John C Klensin wrote:
First, if there were any disagreement between 3696 and 2821,
then 3696 is simply wrong. For that case, one should heap
abuse on the careless author and try to organize an effort
to do a revision.
A "3696bis" integrating its errata should be simple if the
careful author kept his XML source. That author probably
knows the magic spell to get an FYI number for 3696bis (?)
Seriously, IMO there's an intentional disagreement between
2821 and 3696: The latter says how it _should_ be done in
theory, but for 2821 you picked another approach for the
reasons stated by Ned.
Then you and others here considered to fix this in 2821bis,
but I think Ned claimed that it won't fly for MTAs treating
a trailing dot as SMTP syntax error.
And there are certainly no leading, adjacent, or trailing
dots in a 2822 <dot-atom-text>. Admittedly (2)822 allows
TLDs, but without a trailing dot.
Ned quoted RFC 1123 5.2.18 in this thread (2006-04-22):
| Some systems over-qualify domain names by adding a
| trailing dot to some or all domain names in addresses
| or message-ids. This violates RFC-822 syntax.
We could argue forever about this, some folks elsewhere
even think it's a good idea to modify ABNF discussed for
years in AUTH48, but actually you've only three choices:
1: Allow the trailing dot everywhere no matter what 2821
and 2822 say. Maybe say "SHOULD have a dot" for TLDs.
2: Keep it as is in 2821, no trailing dots, and one dot
required. This excludes TLDs, unlike (2)822.
3: Same as (2) with a special rule for TLDs: TLDs MUST
have the trailing dot, other domains MUST NOT have the
trailing dot.
The 3rd choice is not KISS, and it has a high astonishing
factor. Excluding TLDs in the 2nd choice is also somewhat
astonishing, my question about it (2004-11-12) was if this
is a typo, and you said "intentional". 1st and 2nd choice
are KISS.
Frank