From m(_dot_)rose(_at_)radiomail(_dot_)net Wed Jan 20 14:19:05 1993
actually the only sure fire algorithm
for caching is to assign a unique id
to each content--no other algorithm
will work better.
Hmmm. Admittedly I may be improperly reconstructing the problem
from your terse description, but I don't accept your proclamation.
If the reason for caching is to avoid multiple fetches of the same
external-body due to multiple local viewings of the same message,
then constructing an ID via a 1-1 mapping with the parameters of
the retrieval (e.g, access-type;site;directory;name) caches
optimally without requiring that the sender supply an ID. And
MIME-compliance does not require content-ids with external-body
types.
Plus, caching based on retrieval parameters allows you to cache
when the same external-body is referenced by multiple messages.
It is reasonable to expect that independently-constructed messages
will reference the same external-body part. (Suppose people started
using external bodies for complex signatures? There are more
defensible uses.)
You might even be able to deduce file equality across different
access-types, e.g., when produced using multipart/alternatives.
Well, maybe this is blue sky.
There is an issue about cache consistency and external-bodys that
change -- but there are more reasonable heuristics than invalidating
the cache from one message to the next.
experience also tells us that using
non-global ids in a local enviornment
is a loser--ask the people in .na for
an example...
This is a pretty sweeping metaphor, but maybe adding new parts to a
multipart "/layout" or somesuch could be cumbersome or lead to
non-uniqueness. Point taken, I'm less keen on my original suggestion.
Jay