I'm trying to write an example where I'd like to use a frozen dataclass instance during type checking and swap it out with a normal dataclass to avoid paying the instantiation cost of frozen dataclasses. The goal is to ensure that the instance is immutable with the type checker and use a regular dataclass during runtime. Here's the snippet: from dataclasses import dataclass from typing import TYPE_CHECKING from functools import partial if TYPE_CHECKING: frozen = partial(dataclass, frozen=True) else: frozen = dataclass @frozen class Foo: x: int y: int foo = Foo(1, 2) # mypy complains about the number of arguments foo.x = 3 # instead, mypy should complain here This works as expected during runtime, but running mypy raises this error. Pyright gives me the same error as well: foo.py:49: error: Too many arguments for "Foo" [call-arg] In this snippet, the type checker can catch the mutation error: @dataclass(frozen=True) class Foo: x: int y: int foo = Foo(1, 2) foo.x = 3 # mypy correctly catches the error here So, I'm guessing that the type checker doesn't like when I'm aliasing frozen = dataclass or frozen = partial(...). How do I annotate this properly so that the type checker understands that it's a dataclass instance and doesn't complain about mismatched argument count? P.S: This is just an exercise. I know turning on dataclass(frozen=True) is way easier, and I shouldn't care about performance in such cases. I was inspired to try this after reading a blog post by Tin Tvrtković on making attr class instances frozen at compile time. Continue reading...
