http://www.cafepy.com/article/python_attributes_and_methods/python_attributes_and_methods.html
Before You Begin
Some points you should note:
-
This book covers the new-style objects (introduced a long time ago inPython 2.2). Examples are valid for Python 2.5 and all the way to Python3.x.
-
This book is not for absolute beginners. It is for people who alreadyknow Python (some Python at least), and want to know more.
-
You should be familiar with the different kinds of objects inPython and not be confused when you come across the termtype where you expectedclass. You can read the first part of this seriesfor background information -Python Types and Objects.
Happy pythoneering!
What is an attribute? Quite simply, an attribute is a way to get fromone object to another. Apply the power of the almighty dot -objectname.attributename
- and voila! you now havethe handle to another object. You also have the power to createattributes, by assignment: objectname.attributename =notherobject
.
Which object does an attribute access return, though? And where doesthe object set as an attribute end up? These questions are answered inthis chapter.
Example 1.1. Simple attribute access
>>> class C(object): ... classattr = "attr on class"... >>> cobj = C() >>> cobj.instattr = "attr on instance"
>>> >>> cobj.instattr
'attr on instance' >>> cobj.classattr
'attr on class' >>> C.__dict__['classattr']
'attr on class' >>> cobj.__dict__['instattr']
'attr on instance' >>> >>> cobj.__dict__
{'instattr': 'attr on instance'} >>> C.__dict__
{'classattr': 'attr on class', '__module__': '__main__', '__doc__': None}
Attributes can be set on a class. | |
Or even on an instance of a class. | |
Both, class and instance attributes are accessible from aninstance. | |
Attributes really sit inside a dictionary-like | |
|
Ok, I admit 'user-provided attribute' is a term I made up, but I thinkit is useful to understand what is going on. Notethat __dict__
is itself an attribute. We didn't setthis attribute ourselves, but Python provides it. Our old friends__class__
and __bases__
(nonewhich appear to be in __dict__
either) also seem tobe similar. Let's call them Python-providedattributes. Whether an attribute is Python-provided or not depends onthe object in question (__bases__
, for example, isPython-provided only for classes).
We, however, are more interested in user-definedattributes. These are attributes provided by the user, and theyusually (but not always) end up in the __dict__
ofthe object on which they're set.
When accessed (for e.g. printobjectname.attributename
), the following objects aresearched in sequence for the attribute:
-
The object itself(
objectname.__dict__
or anyPython-provided attribute ofobjectname
). -
The object's type(
objectname.__class__.__dict__
). Observe that only__dict__
is searched, which means onlyuser-provided attributes of the class. In otherwordsobjectname.__bases__
may not return anythingeven thoughobjectname.__class__.__bases__
doesexist. -
The bases of the object's class, their bases, and soon. (
__dict__
of each ofobjectname.__class__.__bases__
). More than one basedoes not confuse Python, and should not concern us at the moment. Thepoint to note is that all bases are searched until an attribute isfound.
If all this hunting around fails to find a suitably named attribute,Python raises an AttributeError
. The type of thetype (objectname.__class__.__class__
) is neversearched for attribute access on an object(objectname
in the example).
The built-in dir()
function returns a list ofall attributes of an object. Also look atthe inspectmodule in the standard library for more functions to inspectobjects.
The above section explains the general mechanism forall objects. Even for classes (for exampleaccessing classname.attrname
), with a slightmodification: the bases of the class are searched before theclass of the class (which isclassname.__class__
and for most types, by theway, is <type 'type'>
).
Some objects, such as built-in types and their instances (lists,tuples, etc.) do not have a __dict__
. Consequentlyuser-defined attributes cannot be set on them.
We're not done yet! This was the short version of thestory. There is more to what can happen when setting and gettingattributes. This is explored in the following sections.
