Avoid auto creation of indexes in concat

doc/whats-new.rst

@@ -32,7 +32,10 @@ New Features
 - :py:func:`testing.assert_allclose`/:py:func:`testing.assert_equal` now accept a new argument `check_dims="transpose"`, controlling whether a transposed array is considered equal. (:issue:`5733`, :pull:`8991`)
   By `Ignacio Martinez Vazquez <https://github.com/ignamv>`_.
 - Added the option to avoid automatically creating 1D pandas indexes in :py:meth:`Dataset.expand_dims()`, by passing the new kwarg
-  `create_index=False`. (:pull:`8960`)
+  `create_index_for_new_dim=False`. (:pull:`8960`)
+  By `Tom Nicholas <https://github.com/TomNicholas>`_.
+- Avoid automatically re-creating 1D pandas indexes in :py:func:`concat()`. Also added option to avoid creating 1D indexes for
+  new dimension coordinates by passing the new kwarg `create_index_for_new_dim=False`. (:issue:`8871`, :pull:`8872`)
   By `Tom Nicholas <https://github.com/TomNicholas>`_.
 
 Breaking changes

xarray/core/concat.py

@@ -8,6 +8,7 @@
 
 from xarray.core import dtypes, utils
 from xarray.core.alignment import align, reindex_variables
+from xarray.core.coordinates import Coordinates
 from xarray.core.duck_array_ops import lazy_array_equiv
 from xarray.core.indexes import Index, PandasIndex
 from xarray.core.merge import (
@@ -42,6 +43,7 @@ def concat(
     fill_value: object = dtypes.NA,
     join: JoinOptions = "outer",
     combine_attrs: CombineAttrsOptions = "override",
+    create_index_for_new_dim: bool = True,
 ) -> T_Dataset: ...
 
 
@@ -56,6 +58,7 @@ def concat(
     fill_value: object = dtypes.NA,
     join: JoinOptions = "outer",
     combine_attrs: CombineAttrsOptions = "override",
+    create_index_for_new_dim: bool = True,
 ) -> T_DataArray: ...
 
 
@@ -69,6 +72,7 @@ def concat(
     fill_value=dtypes.NA,
     join: JoinOptions = "outer",
     combine_attrs: CombineAttrsOptions = "override",
+    create_index_for_new_dim: bool = True,
 ):
     """Concatenate xarray objects along a new or existing dimension.
 
@@ -162,6 +166,8 @@ def concat(
 
         If a callable, it must expect a sequence of ``attrs`` dicts and a context object
         as its only parameters.
+    create_index_for_new_dim : bool, default: True
+        Whether to create a new ``PandasIndex`` object when the objects being concatenated contain scalar variables named ``dim``.
 
     Returns
     -------
@@ -217,6 +223,25 @@ def concat(
         x        (new_dim) <U1 8B 'a' 'b'
       * y        (y) int64 24B 10 20 30
       * new_dim  (new_dim) int64 16B -90 -100
+
+    # Concatenate a scalar variable along a new dimension of the same name with and without creating a new index
+
+    >>> ds = xr.Dataset(coords={"x": 0})
+    >>> xr.concat([ds, ds], dim="x")
+    <xarray.Dataset> Size: 16B
+    Dimensions:  (x: 2)
+    Coordinates:
+    * x        (x) int64 16B 0 0
+    Data variables:
+        *empty*
+
+    >>> xr.concat([ds, ds], dim="x").indexes
+    Indexes:
+        x        Index([0, 0], dtype='int64', name='x')
+
+    >>> xr.concat([ds, ds], dim="x", create_index_for_new_dim=False).indexes
+    Indexes:
+        *empty*
     """
     # TODO: add ignore_index arguments copied from pandas.concat
     # TODO: support concatenating scalar coordinates even if the concatenated
@@ -245,6 +270,7 @@ def concat(
             fill_value=fill_value,
             join=join,
             combine_attrs=combine_attrs,
+            create_index_for_new_dim=create_index_for_new_dim,
         )
     elif isinstance(first_obj, Dataset):
         return _dataset_concat(
@@ -257,6 +283,7 @@ def concat(
             fill_value=fill_value,
             join=join,
             combine_attrs=combine_attrs,
+            create_index_for_new_dim=create_index_for_new_dim,
         )
     else:
         raise TypeError(
@@ -439,7 +466,7 @@ def _parse_datasets(
             if dim in dims:
                 continue
 
-            if dim not in dim_coords:
+            if dim in ds.coords and dim not in dim_coords:
                 dim_coords[dim] = ds.coords[dim].variable
         dims = dims | set(ds.dims)
 
