Python Advanced – Building on the Foundation

If you like this presentation – show it...

Slide 0

Python Advanced Building on the foundation

Slide 1

course = { 'title': 'Python Advanced', 'chapters': [ 'Introduction', 'Special Methods & Attributes', 'Unit Testing', 'Function, Object & Class Customisation', 'Object Thinking', 'Containers', 'Iterators & Generators', 'Functional Thinking', 'Concurrency', 'Modules & Packages', 'Outroduction', 'Labs & Homework' ], ... }

Slide 2

course = { ... 'licence': ('Creative Commons Attribution 4.0, ' 'https://creativecommons.org/licenses/by/4.0/'), 'contributors': [ 'Kevlin Henney, kevlin@curbralan.com', ], 'date': '2015-11-05' }

Slide 3

Introduction Everything has a beginning...

Slide 4

Course goals  Increase knowledge of Python 3  This course builds on the themes and concepts of the Python Foundation  Attendance on the Python Foundation course is assumed  The course focus is on the language, the standard library and techniques  Using slides, discussion and hands-on experiments and labs

Slide 5

Coding experiments  Small (or large) coding experiments to explore concepts in practice  Use the IDLE environment, the default python shell or any Python environment and editor combination of your choice!  Some experiments are described in the slides, but others may be suggested  You can carry out experiments individually or in pairs

Slide 6

Programming labs  Labs implemented using Cyber-Dojo  http://cyber-dojo.org/  Labs involve writing both code and tests  Labs are undertaken in pairs  But pairing is not marriage — you will move around  Oh, and there's homework too  Individual rather than paired, and not in Cyber-Dojo

Slide 7

Questions, answers & intros  The best way to have your questions answered is to ask them  Mindreading is not (yet) supported  And now, please introduce yourself!  Briefly...

Slide 8

Special Methods & Attributes Integrating new types with operators & built-in functions

Slide 9

Facts at a glance  Special methods take the place of operator overloading in Python  Built-in functions also build on special methods and special attributes  The set of special names Python is aware of is predefined  Context managers allow new classes to take advantage of with statements

Slide 10

Operator overloading  It is not possible to overload operators in Python directly  There is no syntax for this  But classes can support operators and global built-ins using special methods  They have reserved names, meanings and predefined relationships, and some cannot be redefined, e.g., is and id  Often referred to as magic methods

Slide 11

Comparisons If not defined for a class, == supports reference equality Equality operations __eq__(self, other) __ne__(self, other) self == other self != other If not defined for a class, != is defined as not self == other Ordering operations __lt__(self, __gt__(self, __le__(self, __ge__(self, other) other) other) other) self self self self < other > other <= other >= other These methods return NotImplemented if not provided for a class, but whether TypeError is raised when an operator form is used depends on whether a reversed comparison is also defined, e.g., if __lt__ is defined but __gt__ is not, a > b will be executed as b < a Note that the __cmp__ special method is not supported in Python 3, so consider using the functools.total_ordering decorator to help define the rich comparisons

Slide 12

Arithmetic operations Unary operators and built-in functions __neg__(self) __pos__(self) __invert__(self) __abs__(self) -self +self ~self abs(self) Binary arithmetic operators Reflected forms also exist, e.g., __radd__, where method dispatch should be on the right-hand rather than left-hand operand __add__(self, other) __sub__(self, other) __mul__(self, other) __truediv__(self, other) __floordiv__(self, other) __mod__(self, other) __pow__(self, other) ... self + other self - other self * other self / other self // other self % other self**other

Slide 13

Augmented assignment Note that if these special methods are not defined, but the corresponding binary operations are, these assignment forms are still valid, e.g., if __add__ is defined, then lhs += rhs becomes lhs = lhs.__add__(rhs) Augmented assignment operators __iadd__(self, other) __isub__(self, other) __imul__(self, other) __itruediv__(self, other) __ifloordiv__(self, other) __imod__(self, other) __ipow__(self, other) ... self self self self self self self += other -= other *= other /= other //= other %= other **= other Each method should return the result of the in-place operation (usually self) as the result is used for the actual assignment, i.e., lhs += rhs is equivalent to lhs = lhs.__iadd__(rhs)

Slide 14

Conversions & representation Numeric type conversions __int__(self) __float__(self) __round__(self) __complex__(self) __bool__(self) __hash__(self) And also __round__(self, n) to round to n digits after the decimal point (or before if negative) Should only define if __eq__ is also defined, such that objects comparing equal have the same hash code String type conversions If __str__ is not defined, __repr__ is used by str if present Official string representation, that should ideally be a valid Python expression that constructs the value __str__(self) __format__(self, format) __repr__(self) __bytes__(self)

Slide 15

Containers & iteration Content query __len__(self) len(self) __contains__(self, item) item in self If __contains__ not defined, in uses linear search based on __iter__ or indexing from 0 of __getitem__ Iteration Most commonly invoked indirectly in the context of a for loop __iter__(self) __next__(self) iter(self) next(self) Subscripting __getitem__(self, __setitem__(self, __delitem__(self, __missing__(self, key) key, item) key) key) Hook method used by dict for its subclasses self[key] self[key] = item del self[key]

Slide 16

Interception & lifecycle Support for function calls and attribute access __call__(self, ...) __getattr__(self, name) __getattribute__(self, name) __setattr__(self, name, value) __delattr__(self, name) __dir__(self) Object lifecycle __init__(self, ...) __del__(self) Context management __enter__(self) __exit__(self, exception, value, traceback)

Slide 17

with & context managers  The with statement provides exception safety for paired actions  I.e., where an initial acquire/open/lock action must be paired with a final release/close/unlock action, regardless of exceptions  A context manager is an object that can be used in this way in a with  It is defined by a simple protocol

Slide 18

Context manager anatomy Called on entry to with Result bound to the target of the with class Resource: def __enter__(self): ... return self def __exit__(self, exception, value, trace): ... Exception string or None if no exception Traceback or None if no exception Exception class or None if no exception with Resource() as resource: ... Note that resource still accessible after the with statement

Slide 19

Experiment Write a context manager to time the execution of code within a with statement with Timing(): Code to time the execution of task_to_time() from time import time Record current time using time Print elapsed time, optionally indicating whether an exception was thrown class Timing: def __enter__(self): ... def __exit__(self, exception, value, trace): ...