Continuing our Python experiments:
Example 1.2. A function is more
>>> class C(object): ... classattr = "attr on class" ... def f(self): ... return "function f" ... >>> C.__dict__{'classattr': 'attr on class', '__module__': '__main__', '__doc__': None, 'f': <function f at 0x008F6B70>} >>> cobj = C() >>> cobj.classattr is C.__dict__['classattr']
True >>> cobj.f is C.__dict__['f']
False >>> cobj.f
<bound method C.f of <__main__.C instance at 0x008F9850>> >>> C.__dict__['f'].__get__(cobj, C)
<bound method C.f of <__main__.C instance at 0x008F9850>>
Two innocent looking class attributes, a string 'classattr' and a function 'f'. | |
Accessing the string really gets it from the class's | |
Not so for the function! Why? | |
Hmm, it does look like a different object. (A bound method is acallable object that calls a function ( | |
Here's the spoiler - this is what Python did to create the boundmethod. While looking for an attribute for an instance, if Pythonfinds an object with a |
It is only the presence of the __get__()
methodthat transforms an ordinary function into a boundmethod. There is nothing really special about a functionobject. Anyone can put objects with a __get__()
method inside the class __dict__
and get away withit. Such objects are called descriptors and havemany uses.
Any object with a __get__()
method, and optionally__set__()
and __delete__()
methods, accepting specific parameters is said to follow thedescriptor protocol. Such an object qualifies asa descriptor and can be placed inside a class's__dict__
to do something special when an attributeis retrieved, set or deleted. An empty descriptor is shown below.
Called when attribute is read (eg. | |
Called when attribute is set on an instance(eg. | |
Called when attribute is deleted from an instance(eg. |
What we defined above is a class that can be instantiated to create adescriptor. Let's see how we can create a descriptor, attach it to aclass and put it to work.
Note that when accessed from the class itself, only the__get__()
method comes in the picture, setting ordeleting the attribute will actually replace or remove thedescriptor.
Descriptors work only when attached to classes. Sticking adescriptor in an object that is not a class gives us nothing.
In the previous section we used a descriptor with both__get__()
and __set__()
methods. Such descriptors, by the way, are called datadescriptors. Descriptors with only the__get__()
method are somewhat weaker than theircousins, and called non-data descriptors.
Repeating our experiment, but this time with non-data descriptors, we get:
Calls | |
Puts | |
Surprise!This now returns | |
Deletes theattribute | |
These function identical to a data descriptor. |
Interestingly, not having a __set__()
affects notjust attribute setting, but also retrieval. What is Python thinking?If on setting, the descriptor gets fired and puts the data somewhere,then it follows that the descriptor only knows how to get it back. Whyeven bother with the instance's __dict__
?
Data descriptors are useful for providing fullcontrol over an attribute. This is what one usually wantsfor attributes used to store some piece of data. For example anattribute that gets transformed and savedsomewhere on setting, would usually bereverse-transformed and returned when read. Whenyou have a data descriptor, it controls all access (both read andwrite) to the attribute on an instance. Of course, you could stilldirectly go to the class and replace thedescriptor, but you can't do that from an instance of the class.
Non-data descriptors, in contrast, only provide a value when aninstance itself does not have a value. So setting the attribute on aninstance hides the descriptor. This isparticularly useful in the case of functions (which are non-datadescriptors) as it allows one to hide a function defined in the classby attaching one to an instance.
This is the long version of the attribute access story, includedjust for the sake of completeness.
When retrieving an attribute froman object (print objectname.attrname
) Python followsthese steps:
-
If
attrname
is a special(i.e. Python-provided) attribute forobjectname
,return it. -
Check
objectname.__class__.__dict__
forattrname
. If it exists and isa data-descriptor, return the descriptor result. Search all bases ofobjectname.__class__
for the same case. -
Check
objectname.__dict__
forattrname
, and return if found. Ifobjectname
is a class, search its bases too. If itis a class and a descriptor exists in it or its bases, return thedescriptor result. -
Check
objectname.__class__.__dict__
forattrname
. If it exists and is anon-data descriptor, return the descriptor result. If itexists, and is not a descriptor, just return it. If it exists and is adata descriptor, we shouldn't be here because we would have returnedat point 2. Search all bases ofobjectname.__class__
for samecase. -
Raise
AttributeError
Note that Python first checks for a datadescriptor in the class (and its bases), then for the attribute in theobject __dict__
, and then for anon-data descriptor in the class (and itsbases). These are points 2, 3 and 4 above.
