Add support for funky column names

Author: J-Meyers

duckdb/polars_io.py

@@ -95,7 +95,8 @@ def _pl_tree_to_sql(tree: dict) -> str:
         )
     if node_type == "Column":
         # A reference to a column name
-        return subtree
+        # Wrap in quotes to handle special characters
+        return f'"{subtree}"'
 
     if node_type in ("Literal", "Dyn"):
         # Recursively process dynamic or literal values
@@ -196,7 +197,7 @@ def source_generator(
         duck_predicate = None
         relation_final = relation
         if with_columns is not None:
-            cols = ",".join(with_columns)
+            cols = ",".join(f'"{col}"' for col in with_columns)
             relation_final = relation_final.project(cols)
         if n_rows is not None:
             relation_final = relation_final.limit(n_rows)
@@ -213,7 +214,6 @@ def source_generator(
         while True:
             try:
                 record_batch = results.read_next_batch()
-                df = pl.from_arrow(record_batch)
                 if predicate is not None and duck_predicate is None:
                     # We have a predicate, but did not manage to push it down, we fallback here
                     yield pl.from_arrow(record_batch).filter(predicate)

tests/fast/arrow/test_polars.py

@@ -131,6 +131,29 @@ def test_polars_lazy(self, duckdb_cursor):
         ]
         assert lazy_df.filter(pl.col("b") < 32).select('a').collect().to_dicts() == [{'a': 'Mark'}, {'a': 'Thijs'}]
 
+    def test_polars_column_with_tricky_name(self, duckdb_cursor):
+        # Test that a polars DataFrame with a column name that is non standard still works
+        df_colon = pl.DataFrame({"x:y": [1, 2]})
+        lf = duckdb_cursor.sql("from df_colon").pl(lazy=True)
+        result = lf.select(pl.all()).collect()
+        assert result.to_dicts() == [{"x:y": 1}, {"x:y": 2}]
+        result = lf.select(pl.all()).filter(pl.col("x:y") == 1).collect()
+        assert result.to_dicts() == [{"x:y": 1}]
+
+        df_space = pl.DataFrame({"x y": [1, 2]})
+        lf = duckdb_cursor.sql("from df_space").pl(lazy=True)
+        result = lf.select(pl.all()).collect()
+        assert result.to_dicts() == [{"x y": 1}, {"x y": 2}]
+        result = lf.select(pl.all()).filter(pl.col("x y") == 1).collect()
+        assert result.to_dicts() == [{"x y": 1}]
+
+        df_dot = pl.DataFrame({"x.y": [1, 2]})
+        lf = duckdb_cursor.sql("from df_dot").pl(lazy=True)
+        result = lf.select(pl.all()).collect()
+        assert result.to_dicts() == [{"x.y": 1}, {"x.y": 2}]
+        result = lf.select(pl.all()).filter(pl.col("x.y") == 1).collect()
+        assert result.to_dicts() == [{"x.y": 1}]
+
     @pytest.mark.parametrize(
         'data_type',
         [