Lib/dataclasses.py (part 2)

Source:

cpython 3.14 @ ab2d84fe1023/Lib/dataclasses.py ↗

This annotation covers the __init__ code generation, frozen dataclasses, and the slots=True option introduced in Python 3.10. See lib_dataclasses_detail for @dataclass, field(), and the basic decorator.

Map

Lines	Symbol	Role
1-200	_process_class	Main dataclass transformation
201-400	_init_fn	__init__ code generation via exec
401-550	_frozen_setattr, _frozen_delattr	Frozen class mutation guard
551-700	_hash_fn, _eq_fn	__hash__ and __eq__ generation
701-850	_repr_fn	__repr__ generation
851-1000	slots=True logic	Create a new class with __slots__
1001-1150	kw_only, match_args	PEP 681 and structural pattern matching
1151-1350	fields(), asdict(), astuple(), replace()	Introspection utilities

Reading

__init__ code generation

_init_fn builds the __init__ function by constructing a Python source string and calling exec. This is necessary because the number and names of parameters vary per class.

# CPython: Lib/dataclasses.py:543 _init_fn
def _init_fn(fields, frozen, has_post_init, self_name, globals, slots):
    ...
    body_lines = []
    for f in fields:
        ...
        if f.default is MISSING and f.default_factory is MISSING:
            body_lines.append(f'{self_name}.{f.name} = {f.name}')
        elif f.default_factory is not MISSING:
            body_lines.append(f'if {f.name} is HAS_DEFAULT_FACTORY:')
            body_lines.append(f'  {self_name}.{f.name} = {f.default_factory_name}()')
            body_lines.append(f'else:')
            body_lines.append(f'  {self_name}.{f.name} = {f.name}')
    ...
    return _create_fn('__init__', ..., body_lines, ...)

_create_fn calls exec(f"def {name}({args}):\n {body}") in a temporary namespace, then returns the resulting function.

Frozen classes

For @dataclass(frozen=True), __setattr__ and __delattr__ are replaced:

# CPython: Lib/dataclasses.py:320 _frozen_setattr
def _frozen_setattr(self, name, value):
    if type(self) is cls or type(self) in cls.__mro__:
        raise FrozenInstanceError('cannot assign to field %r' % name)
    object.__setattr__(self, name, value)

This allows inheritance from frozen dataclasses as long as the subclass is not itself frozen.

slots=True (Python 3.10+)

When slots=True, _process_class creates a new class that inherits from the decorated class and adds __slots__ for all defined fields. This reduces memory use and speeds up attribute access.

# CPython: Lib/dataclasses.py:1012 _add_slots
def _add_slots(cls, is_frozen, has_weakref_slot):
    cls_dict = {}
    for key, value in cls.__dict__.items():
        if key not in ('__dict__', '__weakref__'):
            cls_dict[key] = value
    cls_dict['__slots__'] = tuple(f.name for f in fields(cls))
    return type(cls)(cls.__name__, cls.__bases__, cls_dict)

__post_init__

If the class defines __post_init__, the generated __init__ calls it at the end with the InitVar fields as arguments.

gopy notes

dataclasses is a high-priority port because many libraries (pydantic, attrs) use it. The generated __init__ approach via exec requires gopy's exec() to work correctly. An alternative is to generate the __init__ as a Go closure at class creation time rather than compiling Python source. frozen=True maps to overriding __setattr__ in gopy's class creation path.