Re: [Tracker] [Patch] New CVS version

[Jamie McCracken <jamiemcc blueyonder co uk>]

Laurent Aguerreche wrote:
> Le vendredi 25 aoÃt 2006 Ã 13:34 +0100, Jamie McCracken a Ãcrit :
> > Laurent Aguerreche wrote:
> > > Small things:
> > > - a typo in tracker_log()?
> > >   "so now shuting" => "so now shutting"?
>
> Yes, no? :-)

er yes, fixed thanks.

> > sigfillset (&signal_set);
> > sigprocmask (SIG_BLOCK, &signal_set, NULL);
>
> sigprocmask() is only for single-threaded processes. Replace it with
> pthread_sigmask().

yup well spotted. man page for sigprocmask() says its undefined in 
mutlithreaded context and should use pthread_sigmask instead.

Have now fixed in CVS


> > Ergo the only thread that can handle the signals in now the main thread.
>
> I always wondered how masks on threads works and if it could make some
> not-handled signals (i.e. user has to do many Ctrl+C to just see one
> handled...). But I read some docs and it seems that only one thread with
> a correct mask is needed. So it should work!

Well I originally only expected the main thread to handle it as I only 
defined the signal handler and sig mask there. I did not realise that 
created threads inherit the creators sig mask and handler and therefore 
they would need to be set to explicitly block those signals hence the 
cock-up!

> > I believe this should fix all outstanding issues and it should therefore 
> > never crash or hang on exit.
> >
> > Please test latest CVS to confirm this is the case.
>
> Ok, I just wanted to point the problem with sigprocmask().

yes thank you - please keep pointing out any other dodgy stuff :)


--
Mr Jamie McCracken
http://jamiemcc.livejournal.com/