如何让可变变量在静态类型检查中被视为volatile？

问题场景 #

现有如下代码，Worker类的state变量在check_twice方法中被两次检查，两次检查之间会被work方法修改：

import enum


class State(enum.Enum):
    INIT = 0
    DONE = 1


class Worker:
    def __init__(self) -> None:
        self.state = State.INIT

    def work(self):
        self.state = State.DONE

    def check_twice(self) -> None:
        if self.state is State.DONE:
            print("DONE at first try")
            return

        self.work()

        if self.state is State.DONE:
            # mypy 认为此分支不可达，不会检查下面的类型错误
            x: int = "x"
            print("DONE at second try")
            return

        print("not DONE after second try")


if __name__ == "__main__":
    w = Worker()
    w.check_twice()

运行代码时，实际输出为：

$ python3 volatile.py 
DONE at second try

但使用mypy 1.5.0并添加--strict-equality选项检查时，会报错：

$ mypy --strict-equality volatile.py 
volatile.py:23: error: Non-overlapping identity check (left operand type: "Literal[State.INIT]", right operand type: "Literal[State.DONE]")  [comparison-overlap]

原因是mypy通过类型收窄逻辑判定：若首次检查state为DONE则方法直接返回，剩余流程中state只能是INIT，因此第二个if self.state is State.DONE分支被标记为不可达，甚至不会检查分支内明显的类型错误（比如x: int = "x"）。

解决方法 #

1. 显式重置变量类型提示 #

在调用work()之后，给self.state添加类型注释，强制mypy重新识别变量的完整类型：

def check_twice(self) -> None:
    if self.state is State.DONE:
        print("DONE at first try")
        return

    self.work()
    self.state: State  # 重置类型，告知mypy变量可能已被修改

    if self.state is State.DONE:
        x: int = "x"  # 此时mypy会检查到此处的类型错误
        print("DONE at second try")
        return

    print("not DONE after second try")

这样mypy会放弃之前的类型收窄结果，重新将state视为State枚举类型，从而检查后续分支的代码。

2. 使用typing.cast强制转换类型 #

通过cast将self.state强制转换为State类型，打破之前的类型收窄：

from typing import cast

# ...

def check_twice(self) -> None:
    if self.state is State.DONE:
        print("DONE at first try")
        return

    self.work()
    self.state = cast(State, self.state)  # 强制重置类型

    if self.state is State.DONE:
        x: int = "x"  # mypy会检测到类型不匹配
        print("DONE at second try")
        return

    print("not DONE after second try")

此方法和第一种逻辑类似，都是让mypy重新识别变量的类型范围。

3. 将变量封装为属性并添加类型标注 #

如果希望从根源上让类型检查器意识到变量可能被外部修改，可以将state封装为@property，并显式标注其类型为State：

class Worker:
    def __init__(self) -> None:
        self._state = State.INIT

    @property
    def state(self) -> State:
        return self._state

    def work(self):
        self._state = State.DONE

    def check_twice(self) -> None:
        if self.state is State.DONE:
            print("DONE at first try")
            return

        self.work()

        if self.state is State.DONE:
            x: int = "x"  # mypy会正常检查此处错误
            print("DONE at second try")
            return

        print("not DONE after second try")

使用属性后，每次访问self.state都会被视为调用属性方法，mypy不会对其进行类型收窄，自然会考虑到变量可能被修改的情况。

内容的提问来源于stack exchange，提问作者mkrieger1