Python 3 Orientado A Objeto

Python 3 Object Oriented
Programming
Harness the power of Python 3 objects
Dusty Phillips
BIRMINGHAM - MUMBAI
Python 3 Object Oriented Programming

Copyright 2010 Packt Publishing
All rights reserved. No part of this book may be reproduced, stored in a retrieval
system, or transmitted in any form or by any means, without the prior written
permission of the publisher, except in the case of brief quotations embedded in
critical articles or reviews.
Every effort has been made in the preparation of this book to ensure the accuracy
of the information presented. However, the information contained in this book
is sold without warranty, either express or implied. Neither the author nor Packt
Publishing, and its dealers and distributors will be held liable for any damages
caused or alleged to be caused directly or indirectly by this book.
Packt Publishing has endeavored to provide trademark information about all of the
companies and products mentioned in this book by the appropriate use of capitals.
However, Packt Publishing cannot guarantee the accuracy of this information.
First published: July 2010
Production Reference: 1160710
Published by Packt Publishing Ltd.

32 Lincoln Road
Olton
Birmingham, B27 6PA, UK.
ISBN 978-1-849511-26-1
www.packtpub.com
Cover Image by Asher Wishkerman ([email protected] )
Credits
Author
Dusty Phillips
Reviewers
Jason Chu
Editorial Team Leader

Mithun Sehgal
Project Team Leader
Lata Basantani
Michael Driscoll
Dan McGee
Lawrence Oluyede
Acquisition Editor
Steven Wilding
Development Editor
Mayuri Kokate
Technical Editor
Vanjeet D'souza
Indexer
Hemangini Bari
Project Coordinator
Jovita Pinto
Proofreader
Chris Smith
Graphics
Geetanjali Sawant
Production Coordinator
Shantanu Zagade
Cover Work
Shantanu Zagade
About the Author

Dusty Phillips is a Canadian freelance software developer, teacher, martial artist,
and open source aficionado. He is closely affiliated with the Arch Linux community
and other open source projects. He maintains the Arch Linux storefronts, and
compiled the popular Arch Linux Handbook. Dusty holds a Master's degree in
Computer Science specializing in Human-Computer Interaction. He currently
has six different Python interpreters installed on his computer.
I would like to thank my editors, Steven Wilding and Mayuri Kokate
for well-timed encouragement and feedback. Many thanks to friend
and mentor Jason Chu for getting me started in Python and for
patiently answering numerous questions on Python, GIT, and life
over the years. Thanks to my father, C. C. Phillips, for inspiring me
to write while editing his terrific works of fiction. Finally, thanks
to every person who has said they can't wait to buy my book; your
enthusiasm has been a huge motivational force.
About the Reviewers

Jason Chu is the CTO and part founder of Oprius Software Inc. He's developed
software professionally for over 8 years. Chu started using Python in 2003 with
version 2.2. When not developing personal or professional software, he spends his
time teaching karate, playing go, and having fun in his hometown: Victoria, BC,
Canada. You'll often find him out drinking the Back Hand of God Stout at Christie's
Carriage House.
Michael Driscoll has been programming Python for almost 4 years and has
dabbled in other languages since the late nineties. He graduated from university
with a Bachelor's degree in Science, majoring in Management Information Systems.
Michael enjoys programming for fun and profit. His hobbies include biblical
apologetics, blogging about Python at http://www.blog.pythonlibrary.org/,
and learning photography. Michael currently works for the local government
where he programs with Python as much as possible. This is his first book as a
technical reviewer.
I would like to thank my mom without whom I never would have
grown to love learning as much as I do. I would also like to thank
Scott Williams for forcing me to learn Python as, without him, I
wouldn't have even known that the language existed. Most of all, I
want to thank Jesus for saving me from myself.
Dan McGee is a software developer currently living in Chicago, Illinois. He has
several years of experience working full-time in the Chicago area doing primarily
Java web development; however, he has also been spotted working in a variety of
other languages. Dan has also worked on a handful of freelance projects. In 2007,
Dan became a developer for the Arch Linux distribution and has been doing various
projects related to that since, including hacking on the package manager code, being
a part-time system admin, and helping maintain and improve the website.
Lawrence Oluyede is a 26 years old software development expert in Python and
web programming. He's glad that programming is going parallel and functional
languages are becoming mainstream. He has been a co-author and reviewer for the
first Ruby book in Italian (Ruby per applicazioni web) published by Apogeo. He has also
contributed to other books in the past like the Python Cookbook (http://www.amazon.
com/Python-Cookbook-Alex-Martelli/dp/0596007973/) and The Definitive Guide
to Django (http://www.amazon.com/Definitive-Guide-Django-DevelopmentRight/dp/1590597257).
Table of Contents
Preface
Chapter 1: Object-oriented Design
1
7
Object-oriented?
Objects and classes
Specifying attributes and behaviors
Data describes objects
Behaviors are actions
Hiding details and creating the public interface
Composition and inheritance
Inheritance
7
9
11
11
13
14
17
20
Case study
Exercises
Summary
24
31
32
Inheritance provides abstraction

Multiple inheritance
Chapter 2: Objects in Python

Creating Python classes
Adding attributes
Making it do something
Initializing the object
Explaining yourself
Modules and packages
Organizing the modules
Absolute imports
Relative imports
Who can access my data?

Case study
Exercises
Summary
22
23
33
33
35
35
38
41
43
45
46
47
50
53
61
62
Table of Contents
Chapter 3: When Objects are Alike
63
Chapter 4: Expecting the Unexpected
95
Basic inheritance
Extending built-ins
Overriding and super
The diamond problem
Different sets of arguments
Polymorphism
Case study
Exercises
Summary
Raising exceptions
Raising an exception
What happens when an exception occurs?
Handling exceptions
Exception hierarchy
Defining our own exceptions
Exceptions aren't exceptional
Case study
Exercises
Summary
63
66
67
68
71
75
78
80
93
94
95
98
99
101
106
108
109
112
122
123
Chapter 5: When to Use Object-oriented Programming
125
Chapter 6: Python Data Structures
157
Treat objects as objects

Using properties to add behavior to class data
How it works
Decorators: another way to create properties
When should we use properties?
Managing objects
Removing duplicate code
In practice
Or we can use composition
Case study
Exercises
Summary
Empty objects
Tuples and named tuples
Named tuples
Dictionaries
[ ii ]
125
129
132
134
135
137
140
142
145
147
154
156
157
159
161
162
Table of Contents
When should we use dictionaries?

Using defaultdict
Lists
Sorting lists
Sets
Extending built-ins
Case study
Exercises
Summary
166
166
168
171
173
177
182
188
189
Chapter 7: Python Object-oriented Shortcuts
191
Chapter 8: Python Design Patterns I
227
Python built-in functions

Len
Reversed
Enumerate
Zip
Other functions
Comprehensions
List comprehensions
Set and dictionary comprehensions
Generator expressions
Generators
An alternative to method overloading
Default arguments
Variable argument lists
Unpacking arguments
Functions are objects too
Using functions as attributes
Callable objects
Case study
Exercises
Summary
Design patterns
Decorator pattern
Decorator example
Decorators in Python
Observer pattern
Observer example
Strategy pattern
Strategy example
[ iii ]
191
192
192
193
194
196
197
198
200
201
203
205
207
208
212
213
218
219
220
224
225
227
229
230
233
235
235
237
238
Table of Contents
Strategy in Python
State pattern
State example
State versus strategy
Singleton pattern
Singleton implementation
Module variables can mimic singletons
Template pattern
Template example
Exercises
Summary
240
240
241
247
247
248
249
251
252
255
256
Chapter 9: Python Design Patterns II
257
Chapter 10: Files and Strings
283
Adapter pattern
Facade pattern
Flyweight pattern
Command pattern
Abstract factory pattern
Composite pattern
Exercises
Summary
Strings
String manipulation
String formatting
257
260
263
267
271
276
280
281
283
284
287
Escaping braces
Keyword arguments
Container lookups
Object lookups
Making it look right
288
288
289
291
291
Strings are Unicode
294
Mutable byte strings

File IO
Placing it in context
Faking files
Storing objects
Customizing pickles
Serializing web objects
Exercises
Summary
297
299
301
302
303
305
308
310
312
Converting bytes to text

Converting text to bytes
295
296
[ iv ]
Table of Contents
Chapter 11: Testing Object-oriented Programs

Why test?
Test-driven development
Unit testing
Assertion methods
Additional assertion methods in Python 3.1
Reducing boilerplate and cleaning up

Organizing and running tests
Ignoring broken tests
Testing with py.test
One way to do setup and cleanup
A completely different way to set up variables
Test skipping with py.test
py.test extras
How much testing is enough?
Case Study
Implementing it
Exercises
Summary
Chapter 12: Common Python 3 Libraries

Database access
Introducing SQLAlchemy
Adding and querying objects
SQL Expression Language
313
313
315
316
318
319
320
322
323
324
326
329
333
335
336
339
340
345
346
347
348
349
351
352
Pretty user interfaces

TkInter
PyQt
Choosing a GUI toolkit
XML
ElementTree
353
354
358
361
362
362
lxml
CherryPy
366
368
Exercises
Summary
377
378
Constructing XML documents
366
A full web stack?
370
Index
379
[]
Preface
This book will introduce you to the terminology of the object-oriented paradigm,
focusing on object-oriented design with step-by-step examples. It will take you from
simple inheritance, one of the most useful tools in the object-oriented programmer's
toolbox, all the way through to cooperative inheritance, one of the most complicated.
You will be able to raise, handle, define, and manipulate exceptions.
You will be able to integrate the object-oriented and not-so-object-oriented aspects of
Python. You will also be able to create maintainable applications by studying higherlevel design patterns. You'll learn the complexities of string and file manipulation
and how Python distinguishes between binary and textual data. Not one, but
two very powerful automated testing systems will be introduced to you. You'll
understand the joy of unit testing and just how easy unit tests are to create. You'll
even study higher-level libraries such as database connectors and GUI toolkits and
how they apply object-oriented principles.
What this book covers
Chapter 1, Object-oriented Design covers important object-oriented concepts. It deals

mainly with abstraction, classes, encapsulation, and inheritance. We also briefly look
into UML to model our classes and objects.
Chapter 2, Objects in Python discusses classes and objects and how they are used in
Python. We will learn about attributes and behaviors in Python objects, and also the
organization of classes into packages and modules. And lastly we shall see how to
protect our data.
Chapter 3, When Objects are Alike gives us a more in-depth look into inheritance. It
covers multiple inheritance and shows us how to inherit from built-ins. This chapter
also covers polymorphism and duck typing.
Preface
Chapter 4, Expecting the Unexpected looks into exceptions and exception handling. We
shall learn how to create our own exceptions. It also deals with the use of exceptions
for program flow control.
Chapter 5, When to Use Object-oriented Programming deals with objects; when to create
and use them. We will see how to wrap data using properties, and restricting data
access. This chapter also discusses the DRY principle and how not to repeat code.
Chapter 6, Python Data Structures covers object-oriented features of data structures.
This chapter mainly deals with tuples, dictionaries, lists, and sets. We will also see
how to extend built-in objects.
Chapter 7, Python Object-oriented Shortcuts as the name suggests, deals with little
time-savers in Python. We shall look at many useful built-in functions, then move
on to using comprehensions in lists, sets, and dictionaries. We will learn about
generators, method overloading, and default arguments. We shall also see how
to use functions as objects.
Chapter 8, Python Design Patterns I first introduces us to Python design patterns. We
shall then see the decorator pattern, observer pattern, strategy pattern, state pattern,
singleton pattern, and template pattern. These patterns are discussed with suitable
examples and programs implemented in Python.
Chapter 9, Python Design Patterns II picks up where the previous chapter left us. We
shall see the adapter pattern, facade pattern, flyweight pattern, command pattern,
abstract pattern, and composite pattern with suitable examples in Python.
Chapter 10, Files and Strings looks at strings and string formatting. Bytes and byte
arrays are also discussed. We shall also look at files, and how to write and read data
to and from files. We shall look at ways to store and pickle objects, and finally the
chapter discusses serializing objects.
Chapter 11, Testing Object-oriented Programs opens with the use of testing and why
testing is so important. It focuses on test-driven development. We shall see how to
use the unittest module, and also the py.test automated testing suite. Lastly we
shall cover code coverage using coverage.py.
Chapter 12, Common Python 3 Libraries concentrates on libraries and their utilization
in application building. We shall build databases using SQLAlchemy, and user
interfaces TkInter and PyQt. The chapter goes on to discuss how to construct XML
documents and we shall see how to use ElementTree and lxml. Lastly we will use
CherryPy and Jinja to create a web application.
[]
Preface
What you need for this book
In order to compile and run the examples mentioned in this book you require the
following software:
Python version 3.0 or higher
py.test
coverage.py
SQLAlchemy
pygame
PyQt
CherryPy
lxml
Who this book is for
If you're new to object-oriented programming techniques, or if you have basic

Python skills, and wish to learn in depth how and when to correctly apply
object-oriented programming in Python, this is the book for you.
If you are an object-oriented programmer for other languages you will also find
this book a useful introduction to Python, as it uses terminology you are already
familiar with.
Python 2 programmers seeking a leg up in the new world of Python 3 will also find
the book beneficial but you need not necessarily know Python 2.
Conventions
In this book, you will find a number of styles of text that distinguish between
different kinds of information. Here are some examples of these styles, and an
explanation of their meaning.
Code words in text are shown as follows: "We can access other Python modules
through the use of the import statement."
[]
Preface
A block of code is set as follows:

class Friend(Contact):
def __init__(self, name, email, phone):
self.name = name
self.email = email
self.phone = phone
When we wish to draw your attention to a particular part of a code block, the
relevant lines or items are set in bold:
self.name = name
self.email = email
self.phone = phone
Any command-line input or output is written as follows:

>>> e = EmailableContact("John Smith", "[email protected]")
>>> Contact.all_contacts
New terms and important words are shown in bold. Words that you see on the
screen, in menus or dialog boxes for example, appear in the text like this: "We use
this feature to update the label to a new random value every time we click the
Roll! button".
Warnings or important notes appear in a box like this.
Tips and tricks appear like this.
Reader feedback
Feedback from our readers is always welcome. Let us know what you think about
this bookwhat you liked or may have disliked. Reader feedback is important for us
to develop titles that you really get the most out of.
To send us general feedback, simply send an e-mail to [email protected],
and mention the book title via the subject of your message.
[]
Preface
If there is a book that you need and would like to see us publish, please
send us a note in the SUGGEST A TITLE form on www.packtpub.com or
e-mail [email protected].
If there is a topic that you have expertise in and you are interested in either writing
or contributing to a book, see our author guide on www.packtpub.com/authors.
Customer support
Now that you are the proud owner of a Packt book, we have a number of things to
help you get the most from your purchase.
Downloading the example code for this book
You can download the example code files for all Packt books you have
purchased from your account at http://www.PacktPub.com. If you
purchased this book elsewhere, you can visit http://www.PacktPub.
com/support and register to have the files e-mailed directly to you.
Errata
Although we have taken every care to ensure the accuracy of our content, mistakes
do happen. If you find a mistake in one of our booksmaybe a mistake in the text
or the codewe would be grateful if you would report this to us. By doing so, you
can save other readers from frustration and help us improve subsequent versions
of this book. If you find any errata, please report them by visiting http://www.
packtpub.com/support, selecting your book, clicking on the let us know link, and
entering the details of your errata. Once your errata are verified, your submission
will be accepted and the errata will be uploaded on our website, or added to any list
of existing errata, under the Errata section of that title. Any existing errata can be
viewed by selecting your title from http://www.packtpub.com/support.
[]
Preface
Piracy
Piracy of copyright material on the Internet is an ongoing problem across all media.
At Packt, we take the protection of our copyright and licenses very seriously. If you
come across any illegal copies of our works, in any form, on the Internet, please
provide us with the location address or website name immediately so that we can
pursue a remedy.
Please contact us at [email protected] with a link to the suspected
pirated material.
We appreciate your help in protecting our authors, and our ability to bring you
valuable content.
Questions
You can contact us at [email protected] if you are having a problem with

any aspect of the book, and we will do our best to address it.
[]
Object-oriented Design
In software development, design is often considered the step done before
programming. This isn't true; in reality, analysis, programming, and design
tend to overlap, combine, and interweave. In this chapter, we will learn:
What object-oriented means
The difference between object-oriented design and object-oriented

programming
The basic principles of object-oriented design
Basic Unified Modeling Language and when it isn't evil
Object-oriented?
Everyone knows what an object is: a tangible "something" that we can sense, feel, and
manipulate. The earliest objects we interact with are typically baby toys. Wooden
blocks, plastic shapes, and over-sized puzzle pieces are common first objects. Babies
learn quickly that certain objects do certain things. Triangles fit in triangle-shaped
holes. Bells ring, buttons press, and levers pull.
The definition of an object in software development is not so very different. Objects
are not typically tangible somethings that you can pick up, sense, or feel, but they are
models of somethings that can do certain things and have certain things done to them.
Formally, an object is a collection of data and associated behaviors.
So knowing what an object is, what does it mean to be object-oriented? Oriented
simply means directed toward. So object-oriented simply means, "functionally
directed toward modeling objects". It is one of many techniques used for modeling
complex systems by describing a collection of interacting objects via their data
and behavior.
If you've read any hype, you've probably come across the terms object-oriented
analysis, object-oriented design, object-oriented analysis and design, and
object-oriented programming. These are all highly related concepts under
the general object-oriented umbrella.
In fact, analysis, design, and programming are all stages of software development.
Calling them object-oriented simply specifies what style of software development is
being pursued.
Object-oriented Analysis (OOA) is the process of looking at a problem, system,
or task that somebody wants to turn into an application and identifying the objects
and interactions between those objects. The analysis stage is all about what needs
to be done. The output of the analysis stage is a set of requirements. If we were to
complete the analysis stage in one step, we would have turned a task, such as, "I
need a website", into a set of requirements, such as:
Visitors to the website need to be able to (italic represents actions, bold
represents objects):
review our history

apply for jobs
browse, compare, and order our products
Object-oriented Design (OOD) is the process of converting such requirements into

an implementation specification. The designer must name the objects, define the
behaviors, and formally specify what objects can activate specific behaviors on
other objects. The design stage is all about how things should be done. The output
of the design stage is an implementation specification. If we were to complete the
design stage in one step, we would have turned the requirements into a set of
classes and interfaces that could be implemented in (ideally) any object-oriented
programming language.
Object-oriented Programming (OOP) is the process of converting this perfectly
defined design into a working program that does exactly what the CEO
originally requested.
Yeah, right! It would be lovely if the world met this ideal and we could follow these
stages one by one, in perfect order like all the old textbooks told us to. As usual, the
real world is much murkier. No matter how hard we try to separate these stages,
we'll always find things that need further analysis while we're designing. When
we're programming, we find features that need clarification in the design. In the
fast-paced modern world, most development happens in an iterative development
model. In iterative development, a small part of the task is modeled, designed, and
programmed, then the program is reviewed and expanded to improve each feature
and include new features in a series of short cycles.
[]
Chapter 1
The rest of this book is about object-oriented programming, but in this chapter we
will cover the basic object-oriented principles in the context of design. This allows us
to understand these rather simple concepts without having to argue with software
syntax or interpreters.
Objects and classes
So, an object is a collection of data with associated behaviors. How do we tell two
types of objects apart? Apples and oranges are both objects, but it is a common
adage that they cannot be compared. Apples and oranges aren't modeled very often
in computer programming, but let's pretend we're doing an inventory application for
a fruit farm! As an example, we can assume that apples go in barrels and oranges go
in baskets.
Now, we have four kinds of objects: apples, oranges, baskets, and barrels. In
object-oriented modeling, the term used for kinds of objects is class. So, in
technical terms, we now have four classes of objects.
What's the difference between an object and a class? Classes describe objects. They
are like blueprints for creating an object. You might have three oranges sitting on the
table in front of you. Each orange is a distinct object, but all three have the attributes
and behaviors associated with one class: the general class of oranges.
The relationship between the four classes of objects in our inventory system can
be described using a Unified Modeling Language (invariably referred to as UML,
because three letter acronyms are cool) class diagram. Here is our first class diagram:
This diagram simply shows that an Orange is somehow associated with a Basket
and that an Apple is also somehow associated with a Barrel. Association is the most
basic way for two classes to be related.
UML is very popular among managers, and occasionally disparaged by
programmers. The syntax of a UML diagram is generally pretty obvious; you don't
have to read a tutorial to (mostly) understand what is going on when you see one.
UML is also fairly easy to draw, and quite intuitive. After all, many people, when
describing classes and their relationships, will naturally draw boxes with lines
between them. Having a standard based on these intuitive diagrams makes it easy
for programmers to communicate with designers, managers, and each other.
[]
However, some programmers think UML is a waste of time. Citing iterative

development, they will argue that formal specifications done up in fancy UML
diagrams are going to be redundant before they're implemented, and that
maintaining those formal diagrams will only waste time and not benefit anyone.
This is true of some organizations, and hogwash in other corporate cultures.
However, every programming team consisting of more than one person will
occasionally have to sit down and hash out the details of part of the system they are
currently working on. UML is extremely useful, in these brainstorming sessions, for
quick and easy communication. Even those organizations that scoff at formal class
diagrams tend to use some informal version of UML in their design meetings, or
team discussions.
Further, the most important person you ever have to communicate with is yourself.
We all think we can remember the design decisions we've made, but there are
always, "Why did I do that?" moments hiding in our future. If we keep the scraps of
paper we did our initial diagramming on when we started a design, we'll eventually
find that they are a useful reference.
This chapter, however, is not meant to be a tutorial in UML. There are many of
those available on the Internet, as well as numerous books available on the topic.
UML covers far more than class and object diagrams; it also has syntax for use cases,
deployment, state changes, and activities. We'll be dealing with some common class
diagram syntax in this discussion of object-oriented design. You'll find you can pick
up the structure by example, and you'll subconsciously choose UML-inspired syntax
in your own team or personal design sessions.
Our initial diagram, while correct, does not remind us that apples go in barrels or
how many barrels a single apple can go in. It only tells us that apples are somehow
associated with barrels. The association between classes is often obvious and needs
no further explanation, but the option to add further clarification is always there.
The beauty of UML is that most things are optional. We only need to specify as
much information in a diagram as makes sense for the current situation. In a quick
whiteboard session, we might just quickly draw lines between boxes. In a formal
document that needs to make sense in six months, we might go into more detail.
In the case of apples and barrels, we can be fairly confident that the association is,
"many apples go in one barrel", but just to make sure nobody confuses it with, "one
apple spoils one barrel", we can enhance the diagram as shown:
[ 10 ]
Chapter 1
This diagram tells us that oranges go in baskets with a little arrow showing what
goes in what. It also tells us the multiplicity (number of that object that can be used
in the association) on both sides of the relationship. One Basket can hold many
(represented by a *) Orange objects. Any one Orange can go in exactly one Basket.
It can be easy to confuse which side of a relationship the multiplicity goes on. The
multiplicity is the number of objects of that class that can be associated with any one
object at the other end of the association. For the apple goes in barrel association,
reading from left to right, many instances of the Apple class (that is many Apple
objects) can go in any one Barrel. Reading from right to left, exactly one Barrel can
be associated with any one Apple.
Specifying attributes and behaviors
We now have a grasp on some basic object-oriented terminology. Objects are

