fix: to_gbq may swap data columns when replace table (#2532)

Fixes
#2502

Author: chelsea-lin

bigframes/session/bq_caching_executor.py

@@ -247,7 +247,7 @@ def _export_gbq(
         )
         sql = compiled.sql
 
-        if (existing_table is not None) and _if_schema_match(
+        if (existing_table is not None) and _is_schema_match(
             existing_table.schema, array_value.schema
         ):
             # b/409086472: Uses DML for table appends and replacements to avoid
@@ -690,16 +690,16 @@ def _result_schema(
     )
 
 
-def _if_schema_match(
-    table_schema: Tuple[bigquery.SchemaField, ...], schema: schemata.ArraySchema
+def _is_schema_match(
+    table_schema: Tuple[bigquery.SchemaField, ...],
+    schema: schemata.ArraySchema,
 ) -> bool:
     if len(table_schema) != len(schema.items):
         return False
-    for field in table_schema:
-        if field.name not in schema.names:
+    for field, schema_item in zip(table_schema, schema.items):
+        if field.name != schema_item.column:
             return False
-        if bigframes.dtypes.convert_schema_field(field)[1] != schema.get_type(
-            field.name
-        ):
+        _, field_dtype = bigframes.dtypes.convert_schema_field(field)
+        if field_dtype != schema_item.dtype:
             return False
     return True

tests/system/small/test_dataframe_io.py

@@ -631,6 +631,13 @@ def test_to_gbq_if_exists_is_replace(scalars_dfs, dataset_id):
     assert len(gcs_df) == len(scalars_pandas_df)
     pd.testing.assert_index_equal(gcs_df.columns, scalars_pandas_df.columns)
 
+    # When replacing a table with same schema but different column order
+    reordered_df = scalars_df[scalars_df.columns[::-1]]
+    reordered_df.to_gbq(destination_table, if_exists="replace")
+    gcs_df = pandas_gbq.read_gbq(destination_table, index_col="rowindex")
+    assert len(gcs_df) == len(scalars_pandas_df)
+    pd.testing.assert_index_equal(gcs_df.columns, reordered_df.columns)
+
     # When replacing a table with different schema
     partitial_scalars_df = scalars_df.drop(columns=["string_col"])
     partitial_scalars_df.to_gbq(destination_table, if_exists="replace")