baseclasses

Simple dataclasses alternative (beta)

Installation

baseclasses is available on PyPI:

pip install baseclasses

Quick Start

baseclasses is an alternative to dataclasses.

There are many such alternatives, and there are a lot of reasons to use them (including the built-in dataclasses), but if you want a library with:

• No monkey-patched methods generated from strings
• No special decorators, just regular inheritance
• Ability to order fields in subclasses as desired
• Auto-inheritance of parent properties, with ability to change per subclass
• Emphasis on keyword args over positional args
• Optional __pre_init__ ability to mutate kwargs
• Optional ability to reference other kwargs in default_factory func
• A shorthand helper to specify InternalStateField
• Ability to differentiate between str and repr

you can give baseclasses a try. E.g.:

from typing import Dict, Optional

import baseclasses


# No decorator
class Foo(baseclasses.FrozenBaseClass):
    a: int
    b: int
    c: Optional[str] = baseclasses.Field(default="hello", hash=False)
    _d: Dict = baseclasses.InternalStateField(default_factory=dict)


# Auto-inherits FrozenBaseClass properties
class ChildFoo(Foo):
    # No problems with child class field ordering
    x: float

    # Dynamic defaults
    y: int = baseclasses.Field(
        default_factory=lambda **kwargs: kwargs["a"] * 2.0
    )


# Override frozen state per child class
class MutableChildFoo(ChildFoo, frozen=False):
    pass

Comparisons to Alternatives

See tests/test_compare.py.

Like dataclasses, but:

• No issues with adding new optional fields because you're forced to use kwargs in all but the simplest cases
• Uses metaclass and simple inheritance over monkey-patched generated code
• Child classes automatically inherit state from parent, so no need to re-declare
• But you can also change frozen status per child class
• You can use default_factory(lambda **kwargs: ...) to access init fields
• Or you can mutate kwargs with an optional __pre_init__
• More consistent with traditional OOP style (e.g., obj.get_fields() instead of dataclasses.get_fields(obj))

Like dataclassy but:

• Uses metaclass and simple inheritance over monkey-patched generated code

There are others:

• attrs
• pydantic
• traitlets
• param

Note: there are perfectly valid reasons for why the Python community decided to use generated code over simple inheritance. baseclasses is just an alternative that implements things differently.

TODO

• Add init property to field (to mirror dataclasses)
• Expose obj.replace(**kwargs) (to mirror dataclasses)
• Consider serialize property to field
• Take advantage of PEP 681 included in Python 3.11