[ietf-nntp] OVER message-id

[Russ Allbery]

Ken Murchison <ken at oceana.com> writes:
> Clive D.W. Feather wrote:

>> At some point we put "OVER message-id" back into the protocol.  Andrew
>> from Supernews is objecting to this, saying that it's impractical in
>> "every major server design".

> Just out of curiosity, if its impractical in "every major server
> design", shouldn't more than one person be voicing displeasure?  I'm
> sure Russ would've shot this down earlier if INN would have issues with
> it.

Implementing it is a pain because the user is requesting overview
information and overview information is not indexed by message ID.  Really
the only way that you can implement it under most server designs without
adding another overview index that's almost never used is to retrieve the
article headers and cobble together the overview on the fly.  That was how
I was planning on implementing it.

Andrew's cluster has an additional problem that Xref data isn't actually
stored in the article, so he simply doesn't have any good way of answering
an OVER Xref request by message ID.  This is, I believe, somewhat unique
to his software and is not typical of most server designs, but I could be
wrong.  I don't have a lot of familiarity with how Xref works across
servers.

Anyway, given that OVER on message IDs is an innovation of this working
group that we added based on a few people saying it would be useful rather
than based on existing practice, and given the stage that we're at with
this draft, I think we should just drop OVER on message IDs again and not
spend a bunch of time on this.  Servers can easily add it as an extension
since it doesn't conflict with the base command grammar, and if it turns
out to be a common extension, we can revisit it in a future standard
revision.

>> He suggests, instead, an extension to NEWNEWS that would return
>> overview fields along with the message-ids.

> How does this help?  This is overkill if I only want the overview for a
> single article.

You rarely ever want that.  The reason why people would want to run OVER
on a message ID is precisely when they're using message-ID-based article
retrieval, which in practice means they're using NEWNEWS.  So Andrew's
proposal is for a NEWNEWS-like command that returns message IDs plus
additional overview information.

This helps with the server implementation because you no longer need to
map message ID to article; NEWNEWS is generally driven by the overview
database itself or by some other database that has indexes into the
overview database, so you can get overview information and message ID all
at the same time.

-- 
Russ Allbery (rra at stanford.edu)             <http://www.eyrie.org/~eagle/>