Newsgroups: news.software.b
Path: utzoo!utgpu!watserv1!watmath!looking!brad
From: brad@looking.on.ca (Brad Templeton)
Subject: Re: sys file entries for CNEWS
Organization: Looking Glass Software Ltd.
Date: Sat, 23 Mar 91 22:09:33 GMT
Message-ID: <1991Mar23.220933.27563@looking.on.ca>
References: <1991Mar18.170511.6535@st_nik!swindon.ingr.com> <1991Mar19.224950.4653@zoo.toronto.edu> <449@skyking.UUCP> <1991Mar23.193554.25944@looking.on.ca> <AD}=J-%@rpi.edu>

Ok, I admit that sloppy was a bit of a nasty word to use, and in checking
myself, I see I have used all for internal gateway lines in the sys file and
"all,!this,!that" for a few of my single group feeds.   So lazy is the
right word.

As far as I am concerned I see no reason at all to propagate the
articles that say, "Distribution: everyone" or some other such mistake.
Perhaps the local inews program should check for this and warn the user,
but to have any typo or mistake go to the entire world is not the way
to do it in my opinion.

Long ago I pushed that distribution and group name should be disjoint.
This was part of the philosophy behind the great renaming, and the
introduction of the Distribution header field in the first place.

But we went half-way, because we still have these hierarcies that
match distributions.  There is a "ba" (bay area) hierararchy, as in
ba.this, and a "ba" distribution.

We should have split these.   C news took a partially right step by
splitting the newsgroup and distribution parts of the SYS file, but
it caused further harm by letting people conveniently slip "all" in
the latter field.   Under B news, you almost never put "all" in the
combined newsgroups&distributions field, unless you wanted a complete
and utterly full feed.

There is software that handles distributions completely distjointly
(except, sadly for control messages) and that is dynafeed.   In
dynafeed you list the groups you want explicitly, or with regular
expressions, and you list the distributions you want explicitly.
(No regular expressions, although I could do this if sites kept
distributions they carry in a standard place and format)

I had to kludge it a bit.  If you get the distribution "can" then
a distribution of "can.general" on an article is accepted.  And I
had to kludge it for control messages, which in ordinary news are
propagated because the name matches the name hierarchies listed.
(It's difficult in B and C, I believe, to get an individual selection
of comp groups and still get control messages for new comp groups.)

There's no easy fix.  I propose:

a) Completely disassociate distributions and hierarchies.  Rename one or
the other -- probably easiest to rename the distributions.

b) Define some new rules for the propagation of newgroup messages.  It
should be easy to say, "give me newgroups for groups that match the
following patterns without feeding me all the groups that match that
pattern.

c) Define a standard file like the active file that contains the
distributions and information about them, to be used by posting programs
and others.   This file should list distributions in a hierarchical fashion.

Newsclip uses a file called distlist which does this roughly.  It assigns
a numeric value to each distribution that is an approximation of the
number of people who read it, thus ordering them to some degree.  In
retrospect, this is not perfect, because it doesn't tell us how to
combine conflicting hierarchies that a site exists within.  UUNET would
be a royal mess since it lives in every distribution it can get its
hands on.

-- 
Brad Templeton, ClariNet Communications Corp. -- Waterloo, Ontario 519/884-7473
