> It is used since we store pointers to all widgets in the datalist of the
> toplevel widget, by calling gtk_object_set_data_full(). These pointers
> are used for the lookup_widget() utility function Glade provides.
That sounds a bit like treating gtk+ objects like bonobo objects. Whenever
a pointer to the object is kept, ref the pointer. When removed, unref it.
Considering that I would like to add an option to create bonobo objects from
glade top-level widgets, I might be better off keeping the ref/unref
mechanism. I was reading Havoc's book over the weekend, and he gives a
decent description of the three-phase destruction of a gtk+ object. Seems
like a little-used practice based on the code examples in the book, but I
see no harm in using it.
> If we didn't use the ref/unref, and a widget in the interface was
> removed, then using lookup_widget() would return an invalid memory
> address and the app would probably crash.
>
> Though to be honest I'm not sure it is that useful. People rarely
> remove widgets, and if they do they should know not to call
lookup_widget()
> for it.
I suppose if someone did write an app that dynamically creates/removes
widgets, using glade without ref/unref code would be prohibitive.
How about if I put the ref/unref code in the constructor/destructor for each
top-level widget.
typedef struct tag_Whatever
{
GtkWidget *thing1;
GtkWidget *thing2;
} Whatever;
Whatever *whatever_create( ) {
// ref thing1 and thing2
}
void whatever_destroy( Whatever *whatever ) {
// unref thing1 and thing2
}
Sound good?
John
_______________________________________________
Glade-devel maillist - [EMAIL PROTECTED]
http://lists.helixcode.com/mailman/listinfo/glade-devel
