Re: [openpgp] Fingerprint requirements for OpenPGP

[Jon Callas <[e-mail]>]

> On Apr 11, 2016, at 5:40 PM, Daniel Kahn Gillmor <[e-mail]> wrote:
> 
> Is this problem framed correctly?
> 
> If not, what's missing?

Actually, I don't think it's framed correctly.

I don't think distinguishing between local and remote databases is necessary; I know I'm picking a nit here, but the design of many systems is such that there isn't a difference. But moreover, a real remote database is as likely to need to answer the question, "give me the key(s) for <identifier>" where the identifier might be a fingerprint and might be something like an email address.

Like I wrote before, there are two major uses for a "fingerprint," a handle and an auth string. 

Moreover the auth-string use case of fingerprints is actually the auth string of the crypto key that is the top-level signing key of the "PGP key", which is a data structure that usually consists of at least two crypto keys.

But the handle that turns into a key id is presently derived from the fingerprint of a subkey that's used for encryption.

I think it's completely reasonable to update the standard to take the old-style fingerprint into better hash functions than SHA-1. It's imperfect but it's the devil we know.

If you want to move into devils we don't know, then why *not* make an auth-string that is a different thing than handle?

As you're describing the use cases, you've described something that you can't solve. There's nothing wrong with noting that both circles and squares have desirable properties, but you're not going to get space-filling circles or a square that's got the perimeter equidistant from the center. 

	Jon