We have already seen how we can use a dictionary to group related data together, and how we can use functions to create shortcuts for commonly used groups of statements. A function performs an action using some set of input parameters. Not all functions are applicable to all kinds of data. Classes are a way of grouping together related data and functions which act upon that data.
A class is a kind of data type, just like a string, integer or list. When we create an object of that data type, we call it an instance of a class.
As we have already mentioned, in some other languages some entities are objects and some are not. In Python, everything is an object – everything is an instance of some class. In earlier versions of Python a distinction was made between built-in types and user-defined classes, but these are now completely indistinguishable. Classes and types are themselves objects, and they are of type
type. You can find out the type of any object using the
The data values which we store inside an object are called attributes, and the functions which are associated with the object are called methods. We have already used the methods of some built-in objects, like strings and lists.
When we design our own objects, we have to decide how we are going to group things together, and what our objects are going to represent.
Sometimes we write objects which map very intuitively onto things in the real world. For example, if we are writing code to simulate chemical reactions, we might have
Atom objects which we can combine to make a
Molecule object. However, it isn’t always necessary, desirable or even possible to make all code objects perfectly analogous to their real-world counterparts.
Sometimes we may create objects which don’t have any kind of real-world equivalent, just because it’s useful to group certain functions together.
Defining and using a class¶
Here is an example of a simple custom class which stores information about a person:
import datetime # we will use this for date objects class Person: def __init__(self, name, surname, birthdate, address, telephone, email): self.name = name self.surname = surname self.birthdate = birthdate self.address = address self.telephone = telephone self.email = email def age(self): today = datetime.date.today() age = today.year - self.birthdate.year if today < datetime.date(today.year, self.birthdate.month, self.birthdate.day): age -= 1 return age person = Person( "Jane", "Doe", datetime.date(1992, 3, 12), # year, month, day "No. 12 Short Street, Greenville", "555 456 0987", "email@example.com" ) print(person.name) print(person.email) print(person.age())
We start the class definition with the
class keyword, followed by the class name and a colon. We would list any parent classes in between round brackets before the colon, but this class doesn’t have any, so we can leave them out.
Inside the class body, we define two functions – these are our object’s methods. The first is called
__init__, which is a special method. When we call the class object, a new instance of the class is created, and the
__init__ method on this new object is immediately executed with all the parameters that we passed to the class object. The purpose of this method is thus to set up a new object using data that we have provided.
The second method is a custom method which calculates the age of our person using the birthdate and the current date.
__init__ is sometimes called the object’s constructor, because it is used similarly to the way that constructors are used in other languages, but that is not technically correct – it’s better to call it the initialiser. There is a different method called
__new__ which is more analogous to a constructor, but it is hardly ever used.
You may have noticed that both of these method definitions have
self as the first parameter, and we use this variable inside the method bodies – but we don’t appear to pass this parameter in. This is because whenever we call a method on an object, the object itself is automatically passed in as the first parameter. This gives us a way to access the object’s properties from inside the object’s methods.
In some languages this parameter is implicit – that is, it is not visible in the function signature – and we access it with a special keyword. In Python it is explicitly exposed. It doesn’t have to be called
self, but this is a very strongly followed convention.
Now you should be able to see that our
__init__ function creates attributes on the object and sets them to the values we have passed in as parameters. We use the same names for the attributes and the parameters, but this is not compulsory.
age function doesn’t take any parameters except
self – it only uses information stored in the object’s attributes, and the current date (which it retrieves using the
Note that the
birthdate attribute is itself an object. The
date class is defined in the
datetime module, and we create a new instance of this class to use as the birthdate parameter when we create an instance of the
Person class. We don’t have to assign it to an intermediate variable before using it as a parameter to
Person; we can just create it when we call
Person, just like we create the string literals for the other parameters.
Remember that defining a function doesn’t make the function run. Defining a class also doesn’t make anything run – it just tells Python about the class. The class will not be defined until Python has executed the entirety of the definition, so you can be sure that you can reference any method from any other method on the same class, or even reference the class inside a method of the class. By the time you call that method, the entire class will definitely be defined.
