Re: Composability of extensions (was: Re: Preparing for discussion on what to do about the multipath extension milestone)

[Christian Huitema <[e-mail]>]

On 10/1/2020 6:33 AM, Dmitri Tikhonov wrote:
> On Thu, Oct 01, 2020 at 02:07:34PM +0100, Lucas Pardue wrote:
>> This reminded me of the comment I made on the mailing list in January [1]
>> about how QUIC extensions would play together. There was some further
>> discussion on GitHub [2] and in Zurich we decided to do nothing.
>                                ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
>
> That's not quite accurate.  We decided to do nothing with the
> understanding that figuring out extension composability and interplay
> is the responsibility of newer extensions; they'd have to "watch out
> for the dragons" [1].
>
>   - Dmitri.
>
> 1. https://blog.litespeedtech.com/2020/02/11/quic-working-group-interim-zurich-recap/#extension-dragons

We certainly did not "do nothing". The time stamp extension, for
example, was rewritten as an extension that can be composed with the ACK
frame, instead of as an extension changing the format of the ACK frame.
See https://datatracker.ietf.org/doc/draft-huitema-quic-ts/.

I think we should keep this design pattern for future extensions. For
example, instead of "rewrite the retire_connection_id" frame, we should
define an extension frame that composes with it.

-- Christian Huitema