- From: Daniel DuBois <ddubois@rafiki.spyglass.com>
- Date: Sat, 9 Sep 1995 11:57:47 -0500
- To: John Franks <john@math.nwu.edu>, http-wg%cuckoo.hpl.hp.com@hplb.hpl.hp.com
At 11:10 AM 9/9/95 -0500, John Franks wrote: >According to Daniel DuBois: >> >> >- send small Accept without *: risk getting responses with inferior quality >> >> Inferior according to the server, maybe. Not inferior according to the >> client. The client is perfectly happy with what they got, because they >> asked for it! >> > >Maybe you didn't understand the point. It is very expensive for the client >to fully ask for what it wants. If it asks less than fully there is a >good chance the client will not be happy with what it gets. This is >a problem and "Live with it" is not the solution. Harumph! I understand the point perfectly. Don't confuse the different circumstances. Look back again at the exact circumstance I was replying to. It is impossible for a client to get something it's "not happy with" with an Accept: header that has no wildcard specifications. Everything in it's list are the types is presumably thinks are most important. [Explanation lest you again insult my understanding of the subject: Otherwise it gets a 406, and at that point can worry about accepting "lesser variants" ("What?! 406?! Shucks, now I'll have to go to an external viewer or do a save as, depending on the body of this 406"). Presumably a client would have 1) internally supported MIME types, 2) externally supported MIME types, and 3) everything else. Also conceivably it would like q=1 for #1, q=.5 for #2, and q=.01 for #3. Lastly a UA could decide to send an Accept: header that lists its #1s, and doesn't do a */*. Then upon forced reactive negotiation, it could pick and choose from the remaining variants to hopefully get something that's in the #2 list, or let the user decide. If it gets 200 OK on the original request, it most definately IS "happy with what it got". The server might not be thrilled that it had to send out it's crappy rtf2html version, but client is ignorantly bliss.] >I think that the ultimate answer has to be something like always use >reactive negotiation if there are muliple variants of a document. There Reactive negotiation necessitates a 2nd request. This is exactly the negative aspect Koen was complaining about, and I was trying to reiterate with my "live with it/have your cake and eat it to" statement. You simply can't have it both ways. [Both ways = easy, one-trip, fully-specified preemptive negotiation.] ----- Dan DuBois, Software Animal ddubois@spyglass.com (708) 505-1010 x532 http://www.spyglass.com/~ddubois/
Received on Saturday, 9 September 1995 10:00:56 UTC