There's no way to get an ultrahonk verification key through `UltraHonkBackend`

[TomAFrench]

Currently it's impossible to get the verification key for a circuit from the backend without directly using the cbinds. This requires the user to:

1. Spin up another instance of barretenberg
2. base64 decode and decompress their bytecode before passing it to acirWriteVkUltraHonk
3. Know that they need to wrap the response in a RawBuffer and pass it to acirVkAsFieldsUltraHonk

What should be simple then requires copying out bb.js internals and multiple function calls.

[ledwards2225]

Does UltraHonkBackend::getVerificationKey not work for this purpose?

[TomAFrench]

This returns a Uint8Array but in order to feed this into a circuit I need to get the verification key as fields. I don't know offhand how to decompose the binary representation into fields.

[saleel]

generateProofForRecursiveAggregation() got removed in 13217. This method used to return proof and VK in fields which we were able to feed in to nargo.execute() of the recursive circuit IIRC.
I think we need to bring it back.

[ledwards2225]

@saleel Is this something you're able to support us on?

[johnathan79717]

@TomAFrench: Could AztecProtocol/aztec-packages#19718 solve this issue?

[TomAFrench]

That output looks great for feeding into a Nargo.toml!

[johnathan79717]

Now that the bb CLI can read/write JSONs. Does that solve this issue or do we need to hook that functionality into bb.js?

[TomAFrench]

Not the JSON functionality exactly but if we look at UltraHonkBackend, we can see that it's not obvious on how to generate the necessary inputs for recursive verification to be fed into another circuit:

https://github.com/AztecProtocol/aztec-packages/blob/71ffa49969f4c11e42ca685e382c5789a4ed8b2b/barretenberg/ts/src/barretenberg/backend.ts#L157-L277

1. We get a proof back as a byte array which then needs to be converted into fields.
2. We get a vk back as a byte array which then needs to be converted into fields.
3. There's no method to get the verification key hash

We can't assume that the user will know the correct process to convert from the byte to field representations. While it's currently just chunking every 32 bytes, this has changed in the past and we can't just assume that knowledge on the part of the user anyway. We're also requiring the user to know barretenberg internals on how they should calculate the vk hash.

We should have some helper functions convertVkToFields(vk: Uint8Array): string[], convertProofToFields(proof: Uint8Array): string[] and getVkHash(vk: Uint8Array): string to have a fool-proof method of doing these conversions.