- From: Adam Barth <w3c@adambarth.com>
- Date: Wed, 27 Oct 2010 00:31:26 -0700
- To: Mark Nottingham <mnot@mnot.net>
- Cc: Julian Reschke <julian.reschke@gmx.de>, Adrien de Croy <adrien@qbik.com>, HTTP Working Group <ietf-http-wg@w3.org>
There are use cases for including a non-zero length body in an CONNECT request. For example, draft-abarth-websocket-handshake includes some additional data with the CONNECT request. Of course, we could put that data in a header, but it seems to make sense as an entity-body. Adam On Tue, Oct 26, 2010 at 6:58 PM, Mark Nottingham <mnot@mnot.net> wrote: > I think we can specify: > > 1) CONNECT requests MUST have a zero-length body (same language referring to p1 as we used for 205) > 2) CONNECT responses that are successful (2xx) MUST have a zero-length body, because the tunnel begins after the header block. > > Thoughts? > > > On 26/10/2010, at 7:33 PM, Julian Reschke wrote: > >> On 25.10.2010 23:45, Adrien de Croy wrote: >>> ... >>> in relation to CONNECT, I think we can justify giving it special >>> treatment for several reasons: >>> >>> 1. It's not part of the original spec, but an extension designed to >>> enable arbitrary connectivity through a compliant proxy >>> 2. It already has very specific requirements which make it very >>> un-HTTP-like (e.g. the proxy connects and gets out of the way), no HTTP >>> is (necessarily) used upstream. >>> >>> In fact in our code-base the special handling for CONNECT is much more >>> involved than for say HEAD. I find it hard to conceive of a proxy that >>> wouldn't treat CONNECT as a very special case already. >>> >>> In the one case where I've seen a body on a CONNECT method (blu-ray >>> player), if that body were passed through to the end server, it broke >>> things. >>> >>> If you allow bodies on a method, then Content-Length is required. I >>> don't see any Content-Length headers on CONNECT messages, so current >>> browsers would become incompatible. >>> >>> Can we allow Transfer-Encoding: chunked on CONNECT? IMO we can't. >>> >>> Adrien >> >> I think we are in agreement that CONNECT, once we add it to the spec, needs more work (see issues 250 and 251). >> >> Best regards, Julian >> > > -- > Mark Nottingham http://www.mnot.net/ > > > > >
Received on Wednesday, 27 October 2010 07:32:34 UTC