At 15.58 -0800 00-12-14, Harald Alvestrand wrote:
>At 18:20 14/12/2000 -0500, Mark Wickens wrote:
>>Hello,
>>
>>I am interested in opinions on the feasibility of transmitting DSN requests
>>by means of message headers rather than SMTP (as RFC 1891 describes).
>
>This was debated extensively at the time the documents were written,
>and the conclusion at the time was that the layering violation of
>using headers for this was just too distasteful for serious
>consideration.
There are other layering violations which have been accepted. For
example the "Received" RFC822 header obviously belongs to the message
transport layer, but is still part of RFC822. Why is this allowed, if
layering violations are so distaseful?
It's a partly a matter of when things were designed and partly a matter of
costs versus benefits.
The cost of having a separate envelope structure is pretty high. In the
original design there wasn't much envelope information that needed to be
carried around, so the benefit of having a separate structure for it weren't
great. The result is that it made sense to overload the header in this simple
way, even though it is something of a layering violation.
Now, over the years we've added considerably to the amount of envelope
information we carry around, and this begs the question of whether we would
have been better off with an exsensible envelope structure in the beginning.
But frankly, when I look at X.400's envelope/header design, with its many
corner cases, implementation issues, and general confusion, I'm not sure it is
a good idea even now. And as I watch the confusion over this stuff in the
current discussions of instant messaging in general and CPIM in particular, I'm
even less sure. This may be a case of the right being the enemy of the good. We
really may be better off having an envelope that's really hard to extend (and
which has some layering violations inherent in its use) and a header format
that's very easy to extend.
Now consider DSNs. Here the tradeoff is very different: Carrying the
information in the envelope has some costs associated with it, but the benefit
is that the resulting mechanism is reliable. A header-based mechanism is
inherently unreliable. The WG debated this and decided that reliability was
worth the cost. And while I'm sure that the purists were pleased that this
envelope information didn't get added to the header, I don't recall this as
being a deciding factor.
Ned