Re: IESG comments on draft-ietf-ipngwg-default-addr-select-06.txt

[rbrabson]

On Wednesday, 06/12/2002 at 03:20 AST, [EMAIL PROTECTED] wrote:
> I don't quite follow the logic here. Having the stack, by default,
> give preference to temporary over public addresses only means that
> they will be used without applications having explicitely asked for
> their use. For existing applications, that might cause
> surprise/problems in some cases.

Yes, but it is even worse than that.  The same application may run
perfectly fine on one host, but fail on another, if only some of
the hosts support temporary addresses.  And when upgrading an OS from
a level which doesn't support temporary addresses to one which does,
applications may suddenly begin to fail.  And there is absolutely
nothing a user can do to fix it, short of changing their OS to
one which does not support temporary addresses.  These are the types 
of things which drive users, system programmers, and call centers crazy.
I know on the product on which I work, we wouldn't even begin to 
consider making a change like this to our external behavior - our 
customers simply wouldn't let us get away with it.
 
> Future applications, on the other hand, still need to be written, and
> one could argue they will need to make intelligent choices about
> whether the use of temporary addresses will cause problems, and code
> accordingly.

So, how exactly does the application know if temporary addresses will
cause problems?  Do they need to first try connecting using a temporary
address and, if this fails, try to connect using public address?  Do 
they need to support configuration on a per-destination basis so some
poor end user can try to configure the application properly when trying
to access sites which require public addresses?  And will the typical
application programmer even know (s)he needs to include such logic when
converting their application from AF_INET sockets to AF_INET6 sockets?

If we know that applications need code to support temporary addresses
in order for them to work properly in certain cases, wouldn't it make 
sense to have the application indicate it has the necessary support
before using them?

Roy