Re: Subclasses

Hi Espen,

sorry for replying to myself, but the previous message was in part
nonsensical.

Marco Gidde <[email protected]> writes:

> As mentioned in my mail reply clg compiles with errors when
> +struct-alignmen+ is changed to 4 so I kept the old value and debugged
> a bit further. Since I don't know too much about metaclass handling,
> understanding the cgl internals is a bit hard for me. As a result I
> can almost only describe symptoms which may or may not help you
> understand what's going on here.

You were right concerning the +struct-alignmen+ with one important
exception: 64bit values (like pointers) must lie on 8-byte boundaries
so that +struct-alignmen+ not only depends on the previous calculated
slots but also on the current slot. For example:

int         -> offset 0
int         -> offset 4
pointer     -> offset 8

*but*

int         -> offset 0
pointer     -> offset 8
int         -> offset 16

I modified ALIGN-OFFSET and COMPUTE-SLOTS locally to match this
requirement, but it looks too ugly to be published here :-/

> First I went on with GINSPECT and found that it worked partially when
> I added (:GTYPE "GObject") to the definition of
> OBJECT-CONTAINER. After that expanding an item of the tree for the
> first time results in a memory fault caused by the call to
> TREE-VIEW-EXPAND-ROW. When I simply continue from the callback,
> expanding the same item for the second time gives a nice list of all
> the slots. The same happens for any expandable tree item: memory fault
> for the first click, correct expansion for the second.

This effect still remains.

> I then tried to inherit a class from DRAWING-AREA with
>   (:METACLASS GOBJECT-CLASS)
> which resulted in the following messages:
>
> WARNING:
>    GLib-GObject: specified instance size for type `PlotterPlotter' is smaller than the parent type's `GtkDrawingArea' instance size
> WARNING:
>    GLib-GObject: cannot retrieve class for invalid (unclassed) type `<invalid>'
> WARNING: Super class mismatch between CLOS and GObject for PLOTTER
>
> This messages is produced when REGISTER-NEW-TYPE calls
> %TYPE-REGISTER-STATIC and is caused by the fact, that the
> INSTANCE-SIZE of GtkDrawingArea as returned in the TYPE-QUERY object
> is 0. This is also the case for several other classes I tried
> (Gtkwidget, GObject, GtkButton), I suspect it's true for all
> classes. If I hard code a large enough value for INSTANCE-SIZE, the
> message disappears, so it seems that clg and glib disagree on this
> value. The definitions of TYPE-QUERY and TYPE-INFO match the installed
> headers.

All this vanished with the changed alignment handling as described
above.


Regards,

Marco