While looking at #1962 I noticed a similar gap for the signing functions. The BIP340 aux_rand buffer is passed to schnorrsig_sign_custom in the constant-time tests but is never marked as undefined, and ecdsa_sign is only tested with ndata = NULL, so the RFC6979 path that mixes in extra entropy never runs with secret inputs.
Both values end up hashed together with the secret key to derive the nonce, and callers use them to add fresh randomness, so they shouldn't influence control flow or memory access either.
- The first commit marks
aux_randas undefined before bothsign_customcalls (default BIP340 nonce function and the custom one). - The second adds an
ecdsa_signcall with a secretndata.
This is test-only and doesn't overlap with #1969, which handles the ellswift auxrnd32 case from #1962.