BUG: Stop item_from_zerodim from converting subclasses of np.ndarray to float (#62999)

Co-authored-by: Julian Harbeck <[email protected]>

Author: Julian-Harbeck

pandas/_libs/lib.pyx

@@ -322,7 +322,7 @@ def item_from_zerodim(val: object) -> object:
     >>> item_from_zerodim(np.array([1]))
     array([1])
     """
-    if cnp.PyArray_IsZeroDim(val):
+    if cnp.PyArray_IsZeroDim(val) and cnp.PyArray_CheckExact(val):
         return cnp.PyArray_ToScalar(cnp.PyArray_DATA(val), val)
     return val
 

pandas/tests/libs/test_lib.py

@@ -307,3 +307,52 @@ def test_ensure_string_array_list_of_lists():
     # Each item in result should still be a list, not a stringified version
     expected = np.array(["['t', 'e', 's', 't']", "['w', 'o', 'r', 'd']"], dtype=object)
     tm.assert_numpy_array_equal(result, expected)
+
+
+def test_item_from_zerodim_for_subclasses():
+    # GH#62981 Ensure item_from_zerodim preserves subclasses of ndarray
+    # Define a custom ndarray subclass
+    class TestArray(np.ndarray):
+        def __new__(cls, input_array):
+            return np.asarray(input_array).view(cls)
+
+        def __array_finalize__(self, obj) -> None:
+            self._is_test_array = True
+
+    # Define test data
+    val_0_dim = 1
+    val_1_dim = [1, 2, 3]
+
+    # 0-dim and 1-dim numpy arrays
+    arr_0_dim = np.array(val_0_dim)
+    arr_1_dim = np.array(val_1_dim)
+
+    # 0-dim and 1-dim TestArray arrays
+    test_arr_0_dim = TestArray(val_0_dim)
+    test_arr_1_dim = TestArray(val_1_dim)
+
+    # Check that behavior did not change for regular numpy arrays
+    # Test with regular numpy 0-dim array
+    result = lib.item_from_zerodim(arr_0_dim)
+    expected = val_0_dim
+    assert result == expected
+    assert np.isscalar(result)
+
+    # Test with regular numpy 1-dim array
+    result = lib.item_from_zerodim(arr_1_dim)
+    expected = arr_1_dim
+    tm.assert_numpy_array_equal(result, expected)
+    assert isinstance(result, np.ndarray)
+
+    # Check that behaviour for subclasses now is as expected
+    # Test with TestArray 0-dim array
+    result = lib.item_from_zerodim(test_arr_0_dim)
+    expected = test_arr_0_dim
+    assert result == expected
+    assert isinstance(result, TestArray)
+
+    # Test with TestArray 1-dim array
+    result = lib.item_from_zerodim(test_arr_1_dim)
+    expected = test_arr_1_dim
+    assert np.all(result == expected)
+    assert isinstance(result, TestArray)