instances of classes that can be associated with each other. An object instance is a
specific object with its own set of data and behaviors; a specific orange on the table
in front of us is said to be an instance of the general class of oranges. That's simple
enough, but what are these data and behaviors that are associated with each object?
Data describes objects
Let's start with data. Data typically represents the individual characteristics of a
certain object. A class of objects can define specific characteristics that are shared by
all instances of that class. Each instance can then have different data values for the
given characteristics. For example, our three oranges on the table (if we haven't eaten
any) could each have a different weight. The orange class could then have a weight
attribute. All instances of the orange class have a weight attribute, but each orange
might have a different value for this weight. Attributes don't have to be unique
though; any two oranges may weigh the same amount. As a more realistic example,
two objects representing different customers might have the same value for a first
name attribute.
Attributes are frequently referred to as properties. Some authors suggest that
the two terms have different meanings, usually that attributes are settable, while
properties are read only. In Python, the concept of "read only" is not really used, so
throughout this book we'll see the two terms used interchangeably. In addition, as
we'll discuss in Chapter 5, the property keyword has a special meaning in Python for
a particular kind of attribute.
[ 11 ]
In our fruit inventory application, the fruit farmer may want to know what orchard
the orange came from, when it was picked, and how much it weighs. They might
also want to keep track of where each basket is stored. Apples might have a color
attribute and barrels might come in different sizes. Some of these properties may also
belong to multiple classes (we may want to know when apples are picked, too), but
for this first example, let's just add a few different attributes to our class diagram:
Depending on how detailed our design needs to be, we can also specify the type
for each attribute. Attribute types are often primitives that are standard to most
programming languages, such as integer, floating-point number, string, byte, or
boolean. However, they can also represent data structures such as lists, trees, or
graphs, or, most notably, other classes. This is one area where the design stage can
overlap with the programming stage. The various primitives or objects available
in one programming language may be somewhat different from what is available
in other languages. Usually we don't need to concern ourselves with this at the
design stage, as implementation-specific details are chosen during the programming
stage. Use generic names and we'll be fine. If our design calls for a list container
type, the Java programmers can choose to use a LinkedList or an ArrayList when
implementing it, while the Python programmers (that's us!) can choose between the
list built-in and a tuple.
In our fruit farming example, so far, our attributes are all basic primitives. But
there are implicit attributes that we can make explicit: the associations. For a given
orange, we might have an attribute containing the basket that holds that orange.
Alternatively, one basket might contain a list of the oranges it holds. The next
diagram adds these attributes as well as including type descriptions for our
current properties:
[ 12 ]
Chapter 1
Behaviors are actions
Now we know what data is, but what are behaviors? Behaviors are actions that
can occur on an object. The behaviors that can be performed on a specific class of
objects are called methods. At the programming level, methods are like functions
in structured programming, but they magically have access to all the data
associated with that object. Like functions, methods can also accept parameters,
and return values.
Parameters to a method are a list of objects that need to be passed into the method
that is being called. These objects are used by the method to perform whatever
behavior or task it is meant to do. Return values are the results of that task. Here's
a concrete example; if our objects are numbers, the number class might have an add
method that accepts a second number as a parameter. The first number object's add
method will return the sum when the second number is passed to it. Given an object
and it's method name, a calling object can call, or invoke the method on the target
object. Invoking a method, at the programming level, is the process of telling the
method to execute itself by passing it the required parameters as arguments.
We've stretched our, "comparing apples and oranges" example into a basic
(if far-fetched) inventory application. Let's stretch it a little further and see if it
breaks. One action that can be associated with oranges is the pick action. If you
think about implementation, pick would place the orange in a basket by updating
the basket attribute on the orange, and by adding the orange to the oranges list on
the Basket. So pick needs to know what basket it is dealing with. We do this by
giving the pick method a basket parameter. Since our fruit farmer also sells juice,
we can add a squeeze method to Orange. When squeezed, squeeze might return the
amount of juice retrieved, while also removing the Orange from the basket it was in.
[ 13 ]
Basket can have a sell action. When a basket is sold, our inventory system
might update some data on as-yet unspecified objects for accounting and profit
calculations. Alternatively, our basket of oranges might go bad before we can sell
them, so we add a discard method. Let's add these methods to our diagram:
Adding models and methods to individual objects allows us to create a system of

interacting objects. Each object in the system is a member of a certain class. These
classes specify what types of data the object can hold and what methods can be
invoked on it. The data in each object can be in a different state from other objects of
the same class, and each object may react to method calls differently because of the
differences in state.
Object-oriented analysis and design is all about figuring out what those objects are
and how they should interact. The next section describes principles that can be used
to make those interactions as simple and intuitive as possible.
Hiding details and creating the public

interface
The key purpose of modeling an object in object-oriented design is to determine

what the public interface of that object will be. The interface is the collection of
attributes and methods that other objects can use to interact with that object. They
do not need, and are often not allowed, to access the internal workings of the object.
A common real-world example is the television. Our interface to the television is the
remote control. Each button on the remote control represents a method that can be
called on the television object. When we, as the calling object, access these methods,
we do not know or care if the television is getting its signal from an antenna, a cable
connection, or a satellite dish. We don't care what electronic signals are being sent
to adjust the volume, or whether that volume is being output to speakers or a set of
headphones. If we open the television to access the internal workings, for example
to split the output signal to both external speakers and a set of headphones, we will
void the warranty.
[ 14 ]
Chapter 1
This process of hiding the implementation, or functional details, of an object is

suitably called information hiding. It is also sometimes referred to as encapsulation,
but encapsulation is actually a more all-encompassing term. Encapsulated data is not
necessarily hidden. Encapsulation is, literally, creating a capsule, so think of creating
a time capsule. If you put a bunch of information into a time capsule, lock and bury
it, it is both encapsulated and the information is hidden. On the other hand, if the
time capsule has not been buried and is unlocked or made of clear plastic, the items
inside it are still encapsulated, but there is no information hiding.
The distinction between encapsulation and information hiding is largely
irrelevant, especially at the design level. Many practical references use the terms
interchangeably. As Python programmers, we don't actually have or need true
information hiding, (we'll discuss the reasons for this in Chapter 2) so the more
encompassing definition for encapsulation is suitable.
The public interface, however, is very important. It needs to be carefully designed as
it is difficult to change it in the future. Changing the interface will break any client
objects that are calling it. We can change the internals all we like, for example, to
make it more efficient, or to access data over the network as well as locally, and the
client objects will still be able to talk to it, unmodified, using the public interface.
On the other hand, if we change the interface, by changing attribute names that are
publicly accessed or by altering the order or types of arguments that a method can
accept, all client objects will also have to be modified.
Remember, program objects represent real objects, but they are not real objects. They
are models. One of the greatest gifts of modeling is the ability to ignore details that
are irrelevant. A model car may look like a real 1956 Thunderbird on the outside, but
it doesn't run and the driveshaft doesn't turn, as these details are overly complex and
irrelevant to the youngster assembling the model. The model is an abstraction of a
real concept.
[ 15 ]
Abstraction is another object-oriented buzzword that ties in with encapsulation

and information hiding. Simply put, abstraction means dealing with the level of
detail that is most appropriate to a given task. It is the process of extracting a public
interface from the inner details. A driver of a car needs to interact with steering,
gas pedal, and brakes. The workings of the motor, drive train, and brake subsystem
don't matter to the driver. A mechanic, on the other hand works at a different level
of abstraction, tuning the engine and bleeding the breaks. Here's an example of two
abstraction levels for a car:
Now we have several new terms that refer to similar concepts. Condensing all this
jargon into a single sentence, abstraction is the process of encapsulating information
with separate public and private interfaces. The private interfaces can be subject to
information hiding.
The important thing to bring from all these definitions is to make our models
understandable to the other objects that have to interact with them. This means
paying careful attention to small details. Ensure methods and properties have
sensible names. When analyzing a system, objects typically represent nouns in the
original problem, while methods are normally verbs. Attributes can often be picked
up as adjectives, although if the attribute refers to another object that is part of the
current object, it will still likely be a noun. Name classes, attributes, and methods
accordingly. Don't try to model objects or actions that might be useful in the future.
Model exactly those tasks that the system needs to perform and the design will
naturally gravitate towards one that has an appropriate level of abstraction. This
is not to say we should not think about possible future design modifications. Our
designs should be open ended so that future requirements can be satisfied. However,
when abstracting interfaces, try to model exactly what needs to be modeled and
nothing more.
[ 16 ]
Chapter 1
When designing the interface, try placing yourself in the object's shoes and imagine
that the object has a strong preference for privacy. Don't let other objects have access
to data about you unless you feel it is in your best interest for them to have it. Don't
give them an interface to force you to perform a specific task unless you are certain
you want them to be able to do that to you.
This is also a good practice for ensuring privacy on your social networking accounts!
Composition and inheritance
So far, we've learned to design systems as a group of interacting objects, where each
interaction is viewing the objects involved at an appropriate level of abstraction. But
we don't know yet how to create those levels of abstraction. There are a variety of
ways to do this; we'll discuss some advanced design patterns in Chapter 8 and Chapter
9. But even most design patterns rely on two basic principles known as composition
and inheritance.
Composition is the act of collecting together several objects to compose a new one.
Composition is usually a good choice when one object is part of another object.
We've already seen a first hint of composition in the mechanic example. A car is
composed of an engine, transmission, starter, headlights, and windshield, among
numerous other parts. The engine, in turn, is composed of pistons, a crank shaft, and
valves. In this example, composition is a good way to provide levels of abstraction.
The car object can provide the interface required by a driver, while also providing
access to its component parts, which offers a deeper level of abstraction suitable for
a mechanic. Those component parts can, of course, be further broken down if the
mechanic needs more information to diagnose a problem or tune the engine.
This is a common first example of composition, bit it's not a very good one when
it comes to designing computer systems. Physical objects are easy to break into
component objects. People have been doing it at least since the ancient Greeks
originally postulated that atoms were the smallest unit of matter (they, of course,
didn't have access to particle accelerators). Computer systems are generally less
complicated than physical objects, yet identifying the component objects in such
systems does not happen as naturally. The objects in an object-oriented system
occasionally represent physical objects like people, books, or telephones. More often,
however, they represent abstract ideas. People have names, books have titles, and
telephones are used to make calls. Calls, titles, accounts, names, appointments, and
payments are not usually considered objects in the physical world, but they are all
frequently modeled components in computer systems.
[ 17 ]
Let's try modeling a more computer-oriented example to see composition in action.

We'll be looking at the design of a computerized chess game. This was a very
popular pastime among academics in the '80s and '90s. People were predicting
that computers would one day be able to defeat a human chess master. When
this happened in 1997 (IBM's Deep Blue defeated world chess champion, Gary
Kasparov), interest in the problem waned, although there are still contests between
computer and human chess players, and the program has not yet been written that
can defeat a human chess master 100% of the time.
As a basic, high-level analysis: a game of chess is played between two players, using
a chess set featuring a board containing sixty-four positions in an 8x8 grid. The board
can have two sets of sixteen pieces that can be moved, in alternating turns by the
two players in different ways. Each piece can take other pieces. The board will be
required to draw itself on the computer screen after each turn.
I've identified some of the possible objects in the description using italics, and a few
key methods using bold. This is a common first step in turning an object-oriented
analysis into a design. At this point, to emphasize composition, we'll focus on the
board, without worrying too much about the players or the different types of pieces.
Let's start at the highest level of abstraction possible. We have two players
interacting with a chess set by taking turns making moves.
What is that? It doesn't quite look like our earlier class diagrams. That's because it
isn't a class diagram! This is an object diagram, also called an instance diagram. It
describes the system at a specific state in time, and is describing specific instances of
objects, not the interaction between classes. Remember, both players are members of
the same class, so the class diagram looks a little different:
[ 18 ]
Chapter 1
The diagram shows that exactly two players can interact with one chess set. It also
indicates that any one player can be playing with only one chess set at a time.
But we're discussing composition, not UML, so let's think about what the Chess
Set is composed of. We don't care what the player is composed of at this time. We
can assume that the player has a heart and brain, among other organs, but these are
irrelevant to our model. Indeed, there is nothing stopping said player from being
Deep Blue itself, which has neither a heart nor brain.
The chess set, then, is composed of a board and thirty-two pieces. The board is
further comprised of sixty-four positions. You could argue that pieces are not part of
the chess set because you could replace the pieces in a chess set with a different set
of pieces. While this is unlikely or impossible in a computerized version of chess, it
introduces us to aggregation. Aggregation is almost exactly like composition. The
difference is that aggregate objects can exist independently. It would be impossible
for a position to be associated with a different chess board, so we say the board is
composed of positions. But the pieces, which might exist independently of the chess
set, are said to be in an aggregate relationship with that set.
Another way to differentiate between aggregation and composition is to think about
the lifespan of the object. If the composite (outside) object controls when the related
(inside) objects are created and destroyed, composition is most suitable. If the related
object is created independently of the composite object, or can outlast that object,
an aggregate relationship makes more sense. Also keep in mind that composition
is aggregation; aggregation is simply a more general form of composition. Any
composite relationship is also an aggregate relationship, but not vice versa.
Let's describe our current chess set composition and add some attributes to the
objects to hold the composite relationships:
[ 19 ]
The composition relationship is represented in UML as a solid diamond. The hollow

diamond represents the aggregate relationship. You'll notice that the board and
pieces are stored as part of the chess set in exactly the same way a reference to them
is stored as an attribute on the chess set. This shows that once again, in practice,
the distinction between aggregation and composition is often irrelevant once you
get past the design stage. When implemented, they behave in much the same way.
However, it can help to differentiate between the two when your team is discussing
how the different objects interact. Often you can treat them as the same thing, but
when you need to distinguish between them, it's great to know the difference (this is
abstraction at work).
Inheritance
We have discussed three types of relationships between objects: association,

composition, and aggregation. But we have not fully specified our chess set, and
these tools don't seem to give us all the power we need. We discussed the possibility
that a player might be a human or it might be a piece of software featuring artificial
intelligence. It doesn't seem right to say that a Player is associated with a human, or
that the artificial intelligence implementation is part of the Player object. What we
really need is the ability to say that "Deep Blue is a player" or that "Gary Kasparov
is a player".
The is a relationship is formed by inheritance. Inheritance is the most famous,
well-known, and over-used relationship in object-oriented programming.
Inheritance is sort of like a family tree. My grandfather's last name was Phillips and
my father inherited that name. I inherited it from him (along with blue eyes and a
penchant for writing). In object-oriented programming, instead of inheriting features
and behaviors from a person, one class can inherit attributes and methods from
another class.
For example, there are thirty-two chess pieces in our chess set, but there are only
six different types of pieces (pawns, rooks, bishops, knights, king, and queen), each
of which behaves differently when it is moved. All of these classes of piece have
properties, like color and the chess set they are part of, but they also have unique
shapes when drawn on the chess board, and make different moves. See how the six
types of pieces can inherit from a Piece class:
[ 20 ]
Chapter 1
The hollow arrows, of course, indicate that the individual classes of pieces inherit
from the Piece class. All the subtypes automatically have a chess_set and color
attribute inherited from the base class. Each piece provides a different shape
property (to be drawn on the screen when rendering the board), and a different
move method to move the piece to a new position on the board at each turn.
We actually know that all subclasses of the Piece class need to have a move method,
otherwise when the board tries to move the piece it will get confused. It is possible
we want to create a new version of the game of chess that has one additional piece
(the wizard). Our current design would allow us to design this piece without
giving it a move method. The board would then choke when it asked the piece
to move itself.
We can implement this by creating a dummy move method on the Piece class. The
subclasses can then override this method with a more specific implementation. The
default implementation might, for example, pop up an error message that says, That
piece cannot be moved. Overriding methods in subtypes allows very powerful
object-oriented systems to be developed. For example, if we wanted to implement a
player class with artificial intelligence, we might provide a calculate_move method
that takes a Board object and decides which piece to move where. A very basic class
might randomly choose a piece and direction and move it. We could then override
this method in a subclass with the Deep Blue implementation. The first class would
be suitable for play against a raw beginner, the latter would challenge a grand
master. The important thing is that other methods on the class, such as the ones that
inform the board as to which move was chosen would not need to be changed; this
implementation can be shared between the two classes.
[ 21 ]
In the case of chess pieces, it doesn't really make sense to provide a default
implementation of the move method. All we need to do is specify that the move
method is required in any subclasses. This can be done by making Piece an abstract
class with the move method declared abstract. Abstract methods basically say "We
need this method in a subclass, but we are declining to specify an implementation in
this class."
Indeed, it is possible to make a class that does not implement any methods at all. Such
a class would simply tell us what the class should do, but provides absolutely no
advice on how to do it. In object-oriented parlance, such classes are called interfaces.
Inheritance provides abstraction
Now it's time for another long buzzword. Polymorphism is the ability to treat a class
differently depending on which subclass is implemented. We've already seen it in
action with the pieces system we've described. If we took the design a bit further,
we'd probably see that the Board object can accept a move from the player and call
the move function on the piece. The board need not ever know what type of piece it
is dealing with. All it has to do is call the move method and the proper subclass will
take care of moving it as a Knight or a Pawn.
Polymorphism is pretty cool, but it is a word that is rarely used in Python
programming. Python goes an extra step past allowing a subclass of an object to be
treated like a parent class. A board implemented in Python could take any object
that has a move method, whether it is a Bishop piece, a car, or a duck. When move is
called, the Bishop will move diagonally on the board, the car will drive someplace,
and the duck will swim or fly, depending on its mood.
This sort of polymorphism in Python is typically referred to as duck typing: "If it
walks like a duck or swims like a duck, it's a duck". We don't care if it really is a duck
(inheritance), only that it swims or walks. Geese and swans might easily be able to
provide the duck-like behavior we are looking for. This allows future designers to
create new types of birds without actually specifying an inheritance hierarchy for
aquatic birds. It also allows them to create completely different drop-in behaviors
that the original designers never planned for. For example, future designers might
be able to make a walking, swimming penguin that works with the same interface
without ever suggesting that penguins are ducks.
[ 22 ]
Chapter 1
When we think of inheritance in our own family tree, we can see that we inherit
features from more than just one parent. When strangers tell a proud mother that her
son has, "his fathers eyes", she will typically respond along the lines of, "yes, but he
got my nose".
Object-oriented design can also feature such multiple inheritance, which allows a
subclass to inherit functionality from multiple parent classes. In practice, multiple
inheritance can be tricky business, and some programming languages, (most notably,
Java) strictly prohibit it. But multiple inheritance can have its uses. Most often, it
can be used to create objects that have two distinct sets of behaviors. For example,
an object designed to connect to a scanner and send a fax of the scanned document
might be created by inheriting from two separate scanner and faxer objects.
As long as two classes have distinct interfaces, it is not normally harmful for a
subclass to inherit from both of them. But it gets messy if we inherit from two classes
that provide overlapping interfaces. For example, if we have a motorcycle class that
has a move method, and a boat class also featuring a move method, and we want
to merge them into the ultimate amphibious vehicle, how does the resulting class
know what to do when we call move? At the design level, this needs to be explained,
and at the implementation level, each programming language has different ways of
deciding which parent class's method is called, or in what order.
Often, the best way to deal with it is to avoid it. If you have a design showing up like
this, you're probably doing it wrong. Take a step back, analyze the system again, and
see if you can remove the multiple inheritance relationship in favor of some other
association or composite design.
Inheritance is a very powerful tool for extending behavior. It is also one of the most
exciting advancements of object-oriented design over earlier paradigms. Therefore,
it is often the first tool that object-oriented programmers reach for. However, it
is important to recognize that owning a hammer does not turn screws into nails.
Inheritance is the perfect solution for obvious is a relationships but it can be abused.
Programmers often use inheritance to share code between two kinds of objects
that are only distantly related, with no is a relationship in sight. While this is not
necessarily a bad design, it is a terrific opportunity to ask just why they decided to
design it that way, and if a different relationship or design pattern would have been
more suitable.
[ 23 ]
Case study
Let's tie all our new object-oriented knowledge together by going through a few
iterations of object-oriented design on a somewhat real-world example. The system
we'll be modeling is a library catalog. Libraries have been tracking their inventory for
centuries, originally using card catalogs, and, more recently, electronic inventories.
Modern libraries have web-based catalogs that we can query from our home.
Let's start with an analysis. The local librarian has asked us to write a new card
catalog program because their ancient DOS based program is ugly and out of date.
That doesn't give us much detail, but before we start asking for more information,
let's consider what we already know about library catalogs:
Catalogs contain lists of books. People search them to find books on certain subjects,
with specific titles, or by a particular author. Books can be uniquely identified by an
International Standard Book Number (ISBN). Each book has a Dewey Decimal
System (DDS) number assigned to help find it on a particular shelf.
This simple analysis tells us some of the obvious objects in the system. We quickly
identify Book as the most important object, with several attributes already mentioned,
such as author, title, subject, ISBN, and DDS number, and catalog as a sort of
manager for books.
We also notice a few other objects that may or may not need to be modeled in
the system. For cataloging purposes, all we need to search a book by author is an
author_name attribute on the book. But authors are also objects, and we might
want to store some other data about the author. As we ponder this, we might
remember that some books have multiple authors. Suddenly, the idea of having a
single author_name attribute on objects seems a bit silly. A list of authors associated
with each book is clearly a better idea. The relationship between author and book
is clearly association, since you would never say "book is an author" (it's not
inheritance), and saying "book has an author", though grammatically correct, does
not imply that authors are part of books (it's not aggregation). Indeed, any one
author may be associated with multiple books.
We should also pay attention to the noun (nouns are always good candidates for
objects) shelf. Is a shelf an object that needs to be modeled in a cataloging system?
How do we identify an individual shelf. What happens if a book is stored at the end
of one shelf, and later moved to the beginning of the next shelf because another book
was inserted in the previous shelf?
DDS was designed to help locate physical books in a library. As such, storing a
DDS attribute with the book should be enough to locate it, regardless of which shelf
it is stored on. So we can, at least for the moment, remove shelf from our list of
contending objects.
[ 24 ]
Chapter 1
Another questionable object in the system is the user. Do we need to know anything
about a specific user? Their name, address, or list of overdue books? So far the
librarian has told us only that they want a catalog; they said nothing about tracking
subscriptions or overdue notices. In the back of our minds, we also note that authors
and users are both specific kinds of people; there might be a useful inheritance
relationship here in the future.
For cataloging purposes, we decide we don't need to identify the user, for now. We
can assume that a user will be searching the catalog, but we don't have to actively
model them in the system, beyond providing an interface that allows them to search.
We have identified a few attributes on the book, but what properties does a catalog
have? Does any one library have more than one catalog? Do we need to uniquely
identify them? Obviously, the catalog has to have a list of the books it contains,
somehow, but this list is probably not part of the public interface.
What about behaviors? The catalog clearly needs a search method, possibly separate
ones for authors, titles, and subjects. Are there any behaviors on books? Would it
need a preview method? Or could preview be identified by a first pages attribute,
instead of a method?
The questions in the preceding discussion are all part of the object-oriented analysis
phase. But intermixed with the questions, we have already identified a few key
objects that are part of the design. Indeed, what you have just seen is several microiterations between analysis and design. Likely, these iterations would all occur in
an initial meeting with the librarian. Before this meeting, however, we can already
sketch out a most basic design for the objects we have concretely identified:
[ 25 ]
Armed with this basic diagram and a pencil to interactively improve it, we meet
up with the librarian. They tell us that this is a good start, but libraries don't serve
only books, they also have DVDs, magazines, and CDs, none of which have an ISBN
or DDS number. All of these types of items can be uniquely identified by a UPC
number, though. We remind the librarian that they have to find the items on the
shelf, and these items probably aren't organized by UPC. The librarian explains that
each type is organized in a different way. The CDs are mostly audio books and they
only have a couple dozen in stock, so they are organized by the author's last name.
DVDs are divided into genre and further organized by title. Magazines are organized
by title and then refined by volume and issue number. Books are, as we had guessed,
organized by DDS number.
With no previous object-oriented design experience, we might consider adding
separate lists of DVDs, CDs, magazines, and books to our catalog, and search each
one in turn. The trouble is, except for certain extended attributes, and identifying the
physical location of the item, these items all behave in much the same. This is a job
for inheritance! We quickly update our UML diagram:
The librarian understands the gist of our sketched diagram, but is a bit confused
by the locate functionality. We explain using a specific use case where the user is
searching for the word "bunnies". The user first sends a search request to the catalog.
The catalog queries its internal list of items and finds a book and a DVD with
"bunnies" in the title. At this point, the catalog doesn't care if it is holding a DVD,
book, CD or magazine; all items are the same, as far as the catalog is concerned.
But the user wants to know how to find the physical items, so the catalog would be
remiss if it simply returned a list of titles. So it calls the locate method on the two
items it has uncovered. The book's locate method returns a DDS number that can
be used to find the shelf holding the book. The DVD is located by returning the
genre and title of the DVD. The user can then visit the DVD section, find the
section containing that genre, and find the specific DVD as sorted by title.
[ 26 ]
Chapter 1
As we explain, we sketch a UML sequence diagram explaining how the various

objects are communicating:
Where class diagrams describe the relationships between classes, sequence diagrams
describe specific sequences of messages passed between objects. The dashed line
hanging from each object is a lifeline describing the lifetime of the object. The wider
boxes on each lifeline represent active processing in that object (where there's no box,
the object is basically sitting idle, waiting for something to happen). The horizontal
arrows between the lifelines indicate specific messages. The solid arrows represent
methods being called, while the dashed arrows with solid heads represent the
method return values. The half arrowheads indicate asynchronous messages sent to
or from an object. An asynchronous message typically means the first object calls a
method on the second object which returns immediately. After some processing, the
second object calls a method on the first object to give it a value. This is in contrast
to normal method calls, which do the processing in the method, and return a
value immediately.
[ 27 ]
Sequence diagrams, like all UML diagrams, are best used when they are needed.
There is no point in drawing a UML diagram for the sake of drawing a diagram.
But when you need to communicate a series of interactions between two objects,
the sequence diagram is a very useful tool.
Unfortunately, our class diagram so far is still a messy design. We notice that actors
on DVDs and artists on CDs are all types of people, but are being treated differently
from the book authors. The librarian also reminds us that most of their CDs are
audio books, which have authors instead of artists.
How can we deal with different kinds of people that contribute to a title? An obvious
implementation is to create a Person class with the person's name and other relevant
details and then create subclasses of this for the artists, authors, and actors. But is
inheritance really necessary here? For searching and cataloging purposes, we don't
really care that acting and writing are two very different activities. If we were doing
an economic simulation, it would make sense to give separate actor and author
classes different calculate_income and perform_job methods, but for cataloging
purposes, it is probably enough to know how the person contributed to the item.
We recognize that all items have one or more Contributor objects, so we move the
author relationship from the book to its parent class:
The multiplicity of the Contributor/LibraryItem relationship is many-to-many, as

indicated by the * at each end of the relationship. Any one library item might have
more than one contributor (for example, several actors and a director on DVD). And
many authors write many books, so they would be attached to multiple library items.
This little change, while it looks a bit cleaner and simpler has lost some vital
information. We can still tell who contributed to a specific library item, but we don't
know how they contributed. Were they the director or an actor? Did they write the
audio book, or were they the voice that narrated the book?
[ 28 ]
Chapter 1
It would be nice if we could just add a contributor_type attribute on the

