Duck typing with python 3.5 style type-annotations

Question

Supposing I have a function with a signature like:

def foo(self, name:str, stream):
    pass

I want to add an annotation to the "stream" argument so which means "you can have any object x as long as x.readline()->str".

So that means I could use any python file object as an argument here (since it has a readline method), but I could also provide an object which implements nothing but readline and it would be perfectly acceptable.

How could I rewrite this function definition so that I can annotate the 2nd argument?

schot · Accepted Answer · 2019-10-17T15:20:26.250

As ivanl notes, PEP 544 adds Protocols to support 'static duck typing'. This PEP was accepted recently and was added in Python 3.8. You can also try Protocols in Python 3.6 and 3.7 with Mypy using the typing-extensions package.

In your case, you would define a very simple Protocol SupportsReadline with a single method and use this in the annotation of your function arguments:

# Python 3.8+, for 3.6 & 3.7 replace 'typing' with 'typing_extensions'.
from typing import Protocol

class SupportsReadline(Protocol):
    def readline(self) -> str:
        ...

def func(name: str, stream: SupportsReadline) -> None:
    pass

Now any object with a readline method with a compatible signature is an implicit subtype of SupportsReadline and satisfies the annotation of your function argument. Note that LineRepeater does not inherit explicitly from SupportsReadline:

class LineRepeater:
    def readline(self) -> str:
        return "Hello again!"

func("a", LineRepeater())  # OK

The same holds for other objects if the method signature matches exactly:

from io import BytesIO, StringIO

func("a", StringIO())  # OK
func("a", open("foo.txt"))  # OK
func("a", BytesIO())  # ERROR (return type is bytes instead of str)
func("a", [])  # ERROR
func("a", 1)  # ERROR
func("a", object())  # ERROR

score 10 · Answer 2 · answered May 24 '17 at 14:38

Structural subtyping (static duck typing) is proposed by PEP 544 https://www.python.org/dev/peps/pep-0544/. If/when it is accepted you will not need an explicit subclassing, you will be able to simply define your own protocols that will be understood by static type checkers.

score 4 · Answer 3 · answered May 24 '17 at 12:31

This solution is not equivalent to what exactly you're looking for:

you can have any object x as long as x.readline() -> str

Instead we are defining a custom abstract base class that expects readline abstract method to be defined by its child classes. Hence instead of any random object it would only accept instances of this new abstract base class, making it more explicit.

from abc import ABC, abstractmethod

class FileObject(ABC):
    @abstractmethod
    def readline(self):
        raise NotImplementedError()

Now we are going to define a custom type that can work with Python's file objects and instances of FileObject:

from typing import IO, TypeVar


StreamType = TypeVar('StreamType', IO, FileObject)
def func(name: str, stream: StreamType) -> None:
    pass

Now let's test it using mypy:

from io import StringIO, BytesIO


class X(FileObject):
    def readline(self):
        pass


func('a', StringIO())  # passed
func('a', BytesIO())  # passed
func('a', open('foo.txt'))  # passed
func('a', X())  # passed
func('a', object())  # failed
func('a', [])  # failed
func('a', 1)  # failed

Output:

$ mypy so.py
so.py:33: error: Type argument 1 of "func" has incompatible value "object"
so.py:34: error: Type argument 1 of "func" has incompatible value List[None]
so.py:35: error: Type argument 1 of "func" has incompatible value "int"

You may want to implement [`__subclasshook__`](https://docs.python.org/3/library/abc.html#abc.ABCMeta.__subclasshook__) so `isinstance` checks pass at runtime. — user2357112, May 24 '17 at 14:42

Duck typing with python 3.5 style type-annotations

3 Answers3