- From: Rob Hartill <robh@imdb.com>
- Date: Fri, 3 Jan 1997 02:24:09 +0000 (GMT)
- To: http-wg%cuckoo.hpl.hp.com@hplb.hpl.hp.com
Foteos Macrides wrote: > I just tried a variety of servers with: > > Accept: text/html;q=0.000 > >and only Apache v1.2b3 dev paid any attention to it (Apache v1.1.1 >ignores it, like all the others). So I guess Apache v1.2b3 has >defined that situation. :) :) :) By default, Apache ignores the Accept headers unless you compile it with mod_negotiation.c and explicitly use negotiation, e.g. with "Options Multiviews". I've been trying to make use of mod_negotiation.c + Multiviews to a) map /foo to /foo.html so that URLs can get rid of ".html" (it took me a while to catch on to that good idea). b) choose language variants where we've translated some key pages. There are lots of broken/misconfigured browsers out there. Some browsers give users the impression that the order in which they list langauges is important and others give the impression that "en-gb" and "en-us" are reasonable substitutes for "en". It's a mess when people have something like Accept-Language: en-us de Unless Apache has a "en-us" variant I get mail from people asking why they get German pages even though they set English first. As some people have said to me "I know a little German so I added it after my American English preference." For Apache servers, MSIE 3.0 can be a nightmare. It sets the "Accept-Language" based on a Windows local setting. The user doesn't know this has happened and has no way to override it from the browser. When these browsers hit an Apache server that doesn't have an acceptable language they get a "406" response (and scream at the webmaster). I think that an explicit q=0 is reasonably interpreted as "don't send me this". There are plenty of numbers between 0 and 1 to order things without using 0 as a low acceptance value. In the absence of an Accept preference for a particular variant, the only reasonable interpretation (IMO) is an acceptance of the unknown variant with a 'q' value less that anything explicitly accepted. The current Apache approach of mapping missing variants as q=0 just doesn't work in practice. rob
Received on Thursday, 2 January 1997 18:26:48 UTC