Update base branch (#1)

* Fix issue pandas-dev#47101

Bisecting two years ago ( pandas-dev#47101 (comment) ) shows this regression was introduced in b2d54d9 in 2021. Somehow this hasn't been patched since then.

PR pandas-dev#48313 was supposed to address this, but the PR was closed and never merged and the bug has persisted.

* Add a test as per PR guidelines

* Fix typo

* Resolve inconsistent namespace as per PR test

https://results.pre-commit.ci/run/github/858127/1723498369.6V12SWx7T-WpLZDAXXkz0Q

This web UI commit will still fail, as the E501 line-too-long check will fail until the next commit

* Resolve E501 linting errors

https://results.pre-commit.ci/run/github/858127/1723498369.6V12SWx7T-WpLZDAXXkz0Q

* Fix test TypeErrors

np.equal([1,2,3], "") fails

* Quote style for Ruff

* typing and remove code backtick possibly incorrectly triggering ruff formatter

* mpy supression for caught error

* trailing space

pandas/core/missing.py

@@ -124,11 +124,25 @@ def mask_missing(arr: ArrayLike, values_to_mask) -> npt.NDArray[np.bool_]:
                     new_mask = np.zeros(arr.shape, dtype=np.bool_)
                     new_mask[arr_mask] = arr[arr_mask] == x
                 else:
-                    new_mask = arr == x
+                    # GH#47101
+                    # Fix where type bool has no attribute to_numpy() by first
+                    # attempting to broadcast with np.equal for some cases, and then
+                    # an explicit type check when checking the mask for any straggling
+                    # cases. Where a literal comparison would fail np.equal we fall back
+                    # to the original equality check.
+                    try:
+                        # In case of an uncastable type, this will emit TypeError
+                        new_mask = np.equal(arr, x) # type: ignore[arg-type]
+                    except TypeError:
+                        # Old behaviour for uncastable types
+                        new_mask = arr == x
 
                     if not isinstance(new_mask, np.ndarray):
                         # usually BooleanArray
-                        new_mask = new_mask.to_numpy(dtype=bool, na_value=False)
+                        if isinstance(new_mask, bool):
+                            new_mask = np.array([new_mask], dtype= bool)
+                        else:
+                            new_mask = new_mask.to_numpy(dtype=bool, na_value=False)
                 mask |= new_mask
 
     if na_mask.any():

pandas/tests/frame/methods/test_replace.py

@@ -1518,3 +1518,10 @@ def test_replace_object_splitting(self, using_infer_string):
             assert len(df._mgr.blocks) == 2
         else:
             assert len(df._mgr.blocks) == 1
+
+    def test_replace_bool_to_numpy_attributeerror(self):
+        # GH#47101
+        pass_pre_patch = DataFrame({"d":[None]})
+        tm.assert_frame_equal(pass_pre_patch, pass_pre_patch.replace("", pd.NA))
+        fail_pre_patch = DataFrame({"d":[pd.NA]})
+        tm.assert_frame_equal(fail_pre_patch, fail_pre_patch.replace("", pd.NA))