Sorry for the off-topic discussion - I didn't expect it to take so
long...!

>    But RFC 1123 says as below, for example...?
> 
> => you mix up the MTA (which canonizes when it can *) and the MUA
> (which should accept any valid name).

I've always been talking about MTA and never said anything about MUA
in this thread.  Maybe my wording "illegal mail address" caught your
eyes - I should perhaps have said "causing confusion to a major MTA"
or something.

I also quoted the word "legal" in the first place, meaning that at
least some MTAs (such as many configurations of sendmail, i.e., those
with "O DontExpandCnames=False" which is the *default* in many
versions) indeed rewrite CNAME addresses, not only in envelopes but
also in contents, which is well-known to cause confusion.

> => you still mix up MTAs and MUAs. In fact, you suggest *your* MTA is
> wrong (:-)!

Well, as I wrote above, I've never talked about any MUA - I wrote "my
server at Penn" in my first message on this topic.  I agree that one
can argue that some versions/configurations of sendmail are wrong, but
they are very common at least.

>    (I wonder what Dmitry meant by
>    "webmaster@caml.inria.fr is not routable"...)
>    
> => me too.

Hey, you are _so_ picky about his words in your reply...!:-)

| => no E-mail address is routable... What do you mean ? The host
| caml.inria.fr is routable, i.e., the host which has the (alias) name
| caml.inria.fr has a routable IP address (128.93.11.23).

> => but they are not more illegal. BTW the Sender-SMTP is likely a MTA.
> Today the canonization is considered only as useful (if it is not performed,
> the next MTA will have to resolve the alias again) but no more as mandatory
> or necessary.

I can agree that they are no longer stated as illegal in RFC 2821, but
they were in RFC 821 (and 1123) - and there are still many systems
which conform to the latter.

> => I argue that your MUA should accept what you give as soon as it is
> valid and it knows what to do with it.

I agree, but again I've never talked about any MUA.

>    The bottom line is that it is still a bad practice to use aliases in
>    mail address domains for these reasons.
> 
> => not only it is not a bad practice (the first agent on the path can
> resolve the alias, before RFC 2821 it was the first agent using SMTP at
> its sending side) but it is a very common practice. What do you believe
> aliases are for?

I've never said (or even thought) that aliases are bad by themselves -
I'm just saying that aliases *in the "domain" part of mail addresses*
often cause confusion for the reasons above.  You can find many tips
on this issue if you google "sendmail CNAME" or something.  We might
call it a bug of some versions/configurations of sendmail, but they
happen to be still common, unfortunately.

> => I don't understand this comment about A RRs (do you suggest to use
> only litterals?)

> => IMHO MX RRs are a nice idea. With HTTP 1.1 virtual hosting they are
> more than necessary!

Oh, MX is just fine for me, too!:-) Again, I'm just saying that CNAME
domains used in mail addresses are known to be problematic at least in
practice, and I don't think that such uses are so common as you said -
I've only experienced this problem a few times in my billions of
e-mail exchanges for >10 years, even though there are many MTAs with
this problem.  Perhaps my wording "legal/illegal" wasn't correct with
respect to the latest RFC, though.

Best regards,

	Eijiro