- From: Julian Reschke <julian.reschke@gmx.de>
- Date: Mon, 22 Nov 2010 16:24:09 +0100
- To: Mykyta Yevstifeyev <evnikita2@gmail.com>
- CC: ietf-http-wg@w3.org
On 22.11.2010 15:15, Mykyta Yevstifeyev wrote: > Julian, all, > > I have read all these notes. Here are the answers: > > 22.11.2010 12:55, Julian Reschke wrote: >> On 22.11.2010 08:33, Mykyta Yevstifeyev wrote: >>> Hello all, >>> >>> I have recently made an I-D, which, I think, >>> would be interesting for the WG. You can >>> find it here: >>> >>> http://datatracker.ietf.org/doc/draft-yevstifeyev-http-headers-not-recognized/ >>> >>> >>> Could you please review it? >>> ... >> >> Hi Mykyta, >> >> a few thoughts: >> >> - This would be interesting for debugging purposes. Not sure about >> things beyond that. For instance, what's the rational for the >> conformance requirements you make? IMHO, a server MUST continue to >> process the requests (because that's how 1xx status codes work), but >> the actual 103 message should only be a hint to the sender. > Yes, I have mentioned that the server MUST continue processing of the > request. > > If a server sends a response with aforementioned status, > it SHOULD continue processing of client's request. MUST != SHOULD. >> - The ABNF for the header should be a list of comma-separated headers >> (same syntax as for Vary, for instance) >> >> - You'd need IANA considerations for the new header as well. > The information about not-processed headers will be put into the body > of the response. A 103 response doesn't have a body. >> - In many cases, this will be extremely hard to implement, because the >> actual handling of a request requires several layers, and it would >> tricky to find out which headers were processed by whom. Also, in many >> cases, the final recipient might not be *able* to send a 1xx response >> (such as a Java servlet). > Look here: > > If a server receives request with unknown (for it) headers, it*SHOULD* > send a response with 'Some Headers Not Recognized' status. > > If a server is not able to send the 103 code, it won't do, as > we don't set '*MUST*' comformancecriterion here. Understood. I was just trying to explain that for many servers, it will be hard to implement this. Best regards, Julian
Received on Monday, 22 November 2010 15:24:48 UTC