Agreed to this. We went that direction since it was already used as a pandas backend at the time.