Re: #1416 Reinjection - an attempted summary, and a suggested resolution

[Russ Allbery <rra@xxxxxxxxxxxx>]

Charles Lindsey <chl@xxxxxxxxxxxxxxxx> writes:

> Yes, that's a fair summary of my position. For sure, a 3-day limit is
> far too short. Current practice appears to be much longer than that, and
> I can easily imagine situations where injection might be delayed that
> long after composition of the article. And currently (so far as I can
> tell) propagation does not deteriorate until something closer to 8 days.

3-day limit?  Oh, hm:

   3.   It SHOULD reject any article whose Date header field is more
        than 24 hours into the future (and MAY use a margin less than 24
        hours).  It SHOULD reject any article whose Date header appears
        to be stale (more than 72 hours into the past, for example, or
        too old to still be recorded in the database of a relaying agent
        the injecting agent will be using) since not all news servers
        support Injection-Date.

That's really not saying what I wanted to say.  I think I borrowed that
language from something else (probably the staleness checks for
Injection-Date).

What I really wanted to say there was something more like:

    It SHOULD reject any article whose Date header is too far into the
    past (too old to still be recorded in the database of a relaying agent
    the injecting agent is using, for example).  This cutoff SHOULD NOT be
    any shorter than 72 hours into the past.

In other words, the 72 hours was intended to be a minimum, not a suggested
value.

I know you probably still disagree, but I didn't intend to be that harsh.

> I disagree there. Posting agents have no concept of which identity model
> thet are using.

I don't agree with this at all.  I think that you're selling posting
agents short.  Many posting agent authors know very well what identity
model they're using and know that Date plus Message-ID make up the
message's identity, and provide header fields appropriately.  We will
cause problems if we break this.

> Current practice has often been that they do not set the Date, and leave
> it to the injecting agent to set it.

This will continue to work as expected in my proposal (just to be clear)
since in that case the injecting agent can add an Injection-Date as well
(and indeed MUST).

> But that preactice is already changing, since mail user agents tend to
> write the Date at composition time, and lots of mail user agents are
> also news user agents (or think they are :-( ), and hence do the same
> thing.

I think the point of disagreement here is my belief that maintaining the
current identity model and not breaking posting agents that are using it
intentionally is more important than making use of Date for composition
date quietly work without changing posting agents.

-- 
Russ Allbery (rra@xxxxxxxxxxxx)             <http://www.eyrie.org/~eagle/>