Move to using the sqlalchemy mypy plugin

This provides better type checking against the sqlalchemy code
and allows us to remove our stubs and the 'type: ignore's which
were in various places where the stubs didn't match reality.

routemaster/conftest.py

@@ -326,7 +326,7 @@ def database_clear(app: TestApp) -> Iterator[None]:
         with app.new_session():
             for table in metadata.tables:
                 app.session.execute(
-                    f'truncate table {table} cascade',
+                    f'truncate table {table} cascade',  # type: ignore[arg-type]  # noqa: E501
                     {},
                 )
 
@@ -514,8 +514,7 @@ def _inner(label: LabelRef) -> str:
                 label_name=label.name,
                 label_state_machine=label.state_machine,
             ).order_by(
-                # TODO: use the sqlalchemy mypy plugin rather than our stubs
-                History.id.desc(),  # type: ignore[attr-defined]
+                History.id.desc(),
             ).limit(1).scalar()
     return _inner
 

routemaster/db/model.py

@@ -1,7 +1,7 @@
 """Database model definition."""
 import datetime
 import functools
-from typing import Any
+from typing import Any, Dict, List, Optional
 
 import dateutil.tz
 from sqlalchemy import DDL, Table
@@ -16,7 +16,7 @@
     ForeignKeyConstraint,
     func,
 )
-from sqlalchemy.orm import relationship
+from sqlalchemy.orm import Mapped, relationship
 from sqlalchemy.ext.declarative import declarative_base
 from sqlalchemy.dialects.postgresql import JSONB
 
@@ -53,7 +53,8 @@
 class Label(Base):
     """A single label including context."""
 
-    # Note: type annotations for this class are provided by a stubs file
+    # Note: type annotations provided below must be manually kept in sync with
+    # the fields defined in the Table.
 
     __table__ = Table(
         'labels',
@@ -74,6 +75,15 @@ class Label(Base):
         ],
     )
 
+    name: Mapped[str]
+    state_machine: Mapped[str]
+    metadata: Mapped[Dict[str, Any]]
+    metadata_triggers_processed: Mapped[bool]
+    deleted: Mapped[bool]
+    updated: Mapped[datetime.datetime]
+
+    history: List['History'] = relationship('History')
+
     def __repr__(self) -> str:
         """Return a useful debug representation."""
         return (
@@ -84,7 +94,8 @@ def __repr__(self) -> str:
 class History(Base):
     """A single historical state transition of a label."""
 
-    # Note: type annotations for this class are provided by a stubs file
+    # Note: type annotations provided below must be manually kept in sync with
+    # the fields defined in the Table.
 
     __table__ = Table(
         'history',
@@ -115,7 +126,15 @@ class History(Base):
         NullableColumn('new_state', String),
     )
 
-    label = relationship(Label, backref='history')
+    id: Mapped[int]
+    label_name: Mapped[str]
+    label_state_machine: Mapped[str]
+    created: Mapped[datetime.datetime]
+    forced: Mapped[bool]
+    old_state: Mapped[Optional[str]]
+    new_state: Mapped[Optional[str]]
+
+    label: Label = relationship(Label)
 
     def __repr__(self) -> str:
         """Return a useful debug representation."""

routemaster/state_machine/api.py

@@ -108,7 +108,7 @@ def update_metadata_for_label(
         )
 
     # FIXME: handle cases where metadata aren't dicts.
-    new_metadata = dict_merge(existing_metadata, update)  # type: ignore[arg-type]  # noqa: E501
+    new_metadata = dict_merge(existing_metadata, update)
 
     row.metadata = new_metadata
     row.metadata_triggers_processed = not needs_gate_evaluation

routemaster/state_machine/utils.py

@@ -101,11 +101,7 @@ def get_current_history(app: App, label: LabelRef) -> History:
         label_name=label.name,
         label_state_machine=label.state_machine,
     ).order_by(
-        # Our model type stubs define the `id` attribute as `int`, yet
-        # sqlalchemy actually allows the attribute to be used for ordering like
-        # this; ignore the type check here specifically rather than complicate
-        # our type definitions.
-        History.id.desc(),  # type: ignore[attr-defined]
+        History.id.desc(),
     ).first()
 
     if history_entry is None:
@@ -185,14 +181,13 @@ def labels_in_state_with_metadata(
 
     metadata_lookup = Label.metadata
     for part in path:
-        metadata_lookup = metadata_lookup[part]  # type: ignore[call-overload, index]  # noqa: E501
+        metadata_lookup = metadata_lookup[part]  # type: ignore[assignment]  # noqa: E501
 
     return _labels_in_state(
         app,
         state_machine,
         state,
-        # TODO: use the sqlalchemy mypy plugin rather than our stubs file
-        metadata_lookup.astext.in_(values),  # type: ignore[union-attr]
+        metadata_lookup.astext.in_(values),
     )
 
 
@@ -228,11 +223,7 @@ def _labels_in_state(
         History.label_name,
         History.new_state,
         func.row_number().over(
-            # Our model type stubs define the `id` attribute as `int`, yet
-            # sqlalchemy actually allows the attribute to be used for ordering
-            # like this; ignore the type check here specifically rather than
-            # complicate our type definitions.
-            order_by=History.id.desc(),  # type: ignore[attr-defined]
+            order_by=History.id.desc(),
             partition_by=History.label_name,
         ).label('rank'),
     ).filter_by(

routemaster/validation.py

@@ -78,8 +78,7 @@ def _validate_no_labels_in_nonexistent_states(
         History.label_name,
         History.new_state,
         func.row_number().over(
-            # TODO: use the sqlalchemy mypy plugin rather than our stubs file
-            order_by=History.id.desc(),  # type: ignore[attr-defined]
+            order_by=History.id.desc(),
             partition_by=History.label_name,
         ).label('rank'),
     ).filter_by(

setup.cfg

@@ -18,6 +18,7 @@ strict_optional=true
 show_error_codes = true
 enable_error_code = ignore-without-code
 warn_unused_ignores = true
+plugins = sqlalchemy.ext.mypy.plugin
 
 [coverage:run]
 branch=True

setup.py

@@ -50,7 +50,7 @@
         'jsonschema >=3, <5',
         'flask',
         'psycopg2-binary',
-        'sqlalchemy',
+        'sqlalchemy[mypy]',
         'python-dateutil',
         'alembic >=0.9.6',
         'gunicorn >=19.7',