Re: A broken browser

On Wed, 1 Jan 1997, Alexei Kosut wrote:
> [I]t is my understanding that the spec does not define methods 
> of negotiation. As long as the server is careful to set the right 
> cache-controlling headers (in most cases, probably just Vary), the 
> server can decide what to send the browser however the heck it 
> wants.  All the spec defines (and all it should define, IMHO), is the 
> format of Accept, the format of Vary:, the 406 response, etc..., but 
> it doesn't say how the server should interpret them (this is something 
> the transparent conneg drafts have done, but that's something seperate). 

Okay, that makes sense.  I agree that the current language supports this.

> In this case, though, I think Apache is correct. The client sent
> q=0.0, so obviously text/html is "not acceptable". If it really meant
> "not perferable", it should send 0.001 (or anything less than the
> other entries, really). How else are you going to say "I'll take any
> text but HTML" than "Accept: text/*, text/html;q=0.0"?

I agree that Apache is behaving predictably, but I do not agree that it is
behaving desirably.  Almost no browsers give users control of the
Accept string, and very few users even know what an Accept string is.
Apache is denying Lynx users access to text/html when that is clearly not
the intention of the user.  I think Apache should provide a way for server
maintainers to work around browser bugs, and send text/html even if it has
a q value of 0.  (If it has this capability, please inform Brian.  ;)

Albert, this is another reason to allow servers to return documents
indicated to be unacceptable: to work around known browser bugs.  "Be
liberal with what you accept" and all that.....

Marc Hedlund <>

Received on Thursday, 2 January 1997 01:04:57 UTC