From: David Bremner <david@tethera.net>
To: Jameson Graef Rollins <jrollins@finestructure.net>,
	Jani Nikula <jani@nikula.org>, notmuch@notmuchmail.org
Subject: Re: [PATCH] cli: crypto: tell gmime to use gpg-agent
In-Reply-To: <87hakxpwcu.fsf@servo.finestructure.net>
References: <1361950838-22919-1-git-send-email-jani@nikula.org>
	<87hakxpwcu.fsf@servo.finestructure.net>
User-Agent: Notmuch/0.15.1 (http://notmuchmail.org) Emacs/24.2.1
	(x86_64-pc-linux-gnu)
Date: Wed, 27 Feb 2013 13:11:48 -0400
Message-ID: <87r4k1znor.fsf@maritornes.cs.unb.ca>
MIME-Version: 1.0
Content-Type: text/plain
Precedence: list

Jameson Graef Rollins <jrollins@finestructure.net> writes:

> I will argue that the "use-agent" option should be enabled.  If we force
> use of gpg-agent, then we don't allow people to opt out of using it.
> That's not very user friendly, particularly if someone has not enabled
> it for a specific reason.

But right now we force people to enable the agent globally via use-agent
if they want to decrypt mail in notmuch-cli/emacs. The proposed change
allows them to use the agent only for notmuch.

> But I think more to the point we need a little bit of due diligence of
> the effects of this before we enable it.  What happens if gpg-agent is
> not available?  What happens if there is no X session?  Tests that probe
> the various circumstances would be useful.

I don't think we should directly care about the presence of an X session
or not; the agent protocol doesn't depend on how the agent was started
afaik. 

> I do note, though, that the error messages are not very useful.  It
> would be nice if could figure out that the decryption failed because of
> lack of agent and inform the user of that.

Yes, it would be nice to detect a missing/non-responsive agent. And that
could be used by tests.

> We should probably also update the show man page to make explicit that
> an agent may be required.

We probably need to word it more strongly than that.  If the user wants
decryption then notmuch requires an agent; if they want encryption or
signing then message-mode (really probably easypg) requires an agent.
I'm not sure how it manages it, but according to Jani's experiments it
seems that message mode already uses the agent independently of the
user's config; perhaps via the '--use-agent' argument to gpg.  So the
current situation is unfortunately asymmetric.

Of course it would help if there was documentation for the emacs
interface that we could update.