doc: fix msg argument description of nonce function typedefs #1985

The docs for secp256k1_nonce_function and secp256k1_nonce_function_hardened both describe the message argument as the one "being verified". Nonce functions are only ever called from the signing functions, so the first commit changes this to "being signed".

The hardened nonce function docs also say msg is NULL if and only if msglen is 0. That is stricter than what secp256k1_schnorrsig_sign_custom enforces: it only checks msg != NULL || msglen == 0 and passes msg through to the nonce function unchanged, so a custom nonce function can see a non-NULL msg with msglen == 0. The second commit uses the same "Can only be NULL if msglen is 0" wording that sign_custom already has.

The docs for secp256k1_nonce_function and secp256k1_nonce_function_hardened both describe the message argument as the one "being verified". Nonce functions are only ever called from the signing functions, so the first commit changes this to "being signed".

Nice catch, ACK.

The hardened nonce function docs also say msg is NULL if and only if msglen is 0. That is stricter than what secp256k1_schnorrsig_sign_custom enforces: it only checks msg != NULL || msglen == 0 and passes msg through to the nonce function unchanged, so a custom nonce function can see a non-NULL msg with msglen == 0. The second commit uses the same "Can only be NULL if msglen is 0" wording that sign_custom already has.

Hm, I think this is a genuine bug. There's a non-zero probability that there's a custom nonce function out there that relies on the promise and just hasn't been called with an empty message so far. So I tend to think that we should err on the safe side here and rather change the semantics to what the API promised, i.e., add an if (msglen == 0) msg = NULL; with an appropriate comment to secp256k1_schnorrsig_sign_internal(). Ideally, this should get a test, too.