Fix index of Groupby.apply results when it is performed on empty objects

python/cudf/cudf/core/groupby/groupby.py

@@ -21,7 +21,7 @@
 from cudf._lib.sort import segmented_sort_by_key
 from cudf._lib.types import size_type_dtype
 from cudf._typing import AggType, DataFrameOrSeries, MultiColumnAggType
-from cudf.api.types import is_list_like
+from cudf.api.types import is_bool_dtype, is_list_like
 from cudf.core.abc import Serializable
 from cudf.core.column.column import ColumnBase, arange, as_column
 from cudf.core.column_accessor import ColumnAccessor
@@ -1373,7 +1373,16 @@ def mult(df):
         """
 
         if self.obj.empty:
-            return self.obj
+            res = self.obj.copy(deep=True)
+            res.index = self.grouping.keys
+            if function in {"sum", "product"}:
+                # For `sum` & `product`, boolean types
+                # will need to result in `int64` type.
+                for name, col in res._data.items():
+                    if is_bool_dtype(col.dtype):
+                        res._data[name] = col.astype("int")
+            return res
+
         if not callable(function):
             raise TypeError(f"type {type(function)} is not callable")
         group_names, offsets, group_keys, grouped_values = self._grouped()

python/cudf/cudf/tests/test_groupby.py

@@ -3342,6 +3342,35 @@ def test_group_by_pandas_sort_order(groups, sort):
         )
 
 
+@pytest.mark.parametrize(
+    "dtype",
+    ["int32", "int64", "float64", "datetime64[ns]", "timedelta64[ns]", "bool"],
+)
+@pytest.mark.parametrize(
+    "apply_op",
+    ["sum", "min", "max", "idxmax"],
+)
+def test_group_by_empty_apply(request, dtype, apply_op):
+    request.applymarker(
+        pytest.mark.xfail(
+            condition=(dtype == "datetime64[ns]" and apply_op == "sum"),
+            reason=("sum isn't supported for datetime64[ns]"),
+        )
+    )
+    gdf = cudf.DataFrame({"a": [], "b": [], "c": []}, dtype=dtype)
+    pdf = gdf.to_pandas()
+
+    gg = gdf.groupby("a")["c"]
+    pg = pdf.groupby("a")["c"]
+
+    assert_eq(
+        gg.apply(apply_op),
+        pg.apply(apply_op),
+        check_dtype=True,
+        check_index_type=True,
+    )
+
+
 def test_groupby_consecutive_operations():
     df = cudf.DataFrame([[1, np.nan], [1, 4], [5, 6]], columns=["A", "B"])
     pdf = df.to_pandas()