Re: mc, mdi, mem About this list

[jg pa dec com (Jim Gettys)]

> Sender: mark@kiss.uni-lj.si
> From: Marko Macek <Marko.Macek@gmx.net>
> Resent-From: wm-spec-list@gnome.org
> Date: Tue, 03 Aug 1999 19:38:58 +0200
> To: gnome-gui-list@gnome.org, wm-spec-list@gnome.org, jg@pa.dec.com
> Subject: Re: mc, mdi, mem
>              About this list
> -----
> >
> >    > 3. Fantastically slow due to the sucky nature of the X Shape
> >    > Extension.
> >
> >    I repeat again: we have conflicting data on the performance problems
> >    with shape...
> >
> >    Some people see this problem: some do not. From profile data, it looked
> >    to me quite possible there were window manager bugs that were causing
> >    excess ConfigureWindow operations. Anything is slow if you do it much
> >    more often than necessary.
> 
> Using XShape is slow beyond usability when the WM has opaque window move
> and resize enabled. (I like them both very much). Lots of
> configure/expose events are generated during such operations.
> 
> >    IF there is a problem, then it is quite possible to work on region
> performance
> >    in the server, with the only caveat that it is painful to debug if you
> >    get things wrong.
> 
> I am not sure XShape can be improved by orders of magnitude (without
> serious memory overhead). I have recently tried WindowBlinds (for
> Windows) that has E-style shaped themes and it is about as slow as X.
> 
> >    I again make a plea for some serious effort to go into proper analysis
> >    of this problem rather than the mach-1 handwaving that has gone on to
> date.
> 
>

Data, data, data we need real data: you are contributing to mach-1 
handwaving...

The first thing needed is to understand whether the number of Configure 
Window requests is reasonable or not: in the profile I looked at from 
someone who sees the problem, it sure looked like the window manager was 
doing many more Configures than necessary.  Do anything many times more 
than needed and will cause things to feel slow.

As far as I know, this is the only data we have seen on the problem to
date: but I can't answer the question of whether what the window manager
is doing is at all reasonable (particularly since I can't reproduce
this performance problem).

Even then, we need to understand why some people see the problem,
and others do not: it may just be how people are using their desktop,
or it may be other bugs (say in the device dependent code of the server)
that are causing performance problems.  It doesn't seem to happen
on my machines, for example.

If it is how people are using their desktop, it is likely the region code
that the Shape extension uses: this can be fixed (though it is code
that must be touched carefully; I still have painful memories of X11
development and the region code.

If it is not, then we need to get the DDX code fixed on those DDX's.

Waving hands and saying SHAPE is slow isn't contributing to localizing
the problem, which everyone agrees SOME people (maybe even MANY people)
see.

Let's start systematically analyzing the problem rather than handwaving.
					- Jim