`serializable_dataclass` breaks with `from __future__ import annotations` when nested vals

[mivanit]

Details

Description

serializable_dataclass doesn't handle from __future__ import annotations (PEP 563). When the import is present, all annotations become strings at runtime, which causes two problems:

1. Auto-serialization of nested SerializableDataclass fields fails. The generated serialize() doesn't recognize that e.g. list[Inner] contains a SerializableDataclass, so nested objects are left as-is and json.dumps() raises TypeError: Object of type ... is not JSON serializable.

2. Plain defaults after serializable_field cause TypeError: non-default argument follows default argument. When annotations are strings, it appears that the decorator's pre-processing of fields mishandles plain = value defaults, causing dataclasses.dataclass() to reject the field ordering.

Reproducer

from __future__ import annotations

from muutils.json_serialize import (
    SerializableDataclass,
    serializable_dataclass,
    serializable_field,
)
import json


@serializable_dataclass
class Inner(SerializableDataclass):
    value: int = 0


@serializable_dataclass
class Outer(SerializableDataclass):
    items: list[Inner] = serializable_field(default_factory=list)


outer = Outer(items=[Inner(value=42)])
data = outer.serialize()
# data["items"] is [Inner(value=42)] instead of [{"value": 42, ...}]
json.dumps(data)  # TypeError: Object of type Inner is not JSON serializable

Workaround

Add explicit serialization_fn/deserialize_fn for any field containing nested SerializableDataclass objects:

items: list[Inner] = serializable_field(
    default_factory=list,
    serialization_fn=lambda items: [i.serialize() for i in items],
    deserialize_fn=lambda items: [Inner.load(i) for i in items],
)

For the field-ordering issue, use kw_only=True:

@serializable_dataclass(kw_only=True)
class MyConfig(SerializableDataclass):
    ...