Path: utzoo!mnetor!uunet!lll-winken!lll-tis!ames!pasteur!ucbvax!LARRY.MCRCIM.MCGILL.EDU!philipp
From: philipp@LARRY.MCRCIM.MCGILL.EDU (Philip Prindeville [CC])
Newsgroups: comp.protocols.tcp-ip
Subject: Re:  Checksums
Message-ID: <8804031958.AA05134@Larry.McRCIM.McGill.EDU>
Date: 3 Apr 88 19:58:59 GMT
Sender: usenet@ucbvax.BERKELEY.EDU
Organization: The Internet
Lines: 20

[ My apologies for following this up several days late.  I am only now
  starting to catch up on my mail. ]

It seems to me that a lot of the discussion over the last couple of weeks
has been about designing an error detection that does protocol verification
(or aids in debugging, anyway) by doing index-off-by-one detection.  Now
this is a neat idea (on paper), but I wouldn't want to implement it (at
least not entirely in software).  Or if I did, I would do it in a protocol
analyzer, not in a production implementation...

I still believe the greatest aid in finding index-off-by-one and similiar
problems is not done by building elaborate debugging into an implementation
in an on-line environment (eg. the Internet), but in the TCP/IP bake-offs.
I just wish more implementors would perform complete tests (options and
everything) against a real variety of implementations before going to market.

"Do I work against a Sun?  Check.  Do I work against a Wollongong?  Check..."

In a perfect world...

-Philip