On 2/6/2004 11:03 AM, Paul Hoffman / IMC wrote:
+ mixed parts
: marked for human consumption
: marked for machine consumption
Currently, I like Set C the best because it seems that something that
can read the body can read the markings and decide what to do with it.
That's the right track definitely, since it also allows for non-listed
categories like human-to-machine and machine-to-human that are needed for
certain applications and extensions. It probably needs refinement or
consideration, but it's definitely a good cut.
It is completely clear that whatever we define has to be extensible
because we can't know what users in the future of what we define will
want. The question is how much we define initially, and how the
extension mechanism works.
That's right. What I'd actually like to see would be several application
threads getting spawned off, and have their requirements feed back into
the base transport requirements. As stated already, the transport should
provide the mimimal necessary mechanics and nothing more. Fewest verbs,
simplest structures, etc.
I also want to reiterate that human-to-human comms is a very specific
application of the transport network. Legacy mail was built around an
implicit assumption of h2h, and that assumption triggered all kinds of
latent problems that had to be dealt with later (not the least of which
were problems related to encoding binary content within an h2h stream).
Eric A. Hall http://www.ehsco.com/
Internet Core Protocols http://www.oreilly.com/catalog/coreprot/