Simon Josefsson <jas@extundo.com> wrote:

Hi Simon,

first of all, fore some reasons you got lost from the list of CC's so
you missed some important messages in this thread, including a more
elaborated version of my patch...

> 1) Non-agent mode didn't work at all (the args list contained 'nil' so
>    invoking gpg failed with a lisp error).

Yes.  This was fixed in my second patch.  (and in the attached, of cause)

> 2) After fixing the above, the patch made the passphrase be 't', which
>    broke the passphrase cache, causing a lisp error.

Thanks, I used your strategy to fix this issue in my new patch (which
I attached).

> How about this patch instead?  It works for me.  I've also improved
> the pgg-use-agent default value.

This is very error prone.  As I wrote before on emacs-devel: there are
certain situations in which using the agent will fail -- even if it is
available: for example: on an text console running the standard
pin-entry program (which uses curses) from within emacs won't work.
That's why I think this option should be explicitly enabled by the user.

Another problem is, that checking for $GPG_AGENT_INFO is not
sufficient, the environment variable will stay set, even is the agent
was killed.

Could you please consider committing the attached patch, which
reverses the default of pgg-gpg-use-agent to nil, and adds an check if
the agent is _really_ available, when pgg-gpg-use-agent is set to t?

(The patch is against the latest CVS, including your patch that is)

cheers
sascha