Contributor class, but this will fall apart when dealing with multi-talented people
who have both authored books and directed movies.
One option is to add attributes to each of our LibraryItem subclasses that hold the
information we need, such as Author on Book, or Artist on CD, and then make the
relationship to those properties all point to the Contributor class. The problem with
this is that we lose a lot of polymorphic elegance. If we want to list the contributors
to an item, we have to look for specific attributes on that item, such as Authors
or Actors. We can alleviate this by adding a GetContributors method on the
LibraryItem class that subclasses can override. Then the catalog never has to know
what attributes the objects are querying; we've abstracted the public interface:
Just looking at this class diagram, it feels like we are doing something wrong. It is
bulky and fragile. It may do everything we need, but it feels like it will be hard to
maintain or extend. There are too many relationships, and too many classes would
be affected by modifications to any one class. It looks like spaghetti and meatballs.
[ 29 ]
Now that we've explored inheritance as an option, and found it wanting, we might
look back at our previous composition-based diagram, where Contributor was
attached directly to LibraryItem. With some thought, we can see that we actually
only need to add one more relationship to a brand-new class to identify the type of
contributor. This is an important step in object-oriented design. We are now adding a
class to the design that is intended to support the other objects, rather than modeling
any part of the initial requirements. We are refactoring the design to facilitate the
objects in the system, rather than objects in real life. Refactoring is an essential
process in the maintenance of a program or design. The goal of refactoring is to
improve the design by moving code around, removing duplicate code or complex
relationships in favor of simpler, more elegant designs.
This new class is composed of a Contributor and an extra attribute identifying
the type of contribution the person has made to the given LibraryItem. There can
be many such contributions to a particular LibraryItem, and one contributor can
contribute in the same way to different items. The diagram communicates this
design very well:
At first, this composition relationship looks less natural than the inheritancebased relationships. But it has the advantage of allowing us to add new types of
contributions without adding a new class to the design. Inheritance is most useful
when the subclasses have some kind of specialization. Specialization is creating
or changing attributes or behaviors on the subclass to make it somehow different
from the parent class. It seems silly to create a bunch of empty classes solely for
identifying different types of objects (this attitude is less prevalent among Java and
other "everything is an object" programmers, but it is common among more practical
Python designers). If we look at the inheritance version of the diagram, we can see a
bunch of subclasses that don't actually do anything:
[ 30 ]
Chapter 1
Sometimes it is important to recognize when not to use object-oriented principles.

This example of when not to use inheritance is a good reminder that objects are just
tools, and not rules.
Exercises
This is a practical book, not a textbook. As such, I'm not about to assign you a bunch
of fake object-oriented analysis problems to create designs for. Instead, I want to give
you some things to think about that you can apply to your own projects. If you have
previous object-oriented experience, you won't need to put much effort into these.
But they are useful mental exercises if you've been using Python for a while but
never really cared about all that class stuff.
First, think about a recent programming project you've completed. Identify the most
prominent object in the design. Try to think of as many attributes for this object as
possible. Did it have: Color? Weight? Size? Profit? Cost? Name? ID number? Price?
Style? Think about the attribute types. Were they primitives or classes? Were some
of those attributes actually behaviors in disguise? Sometimes what looks like data
is actually calculated from other data on the object, and you can use a method to
do those calculations. What other methods or behaviors did the object have?
What objects called those methods. What kinds of relationships did they have
to this object?
[ 31 ]
Now think about an upcoming project. It doesn't matter what the project is; it might
be a fun free-time project or a multi-million dollar contract. It doesn't have to be a
complete application; it could just be one subsystem. Perform a basic object-oriented
analysis. Identify the requirements and the interacting objects. Sketch out a class
diagram featuring the very highest level of abstraction on that system. Identify
the major interacting objects. Identify minor supporting objects. Go into detail
for the attributes and methods of some of the most interesting ones. Take different
objects to different levels of abstraction. Look for places you can use inheritance
or composition. Look for places you should avoid inheritance.
The goal is not to design a system (although you're certainly welcome to do so if
inclination meets both ambition and available time). The goal is to think about
object-oriented designs. Focusing on projects that you have worked on or are
expecting to work on in the future simply makes it real.
Now visit your favorite search engine and look up some tutorials on UML. There are
dozens, so find the one that suits your preferred method of study. Sketch some class
diagrams or a sequence diagram for the objects you identified earlier. Don't get too
hung up on memorizing the syntax (after all, if it is important, you can always look it
up again), just get a feel for the language. Something will stay lodged in your brain,
and it can make communicating a bit easier if you can quickly sketch a diagram for
your next OOP discussion.
Summary
In this chapter, we took a whirlwind tour through the terminology of the objectoriented paradigm, focusing on object-oriented design. We learned how to separate
different objects into a taxonomy of different classes and to describe the attributes
and behaviors of those objects via the class interface. In particular, we covered:
Classes and objects
Abstraction, encapsulation, and information hiding
Designing a public interface
Object relations: association, composition, and inheritance
Basic UML syntax for fun and communication
In the next chapter, we'll explore how to implement classes and methods in Python.
[ 32 ]
Objects in Python
So, we now have a design in hand and are ready to turn that design into a working
program! Of course, it doesn't usually happen that way, but this book is about
programming in Python. We'll be seeing examples and hints for good software
design throughout the book, but our focus is object-oriented programming. So let's
have a look at the Python syntax that allows us to create object-oriented software.
After completing this chapter we will understand:
How to create classes and instantiate objects in Python
How to add attributes and behaviors to Python objects
How to organize classes into packages and modules
How to suggest people don't clobber our data
Creating Python classes
We don't have to write much Python code to realize that Python is a very "clean"
language. When we want to do something, we just do it, without having to go
through a lot of setup. The ubiquitous, "hello world" in Python, as you've likely
seen, is only one line.
Similarly, the simplest class in Python 3 looks like this:
class MyFirstClass:
pass
There's our first object-oriented program! The class definition starts with the class
keyword. This is followed by a name (of our choice) identifying the class, and is
terminated with a colon.
Objects in Python
The class name must follow standard Python variable naming rules
(must start with a letter or underscore, can only be comprised of letters,
underscores, or numbers). In addition, the Python style guide (search
the web for "PEP 8"), recommends that classes should be named using
CamelCase notation (start with a capital letter, any subsequent words
should also start with a capital).
The class definition line is followed by the class contents, indented. As with other
Python constructs, indentation is used to delimit the classes, rather than braces or
brackets as many other languages use. Use four spaces for indentation unless you
have a compelling reason not to (such as fitting in with somebody else's code that
uses tabs for indents). Any decent programming editor can be configured to insert
four spaces whenever the Tab key is pressed.
Since our first class doesn't actually do anything, we simply use the pass keyword
on the second line to indicate that no further action needs to be taken.
We might think there isn't much we can do with this most basic class, but it does
allow us to instantiate objects of that class. We can load the class into the Python 3
interpreter so we can play with it interactively. To do this, save the class definition
mentioned earlier into a file named first_class.py and then run the command
python -i first_class.py. The -i argument tells Python to "run the code and then
drop to the interactive interpreter". The following interpreter session demonstrates
basic interaction with this class:
>>> a = MyFirstClass()
>>> b = MyFirstClass()
>>> print(a)
<__main__.MyFirstClass object at 0xb7b7faec>
>>> print(b)
<__main__.MyFirstClass object at 0xb7b7fbac>
>>>
This code instantiates two objects from the new class, named a and b. Creating an
instance of a class is a simple matter of typing the class name followed by a pair
of parentheses. It looks much like a normal function call, but Python knows we're
"calling" a class and not a function, so it understands that its job is to create a new
object. When printed, the two objects tell us what class they are and what memory
address they live at. Memory addresses aren't used much in Python code, but here,
it demonstrates that there are two distinctly different objects involved.
[ 34 ]
Chapter 2
Adding attributes
Now, we have a basic class, but it's fairly useless. It doesn't contain any data, and it
doesn't do anything. What do we have to do to assign an attribute to a given object?
It turns out that we don't have to do anything special in the class definition. We can
set arbitrary attributes on an instantiated object using the dot notation:
class Point:
pass
p1 = Point()
p2 = Point()
p1.x = 5
p1.y = 4
p2.x = 3
p2.y = 6
print(p1.x, p1.y)
print(p2.x, p2.y)
If we run this code, the two print statements at the end tell us the new attribute
values on the two objects:
5 4
3 6
This code creates an empty Point class with no data or behaviors. Then it creates
two instances of that class and assigns each of those instances x and y coordinates to
identify a point in two dimensions. All we need to do to assign a value to an attribute
on an object is use the syntax <object>.<attribute> = <value>. This is sometimes
referred to as dot notation. The value can be anything: a Python primitive, a
built-in data type, another object. It can even be a function or another class!
Making it do something
Now, having objects with attributes is great, but object-oriented programming is

really about the interaction between objects. We're interested in invoking actions that
cause things to happen to those attributes. It is time to add behaviors to our classes.
[ 35 ]
Objects in Python
Let's model a couple of actions on our Point class. We can start with a method
called reset that moves the point to the origin (the origin is the point where x
and y are both zero). This is a good introductory action because it doesn't require
any parameters:
class Point:
def reset(self):
self.x = 0
self.y = 0
p = Point()
p.reset()
print(p.x, p.y)
That print statement shows us the two zeros on the attributes:

0 0
A method in Python is identical to defining a function. It starts with the keyword

def followed by a space and the name of the method. This is followed by a set
of parentheses containing the parameter list (we'll discuss the self parameter in
just a moment), and terminated with a colon. The next line is indented to contain
the statements inside the method. These statements can be arbitrary Python code
operating on the object itself and any parameters passed in as the method sees fit.
The one difference between methods and normal functions is that all methods
have one required argument. This argument is conventionally named self; I've
never seen a programmer use any other name for this variable (convention is a very
powerful thing). There's nothing stopping you, however, from calling it this or
even Martha.
The self argument to a method is simply a reference to the object that the method
is being invoked on. We can access attributes and methods of that object as if it were
any other object. This is exactly what we do inside the reset method when we set
the x and y attributes of the self object.
Notice that when we call the p.reset() method, we do not have to pass the self
argument into it. Python automatically takes care of this for us. It knows we're
calling a method on the p object, so it automatically passes that object to the method.
However, the method really is just a function that happens to be on a class. Instead of
calling the method on the object, we could invoke the function on the class, explicitly
passing our object as the self argument:
p = Point()
Point.reset(p)
print(p.x, p.y)
[ 36 ]
Chapter 2
The output is the same as the previous example because, internally, the exact same
process has occurred.
What happens if we forget to include the self argument in our class definition?
Python will bail with an error message:
>>> class Point:
...
def reset():
...
pass
...
>>> p = Point()
>>> p.reset()
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
TypeError: reset() takes no arguments (1 given)
The error message is not as clear as it could be ("You silly fool, you forgot the self
argument" would be more informative). Just remember that when you see an error
message that indicates missing arguments, the first thing to check is whether you
forgot self in the method definition.
So how do we pass multiple arguments to a method? Let's add a new method that
allows us to move a point to an arbitrary position, not just the origin. We can also
include one that accepts another Point object as input and returns the distance
between them:
import math
class Point:
def move(self, x, y):
self.x = x
self.y = y
def reset(self):
self.move(0, 0)
def calculate_distance(self, other_point):
return math.sqrt(
(self.x - other_point.x)**2 +
(self.y - other_point.y)**2)
# how to use it:
point1 = Point()
[ 37 ]
Objects in Python
point2 = Point()
point1.reset()
point2.move(5,0)
print(point2.calculate_distance(point1))
assert (point2.calculate_distance(point1) ==
point1.calculate_distance(point2))
point1.move(3,4)
The print statements at the end give us the following output:

5.0
4.472135955
0.0
A lot has happened here. The class now has three methods. The move method accepts
two arguments, x and y, and sets the values on the self object, much like the old
reset method from the previous example. The old reset method now calls move,
since a reset is just a move to a specific known location.
The calculate_distance method uses the not-too-complex Pythagorean Theorem
to calculate the distance between two points. I hope you understand the math (**
means squared, and math.sqrt calculates a square root), but it's not a requirement
for our current focus: learning how to write methods.
The example code at the end shows how to call a method with arguments; simply
include the arguments inside the parentheses, and use the same dot notation to
access the method. I just picked some random positions to test the methods. The test
code calls each method and prints the results on the console. The assert function
is a simple test tool; the program will bail if the statement after assert is False (or
zero, empty, or None). In this case, we use it to ensure that the distance is the same
regardless of which point called the other point's calculate_distance method.
Initializing the object
If we don't explicitly set the x and y positions on our Point object, either using move
or by accessing them directly, we have a broken point with no real position. What
will happen when we try to access it?
[ 38 ]
Chapter 2
Well, let's just try it and see. "Try it and see" is an extremely useful tool for Python
study. Open up your interactive interpreter and type away. The following interactive
session shows what happens if we try to access a missing attribute. If you saved the
previous example as a file or are using the examples distributed with the book, you
can load it into the Python interpreter with the command python -i filename.py.
>>> point = Point()
>>> point.x = 5
>>> print(point.x)
5
>>> print(point.y)
AttributeError: 'Point' object has no attribute 'y'
Well, at least it threw a useful exception. We'll cover exceptions in detail in Chapter
4. You've probably seen them before (especially the ubiquitous SyntaxError, which
means you typed something incorrectly!). At this point, simply be aware that it
means something went wrong.
The output is useful for debugging. In the interactive interpreter it tells us the error
occurred at line 1, which is only partially true (in an interactive session, only one line
is executed at a time). If we were running a script in a file, it would tell us the exact line
number, making it easy to find the offending code. In addition, it tells us the error is an
AttributeError, and gives a helpful message telling us what that error means.
We can catch and recover from this error, but in this case, it feels like we should have
specified some sort of default value. Perhaps every new object should be reset()
by default or maybe it would be nice if we could force the user to tell us what those
positions should be when they create the object.
Most object-oriented programming languages have the concept of a constructor,
a special method that creates and initializes the object when it is created. Python
is a little different; it has a constructor and an initializer. Normally, the constructor
function is rarely ever used unless you're doing something exotic. So we'll start our
discussion with the initialization method.
The Python initialization method is the same as any other method, except it has a
special name: __init__. The leading and trailing double underscores mean, "this is
a special method that the Python interpreter will treat as a special case". Never name
a function of your own with leading and trailing double underscores. It may mean
nothing to Python, but there's always the possibility that the designers of Python will
add a function that has a special purpose with that name in the future, and when
they do, your code will break.
[ 39 ]
Objects in Python
Let's start with an initialization function on our Point class that requires the user to
supply x and y coordinates when the Point object is instantiated:
class Point:
def __init__(self, x, y):
self.move(x, y)
self.x = x
self.y = y
def reset(self):
self.move(0, 0)
# Constructing a Point
point = Point(3, 5)
print(point.x, point.y)
Now, our point can never go without a y coordinate! If we try to construct a

point without including the proper initialization parameters, it will fail with a not
enough arguments error similar to the one we received earlier when we forgot the
self argument.
What if we don't want to make those two arguments required? Well then we can use
the same syntax Python functions use to provide default arguments. The keyword
argument syntax appends an equals sign after each variable name. If the calling
object does not provide that argument, then the default argument is used instead; the
variables will still be available to the function, but they will have the values specified
in the argument list. Here's an example:
class Point:
def __init__(self, x=0, y=0):
self.move(x, y)
Most of the time, we put our initialization statements in an __init__ function. But as
mentioned earlier, Python has a constructor in addition to its initialization function.
You may never need to use the other Python constructor, but it helps to know it
exists, so we'll cover it briefly.
The constructor function is called __new__ as opposed to __init__, and accepts
exactly one argument, the class that is being constructed (it is called before the object
is constructed, so there is no self argument). It also has to return the newly created
object. This has interesting possibilities when it comes to the complicated art of
meta-programming, but is not very useful in day-to-day programming. In practice,
you will rarely, if ever, need to use __new__, and __init__ will be sufficient.
[ 40 ]
Chapter 2
Explaining yourself
Python is an extremely easy-to-read programming language; some might say it

is self-documenting. However, when doing object-oriented programming, it is
important to write API documentation that clearly summarizes what each object and
method does. Keeping documentation up-to-date is difficult; the best way to do it is
to write it right into our code.
Python supports this through the use of docstrings. Each class, function, or
method header can have a standard Python string as the first line following the
definition (the line that ends in a colon). This line should be indented the same
as the following code.
Docstrings are simply Python strings enclosed with apostrophe (') or quote (")
characters. Often, docstrings are quite long and span multiple lines (the style guide
suggests that line-length should not exceed 80 characters), which can be formatted
as multi-line strings, enclosed in matching triple apostrophe (''') or triple quote
(""") characters.
A docstring should clearly and concisely summarize the purpose of the class or
method it is describing. It should explain any parameters whose usage is not
immediately obvious, and is also a good place to include short examples of how to
use the API. Any caveats or problems an unsuspecting user of the API should be
aware of should also be noted.
To illustrate the use of docstrings, we will end this section with our completely
documented Point class:
import math
class Point:
'Represents a point in two-dimensional geometric coordinates'
def __init__(self, x=0, y=0):
'''Initialize the position of a new point. The x and y
coordinates can be specified. If they are not, the point
defaults to the origin.'''
self.move(x, y)
"Move the point to a new location in two-dimensional space."
self.x = x
self.y = y
[ 41 ]
Objects in Python
def reset(self):
'Reset the point back to the geometric origin: 0, 0'
self.move(0, 0)
def calculate_distance(self, other_point):
"""Calculate the distance from this point to a second point
passed as a parameter.
This function uses the Pythagorean Theorem to calculate
the distance between the two points. The distance is returned
as a float."""
return math.sqrt(
(self.x - other_point.x)**2 +
(self.y other_point.y)**2)
Try typing or loading (remember, it's python -i filename.py) this file into the
interactive interpreter. Then enter help(Point)<enter> at the Python prompt.
You should see nicely formatted documentation for the class, as shown in the
following screenshot:
[ 42 ]
Chapter 2
Now that we know how to create classes and instantiate objects, it is time to think
about organizing them. For small programs, we can just put all our classes into one
file and put some code at the end of the file to start them interacting. However, as
our projects grow, it can become difficult to find one class that needs to be edited
among the many classes we've defined. This is where modules come in. Modules are
simply Python files, nothing more. The single file in our small program is a module.
Two Python files are two modules. If we have two files in the same folder, we can
load a class from one module for use in the other module.
For example, if we are building an e-commerce system, we will likely be storing a
lot of data in a database. We can put all the classes and functions related to database
access into a separate file (we'll call it something sensible: database.py). Then our
other modules (for example: customer models, product information, and inventory)
can import classes from that module in order to access the database.
The import statement is used for importing modules or specific classes or functions
from modules. We've already seen an example of this in our Point class in the
previous section. We used the import statement to get Python's built-in math
module so we could use its sqrt function in our distance calculation.
Here's a concrete example. Assume we have a module called database.py that
contains a class called Database, and a second module called products.py that is
responsible for product-related queries. At this point, we don't need to think too
much about the contents of these files. What we know is that products.py needs
to instantiate the Database class from database.py so it can execute queries on the
product table in the database.
There are several variations on the import statement syntax that can be used to
access the class.
import database
db = database.Database()
# Do queries on db
This version imports the database module into the products namespace
(the list of names currently accessible in a module or function), so any class or
function in the database module can be accessed using database.<something>
notation. Alternatively, we can import just the one class we need using the from...
import syntax:
from database import Database
db = Database()
# Do queries on db
[ 43 ]
Objects in Python
If, for some reason, products already has a class called Database, and we don't
want the two names to be confused, we can rename the class when used inside the
products module:
from database import Database as DB
db = DB()
# Do queries on db
We can also import multiple items in one statement. If our database module also
contains a Query class, we can import both classes using:
from database import Database, Query
Some sources say that we can even import all classes and functions from the
database module using this syntax:
from database import *
Don't do this. Every experienced Python programmer will tell you that you should
never use this syntax. They'll use obscure justifications like, "it clutters up the
namespace", which doesn't make much sense to beginners. One way to learn why
to avoid this syntax is to use it and try to understand your code two years later.
But we can save some time and two years of poorly written code with a quick
explanation now!
When we explicitly import the database class at the top of our file using from
database import Database, we can easily see where the Database class comes
from. We might use db = Database() 400 lines later in the file, and we can quickly
look at the imports to see where that Database class came from. Then if we need
clarification as to how to use the Database class, we can visit the original file (or
import the module in the interactive interpreter and use the help(database.
Database) command). However, if we use from database import * syntax, it
takes a lot longer to find where that class is located. Code maintenance becomes
a nightmare.
In addition, many editors are able to provide extra functionality, such as reliable
code completion or the ability to jump to the definition of a class if normal imports
are used. The import * syntax usually completely destroys their ability to do
this reliably.
Finally, using the import * syntax can bring unexpected objects into our local
namespace. Sure, it will import all the classes and functions defined in the module
being imported from, but it will also import any classes or modules that were
themselves imported into that file!
[ 44 ]
Chapter 2
In spite of all these warnings, you may think, "if I only use from X import * syntax
for one module, I can assume any unknown imports come from that module". This
is technically true, but it breaks down in practice. I promise that if you use this
syntax, you (or someone else trying to understand your code) will have extremely
frustrating moments of, "Where on earth can this class be coming from?" Every name
used in a module should come from a well-specified place, whether it is defined in
that module, or explicitly imported from another module. There should be no magic
variables that seem to come out of thin air. We should always be able to immediately
identify where the names in our current namespace originated.
Organizing the modules
As a project grows into a collection of more and more modules, we may find that
we want to add another level of abstraction, some kind of nested hierarchy on our
modules' levels. But we can't put modules inside modules; one file can only hold one
file, after all, and modules are nothing more than Python files.
Files, however, can go in folders and so can modules. A package is a collection of
modules in a folder. The name of the package is the name of the folder. All we need
to do to tell Python that a folder is a package and place a (normally empty) file in the
folder named __init__.py. If we forget this file, we won't be able to import modules
from that folder.
Let's put our modules inside an ecommerce package in our working folder, which
will also contain a main.py to start the program. Let's additionally add another
package in the ecommerce package for various payment options. The folder
hierarchy will look like this:
parent_directory/
main.py
ecommerce/

