+ II)
+
+ The design of scrolling in wxWindows is markedly different from that offered
+ by the GTK itself and therefore we cannot simply take it as it is. In GTK,
+ clicking on a scrollbar belonging to scrolled window will inevitably move
+ the window. In wxWindows, the scrollbar will only emit an event, send this
+ to (normally) a wxScrolledWindow and that class will call ScrollWindow()
+ which actually moves the window and its subchildren. Note that GtkPizza
+ memorizes how much it has been scrolled but that wxWindows forgets this
+ so that the two coordinates systems have to be kept in synch. This is done
+ in various places using the pizza->xoffset and pizza->yoffset values.
+
+ III)
+
+ Singularily the most broken code in GTK is the code that is supposes to
+ inform subwindows (child windows) about new positions. Very often, duplicate
+ events are sent without changes in size or position, equally often no
+ events are sent at all (All this is due to a bug in the GtkContainer code
+ which got fixed in GTK 1.2.6). For that reason, wxGTK completely ignores
+ GTK's own system and it simply waits for size events for toplevel windows
+ and then iterates down the respective size events to all window. This has
+ the disadvantage, that windows might get size events before the GTK widget
+ actually has the reported size. This doesn't normally pose any problem, but
+ the OpenGl drawing routines rely on correct behaviour. Therefore, I have
+ added the m_nativeSizeEvents flag, which is true only for the OpenGL canvas,
+ i.e. the wxGLCanvas will emit a size event, when (and not before) the X11
+ window that is used for OpenGl output really has that size (as reported by
+ GTK).
+
+ IV)
+
+ If someone at some point of time feels the immense desire to have a look at,
+ change or attempt to optimse the Refresh() logic, this person will need an
+ intimate understanding of what a "draw" and what an "expose" events are and
+ what there are used for, in particular when used in connection with GTK's
+ own windowless widgets. Beware.
+
+ V)
+
+ Cursors, too, have been a constant source of pleasure. The main difficulty
+ is that a GdkWindow inherits a cursor if the programmer sets a new cursor
+ for the parent. To prevent this from doing too much harm, I use idle time
+ to set the cursor over and over again, starting from the toplevel windows
+ and ending with the youngest generation (speaking of parent and child windows).
+ Also don't forget that cursors (like much else) are connected to GdkWindows,
+ not GtkWidgets and that the "window" field of a GtkWidget might very well
+ point to the GdkWindow of the parent widget (-> "window less widget") and
+ that the two obviously have very different meanings.
+