Ah. Using the first round of RCPT TOs as capabilities checks only and
sending
the RSET unconditionally is really quite cute.
"cute" is rather opposite of the four-letter word I would have chosen.
It still has the added
complexity of having multiple commands per recipient, but now it is the
same
command repeated, which eliminates a bunch of the edge conditions and
makes the
silly states very silly indeed.
well, I'll admit I'm curious - how does using the same command in two
different
ways eliminate a bunch of edge conditions? (if we had only one command in
SMTP
and used it over and over would that eliminate even more edge conditions?)
The edge conditions arise when the responses from the different commands
disagree in some way. And this is likely to happen when there are two commands
and two different code paths. The usual outcome then is that things which
shouldn't make a difference do make a difference, and that then leads to the
time honored whine of "but it works when I do it over there so this stuff here
must be broken".
ah. well it was sort of obvious to me that RCAP responses should be defined to
provide no indication whatsoever as to whether the recipient is valid, so you
couldn't get a conflict between a RCPT indication and an RCAP indication.
on the other hand either command could return a value that would prevent
delivery
of the message - RCPT could return 5xx and RCAP could return "this recipient
supports no content types you've ever heard of", and either one could cause a
temporary failure of message delivery. I don't consider that a silly state -
it's a natural consequence of being able to specify what content types you're
willing to accept.
however I really don't have a problem with an option to RCPT to return
UA capabilities as long as it is an extensible mechanism (other options could
return other information and could be asserted at the same time as CONNEG)
and as long as it doesn't affect the response and status codes.
Keith