Explain what the following variables refer to, and their scope:
age(the function name)
age(the variable used inside the function)
It is important to note that the attributes set on the object in the
__init__ function do not form an exhaustive list of all the attributes that our object is ever allowed to have.
In some languages you must provide a list of the object’s attributes in the class definition, placeholders are created for these allowed attributes when the object is created, and you may not add new attributes to the object later. In Python, you can add new attributes, and even new methods, to an object on the fly. In fact, there is nothing special about the
__init__ function when it comes to setting attributes. We could store a cached age value on the object from inside the
def age(self): if hasattr(self, "_age"): return self._age today = datetime.date.today() age = today.year - self.birthdate.year if today < datetime.date(today.year, self.birthdate.month, self.birthdate.day): age -= 1 self._age = age return age
Starting an attribute or method name with an underscore (
_) is a convention which we use to indicate that it is a “private” internal property and should not be accessed directly. In a more realistic example, our cached value would sometimes expire and need to be recalculated – so we should always use the
age method to make sure that we get the right value.
We could even add a completely unrelated attribute from outside the object:
person.pets = ['cat', 'cat', 'dog']
It is very common for an object’s methods to update the values of the object’s attributes, but it is considered bad practice to create new attributes in a method without initialising them in the
__init__ method. Setting arbitrary properties from outside the object is frowned upon even more, since it breaks the object-oriented paradigm (which we will discuss in the next chapter).
__init__ method will definitely be executed before anything else when we create the object – so it’s a good place to do all of our initialisation of the object’s data. If we create a new attribute outside the
__init__ method, we run the risk that we will try to use it before it has been initialised.
age example above we have to check if an
_age attribute exists on the object before we try to use it, because if we haven’t run the
age method before it will not have been created yet. It would be much tidier if we called this method at least once from
__init__, to make sure that
_age is created as soon as we create the object.
Initialising all our attributes in
__init__, even if we just set them to empty values, makes our code less error-prone. It also makes it easier to read and understand – we can see at a glance what attributes our object has.
__init__ method doesn’t have to take any parameters (except
self) and it can be completely absent.
What if we want to get or set the value of an attribute of an object without hard-coding its name? We may sometimes want to loop over several attribute names and perform the same operation on all of them, as we do in this example which uses a dictionary:
for key in ["a", "b", "c"]: print(mydict[key])
How can we do something similar with an object? We can’t use the
. operator, because it must be followed by the attribute name as a bare word. If our attribute name is stored as a string value in a variable, we have to use the
getattr function to retrieve the attribute value from an object:
for key in ["a", "b", "c"]: print(getattr(myobject, key, None))
getattr is a built-in function, not a method on the object: it takes the object as its first parameter. The second parameter is the name of the variable as a string, and the optional third parameter is the default value to be returned if the attribute does not exist. If we do not specify a default value,
getattr will raise an exception if the attribute does not exist.
setattr allows us to set the value of an attribute. In this example, we copy data from a dictionary to an object:
for key in ["a", "b", "c"]: setattr(myobject, key, mydict[key])
The first parameter of
setattr is the object, the second is the name of the function, and the third is the new value for the attribute.
As we saw in the previous
age function example,
hasattr detects whether an attribute exists.
There’s nothing preventing us from using
getattr on attributes even if the name can be hard-coded, but this is not recommended: it’s an unnecessarily verbose and round-about way of accessing attributes:
getattr(myobject, "a") # means the same thing as myobject.a
You should only use these functions if you have a good reason to do so.
- Rewrite the
Personclass so that a person’s age is calculated for the first time when a new person instance is created, and recalculated (when it is requested) if the day has changed since the last time that it was calculated.
All the attributes which are defined on a
Person instance are instance attributes – they are added to the instance when the
__init__ method is executed. We can, however, also define attributes which are set on the class. These attributes will be shared by all instances of that class. In many ways they behave just like instance attributes, but there are some caveats that you should be aware of.
We define class attributes in the body of a class, at the same indentation level as method definitions (one level up from the insides of methods):
class Person: TITLES = ('Dr', 'Mr', 'Mrs', 'Ms') def __init__(self, title, name, surname): if title not in self.TITLES: raise ValueError("%s is not a valid title." % title) self.title = title self.name = name self.surname = surname
As you can see, we access the class attribute
TITLES just like we would access an instance attribute – it is made available as a property on the instance object, which we access inside the method through the
Person objects we create will share the same
TITLES class attribute.
