flake8-sqlalchemy2 0.3.0

flake8 plugin for SQLAlchemy 2.0

flake8-sqlalchemy2

flake8 plugin to enforce modern, typed SQLAlchemy 2.0.

Installation

Use uvx for a one-time check of your code base:

uvx --with flake8-sqlalchemy2 flake8 --select SA2

Install via pip for using as "permanent" flake8 plugin:

pip install flake8-sqlalchemy2

Rules

missing-mapped-type-annotation (SA201)

What it does

Checks for existence of Mapped or other ORM container class type annotations in SQLAlchemy models.

Why is this bad?

If an annotation is missing, type checkers will treat the corresponding field as type Any.

Example

from sqlalchemy import Integer
from sqlalchemy.orm import DeclarativeBase
from sqlalchemy.orm import Mapped
from sqlalchemy.orm import mapped_column

class Base(DeclarativeBase):
    pass


class MyModel(Base):
    __tablename__ = "my_model"
    id: Mapped[int] = mapped_column(primary_key=True)

    count = mapped_column(Integer)


m = MyModel()
reveal_type(m.count)  #  note: Revealed type is "Any"

Use instead:

from sqlalchemy.orm import DeclarativeBase
from sqlalchemy.orm import Mapped
from sqlalchemy.orm import mapped_column

class Base(DeclarativeBase):
    pass


class MyModel(Base):
    __tablename__ = "my_model"
    id: Mapped[int] = mapped_column(primary_key=True)

    count: Mapped[int]


m = MyModel()
reveal_type(m.count)  # note: Revealed type is "builtins.int"

legacy-collection (SA202)

What it does

Checks for existence of DynamicMapped.

Why is this bad?

DynamicMapped is considered legacy and exposes the legacy query API.

Example

from sqlalchemy.orm import DeclarativeBase
from sqlalchemy.orm import DynamicMapped
from sqlalchemy.orm import mapped_column
from sqlalchemy.orm import Mapped
from sqlalchemy.orm import relationship

class Base(DeclarativeBase):
    pass


class MyModel(Base):
    __tablename__ = "my_model"
    id: Mapped[int] = mapped_column(primary_key=True)

    children: DynamicMapped["Child"] = relationship()

Use instead:

from sqlalchemy.orm import DeclarativeBase
from sqlalchemy.orm import mapped_column
from sqlalchemy.orm import Mapped
from sqlalchemy.orm import relationship
from sqlalchemy.orm import WriteOnlyMapped

class Base(DeclarativeBase):
  pass


class MyModel(Base):
    __tablename__ = "my_model"
    id: Mapped[int] = mapped_column(primary_key=True)

    children: WriteOnlyMapped["Child"] = relationship()

legacy-relationship (SA203)

What it does

Checks for existence of relationship definition with backref keyword argument.

Why is this bad?

backref is considered legacy. It adds dynamic attributes that type checkers and code completion cannot understand.

Example

from typing import List

from sqlalchemy import ForeignKey

from sqlalchemy.orm import DeclarativeBase
from sqlalchemy.orm import mapped_column
from sqlalchemy.orm import Mapped
from sqlalchemy.orm import relationship

class Base(DeclarativeBase):
    pass


class Parent(Base):
    __tablename__ = "parent"
    id: Mapped[int] = mapped_column(primary_key=True)

    children: Mapped[List["Child"]] = relationship(backref="parent")


class Child(Base):
    __tablename__ = "child"
    id: Mapped[int] = mapped_column(primary_key=True)

    parent_id: Mapped[int] = mapped_column(ForeignKey("parent.id"))


c = Child()
p = Parent(children=[c])
c.parent  # error: "Child" has no attribute "parent"; maybe "parent_id"?  [attr-defined]

Use instead:

from typing import List

from sqlalchemy import ForeignKey

from sqlalchemy.orm import DeclarativeBase
from sqlalchemy.orm import mapped_column
from sqlalchemy.orm import Mapped
from sqlalchemy.orm import relationship

class Base(DeclarativeBase):
  pass


class Parent(Base):
    __tablename__ = "parent"
    id: Mapped[int] = mapped_column(primary_key=True)

    children: Mapped[List["Child"]] = relationship(back_populates="parent")


class Child(Base):
    __tablename__ = "child"
    id: Mapped[int] = mapped_column(primary_key=True)

    parent_id: Mapped[int] = mapped_column(ForeignKey("parent.id"))
    parent: Mapped["Parent"] = relationship(back_populates="children")

Note on ruff

Q: Why still use flake8 when there is ruff!?

A: For rules not supported by ruff. There is a proposed merge request to bring the first SQLAlchemy linting rule (SA201) to ruff ("needs-decision" tagged).

Note on flake8-sqlalchemy

Q: Why not integrate these rules into flake8-sqlalchemy?

A: The focus of this package are rules for modern, typed SQLAlchemy. Furthermore, I wanted to learn something new.