Search code examples
pythonloopspython-typingpython-itertools

Python: why use overloading instead of *args in a function (especially when the type of arguments do not affect how the function runs)


Edited: I was looking at the type annotations of the built-in zip() function.

I understand that overload (in the context of type checking), can modify the behaviour of a function depending on the type of parameters it's given.

    @overload
    def zip(__iter1: Iterable[_T1]) -> List[Tuple[_T1]]: ...
    @overload
    def zip(__iter1: Iterable[_T1],
            __iter2: Iterable[_T2]) -> List[Tuple[_T1, _T2]]: ...
    @overload
    def zip(__iter1: Iterable[_T1], __iter2: Iterable[_T2],
            __iter3: Iterable[_T3]) -> List[Tuple[_T1, _T2, _T3]]: ...
    @overload
    def zip(__iter1: Iterable[_T1], __iter2: Iterable[_T2], __iter3: Iterable[_T3],
            __iter4: Iterable[_T4]) -> List[Tuple[_T1, _T2, _T3, _T4]]: ...
    @overload
    def zip(__iter1: Iterable[_T1], __iter2: Iterable[_T2], __iter3: Iterable[_T3],
            __iter4: Iterable[_T4], __iter5: Iterable[_T5]) -> List[Tuple[_T1, _T2, _T3, _T4, _T5]]: ...
    @overload
    def zip(__iter1: Iterable[Any], __iter2: Iterable[Any], __iter3: Iterable[Any],
            __iter4: Iterable[Any], __iter5: Iterable[Any], __iter6: Iterable[Any],
            *iterables: Iterable[Any]) -> List[Tuple[Any, ...]]: ...
  1. In this example, however, the type is the same (Any). So what purpose does overload serve?

  2. Why not just use the last function, which takes in an arbitrary number of parameters. Why create the first five which essentially say: if there's one argument do this, if there's two do that, if there's 3 ... This seems to violate the DRY principle.

    @overload
    def zip(__iter1: Iterable[Any], __iter2: Iterable[Any], __iter3: Iterable[Any],
            __iter4: Iterable[Any], __iter5: Iterable[Any], __iter6: Iterable[Any],
            *iterables: Iterable[Any]) -> List[Tuple[Any, ...]]: ...

Solution

  • The overload decorator is for static type analysis, not for implementation. In fact, the code shown in the question are just the type annotations - zip is a builtin, it is not implemented in Python.

    The purpose of the various overloads is to preserve the number and type of arguments. For example, it states that a zip over three iterables yields tuples with three elements matching the type of the iterable elements. The final overload with variadic *args and Any is merely a catch-all for unspecified cases.