Question

I want to know that if I have class like this:

class TestClass(object):
    def __init__(self):
        self.a = 20
        self.b = 30
    
obj = TestClass()

When I write obj.a, which of the following is called first?

• TestClass.__getattribute__("a")
• TestClass.__getattr__("a")
• getattr(obj, "a")
• obj.__dict__['a']

I have a similar question for setattr

According to Python 2.7 docs:

object._getattr_(self, name)

Called when an attribute lookup has not found the attribute in the usual places (i.e. it is not an instance attribute nor is it found in the class tree for self). name is the attribute name. This method should return the (computed) attribute value or raise an AttributeError exception.

It says "not found […] in usual places". What are the "usual places". I want to know when __getattr__ is called.

Also, what is the difference between __getattr__ and __getattribute__?

Can anyone give me example where all of these are used?

Answer 1

It’s a bit complicated. Here’s the sequence of checks Python does if you request an attribute of an object.

First, Python will check if the object’s class has a __getattribute__ method. If it doesn’t have one defined, it will inherit object.__getattribute__ which implements the other ways of finding the attribute’s values.

The next check is in the object’s class’s __dict__. However, even if a value is found there, it may not be the result of the attribute lookup! Only "data descriptors" will take precedence if found here. The most common data descriptor is a property object, which is a wrapper around a function that will be called each time an attribute is accessed. You can create a property with a decorator:

class foo(object):
    @property
    def myAttr(self):
        return 2

In this class, myAttr is a data descriptor. That simply means that it implements the descriptor protocol by having both __get__ and __set__ methods. A property is a data descriptor.

If the class doesn’t have anything in its __dict__ with the requested name, object.__getattribute__ searches through its base classes (following the MRO) to see if one is inherited. An inherited data descriptor works just like one in the object’s class.

If a data descriptor was found, its __get__ method is called and the return value becomes the value of the attribute lookup. If an object that is not a data descriptor was found, it is held on to for a moment, but not returned just yet.

Next, the object’s own __dict__ is checked for the attribute. This is where most normal member variables are found.

If the object’s __dict__ didn’t have anything, but the earlier search through the class (or base classes) found something other than a data descriptor, it takes next precedence. An ordinary class variable will be simply returned, but "non-data descriptors" will get a little more processing.

A non-data descriptor is an object with a __get__ method, but no __set__ method. The most common kinds of non-data descriptors are functions, which become bound methods when accessed as a non-data descriptor from an object (this is how Python can pass the object as the first argument automatically). The descriptor’s __get__ method will be called and it’s return value will be the result of the attribute lookup.

Finally, if none of the previous checks succeeded, __getattr__ will be called, if it exists.

Here are some classes that use steadily increasing priority attribute access mechanisms to override the behavior of their parent class:

class O1(object):
    def __getattr__(self, name):
        return "__getattr__ has the lowest priority to find {}".format(name)

class O2(O1):
    var = "Class variables and non-data descriptors are low priority"
    def method(self): # functions are non-data descriptors
        return self.var

class O3(O2):
    def __init__(self):
        self.var = "instance variables have medium priority"
        self.method = lambda: self.var # doesn't recieve self as arg

class O4(O3):
    @property # this decorator makes this instancevar into a data descriptor
    def var(self):
        return "Data descriptors (such as properties) are high priority"

    @var.setter # I'll let O3's constructor set a value in __dict__
    def var(self, value):
        self.__dict__["var"]  = value # but I know it will be ignored

class O5(O4):
    def __getattribute__(self, name):
        if name in ("magic", "method", "__dict__"): # for a few names
            return super(O5, self).__getattribute__(name) # use normal access
        
        return "__getattribute__ has the highest priority for {}".format(name)

And, a demonstration of the classes in action:

O1 (__getattr__):

>>> o1 = O1()
>>> o1.var
'__getattr__ has the lowest priority to find var'

O2 (class variables and non-data descriptors):

>>> o2 = O2()
>>> o2.var
'Class variables and non-data descriptors are low priority'
>>> o2.method
<bound method O2.method of <__main__.O2 object at 0x000000000338CD30>>
>>> o2.method()
'Class variables and non-data descriptors are low priority'

O3 (instance variables, including a locally overridden method):

>>> o3 = O3()
>>> o3.method
<function O3.__init__.<locals>.<lambda> at 0x00000000034AAEA0>
>>> o3.method()
'instance variables have medium priority'
>>> o3.var
'instance variables have medium priority'

O4 (data descriptors, using the property decorator):

>>> o4 = O4()
>>> o4.method()
'Data descriptors (such as properties) are high priority'
>>> o4.var
'Data descriptors (such as properties) are high priority'
>>> o4.__dict__["var"]
'instance variables have medium priority'

O5 (__getattribute__):

>>> o5 = O5()
>>> o5.method
<function O3.__init__.<locals>.<lambda> at 0x0000000003428EA0>
>>> o5.method()
'__getattribute__ has the highest priority for var'
>>> o5.__dict__["var"]
'instance variables have medium priority'
>>> o5.magic
'__getattr__ has the lowest priority to find magic'