Re: [quicwg/base-drafts] Consider making h3 frame types varint (#2253)

[Kazuho Oku <[e-mail]>]

@DavidSchinazi 
> The added ability to perform experiments without coordination is worth the addition of a 5-line inline function.

IMO, the endpoints willing to do experiments should spend their 5-line worth of coding effort to negotiate the experiment, rather than asking every implementation to accept the complexity.

As suggested in the last paragraph of [HTTP/3 section 7](
https://quicwg.org/base-drafts/draft-ietf-quic-http.html#rfc.section.7), endpoints that are willing to do an experiment can exchange their own SETTINGS value to confirm that the peer supports the experiment, then use any of the unused 8-bit frame types for the experiment.

The benefits of the approach are:
* does not introduce the need to deal with varint types
* less network overhead in experiments (due to being able to use 8-bit types)
* less chance of the experiment becoming a de-facto standard

The last argument is based on the fact that the experiments use negotiation. The fear of doing experiments is that sometimes the code point used for the experiment gets ossified; you cannot decide when to stop sending the extensions. Negotiating the use is one way to mitigate the problem, because with negotiation you can observe if the extension is actually used by the peer.

-- 
You are receiving this because you are subscribed to this thread.
Reply to this email directly or view it on GitHub:
https://github.com/quicwg/base-drafts/issues/2253#issuecomment-449774738