- From: Roy T. Fielding <fielding@gbiv.com>
- Date: Thu, 15 Jul 2010 22:39:01 -0700
- To: Adrien de Croy <adrien@qbik.com>
- Cc: "Thomson, Martin" <Martin.Thomson@andrew.com>, Willy Tarreau <w@1wt.eu>, HTTP Working Group <ietf-http-wg@w3.org>
On Jul 15, 2010, at 10:16 PM, Adrien de Croy wrote: > > just to belabour the point > > RFC2616 S 14.10 > > Message headers listed in the Connection header MUST NOT include end- > to-end headers, such as Cache-Control. Ah, excellent example of a nonsense requirement. Note how it is phrased as an existence test rather than a requirement on senders (not to send) and on recipients (what the heck we are supposed to do when we receive it). > This should either > > a) be removed from HTTPbis if we are to take the approach suggested by Roy; or > b) be elaborated further to specify what an agent should do upon receiving a message that violates this MUST level requirement Well, given that any header sent from the one end and received at the other end is an end-to-end header, and that the only thing that makes a header other than Connection a "hop-by-hop" header is the fact that it is listed in Connection (this was part of my original HTTP/1.1 proposals regarding keep-alive), we should remove that requirement because it requires omnipotence on the part of implementations (they must know the purpose of all future headers) and serves no useful purpose (because the requirement to not forward any header named in Connection will override this pseudo-requirement that says the field value MUST NOT exist). In fact, I already made this comment in ticket #60, but it was part of an editorial discussion on a separate issue. ....Roy
Received on Friday, 16 July 2010 05:39:30 UTC