@@ -456,6 +483,7 @@ def _dataset_concat(
     fill_value: Any = dtypes.NA,
     join: JoinOptions = "outer",
     combine_attrs: CombineAttrsOptions = "override",
+    create_index_for_new_dim: bool = True,
 ) -> T_Dataset:
     """
     Concatenate a sequence of datasets along a new or existing dimension
@@ -489,7 +517,6 @@ def _dataset_concat(
         datasets
     )
     dim_names = set(dim_coords)
-    unlabeled_dims = dim_names - coord_names
 
     both_data_and_coords = coord_names & data_names
     if both_data_and_coords:
@@ -502,15 +529,18 @@ def _dataset_concat(
 
     # case where concat dimension is a coordinate or data_var but not a dimension
     if (dim in coord_names or dim in data_names) and dim not in dim_names:
-        datasets = [ds.expand_dims(dim) for ds in datasets]
+        datasets = [
+            ds.expand_dims(dim, create_index_for_new_dim=create_index_for_new_dim)
+            for ds in datasets
+        ]
 
     # determine which variables to concatenate
     concat_over, equals, concat_dim_lengths = _calc_concat_over(
         datasets, dim, dim_names, data_vars, coords, compat
     )
 
     # determine which variables to merge, and then merge them according to compat
-    variables_to_merge = (coord_names | data_names) - concat_over - unlabeled_dims
+    variables_to_merge = (coord_names | data_names) - concat_over
 
     result_vars = {}
     result_indexes = {}
@@ -567,7 +597,8 @@ def get_indexes(name):
                 var = ds._variables[name]
                 if not var.dims:
                     data = var.set_dims(dim).values
-                    yield PandasIndex(data, dim, coord_dtype=var.dtype)
+                    if create_index_for_new_dim:
+                        yield PandasIndex(data, dim, coord_dtype=var.dtype)
 
     # create concatenation index, needed for later reindexing
     file_start_indexes = np.append(0, np.cumsum(concat_dim_lengths))
@@ -646,29 +677,33 @@ def get_indexes(name):
             # preserves original variable order
             result_vars[name] = result_vars.pop(name)
 
-    result = type(datasets[0])(result_vars, attrs=result_attrs)
-
-    absent_coord_names = coord_names - set(result.variables)
+    absent_coord_names = coord_names - set(result_vars)
     if absent_coord_names:
         raise ValueError(
             f"Variables {absent_coord_names!r} are coordinates in some datasets but not others."
         )
-    result = result.set_coords(coord_names)
-    result.encoding = result_encoding
 
-    result = result.drop_vars(unlabeled_dims, errors="ignore")
+    result_data_vars = {}
+    coord_vars = {}
+    for name, result_var in result_vars.items():
+        if name in coord_names:
+            coord_vars[name] = result_var
+        else:
+            result_data_vars[name] = result_var
 
     if index is not None:
-        # add concat index / coordinate last to ensure that its in the final Dataset
         if dim_var is not None:
             index_vars = index.create_variables({dim: dim_var})
         else:
             index_vars = index.create_variables()
-        result[dim] = index_vars[dim]
+
+        coord_vars[dim] = index_vars[dim]
         result_indexes[dim] = index
 
-    # TODO: add indexes at Dataset creation (when it is supported)
-    result = result._overwrite_indexes(result_indexes)
+    coords_obj = Coordinates(coord_vars, indexes=result_indexes)
+
+    result = type(datasets[0])(result_data_vars, coords=coords_obj, attrs=result_attrs)
+    result.encoding = result_encoding
 
     return result
 
@@ -683,6 +718,7 @@ def _dataarray_concat(
     fill_value: object = dtypes.NA,
     join: JoinOptions = "outer",
     combine_attrs: CombineAttrsOptions = "override",
+    create_index_for_new_dim: bool = True,
 ) -> T_DataArray:
     from xarray.core.dataarray import DataArray
 
@@ -719,6 +755,7 @@ def _dataarray_concat(
         fill_value=fill_value,
         join=join,
         combine_attrs=combine_attrs,
+        create_index_for_new_dim=create_index_for_new_dim,
     )
 
     merged_attrs = merge_attrs([da.attrs for da in arrays], combine_attrs)

xarray/core/dataarray.py

@@ -2558,7 +2558,7 @@ def expand_dims(
         self,
         dim: None | Hashable | Sequence[Hashable] | Mapping[Any, Any] = None,
         axis: None | int | Sequence[int] = None,
-        create_index: bool = True,
+        create_index_for_new_dim: bool = True,
         **dim_kwargs: Any,
     ) -> Self:
         """Return a new object with an additional axis (or axes) inserted at
@@ -2569,7 +2569,7 @@ def expand_dims(
         coordinate consisting of a single value.
 
