getaddrinfo canonicalization (Re: host name resolution ...)

[Ken Raeburn]

"Donn Cave" <donn at u.washington.edu> writes:

> That does fix (or break, depending on point of view) the GNU C library
> version, but doesn't fix the AIX problem - I still get a zero-length
> string in ai_canonname (nor would I expect that to be affected from
> what I see in fake-addrinfo.h.)

For a numeric or text hostname input?  And, what version of AIX?  I've
only got 4.3.3 to test with right now.

> I see RFC 2553 doesn't get into what "canonical" means, must have
> seemed obvious to them (as I would have thought it was, until today.)

Yep.  The IETF is following the Open Group's lead in this
specification, and I've already raised the issue with them.  I should
check if it's been changed...

The Open Group spec is written so as to suggest what might happen with
DNS, but not to require DNS.  So "canonical name" needs to be either a
generic term, or specific to the underlying mechanism.  From the use
of the terminology in the DNS RFCs, I think in that context it's
pretty clear the CNAME record defines an alias for the canonical name,
and PTR records have nothing to do with it.

One key point in the getaddrinfo context is that the "canonical name"
issue has nothing to do with actually connecting to the host.  So the
behavior of getaddrinfo is a separate question from which name to use
with Kerberos (which is addressed in kerberos-clarifications, and not
properly implemented in the MIT code at this time).

Ken