Curve25519 with the prefix 0x40
gniibe at fsij.org
Thu Dec 3 01:01:55 CET 2015
On 11/26/2015 12:02 PM, NIIBE Yutaka wrote:
> I review our implementation of Curve25519 again, and I think that we
> should add the prefix 0x40 so that it matches the practice of Ed25519.
> While my patch for libgcrypt is on review, and here is the patch for
> GnuPG. Well, it seems that only change for scdaemon.
I pushed it to master, since the impact is only scdaemon, and
it means that it will be Gnuk Token users.
I think that having the prefix 0x40 for point representation of
Curve25519 makes sense and it's better than no prefix, because when we
have the prefix, it is safe to handle the octets as plain MPI. If no
prefix, some code in GnuPG for MPI puts 0x00 at the beginning, it
changes the value for the point for Curve25519.
More information about the Gnupg-devel