On Fri 2017-04-28 08:50:44 -0700, Joe Touch wrote:
> On 4/28/2017 12:14 AM, Daniel Kahn Gillmor wrote:
>> In particular, the approach i've described in the draft only works for
>> client-speaks-first, stream-based protocols.
>
> I don't think you can assert that it works except where you have
> checked, and only for the *current* definition of those protocols.
right, it only "works" in the sense that i've documented how existing
versions are demultiplexable. In the text you quote above, i wasn't
saying it works for *all* client-speaks-first, stream-based protocols, i
was saying that it *can't* work for non-stream protcols, or for
protocols where the server speaks first.
> Keep in mind that ports are assigned to current *and all future*
> versions of a protocol, so all bets are off the instant you stop
> looking. Which means, effectively, that you can never assert that this
> works on an existing port assignment UNTIL you coordinate with that
> port's assignee, and they confirm that their service will never conflict
> with your definition.
i think we're in violent agreement here :)
> That also implies that this service cannot be defined as valid for any
> service that isn't already assigned. That would effectively be squatting
> on the entire port space.
I don't think i understand what you're saying here. The document is
about how a server can distinguish between existing versions of DNS and
existing versions of HTTP. it's not making any claims about any other
service. are you suggesting that it is making such a claim?
--dkg
_______________________________________________
dns-privacy mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/dns-privacy
