Even worse: users of TweetNaCl can easily switch over to the real deal (performance-wise) — NaCl or libsodium — without changing anything, while with this library they'll have to grab pieces of optimized code from various sources, and then dig into insides Ed25519 implementation to replace SHA-512 with BLAKE2b. Unless the author implements an optimized version of Monocypher, of course.
> Unless the author implements an optimized version of Monocypher, of course.
We may be able to compete with ref10 without too much code. I haven't found a way yet (very frustrating). In the mean time, you have the Donna implementation, which can be used with any hash (it's a compilation option).