The descriptor result above impliesthe result of calling the __get__()
method of thedescriptor with appropriate arguments. Also, checking a__dict__
for attrname
meanschecking if __dict__["attrname"]
exists.
Now, the steps Python follows when settinga user-defined attribute (objectname.attrname =something
):
-
Check
objectname.__class__.__dict__
forattrname
. If it exists and isa data-descriptor, use the descriptor to set the value. Search all bases ofobjectname.__class__
for the same case. -
Insert
something
intoobjectname.__dict__
for key"attrname"
. -
Think "Wow, this was much simpler!"
What happens when setting a Python-provided attribute depends onthe attribute. Python may not even allow some attributes to beset. Deletion of attributes is very similar to setting asabove.
Before you rush to the mall and get yourself some expensivedescriptors, note that Python ships with some very useful ones thatcan be found by simply looking in the box.
Example 1.7. Built-in descriptors
class HidesA(object): def get_a(self): return self.b - 1 def set_a(self, val): self.b = val + 1 def del_a(self): del self.b a = property(get_a, set_a, del_a, "docstring")def cls_method(cls): return "You called class %s" % cls clsMethod = classmethod(cls_method)
def stc_method(): return "Unbindable!" stcMethod = staticmethod(stc_method)
![]()
Aproperty provides an easy way to call functionswhenever an attribute is retrieved, set or deleted on theinstance. When the attribute is retrieved from the class, the gettermethod is not called but the property object itself is returned. Adocstring can also be provided which is accessible as | |
Aclassmethod is similar to a regular method,except that is passes the class (and not the instance) as the firstargument to the function. The remaining arguments are passed throughas usual. It can also be called directly on the class and it behavesthe same way. The first argument is named | |
Astaticmethod is just like a function outside theclass. It is never bound, which means no matterhow you access it (on the class or on an instance), it gets called withexactly the same arguments you pass. No object is inserted as thefirst argument. |
As we saw earlier, Python functions are descriptors too. They weren'tdescriptors in earlier versions of Python (as there were nodescriptors at all), but now they fit nicely into a more genericmechanism.
A property is always a data-descriptor, but not all arguments arerequired when defining it.
Can be set, retrieved, or deleted. | |
Attempting todelete this attribute from an instance will raise | |
Attempting toset or delete this attribute from an instance will raise |
The getter and setter functions need not be defined in the classitself, any function can be used. In any case, the functions will becalled with the instance as the first argument. Note thatwhere the functions are passed to the propertyconstructor above, they are not bound functions anyway.
Another useful observation would be to note that subclassing theclass and redefining the getter (or setter) functions is not going tochange the property. The property object is holdingon to the actual functions provided. When kicked, it isgoing to say "Hey, I'm holding this function I was given, I'll justcall this and return the result.", and not "Hmm, let me look up thecurrent class for a method called 'get_a' andthen use that". If that is what one wants, then defining a newdescriptor would be useful. How would it work? Let's say it isinitialized with a string (i.e. the name of the method to call). Onactivation, it does a getattr()
for the method nameon the class, and use the method found. Simple!
Classmethods and staticmethods are non-data descriptors, and so can behidden if an attribute with the same name is setdirectly on the instance. If you are rolling your own descriptor (andnot using properties), it can be made read-only by giving it a__set__()
method but raisingAttributeError
in the method. This is how aproperty behaves when it does not have a setter function.
