Advanced Python (sort of)

Brief

Learn English first

Who am I

Andrii Ursulenko
not working at Evo
but worked here through June 2013 till October 2016
been team lead, worked on prom family projects, kabanchik, goodini and couple more
this is my third time summer python lab participation
have 3 years of Java, 4 years of python and 2 years of go
having fun with clojure and erlang

github.com/a-urth
a-urth.github.io
a.ursulenko@gmail.com
skype: a.ursulenko (but I dare You not)

Why this is best (IMO) place to start

Its hard to get here
There are tons of cool and useful stuff to learn
Definitely You'll learn something new

What we will talk about

Python's object model
Classes
Metaclasses
Descriptors

Disclaimer #1

CPython, Python3

Disclaimer #2

I'm not the highest instance and not an expert
You may know everything already
Where are a lot of things in here
You may remember only 10% of that
It may be overwhelming, but bear with me
All of this is just my point of view

Goal from the capital G

Inspire You and clarify some details

And yeah, I'm struggling of jet lag...

Python's object/data model

https://docs.python.org/3/reference/datamodel.html

Data types in python?

None, NotImplemented, Ellipsis, int, bool, float, complex, string, tuple, byte, list, byte array, set, frozen set, dict, user function, instance method, generator function, coroutine function, async generator function, built-in function, built-in method, class, class instance, module, file object, code object, frame object, traceback object, slice object, static method, class method,

What does it mean?

Dynamic typing
No primitive types
Every object is PyObject

PyObject

typedef struct _object {
    PyObject_HEAD
} PyObject;

struct _object *_ob_next;
struct _object *_ob_prev;
Py_ssize_t ob_refcnt;
struct _typeobject *ob_type;

id, is

# id (built-in function)
# returns object's address in memory

In [1]: id(1)
Out[1]: 4452383728

In [2]: id('')
Out[2]: 4453206704

In [3]: id(None)
Out[3]: 4452107176

In [4]: id([])
Out[4]: 4477498568

In [5]: id(...)
Out[5]: 4452118880

# is (statement) compares objects
# by their identity i.e. by their
# address in memory

In [6]: 1 is 1
Out[6]: True

In [7]: -6 is -6
Out[7]: False

In [9]: 'asd' is 'asd'
Out[9]: True

In [10]: a, b = 'asd!', 'asd!'

In [11]: a is b
Out[11]: True

In [12]: a = 'asd!'

In [13]: b = 'asd!'

In [14]: a is b
Out[14]: False

int

int is array of uint_32, float - array of C double
static list for integers at startup to reduce MALLOC
interns small integers from -5 to 256 inclusive

In [28]: -6 is -6
Out[28]: False

In [29]: 257 is 257
Out[29]: True

In [36]: a = 257

In [37]: b = 257

In [38]: a is b
Out[38]: False

str

immutability matters
string interning FTW
empty and one letter strings interned
strings till 20 alpha characters interned
interning is done again, during compilation
strings compared and interned by hash

In [1]: 'a' * 20 is 'aaaaaaaaaaaaaaaaaaaa'
Out[1]: True

In [2]: 'a' * 21 is 'aaaaaaaaaaaaaaaaaaaaa'
Out[2]: False

In [11]: 'a' + 'sd' is 'asd'
Out[11]: True

In [12]: ''.join(['a', 's', 'd']) is 'asd'
Out[12]: False

In [3]: 'asd!' is 'asd!'
Out[3]: True

In [6]: a, b = 'asd!', 'asd!'

In [7]: a is b
Out[7]: True

In [8]: a = 'asd!'

In [9]: b = 'asd!'

In [10]: a is b
Out[10]: False

list, tuple, dict

lists are dynamically extendable and shrinkable
tuples preallocated same as ints (sort of) - for each tuple size (from 0 to 20) up to 2000 tuples preallocated
dicts are hash tables...

bool, None, ellipsis, NotImplemented

bools are ints, literally
True, False, None, NotImplemented and ellipsis - singletons
and thats why you should ALWAYS compare with any of them by means of is
NotImplemented is not NotImplementedError and its not an exception, its special value to be returned
ellipsis is pretty useful, but don't abuse it

In [1]: True + True == 2 - False
Out[1]: True

In [2]: a = True

In [3]: a is True
Out[3]: True

In [4]: a = ...

