[fantastisch]

Great question, and thank you, Salvatore!

Some reasons for that are: different hashes for s and s\0 (without explicitly using length) to make it harder to create collisions, related to as you say better distribution/avalanche for short strings, also design-wise I like the asymmetry that strings not perfectly divisible by 8 will be treated a little differently. I just think that makes it better overall.

[antirez]

Thanks for the replies!

[fantastisch]

No probs. I think your idea of unifying the two loops is a good one.

A new version that uses the idea and overcomes the challenges would be great, and would simplify the code even more.

If you had a way to make a start on it, I welcome your PR.