dataclass init=False field with default works but default_factory does not

[simplecoder878]

BPO	45366
Nosy	@ericvsmith, @sobolevn, @akulakov

^{Note: these values reflect the state of the issue at the time it was migrated and might not reflect the current state.}

Show more details

GitHub fields:

assignee = None
closed_at = None
created_at = <Date 2021-10-04.20:17:33.399>
labels = ['ctypes', 'type-bug', '3.9']
title = 'dataclass init=False field with default works but default_factory does not'
updated_at = <Date 2021-11-08.01:28:02.575>
user = 'https://bugs.python.org/simplecoder878'

bugs.python.org fields:

activity = <Date 2021-11-08.01:28:02.575>
actor = 'andrei.avk'
assignee = 'none'
closed = False
closed_date = None
closer = None
components = ['ctypes']
creation = <Date 2021-10-04.20:17:33.399>
creator = 'simple_coder878'
dependencies = []
files = []
hgrepos = []
issue_num = 45366
keywords = []
message_count = 4.0
messages = ['403182', '403183', '403206', '405920']
nosy_count = 4.0
nosy_names = ['eric.smith', 'sobolevn', 'andrei.avk', 'simple_coder878']
pr_nums = []
priority = 'normal'
resolution = None
stage = None
status = 'open'
superseder = None
type = 'behavior'
url = 'https://bugs.python.org/issue45366'
versions = ['Python 3.9']

[simplecoder878]

Simple example
from dataclasses import dataclass, field

@dataclass(init=False)
class TestObject(object):
    m: str = field(default='hi')
    k: list = field(default_factory=list)

    def test(self):
        print(f'm is {self.m} ')
        print(f'k is {self.k}')

if __name__ == '__main__':
    myobject = TestObject()
    myobject.test()

Produces:

Traceback (most recent call last):
  File "H:\unit_test\tests_dataclass.py", line 81, in <module>
    myobject.test()
  File "H:\unit_test\tests_dataclass.py", line 76, in test
    print(f'k is {self.k}')
AttributeError: 'TestObject' object has no attribute 'k'
m is hi 

So m is initialized to hi but k just disappears

But wait there's more!

If i do

from dataclasses import dataclass, field

@dataclass(init=False)
class TestObject(object):
    m: str = field(default='hi')
    k: list = field(default_factory=list)

    def test(self):
        print(f'm is {self.m} ')
        print(f'k is {self.k}')


@dataclass
class InheritedTestObject(TestObject):

    def __post_init__(self):
        super().__init__()
        print(f'Inherited m is {self.m} ')
        print(f'Inherited k is {self.k}')
        print(f'Inherited g is {self.k}')


if __name__ == '__main__':
    myobject = InheritedTestObject()
    myobject.test()

It produces:

Inherited m is hi
Inherited k is []
Inherited g is []
m is hi
k is []

Process finished with exit code 0

NO ERRORS!

It seems like a bug to me, but what is the expected behavior in this case? I would expect the first case to not error out and should have an empty list.

I've only tested this on Python 3.9 so far.

[simplecoder878]

Also wanted to add that I did try another variation of the first example where I set the default_factory field's init value to False and I got the same error.

from dataclasses import dataclass, field

@dataclass(init=False)
class TestObject(object):
    m: str = field(default='hi')
    k: list = field(init=False, default_factory=list)

    def test(self):
        print(f'm is {self.m} ')
        print(f'k is {self.k}')

if __name__ == '__main__':
    myobject = TestObject()
    myobject.test()

Also produces

Traceback (most recent call last):
  File "H:\unit_test\tests_dataclass.py", line 81, in <module>
    myobject.test()
  File "H:\unit_test\tests_dataclass.py", line 76, in test
    print(f'k is {self.k}')
AttributeError: 'TestObject' object has no attribute 'k'
m is hi

[sobolevn]

Right now dataclasses.py has this explanation: https://github.com/python/cpython/blame/07cf10bafc8f6e1fcc82c10d97d3452325fc7c04/Lib/dataclasses.py#L962-L966

         if isinstance(getattr(cls, f.name, 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.
                delattr(cls, f.name)
            else:
                setattr(cls, f.name, f.default)

This is why: imagine, that we execute .default_factory there.
It will be vulnerable to "default mutable" problem:

from dataclasses import dataclass, field

@dataclass(init=False)
class TestObject(object):
    m: str = field(default='hi')
    k: list = field(default_factory=list)

    def test(self):
        print(f'm is {self.m} ')
        self.k.append(1)
        print(f'k is {self.k}')

if __name__ == '__main__':
    myobject = TestObject()
    print(TestObject.m)  # hi
    print(TestObject.k)  # []

    myobject.test()
    # m is hi 
    # k is [1]

    other_object = TestObject()
    other_object.test()
    # m is hi 
    # k is [1, 1]

Another, more complex solution is to track fields with default_factory and still generate __init__ / __new__ / etc for them to run their default_factoriess when object is created.

[akulakov]

I think a good possible solution is to raise an error if default_factory is provided on a init=False dataclass that doesn't have a __init__() defined. However, it will create a slight inconsistency because there will be an error when __init__ is not defined, but no error otherwise -- to allow calling the factory in custom defined __init__().

The error would be "Error: default_factory argument needs init=True to be set on dataclass because default_factory value is created in the generated init() method".