In [5]: a is ...
Out[5]: True

In [6]: a = None

In [7]: a is None
Out[7]: True

# ellipsis is useful when You need default
# value but None is also an option

default = ...


def update_user(name, email, address=default):
    user = db.get_user(email)
    if address is not default:
        user.address = address

    user.name = name
    user.email = email

    db.update_user(user)

P.S.

If You would ever consider track object's uniqueness by their id without storing links on them

ids = set()  # just a set for some object's identifiers


def f():
    l2 = [1, 2, 3]  # create a list

    ids.add(id(l2))  # store its address in global registry


f()  # call function

l3 = ['1', '2', '3', '4']  # create new list with other data


print(id(l3) in ids)

Writing code with a class

Class, OOP?

Why?

decoupling
logic extraction
redundancy reducing
abstraction

What?

encapsulation
inheritance
polymorphism
abstraction(?)

Old style vs new style

python2 relevant only
not inherited from object
MRO depth first
not have a lot of things
somewhat deprecated

Old

New

subclassed from object (explicitly for py2, implicitly for py3
MRO breadth first
have all magic methods (basic ones)
somewhat faster
somewhat bigger

class Person:  # always uppercased camel-case (pep-8)

    name = 'Person'

    def __init__(self, name):
        self._name = name
        self.__secret_name = 'secret-{}'.format(name)

    def get_name(self, *args, **kwargs):
        print('I am {}'.format(self))

    @classmethod
    def get_class_name(cls, *args, **kwargs):
        print('I am {}'.format(cls))

    @staticmethod
    def get_static_name():
        print('I have no name actually =(')


person = Person('Bob')
person.get_name()

Person.get_class_name()

Person.get_static_name()

print(person.name, person._name)
print(person.__secret_name)

I am <__main__.Person object at 0x10bbb8cf8>

I am <class '__main__.Person'>

I have no name actually =(

Person Bob

Traceback (most recent call last):
  File "t.py", line 29, in <module>
    print(person.__secret_name)
AttributeError: 'Person' object has no attribute '__secret_name'

it is still subclassed from object
__init__ is a magic method and there are A LOT of them
__secret_name is sort of private field
self is just a name
first argument to instance methods is instance itself
@classmethod decorator turns first argument to class objects
@staticmethod turns instance method to sort of function inside a class - no additional arguments

Constructor, destructor?

Java and C++ people where you at?

init

constructor
creates and returns new class instance usually by means of object.__new__
called first

sort of destructor
returns nothing
called after all links to objects lost

class Object:

    def __new__(cls, *args, **kwargs):
        print(cls)
        return super().__new__(cls)

    def __init__(self, *args, **kwargs):
        print(self)

    def __del__(self):
        print('{} is down!'.format(self))


b = Object()

a = Object()
del a

Object()

new

init

del

just a place for initialization
returns nothing
called after __new__

<class '__main__.Object'>
<__main__.Object object at 0x10c6c9c18>

<class '__main__.Object'>
<__main__.Object object at 0x10c6c9c50>
<__main__.Object object at 0x10c6c9c50> is down!

<class '__main__.Object'>
<__main__.Object object at 0x10c6c9c50>
<__main__.Object object at 0x10c6c9c50> is down!

<__main__.Object object at 0x10c6c9c18> is down!

__new__ - control over instances creation, e.g. cache, singleton, etc.
__init__ - control over instances initialization, e.g. fields instantiation, resource opening, validation, etc.
__del__ - resource freeing, warnings, etc. CAUTION, self may not exists when it will be called.

Used for

del smth != smth.del

Quick mention on pythons GC

each object holds number of references to it
when it reaches 0 memory is freed
additional periodical cyclic reference resolve
generations concept

and weakref

reference to an object
but discards itself when all strong refs are gone
use - when You need to track objects but to not intrude into their lifecycle

from weakref import WeakSet


class A:
    pass


references = set()
weak_references = WeakSet()

a1, a2 = A(), A()

references.add(a1)
weak_references.add(a2)


print(references)
print(weak_references.data)

del a1
del a2

print(references)
print(weak_references.data)

{<__main__.A object at 0x10b532b38>}
{<weakref at 0x10b537b88; to 'A' at 0x10b532ba8>}

