Thank you all for very insightful discussion.
My conclusion for the initial question is that we need to fully support
single-component domain names and ABNF production in RFC2821 is an error.
Supporting single-component domain names should not be a problem as long as it
is clear from the context that this is indeed FQDN and not an abbreviation.
For the trailing dot in a domain name, I understand that accepting it can be
desirable, although it is legal/fine to be stricter and reject such domain
names as invalid. Of course it must never be produced.
Please let me know if my interpretation is incorrect.
From: John C Klensin [mailto:john+smtp(_at_)jck(_dot_)com]
Ack. Unless others disagree with Ned's (IMO extremely coherent)
analysis, I'll pull the changes out of 2821bis.