Re: Minor isn't. It's a pardigm change

SM wrote:
> Hi Paul,
> At 02:39 31-03-2008, Paul Smith wrote:
> > Either:
> > - toaster.example.org is going to send a message with the return-path 
> > of something(_at_)toaster(_dot_)example(_dot_)org, in which case I'd EXPECT 
> > toaster.example.org to have an SMTP server running on it, hence an MX 
> > record would not be a bad thing, or
> > - toaster.example.org is going to send a message with a null return 
> > path, in which case what does it matter..
> The discussion have been whether the receiver should explicitly specify 
> a MX RR to determine whether it accepts mail.  If we stretch the MX RR 
> requirement to a forward and reverse model, then the sender 
> (toaster.example.org) also has to have a MX RR.  This means that the 
> change affects both senders and receivers.  If we use a null 
> return-path, the sender won't know whether there was a delivery failure.
<snip>

I think that stretching the MX RR requirement to both a forward and 
reverse model is not something we should necessarily consider.  I 
believe that would change the architecture too fundamentally.
I do stand on the side that MX records should be required for IPv6 only 
hosts that wish to receive mail at a FQDN.
The reason this must be discussed now, is because if 2821bis is 
published allowing AAAA fall-back, then we can never* change it in the 
future.
- Willie

* Obviously by never, I mean that it wouldn't ever happen.  That that it 
 theoretically couldn't happen.