Skip to main content

A plugin for flake8 to enable linting .pyi stub files.

Project description

flake8-pyi

A plugin for Flake8 that provides specializations for type hinting stub files, especially interesting for linting typeshed.

Refer to this documentation for more details on stub files.

Functionality

  1. Adds the .pyi extension to the default value of the --filename command-line argument to Flake8. This means stubs are linted by default with this plugin enabled, without needing to explicitly list every file.

  2. Modifies PyFlakes runs for .pyi files to defer checking type annotation expressions after the entire file has been read. This enables support for first-class forward references that stub files use.

  3. Provides a number of .pyi-specific warnings that enforce typeshed's style guide.

Note: Be careful when using this plugin in the same environment as other flake8 plugins, as they might generate errors that are inappropriate for .pyi files (e.g., about missing docstrings). We recommend running flake8-pyi in a dedicated environment in your CI.

List of warnings

This plugin reserves codes starting with Y0. The following warnings are currently emitted:

Code Description
Y001 Names of TypeVars, ParamSpecs and TypeVarTuples in stubs should usually start with _. This makes sure you don't accidentally expose names internal to the stub.
Y002 If test must be a simple comparison against sys.platform or sys.version_info. Stub files support simple conditionals to indicate differences between Python versions or platforms, but type checkers only understand a limited subset of Python syntax, and this warning triggers on conditionals that type checkers will probably not understand.
Y003 Unrecognized sys.version_info check. Similar, but triggers on some comparisons involving version checks.
Y004 Version comparison must use only major and minor version. Type checkers like mypy don't know about patch versions of Python (e.g. 3.4.3 versus 3.4.4), only major and minor versions (3.3 versus 3.4). Therefore, version checks in stubs should only use the major and minor versions. If new functionality was introduced in a patch version, pretend that it was there all along.
Y005 Version comparison must be against a length-n tuple.
Y006 Use only < and >= for version comparisons. Comparisons involving > and <= may produce unintuitive results when tools do use the full sys.version_info tuple.
Y007 Unrecognized sys.platform check. Platform checks should be simple string comparisons.
Y008 Unrecognized platform. To prevent you from typos, we warn if you use a platform name outside a small set of known platforms (e.g. "linux" and "win32").
Y009 Empty body should contain ..., not pass. This is just a stylistic choice, but it's the one typeshed made.
Y010 Function body must contain only .... Stub files should not contain code, so function bodies should be empty.
Y011 Only simple default values (int, float, complex, bytes, str, bool, None or ...) are allowed for typed function arguments. Type checkers ignore the default value, so the default value is not useful information for type-checking, but it may be useful information for other users of stubs such as IDEs. If you're writing a stub for a function that has a more complex default value, use ... instead of trying to reproduce the runtime default exactly in the stub.
Y012 Class body must not contain pass.
Y013 Non-empty class body must not contain ....
Y014 Only simple default values are allowed for any function arguments. A stronger version of Y011 that includes arguments without type annotations.
Y015 Only simple default values are allowed for assignments. Similar to Y011, but for assignments rather than parameter annotations.
Y016 Unions shouldn't contain duplicates, e.g. str | str is not allowed.
Y017 Stubs should not contain assignments with multiple targets or non-name targets.
Y018 A private TypeVar should be used at least once in the file in which it is defined.
Y019 Certain kinds of methods should use _typeshed.Self instead of defining custom TypeVars for their return annotation. This check currently applies for instance methods that return self, class methods that return an instance of cls, and __new__ methods.
Y020 Quoted annotations should never be used in stubs.
Y021 Docstrings should not be included in stubs.
Y022 The typing and typing_extensions modules include various aliases to stdlib objects. Use these as little as possible (e.g. prefer builtins.list over typing.List, collections.Counter over typing.Counter, etc.).
Y023 Where there is no detriment to backwards compatibility, import objects such as ClassVar and NoReturn from typing rather than typing_extensions.
Y024 Use typing.NamedTuple instead of collections.namedtuple, as it allows for more precise type inference.
Y025 Always alias collections.abc.Set when importing it, so as to avoid confusion with builtins.set.
Y026 Type aliases should be explicitly demarcated with typing.TypeAlias.
Y028 Always use class-based syntax for typing.NamedTuple, instead of assignment-based syntax.
Y029 It is almost always redundant to define __str__ or __repr__ in a stub file, as the signatures are almost always identical to object.__str__ and object.__repr__.
Y030 Union expressions should never have more than one Literal member, as Literal[1] | Literal[2] is semantically identical to Literal[1, 2].
Y031 TypedDicts should use class-based syntax instead of assignment-based syntax wherever possible. (In situations where this is not possible, such as if a field is a Python keyword or an invalid identifier, this error will not be raised.)
Y032 The second argument of an __eq__ or __ne__ method should usually be annotated with object rather than Any.
Y033 Do not use type comments (e.g. x = ... # type: int) in stubs. Always use annotations instead (e.g. x: int).
Y034 Y034 detects common errors where certain methods are annotated as having a fixed return type, despite returning self at runtime. Such methods should be annotated with _typeshed.Self. This check looks for:

