CODEWITHSUNDEEP
pythonpython-typingmypy
Jean-Paul Calderone
Jean-Paul Calderone

Reputation: 48335

What is the correct type annotation for a function that returns a Sequence of the same type as one of its inputs?

For the Python code

from typing import TypeVar, Sequence, Iterator

S = TypeVar("S", bound=Sequence)

def chunk(data: S) -> S:
    return data[:]

mypy 0.971 reports the error

simple_8.py:6:12: error: Incompatible return value type (got "Sequence[Any]", expected "S") [return-value]

Where is the type annotation error in this example and what is the correct, precise way to annotate this function?

Upvotes: 0

Views: 1499

Answers (1)

jonrsharpe
jonrsharpe

Reputation: 122154

If you look at the Typeshed definition of Sequence:

class Sequence(Collection[_T_co], Reversible[_T_co], Generic[_T_co]):
    @overload
    @abstractmethod
    def __getitem__(self, index: int) -> _T_co: ...
    @overload
    @abstractmethod
    def __getitem__(self, index: slice) -> Sequence[_T_co]: ...

you can see that seq[:] is guaranteed to return some Sequence with the same generic type _T_Co as the source, but that it's not necessarily the same type of sequence. Although the built-in sequence types generally have this behaviour, e.g. for list:

    @overload
    def __getitem__(self, __s: slice) -> list[_T]: ...
                                       # ^ concrete list not abstract Sequence

it's not a requirement of the interface.

As you don't supply the generic type to the Sequence in your TypeVar, it's the default Any, hence the error:

error: Incompatible return value type (got "Sequence[Any]", expected "S") [return-value]

The slice data[:] gives a Sequence[Any], which might be the same as S but is not required to be. So if you want to support any Sequence, the most precise you can be is:

from typing import Sequence, TypeVar

T = TypeVar("T")

def chunk(data: Sequence[T]) -> Sequence[T]:
    return data[:]

Playground

Alternatively, if you define a Protocol with the stricter typing that seq[:] must return the same type:

from typing import Protocol, TypeVar

T = TypeVar("T")

class StrictSlice(Protocol):
    def __getitem__(self: T, index: slice) -> T: ...

you can then use this as the bound type in S and get the type-preserving behaviour:

S = TypeVar("S", bound=StrictSlice)

def chunk(data: S) -> S:
    return data[:]

l: list[int] = chunk([1, 2, 3])
s: str = chunk("Hello, world!")

If you tried this with the Sequence[T], you'd get e.g. expression has type "Sequence[int]", variable has type "List[int]".

Playground

Upvotes: 2

Related Questions