Recognize many=True serializer constructors as ListSerializer

[federicobond]

I have made things!

At runtime, SomeSerializer(many=True, ...) is swapped for a ListSerializer via many_init, but the stubs type it as the original serializer, so valid calls are rejected and downstream attributes lie about their types.

This adds a get_function_signature_hook on every registered serializer class that, when many=True is literally passed:

• accepts the ListSerializer-only kwargs min_length and max_length;
• rewrites the constructor's return type to ListSerializer[Collection[_IN]], extracting _IN by mapping the child serializer's instance to BaseSerializer in its MRO.

[intgr]

Thanks! Sorry for the long delay in responding.

This is a known issue, also documented at #827

Ideally, it would also be possible to express this with an overload of __new__() -> ListSerializer return type.

Problem is, right now mypy constrains the __new__ return type to be a subclass of Self. But there's an open issue to relax this constraint on mypy, with lots of users requesting it. I suspect mypy upstream would accept to change it:

• Honor return type of __new__ even if not a subclass python/mypy#15182

I'd prefer to avoid maintaining a mypy plugin if possible. Would you be up to implementing this in mypy instead?

[federicobond]

Will take a look and see what I can do!

[intgr]

Great, thanks. If you see any reason why it wouldn't work then I can reconsider

[intgr]

@federicobond Not sure if you're aware, the blocker was solved in mypy: python/mypy#21441

[federicobond]

Amazing. I was not aware!