Convert Column Name to String Before Using Struct Column Factory (NVI…

…DIA#10156)

Closes NVIDIA#10155 

`build_struct_column` requires that the field names to be strings. But dataframe column names can be any hashable types. Passing in column names as field names in `to_struct` is thus unsafe. This PR adds a check and raise a warning if the cast to string is required to take place.

Authors:
  - Michael Wang (https://github.com/isVoid)
  - Bradley Dice (https://github.com/bdice)

Approvers:
  - GALI PREM SAGAR (https://github.com/galipremsagar)
  - Sheilah Kirui (https://github.com/skirui-source)
  - Bradley Dice (https://github.com/bdice)

URL: rapidsai/cudf#10156

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

@@ -1602,8 +1602,8 @@ def build_struct_column(
 
     Parameters
     ----------
-    names : list-like
-        Field names to map to children dtypes
+    names : sequence of strings
+        Field names to map to children dtypes, must be strings.
     children : tuple
 
     mask: Buffer

python/cudf/cudf/core/dataframe.py

@@ -5864,8 +5864,16 @@ def to_struct(self, name=None):
         -----
         Note that a copy of the columns is made.
         """
+        if not all(isinstance(name, str) for name in self._data.names):
+            warnings.warn(
+                "DataFrame contains non-string column name(s). Struct column "
+                "requires field name to be string. Non-string column names "
+                "will be casted to string as the field name."
+            )
+        field_names = [str(name) for name in self._data.names]
+
         col = cudf.core.column.build_struct_column(
-            names=self._data.names, children=self._data.columns, size=len(self)
+            names=field_names, children=self._data.columns, size=len(self)
         )
         return cudf.Series._from_data(
             cudf.core.column_accessor.ColumnAccessor(

python/cudf/cudf/tests/test_struct.py

@@ -205,6 +205,14 @@ def test_dataframe_to_struct():
     df["a"][0] = 5
     assert_eq(got, expect)
 
+    # check that a non-string (but convertible to string) named column can be
+    # converted to struct
+    df = cudf.DataFrame([[1, 2], [3, 4]], columns=[(1, "b"), 0])
+    expect = cudf.Series([{"(1, 'b')": 1, "0": 2}, {"(1, 'b')": 3, "0": 4}])
+    with pytest.warns(UserWarning, match="will be casted"):
+        got = df.to_struct()
+    assert_eq(got, expect)
+
 
 @pytest.mark.parametrize(
     "series, slce",