openpgp: prevent panic for missing self-signature

[bitfehler]

There can be public keys where an identity has no self-signature. Since
the order of identities as tested against shouldPreferIdentity is
unstable (due to being based on a map iteration), such a situation leads
to a panic in a certain percentage of runs.

Avoid the panic by preferring the potential new identity. This can still
lead to unusable results, but they will returned as a proper error,
allowing an application to handle it.

[twiss]

Hey 👋 Thanks for catching this!

I agree this is an issue, that was introduced by not considering revocation signatures as self-signatures anymore (and storing them separately), but still accepting User IDs with only a revocation signature; breaking the assumption that all Identities have a SelfSignature.

However, I think there are a few other places that have this assumption - for example, in EncryptionKey() and SigningKeyById() (which are called by most functions that use the key), the PrimaryIdentity()'s SelfSignature is also used without checking whether it exists. Probably checks should be introduced there as well? Or did you already test this?

[bitfehler]

Thanks for the feedback! I did not check this, I was in fact mostly trying to mitigate an issue I was encountering in the wild, I wasn't aware of the background. I can look into this and put a similar check in place in other places where I find this.

[twiss]

👍 Thanks! That would be great 🙏