Date:        Mon, 14 Feb 2000 11:52:35 +0900
    From:        itojun@iijlab.net
    Message-ID:  <10613.950496755@coconut.itojun.org>

  | 	is this the right place for this discussion? :-P

Probably not...

  | 	It's in protocol overview section and there's no explicit mention
  | 	that this is only about mechanisms (correct me if I'm wrong).

No, it it is 2462 - it is talking only about what is in 2462...

  | 	Please let me know how you would handle autoconfigured default
  | 	router list on a router?

Don't know yet - certainly not from anything in RA packets.
There's lots more that could usefully be autoconfigured as well,
I don't know wha the mechanism for any of that will be either - but
I very much doubt than any of it will be RAs.

  |	When the router accept RAs from multiple
  | 	interface, what would you do?

I don't think routers should be looking at RAs at all, other than
perhaps to verify that they're not inconsistent with its view of
the world and issue a diagnostic (syslog/kprintf) if a problem is found.
(And even that takes care, as not all differences are inconsistencies).

  | 	Also, accepting RA from the router itself (which was the topic we
  | 	started this thread) is very strange,

Yes, I agree, and I thought I made it clear in the first message
that as things are now, attempting to autoconfig a router would not
be intelligent.

  | 	I still believe RFC2462 assumes the following:
  | 	- the box to be autoconfigured is a host, not a router

Yes (though just to confuse things, most routers are hosts before
routing is enabled... eg: routers sometimes boot over the net, or
load configuration over the net - when they're doing that, they're
hosts).

  | 	- the host has only single external interface

That one I don't think is assumed.   Or I certainly don't recall
it as intended.  Where do you see that assumption?  I know there
is still work to be done on multi-homed hosts, but I didn't think
any of that would affect the autoconf part.

Note I only sent my first message on this because it seemed to me that
some people who saw your earlier message had read it as meaning that there
was some IPv6 architectural design decision somewhere along the way that
said "no IPv6 routers can ever be autoconfigured".  That isn't the case.
On the other hand, we have currently no mechanisms at all to use to
autoconfigure a router, so anyone who wants to try is either going to have
to wait for someone to define the mechanisms - or define them themselves.

kre