I'm once again trying to port some python-dbus stuff over to GDBus with GI. The part where this currently fails is registering an object on the Bus:

$ python -c "from gi.repository import Gio; con=Gio.bus_get_sync(Gio.BusType.SESSION, None); con.register_object('/org/gtk/Test', Gio.DBusInterfaceInfo(), None, 'my_data')"
Traceback (most recent call last):

+ Trace 227832

Now, I don't understand where the 6th argument comes from. The C API 

  http://developer.gnome.org/gio/2.29/GDBusConnection.html#g-dbus-connection-register-object

has the GDestroyNotify as 6th argument, and the GError** as 7th, both of which are usually provided internally by pygobject. The corresponding GIR from this:

        <parameters>
          <parameter name="object_path" transfer-ownership="none">
            <doc xml:whitespace="preserve">The object path to register at.</doc>
            <type name="utf8" c:type="gchar*"/>
          </parameter>
          <parameter name="interface_info" transfer-ownership="none">
            <doc xml:whitespace="preserve">Introspection data for the interface.</doc>
            <type name="DBusInterfaceInfo" c:type="GDBusInterfaceInfo*"/>
          </parameter>
          <parameter name="vtable" transfer-ownership="none" allow-none="1">
            <doc xml:whitespace="preserve">A #GDBusInterfaceVTable to call into or %NULL.</doc>
            <type name="DBusInterfaceVTable" c:type="GDBusInterfaceVTable*"/>
          </parameter>
          <parameter name="user_data" transfer-ownership="none">
            <doc xml:whitespace="preserve">Data to pass to functions in @vtable.</doc>
            <type name="gpointer" c:type="gpointer"/>
          </parameter>
          <parameter name="user_data_free_func"
                     transfer-ownership="none"
                     scope="async">
            <doc xml:whitespace="preserve">Function to call when the object path is unregistered.</doc>
            <type name="GLib.DestroyNotify" c:type="GDestroyNotify"/>
          </parameter>
        </parameters>

There are two odd things here:

 * The GError isn't specified at all. Other methods in /usr/share/gir-1.0/Gio-2.0.gir do specify this. Is this a bug in g-i?

 * The GDestroyNotify seems to be right. Many methods in Gio's GIR declare it as allow-none="1", a lot of others don't. Which one is right here?

Trying to supply an extra None doesn't work:

python -c "from gi.repository import Gio; con=Gio.bus_get_sync(Gio.BusType.SESSION, None); con.register_object('/org/gtk/Test', Gio.DBusInterfaceInfo(), None, 'my_data', None)"

But specifying a bogus GDestroyNotify seems to "work":

python -c "from gi.repository import Gio; con=Gio.bus_get_sync(Gio.BusType.SESSION, None); con.register_object('/org/gtk/Test', Gio.DBusInterfaceInfo(), None, 'my_data', lambda *x: None)" 

(process:7712): GLib-GIO-CRITICAL **: g_dbus_is_interface_name: assertion `string != NULL' failed

(process:7712): GLib-GIO-CRITICAL **: g_dbus_connection_register_object: assertion `g_dbus_is_interface_name (interface_info->name)' failed

(These warnings seem justified, as the passed InterfaceInfo() is bogus)

So it seems pygobject doesn't properly see the GDestroyNotify here?

Note that this isn't related to the invoke-rewrite branch. It happens just as well in 2.28, or earlier master versions.