difference between variables inside and outside of __init__() (class and instance attributes)

Variable set outside __init__ belong to the class. They're shared by all instances.

Variables created inside __init__ (and all other method functions) and prefaced with self. belong to the object instance.

Without Self

Create some objects:

class foo(object):
    x = 'original class'

c1, c2 = foo(), foo()

I can change the c1 instance, and it will not affect the c2 instance:

c1.x = 'changed instance'
c2.x
>>> 'original class'

But if I change the foo class, all instances of that class will be changed as well:

foo.x = 'changed class'
c2.x
>>> 'changed class'

Please note how Python scoping works here:

c1.x
>>> 'changed instance'

With Self

Changing the class does not affect the instances:

class foo(object):
    def __init__(self):
        self.x = 'original self'

c1 = foo()
foo.x = 'changed class'
c1.x
>>> 'original self'

I would like to add something to the responses that I read in this thread and this thread (which references this one).

Disclaimer: this remarks come from the experiments I ran

Variables outside __init__:

These are, in fact, static class variables and are, therefore, accesible to all instances of the class.

Variables inside __init__:

The value of these instance variables are only accesible to the instance at hand (through the self reference)

My contribution:

One thing that programmers must consider when using static class variables is that they can be shadowed by instance variables (if you are accessing the static class variables through the self reference).

Explanation:

Previously, I thought that both ways of declaring the variables were exactly the same (silly me), and that was partly because I could access both kind of variables through the self reference. It was now, when I ran into trouble, that I researched the topic and cleared it up.

The problem with accessing static class variables through the self reference is that it only references the static class variable if there is no instance variable with the same name, and to make things worse, trying to redefine a static class variable through the self reference does not work because an instance variable is created which then shadows the previously-accesible static class variable.

To get around this problem, you should always reference static class variables through the name of the class.

Example:

#!/usr/bin/env python

class Foo:
    static_var = 'every instance has access'

    def __init__(self,name):
        self.instance_var = 'I am %s' % name

    def printAll(self):
        print 'self.instance_var = %s' % self.instance_var
        print 'self.static_var = %s' % self.static_var
        print 'Foo.static_var = %s' % Foo.static_var

f1 = Foo('f1')

f1.printAll()

f1.static_var = 'Shadowing static_var'

f1.printAll()

f2 = Foo('f2')

f2.printAll()

Foo.static_var = 'modified class'

f1.printAll()
f2.printAll()

Output:

self.instance_var = I am f1
self.static_var = every instance has access
Foo.static_var = every instance has access
self.instance_var = I am f1
self.static_var = Shadowing static_var
Foo.static_var = every instance has access
self.instance_var = I am f2
self.static_var = every instance has access
Foo.static_var = every instance has access
self.instance_var = I am f1
self.static_var = Shadowing static_var
Foo.static_var = modified class
self.instance_var = I am f2
self.static_var = modified class
Foo.static_var = modified class

I hope this is helpful to someone

further to S.Lott's reply, class variables get passed to metaclass new method and can be accessed through the dictionary when a metaclass is defined. So, class variables can be accessed even before classes are created and instantiated.

for example:

class meta(type):
    def __new__(cls,name,bases,dicto):
          # two chars missing in original of next line ...
          if dicto['class_var'] == 'A':
             print 'There'
class proxyclass(object):
      class_var = 'A'
      __metaclass__ = meta
      ...
      ...