Re: HTTP Caching Model?

+--- On Mon, 12 Dec 1994, Daniel W. Connolly wrote:
| [...] some information providers are using, of all things, the
| User-Agent field to customize their documents: they server up
| different stuff for MacMosaic, WinMosaic, Netscape, etc.
[...]
| User-Agent shouldn't affect the retuned data. (The fact that it
| does is a wart that we'll have to deal with somehow.)
| 
| It means that introducing new headers that can affect the returned
| data (like the recently proposed Accept-Charset: header) can't be done
| with correct backwards compatibility. It might be wise to say that all
| headers matching Accept-*: are allowed to affect the returned data.
+---

aargh.  I think you're placing undue blame on information providers for 
this problem.  I agree that content should not be affected by the 
User-Agent header; and that it should instead be affected by the Accept 
header.

Acknowledging that I am rather new to this working group and that I may 
be missing some history that would make this more clear, I consider the 
following line from the HTTP-1.0 draft to be at fault:

-----included text follows-----

5.5.8 ACCEPT

[...] In order to save time, and also allow clients to receive content 
types of which they may not be aware, an asterisk "*" may be used in 
place of either the type token and/or the subtype token. [...]

-----included text ends-----

For an information provider, an Accept header of "*/*" is useless.  Other 
than to save time (network time? implementation time?), I don't see why 
this sentence couldn't be amended to restrict wildcarding to the subtype 
token only.  With such a restriction in place, a provider could search 
for "image/" within the Accept header, and modify content based on its 
presence or absence; and a client sending "image/*" could still receive 
image subtypes of which it may not be aware.

Part of the problem also lies with the ALT value limitations of the <img> 
tag, which I understand are addressed by <fig>.  Since Doug Stevenson's 
monthly browser survey indicates that plenty of people are using 
text-only browsers, and since ALT and Accept are currently inadequate 
solutions, what is there other than User-Agent?

I agree with Dan that a specification for which request headers may affect 
returned content would be helpful, and that the Accept-* headers are the 
obvious candidate.  I also think that this information belongs in the CGI 
spec, which information providers are more likely to consult.  (Is the CGI
specification handled by this working group, or some other?  Or none?)

</marc>

Received on Monday, 12 December 1994 11:40:02 UTC