John C Klensin wrote (2007-07-17):
Arnt Gulbrandsen <arnt(_at_)oryx(_dot_)com> wrote:
My code to implement this does
if all rejected
answer "CR LF . CR LF" with 5xx
else if all accepted
2xx
else
2xx, and send a DSN with the exact details
The third case is very, very uncommon in practice, so I'm not
all that worried by its ill effects.
yep. It is also no worse than "accept everything, send DSNs as
needed later" which, for better or worse, 2821 explicitly
contemplates.
BTW, while I'm still catching up with old messages: The BSI, a
federal agency for IT security in Germany, allegedly recommends
that all federal organizations in Germany cease to send NDRs,
ignoring RFC 2821 for better or worse:
<http://www.heise.de/ix/news/meldung/96350> (German article)
<http://purl.net/net/de2en/heise.de/ix/news/meldung/print/96350>
The second link is a crude translation provided by Google.
Frank