- From: Koen Holtman <koen@win.tue.nl>
- Date: Fri, 8 Sep 1995 01:27:04 +0200 (MET DST)
- To: Brian Behlendorf <brian@organic.com>
- Cc: koen@win.tue.nl, fielding@beach.w3.org, http-wg%cuckoo.hpl.hp.com@hplb.hpl.hp.com
Brian Behlendorf: >On Wed, 6 Sep 1995, Koen Holtman wrote: >> For example: >> >> Accept-Color: monochrome, multicolor;qco=0.5 >> Accept-Color: 8bit, 2bit;qco=0.5, 24bit;qco=0.7 > >We could come up with attributes of different media types and make >them separate Accept: parameters all day, but I don't think that would >get us very far. I agree that the number of Accept-* headers is getting a bit large. As for getting us far: I happen to think that preemptive color negotiation is a very important feature to have, much more important than, say, preemptive charset negotiation. The accept-parameter: suggestion by Larry Masinter done elsewhere in this thread would eliminate Accept-* header bloat by generalizing all Accept-* headers. I support this suggestion, it sounds like a very good thing to have to me. > Instead let's roll this in with the 406 response No, that won't do: it would require an extra request-response round trip for each (inline) image having color variants, which would be hideously inefficient. (Right now, I am developing a system in which a page can have 20 different small (12x16 pixel) inline icons on it. Imagine the round trip overhead.) Also, (see my `Preemptive and reactive content'), I'd like to have an orthogonal protocol, in which Accept headers (allowing preemptive content negotiation) are a way to speed up the most common cases in reactive content negotiation. Limiting the things that could be negotiated on in preemptive negotiation to a subset of the things that can be negotiated on in reactive negotiation seems to me 1) to be completely unnecessary (even less so with accept-parameter:) 2) to introduce all kinds of tradeoffs (is color negotiation important enough? Is charset negotiation important enough?) that would have to be made by us, http-wg, instead of by individual users. [...] >It looks like the "color" or "bitdepth" or any >attribute is going to be at least mime-major-type specific, perhaps >subtype-specific too. "color" is certainly not mime-major-type specific. Mime types like application/postscript (color postscript!) and text/html (html with color extensions) would also benefit from it. The more I think about it, the more I like accept-parameter: color=8bit, color=*;q=0.5, encoding=gzip headers. Having these would allow for a painless introduction of new attributes too: once a new attribute ( (e.g. screen-width) becomes popular in reactive negotiation, a browser (user) could move it into the accept-parameter header to allow for faster preemptive negotiation. Currently, such a move would require an update of the spec and rewrites of server software. > Brian Koen.
Received on Thursday, 7 September 1995 16:30:53 UTC