Re: GtkGC shared auto-release

[muppet <scott asofyet org>]

On Jan 14, 2008, at 7:59 PM, Kevin Ryde wrote:

> Torsten Schoenfeld <kaffeetisch gmx de> writes:
> > +  my $two = Gtk2::GC -> get(32, $colormap, $values);
> > +  Gtk2::GC -> get(32, $colormap, $values);
> > +  Gtk2::GC -> get(32, $colormap, $values);
> > +  # Gtk2::GC -> release($two) should now be called three times,  
> > but how do we
> > +  # verify that?
> > +}
>
> Too few releases should leave the pool with a ref on the GC, so  
> perhaps
> keep only a weakened copy of the GC and check that goes undef when the
> pool lets go.

This is not easy to test.  GCs are GObjects, not GtkObjects, so we  
don't get a "destroy" signal.  You need g_object_weak_ref() to get  
destruction notifications on plain GObjects.  Unfortunately, Glib does  
not bind the weakref stuff for technical reasons (how do you map  
between the scalar and the GObject without creating a potential  
crash?), and from what i can tell, stuff like Devel::WeakRef and  
Test::Weaken is not part of the perl core.  That would make the  
testing a little difficult.

I did, however, use g_object_weak_ref() in my XS code, with a callback  
that simply prints the dead pointer value, to verify that the GC  
objects do, in fact, go away at the right time.

Some quick experiments with using a normal perl object, stored in the  
Gtk2::GC's wrapper hash, as a destroy notifier were frustrating, as  
the objects seemed to live too long.

--
So this new album took us quite a while, to get it together, to find a  
title and things like that - and we were travelling quite a bit - we  
made a few sort of gestures in the East, and a few gestures in the  
West, and then we got thrown out because of our gestures - this is  
something that we decided was an apt title for a thing that's called  
'The Song Remains The Same'.
  -- Robert Plant, rambling to introduce a song