switch MurmurHash3 to rapidhash ?

[adienes]

probably better as an issue than a zulip thread

rapidhash is a non-cryptographic hash I've seen picking up some steam in a lot of other ecosystems.

its predecessor, wyhash, is already the default hashing algorithm in Zig, Nim, and Go. and in comparison rapidhash is both faster and a higher quality hash.

I did a sloppy mockup (pure Julia) and found suggestions that hashing UInt64 can be already 30% faster than current Base.hash and since pretty much all the other hash methods end up calling hash(::UInt64) then performance improvements here could propagate a long way. And due to the higher quality hash, seeing fewer collisions can also be a big performance benefit for many data structures

It's also a keyed hash with secrets easy enough to generate on startup if desired, so it could help address #37166 in practice (although not fully cryptographically)

if this seems like something that could plausibly be merged in a 1.x release I'd be interested in working on this.

[oscardssmith]

We very specifically exclude outputs of hash (and rand) from our semver specification, so this is definitely feasible in 1.x.

[adienes]

#53855 would probably be pretty useful here

[JeffBezanson]

Great, let's do it :)