ietf-nntp DEBUG command & x9x (was: 9xx)

[Russ Allbery]

Chin Chee-Kai <cheekai at SoftML.net> writes:

> If I recall correctly, Russ made a suggestion (Message-ID:
> <ylu2de5dfd.fsf at windlord.stanford.edu>) on using x9x for any and all
> private uses (including debugging) and x8x for authentication, to which
> Andrew also agreed.

I think this is the best approach even though it differs from RFC 977.
Authentication is a pretty huge part of the protocol; a lot of sites use
it and would find NNTP completely unusable without authentication.  I
think it therefore passes the level of critical problem that justifies
changing RFC 977 based on existing practice.  (I entirely agree with Stan
that we shouldn't do this lightly.)

> I pointed out the lack of specifying what a client should behave on
> receiving x9x messages in both RFC977 and this draft (Message-ID:
> <Pine.LNX.3.93.1000725120021.1647B-100000 at one.softml.net>), and
> suggested that clients treat all x9x responses as 500 errors.  Clive
> commented that its a particular case of a more general case of what to
> do when client confronts an unexpected response.  Clive also suggested
> to make a best guess based on first digit.

The only problem that I have with making decisions based on the first
digit is what to do with commands that normally return multiline
responses?  If you guess that an unrecognized code is actually the
expected return (the beginning of a multiline response) and that code
isn't actually beginning a multiline response, you get into protocol
trouble.

In retrospect, it would have been nice if there were a Nxx message class
reserved specifically for introducing multiline responses, but
unfortunately we don't have that.

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