Re: AD review of draft-ietf-httpbis-http2bis-05

[Martin Thomson <[e-mail]>]

On Sat, Nov 13, 2021, at 21:49, Francesca Palombini wrote:
> Hi Martin! Thanks to you and Cory for considering my comments, I know 
> these were mostly straightforward, and sorry it was tedious, I do hope 
> they bring some additional value, however small, to the document.

Only tedious because that's the nature of the job.  I do appreciate how clear and actionable your input has been.

> [...] the TCPbis doc currently sits with a big number of 
> DISCUSS comments since end of September. 

My opinion: we should drop the dependency unless TCPbis suddenly resolves things.  We can keep the pull request open until that happens.  We don't depend on any of the details of TCP, so it's OK to have an outdated reference.
 
>>> 7. ----- https://github.com/httpwg/http2-spec/pull/983
>>I've added a reference to the priority deprecation text to the first of these newly deprecated fields.
> 
> FP: Great, thank you. I had some questions about what part of the 
> handling for priorities is still mandatory to support, following my 
> reading of draft-ietf-httpbis-priority (see point 2 of 
> https://mailarchive.ietf.org/arch/msg/httpbisa/mLM0RujKL6ZXY4eQtbQo2Xomf7c/ 
> ). Maybe you could help clarifying it for me? I read this section as 
> well as 5.3, and although the fields are defined, since they are 
> deprecated there is no processing defined, except for errors. What am I 
> missing?

I think that's a problem in the -priority draft more than this one, though maybe Section 5.3 could be a tiny bit clearer.

What this is supposed to say, and what I think that it does say, is:

~~~
 HTTP/2 signals are pretty bad, but no signals is almost certainly worse. So,

1. Use newer, better signals if you can.
2. Use HTTP/2 signals if you don't have anything better.
3. Use ANYTHING else in place of RFC 7540 default priorities.

"Use" being both send and receive.
~~~

I think that the priority draft wants to say the same thing, but you caught some unfortunate word choices that might make that less clear.  No doubt there are similar choices in this document that I'm just not seeing.