Re: Changes to the GLib main loop [ g_main_iterate() ]

[Owen Taylor <otaylor redhat com>]

Sebastian Wilhelmi <wilhelmi ira uka de> writes:

> > My other fear would be that people would write:
> > 
> >  while (i_dont_have_a_message)
> >    g_main_context_iteration (context, TRUE);
> > 
> > Which, in a non-owner thread is hideously expensive. Just like the
> > case of the g_main_context_run(). 
> 
> I don't think it is that expensive. Per iteration it costs one
> g_cond_broadcast in the owning thread, one context switch, and then
> the thread falls asleep again. Given that this happens only when a
> new event arives, it isn't that bad.

Just for the record:

On one system, the rough timing was (did it a week or so ago,
so from memory):

 - Time for a context switch 2us, so the time to signal a thread 
   and let it go back to sleep, around 5us
 
 - Time for iterations, including poll(), 10us

But remember that when dispatching a batch of pending events,
the poll() is skipped, so the main loop iteration is considerably
faster. [ No timing at hand ]

Also, if you have more than one thread waiting in that fashion,
they all wake up.

So, yes, I would consider this an expensive thing to do.

Regards,
                                                 Owen