Google's implementation uses a 1:1 mapping between
an h3 ALPN and a QUIC version. Because of this, when
we ship QUIC 0x00000001, it'll be with ALPN=h3.

Our code supports v1/h3 already, but v1/h3 is disabled by default.
We'd like to align with everyone to pick a date when we start
enabling v1/h3 in production though.

>From the conversations I've had, I think everyone agrees that
when draft-ietf-quic-http ships as RFC, everyone will be allowed
to ship v1/h3. I think everyone also agrees that we shouldn't do
that before draft-ietf-quic-transport ships as RFC.

The open question is: do we wait for draft-ietf-quic-http or do we
move forward when draft-ietf-quic-transport ships?

David

On Tue, Apr 27, 2021 at 4:04 PM Martin Duke <[email protected]> wrote:

> QUIC, sorry the confusion. The original message in this thread included
> HTTPbis, and you should reply to that one to keep everyone in the loop.
>
> On Tue, Apr 27, 2021 at 3:59 PM Martin Duke <[email protected]>
> wrote:
>
>> Damn it, wrong http
>>
>> On Tue, Apr 27, 2021 at 3:40 PM Martin Duke <[email protected]>
>> wrote:
>>
>>> In the quicdev slack channel today, we realized that we had a disconnect
>>> on what ALPN to use in the interval between the QUIC RFCs publishing and
>>> the HTTP/3 RFCs being ready (due to a MISREF with http-semantics, etc).
>>>
>>> It's lost in the slack archives now, but I *think* we had concluded that
>>> once the QUIC RFCs ship the endpoints should use 0x00000001/h3, not h3-29
>>> or h3-32, because the chance of something in http-semantics breaking
>>> interoperability was nil. I personally don't really care how we converge,
>>> as long as we converge.
>>>
>>> To summarize the choices, in the ~months between the RFCs, are endpoints
>>> doing a QUIC version + ALPN of
>>> 1) 0x00000001/h3 or
>>> 2) 0x00000001/h3-xx
>>>
>>> Can we come to an agreement on this point?
>>>
>>> Martin
>>>
>>

Reply via email to