On Sat, 08 Sep 2001, D. J. Bernstein wrote:
> Details: qmail, maildir, QMTP, and QMQP support an extended full-binary
> message format. This format cannot be delivered through SMTP, even with
> qmail on both ends, because of a fundamental SMTP design flaw.
Dan, I'm well aware that the mail I presented to qmail-qmqpd cannot be
delivered across SMTP. SMTP has been fixed almost 6 years ago,
experimental RFC-1830 has been obsoleted by RFC-3030. ("CHUNKING")
However, you're confused if you write "support an extended full-binary
message format" in context with qmail-send. qmail-send has no notion of
the binary cleanness of its transport backend. It must not assume so
when creating bounces. qmail-send has no idea where the mail it is to
deliver came from, and must not assume any particular format. qmail-send
must be able to notify me of non-delivery unless I give a bogus envelope
sender (which I did not).
> But this has no relevance to the delivery of your mail. Your mail does
> not use that extension.
Evidently, the test mail that failed does.
> As for double bounces: I introduced double bounces in qmail in 1995 as a
> last-resort mechanism for saving messages that couldn't be delivered and
> couldn't be bounced. Of course, if you forward these messages to an
> address that can't handle them, qmail will simply throw them away. In
The address is a regular mail address and does not impose any particular
restriction. Just send it a properly-formatted mail and it will take it.
qmail-send creates bounces that are not properly-formatted in that they
don't pass SMTP.
> particular, if you forward everything through SMTP, not even delivering
> double bounces locally, then extended messages won't be deliverable at
> all, and qmail will throw them away.
I expect messages that qmail creates be routable through SMTP, qmail
cannot assume it can send the bounce through QMQP.
If it cannot bounce the binary stuff of the mail body, it can still
bounce the mail headers.
That's what RFC-1123 has called losing mail for frivulous reasons, which
is expressly forbidden.
If the transition to binary-clean message transports comes at the expense
of breaking non-delivery notifications, that's a fundamental design
flaw.
Looks like QSBMF is not SMTP clean if you assume QSBMF also applies to
binary mail. Not pretty useful then.
--
Matthias Andree
