Re: FWS problem

[Frank Ellermann <nobody@xxxxxxxxxxxxxxxxx>]

Charles Lindsey wrote:
 
> I am not sure how we ever got away from square one.

I wanted the ABNF to reflect your "MUST contain at least one
non-whitespace", and Bruce claimed that this is not the case
for "name: SP CRLF SP body CRLF".  But now "we" (you and me ;-)
obviously agree that this is really illegal.
 
> we have always concluded it needs to be so.

Okay, maybe it's a "we" excl. Bruce, and I can live with it,
if "we" (= you ;-) replace all corresponding [FWS] by *WSP in
the ABNF.

> It has indeed been so since Son-of-1036.

You can always convince me with this line.  But s-o-1036 says:

| header          = start-line *continuation
| start-line      = header-name ":" space [ nonblank-text ] eol
[...]
| space           = 1*( <HT (ASCII 9)> / <blank (ASCII 32)> )

I don't see any mandatory SP in this syntax, I only see a WSP.

Are you absolutely sure that we want to go back to the "blank"
in 1036 ?  It's the old problem of "does RfC 2822 update 1036
in some way ?"  IMHO a "name: 1*WSP body CRLF" is better than
your "name: SP *WSP body CRLF".

>> should be "name: SP *WSP header content CRLF" everywhere.

> Maybe for our own news headers, but in some cases the syntax
> for the [*WSP header] bit comes straight from RFC 2822, and
> so it is not ours to change.

Let's fix it at least for our headers, replace [FWS] by *WSP
before and after any header content of "our" headers.

>> The syntax "name: SP [FWS] header content CRLF" is wrong,
>> because "name: SP CRLF SP header content CRLF" is illegal.
 
> I don't think this prohibition can be made by syntax. The
> corresponding provision in RFC 2822 (the one against empty
> folded lines) was not made by syntax, and we are following
> RFC 2822 wherever possible as a matter of policy.

We're following the _intent_ of RfC 2822 whereever possible,
but you're not forced to copy inaccurate syntax.  If there's a
way to get it right in the ABNF, just do it please.  And as a
matter of policy I'm not happy with making a TAB illegal in a
place where it's allowed by both s-o-1036 and RfC 2822.

> The draft makes it perfectly clear what the rule is, and that
> it is different from RFC 2822.

Yes, but if message/news is a proper subset of message/rfc822,
then why remove it from the IANA registry ?  It's a practical
problem at least for gateways and moderated newsgroups, or in
your terminology for all injecting agents:

For "name: HT CRLF HT body CRLF" the injecting agent is forced
to format it as "name: SP body CRLF" somehow, and that's not
always trivial, e.g. what about a field body starting with a
maximal length encoded word ?
                                Bye, Frank