Why do we need Method Resolution Order? Let's say:
-
We're happily doing OO programming and building aclass hierarchy.
-
Our usual technique to implement the
do_your_stuff()
method is to first calldo_your_stuff()
on the base class, and then doour stuff.Example 2.1. Usual base call technique
class A(object): def do_your_stuff(self): # do stuff with self for A return class B(A): def do_your_stuff(self): A.do_your_stuff(self) # do stuff with self for B return class C(A): def do_your_stuff(self): A.do_your_stuff(self) # do stuff with self for C return
-
We subclass a new class from two classes and end uphaving the same superclass being accessible through two paths.
Example 2.2. Base call technique fails
class D(B,C): def do_your_stuff(self): B.do_your_stuff(self) C.do_your_stuff(self) # do stuff with self for D return
-
Now we're stuck if we want to implement
do_your_stuff()
. Using our usual technique, if wewant to call bothB
andC
, weend up callingA.do_your_stuff()
twice. And we allknow it might be dangerous to haveA
do its stufftwice, when it is only supposed to be done once. The other optionwould leave eitherB
's stuff orC
's stuff not done, which is not what we wanteither.
There are messy solutions to this problem, and clean ones. Python,obviously, implements a clean one which is explained in the next section.
Let's say:
-
For each class, we arrange allsuperclasses into an ordered list without repetitions, and insert theclass itself at the start of the list. We put this list in an classattribute called
next_class_list
for our uselater.Example 2.3. Making a "Who's Next" list
B.next_class_list = [B,A] C.next_class_list = [C,A] D.next_class_list = [D,B,C,A]
-
We use a different technique to implement
do_your_stuff()
for our classes.Example 2.4. Call next method technique
class B(A): def do_your_stuff(self): next_class = self.find_out_whos_next() next_class.do_your_stuff(self) # do stuff with self for B def find_out_whos_next(self): l = self.next_class_list # l depends on the actual instance mypos = l.index(B)
# Find this class in the list return l[mypos+1] # Return the next one
The interesting part is how we
find_out_whos_next()
, which depends on whichinstance we are working with. Note that:-
Depending on whether we passed an instance of
D
or ofB
,next_class
above will resolve to eitherC
orA
. -
We have to implement
find_out_whos_next()
for each class, since it hasto have the class name hardcoded in it (seeabove). We cannot use
self.__class__
here. If we have calleddo_your_stuff()
on an instance ofD
, and the call is traversing up the hierarchy,thenself.__class__
will beD
here.
-
Using this technique, each method is called only once. Itappears clean, but seems to require too much work. Fortunately for us,we neither have to implement find_out_whos_next()
for each class, nor set the next_class_list
, asPython does both of these things.
Python provides a class attribute __mro__
foreach class, and a type called super
. The__mro__
attribute is a tuple containing the classitself and all of its superclasses without duplicates in a predictableorder. A super
object is used in place of thefind_out_whos_next()
method.
If we're using a class method, we don't have aninstance self
to pass into thesuper
call. Fortunately for us,super
works even with a class as the secondargument. Observe that above, super
usesself
only to get atself.__class__.__mro__
. The class can be passeddirectly tosuper
as shown below.
Example 2.6. Using super with a class method
class A(object): @classmethoddef say_hello(cls): print 'A says hello' class B(A): @classmethod def say_hello(cls): super(B, cls).say_hello()
print 'B says hello' class C(A): @classmethod def say_hello(cls): super(C, cls).say_hello() print 'C says hello' class D(B, C): @classmethod def say_hello(cls): super(D, cls).say_hello() print 'D says hello' B.say_hello()
D.say_hello()
![]()
This example is for classmethods (not instancemethods). | |
Note we pass | |
This prints out: A says hello | |
This prints out (observe each method is called only once): A says hello |
There is yet another way to use super
:
When created with only a type, the super
instancebehaves like a descriptor. This means (if d
is aninstance of D
) thatsuper(B).__get__(d)
returns the same thing assuper(B,d)
. In above, we munge anattribute name, similar to what Python does for names starting withdouble underscore inside the class. So this isaccessible as
self.__super
within the body of theclass. If we didn't use a class specific attribute name, accessing theattribute through the instance self
might return anobject defined in a subclass.
