- From: Willy Tarreau <w@1wt.eu>
- Date: Wed, 6 Jul 2011 01:01:05 +0200
- To: Poul-Henning Kamp <phk@phk.freebsd.dk>
- Cc: Brian Pane <brianp@brianp.net>, httpbis Group <ietf-http-wg@w3.org>
On Tue, Jul 05, 2011 at 08:40:19PM +0000, Poul-Henning Kamp wrote: > In message <CAAbTgTtK3N=m-Hdz64ybr5xCsjXyVdZeF0AsnqHTxwgxGSO07Q@mail.gmail.com> > , Brian Pane writes: > > >Should the HTTP/1.1 spec thus prohibit the use of the Upgrade header > >in pipelined requests, or is the issue too obvious to document > >explicitly? > > I have always read the text such that after a 101 response you > no longer run HTTP, and it follows pretty obviously there from, > that you cannot pipeline anything near an upgrade, which by > definition must be a synchronous and synchronizing operation. In fact we should be more precise on this in order to allow future websocket implementations to save one round trip. We should state that whatever is sent after the request will be handled by the next protocol, whether it's the upgraded protocol (in case of success) or HTTP (in case of upgrade failure). Right now a client has to wait for a 101 before daring send anything and I don't think that's an unsolvable situation. Willy
Received on Tuesday, 5 July 2011 23:01:40 UTC