In message <C9C0CAF9-8BB3-41A9-AFBC-9B07B4B8AA1D@shagadelic.org>,
Jason Thorpe writes:

>
>On Apr 23, 2005, at 12:51 AM, Jonathan Stone wrote:
>
>> Try intsead to articulate what you *actually* want.
>
>Let me articulate what *I* want, in abstract terms, so we can get  
>away from the implementation details.
>
>I want the source address of an outgoing packet to have the same  
>"scope" as the destination address, when possible.

Jason,

That would be a change to the semantics of IPv4 addressing as
described in RFC-1122.  RFC-1918 addresses simply do *not* have
IPv6-style scoped semantics. That's just a fact.  10/8, 172.16/12,
and 192.168/16 are perfectly ordinary IP fadfresses reserved for
private internets (note the small i).

Forcing IPv6-style scoped semantics onto IPv4 addresses violates
RFC-1122.  Therefore, if you want to change that, the apprproiate
action is to go through the IETF process: draft an RFC specifying
scoping rules and get it onto the standards track.

Personally, I think that's reaonable for link-local scope addresses
(RFC-1918 addresses are another story). But either way, it *is* a
change. it *does* break existing code which relies on the
long-standing prior art semantics of ``if no local address bound, use
address of outgoing interface'', for however well that is defined.

I'm not fussed about implementation details.  I think the idea of
introducing explicitly-marked ``first-class'' and ``second-class''
local addresses can provide what you want, and what Manuel wants, and
what Tom said he wants, too.