Hi folks--

On Sun 2020-06-28 08:33:42 -0300, David Bremner wrote:
> I dug a bit further down, and this is what is returned from gpgme
> (line 345 in g_mime_gpgme_get_signatures)
>
> sig = {next = 0x0,
>     summary = GPGME_SIGSUM_KEY_MISSING, 
>     fpr = 0x4ac480 "5AEAB11F5E33DCE875DDB75B6D92612D94E46381", status = 9, 
>     notations = 0x0, timestamp = 1559167762, exp_timestamp = 0, wrong_key_usage = 0, 
>     pka_trust = 0, chain_model = 0, is_de_vs = 0, _unused = 0, 
>     validity = GPGME_VALIDITY_UNKNOWN, validity_reason = 0, 
>     pubkey_algo = GPGME_PK_RSA, hash_algo = GPGME_MD_SHA256, pka_address = 0x0, 
>     key = 0x0}
>
> At this point I'm leaning towards declaring it a gpgme problem in
> fedora32, and suggesting that relevant distros mark the test broken. I
> am of course open to more informed opinions.

The problem does indeed appear to be with gpgme, in versions 1.13.0 and
1.13.1.

In particular, it is a problem with the resolution of
https://dev.gnupg.org/T3464, which is ultimately fixed upstream, but is
not yet fixed in a released version of gpgme.

The upstream commit, which should be patched into gpgme on Fedora and
OpenSUSE (and anywhere else that depends on gpgme) is:

    https://dev.gnupg.org/rMae4d7761a15b82eb98b0bcc72af2ae2e8973e1f9

(patch attached here as well)

We don't see this on Debian because gpgme in debian has carried this
patch for over a year now.

In gpgme 1.12.0 and earlier, this bug did not exist.  But gpgme 1.13.0
introduced the bug in an attempt to avoid error diagnostics when *not*
trying to verify a signature while using a session key.  The fix in
1.13.0 inadvertently introduced an error when the caller does actually
try to verify a signature, which is what we see here.

      --dkg