Remove supertype in Scope?

[jl-wynen]

Can we remove the supertype in scope definitions? It does not seem to serve any purpose any more. Or am I missing something?

For example, when I remove SUPER from Scope:

class Scope(Generic[PARAM]):
    def __init_subclass__(cls, **kwargs: Any) -> None:
        return super().__init_subclass__(**kwargs)

    def __new__(cls, x: T) -> T:  # type: ignore[misc]
        return x

this runs fine and passes type checks:

from typing import NewType, TypeVar
import sciline as sl

A = NewType('A', int)
B = NewType('B', int)
T = TypeVar('T', A, B)

class X(sl.Scope[T], int): ...
class Y(sl.Scope[T], int): ...

def foo(x: X[T]) -> Y[T]:
    return Y[T](x + 1)

pl = sl.Pipeline([foo], params={X[A]: 3, X[B]: 4})
print(pl.visualize())
print(pl.compute(X[A]))

Is the supertype simply a leftover from an earlier version of Sciline or do we still need it?

Of course, doping this would require a lot of changes in downstream code. But I think the reduced duplication would be worth it.

[SimonHeybrock]

I don't think it is (or ever was) for type checks but for IDEs.

[jl-wynen]

Why do IDEs need it? Mine doesn't care from what I can tell.

[SimonHeybrock]

Hmm, but there was a reason for it.

[jl-wynen]

Maybe that reason is no longer valid. Do you have any old documents or tests to check?

[SimonHeybrock]

Maybe that reason is no longer valid. Do you have any old documents or tests to check?

#22 claims that more issues can be detected by mypy?

[jl-wynen]

True. Here is an example:

from typing import NewType, TypeVar
import sciline as sl

T1 = NewType('T1', int)
T2 = NewType('T2', int)
T = TypeVar('T', T1, T2)

class Int(sl.Scope[T, int], int): ...
class Str(sl.Scope[T, str], str): ...

def bad_return(i: Int[T]) -> Str[T]:
    return Str[T](i)

Mypy correctly detects that the argument to Str[T] in the return statement has an incorrect type. If we remove the superclass param, it no longer detects this.

But it seems to me like there has to be a better solution.