RE: site local addresses (was Re: Fw: Welcome to the InterNAT...)

--On Wednesday, 26 March, 2003 15:13 -0800 Tony Hain 
<alh-ietf(_at_)tndh(_dot_)net> wrote:
> Michael Mealling wrote:
> > Its not that 'we don't want to change because its to much
> > work'. Its that the Internet architecture assured us that the
> > hour glass model applied, that the network topology would
> > remain abstracted within what to us is an opaque address
> > space. One of the number one reasons its so easy for new
> > application layer technologies to be deployed is that a
> > developer doesn't need to know or care about any layer below
> > TCP (or, in rare cases, UDP). If the lower layers want to
> > change that hour glass model then we're talking about a
> > serious breach of contract with the layers above it and a
> > dangerous blow to the hour glass model.
> You really don't want to go there, since it is in fact the
> violation of the layering by the apps that has created some of
> the mobility and renumbering challenges. Apps know all too
> much about what is going on below them, which is why the app
> developer sees any change as a lot of work.
Tony,

I am probably missing something, but I don't see this. 
Partially to aid transition to v6, we've been telling apps 
developers for several years that they should rely on DNS names, 
paying at little attention as possible to the format or content 
of addresses, their topology implications, etc.  As an 
application-writer, I'm not supposed to know anything about 
prefixes, address-masks, or anything else about how addresses 
are put together.  Yes, if I start doing tricky things about 
firewall or NAT-traversal, I can get myself into a lot of 
trouble, but that is part of why we have always considered those 
sorts of things to be layer-violating if not worse.
Ignoring the format of addresses has worked well for 1918 
addresses (loathsome as they might be) because the assumption is 
that filtering (so that they don't leak onto the public network) 
is the responsibility of anything that connects a 1918 network 
to the public Internet.  And CIDR was deployable precisely 
because the applications themselves didn't know about classes.
Now, if I correctly understood Margaret's presentation last week 
(and some other things), if there is a possibility that a 
particular address might be site-local but that the target might 
also be associated with publicly-routable prefixes, the 
application has to figure that out, figure out whether the 
target is local, and then do the right thing.  That isn't a 
matter of filtering or masking external to the application: the 
application has to do the work, or the architecture that 
surrounds the applications has to change.  That is a fairly big 
deal and, I think, involves more layering problems than what we 
are doing today.
But, obviously, I'm not understanding something.  Could you 
explain?
   john