copy.deepcopy raises TypeError on objects with self-defined __new__() method

Question

I want to implement a symbol type, which keeps track of the symbols we already have(saved in _sym_table), and return them if they exist, or create new ones otherwise. The code:

# -*- coding: utf-8 -*-

_sym_table = {}

class Symbol(object):
    def __new__(cls, sym):
        if sym not in _sym_table:
            return super().__new__(cls)
        else:
            return _sym_table[sym]

    def __init__(self, sym):
        self.sym = sym
        _sym_table[sym] = self

    def __str__(self):
        return self.sym

    def __cmp__(self, other):
        return self is other

    def __hash__(self):
        return self.sym.__hash__()

But when I call copy.deepcopy on a list of such Symbol instances, exception is raised:

a = Symbol('a')
b = Symbol('b')
s = [a, b]
t = copy.deepcopy(s)

Error messages:

Traceback (most recent call last):
  File "xxx.py", line 7, in <module>
    t = copy.deepcopy(s)
  File "/usr/lib/python3.2/copy.py", line 147, in deepcopy
    y = copier(x, memo)
  File "/usr/lib/python3.2/copy.py", line 209, in _deepcopy_list
    y.append(deepcopy(a, memo))
  File "/usr/lib/python3.2/copy.py", line 174, in deepcopy
    y = _reconstruct(x, rv, 1, memo)
  File "/usr/lib/python3.2/copy.py", line 285, in _reconstruct
    y = callable(*args)
  File "/usr/lib/python3.2/copyreg.py", line 88, in __newobj__
    return cls.__new__(cls, *args)
TypeError: __new__() takes exactly 2 arguments (1 given)

So my questions are:

• How can I make a deep copy on these objects with self-defined __new__ methods?
• And any suggestions about when and how to use copy.deepcopy?

Thanks a lot!

Answer 1

one problem is that deepcopy and copy have no way of knowing which arguments to pass to __new__, therefore they only work with classes that don't require constructor arguments.

the reason why you can have __init__ arguments is that __init__ isn't called when copying an object, but __new__ must be called to create the new object.

so if you want to control copying, you'll have to define the special __copy__ and __deepcopy__ methods:

def __copy__(self):
    return self

def __deepcopy__(self, memo):
    return self

by the way, singletons are evil and not really needed in python.

Answer 2

Seems to me you want the Symbol instances to be singletons. Deepcopy, however is supposed to be used when you want an exact copy of an instance, i.e. a different instance that is equal to the original.

So the usage here kinda contradicts the purpose of deepcopy. If you want to make it work anyhow, you can define the __deepcopy__ method on Symbol.

Answer 3

Define __getnewargs__ — that way you will not only be able to copy and deepcopy, but you'll also be able to pickle.