Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Posting-Version: version B 2.10.2 9/18/84 SMI; site sun.uucp
Path: utzoo!linus!philabs!cmcl2!seismo!harvard!talcott!panda!genrad!decvax!decwrl!sun!guy
From: guy@sun.uucp (Guy Harris)
Newsgroups: net.news
Subject: Re: Suggestion for checksum on articles
Message-ID: <2464@sun.uucp>
Date: Sun, 21-Jul-85 18:47:42 EDT
Article-I.D.: sun.2464
Posted: Sun Jul 21 18:47:42 1985
Date-Received: Wed, 24-Jul-85 07:26:24 EDT
References: <514@oliveb.UUCP>
Distribution: net
Organization: Sun Microsystems, Inc.
Lines: 35

> The only hard part of this comming up with a checksum that is portable
> across the wide varity of machines on the net.

Given that the vast majority of UUCP connections on USENET run over serial
lines using the checksummed "g" protocol, I think we have an existence proof
for such a protocol.  We may not want to use that protocol, though, due to

	1) trade secret restrictions, although Lauren Weinstein could
	   just tell us what the algorithm is

and

	2) the fact that at least some implementations of it within
	   UUCP don't compile into correct code with some C compilers.

To make the checksum portable:

	1) always checksum bytes, not anything larger - but since we're
	   processing text here, it's unlikely that anybody'd go through
	   the trouble to accumulate two bytes and do checksumming on two-
	   byte quantities.

	2) don't use "native" arithmetic operations like addition and
	   subtraction; there is, I believe, at least one Univac 1100 on
	   the net and it's a one's-complement machine.

	3) don't assume that characters are signed or that they're
	   unsigned - there are lots of signed-character VAXes on the net
	   and there are lots of unsigned-character 3Bs on the net.

	4) don't assume that casts will be done in the order you think
	   they will - I believe that was the cause of lots of porting
	   problems for the UUCP "g" protocol checksum code.

	Guy Harris