--On Sunday, January 25, 2009 0:14 +0000 John Levine
A friend* wonders about the SMTP treatment of addresses in
TLDs. Historically there have been a smattering of such
addresses, most famously n(_at_)ai, and there are 21 TLDs with MX
records, but spot checking shows that most appear to be config
mistakes, with the MTAs not actually accepting mail to
If ICANN does what they claim they're going to do and starts
selling thousands of new TLDs, we're likely to see a lot more
dotless addresses. bob(_at_)aol, anyone?
As best I understand it, the prose description of addresses in
2821 allowed TLDs in addresses, but the ABNF didn't. In 5321
the ABNF changed to match the prose. Although the DNS allows
you to put a dot at the end of a domain name to make it clear
that it's an absolute address, SMTP has never allowed that.
Correct on all counts. In the 2821bis work, the possibility of
changing the syntax to match the prose was considered too. That
change was actually made to the document at one point. The
group then decided it was a bad idea and the change was un-done.
My impression is that in the real world, mail addressed to
postmaster(_at_)va is far more likely to be rewritten to
postmaster(_at_)va(_dot_)example(_dot_)com than delivered to the Vatican
(which, as it happens, will accept it.) Mail addressed to
postmater(_at_)va(_dot_) with a dot isn't valid under 821, 2821, or 5321,
so some MTAs accept it, some reject it.
Yes. And, again, the decision to leave it that way was quite
extensively reviewed and discussed during the run-up to 5321.
FWIW, we did explain the situation to ICANN at the time, and
took a little poll to find out which TLDs thought they might be
hurt if there was no way to send mail from some conforming
clients to mailboxname(_at_)TLD(_dot_) We found out that a few TLDs tried
to permit it (as you obviously know already). But no one,
including those TLDs, seemed very concerned if it didn't work.
I have no way to know whether ICANN remembers that discussion as
they launch into the world of more TLDs than they can easily
keep track of.
Is the current situation with TLDs deliberate, or was the
change in 5321 just aesthetic tidying up? Is there any reason
not to permit a trailing dot in 5321-bis to bring it in line
with DNS rules?
Both pieces of this -- requiring a domain name at the second
level and lower and not permitting the trailing dot -- were
explicit decisions after consideration of relevant
implementation, installed base, and operational issues.
It is not a matter of DNS rules, either. The DNS specs (and
1123) say, essentially, "there has to be a way to distinguish an
FQDN from a short name, and a trailing period is such a way".
The SMTP spec says (again, not in these words) "if it is a
domain name, and you put it on the wire, it MUST be an FQDN".
So there is no need for the trailing period in SMTP and it is
conformant with the DNS rules. That is, of course, not a reason
to prohibit it. Others can probably explain that one better
than my memory permits right now, but I think history and
existing implementations had a lot to do with it.