Cache does not work when passing numpy arrays to objmode inside overload

Reporting a bug

• I have tried using the latest released version of Numba (most recent is visible in the change log (https://github.com/numba/numba/blob/master/CHANGE_LOG).
• I have included below a minimal working reproducer (if you are unsure how to write one see http://matthewrocklin.com/blog/work/2018/02/28/minimal-bug-reports).

Hello, I found this bug with cache=True when using overloads. Replicator is below:

import numpy as np
import numba
from numba.extending import overload


def print_this(a, b):
    print(a)
    print(b)


def do_something(a, b):
    print(a)
    print(b)


@overload(do_something)
def overload_do_something(a, b):
    def _do_something_impl(a, b):
        with numba.objmode:
            print_this(a, b)

    return _do_something_impl


@numba.njit(cache=True)
def test_caching():
    a = np.arange(20)
    b = np.arange(50)
    #a = 1  # works if pass this instead of numpy array
    #b = 3
    do_something(a, b)


test_caching()

Thanks!

[stuartarchibald]

Holding a reference to env https://github.com/numba/numba/blob/ebbae83720e6c29ef6fbb756298211d24869279f/numba/core/environment.py#L57 stops the segfault.

[sklam]

Thanks @juanjgalvez for debugging it. A null pointer in Python C-API can be caused by a previous C-API failed and raised an exception. The code may have forgotten to check for the error.