BUG: DTA/TDA/PA setitem incorrectly allowing i8 (pandas-dev#33717)

Author: jbrockmendel

pandas/core/arrays/datetimelike.py

@@ -740,7 +740,7 @@ def _validate_fill_value(self, fill_value):
         return fill_value
 
     def _validate_shift_value(self, fill_value):
-        # TODO(2.0): once this deprecation is enforced, used _validate_fill_value
+        # TODO(2.0): once this deprecation is enforced, use _validate_fill_value
         if is_valid_nat_for_dtype(fill_value, self.dtype):
             fill_value = NaT
         elif isinstance(fill_value, self._recognized_scalars):
@@ -782,6 +782,9 @@ def _validate_searchsorted_value(self, value):
         elif isinstance(value, self._recognized_scalars):
             value = self._scalar_type(value)
 
+        elif isinstance(value, type(self)):
+            pass
+
         elif is_list_like(value) and not isinstance(value, type(self)):
             value = array(value)
 
@@ -791,7 +794,7 @@ def _validate_searchsorted_value(self, value):
                     f"not {type(value).__name__}"
                 )
 
-        if not (isinstance(value, (self._scalar_type, type(self))) or (value is NaT)):
+        else:
             raise TypeError(f"Unexpected type for 'value': {type(value)}")
 
         if isinstance(value, type(self)):
@@ -803,25 +806,41 @@ def _validate_searchsorted_value(self, value):
         return value
 
     def _validate_setitem_value(self, value):
-        if lib.is_scalar(value) and not isna(value):
-            value = com.maybe_box_datetimelike(value)
 
         if is_list_like(value):
-            value = type(self)._from_sequence(value, dtype=self.dtype)
-            self._check_compatible_with(value, setitem=True)
-            value = value.asi8
-        elif isinstance(value, self._scalar_type):
-            self._check_compatible_with(value, setitem=True)
-            value = self._unbox_scalar(value)
+            value = array(value)
+            if is_dtype_equal(value.dtype, "string"):
+                # We got a StringArray
+                try:
+                    # TODO: Could use from_sequence_of_strings if implemented
+                    # Note: passing dtype is necessary for PeriodArray tests
+                    value = type(self)._from_sequence(value, dtype=self.dtype)
+                except ValueError:
+                    pass
+
+            if not type(self)._is_recognized_dtype(value):
+                raise TypeError(
+                    "setitem requires compatible dtype or scalar, "
+                    f"not {type(value).__name__}"
+                )
+
+        elif isinstance(value, self._recognized_scalars):
+            value = self._scalar_type(value)
         elif is_valid_nat_for_dtype(value, self.dtype):
-            value = iNaT
+            value = NaT
         else:
             msg = (
                 f"'value' should be a '{self._scalar_type.__name__}', 'NaT', "
                 f"or array of those. Got '{type(value).__name__}' instead."
             )
             raise TypeError(msg)
 
+        self._check_compatible_with(value, setitem=True)
+        if isinstance(value, type(self)):
+            value = value.asi8
+        else:
+            value = self._unbox_scalar(value)
+
         return value
 
     def _validate_insert_value(self, value):

pandas/tests/arrays/test_datetimelike.py

@@ -241,6 +241,16 @@ def test_setitem(self):
         expected[:2] = expected[-2:]
         tm.assert_numpy_array_equal(arr.asi8, expected)
 
+    def test_setitem_str_array(self, arr1d):
+        if isinstance(arr1d, DatetimeArray) and arr1d.tz is not None:
+            pytest.xfail(reason="timezone comparisons inconsistent")
+        expected = arr1d.copy()
+        expected[[0, 1]] = arr1d[-2:]
+
+        arr1d[:2] = [str(x) for x in arr1d[-2:]]
+
+        tm.assert_equal(arr1d, expected)
+
     def test_setitem_raises(self):
         data = np.arange(10, dtype="i8") * 24 * 3600 * 10 ** 9
         arr = self.array_cls(data, freq="D")
@@ -252,6 +262,17 @@ def test_setitem_raises(self):
         with pytest.raises(TypeError, match="'value' should be a.* 'object'"):
             arr[0] = object()
 
+    @pytest.mark.parametrize("box", [list, np.array, pd.Index, pd.Series])
+    def test_setitem_numeric_raises(self, arr1d, box):
+        # We dont case e.g. int64 to our own dtype for setitem
+
+        msg = "requires compatible dtype"
+        with pytest.raises(TypeError, match=msg):
+            arr1d[:2] = box([0, 1])
+
+        with pytest.raises(TypeError, match=msg):
+            arr1d[:2] = box([0.0, 1.0])
+
     def test_inplace_arithmetic(self):
         # GH#24115 check that iadd and isub are actually in-place
         data = np.arange(10, dtype="i8") * 24 * 3600 * 10 ** 9