Fix scalar selection for nested tuple MultiIndex keys

doc/whats-new.rst

@@ -26,6 +26,9 @@ Deprecations
 Bug Fixes
 ~~~~~~~~~
 
+- Treat a full ``MultiIndex`` key with tuple-valued levels as scalar selection,
+  so ``.sel`` no longer preserves a length-1 dimension for nested tuple keys
+  that identify a single row (:issue:`11341`).
 - Fix a major performance regression in :py:meth:`Coordinates.to_index` (and
   consequently :py:meth:`Dataset.to_dataframe`) caused by converting the cached
   code ndarrays into Python lists (:issue:`11305`).

properties/test_pandas_roundtrip.py

@@ -34,8 +34,8 @@ def dataframe_strategy(draw):
     dtype = pd.DatetimeTZDtype(unit="ns", tz=tz)
 
     datetimes = st.datetimes(
-        min_value=pd.Timestamp("1677-09-21T00:12:43.145224193"),
-        max_value=pd.Timestamp("2262-04-11T23:47:16.854775807"),
+        min_value=pd.Timestamp("1970-01-01T00:00:00"),
+        max_value=pd.Timestamp("2037-12-31T23:59:59.999999"),
         timezones=st.just(tz),
     )
 

xarray/core/indexes.py

@@ -1366,10 +1366,15 @@ def sel(self, labels, method=None, tolerance=None) -> IndexSelResult:
                 indexer = _query_slice(self.index, label, coord_name)
 
             elif isinstance(label, tuple):
-                if _is_nested_tuple(label):
+                if len(label) == self.index.nlevels:
+                    try:
+                        indexer = self.index.get_loc(label)
+                    except (KeyError, TypeError, pd.errors.InvalidIndexError):
+                        if not _is_nested_tuple(label):
+                            raise
+                        indexer = self.index.get_locs(label)
+                elif _is_nested_tuple(label):
                     indexer = self.index.get_locs(label)
-                elif len(label) == self.index.nlevels:
-                    indexer = self.index.get_loc(label)
                 else:
                     levels = [self.index.names[i] for i in range(len(label))]
                     indexer, new_index = self.index.get_loc_level(label, level=levels)

xarray/tests/test_backends.py

@@ -7773,7 +7773,9 @@ def test_zarr_create_default_indexes(tmp_path, create_default_indexes) -> None:
 def test_raises_key_error_on_invalid_zarr_store(tmp_path):
     root = zarr.open_group(tmp_path / "tmp.zarr")
     if Version(zarr.__version__) < Version("3.0.0"):
-        root.create_dataset("bar", shape=(3, 5), dtype=np.float32)
+        getattr(root, "create_dataset")(  # noqa: B009
+            "bar", shape=(3, 5), dtype=np.float32
+        )
     else:
         root.create_array("bar", shape=(3, 5), dtype=np.float32)
     with pytest.raises(KeyError, match=r"xarray to determine variable dimensions"):

xarray/tests/test_dataarray.py

@@ -1463,6 +1463,20 @@ def test_sel(
 
         assert_identical(mdata.sel(x={"one": "a", "two": 1}), mdata.sel(one="a", two=1))
 
+    def test_selection_multiindex_nested_tuple_level_value(self) -> None:
+        level_0 = pd.Index(
+            [(1, 1), (1, 1), (2, 2), (3, 3)], name="a", tupleize_cols=False
+        )
+        level_1 = pd.Index([1, 2, 10, 20], name="b")
+        midx = pd.MultiIndex.from_arrays([level_0, level_1])
+        coords = Coordinates.from_pandas_multiindex(midx, "index")
+        data = DataArray(np.arange(4), dims=("index",), coords=coords)
+
+        actual = data.sel(index=((1, 1), 2))
+        expected = data.isel(index=1)
+
+        assert_identical(actual, expected)
+
     def test_selection_multiindex_remove_unused(self) -> None:
         # GH2619. For MultiIndex, we need to call remove_unused.
         ds = xr.DataArray(