In rolling().apply() of pandas, are args cached when enigne="numba"

In rolling().apply() of pandas, Args are cached when enigne="numba".
Is this correct behavior?
Is there any way to prevent args from being cached?

The sample code is as follows

import pandas as pd
import numba as nb

@nb.jit
def test_func(x, c):
    return c

@nb.jit
def test_func2(x, c):
    return c

s = pd.Series(range(5))
print("s")
print(s)

rolling_s1 = s.rolling(window=3).apply(test_func, raw=True, engine="numba", args=(100,))
print("rolling_s1")
display(rolling_s1)

rolling_s2 = s.rolling(window=3).apply(test_func, raw=True, engine="numba", args=(-100,))
print("rolling_s2")
display(rolling_s2)

rolling_s3 = s.rolling(window=3).apply(test_func2, raw=True, engine="numba", args=(-100,))
print("rolling_s3")
display(rolling_s3)

print("test_func(None, 200)")
print(test_func(None, 200))

print("test_func2(None, -200)")
print(test_func2(None, -200))

The results of the execution are as follows

s
0    0
1    1
2    2
3    3
4    4
dtype: int64
rolling_s1
0      NaN
1      NaN
2    100.0
3    100.0
4    100.0
dtype: float64
rolling_s2
0      NaN
1      NaN
2    100.0
3    100.0
4    100.0
dtype: float64
rolling_s3
0      NaN
1      NaN
2   -100.0
3   -100.0
4   -100.0
dtype: float64
test_func(None, 200)
200
test_func2(None, -200)
-200

The value of rollig_s2 should be -100, but it is 100.