Re: i93: Repeating Single-value headers

sön 2007-12-23 klockan 22:53 +1100 skrev Mark Nottingham:

>    1. There is no one "right" answer; different kinds of clients -- or  
> the same client in different situations -- may need to act differently  
> upon an error, for most kinds of errors (there may be exceptions), and
> 
>    2. Specifying behaviour for handling every conceivable error isn't  
> possible, attempting to do so will seriously bloat the specification,  
> and will not significantly improve interoperability.
> 
> We'll wait for Roy's text to see if there's consensus, but if you have  
> a problem with this, speaking up sooner rather than later would be good.

I agree with the view expressed above, with one exception.
Content-Length deserves an explicit mention due to it's use in parsing
of the protocol itself.

For all other single-valued headers I know about it's sufficient to say
that one SHOULD NOT send more than one value of a header defined as
single-valued, except for proxies & gateways which MAY if the
request/response they received had multiple values for the header in
question.

How recipients deal with multiple values is better left implementation
dependent, with maybe a note to remind implementers that nothing stops
others from sending them malformed data such as multiple values of a
single-valued header, and general MAY rules on how clients respektive
serves should respond when detecting such messages if it matters to the
implementation.

Regards
Henrik

Received on Friday, 28 December 2007 14:39:55 UTC