Variadic tuple type parameter

Question

I am trying to annotate a function which receives a tuple of types, and returns a value of one of the provided types.

For example:

@overload
def func[T1](types: tuple[type[T1]]) -> T1: ...
@overload
def func[T1, T2](types: tuple[type[T1], type[T2]]) -> T1 | T2: ...
@overload
def func[T1, T2, T3](types: tuple[type[T1], type[T2], type[T3]]) -> T1 | T2 | T3: ...

Conceptually analogous to the following invalid code:

# invalid
def func[*Ts](types: tuple[type[*Ts]]) -> Union[*Ts]: ...

Is something like this supported?

Not directly relevant to the problem, but should it really be tuple and not collections.abc.Collection? There's rarely a reason not to allow any kind of collection. — Barmar
– Barmar, Commented Oct 2 at 20:58
@Barmar tuple captures each distinct member in the variadic pack, which is the point of the question, whereas most other kinds of type annotation don't. — alecov
– alecov, Commented Oct 3 at 5:28
Right. After I posted that I realized that Collection is for collections of homogeneous types. — Barmar
– Barmar, Commented Oct 3 at 14:18

InSync · Accepted Answer · 2025-11-06 05:35:16Z

2

Depending on what type checker you use, something like this might work:

_{(playground: Mypy, Pyright, Pyrefly, ty)}

# Either
def func[T](types: tuple[type[T], ...]) -> T: ...
# or
def func[T](types: Iterable[type[T]]) -> T: ...

reveal_type(f((int,)))      # mypy, pyright => int
reveal_type(f((int, str)))  # mypy          => object
                            # pyright       => int | str

# Pyrefly errors out; ty infers Unknown.

This works due to a property of type[]: It distributes over unions. That is, type[A] | type[B] is equivalent to type[A | B].

A type checker might infer the type tuple[type[int], type[str]] for (int, str) and then cast it to tuple[type[int | str], ...] or Iterable[type[int | str]], which leads to T being resolved as int | str. This behaviour is not guaranteed across type checkers, however, as seen above.

edited Nov 6 at 5:35

answered Oct 3 at 5:08

InSync

12.2k5 gold badges22 silver badges60 bronze badges

Sign up to request clarification or add additional context in comments.

1 Comment

alecov Oct 3 at 5:26

This looks very promising, thanks for sharing. Getting object from mypy is a sad situation but better than nothing.

ti7 · Accepted Answer · 2025-10-02 19:07:34Z

-1

I'd be most tempted to use the ellipsis and avoid bothering to associate the length of the tuple with the result length

def func(tuples: Tuple[T, ...]) -> Tuple[T, ...]:

See also my Answer here Pydantic issue for tuple length

answered Oct 2 at 19:07

ti7

19.8k8 gold badges50 silver badges87 bronze badges

4 Comments

Barmar Oct 2 at 19:20

The return value isn't a tuple, it's an object of one of the types in the input tuple.

Barmar Oct 2 at 19:20

An example would be something like random.choice()

ti7 Oct 2 at 20:52

sheesh, indeed - I have trouble reading when my @overload allergy flares up..

Barmar Oct 2 at 20:54

I also missed that the elements are type[Tn] rather than just Tn on my first reading. So it's a function that takes a collection of types, and returns an instance of one of those types.

Collectives™ on Stack Overflow

Variadic tuple type parameter

2 Answers 2

1 Comment

4 Comments

Your Answer

Linked

Hot Network Questions

Collectives™ on Stack Overflow

2 Answers 2

1 Comment

4 Comments

Your Answer

Sign up or log in

Post as a guest

Linked

Related