__init__.py

database.py

products.py

payments/

__init__.py

paypal.py

authorizenet.py
When importing modules or classes between packages, we have to be cautious about

the syntax. In Python 3, there are two ways of importing modules: absolute imports
and relative imports.
[ 45 ]
Objects in Python
Absolute imports
Absolute imports specify the complete path to the module, function, or path we
want to import. If we need access to the Product class inside the products module,
we could use any of these syntaxes to do an absolute import:
import ecommerce.products
product = ecommerce.products.Product()
or
from ecommerce.products import Product
product = Product()
or
from ecommerce import products
product = products.Product()
The import statements separate packages or modules using the period as

a separator.
These statements will work from any module. We could instantiate a Product using
this syntax in main.py, in the database module, or in either of the two payment
modules. Indeed, so long as the packages are available to Python, it will be able
to import them. For example, the packages can also be installed to the Python site
packages folder, or the PYTHONPATH environment variable could be customized
to dynamically tell Python what folders to search for packages and modules it is
going to import.
So with these choices, which syntax do we choose? It depends on your personal
taste and the application at hand. If there are dozens of classes and functions inside
the products module that I want to use, I generally import the module name using
the from ecommerce import products syntax and then access the individual classes
using products.Product. If I only need one or two classes from the products
module, I import them directly using the from ecommerce.proucts import Product
syntax. I don't personally use the first syntax very often unless I have some kind of
name conflict (for example, I need to access two completely different modules called
products and I need to separate them). Do whatever you think makes your code
look more elegant.
[ 46 ]
Chapter 2
Relative imports
When working with related modules in a package, it seems kind of silly to specify
the full path; we know what our parent module is named. This is where relative
imports come in. Relative imports are basically a way of saying "find a class,
function, or module as it is positioned relative to the current module". For example, if
we are working in the products module and we want to import the Database class
from the database module "next" to it, we could use a relative import:
from .database import Database
The period in front of database says, "Use the database module inside the
current package". In this case, the current package is the package containing the
products.py file we are currently editing, that is, the ecommerce package.
If we were editing the paypal module inside the ecommerce.payments package, we
would want to say, "Use the database package inside the parent package", instead.
That is easily done with two periods:
from ..database import Database
We can use more periods to go further up the hierarchy. Of course, we can also
go down one side and back up the other. We don't have a deep enough example
hierarchy to illustrate this properly, but the following would be a valid import if
we had a ecommerce.contact package containing an email module and wanted
to import the send_mail function into our paypal module:
from ..contact.email import send_mail
This import uses two periods to say, "the parent of the payments package", then uses
normal package.module syntax to go back "up" into the contact package.
Inside any one module, we can specify variables, classes, or functions. They can be
a handy way of storing global state without namespace conflicts. For example, we
have been importing the Database class into various modules and then instantiating
it, but it might make more sense to have only one database object globally available
from the database module. The database module might look like this:
class Database:
# the database implementation
pass
database = Database()
[ 47 ]
Objects in Python
Then we can use any of the import methods we've discussed to access the database
object, for example:
from ecommerce.database import database
A problem with the above class is that the database object is created immediately
when the module is first imported, which is usually when the program starts up.
This isn't always ideal, since connecting to a database can take a while, slowing
down startup, or the database connection information may not yet be available.
We could delay creating the database until it is actually needed by calling an
initialize_database function to create the module-level variable:
class Database:
# the database implementation
pass
database = None
def initialize_database():
global database
database = Database()
The global keyword tells Python that the database variable inside initialize_
database is the module-level one we just defined. If we had not specified the
variable as global, Python would have created a new local variable that would be
discarded when the method exits, leaving the module-level value unchanged.
As these two examples illustrate, all code in a module is executed immediately at the
time it is imported. However, if it is inside a method or function, the function will
be created, but its internal code will not be executed until the function is called. This
can be a tricky thing for scripts (like the main script in our e-commerce example)
that perform execution. Often, we will write a program that does something useful,
and then later find that we want to import a function or class from that module in
a different program. But as soon as we import it, any code at the module level is
immediately executed. If we are not careful, we can end up running the first program
when we really only meant to access a couple functions inside that module.
To solve this, we should always put our startup code in a function (conventionally
called main) and only execute that function when we know we are executing as a
script, but not when our code is being imported from a different script. But how do
we know that?:
class UsefulClass:
'''This class might be useful to other modules.'''
pass
[ 48 ]
Chapter 2
def main():
'''creates a useful class and does something with it for our
module.'''
useful = UsefulClass()
print(useful)
if __name__ == "__main__":
main()
Every module has a __name__ special variable (remember, Python uses double
underscores for special variables, like a class's __init__ method) that specifies the
name of the module when it was imported. But when the module is executed directly
with python module.py, it is never imported, so the __name__ is set to the string
"__main__". Make it a policy to wrap all your scripts in an if __name__ ==
"__main__": test, just in case you write a function you will find useful to be
imported by other code someday.
So methods go in classes, which go in modules, which go in packages. Is that all
there is to it?
Actually, no. That is the typical order of things in a Python program, but it's not the
only possible layout. Classes can be defined anywhere. They are typically defined at
the module level, but they can also be defined inside a function or method, like so:
def format_string(string, formatter=None):
'''Format a string using the formatter object, which
is expected to have a format() method that accepts
a string.'''
class DefaultFormatter:
'''Format a string in title case.'''
def format(self, string):
return str(string).title()
if not formatter:
formatter = DefaultFormatter()
return formatter.format(string)
hello_string = "hello world, how are you today?"
print(" input: " + hello_string)
print("output: " + format_string(hello_string))
[ 49 ]
Objects in Python
Output:
input: hello world, how are you today?
output: Hello World, How Are You Today?
The format_string function accepts a string and optional formatter object, and then
applies the formatter to that string. If no formatter is supplied, it creates a formatter
of its own as a local class and instantiates it. Since it is created inside the scope of
the function, this class cannot be accessed from anywhere outside of that function.
Similarly, functions can be defined inside other functions as well; in general, any
Python statement can be executed at any time. These "inner" classes and functions
are useful for "one-off" items that don't require or deserve their own scope at the
module level, or only make sense inside a single method.
Who can access my data?
Most object-oriented programming languages have a concept of "access control".

This is related to abstraction. Some attributes and methods on an object are marked
"private", meaning only that object can access them. Others are marked "protected",
meaning only that class and any subclasses have access. The rest are "public",
meaning any other object is allowed to access them.
Python doesn't do that. Python doesn't really believe in enforcing laws that might
someday get in your way. Instead, it provides unenforced guidelines and best
practices. Technically, all methods and attributes on a class are publicly available. If
we want to suggest that a method should not be used publicly, we can put a note in
docstrings indicating if a method is meant for internal use only, (preferably with an
explanation of how the public-facing API works!).
By convention, we can also prefix an attribute or method with an underscore
character: _. Most Python programmers will interpret this as, "This is an internal
variable, think three times before accessing it directly". But there is nothing stopping
them from accessing it if they think it is in their best interest to do so. Yet, if they
think so, why should we stop them? We may not have any idea what future uses
our classes may be put to.
[ 50 ]
Chapter 2
There's another thing you can do to strongly suggest that outside objects don't access
a property or method. Prefix it with a double underscore: __. This will perform name
mangling on the attribute in question. This basically means that the method can still
be called by outside objects if they really want to do it, but it requires extra work
and is a strong indicator that you think your attribute should remain private.
For example:
class SecretString:
'''A not-at-all secure way to store a secret string.'''
def __init__(self, plain_string, pass_phrase):
self.__plain_string = plain_string
self.__pass_phrase = pass_phrase
def decrypt(self, pass_phrase):
'''Only show the string if the pass_phrase is correct.'''
if pass_phrase == self.__pass_phrase:
return self.__plain_string
else:
return ''
If we load this class and test it in the interactive interpreter, we can see that it hides
the plaintext string from the outside world:
>>> secret_string = SecretString("ACME: Top Secret", "antwerp")
>>> print(secret_string.decrypt("antwerp"))
ACME: Top Secret
>>> print(secret_string.__plain_text)
AttributeError: 'SecretString' object has no attribute
'__plain_text'
>>>
It looks like it works. Nobody can access our plain_text attribute without the
passphrase, so it must be safe. Before we get too excited, though, let's see how
easy it can be to hack our security:
>>> print(secret_string._SecretString__plain_string)
ACME: Top Secret
[ 51 ]
Objects in Python
Oh No! Somebody has hacked our secret string. Good thing we checked! This is
Python name mangling at work. When we use a double underscore, the property
is prefixed with _<classname>. When methods in the class internally access the
variable, they are automatically unmangled. When external classes wish to access
it, they have to do the name mangling themselves. So name mangling does not
guarantee privacy, it only strongly recommends it. Most Python programmers
will not touch a double-underscore variable on another object unless they have
an extremely compelling reason to do so.
However, most Python programmers will not touch a single-underscore variable
without a compelling reason either. For the most part, there is no good reason to
use a name-mangled variable in Python, and doing so can cause grief. For example,
a name-mangled variable may be useful to a subclass, and it would have to do the
mangling itself. Let other objects access your hidden information if they want to, just
let them know, using a single-underscore prefix or some clear docstrings that you
think this is not a good idea.
Finally, we can import code directly from packages, as opposed to just
modules inside packages. In our earlier example, we had an ecommerce package
containing two modules named database.py and products.py. The database
module contains a db variable that is accessed from a lot of places. Wouldn't it be
convenient if this could be imported as import ecommerce.db instead of import
ecommerce.database.db?
Remember the __init__.py file that defines a directory as a package? That file
can contain any variables or class declarations we like, and they will be available
as part of the package. In our example, if the ecommerce/__init__.py file
contained this line:
from .database import db
We could then access the db attribute from main.py or any other file using
this import:
from ecommerce import db
It might help to think of the __init__.py file as if it was an ecommerce.py file if that
file were a module instead of a package. This can also be useful if you put all your
code in a single module and later decide to break it up into a package of modules.
The __init__.py file for the new package can still be the main point of contact for
other modules talking to it, but the code can be internally organized into several
different modules or subpackages.
[ 52 ]
Chapter 2
Case study
To tie it all together, let's build a simple command-line notebook application. This is
a fairly simple task, so we won't be experimenting with multiple packages. We will,
however, see common usage of classes, functions, methods, and docstrings.
Let's start with a quick analysis: Notes are short memos stored in a notebook. Each
note should record the day it was written and can have tags added for easy querying.
It should be possible to modify notes. We also need to be able to search for notes. All
of this should be done from the command-line.
The obvious object is the Note. Less obvious is a Notebook container object. Tags and
dates also seem to be objects, but we can use dates from Python's standard library
and a comma-separated string for tags. To avoid complexity at this point, let's not
define separate classes for these objects.
Note objects have attributes for the memo itself, tags, and a creation_date. Each
note will also need a unique integer id, so that users can select them in a menu
interface. Notes could have a method to modify note content and another for tags,
or we could just let the notebook access those attributes directly. To make searching
easier, we should put a match method on the Note. This method will accept a string
and can tell us if a note matches the string without accessing the attributes directly.
That way, if we want to modify the search parameters (to search tags instead of note
contents, for example, or to make the search case-insensitive), we only have to do it
in one place.
The Notebook obviously has the list of notes as an attribute. It will also need a
search method that returns a list of filtered notes.
But how do we interact with these objects? We've specified a command-line app,
which can either mean we run the program with different options to add or edit
commands, or we could have some kind of a menu that allows us to pick different
things to do to the notebook. It would be nice if we could design it so that either
interface was allowed, or we could add other interfaces such as a GUI toolkit or a
web-based interface in the future.
As a design decision, we'll implement the menu interface now, but will keep the
command-line options version in mind to ensure we design our Notebook class with
extensibility in mind.
So if we have two command-line interfaces each interacting with the Notebook, then
Notebook is going to need some methods for them to interact with. We'll need to be
able to add a new note, and modify an existing note by id, in addition to the search
method we've already discussed. The interfaces will also need to be able to list all
notes, but they can do that by accessing the notes list attribute directly.
[ 53 ]
Objects in Python
We may be missing a few details, but that gives us a really good overview of the
code we need to write. We can summarize all this in a simple class diagram:
Before writing any code, let's define the folder structure for this project. The menu
interface should clearly be in its own module, since it will be an executable script,
and we may have other executable scripts accessing the notebook in the future. The
Notebook and Note objects can live together in one module. These modules can both
exist in the same top-level directory without having to put them in a package. An
empty command_option.py module can help remind us in the future that we were
planning to add new user interfaces.
parent_directory/
notebook.py
menu.py
command_option.py
Now, on to the code. Let's start by defining the Note class, as it seems simplest.
The following example presents Note in its entirety. Docstrings within the example
explain how it all fits together.
import datetime
# Store the next available id for all new notes
last_id = 0
class Note:
'''Represent a note in the notebook. Match against a
string in searches and store tags for each note.'''
def __init__(self, memo, tags=''):
[ 54 ]
Chapter 2
'''initialize a note with memo and optional
space-separated tags. Automatically set the note's
creation date and a unique id.'''
self.memo = memo
self.tags = tags
self.creation_date = datetime.date.today()
global last_id
last_id += 1
self.id = last_id
def match(self, filter):
'''Determine if this note matches the filter
text. Return True if it matches, False otherwise.
Search is case sensitive and matches both text and
tags.'''
return filter in self.memo or filter in self.tags
Before continuing, we should quickly fire up the interactive interpreter and test our
code so far. Test frequently and often, because things never work the way you expect
them to. Indeed, when I tested my first version of this example I found out I had
forgotten the self argument in the match function! We'll discuss automated testing
in Chapter 10; for now, it suffices to check a few things using the interpreter:
>>> from notebook import Note
>>> n1 = Note("hello first")
>>> n2 = Note("hello again")
>>> n1.id
1
>>> n2.id
2
>>> n1.match('hello')
True
>>> n2.match('second')
False
>>>
[ 55 ]
Objects in Python
It looks like everything is behaving as expected. Let's create our notebook next:
class Notebook:
'''Represent a collection of notes that can be tagged,
modified, and searched.'''
def __init__(self):
'''Initialize a notebook with an empty list.'''
self.notes = []
def new_note(self, memo, tags=''):
'''Create a new note and add it to the list.'''
self.notes.append(Note(memo, tags))
def modify_memo(self, note_id, memo):
'''Find the note with the given id and change its
memo to the given value.'''
for note in self.notes:
if note.id == note_id:
note.memo = memo
break
def modify_tags(self, note_id, tags):
tags to the given value.'''
note.tags = tags
break
def search(self, filter):
'''Find all notes that match the given filter
string.'''
return [note for note in self.notes if
note.match(filter)]
We'll clean that up in a minute. First let's test it to make sure it works:
>>> from notebook import Note, Notebook
>>> n = Notebook()
>>> n.new_note("hello world")
>>> n.new_note("hello again")
>>> n.notes
[ 56 ]
Chapter 2
[<notebook.Note object at 0xb730a78c>, <notebook.Note object at
0xb73103ac>]
>>> n.notes[0].id
1
>>> n.notes[1].id
2
>>> n.notes[0].memo
'hello world'
>>> n.search("hello")
[<notebook.Note object at 0xb730a78c>, <notebook.Note object at
0xb73103ac>]
>>> n.search("world")
[<notebook.Note object at 0xb730a78c>]
>>> n.modify_memo(1, "hi world")
>>> n.notes[0].memo
'hi world'
>>>
It does work. The code is a little messy though; our modify_tags and modify_memo
methods are almost identical. That's not good coding practice. Let's see if we
can fix it.
Both methods are trying to identify the note with a given id before doing something
to that note. So let's add a method to locate the note with a specific id. We'll prefix
the method name with an underscore to suggest that the method is for internal use
only, but of course, our menu interface can access the method if it wants to.
def _find_note(self, note_id):
'''Locate the note with the given id.'''
return note
return None
self._find_note(note_id).memo = memo
[ 57 ]
Objects in Python
That should work for now; let's have a look at the menu interface. The interface
simply needs to present a menu and allow the user to input choices. Here's
a first try:
import sys
from notebook import Notebook, Note
class Menu:
'''Display a menu and respond to choices when run.'''
def __init__(self):
self.notebook = Notebook()
self.choices = {
"1": self.show_notes,
"2": self.search_notes,
"3": self.add_note,
"4": self.modify_note,
"5": self.quit
}
def display_menu(self):
print("""
Notebook Menu
1. Show all Notes
2. Search Notes
3. Add Note
4. Modify Note
5. Quit
""")
def run(self):
'''Display the menu and respond to choices.'''
while True:
self.display_menu()
choice = input("Enter an option: ")
action = self.choices.get(choice)
if action:
action()
else:
print("{0} is not a valid choice".format(choice))
def show_notes(self, notes=None):
if not notes:
notes = self.notebook.notes
[ 58 ]
Chapter 2
for note in notes:
print("{0}: {1}\n{2}".format(
note.id, note.tags, note.memo))
def search_notes(self):
filter = input("Search for: ")
notes = self.notebook.search(filter)
self.show_notes(notes)
def add_note(self):
memo = input("Enter a memo: ")
self.notebook.new_note(memo)
print("Your note has been added.")
def modify_note(self):
id = input("Enter a note id: ")
memo = input("Enter a memo: ")
tags = input("Enter tags: ")
if memo:
self.notebook.modify_memo(id, memo)
if tags:
self.notebook.modify_tags(id, tags)
def quit(self):
print("Thank you for using your notebook today.")
sys.exit(0)
if __name__ == "__main__":
Menu().run()
This code first imports the notebook objects using an absolute import. Relative
imports wouldn't work because we haven't placed our code inside a package. The
Menu class's run method repeatedly displays a menu and responds to choices by
calling functions on the notebook. This is done using an idiom that is rather peculiar
to Python. The choices entered by the user are strings. In the menu's __init__ we
create a dictionary that maps strings to functions on the menu object itself. Then
when the user makes a choice, we retrieve the object from the dictionary. The action
variable actually refers to a specific method and is called by appending empty
brackets (since none of the methods require parameters) to the variable. Of course,
the user might have entered an inappropriate choice, so we check if the action really
exists before calling it.
[ 59 ]
Objects in Python
The various methods each request user input and call appropriate methods on the
Notebook object associated with it. For the search implementation, we notice that
after we've filtered the notes, we need to show them. So we make the show_notes
function serve double duty; it accepts an optional notes parameter. If it's supplied,
it displays only the filtered notes, but if it's not, it displays all notes. Since the notes
parameter is optional, show_notes can still be called with no parameters as an empty
menu item.
If we test this code, we'll find that modifying notes doesn't work. There are two
bugs, namely:
The notebook crashes when we enter a note id that does not exist. We should
never trust our users to enter correct data!
Even if we enter a correct id, it will crash because the note ids are integers,
but our menu is passing a string.
The latter bug can be solved by modifying the Notebook class's _find_note
method to compare the values using strings instead of the integers stored in
the note, as follows:
def _find_note(self, note_id):
'''Locate the note with the given id.'''
if str(note.id) == str(note_id):
return note
return None
We simply convert both the input (note_id) and the note's id to strings before
comparing them. We could also convert the input to an integer, but then we'd have
trouble if the user had entered the letter "a" instead of the number "1".
The problem with users entering note ids that don't exist can be fixed by changing
the two modify methods on the notebook to check if _find_note returned a note or
not, like this:
note = self._find_note(note_id)
if note:
note.memo = memo
return True
return False
[ 60 ]
Chapter 2
This method has been updated to return True or False, depending on whether a
note has been found. The menu could use this return value to display an error if the
user entered an invalid note. This code is a bit unwieldy though; it would look a bit
better if it raised an exception instead. We'll cover those in Chapter 4.
Exercises
Write some object-oriented code. The goal is to use the principles and syntax you
learned in this chapter to ensure you can use it, instead of just reading about it. If
you've been working on a Python project, go back over it and see if there are some
objects you can create and add properties or methods to. If it's large, try dividing it
into a few modules or even packages and play with the syntax.
If you don't have such a project, try starting a new one. It doesn't have to be
something you intend to finish, just stub out some basic design parts. You don't
need to fully implement everything, often just a print("this method will do
something") is all you need to get the overall design in place. This is called
top-down design, when you work out the different interactions and describe
how they should work before actually implementing what they do. The converse,
bottom-up design, implements details first and then ties them all together. Both
patterns are useful at different times, but for understanding object-oriented
principles, a top-down workflow is more suitable.
If you're having trouble coming up with ideas, try writing a TO DO application.
(Hint: It would be similar to the design of the notebook application, but with extra
date management methods.) It can keep track of things you want to do each day, and
allow you to mark them as completed.
Now, try designing a bigger project; it doesn't have to actually do anything, but
make sure you experiment with the package and module importing syntax. Add
some functions in various modules and try importing them from other modules and
packages. Use relative and absolute imports. See the difference, and try to imagine
scenarios where you would want to use each one.
[ 61 ]
Objects in Python
Summary
In this chapter, we learned how simple it is to create classes and assign properties
and methods in Python. We also covered access control and different levels of scope
(packages, modules, classes, and functions). In particular, we covered:
Class syntax
Attributes and methods
Initializers and constructors
Relative and absolute imports
Access control and its limitations
In the next chapter, we'll learn how to share implementation using inheritance.
[ 62 ]
When Objects are Alike

