指定在 Python 中调用抽象方法的包装函数的返回类型

Question

对于此示例，请考虑简化的场景，其中 Solver 将返回 Solution。

我们有Solutions：

class Solution(ABC):
    pass


class AnalyticalSolution(Solution):
    pass


class NumericalSolution(Solution):
    def get_mesh_size(self) -> float:
        return 0.12345

还有Solvers：

class Solver(ABC):
    def solve(self, task: int) -> Solution:
        # Do some pre-processing with task
        # ...
        return self._solve(task)

    @abstractmethod
    def _solve(self, task: int) -> Solution:
        pass


class NumericalSolver(Solver):
    def _solve(self, task: int) -> NumericalSolution:
        return NumericalSolution()


class AnalyticalSolver(Solver):
    def _solve(self, task: int) -> AnalyticalSolution:
        return AnalyticalSolution()

我遇到的问题是由包装器方法solve 的实现引起的，然后调用抽象方法_solve。 我经常遇到这样的情况，我想在 solve 方法中做一些预处理，这对所有求解器都是一样的，但是 _solve 的实际实现可能会有所不同。

如果我现在调用数值求解器并调用 get_mesh_size() 方法，Pylance（正确地）告诉我 Solution 对象没有 get_mesh_sizemember。

if __name__ == "__main__":
    solver = NumericalSolver()
    solution = solver.solve(1)
    print(solution.get_mesh_size())

我知道 Pylance 只看到 solve 的接口，这表明返回类型是一个 Solution 对象，不需要 get_mesh_size 方法。 我也知道这个例子在运行时有效。

我试着像这样使用TypeVar（实际上，因为 ChatGPT 建议这样做）：

class Solution(ABC):
    pass
T = TypeVar("T", bound=Solution)

然后重写Solver类：

class Solver(ABC):
    def solve(self, task: int) -> T:
        # Do some pre-processing with task
        # ...
        return self._solve(task)

    @abstractmethod
    def _solve(self, task: int) -> T:
        pass

但是 Pylance 现在告诉我TypeVar "T" appears only once in generic function signature。所以这不是解决方案。

我如何输入才能使用此示例？

Answer 1

您可以使用 Generic[T] 作为 Solver 的基础，然后按如下方式扩展它

from abc import ABC, abstractmethod
from typing import TypeVar, Generic


class Solution(ABC):
    pass


class AnalyticalSolution(Solution):
    pass


class NumericalSolution(Solution):
    def get_mesh_size(self) -> float:
        return 0.12345


SolutionGeneric = TypeVar("SolutionGeneric", bound=Solution)


class Solver(ABC, Generic[SolutionGeneric]):
    def solve(self, task: int) -> SolutionGeneric:
        # Do some pre-processing with task
        # ...
        return self._solve(task)

    @abstractmethod
    def _solve(self, task: int) -> SolutionGeneric:
        pass


class NumericalSolver(Solver[NumericalSolution]):
    def _solve(self, task: int) -> NumericalSolution:
        return NumericalSolution()


class AnalyticalSolver(Solver):
    def _solve(self, task: int) -> AnalyticalSolution:
        return AnalyticalSolution()


if __name__ == "__main__":
    solver = NumericalSolver()
    solution = solver.solve(1)
    print(solution.get_mesh_size())