notmuch reply From:-guessing heuristics

Subject: notmuch reply From:-guessing heuristics

Date: Fri, 09 Dec 2011 10:53:04 +1300

To: notmuch@notmuchmail.org

Cc:

From: Michael Hudson-Doyle


Hi all,

I have a specific example where "notmuch reply" does not set a useful
from: header.  The set is:

1) I have a google apps-for-your-domain address,
   michael.hudson@linaro.org

2) There is a group/alias, validation@linaro.org that only accepts mail
   from members

3) michael.hudson@linaro.org forwards to michael.hudson@canonical.com,
   from where I get email with offlineimap.

4) michael.hudson@linaro.org and michael.hudson@canonical.com are both
   configured as addresses in ~/.notmuch-config with the linaro.org
   address being primary.

When I reply to a mail to validation@linaro.org, the guessed address is
michael.hudson@canonical.com and unless I change it, the reply bounces,
which is obviously a bit annoying.

Looking at the headers (and the notmuch source),
michael.hudson@canonical.com seems to be being found in the Received
headers.  michael.hudson@linaro.org is not in any Received header, which
is a bit strange, but in any case I think the code would find
michael.hudson@canonical.com first (would it perhaps make more sense to
search backwards through received?  An earlier header is probably closer
to what was intended, in some sense).

michael.hudson@linaro.org does appear in a Delivered-To: header, so
maybe those could be considered (it seems that some MTAs add a
Delivered-To header from the envelope address, so it would have some
legitimacy)?  It would work in this case, but only because the
Delivered-To that the canonical.com MTA adds is an internal address
that's not configured as an address for me in notmuch...

Alternatively, *I* wouldn't mind if notmuch stopped trying at all hard,
and just used the primary address if there was nothing matching in
to:/cc: but I guess that wouldn't work on mailing lists at all...

Hm, I guess I've argued myself around to thinking that considering
Delivered-To as a source of potential from addresses would be an
improvement.  What do you guys think?

Cheers,
mwh

Thread: