Provide a constructor for Hash

[tailhook]

It would be nice to be able to construct Hash from string or bytes to compare it to the computed hash, instead of comparing the string/bytes representation.

Optional serde support would be even better.

[oconnor663]

I forgot that the Hash type in these crates were missing the From<[u8; N]> impl. I should probably just add it.

The version of this type in the blake3 crate has more methods, and probably all of them would make sense here too. I wonder if it's time to consider splitting it out into a standalone crate, which could be shared by many different crates? On the other hand, would we rather keep the property that comparing a BLAKE2 hash to a BLAKE3 hash with == is a type error? Not sure.

Also I'm sure the RustCrypto folks (who maintain the digest crate) have some thoughts about this. @tarcieri is working on merging implementations from this repo into their blake2 crate also. Tony, have you guys ever considered defining a shared Hash type or something like that? I know the Digest trait works with GenericArray, but now that const generics don't require any sort of hacks, it might make sense to define a new type (with constant-time equality, hexification, and so on).

[oconnor663]

@tailhook what do you think of #26 as a first step?

[tailhook]

Yes. This is a good start! At least I can do all other stuff (parse from string) myself now.

On the other hand, would we rather keep the property that comparing a BLAKE2 hash to a BLAKE3 hash with == is a type error?

I like this idea. It's better than having hash type by length (i.e. Hash32/Hash64), and clearer than using generic ConstTimeCmpArray<u8, 32>.

[tarcieri]

Tony, have you guys ever considered defining a shared Hash type or something like that?

This is what we presently have in the digest crate:

https://docs.rs/digest/0.9.0/digest/type.Output.html

I know the Digest trait works with GenericArray, but now that const generics don't require any sort of hacks...

Unfortunately min_const_generics are not sufficient for our use cases. See ongoing discussion here:

RustCrypto/hashes#303

[tarcieri]

define a new type (with constant-time equality, hexification, and so on).

This sounds a bit more like the current crypto_mac::Output newtype:

https://docs.rs/crypto-mac/0.11.1/crypto_mac/struct.Output.html

I believe @newpavlov has plans to unify digest and crypto_mac, although I don't know the plans regarding the Output types specifically.

[oconnor663]

I just landed the PR above. As part of publishing these changes, I'm leaning towards just tagging these crates v1.0.0 and calling everything stable. (Which wouldn't rule out any future enhancements to the Hash type, of course.) @tarcieri would that be alright with you? Any impact to blake2 from doing that?

[tarcieri]

That sounds fine to me. I guess the only thing to worry about is porting over the changes to blake2.

[oconnor663]

Published blake2b_simd, blake2s_simd, and blake2_bin versions 1.0.0.