Bruce Lilly wrote:
With optional FWS you get ASCII art. With LWS you would
get "apparently empty lines"
No, you get a list with empty elements
LWS in 2068 ff. is [CRLF] 1*( SP | HT ), and "implied *LWS"
means zero or more LWS. That's the same as LWSP in 4234bis:
| LWSP = *(WSP / CRLF WSP)
A sequence of zero or more SP, HT, CRLF in any order, where
each CRLF is followed by either SP or HT.
That means "apparently empty lines" CRLF SP CRLF or similar
can occur within LWSP or *LWS. Such constructs are far too
fragile for an environment such as mail header fields sent
via multiple relays and passing through dubious software
doing "something" with messages.
It's no problem where "really empty lines" CRLF CRLF have no
different effect from "apparently empty lines". But that's
not the case for mail, where a "really empty line" indicates
the end of the header.
Like NUL in various obs-rules the LWSP in <obs-FWS> *cannot*
work in practice, it was never a good idea, 30 years ago or
in 2008.
You don't neeed to quote 2821-rules about not modifying the
header in transit, everybody knows this. And apparently some
folks feel that "mailing lists" are entitled to ignore RFCs
as it pleases them.
IMO you can't get rid of "sh*t happens" by a MUST NOT, that
would be irresponsible if you know that it will nevertheless
happen, as a plausible bug.
Your hypothetical "something" is nonconforming and
non-interoperable.
IBTD, we wouldn't need 2822(upd) if RFC 822 was good enough.
Frank