How to accept Annotated types as well as normal types?

[superlopuh]

Here's a minimal snippet:

from typing import Annotated, TypeAlias


class MyClass: ...


def my_function(arg: type[MyClass]) -> None: ...


MyTypealias: TypeAlias = Annotated[MyClass, 1]

# works
my_function(MyClass)
# doesn't work
my_function(MyTypealias)

The error:

.../hello.py:15:13 - error: Argument of type "MyTypealias" cannot be assigned to parameter "arg" of type "type[MyClass]" in function "my_function"
    Type "type[Annotated]" is incompatible with type "type[MyClass]" (reportArgumentType)

I'm surprised, I thought that Annotated was supposed to be transparent wrt the type system. The issue is that type[Annotated] seems to not be something I can spell, if I add it to the list of types accepted by my_function, I get two errors, the existing one, as well as Expected type arguments for generic class "Annotated".

Is there any way to make this work?

[erictraut]

Annotated is intended to be used in type expressions (i.e. type annotations). You're using it here in a value expression. See this issue for more details.

What are you doing in my_function with the value you pass to it? The value MyTypealias is not a valid class object, so my_function is not going to be able to use it as a normal type. For example, if there is a class variable defined on MyClass, you won't be able to access it using arg.my_class_var.

[superlopuh]

Ah, I see, thank you.

We use this as part of a declarative constraint system. my_function is a dataclass field descriptor that leverages the type system for static/IDE checking, and we use Annotated to add additional constraints that the type system doesn't support, but we can in our custom verify function. This pattern is currently used to enforce a same-type constraint when multiple fields must have the same type. I'll try to find a different API for this.

[erictraut]

You may be interested in the draft PEP discussed in this thread.