Newsgroups: news.software.b
Path: utzoo!henry
From: henry@utzoo.uucp (Henry Spencer)
Subject: Re: cancel propagation 
Message-ID: <1989Sep6.214313.27685@utzoo.uucp>
Organization: U of Toronto Zoology
References: <5200@looking.on.ca> <536@logicon.arpa> <3246@deimos.cis.ksu.edu> <1989Aug30.174430.20687@anise.acc.com> <1989Aug31.034105.2177@utstat.uucp> <6233@looking.on.ca>
Date: Wed, 6 Sep 89 21:43:13 GMT

In article <6233@looking.on.ca> brad@looking.on.ca (Brad Templeton) writes:
>You never needed to send the cancel message to the downstream sites, because
>you *know* you are never going to feed them an article with the specified
>message-id.  Why do they need to keep it around in their databases?

Brad, you are assuming that the feed system is an acyclic graph composed
entirely of well-behaved links.  RFC1036 makes the same assumption.  The
assumption is wrong.  The real feed structure is *full* of loops, back doors,
sneak paths, intermittent links, unidirectional links, partial feeds, and
other perversions.  Not to mention a significant incidence of dropped
articles.  The issue is not theoretical adequacy, but practical robustness
in a messy, unpredictable, and very non-ideal world.
-- 
V7 /bin/mail source: 554 lines.|     Henry Spencer at U of Toronto Zoology
1989 X.400 specs: 2200+ pages. | uunet!attcan!utzoo!henry henry@zoo.toronto.edu