I'm not sure if this is intended behavior or an error. I'm creating dataclasses dynamically and trying to pickle those classes or objects containing instances of those classes. This was resulting in an error, so I trimmed it down to this example.

```
import pickle as pk
import dataclasses as dc

@dc.dataclass
class A:
     pass
pk.dumps(A)  # --> this is fine

B = dc.make_dataclass('B', [], bases=(A,))
pk.dumps(B)  # --> results in an error

# PicklingError: Can't pickle <class 'types.B'>: attribute lookup B on types failed
```

is this expected behavior? and if so, is there a way to create a dynamic dataclass that pickles?

You need to set the __module__ attribute.

B.__module__ = __name__

thank you + serhiy.storchaka

while that works in an interactive namespace, it fails in the following setting, which is closer to my dynamic use case.

```
class A:
    def __init__(self, fields=None):
        self.B = dc.make_dataclass('B', fields or [])
        self.B.__module__ = __name__
        # ...
        self.C = self.B()
```

now pickling A() fails.

Dataclasses are pickled by name, as well as other classes. Pickling classes which can not be accessed by name is not supported.

Thank you.