[BUG] Using a dataclass with positional and non-positional args causes a runtime exception

[Mnkras]

Describe the bug
Execute the code below after cythonizing it and it will generate a runtime exception.
Vanilla Python has no issues

To Reproduce
Code to reproduce the behaviour:

from dataclasses import dataclass
from datetime import datetime
from typing import Optional


@dataclass
class PipelineTiming:
    initiated: datetime
    started: Optional[datetime] = None
    stopped: Optional[datetime] = None


def testing():
    return PipelineTiming(datetime.utcnow())

print(testing())

>>> import test
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "test.py", line 16, in init test
    print(testing())
  File "test.py", line 14, in test.testing
    return PipelineTiming(datetime.utcnow())
TypeError: __init__() takes 1 positional argument but 2 were given

Expected behavior
I expect the code to run, and not crash at runtime

PipelineTiming(initiated=datetime.datetime(2022, 7, 28, 18, 14, 8, 100560), started=None, stopped=None)

Environment (please complete the following information):

• OS: Linux and macOS
• Python version: 3.8.4
• Cython version: 0.29.31

Additional context
There seems to be other weirdness with dataclasses as well, I can file issues for those as well.

[da-woods]

Cython 0.29.x doesn't support class annotations (and won't support class annotations). Therefore dataclasses will not usefully work with it.

They will work with the Cython 3 alpha branch.

Duplicate of #2552

[da-woods]

It just doesn't generate the __annotations__ attribute so the dataclass decorator has nothing to work with. There's some useful workarounds in the linked issue, but the best advice is to use the Cython 3 alpha branch.

[Mnkras]

Is 3 considered stable enough? I would love to use these python features, but not at the risk of destabilizing our application

[da-woods]

Considered by whom? 0.29.31 isn't a great story of not breaking things right now...

It's reasonably stable I think, but it obviously depends on what the risk is. The current 3.0a10 release has been out for a while without major issue. The next release may well break some typing related things (because it's starting to try to use those intelligently, and new features obviously come with a bit of risk)

[Mnkras]

I'm on an older cython than 0.29.31, I just tried to to make sure it wasn't fixed in the latest.

Thanks, ill give it a try, is 3 published to anaconda?

[da-woods]

is 3 published to anaconda?

It doesn't look it it unfortunately