Slide 20

Special attributes  A number of special attributes can also be defined, including...  __doc__ is the documentation string and is used by the built-in help function  __name__ is used to hold the name for functions and classes  __dict__ refers to the dictionary of values held by any object  __module__ names the defining module

Slide 21

Unit Testing Programming unittest with GUTs

Slide 22

Facts at a glance  Python has many testing frameworks available, including unittest  unittest is derived from JUnit and has a rich vocabulary of assertions  Good Unit Tests (GUTs) are structured around explanation and intention  The outcome of a unit test depends solely on the code and the tests

Slide 23

unittest  The unittest framework is a standard JUnit-inspired framework  It is organised around test case classes, which contain test case methods  Naming follows Java camelCase  Runnable in other frameworks, e.g., pytest  Python has no shortage of available testing frameworks!  E.g., doctest, nose, pytest

Slide 24

Good Unit Tests (GUTs) Very many people say "TDD" when they really mean, "I have good unit tests" ("I have GUTs"?). Ron Jeffries tried for years to explain what this was, but we never got a catch-phrase for it, and now TDD is being watered down to mean GUTs. Alistair Cockburn http://alistair.cockburn.us/The+modern+programming+professional+has+GUTs

Slide 25

GUTs in practice  Good unit tests (GUTs)...  Are fully automated, i.e., write code to test code  Offer good coverage of the code under test, including boundary cases and errorhandling paths  Are easy to read and to maintain  Express the intent of the code under test — they do more than just check it

Slide 26

Not-so-good unit tests  Problematic test styles include...  Monolithic tests: all test cases in a single function, e.g., test  Ad hoc tests: test cases arbitrarily scattered across test functions, e.g., test1, test2, ...  Procedural tests: test cases bundled into a test method that correspond to target method, e.g., test_foo tests foo

Slide 27

A unittest example Test case method names must start with test Derivation from base class is necessary from leap_year import is_leap_year import unittest class LeapYearTests(unittest.TestCase): def test_that_years_not_divisible_by_4_are_not_leap_years(self): self.assertFalse(is_leap_year(2015)) def test_that_years_divisible_by_4_but_not_by_100_are_leap_years(self): self.assertTrue(is_leap_year(2016)) def test_that_years_divisible_by_100_but_not_by_400_are_not_leap_years(self): self.assertFalse(is_leap_year(1900)) def test_that_years_divisible_by_400_are_leap_years(self): self.assertTrue(is_leap_year(2000)) if __name__ == '__main__': unittest.main()

Slide 28

Assertions assertEqual(lhs, rhs) assertNotEqual(lhs, rhs) assertTrue(result) assertFalse(result) assertIs(lhs, rhs) assertIsNot(lhs, rhs) assertIsNone(result) assertIsNotNone(result) assertIn(lhs, rhs) assertNotIn(lhs, rhs) assertIsInstance(lhs, rhs) assertNotIsInstance(lhs, rhs) ... All assertions also take an optional msg argument

Slide 29

More assertions assertLess(lhs, rhs) assertLessEqual(lhs, rhs) assertGreater(lhs, rhs) assertGreaterEqual(lhs, rhs) assertRegex(result) assertNotRegex(result) assertRaises(exception) assertRaises(exception, callable, *args, *kwargs) fail() assertAlmostEqual(lhs, rhs) assertAlmostNotEqual(lhs, rhs) ... By default, approximate equality is established to within 7 decimal places — this can be changed using the places keyword argument or by providing a delta keyword argument

Slide 30

assertRaises  assertRaises can be used as an ordinary assertion call or with with  If a callable argument is not provided, assertRaises returns a context manager  As a context manager, assertRaises fails if associated with body does not raise self.assertRaises(ValueError, lambda: is_leap_year(-1)) with self.assertRaises(ValueError) as context: is_leap_year(-1) Optional, but can be used to access raised exception details

Slide 31

assertRaisesRegex  assertRaisesRegex also matches string representation of raised exception  Equivalent to regex search  Like assertRaises, assertRaisesRegex can be used as function or context manager self.assertRaisesRegex( ValueError, 'invalid', lambda: is_leap_year(-1)) with self.assertRaisesRegex(ValueError, 'invalid') : is_leap_year(-1)

Slide 32

Tests as explanation So who should you be writing the tests for? For the person trying to understand your code. Good tests act as documentation for the code they are testing. They describe how the code works. For each usage scenario, the test(s):  Describe the context, starting point, or preconditions that must be satisfied  Illustrate how the software is invoked  Describe the expected results or postconditions to be verified Different usage scenarios will have slightly different versions of each of these. Gerard Meszaros "Write Tests for People"

Slide 33

Test anatomy  Example-based tests ideally have a simple linear flow: arrange, act, assert  Given: set up data  When: perform the action to be tested  Then: assert desired outcome  Tests should be short and focused  A single objective or outcome — but not necessarily a single assertion — that is reflected in the name

Slide 34

Common fixture code  Common test fixture code can be factored out...  By defining setUp and tearDown methods that are called automatically before and after each test case execution  By factoring out common initialisation code, housekeeping code or assertion support code into its own methods, local to the test class

Slide 35

Tests as specifications Tests that are not written with their role as specifications in mind can be very confusing to read. The difficulty in understanding what they are testing can greatly reduce the velocity at which a codebase can be changed. Nat Pryce & Steve Freeman "Are your tests really driving your development"

Slide 36

What is a unit test? A test is not a unit test if:  It talks to the database  It communicates across the network  It touches the file system  It can't run at the same time as any of your other unit tests  You have to do special things to your environment (such as editing config files) to run it. Tests that do these things aren't bad. Often they are worth writing, and they can be written in a unit test harness. However, it is important to be able to separate them from true unit tests so that we can keep a set of tests that we can run fast whenever we make our changes. Michael Feathers http://www.artima.com/weblogs/viewpost.jsp?thread=126923

Slide 37

External dependencies  External resources are a common source of non-unit dependencies      Registries and environment variables Network connections and databases Files and directories Current date and time Hardware devices  Externals can often be replaced by test doubles or pre-evaluated objects

Slide 38

A taxonomy of test doubles  It's not just about mocks...  A test stub is used to substitute input and can be used for fault injection  A test spy offers input and captures output behaviour  A mock object validates expectations  A fake object offers a usable alternative to a real dependency  A dummy object fulfils a dependency

Slide 39

Doubles in practice  Mocking and related techniques are easier in Python than many languages  Duck typing means dependencies can be substituted without changing class design  Passing functions as objects allows for simple pluggability  It is easy to write interception code for new classes and to add interception code on existing objects or use existing libraries, e.g., unittest.mock

Slide 40

Function, Object & Class Customisation Ad hoc objects, metaclasses, monkey patching & decorators

Slide 41

Facts at a glance  Functions are objects with state  Classes are mutable after definition, so methods can be added and removed  Objects can have attributes added and removed after creation  Decorators wrap functions on definition  A metaclass is the class of a class, defining how the class behaves

Slide 42

It's objects all the way down  Everything in Python is expressed as objects, from functions to modules  Functions are callable objects — with associated state and metadata — and are instantiated with a def statement  And Python is a very dynamic language, not just dynamically typed  Python's object model can be fully accessed at runtime

Slide 43

Object creation  Objects are normally created based on calling their class name  But the class name does not have to be present, only the class object  E.g., type(parrot)() creates a new object with the same type as parrot  This eliminates the need for some of the factory patterns and convoluted tricks employed in other languages