Class attributes are often used to define constants which are closely associated with a particular class. Although we can use class attributes from class instances, we can also use them from class objects, without creating an instance:
# we can access a class attribute from an instance person.TITLES # but we can also access it from the class Person.TITLES
Note that the class object doesn’t have access to any instance attributes – those are only created when an instance is created!
# This will give us an error Person.name Person.surname
Class attributes can also sometimes be used to provide default attribute values:
class Person: deceased = False def mark_as_deceased(self): self.deceased = True
When we set an attribute on an instance which has the same name as a class attribute, we are overriding the class attribute with an instance attribute, which will take precedence over it. If we create two
Person objects and call the
mark_as_deceased method on one of them, we will not affect the other one. We should, however, be careful when a class attribute is of a mutable type – because if we modify it in-place, we will affect all objects of that class at the same time. Remember that all instances share the same class attributes:
class Person: pets =  def add_pet(self, pet): self.pets.append(pet) jane = Person() bob = Person() jane.add_pet("cat") print(jane.pets) print(bob.pets) # oops!
What we should do in cases like this is initialise the mutable attribute as an instance attribute, inside
__init__. Then every instance will have its own separate copy:
class Person: def __init__(self): self.pets =  def add_pet(self, pet): self.pets.append(pet) jane = Person() bob = Person() jane.add_pet("cat") print(jane.pets) print(bob.pets)
Note that method definitions are in the same scope as class attribute definitions, so we can use class attribute names as variables in method definitions (without
self, which is only defined inside the methods):
class Person: TITLES = ('Dr', 'Mr', 'Mrs', 'Ms') def __init__(self, title, name, surname, allowed_titles=TITLES): if title not in allowed_titles: raise ValueError("%s is not a valid title." % title) self.title = title self.name = name self.surname = surname
Can we have class methods? Yes, we can. In the next section we will see how to define them using a decorator.
Explain the differences between the attributes
profession, and what values they can have in different instances of this class:
class Smith: surname = "Smith" profession = "smith" def __init__(self, name, profession=None): self.name = name if profession is not None: self.profession = profession
In the previous chapter we learned about decorators – functions which are used to modify the behaviour of other functions. There are some built-in decorators which are often used in class definitions.
Just like we can define class attributes, which are shared between all instances of a class, we can define class methods. We do this by using the
@classmethod decorator to decorate an ordinary method.
A class method still has its calling object as the first parameter, but by convention we rename this parameter from
cls. If we call the class method from an instance, this parameter will contain the instance object, but if we call it from the class it will contain the class object. By calling the parameter
cls we remind ourselves that it is not guaranteed to have any instance attributes.
What are class methods good for? Sometimes there are tasks associated with a class which we can perform using constants and other class attributes, without needing to create any class instances. If we had to use instance methods for these tasks, we would need to create an instance for no reason, which would be wasteful. Sometimes we write classes purely to group related constants together with functions which act on them – we may never instantiate these classes at all.
Sometimes it is useful to write a class method which creates an instance of the class after processing the input so that it is in the right format to be passed to the class constructor. This allows the constructor to be straightforward and not have to implement any complicated parsing or clean-up code:
class Person: def __init__(self, name, surname, birthdate, address, telephone, email): self.name = name # (...) @classmethod def from_text_file(cls, filename): # extract all the parameters from the text file return cls(*params) # this is the same as calling Person(*params)
A static method doesn’t have the calling object passed into it as the first parameter. This means that it doesn’t have access to the rest of the class or instance at all. We can call them from an instance or a class object, but they are most commonly called from class objects, like class methods.
If we are using a class to group together related methods which don’t need to access each other or any other data on the class, we may want to use this technique. The advantage of using static methods is that we eliminate unnecessary
self parameters from our method definitions. The disadvantage is that if we do occasionally want to refer to another class method or attribute inside a static method we have to write the class name out in full, which can be much more verbose than using the
cls variable which is available to us inside a class method.
