Request for dataclass backport support in 3.6 (using 3.6.6).

[bbarker]

• Are you reporting a bug, or opening a feature request?

This is a bug for 3.6, supposedly works in 3.7. So, I'm requesting a backport of dataclass functionality for 3.6 if not too difficult.

Our initial use case looks like this, with a simple dataclass definition, and can be compared to the even simpler Point example in #3937, where we get error: Too many arguments for the constructor.

@dataclass(frozen=True)
class AuthorList:
    """Represents author names."""

    """Raw author field string."""
    raw: str = field(default_factory=str)

    def __str__(self) -> str:
        return self.raw

And a simple dataclass use:

        # TODO type ignore: possibly mypy #3937
        fields['authors'] = AuthorList(fields['authors'])  # type: ignore

• What are the versions of mypy and Python you are using?

3.6.6 and 0.620

Do you see the same issue after installing mypy from Git master?

I have not tried.

• What are the mypy flags you are using? (For example --strict-optional)

[mypy]

#
# Covered by --strict, with some turned off:
#
disallow_untyped_defs=True
check_untyped_defs=True
# currently an issue with sql alchemy
disallow_subclassing_any=false
# Need to experiment/think about this one:
disallow_any_decorated=false
warn_redundant_casts=True
warn_return_any=True
warn_unused_ignores=True
# this seems to be at least somewhat non-functioning:
warn_unused_configs=True
#may be worth reconsidering this one:
no_implicit_optional=True
strict_optional=True

#
# Other: 
#
ignore_missing_imports=True

[browse.*]
disallow_untyped_calls=True

[gvanrossum]

I think all this requires is a change to typeshed, copying dataclasses.pyi from stdlib/3.7 to third_party/3.

[dsully]

I'm running into this as well. Not quite able to use 3.7