Slide 44

Ad hoc data structures  Keyword arguments make dicts easy to use as ad hoc data structures  Keyword arguments are passed as dicts, with keywords becoming keys  See also collections.namedtuple sputnik_1 = dict(year=1957, month=10, day=4) {'day': 4, 'month': 10, 'year': 1957} origin = dict(x=0, y=0) {'y': 0, 'x': 0}

Slide 45

Dynamic object attributes  It is possible to add attributes to an object after it has been created  Object attributes are dynamic and implemented as a dictionary  An attribute can be removed using del Ad hoc data structures with named attributes can be created easily class Spam: pass meal = Spam() meal.egg = True meal.bacon = False

Slide 46

Monkey patching A monkey patch is a way for a program to extend or modify supporting system software locally (affecting only the running instance of the program). This process has also been termed duck punching. The definition of the term varies depending upon the community using it. In Ruby, Python, and many other dynamic programming languages, the term monkey patch only refers to dynamic modifications of a class or module at runtime. http://en.wikipedia.org/wiki/Monkey_patch

Slide 47

Modifying existing classes  Methods and attributes can be attached to existing classes  But be careful doing so! class Menu: pass menu = Menu() Menu.options = {'spam', 'egg', 'bacon', 'sausage'} def order(self, *choice): return set(choice) <= self.options Menu.order = order assert menu.order('egg') assert not menu.order('toast')

Slide 48

Exploring the object model  A number of functions help to explore the runtime object model  dir: a sorted list of strings of names in the current or specified scope  locals: dict of current scope's variables  globals: dict of current scope's global variables  vars: dict of an object's state, i.e., __dict__, or locals if no object specified

Slide 49

Object attribute operations  Object attributes can be accessed using hasattr, getattr and setattr  These global functions are preferred to manipulating an object's __dict__ directly  An object can provide special methods to intercept attribute access  Query is via __getattr__ (if not present) and __getattribute__ (unconditional)  Modify via __setattr__ and __delattr__

Slide 50

Decorators  A function definition may be wrapped in decorator expressions  A decorator is a callable object that results in a callable object given a callable object or specified arguments  On definition the function is passed to the decorator and the decorator's result is used in place of the function  This forms a chain of wrapping if more than one decorator is specified

Slide 51

Basic decoration def non_zero_args(wrapped): def wrapper(*args): if all(args): return wrapped(*args) else: raise ValueError return wrapper @non_zero_args def date(year, month, day): return year, month, day date(1957, 10, 4) date(2000, 0, 0) A nested function that performs the validation, executing the wrapped function with arguments if successful (Note: for brevity, keyword arguments are not handled) Decorator Returns (1957, 10, 4) Raises ValueError

Slide 52

Refined decoration from functools import wraps def non_zero_args(wrapped): @wraps(wrapped) def wrapper(*args): if all(args): return wrapped(*args) else: raise ValueError return wrapper @non_zero_args def date(year, month, day): return year, month, day Ensures the resulting wrapper has the same principal attributes as the wrapped function, e.g., __name__ and __doc__

Slide 53

Parameterised decoration def check_args(checker): def decorator(wrapped): def wrapper(*args): if checker(args): return wrapped(*args) else: raise ValueError return wrapper return decorator Nested function that curries the checker argument Nested nested function that performs the validation @check_args(all) def date(year, month, day): return year, month, day date(1957, 10, 4) date(2000, 0, 0) Returns (1957, 10, 4) Raises ValueError

Slide 54

Metaclasses  A metaclass can be considered the class of class  A class defines how objects behaves; a metaclass defines how classes behaves  Metaclasses are most commonly used as class factories  Customise class creation and execution  A class is created from a triple of name, base classes and a dictionary of attributes

Slide 55

abc.ABCMeta Use of the ABCMeta metaclass to define abstract classes is one of the more common metaclass examples Instantiation for Visitor is disallowed because of ABCMeta and the presence of at least one @abstractmethod @abstractmethod only has effect if ABCMeta is the metaclass from abc import ABCMeta from abc import abstractmethod class Visitor(metaclass=ABCMeta): @abstractmethod def enter(self, visited): pass @abstractmethod def exit(self, visited): pass @abstractmethod def visit(self, value): pass

Slide 56

type  type is the default metaclass in Python  And type is its own class, i.e., type(type) is an identity operation  type is most often used as a query, but it can also be used to create a new type def init(self, options): self.__options = options def options(self): return self.__options Menu = type('Menu', (), {'__init__': init, 'options': options}) menu = Menu({'spam', 'egg', 'bacon', 'sausage'}) The base defaults to object

Slide 57

Adding new code at runtime  New code, from source, can be added to the Python runtime  The compile function returns a code object from a module, statement or expression  The exec function executes a code object  The eval function evaluates source code and returns the resulting value  Here be dragons!

Slide 58

Object Thinking Protocols, polymorphism, patterns & practical advice

Slide 59

Facts at a glance  Object usage defined more by protocols than by class relationships  Substitutability is strong conformance  Polymorphism is a fundamental consideration (but inheritance is not)  Values can be expressed as objects following particular conventions  Enumeration types have library support

Slide 60

Not all objects are equal  Be careful how you generalise your experience!  Patterns of practice are context sensitive  Python's type system means that...  Some OO practices from other languages and design approaches travel well  Some do not translate easily or well — or even at all  And some practices are specific to Python

Slide 61

Protocols and conformance  An expected set of method calls can be considered to form a protocol  Protocols may be named informally, but they are not part of the language  Conformance to a protocol does not depend on inheritance  Dynamic binding means polymorphism and subclassing are orthogonal  Think duck types not declared types

Slide 62

Liskov Substitution Principle A type hierarchy is composed of subtypes and supertypes. The intuitive idea of a subtype is one whose objects provide all the behavior of objects of another type (the supertype) plus something extra. What is wanted here is something like the following substitution property: If for each object o1 of type S there is an object o2 of type T such that for all programs P defined in terms of T, the behavior of P is unchanged when o1 is substituted for o2, then S is a subtype of T. Barbara Liskov "Data Abstraction and Hierarchy"

Slide 63

Contracts & consequences  Substitutability is the strongest form of conformance to a protocol  A contract of behavioural equivalence  Following the contract means...  An overridden method cannot result in a wider range or cover a narrower domain  Concrete classes should not be bases  A derived class should support the same initialisation protocol as its base

Slide 64

Inheritance & composition  A class can respect logical invariants  Assertions true of its methods and state  A derived class should respect the invariants of its base classes  And may strengthen them  Consider composition and forwarding  Especially if derivation would break invariants and lead to method spam from the base classes

Slide 65

Object structure & traversal  The following patterns combine and complement one another:  Composite: recursive–whole part structure  Visitor: dispatch based on argument type  Enumeration Method: iteration based on inversion of control flow  Lifecycle Callback: define actions for object lifecycle events as callbacks

Slide 66

Composite pattern class Node(metaclass=ABCMeta): @abstractmethod def children(self): pass def name(self): return self._name ... class Primitive(Node): def __init__(self, name): self._name = name def children(self): return () ... * class Group(Node): def __init__(self, name, children): self._name = name self._children = tuple(children) def children(self): return self._children ...

