Subject: Re: uugetty for NetBSD
To: der Mouse <mouse@Holo.Rodents.Montreal.QC.CA>
From: Greg A. Woods <>
List: current-users
Date: 11/03/1996 20:17:49
[ On Sun, November  3, 1996 at 19:13:38 (-0500), der Mouse wrote: ]
> Subject: Re: uugetty for NetBSD
> Speaking personally, I would say it didn't do it because it shouldn't
> do it.  If you really must use the same hardware for dialin and
> dialout, you should be running getty on the dialin device and uucp on
> the dialout device, and let the kernel do the interlocking _right_,
> instead of half-baked userland schemes that can go wrong in far too
> many ways.

Speaking from hard-won experience, I can very strongly say the *many*
of us would prefer to keep such locks out in user-land.  I've yet to see
a kernel that can properly handle all the stupidities various kinds of
hardware (i.e. modems and UARTS, etc.) can throw at it.  The day async
driver writers can manage to write a truely robust state machine that
both properly implements RS-232 (or whatever) state transitions, *and*
can handle all the illegal transitions that might possibly hget thrown
at it, then and only then will I trust such things in the kernel.  I do
not ever want to have to contemplate rebooting just because some stray
signal threw the tty driver into a fit.

At least when user-land locking schemes go wrong, or are not quite
robust in the face of un-expected events, they can quickly and quietly
be fixed without interfering with any other current activities on the

							Greg A. Woods

+1 416 443-1734			VE3TCP			robohack!woods
Planix, Inc. <>; Secrets Of The Weird <>