Default value to dataclass field is ignored if setting up getter/setter

[subercui]

Bug report

As the title describes, for example, if I run the following,

from dataclasses import dataclass, InitVar


@dataclass
class Test:
    b: int = 6
    id: InitVar[int]=1

    def __post_init__(self, id=0):
        print(f"the value of id is {id} in the post_init")
        print(f"is id a property: {isinstance(id, property)}")
        self._id = id

    @property
    def id(self):
        print("using the getter")
        return self._id

    @id.setter
    def id(self, to):
        print("using the setter")
        self._id = to


print("\nInit with position argument")
a = Test(1)
print("a is", a)
print("a.id is", a.id)

print("\nInit with explicit key")
a4 = Test(id=4)
print("a4.id is", a4.id)

In the position argument initialization case, I expect to have a.id been initialized as either 1 or 0 (I think either behavior is acceptable). However, turns out the value is never set. The actual print-out is as follows:

Init with position argument
the value of id is <property object at 0x7f03526be450> in the post_init
is id a property: True
a is Test(b=1)
using the getter
a.id is <property object at 0x7f03526be450>

Init with explicit key
the value of id is 4 in the post_init
is id a property: False
using the getter
a4.id is 4

I've also tested setting id as an actual field(init=True/False), and the same issue occurs.

Your environment

• CPython versions tested on: 3.9.5
• Operating system and architecture: Ubuntu 18.04.6 LTS, x86_64

[subercui]

Here is the case for id as an actual Field

from dataclasses import dataclass, InitVar


@dataclass
class Test:
    b: int = 6
    id: int = 1

    def __post_init__(self):
        print(f"the value of id is {self.id} in the post_init")
        print(f"is id a property: {isinstance(self.id, property)}")
        self._id = self.id

    @property
    def id(self):
        print("using the getter")
        return self._id

    @id.setter
    def id(self, to):
        print("using the setter")
        self._id = to


print("\nInit with position argument")
a = Test(1)
print("a is", a)
print("a.id is", a.id)

print("\nInit with explicit key")
a4 = Test(id=4)
print("a4.id is", a4.id)

Output:

Init with position argument
using the setter
using the getter
the value of id is <property object at 0x7f03526c55e0> in the post_init
using the getter
is id a property: True
using the getter
a is using the getter
Test(b=1, id=<property object at 0x7f03526c55e0>)
using the getter
a.id is <property object at 0x7f03526c55e0>

Init with explicit key
using the setter
using the getter
the value of id is 4 in the post_init
using the getter
is id a property: False
using the getter
using the getter
a4.id is 4

[ericvsmith]

This is essentially what's happening:

from dataclasses import dataclass, InitVar


class Test:
    b: int = 6
    id: int = 1

    def __post_init__(self):
        print(f"the value of id is {self.id} in the post_init")
        print(f"is id a property: {isinstance(self.id, property)}")
        self._id = self.id

    @property
    def id(self):
        print("using the getter")
        return self._id

    @id.setter
    def id(self, to):
        print("using the setter")
        self._id = to


def init(self, b=6, id=Test.id):
    self.b = b
    self.id = id
    self.__post_init__()


Test.__init__ = init
Test = dataclass(Test)


print("\nInit with position argument")
a = Test(1)
print("a is", a)
print("a.id is", a.id)

print("\nInit with explicit key")
a4 = Test(id=4)
print("a4.id is", a4.id)

[ericvsmith]

There's nothing that can be done about this. After the class is created, and before @dataclass() runs. the value of Test.id is the property, and the value originally assigned to id is overwritten and lost. Consider:

class Test1:
    id: int = 1
print(Test1.__dict__)

Which prints {'__module__': '__main__', '__annotations__': {'id': <class 'int'>}, 'id': 1, '__dict__': <attribute '__dict__' of 'Test1' objects>, '__weakref__': <attribute '__weakref__' of 'Test1' objects>, '__doc__': None}
and

class Test2:
    id: int = 1
    @property
    def id(self): pass
print(Test2.__dict__)

which prints {'__module__': '__main__', '__annotations__': {'id': <class 'int'>}, 'id': <property object at 0x000001B58BE13EE0>, '__dict__': <attribute '__dict__' of 'Test2' objects>, '__weakref__': <attribute '__weakref__' of 'Test2' objects>, '__doc__': None}

Note that in the Test2 case, the default value of 1 has been overwritten, and is not available to @dataclass().

[subercui]

Thanks for the clarification!
I guess for the InitVar case, is it like:

def init(self, b=6, id=Test.id):
    self.b = b
    self.__post_init__(id)

So if I understand correctly, using a default value is kind of difficult when working with getter/setter. A workaround can be, in the __post_init__ have

if isinstance(self.id, property):
    self._id = put_default_value_here

[subercui]

So generally speaking, is there an easier way to use property within dataclass?

[ericvsmith]

Sure, you can do anything you want in __post_init__.

[ericvsmith]

So generally speaking, is there an easier way to use property within dataclass?

Somewhere there's an issue about changing the name of a field (maybe by using a field parameter?), but I can't find it and I don't recall if it was intended to solve this case.

[hauntsaninja]

Closing, since there doesn't seem to be anything actionable here