Slide 67

Visitor pattern class Node(metaclass=ABCMeta): ... @abstractmethod def accept(self, visitor): pass ... class Primitive(Node): ... def accept(self, visitor): visitor.visit_primitive(self) ... class Group(Node): ... def accept(self, visitor): visitor.visit_group(self) ... class Visitor: def visit_primitive(self, visited): ... def visit_group(self, visited): ...

Slide 68

Enumeration Method pattern class Node(metaclass=ABCMeta): @abstractmethod def map(self, callback): pass ... class Primitive(Node): ... def map(self, callback): pass ... * class Group(Node): def __init__(self, name, children): self._name = name self._children = tuple(children) def map(self, callback): for child in self._children: callback(child) ...

Slide 69

Lifecycle Callback pattern class Visitor: def enter(self, group): ... def leave(self, group): ... def visit(self, primitive): ... class Printer: def __init__(self): self.__depth = 0 def enter(self, group): print(self.__depth * ' ' + '<group>') self.__depth += 1 def leave(self, group): self.__depth -= 1 print(self.__depth * ' ' + '</group>') def visit(self, primitive): print(self.__depth * ' ' + '<primitive/>')

Slide 70

Patterns in combination class Node(metaclass=ABCMeta): @abstractmethod def for_all(self, visitor): pass ... class Primitive(Node): ... def for_all(self, visitor): visitor.visit(self) ... * class Group(Node): ... def for_all(self, visitor): visitor.enter(self) for child in self._children: child.for_all(visitor) visitor.leave(self) ...

Slide 71

Object forwarding  Many object forwarding patterns have a simple and general form in Python  Proxy: offer transparent forwarding by one object to another, supporting the target object's protocol  Null Object: represent absence of an object with an object that realises the object protocol with do-nothing or returndefault implementations for its methods

Slide 72

Proxy pattern class TimingProxy: def __init__(self, target, report=print): self.__target = target self.__report = report def __getattr__(self, name): attr = getattr(self.__target, name) def wrapper(*args, **kwargs): start = time() try: return attr(*args, **kwargs) finally: end = time() self.__report(name, end – start) return wrapper if callable(attr) else attr

Slide 73

Null Object pattern A generic Null Object implementation useful, for example, as for test dummy objects, particularly if no specific return values are expected or tested class NullObject: def __call__(self, *args, **kwargs): return self def __getattr__(self, name): return self Method calls on NullObject are chainable as NullObject is callable

Slide 74

Realising values as objects  It is worth differentiating between objects that represent mechanisms...  And may therefore have changing state  And objects that represent values  Their focus is information (e.g., quantities), rather than being strongly behavioural (e.g., tasks), or entity-like (e.g., users)  They should be easily shared and consistent, hence immutable (i.e., like str)

Slide 75

Immutable Value pattern  For a class of value objects...  Provide for rich construction to ensure well-formed objects are easy to create  Provide query but not modifier methods  Consider @property for zero-parameter query methods so they look like attributes  Define __eq__ and __hash__ methods  Provide relational operators if values are ordered (see functools.total_ordering)

Slide 76

In the money @total_ordering class Money: def __init__(self, units, hundredths): self.__units = units self.__hundredths = hundredths @property def units(self): return self.__units @property def hundredths(self): return self.__hundredths def __eq__(self, other): return (self.units == other.units and self.hundredths == other.hundredths) def __lt__(self, other): return ((self.units, self.hundredths) < (other.units, other.hundredths)) ...

Slide 77

On the money @total_ordering class Money: def __init__(self, units, hundredths): self.__total_hundredths = units * 100 + hundredths @property def units(self): return self.__total_hundredths // 100 @property def hundredths(self): return self.__total_hundredths % 100 def __eq__(self, other): return (self.units == other.units and self.hundredths == other.hundredths) def __lt__(self, other): return ((self.units, self.hundredths) < (other.units, other.hundredths)) ...

Slide 78

Enumeration types  Python (as of 3.4) supports enum types  They are similar — but also quite different — to enum types in other languages  Support comes from the enum module  An enumeration type must inherit from either Enum or IntEnum  Enum is the base for pure enumerations  IntEnum is the base class for integercomparable enumerations

Slide 79

Enumerated from enum import Enum class Suit(Enum): spades = 1 hearts = 2 diamonds = 3 clubs = 4 Use IntEnum if easily comparable enumerations are needed Enumeration names are accessible as strings are accessible via the name property and the integer via value values = ['A'] + list(range(2, 11)) + ['J', 'Q', 'K'] [(value, suit.name) for suit in Suit for value in values] Enumeration types are iterable

Slide 80

Ad hoc enums  Enum and IntEnum can be used to create enums on the fly  They are both callable Colour = Enum('Colour', 'red green blue') Colour = Enum('Colour', ('red', 'green', 'blue')) Colour = Enum('Colour', {'red': 1, 'green': 2, 'blue': 3}) class Colour(Enum): red = 1 green = 2 blue = 3

Slide 81

Containers Built-in containers, library collections, tips & tricks

Slide 82

Facts at a glance  Built-in containers address most needs  There are some common iteration patterns to follow (and avoid)  Comprehensions address many common container iteration needs  collections module offers variations on standard sequence and lookup types  collections.abc supports container usage and definition

Slide 83

Container guidance  Although everything is, at one level, a dict, dict is not always the best choice  Nor is list  Choose containers based on usage patterns and mutability  E.g., tuple is immutable whereas list is not  Iteration is the most common container activity, so favour the most direct and efficient iteration style

Slide 84

Built-in sequence types list tuple range list() [] [0, 0x33, 0xCC] ['Albert', 'Einstein', [1879, 3, 14]] [random() for _ in range(42)] tuple() () (42,) ('red', 'green', 'blue') ((0, 0), (3, 4)) range(42) range(1, 100) range(100, 0, -1)

Slide 85

From str to bytes & back  bytes and str are immutable sequence types used for holding strings  str is Unicode and bytes holds, well, bytes  By default, encoding and decoding between str and bytes is based on UTF-8 str encode decode bytes

Slide 86

Built-in lookup types dict dict() {} {'Bohr': True, {n: n**2 for n set set() {'1st', '2nd', {n**2 for n in frozenset frozenset() frozenset({15, frozenset(n**2 'Einstein': False} in range(0, 100)} '3rd'} range(1, 100)} 30, 40}) for n in range(0, 100))

Slide 87

How not to iterate currencies 'EUR': 'GBP': 'NOK': } = { 'Euro', 'British pound', 'Norwegian krone', for code in currencies: print(code, currencies[code]) ordinals = ['first', 'second', 'third'] for index in range(0, len(ordinals)): print(ordinals[index]) for index in range(0, len(ordinals)): print(index + 1, ordinals[index])

Slide 88

How to iterate currencies 'EUR': 'GBP': 'NOK': } = { 'Euro', 'British pound', 'Norwegian krone', for code, name in currencies.items(): print(code, name) ordinals = ['first', 'second', 'third'] for ordinal in ordinals: print(ordinal) for index, ordinal in enumerate(ordinals, 1): print(index, ordinal)

