Re: p6: Returning the freshest response

[Mark Nottingham <[e-mail]>]

On 03/04/2013, at 8:27 PM, Amos Jeffries <[e-mail]> wrote:

> On 3/04/2013 5:32 p.m., Mark Nottingham wrote:
>> On 31/03/2013, at 11:30 AM, Amos Jeffries <[e-mail]> wrote:
>> 
>>> Perhapse this is a bit better?
>>> 
>>> "
>>> 
>>> If multiple selected responses are available, the cache will need to choose one to use. When a selecting header has a known mechanism for doing so (e.g., qvalues on Accept and similar request headers), that mechanism SHOULD be used to eliminate unwanted responses; of the remainder, the most recent response (as determined by the Date header field) is used, as per Section 4.
>>> 
>>> "
>>> 
>>> Making it clear that Date mechanism still applies, but only after the negotiation filtering has been done. AFAIK that is how it always gets done in practice anyway.
>> I'm OK with your intent, but upgrading the MAY to a SHOULD is going to make implementations non-conformant.
> 
> How? and which ones?
> 
> I would think upgrading it to MUST would make things non-conformant but SHOULD is not absolute.


That's not what SHOULD means. A SHOULD is still a requirement; it's just that there are situations where there's ample justification for violating it.

Also, with a very few exceptions, RFC2119 requirements are used to improve interoperability. This isn't an interoperability requirement.

We can't impose a SHOULD -- especially on something like this, when many implementations are likely to violate it today -- just because we really want something to happen. 

Cheers,


--
Mark Nottingham   http://www.mnot.net/