questions about a device driver

[Dan Christensen]

Everything below was tried on an IRIS 4D/120GTX unless otherwise stated.

I am working on connecting an input device to the Iris through a serial
port.  Conceptually, the device has several valuators and several
buttons.  I want to transparently let an application use this device
just as it uses a device such as the mouse, or the dial and button
box.  There are several possible methods of doing this, each with
problems.

1) Create new valuators and buttons for this device.  This would be
   an optimal solution but would probably require that the graphics
   library be recompiled, so I doubt that it is possible.

2) Put events onto event queues directly.  This would require knowledge
   of which process currently has the input focus and what its input
   channel number is.  This could probably be done with the routines
   gl_inchanget and gl_anyqenter, but unfortunately these routines are
   undocumented.  Can anyone explain what these routines do, what
   parameters they take, etc?  The problem with this approach is that
   polling the device wouldn't work, but I would be willing to sacrifice
   this.

3) Put events onto the NeWS event queue.  This is easy to do from
   PostScript but most likely the graphics library only filters out
   certain events from this queue to put onto GL clients' event queues.
   To change this would probably require that the graphics library be
   recompiled, so I doubt that it is possible.

4) Override existing valuators and buttons with setvaluator and
   gl_changedevice respectively.  This does work but has a couple of
   problems.  First of all, the message
   
     ERROR #23 in kernal: severity=2, sevaluator: <some number>
   
   appears in the console window once every couple of seconds, especially
   when the MOUSEX and MOUSEY valuators are being overridden and the
   tracker crosses a window boundary.  This only started to happen since
   we switched from an IRIS 4D/70G to our IRIS 4D/120GTX.  The second
   problem is that we are limited in the number of devices we can use
   concurrently since this uses up valuators and buttons from a limited
   supply.  Also, gl_changedevice is an undocumented routine, although
   it is used in /usr/people/4Dgifts/examples/devices/tabletd.c.

5) I found this section difficult to explain because I don't really
   understand how valuators work internally.  Maybe someone who does
   understand how they work will understand what I am saying.
   
   Override existing valuators and buttons using only gl_changedevice.
   Using gl_changedevice for both the valuators and the buttons
   eliminates the error message produced by setvaluator.  There are a
   couple of problems with this method also.  The correlation between
   the value sent to gl_changedevice for a valuator and the value that
   the valuator gets set to is difficult to understand.  Unless there
   is a routine like gl_readdevice, it is difficult to reuse an
   existing valuator.  For example, if a valuator can take on values
   between 0 and 100 and the gl_changedevice values that make it do
   this are between 1000 and 1100 (call this its 'device range') then
   when I try to set the valuator to 50 with gl_changedevice it gets
   set to 0 and its device range changes to 50 to 150.  I need to find
   out the difference between the gl_changedevice values and the
   setvaluator values and I don't know how to do this.  If more that
   one process is trying to control a valuator then using
   gl_changedevice seems out of the question.  If the device ranges
   they use are not identical, then the valuator will not be set
   correctly.

How should I do this?  If 1) could be done that would be great, but I
doubt it is possible.  2) looks feasable if I can find out how those
gl_* routines work.  3) doesn't look possible.  4) looks possible if
that error message could be removed.  I reported this to the SGI
Hotline and the person I spoke to couldn't explain why this was
happening.  5) looks tough.  4) and 5) aren't as good as the other
because they reuse existing valuators.  In the future will it be
easier to create new valuators?

Thanks in advance for any help.  I've been working on this for months.

----
Dan Christensen, Computer Graphics Lab,	         jdchrist at watcgl.uwaterloo.ca
University of Waterloo, Waterloo, Ont.	         jdchrist at watcgl.waterloo.edu