Library is not compatible with "from __future__ import annotations" (PEP 563)

[lidatong]

Library is not compatible with "from __future__ import annotations" (PEP 563)

[chrahunt]

To get the types we must call typing.get_type_hints and passing the class. This returns a mapping that can be used to retrieve the type by passing field.name. For example:

$ python
Python 3.7.1 (default, Jan  1 2019, 15:19:53)
[GCC 5.5.0 20171010] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> from __future__ import annotations
>>> from dataclasses import dataclass, fields
>>> import typing
>>> @dataclass
... class C:
...  i: int
...
>>> fields(C)
(Field(name='i',type='int',default=<dataclasses._MISSING_TYPE object at 0x7f62e8594780>,default_factory=<dataclasses._MISSING_TYPE object at 0x7f62e8594780>,init=True,repr=True,hash=None,compare=True,metadata=mappingproxy({}),_field_type=_FIELD),)
>>> fields(C)[0].type
'int'
>>> fields(C)[0].name
'i'
>>> typing.get_type_hints(C)
{'i': <class 'int'>}
>>> typing.get_type_hints(C)[fields(C)[0].name]
<class 'int'>

The only caveat with this approach is that for classes that use forward references (with the class name itself) the typing.get_type_hints will raise a NameError if called before the class has been actually initialized in its module (such as directly in a decorator).

[lidatong]

Hi, yes, I am aware of this issue and it is being tracked in #31, specifically #5. Thanks for reporting.

My instinct is there isn't a good way to address this (as described in #5) without improvements to the typing API, but let me know if you have suggestions.