Advanced dataclass support

[sharkdp]

We already have initial support for dataclasses, but some more advanced features are still missing. The following list is most probably not complete:

• Make sure that dataclass can be used as a non-decorator: class C: …; C = dataclass(C). See empty test here.
• Emit an error if >=2 variables in a dataclass class body are annotated with KW_ONLY (failing test here)
• Support for dataclasses.KW_ONLY
• Support for Final[…] fields and ClassVar[Final[…]] fields.
• Improve support for frozen dataclasses to handle subclasses of frozen dataclasses also
• Support for dataclasses.InitVar [ty] Support dataclasses.InitVar ruff#19527
• Support for dataclass fields
• Add support for other synthesized functions / arguments
  • unsafe_hash
  • match_args
  • kw_only
  • slots: [ty] initial support for slots=True in dataclasses ruff#20278
  • weakref_slot
  • The synthesized __replace__ method on Python 3.13+ [ty] synthesize __replace__ for dataclasses (>=3.13) ruff#19545
• Emit diagnostic when defining a field without a default after a field with a default, see existing TODO: [ty] diagnostic for dataclass field order ruff#19825
• Emit diagnostic when setting order=True on a dataclass that has a custom __lt__ (or similar), see existing TODO
• Verify signature of methods such as __post_init__, see this comment
• Emit diagnostic if frozen=True but the class has a custom __setattr__ or __delattr__ method in the class body (this raises TypeError at runtime)
• Add support for dataclasses.field(kw_only=True)

[AlexWaygood]

We'll eventually want to support a generalised feature for slotted classes, where something like this is reported as an error:

class Foo:
    __slots__ = ("a,")
    b: int

Foo cannot declare b as an instance variable here, because Foo is a slotted class and b is not present in Foo.__slots__; an assignment such as Foo().b = 42 will therefore fail. When implementing that feature, we'll want to make sure that we understand slotted dataclasses (both those that use slots=True and those that manually define __slots__ in the class namespace) consistently with how we understand other slotted classes.

[thejchap]

@sharkdp once i wrap astral-sh/ruff#17697 i'd be interested in taking a look at this, in particular:

Add support for frozen dataclasses

would it make sense to spin up individual issues for each of these features?

[sharkdp]

I think it's a good idea to attempt to solve these items separately. I think it's fine to keep just this one ticket but implement it in several PRs. However, if you think that there is significant discussion needed for a sub-item, please feel free to create additional issues and we can link them here.