Hector Santos <hsantos(_at_)santronics(_dot_)com> wrote:
And this is why I was proposing the text that if a IPv6 aware client was
sending to mail out to a IPv4 client (because it used a IPv4 socket
connection), it SHOULD NOT make any presumption that the receiver would
be able to *communicate* back (in sending a bounce or user reply) using
IPv6. If it doesn't provide a compatible IPv4 layer, then its should
expect loss of mail.
(I shall try -- really I shall -- to avoid an IPv6 religious war here.)
Unless you have a middle man or proxie handling all the outbound and
inbound for you, you will need to have a dual stack smtp system for
direct contacts from IPv4 systems.
My hope has been that an IPv6-only host speaking SMTP to the rest of
the world would:
- look for an MX RR pointing to an AAAA RR
- if it finds one, use that AAAA RR
- if not, look for an MX (explicit or implied) pointing to an A RR
- if it finds one, pass the email to a friendly relay speaking IPv4
_ if not, give the usual error
- advertise an MX RR pointing to an A RR
(in addition to any pointing to AAAA RRs)
This strikes me as a more reasonable long-term algorithm than
requiring all mail from an IPv6 user to go through a SMTP server with
both IPv4 and IPv6 connectivity.
It may be that the consensus here prefers Hector's solution: if so,
I suppose I should shut up. But please think long-term: we want something
that can work today and continue working for 20 years, by which time
IPv4 should be as rare as IPv6 is today.
--
John Leslie <john(_at_)jlc(_dot_)net>