While using super
we typically use only onesuper
call in one method even if the class hasmultiple bases. Also, it is a good programming practice to usesuper
instead of calling methods directly on a baseclass.
A possible pitfall appears if do_your_stuff()
accepts different arguments for C
andA
. This is because, if we usesuper
in B
to calldo_your_stuff()
on the nextclass, we don't know if it is going to be called onA
or C
. If this scenario isunavoidable, a case specific solution might be required.
One question as of yet unanswered is how does Python determinethe __mro__
for a type? A basic idea behind thealgorithm is provided in this section. This is not essential for justusing super
, or reading following sections, so youcan jump to the next section if you want.
Python determines the precedence of types(or the order in which they should be placed in any__mro__
) from two kinds of constraints specified bythe user:
-
If
A
is a superclass ofB
, thenB
has precedence overA
. Or,B
should always appearbeforeA
in all__mro__
s (that contain both). In short let's denotethis asB > A
. -
If
C
appears beforeD
in the list of bases in a class statement(eg.class Z(C,D):
), thenC > D
.
In addition, to avoid being ambiguous, Python adheres to thefollowing principle:
-
If
E > F
in one scenario (or one__mro__
), then it should be thatE >F
in all scenarios (or all__mro__
s).
We can satisfy the constraints if we build the__mro__
for each new class C
weintroduce, such that:
-
All superclasses of
C
appear in theC.__mro__
(plusC
itself, at the start), and -
The precedence of types in
C.__mro__
does not conflict with the precedence oftypes inB.__mro__
for eachB
inC.__bases__
.
Here the same problem is translated into a game. Consider a classhierarchy as follows:
Since only single inheritance is in play, it is easy to find the__mro__
of these classes. Let's say we define a newclass as class N(A,B,C)
. To compute the__mro__
, consider a game using abacus style beadsover a series of strings.
Beads can move freely over the strings, but the strings cannot be cutor twisted. The strings from left to right contain beads in the orderof __mro__
of each of the bases. The rightmoststring contains one bead for each base, in the order the bases arespecified in the class statement.
The objective is to line up beads in rows, so that each row containsbeads with only one label (as done with the O
beadin the diagram). Each string represents an ordering constraint, and ifwe can reach the goal, we would have an order that satisfies allconstraints. We could then just read the labels off rows from thebottom up to get the __mro__
forN
.
Unfortunately, we cannot solve this problem. The last two strings haveC
and B
in differentorders. However, if we change our class definition to classN(A,C,B)
, then we have some hope.
We just found out that N.__mro__
is(N,A,C,B,object)
(note we insertedN
at the head). The reader can try out thisexperiment in real Python (for the unsolvable case above, Pythonraises an exception). Observe that we even swapped the position of twostrings, keeping the strings in the same order as the bases arespecified in the class statement. The usefulness of this is seenlater.
Sometimes, there might be more than one solution, as shown inthe figure below. Consider four classes classA(object)
, class B(A)
, classC(object)
and class D(C)
. If a new classis defined as class E(B, D)
, there are multiplepossible solutions that satisfy all constraints.
Possible positions for A
are shown as thelittle beads. The order can be kept unambiguous (morecorrectly, monotonic) if the following policiesare followed:
-
Arrange strings from left to right in order ofappearance of bases in the class statement.
-
Attempt to arrange beads in rows moving from bottomup, and left to right. What this means is that the MROof
class E(B, D)
will be setto:(E,B,A,D,C,object)
. This isbecauseA
, being left ofC
, willbe selected first as a candidate for the second row frombottom.