Slide 89

Comprehensive containers List comprehension Set comprehension [n for n in range(2, 100) if n not in {m for l in range(2, 10) for m in range(l*2, 100, l)}] [2, 3, 5, 7, 11, 13, 17, 19, 23, 29, 31, 37, 41, 43, 47, 53, 59, 61, 67, 71, 73, 79, 83, 89, 97]

Slide 90

collections  Built-in container types meet many common and initial needs  But sometimes a different data structure will cut down the amount of code written or the runtime resources used  Built-in container types supplemented by the standard collections module  Container types in collections are not built in, so they do not have display forms

Slide 91

defaultdict  defaultdict offers on-demand creation for keys that are not already present  Derives from dict  Uses a given factory function, such as a class name, to create default values def histogram(data): result = defaultdict(str) for item in data: result[word] += '#' return result Inserts a value of str() for each key looked up that is not already present

Slide 92

Counter  Counter tracks occurrences of a key  It derives from dict but behaves a lot like a bag or a multiset  Counts can be looked up for a key  Each key 'occurrence' can be iterated with open(filename) as input: words = input.read().split() counted = Counter(votes) most_common = counted.most_common(1) most_to_least_common = counted.most_common()

Slide 93

OrderedDict  OrderedDict preserves the order in which keys were inserted  Derives from dict (with some LSP issues...)  One of the most useful applications is to create sorted dictionaries, i.e., initialise from result of sorted on a dict def word_counts(words): counts = OrderedDict() for word in words: counts[word] = 1+ counts.setdefault(word, 0) return counts

Slide 94

ChainMap  ChainMap provides a view over a number of mapping objects  Lookups are in sequence along the chain  Updates apply only to the first mapping — avoid side effects by supplying {} defaults = {'drink': 'tea', 'snack': 'biscuit'} preferences = {'drink': 'coffee'} options = ChainMap(preferences, defaults) for key, value in options.items(): print(key, value, sep=': ') drink: coffee snack: biscuit

Slide 95

deque  A deque is a double-ended queue  Supports efficient append and pop and appendleft and popleft operations  Can be bounded, with overflow causing a pop from the opposite end to the append def tail(filename, lines=10): with open(filename) as source: return deque(source, lines)

Slide 96

namedtuple  namedtuple allows the creation of a tuple type with named attributes  Can still be indexed and iterated  namedtuple is subclass of tuple Date = namedtuple('Date', 'year month day') Date = namedtuple('Date', ('year', 'month', 'day')) sputnik_1 = Date(1957, 10, 4) sputnik_1 = Date(year=1957, month=10, day=4) year, month, day = sputnik_1 year = sputnik_1.year month = sputnik_1[1]

Slide 97

collections.abc  The collections.abc module provides abstract container classes  Using isinstance, these can be used to check whether an object conforms to a protocol, e.g., Container, Hashable, Iterable, MutableSequence, Set  These abstract classes can also be used as mixin base classes for new container classes

Slide 98

Iterators & Generators Iteration abstractions, iterators, generators & wrapped control

Slide 99

Facts at a glance  Iterators enjoy direct protocol and control structure support in Python  Iterators can be defined as classes  Generators are functions that automatically create iterators  Generator expressions support a simple way of generating generators  Generators can abstract with logic

Slide 100

Iteration  An iterator is an object that iterates, following the iterator protocol  An iterable object can be used with for  Iterators and generators are lazy, returning values on demand  You don't need to resolve everything into a list in order to use a series of values  Yield values in sequence, so can linearise complex traversals, e.g., tree structures

Slide 101

Iterables & iterators An object is iterable if it supports the __iter__ special method, which is called by the iter function All iterators are iterable, so the __iter__ method is an identity operation The __next__ special method, called by next, advances the iterator Iteration is terminated by raising StopIteration class Iterable: ... def __iter__(self): return Iterator(...) ... class Iterator: ... def __iter__(self): return self def __next__(self): ... raise StopIteration ... ...

Slide 102

Defining iterators  There are many ways to provide an iterator...  Define a class that supports the iterator protocol directly  Return an iterator from another object  Compose an iterator with iter, using an action and a termination value  Define a generator function  Write a generator expression

Slide 103

iter  Use iter to create an iterator from a callable object and a sentinel value  Or to create an iterator from an iterable def pop_until(stack, end): return iter(stack.pop, end) for popped in pop_until(history, None): print(popped) def repl(): for line in iter(lambda: input('> '), 'exit'): print(evaluate(line))

Slide 104

next  Iterators can be advanced manually using next  Calls the __next__ method  Watch out for StopIteration at the end... def repl(): try: lines = iter(lambda: input('> '), 'exit') while True: line = next(lines) print(evaluate(line)) except StopIteration: pass

Slide 105

Generator expressions  A comprehension-based expression that results in an iterator object  Does not result in a container of values  Must be surrounded by parentheses unless it is the sole argument of a function  May be returned as the result of a function numbers = (random() for _ in range(42)) sum(numbers) sum(random() for _ in range(42))

Slide 106

Generator functions  A generator is an ordinary function that returns an iterator as its result  The presence of a yield or yield from makes a function a generator, and can only be used within a function  yield returns a single value  yield from takes values from another iterator, advancing by one on each call  return in a generator raises StopIteration, passing it any return value specified

Slide 107

Yielding a winner for medal in medals(): print(medal) def medals(): yield 'Gold' yield 'Silver' yield 'Bronze' def medals(): for result in 'Gold', 'Silver', 'Bronze': yield result def medals(): yield from ['Gold', 'Silver', 'Bronze']

Slide 108

@contextmanager  The contextmanager decorator offers simple definition of context managers  Defined in the standard contextlib module  Decorate a generator that holds a single yield statement  The entry action comes before the yield  The exit action follows the yield  If yield has a value it will be bound to the target of the with

Slide 109

It's all about timing class timing: def __init__(self, report=print): self.__report = report def __enter__(self): self.__start = time() def __exit__(self, *args): self.__report(time() – self.__start) with timing(): task_to_time() @contextmanager def timing(report=print): start = time() yield report(time() – start)

Slide 110

Trading places print(os.getcwd()) with pushd('/tmp'): print(os.getcwd()) task() print(os.getcwd()) @contextmanager def pushd(new): old = os.getcwd() os.chdir(new) try: yield finally: os.chdir(old) Should print '/tmp' Should print the same directory before and after the with Without the finally, any exception would propagate out from the yield, terminating the generator and bypassing the directory restore action

Slide 111

Experiment  Define a generator that returns successive days from a start date  Use date and timedelta from datetime  Consider using date.today() as default  Use it to explore different ways of selecting dates that fall on Friday 13th  Use a generator expression  Use a container comprehension  Use filter

