flake8-type-checking

Lets you know which imports to move in or out of type-checking blocks.

The plugin assumes that the imports you only use for type hinting are not required at runtime. When imports aren't strictly required at runtime, it means we can guard them.

Guarding imports provides 3 major benefits:

• 🔧  It reduces import circularity issues,
• 🧹  It organizes imports, and
• 🚀  It completely eliminates the overhead of type hint imports at runtime

Essentially, this code:

import pandas  # 15mb library

x: pandas.DataFrame

becomes this:

from typing import TYPE_CHECKING

if TYPE_CHECKING:
    import pandas  # <-- no longer imported at runtime

x: "pandas.DataFrame"

More examples can be found in the examples section.

If you're using pydantic, fastapi, cattrs, or injector see the configuration for how to enable support.

Primary features

The plugin will:

• Tell you when an import should be moved into a type-checking block
• Tell you when an import should be moved out again

And depending on which error code range you've opted into, it will tell you

• Whether you need to add a from __future__ import annotations import
• Whether you need to quote an annotation
• Whether you can unquote a quoted annotation

Error codes

Choosing how to handle forward references

You need to choose whether to opt-into using the TC100- or the TC200-range of error codes.

They represent two different ways of solving the same problem, so please only choose one.

TC100 and TC101 manage forward references by taking advantage of postponed evaluation of annotations.

TC200 and TC201 manage forward references using string literals.

Enabling error ranges

Add TC and TC1 or TC2 to your flake8 config like this:

[flake8]
max-line-length = 80
max-complexity = 12
...
ignore = E501
# You can use 'extend-select' (new in flake8 v4):
extend-select = TC, TC2
# OR 'select':
select = C,E,F..., TC, TC2  # or TC1
# OR 'enable-extensions':
enable-extensions = TC, TC2  # or TC1

If you are unsure which TC range to pick, see the rationale for more info.

Installation

pip install flake8-type-checking

Configuration

These options are configurable, and can be set in your flake8 config.

Exempt modules

If you wish to exempt certain modules from needing to be moved into type-checking blocks, you can specify which modules to ignore.

• setting name: type-checking-exempt-modules
• type: list

[flake8]
type-checking-exempt-modules = typing_extensions  # default []

Strict

The plugin, by default, will report TC00[1-3] errors for imports if there aren't already other imports from the same module. When there are other imports from the same module, the import circularity and performance benefits no longer apply from guarding an import.

When strict mode is enabled, the plugin will flag all imports that can be moved.

• setting name: type-checking-strict
• type: bool

[flake8]
type-checking-strict = true  # default false

Pydantic support

If you use Pydantic models in your code, you should enable Pydantic support. This will treat any class variable annotation as being needed during runtime.

• name: type-checking-pydantic-enabled
• type: bool

[flake8]
type-checking-pydantic-enabled = true  # default false

Pydantic support base-class passlist

Disabling checks for all class annotations is a little aggressive.

If you feel comfortable that all base classes named, e.g., NamedTuple are not Pydantic models, then you can pass the names of the base classes in this setting, to re-enable checking for classes which inherit from them.

• name: type-checking-pydantic-enabled-baseclass-passlist
• type: list

[flake8]
type-checking-pydantic-enabled-baseclass-passlist = NamedTuple, TypedDict  # default []

FastAPI support

If you're using the plugin for a FastAPI project, you should enable support. This will treat the annotations of any decorated function as needed at runtime.

Enabling FastAPI support will also enable Pydantic support.

• name: type-checking-fastapi-enabled
• type: bool

[flake8]
type-checking-fastapi-enabled = true  # default false

One more thing to note for FastAPI users is that dependencies (functions used in Depends) will produce false positives, unless you enable dependency support as described below.

FastAPI dependency support

In addition to preventing false positives for decorators, we can prevent false positives for dependencies. We are making a pretty bad trade-off however: by enabling this option we treat every annotation in every function definition across your entire project as a possible dependency annotation. In other words, we stop linting all function annotations completely, to avoid the possibility of false positives. If you prefer to be on the safe side, you should enable this - otherwise it might be enough to be aware that false positives can happen for functions used as dependencies.

Enabling dependency support will also enable FastAPI and Pydantic support.

