Do users need to set `mml-secure-signer-names' (etc) now?

Do users need to set `mml-secure-signer-names' (etc) now?

[Karl Fogel]

For a while now, sending GPG-signed emails from message mode (just signed, not encrypted) has been failing for me with an error from `mml-secure-epg-sign':

  "Couldn't find any signer names."

This is a regression -- in the past, message mode would correctly figure out the signer identity from the message's sender.

(Note that an older version of that error message suggested setting `mml-secure-smime-sign-with-sender'.  Because I encountered the error when `mml-secure-smime-sign-with-sender' was set to non-nil, I improved the error message in commit 06cb8350c69 on 4 Nov 2019: now it only makes that suggestion if the variable is nil.  However, even with that variable set, the problem still occurrs for me, and I said I would debug further -- see [1] for more context.)

I've now debugged enough to reach a question:

Is it intended that users should set at least one of `mml-secure-signer-names' or `mml-secure-openpgp-sign-with-sender' in order for message-signing to work?  

I would hope not, because using the message sender as the signing party seems like a reasonable default behavior.  Anyway, I can't proceed further without knowing the intended behavior -- I don't yet know if I'm looking at a bug or if it's simply that user-configuration requirements changed and I didn't get the memo.

Here is what led me to this question:

In lisp/gnus/mml-sec.el, `mml-secure-signer-names' returns nil if `mml-secure-openpgp-signers' is not set *and* `mml-secure-openpgp-sign-with-sender' is not set.  (This is all in the `OpenPGP' protocol case; I'm not concerned with the non-OpenPGP case right now.)  Naturally, `mml-secure-epg-sign' fails with the above error when `mml-secure-signer-names' returns nil.

So if I set `mml-secure-openpgp-sign-with-sender', then I can send signed messages again (that variable was nil by default for me).  I have never set `mml-secure-openpgp-signers' in my .emacs, and neither it nor `mml-secure-smime-sign-with-sender' are mentioned in any NEWS file.  However, `mml-secure-openpgp-sign-with-sender' is mentioned in NEWS.27:

  > *** 'mml-secure-openpgp-sign-with-sender' sets also "gpg --sender".
  > When 'mml-secure-openpgp-sign-with-sender' is non-nil, message sender's
  > email address (in addition to its old behavior) will also be used to
  > set gpg's "--sender email@domain" option.
  > 
  > The option is useful for two reasons when verifying the signature:
  > 
  >  1. GnuPG's TOFU statistics are updated for the specific user id
  >     (email) only.  See gpg(1) man page about "--sender".
  > 
  >  2. GnuPG's "--auto-key-retrieve" functionality can use WKD (web key
  >     directory) method for finding the signer's key.  You need GnuPG
  >     2.2.17 to fully benefit from this feature.  See gpg(1) man page for
  >     "--auto-key-retrieve".

I don't fully understand that NEWS entry, but the "(in addition to its old behavior)" implies to me that defaulting-to-sender-as-signer was perhaps never intended to break?  I can't tell for sure, because I'm not sure what the old behavior was.  The only other mention of that variable in a NEWS entry is this, chronologically earlier (lower down) in NEWS.27:

  > ---
  > *** 'epg-context' structure supports now 'sender' slot.
  > The value of the new 'sender' slot (if a string) is used to set gpg's
  > "--sender" option.  This feature is used by
  > 'mml-secure-openpgp-sign-with-sender'.  See gpg(1) manual page about
  > "--sender" for more information.

So do we intend that users should set at least one of `mml-secure-signer-names' or `mml-secure-openpgp-sign-with-sender' in order to get use-sender-as-signer behavior?  Or should Emacs just derive the signer from the sender automagically (as I believe it used to)?

This is all with recent `master'.  I haven't tested with the Emacs 27 branch yet.

Best regards,
-Karl

[1] See these two posts:

    From: Karl Fogel
    To: Emacs Devel
    Subject: [PATCH] Improve signer name error in `mml-secure-epg-sign'.
    Date: Mon, 04 Nov 2019 14:38:00 -0600
    Message-ID: <87tv7j74yf.fsf@red-bean.com>

    From: Karl Fogel
    To: Emacs Devel
    Subject: Re: [PATCH] Improve signer name error in `mml-secure-epg-sign'.
    Date: Sat, 09 Nov 2019 15:38:55 -0500
    Message-ID: <87d0e0kco0.fsf@red-bean.com>

Re: Do users need to set `mml-secure-signer-names' (etc) now?

[Eli Zaretskii]

> From: Karl Fogel <kfogel@red-bean.com>
> Date: Tue, 28 Apr 2020 10:59:18 -0500
> 
> For a while now, sending GPG-signed emails from message mode (just signed, not encrypted) has been failing for me with an error from `mml-secure-epg-sign':
> 
>   "Couldn't find any signer names."
> 
> This is a regression -- in the past, message mode would correctly figure out the signer identity from the message's sender.

Isn't this bug#40118?

Re: Do users need to set `mml-secure-signer-names' (etc) now?

[Karl Fogel]

On 28 Apr 2020, Eli Zaretskii wrote:
>> From: Karl Fogel <kfogel@red-bean.com>
>> Date: Tue, 28 Apr 2020 10:59:18 -0500
>> 
>> For a while now, sending GPG-signed emails from message mode (just signed, not encrypted) has been failing for me with an error from `mml-secure-epg-sign':
>> 
>>   "Couldn't find any signer names."
>> 
>> This is a regression -- in the past, message mode would correctly figure out the signer identity from the message's sender.
>
>Isn't this bug#40118?

Thank you, Eli!  I didn't think to search in the tracker this time around; I should have.

The conversation there is what I should track, so I'll start doing so now.

Best regards,
-Karl

Re: Do users need to set `mml-secure-signer-names' (etc) now?

[Robert Pluim]

>>>>> On Tue, 28 Apr 2020 10:59:18 -0500, Karl Fogel <kfogel@red-bean.com> said:

    Karl> For a while now, sending GPG-signed emails from message mode (just
    Karl> signed, not encrypted) has been failing for me with an error from
    Karl> `mml-secure-epg-sign':

    Karl>   "Couldn't find any signer names."

    Karl> This is a regression -- in the past, message mode would
    Karl> correctly figure out the signer identity from the message's
    Karl> sender.

No, it wouldnʼt, it just looked like it did. But see bug#40118, as Eli
pointed out. That should answer (most of) your other questions.

Robert