Support Series.__setitem__ with key to a new row (rapidsai#7443)

Closes rapidsai#7290 

Supports assigning to a new row (specified by a new label) in a series.

Authors:
  - Michael Wang (@isVoid)

Approvers:
  - @brandon-b-miller
  - GALI PREM SAGAR (@galipremsagar)

URL: rapidsai#7443

python/cudf/cudf/core/indexing.py

@@ -7,9 +7,12 @@
 from nvtx import annotate
 
 import cudf
+from cudf._lib.concat import concat_columns
 from cudf._lib.scalar import _is_null_host_scalar
-from cudf._typing import DataFrameOrSeries, ScalarLike
+from cudf._typing import ColumnLike, DataFrameOrSeries, ScalarLike
+from cudf.core.column.column import as_column
 from cudf.utils.dtypes import (
+    find_common_type,
     is_categorical_dtype,
     is_column_like,
     is_list_like,
@@ -142,7 +145,19 @@ def __getitem__(self, arg: Any) -> Union[ScalarLike, DataFrameOrSeries]:
         return self._sr.iloc[arg]
 
     def __setitem__(self, key, value):
-        key = self._loc_to_iloc(key)
+        try:
+            key = self._loc_to_iloc(key)
+        except KeyError as e:
+            if (
+                is_scalar(key)
+                and not isinstance(self._sr.index, cudf.MultiIndex)
+                and is_scalar(value)
+            ):
+                _append_new_row_inplace(self._sr.index._values, key)
+                _append_new_row_inplace(self._sr._column, value)
+                return
+            else:
+                raise e
         if isinstance(value, (pd.Series, cudf.Series)):
             value = cudf.Series(value)
             value = value._align_to_index(self._sr.index, how="right")
@@ -481,3 +496,14 @@ def _normalize_dtypes(df):
         for name, col in df._data.items():
             df[name] = col.astype(normalized_dtype)
     return df
+
+
+def _append_new_row_inplace(col: ColumnLike, value: ScalarLike):
+    """Append a scalar `value` to the end of `col` inplace.
+       Cast to common type if possible
+    """
+    to_type = find_common_type([type(value), col.dtype])
+    val_col = as_column(value, dtype=to_type)
+    old_col = col.astype(to_type)
+
+    col._mimic_inplace(concat_columns([old_col, val_col]), inplace=True)

python/cudf/cudf/tests/test_indexing.py

@@ -1043,6 +1043,10 @@ def test_series_setitem_string(key, value):
     [
         ("a", 4),
         ("b", 4),
+        ("b", np.int8(8)),
+        ("d", 4),
+        ("d", np.int8(16)),
+        ("d", np.float32(16)),
         (["a", "b"], 4),
         (["a", "b"], [4, 5]),
         ([True, False, True], 4),
@@ -1058,6 +1062,27 @@ def test_series_setitem_loc(key, value):
     assert_eq(psr, gsr)
 
 
+@pytest.mark.parametrize(
+    "key, value",
+    [
+        (1, "d"),
+        (2, "e"),
+        (4, "f"),
+        ([1, 3], "g"),
+        ([1, 3], ["g", "h"]),
+        ([True, False, True], "i"),
+        ([False, False, False], "j"),
+        ([True, False, True], ["k", "l"]),
+    ],
+)
+def test_series_setitem_loc_numeric_index(key, value):
+    psr = pd.Series(["a", "b", "c"], [1, 2, 3])
+    gsr = cudf.from_pandas(psr)
+    psr.loc[key] = value
+    gsr.loc[key] = value
+    assert_eq(psr, gsr)
+
+
 @pytest.mark.parametrize(
     "key, value",
     [