Allow omitting initializer for Final attributes in dataclasses

[ilevkivskyi]

Dataclasses typically don't have __init__, so this is rejected, but is probably valid:

@dataclass
class C:
    x: int
    y: Final[int]

A possible solution is that plugin should set the final_set_in_init flag manually for the corresponding field Vars.

[jcrist]

I just ran into this. pyright allows these type annotations as is (Final indicates an instance variable), and discussion in python/cpython#89547 seems to indicate this is what'd be most ergonomic to support. It also matches how dataclasses itself handles these annotations.

For now, a workaround is to add # type: ignore on the annotation line. The dataclass definition still seems to be processed correctly, this just skips the error about a missing initializer.

import dataclasses
from typing import Final


@dataclasses.dataclass
class Test:
    # The `type: ignore` is needed here to avoid an error due to a missing
    # initializer
    x: Final[int]  # type: ignore


t = Test(1)  # Signature still inferred properly
reveal_type(t.x)
t2 = Test("bad")  # Which means mypy marks this as an error

$ mypy ex.py 
ex.py:13: note: Revealed type is "builtins.int"
ex.py:14: error: Argument 1 to "Test" has incompatible type "str"; expected "int"  [arg-type]
Found 1 error in 1 file (checked 1 source file)

For my own use cases, it'd be nice if the fix here also applied to the recent dataclass_tranform support, so this fix could work for other dataclass-like things as well.

[tmke8]

The typing spec now specifies that this should be allowed: python/typing#1669