Who is using experimental codepoints in HTTP/2?

https://github.com/httpwg/http2-spec/pull/853 proposes that we do away with the "Experimental Use" ranges for frame types and settings.  The result would be that new registrations could use these values.

This brings things more in line with recent changes in how we've seen protocols, like TLS and QUIC, do codepoint allocation.  There, we've learned (as we learned with RFC 6648 for textual registries) that codepoints used for experiments don't change when the experiment is successful.

What we're seeing instead new allocations that take _random_ unassigned codepoints and experiment with them without much coordination.  Registration policies that are welcoming (no review process, just liberally applied approval) ensure that these are entered into registries early.  As the policies (except for frame types) are already quite open for HTTP/2, removing this "Experimental Use" rule should make it easier rather than harder to experiment ... and transition the experiment into mainstream usage.

Of course, if extending HTTP/2 is as bad as Ian Swett would have it, then no one is experimenting because they can't.  Even if that might render this change moot, it's probably still worth doing the right thing anyway.

The other part of this, which might be worthwhile (I'm undecided here) is whether adding provisional registration makes sense.  I haven't written that up, but if people really want that, then we can do that too.

Does anyone have any objection to making this change?

Received on Monday, 24 May 2021 10:11:29 UTC