signDigest occasionally generates invalid signatures

Perhaps I’m doing this wrong but I get different results when using web3.eth.sign (the version that Truffle 3.x uses - have not updated to 4.x yet) versus ethers.js signDigest:

Here’s my web3 code:

function generateSignedData(msg, account) {
    let seedData = web3.sha3(msg);

    let signedData = web3.eth.sign(account, seedData);

    let r = signedData.substr(0, 66);
    let s = '0x' + signedData.substr(66, 64);
    let v = parseInt(signedData.substr(130)) + 27;

    return { hash: seedData, v: v, r: r, s: s };
}

This is the version with ethers.js (slightly complicated due to having to prefix the ethereum signed message string to the hash):

function generateEthersSignedData(msg, signingKey) {
  var prefix = "\x19Ethereum Signed Message:\n32";
  var msgBytes = ethers.utils.toUtf8Bytes(msg);
  var msgDigest = ethers.utils.keccak256(msgBytes);

  var final = ethers.utils.solidityKeccak256(['string', 'bytes32'], [prefix, msgDigest]);

  var signature = signingKey.signDigest(final);

  return { hash: msgDigest, v: signature.recoveryParam + 27, r: signature.r, s: signature.s };
}

As an example: msg = '0x7f23b5eed5bc7e89f267f339561b2697faab234a2' test private key used: 0x09a11afa58d6014843fd2c5fd4e21e7fadf96ca2d8ce9934af6b8e204314f25c

web3 result:

signed data: {
    "hash": "0x69aff0e8e6bad68d84a1edb4175a4396d5a78d4d8b9d17e08034e2785bc8d2d7",
    "v": 28,
    "r": "0x7222038446034a0425b6e3f0cc3594f0d979c656206408f937c37a8180bb1bea",
    "s": "0x47d061e4ded4aeac77fa86eb02d42ba7250964ac3eb9da1337090258ce79849"
}

ethers.js result:

ethers.js signed data: {
    "hash": "0x69aff0e8e6bad68d84a1edb4175a4396d5a78d4d8b9d17e08034e2785bc8d2d7",
    "v": 28,
    "r": "0x7222038446034a0425b6e3f0cc3594f0d979c656206408f937c37a8180bb1bea",
    "s": "0x047d061e4ded4aeac77fa86eb02d42ba7250964ac3eb9da1337090258ce79849"
}

Note the extra 0 prefixed for s with the ethers.js result. Testing these results via ecrecover in a contract that I’ve wrote I was able to successfully extract the owning address from the web3 version but not ethers.js (due to the extra 0).

I’ve encountered extraneous 0 being added to r, s with certain values of msg. I can provide more sample data if you’d like.

[starlabs007]

I’ve finally had some extended time to investigate and test this further. So it turns out the issue was on my side as the contract had a modifier clause that caused the ethers.js signing version of my test to fail - something I totally overlooked.

SIgning works 100% now. Thanks for a great ethereum library and keep up the good work!

[ricmoo]

For another similar bug in another issue we resolved it down to being an issue with how the v was being adjusted.

Can you include the JavaScript you are actually using to pass the signature into the contract? Are you suing the ethers.Contract object?