Moving all the RTL flipping bugs to the 2.4.0 milestone; in some
cases the changes should be easy to do. (GtkPaned comes to mind)
In those cases, we can move the bugs back to 2.2.0 if patches occur.

Actually, hpaned turned out to be one of the hardest so far, since a
good part of the position calculation is done in paned.c and shared
with   vpaned. In the end, I had to enhance gtk_paned_compute_position
() to support flipping child1 and child2 by adding a new boolean
parameter.
This should not count as an api change, however, since
gtk_paned_compute_position is an internal function (which should
really be named _gtk_paned_compute_position and not be exported at all).

Created attachment 11901 [details] [review]
patch

Can't you compute the inverted parameter using 
paned->orientation and gtk_widget_get_direction()?

(The "encapsulation" of the orientation in the subclasses
is already incomplete.)

Yes, I could. If I would have been totally confused by 
hpaned->orientation being vertical...new patch coming

Created attachment 11974 [details] [review]
less invasive patch

It definitely bothers me that in this patch, when the hpaned
is flipped, paned->child1_size refers to the size of 
paned->child2.

Not only that, but the effect of gtk_paned_set_position()
will be wrong if it effects the size of the left pane
not the size of pane1.

I think it would all come out a whole lot cleaner if the
child1_size is left the size of child 1, and flipping 
is done:

 - When getting event coordinates (update_drag(), 
   gtk_paned_button_press())

 - When allocating the children gtk_hpaned_size_allocate() - 
   it should be possible to add helper functions:

    compute_screen_rectangle (paned, &unflipped_rectangle);
    allocate_child (paned, child, &unflipped_allocation);

   and do it pretty cleanly.

Removing PATCH since the attached patch is not acceptable and I have not been able to 
produce a better one, despite several attempts. Owen, since you think this is simple, 
maybe you could give it a try...

Created attachment 20096 [details] [review]
patch

- I did it pretty much as Owen suggested, except that flipping
  was not necessary in the button_press() handler.

- It is not necessary to flip during f6 and f8 navigation because 
  these routines selects the handles to activate based on child1/child2,
  not on aboslute position.

Looks plausible; if it works, I'd say commit it.

Fri Sep 19 18:15:31 2003  Soeren Sandmann  <sandmann@daimi.au.dk>

	* gtk/gtkpaned.c, gtk/gtkhpaned.c: implement RTL flipping
	(#96632)