I feel like I'm going to gain a reputation as the guy who only ever asks about type safety but... I've got pyright integrated into my environment now, which is working far better with Prefect than Mypy did. It just requires one small hack, and it makes me wonder why this isn't part of the library: I had to add

py.typed

on my own. I see that there is a

py.typed

on the master branch; was there a conscious decision made not to include this marker file in 2.0? If so, why? Without the marker file, I get:

from prefect.flows import flow

reveal_type(flow)  # Unknown

With the marker file, I get:

from prefect.flows import flow

reveal_type(flow)  # Overload[(__fn: (**P@flow) -> R@flow, /) -> Flow[P@flow, R@flow], (*, name: str = None, version: str = None, task_runner: BaseTaskRunner = ConcurrentTaskRunner, description: str = None, timeout_seconds: int | float = None, validate_parameters: bool = True) -> (((**P@flow) -> R@flow) -> Flow[P@flow, R@flow]), (__fn: Unknown | None = None, *, name: str = None, version: str = None, task_runner: BaseTaskRunner = ConcurrentTaskRunner, description: str = None, timeout_seconds: int | float = None, validate_parameters: bool = True) -> (Flow[P, R] | (((**P) -> R) -> Flow[P, R]))]

which is much better. For example, now when I decorate a function with

@task

and use it in an

@flow

, the return type of the task is known to have a

.result()

method. When it was just an Unknown, pyright would complain that I'm accessing a method that doesn't exist. For someone who enforces fully passing mypy (now pyright) checks on every pull request, this is kind of a necessity.

Oh we’re not omitting that on purpose

So is it fine for now for me to just

touch py.typed

Nice, is this typing not recognized by mypy then?

In mypy, presumably due to their lack of full support for [PEP612](https://peps.python.org/pep-0612/), when it tries to type the result of

Task.__call__

(i.e. type your function you’ve decorated with

@task

), it always chooses the [first overload](https://github.com/PrefectHQ/prefect/blob/orion/src/prefect/tasks.py#L231-L255),

Task[P, NoReturn]

, whose return type is

PrefectFuture[None, Sync]

. Basically, it assumes that every single user function decorated with

@task

has a return type of

None

, which causes

.result()

to always be None