         The automatic creation of indexes to back new 1D coordinate variables
-        controlled by the create_index kwarg.
+        controlled by the create_index_for_new_dim kwarg.
 
         Parameters
         ----------
@@ -2586,8 +2586,8 @@ 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_index : bool, default is True
-            Whether to create new PandasIndex objects for any new 1D coordinate variables.
+        create_index_for_new_dim : bool, default is True
+            Whether to create new ``PandasIndex`` objects when the object being expanded contains scalar variables with names in ``dim``.
         **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
@@ -2651,7 +2651,9 @@ def expand_dims(
             dim = {dim: 1}
 
         dim = either_dict_or_kwargs(dim, dim_kwargs, "expand_dims")
-        ds = self._to_temp_dataset().expand_dims(dim, axis, create_index=create_index)
+        ds = self._to_temp_dataset().expand_dims(
+            dim, axis, create_index_for_new_dim=create_index_for_new_dim
+        )
         return self._from_temp_dataset(ds)
 
     def set_index(

xarray/core/dataset.py

@@ -4513,7 +4513,7 @@ def expand_dims(
         self,
         dim: None | Hashable | Sequence[Hashable] | Mapping[Any, Any] = None,
         axis: None | int | Sequence[int] = None,
-        create_index: bool = True,
+        create_index_for_new_dim: bool = True,
         **dim_kwargs: Any,
     ) -> Self:
         """Return a new object with an additional axis (or axes) inserted at
@@ -4524,7 +4524,7 @@ def expand_dims(
         coordinate consisting of a single value.
 
         The automatic creation of indexes to back new 1D coordinate variables
-        controlled by the create_index kwarg.
+        controlled by the create_index_for_new_dim kwarg.
 
         Parameters
         ----------
@@ -4541,8 +4541,8 @@ 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_index : bool, default is True
-            Whether to create new PandasIndex objects for any new 1D coordinate variables.
+        create_index_for_new_dim : bool, default is True
+            Whether to create new ``PandasIndex`` objects when the object being expanded contains scalar variables with names in ``dim``.
         **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
@@ -4612,6 +4612,33 @@ def expand_dims(
         Data variables:
             temperature  (y, x, time) float64 96B 0.5488 0.7152 0.6028 ... 0.7917 0.5289
 
+        # Expand a scalar variable along a new dimension of the same name with and without creating a new index
+
+        >>> ds = xr.Dataset(coords={"x": 0})
+        >>> ds
+        <xarray.Dataset> Size: 8B
+        Dimensions:  ()
+        Coordinates:
+            x        int64 8B 0
+        Data variables:
+            *empty*
+
+        >>> ds.expand_dims("x")
+        <xarray.Dataset> Size: 8B
+        Dimensions:  (x: 1)
+        Coordinates:
+        * x        (x) int64 8B 0
+        Data variables:
+            *empty*
+
+        >>> ds.expand_dims("x").indexes
+        Indexes:
+            x        Index([0], dtype='int64', name='x')
+
+        >>> ds.expand_dims("x", create_index_for_new_dim=False).indexes
+        Indexes:
+            *empty*
+
         See Also
         --------
         DataArray.expand_dims
@@ -4663,7 +4690,7 @@ def expand_dims(
                 # value within the dim dict to the length of the iterable
                 # for later use.
 
-                if create_index:
+                if create_index_for_new_dim:
                     index = PandasIndex(v, k)
                     indexes[k] = index
                     name_and_new_1d_var = index.create_variables()
@@ -4705,14 +4732,14 @@ def expand_dims(
                     variables[k] = v.set_dims(dict(all_dims))
             else:
                 if k not in variables:
-                    if k in coord_names and create_index:
+                    if k in coord_names and create_index_for_new_dim:
                         # 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:
-                        if create_index:
+                        if create_index_for_new_dim:
                             warnings.warn(
                                 f"No index created for dimension {k} because variable {k} is not a coordinate. "
                                 f"To create an index for {k}, please first call `.set_coords('{k}')` on this object.",
@@ -5400,7 +5427,7 @@ def to_stacked_array(
                [3, 4, 5, 7]])
         Coordinates:
           * z         (z) object 32B MultiIndex
-          * variable  (z) object 32B 'a' 'a' 'a' 'b'
+          * variable  (z) <U1 16B 'a' 'a' 'a' 'b'
           * y         (z) object 32B 'u' 'v' 'w' nan
         Dimensions without coordinates: x