Impact

All LLVM versions since 2015 fold (xor (shl 1, x), -1) to (rotl ~1, x) if run with optimizations enabled.
Here ~1 is generated as an unsigned 64 bits number (2^64-1). This number is zero-extended to 256 bits on EraVM target while it should have been sign-extended. Thus instead of producing roti 2^256 - 1, x the compiler produces rotl 2^64 - 1, x.

In Solidity, the bug has been confirmed to be reproducible with the following pattern:

    function bug(
        uint256 data,
        uint256 index,
    ) public pure returns(uint256) {
        uint256 bit = 1 << index;
        data &= ~bit;
        return data;
    }

Thus, similar patterns in Vyper must be vulnerable to this bug as well.

Analysis has shown that no contracts were affected by the date of publishing this advisory.

Patches

Fixed in version 1.5.4.

Workarounds

Upgrading and redeploying affected contracts is the only way.