plans for XplatUIX11.cs:

	- 1 thread watching X socket, doing nothing but
	  XNextEvent/hwnd.Queue.Enqueue (with minimal motion
	  compression/key repeat detection)

	- figure out how to deal with the selection events, since they
	  busy loop waiting on a response from the X socket.  maybe
	  block that thread with an event and wake it up when the
	  event thread get enough info?

	- Idle event raised once regardless of how many thread queues
	  we have (should it come from the X event thread?)

	- X11Hwnd which caches (amongst other things) the properties
	  for the toplevel window so we can avoid doing all those
	  XChangeProperty things on every single UpdateStyle call.

	- have RULES for things (like "do not call this function
          except from ...", etc) and make sure they're enforced.

	- SendMessage.. can't we just insert the right message on the
          hwnd's queue instead of using SendAsyncMethod?

	- what's the difference between PostMessage and SendMessage?

	- clean up the MESS that is:
		GetWindowState
		SetWindowState
		
		and especially:
		SetHwndStyles
		SetWMStyles
		DeriveStyles
		the Style handling in CreateWindow

	- multiple screens?  do we care?  xinerama foo?

	- fix the mostly static, somewhat instance nature of
          XplatUIX11.cs.  Either make it static, or make it an
          instance.

	- we need to rethink the expose event handling.  right now
	  expose events are only handled once all the other events in
	  the queue are handled.  So, if you're resizing quickly you
	  get no redraws until you're done.  maybe look at the
	  _NET_WM_SYNC_REQUEST protocol (for the resize/redraw case
	  specifically).  Also, the current expose handling has us
	  accessing the invalid list/rectangle from multiple threads.

	- factor out ALL knowledge of the multiple windows
	  (client/whole) into the X11Hwnd class, so we can try to
	  factor it out.

	- clean up the systray icon stuff - right now it destroys (!)
	  the client_window and sets client_window == whole_window.
	  ugh.

	- hwnds should keep track of the width/height they report to
          SWF as well as the x window's width/height.  That is, we
          should allow hwnds with 0,0 width/height even though X
          doesn't allow it.  There's some "zero_size" stuff in Hwnd,
	  but it still appears impossible to have 0,0 sized controls.

	- window (subwindow, really) destruction should move to Hwnd
          if possible.

	- remove 90% of "lock (XlibLock)" usage.  It's not really
	  needed at all, except perhaps to synchronize the
	  XEventThread and the PeekMessage/GetMessage methods.

	- the big synchronization problem now is the XEventQueue
	  class.  I think all queue operations need to be done while
	  holding a lock, as the events are queued from the
	  XEventThread and dequeued from GetMessage.

possible classes:

	- XplatUIX11 (creates the first (only?) X11Display class.
          Most calls should just get the hwnd/display and make a
          method call)

	- X11Display (manages the display handle, creates the event
	  thread, keeps ref to an X11Atoms and the X11RootHwnd)

	- XEventThread (does this need to be a separate class?  can
	  probably just be a method in X11Display)

	- X11Hwnd (move all the style translation here?  handles
	  property changes and caches them.  window title here too)

		- X11RootHwnd (add the special root window property stuff
		  here)

	- X11Atoms (per display atom list.  should these be properties
          or fields?  fields probably, with the builtins as constants).

	- Xlib (the pinvoke wrapper for Xlib calls)

window managers we need to test again:

	- metacity on SLED/SUSE
	- metacity on ubuntu edgy
	- compiz on SLED
	- compiz HEAD
	- beryl?  or should compiz be enough?

specs:

Extended Window Manager Hints:  http://standards.freedesktop.org/wm-spec/wm-spec-latest.html
XEmbed spec:                    http://standards.freedesktop.org/xembed-spec/latest/
Systray spec:                   http://standards.freedesktop.org/systemtray-spec/systemtray-spec-0.2.html