{<__main__.A object at 0x10b532b38>}
set()

super

built in function to get base/super class
since python3 works without arguments
but still may take them if You need
its not that simple in terms how it works
because You know, multiple inheritance
and totally relies on MRO (later on that)
and its not a function, its a class and returns proxy object

class A:

    def foo(self):
        print('from A')


class B(A):

    def foo(self):
        print('from B')
        super().foo()


class C(A):

    def foo(self):
        print('from C')
        super().foo()


class D(B, C):

    def foo(self):
        print('from D')
        super().foo()


D().foo()

from D

from B

from C

from A

MRO

stands for method resolution order
algorithm is called C3
generally breadth first tree traversal
accessed from class by .mro() method
super without arguments uses first class in called class mro list
super with arguments uses second argument (which is current self ) mro and first argument (which is class) for resolving current step of mro, then does same

D.mro()

[
    <class '__main__.D'>,
    <class '__main__.B'>,
    <class '__main__.C'>,
    <class '__main__.A'>,
    <class 'object'>
]

class A:
    def foo(self):
        print('from A')


class B(A):
    def foo(self):
        print('from B')
        super().foo()
        # eq to super(B, self).foo()


class C(A):
    def foo(self):
        print('from C')
        super().foo()
        # eq to super(C, self).foo()


class D(B, C):
    def foo(self):
        print('from D')
        super().foo()
        # eq to super(D, self).foo()


D().foo()

(1) So its D class and super will give B

(2) So its B class and super will give C

(3) So its C class and super will give A

(4) So its A class

So, can You redefine mro?

class A:
    def foo(self):
        print('from A')


class B(A):
    def foo(self):
        print('from B')
        super().foo()


class C(A):
    def foo(self):
        print('from C')
        super().foo()


class D(B, C):

    @classmethod
    def mro(cls):
        return [D, B, A]

    def foo(self):
        print('from D')
        super().foo()


print(D.mro())

D().foo()

Yes, but not the way You'd expect it to be

And I dare You don't

[
    <class '__main__.D'>,
    <class '__main__.B'>,
    <class '__main__.A'>
]

from D
from B
from C
from A

class meta(type):

    @classmethod
    def mro(cls):
        return [B, A, object]


class A:
    def foo(self):
        print('from A')


class B(A):
    def foo(self):
        print('from B')
        super().foo()


class C(A):
    def foo(self):
        print('from C')
        super().foo()


class D(B, C, metaclass=meta):

    def foo(self):
        print('from D')
        super().foo()


D().foo()

from B
from A

Since mro for D class doesn't contain it, its namespace doesn't even looked up

So it just goes for B and A classes

If You won't include object You won't be able to create objects at all

dict

Every object has __dict__ field
It contains all attributes, which belongs to this object
Its a regular dict, so its mutable
So You can add attributes dynamically

(1) True
(2)
{
    'foo': <function A.foo at 0x10dd82158>,
    'a': None,
    '__dict__': <attribute '__dict__' of 'A' objects>,
    '__weakref__': <attribute '__weakref__' of 'A' objects>,
    '__module__': '__main__',
    '__init__': <function A.__init__ at 0x10dd820d0>,
    '__doc__': 'This class literally does nothing'
}

(3) {'name': 'Bill'}
(4) {'surname': 'Billinsons', 'name': 'Bill'}
(5) J

class A:
    '''This class literally does nothing'''

    a = None

    def __init__(self, name, surname=None):
        self.name = name

        if surname is not None:
            self.surname = surname

    def foo(self):
        print('My name is {}'.format(self.name))


print('__dict__' in dir(A))  # 1

print(A.__dict__)  # 2

print(A('Bill').__dict__)  # 3

print(A('Bill', 'Billinsons').__dict__)  # 4

john = A('John')
john.__dict__['middlename'] = 'J'
print(john.middlename)  # 5

The problem

dict is hash table which starts with fixed size (?) and double/quadruples its size on reaching limit
so, even if You have class with one, two or no attributes dict will be created with preallocated memory
a lot of classes -> a lot of dicts -> a lot of arrays -> a lot of memory
dict access may be slow
mutability not always good

What to do?

slots to the rescue!

attributes are fixed
attributes are stored in array of required length
somewhat faster access
less memory (it depends but in general cases 40-50% less)

class A:

    __slots__ = ('a', 'name')

    def __init__(self, name):
        self.a = 'A!'
        self.name = name

    def foo(self):
        print('My name is {}'.format(self.name))


print('__dict__' in dir(A))  # 1

print(A.__dict__)  # 2

bill = A('Bill')

print(bill.a, bill.name)  # 3

bill.foo()  # 4

print(bill.__dict__)  # 5

(1) False
(2)
{
    'name': <member 'name' of 'A' objects>,
    'foo': <function A.foo at 0x1065fd158>,
    'a': <member 'a' of 'A' objects>,
    '__module__': '__main__',
    '__doc__': None,
    '__slots__': ('a', 'name'),
    '__init__': <function A.__init__ at 0x1065fd0d0>
}

(3) A! Bill

(4) My name is Bill
(5)
Traceback (most recent call last):
  File "t.py", line 21, in <module>
    print(bill.__dict__)
AttributeError: 'A' object has no attribute '__dict__'

But what about interfaces?

