Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Posting-Version: version B 2.10.2 9/3/84; site enmasse.UUCP
Path: utzoo!watmath!clyde!burl!ulysses!allegra!mit-eddie!genrad!panda!enmasse!dave
From: dave@enmasse.UUCP (Dave Brownell)
Newsgroups: net.unix-wizards,net.lan
Subject: STREAMS
Message-ID: <480@enmasse.UUCP>
Date: Fri, 11-Oct-85 10:38:52 EDT
Article-I.D.: enmasse.480
Posted: Fri Oct 11 10:38:52 1985
Date-Received: Mon, 14-Oct-85 03:26:23 EDT
Reply-To: dave@enmasse.UUCP (Dave Brownell)
Distribution: net
Organization: Enmasse Computer Corp., Acton, Mass.
Lines: 24
Xref: watmath net.unix-wizards:15211 net.lan:1085
Summary:  Jeez -- how about USER programs?

In article <455@basser.oz> boyd@basser.OZ (Boyd Roberts) writes:
> ...  Sockets are all wrong.  You don't need that mess in your kernel.

As the person who originally posted the query, I can't help but be
amused by what's come out of it.  It seems like most people here are
more concerned with elegance in the kernel than in user interfaces.

So far as I'm concerned, the ONLY interesting thing about UCB sockets is
the user interface:  the socket() system call and its friends connect(),
bind(), accept(), and so on.  With these standard routines, people can
finally write reasonably protocol-independant applications.  (What sells
computers, by the way -- elegant kernels or good applications?)

Kernel implementation -- either by addmittedly flexible streams, or
BBN/UCB socket code -- is immaterial except insofar as it provides
performance and maintainability.

In all the foofaraw, I STILL haven't seen any discussion about how STREAMS
provide a good user interface.  (Do they?)  I think I'd put the socket
system calls on top of them, myself ...
-- 
David Brownell
EnMasse Computer Corp
...!{harvard,talcott,genrad}!enmasse!dave