In the programming world, duplicate code is considered evil. We should not have
multiple copies of the same, or similar code in different places.
There are many ways to merge similar pieces of code or objects with similar
functionality. In this chapter, we'll be covering the most famous object-oriented
principle: inheritance. As discussed in Chapter 1, inheritance allows us to create
"is a" relationships between two or more classes, abstracting common details into
superclasses and storing specific ones in the subclass. In particular, we'll be covering
the Python syntax and principles for:
Basic inheritance
Inheriting from built-ins
Polymorphism and duck typing
Basic inheritance
Technically, every class we create uses inheritance. All Python classes are subclasses
of the special class named object. This class provides very little in terms of data
and behaviors (those behaviors it does provide are all double-underscore methods
intended for internal use only), but it does allow Python to treat all objects in the
same way.
If we don't explicitly inherit from a different class, our classes will automatically
inherit from object. However, we can openly state that our class derives from
object using the following syntax:
class MySubClass(object):
pass
This is inheritance! This example is, technically, no different from our very first
example in Chapter 2, since Python 3 automatically inherits from object if we don't
explicitly provide a different superclass. A superclass, or parent class, is a class that
is being inherited from. A subclass is a class that is inheriting from a superclass. In
this case, the superclass is object, and MySubClass is the subclass. A subclass is also
said to be derived from its parent class or that the subclass extends the parent.
As you've probably figured out from the example, inheritance requires a minimal
amount of extra syntax over a basic class definition. Simply include the name of the
parent class inside a pair of parentheses after the class name, but before the colon
terminating the class definition. This is all we have to do to tell Python that the new
class should be derived from the given superclass.
How do we apply inheritance in practice? The simplest and most obvious use of
inheritance is to add functionality to an existing class. Let's start with a simple
contact manager that tracks the name and e-mail address of several people. The
contact class is responsible for maintaining a list of all contacts in a class variable,
and for initializing the name and address, in this simple class:
class Contact:
all_contacts = []
def __init__(self, name, email):
self.name = name
self.email = email
Contact.all_contacts.append(self)
This example introduces us to class variables. The all_contacts list, because

it is part of the class definition, is actually shared by all instances of this class.
This means that there is only one Contact.all_contacts list, and if we call
self.all_contacts on any one object, it will refer to that single list. The code
in the initializer ensures that whenever we create a new contact, the list will
automatically have the new object added. Be careful with this syntax, for if you
ever set the variable using self.all_contacts, you will actually be creating a
new instance variable on the object; the class variable will still be unchanged and
accessible as Contact.all_contacts.
[ 64 ]
Chapter 3
This is a very simple class that allows us to track a couple pieces of data about our
contacts. But what if some of our contacts are also suppliers that we need to order
supplies from? We could add an order method to the Contact class, but that would
allow people to accidentally order things from contacts who are customers or family
friends. Instead, let's create a new Supplier class that acts like a Contact, but has an
additional order method:
class Supplier(Contact):
def order(self, order):
print("If this were a real system we would send "
"{} order to {}".format(order, self.name))
Now, if we test this class in our trusty interpreter, we see that all contacts, including
suppliers, accept a name and e-mail address in their __init__, but only suppliers
have a functional order method:
>>> c = Contact("Some Body", "[email protected]")
>>> s = Supplier("Sup Plier", "[email protected]")
>>> print(c.name, c.email, s.name, s.email)
Some Body [email protected] Sup Plier [email protected]
>>> c.all_contacts
[<__main__.Contact object at 0xb7375ecc>,
<__main__.Supplier object at 0xb7375f8c>]
>>> c.order("Ineed pliers")
AttributeError: 'Contact' object has no attribute 'order'
>>> s.order("I need pliers")
If this were a real system we would send I need pliers order to
Supplier
>>>
So now our Supplier class can do everything a Contact can do (including adding
itself to the list of all_contacts) and all the special things it needs to handle as a
supplier. This is the beauty of inheritance.
[ 65 ]
Extending built-ins
One of the most interesting uses of this kind of inheritance is adding functionality
to built-in classes. In the Contact class seen earlier, we are adding contacts to a list
of all contacts. What if we also wanted to search that list by name? Well, we could
add a method on the Contact class to search it, but it feels like this method actually
belongs on the list itself. We can do this using inheritance:
class ContactList(list):
def search(self, name):
'''Return all contacts that contain the search value
in their name.'''
matching_contacts = []
for contact in self:
if name in contact.name:
matching_contacts.append(contact)
return matching_contacts
class Contact:
all_contacts = ContactList()
def __init__(self, name, email):
self.name = name
self.email = email
self.all_contacts.append(self)
Instead of instantiating a normal list as our class variable, we create a new

ContactList class that extends the built-in list. Then we instantiate this subclass
as our all_contacts list. We can test the new search functionality as follows:
>>> c1 = Contact("John A", "[email protected]")
>>> c2 = Contact("John B", "[email protected]")
>>> c3 = Contact("Jenna C", "[email protected]")
>>> [c.name for c in Contact.all_contacts.search('John')]
['John A', 'John B']
>>>
Are you wondering how we changed the built-in syntax [] into something we can
inherit from? Creating an empty list with [] is actually a shorthand for creating an
empty list using list(); the two syntaxes are identical:
>>> [] == list()
True
[ 66 ]
Chapter 3
So, the list data type is like a class that we can extend, not unlike object.
As a second example, we can extend the dict class, which is the long way of creating
a dictionary (the {:} syntax).
class LongNameDict(dict):
def longest_key(self):
longest = None
for key in self:
if not longest or len(key) > len(longest):
longest = key
return longest
This is easy to test in the interactive interpreter:

>>> longkeys = LongNameDict()
>>> longkeys['hello'] = 1
>>> longkeys['longest yet'] = 5
>>> longkeys['hello2'] = 'world'
>>> longkeys.longest_key()
'longest yet'
Most built-in types can be similarly extended. Commonly extended built-ins are
object, list, set, dict, file, and str. Numerical types such as int and float
are also occasionally inherited from.
Overriding and super
So inheritance is great for adding new behavior to existing classes, but what about
changing behavior? Our contact class allows only a name and an e-mail address.
This may be sufficient for most contacts, but what if we want to add a phone number
for our close friends?
As we saw in Chapter 2, we can do this easily by just setting a phone attribute on the
contact after it is constructed. But if we want to make this third variable available on
initialization, we have to override __init__. Overriding is altering or replacing a
method of the superclass with a new method (with the same name) in the subclass.
No special syntax is needed to do this; the subclass's newly created method is
automatically called instead of the superclass's method. For example:
self.name = name
self.email = email
self.phone = phone
[ 67 ]
Any method can be overridden, not just __init__. Before we go on, however,
we need to correct some problems in this example. Our Contact and Friend
classes have duplicate code to set up the name and email properties; this can make
maintenance complicated, as we have to update the code in two or more places.
More alarmingly, our Friend class is neglecting to add itself to the all_contacts
list we have created on the Contact class.
What we really need is a way to call code on the parent class. This is what the super
function does; it returns the object as an instance of the parent class, allowing us to
call the parent method directly:
super().__init__(name, email)
self.phone = phone
This example first gets the instance of the parent object using super, and calls
__init__ on that object, passing in the expected arguments. It then does its own
initialization, namely setting the phone attribute.
A super() call can be made inside any method, not just __init__. This means all
methods can be modified via overriding and calls to super. The call to super can also
be made at any point in the method; we don't have to make the call as the first line
in the method. For example, we may need to manipulate the incoming parameters
before forwarding them to the superclass.
Multiple inheritance is a touchy subject. In principle, it's very simple: a subclass

that inherits from more than one parent class is able to access functionality from
both of them. In practice, this is much less useful than it sounds and many expert
programmers recommend against using it. So we'll start with a warning:
As a rule of thumb, if you think you need multiple inheritance, you're
probably wrong, but if you know you need it, you're probably right.
The simplest and most useful form of multiple inheritance is called a mixin. A mixin
is generally a superclass that is not meant to exist on its own, but is meant to be
inherited by some other class to provide extra functionality. For example, let's say
we wanted to add functionality to our Contact class that allows sending an e-mail
to self.email. Sending e-mail is a common task that we might want to use on many
other classes. So we can write a simple mixin class to do the e-mailing for us:
[ 68 ]
Chapter 3
class MailSender:
def send_mail(self, message):
print("Sending mail to " + self.email)
# Add e-mail logic here
For brevity, we won't include the actual e-mail logic here; if you're interested in
studying how it's done, see the smtplib module in the Python standard library.
This class doesn't do anything special (in fact, it can barely function as a
stand-alone class), but it does allow us to define a new class that is both a
Contact and a MailSender, using multiple inheritance:
class EmailableContact(Contact, MailSender):
pass
The syntax for multiple inheritance looks like a parameter list in the class definition.
Instead of including one base class inside the parenthesis, we include two (or more),
separated by a comma. We can test this new hybrid to see the mixin at work:
>>> e = EmailableContact("John Smith", "[email protected]")
>>> Contact.all_contacts
[<__main__.EmailableContact object at 0xb7205fac>]
>>> e.send_mail("Hello, test e-mail here")
Sending mail to [email protected]
The Contact initializer is still adding the new contact to the all_contacts list, and
the mixin is able to send mail to self.email so we know everything is working.
That wasn't so hard, and you're probably wondering what the dire warnings about
multiple inheritance are. We'll get into the complexities in a minute, but let's consider
what options we had, other than using a mixin here:
We could have used single inheritance and added the send_mail function to
the subclass. The disadvantage here is that the e-mail functionality then has
to be duplicated for any other classes that need e-mail.
We can create a stand-alone Python function for sending mail, and just call
that, with the correct e-mail address supplied as a parameter, when e-mail
needs to be sent.
We could monkey-patch (we'll briefly cover monkey-patching in

Chapter 7) the Contact class to have a send_mail method after the class
has been created. This is done by defining a function that accepts the self
argument, and setting it as an attribute on an existing class.
[ 69 ]
Multiple inheritance works all right when mixing methods from different classes, but
it gets very messy when we have to work with calling methods on the superclass.
Why? Because there are multiple superclasses. How do we know which one to call?
How do we know what order to call them in?
Let's explore these questions by adding a home address to our Friend class. What
are some ways we could do this? An address is a collection of strings representing
the street, city, country, and other related details of the contact. We could pass each
of these strings as parameters into the Friend class's __init__ method. We could
also store these strings in a tuple or dictionary and pass them into __init__ as a
single argument. This is probably the best course of action if there is no additional
functionality that needs to be added to the address.
Another option would be to create a new Address class to hold those strings
together, and then pass an instance of this class into the __init__ in our Friend
class. The advantage of this solution is that we can add behavior (say, a method to
give directions to that address or to print a map) to the data instead of just storing it
statically. This would be utilizing composition, the "has a" relationship we discussed
in Chapter 1. Composition is a perfectly viable solution to this problem and allows
us to reuse Address classes in other entities such as buildings, businesses,
or organizations.
However, inheritance is also a viable solution, and that's what we want to explore, so
let's add a new class that holds an address. We'll call this new class AddressHolder
instead of Address, because inheritance defines an "is a" relationship. It is not correct
to say a Friend is an Address, but since a friend can have an Address, we can argue
that a Friend is an AddressHolder. Later, we could create other entities (companies,
buildings) that also hold addresses. Here's our AddressHolder class:
class AddressHolder:
def __init__(self, street, city, state, code):
self.street = street
self.city = city
self.state = state
self.code = code
Very simple; we just take all the data and toss it into instance variables
upon initialization.
[ 70 ]
Chapter 3
The diamond problem
But how can we use this in our existing Friend class, which is already inheriting
from Contact? Multiple inheritance, of course. The tricky part is that we now have
two parent __init__ methods that both need to be initialized. And they need to be
initialized with different arguments. How do we do that? Well, we could start with
the nave approach:
class Friend(Contact, AddressHolder):
def __init__(self, name, email, phone,
street, city, state, code):
Contact.__init__(self, name, email)
AddressHolder.__init__(
self, street, city, state, code)
self.phone = phone
In this example, we directly call the __init__ function on each of the superclasses
and explicitly pass the self argument. This example technically works; we can
access the different variables directly on the class. But there are a few problems.
First, it is possible for a superclass to go uninitialized if we neglect to explicitly call
the initializer. This is not bad in this example, but it could cause bad program crashes
in common scenarios. Imagine, for example, trying to insert data into a database that
has not been connected to.
Second, and more sinister, is the possibility of a superclass being called
multiple times, because of the organization of the class hierarchy. Look at
this inheritance diagram:
The __init__ method from the Friend class first calls __init__ on Contact which
implicitly initializes the object superclass (remember, all classes derive from
object). Friend then calls __init__ on AddressHolder, which implicitly initializes
the object superclass... again. The parent class has been set up twice. In this case,
that's relatively harmless, but in some situations, it could spell disaster. Imagine
trying to connect to a database twice for every request! The base class should only be
called once. Once, yes, but when? Do we call Friend then Contact then Object then
AddressHolder? Or Friend then Contact then AddressHolder then Object?
[ 71 ]
Technically, the order in which methods can be called can be adapted on

the fly by modifying the __mro__ (Method Resolution Order) attribute
on the class. This is beyond the scope of this book. If you think you need
to understand it, I recommend Expert Python Programming, Tarek Ziad,
Packt Publishing, or read the original documentation on the topic at:
http://www.python.org/download/releases/2.3/mro/
Let's look at a second contrived example that illustrates this problem more clearly.
Here we have a base class that has a method named call_me. Two subclasses
override that method, and then another subclass extends both of these using multiple
inheritance. This is called diamond inheritance because of the diamond shape of the
class diagram:
Diamonds are what makes multiple inheritance tricky. Technically, all multiple
inheritance in Python 3 is diamond inheritance, because all classes inherit from
object. The previous diagram, using object.__init__ is also such a diamond.
Converting this diagram to code, this example shows when the methods are called:
class BaseClass:
num_base_calls = 0
def call_me(self):
print("Calling method on Base Class")
self.num_base_calls += 1
class LeftSubclass(BaseClass):
num_left_calls = 0
def call_me(self):
BaseClass.call_me(self)
print("Calling method on Left Subclass")
self.num_left_calls += 1
[ 72 ]
Chapter 3
class RightSubclass(BaseClass):
num_right_calls = 0
def call_me(self):
BaseClass.call_me(self)
print("Calling method on Right Subclass")
self.num_right_calls += 1
class Subclass(LeftSubclass, RightSubclass):
num_sub_calls = 0
def call_me(self):
LeftSubclass.call_me(self)
RightSubclass.call_me(self)
print("Calling method on Subclass")
self.num_sub_calls += 1
This example simply ensures each overridden call_me method directly calls the
parent method with the same name. Each time it is called, it lets us know by printing
the information to the screen, and updates a static variable on the class to show how
many times it has been called. If we instantiate one Subclass object and call the
method on it once, we get this output:
>>> s = Subclass()
>>> s.call_me()
Calling method on Base Class
Calling method on Left Subclass
Calling method on Right Subclass
Calling method on Subclass
>>> print(s.num_sub_calls, s.num_left_calls, s.num_right_calls,
s.num_base_calls)
1 1 1 2
>>>
The base class's call_me method has been called twice. This isn't expected behavior
and can lead to some very difficult bugs if that method is doing actual worklike
depositing into a bank account twice.
[ 73 ]
The thing to keep in mind with multiple inheritance is that we only want to call
the "next" method in the class hierarchy, not the "parent" method. In fact, that next
method may not be on a parent or ancestor of the current class. The super keyword
comes to our rescue once again. Indeed, super was originally developed to make
complicated forms of multiple inheritance possible. Here is the same code written
using super:
class BaseClass:
num_base_calls = 0
def call_me(self):
print("Calling method on Base Class")
self.num_base_calls += 1
class LeftSubclass(BaseClass):
num_left_calls = 0
def call_me(self):
super().call_me()
print("Calling method on Left Subclass")
self.num_left_calls += 1
class RightSubclass(BaseClass):
num_right_calls = 0
def call_me(self):
super().call_me()
print("Calling method on Right Subclass")
self.num_right_calls += 1
class Subclass(LeftSubclass, RightSubclass):
num_sub_calls = 0
def call_me(self):
super().call_me()
print("Calling method on Subclass")
self.num_sub_calls += 1
The change is pretty minor; we simply replaced the nave direct calls with calls to
super(). This is simple enough, but look at the difference when we execute it:
>>> s = Subclass()
>>> s.call_me()
Calling method on Right Subclass
Calling method on Left Subclass
Calling method on Subclass
[ 74 ]
Chapter 3
>>> print(s.num_sub_calls, s.num_left_calls, s.num_right_calls,
s.num_base_calls)
1 1 1 1
Looks good, our base method is only being called once. But what is super() actually
doing here? Since the print statements are executed after the super calls, the printed
output is in the order each method is actually executed. Let's look at the output from
back to front to see who is calling what.
First call_me of Subclass calls super().call_me(), which happens to refer to
LeftSubclass.call_me(). LeftSubclass.call_me() then calls super().call_
me(), but in this case, super() is referring to RightSubclass.call_me(). Pay
particular attention to this; the super call is not calling the method on the superclass
of LeftSubclass (which is BaseClass), it is calling RightSubclass, even though
it is not a parent of LeftSubclass! This is the next method, not the parent method.
RightSubclass then calls BaseClass and the super calls have ensured each method
in the class hierarchy is executed once.
Different sets of arguments
Can you see how this is going to make things complicated when we return to our
Friend multiple inheritance example? In the __init__ method for Friend, we were
originally calling __init__ for both parent classes, with different sets of arguments:
Contact.__init__(self, name, email)
AddressHolder.__init__(self, street, city, state, code)
How can we convert this to using super? We don't necessarily know which class
super is going to try to initialize first. Even if we did, we need a way to pass the
"extra" arguments so that subsequent calls to super, on other subclasses, have the
right arguments.
Specifically, if the first call to super passes the name and email arguments
to Contact.__init__, and Contact.__init__ then calls super, it needs to
be able to pass the address related arguments to the "next" method, which is
AddressHolder.__init__.
This is a problem whenever we want to call superclass methods with the same
name, but different sets of arguments. Most often, the only time you would want
to call a superclass with a completely different set of arguments is in __init__, as
we're doing here. Even with regular methods, though, we may want to add optional
parameters that only make sense to one subclass or a set of subclasses.
[ 75 ]
Sadly, the only way to solve this problem is to plan for it from the beginning. We
have to design our base class parameter lists so that they accept keyword arguments
for any argument that is not required by every subclass implementation. We also
have to ensure the method accepts arguments it doesn't expect and pass those on in
its super call, in case they are necessary to later methods in the inheritance order.
Python's function parameter syntax provides all the tools we need to do this, but it
makes the overall code cumbersome. Have a look at the proper version of the Friend
multiple inheritance code:
class Contact:
all_contacts = []
def __init__(self, name='', email='', **kwargs):
super().__init__(**kwargs)
self.name = name
self.email = email
self.all_contacts.append(self)
class AddressHolder:
def __init__(self, street='', city='', state='', code='',
**kwargs):
self.street = street
self.city = city
self.state = state
self.code = code
class Friend(Contact, AddressHolder):
def __init__(self, phone='', **kwargs):
self.phone = phone
We've changed all arguments to keyword arguments by giving them an empty string
as a default value. We've also ensured that a **kwargs parameter is included to
capture any additional parameters that our particular method doesn't know what to
do with. It passes these parameters up to the next class with the super call.
If you aren't familiar with the **kwargs syntax, it basically collects any
keyword arguments passed into the method that were not explicitly
listed in the parameter list. These arguments are stored in a dictionary
named kwargs (we can call the variable whatever we like, but convention
suggests kw, or kwargs). When we call a different method (for example:
super().__init__) with a **kwargs syntax, it unpacks the dictionary
and passes the results to the method as normal keyword arguments. We'll
cover this in detail in Chapter 7.
[ 76 ]
Chapter 3
The previous example does what it is supposed to do. But it's starting to look messy,
and it has become difficult to answer the question, "What arguments do we need to
pass into Friend.__init__?" This is the foremost question for anyone planning to
use the class, so a docstring should be added to the method to explain what
is happening.
Further, even this implementation is insufficient if we want to "reuse" variables in
parent classes. When we pass the **kwargs variable to super, the dictionary does
not include any of the variables that were included as explicit keyword arguments.
For example, in Friend.__init__, the call to super does not have phone in the
kwargs dictionary. If any of the other classes need the phone parameter, we need to
ensure it is in the dictionary that is passed. Worse, if we forget to do that, it will be
tough to debug, because the superclass will not complain, but will simply assign the
default value (in this case, an empty string) to the variable.
There are a few ways to ensure that the variable is passed upwards. Assume the
Contact class does, for some reason, need to be initialized with a phone parameter,
and the Friend class will also need access to it. We can do any of the following:
Don't include phone as an explicit keyword argument. Instead, leave it

in the kwargs dictionary. Friend can look it up using the syntax
kwargs['phone']. When it passes **kwargs to the super call,
phone will still be in the dictionary.
Make phone an explicit keyword argument but update the kwargs

dictionary before passing it to super, using the standard dictionary
syntax kwargs['phone'] = phone.
Make phone an explicit keyword argument, but update the kwargs

dictionary using the kwargs.update method. This is useful if you have
several arguments to update. You can create the dictionary passed into
update using either the dict(phone=phone) constructor, or the dictionary
syntax {'phone': phone}.
Make phone an explicit keyword argument, but pass it to the super call
explicitly with the syntax super().__init__(phone=phone, **kwargs).
We have covered many of the caveats involved with multiple inheritance in Python.
When we need to account for all the possible situations, we have to plan for them
and our code will get messy. Basic multiple inheritance can be handy, but in many
cases, we may want to choose a more transparent way of combining two disparate
classes, usually using composition or one of the design patterns we'll be covering in
Chapter 8 and Chapter 9.
[ 77 ]
Polymorphism
We introduced polymorphism in Chapter 1. It is a fancy name describing a simple

concept; different behaviors happen depending on which subclass is being used,
without having to explicitly know what the subclass actually is. As an example,
imagine a program that plays audio files. A media player might need to load an
AudioFile object and then play it. We'd put a play() method on the object, which
is responsible for decompressing or extracting the audio and routing it to the sound
card and speakers. The act of playing an AudioFile could feasibly be as simple as:
audio_file.play()
However the process of decompressing and extracting an audio file is very different
for different types of files. The .wav files are stored uncompressed, while .mp3, .wma,
and .ogg files all have very different compression algorithms.
We can use inheritance with polymorphism to simplify the design. Each type of
file can be represented by a different subclass of AudioFile, for example, WavFile,
MP3File. Each of these would have a play() method, but that method would be
implemented differently for each file to ensure the correct extraction procedure is
followed. The media player object would never need to know which subclass of
AudioFile it is referring to; it just calls play() and polymorphically lets the object
take care of the actual details of playing. Let's look at a quick skeleton showing how
this might look:
class AudioFile:
def __init__(self, filename):
if not filename.endswith(self.ext):
raise Exception("Invalid file format")
self.filename = filename
class MP3File(AudioFile):
ext = "mp3"
def play(self):
print("playing {} as mp3".format(self.filename))
class WavFile(AudioFile):
ext = "wav"
def play(self):
print("playing {} as wav".format(self.filename))
class OggFile(AudioFile):
ext = "ogg"
def play(self):
print("playing {} as ogg".format(self.filename))
[ 78 ]
Chapter 3
All audio files check to ensure that a valid extension was given upon initialization.
But notice how the __init__ method in the parent class is able to access the ext
class variable from different subclasses? That's polymorphism at work. If the
filename doesn't end with the correct name, it raises an exception (exceptions will
be covered in detail in the next chapter). The fact that AudioFile doesn't actually
store a reference to the ext variable doesn't stop it from being able to access it on
the subclass.
In addition, each subclass of AudioFile implements play() in a different way
(this example doesn't actually play the music; audio compression algorithms really
deserve a separate book!). This is also polymorphism in action. The media player can
use the exact same code to play a file, no matter what type it is; it doesn't care what
subclass of AudioFile it is looking at. The details of decompressing the audio file are
encapsulated. If we test this example, it works as we would hope:
>>> ogg = OggFile("myfile.ogg")
>>> ogg.play()
playing myfile.ogg as ogg
>>> mp3 = MP3File("myfile.mp3")
>>> mp3.play()
playing myfile.mp3 as mp3
>>> not_an_mp3 = MP3File("myfile.ogg")
File "polymorphic_audio.py", line 4, in __init__
Exception: Invalid file format
See how AudioFile.__init__ is able to check the file type without actually
knowing what subclass it is referring to?
Polymorphism is actually one of the coolest things about object-oriented
programming, and it makes some programming designs obvious that weren't
possible in earlier paradigms. However, Python makes polymorphism less cool
because of duck typing. Duck typing in Python allows us to use any object that
provides the required behavior without forcing it to be a subclass. The dynamic
nature of Python makes this trivial. The following example does not extend
AudioFile, but it can be interacted with in Python using the exact same interface:
class FlacFile:
if not filename.endswith(".flac"):
[ 79 ]

