On Fri, Apr 05, 2013 at 09:23:42AM -0400, Wietse Venema wrote:
> Wietse Venema:
> > I've done a proof-of-concept implementation that works as documented
> > below the signature.
>
> I was able to simplify this further. The result is below.
> Comments are welcome.
One issue this does not discuss is the handling of:
propagate_unmatched_extensions = canonical, virtual
a relay that accepts multiple extensions and validates addresses
via relay_recipient_maps, may forward mail via SMTP to downstream
destinations which handle a subset (possibly none) of the supported
extensions. This can create bouncebacks.
Even with a single recipient delimiter (say "+"), I've had to set:
propagate_unmatched_extensions = canonical
so that envelope recipients forwarded to Microsoft Exchange were
not extended, since Exchange does not support extensions.
The general picture is more complex, since while MUAs only need
extensions in headers to help users sort incoming mail, the delivery
MTA (e.g. qmail or Postfix via forward_path, ...) uses the envelope
recipient.
So a complete implementation possibly needs to be able to determine
the correct downstream recipient delimiter based on the destination
nexthop or transport:nexthop. In recursive virtual (or canonical)
expansion this logic need only apply to the final address.
I'm also concerned that matching the first delimiter is problematic
in mixed environments. When a relay sits in front of two domains
example.com (whose extension is "+") and example.net (whose extension
is "-") we don't get correct behaviour:
postfix-users+extens...@example.com
user-extension+more-extens...@example.net
the relay would bounce the "postfix-users+extension" mail, as it
would misinterpret this as being addressed to "postfix", unless in
fact multiple lookups are made, and the recipient delimiter is
inferred from the shortest match (try "postfix" - "...", then
"postfix-users" + "...").
If we do add support for destination specific address extensions
on output, what should be done with the wrong extension on input?
Hypothetical:
postfix-users-mum...@example.com
is just an invalid address when example.com is a "+" delimiter domain.
So I'm not entirely convinced we're not opening up a bit of a can
of worms.
--
Viktor.
