gh-148941: Skip generating __init__ when class already defines it

[nahcmon]

Bug

When @dataclass(init=True) (the default) is applied to a class that already defines __init__ in its own __dict__, the generated __init__ is always discarded by _set_new_attribute. However, _init_fn was still called unconditionally, and its field-ordering validation raised TypeError when inherited fields with defaults were followed by fields without defaults:

from dataclasses import dataclass

@dataclass
class Base:
    x: int = 0  # has a default

@dataclass
class Child(Base):
    y: int          # no default — would produce invalid generated signature
    def __init__(self, y):
        self.y = y
# TypeError: non-default argument 'y' follows default argument 'x'

The docs state: "If the class already defines __init__(), this parameter is ignored." The design table in dataclasses.py itself also documents this as the intended behaviour (the init=True / class-has-__init__ cell is intentionally blank/no-op).

Fix

Add and '__init__' not in cls.__dict__ to the guard before calling _init_fn in _process_class. This skips the generation (and its validation) entirely when the class's own __init__ would have been preserved anyway.

Tests

test_overwriting_init_with_field_ordering_conflict is added to TestInit in Lib/test/test_dataclasses/__init__.py:

• Fails before the fix: TypeError: non-default argument 'y' follows default argument 'x'
• Passes after the fix: Child(5).y == 5, class's own __init__ is used

All 282 dataclasses tests pass.

NEWS

Misc/NEWS.d/next/Library/2026-06-09-16-41-45.gh-issue-148941.aB3kLm.rst added.

CLA

⚠️ The GitHub account nahcmon may not have a signed PSF CLA on file. If the CLA bot flags this PR, the contributor will need to sign at https://www.python.org/psf/contrib/contrib-form/ before it can be merged.

• Issue: dataclass tries to generate __init__() even though the class already defines one #148941

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.
To continue using code reviews, you can upgrade your account or add credits to your account and enable them for code reviews in your settings.

[StanFromIreland]

Please revert all unrelated changes.

[DavidCEllis]

I don't think we should make this change. Certainly not in isolation like this.

Currently all of the main methods are only checked after generation when they are added to the class via the _set_new_attribute function. Adding this check early now with no other changes means checking __init__ twice, preventing its construction while not doing the same for any of the other methods.

Either we should keep the status quo of only checking after generation or we should check all of the methods on the class before generating.

I'm also not certain we shouldn't treat this as an error if the user hasn't declared init=False as I mentioned on the issue.