In short - there are no such things.

Protocols are, but its not more than a convention

class ContextManager:

    def __enter__(self):
        pass

    def __exit__(self, *exc_info):
        pass

class Iterable:

    def __iter__(self):
        return iterator


class Iterator:

    def __next__(self):
        return next_value

Abstract classes

Classes with at least one abstract method
Describe base usage, intention for further implementation
Objects of which don't meant to be created
Only purpose is to be inherited from

from abc import ABCMeta, abstractmethod


class BaseMerger(metaclass=ABCMeta):

    @abstractmethod
    def get(self, _id):
        pass

    @abstractmethod
    def create(self, data):
        pass

    @abstractmethod
    def update(self, _id, data):
        pass

    def merge(self, _id, new_data):
        old_obj = self.get(_id)
        if old_obj is None:
            return self.create(new_data)

        return self.update(_id, new_data)

Wrong

class BaseMerger:

    def get(self, _id):
        raise NotImplementedError()

    def create(self, data):
        raise NotImplementedError()

    def update(self, _id, data):
        raise NotImplementedError()

    def merge(self, _id, new_data):
        old_obj = self.get(_id)
        if old_obj is None:
            return self.create(new_data)

        return self.update(_id, new_data)

Correct

somewhat works
objects still can be created
implementation for abstract method is not restricted

works
everything is controlled by meta

And one more thing, actually more than one

Do not use mutable objects as class attributes
Treat class attributes as constants or default values
Do not update class attributes
Do not mix class and instance members
functions > staticmethods
Do not use type(smth) == Smth, use isinstance(smth, Smth)

class Storage:

    __storage = []

    def store(self, val):
        self.__storage.append(val)

    def __repr__(self):
        return str(self.__storage)


class FutureStorage(Storage):
    pass


s = Storage()
s.store(1)
s.store(2)
print(s)  # 1

ss = Storage()
print(ss)  # 2

fs = FutureStorage()
print(fs)  # 3

(1) [1, 2]

(2) [1, 2]

(3) [1, 2]

Wrong

Correct

class Storage:

    __storage = None

    def __init__(self):
        self.__storage = []

    def store(self, val):
        self.__storage.append(val)

    def __repr__(self):
        return str(self.__storage)


class FutureStorage(Storage):
    pass


s = Storage()
s.store(1)
s.store(2)
print(s)

ss = Storage()
print(ss)

fs = FutureStorage()
print(fs)

(1) [1, 2]

(2) []

(3) []

Metaclasses

Have nothing to do with meta-programming
Mainly used to create classes same as classes for objects
Control over classes in general
Sometimes contain logic shared between classes
Also may be used to control object creation
Used rarely
Callable - may be classes but may be functions
If You don't know what metaclasses are 99% You don't actually need them

Where is one metaclass which all of You've already used

type()

type(class_name, base_classes, fields)

creates class objects, not only returns them
type != class statement
main metaclass by means of which object is created
instance of object class
think on that

Person = type('Person', (), {'name': '', 'surname': ''})

print(type(Person))  # 1
print(Person)  # 2

# p = Person('Patrick', 'Oneil')
# p = Person(name='Patrick', surname='Oneil')
p = Person()
p.name, p.surname = 'Patrick', 'Oneil'

print(type(p))  # 3
print(p)  # 4
print(p.name, p.surname)  # 5

(1) <class 'type'>
(2) <class '__main__.Person'>
(3) <class '__main__.Person'>
(4) <__main__.Person object at 0x10a373be0>
(5) Patrick Oneil

class Versioned(type):

    def __init__(cls, name, bases, params, **kwds):
        super().__init__(name, bases, params)

    def __prepare__(name, bases, version=None, **kwds):
        namespace = {}
        if version is not None:
            namespace['version'] = version

        return namespace

    def __new__(cls, cls_name, bases, params, **kwds):
        print('Creating class object for {}'.format(cls_name))
        return super().__new__(cls, cls_name, bases, params)


class Person(metaclass=Versioned, version=1):
    pass


print(Person().version)

