dataclass-compat

统一API，用于与多个类似dataclass库一起工作

dataclass模式

有许多库实现了类似dataclass模式！

dataclasses.dataclass

import dataclasses

@dataclasses.dataclass
class SomeDataclass:
    a: int = 0
    b: str = "b"
    c: list[int] = dataclasses.field(default_factory=list)

pydantic.BaseModel

import pydantic

class SomePydanticModel(pydantic.BaseModel):
    a: int = 0
    b: str = "b"
    c: list[int] = pydantic.Field(default_factory=list)

attrs.define

import attr

@attr.define
class SomeAttrsModel:
    a: int = 0
    b: str = "b"
    c: list[int] = attr.field(default=attr.Factory(list))

msgspec.Struct

import msgspec

class SomeMsgspecStruct(msgspec.Struct):
    a: int = 0
    b: str = "b"
    c: list[int] = msgspec.field(default_factory=list)

等等...

统一API

这些都是非常棒的库，每个库都有其自身的优点和缺点。有时，你只是想能够查询类似dataclass对象的基本信息，例如获取字段名或类型，或将其转换为字典。

dataclass-compat提供了对这些操作的统一API（尽可能遵循或扩展dataclasses的API）。

def fields(obj: Any) -> tuple[Field, ...]:
    """Return a tuple of dataclass_compat.Field objects for the object."""

def replace(obj: Any, /, **changes: Any) -> Any:
    """Return a copy of obj with the specified changes."""

def asdict(obj: Any) -> dict[str, Any]:
    """Return a dict representation of obj."""

def astuple(obj: Any) -> tuple[Any, ...]:
    """Return a tuple representation of obj."""

def params(obj: Any) -> DataclassParams:
    """Return parameters used to define the dataclass."""

dataclass_compat.Field和dataclass_compat.DataclassParam对象是简单的dataclass，分别匹配dataclasses.Field协议和（私有）dataclasses._DataclassParams对象。字段对象还添加了一个native_field属性，该属性是来自底层库的原始字段对象。

示例

from dataclass_compat import Field, fields

standardized_fields = (
    Field(name="a", type=int, default=0),
    Field(name="b", type=str, default="b"),
    Field(name="c", type=list[int], default_factory=list),
)

assert (
    fields(SomeDataclass)
    == fields(SomePydanticModel)
    == fields(SomeAttrsModel)
    == fields(SomeMsgspecStruct)
    == standardized_fields
)

支持的库

• dataclasses
• collections.namedtuple
• pydantic（v1和v2）
• attrs
• msgspec
• dataclassy
• sqlmodel（它是pydantic）

...也许有一天？

• pyfields
• marshmallow
• sqlalchemy
• django
• peewee
• pyrsistent
• recordclass