On Wed, May 26, 2021 at 09:27:13AM +0300, Nikolay Borisov wrote:

On 26.05.21 г. 8:51, Neal Gompa wrote:

quoted

I just read this article on BLAKE3 1.0 coming soon[1], and the
performance of BLAKE3 looks extremely compelling. With the context of
thinking about Btrfs authenticated hashes and fsverity[2], it seems
like it would make sense to see if BLAKE3 can provide us roughly
similar performance to xxhash or crc32. If so, that would make it
highly attractive for a new default hash, too[3].

Especially in the context of security I'd say we shouldn't be quick to
jump on the bandwagon. Let the algorithm be ratified, let it simmer in
the public for a while to see if any weakness are found and then we can
consider adding it.

Agreed, this was the idea behind choosing BLAKE2 a few years after it
got adopted by other projects. In a ~five year horizon we can evaluate
the hashes again. Meanwhile I'd like to make some progress with the
authenticated hashes and for that we have enough to choose from.