def play(self):
print("playing {} as flac".format(self.filename))
Our media player can play this object just as easily as one that extends AudioFile.
Polymorphism is one of the most important reasons to use inheritance in many
object-oriented contexts. Because any objects that supply the correct interface can
be used interchangeably in Python, it reduces the need for polymorphic common
superclasses. Inheritance can still be useful for sharing code, but if all that is being
shared is the public interface, duck typing is all that is required. This reduced need
for inheritance also reduces the need for multiple inheritance; often, when multiple
inheritance appears to be a valid solution, we just can use duck typing to mimic one
of the multiple superclasses.
Of course, just because an object satisfies a particular interface (by providing
required methods or attributes) does not mean it will simply work in all situations.
It has to fulfill that interface in a way that makes sense in the overall system. Just
because an object provides a play() method does not mean it will automatically
work with a media player. For example, our chess AI object from Chapter 1 may have
a play() method that moves a chess piece. Even though it satisfies the interface,
this class would likely break in spectacular ways if we tried to plug it into a
media player!
Another useful feature of duck typing is that the duck-typed object only needs to
provide those method and attributes that are actually being accessed. For example,
if we needed to create a fake file object to read data from, we can create a new object
that has a read() method; we don't have to override the write method if the code
that is going to interact with the object will only be reading from the file. More
succinctly, duck typing doesn't need to provide the entire interface of an object
that is available, it only needs to fulfill the interface that is actually used.
Case study
Let's try to tie everything we've learned together with a larger example. We'll be
designing a simple real estate application that allows an agent to manage properties
available for purchase or rent. There will be two types of properties: apartments
and houses. The agent needs to be able to enter a few relevant details about new
properties, list all currently available properties, and mark a property as sold or
rented. For brevity, we won't worry about editing property details or reactivating
a property after it is sold.
[ 80 ]
Chapter 3
The project will allow the agent to interact with the objects using the Python
interpreter prompt. In this world of graphical user interfaces and web applications,
you might be wondering why we're creating such old-fashioned looking programs.
Simply put, both windowed programs and web applications require a lot of
overhead knowledge and boilerplate code to make them do what is required. If we
were developing software using either of these paradigms, we'd get so lost in "GUI
programming" or "web programming" that we'd lose sight of the object-oriented
principles we're trying to master.
Luckily, most GUI and web frameworks utilize an object-oriented approach, and
the principles we're studying now will help in understanding those systems in the
future. We'll discuss them both briefly, in Chapter 12, but complete details are far
beyond the scope of a single book.
Looking at our requirements it seems like there are quite a few nouns that might
represent classes of objects in our system. Clearly we'll need to represent a Property.
Houses and apartments may need separate classes. Rentals and purchases also seem
to require separate representation. Since we're focusing on inheritance right now,
we'll be looking at ways to share behavior using inheritance or multiple inheritance.
Clearly, House and Apartment are both types of properties, so Property can be
a superclass of those two classes. Rental and Purchase will need some extra
thought; if we use inheritance, we'll need to have separate classes, for example, for
HouseRental and HousePurchase, and use multiple inheritance to combine them.
This feels a little clunky compared to a composition/association-based design, but
let's run with it and see what we come up with.
Now then, what attributes might be associated with a Property? Regardless of
whether it is an apartment or a house, most people will want to know the square
footage, number of bedrooms, and number of bathrooms. (There are numerous
other attributes that might be modeled, but we'll keep it simple for our prototype.)
If the property is a house, it will want to advertise the number of stories, whether
it has a garage (attached, detached, or none), and whether the yard is fenced. An
apartment will want to indicate if it has a balcony, and if laundry is en-suite, coin,
or off-site.
Both property types will require a method to display the characteristics of that
property. At the moment, no other behaviors are apparent.
[ 81 ]
Rental properties will need to store the rent per month, whether the property is
furnished, and whether utilities are included, and if not, what they are estimated
to be. Properties for purchase will need to store the purchase price and estimated
annual property taxes. For our application, we'll only need to display this data, so
we can get away with just adding a display() method similar to that used in the
other classes.
Finally, we'll need an Agent object that holds a list of all properties, displays
those properties, and allows us to create new ones. Creating properties will entail
prompting the user for the relevant details for each property type. This could be
done in the Agent object, but then Agent would need to know a lot of information
about the types of properties. This is not taking advantage of polymorphism.
Another alternative would be to put the prompts in the initializer or even a
constructor for each class, but this would not allow the classes to be applied later in
a GUI or web application in the future. A better idea is to create a static method that
does the prompting and returns a dictionary of the prompted parameters. Then all
the Agent has to do is prompt the user for the type of property and payment method,
and ask the correct class to instantiate itself.
That's a lot of designing for a seemingly simple app! The following class diagram
may communicate our design decisions a little more clearly:
[ 82 ]
Chapter 3
Wow, That's a lot of inheritance arrows! I don't think it would be possible to add
another level of inheritance without crossing arrows. Multiple inheritance is messy
business, even at the design stage.
Clearly, the trickiest aspects of these classes is going to be ensuring superclass
methods get called in the inheritance hierarchy. Let's start with the Property
implementation:
class Property:
def __init__(self, square_feet='', beds='',
baths='', **kwargs):
self.square_feet = square_feet
self.num_bedrooms = beds
self.num_baths = baths
def display(self):
print("PROPERTY DETAILS")
print("================")
print("square footage: {}".format(self.square_feet))
print("bedrooms: {}".format(self.num_bedrooms))
print("bathrooms: {}".format(self.num_baths))
print()
def prompt_init():
return dict(square_feet=input("Enter the square feet: "),
beds=input("Enter number of bedrooms: "),
baths=input("Enter number of baths: "))
prompt_init = staticmethod(prompt_init)
This class is pretty straightforward. We've already added the extra **kwargs
parameter to __init__ because we know it's going to be used in a multiple
inheritance situation. We've also included a call to super().__init__ in case we are
not the last call in the multiple inheritance chain. In this case, we're "consuming" the
keyword arguments because we know they won't be needed at other levels of the
inheritance hierarchy.
We see something new in the prompt_init method. This method is made into a
static method immediately after it is initially created. Static methods are associated
only with a class (much like class variables), rather than a specific object instance.
Hence, they have no self argument. Because of this, the super keyword won't work
(there is no parent object, only a parent class), so we simply call the static method
on the parent class directly. This method uses the Python dict constructor to create
a dictionary of values that can be passed into __init__. The value for each key is
prompted with a call to input.
[ 83 ]
The Apartment class extends Property, and is similar in structure:

class Apartment(Property):
valid_laundries = ("coin", "ensuite", "none")
valid_balconies = ("yes", "no", "solarium")
def __init__(self, balcony='', laundry='', **kwargs):
self.balcony = balcony
self.laundry = laundry
def display(self):
super().display()
print("APARTMENT DETAILS")
print("laundry: %s" % self.laundry)
print("has balcony: %s" % self.balcony)
parent_init = Property.prompt_init()
laundry = ''
while laundry.lower() not in \
Apartment.valid_laundries:
laundry = input("What laundry facilities does "
"the property have? ({})".format(
", ".join(Apartment.valid_laundries))
balcony = ''
while balcony.lower() not in \
Apartment.valid_balconies:
balcony = input(
"Does the property have a balcony? "
"({})".format(
", ".join(Apartment.valid_balconies))
parent_init.update({
"laundry": laundry,
"balcony": balcony
})
return parent_init
The display() and __init__() methods call their respective parent class methods
using super() to ensure the Property class is properly initialized.
[ 84 ]
Chapter 3
The prompt_init static method is now getting dictionary values from the parent
class, and then adding some additional values of its own. It calls the dict.update
method to merge the new dictionary values into the first one. However, that
prompt_init method is looking pretty ugly; it twice loops until the user enters a
valid input using structurally similar code but different variables. It would be nice to
extract this validation logic so we can maintain it in only one location; it will likely
also be useful to later classes.
With all the talk on inheritance, we might think this is a good place to use a mixin.
Instead, we have a chance to study a situation where inheritance is not the best
solution. The method we want to create will be used in a static method. If we were
to inherit from a class that provided validation functionality, the functionality would
also have to be provided as a static method that did not access any instance variables
on the class. If it doesn't access any instance variables, what's the point of making it
a class at all? Why don't we just make this validation functionality a module-level
function that accepts an input string and list of valid answers, and leave it at that?
Let's explore what this validation function would look like:
def get_valid_input(input_string, valid_options):
input_string += " ({}) ".format(", ".join(valid_options))
response = input(input_string)
while response.lower() not in valid_options:
response = input(input_string)
return response
We can test this function in the interpreter, independently of all the other classes
we've been working on. This is a good sign, it means different pieces of our design
are not tightly coupled to each other and can later be improved independently,
without affecting other pieces of code.
>>> get_valid_input("what laundry?", ("coin", "ensuite", "none"))
what laundry? (coin, ensuite, none) hi
what laundry? (coin, ensuite, none) COIN
'COIN'
Now, let's quickly update our Apartment.prompt_init method to use this new
function for validation:
def prompt_init():
laundry = get_valid_input(
"What laundry facilities does "
"the property have? ",
Apartment.valid_laundries)
[ 85 ]

balcony = get_valid_input(
"Does the property have a balcony? ",
Apartment.valid_balconies)
"laundry": laundry,
"balcony": balcony
})
return parent_init
That's much easier to read (and maintain!) than our original version. Now we're
ready to build the House class. This class has a parallel structure to Apartment, but
refers to different prompts and variables:
class House(Property):
valid_garage = ("attached", "detached", "none")
valid_fenced = ("yes", "no")
def __init__(self, num_stories='',
garage='', fenced='', **kwargs):
self.garage = garage
self.fenced = fenced
self.num_stories = num_stories
def display(self):
super().display()
print("HOUSE DETAILS")
print("# of stories: {}".format(self.num_stories))
print("garage: {}".format(self.garage))
print("fenced yard: {}".format(self.fenced))
def prompt_init():
fenced = get_valid_input("Is the yard fenced? ",
House.valid_fenced)
garage = get_valid_input("Is there a garage? ",
House.valid_garage)
num_stories = input("How many stories? ")
"fenced": fenced,
"garage": garage,
[ 86 ]
Chapter 3
"num_stories": num_stories
})
return parent_init
There's nothing new to explore here, so let's move on to the Purchase and Rental
classes. In spite of having apparently different purposes, they are also similar in
design to the ones we just discussed:
class Purchase:
def __init__(self, price='', taxes='', **kwargs):
self.price = price
self.taxes = taxes
def display(self):
super().display()
print("PURCHASE DETAILS")
print("selling price: {}".format(self.price))
print("estimated taxes: {}".format(self.taxes))
def prompt_init():
return dict(
price=input("What is the selling price? "),
taxes=input("What are the estimated taxes? "))
class Rental:
def __init__(self, furnished='', utilities='',
rent='', **kwargs):
self.furnished = furnished
self.rent = rent
self.utilities = utilities
def display(self):
super().display()
print("RENTAL DETAILS")
print("rent: {}".format(self.rent))
print("estimated utilities: {}".format(
self.utilities))
print("furnished: {}".format(self.furnished))
[ 87 ]

def prompt_init():
return dict(
rent=input("What is the monthly rent? "),
utilities=input(
"What are the estimated utilities? "),
furnished = get_valid_input(
"Is the property furnished? ",
("yes", "no")))
These two classes don't have a superclass (other than object), but we still call
super().__init__ because they are going to be combined with the other classes,
and we don't know what order the super calls will be made in. The interface
is similar to that used for House and Apartment, which is very useful when we
combine the functionality of these four classes in separate subclasses. For example:
class HouseRental(Rental, House):
def prompt_init():
init = House.prompt_init()
init.update(Rental.prompt_init())
return init
This is slightly surprising, as it has neither an __init__ nor display method!

Because both parent classes appropriately call super in these methods, we only have
to extend those classes and the classes will behave in the correct order. This is not
the case with prompt_init, of course, since it is a static method that does not call
super, so we implement this one explicitly. We should test this class to make sure it
is behaving properly before we write the other three combinations:
>>> init = HouseRental.prompt_init()
Enter the square feet: 1
Enter number of bedrooms: 2
Enter number of baths: 3
Is the yard fenced?
Is there a garage?
(yes, no) no
(attached, detached, none) none
How many stories? 4

What is the monthly rent? 5
What are the estimated utilities? 6
Is the property furnished?
(yes, no) no
>>> house = HouseRental(**init)

>>> house.display()
[ 88 ]
Chapter 3
PROPERTY DETAILS
================
square footage: 1
bedrooms: 2
bathrooms: 3
HOUSE DETAILS
# of stories: 4
garage: none
fenced yard: no
RENTAL DETAILS
rent: 5
estimated utilities: 6
furnished: no
It looks like it is working fine. The prompt_init method is prompting for

initializers to all the super classes, and display() is also cooperatively calling
all three superclasses.
Note: The order of the inherited classes in the preceding example is
important. If we had written class HouseRental(House, Rental)
instead of class HouseRental(Rental, House), display() would
not have called Rental.display()! When display is called on our
version of HouseRental, it refers to the Rental version of the method,
which calls super.display() to get the House version, which again
calls super.display() to get the property version. If we reverse it,
display would refer to the House class's display(). When super is
called, it calls the method on the Property parent class. But Property
does not have a call to super in its display method. This means
Rental class's display method would not be called! By placing the
inheritance list in the order we did, we ensure that Rental calls super,
which then takes care of the House side of the hierarchy. You might think
we could have added a super call to Property.display(), but that
will fail because the next superclass of Property is object, and object
does not have a display method. Another way to fix this is to allow
Rental and Purchase to extend the Property class instead of deriving
directly from object. (Or we could modify the method resolution order
dynamically, but that is beyond the scope of this book.)
[ 89 ]
Now that we have tested it, we are prepared to create the rest of our
combined subclasses:
class ApartmentRental(Rental, Apartment):
def prompt_init():
init = Apartment.prompt_init()
init.update(Rental.prompt_init())
return init
class ApartmentPurchase(Purchase, Apartment):
def prompt_init():
init = Apartment.prompt_init()
init.update(Purchase.prompt_init())
return init
class HousePurchase(Purchase, House):
def prompt_init():
init = House.prompt_init()
init.update(Purchase.prompt_init())
return init
That should be the most intense designing out of our way! Now all we have to do
is create the Agent, which is responsible for creating new listings and displaying
existing ones. Let's start with the simpler storing and listing of properties:
class Agent:
def __init__(self):
self.property_list = []
def display_properties(self):
for property in self.property_list:
property.display()
Adding a property will require first querying the type of property and whether
property is for purchase or rental. We can do this by displaying a simple menu. Once
this has been determined, we can extract the correct subclass and prompt for all the
details using the prompt_init hierarchy we've already developed. Sound simple? It
is. Let's start by adding a dictionary class variable to the Agent class:
type_map = {
("house", "rental"): HouseRental,
("house", "purchase"): HousePurchase,
("apartment", "rental"): ApartmentRental,
("apartment", "purchase"): ApartmentPurchase
}
[ 90 ]
Chapter 3
That's some pretty funny looking code. This is a dictionary, where the keys are tuples
of two distinct strings, and the values are class objects. Class objects? Yes, classes
can be passed around, renamed, and stored in containers just like "normal" objects
or primitive data types. With this simple dictionary, we can simply hijack our earlier
get_valid_input method to ensure we get correct dictionary keys and look up the
appropriate class, like so:
def add_property(self):
property_type = get_valid_input(
"What type of property? ",
("house", "apartment")).lower()
payment_type = get_valid_input(
"What payment type? ",
("purchase", "rental")).lower()
PropertyClass = self.type_map[
(property_type, payment_type)]
init_args = PropertyClass.prompt_init()
self.property_list.append(PropertyClass(**init_args))
This may look a bit funny, too! We look up the class in the dictionary and store it
in a variable named PropertyClass. We don't know exactly which class we have
available, but the class knows itself, so we can polymorphically call prompt_init to
get a dictionary of values appropriate to pass into the constructor. Then we use the
keyword argument syntax to convert the dictionary into arguments and construct
the new object to load the correct data.
Now our user can use this Agent class to add and view lists of properties. It wouldn't
take much work to add features to mark a property as available or unavailable or to
edit and remove properties. Our prototype is now in a good enough state to take to
a real estate agent and demonstrate its functionality. Here's how a demo session
might work:
>>> agent = Agent()
>>> agent.add_property()
What type of property?
What payment type?
(house, apartment) house
(purchase, rental) rental

Enter number of baths: one and a half
Is the yard fenced?
Is there a garage?
(yes, no) yes

(attached, detached, none) detached
How many stories? 1

[ 91 ]

What is the monthly rent? 1200
What are the estimated utilities? included
Is the property furnished?
(yes, no) no
>>> agent.add_property()
What type of property?
What payment type?
(house, apartment) apartment
(purchase, rental) purchase

Enter number of baths: 2
What laundry facilities does the property have?
(coin, ensuite,
one) ensuite
Does the property have a balcony? (yes, no, solarium) yes
What is the selling price? $200,000
What are the estimated taxes? 1500
>>> agent.display_properties()
PROPERTY DETAILS
================
square footage: 900
bedrooms: 2
bathrooms: one and a half
HOUSE DETAILS
# of stories: 1
garage: detached
fenced yard: yes
RENTAL DETAILS
rent: 1200
estimated utilities: included
furnished: no
PROPERTY DETAILS
================
square footage: 800
bedrooms: 3
bathrooms: 2
APARTMENT DETAILS
[ 92 ]
Chapter 3
laundry: ensuite
has balcony: yes
PURCHASE DETAILS
selling price: $200,000
estimated taxes: 1500
>>>
Exercises
Look around you at some of the physical objects in your workspace and see if you
can describe them in an inheritance hierarchy. Humans have been dividing the
world into taxonomies like this for centuries, so it shouldn't be difficult. Are there
any non-obvious inheritance relationships between classes of objects? If you were
to model these objects in a computer application, what properties and methods
would they share? Which ones would have to be polymorphically overridden?
What properties would be completely different between them?
Now, write some code. No, not for the physical hierarchy; that's boring. Physical
items have more properties than methods. Just think about a pet programming
project you've wanted to tackle in the past year, but never got around to. For
whatever problem you want to solve, try to think of some basic inheritance
relationships. Then implement them. Make sure you pay attention to the sorts of
relationships that you actually don't need to use inheritance for, too! Are there any
places where you might want to use multiple inheritance? Are you sure? Can you see
any place you would want to use a mixin? Try to knock together a quick prototype.
It doesn't have to be useful or even partially working. You've seen how you can test
code using python3 -i already; just write some code and test it in the interactive
interpreter. If it works, write some more. If it doesn't, fix it!
Now take a look at the real estate example. This turned out to be a quite effective use
of multiple inheritance. I have to admit though, I had my doubts when I started the
design. Have a look at the original problem and see if you can come up with another
design to solve it that uses only single inheritance. What about a design that doesn't
use inheritance at all? Of the three, which do you think is the most elegant solution?
Elegance is a primary goal in Python development, but each programmer has a
different opinion as to what is the most elegant solution. Some people tend to think
and understand problems most clearly using composition, while others find multiple
inheritance to be the most useful tool.
[ 93 ]
Finally, try adding some new features to the three designs. Whatever features strike
your fancy are fine. I'd like to see a way to differentiate between available and
unavailable properties, for starters. It's not much use to me if it's already rented!
Which design is easiest to extend? Which is hardest? If somebody asked you why
you thought that, would you be able to explain yourself?
Summary
We've gone from simple inheritance, one of the most useful tools in the
object-oriented programmer's toolbox, all the way through to multiple
inheritance, one of the most complicated. We learned how to:
Add functionality to existing classes and built-ins using inheritance
Share similar code between classes by abstracting it into a parent class
Combine multiple threads of functionality using multiple inheritance
Call parent methods using super
Format argument lists in multiple inheritance so super doesn't choke
In the next chapter, we'll cover the subtle art of handling exceptional circumstances.
[ 94 ]
Expecting the Unexpected

Programs are very fragile. It would be nice if code always returned a valid result, but
sometimes a valid result can't be calculated. It's not possible to divide by zero, or to
access the eighth item in a five-item list, for example.
In the old days, the only way around this was to rigorously check the inputs for
every function to make sure they made sense. Typically functions had special return
values to indicate an error condition; for example, they could return a negative
number to indicate that a positive value couldn't be calculated. Different numbers
might mean different errors occurred. Any code that called this function would have
to explicitly check for an error condition and act accordingly. A lot of code didn't
bother to do this, and programs simply crashed.
Not so in the object-oriented world! In this chapter we will study exceptions, special
error objects that only need to be handled when it makes sense to handle them. In
particular, we will cover:
How to cause an exception to occur
How to recover when an exception has occurred
How to handle different exceptions with different code
Cleaning up when an exception has occurred
Creating new exceptions
Using the exception syntax for flow control
Raising exceptions
So what is an exception, really? Technically, an exception is just an object. There are

many different exception classes available and we can easily define more of our own.
The one thing they all have in common is that they derive from a built-in class called
BaseException.
These exception objects become special when they are handled inside the program's
flow of control. When an exception occurs, everything that was supposed to happen
doesn't happen, unless it was supposed to happen when an exception occurred.
Make sense? Don't worry, it will!
So then, how do we cause an exception to occur? The easiest way is to do something
stupid! Chances are you've done this already and seen the exception output. For
example, any time Python encounters a line in your program that it can't understand,
it bails with a SyntaxError, which is a type of exception. Here's a common one:
>>> print "hello world"
File "<stdin>", line 1
print "hello world"
^
SyntaxError: invalid syntax
That print statement was a valid command in Python 2 and previous versions,
but in Python 3, because print is a function, we have to enclose the arguments
in parenthesis. So if we type the above into a Python 3 interpreter, we get the
SyntaxError exception.
A SyntaxError, while common, is actually a special exception, because we can't
handle it. It tells us that we typed something wrong and we better figure out what
it is. Some other common exceptions, which we can handle, are shown in the
following example:
>>> x = 5 / 0
ZeroDivisionError: int division or modulo by zero
>>> lst = [1,2,3]
>>> print(lst[3])
IndexError: list index out of range
>>> lst + 2
[ 96 ]
Chapter 4
TypeError: can only concatenate list (not "int") to list
>>> lst.add
AttributeError: 'list' object has no attribute 'add'
>>> d = {'a': 'hello'}
>>> d['b']
KeyError: 'b'
>>> print(this_is_not_a_var)
NameError: name 'this_is_not_a_var' is not defined
>>>
Sometimes these exceptions are indicators of something wrong in our program

