Python dataclass 源码阅读与分析


In [1]: class FooDataClass:
   ...:     def __init__(self, name, num=1):
   ...: = name
   ...:         self.num = num

In [2]: from dataclasses import dataclass

In [3]: @dataclass
   ...: class BarDataClass:
   ...:     name: str
   ...:     num: int = 1

In [4]: FooDataClass("jhon")
Out[4]: <__main__.FooDataClass at 0x7f1fe26964a8>

In [5]: BarDataClass("jhon")
Out[5]: BarDataClass(name='jhon', num=1)

没有dataclass之前,主要都是如1中使用,或者是使用namedtuple,但是现在有了dataclass就方便了。但是这玩意儿强制使用type signature。


当然,宇宙惯例,直接去看 dataclass 装饰器怎么实现的。

def dataclass(_cls=None, *, init=True, repr=True, eq=True, order=False,
              unsafe_hash=False, frozen=False):
    """Returns the same class as was passed in, with dunder methods
    added based on the fields defined in the class.

    Examines PEP 526 __annotations__ to determine fields.

    If init is true, an __init__() method is added to the class. If
    repr is true, a __repr__() method is added. If order is true, rich
    comparison dunder methods are added. If unsafe_hash is true, a
    __hash__() method function is added. If frozen is true, fields may
    not be assigned to after instance creation.

    def wrap(cls):
        return _process_class(cls, init, repr, eq, order, unsafe_hash, frozen)

    # See if we're being called as @dataclass or @dataclass().
    if _cls is None:
        # We're called with parens.
        return wrap

    # We're called as @dataclass without parens.
    return wrap(_cls)

按照我们上面的例子,那就是调用了 _process_class,所以我们继续进去看:

def _process_class(cls, init, repr, eq, order, unsafe_hash, frozen):
    # Now that dicts retain insertion order, there's no reason to use
    # an ordered dict.  I am leveraging that ordering here, because
    # derived class fields overwrite base class fields, but the order
    # is defined by the base class, which is found first.
    fields = {}

    setattr(cls, _PARAMS, _DataclassParams(init, repr, eq, order,
                                           unsafe_hash, frozen))

    # Find our base classes in reverse MRO order, and exclude
    # ourselves.  In reversed order so that more derived classes
    # override earlier field definitions in base classes.  As long as
    # we're iterating over them, see if any are frozen.
    any_frozen_base = False
    has_dataclass_bases = False
    for b in cls.__mro__[-1:0:-1]:
        # Only process classes that have been processed by our
        # decorator.  That is, they have a _FIELDS attribute.
        base_fields = getattr(b, _FIELDS, None)
        if base_fields:
            has_dataclass_bases = True
            for f in base_fields.values():
                fields[] = f
            if getattr(b, _PARAMS).frozen:
                any_frozen_base = True

    # Annotations that are defined in this class (not in base
    # classes).  If __annotations__ isn't present, then this class
    # adds no new annotations.  We use this to compute fields that are
    # added by this class.
    # Fields are found from cls_annotations, which is guaranteed to be
    # ordered.  Default values are from class attributes, if a field
    # has a default.  If the default value is a Field(), then it
    # contains additional info beyond (and possibly including) the
    # actual default value.  Pseudo-fields ClassVars and InitVars are
    # included, despite the fact that they're not real fields.  That's
    # dealt with later.
    cls_annotations = cls.__dict__.get('__annotations__', {})

    # Now find fields in our class.  While doing so, validate some
    # things, and set the default values (as class attributes) where
    # we can.
    cls_fields = [_get_field(cls, name, type)
                  for name, type in cls_annotations.items()]
    for f in cls_fields:
        fields[] = f

        # If the class attribute (which is the default value for this
        # field) exists and is of type 'Field', replace it with the
        # real default.  This is so that normal class introspection
        # sees a real default value, not a Field.
        if isinstance(getattr(cls,, None), Field):
            if f.default is MISSING:
                # If there's no default, delete the class attribute.
                # This happens if we specify field(repr=False), for
                # example (that is, we specified a field object, but
                # no default value).  Also if we're using a default
                # factory.  The class attribute should not be set at
                # all in the post-processed class.
                setattr(cls,, f.default)

    # Do we have any Field members that don't also have annotations?
    for name, value in cls.__dict__.items():
        if isinstance(value, Field) and not name in cls_annotations:
            raise TypeError(f'{name!r} is a field but has no type annotation')

    # Check rules that apply if we are derived from any dataclasses.
    if has_dataclass_bases:
        # Raise an exception if any of our bases are frozen, but we're not.
        if any_frozen_base and not frozen:
            raise TypeError('cannot inherit non-frozen dataclass from a '
                            'frozen one')

