I'm running 2.3.4

I was reading the documentation for classes & types
       http://www.python.org/2.2.3/descrintro.html
And stumbled on this paragraph:

"""
__new__ must return an object. There's nothing that requires that it return a
new object that is an instance of its class argument, although that is the
convention. If you return an existing object, the constructor call will still
call its __init__ method. If you return an object of a different class, its
__init__ method will be called.
"""

The quote implies that when I call carol, b.__init__ should be called.
However, this does not seem to be the case (see code below). What am I not
understanding? Shouldn't the interpreter call b.__init__ when b is returned
from carol.__new__?

James

py> class bob(object):
...   def __init__(self):
...     print self.x
...   x = 2
...
py> class carol(object):
...   def __new__(cls):
...     return b
...
py> b=bob()
py> b.x
2
py> c = carol()   # should print "2"
py> c
<__main__.bob object at 0x404333cc>

--
James Stroud
UCLA-DOE Institute for Genomics and Proteomics
Box 951570
Los Angeles, CA 90095

Code:

    class bob(object):
        def __init__(self):
            print self.x
        x = 2

    class carol(object):
        def __new__(cls):
            return b
    b=bob()
    print b.x
    c = carol()
    c

Output:

This code produces the same output:

    class bob(object):
        def __init__(self):
            print self.x
        x = 2
    ##    class carol(object):
    ##        def __new__(cls):
    ##            return b
    b=bob()
    print b.x
    ##    c = carol()
    ##c

"""
Each new-style class has a static method named __new__. When you call
C(*args, **kwds) to create a new instance of a new-style class C, Python
invokes C.__new__(C, *args, **kwds). Python uses __new__'s return value x
as the newly created instance. Then, Python calls C.__init__(x, *args,
**kwds), but only when x is indeed an instance of C (otherwise, x's state
is as __new__ had left it). Thus, for a new-style class C, the statement
x=C(23) is equivlent to the following code:

    x = C.__new__(C, 23)
    if isinstance(x, C): C.__init__(x, 23)
"""

If the following code says what I think it does

class A(object):
    def __init__(self):
        print "init A"
    def __new__(cls):
        return b

class B(A):
    def __init__(self):
        print "init B"

b = object.__new__(B)

print "---"
A()              # prints 'init B'
A.__init__(b)    # prints 'init A'
print "---"
b = 42
A()              # prints nothing

the "Nutshell" example should be changed to

    x = C.__new__(C, 23)
    if isinstance(x, C): x.__init__(23)

to comply with the current implementation (I used Python 2.4).

class B(A):
    def __init__(self):
        print "init B"
        def f(): print 'from instance'
        self.__init__ = f

you'll see that while b.__init__() would print 'from instance',
instantiating A will not.  I think the right correction to the Nutshell
is therefore:

     x = C.__new__(C, 23)
     if isinstance(x, C): type(x).__init__(x, 23)

and this is how I plan to have it in the 2nd edition.

Anyway, the current docs corrected this mistake[1]

"""
If __new__() returns an instance of cls, then the new instance's
__init__() method will be invoked like "__init__(self[, ...])", where
self is the new instance and the remaining arguments are the same as
were passed to __new__().

If __new__() does not return an instance of cls, then the new instance's
__init__() method will not be invoked.
"""

[1]http://docs.python.org/ref/customization.html