Feature request: ForwardRef support for dataclasses

[ceelian]

Bug

With python 3.7+ from __future__ import annotations the example is broken and raises an exception.

• OS: OS X 10.13.6 (17G4015)
• Python version: 3.7.1
• Pydantic version: 0.19

# only the following line was changed compared to the example
# in the docs https://pydantic-docs.helpmanual.io/#nested-dataclasses
from __future__ import annotations 

from pydantic import UrlStr
from pydantic.dataclasses import dataclass

@dataclass
class NavbarButton:
    href: UrlStr

@dataclass
class Navbar:
    button: NavbarButton

navbar = Navbar(button=('https://example.com',))
print(navbar)

 python foo.py
Traceback (most recent call last):
  File "foo.py", line 13, in <module>
    navbar = Navbar(button=('https://example.com',))
  File "<string>", line 3, in __init__
  File "/tmp/env/lib/python3.7/site-packages/pydantic/dataclasses.py", line 25, in _pydantic_post_init
    d = validate_model(self.__pydantic_model__, self.__dict__)
  File "/tmp/env/lib/python3.7/site-packages/pydantic/main.py", line 508, in validate_model
    f"field {field.name} not yet prepared and type is still a ForwardRef, "
pydantic.errors.ConfigError: field button not yet prepared and type is still a ForwardRef, you'll need to call MetaModel.update_forward_refs()

[samuelcolvin]

This isn't a bug, it's a feature request.

the example is broken and raises an exception

Wrong, this isn't the example. This is some other code that looks like the example, adding lazy type evaluation via from __future__ import annotations completely changes the context.

[samuelcolvin]

should be fixed by #399, let me know if this fixes the issue for you.

[ceelian]

Well, the example code works in a standalone script with the new version, but not in my pytest in the project I use it. Can't figure it out why at the moment, but I am pretty sure that might not be related to pydantic. Although the error I get is from the new version:

pydantic.errors.ConfigError: field "button" not yet prepared so type is still a ForwardRef, you might need to call MetaModel.update_forward_refs().

and the python env with pytest is exactly the same as with the standalone script.

[samuelcolvin]

Because, for pydantic to be able to resolve type annotations, they need to be global in the module.

[ceelian]

Ah, got it. Thank you for clarification and the fast implementation of the new feature!

[dpriskorn]

Because, for pydantic to be able to resolve type annotations, they need to be global in the module.

I'm a beginner of pydantic. I don't understand "global in the module" and run into the same error as @ceelian. Could someone elaborate?