Re: Why some GTK objects are not GObject?

[David Nečas (Yeti) <yeti physics muni cz>]

On Thu, May 11, 2006 at 11:17:59AM +0700, Marc Krisnanto wrote:
> Why are there "objects" in GTK which are not GObject?
> For example, GtkTreeModel, GtkTreeIter, GtkTreePath etc.
> I am speaking for GTK+2.0 because I see G_IS_OBJECT assertions
> fail.
> 
> Will they be GObject in the future or will they stay as they are?

GtkTreeModel is an interface.  But anything implementing
this interface *is* a GObject.  How do you make the
assertions fail?  The interface typecasting machinery is
not a problem:

    GtkListStore *store;
    GtkTreeModel *model;

    store = gtk_list_store_new(1, G_TYPE_INT);
    model = GTK_TREE_MODEL(store);

    g_print("G_IS_OBJECT(store): %d\n", G_IS_OBJECT(store));
    g_print("G_IS_OBJECT(model): %d\n", G_IS_OBJECT(model));

GtkTreeIter is a very simple structure intended to be
commonly allocated on stack (i.e., be an automatic
variable) and copied by value:

    GtkTreeIter iter1, iter2;

    set_iter_to_something(&iter1);
    iter2 = iter1;

Obviously, it cannot be a GObject then.

GtkTreePath is another very simple data structure which does
not make much sense to be an GObject.

> This makes those kind of objects difficult to implement because
> it makes an inconsistent model. I am referring to implementing
> GTK via direct calls from a programming language without going
> through C (for example, via a JIT engine such as GNU Lightning).
> 
> With GObject, I can easily automate "self" while with those kind
> of "object" I am forced to pass a pointer to the struct, which
> means I have to take care of details to maintain, synchronize,
> and free the pointer.

The same for any simple C structure you may happen to use
like GdkRectangle or GtkAllocation, for GList items, etc.
They are not `objects'.  If all these simple data types were
objects it might make Java programmers happy but I doubt any
C programmer would use Gtk+ then.

Yeti


--
Anonyms eat their boogers.