Re: Compatibility wit Netnews

Charles Lindsey wrote:

 [canonical Message-ID]
> The essential feature required for Netnews was that, in order to
> compare two <mesg-id>s for equality, a simple octet-by-octet
> comparison would always suffice. Such comparisons are at the very
> heart of the Netnews protocol, though they hardly arise at all in
> email except for the purpose of threading articles in MUAs using
> the <msg-id>s that occur in References header fields. Even there,
> I suspect, many threaders have simply implemented an octet-by-octet
> comparison.

+1  And it's not only about "comparing" Message-IDs, it's also about
using something that's allowed in NNTP, i.e. no raw control char.s
(= NO-WS-CTL), and no ">" even within <no-fold-quote> (quoted LHS) or
<no-fold-literal> (domain literal as RHS).

Another application are Message-IDs in mailing list archives used as
Archived-AT URIs, or Message-IDs in NetNews used in news: URIs.  It's
possible to percent-encode any NO-WS-CTL, but it would be a pain.

Percent-encoding is always a kludge and harms interoperability as
soon as applications try to percent-encode something that is already
partially percent-encoded for local purposes, and then let another
application try to undo this.  The PURL resolver is an example how
that can get too easily out of hand.

 [magic SP]
> The other principle difference is that, in each header field, an
> SP is obligatory after the ":". Since this is already the universal
> practice within email, I would suggest that there should be no
> problem in REQUIRING the same in the case of email, and to banish
> the possibility of omitting that SP to the obs-syntax.

I'd support a mandatory FWS after the colon.  But I can't support a
"magic SP" also in mail, IMO any significant "trailing white space"
is harmful.  Let's take an example header field:

   foo = "foo:" [CFWS] bar [CFWS] CRLF

Adding a magic SP results in:

   foo = "foo:" SP [CFWS] bar [CFWS] CRLF

Now the following folding would be syntactically wrong:

| foo:<CRLF>
| <WSP>bar<CRLF>

The NetNews RFC guarantees that it's never allowed to fold a header
field before the first non-blank character in the header field body,
here <bar>.  But 2822 doesn't guarantee this, it only says that a
folded line containing _only_ trailing white space isn't allowed.

And it's okay that 2822 doesn't guarantee this, sometimes it's very
convenient to fold a header field directly at the colon.  Extremely
long 2047-encoded words are an example, it would be odd to have a
special rule for the first line of a folded header field depending
on the length of the field name.

So I'd propose to add a mandory FWS to unstructured fields and the
<date-time>, i.e. all header field bodies not starting with [CFWS].

All other header field bodies start with [CFWS], and we could make
it mandatory, for the example above:

   foo = "foo:" CFWS bar [CFWS] CRLF

A magic SP matches CFWS, you can do what you claim everybody does
already.  An initial folding also matches CFWS, theerfore I can
fold header fields directly after the colon.  Admittedly this
still allows silly constructs:

| foo:(ugh)bar<CRLF>

OTOH it's no new ugliness, 2822 already allows this.  In essence
the only ugliness allowed by 2822 which would be "eliminated" by
a mandatory CFWS is this:

| foo:bar<CRLF>

> Life would be much simpler if Email could now be brought into
> line with Netnews in these regards (the present practices would
> just move to the obs-syntax, of course).

USEFOR had compelling reasons why it didn't copy all 2822 ideas.
I think the opposite is also true.

Frank