default_factory in dataclasses behaves strange when set to a callable object instead of a function

[frnhr]

Initial Checks

• I confirm that I'm using Pydantic V2

Description

There seems to be an unexpected difference with default_factory with pydantic.dataclasses in a very specific case.

It happens in this case:

• Using nested dataclasses (a parent and a child).
• A field in the parent and a field in the child dataclass have default_factory set to the same callable object.
• This callable is a class instance and not a function.
• Initializing parent and child objects in a single call by passing child data as dict.

It seems that the callable object (used for default_factory) gets copied (duplicated) at some point before being called. This can cause some unexpected behaviour, as shown in the tests below.

Example Code

import dataclasses

import pydantic


# First, we have a dead-simple auto-increment int class:


class UniqueId:
    """
    >>> unique_id = UniqueId()
    >>> unique_id()
    1
    >>> unique_id()
    2
    >>> another_unique_id = UniqueId()
    >>> another_unique_id()
    1
    >>> another_unique_id()
    2
    >>> unique_id()
    3
    >>> another_unique_id()
    3
    """
    def __init__(self):
        self.counter = 0

    def __call__(self) -> int:
        self.counter += 1
        return self.counter


# Here is the test case which fails:


def test_1():
    unique_id = UniqueId()

    @pydantic.dataclasses.dataclass
    class User:
        name: str
        id_: int = dataclasses.field(default_factory=unique_id)

    @pydantic.dataclasses.dataclass
    class Group:
        name: str
        users: list[User]
        id_: int = dataclasses.field(default_factory=unique_id)

    group = Group(name="Group 1", users=[{"name": "John Smith"}, {"name": "John Doe"}])

    assert group.users[0].id_ == 1
    assert group.users[1].id_ == 2
    assert group.id_ == 3  # Fails: 1 != 3


# Below are some variations of this test.
# All tests below are OK!


def test_2():
    unique_id = UniqueId()

    @pydantic.dataclasses.dataclass
    class User:
        name: str
        id_: int = dataclasses.field(default_factory=unique_id)

    @pydantic.dataclasses.dataclass
    class Group:
        name: str
        users: list[User]
        id_: int = dataclasses.field(default_factory=unique_id)

    group = Group(name="Group 1", users=[User(name="John Smith"), User(name="John Doe")])
    #  ^^^^^ Change in this line, users are now inited explicitly, not dicts ^^^^^

    assert group.users[0].id_ == 1
    assert group.users[1].id_ == 2
    assert group.id_ == 3  # OK!



def test_3():
    unique_id = UniqueId()

    @pydantic.dataclasses.dataclass
    class User:
        name: str
        id_: int = dataclasses.field(default_factory=lambda: unique_id())
        #  ^^^^^ Change in this line, using a lambda ^^^^^

    @pydantic.dataclasses.dataclass
    class Group:
        name: str
        users: list[User]
        id_: int = dataclasses.field(default_factory=unique_id)

    group = Group(name="Group 1", users=[{"name": "John Smith"}, {"name": "John Doe"}])

    assert group.users[0].id_ == 1
    assert group.users[1].id_ == 2
    assert group.id_ == 3  # OK!


def test_4():
    unique_id_instance = UniqueId()

    def unique_id():
        return unique_id_instance()

    # ^^^^^ Changes above, using a named function instead of a class instance ^^^^^

    @pydantic.dataclasses.dataclass
    class User:
        name: str
        id_: int = dataclasses.field(default_factory=unique_id)

    @pydantic.dataclasses.dataclass
    class Group:
        name: str
        users: list[User]
        id_: int = dataclasses.field(default_factory=unique_id)

    group = Group(name="Group 1", users=[{"name": "John Smith"}, {"name": "John Doe"}])

    assert group.users[0].id_ == 1
    assert group.users[1].id_ == 2
    assert group.id_ == 3  # OK!

Python, Pydantic & OS Version

pydantic version: 2.5.3
        pydantic-core version: 2.14.6
          pydantic-core build: profile=release pgo=true
                 install path: /Users/fran/.pyenv/versions/3.11.4/envs/someenvname/lib/python3.11/site-packages/pydantic
               python version: 3.11.4 (main, Sep 28 2023, 13:58:48) [Clang 14.0.3 (clang-1403.0.22.14.1)]
                     platform: macOS-14.1-arm64-arm-64bit
             related packages: mypy-1.6.1 typing_extensions-4.8.0

[sydney-runkle]

@frnhr,

Thanks for reporting this. Definitely a bug! We'll look into fixing this 🐛.

[sydney-runkle]

Such an odd bug. Fixed by #9114.

[sydney-runkle]

Fix should be released in 2.7 soon:

import dataclasses

import pydantic

class UniqueId:
    """
    >>> unique_id = UniqueId()
    >>> unique_id()
    1
    >>> unique_id()
    2
    >>> another_unique_id = UniqueId()
    >>> another_unique_id()
    1
    >>> another_unique_id()
    2
    >>> unique_id()
    3
    >>> another_unique_id()
    3
    """
    def __init__(self):
        self.counter = 0

    def __call__(self) -> int:
        self.counter += 1
        return self.counter

unique_id = UniqueId()

@pydantic.dataclasses.dataclass
class User:
    name: str
    id_: int = dataclasses.field(default_factory=unique_id)

@pydantic.dataclasses.dataclass
class Group:
    name: str
    users: list[User]
    id_: int = dataclasses.field(default_factory=unique_id)

group = Group(name="Group 1", users=[{"name": "John Smith"}, {"name": "John Doe"}])

assert group.users[0].id_ == 1
assert group.users[1].id_ == 2
assert group.id_ == 3