> Why push this complexity into the kernel?  Things like the X server
> already need to be able to cope with different streams of input, etc.
> It seems to me that it would most logically be handled there.
Well, it's always a problem where functionality should go.
You could argue that the mouse protocol decoding should be
done in the X server (as it partly is today) rather than in
the kernel, but the current NetBSD trend is to move that to
the kernel.  I think this situation is similar.  Yes, the X
server could try and open /dev/wmsN (with N = 0 .. 15 say)
regularely, and take care of new devices that show up.
But personally I think putting this kind of functionality 
in the kernel more pleasing.  Too bad our configuration code
can't do it in a simple way.

        -- Lennart