fix: close oracle round 14 gaps

Author: yeongseon

docs/DEVELOPMENT.md

@@ -120,7 +120,7 @@ sqlalchemy-excel/
 │   ├── types.py              # ExcelTypeCompiler (type mappings)
 │   ├── reflection.py         # ExcelInspectionMixin (schema inspection)
 │   └── py.typed              # PEP 561 marker file
-├── tests/                    # 18 test modules (402 passed, 2 xfailed)
+├── tests/                    # 18 test modules (406 passed, 1 xfailed)
 │   ├── conftest.py           # Shared fixtures
 │   ├── test_alter_table.py
 │   ├── test_dialect.py

docs/USAGE.md

@@ -386,7 +386,7 @@ sqlalchemy-excel has some limitations due to the nature of Excel as a database:
 - **ALTER TABLE**: Supports `ADD COLUMN`, `DROP COLUMN`, and `RENAME COLUMN` via raw SQL through the driver.
 - **Raw CREATE/DROP reflection**: Raw `CREATE TABLE` writes declared schema metadata and raw `DROP TABLE` removes it.
 - **Schema guards scope**: Schema validation guards apply only to SQLAlchemy-compiled SQL. Raw SQL sent through `exec_driver_sql()` is forwarded directly to excel-dbapi without schema validation.
-- **ORM relationship limits**: Lazy one-to-many relationship loading can return empty collections; use eager loading (`joinedload`) for reliable one-to-many reads.
+- **ORM relationship limits**: Simple one-to-many lazy loading now works correctly. Use eager loading (`joinedload`) for more complex relationship patterns.
 - **Many-to-many loading is unsupported**: Association table persistence works, but relationship loader SQL for many-to-many is not fully supported.
 - **No foreign keys or indexes**: Excel has no concept of these.
 - **UNIQUE/CHECK/FOREIGN KEY are not enforced**: They are accepted for SQLAlchemy compatibility, ignored by the backend, and compile-time warnings are emitted for `CREATE TABLE` and `ALTER TABLE ... ADD COLUMN`.

src/sqlalchemy_excel/dialect.py

@@ -34,7 +34,13 @@ def _after_create(
         return
 
     for col in target.columns:
-        if col.server_default is not None:
+        # Skip server_default warning if column has Computed or Identity,
+        # as those are already warned in DDL compilation
+        if (
+            col.server_default is not None
+            and getattr(col, "computed", None) is None
+            and getattr(col, "identity", None) is None
+        ):
             warnings.warn(
                 "Excel dialect does not support server_default; value will be ignored",
                 stacklevel=2,
@@ -44,16 +50,8 @@ def _after_create(
                 "Excel dialect does not support autoincrement=True; value must be set explicitly",
                 stacklevel=2,
             )
-        if getattr(col, "computed", None) is not None:
-            warnings.warn(
-                f"Column '{col.name}': Computed columns are not supported by excel dialect; the expression will be ignored",
-                stacklevel=2,
-            )
-        if getattr(col, "identity", None) is not None:
-            warnings.warn(
-                f"Column '{col.name}': Identity columns are not supported by excel dialect; auto-increment will not be applied",
-                stacklevel=2,
-            )
+        # Skip Computed and Identity warnings here - they're already warned in DDL compilation
+        # to avoid duplicate warnings
 
     import excel_dbapi
 

tests/test_orm_relationships.py

@@ -4,7 +4,7 @@
 - one-to-many inserts and joined eager loading
 
 Known limitations:
-- lazy relationship loading can return incorrect results
+- many-to-many relationship loading can return incorrect results
 - many-to-many relationship loaders generate SQL patterns the backend can't parse
 """
 
@@ -113,9 +113,6 @@ def test_one_to_many_relationship_round_trip(relationship_engine: Engine) -> Non
         assert [child.name for child in parent.children] == ["child-1", "child-2"]
 
 
-@pytest.mark.xfail(
-    reason="Lazy one-to-many loaders currently return empty collections with this backend.",
-)
 def test_one_to_many_lazy_loading_boundary(relationship_engine: Engine) -> None:
     with Session(relationship_engine) as session:
         session.add(