Cross-platform test failures due to `np.int_` type differences in numba-jitted functions

[damar-wicaksono]

Several tests fail on certain platforms (e.g., Windows) due to type mismatches between test data and numba-jitted functions. Numba's njit requires explicit type declarations, which in this context is np.int_. However, the actual integer type varies across platforms and NumPy versions. When generating test data with np.random.randint(), the default dtype is also platform-dependent, which can create mismatches with the expected np.int_ type.

Affected tests

The affected tests are related to the following functions:

• get_max_columnwise()
• is_lex_smaller_or_equal()
• is_lex_sorted()

Example

def test_get_max_columnwise():
    """Test getting the column-wise max of a two-dimensional integer array."""
    num_rows = np.random.randint(low=100, high=1000)
    num_cols = np.random.randint(low=1, high=10)
    xx = np.random.randint(low=0, high=100, size=(num_rows, num_cols))
    # xx.dtype differs between Linux (int64) and Windows (int32)

    max_ref = np.max(xx, axis=0)
    max_iter = get_max_columnwise(xx)  # Type mismatch on some platforms
    
    assert np.array_equal(max_ref, max_iter)

Proposed Solution

Explicitly specify dtype=INT_DTYPE when generating test data to match the numba function signatures.

Notes

This is strictly a testing issue. The functions themselves work correctly when called with appropriately typed arrays. The fix ensures test data matches the function signatures across all platforms.

[damar-wicaksono]

This Issue has been resolved by PR #60.