The Mouse Detective :-)

This is an incomplete program for mouse detection. The idea is to
use it in some kind of dialog that says: try to move your mouse and
click on the OK button. If this doesn't work, enter return [the
keyboard focus is on cancel].

Use it like this:

	$ ./mdetect -v
	Found the following devices:
	   /dev/ttyS1
	   /dev/ttyS0
	   /dev/psaux
   [start moving mouse]
	microsoft:     105 102
	logitech: LM  -27 27
	microsoft: L   6  -43
	logitech: LM  93 191
	microsoft:     7  -7
	Detected microsoft mouse on /dev/ttyS0

You can see it switching between logitech and microsoft mode until it settles
on microsoft mode. Invoking it without the -v switch disables all output
except for the `Detected ... mouse' line.

Currently, the following mice are reported:

	psaux		PS/2
	mousesystems	MouseSystems compatible
	microsoft	Old Microsoft protocol
	logitech	Some (old) logitech mice
	mouseman	MouseMan (microsoft with 3/4 packet extension)

Here's a framework that I think might work

 1.	Start the mouse detection program
 2.	Start X server, display dialog
 3.	User starts moving the mouse
 4.	Detection program tries to understand what the mouse
	tries to say (this is stage1). It should arrive at a
	decision within 1 sec.
 5a.	Detection program has decided on a driver. Enters stage2,
	where it writes the mouse packets it received into a fifo
	(but now, in a mouse-type independent packet format). This is
	basically what gpm -R does.

	Pointer starts to move on the screen.

 5b.	Detection Program wasn't able to detect mouse type. It doesn't
	do anything. Pointer doesn't move, user has to select mouse
	manually.

NB: How does XF86Config set the mouse type? I assume there's an X
extension?  If so, in step 5, we can also tell the X server what mouse
to use (if we know what to use), and exit. This is a lot better than
the fifo approach.



Mode of Operation

What the program does is open all available devices for reading.
Whenever it gets a byte, it lets all drivers (currently psaux, microsoft,
logitech) have a look at it. If the driver can make sense of the
packet, fine. If any driver is able to successfully decode 20 mouse
packets in a row, it is considered _the_ driver.

Serial mice are a bit special, since they use different line settings
etc. So there's a serial 'meta' driver that sets the line characteristics
to those of each driver in turn.

Currently, I'm able to detect either ps2 or an M$-compatible mouse
fairly reliably. I haven't tested the logitech code. If you don't jiggle
the mouse too hard, detection is faster; otherwise, it will start with
a few false guesses before settling on the right one. Good packets are
those that can be decoded, have reasonable dx/dy values and no mouse
button pressed. (this might be improved by looking at the second
derivative, so to speak: changes in dx/dy values sent by the mouse
shouldn't be too large).

The question is whether this is really worth the trouble. Maybe we will
have as high a success rate if we use ps2 when able to open the device;
and ttyS0 with the M$ protocol otherwise.