Re: SubstruvtorNotify and DestroyEvent

[Owen Taylor <otaylor redhat com>]

Michael Lausch <mla@gams.co.at> writes:

> I need to get SubstructureNotify events from the root window.
> What i did was:
> 
>   root_widget = gtk_window_new(GTK_WINDOW_POPUP);
>   fprintf(stderr,"root_widget is at %p\n", root_widget);
>   gtk_widget_realize(root_widget);
>   fprintf(stderr,"root_widget realized\n");
>   gdk_window_set_user_data(GDK_ROOT_PARENT(), root_widget);
>   gdk_window_ref(GDK_ROOT_PARENT());
>   gdk_xid_table_insert(&(GDK_WINDOW_XWINDOW(GDK_ROOT_PARENT())), 
> 		       GDK_ROOT_PARENT());
> 
> and then
> 
>   gtk_signal_connect(GTK_OBJECT(root_widget), "event",
> 		     GTK_SIGNAL_FUNC(substruct_notify), 0);
>   fprintf(stderr,"root_widget signal connected\n");
>   /* i'm not sure if this is really necessary */
>   XSelectInput(GDK_DISPLAY(), GDK_WINDOW_XWINDOW(GDK_ROOT_PARENT()), 
> 	       SubstructureNotifyMask);
> 
> with this setup, i observ DestroyNotify events on the root window. A
> closer look into the XDestroyWindowEvent and the XAnyEvent part of the
> raw X event showed me, that the gdk_translate_event function takes the 
> XanyEvent.window part as the target window, which is the root
> window. But the destroyed window is not the root window. It's stored
> in the XDestroyWindowEvent.window data. Is this on purpose and do i
> miss something? If this is so, how do i call a signal function on a
> ClientMessage event, which is sent with an X event mask of
> SubStructureNotifyMask?

First, using SubStructureNotifyMask for that is a bit bizzare.
I would think StructureNotify would be a more reasonable choice.

Basically, something like SubstructureNotify mask goes beyond the
level of detail in which I think GDK should be expected to report X
events. GDK can't report two separate windows for a destroy event, so
it's current behavior IMO, isn't wrong, though it isn't useful.

I'd probably advise using the facilities provided by
gdk_window_add_filter(); redirecting the events from the root window
to a widget is a tricky business and probably not worth the bother.

(GDK probably though, should ignore destroy events with event->event
 != event->window. Patches accepted)
 
Regards,
                                        Owen