Re: revised image prop, icon patch

[Havoc Pennington <hp redhat com>]

Owen Taylor <otaylor redhat com> writes: 
> Is this still necessary? It means that you can have a pixmap storage
> type with a null pixmap, which seems like a bad idea to me. Avoiding
> this for icon-size is why I suggested the storage change, really.
> 
> I guess the downside is that if you had pixmap/mask and switched
> to pixbuf without freeing the mask, the mask wouldnt' be freed.
> I don't think this really matters ... people won't be changing
> the storage type of GtkImage widgets at all often, if ever.

I ended up doing it somewhat differently... we can sort it out
tomorrow.

I don't like breaking the invariant that the mask is part of the
PIXMAP or IMAGE types, i.e. it should always be NULL if you are a
PIXBUF type. So instead I changed it so that the mask and icon size
are allowed to exist in EMPTY images, but get deleted/reset if you
switch to PIXBUF or something. If you set the "mask" property while
not in pixmap/image mode, then you get a GTK_IMAGE_EMPTY with a
mask. Saving the mask in EMPTY is a hidden property-system-only
feature, there's no C API to retrieve it.

Anyhow, this ensures that pixmap != NULL if storage_type ==
GTK_IMAGE_PIXMAP, but also frees the mask if you switch to another
storage type.

Havoc