Feature request: Add support for dataclasses

[Dementiy]

At the moment numpydoc will ignore "Attributes" section for dataclasses (or maybe I don't quite understand how to document dataclasses properly):

@dataclass
class Klass:
    """
    Klass description.

    Attributes
    ----------
    attr: int
        Some attribute description.
    """

    attr: int

It would be great to have such a feature.

[mattgebert]

Whilst not directly related, I also notice that while the ability to track decorator functions has been added a while ago, specific support for the property decorator and it's subsequent getter, setter and deleter sub-functions are also not supported by numpydoc.

For example, after defining @property def myprop, the myprop.setter function decorator will still require a docstring for the setter function according to numpydoc, even though the docstring of the property class really ought to specify the whole property in one go. Usually there's only one reference/docstring point, so having a single docstring becomes a necessity.

Given these are standard library modules, can we implement module specific support?

I would be happy to give this a crack for both property attributes and dataclasses.

[mattgebert]

Also noticed this pull request related to long running issue #196 which touches on the use of type hinting to avoid requiring numpydoc to specify the type hinting with redundancy. This seems relevant for the dataclass decorator, where you are going to pull the implied constructor from the type hinting attributes.

[RastislavTuranyi]

Hi, i only noticed this from the mentioned PR, but i wanted ask what the problem is? (maybe i might be able to help)

Running your example with numpydoc==1.8.0 and sphinx==8.1.3 (and rtd theme) gives me:

which is what i expect when documenting both dataclasses and normal classes, e.g.:

class Klass2:
    """
    Klass description.

    Attributes
    ----------
    attr: int
        Some attribute description.
    """

    attr: int

gives