Creating class object for Person
1

__prepare__ creates namespace before class creation, returns dict-like object
all keywords passed to class statement are passed through class creation process (metaclass methods)

To create class

subclass from type or and/or call it directly
redefine __new__ and/or __init__

To create instance

subclass from type or and/or call it directly
redefine __call__

class Singletoned(type):

    @classmethod
    def __prepare__(cls, name, bases, **kwds):
        return {'_obj': None}

    def __call__(cls, *args, **kwargs):
        if cls._obj is None:
            print('Creating new object for {}'.format(cls))
            cls._obj = super().__call__(*args, **kwargs)

        return cls._obj


class A(metaclass=Singletoned):
    pass


class B(metaclass=Singletoned):
    pass


a = A()
b = A()

print(a)  # 1
print(b)  # 2

a = B()
b = B()

print(a)  # 3
print(b)  # 4

Creating new object for <class '__main__.A'>
(1) <__main__.A object at 0x10c9b0c88>
(2) <__main__.A object at 0x10c9b0c88>

Creating new object for <class '__main__.B'>
(3) <__main__.B object at 0x10c9b0cc0>
(4) <__main__.B object at 0x10c9b0cc0>

Attributes lookup

everything from instance is looked up in the class if doesn't found
everything from class is looked up in metaclass if doesn't found
except __new__ and __init__ - they are used for class creation, and if not found on class they are never taken from metaclass
this shit is hard

Instance attribute lookup

Class attribute lookup

Summary

ORMs
template engines
interviews
very few other places
they are rare in general
as previously said - most of all You won't use them directly
but You'd better know about them because highly likely that You'll face them

Metaclass in a nutshell

hard to learn
hard to pull off
very powerful
looks awesome
easy avoidable

Descriptors

You know, python's, not the file ones

Why?

Very handy control over class/instance fields
Property/binding behavior
"Learning about descriptors not only provides access to a larger toolset, it creates a deeper understanding of how Python works and an appreciation for the elegance of its design." Python documentation

What?

Class implementing a protocol

How?

__get__
__set__
__delete__

class Person:

    name = ''
    age = 0

    def __check_name(self, name):
        if not isinstance(name, str):
            raise TypeError

        return name

    def __check_age(self, age):
        if not isinstance(age, int):
            raise TypeError

        return age

    def __init__(self, name, age):
        self.age = age
        self.name = name

    def __setattr__(self, name, value):
        if name == 'name':
            value = self.__check_name(value)
        elif name == 'age':
            value = self.__check_age(value)

        super().__setattr__(name, value)

    def __repr__(self):
        return 'Person: {} {}'.format(self.name, self.age)


p = Person('Bill', 29)
print(p)

p.name = 123

Person: Bill 29

Traceback (most recent call last):
  File "t.py", line 38, in <module>
    p.name = 123
  File "t.py", line 25, in __setattr__
    value = self.__check_name(value)
  File "t.py", line 8, in __check_name
    raise TypeError
TypeError

Problem

redundancy
not full control
looks ugly

class Typed:

    def __init__(self, _type, default=...):
        self.val = default if default is not ... else _type
        self.type = _type

    def __set__(self, obj, value):
        if not isinstance(value, self.type):
            raise TypeError

        self.val = value

    def __get__(self, obj, type=None):
        return self.val

    def __delete__(self, obj):
        raise AttributeError


class Person:

    name = Typed(str)
    age = Typed(int)

    def __init__(self, name, age):
        self.age = age
        self.name = name

    def __repr__(self):
        return 'Person: {} {}'.format(self.name, self.age)


p = Person('Bill', 29)
print(p)

p.name = 123

Person: Bill 29

Traceback (most recent call last):
  File "t.py", line 36, in <module>
    p.name = 123
  File "t.py", line 9, in __set__
    raise TypeError
TypeError

Important

Non-data descriptors have only __get__
Data descriptors - have both __get__ and __set__
Read only descriptors have __get__ and __set__ but raise exception on __set__
Descriptors work though __getattribute__
Data descriptors always override instance dictionaries.

class Descriptor:

    def __set__(self, obj, value):
        pass

    def __get__(self, obj, type=None):
        return 'Look at me!'


