CLN: ensure we pass correct type to DTI/TDI shallow_copy

pandas/core/algorithms.py

@@ -696,6 +696,8 @@ def factorize(
 
     # return original tenor
     if isinstance(original, ABCIndexClass):
+        if original.dtype.kind in ["m", "M"] and isinstance(uniques, np.ndarray):
+            uniques = type(original._data)._simple_new(uniques, dtype=original.dtype)
         uniques = original._shallow_copy(uniques, name=None)
     elif isinstance(original, ABCSeries):
         from pandas import Index
@@ -1650,7 +1652,8 @@ def take_nd(
     """
     mask_info = None
 
-    if is_extension_array_dtype(arr):
+    if isinstance(arr, ABCExtensionArray):
+        # Check for EA to catch DatetimeArray, TimedeltaArray
         return arr.take(indexer, fill_value=fill_value, allow_fill=allow_fill)
 
     arr = extract_array(arr)
@@ -2043,7 +2046,7 @@ def safe_sort(
             "Only list-like objects are allowed to be passed to safe_sort as values"
         )
 
-    if not isinstance(values, np.ndarray) and not is_extension_array_dtype(values):
+    if not isinstance(values, (np.ndarray, ABCExtensionArray)):
         # don't convert to string types
         dtype, _ = infer_dtype_from_array(values)
         values = np.asarray(values, dtype=dtype)

pandas/core/indexes/base.py

@@ -2664,6 +2664,11 @@ def _union(self, other, sort):
         return self._shallow_copy(result)
 
     def _wrap_setop_result(self, other, result):
+        if isinstance(self, (ABCDatetimeIndex, ABCTimedeltaIndex)) and isinstance(
+            result, np.ndarray
+        ):
+            result = type(self._data)._simple_new(result, dtype=self.dtype)
+
         name = get_op_result_name(self, other)
         if isinstance(result, Index):
             if result.name != name:
@@ -2740,10 +2745,10 @@ def intersection(self, other, sort=False):
             indexer = algos.unique1d(Index(rvals).get_indexer_non_unique(lvals)[0])
             indexer = indexer[indexer != -1]
 
-        result = other.take(indexer)
+        result = other.take(indexer)._values
 
         if sort is None:
-            result = algos.safe_sort(result.values)
+            result = algos.safe_sort(result)
 
         return self._wrap_setop_result(other, result)
 
@@ -2800,7 +2805,7 @@ def difference(self, other, sort=None):
         indexer = indexer.take((indexer != -1).nonzero()[0])
 
         label_diff = np.setdiff1d(np.arange(this.size), indexer, assume_unique=True)
-        the_diff = this.values.take(label_diff)
+        the_diff = this._values.take(label_diff)
         if sort is None:
             try:
                 the_diff = algos.safe_sort(the_diff)

pandas/core/indexes/datetimelike.py

@@ -696,9 +696,6 @@ def _shallow_copy(self, values=None, name: Label = lib.no_default):
         name = self.name if name is lib.no_default else name
 
         if values is not None:
-            # TODO: We would rather not get here
-            if isinstance(values, np.ndarray):
-                values = type(self._data)(values, dtype=self.dtype)
             return self._simple_new(values, name=name)
 
         result = self._simple_new(self._data, name=name)

pandas/tests/indexes/test_common.py

@@ -320,6 +320,10 @@ def test_get_unique_index(self, index):
             vals[0] = np.nan
 
         vals_unique = vals[:2]
+        if index.dtype.kind in ["m", "M"]:
+            # i.e. needs_i8_conversion but not period_dtype, as above
+            vals = type(index._data)._simple_new(vals, dtype=index.dtype)
+            vals_unique = type(index._data)._simple_new(vals_unique, dtype=index.dtype)
         idx_nan = index._shallow_copy(vals)
         idx_unique_nan = index._shallow_copy(vals_unique)
         assert idx_unique_nan.is_unique is True