Dude :-(

[Martin Baulig <martin home-of-linux org>]

Hi guys,

next time I'm on a masochist trip I'll probably fight against some Cactus ...

I already said a couple of times that I don't like the idea of doing larger
changes to any module and then immediately freezing.

So now we're in the situation that I have a source and binary incompatible
chane in libbonobo - and it *must* go in, unless we want to live with a
totally unusable Bonobo:PropertyBag, of course ....

Today was yet another of these very frustrating moments where I spend hours
to debug my code to find out what was going wrong - first I suspected that
there was a bug in my guile bindings, after writing a small C test program
I quickly realized that this wasn't the case so I went on debugging my
component.

After hacking and debugging a lot in my guile bindings, I finally reached the
point where there was a 

	BONOBO_ARG_SET_STRING (arg, text);

in my component's PropertyBag's get_prop() function, with text == "San Francisco",
but what "arrived" in my container was an empty string ("") - so I looked at
ORBit2's marshalling code, without any luck.

After three Koelsch and two films in TV I returned and then I found:

=====
void
bonobo_closure_invoke_va_list (GClosure *closure,
			       GValue   *retval,
			       GType     first_type,
			       va_list   var_args)
{
	GArray *params;
	GType   type;
	int     i;
  
	g_return_if_fail (closure != NULL);

	params = g_array_sized_new (FALSE, TRUE, sizeof (GValue), 6);

	for (type = first_type; type; type = va_arg (var_args, GType)) {
		GValue value;
		gchar *error;

		value.g_type = 0;
		g_value_init  (&value, type);

		G_VALUE_COLLECT (&value, var_args, 0, &error);
		if (error) {
			g_warning ("%s: %s", G_STRLOC, error);
			g_free (error);
			break;
		}
      
		g_array_append_val (params, value);
	}

	g_closure_invoke (closure,
			  retval,
			  params->len,
			  (GValue *)params->data,
			  NULL);

	for (i = 0; i < params->len; i++)
		g_value_unset (&g_array_index (params, GValue, i));
}
======

which is used on a function which looks like this:

======
typedef void (*BonoboPropertyGetFn) (BonoboPropertyBag *bag,
				     BonoboArg         *arg,
				     guint              arg_id,
				     CORBA_Environment *ev,
				     gpointer           user_data);
======

Since this is not the first problem with broken marshallers in Cactus's
closure code - let's make this like in a quiz show: what's broken with
this code ?

Answer + patch follows tomorrow .....

-- 
Martin Baulig
martin gnome org (private)
baulig suse de (work)