class_schema() vs .Schema

[vkruglik-aka]

cc @lovasoa

Can class_schema() and .Schema be used interchangeably on a Marshmallow Dataclass?

The documentation, such as in https://pypi.org/project/marshmallow-dataclass/, suggests to use the class_schema() function to generate a marshmallow Schema class from a python builtin dataclass. The doc doesn't say anything about the use of class_schema() on a Marshmallow Dataclass.

Is there any reason not to use class_schema() on a marshmallow dataclass?

The reason I ask is that .Schema confuses type checkers, and the documentation-suggested addition of Schema: ClassVar[Type[Schema]] = Schema in every marshmallow dataclass to counter this confusion just seems like unnecessary boilerplate, if class_schema() provides the same capability (and it already has the proper type annotations)

[dairiki]

I believe it will work to use class_schema on a marshmallow_dataclass.dataclass. Doing so, however, does not use the .Schema attribute. If the .Schema attribute is never evaluated there is a possibility of memory leakage in certain circumstances.

The only real reason to use a marshmallow_dataclass.dataclass over a dataclasses.dataclass is to add the .Schema attribute. If you're not going to use .Schema, just use a regular dataclass and manually call class_schema on it. (That's what I do in my code.)

There are also some comments that address this topic on #198.

[vkruglik-aka]

Thank you @dairiki