Slide 112

Functional Thinking Functional programming techniques, tips & tricks

Slide 113

Facts at a glance  Functions are the principal units of composition  Functions are pure, i.e., no side effects  Immutability, i.e., query and create state rather than modify it  Declarative over imperative style, e.g., use of comprehensions over loops  Deferred evaluation, i.e., be lazy

Slide 114

Functional guidance  Immutability...  Prefer to return new state rather than modifying arguments and attributes  Resist the temptation to match every query or get method with a modifier or set  Expressiveness...  Consider where loops and intermediate variables can be replaced with comprehensions and existing functions

Slide 115

Mutability pitfalls  Python is reference-based language, so objects are shared by default  Easily accessible data or state-modifying methods can give aliasing surprises  Note that true and full object immutability is not possible in Python  Default arguments should be of immutable type, e.g., use tuple not list  Changes persist between function calls

Slide 116

Fewer explicit loops  A common feature of functional programming is fewer explicit loops  Recursion, but note that Python does not support tail recursion optimisation  Comprehensions — very declarative, very Pythonic  Higher-order functions, e.g., map applies a function over an iterable sequence  Existing algorithmic functions, e.g., min, str.split, str.join

Slide 117

Recursion  Recursion may be a consequence of data structure traversal or algorithm  E.g., iterating over a tree structure  E.g., quicksort  But it can be used as an alternative to looping in many simple situations  But beware of efficiency concerns and Python's limits (see sys.getrecursionlimit)

Slide 118

Reducing modifiable state def factorial(n): result = 1 while n > 0: result *= n n -= 1 return result Two variables being modified explicitly def factorial(n): result = 1 for i in range(1, n + 1): result *= i return result def factorial(n): if n > 0: return n * factorial(n – 1) else: return 1 No variables modified

Slide 119

Expressions versus statements There is a tendency in functional programming to favour expressions... def factorial(n): if n > 0: return n * factorial(n – 1) else: return 1 def factorial(n): return n * factorial(n – 1) if n > 0 else 1 factorial = lambda n: n * factorial(n – 1) if n > 0 else 1 But using a lambda bound to a variable instead of a single-statement function is not considered Pythonic and means factorial lacks some metadata of a function, e.g., a good __name__.

Slide 120

Comprehensions  A comprehension is used to define a sequence of values declaratively  Sequence of values defined by intension as an expression, rather than procedurally in terms of loops and modifiable state  They have a select...from...where structure  Many common container-based looping patterns are captured in the form of container comprehensions

Slide 121

Imperative versus declarative def is_leap_year(year): return year % 4 == 0 and year % 100 != 0 or year % 400 == 0 leap_years = [] for year in range(2000, 2030): if is_leap_year(year): leap_years.append(year) [2000, 2004, 2008, 2012, 2016, 2020, 2024, 2028] Imperative list initialisation leap_years = [year for year in range(2000, 2030) if is_leap_year(year)] [2000, 2004, 2008, 2012, 2016, 2020, 2024, 2028] List comprehension

Slide 122

Higher-order functions  A higher-order function...  Takes one or more functions as arguments  Returns one or more function as its result  Takes and returns functions  Higher-order functions often used to abstract common iteration operations  Hides the mechanics of repetition  Comprehensions are often an alternative to such higher-order functions

Slide 123

map  map applies a function over an iterable to produce a new iterable  Can sometimes be replaced with a comprehension or generator expression that has no predicate  Often shorter if no lambdas are involved map(len, 'The cat sat on the mat'.split()) (len(word) for word in 'The cat sat on the mat'.split())

Slide 124

filter  filter includes only values that satisfy a given predicate in its generated result  Can sometimes be replaced with a comprehension or generator expression that has a predicate  Often shorter if no lambdas are involved filter(lambda score: score > 50, scores) (score for score in scores if score > 50)

Slide 125

reduce  functools.reduce implements what is know as a fold left operation  Reduces a sequence of values to a single value, left to right, with the accumulated value on the left and the other on the right def factorial(n): return reduce(lambda l, r: l*r, range(1, n+1), 1) def factorial(n): return reduce(operator.mul, range(1, n+1), 1) int.__mul__ would be a less general alternative

Slide 126

The operator module operator exports named functions corresponding to operators Comparison operations eq(a, b) a ne(a, b) a lt(a, b) a le(a, b) a gt(a, b) a ge(a, b) a is_(a, b) a is_not(a, b) a contains(a, b) == b != b < b <= b > b >= n is b is not b a in b Binary arithmetic operations add(a, b) sub(a, b) mul(a, b) truediv(a, b) floordiv(a, b) mod(a, b) pow(a, b) a + b a - b a * b a / b a // b a % b a**b Unary operations pos(a) +a neg(a) -a invert(a) ~a

Slide 127

Experiment  Use reduce to implement your own version of map  Can be written as a single-line method  Return the result as a list def my_map(func, iterable): ... my_map(lambda n: n*2, [3, 1, 4, 1, 5, 9]) [6, 2, 8, 2, 10, 18]

Slide 128

Currying In mathematics and computer science, currying is the technique of translating the evaluation of a function that takes multiple arguments (or a tuple of arguments) into evaluating a sequence of functions, each with a single argument (partial application). It was introduced by Moses Schönfinkel and later developed by Haskell Curry. http://en.wikipedia.org/wiki/Currying

Slide 129

Nested functions & lambdas  Nested functions and lambdas bind to their surrounding scope  I.e., they are closures def curry(function, first): def curried(second): return function(first, second) return curried def curry(function, first): return lambda second: function(first, second) hello = curry(print, 'Hello') hello('World')

Slide 130

Function wrapping Force non-positional use of subsequent parameters def timed_function(function, *, report=print): from time import time def wrapper(*args): start = time() try: function(*args) finally: report(time() – start) return wrapper wrapped = timed_function(long_running) wrapped(arg_for_long_running)

Slide 131

partial application  Values can be bound to a function's parameters using functools.partial  Bound positional and keyword arguments are supplied on calling the resulting callable, other arguments are appended quoted = partial(print, '>') quoted() print('>') quoted('Hello, World!') print('>', 'Hello, World!') quoted('Hello, World!', end=' <\n') print('>', 'Hello, World!', end=' <\n')

Slide 132

Built-in algorithmic functions min(iterable) min(iterable, default=value) min(iterable, key=function) max(iterable) max(iterable, default=value) max(iterable, key=function) sum(iterable) sum(iterable, start) any(iterable) all(iterable) sorted(iterable) sorted(iterable, key=function) sorted(iterable, reverse=True) zip(iterable, ...) ... Default used if iterable is empty The function to transform the values before determining the lowest one One or more iterables can be zipped together as tuples, i.e., effectively converting rows to columns, but zipping two iterables is most common

Slide 133

Functions from itertools chain(iterable, ...) compress(iterable, selection) dropwhile(predicate, iterable) takewhile(predicate, iterable) count() count(start) count(start, step) islice(iterable, stop) islice(iterable, start, stop) islice(iterable, start, stop, step) cycle(iterable) repeat(value) repeat(value, times) zip_longest(iterable, ...) zip_longest(iterable, ..., fillvalue=fill) ...

