0

As by the following example, the id of the deep-copied objects changes, but not if deepcopy is applied to an object of objects. Why?


import copy
import numpy

class A:
    data = numpy.array([])

    def __init__(self, numbers):
        self.data = numpy.array(numbers)
class B:
    As = []
    
    def __init__(self,*many_A):
        for a in many_A:
            self.As.append(a)
            
A1 = A([1,2,3])
print(["A1:",id(A1.data)]) # OK

A2 = A([4,5,6])
print(["A2:",id(A2.data)]) # Also OK

A3 = copy.deepcopy(A2)
print(["A3:",id(A3.data)]) # Ok, deepcopy changes id!

B1 = B(A1,A2)
print(["B1.A[0]:",id(B1.As[0].data)]) # ok, same id as A1
print(["B1.A[1]:",id(B1.As[1].data)]) # ok, same id as A2

B2 = copy.deepcopy(B1)
print(["B2.A[0]:",id(B2.As[0].data)]) # wtf???? why does deepcopy not change id's here?
print(["B2.A[1]:",id(B2.As[1].data)]) # wtf????

The output of the code:

['A1:', 1557465682928]
['A2:', 1557465683120]
['A3:', 1557465682832]
['B1.A[0]:', 1557465682928]
['B1.A[1]:', 1557465683120]
['B2.A[0]:', 1557465682928]
['B2.A[1]:', 1557465683120]

This is very strange behaviour. Deepcopy knows how to deepcopy A, but not B, which only consists of a list of A's.

Steffen Wolp
  • 31
  • 5
  • 1
    ``B.As``, ``B1.As`` and ``B2.As`` all are the same list *owned by the class*. ``copy.deepcopy`` merely copies *instance* data. You can check that ``B.As`` contains the ``A``s, yet ``B1.__dict__`` and ``B2.__dict__`` are empty. – MisterMiyagi Feb 16 '21 at 14:29
  • `As = []` is your culprit. if you move it into the `__init__` it will be fine. be careful using class level vs instance level parameters – Nullman Feb 16 '21 at 14:32

0 Answers0