How to create a class instance without calling initializer?

Question

Is there any way to avoid calling __init__ on a class while initializing it, such as from a class method?

I am trying to create a case and punctuation insensitive string class in Python used for efficient comparison purposes but am having trouble creating a new instance without calling __init__.

>>> class String:

    def __init__(self, string):
        self.__string = tuple(string.split())
        self.__simple = tuple(self.__simple())

    def __simple(self):
        letter = lambda s: ''.join(filter(lambda s: 'a' <= s <= 'z', s))
        return filter(bool, map(letter, map(str.lower, self.__string)))

    def __eq__(self, other):
        assert isinstance(other, String)
        return self.__simple == other.__simple

    def __getitem__(self, key):
        assert isinstance(key, slice)
        string = String()
        string.__string = self.__string[key]
        string.__simple = self.__simple[key]
        return string

    def __iter__(self):
        return iter(self.__string)

>>> String('Hello, world!')[1:]
Traceback (most recent call last):
  File "<pyshell#2>", line 1, in <module>
    String('Hello, world!')[1:]
  File "<pyshell#1>", line 17, in __getitem__
    string = String()
TypeError: __init__() takes exactly 2 positional arguments (1 given)
>>> 

What should I replace string = String(); string.__string = self.__string[key]; string.__simple = self.__simple[key] with to initialize the new object with the slices?

EDIT:

As inspired by the answer written below, the initializer has been edited to quickly check for no arguments.

def __init__(self, string=None):
    if string is None:
        self.__string = self.__simple = ()
    else:
        self.__string = tuple(string.split())
        self.__simple = tuple(self.__simple())

Answer 1

A trick the standard pickle and copy modules use is to create an empty class, instantiate the object using that, and then assign that instance's __class__ to the "real" class. e.g.

>>> class MyClass(object):
...     init = False
...     def __init__(self):
...         print 'init called!'
...         self.init = True
...     def hello(self):
...         print 'hello world!'
... 
>>> class Empty(object):
...     pass
... 
>>> a = MyClass()
init called!
>>> a.hello()
hello world!
>>> print a.init
True
>>> b = Empty()
>>> b.__class__ = MyClass
>>> b.hello()
hello world!
>>> print b.init
False

But note, this approach is very rarely necessary. Bypassing the __init__ can have some unexpected side effects, especially if you're not familiar with the original class, so make sure you know what you're doing.

Answer 2

When feasible, letting __init__ get called (and make the call innocuous by suitable arguments) is preferable. However, should that require too much of a contortion, you do have an alternative, as long as you avoid the disastrous choice of using old-style classes (there is no good reason to use old-style classes in new code, and several good reasons not to)...:

   class String(object):
      ...

   bare_s = String.__new__(String)

This idiom is generally used in classmethods which are meant to work as "alternative constructors", so you'll usually see it used in ways such as...:

@classmethod 
def makeit(cls):
    self = cls.__new__(cls)
    # etc etc, then
    return self

(this way the classmethod will properly be inherited and generate subclass instances when called on a subclass rather than on the base class).

Answer 3

Using a metaclass provides a nice solution in this example. The metaclass has limited use but works fine.

>>> class MetaInit(type):

    def __call__(cls, *args, **kwargs):
        if args or kwargs:
            return super().__call__(*args, **kwargs)
        return cls.__new__(cls)

>>> class String(metaclass=MetaInit):

    def __init__(self, string):
        self.__string = tuple(string.split())
        self.__simple = tuple(self.__simple())

    def __simple(self):
        letter = lambda s: ''.join(filter(lambda s: 'a' <= s <= 'z', s))
        return filter(bool, map(letter, map(str.lower, self.__string)))

    def __eq__(self, other):
        assert isinstance(other, String)
        return self.__simple == other.__simple

    def __getitem__(self, key):
        assert isinstance(key, slice)
        string = String()
        string.__string = self.__string[key]
        string.__simple = self.__simple[key]
        return string

    def __iter__(self):
        return iter(self.__string)

>>> String('Hello, world!')[1:]
<__main__.String object at 0x02E78830>
>>> _._String__string, _._String__simple
(('world!',), ('world',))
>>> 

Addendum:

After six years, my opinion favors Alex Martelli's answer more than my own approach. With meta-classes still on the mind, the following answer shows how the problem can be solved both with and without them:

#! /usr/bin/env python3
METHOD = 'metaclass'


class NoInitMeta(type):
    def new(cls):
        return cls.__new__(cls)


class String(metaclass=NoInitMeta if METHOD == 'metaclass' else type):
    def __init__(self, value):
        self.__value = tuple(value.split())
        self.__alpha = tuple(filter(None, (
            ''.join(c for c in word.casefold() if 'a' <= c <= 'z') for word in
            self.__value)))

    def __str__(self):
        return ' '.join(self.__value)

    def __eq__(self, other):
        if not isinstance(other, type(self)):
            return NotImplemented
        return self.__alpha == other.__alpha

    if METHOD == 'metaclass':
        def __getitem__(self, key):
            if not isinstance(key, slice):
                raise NotImplementedError
            instance = type(self).new()
            instance.__value = self.__value[key]
            instance.__alpha = self.__alpha[key]
            return instance
    elif METHOD == 'classmethod':
        def __getitem__(self, key):
            if not isinstance(key, slice):
                raise NotImplementedError
            instance = self.new()
            instance.__value = self.__value[key]
            instance.__alpha = self.__alpha[key]
            return instance

        @classmethod
        def new(cls):
            return cls.__new__(cls)
    elif METHOD == 'inline':
        def __getitem__(self, key):
            if not isinstance(key, slice):
                raise NotImplementedError
            cls = type(self)
            instance = cls.__new__(cls)
            instance.__value = self.__value[key]
            instance.__alpha = self.__alpha[key]
            return instance
    else:
        raise ValueError('METHOD did not have an appropriate value')

    def __iter__(self):
        return iter(self.__value)


def main():
    x = String('Hello, world!')
    y = x[1:]
    print(y)


if __name__ == '__main__':
    main()