Here is a brief example comparing the three method types:
class Person: TITLES = ('Dr', 'Mr', 'Mrs', 'Ms') def __init__(self, name, surname): self.name = name self.surname = surname def fullname(self): # instance method # instance object accessible through self return "%s %s" % (self.name, self.surname) @classmethod def allowed_titles_starting_with(cls, startswith): # class method # class or instance object accessible through cls return [t for t in cls.TITLES if t.startswith(startswith)] @staticmethod def allowed_titles_ending_with(endswith): # static method # no parameter for class or instance object # we have to use Person directly return [t for t in Person.TITLES if t.endswith(endswith)] jane = Person("Jane", "Smith") print(jane.fullname()) print(jane.allowed_titles_starting_with("M")) print(Person.allowed_titles_starting_with("M")) print(jane.allowed_titles_ending_with("s")) print(Person.allowed_titles_ending_with("s"))
Sometimes we use a method to generate a property of an object dynamically, calculating it from the object’s other properties. Sometimes you can simply use a method to access a single attribute and return it. You can also use a different method to update the value of the attribute instead of accessing it directly. Methods like this are called getters and setters, because they “get” and “set” the values of attributes, respectively.
In some languages you are encouraged to use getters and setters for all attributes, and never to access their values directly – and there are language features which can make attributes inaccessible except through setters and getters. In Python, accessing simple attributes directly is perfectly acceptable, and writing getters and setters for all of them is considered unnecessarily verbose. Setters can be inconvenient because they don’t allow use of compound assignment operators:
class Person: def __init__(self, height): self.height = height def get_height(self): return self.height def set_height(self, height): self.height = height jane = Person(153) # Jane is 153cm tall jane.height += 1 # Jane grows by a centimetre jane.set_height(jane.height + 1) # Jane grows again
As we can see, incrementing the height attribute through a setter is much more verbose. Of course we could write a second setter which increments the attribute by the given parameter – but we would have to do something similar for every attribute and every kind of modification that we want to perform. We would have a similar issue with in-place modifications, like adding values to lists.
Something which is often considered an advantage of setters and getters is that we can change the way that an attribute is generated inside the object without affecting any code which uses the object. For example, suppose that we initially created a
Person class which has a
fullname attribute, but later we want to change the class to have separate
surname attributes which we combine to create a full name. If we always access the
fullname attribute through a setter, we can just rewrite the setter – none of the code which calls the setter will have to be changed.
But what if our code accesses the
fullname attribute directly? We can write a
fullname method which returns the right value, but a method has to be called. Fortunately, the
@property decorator lets us make a method behave like an attribute:
class Person: def __init__(self, name, surname): self.name = name self.surname = surname @property def fullname(self): return "%s %s" % (self.name, self.surname) jane = Person("Jane", "Smith") print(jane.fullname) # no brackets!
There are also decorators which we can use to define a setter and a deleter for our attribute (a deleter will delete the attribute from our object). The getter, setter and deleter methods must all have the same name:
class Person: def __init__(self, name, surname): self.name = name self.surname = surname @property def fullname(self): return "%s %s" % (self.name, self.surname) @fullname.setter def fullname(self, value): # this is much more complicated in real life name, surname = value.split(" ", 1) self.name = name self.surname = surname @fullname.deleter def fullname(self): del self.name del self.surname jane = Person("Jane", "Smith") print(jane.fullname) jane.fullname = "Jane Doe" print(jane.fullname) print(jane.name) print(jane.surname)
Create a class called
Numbers, which has a single class attribute called
MULTIPLIER, and a constructor which takes the parameters
y(these should all be numbers).
- Write a method called
addwhich returns the sum of the attributes
- Write a class method called
multiply, which takes a single number parameter
aand returns the product of
- Write a static method called
subtract, which takes two number parameters,
c, and returns
- Write a method called
valuewhich returns a tuple containing the values of
y. Make this method into a property, and write a setter and a deleter for manipulating the values of
- Write a method called
Inspecting an object¶
We can check what properties are defined on an object using the
class Person: def __init__(self, name, surname): self.name = name self.surname = surname def fullname(self): return "%s %s" % (self.name, self.surname) jane = Person("Jane", "Smith") print(dir(jane))
Now we can see our attributes and our method – but what’s all that other stuff? We will discuss inheritance in the next chapter, but for now all you need to know is that any class that you define has
object as its parent class even if you don’t explicitly say so – so your class will have a lot of default attributes and methods that any Python object has.
in Python 2 we have to inherit from
object explicitly, otherwise our class will be almost completely empty except for our own custom properties. Classes which don’t inherit from
object are called “old-style classes”, and using them is not recommended. If we were to write the person class in Python 2 we would write the first line as
This is why you can just leave out the
__init__ method out of your class if you don’t have any initialisation to do – the default that you inherited from
object (which does nothing) will be used instead. If you do write your own
__init__ method, it will override the default method. Sometimes we also call this overloading.
Many default methods and attributes that are found in built-in Python objects have names which begin and end in double underscores, like
__str__. These names indicate that these properties have a special meaning – you shouldn’t create your own methods or attributes with the same names unless you mean to overload them. These properties are usually methods, and they are sometimes called magic methods.
