Fix for stack+groupby+apply w/ non-increasing coord (#3906)

* WIP fix for stack+groupby+apply w/ non-increasing coord

* Only sort if multiindex

* store safe_cast_to_index call to avoid double call

* black formatting

Author: Spencer Hill

doc/whats-new.rst

@@ -45,6 +45,9 @@ New Features
 
 Bug fixes
 ~~~~~~~~~
+- Fix renaming of coords when one or more stacked coords is not in
+  sorted order during stack+groupby+apply operations. (:issue:`3287`,
+  :pull:`3906`) By `Spencer Hill <https://github.com/spencerahill>`_
 - Fix a regression where deleting a coordinate from a copied :py:class:`DataArray`
   can affect the original :py:class:`Dataarray`.  (:issue:`3899`, :pull:`3871`)
   By `Todd Jennings <https://github.com/toddrjen>`_
@@ -96,13 +99,13 @@ New Features
 - Added support for :py:class:`pandas.DatetimeIndex`-style rounding of
   ``cftime.datetime`` objects directly via a :py:class:`CFTimeIndex` or via the
   :py:class:`~core.accessor_dt.DatetimeAccessor`.
-  By `Spencer Clark <https://github.com/spencerkclark>`_ 
+  By `Spencer Clark <https://github.com/spencerkclark>`_
 - Support new h5netcdf backend keyword `phony_dims` (available from h5netcdf
   v0.8.0 for :py:class:`~xarray.backends.H5NetCDFStore`.
   By `Kai Mühlbauer <https://github.com/kmuehlbauer>`_.
 - Add partial support for unit aware arrays with pint. (:pull:`3706`, :pull:`3611`)
   By `Justus Magin <https://github.com/keewis>`_.
-- :py:meth:`Dataset.groupby` and :py:meth:`DataArray.groupby` now raise a 
+- :py:meth:`Dataset.groupby` and :py:meth:`DataArray.groupby` now raise a
   `TypeError` on multiple string arguments. Receiving multiple string arguments
   often means a user is attempting to pass multiple dimensions as separate
   arguments and should instead pass a single list of dimensions.
@@ -120,7 +123,7 @@ New Features
   By `Maximilian Roos <https://github.com/max-sixty>`_.
 - ``skipna`` is available in :py:meth:`Dataset.quantile`, :py:meth:`DataArray.quantile`,
   :py:meth:`core.groupby.DatasetGroupBy.quantile`, :py:meth:`core.groupby.DataArrayGroupBy.quantile`
-  (:issue:`3843`, :pull:`3844`) 
+  (:issue:`3843`, :pull:`3844`)
   By `Aaron Spring <https://github.com/aaronspring>`_.
 
 Bug fixes
@@ -814,7 +817,7 @@ Bug fixes
 Documentation
 ~~~~~~~~~~~~~
 
-- Created a `PR checklist <https://xarray.pydata.org/en/stable/contributing.html/contributing.html#pr-checklist>`_ 
+- Created a `PR checklist <https://xarray.pydata.org/en/stable/contributing.html/contributing.html#pr-checklist>`_
   as a quick reference for tasks before creating a new PR
   or pushing new commits.
   By `Gregory Gundersen <https://github.com/gwgundersen>`_.

xarray/core/groupby.py

@@ -370,8 +370,10 @@ def __init__(
                 group = group.dropna(group_dim)
 
             # look through group to find the unique values
+            group_as_index = safe_cast_to_index(group)
+            sort = bins is None and (not isinstance(group_as_index, pd.MultiIndex))
             unique_values, group_indices = unique_value_groups(
-                safe_cast_to_index(group), sort=(bins is None)
+                group_as_index, sort=sort
             )
             unique_coord = IndexVariable(group.name, unique_values)
 

xarray/tests/test_dataarray.py

@@ -1202,6 +1202,25 @@ def test_selection_multiindex_from_level(self):
         expected = data.isel(xy=[0, 1]).unstack("xy").squeeze("y").drop_vars("y")
         assert_equal(actual, expected)
 
+    def test_stack_groupby_unsorted_coord(self):
+        data = [[0, 1], [2, 3]]
+        data_flat = [0, 1, 2, 3]
+        dims = ["x", "y"]
+        y_vals = [2, 3]
+
+        arr = xr.DataArray(data, dims=dims, coords={"y": y_vals})
+        actual1 = arr.stack(z=dims).groupby("z").first()
+        midx1 = pd.MultiIndex.from_product([[0, 1], [2, 3]], names=dims)
+        expected1 = xr.DataArray(data_flat, dims=["z"], coords={"z": midx1})
+        xr.testing.assert_equal(actual1, expected1)
+
+        # GH: 3287.  Note that y coord values are not in sorted order.
+        arr = xr.DataArray(data, dims=dims, coords={"y": y_vals[::-1]})
+        actual2 = arr.stack(z=dims).groupby("z").first()
+        midx2 = pd.MultiIndex.from_product([[0, 1], [3, 2]], names=dims)
+        expected2 = xr.DataArray(data_flat, dims=["z"], coords={"z": midx2})
+        xr.testing.assert_equal(actual2, expected2)
+
     def test_virtual_default_coords(self):
         array = DataArray(np.zeros((5,)), dims="x")
         expected = DataArray(range(5), dims="x", name="x")