(in which case we would go to the indicated line number and fix it), but they also
occur in legitimate situations. A ZeroDivisionError doesn't always mean we
received invalid input, just different input. The user may have entered a zero by
mistake, or on purpose, or it may represent a legitimate value such as an empty
bank account or the age of a newborn child.
You may have noticed all the above built-in exceptions end in the name Error. In
Python, the words "error" and "exception" are used almost interchangeably. Errors
are sometimes considered more dire than exceptions, but they are dealt with in
exactly the same way. Indeed, all the error classes above have Exception (which
extends BaseException) as their superclass.
[ 97 ]
Raising an exception
Now, then, what do we do if we're writing a program that needs to inform the
user or a calling function that the inputs are somehow invalid? It would be nice
if we could use the same mechanism that Python uses and we can! Want to
see how? Here's a simple class that adds items to a list only if they are even
numbered integers:
class EvenOnly(list):
def append(self, integer):
if not isinstance(integer, int):
raise TypeError("Only integers can be added")
if integer % 2:
raise ValueError("Only even numbers can be added")
super().append(integer)
This class extends the list built-in, as we discussed in Chapter 2, and overrides
the append method to check two conditions that ensure the item is an even integer.
We first check if the input is an instance of the int type, and then use the modulus
operator to ensure it is divisible by two. If either of the two conditions is not met, the
raise keyword is used to cause an exception to occur. The raise keyword is simply
followed by the object being raised as an exception. In the example above, two
objects are newly constructed from the built-in classes TypeError and ValueError.
The raised object could just as easily be an instance of a new exception class we
create ourselves (we'll see how shortly), an exception that was defined elsewhere,
or even an exception object that has been previously raised and handled.
If we test this class in the Python interpreter, we can see that it is outputting useful
error information when exceptions occur, just as before:
>>> e = EvenOnly()
>>> e.append("a string")
File "even_integers.py", line 7, in add
raise TypeError("Only integers can be added")
TypeError: Only integers can be added
>>> e.append(3)
File "even_integers.py", line 9, in add
[ 98 ]
Chapter 4
raise ValueError("Only even numbers can be added")
ValueError: Only even numbers can be added
>>> e.append(2)
Note: While this class is effective for demonstrating exceptions in action,

it isn't very good at its job. It is still possible to get other values into
the list using index notation or slice notation. This can all be avoided
by overriding other appropriate methods, some of which are
double-underscore methods.
What happens when an exception occurs?
When an exception is raised, it appears to stop program execution immediately. Any

lines that were supposed to happen after the exception are not executed, and, unless
the exception is dealt with, the program will exit with an error message. Take a look
at this simple function:
def no_return():
print("I am about to raise an exception")
raise Exception("This is always raised")
print("This line will never execute")
return "I won't be returned"
If we execute this function, we see that the first print call is executed and then the
exception is raised. The second print statement is never executed, and the return
statement never executes either:
>>> no_return()
I am about to raise an exception
File "exception_quits.py", line 3, in no_return
Exception: This is always raised
[ 99 ]
Further, if we have a function that calls a second function that raises an exception,
nothing will be executed in the first function after the point where the second
function was called. Raising an exception stops all execution right up the function
call stack until it is either handled, or forces the interpreter to exit. To demonstrate,
let's add a second function that calls our first one:
def call_exceptor():
print("call_exceptor starts here...")
no_return()
print("an exception was raised...")
print("...so these lines don't run")
When we call this function, we see that the first print statement executes as well as
the first line in the no_return function. But once the exception is raised, nothing
else executes:
>>> call_exceptor()
call_exceptor starts here...
File "method_calls_excepting.py", line 9, in call_exceptor
no_return()
File "method_calls_excepting.py", line 3, in no_return
Exception: This is always raised
We'll soon see that the interpreter is not actually taking a shortcut and exiting
immediately; the exception can be handled inside either method. Indeed, exceptions
can be handled at any level after they are initially raised. If we look at the exception's
output (called a traceback) from bottom to top, we see both methods listed. Inside
no_return, the exception is initially raised. Then just above that, we see that inside
call_exceptor, that pesky no_return function was called and the exception
"bubbled up" to the calling method. From there it went up one more level to
the main interpreter, which finally printed the traceback.
[ 100 ]
Chapter 4
Handling exceptions
Now let's look at the tail side of the exception coin. Namely, if we encounter an
exception situation, how should our code react to, or recover from it? We handle
exceptions by wrapping any code that might throw one (whether it is exception code
itself, or a call to any function or method that may have an exception raised inside it)
inside a try...except clause. The most basic syntax looks like this:
try:
no_return()
except:
print("I caught an exception")
print("executed after the exception")
If we run this simple script using our existing no_return function, which we know,
very well, always throws an exception, we get this output:
I caught an exception
executed after the exception
The no_return function happily informs us that it is about to raise an exception.

But we fooled it and caught the exception. Once caught, we were able to clean up
after ourselves (in this case, by outputting that we were handling the situation),
and continue on our way, with no interference from the offending function. The
remainder of the code in the no_return function still went unexecuted, but the
code that called the function was able to recover and continue.
Note the indentation around try and except. The try clause wraps any code that
might throw an exception. The except clause is then back on the same indentation
level as the try line. Any code to handle the exception is indented after the except
clause. Then normal code resumes at the original indentation level.
The problem with the above code is that it will catch any type of exception.
What if we were writing some code that could raise both a TypeError and a
ZeroDivisionError? We might want to catch the ZeroDivisionError, but let the
TypeError propagate to the console. Can you guess the syntax? Here's a rather silly
function that does just that:
def funny_division(anumber):
try:
return 100 / anumber
except ZeroDivisionError:
return "Silly wabbit, you can't divide by zero!"
[ 101 ]

print(funny_division(0))
print(funny_division(50.0))
print(funny_division("hello"))
The function is tested with print statements that show it behaving as expected:
Silly wabbit, you can't divide by zero!
2.0
File "catch_specific_exception.py", line 9, in <module>
print(funny_division("hello"))
File "catch_specific_exception.py", line 3, in funny_division
TypeError: unsupported operand type(s) for /: 'int' and 'str'
The first line of output shows that if we enter 0, we get properly mocked. If we
call with a valid number (note that it's not an integer, but it's still a valid divisor),
it operates correctly. Yet if we enter a string (you were wondering how to get a
TypeError, weren't you?), it fails with an exception. If we had used an empty
except clause, that didn't specify a ZeroDivisionError, it would have accused us
of dividing by zero when we sent it a string, which is not proper behavior at all.
We can even catch two or more different exceptions and handle them with the same
code. Here's an example that raises three different types of exceptions. It handles
TypeError and ZeroDivisionError with the same exception handler, but it may
also raise a ValueError if you supply the number 13:
def funny_division2(anumber):
try:
if anumber == 13:
raise ValueError("13 is an unlucky number")
except (ZeroDivisionError, TypeError):
return "Enter a number other than zero"
for val in (0, "hello", 50.0, 13):
print("Testing {}:".format(val), end=" ")
print(funny_division2(val))
[ 102 ]
Chapter 4
The for loop at the bottom simply loops over several test inputs and prints the
results. If you're wondering about that end argument in the print statement, it simply
turns the default trailing newline into a space so it's joined with the output from the
next line. Here's a run of the script :
Testing 0: Enter a number other than zero
Testing hello: Enter a number other than zero
Testing 50.0: 2.0
Testing 13: Traceback (most recent call last):
File "catch_multiple_exceptions.py", line 11, in <module>
print(funny_division2(val))
File "catch_multiple_exceptions.py", line 4, in funny_division2
ValueError: 13 is an unlucky number
The number 0 and the string are both caught by the except clause and a suitable
error message is printed. The exception from the number 13 is not caught, because it
is a ValueError, which was not included in the types of exceptions being handled.
This is all well and good, but what if we want to catch different exceptions and do
different things with them? Or maybe we want to do something with an exception
and then allow it to continue to bubble up to the parent function, as if it had never
been caught? We don't need any new syntax to deal with these cases. It's possible
to stack except clauses, and only one will be executed. For the second question, the
raise keyword, with no arguments, will re-raise the last exception if we're already
inside an exception handler, observe:
def funny_division3(anumber):
try:
if anumber == 13:
return "Enter a number other than zero"
except TypeError:
return "Enter a numerical value"
except ValueError:
print("No, No, not 13!")
raise raise
[ 103 ]
The last line re-raises the ValueError, so after outputting No, No, not 13!, it will
raise the exception again; we'll still get the stack trace on the console.
If we stack exception clauses like we did above, only the first matching clause will
be run, even if more than one of them fit. How can more than one clause match?
Remember that exceptions are objects, and inheritance can take place. As we'll see in
the next section, most exceptions extend the Exception class (which is, itself, derived
from BaseException). If we catch Exception before we catch TypeError, then only
the Exception handler will be executed, because TypeError is an Exception,
by inheritance.
This can come in handy in cases where we want to handle some exceptions
specifically, and then handle all remaining exceptions in a more general case. We
can simply catch Exception after catching any specific exceptions and handle the
general case there.
Sometimes when we catch an exception, we need a reference to the Exception
object itself. This most often happens if we define our own exceptions with custom
arguments, but can also be relevant with standard exceptions. Most exception classes
accept a set of arguments in their constructor, and we might want to access those
attributes in the exception handler. If we define our own exception class, we can even
call custom methods on it when we catch it. The syntax for capturing an exception as
a variable uses the as keyword:
try:
raise ValueError("This is an argument")
except ValueError as e:
print("The exception arguments were", e.args)
If we run this simple snippet, it prints out the string argument that we passed into
ValueError upon initialization.
In Python 2.5 and earlier, the as keyword did not work for naming an
exception. Instead a comma was used, so the previous example would
look like except ValueError, e: This was changed in Python 3.0 to
avoid confusion with the syntax for catching multiple types of exceptions
(as in except ValueError, TypeError), and the change was
backported to Python 2.6. The as keyword is therefore supported in most
modern versions, but if you're working with an older interpreter, keep
this syntax difference in mind, as it is bound to catch you.
[ 104 ]
Chapter 4
We've seen several variations on the syntax for handling exceptions, but we still
don't know how to execute code regardless of whether or not an exception has
occurred. We also can't specify code that should be executed only if an exception
does not occur. Two more keywords, finally and else, can provide the missing
pieces. Neither one takes any extra arguments. The following example randomly
picks an exception to throw and raises it. Then some not-so-complicated exception
handling code is run that illustrates the newly introduced syntax:
import random
some_exceptions = [ValueError, TypeError, IndexError, None]
try:
choice = random.choice(some_exceptions)
print("raising {}".format(choice))
if choice:
raise choice("An error")
except ValueError:
print("Caught a ValueError")
except TypeError:
print("Caught a TypeError")
except Exception as e:
print("Caught some other error: %s" %
( e.__class__.__name__))
else:
print("This code called if there is no exception")
finally:
print("This cleanup code is always called")
If we run this examplewhich illustrates almost every conceivable exception

handling scenarioa few times, we'll get different output each time, depending
on which exception random chooses. Here are some example runs:
$ python finally_and_else.py
raising None
This code called if there is no exception
This cleanup code is always called
raising <class 'TypeError'>
Caught a TypeError
[ 105 ]

raising <class 'IndexError'>
Caught some other error: IndexError
raising <class 'ValueError'>
Caught a ValueError
Note how the print statement in the finally clause is executed no matter what.
This is extremely useful, for example, when we need to clean up an open database
connection, close an open file, or send a closing handshake over the network when
our code has finished running, even if an exception has occurred. This can also be
applied in interesting ways if we are inside a try clause when we return from a
function; the finally handle will still be executed upon return.
Also pay attention to the output when no exception is raised; both the else and the
finally clauses are executed. The else clause may seem redundant, as it appears
that code that should be executed only when no exception is raised can just be placed
after the entire try...except block. However, this code would still be executed in
the case where an exception is caught and handled. We'll see more on this when
we discuss using exceptions as flow control, shortly.
Any of the except, else, and finally clauses can be omitted after a try block
(although, else, by itself is invalid). If you include more than one, the except
clauses must come first, then the else clause, with the finally clause at the end.
The order of the except clauses normally goes from most specific to most generic.
Exception hierarchy
We've already encountered many of the most common built-in exceptions, and you'll
probably encounter the rest over the course of your regular Python development.
As we noticed above, most exceptions are subclasses of the Exception class. But
this is not true of all exceptions. Exception itself actually inherits from a class
called BaseException (In fact, all exceptions must extend the BaseException
class or one of its subclasses). There are two key exceptions, SystemExit and
KeyboardInterrupt, that derive directly from BaseException instead
of Exception.
[ 106 ]
Chapter 4
SystemExit is an exception that is raised whenever the program exits naturally,

typically because we called the sys.exit function somewhere in our code (for
example, because the user selected an exit menu item, clicked the "close" button on
a window, or entered a command to shut down a server). The exception is designed
to allow us to clean up code before the program ultimately exits, so we generally
don't need to handle it explicitly (because cleanup code happens inside a finally
clause, right?). If we do handle it, we would normally re-raise the exception, since
catching it would stop the program from exiting. There are, of course, situations
where we might want to stop the program exiting, for example if there are unsaved
changes and we want to prompt the user if they really want to exit. Usually, if we
handle SystemExit at all, it's because we want to do something special with it or
are anticipating it directly. We especially don't want it to be accidentally caught in
generic clauses that catch all normal exceptions. This is why it derives directly from
BaseException.
The KeyboardInterrupt exception is common in command-line programs.
It is thrown when the user explicitly interrupts program execution with an
OS-dependent key combination (normally Ctrl + C). This is a standard way for the
user to deliberately interrupt a running program, and like SystemExit, it should
almost always respond by terminating the program. Also like SystemExit, it
should handle any cleanup tasks inside finally blocks.
Here is a class diagram that fully illustrates the exception hierarchy:
When we use the except: clause without specifying any type of exception, it will
catch all subclasses of BaseException, which is to say, it will catch all exceptions,
including the two "special" ones. Generally, we don't want to catch them, so, it is
unwise to use the except: statement without any arguments. If you want to catch
all exceptions other than the two we just discussed, explicitly catch Exception. As
we discussed earlier, make sure this is the last except clause if you're trying to catch
other more specific exceptions separately.
[ 107 ]
Further, if you do want to catch all exceptions, I suggest using the syntax except
BaseException: instead of a raw except: This helps explicitly tell the readers of
your code that you really meant to handle the special case exceptions.
Defining our own exceptions
Often, when we want to raise an exception, we find that none of the built-in
exceptions are quite what we need. Luckily, it's very easy to define new
exceptions of our own. The name of the class is usually designed to communicate
what went wrong, and we can provide arbitrary arguments in the initializer to
add additional information.
All we have to do is inherit from the Exception class. We don't even have to add any
content to the class! We can, of course, extend BaseException directly, but then it
will not be caught by generic except Exception clauses.
Without further ado, here's a simple exception we might use in a banking application:
class InvalidWithdrawal(Exception):
pass
raise InvalidWithdrawal("You don't have $50 in your account")
The last line illustrates how to raise the newly defined exception. We are able to pass
an arbitrary number of arguments (often a string message, but any useful object can
be stored) into the exception. The Exception __init__ method is designed to accept
any arguments and store them as a tuple in an attribute named args. This makes
exceptions easier to define without needing to override __init__.
Of course, if we do want to customize the initializer, we are free to do so. Here's an
exception whose initializer demands the current balance and the amount the user
wanted to withdraw. In addition, it adds a method to calculate how much over the
withdrawal was:
class InvalidWithdrawal(Exception):
def __init__(self, balance, amount):
super().__init__("account doesn't have ${}".format(
amount))
self.amount = amount
self.balance = balance
def overage(self):
return self.amount - self.balance
raise InvalidWithdrawal(25, 50)
[ 108 ]
Chapter 4
The raise statement at the end illustrates how to construct the exception. As you can
see, we can do anything with an exception that we would do with other objects. We
could catch an exception and pass it around as a working object, although it is more
common to include a reference to the working object as an attribute on an exception
and pass that around instead.
Here's how we would handle an InvalidWithdrawal if one was raised:
try:
raise InvalidWithdrawal(25, 50)
except InvalidWithdrawal as e:
print("I'm sorry, but your withdrawal is "
"more than your balance by "
"${}".format(e.overage()))
Here we see a valid use of the as keyword. By convention, most Python coders name
the exception variable e, although, as usual, you are free to call it ex, exception, or
aunt_sally if you prefer.
Now that we have full control over exception definitions, including initializer,
attributes, and methods, and we can access an exception instance when it is being
handled, we have absolute power over the information passed with an exception.
There are many reasons for defining our own exceptions. It is often useful to
add information to the exception or log it in some way. But the utility of custom
exceptions truly comes to light when creating a framework, library, or API that is
intended for access by other users. In that case, be careful to ensure your code is
raising exceptions that make sense to the client programmer, are easy to handle,
and clearly describe what went wrong so they can fix it (if it is a bug in their code)
or handle it (if it's a situation they need to be made aware of).
Exceptions aren't exceptional
Novice programmers tend to think of exceptions as only useful for "exceptional

circumstances". However, the definition of "exceptional circumstances" can be
vague and subject to interpretation. Consider the following two functions:
def divide_with_exception(number, divisor):
try:
print("{} / {} = {}".format(
number, divisor, number / divisor * 1.0))
print("You can't divide by zero")
[ 109 ]

def divide_with_if(number, divisor):
if divisor == 0:
print("You can't divide by zero")
else:
print("{} / {} = {}".format(
number, divisor, number / divisor * 1.0))
These two functions behave identically. If divisor is zero, an error message is

printed, otherwise, a message printing the result of division is displayed. Clearly,
we could avoid a ZeroDivisionError ever being thrown by testing for it with an if
statement. Similarly, we can avoid an IndexError by explicitly checking whether or
not the parameter is within the confines of the list, and a KeyError by checking if the
key is in a dictionary.
But we shouldn't do this. For one thing, we might write an if statement that checks
if the index is lower than the parameters of the list, but forget to check negative
values (remember, Python lists support negative indexing; -1 refers to the last
element in the list). Eventually we would discover this and have to find all the
places where we were checking code. But if we'd simply caught the IndexError
and handled it, our code would just work.
In short, use exceptions for exceptional circumstances, even if those circumstances
are only a little bit exceptional.
Turning this argument around, we can see that exception syntax is also effective
for flow control. Like an if statement, exceptions can be used for decision making,
branching, and message passing.
Imagine an inventory application for a company that sells widgets and gadgets.
When a customer makes a purchase, the item can either be available, in which case
the item is removed from inventory and the number of items left is returned, or it
might be out of stock. Now, being out of stock is a perfectly normal thing to happen
in an inventory application. It is certainly not an exceptional circumstance. But what
do we return if it's out of stock? A string saying "out of stock"? A negative number?
In both cases, the calling method would have to check whether the return value is a
positive integer, or something else to determine if it is out of stock. That seems a bit
messy. Instead, we can raise an OutOfStockException and use the try statement to
direct program flow control. Make sense?
[ 110 ]
Chapter 4
In addition, we want to make sure we don't sell the same item to two different
customers, or sell an item that isn't in stock yet. One way to facilitate this is to lock
each type of item to ensure only one person can update it at a time. The user must
lock the item, manipulate the item (purchase, add stock, count items left), and
then unlock the item. Here's a non-functional Inventory object with docstrings that
describes what some of the methods should do:
class Inventory:
def lock(self, item_type):
'''Select the type of item that is going to
be manipulated. This method will lock the
item so nobody else can manipulate the
inventory until it's returned. This prevents
selling the same item to two different
customers.'''
pass
def unlock(self, item_type):
'''Release the given type so that other
customers can access it.'''
pass
def purchase(self, item_type):
'''If the item is not locked, raise an
exception. If the itemtype does not exist,
raise an exception. If the item is currently
out of stock, raise an exception. If the item
is available, subtract one item and return
the number of items left.'''
pass
We could hand this object prototype to a developer and have them implement the
methods to do exactly as they say while we work on the code that needs to make
a purchase. We'll use Python's robust exception handling to consider different
branches depending on how the purchase was made:
item_type = 'widget'
inv = Inventory()
inv.lock(item_type)
try:
num_left = inv.purchase(item_type)
except InvalidItemType:
print("Sorry, we don't sell {}".format(item_type))
[ 111 ]

except OutOfStock:
print("Sorry, that item is out of stock.")
else:
print("Purchase complete. There are "
"{} {}s left".format(num_left, item_type))
finally:
inv.unlock()
Pay attention to how all the possible exception handling clauses are used to ensure
the correct actions happen at the correct time. Even though OutOfStock is not
a terribly exceptional circumstance, we are able to use an exception to handle it
suitably. This same code could be written with an if...elif...else structure, but it
wouldn't be as easy to read or maintain.
We can also use exceptions to pass messages between different methods. For
example, if we wanted to inform the customer as to what date the item is expected to
be in stock again, we could ensure our OutOfStock object requires a back_in_stock
parameter when it is constructed. Then when we handle the exception we can check
that value and provide additional information to the customer. The information
attached to the object can be easily passed between two different parts of the
program. The exception could even provide a method that instructs the inventory
object to reorder or backorder an item.
Using exceptions for flow control can make for some handy program designs. The
important thing to bring from this discussion is that exceptions are not a bad thing
that we should try to avoid. Having an exception occur does not say, "you should
have prevented this exceptional circumstance from happening". Rather, it is just a
powerful way to communicate information between two sections of code that may
not be directly calling each other.
Case study
We've been looking at the use and handling of exceptions at a fairly low level of
detail; syntax and definitions. This case study will help tie it all in with our previous
chapters so we can see how exceptions should be used in the larger context of
objects, inheritance, and modules.
Today, we'll be designing a simple, central authentication and authorization system.
The entire system will be placed in one module, and other code will be able to query
that module object for authentication and authorization purposes. We should admit,
from the start, that we aren't security experts, and that the system we are designing
may be full of security holes. It will be sufficient, however, for a basic login and
permission system that other code can interact with. Later, if that other code needs
to be made more secure, we can have a security or cryptography expert review or
rewrite our module, without changing the API.
[ 112 ]
Chapter 4
Authentication is the process of ensuring a user is really the person they say they
are. We'll follow the lead of common web systems today, which use a username
and private password combination. Other methods of authentication include voice
recognition, fingerprint or retinal scanners, and identification cards.
Authorization, on the other hand, is all about determining whether a given
(authenticated) user is permitted to perform a specific action. We'll create a basic
permission list system that stores a list of the specific people allowed to perform
each action.
In addition, we'll add some administrative features to allow new users to be added
to the system. For brevity, we'll leave out editing of passwords or changing of
permissions once they've been added, but these (highly necessary) features can
certainly be added in the future.
There's a simple analysis; now let's proceed with design. We're obviously going to
need a User class that stores the username and an encrypted password. This class
will also allow a user to log in by checking whether an entered password is valid.
We probably won't need a Permission class, as those are just strings mapped to
a list of users using a dictionary. We should have a central Authenticator class
that handles user management and logging in or out. The last piece of the puzzle
is an Authorizor class that deals with permissions and checking whether a user
can perform an activity. We'll provide a single instance of each of these classes in
the auth module so that other modules can use this central mechanism for all their
authentication and authorization needs. Of course, if they want to instantiate private
instances of these classes, for non-central authorization activities, they are free to
do so.
We'll also be defining several exceptions as we go along. We'll start with a special
AuthException base class that accepts a username and optional user object as
parameters; most of our self-defined exceptions will inherit from this one.
Let's build the User class first; it seems simple enough. A new user can be initialized
with a username and password. The password will be stored encrypted to reduce
the chances of its being stolen. Our purpose is to study exceptions, not to secure a
system. You have been warned! We'll also need a check_password method to test if
a supplied password is the correct one. Here is the class in full:
import hashlib
class User:
def __init__(self, username, password):
'''Create a new user object. The password
will be encrypted before storing.'''
self.username = username
[ 113 ]

self.password = self._encrypt_pw(password)
self.is_logged_in = False
def _encrypt_pw(self, password):
'''Encrypt the password with the username and return
the sha digest.'''
hash_string = (self.username + password)
hash_string = hash_string.encode("utf8")
return hashlib.sha256(hash_string).hexdigest()
def check_password(self, password):
'''Return True if the password is valid for this
user, false otherwise.'''
encrypted = self._encrypt_pw(password)
return encrypted == self.password
Since the code for encrypting a password is required in both __init__ and
check_password, we pull it out to its own method. That way, it only needs to be
changed in one place if someone realizes it is insecure and needs improvement. This
class could easily be extended to include mandatory or optional personal details such
as names, contact information, and birth dates.
Before we write code to add users (which will happen in the as-yet undefined
Authenticator class), we should examine some use cases. If all goes well, we can
add a user with a username and password; the User object is created and inserted
into a dictionary. But in what ways can all not go well? Well, clearly, we don't want
to add a user if that username already exists in the dictionary. Otherwise we'd
overwrite an existing user's data and the new user might have access to that user's
privileges. So we'll need a UsernameAlreadyExists exception. Also, for security's
sake, we should probably raise an exception if the password is too short. Both of
these exceptions will extend AuthException, which we mentioned earlier. So,
before writing the Authenticator, let's define these three exception classes.
class AuthException(Exception):
def __init__(self, username, user=None):
super().__init__(username, user)
self.user = user
class UsernameAlreadyExists(AuthException):
pass
class PasswordTooShort(AuthException):
pass
[ 114 ]
Chapter 4
The AuthException requires a username and has an optional user parameter.

