It would be nice to have tests for behavior of SHL, SHR and SAR when the shift is something between 2^64 and 2^256. This would have caught at least one real bug.