Path: utzoo!utgpu!news-server.csri.toronto.edu!clyde.concordia.ca!uunet!snorkelwacker!bloom-beacon!HARVARD.HARVARD.EDU!jimf%saber
From: jimf%saber@HARVARD.HARVARD.EDU
Newsgroups: comp.windows.x
Subject: Re: Invisible windows (Technical Limitations of X?)
Message-ID: <9003221848.AA02726@armory>
Date: 22 Mar 90 18:48:47 GMT
Sender: daemon@athena.mit.edu (Mr Background)
Organization: The Internet
Lines: 47

|In particular, consider the situation where we have to produce
|complicated graphics into a window.  The desired effect is to have
|whatever is behind the window show through (no matter where the window
|is currently positioned).  The proposed solutions requires one to
|
|	Translate to the root window, draw the graphics, and undraw
|them when the window moves!!

Won't work if there's a repaint from a window below yours; X will clip
around your "transparent" window and no update will happen, or the
window will draw through your window -- destroying it -- and you'll
not be told to update it.

|I (admittedly, a neophyte) don't understand what the technical
|difficulties are in providing this capability.

There are no technical difficulties, only synchronization problems
when you're using a non-retained windowing system.  If X had been
designed as a retained system this would be a snap, but it would
require a lot more memory to run.  Currently the shape extension can
do what you want but it slows down a LOT if you use a complex shape or
change things often.

The memory problem can be partially alleviated by creating an
object-oriented environment where your objects are some type of higher
complexity than a pixel (eg rectangle, ellipse, whatever).  You can
rebuild the pixel image by redrawing the objects and all you've lost
is some CPU time.  This is what you end up doing with X anyway so you
really don't loose anything and you simplify your applications
dramatically (they don't have to remember how to redraw everything
because they never see a repaint [expose] event).

Once you have a retained layer, creating transparent windows is as
simple as drawing from the bottom up and clipping around things as
necessary.  The code to do so with good efficiency is remarkably
simple.

I implemented a library layer which created a retained object-oriented
environment under X11R2 about a year and a half ago, with excellent
success in both memory requirements and performance (and portability
-- a co-worker [hi Joe] ported it to SGI GL in about four hours).
Hopefully someday I'll have the time to reimplement a public-domain
version....

jim frost
saber software
jimf@saber.com