        # Raise an exception if we're frozen, but none of our bases are.
        if not any_frozen_base and frozen:
            raise TypeError('cannot inherit frozen dataclass from a '
                            'non-frozen one')

    # Remember all of the fields on our class (including bases).  This
    # also marks this class as being a dataclass.
    setattr(cls, _FIELDS, fields)

    # Was this class defined with an explicit __hash__?  Note that if
    # __eq__ is defined in this class, then python will automatically
    # set __hash__ to None.  This is a heuristic, as it's possible
    # that such a __hash__ == None was not auto-generated, but it
    # close enough.
    class_hash = cls.__dict__.get('__hash__', MISSING)
    has_explicit_hash = not (class_hash is MISSING or
                             (class_hash is None and '__eq__' in cls.__dict__))

    # If we're generating ordering methods, we must be generating the
    # eq methods.
    if order and not eq:
        raise ValueError('eq must be true if order is true')

    if init:
        # Does this class have a post-init function?
        has_post_init = hasattr(cls, _POST_INIT_NAME)

        # Include InitVars and regular fields (so, not ClassVars).
        flds = [f for f in fields.values()
                if f._field_type in (_FIELD, _FIELD_INITVAR)]
        _set_new_attribute(cls, '__init__',
                                    # The name to use for the "self"
                                    # param in __init__.  Use "self"
                                    # if possible.
                                    '__dataclass_self__' if 'self' in fields
                                            else 'self',

    # Get the fields as a list, and include only real fields.  This is
    # used in all of the following methods.
    field_list = [f for f in fields.values() if f._field_type is _FIELD]

    if repr:
        flds = [f for f in field_list if f.repr]
        _set_new_attribute(cls, '__repr__', _repr_fn(flds))

    if eq:
        # Create _eq__ method.  There's no need for a __ne__ method,
        # since python will call __eq__ and negate it.
        flds = [f for f in field_list if]
        self_tuple = _tuple_str('self', flds)
        other_tuple = _tuple_str('other', flds)
        _set_new_attribute(cls, '__eq__',
                           _cmp_fn('__eq__', '==',
                                   self_tuple, other_tuple))

    if order:
        # Create and set the ordering methods.
        flds = [f for f in field_list if]
        self_tuple = _tuple_str('self', flds)
        other_tuple = _tuple_str('other', flds)
        for name, op in [('__lt__', '<'),
                         ('__le__', '<='),
                         ('__gt__', '>'),
                         ('__ge__', '>='),
            if _set_new_attribute(cls, name,
                                  _cmp_fn(name, op, self_tuple, other_tuple)):
                raise TypeError(f'Cannot overwrite attribute {name} '
                                f'in class {cls.__name__}. Consider using '

    if frozen:
        for fn in _frozen_get_del_attr(cls, field_list):
            if _set_new_attribute(cls, fn.__name__, fn):
                raise TypeError(f'Cannot overwrite attribute {fn.__name__} '
                                f'in class {cls.__name__}')

    # Decide if/how we're going to create a hash function.
    hash_action = _hash_action[bool(unsafe_hash),
    if hash_action:
        # No need to call _set_new_attribute here, since by the time
        # we're here the overwriting is unconditional.
        cls.__hash__ = hash_action(cls, field_list)

    if not getattr(cls, '__doc__'):
        # Create a class doc-string.
        cls.__doc__ = (cls.__name__ +
                       str(inspect.signature(cls)).replace(' -> None', ''))

    return cls

看到 if init 那里,就很明显了,调用了 _set_new_attribute,那我们继续进去看:

def _set_new_attribute(cls, name, value):
    # Never overwrites an existing attribute.  Returns True if the
    # attribute already exists.
    if name in cls.__dict__:
        return True
    setattr(cls, name, value)
    return False

_init_fn 做的事情就是初始化了。


  • Redis源码阅读:AOF重写
  • Redis源码阅读:AOF持久化
  • Redis源码阅读:key是怎么过期的
  • Redis源码阅读:字典是怎么实现的
  • Redis源码阅读:执行命令
  • Redis源码阅读:启动过程
  • WAL(Write-ahead logging)的套路
  • 搞定CORS问题
  • 如何定位程序问题所在
  • 设计一个IM归档系统
  • logrotate read only filesystem问题
  • Golang GIN写单测时,愉快的使用返回值
  • Python Queue源码分析
  • Go里优雅的使用全局配置
  • Golang sync.Map源码分析