[arin-ppml] Is this more desired than aTransferPolicy? Needinput

[Owen DeLong]

>>
>> Really?  Show me what an IPv6 capable Mac or PC does when it  
>> receives an
>> AAAA record in a DNS request submitted on an IPv4 network.
>> (no, don't show me.  But acknowledge that this isn't as simple as  
>> you're
>> making it)
>
> Actually, the host shouldn't have asked for an AAAA if it had no IPv6
> connectivity ... if it does have some, or atleast thinks it does, it  
> will
> use it (or try to).

The problem is that by default, IPv6 capable hosts always assume they
have IPv6 connectivity whether they do or not. Here's what happens...


Default configuration is that IPv6 is enabled and stateless  
autoconfiguration
is enabled.

Unfortunately, there's no test done in any IPv6 implementation I've
seen that even goes so far as the common sense "Do I have any IPv6
information that came from ANY external source?" before deciding
it has IPv6 connectivity, so, if an interface comes up and can give
itself a local "autoconf" address, bingo, the host thinks it has IPv6
connectivity whether it does or not.

Valid tests I can imagine:

	Have I seen an IPv6 packet on the wire? (A questionable test at best)
	Have I received an IPv6 RDISC packet?
	Do I have an IPv6 default route configured through any mechanism?
	Do I have an address from any mechanism other than stateless autoconf?
	Have I received a valid response to any IPv6 datagram?

Now, none of these tests is foolproof, but, if we made sure to satisfy  
at least
one of the last 4 criteria, I think it'd be a dramatic improvement to  
the current
situation.  If we managed even the first criterion, I think it would  
make things
quite a bit better.

Owen