Unable to deserialize nested types of union and tuple

Python 3.9 pyserde[toml]==0.3.1

from dataclasses import dataclass
from typing import Union
from serde import toml, deserialize


@deserialize
@dataclass
class TestClass:
    a: Union[None, Union[int, float], tuple[Union[int, float], Union[int, float]]]


if __name__ == '__main__':
    test_class = toml.from_toml(TestClass, 'a = [4, 1]')
    print(test_class)

Running the above code will throw an error:

serde.core.SerdeError: Can not deserialize [4, 1] of type List into Union[NoneType, int, float, Tuple[Union[int, float], Union[int, float]]]. Reason: 'union_de_int_float'

[ydylla]

It seems this loop runs only once. We need to expand union args and recursively generate function IMO. @ydylla What do you think?

Yes, but the recursion should happen in iter_unions there it should be:

if is_union(cls):
    yield cls
    for arg in type_args(cls):
        yield from iter_unions(arg)

I did not really test Unions in Unions, the only existing test is too simple and python is actually optimizing it away 😃 Sadly the fix in iter_unions produces a follow up issue. A nested tuple is flattened, but it shouldn’t.

@deserialize
@serialize
@dataclass
class B:
    v: Union[UUID, Tuple[Union[UUID, int]]]

to_dict(B((1,)))
# becomes {'v': 1} but should be {'v': (1,)}

from_dict(B,{"v":(1,)})
# becomes B(v=1) but should be B(v=(1,))

We probably need an own render mode that is used within union functions similar to DeField.iterbased to prevent this.

[yukinarit]

Hi @uheee

Just wanted to let you know v0.3.2 is pushed to PyPI! 🙂