Charles Lindsey wrote:
The whole of the <msg-id> syntax in USEFOR is a "mess",
and we knew it was so when we wrote it.
Well, it used to be completely wrong for some years, simply
copying whatever RFC 2822 said without checking very basic
requirements like "does it work anywhere, e.g. with NNTP ?"
After it was clear that the USEFOR WG didn't bother for six
years to get at least this fundamental concept for NetNews
right, the Message-ID syntax was fixed to reflect a minimal
subset of RFC 2822 still expected to survive NNTP: Killing
NO-WS-CTL and unnecessary <quoted-pair> everywhere, keeping
fictitious <no-fold-literal>s with \[, \\, or \], because
RFC 2822 demanded that they are "valid".
As far as that was still messy it was the RFC 2822 mess, and
IIRC you wanted it in the spec. Similarly some wanted [FWS]
in places where it made no sense, until I posted an erratum
about one pointless [FWS] in RFC 2822, and it was pulled.
If you are now ready to pull also \[, \\, and \] please do,
but don't introduce a new bare backslash in domain literals
while you're at it.
I hope we will find some way (AUTH48 or otherwise) to fix
USEFOR so as to match <msg-id> in 2822bis exactly.
I'll believe it when I see it, confirmed by the responsible
area director. It took ten years so far, what the approved
RFC says is already supposed to work, plausible Message-IDs
up to very ugly ideas match the approved syntax, while worse
constructs breaking NNTP won't match.
My wording will still make an improvement in
interoperability in 99% of cases, and I don't think we are
going to persuade this List to go beyond that.
I've no idea how you determined that 99% of all cases wihout
SP after the colon are a missing "magic SP". The only cases
where I ever did this in e-mail were related to colon-CRLF-SP,
or in other words folding, e.g. when that's reqired for long
2047-encoded words due to 2047 line length limits. An issue
also relevant for 2231-encoded words with long 4646 language
tags, trapped in its "magic SP" heritage USEFOR ignored this.