Re: Submenu breakage

[Tim Janik <timj gtk org>]

On 7 Feb 2001, Owen Taylor wrote:

> 
> Alexander Larsson <alla lysator liu se> writes:
> 
> > Has anyone actually looked deeper in the submenu horkage? Is it 100% sure
> > that it is the boxed type changes that causes it. 
> 
> Well, the boxed typed stuff BREAKS a lot of stuff BADLY, so it's probably 
> not worth debugging until we get the boxed type fixed.

that's exagerating, boxed copying breaks signals that pass boxed
arguments in to signal handlers and rely on them to make modifications
which have to be available to the caller after emission.
that in itself should not break menus as menus don't have strange
text iterators or non-NULL terminated strings.

however the boxed type copying uses gdk_event_copy() upon GtkWidget::event
now, and that relies on the event structures properly setup.
now gtk_menu_motion_notify() creates a syntesized event that's
only half setup:

          GdkEvent send_event;

          send_event.crossing.type = GDK_ENTER_NOTIFY;
          send_event.crossing.window = event->window;
          send_event.crossing.time = event->time;
          send_event.crossing.send_event = TRUE;
          send_event.crossing.x_root = event->x_root;
          send_event.crossing.y_root = event->y_root;
          send_event.crossing.x = event->x;
          send_event.crossing.y = event->y;

specifically send_event.crossing.subwindow can contain garbage this way,
which will obviously break gdk_window_ref (event->crossing.subwindow) in 
gdk_event_copy().

i just had to figure that plain:

          GdkEvent send_event = { 0, };

won't fix this because default-zero-initialization of unions only
applies to the first member, so that code actually needs to read:

          GdkEvent send_event;
          memset (&send_event, 0, sizeof (send_event));

i cought a few other code portions that have this problem.

> 
> Regards,
>                                         Owen
> 

---
ciaoTJ