Re: Proposed change to 14.6 (Age)

Jeffrey Mogul:
     [Koen Holtman:]
>
>    Based on all this, I propose that the sentence
>    
>      HTTP/1.1 caches MUST send an Age header in every response.
>    
>    in 16.4 is changed into
>    
>      HTTP/1.1 caches MUST send an Age header in every non-first-hand
>      response.
>    
>I'd only support this change if it became:
>
>      An HTTP/1.1 cache MUST send an Age header in every response,
>      except that it MAY omit the Age header if the cache has
>      unambiguous proof that the response is firsthand, has not
>      flowed through any cache that is not compliant with HTTP/1.1,
>      and is not being transmitted to a client whose HTTP-Version
>      is less than 1.1.

The last part of the above requirement is a bit too cautious for my
taste, but I could live with your proposal for the change as a whole.

>but frankly, I think this is pointless.  What's the cost of adding
>the Age header, given that a compliant cache implementation MUST have
>the code to add Age to non-first-hand responses anyway?  It would
>take far more code to check all of the necessary conditions.

I'm not very worried about the cost of adding the header.  By the way,
I think that the presence of most of the machinery for detecting the
conditions you list above is already required by other aspects of the
protocol.

I want this change to make the age header useful as a detection device
for first-hand responses.  I had some private discussions with John
Klensin in which he convinced me that a way of detecting definite
first-hand-ness would be a desirable protocol feature. 

>Let's keep it simple.

NOTE TO THE EDITOR: if you decide _not_ to change the requirement in
the pre-04 Age header section, please change the wording in the note
at the end of 13.2.3.  See my original message in this thread.

>-Jeff

Koen.

Received on Friday, 31 May 1996 16:37:14 UTC