Fix min_count behaviour with flox. (#7809)

* Fix `min_count` behaviour with flox.

Closes #7808

* Apply suggestions from code review

Co-authored-by: Illviljan <14371165+Illviljan@users.noreply.github.com>

---------

Co-authored-by: Illviljan <14371165+Illviljan@users.noreply.github.com>

Author: dcherian

doc/whats-new.rst

@@ -42,6 +42,8 @@ Bug fixes
 ~~~~~~~~~
 - Fix groupby binary ops when grouped array is subset relative to other. (:issue:`7797`).
   By `Deepak Cherian <https://github.com/dcherian>`_.
+- Fix groupby sum, prod for all-NaN groups with ``flox``. (:issue:`7808`).
+  By `Deepak Cherian <https://github.com/dcherian>`_.
 
 Documentation
 ~~~~~~~~~~~~~

xarray/core/groupby.py

@@ -964,6 +964,13 @@ def _flox_reduce(
         else:
             non_numeric = {}
 
+        if "min_count" in kwargs:
+            if kwargs["func"] not in ["sum", "prod"]:
+                raise TypeError("Received an unexpected keyword argument 'min_count'")
+            elif kwargs["min_count"] is None:
+                # set explicitly to avoid unncessarily accumulating count
+                kwargs["min_count"] = 0
+
         # weird backcompat
         # reducing along a unique indexed dimension with squeeze=True
         # should raise an error

xarray/tests/test_groupby.py

@@ -1,6 +1,7 @@
 from __future__ import annotations
 
 import datetime
+import operator
 import warnings
 
 import numpy as np
@@ -17,6 +18,7 @@
     assert_identical,
     create_test_data,
     has_cftime,
+    has_flox,
     has_pandas_version_two,
     requires_dask,
     requires_flox,
@@ -2336,3 +2338,37 @@ def test_groupby_binary_op_regression() -> None:
     anom_gb = x_slice.groupby("time.month") - clim
 
     assert_identical(xr.zeros_like(anom_gb), anom_gb)
+
+
+@requires_flox
+@pytest.mark.parametrize("func", ["sum", "prod"])
+@pytest.mark.parametrize("skipna", [True, False])
+@pytest.mark.parametrize("min_count", [None, 1])
+def test_min_count_vs_flox(func: str, min_count: int | None, skipna: bool) -> None:
+    da = DataArray(
+        data=np.array([np.nan, 1, 1, np.nan, 1, 1]),
+        dims="x",
+        coords={"labels": ("x", np.array([1, 2, 3, 1, 2, 3]))},
+    )
+
+    gb = da.groupby("labels")
+    method = operator.methodcaller(func, min_count=min_count, skipna=skipna)
+    with xr.set_options(use_flox=True):
+        actual = method(gb)
+    with xr.set_options(use_flox=False):
+        expected = method(gb)
+    assert_identical(actual, expected)
+
+
+@pytest.mark.parametrize("use_flox", [True, False])
+def test_min_count_error(use_flox: bool) -> None:
+    if use_flox and not has_flox:
+        pytest.skip()
+    da = DataArray(
+        data=np.array([np.nan, 1, 1, np.nan, 1, 1]),
+        dims="x",
+        coords={"labels": ("x", np.array([1, 2, 3, 1, 2, 3]))},
+    )
+    with xr.set_options(use_flox=use_flox):
+        with pytest.raises(TypeError):
+            da.groupby("labels").mean(min_count=1)