Decoder Fails to Deserialize Union with Dataclass variants

[mao3267]

• mashumaro version: 3.13.1
• Python version: 3.12.2
• Operating System: Sequoia 15.0

Description

I'm encountering an issue when using mashumaro to serialize and deserialize a nested dataclass with Union properties that include dataclasses and other non-None variants. While the serialization works as expected, the deserialization does not properly decode the union property. Instead of the expected object, the decoded result falls back to a dict type.

What I Did

Here is a minimal reproducible example:

from mashumaro.codecs.msgpack import MessagePackDecoder, MessagePackEncoder
from dataclasses import dataclass
from typing import Any, Union
import msgpack

def _default_msgpack_decoder(data: bytes) -> Any:
    return msgpack.unpackb(data, raw=False, strict_map_key=False)

@dataclass
class A:
    a: int

@dataclass
class InnerDC:
    a: Union[None, str, A] # or Union[str, A]

@dataclass
class DC:
    inner_dc: InnerDC 
    

# Encoding works
encoder = MessagePackEncoder(DC)
msgpack_bytes = encoder.encode(DC(inner_dc=InnerDC(a=A(a=1))))

# Decoding fails
decoder = MessagePackDecoder(DC, pre_decoder_func=_default_msgpack_decoder)
python_val = decoder.decode(msgpack_bytes)

assert python_val == DC(inner_dc=InnerDC(a=A(a=1)))

Actual Result

The decoded result is:
DC(inner_dc=InnerDC(a={'a': 1}))

The union property a is incorrectly decoded as a dictionary instead of the expected A object.

Expected Result

The decoded result should be:
DC(inner_dc=InnerDC(a=A(a=1)))

It would be helpful to understand whether this is a bug or if additional configuration is required.

[Future-Outlier]

Hi, @Fatal1ty
Do you know how to fix it?
If you don't have time, @mao3267 can create a PR to do this.
but we need some guidance, for example point out where might exist the bug

[Fatal1ty]

Hi @mao3267 @Future-Outlier

This is an old issue related to the default str deserialization as it was described here.

Good news, yesterday I finally decided to put an end to this problem and change the logic for deserializing unions containing str, bool, so we can expect a solution soon. In the next release, when deserializing unions, we will check the type of the input value before returning it as is as a string.

[Future-Outlier]

Hi @mao3267 @Future-Outlier

This is an old issue related to the default str deserialization as it was described here.

Good news, yesterday I finally decided to put an end to this problem and change the logic for deserializing unions containing str, bool, so we can expect a solution soon. In the next release, when deserializing unions, we will check the type of the input value before returning it as is as a string.

really love this solution, thank you very much, we will definitely test more if you need us, feel free to ping us <3

[Fatal1ty]

we will definitely test more if you need us, feel free to ping us <3

I would appreciate if you check the branch from this PR:

• Make Union deserialization algorithm more robust #256

[Fatal1ty]

The fix is ​​ready, will be in the next release.

[Future-Outlier]

The fix is ​​ready, will be in the next release.

Love you, you are a hero