ietf-nntp Currently outstanding issues

[Russ Allbery]

Charles Lindsey <chl at clerew.man.ac.uk> writes:
> "Clive D.W. Feather" <clive at demon.net> writes:
>> Russ Allbery said:

>>> The exception isn't required for NNTP; I think we should drop it from
>>> this specification.  (The line has to contain some character, but
>>> whitespace-only lines don't break anything in the NNTP protocol.  If
>>> RFC 1036 or USEFOR wants to outlaw them separately, that's great.)

> No, I don't think that is right. NNTP has exactly the same problem as
> 1036 and Usefor here. This recently came up again on the Usefor list,
> and the opinion seemed to be that our previous decision was correct.

> The concern is headers that are folded in such a way that a line
> contains nothing but whitespace (and in News, there is the related case
> where the first line of the header contains nothing but the header-name
> and the colon plus its SP). Even RFC 2822 outlaws the first of those
> cases.

The reason to allow folding immediately after the header name, since this
is used in practice and lack of support for this in NNTP causes problems
with, e.g., mail to news gateways.

The main issue that came up were worries from Henry that trailing
whitespace could be stripped.

The current language allows folding *before* the space after the colon, so
there's no worry that space will be lost unless the next line is all
whitespace (which I'll cover below).  However, that means that one can
generate articles where the first line of the header has no space after
the colon (but will have a space after the colon after the header has been
unfolded).

I'm inclined to stick with this and tell people to fix their header
parsers; this is an edge case that different servers handle in different
ways and standardizing it at something sensible would be a benefit for
everyone.  Standardizing this particular algorithm would allow for easier
gating of mail to news, which I think would be good.

The case of a line containing nothing but whitespace is more problematic,
since if that whitespace is stripped, you're left with what looks like the
end of headers and the rest of the data is reinterpreted in ways that may
be badly incorrect.  I think the chances of this happening are somewhat
slim, but the breakage caused is much more severe.  I'm therefore content
to ban that and require that at least one non-whitespace character be on
every header line.

But I think we should stick with the new text for folding after the colon.

> Note also that Usefor does not permit folding before the SP following
> the colon. It might be safer not to permit it in NNTP. We know that some
> existing software is fussy about that SP (which is why we still insist
> on it). Can we be sure that existing NNTP software would not be
> similarly confused?

Worst case scenario is that the article will be rejected.  The breakage is
therefore fairly small, particularly given that USEFOR is going to
restrict this area further.

> Wouldn't is be simpler to say that MODE may be followed by any
> word-parameter. The only one defined in our document is READER, but
> extensions MAY define additional ones. If it is one the implementation
> does not recognize, it SHOULD/MUST ignore it.

LIST has the same issue.  Whatever we do for one we should also do for the
other.  I think we should either require that commands be registered or
use X, or we shouldn't; I don't like halfway solutions where regular
commands have to be registered but you can sneak in a command via MODE or
LIST without registering it.  It seems inconsistent.

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