Support several dataclasses in a union

[sobolevn]

Description

Currently msgspec has a very noticable limitation: it cannot convert an object with a field defined as a union of two dataclasses. Which is very common. For example:

>>> import msgspec
>>> import dataclasses
>>> @dataclasses.dataclass
... class User:
...     username: str
...     
>>> @dataclasses.dataclass
... class Reference:
...     instance_id: int
...     
>>> @dataclasses.dataclass
... class Project:
...     author: User | Reference
...     
>>> msgspec.convert({'author': {'username': 'example'}}, Project)
Traceback (most recent call last):
  File "<python-input-5>", line 1, in <module>
    msgspec.convert({'author': {'username': 'example'}}, Project)
    ~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
TypeError: Type unions may not contain more than one dataclass or attrs type - type `__main__.User | __main__.Reference` is not supported

pydantic does not have such a limitation. It works for this case.
I would love to have the same feature in msgspec.

See https://docs.pydantic.dev/2.2/usage/types/unions/

Real use-case

We in django-modern-rest have a lot of cases where unions are very specific. For example, in our OpenAPI integration. https://github.com/wemake-services/django-modern-rest/blob/04632fcc87ddb5d067ec22effec6d93bc07ba3b8/dmr/openapi/objects/schema.py#L33

We cannot model it any other way, because:

1. We need them to be dataclasses, so both msgspec and pydantic will understand it
2. We can't remove the union type, because this is how the OpenAPI spec defines this field

[sobolevn]

The same happens with TypedDict types:

>>> from typing import TypedDict
>>> class A(TypedDict):
...     a: int
...     
>>> class B(TypedDict):
...     b: str
...     
>>> import msgspec
>>> msgspec.convert({'a': 1}, A | B)
Traceback (most recent call last):
  File "<python-input-4>", line 1, in <module>
    msgspec.convert({'a': 1}, A | B)
    ~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^
TypeError: Type unions may not contain more than one TypedDict type - type `__main__.A | __main__.B` is not supported

[ofek]

As a temporary workaround, would you be able to use a tagged union? Or define a custom dec/enc_hook that would automatically attempt coercion into each type in the union?

[sobolevn]

would you be able to use a tagged union?

I can't do that unfortunatelly :(
My models must be independent from the serializers. Because users might use my framework with pydantic or msgspec.

I can try the custom hook though.