Fix multiindex level serialization after reset_index

doc/whats-new.rst

@@ -40,6 +40,9 @@ Deprecations
 Bug fixes
 ~~~~~~~~~
 
+- Fixed a regression that prevented multi-index level coordinates being
+  serialized after resetting or dropping the multi-index (:issue:`8628`, :pull:`8672`).
+  By `Benoit Bovy <https://github.com/benbovy>`_.
 - Fix bug with broadcasting when wrapping array API-compliant classes. (:issue:`8665`, :pull:`8669`)
   By `Tom Nicholas <https://github.com/TomNicholas>`_.
 - Ensure :py:meth:`DataArray.unstack` works when wrapping array API-compliant classes. (:issue:`8666`, :pull:`8668`)

xarray/conventions.py

@@ -16,7 +16,7 @@
 )
 from xarray.core.pycompat import is_duck_dask_array
 from xarray.core.utils import emit_user_level_warning
-from xarray.core.variable import Variable
+from xarray.core.variable import IndexVariable, Variable
 
 CF_RELATED_DATA = (
     "bounds",
@@ -84,13 +84,17 @@ def _infer_dtype(array, name=None):
 
 
 def ensure_not_multiindex(var: Variable, name: T_Name = None) -> None:
+    # only the pandas multi-index dimension coordinate cannot be serialized (tuple values)
     if isinstance(var._data, indexing.PandasMultiIndexingAdapter):
-        raise NotImplementedError(
-            f"variable {name!r} is a MultiIndex, which cannot yet be "
-            "serialized. Instead, either use reset_index() "
-            "to convert MultiIndex levels into coordinate variables instead "
-            "or use https://cf-xarray.readthedocs.io/en/latest/coding.html."
-        )
+        if name is None and isinstance(var, IndexVariable):
+            name = var.name
+        if var.dims == (name,):
+            raise NotImplementedError(
+                f"variable {name!r} is a MultiIndex, which cannot yet be "
+                "serialized. Instead, either use reset_index() "
+                "to convert MultiIndex levels into coordinate variables instead "
+                "or use https://cf-xarray.readthedocs.io/en/latest/coding.html."
+            )
 
 
 def _copy_with_dtype(data, dtype: np.typing.DTypeLike):

xarray/tests/test_backends.py

@@ -1245,6 +1245,11 @@
             with self.roundtrip(ds):
                 pass
 
+        # regression GH8628 (can serialize reset multi-index level coordinates)
+        ds_reset = ds.reset_index("x")
+        with self.roundtrip(ds_reset) as actual:
+            assert_identical(actual, ds_reset)
+
 
 class NetCDFBase(CFEncodedBase):
     """Tests for all netCDF3 and netCDF4 backends."""
@@ -4467,14 +4472,14 @@
             with open_dataset(tmp, chunks=chunks) as dask_ds:
                 assert_identical(data, dask_ds)
                 with create_tmp_file() as tmp2:
                     dask_ds.to_netcdf(tmp2)
                     with open_dataset(tmp2) as on_disk:
                         assert_identical(data, on_disk)
 
     def test_deterministic_names(self) -> None:
         with create_tmp_file() as tmp:
             data = create_test_data()
             data.to_netcdf(tmp)
             with open_mfdataset(tmp, combine="by_coords") as ds:
                 original_names = {k: v.data.name for k, v in ds.data_vars.items()}
             with open_mfdataset(tmp, combine="by_coords") as ds:
@@ -4492,7 +4497,7 @@
         computed = actual.compute()
         assert not actual._in_memory
         assert computed._in_memory
         assert_allclose(actual, computed, decode_bytes=False)
 
     def test_save_mfdataset_compute_false_roundtrip(self) -> None:
         from dask.delayed import Delayed
@@ -4505,7 +4510,7 @@
                     datasets, [tmp1, tmp2], engine=self.engine, compute=False
                 )
                 assert isinstance(delayed_obj, Delayed)
                 delayed_obj.compute()
                 with open_mfdataset(
                     [tmp1, tmp2], combine="nested", concat_dim="x"
                 ) as actual:
@@ -4514,7 +4519,7 @@
     def test_load_dataset(self) -> None:
         with create_tmp_file() as tmp:
             original = Dataset({"foo": ("x", np.random.randn(10))})
             original.to_netcdf(tmp)
             ds = load_dataset(tmp)
             # this would fail if we used open_dataset instead of load_dataset
             ds.to_netcdf(tmp)
@@ -4522,7 +4527,7 @@
     def test_load_dataarray(self) -> None:
         with create_tmp_file() as tmp:
             original = Dataset({"foo": ("x", np.random.randn(10))})
             original.to_netcdf(tmp)
             ds = load_dataarray(tmp)
             # this would fail if we used open_dataarray instead of
             # load_dataarray
@@ -4535,7 +4540,7 @@
     def test_inline_array(self) -> None:
         with create_tmp_file() as tmp:
             original = Dataset({"foo": ("x", np.random.randn(10))})
             original.to_netcdf(tmp)
             chunks = {"time": 10}
 
             def num_graph_nodes(obj):
@@ -4588,7 +4593,7 @@
             yield actual, expected
 
     def test_cmp_local_file(self) -> None:
         with self.create_datasets() as (actual, expected):
             assert_equal(actual, expected)
 
             # global attributes should be global attributes on the dataset
@@ -4622,7 +4627,7 @@
 
     def test_compatible_to_netcdf(self) -> None:
         # make sure it can be saved as a netcdf
         with self.create_datasets() as (actual, expected):
             with create_tmp_file() as tmp_file:
                 actual.to_netcdf(tmp_file)
                 with open_dataset(tmp_file) as actual2:
@@ -4631,7 +4636,7 @@
 
     @requires_dask
     def test_dask(self) -> None:
         with self.create_datasets(chunks={"j": 2}) as (actual, expected):
             assert_equal(actual, expected)