
> > SIG_IGN doesn't work properly.
> 
> I guess your debugger gets triggered because it registers the signals
> itself for being able to handle them. The default action for SIGPIPE is
> terminate the process. So if there'd be a problem GtkG should
> occasionally crash with "Broken pipe".

Thats the bit that confused me, YAMD only installs a signal handler for
SIGSEV (to trap buffer over-runs etc) so I assumed the GtkG signal
handler should behave as before. I can get round it with linking the
YAMD library statically and you'll be pleased to know it currently does
not detect any memory trickery.

> > Anyone able to eulucidate me? Point me to a decent explanation?
> 
> You might want to read "Advanced Programming in the UNIX Environment" by
> Richard W. Stevens.

Another book to add to my Amazon wish list, thanks :-)
-- 
Alex@Bennee.com
http://www.bennee.com/~alex/