Implement UpdateClass

[dnwpark]

Lets us write things like:

type AttrsAsSets[T] = UpdateClass[
    *[Member[GetName[m], set[GetType[m]]] for m in Iter[Attrs[T]]]
]

class A:
    def __init_subclass__[T](
        cls: type[T],
    ) -> AttrsAsSets[T]:
        super().__init_subclass__()

class B(A):
    n: int

B.n will be interpretted as set[int]

Note: There is an unfortunate part of python in general, where we must wrap any for/if statements in annotations in a generic type alias.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
python-typemap	Ready	Preview, Comment	Feb 6, 2026 2:45am

[msullivan]

Note: There is an unfortunate part of python in general, where we must wrap any for/if statements in annotations in a generic type alias.

I think this is fixable by deferring the signature inspection into the inside of the GenericCallable. That's what I was getting at earlier

Though there might still be snags going the other direction

[dnwpark]

@msullivan I'm not sure that's possible during runtime. It's possible at type check time at least, but if I try to look at the annotations of something like def f[T](self, x: T) -> set[T] if IsAssignable[T, int] else T: ..., it is already collapsed to T.

[msullivan]

@msullivan I'm not sure that's possible during runtime. It's possible at type check time at least, but if I try to look at the annotations of something like def f[T](self, x: T) -> set[T] if IsAssignable[T, int] else T: ..., it is already collapsed to T.

Right, we'll need to do it inside the GenericCallable lambda, and we'll need to do it by calling the __annotate__ function with the actual type parameters, not via inspect, I think.

Kind of a pain; not the highest priority I suppose

[msullivan]

• Did you test adding methods?
• Does format_class still work with __init_subclass__ or is that the end of the world

[msullivan]

Instead of grabbing __annotations__ directly, we should call __annotate__ with a special environment, like we do elsewhere. That way we can arrange to put the correct type variable in at the start, and have it evaluate correctly.

... maybe; there are still snags in other places, and I'm not 100% sure we can overcome them.

[msullivan]

That could be a follow-up though

[msullivan]

I don't totally follow this?

[dnwpark]

I was also confused. I have a followup that will make things more straightforward.

[dnwpark]

Added some tests for adding callables. The format class stuff worked, it was just kind of annoying to keep in sync as I was tweaking things. I've added them back in.