Allow fillna to validate for CategoricalColumn.fillna (#15683)

Fixes: #15666 

This PR validates values passed to `fillna` even if there are no null values in a categorical column.

Forks from #14534

Authors:
  - GALI PREM SAGAR (https://github.com/galipremsagar)

Approvers:
  - Matthew Roeschke (https://github.com/mroeschke)

URL: #15683

python/cudf/cudf/core/column/categorical.py

@@ -1045,9 +1045,6 @@ def fillna(
         """
         Fill null values with *fill_value*
         """
-        if not self.nullable:
-            return self
-
         if fill_value is not None:
             fill_is_scalar = np.isscalar(fill_value)
 
@@ -1079,6 +1076,11 @@ def fillna(
                     self.codes.dtype
                 )
 
+        # Validation of `fill_value` will have to be performed
+        # before returning self.
+        if not self.nullable:
+            return self
+
         return super().fillna(fill_value, method=method)
 
     def indices_of(

python/cudf/cudf/core/frame.py

@@ -762,10 +762,17 @@ def fillna(
             else:
                 replace_val = None
             should_fill = (
-                col_name in value
-                and col.has_nulls(include_nan=True)
-                and not libcudf.scalar._is_null_host_scalar(replace_val)
-            ) or method is not None
+                (
+                    col_name in value
+                    and col.has_nulls(include_nan=True)
+                    and not libcudf.scalar._is_null_host_scalar(replace_val)
+                )
+                or method is not None
+                or (
+                    isinstance(col, cudf.core.column.CategoricalColumn)
+                    and not libcudf.scalar._is_null_host_scalar(replace_val)
+                )
+            )
             if should_fill:
                 filled_data[col_name] = col.fillna(replace_val, method)
             else:

python/cudf/cudf/tests/test_categorical.py

@@ -859,3 +859,19 @@ def test_cat_from_scalar(scalar):
     gs = cudf.Series(scalar, dtype="category")
 
     assert_eq(ps, gs)
+
+
+def test_cat_groupby_fillna():
+    ps = pd.Series(["a", "b", "c"], dtype="category")
+    gs = cudf.from_pandas(ps)
+
+    with pytest.warns(FutureWarning):
+        pg = ps.groupby(ps)
+    gg = gs.groupby(gs)
+
+    assert_exceptions_equal(
+        lfunc=pg.fillna,
+        rfunc=gg.fillna,
+        lfunc_args_and_kwargs=(("d",), {}),
+        rfunc_args_and_kwargs=(("d",), {}),
+    )