Re: Any way to captuer smtp output when sending mail?

[W. Greenhouse <wgreenhouse-sGOZH3hwPm2sTnJN9+BGXg@public.gmane.org>]

Harry Putnam <reader-kFrNdAxtuftBDgjK7y7TUQ@public.gmane.org> writes:

[...]

> Perhaps there are good reasons to do it and I would be better off
> spending my time getting setup for that.
>
> Any speed or cpu differences of note?
>
> Any other points infavor of all smtpmail.el?

As I see it, the pros of smtpmail.el are:

1. Fewer moving parts--sometimes following the "UNIX philosophy"
   to the extreme for a problem like email means that it's hard to know
   what has failed when you change something in your delivery pipeline
   and it stops working:
   
   Is your MTA using the right smtp server for each of your outgoing
   mail addresses?

   Is mail getting stuck in a queue somewhere and you'd never know
   unless you poked around in your logs?

   ...etc, etc.

2. Emacs can store SMTP authentication credentials in an encrypted file,
   instead of leaving them sitting in plaintext on the disk (I know that
   some MTAs can store them as salted and hashed values after which you
   can delete the plaintext original, but Emacs with authinfo.gpg is
   even better).

3. Fairly tuneable behavior as far as TLS usage. Emacs unfortunately has
   nasty "fail open" behavior as a TLS client per default, and there are
   a couple of bug reports open about this, but it is possible to tune
   it to implement a "Trust on First Use" or certificate pinning scheme,
   which will "fail closed" if an unfamiliar certificate is seen
   (similar to the behavior of openssh when seeing a new host). See

   https://blogs.fsfe.org/jens.lechtenboerger/2014/03/23/certificate-pinning-for-gnu-emacs/

   for an example.

The cons are:

1. Delivery through smtpmail.el is synchronous, blocks your Emacs, and
   can be slow. Normally when using the /usr/sbin/sendmail binary on a
   *nix system, you are only copying standard output to a queue file,
   and delivery happens separately in the background when the mailserver
   flushes the queue, so this can be faster (and more importantly feels
   faster because you aren't sitting there waiting for your Emacs to
   return control to you :-) ). John Wiegley's async.el
   library comes with an example which works around the smtpmail
   blocking problem:

   https://github.com/jwiegley/emacs-async

   but because that library relies on using a second Emacs instance to
   do the background work, there can be issues with error reporting to
   the "master" Emacs, and getting this to work with encrypted auth
   credentials is pretty painful.

2. There is no easy recipe built into Gnus and smtpmail.el for the very
   common modern case of having several mail identities and needing to
   use a different outgoing smtp server for each. For that you need to
   figure out (info "(gnus) Posting Styles") and (info "(message) Mail
   Variables") and the care and feeding of the "X-Message-SMTP-Method"
   header, or else use an addon library like

   http://www.emacswiki.org/emacs/GnusAlias

   to make it easier to switch ougoing servers. Most MTAs have some
   example configuration for the "sender dependent smarthost" problem
   and it just works transparently after you set it up. Or people use a
   special tiny remote-client-only MTA like msmtp, which handles this
   multiple-outgoing-servers case brilliantly using just the -f flag to
   /usr/sbin/sendmail.