[Python] How can I avoid getting the wrong answer when calculating with njit in Python?

I use a njit library from numba to improve the speed of my code in Python. I got a wrong answer. At the same time, if I do not use njit, my code gives a correct answer. Here are two examples

from numba import njit


@njit
def main_with_njit():
a = 94906267
print(f'main_with_njit: a**2 = {a ** 2}') # -> a**2 = 9007199515875288
# the type is int64


main_with_njit()


def main_without_njit():
a = 94906267
print(f'main_without_njit: a**2 = {a ** 2}') # -> a**2 = 9007199515875289
# the type is <class 'int'>


main_without_njit()


How can I use njit without getting a wrong answer?


I use Python 3.11.3 with PyCharm 2023.1.2 (Community Edition) on Windows, 64 bit, 0.61.0 version of numba

I can reproduce the problem originally described by Roshan N using the math library.

import math


print(math.pow(94906267, 2)) # -> 9007199515875288.0

Continue reading...