> From: Owen Taylor [mailto:otaylor@redhat.com] > > On Thu, 2003-06-26 at 02:54, Murray.Cumming@Comneon.com wrote: > > I believe that the callback given to g_object_set_qdata_full() > > should be called when the GObject is destroyed, but it > doesn't seem to work > > in this simple test case: > > http://bugzilla.gnome.org/showattachment.cgi?attach_id=17788 > > > > Have I made a simple mistake in that test code, or am I > misunderstanding > > something? > > > > We use this in the internals of gtkmm and I'm sure it works > most of the > > time, so I am confused. We can of course just connect to > the "destroy" > > signal if we can't get this to work consistently. > > qdata is freed on finalize, not on destroy. Your example leaks > the button, so the qdata will never be freed. I thought it might be something like this. I guess we'll watch for destroy rather than finalize. > The reason that it leaks is probably most apparent if you > understand why it is *not* leaked in the normal case: > > create button: reference count 1, floating > add button to container: container adopts button, reference count 1, > not floating > destroy button: button is removed from parent, reference > count drops to 0, finalized > > But if you destroy a floating button, nothing will happen to > the reference count. OK, thanks. Out of interest, can we explicitly finalize the button, or do we have to add it to a container, then destroy it, in order to avoid the leak. Murray Cumming murrayc@usa.net www.murrayc.com
