Re: Bonobo::Unknown

[Maciej Stachowiak <mjs eazel com>]

Mathieu Lacage <lacage@email.enst.fr> writes:

> Maciej Stachowiak <mjs@eazel.com> writes:
> 
> 
> > 
> > > Also, imagine I have A, B and C with interfaces IA, IB, and IC.
> > > 
> > > I make: 
> > > bonobo_object_add_interface (A, B)
> > > bonobo_object_add_interface (A, C)
> > > B = bonobo_object_QI (A, IB)
> > > bonobo_object_QI (B, IC)   ------> this will not return C.
> > > 
> > > object B does not know about object C.
> > 
> > That is not correct. the final expression in your list above _will_
> > return C. `query_interface' is reflexive, transitive, and symmetric,
> > i.e. it is an equivalence relation.
> 
> I know what QI is supposed to do and what its properties are supposed to be. 
> My point is that the current implementation does not look like it works in 
> the case described above.

Did you try it?
 
> look at the folowing: 
> 
>        for (l = newobj->priv->ao->objs; l; l = l->next) {
>                if (!g_list_find (object->priv->ao->objs, l->data)) {
>                        object->priv->ao->objs = g_list_prepend (object->priv->ao->objs, l->data);
>                        ((BonoboObject *)l->data)->priv->ao = object->priv->ao;
                         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
>                }
>        }
> 
> If you use this code to merge A and B and then to merge A and C, 
> A's linked list will be updated together with C's one but
> the B linked list will not be updated. B will still hold a pointer
> to the old linked list (which is gfree'd actually, so, we may have 
> a segfault)

The line of code I have marked sets the priv->ao field of every object
in B's AggregateObject (which includes a pointer back to B itself) to
A's priv->ao field. The `ao' struct pointer contains the `objs' linked
list of interface pointers, so A and B will always share linked lists
no matter how they get changed. In fact you could say that object
identity in the libbonobo Bonobo::Unknown implementation is having
equal ao pointers.

> Unless I missed something really important, I doubt I am wrong.
> 

I think you did, but the code is pretty unclear. Perhaps the line I
marked should be commented.

In the future it might be a good idea to test such things before
reporting bugs. :-)

 - Maciej