Re: [gtkmm] Spawing a process and being notified of its termination

[Chris Vine <chris cvine freeserve co uk>]

On Sunday 22 August 2004 22:56, Enrico Zini wrote:
> On Sun, Aug 22, 2004 at 11:24:51PM +0200, Enrico Zini wrote:
> > Almost there. I wrote code to create a pipe and set it in nonblocking
> > mode, to avoid the signal handler to block if the Gtk main loop has
> > something else to do and isn't reading from the pipe.
> > Now I'd like to add reading the pipe in the Gtk event loop, but:
>
> I reply myself, as I solved the problem: while desperately looking at
> the Gtkmm namespace in search for some clue, I stumbled on the
> Glib::Dispatcher class, which does exactly what I want: you call its
> emit() method anywhere and it activates a slot in the correct place in
> the Glib event loop.  Or at least this is what I understood.

As I understand it Glib::Dispatcher is really intended for communicating 
between threads.  The functions which IEEE Std 1003.1 requires to be 
async-signal-safe (which may be called in an asynchronous signal handler) are 
much smaller in number than those required to be thread safe, so you may 
encounter problems.  Compare 
http://www.opengroup.org/onlinepubs/000095399/functions/xsh_chap02_04.html 
with 
http://www.opengroup.org/onlinepubs/000095399/functions/xsh_chap02_09.html.

Although Glib::Dispatcher uses a pipe to implement the signalling (so it may 
be kosher), it would be a good idea to check the code and see what it gets up 
to, or (better in my view) implement a blocking pipe yourself, which is very 
easy to do.  You would normally plumb the pipe into the main program loop 
using Glib::signal_io().connect();

An alternative is for the signal handler just to set a flag (of type volatile 
sigatomic_t) and then to check for it in a timer slot (you would normally 
connect to the slot using using Glib::signal_timeout().connect()).

Chris