[java-gnome-hackers] Function pointers

[Andrew C. <an...@op...>]

GtkTreeModelFilter revolves around using
gtk_tree_model_filter_set_visible_func() to register a callback to a
function of signature (*GtkTreeModelVisibleFilterVisibleFunc). As most
of you know, we don't have a general mechanism for handling function
pointers, and indeed blacklist any method with one as an argument.

We already have a very powerful C to Java callback mechanism, and that's
the marshaling code used to deal with GObject signals. I've long had
bigger plans for that code, because it works well, and doing the C to
Java thing is ninja voodoo as it is, and nasty when you consider it in
the general case.

That observed, my general strategy for some time for dealing with non
signal callbacks has been to merely find a way to turn it into a signal
emission and then deal with it down the existing code path.

It took a lot of futzing to find the magic incantation necessary, but
I've worked something out for TreeModelFilter which C side registers a
custom signal and then the function which meets
(*GtkTreeModelVisibleFilterVisibleFunc) "simply" emits the signal.

It's all done by hand, of course, and I'm not sure how amenable this
will be to code generation (indeed, we don't have the required
signatures in our .defs data), but it seems to work fairly well. The
public API Java side looks right:
http://java-gnome.sourceforge.net/4.0/doc/api/org/gnome/gtk/TreeModelFilter=
.html

and here is the implementing C code:
http://research.operationaldynamics.com/bzr/java-gnome/mainline/src/binding=
s/org/gnome/gtk/GtkTreeModelFilterOverride.c

The real trick was the insight that we didn't need to figure out the
marshaling function when registering the signal after all - which is
where I'd been stuck before, because there wasn't an existing
BOOLEAN:OBJECT,BOXED marshaller.

AfC
Sydney