Yet another attempt at adopting IETF BLS Standards

[CarlBeek]

This is an alternate approach to #1398 and #1499 with the goal of having complete separation between BLS libraries and the spec.

Initially I attempted to include @JustinDrake's requests in #1499, but it got very ugly very quickly. So this is a rewrite that deprecates #1499.

@JustinDrake, I believe this addresses your concerns from #1499, please let me know whether it does.

[djrtwo]

Awesome! This looks and feels clean

How far are we from having a conforming py_ecc?

[CarlBeek]

How far are we from having a conforming py_ecc?

Pretty darn close. I opened a big PR this morning. It passes all tests etc.

Annother point I'd like to raise for clarity is whether we should treat BLS as a module. Is bls.Verify() clearer than Verify? In a similar vein, we could alias all bls functions: bls_verify = Verify etc.

[djrtwo]

I like the idea of treating it like a module for readability. You can easily jump to a random part of the spec and see bls.Verify and know what is going on.

A quick note that they are used under the bls module in the "BLS Signatures" section for added clarification.

As for bls.Verify vs bls_verify... Keeping the names in line with the standard seems fine even though it breaks pythonic convention. I like the easy mapping to the standards spec. I find myself searching that doc from time to time and it's easy to see what to search for.

[CarlBeek]

BLS is now a module and I think it is much clearer.

If we can reach agreement on what to call roots before they are signed/verified then we can get this thing merged. 🎉@djrtwo, can I ask for your input on our comment chain above re: wrapped_root.

[djrtwo]

wrapped_root is reasonable but wrapped doesn't give too much description on what the wrapping is.

message is reasonable except that the signed containers use message for the entire object. message_root might work well because it is the root of the message from the signed container.

I'm okay with message or message_root

[CarlBeek]

The way domain-bytes were handled was a hack which I perpetuated with the change of domain into bytes in lieu of ints cast to bytes.

BLS_WITHDRAWAL_PREFIX had a similar problem and so I fixed both in #1551

[CarlBeek]

Having replaced all the domain_wrappers with signing_roots, I think this PR is semantically complete. 🎉 The only thing that should change first is the BLS API in the tests. As soon as ethereum/py_ecc#83 is merged and a new release is cut, I can change the API in this PR and then merge.

[djrtwo]

Very happy with this.
Thanks @CarlBeek !

edit: waiting on the py_ecc release before merge

[djrtwo]

ethereum/py_ecc#83 is merged!

[JustinDrake]

@CarlBeek: What is the status of PoP separation domain? Is that nailed or still somewhat in flux relative to the standardisation effort?

waiting on the py_ecc release before merge

Do we also want basic tests?

[djrtwo]

Do we also want basic tests

@JustinDrake once we integrate py_ecc here and all existing tests run, we'll have satisfied the testing requirement. [py_ecc independent tests and all of the existing eth2 test suite]

[djrtwo]

• added version bump in a couple more places
• bumped circlecli cache version to resolve CI failure
• fixed merge conflicts

LGTM!

[CarlBeek]

• added version bump in a couple more places
• bumped circlecli cache version to resolve CI failure
• fixed merge conflicts

I was just trying to push these changes myself, but I see you've already done them. Looks good to me! Ready for merge.