It would be changing the semantics of UPGRADE to be stream instead of connection, but yes. The main source of complexity here is that one cannot jettison the HTTP/2 parsing, etc. state machine when one gets an UPGRADE, but must concatenate the state machines. I can understand use-cases for wanting to do these kinds of things (e.g. if TLS 1.3 doesn't happen or doesn't have an encrypted handshake and proxies are mucking with the list of negotiated protocols). Alternately, it could mean for the entire connection, but that has its own uglyness-- the server would likely want to hold off on doing the upgrade until after it has served the responses to the requests... Anyway you cut it, there is some uglyness here. -=R On Tue, Aug 13, 2013 at 2:41 PM, William Chan (陈智昌) <willchan@chromium.org>wrote: > I read this email and was like...yes...those are factual statements...why > are you saying them... > > Then I re-read the earlier emails and realized you haven't been talking > about CONNECT at all, you want to do HTTP Upgrade within a stream. My bad > for misunderstanding you earlier. > > I do believe it's doable...although I'm not sure it's desirable :) > > > On Tue, Aug 13, 2013 at 11:33 PM, Ilari Liusvaara < > ilari.liusvaara@elisanet.fi> wrote: > >> On Tue, Aug 13, 2013 at 11:22:09PM +0200, William Chan (陈智昌) wrote: >> > As far as I'm concerned, you're welcome to do whatever you want with a >> > CONNECT tunnel. If you want to layer WebSockets on top of it, go for it. >> >> There is a fundamential difference between CONNECT and upgrade: CONNECT >> designates a target host, whereas upgrade does not. >> >> Also, existing websockets runs using upgrade, not CONNECT... >> >> -Ilari >> > >
Received on Tuesday, 13 August 2013 21:55:07 UTC