Re: RFC: Reference counts on Gtk2::Tooltips

[Andrew Pimlott <gtk-perl-list andrew pimlott net>]

On Mon, May 05, 2003 at 04:16:42PM -0400, muppet wrote:
> Andrew Pimlott said:
> > On Mon, May 05, 2003 at 03:46:43PM -0400, muppet wrote:
> > > well, there's not really a way to leak the tooltips reference in perl,
> > > because the wrapper will clean up the object when the time comes.  in C,
> > > this object is leaked simply by virtue of not calling _unref on it.
> >
> > When does the time come?
>
> G::Object::DESTROY *always* calls g_object_unref (provided the pointer is not
> NULL (which shouldn't happen)).
>
> DESTROY is called when the last reference to the perl scalar goes away.  the
> perlobj manpage describes this.  (i would swear there's another place that
> gives examples but i can't find it.)

That's what I thought you meant, ok.

> > I'm confused, but maybe I'm just thinking of the terms differently. If you
> > put the tooltip in a global variable (or inside some global data structure)
> > and then forget about it, that's a leak to me.
>
> but it's not leaking for real, because perl still knows that it exists and
> will call DESTROY on it when that reference goes away.  the leak to which i
> refer is the genuinely unrecoverable leak for which C is so famous and which
> can cause resource exhaustion and crashes (you can't free it if you don't know
> its address).

That's what I thought, too.  This sort of dissonance always seems to
come up around the subject of memory leaks.  :-)

> but have no fear --- your argument that providing the workaround doesn't hurt
> anybody, doesn't change the behavior, *and* helps avoid a common and obviously
> hard-to-explain pitfall is the best one, so the fix is going in.

Great!

Andrew