gpgme 1.5.5 tests failing

[Daniel Kahn Gillmor]

On Fri 2015-06-12 15:43:41 -0400, Daiki Ueno wrote:
> Daniel Kahn Gillmor <dkg at fifthhorseman.net> writes:
>
>>>> t-encrypt.c:60: GPGME: Inappropriate ioctl for device
>>>
>>> That is:
>>>
>>>   err = gpgme_get_key (ctx, "A0FF4590BB6122EDEF6E3C542D727CC768697734",
>>> 		       &key[0], 0);
>>>   fail_if_err (err);
>>>
>>> Any other special configuration of gpg or gpg-agent?
>>
>> Not that i'm aware of.  The headless build system where these errors
>> show up is using cowbuilder inside a chroot.  This means no
>> pseudo-terminal is directly attached to the build/test process -- i
>> don't know whether that's relevant.  i'm stumped by this right now :/
>
> The error looks similar to the one I got for ruby-gpgme, when running
> on Heroku:
> https://gist.github.com/vovayartsev/a466aa82320b84f32b13
> (the original issue: https://github.com/ueno/ruby-gpgme/issues/26)
> where it seems to be failing in ttyname_r usage in GPGME.
>
> I have not tried it by myself, but ArchLinux people tweak the mount
> option of /dev/pts to work around this:
> https://lists.archlinux.org/pipermail/arch-projects/2011-October/001953.html
> https://lists.archlinux.org/pipermail/arch-projects/2011-October/001994.html

Wow, thanks for finding this!  it does look relevant.  I'll poke around
in cowbuilder and pbuilder to see if there's a /dev/pts issue i can sort
out there.

But...  should gpgme require the existence or proper configuration of
/dev/pts in some particular mode to be able to work?  It seems to me
that a system without /dev/pts at all should still be able to use gpgme,
right?

        --dkg