Description

The BatchEntry::dummy function in circuits/src/batch_verify/universal/types.rs file is implemented as follows:

/// Creates a dummy [`BatchEntry`] for `config`.
pub fn dummy(config: &UniversalBatchVerifierConfig) -> Self {
    let num_public_inputs = config.max_num_public_inputs as usize;
    let len = F::from(num_public_inputs as u64);
    // There's no difference passing `true` or `false` here.
    let has_commitment = true;
    let vk = VerificationKey::default_with_length(
        num_public_inputs,
        has_commitment,
    );
    let proof = Proof::default_with_commitment(has_commitment);
    let inputs = PublicInputs::default_with_length(num_public_inputs);
    Self {
        len,
        has_commitment,
        vk,
        proof,
        inputs,
        commitment_hash: Default::default(),
    }
}

In the BatchEntry that is returned, len is set to config.max_num_public_inputs, which is also how many entries inputs has. However, has_commitments is true, so inputs should also include the hash of the commitment. From the rest of the code (for example, from_ubv_input_and_config) it can be seen that len should instead be only the number of public inputs apart from the hash of the commitment.

Impact

The BatchEntry::dummy returns BatchEntrys that are malformed. Due to the way BatchEntry::dummy is used in practice (for key generation), this should not have any actual impact, however.

Recommendations

To ensure the return value of BatchEntry::dummy is consistent with the rest of the code, reduce len by one:

let num_public_inputs = config.max_num_public_inputs as usize;
- let len = F::from(num_public_inputs as u64);
+ let len = F::from((num_public_inputs - 1) as u64);

Remediation

This issue has been acknowledged by Nebra, and a fix was implemented in commit 3af8a51e↗.