Re: [gtk-list] Re: querying object args and gtk+ class initialis

[Tim Janik <timj gtk org>]

On Thu, 29 Apr 1999, Michael Beach wrote:

> >> 
> >> Paradoxically though, gtk_args_query() contains the following snippet of
> >> code...
> >> 
> >>   g_return_val_if_fail (arg_info_hash_table != NULL, NULL);
> >> 
> >>   /* make sure the types class has been initialized, because
> >>    * the argument setup happens in the gtk_*_class_init() functions.
> >>    */
> >>   gtk_type_class (class_type);
> >> 
> >> ie it wants to call gtk_type_class() to ensure that class initialisation is
> >> done, but unless some class has been previously initialised through some
> >> other
> >> path of execution it can never get that far, as it will bomb out due to
> >> arg_info_hash_table being uninitialised. Indeed, when the call to
> >> gtk_type_class() is uncommented in my code fragment above, the error goes
> >> away,
> >> as the call results in object_arg_info_ht being initialised.
> >> 
> >> To me, all of this looks like a bit of a buglet.
> > [...]
> > 
> > yeah, you're absolutely correct, i added that line
> >>   gtk_type_class (class_type);
> > when we had a similar problem with signals, where a query function needed
> > the class to be initialized, and didn't really think about the hash table
> > constrains of the argument code.
> > 
> > the proper fix would be:
> > 
> > @@ gtkarg.c:gtk_args_query
> > -  /* make sure the types class has been initialized, because
> > -   * the argument setup happens in the gtk_*_class_init() functions.
> > -   */
> > -  gtk_type_class (class_type);
> > 
> > @@ gtkarg.h:gtk_args_query
> > +  /* make sure the types class has been initialized, so a
> > +   * arg_info_hash_table != NULL is passed into this function
> > +   */
> >  GtkArg*         gtk_args_query           (GtkType       class_type,
> > 
> > @@ gtkcontainer.h:gtk_container_query_child_args
> > +  gtk_type_class (class_type);
> > +
> >    return gtk_args_query (class_type, container_child_arg_info_ht, arg_flags,
> > n_args);
> > 
> > @@ gtkobject.h:gtk_object_query_args
> > +  gtk_type_class (class_type);
> > +
> >    return gtk_args_query (class_type, object_arg_info_ht, arg_flags, n_args);
> > 
> 
> Great, those patches look like they'll do the trick. However I've not
> encountered the function gtk_container_query_child_args before. So, what's the
> difference between an ordinary arg, and a "child_arg" of a container? If the
> answer is long, a pointer to something to read would be fine.

unfortunately there's no specific doumentation available yet, though the
gtk-list archives might reveal the ongoing discussion before the actuall
implementation happened. basically the normal object args cover things
like a label's text string, or the GtkWidget::visible state. what child-args
care about is packing flags like in boxes, e.g. expand or fill, as such a
given child arg only exists for a specific container<->child combination.

you can easily grasp the funcitonality by plaing around with GLE's widget 
editor, it's a gtk module available from gnome cvs or its homepage
http://www.SoftHome.net/pub/users/timj/gle/index.htm
the latest version should work fine with the gtk+ 1.2.2 version.

> 
> ----------------------------------
> Michael Beach

---
ciaoTJ