Slide 134

Concurrency Coroutines, threading, processes & futures

Slide 135

Facts at a glance  Generators can be used as coroutines  Threading is based on native threads  Python's threading is subject to GIL constraints on many implementations  Multiprocessing has API-compatible support plus IPC mechanisms  Executors and futures offer a simple task-based concurrency model

Slide 136

Concurrency  Concurrency can...       Be implicit in the problem domain Be a by-product of the implementation Improve performance Worsen performance Simplify and decouple code Complicate and couple code  It's all a question of implementation mechanisms and choices!

Slide 137

Coroutines Coroutines are computer program components that generalize subroutines for nonpreemptive multitasking, by allowing multiple entry points for suspending and resuming execution at certain locations. Coroutines are well-suited for implementing more familiar program components such as cooperative tasks, exceptions, event loop, iterators, infinite lists and pipes. https://en.wikipedia.org/wiki/Coroutine

Slide 138

Generators as coroutines  yield and yield from cause suspension of the current function execution  On resumption, e.g., via next, execution continues where it left off, so that local variable state is retained  The alternative to using coroutines is often more complex  E.g., a state machine to remember where a task was last time it was executed

Slide 139

Reducing complexity Symmetry is a complexity-reducing concept (co-routines include subroutines); seek it everywhere. Alan Perlis "Epigrams in Programming"

Slide 140

Coroutine communication  Can communicate with a generator     send passes a value to it throw raises an exception within it close terminates it send and throw return next yielded value  Although often used as statements, yield and yield from are expressions  When invoked by next, they return None  When send is used, value sent is returned

Slide 141

Coroutine action def evaluate(): input = '' for n in count(): input = (yield str(n) + ' ' + evaluated(input)) def repl(): evaluator = evaluate() next(evaluator) try: for line in iter(lambda: input('> '), 'exit'): print(evaluator.send(line)) finally: evaluator.close()

Slide 142

Multithreading Multithreading is just one damn thing before, after, or simultaneous with another. Andrei Alexandrescu

Slide 143

threading  The threading module offers constructs for creating threads and locking  Threads are built on native OS threads  Locking primitives are based on binary semaphores, which are in turn based on OS mutexes  Other synchronisation primitives, such as events, and thread-based constructs, such as timers and thread-local data, are also defined

Slide 144

Thread definition Inheritance-based approach Composition-based approach task = Task() task.start() ... task.join() task = Thread(target=run) task.start() ... task.join() class Task(Thread): def run(): ... def run(): ...

Slide 145

Global Interpreter Lock (GIL)  In CPython, the GIL introduces a concurrency bottleneck  It is a mutex that restricts system threads from concurrently executing Python code  The thread-unsafe reference-counted memory management model is the primary reason for the GIL's existence  A feature of CPython but not necessarily other implementations, e.g., Jython relies on JVM GC and IronPython relies on .NET

Slide 146

Synchronisation  Python offers a number of different types of synchronisation primitive  E.g., Lock, RLock, Condition, Semaphore, BoundedSemaphore, Event, Barrier  All the locking-related primitives are context managers  Use with rather than acquire and release unless for non-blocked acquisition or a timeout is required

Slide 147

Sharing & synchronisation Shared memory is like a canvas where threads collaborate in painting images, except that they stand on the opposite sides of the canvas and use guns rather than brushes. The only way they can avoid killing each other is if they shout "duck!" before opening fire. Bartosz Milewski http://bartoszmilewski.com/2013/12/10/functional-data-structures-and-concurrency-in-c/

Slide 148

Multiprocessing  An alternative to threads is to use OS processes to express concurrency  A standard workaround for GIL issues  Processes naturally give better isolation and long-term concurrency  But may not be effective for fine-grained concurrency  Any values exchanged between processes must be picklable

Slide 149

multiprocessing  The multiprocessing module has similar API to threading  Including locking primitives  Also includes IPC mechanisms  Different types of inter-process queues and Pipe, which results in a pair of Connection objects  Value and Array can be used to create and access objects in shared memory

Slide 150

Process action def evaluate(client): try: for n in count(1): input = client.recv() client.send(str(n) + ' ' + evaluated(input)) except EOFError: pass def repl(): here, there = Pipe() evaluator = Process(target=evaluate, args=(there,)) evaluator.start() for line in iter(lambda: input('> '), 'exit'): here.send(line) print(here.recv()) here.close() evaluator.join()

Slide 151

Asynchronous functions  Asynchronous function approaches simplify use of threads and processes  Instead of blocking on a function call, a function is executed concurrently  A future — a virtual proxy also known as an IOU or a deferred — is a queryable place holder for the function's result  asyncio and concurrent.futures are based on this model

Slide 152

Executors & futures  Executors simplify use of concurrency execution resources  One or more tasks can be submitted to an executor for execution  Execution mechanics and management are hidden from the submitter  The caller uses a future to synchronise with and collect execution results  Reduces need for other synchronisation

Slide 153

concurrent.futures  An executor is a pools of workers  Either threads, in the case of ThreadPoolExecutor, or processes, in the case of ProcessPoolExecutor  Executors are context managers  To submit work...  Call submit to submit an individual item  Call map to submit multiple items — a returned generator plays the future role

Slide 154

In the future... with ThreadPoolExecutor(max_workers=1) as executor: future = executor.submit(task, data) print(future.result()) with ThreadPoolExecutor(max_workers=4) as executor: futures = [ executor.submit(task, data) for data in work ] for future in futures: print(future.result()) with ThreadPoolExecutor(max_workers=4) as executor: for result in executor.map(task, work): print(result)

Slide 155

Synchronisation quadrant Mutable Unshared mutable data needs no synchronisation Mutable data shared between threads needs synchronisation Unshared Shared Unshared immutable data needs no synchronisation Shared immutable data needs no synchronisation Immutable

Slide 156

Modules & Packages The mechanics of organising source code

Slide 157

Facts at a glance  A package hierarchically organises modules and other packages  Where a module corresponds to a file, a regular package corresponds to a directory  Modules and packages define global namespaces  Extension modules allow extension of Python itself

Slide 158

Modules  A module in Python corresponds to a file with a .py extension  import is used to access features in another module  A module's __name__ corresponds to its file name without the extension  When run as a script (e.g., using the –m option), the root module has '__main__' as its __name__

Slide 159

Extensions & embedding  The Python interpreter can be extended using extension modules  Python has a C API that allows programmatic access in C (or C++) and, therefore, any language callable from C  Python can also be embedded in an application as a scripting language  E.g., allow an application's features to be scripted in Python

Slide 160

Packages  Packages are a way of structuring the module namespace  A submodule bar within a package foo represents the module foo.bar  A regular package corresponds to a directory of modules (and packages)  A regular package directory must have an __init__.py file (even if it's empty)  The package name is the directory name

Slide 161

Namespaces  A namespace is a mapping from names to objects, e.g., variables  A namespace is implemented as a dict  Global, local, built-in and nested  Each module gets its own global namespace, as does each package  import pulls names into a namespace  Within a scope, names in a namespace can be accessed without qualification

Slide 162

Packages & import  A submodule can be imported with respect to its package  E.g., import package.submodule  Relative naming can be used to navigate within a package  Main modules must use absolute imports  Submodules seen by wildcard import can be specified by assigning a list of module names (as strings) to __all__ in __init__.py

Slide 163

Packages, modules & imports Top-level regular package Subpackage root/ __init__.py first/ __init__.py foo.py bar.py second/ __init__.py foo.py bar.py # empty, but necessary __all__ = ['foo'] from . import bar from .. import first from ..first import foo

Slide 164

Experiment  Write a simple script that lists the package and module hierarchy  Use current directory if no path supplied  Use os.listdir, os.walk or pathlib for traversal  No need to open and load modules, just detect packages by matching __init__.py and modules by matching *.py  List the results using indentation and/or in some XML-like form

Slide 165

Outroduction All good things...

