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?

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.

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.

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.

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.

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.

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.

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.