Add finish_buf output function for 128-bit hasher

[dhardy]

This is a proof of concept.

The idea is to make it easy to reproducibly seed any PRNG (see Seed type) from any input with a "good quality" seed.

To make it portable it also needs Hasher to be portable see this thread.

It's not intended for cryptography (and definitely not for password hashing).

[jedisct1]

Hi Diggory,

I'm not a big fan of this.

The point over the current finish() function is not obvious. It seems to be designed for a very niche use case that I don't completely understand (are you willing to use SipHash-core as a stream cipher?) and the implementation doesn't look very idiomatic.

[dhardy]

Hi Frank,

You're right, this is probably very niche in usage. This is my original attempt to build a universal PRNG seeder; I also experimented with the Keccak team's KangarooTwelve. I don't want to include a complete hash function in rand or rand_core though, so tacking the functionality required onto an existing crate has some sense. Alternatively maybe I should create a new crate for this.

I suspect SipHash's core will work fine as a stream cipher, though I guess I should at least pass the output through PractRand.

It's also terrible code style using so many raw pointers, but this was more a proof of concept (I tried to use split_at_mut but had lifetime issues).

I pushed a second commit which is more generally applicable. BTW you really ought to add CI tests on a BE platform (e.g. like this; it also needs the global install and script commands). I believe both versions of the function are correct however.

[dhardy]

Well, I created a new crate instead: rust-random/rand#554