Slide 166

The end?  The foundation is all there  Python Foundation course and this course  But a foundation means little unless you build on it  There is more to the language...  There is more to the library...  There are more libraries...  Enjoy!

Slide 167

Labs & Homework To do...

Slide 168

Guidance  Labs should be undertaken in pairs using Cyber-Dojo  Swap between driving and navigator roles within the same pair  Homework is carried out individually in your own environment  Do not spend too much time on this — an hour or so — and try not to overengineer the solution!

Slide 169

Lab: Day 1 Money Write a class (and unit tests) for a class that represents a monetary value:  It should hold an amount that can be queried as units and hundredths (e.g., pounds and pence, euros and cents, dollars and cents).  It should be possible to compare money values for equality and for ordering using relational operators.  It should support non-lossy addition and subtraction, i.e., don't use float to hold the amount.  Negative money is allowed.  Money values should not offer modifier methods.  Don't worry about different currencies.

Slide 170

Lab: Day 1 Option 1: Incompatible currencies Refine your money class and tests so that instances have a currency code associated with them, e.g., 'NOK' for Norwegian krone, 'EUR' for Euro:  Only money values with the same currency code can be added or subtracted from one another.  Money values of different currencies can be compared for equality, but always return False.  Money values of different currencies cannot be ordered.

Slide 171

Lab: Day 1 Option 2: Compatible currencies Allow money values of different currencies to be added together, but so that they result in composite type, i.e., a money bag:  They support addition, but do not resolve down to a single currency, e.g., 100 NOK + 10 EUR + 200 NOK + 5 EUR results in a money bag of 300 NOK + 15 EUR.  Money bags can be compared for equality, but cannot be ordered.

Slide 172

Homework A simple testing framework Write a small testing framework that executes all the tests defined within a given class:  Provide a run function that takes a class of test cases as its argument.  For each test method in the given class, create an instance of the class and execute the test method against it, reporting back the result to the console, and then reporting a summary of the result at the end of the test run.  A test method is one whose name begins with test.  Use the built-in assert statement for assertions within test methods.

Slide 173

Homework An example of use For the following function, which determines whether a year is a leap year or not: def is_leap_year(year): ... The test framework should support the following test code: class LeapYearTests: def test_years_not_divisible_by_4_are_not_leap_years(self): assert not is_leap_year(2015) def test_years_divisible_by_4_but_not_by_100_are_leap_years(self): assert is_leap_year(2016) def test_years_divisible_by_100_but_not_by_400_are_not_leap_years(self): assert not is_leap_year(1900) def test_years_divisible_by_400_are_leap_years(self): assert is_leap_year(2000) Executed as follows: run(LeapYearTests)

Slide 174

Homework Option 1: Using decorators to mark test methods If you have enough time, consider changing the test runner so that instead of relying on a method prefix, a decorator is used to indicate that a method is a test method: class LeapYearTests: @test def years_not_divisible_by_4_are_not_leap_years(self): assert not is_leap_year(2015) @test def years_divisible_by_4_but_not_by_100_are_leap_years(self): assert is_leap_year(2016) ... This is syntactically similar to the annotations used in JUnit, but in Python it will work by a different effect. Define a test decorator that adds an attribute to the test method; it is this attribute that the run function will use to select the test methods for execution.

Slide 175

Homework Option 2: Using decorators to pass data to a test method If you have enough time, consider refining your decorator design so that data can optionally be passed into the test case: class LeapYearTests: @test @data(2015, 1999, 1) def years_not_divisible_by_4_are_not_leap_years(self, year): assert not is_leap_year(year) @test @data(2016) @data(1984) @data(4) def years_divisible_by_4_but_not_by_100_are_leap_years(self, year): assert is_leap_year(year) @test def years_divisible_by_100_but_not_by_400_are_not_leap_years(self): assert not is_leap_year(1900) @test @data(*range(400, 2401, 400)) def years_divisible_by_400_are_leap_years(self, year): assert is_leap_year(year)

Slide 176

Homework Further variations If you still have enough time, or would prefer something different to the previous two options, here are some more ideas to play with:  Use a keyword argument to specify the output for test results, with the sys.stdout as default.  Use an Enumeration Method callback from the test runner function to communicate test execution results (default to print).  Use multiple callbacks to indicate different events in a test run's lifecycle, i.e., the Lifecycle Callback pattern. Either pass in callables via keyword arguments or use the Visitor pattern.  Instead of instantiating a test object, consider using module-level functions or static methods as the basis for test cases.  Define a decorator that checks that an exception has been thrown from a test case, failing if no or the wrong type is thrown.  Execute test cases concurrently instead of sequentially.

Slide 177

Lab: Day 2 Recently used list Write a class (and unit tests) for a container that represents a recently used list, i.e., a sequence ordered by reverse insertion order and restricted by uniqueness:  A recently used list is initially empty.  Strings can be added to the list, but non-string values (including None) are disallowed.  Items can be indexed, which counts from zero.  Additions are retained in stack ordering (i.e., LIFO), so the most recently added item is first, the least recently added item is last, and so on.  Items in the list are unique, so duplicate insertions are moved to the head rather than added, i.e., the length will not increase.

Slide 178

Lab: Day 2 Option 1: Support for non-string items Generalise the implementation so that addition of non-string values is supported. This is not, however, simply a case of removing the type restriction in add:  Because the list is set-like — i.e., items are unique — items also play the role of keys, which means they should be immutable.  There is no guarantee of a type's immutability, but a common convention is that hashable types are immutable and, therefore, non-hashable types are not.  It should therefore be possible to add tuples and integers to a recently used list, but not lists and sets.  But note that addition of None is still disallowed.

Slide 179

Lab: Day 2 Option 2: Additional container operations Add methods to improve support for a more typical container protocol:  A clear method to empty the list.  A pop method to remove the oldest item in the list.  A remove method to remove an item by value.  Support common lookup operations such as in, not in and index.  Support comparison for equality and inequality.  Support the iterator protocol.  Allow items to be deleted by index.

Slide 180