Re: Proposal for i23: no-store invalidation

You wrote:

> Not sure. It would make sense to generalize the HEAD invalidation to
> also apply to GET responses. i.e. copy the text from HEAD to GET as
> well. There is a whole family of uncachable responses, and the result
> things can get very surprising  if caches do not invalidate the older
> copy when the resource has obviously changed even if the current
> representation can not be cached..
>
> But no-store in itself is not a parameter to that decision, only the
> fact that the response is significantly different from the cached  
> copy.
>
> And it's not about caches deleting the cached copy, only  
> invalidating it
> (threat it as stale on future requests).

Can you give some examples? 4xx to a GET shouldn't invalidate the  
cache, and a cache is allowed to return a cached response when  
encountering a 5xx unless must-revalidate is present.

Or are you just saying that 5xx + must-revalidate = invalidate?

I can see a case for augmenting the text in p6-cache regarding HEAD  
invalidations (currently only mentioned in p2-semantics), and for  
aligning the HEAD text with that for 304 responses (p6-cache 7.2). Is  
that what you want? I suspect that's going to be part of a larger p6  
rewrite...

In any case, I believe we can close this issue with no spec change; we  
may change text regarding cache invalidation separately.


On 08/04/2008, at 5:46 AM, Henrik Nordstrom wrote:

>
> tis 2008-04-08 klockan 10:19 +1000 skrev Mark Nottingham:
>> That seems like a different issue.
>
> Yes, but related as it's mainly visible when using no-store. I would  
> say
> it's the root issue of the no-store invalidation question.
>
> Regards
> Henrik
>
>


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

Received on Thursday, 17 April 2008 01:59:48 UTC