Re: Removals from HTTP/2

On Thu, Apr 22, 2021 at 04:30:58PM +1000, Martin Thomson wrote:
> We're pointing to the new priority draft, which uses a different frame type.
> Do you think it worthwhile saying "frames with this type cannot be reassigned
> different semantics?"  Technically you can negotiate different format or
> semantics for any frame; in practice, this is probably a terrible idea and so
> we'd look to discourage that.

I think we should. Since the document explicitly suggests that "some
might or some might not use them", "removing support is probably worse
than keeping it", "each side ought to do what it thinks is best", I'm
pretty sure it will give a signal to implementations that they can start
to deviate a little bit from a deprecated standard if they figure it helps
in their situation. And we've seen in the past that dependencies on closed
streams and such things could have horrible effects. For this reason I'd
rather see one sentence like:

   "Although deprecated, PRIORITY frames, if sent, MUST strictly conform
    to their definition in RFC7540 and may in no way be used to experiment
    with alternate mechanisms."

It's cheap, and there are plenty of frame types available to play with.

Willy

Received on Thursday, 22 April 2021 06:39:23 UTC