We can use
dir on any object. You can try to use it on all kinds of objects which we have already seen before, like numbers, lists, strings and functions, to see what built-in properties these objects have in common.
Here are some examples of special object properties:
__init__: the initialisation method of an object, which is called when the object is created.
__str__: the string representation method of an object, which is called when you use the
strfunction to convert that object to a string.
__class__: an attribute which stores the the class (or type) of an object – this is what is returned when you use the
typefunction on the object.
__eq__: a method which determines whether this object is equal to another. There are also other methods for determining if it’s not equal, less than, etc.. These methods are used in object comparisons, for example when we use the equality operator
==to check if two objects are equal.
__add__is a method which allows this object to be added to another object. There are equivalent methods for all the other arithmetic operators. Not all objects support all arithemtic operations – numbers have all of these methods defined, but other objects may only have a subset.
__iter__: a method which returns an iterator over the object – we will find it on strings, lists and other iterables. It is executed when we use the
iterfunction on the object.
__len__: a method which calculates the length of an object – we will find it on sequences. It is executed when we use the
lenfunction of an object.
__dict__: a dictionary which contains all the instance attributes of an object, with their names as keys. It can be useful if we want to iterate over all the attributes of an object.
__dict__does not include any methods, class attributes or special default attributes like
Create an instance of the
Personclass from example 2. Use the
dirfunction on the instance. Then use the
dirfunction on the class.
- What happens if you call the
__str__method on the instance? Verify that you get the same result if you call the
strfunction with the instance as a parameter.
- What is the type of the instance?
- What is the type of the class?
- Write a function which prints out the names and values of all the custom attributes of any object that is passed in as a parameter.
- What happens if you call the
Overriding magic methods¶
We have already seen how to overload the
__init__ method so that we can customise it to initialise our class. We can also overload other special methods. For example, the purpose of the
__str__ method is to output a useful string representation of our object. but by default if we use the
str function on a person object (which will call the
__str__ method), all that we will get is the class name and an ID. That’s not very useful! Let’s write a custom
__str__ method which shows the values of all of the object’s properties:
import datetime class Person: def __init__(self, name, surname, birthdate, address, telephone, email): self.name = name self.surname = surname self.birthdate = birthdate self.address = address self.telephone = telephone self.email = email def __str__(self): return "%s %s, born %s\nAddress: %s\nTelephone: %s\nEmail:%s" % (self.name, self.surname, self.birthdate, self.address, self.telephone, self.email) jane = Person( "Jane", "Doe", datetime.date(1992, 3, 12), # year, month, day "No. 12 Short Street, Greenville", "555 456 0987", "firstname.lastname@example.org" ) print(jane)
Note that when we insert the birthdate object into the output string with
%s it will itself be converted to a string, so we don’t need to do it ourselves (unless we want to change the format).
It is also often useful to overload the comparison methods, so that we can use comparison operators on our person objects. By default, our person objects are only equal if they are the same object, and you can’t test whether one person object is greater than another because person objects have no default order.
Suppose that we want our person objects to be equal if all their attributes have the same values, and we want to be able to order them alphabetically by surname and then by first name. All of the magic comparison methods are independent of each other, so we will need to overload all of them if we want all of them to work – but fortunately once we have defined equality and one of the basic order methods the rest are easy to do. Each of these methods takes two parameters –
self for the current object, and
other for the other object:
class Person: def __init__(self, name, surname): self.name = name self.surname = surname def __eq__(self, other): # does self == other? return self.name == other.name and self.surname == other.surname def __gt__(self, other): # is self > other? if self.surname == other.surname: return self.name > other.name return self.surname > other.surname # now we can define all the other methods in terms of the first two def __ne__(self, other): # does self != other? return not self == other # this calls self.__eq__(other) def __le__(self, other): # is self <= other? return not self > other # this calls self.__gt__(other) def __lt__(self, other): # is self < other? return not (self > other or self == other) def __ge__(self, other): # is self >= other? return not self < other
other is not guaranteed to be another person object, and we haven’t put in any checks to make sure that it is. Our method will crash if the other object doesn’t have a
surname attribute, but if they are present the comparison will work. Whether that makes sense or not is something that we will need to think about if we create similar types of objects.
