- From: Mark Nottingham <mnot@mnot.net>
- Date: Fri, 11 Jul 2014 18:03:48 +1000
- To: Jeff Pinner <jpinner@twitter.com>
- Cc: HTTP Working Group <ietf-http-wg@w3.org>
Hi Jeff, Thanks. Based on the discussion today, I'm open to this. From where I sit, the only one that seems controversial is #4 - and we've already seen a proposed modification from Greg. Personally, I wonder if we can just ditch #4 and still have reasonable properties; an implementation that receives too many CONTINUATIONS can reset the stream and continue to process other streams, correct? Let's talk it through. Cheers, On 11 Jul 2014, at 5:32 pm, Jeff Pinner <jpinner@twitter.com> wrote: > How do people feel about the following compromise: > > 1) Increase frame size to 16-bits. > 2) Remove reference set from HPACK allowing for "streaming" decoding. > 3) Requiring that all ":"-headers appear first. > 4) Only allowing CONTINUATION if the previous frame is 2^16-1 bytes. > 5) Allowing interleaving of CONTINUATION frames with other frames. > > For implementors that know that they will never accept more than 64kb > of headers, they don't have to implement CONTINUATION frames. > > For implementors that are concerned about not receiving routing > headers and having to buffer full frames, they will be the first > things that show up and the rest of the headers can be proxied without > any buffering. > > For implementors that need to support headers > 64kb they can do so > using CONTINUATION frames. > > For implementors concerned about interleaving headers between multiple > streams, that is now allowed because the reference set does not > coincide with the end of the header block. > > For implementors that are concerned with spec complexity, the entire > notion of "header block fragments" is removed. > > Thoughts? > -- Mark Nottingham https://www.mnot.net/
Received on Friday, 11 July 2014 08:04:18 UTC