fix __post_init__ cause infinite recursion in inheritance

[Hanaasagi]

Change Summary

• Move _pydantic_post_init into _process_class, so it can use the closure feature to capture parent scope post_init_original method.
• Remove __post_init_original__ and __post_init_post_parse__ method.

Related issue number

It will resolve #536

Checklist

• Unit tests for the changes exist
• Tests pass on CI and coverage remains at 100%
• Documentation reflects the changes where applicable
• HISTORY.rst has been updated
  • if this is the first change since a release, please add a new section
  • include the issue number or this pull request number #<number>
  • include your github username @<whomever>

Why it can resolve the problem

Sorry for my pool English, it is not my native language.

Firstly, we should know

class A:
    def test(self):
        print(self)


class B(A):
    def test(self):
        super().test()


B().test()

Above code will print <__main__.B object at 0x7fdb1d7a1908>, it is a instance of Class B that we pass to A.test.

In pydantic, if we use the dataclass decorator, it will replace user's __post_init__ to _pydantic_post_init, and save original __post_init__ in __post_init_original__

pydantic/pydantic/dataclasses.py

Lines 74 to 86 in dd44fda

	post_init_original = getattr(_cls, '__post_init__', None)
	post_init_post_parse = getattr(_cls, '__post_init_post_parse__', None)
	if post_init_original and post_init_original.__name__ == '_pydantic_post_init':
	post_init_original = None
	_cls.__post_init__ = _pydantic_post_init
	cls = dataclasses._process_class(_cls, init, repr, eq, order, unsafe_hash, frozen) # type: ignore
	fields: Dict[str, Any] = {
	field.name: (field.type, field.default if field.default != dataclasses.MISSING else Required)
	for field in dataclasses.fields(cls)
	}
	cls.__post_init_original__ = post_init_original
	cls.__post_init_post_parse__ = post_init_post_parse

According to code of issue #536,

from pydantic.dataclasses import dataclass

@dataclass
class A:
    a: int
    
    def __post_init__(self):
        print("A")

@dataclass
class B(A):
    b: int
    
    def __post_init__(self):
        super().__post_init__()
        print("B")

B(b=1, a=2)

After decorated by dataclass,

In [1]: A.__post_init__
Out[1]: <function pydantic.dataclasses._pydantic_post_init(self: 'DataclassType', *initvars: Any) -> None>

In [2]: A.__post_init_original__
Out[2]: <function __main__.A.__post_init__(self)>

In [3]: B.__post_init__
Out[3]: <function pydantic.dataclasses._pydantic_post_init(self: 'DataclassType', *initvars: Any) -> None>

In [4]: B.__post_init_original__
Out[4]: <function __main__.B.__post_init__(self)>

In [5]: A.__post_init__ is B.__post_init__
Out[5]: True

So when we create an instance of class B, like B(a=1, b=2). Python stdlib's dataclass will call __post_init__ automatically. But it calls pydantic.dataclasses._pydantic_post_initactually(we replace it).

In _pydantic_post_init, it will call original __post_init__ through __post_init_original__

pydantic/pydantic/dataclasses.py

Lines 28 to 35 in dd44fda

	def _pydantic_post_init(self: 'DataclassType', *initvars: Any) -> None:
	if self.__post_init_original__:
	self.__post_init_original__(*initvars)
	d = validate_model(self.__pydantic_model__, self.__dict__, cls=self.__class__)[0]
	object.__setattr__(self, '__dict__', d)
	object.__setattr__(self, '__initialised__', True)
	if self.__post_init_post_parse__:
	self.__post_init_post_parse__()

In above code, we call super().__post_init__() in __post_init_original__, Python will find A.__post_init__ and pass instance b. Currently A.__post_init__ is also a pydantic.dataclasses._pydantic_post_init. So it will get __post_init_original__ by instance b, we fall into a infinite recursion.

[codecov]

Codecov Report

Merging #606 into master will not change coverage.
The diff coverage is 100%.

@@          Coverage Diff          @@
##           master   #606   +/-   ##
=====================================
  Coverage     100%   100%           
=====================================
  Files          15     15           
  Lines        2584   2582    -2     
  Branches      510    510           
=====================================
- Hits         2584   2582    -2

[samuelcolvin]

otherwise this looks great.

[samuelcolvin]

is there any reason we can't call the first argument self?

[samuelcolvin]

great, thank you.