Replace kilic/bls12-381 with gnark-crypto/ecc/bls12-381

[amsanghi]

Fixes : NIT-2721

[PlasmaPower]

This looks correct to me, but I have a few recommended simplifications and test cases

[PlasmaPower]

I pushed up a fuzzer for the new impl vs the old impl to the new-bls-impl-fuzzer branch. It's currently failing, I believe because the new impl accepts compressed points but the old impl does not. For the public key and signature deserialization in the new impl we should ensure the compressed bit (I think the most significant bit from the docs) is not set. Private key deserialization isn't relevant to this though, as it's only done with local trusted input.

[amsanghi]

I pushed up a fuzzer for the new impl vs the old impl to the new-bls-impl-fuzzer branch. It's currently failing, I believe because the new impl accepts compressed points but the old impl does not. For the public key and signature deserialization in the new impl we should ensure the compressed bit (I think the most significant bit from the docs) is not set. Private key deserialization isn't relevant to this though, as it's only done with local trusted input.

Thanks for running the fuzzer on this, updated!

[PlasmaPower]

The new impl is now panicking with panic: runtime error: index out of range [0] with length 0 if passed an empty signature or public key bytes. I think your new if statements should be prefixed with e.g. if len(in) == 0 || .

You can also run the fuzzer yourself locally by checking out the new-bls-impl-fuzzer branch and running go test ./blsSignaturesNew -v -fuzz FuzzSignatureSerialization . The fuzzer also catches this new panic.

[amsanghi]

FuzzSignatureSerialization

Thanks for catching this, updated! Also started running fuzzer to see if it catches something new.

[amsanghi]

FuzzSignatureSerialization

Thanks for catching this, updated! Also started running fuzzer to see if it catches something new.

Its failing with
new_vs_old_fuzz_test.go:25: signature mismatch: 000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000 vs 400000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000

It cause the first byte is being set to mUncompressedInfinity in the new setup, any thoughts on how to handle this?

[PlasmaPower]

I think that's a cofactor issue.. Maybe try calling ClearCofactor after deserializing points in the new impl? I'm also getting a different fuzz failure locally with the input 00000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000030 . I think we need to require it's exactly 96 bytes in the new version.