- From: Amos Jeffries <squid3@treenet.co.nz>
- Date: Thu, 24 Nov 2011 11:29:28 +1300
- To: <ietf-http-wg@w3.org>
On Wed, 23 Nov 2011 17:22:29 +0100, Harald Alvestrand wrote: > On 11/23/2011 04:55 PM, Julian Reschke wrote: >> On 2011-11-23 16:41, Harald Alvestrand wrote: >>> I ran across this in a discussion, and went to check it up; it >>> might >>> need fixing. >>> ... >> >> I'm not sure that "fixing" is the right term; HTTP doesn't assign >> semantics to the ordering, and AFAIU never has. >> >> How is this a problem in practice? > > Well... either the semantics of the header > > Accept-language: en, no > > is that the languages are ordered by preference, or it does not imply > an ordering. > > As of the time I wrote the RFC (2002), I observed multiple browsers > offering an UI that let people rank their preferences for language, > and observed the relevant Accept-Language: headers being sent without > q= values. > > When I inquired, the response was that "leftmost language wins". > If this is still "what people do", it should be documented. > > Note: Firefox and Chrome seem to send q= values in my current > versions. The world might have changed. > But digging around in server-side code, I find that some code (which > I hope is not used) actually ignores q values totally and just picks > languages starting at left. Perhaps it hasn't. > > > Harald I've been keeping an eye on this since implementing language negotiation in Squid. It appears that nearly all agents are sending the language codes sorted by q value anyway. Whether they send the q value or not it is still possible to optimize by using the left-most wins assumption. If anyone is interested in doing a deeper analysis I have a dataset available covering the last year on several networks linking the Accept-Language and User-Agent header pair. Amos
Received on Wednesday, 23 November 2011 22:30:17 UTC