At 14:01 06/07/04 -0600, Alex Rousskov wrote:
The above "you do not know" assertion is false for some kinds of
e-mail transformations (e.g., detecting a known spam by looking at
Subject lines or headers) and true for some kinds of HTTP
transformations (e.g., detecting a virus by examining sensitive octets
at the end of an HTTP entity).
Overall, there is nothing special about SMTP here, IMO: some OPES
adaptations require the whole message, some do not. Sometimes OPES
processing overheads delay the entire message; sometimes they delay a
portion of the message.
Why is SMTP special at this level of thinking?
(Trying to answer this question, without making any statement about OPES
applicabiluty...)
The delivery requirements associated with SMTP effectively require a mail
relay to store an entire message until it has been successfully transferred
to the next relay. This is more than just "best effort" delivery -- I have
seen the term "heroic effort" applied here. A message is expected to
survive a complete restart of an MTA system while an SMTP transaction is
in-progress.
It's true that the spec doesn't actually say that messages must be stored,
but that does seem to be a consequence of the responsibility that the spec
places on MTAs.
In practice, I think this means that most real-world MTAs do store messages
in a non-volatile message store, where they might arguably (I won't say
"reasonably") be regarded as not in-transit.
#g
------------
Graham Klyne
For email:
http://www.ninebynine.org/#Contact