- From: James M Snell <jasnell@gmail.com>
- Date: Tue, 13 Aug 2013 14:14:54 -0700
- To: William Chan (陈智昌) <willchan@chromium.org>
- Cc: Tatsuhiro Tsujikawa <tatsuhiro.t@gmail.com>, "ietf-http-wg@w3.org" <ietf-http-wg@w3.org>
While I'm certainly not going to spend any time arguing this one much, my vote is to deprecate CONNECT in HTTP/2 with the recognition that doing so removes a feature that currently exists in HTTP/1. Later, a better approach for handling the use case can be determined. On Tue, Aug 13, 2013 at 7:42 AM, William Chan (陈智昌) <willchan@chromium.org> wrote: > Great catch again! You're describing behavior that we've implemented in > Chromium > (https://code.google.com/p/chromium/codesearch#chromium/src/net/spdy/spdy_proxy_client_socket.cc&q=spdy_proxy_client_socket.cc&sq=package:chromium&l=368) > to get SPDY proxies to work with CONNECT, but we never wrote out in the SPDY > spec, since we've been bad people and have forgotten to write everything > down. I think we may need to separately discuss HTTP CONNECT mapping to > HTTP/2. I suggest opening an issue for this. > > > On Tue, Aug 13, 2013 at 4:23 PM, Tatsuhiro Tsujikawa <tatsuhiro.t@gmail.com> > wrote: >> >> Hi, >> >> draft-05 says that :scheme MUST be present in HTTP request. >> If client sends to CONNECT method to the middle box, which value >> should be sent as :scheme? http? >> For the reference, the current SPDY proxy (aka secure proxy) omits >> :scheme. For example, CONNECT to remote example.org:443 via >> proxy, chrome sends the followings: >> >> :method => CONNECT >> :host => example.org >> :path => example:org:443 >> :version => HTTP/1.1 >> >> :version is not omitted in HTTP/2.0, so it can be ignored. >> >> Best regards, >> >> Tatsuhiro Tsujikawa >> >
Received on Tuesday, 13 August 2013 21:15:40 UTC