Option to not auto-create index during expand_dims

xarray/core/dataset.py

@@ -4497,6 +4497,7 @@ def expand_dims(
         self,
         dim: None | Hashable | Sequence[Hashable] | Mapping[Any, Any] = None,
         axis: None | int | Sequence[int] = None,
+        create_1d_index: bool = True,
         **dim_kwargs: Any,
     ) -> Self:
         """Return a new object with an additional axis (or axes) inserted at
@@ -4506,6 +4507,9 @@ def expand_dims(
         If dim is already a scalar coordinate, it will be promoted to a 1D
         coordinate consisting of a single value.
 
+        The automatic creation of indexes to back new 1D coordinate variables
+        controlled by the create_1d_index kwarg.
+
         Parameters
         ----------
         dim : hashable, sequence of hashable, mapping, or None
@@ -4521,6 +4525,8 @@ def expand_dims(
             multiple axes are inserted. In this case, dim arguments should be
             same length list. If axis=None is passed, all the axes will be
             inserted to the start of the result array.
+        create_1d_index : bool, default is True
+            Whether to create new PandasIndex objects for any new 1D coordinate variables.
         **dim_kwargs : int or sequence or ndarray
             The keywords are arbitrary dimensions being inserted and the values
             are either the lengths of the new dims (if int is given), or their
@@ -4640,6 +4646,8 @@ def expand_dims(
                 # save the coordinates to the variables dict, and set the
                 # value within the dim dict to the length of the iterable
                 # for later use.
+
+                # TODO should we have an option to not create a variable here?
                 index = PandasIndex(v, k)
                 indexes[k] = index
                 variables.update(index.create_variables())
@@ -4678,11 +4686,16 @@ def expand_dims(
                     variables[k] = v.set_dims(dict(all_dims))
             else:
                 if k not in variables:
-                    # If dims includes a label of a non-dimension coordinate,
-                    # it will be promoted to a 1D coordinate with a single value.
-                    index, index_vars = create_default_index_implicit(v.set_dims(k))
-                    indexes[k] = index
-                    variables.update(index_vars)
+                    if k in coord_names and create_1d_index:
+                        # If dims includes a label of a non-dimension coordinate,
+                        # it will be promoted to a 1D coordinate with a single value.
+                        index, index_vars = create_default_index_implicit(v.set_dims(k))
+                        indexes[k] = index
+                        variables.update(index_vars)
+                    else:
+                        # create 1D variable without creating a new index
+                        new_1d_var = v.set_dims(k)
+                        variables.update({k: new_1d_var})
 
         return self._replace_with_new_dims(
             variables, coord_names=coord_names, indexes=indexes

xarray/tests/test_dataset.py

@@ -3431,6 +3431,36 @@ def test_expand_dims_kwargs_python36plus(self) -> None:
         )
         assert_identical(other_way_expected, other_way)
 
+    @pytest.mark.parametrize("create_1d_index_flag", [True, False])
+    def test_expand_dims_create_index_data_variable(self, create_1d_index_flag):
+        # data variables should not gain an index ever
+        ds = Dataset({"x": 0})
+        expanded = ds.expand_dims("x", create_1d_index=create_1d_index_flag)
+
+        # TODO I can't just create the expected dataset directly using constructor because of GH issue 8959
+        # expected = Dataset(data_vars={"x": ("x", [0])})
+        expected = Dataset({"x": ("x", [0])}).drop_indexes("x").reset_coords("x")
+
+        # TODO also can't just assert equivalence because it will fail internal invariants default indexes checks
+        # assert_identical(expanded, expected)
+        assert expected.data_vars == {"x": Variable(data=[0], dims=["x"])}
+        assert expanded.indexes == {}
+
+    def test_expand_dims_create_index_coordinate_variable(self):
+        # coordinate variables should gain an index only if create_1d_index is True (the default)
+        ds = Dataset(coords={"x": 0})
+        expanded = ds.expand_dims("x")
+        expected = Dataset({"x": ("x", [0])})
+        assert_identical(expanded, expected)
+
+        expanded_no_index = ds.expand_dims("x", create_1d_index=False)
+        expected = Dataset(coords={"x": ("x", [0])}).drop_indexes("x")
+
+        # TODO also can't just assert equivalence because it will fail internal invariants default indexes checks
+        # assert_identical(expanded, expected)
+        assert expanded_no_index.coords == {"x": Variable(data=[0], dims=["x"])}
+        assert expanded_no_index.indexes == {}
+
     @requires_pandas_version_two
     def test_expand_dims_non_nanosecond_conversion(self) -> None:
         # Regression test for https://github.com/pydata/xarray/issues/7493#issuecomment-1953091000