Creating a deepcopy of class instance with nested weakref to it

问题

I have two classes: a parent class and a container class. The parent class instance has matching container class instance as a weak reference.

There is a problem while deep copying the parent instance, the weakref is still linking to the original instance. Here is a minimal example:

import weakref
from copy import deepcopy


class Container:    
    def __init__(self, parent):
        self.parent = weakref.ref(parent)


class Parent:
    def __init__(self):
        self.container = Container(self)


if __name__ == '__main__':
    parent1 = Parent()
    assert(parent1 is parent1.container.parent())

    parent2 = deepcopy(parent1)
    assert(parent2 is parent2.container.parent())

The second assertion fails.

I suspect that the __deepcopy__ magic method can be implemented, but not sure how exactly.

回答1:

The problem is that deepcopy won't follow the weakref.ref link. It doesn't even copy the weakref.ref:

>>> from copy import deepcopy
>>> import weakref
>>>
>>> parent1 = Parent()
>>> ref1 = weakref.ref(parent1)
>>> ref2 = deepcopy(ref1)
>>> ref1 is ref2
True

That is explicitly hardcoded in the copy module. I don't know why this is but I suspect they had their reasons.

However you can implement a __deepcopy__ method:

import weakref
from copy import deepcopy

class Container:    
    def __init__(self, parent):
        self.parent = weakref.ref(parent)

class Parent:
    def __init__(self):
        self.container = Container(self)

    def __deepcopy__(self, memo):
        # set __deepcopy__ element to "false"-ey value so we don't go into
        # recusion.
        self.__deepcopy__ = None
        try:
            new = deepcopy(self, memo)
        finally:
            # Always delete the self.__deepcopy__ again, even if deepcopying failed
            del self.__deepcopy__
        del new.__deepcopy__  # remove the copied __deepcopy__ attribute
        new.container.parent = weakref.ref(new)
        return new

if __name__ == '__main__':
    parent1 = Parent()
    assert parent1 is parent1.container.parent()

    parent2 = deepcopy(parent1)
    assert parent2 is parent2.container.parent()

    parent3 = deepcopy(parent2)
    assert parent3 is parent3.container.parent()

It's a bit ugly because of the temporary __deepcopy__ instance attribute. But it allows to use the normal deepcopy function on self without going into infinite recursions and then you only have to manually create a new weak reference to your parent.

You could even do it without temporarily setting __deepcopy__ (it should work correctly):

import weakref
from copy import deepcopy

class Container:    
    def __init__(self, parent):
        self.parent = weakref.ref(parent)

class Parent:
    def __init__(self):
        self.container = Container(self)

    def __deepcopy__(self, memo):
        # Create a new class
        new = object.__new__(type(self))
        memo[id(self)] = new   # add the new class to the memo
        # Insert a deepcopy of all instance attributes
        new.__dict__.update(deepcopy(self.__dict__, memo))
        # Manually update the weakref to be correct
        new.container.parent = weakref.ref(new)
        return new

if __name__ == '__main__':
    parent1 = Parent()
    assert parent1 is parent1.container.parent()

    parent2 = deepcopy(parent1)
    assert parent2 is parent2.container.parent()

    parent3 = deepcopy(parent2)
    assert parent3 is parent3.container.parent()

来源：https://stackoverflow.com/questions/46203335/creating-a-deepcopy-of-class-instance-with-nested-weakref-to-it