Re: #290 [was: SHOULD-level requirements in p6-caching]

On 05/05/2011, at 11:37 AM, Roy T. Fielding wrote:

>> As I've said a few times now, according to the spec, that's currently very clearly equivalent to:
>> GET /foo HTTP/1.1
>> Host:
>> Cache-Control: max-age=45, no-cache
>> Which I *hope* we can agree should be interpreted conservatively.
> No, that's not what the client wants.  The client wants an HTTP/1.1
> compliant cache to obey max-age=45, not no-cache, but fears the
> presence of non-compliant caches that only understand Pragma.
> Hence, both headers are sent.  If the client wanted no-cache
> semantics, it would have sent no-cache in *both* fields.
> Pragma has the same relation to CC as Expires.  If that isn't defined
> in 2616, then it is surely an error in 2616.

OK. How do we specify this? Saying "When CC is present, you can ignore Pragma" brings about the case where:

GET /foo HTTP/1.1
Cache-Control: myweirdextension=foo
Pragma: no-cache

means that a HTTP/1.1 cache that doesn't implement myweirdextension now silently ignores the no-cache.

We do already say

> A client SHOULD include both header fields when a no-cache request is sent to a server not known to be HTTP/1.1 compliant.

... but that's only a SHOULD (hey, the Subject *is* relevant!).

If we go in this direction, AFAICT we'd end up changing 

>    When the no-cache directive is present in a request message, a cache
>    SHOULD forward the request toward the origin server even if it has a
>    stored copy of what is being requested.  This pragma directive has
>    the same semantics as the no-cache response directive (see Section 3.2.2
>    ) and is defined here for backward compatibility with
>    HTTP/1.0.  A client SHOULD include both header fields when a no-cache
>    request is sent to a server not known to be HTTP/1.1 compliant.  A
>    cache SHOULD treat "Pragma: no-cache" as if the client had sent
>    "Cache-Control: no-cache".

to something like

When the no-cache pragma directive is present in a request, a cache MUST treat the request as if it contains the no-cache cache-control directive [ref], unless the Cache-Control header is also present in the request. 

A client SHOULD include both pragma and cache-control directives when sending a no-cache request , unless Cache-Control: no-cache is omitted to target other Cache-Control directives at HTTP/1.1 caches, while precluding HTTP/1.0 implementations from using a cached response.

Is that workable?

Personally, I think that this approach is going to surprise more people than it will help, but I'm not going to lie down in the road about it; highly refined CC request directives are pretty unusual.


Mark Nottingham

Received on Thursday, 5 May 2011 02:01:33 UTC