[AFS3-std] Re: Encoding IPvN addresses

[Andrew Deason]

On Wed, 9 Feb 2011 20:36:53 -0500
Derrick Brashear <shadow@gmail.com> wrote:

> > I support having port numbers so that hosting multiple cells on the
> > same host may be possible. How would port numbers be handled, if not
> > alongside the address?
> 
> as another element in an RPC payload?

But if someone has an endpoint with a different 'port' width, or some
entirely different equivalent of a 'port', we have to rev the whole
protocol?

If we bake all of this stuff into the address type, and something
doesn't need it, the downside is that we waste space. If we have those
extra details alongside the address type, and we need to change them, we
make it impossible without changing all of the RPCs.

While I agree that something may want to represent "just an address", it
seems very rare, at least currently, so some overhead doesn't seem like
the end of the world. It doesn't even need to incur overhead; you can
just have an RX_IPV4_UDP address type that includes a port, and a
RX_IPV4 address type that doesn't include one. Or just define an
"endpoint" type and an "address" type? 

-- 
Andrew Deason
adeason@sinenomine.net