Re: client side windows branch now in gnome git

[Alexander Larsson <alexl redhat com>]

On Fri, 2009-04-03 at 16:32 +0300, Kalle Vahlman wrote:
> 2009/4/3 Alexander Larsson <alexl redhat com>:
> > The client side windows code is now availible in the client-side-windows
> > branch of gtk+ on git.gnome.org, rebased on the latest version of gtk+.
> [...]
> > So, go play with it. It should be much easier to do now, with it being
> > it the official repository, based on the official git tree.
> 
> yay!
> 
> I already did that from the gtk-shallow one though, and ran into some
> trouble with WebKit. I haven't had time to pinpoint it exactly, but
> the assertion at
> 
>   http://git.gnome.org/cgit/gtk+/tree/gdk/gdkwindow.c?h=client-side-windows#n2044
> 
> fails whenever I happen to have flash in a web page. WebKit-GTK+
> embeds the flash plugin window with an unholy mix of GtkPlug and XT
> which I don't fully grasp yet.
> 
> From my debugs I gather that it is an unmatched calls to start/end
> implicit paint or least flushing in between (which, guessing from that
> assert, is not supposed to happen...). The GtkPlug tester seems to
> work ok though, so it very well could be an application bug too.

This was caused by webkit resizing/moving a GdkWindow during a
(double-buffered) expose. I don't recommend doing this, as the exact
behaviour of that is kinda weird (the double buffer paint stack sort of
depends on the window positions, etc and if they change after
begin_paint things go slightly weird). However, its not a hard error (it
used to "work" (i.e. not assert, do something), so I fixed the csw
branch to handle this.