class A:

    a = Descriptor()

    def __init__(self):
        self.a = 'Hey ya!'


a = A()
print(a.a)

Look at me!

class Property(object):

    def __init__(self, fget=None, fset=None, fdel=None, doc=None):
        self.fget = fget
        self.fset = fset
        self.fdel = fdel
        if doc is None and fget is not None:
            doc = fget.__doc__
        self.__doc__ = doc

    def __get__(self, obj, objtype=None):
        if obj is None:
            return self
        if self.fget is None:
            raise AttributeError("unreadable attribute")
        return self.fget(obj)

    def __set__(self, obj, value):
        if self.fset is None:
            raise AttributeError("can't set attribute")
        self.fset(obj, value)

    def __delete__(self, obj):
        if self.fdel is None:
            raise AttributeError("can't delete attribute")
        self.fdel(obj)

class Square:

    def __init__(self, width, height):
        self.width = width
        self.height = height

    def _area(self):
        return self.width * self.height

    area = Property(_area)

    del _area


box = Square(7, 15)

print(box.area)
box.area = 150

105

Traceback (most recent call last):
  File "t.py", line 56, in <module>
    box.area = 150
  File "t.py", line 21, in __set__
    raise AttributeError("can't set attribute")
AttributeError: can't set attribute

del _area just to clean namespace a little
use @property decorator instead

Usecases

A lot
Non data descriptors transform Your method calls from obj.f(*args) to f(obj, *args)
@property, @classmethod, @staticmethod
ORMs
Forms and other validators
etc, etc, etc

https://docs.python.org/3/reference/datamodel.html
http://spyhce.com/blog/cpython-data-structures
https://docs.python.org/3.6/c-api/structures.html
http://guilload.com/python-string-interning/
https://docs.python.org/3/tutorial/classes.html
https://wiki.python.org/moin/NewClassVsClassicClass
http://lucumr.pocoo.org/2014/8/16/the-python-i-would-like-to-see/
http://blog.thedigitalcatonline.com/blog/2014/08/20/python-3-oop-part-1-objects-and-types/#.WV9COsaB1P0
https://jeffknupp.com/blog/2014/06/18/improve-your-python-python-classes-and-object-oriented-programming/
http://minhhh.github.io/posts/a-guide-to-pythons-magic-methods
http://sixty-north.com/blog/pythons-super-not-as-simple-as-you-thought
https://www.quora.com/How-does-garbage-collection-in-Python-work
https://blog.ionelmc.ro/2015/02/09/understanding-python-metaclasses/
http://eli.thegreenplace.net/2011/08/14/python-metaclasses-by-example
https://docs.python.org/3.6/howto/descriptor.html
http://intermediatepythonista.com/classes-and-objects-ii-descriptors
https://www.youtube.com/watch?v=HTLu2DFOdTg
https://www.youtube.com/watch?v=p33CVV29OG8

Advanced Python (sort of)

Brief

Learn English first

Who am I

Why this is best (IMO) place to start

What we will talk about

Disclaimer #1

CPython, Python3

Disclaimer #2

I'm not the highest instance and not an expert

You may know everything already

Where are a lot of things in here

You may remember only 10% of that

It may be overwhelming, but bear with me

All of this is just my point of view

Goal from the capital G

Inspire You and clarify some details

And yeah, I'm struggling of jet lag...

Python's object/data model

https://docs.python.org/3/reference/datamodel.html

Data types in python?

What does it mean?

PyObject

id, is

int

str

list, tuple, dict

bool, None, ellipsis, NotImplemented

P.S.

Writing code with a class

Class, OOP?

Why?

What?

Old style vs new style

Old

New

Constructor, destructor?

__init__

__new__

__init__

__del__

Used for

del smth != smth.__del__

Quick mention on pythons GC

and weakref

super

MRO

So, can You redefine mro?

__dict__

The problem

What to do?

__slots__ to the rescue!

But what about interfaces?

Abstract classes

And one more thing, actually more than one

Metaclasses

Where is one metaclass which all of You've already used

type()

type(class_name, base_classes, fields)

To create class

To create instance

Attributes lookup

Summary

Metaclass in a nutshell

Descriptors

Why?

What?

How?

Problem

Important

Usecases

Advanced Python (sort of)

More from Andrii Ursulenko

init

new

init

del

del smth != smth.del

dict

slots to the rescue!