Feature/handle generics

[mciszczon]

This is a draft PR for introducing the handling of typing generics. Feel free to review this PR, but keep in mind it is not yet finished—publishing it just for the sake of code reviews and brainstorming.

Fixes #131
Fixes #157
Fixes #160

TODO

• Add much more test cases that use typing.Generic and typing.TypeVar.
• Consider handling other typing shenanigans, like Annotated, TypeAlias, TypeVarTuple.
• Backport typing.get_origin and typing.get_args for lower Python versions.
• Refactor and clean up the code, it's super ugly at the moment.
• Consider using typing. TypeGuard for typehinting the is_instance and its related functions.

[mciszczon]

@konradhalas After consideration I think it could already be merged as is feature-wise. Maybe we might want to write a few more tests, but that would be it. What do you think?

[object-Object]

I'm using this branch for a project and found this. This block doesn't make sense and also doesn't work.

get_type_hints gets a mapping from field name to field type for the dataclass value. You're iterating over that, but you only ever check the first item because both branches end with return. I'm pretty sure both of those returns in the loop should only be returning if the value is False.

On the first branch, if field_type is a TypeVar, you check the field value against all the arguments for the subscripted generic type_. Why is get_args in the loop? Why is getattr in the inner loop? Neither one is changing at that point. I believe this would also fail to detect an incorrect type if, say, a field is hinted AnyStr, the value is None, and the dataclass has two type arguments, one str and one None.

I assume the second branch is trying to check if the field value matches the field type? But what you're actually doing is checking if the dataclass matches the subscripted dataclass type. This crashes with the exception TypeError: Subscripted generics cannot be used with class and instance checks for obvious reasons.

For the false negative issue with multiple type arguments, possibly something could be done with __orig_bases__? Not sure.

Here's my suggestion for a fixed version. I haven't written test cases, but I did try it with my fairly complicated use case and it seems to work.

Suggested change

	type_hints = cache(get_type_hints)(type(value))
	for field_name, field_type in type_hints.items():
	if isinstance(field_type, TypeVar):
	args = get_args(type_)
	return True if not args else any(isinstance(getattr(value, field_name, None), arg) for arg in args)
	else:
	return isinstance(value, type_)
	type_args = get_args(type_)
	type_hints = cache(get_type_hints)(type(value))
	for field_name, field_type in type_hints.items():
	field_value = getattr(value, field_name, None)
	if isinstance(field_type, TypeVar):
	# TODO: this will fail to detect incorrect type in some cases
	# see comments on https://github.com/konradhalas/dacite/pull/209
	if not any(is_instance(field_value, arg) for arg in type_args):
	return False
	elif get_origin(field_type) is not ClassVar:
	if not is_instance(field_value, field_type):
	return False