> Another way for a user level program to signal to the kernel
> that it wants different behaviour from a device is to set a
> bit in the flags to the open(2) call.  This seems cleaner to me
> than cluttering up /dev with ever expanding numbers of device
> entries.  To my way of thinking, one physical device should have
> one entry in /dev.  If you want the open to behave differently you
> pass a flag to open(2).

Excuse the sarcasm, but:

Great idea!  It's _much_ better to have infinitely expanding numbers
of open arguments!


Solving this particular problems with device nodes is dumb, but if
that's true, then doing it with an open() flag is a large factor
dumber.  (The 'large factor' is the number of meanings for spare
device node minor number bits than you can have if you use open
bits...  and don't forget, different types of devices can use those
device number bits in different ways, _and_ one day NetBSD should
migrate to 32-bit devices, hopefully split 12/20.)


It seems to me that:
	(1) this should be handled on a case-by-case basis, and
	(2) since there's an obvious analogue in this case for
	    non-permanent settings (i.e. the ioctls used to set
	    current termios state), if the ability to set default
	    device parameters is desired then it would make the
	    most sense to use ioctls similar to the termios ones to
	    do it.  That way, you could even make stty do the
	    right thing easily.


cgd