> If it's a small range of gnupg versions which fail badly when the GNUPGHOME
> dir is removed, then there's far less reason for git to do much more than
> make an effort to kill the agent.
> 

Yeah. FWIW, it may be reasonable to consider dropping the patch once we are
certain distros don't ship this range anymore.

> It seems like all the gnupg versions which may suffer from the bug also
> support gpgconf --kill, which would make any further change in the test to
> handle versions which lack the --kill option moot.
> 

This is also true. We should definitely not litter the stdout with ENOENT-like
error messages though...

Thanks again for catching this!
-Santiago.