Add two Poseidon Filecoin variant multihashes

[rvagg]

Reserving the 0xb400 range for Poseidon variants. These entries allow FIL to iterate on the fcX extension of the name where they stay with BLS12-381 and arity=2. High security variant is for extra circuits that are usable in case new attacks arise from the standard variant.

• The primary ways of differentiating Poseidon implementations are curve and arity, hence their encoding in the name.
• There are additional parameters, hence the need for an extra differentiator fc1.
• Filecoin may change these additional parameters in future versions, and therefore need to differentiate and iterate on the fcX extension.
• When parameters are initially generated, extra circuits may also be generated, providing a high security, but more costly, variant at the same time. Hence the sc addition. This may or may not be used, but will exist alongside the standard variant and be deployable if new attacks render the standard variant less secure.

Implementation: https://github.com/filecoin-project/neptune

Ref: #161
Ref: https://eprint.iacr.org/2019/458.pdf

R= @vmx @mikeal @dignifiedquire @porcuquine @Stebalien

[porcuquine]

The above is a good summary of our long discussion earlier today.

[vmx]

I use some automatic conversion in JS libs in regards to dashes and underscores. I meant to document those, but haven't really. But the rules are (IIRC) only use alphanumeric characters and dashes. Would it be ok to switch this to the slightly less readable `poseidon-bls12-381-a2-fc1?

[rvagg]

Is that able to be changed? the reason that change bothers me is that you can no longer split it by logical sections. If we end up with 50 of these, - is no longer a useful separator without additional logic. Replacements for BLS12-381 include bn and ed for now, possibly more in the future, including other BLS12-XXX variants. But you can see with bn how the scheme would break down. poseidon-bn-a32-boop1?. bls12381 is an option too, just less clear I think.

[rvagg]

Feedback from FIL team is bls12381 would not be the worst thing in the world, bls12q381 is an option too, but still not optimal; a clear separator is preferred.

[vmx]

To explain what I'm doing. I create constants out of the names. Constants (in many languages) are all uppercase with underscore as separator. Converting it one way is easy poseidon-bls12_381-a2-fc1 to POSEIDON_BLS12_381_A2_FC1. Probably you would keep the original around in order to use it in error message.

I just though it would be nice to have the option to have a two-way conversion back to the original name, but that isn't a must. I thought restricting the characters would make things simpler, but if it makes things harder to use/read, we can surely have more characters.

[rvagg]

Ah yeah, I can feel the awkwardness of that. And the significance of - vs _ would get lost in the conversion because they'd all be _ anyway. Maybe we just need to force bls12q381 on them.

[porcuquine]

One quick thought, in case it can rescue this: what if you convert _ to __ (double underscore)? As long as we also established the rule that source names will never contain two of either - or _ in a row, then that would be unambiguous and fairly legible — preserving the distinction visually as well.

[rvagg]

yeah, although __ can get out of hand really quickly if people aren't conscious of it in their naming - if @vmx is the one responsible for converting and the people adding names here aren't consuming those constants then they're not going to feel the pain of FOO__BAR_2_3__456__BLIP_BORK. I'll leave the call to @vmx because it'd be a policy change.

[vmx]

Given that the conversion from a constant back into the name is rather hypothetical (I checked the JS code again, I don't think I ever do that), I think it's fine to have deterministic rules to convert from name into an all uppercase constant name. Hence I suggest that we just allow underscored, which will stay underscores. Allowed characters would then be [a-z0-9-_].

So I would expect in most languages poseidon-bls12_381-a2-fc1 being used as a constant named POSEIDON_BLS12_381_A2_FC1.

Give a thumbs up if that sounds like a plan.