[Top] [All Lists]

Re: domain name definition in RFC2821

2007-03-20 09:59:21

John C Klensin wrote:
This is a bit more complex than you suggest.

Oops.   We're discussing it for about three years
now after it somehow showed up in a USEFOR debate,
no problem if it takes some more time.

[[ But I do want it settled before 4408bis :-) ]]

I've reread the relevant notes.

Me too, Ned's main argument was apparently this,
almost a year ago:

| Considerable effort has been expended to switch away from,
| say, ned(_dot_)freed(_at_)west to FQDN forms like 
| But an essential ingredient in making this work is to 
| canonicalize short form names into FQDNs. And this applies
| not only to email but also to web services.
| If you told the folks in SunIT that, say, postmaster(_at_)ws
| needs to work "correctly" my guess is they'd just about
| fall over laughing. And they'd only laugh harder if you
| pointed out that the current behavior is a standards
| violation.

I do not believe that there is any remaining question
about trailing periods on multiple-label FQDNs.  They
are prohibited.

+1 wrt SMTP (wrt SPF policies it's a different story)

There is also no question about whether anything other
than an FQDN is permitted between SMTP servers on the


The question is how the email address
is to be interpreted.

Yes, or a similar "HELO foo" (no dot), or "EHLO museum." 
(with dot).  At some point in the recent thread on the
SPF list we almost arrived at the compromise that TLDs
MUST have a trailing dot (inspired by 3696 and 2821bis).

the intent of the original remark in 2.3.5 was to permit
for the one-label case only.

Cando (wrt SPF errata), if the <target-name> is anyway
underspecified there's some wiggle room.  Too late for
SPF's <domain-spec>, and USEFOR also treats TLDs as not
good enough for a "non-legacy" (non-UUCP) <path-identity>
of a news server.

(1) I replace the existing text in 2.3.5 by a comment
reminding people that single-label domains MUST NOT
appear between SMTP servers unless they refer to TLDs
and that any implementation intending to use single-label
domains as abbreviations must be sure to have a mechanism
for distinguishing between such abbreviations and FQDNs.

The only permitted way (in RFC 3696) to distinguish this is
to ask the root server, but that would be "suboptimal" for
crap like "HELO oemcomputer" (without dot).

(2) I tune the 2.3.5 text slightly to clearly permit the
trailing period on single-label domains, i.e., for TLDs

It's IMO more about "require" than "permit".

My instinct and understanding right now is that the first
option reflects the dominant current practice as we
understand it and ought to be the choice here.  Anyone
who disagrees should please speak up quickly and clearly.

I can't say that I disagree, but RFC 3696 "forbids" (as far
as that's possible in an informational RFC) to "hardwire"
TLDs in an application, with a compelling rationale.  If
the TLDs are not hardwired there have to be DNS queries.

Are you sure that you want this (and sure that the DNSOP
folks will let you... :-) ?