- From: Jeffrey Mogul <mogul@pa.dec.com>
- Date: Thu, 11 Apr 96 12:19:24 MDT
- To: Paul Leach <paulle@microsoft.com>
- Cc: http-caching@pa.dec.com
Jeff's second to the last message finally got thru to me. The only
overloading is my brain because of the name. It's not "must-revalidate";
it's "must-verify" with the user if you're going to over-ride when
max-age=0. The confusion is because max-age=0 already means
"must-revalidate" (except when user preferences say not to.).
This misunderstanding based on the name has confused at least Henry and
me. Maybe it's the Redmond air.
Oh dear, I guess that message of mine confused you even more. (Or
maybe it IS that Redmond air.)
This "must-verify" interpretation perhaps is correct for the end-user
cache (the one inside the PDA, for example), but it makes no sense
for an intermediate (proxy) cache. Since such a cache has no
way to query the "user" for a verification. Since it can't do so,
it has no way to avoid the "must revalidate" rule.
Consider the case where the PDA has retrieved an entity via the
proxy cache, then (because the PDA has almost no storage) the PDA has
deleted the value from its own cache. Later it does a new request
for the entity; since it isn't in the PDA's cache, it does a
non-conditional GET. The proxy cache is now the only place to
enforce Koen's requirement that some message actually goes all
the way to the origin server. So here "must-revalidate" is clearly
the only possible interpretation of what has to happen.
-Jeff
Received on Thursday, 11 April 1996 19:36:23 UTC