Sometimes it makes sense to exit with an error if the other object is not of the same type as our object, but sometimes we can compare two compatible objects even if they are not of the same type. For example, it makes sense to compare
2.5 because they are both numbers, even though one is an integer and the other is a float.
Python 2 also has a
__cmp__ method which was introduced to the language before the individual comparison methods (called rich comparisons) described above. It is used if the rich comparisons are not defined. You should overload it in a way which is consistent with the rich comparison methods, otherwise you may encounter some very strange behaviour.
- Write a class for creating completely generic objects: its
__init__function should accept any number of keyword parameters, and set them on the object as attributes with the keys as names. Write a
__str__method for the class – the string it returns should include the name of the class and the values of all the object’s custom instance attributes.
Answers to exercises¶
Answer to exercise 1¶
Personis a class defined in the global scope. It is a global variable.
personis an instance of the
Personclass. It is also a global variable.
surnameis a parameter passed into the
__init__method – it is a local variable in the scope if the
selfis a parameter passed into each instance method of the class – it will be replaced by the instance object when the method is called on the object with the
.operator. It is a new local variable inside the scope of each of the methods – it just always has the same value, and by convention it is always given the same name to reflect this.
ageis a method of the
Personclass. It is a local variable in the scope of the class.
age(the variable used inside the function) is a local variable inside the scope of the
self.emailisn’t really a separate variable. It’s an example of how we can refer to attributes and methods of an object using a variable which refers to the object, the
.operator and the name of the attribute or method. We use the
selfvariable to refer to an object inside one of the object’s own methods – wherever the variable
selfis defined, we can use
person.emailis another example of the same thing. In the global scope, our person instance is referred to by the variable name
personis defined, we can use
Answer to exercise 2¶
Here is an example program:
import datetime class Person: def __init__(self, name, surname, birthdate, address, telephone, email): self.name = name self.surname = surname self.birthdate = birthdate self.address = address self.telephone = telephone self.email = email # This isn't strictly necessary, but it clearly introduces these attributes self._age = None self._age_last_recalculated = None self._recalculate_age() def _recalculate_age(self): today = datetime.date.today() age = today.year - self.birthdate.year if today < datetime.date(today.year, self.birthdate.month, self.birthdate.day): age -= 1 self._age = age self._age_last_recalculated = today def age(self): if (datetime.date.today() > self._age_last_recalculated): self._recalculate_age() return self._age
Answer to exercise 3¶
nameis always an instance attribute which is set in the constructor, and each class instance can have a different name value.
surnameis always a class attribute, and cannot be overridden in the constructor – every instance will have a surname value of
professionis a class attribute, but it can optionally be overridden by an instance attribute in the constructor. Each instance will have a profession value of
smithunless the optional
surnameparameter is passed into the constructor with a different value.
Answer to exercise 4¶
Here is an example program:
class Numbers: MULTIPLIER = 3.5 def __init__(self, x, y): self.x = x self.y = y def add(self): return self.x + self.y @classmethod def multiply(cls, a): return cls.MULTIPLIER * a @staticmethod def subtract(b, c): return b - c @property def value(self): return (self.x, self.y) @value.setter def value(self, xy_tuple): self.x, self.y = xy_tuple @value.deleter def value(self): del self.x del self.y
Answer to exercise 5¶
You should see something like
'<__main__.Person object at 0x7fcb233301d0>'.
__main__is Python’s name for the program you are executing.
<class 'type'>– any class has the type
Here is an example program:
def print_object_attrs(any_object): for k, v in any_object.__dict__.items(): print("%s: %s" % (k, v))
Answer to exercise 6¶
Here is an example program:
class AnyClass: def __init__(self, **kwargs): for k, v in kwargs.items(): setattr(self, k, v) def __str__(self): attrs = ["%s=%s" % (k, v) for (k, v) in self.__dict__.items()] classname = self.__class__.__name__ return "%s: %s" % (classname, " ".join(attrs))