Compatibility with RustCrypto

[daxpedda]

Hi!

I'm currently working on an implementation of the Balloon hashing algorithm for the RustCrypto organization: RustCrypto/password-hashes#232. In an effort to make this viable, I am seeking to provide compatibility between different implementations.

Only two changes are required to provide compatibility:

• Change encoding of values to little-endian as is done in the prototype implementation
• The ints_to_block function was interpreted as hashing these values and produce a "block" of bytes; the prototype implementation uses some sort of encryption here, I merely replaced it with the hash already used

Any input is welcome! I added test-vectors to assure compatibility. I am willing to contribute a CI setup, just say the word!
I have never programmed in Go before!

Another library I provided compatibility for: nachonavarro/balloon-hashing#2

Addresses #1.

[daxpedda]

math/big only takes big-endian, I copied this reverse slice implementation from here:
https://github.com/golang/go/wiki/SliceTricks#reversing

[darenliang]

I was messing around with this project and found out that there exists a function in math/big that takes little-endian.

I've submitted a pull request on your fork a while ago, but sadly I think this original repo has gotten stale. If there's more interest, I might spend some time to optimize the implementation.

khonsulabs#1

[daxpedda]

Just saw your comment here, sorry for the delay!
@nogoegst has to look at this PR, I really don't feel comfortable reviewing unsafe Go code.

@darenliang personally I have no stake in this library, I was only trying to help multiple Balloon implementations be compatible with each other, specifically the Rust library I contributed to. So if you are more interested in using Balloon with Go and want to have common compatibility, feel free to contact me.

[unkaktus]

Hi!
Thanks for your pull request, looks good to me!
I will only replace the slice inversion via more straightforward function - there should be one in math.big.

[daxpedda]

I did try to replace the slice inversion. What I did was basically make a new slice with the math.big.Word type, copied over the bytes, then used math.big.Int.SetBits to convert it. This involved a bit of trickery and didn't really improve the code, I'm open to suggestions!