Re: [Cfrg] A draft merging rpgecc and thecurve25519function.

[Alexey Melnikov <[e-mail]>]

On 02/01/2015 01:16, Adam Langley wrote:
> On Thu, Jan 1, 2015 at 4:51 PM, Adam Langley <[e-mail]> wrote:
>> However, I don't agree that the algorithm is simply ignored. It's
>> clearly important to some that the generation algorithm be explicit
>> and transparent. Yes, it's odd that at the end we have to do an
>> arbitrary isogeny but I think the motivation for that is clear and
>> it's a "safe" step (i.e. we can't have hidden anything dodgy in there
>> to my knowledge.)
>>
>> The algorithm could be written to reflect how curve25519 was
>> developed: i.e. that it output a Montgomery curve and then the twisted
>> Edwards is the obvious isomorphism from there (if needed). That might
>> well be clearer if we don't recommend any other curves.
>>
>> But dealing with Edwards curves and having the algorithm generate them
>> makes it easier if we recommend another, or in the optimistic scenario
>> that we also end up specifying a signature scheme.
> I think I've fallen into the "details" trap in that reply.
>
> However the generation is structured to make it most clear, and
> whether or not we mention (twisted) Edwards curves, I think that the
> important point is this:
>
> Unless the chairs believe, and state within weeks, that we can move
> forward with something shaped like this draft (whether it's based on
> it or not)
Chairs believe that your draft is a very reasonable compromise  (in IETF 
sense, not in security sense) between technically sound proposals and 
will initiate Last Call on acceptance of your document as a starting 
point for CFRG document shortly.
> then I don't think that this RG will be able to recommend
> any specific curves. I don't think anything will be more generally
> agreeable.
Agreed.
> If we can't recommend curves then the best that we might be able to do
> is to put out a document that states the requirements for safe
> elliptic curves and mentions that curve25519, curve25519-isogeny,
> 41417, Goldilocks, E-521, etc are all fine, cryptographically. Then
> other WGs will have to do the picking.
Indeed. This outcome would be suboptimal (IMHO), but still a bit better 
than "no consensus on anything in CFRG".