Support for optional dataclass attributes

[lidatong]

Support for optional dataclass attributes

[ElmoVanKielmo]

Of course I meant to_json() and not to_dict() in description of my second commit ;)

[lidatong]

Thanks for raising the issue and the corresponding PR!

I've given this some thought and I actually don't agree that {"bar": 7.45} should be a valid json representation of the Foo class in your example.

When considering the distinction between:
{"bar": 7.45}
vs.
{"bar": 7.45, "baz": None}

The main problem I see with making the former the json representation of a Foo is it would lead to equivalent json representations of two structurally dissimilar types:

class Foo:
    bar: float

class Foo:
    bar: float
    baz: Optional[float]

# these two classes end up having the same json repr

However, I'm in favor of adding an option to from_json to "infer missing keys as none" (which would be similar to your pr). Technically, this violates a (potential) invariant that from_json and to_json are inverses, which as nice that may be to provide, supplementing the API to support this usecase seems the more practical approach.

[ElmoVanKielmo]

I completely agree.

[lidatong]

really sorry for the long delay. infer_missing feature provided as a kwarg to from_json in https://github.com/lidatong/dataclasses-json/releases/tag/v0.0.7 along with fixes to make this lib work in 3.7