This, essentially, is the idea behind the algorithm used byPython to generate the __mro__
for any newtype. The formal algorithm is formally explained elsewhere [mro-algorithm].
This chapter includes usage notes that do not fit in otherchapters.
In Python, we can use methods with special name like__len__()
, __str__()
and__add__()
to make objects convenient to use (forexample, with the built-in functions len()
,str()
or with the '+
' operator,etc.)
Usually we putthe special methods in a class. | |
We can try to putthem in the instance itself, but it doesn't work. | |
This goesstraight to the class (calls |
The same is true for all such methods, putting them on the instance wewant to use them with does not work. If it did go to the instance theneven something like str(C)
(str
of the class C
) would go toC.__str__()
, which is a method defined for aninstance of C
, and notC
itself.
A simple technique to allow defining such methods for each instanceseparately is shown below.
Subclassing built-in types is straightforward. Actually we have beendoing it all along (whenever we subclass <type 'object'>
). Some built-intypes (types.FunctionType
, for example) are notsubclassable (not yet, at least). However, here we talk aboutsubclassing <type 'list'>
, <type'tuple'>
and other basic data types.
A regular class statement. | |
Define the method tobe overridden. In this case we will convert all items passed through | |
Upcall to the base if required. | |
Append a float and... | |
watch it automatically become an integer. | |
Otherwise, it behaves like any other list. | |
This doesn't gothrough append. We would have to define | |
We can set attributeson our instance. This is because it has a |
Basic lists do not have __dict__
(and so nouser-defined attributes), but ours does. This is usually not a problemand may even be what we want. If we use a verylarge number of MyList
s, however, we could optimizeour program by telling Python not to create the__dict__
for instances ofMyList
.
Example 3.4. Using __slots__
for optimization
class MyList(list): "A list subclass disallowing any user-defined attributes" __slots__ = []ml = MyList() ml.color = 'red' # raises exception!
class MyListWithFewAttrs(list): "A list subclass allowing specific user-defined attributes" __slots__ = ['color']
mla = MyListWithFewAttrs() mla.color = 'red'
mla.weight = 50 # raises exception!
![]()
The | |
Setting any attribute on this raises an exception. | |
| |
Now, if an attribute has space reserved, it can be used. | |
Otherwise, it cannot. This will raise an exception. |
The purpose and recommended use of __slots__
is for optimization. After a type is defined, its slots cannot bechanged. Also, every subclass must define __slots__
,otherwise its instances will end up having __dict__
.
We can create a list even by instantiating it like any other type:list([1,2,3])
. This meanslist.__init__()
accepts the same argument (i.e. anyiterable) and initializes a list. We can customize initialization in asubclass by redefining __init__()
andupcalling __init__()
on thebase.
Tuples are immutable and different from lists. Once an instanceis created, it cannot be changed. Note that the instance of a typealready exists when __init__()
is called (in factthe instance is passed as the first argument). The__new__()
static method of a type is called tocreate an instance of the type. It is passed thetype itself as the first argument, and passed through other initialarguments (similar to __init__()
). We use this tocustomize immutable types like a tuple.
For a list, we massage the arguments and hand them over to list.__init__() . | |
For a tuple, we have to override __new__() . | |
A __new__() should always return. It is supposed to return an instance of the type. |
The __new__()
method is not special to immutabletypes, it is used for all types. It is also converted to a staticmethod automatically by Python (by virtue of its name).
[descrintro] Unifying types and classes in Python 2.2.
[pep-252] Making Types Look More Like Classes.
[pep-253] Subclassing Built-in Types.
[descriptors-howto] How-To Guide for Descriptors.
[mro-algorithm] The Python 2.3 Method Resolution Order.
Colophon
This book was written in DocBook XML. TheHTML version was produced using DocBook XSL stylesheets andxsltproc
. The PDF version was produced usinghtmldoc
. The diagrams were drawn using OmniGraffe[1]. Theprocess was automated usingPaver [2].