  1.  Any in-place BinOp dunder methods (__iadd__, __ior__, etc.) that do not return Self.
  2.  __new__, __enter__ and __aenter__ methods that return the class's name unparameterised.
  3.  __iter__ methods that return Iterator, even if the class inherits directly from Iterator.
  4.  __aiter__ methods that return AsyncIterator, even if the class inherits directly from AsyncIterator.

This check excludes methods decorated with @overload or @abstractmethod.
Y035 __all__ and __match_args__ in a stub file should always have values, as these special variables in a .pyi file have identical semantics to __all__ and __match_args__ in a .py file. E.g. write __all__ = ["foo", "bar"] instead of __all__: list[str].
Y036 Y036 detects common errors in __exit__ and __aexit__ methods. For example, the first argument in an __exit__ method should either be annotated with object or type[BaseException] | None.
Y037 Use PEP 604 syntax instead of typing.Union and typing.Optional. E.g. use str | int instead of Union[str, int], and use str | None instead of Optional[str].
Y038 Use from collections.abc import Set as AbstractSet instead of from typing import AbstractSet.
Y039 Use str instead of typing.Text.
Y040 Never explicitly inherit from object, as all classes implicitly inherit from object in Python 3.
Y041 Y041 detects redundant numeric unions in the context of parameter annotations. For example, PEP 484 specifies that type checkers should allow int objects to be passed to a function, even if the function states that it accepts a float. As such, int is redundant in the union int | float in the context of a parameter annotation. In the same way, int is sometimes redundant in the union int | complex, and float is sometimes redundant in the union float | complex.
Y042 Type alias names should use CamelCase rather than snake_case
Y043 Do not use names ending in "T" for private type aliases. (The "T" suffix implies that an object is a TypeVar.)
Y044 from __future__ import annotations has no effect in stub files, since type checkers automatically treat stubs as having those semantics.
Y045 __iter__ methods should never return Iterable[T], as they should always return some kind of iterator.
Y046 A private Protocol should be used at least once in the file in which it is defined.
Y047 A private TypeAlias should be used at least once in the file in which it is defined.
Y048 Function bodies should contain exactly one statement. (Note that if a function body includes a docstring, the docstring counts as a "statement".)
Y049 A private TypedDict should be used at least once in the file in which it is defined.
Y050 Prefer typing_extensions.Never over typing.NoReturn for argument annotations. This is a purely stylistic choice in the name of readability.
Y051 Y051 detect redundant unions between Literal types and builtin supertypes. For example, Literal[5] is redundant in the union int | Literal[5], and Literal[True] is redundant in the union Literal[True] | bool.

Many error codes enforce modern conventions, and some cannot yet be used in all cases:

  • Y037 (enforcing PEP 604 syntax everywhere) is not yet fully compatible with the mypy type checker, which has a few bugs regarding PEP 604 type aliases.

Note that several error codes recommend using types from typing_extensions or _typeshed. Strictly speaking, these packages are not part of the standard library. However, these packages are included in typeshed's stdlib/ directory, meaning that type checkers believe them to be part of the standard library even if this does not reflect the reality at runtime. As such, since stubs are never executed at runtime, types from typing_extensions and _typeshed can be used freely in a stubs package, even if the package does not have an explicit dependency on either typing_extensions or typeshed.

Flake8-pyi's checks may produce false positives on stubs that aim to support Python 2.

License

MIT

Authors

Originally created by Łukasz Langa and now maintained by Jelle Zijlstra, Alex Waygood, Sebastian Rittau, Akuli, and Shantanu.

See also

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

flake8_pyi-23.1.0.tar.gz (44.4 kB view hashes)

Uploaded Source

Built Distribution

flake8_pyi-23.1.0-py3-none-any.whl (24.5 kB view hashes)

Uploaded Python 3

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page