• name: type-checking-fastapi-dependency-support-enabled
• type: bool

[flake8]
type-checking-fastapi-dependency-support-enabled = true  # default false

SQLAlchemy 2.0+ support

If you're using SQLAlchemy 2.0+, you can enable support. This will treat any Mapped[...] types as needed at runtime. It will also specially treat the enclosed type, since it may or may not need to be available at runtime depending on whether or not the enclosed type is a model or not, since models can have circular dependencies.

• name: type-checking-sqlalchemy-enabled
• type: bool

type-checking-sqlalchemy-enabled = true  # default false

SQLAlchemy 2.0+ support mapped dotted names

Since it's possible to create subclasses of sqlalchemy.orm.Mapped that define some custom behavior for the mapped attribute, but otherwise still behave like any other mapped attribute, i.e. the same runtime restrictions apply it's possible to provide additional dotted names that should be treated like subclasses of Mapped. By default we check for sqlalchemy.orm.Mapped, sqlalchemy.orm.DynamicMapped and sqlalchemy.orm.WriteOnlyMapped.

If there's more than one import path for the same Mapped subclass, then you need to specify each of them as a separate dotted name.

• name: type-checking-sqlalchemy-mapped-dotted-names
• type: list

type-checking-sqlalchemy-mapped-dotted-names = a.MyMapped, a.b.MyMapped  # default []

Cattrs support

If you're using the plugin in a project which uses cattrs, you can enable support. This will treat the annotations of any decorated attrs class as needed at runtime, since cattrs.unstructure calls will fail when loading classes where types are not available at runtime.

Note: the cattrs support setting does not yet detect and ignore class var annotations on dataclasses or other non-attrs class types. This can be added in the future if needed.

• name: type-checking-cattrs-enabled
• type: bool

[flake8]
type-checking-cattrs-enabled = true  # default false

Injector support

If you're using the injector library, you can enable support. This will treat any Inject[Dependency] types as needed at runtime.

• name: type-checking-injector-enabled
• type: bool

type-checking-injector-enabled = true  # default false

Rationale

Why did we create this plugin?

Good type hinting typically requires a lot of project imports, which can increase the risk of import cycles in a project. The recommended way of preventing this problem is to use typing.TYPE_CHECKING blocks to guard these types of imports. In particular, TC001 helps protect against this issue.

Once imports are guarded, they will no longer be evaluated/imported during runtime. The consequence of this is that these imports can no longer be treated as if they were imported outside the block. Instead we need to use forward references.

For Python version >= 3.7, there are actually two ways of solving this issue. You can either make your annotations string literals, or you can use a __futures__ import to enable postponed evaluation of annotations. See this excellent stackoverflow answer for a great explanation of the differences.

Examples

import pandas


def dataframe_length(df: pandas.DataFrame) -> int:
    return len(df)

from typing import TYPE_CHECKING


if TYPE_CHECKING:
    import pandas  # <-- no longer imported at runtime


def dataframe_length(df: "pandas.DataFrame") -> int:
    return len(df)

from models.b import B

class A(Model):
    def foo(self, b: B): ...

from models.a import A

class B(Model):
    def bar(self, a: A): ...

>> a.py: TC002 Move third-party import 'models.b.B' into a type-checking block
>> b.py: TC002 Move third-party import 'models.a.A' into a type-checking block

>> a.py: TC100 Add 'from __future__ import annotations' import
>> b.py: TC100 Add 'from __future__ import annotations' import

>> a.py: TC200 Annotation 'B' needs to be made into a string literal
>> b.py: TC200 Annotation 'A' needs to be made into a string literal

# TC1
from __future__ import annotations

from typing import TYPE_CHECKING

if TYPE_CHECKING:
    from models.b import B

class A(Model):
    def foo(self, b: B): ...

# TC2
from typing import TYPE_CHECKING

if TYPE_CHECKING:
    from models.b import B

class A(Model):
    def foo(self, b: 'B'): ...

# TC1
from __future__ import annotations

from typing import TYPE_CHECKING

if TYPE_CHECKING:
    from models.a import A

class B(Model):
    def bar(self, a: A): ...

# TC2
from typing import TYPE_CHECKING

if TYPE_CHECKING:
    from models.a import A

class B(Model):
    def bar(self, a: 'A'): ...

Running the plugin as a pre-commit hook

You can run this flake8 plugin as a pre-commit hook:

- repo: https://github.com/pycqa/flake8
  rev: 4.0.1
  hooks:
    - id: flake8
      additional_dependencies:
        - flake8-type-checking

Contributing

Please feel free to open an issue or a PR 👏