[openpgp] Fingerprint schemes versus what to fingerprint

[Bryan Ford <[e-mail]>]

To followup on today’s in-meeting discussion of evolving OpenPGP fingerprints beyond SHA-1, I want to propose that there are at least two orthogonal issues to decide (and I’m probably not the first to suggest this):

1. What fingerprint scheme(s) should OpenPGP move to going forward?

2. What exactly should the OpenPGP “application” fingerprint with that scheme?

To clarify, I propose to define a “fingerprint scheme” as an algorithm that takes a raw octet string and produces an ASCII string of some kind for users to cut-and-paste, compare, read off over the phone, etc.  By this definition, just like a cryptographic “hash scheme” or “signature scheme”, the “fingerprint scheme” itself doesn’t need to know or care what octet string gets fed into it.

As such there’s no reason such a “fingerprint scheme” itself needs to be in any way specific to OpenPGP, and I would support the proposals that Phillip and others have made that it would be ideal to standardize future fingerprint scheme(s) independently of particular protocols such as OpenPGP, and just have OpenPGP use that scheme.  CFRG might be the obvious place to do this.  Of course, I understand the logistical downsides of having an OpenPGP work-item depend on work elsewhere (e.g., CFRG) that isn’t even started yet…  But this approach might still be worth considering from a “get it right” perspective if there isn’t currently some kind of severe time pressure on the OpenPGP side.

The other, very OpenPGP-specific, question is of course what exact octet-string should get fed into whatever fingerprint scheme is chosen.  DKG  brought up the question of whether that octet-string should still include the Unix timestamp like it currently does.  I think that question leads to a bigger set of issues that I’ll try to tease apart in a subsequent E-mail.

But first I just wanted to propose this explicit separation of the two questions, “which fingerprint scheme?” (i.e., which function from octet-strings to ASCII-strings), and “what to fingerprint?” (how does OpenPGP get from a key to the octet-string to feed the fingerprint scheme?).

Thanks
Bryan