Fix @dataclass to catch default / non-default fields order

[sobolevn]

Closes#12137

[sobolevn]

So, here's what happens with overriding kw-only with a regular field.

fromdataclassesimportdataclass, field@dataclassclassSome:
x: int=field(kw_only=True)
@dataclassclassOther(Some):
x: intprint(Other(1)) # Other(x=1)

Two fields (there was a difference in the original bug report):

fromdataclassesimportdataclass, field@dataclassclassSome:
x: int=field(kw_only=True)
@dataclassclassOther(Some):
x: inty: intprint(Other(1, 2))

Mypy is also happy with this.
So, I've just added a test case to cement this behavior (which I believe is the correct one).

Funny thing

But, while testing this, I found a different problem:

fromdataclassesimportdataclass, field, KW_ONLY@dataclassclassBase:
x: inty: int=field(kw_only=True)
_: KW_ONLYz: int=0@dataclassclassOther(Base): # E: Attributes without a default cannot follow attributes with onex: inty: intz: int

I will open a new issue for it and address in a separate PR (if this is a problem at all, some more analysis is needed).

[github-actions]

According to mypy_primer, this change has no effect on the checked open source code. 🤖🎉

[sobolevn]

This is a new regression: #12173
It is not happening on master

[JelleZijlstra]

Interestingly the runtime accepts this one. It's pretty weird, though, and I'm OK with mypy giving an error for it.

[sobolevn]

This still has a couple of regressions 😢
I will fix them later.