This second parameter should be an instance of the User class associated with
that username. The two specific exceptions we're defining simply need to inform
the calling class of an exceptional circumstance, so we don't need to add any extra
methods to them.
Now, let's start on the Authenticator class. It can simply be a mapping of
usernames to user objects, so we'll start with a dictionary in the initialization
function. The method for adding a user needs to check the two conditions
(password length and previously existing users) before creating a new User
instance and adding it to the dictionary:
class Authenticator:
def __init__(self):
'''Construct an authenticator to manage
users logging in and out.'''
self.users = {}
def add_user(self, username, password):
if username in self.users:
raise UsernameAlreadyExists(username)
if len(password) < 6:
raise PasswordTooShort(username)
self.users[username] = User(username, password)
We could, of course, extend the password validation to raise exceptions for

passwords that are too easy to crack in other ways, if we desired.
Now let's prepare the login method. If we weren't thinking about exceptions just
now, we might just want the method to return True or False, depending on whether
the login was successful or not. But we are thinking about exceptions, and this could
be a good place to use them for a not-so-exceptional circumstance. We could raise
different exceptions, for example, if the username does not exist or the password
does not match. This will allow anyone trying to log a user in to elegantly handle the
situation using a try/except/else clause. So first we add these new exceptions:
class InvalidUsername(AuthException):
pass
class InvalidPassword(AuthException):
pass
[ 115 ]
Then we can define a simple login method to our Authenticator class that raises
these exceptions if necessary. If not, it flags the user as logged in and returns:
def login(self, username, password):
try:
user = self.users[username]
except KeyError:
raise InvalidUsername(username)
if not user.check_password(password):
raise InvalidPassword(username, user)
user.is_logged_in = True
return True
Notice how the KeyError is handled. This could have been handled using if
username not in self.users: instead, but we choose to handle the exception
directly. We end up eating up this first exception and raising a brand new one
of our own that fits the user facing API better.
We can also add a method to check if a particular username is logged in. Deciding
whether to use an exception here is trickier, should we raise an exception if the
username does not exist? Should we raise an exception if the user is not logged in?
To answer these questions, we need to think about how the method would be
accessed. Most often, this method will be used to answer the yes/no question,
"Should I allow them access to <something>?" The answer will either be, "yes, the
username is valid and they are logged in", or, "no, the username is not valid or they
are not logged in". Therefore, a Boolean return value is sufficient. There is no need to
use exceptions here, just for the sake of using an exception.
def is_logged_in(self, username):
if username in self.users:
return self.users[username].is_logged_in
return False
Finally, we can add a default authenticator instance to our module so that client code
can access it easily using auth.authenticator:
authenticator = Authenticator()
[ 116 ]
Chapter 4
This line goes at the module level, outside any class definition, so the authenticator
variable can be accessed as auth.authenticator. Now we can start on the
Authorizor class, which maps permissions to users. The Authorizor should
not permit a user access to a permission if they are not logged in, so they'll need
a reference to a specific authenticator. We'll also need to set up the permission
dictionary upon initialization:
class Authorizor:
def __init__(self, authenticator):
self.authenticator = authenticator
self.permissions = {}
Now we can write methods to add new permissions and to set up which users are
associated with each permission.
def add_permission(self, perm_name):
'''Create a new permission that users
can be added to'''
try:
perm_set = self.permissions[perm_name]
except KeyError:
self.permissions[perm_name] = set()
else:
raise PermissionError("Permission Exists")
def permit_user(self, perm_name, username):
'''Grant the given permission to the user'''
try:
except KeyError:
raise PermissionError("Permission does not exist")
else:
if username not in self.authenticator.users:
raise InvalidUsername(username)
perm_set.add(username)
The first method allows us to create a new permission, unless it already exists, in
which case, an exception is raised. The second allows us to add a username to a
permission, unless either the permission or the username doesn't yet exist.
We use a set instead of a list for usernames, so that even if you grant a user
permission more than once, the nature of sets means the user is only in the set once.
Sets are sequences, like lists, but unlike lists, they are unordered, and store unique
values. No matter how many times we add a value to a set, it will only be stored in
the set once.
[ 117 ]
A PermissionError is raised in both methods. This new error doesn't require

a username, so we'll make it extend Exception directly, instead of our custom
AuthException:
class PermissionError(Exception):
pass
Finally, we can add a method to check whether a user has a specific permission
or not. In order for them to be granted access, they have to be both logged into the
authenticator and in the set of people who have been granted access to that privilege.
If either of these conditions is not met, an exception is raised:
def check_permission(self, perm_name, username):
if not self.authenticator.is_logged_in(username):
raise NotLoggedInError(username)
try:
except KeyError:
else:
if username not in perm_set:
raise NotPermittedError(username)
else:
return True
There are two new exceptions in here; they both take usernames, so we'll define them
as subclasses of AuthException:
class NotLoggedInError(AuthException):
pass
class NotPermittedError(AuthException):
pass
Finally, we can add a "default" authorizor to go with our default authenticator:

authorizor = Authorizor(authenticator)
And that completes a basic, but complete authentication/authorization system. We

can test the system at the Python prompt, checking to see if a user, joe, is permitted
to do tasks in the paint department:
>>> import auth
>>> auth.authenticator.add_user("joe", "joepassword")
>>> auth.authorizor.add_permission("paint")
>>> auth.authorizor.check_permission("paint", "joe")
[ 118 ]
Chapter 4
File "auth.py", line 109, in check_permission
raise NotLoggedInError(username)
auth.NotLoggedInError: joe
>>> auth.authenticator.is_logged_in("joe")
False
>>> auth.authenticator.login("joe", "joepassword")
True
raise NotPermittedError(username)
auth.NotPermittedError: joe
>>> auth.authorizor.check_permission("mix", "joe")
KeyError: 'mix'
During handling of the above exception, another exception occurred:
auth.PermissionError: Permission does not exist
>>> auth.authorizor.permit_user("mix", "joe")
File "auth.py", line 99, in permit_user
KeyError: 'mix'
During handling of the above exception, another exception occurred:
[ 119 ]

File "auth.py", line 101, in permit_user
auth.PermissionError: Permission does not exist
>>> auth.authorizor.permit_user("paint", "joe")
True
The above output shows all of our code and most of our exceptions in action, but to
really understand the API we've defined, we should write some exception handling
code that actually uses it. Here's a basic menu interface that allows certain users to
change or test a program:
import auth
# Set up a test user and permission
auth.authenticator.add_user("joe", "joepassword")
auth.authorizor.add_permission("test program")
auth.authorizor.add_permission("change program")
auth.authorizor.permit_user("test program", "joe")
class Editor:
def __init__(self):
self.username = None
self.menu_map = {
"login": self.login,
"test": self.test,
"change": self.change,
"quit": self.quit
}
def login(self):
logged_in = False
while not logged_in:
username = input("username: ")
password = input("password: ")
try:
logged_in = auth.authenticator.login(
username, password)
except auth.InvalidUsername:
print("Sorry, that username does not exist")
[ 120 ]
Chapter 4
except auth.InvalidPassword:
print("Sorry, incorrect password")
else:
def is_permitted(self, permission):
try:
auth.authorizor.check_permission(
permission, self.username)
except auth.NotLoggedInError as e:
print("{} is not logged in".format(e.username))
return False
except auth.NotPermittedError as e:
print("{} cannot {}".format(
e.username, permission))
return False
else:
return True
def test(self):
if self.is_permitted("test program"):
print("Testing program now...")
def change(self):
if self.is_permitted("change program"):
print("Changing program now...")
def quit(self):
raise SystemExit()
def menu(self):
try:
answer = ""
while True:
print("""
Please enter a command:
\tlogin\tLogin
\ttest\tTest the program
\tchange\tChange the program
\tquit\tQuit
""")
answer = input("enter a command: ").lower()
try:
func = self.menu_map[answer]
[ 121 ]

except KeyError:
print("{} is not a valid option".format(
answer))
else:
func()
finally:
print("Thank you for testing the auth module")
Editor().menu()
This rather long example is actually very simple. The is_permitted method is
probably the most interesting; this is a mostly internal method that is called by both
test and change to ensure the user is permitted access before continuing. Of course,
those two methods are stubs, but we aren't writing an editor here, we're illustrating
the use of exceptions and exception handlers by testing an authentication and
authorization framework!
Exercises
If you've never dealt with exceptions before, the first thing you need to do is look at
old Python code you've written and notice if there are places you should have been
handling exceptions. How would you handle them? Do you need to handle them
at all? Sometimes letting the exception propagate to the console is the best thing to
do. Sometimes you can recover from the error and allow the program to continue.
Sometimes you can only reformat the error into something the user can understand
and display it to them.
Some common places to look are file I/O (is it possible your code will try to read a
file that doesn't exist?), mathematical expressions (is it possible that a value you are
dividing by is zero?) list indices (is the list empty?) and dictionaries (does the key
exist?). Ask yourself if you should ignore the problem, handle it by checking
values first, or handle it with an exception. Pay special attention to areas where
you might have used finally and else to ensure the correct code is executed
under all conditions.
[ 122 ]
Chapter 4
Now write some new code. Think of a program that requires authentication and
authorization, and try writing some code that uses the auth module we built in the
case study. Feel free to modify the module if it's not flexible enough. Try to handle
all the exceptions in a sensible way. If you're having trouble coming up with
something that requires authentication, try adding authorization to the notepad
example from Chapter 2. Or add authorization to the auth module itselfit's not a
terribly useful module if just anybody can start adding permissions! Maybe require
an administrator username and password before allowing privileges to be added
or changed.
Finally, try to think of places in your code where you can raise exceptions. It can
be in code you've written or are working on; or write a new project as an exercise.
You'll probably have the best luck designing a small framework or API that is meant
to be used by other people; exceptions are a terrific communication tool between
your code and someone else's. Remember to design and document any self-raised
exceptions as part of the API, or they won't know if or how to handle them!
Summary
In this chapter, we went into the gritty details of raising, handling, defining, and
manipulating exceptions. Exceptions are a powerful way to communicate unusual
circumstances or error conditions without requiring a calling function to explicitly
check return values. Specifically, we covered:
Built-in exceptions and raising exceptions
Several ways to handle specific exceptions
Defining new exceptions
Using exceptions for unexceptional circumstances
In the next chapter, everything we've studied so far will come together as we discuss
how object-oriented programming principles and structures should best be applied
in Python applications.
[ 123 ]
When to Use Object-oriented

Programming
In the previous chapters, we've covered many of the defining features of
Object-oriented Programming. We now know the principles and paradigms
of Object-oriented Design, and we've covered the syntax of Object-oriented
Programming in Python.
Yet, things are still hazy when it comes to knowing how and when to apply these
principles and syntax. Here, we'll discuss some of the more useful applications of the
knowledge we've gained, while in Chapter 7, we'll cover something just as important:
when not to use them! In this chapter we'll cover:
How to recognize objects
Data and behaviors, once again
Wrapping data in behavior using properties
Restricting data using behavior
The Don't Repeat Yourself principle
Recognizing code that is the same
Treat objects as objects
This may seem obvious, but you should generally give separate objects in your
problem domain a special class in your code. We've seen examples of this in the
case studies in previous chapters; the process is generally to identify objects in the
problem and then model their data and behaviors.
When to Use Object-oriented Programming
Identifying objects is a very important task in object-oriented analysis and

programming. But it isn't always as easy as counting the nouns in a short paragraph,
as we've been doing. Remember, objects are things that have both data and behavior.
If we are working with only data, we are often better off storing it in a list, set,
dictionary, or some other Python data structure (which we'll be covering thoroughly
in the next chapter). On the other hand, if we are working with only behavior, with
no stored data, a simple function is more suitable.
An object, however, has both data and behavior. Most Python programmers use
built-in data structures unless (or until) there is an obvious need to define a class.
This is a good thing; there is no reason to add an extra level of abstraction if it doesn't
help organize our code. Sometimes, though, the "obvious" need is not so obvious.
A Python programmer often starts by storing data in a few variables. As our
program expands, we will later find that we are passing the same set of related
variables to different functions. This is the time to think about grouping both
variables and functions into a class. If we are designing a program to model
polygons in two-dimensional space, we might start with each polygon being
represented as a list of points. The points would be modeled as two-tuples (x,y)
describing where that point is located. This is all data, stored in two nested data
structures (specifically, a list of tuples):
square = [(1,1), (1,2), (2,2), (2,1)]
Now, if we want to calculate the distance around the perimeter of the polygon, we
simply need to sum the distances between the two points, but to do that, we need a
function to calculate the distance between two points. Here are two such functions:
import math
def distance(p1, p2):
return math.sqrt((p1[0]-p2[0])**2 + (p1[1]-p2[1])**2)
def perimeter(polygon):
perimeter = 0
points = polygon + [polygon[0]]
for i in range(len(polygon)):
perimeter += distance(points[i], points[i+1])
return perimeter
Now, as object-oriented programmers, we clearly recognize that a polygon class

could encapsulate the list of points (data) and the perimeter function (behavior).
Further, a point class, such as we defined in Chapter 2 might encapsulate the x and y
coordinates and the distance method. But should we do this?
[ 126 ]
Chapter 5
For the previous code, maybe, maybe not. We've been studying object-oriented
principles long enough that we can now write the object-oriented version in
record time:
import math
class Point:
def __init__(self, x, y):
self.x = x
self.y = y
def distance(self, p2):
return math.sqrt((self.x-p2.x)**2 + (self.y-p2.y)**2)
class Polygon:
def __init__(self):
self.vertices = []
def add_point(self, point):
self.vertices.append((point))
def perimeter(self):
perimeter = 0
points = self.vertices + [self.vertices[0]]
for i in range(len(self.vertices)):
perimeter += points[i].distance(points[i+1])
return perimeter
As you can see from the highlighted sections, there is twice as much code here as
there was in our earlier version, although we can argue that the add_point method
is not strictly necessary.
Now, to understand the difference a little better, let's compare the two APIs in use.
Here's how to calculate the perimeter of a square using the object-oriented code:
>>> square = Polygon()
>>> square.add_point(Point(1,1))
>>> square.perimeter()
4.0
[ 127 ]
That's fairly succinct and easy to read, you might think, but let's compare it to the
function-based code:
>>> square = [(1,1), (1,2), (2,2), (2,1)]
>>> perimeter(square)
4.0
Hmm, maybe the object-oriented API isn't so compact! On the other hand, I'd argue
that it was easier to read than the function example: How do we know what the list
of tuples is supposed to represent in the second version? How do we remember what
kind of object (a list of two-tuples? That's not intuitive!) we're supposed to pass into
the perimeter function? We would need a lot of external documentation to explain
how these functions should be used.
In contrast, the object-oriented code is relatively self documenting, we just have to
look at the list of methods and their parameters to know what the object does and
how to use it. By the time we wrote all the documentation for the functional version,
it would probably be longer than the object-oriented code.
Besides, code length is a horrible indicator of code complexity. Some programmers
(thankfully, not many of them are Python coders) get hung up on complicated, "one
liners", that do incredible amounts of work in one line of code. One line of code
that even the original author isn't able to read the next day, that is. Always focus on
making your code easier to read and easier to use, not shorter.
As a quick exercise, can you think of any ways to make the object-oriented Polygon
as easy to use as the functional implementation? Pause a moment and think about it.
Really, all we have to do is alter our Polygon API so that it can be constructed with
multiple points. Let's give it an initializer that accepts a list of Point objects. In fact,
let's allow it to accept tuples too, and we can construct the Point objects ourselves,
if needed:
def __init__(self, points = []):
self.vertices = []
for point in points:
if isinstance(point, tuple):
point = Point(*point)
self.vertices.append(point)
This example simply goes through the list and ensures that any tuples are converted
to points. If the object is not a tuple, we leave it as is, assuming that it is either a
Point already, or an unknown duck typed object that can act like a Point.
[ 128 ]
Chapter 5
As we can see, it's not always easy to identify when an object should really be
represented as a self-defined class. If we have new functions that accept a polygon
argument, such as area(polygon) or point_in_polygon(polygon, x, y), the
benefits of the object-oriented code become increasingly obvious. Likewise, if we add
other attributes to the polygon, such as color or texture, it makes more and more
sense to encapsulate that data into a class.
The distinction is a design decision, but in general, the more complicated a set of
data is, the more likely it is to have functions specific to that data, and the more
useful it is to use a class with attributes and methods instead.
When making this decision, it also pays to consider how the class will be used. If
we're only trying to calculate the perimeter of one polygon in the context of a much
greater problem, using a function will probably be quickest to code and easiest to use
"one time only". On the other hand, if our program needs to manipulate numerous
polygons in a wide variety of ways (calculate perimeter, area, intersection with other
polygons, and more), we have most certainly identified an object; one that needs to
be extremely versatile.
Pay additional attention to the interaction between objects. Look for inheritance
relationships; inheritance is impossible to model elegantly without classes, so make
sure to use them. Look for the other types of relationships we discussed in Chapter 1:
association and composition. Composition can, technically, be modeled using only
data structures; for example, we can have a list of dictionaries holding tuple values,
but it is often less complicated to create an object, especially if there is behavior
associated with the data.
Don't rush to use an object just because you can use an object, but never
neglect to create a class when you need to use a class.
Using properties to add behavior to class

data
Throughout this book, we've been focusing on the separation of behavior and data.
This is very important in object-oriented programming, but we're about to see that,
in Python, the distinction can be eerily blurry. Python is very good at blurring
distinctions; it doesn't exactly help us to "think outside the box". Rather, it teaches
us that the box is in our own head; "there is no box".
[ 129 ]
Before we get into the details, let's discuss some bad object-oriented theory. Many
object-oriented languages (Java is the most guilty) teach us to never access attributes
directly. They teach us to write attribute access like this:
class Color:
def __init__(self, rgb_value, name):
self._rgb_value = rgb_value
self._name = name
def set_name(self, name):
self._name = name
def get_name(self):
return self._name
The variables are prefixed with an underscore to suggest that they are private (in other
languages it would actually force them to be private). Then the get and set methods
provide access to each variable. This class would be used in practice as follows:
>>> c = Color("#ff0000", "bright red")
>>> c.get_name()
'bright red'
>>> c.set_name("red")
>>> c.get_name()
'red'
This is not nearly as readable as the direct access version that Python favors:
class Color:
self.rgb_value = rgb_value
self.name = name
c = Color("#ff0000", "bright red")
print(c.name)
c.name = "red"
So why would anyone recommend the method-based syntax? Their reasoning is

that someday we may want to add extra code when a value is set or retrieved. For
example, we could decide to cache a value and return the cached value, or we might
want to validate that the value is a suitable input. In code, we could decide to change
the set_name() method as follows:
def set_name(self, name):
if not name:
raise Exception("Invalid Name")
self._name = name
[ 130 ]
Chapter 5
Now, in Java and similar languages, if we had written our original code to do direct
attribute access, and then later changed it to a method like the above, we'd have a
problem: Anyone who had written code that accessed the attribute directly would
now have to access the method; if they don't change the access style, their code
will be broken. The mantra in these languages is that we should never make public
members private. This doesn't make much sense in Python since there isn't any
concept of private members!
Indeed, the situation in Python is much better. We can use the Python property
keyword to make methods look like a class attribute. If we originally wrote our
code to use direct member access, we can later add methods to get and set the
name without changing the interface. Let's see how it looks:
class Color:
self.rgb_value = rgb_value
self._name = name
def _set_name(self, name):
if not name:
self._name = name
def _get_name(self):
return self._name
name = property(_get_name, _set_name)
If we had started with the earlier non-method-based class, which set the name
attribute directly, we could later change the code to look like the above. We first
change the name attribute into a (semi-) private _name attribute. Then we add two
more (semi-) private methods to get and set that variable, doing our validation when
we set it.
Finally, we have the property declaration at the bottom. This is the magic. It creates
a new attribute on the Color class called name, which now replaces the previous
name attribute. It sets this attribute to be a property, which calls the two methods
we just created whenever the property is accessed or changed. This new version of
the Color class can be used exactly the same way as the previous version, yet it now
does validation when we set the name:
>>> c = Color("#0000ff", "bright red")
>>> print(c.name)
bright red
[ 131 ]

>>> c.name = "red"
>>> print(c.name)
red
>>> c.name = ""
File "setting_name_property.py", line 8, in _set_name
Exception: Invalid Name
So if we'd previously written code to access the name attribute, and then changed
it to use our property object, the previous code would still work, unless it was
sending an empty property value, which is the behavior we wanted to forbid
in the first place. Success!
Bear in mind that even with the name property, the previous code is not 100% safe.
People can still access the _name attribute directly and set it to an empty string if
they wanted to. But if they access a variable we've explicitly marked with an
underscore to suggest it is private, they're the ones that have to deal with the
consequences, not us.
How it works
So, what exactly is that property object doing? Think of the property function
as returning an object that proxies any requests to set or access the attribute value
through the methods we have specified. The property keyword is like a constructor
for such an object.
This property constructor can actually accept two additional arguments, a deletion
function and a docstring for the property. The delete function is rarely supplied in
practice, but it can be useful for logging that a value has been deleted, or possibly
to veto deleting if we have reason to do so. The docstring is just a string describing
what the property does, no different from the docstrings we discussed in Chapter
2. If we do not supply this parameter, the docstring will instead be copied from the
docstring for the first argument: the getter method.
Here is a silly example that simply states whenever any of the methods are called:
class Silly:
def _get_silly(self):
print("You are getting silly")
return self._silly
def _set_silly(self, value):
print("You are making silly {}".format(value))
[ 132 ]
Chapter 5
self._silly = value
def _del_silly(self):
print("Whoah, you killed silly!")
del self._silly
silly = property(_get_silly, _set_silly,
_del_silly, "This is a silly property")
If we actually use this class, it does indeed print out the correct strings when we ask
it to:
>>> s = Silly()
>>> s.silly = "funny"
You are making silly funny
>>> s.silly
You are getting silly
'funny'
>>> del s.silly
Whoah, you killed silly!
Further, if we look at the help file for the Silly class (by issuing help(silly) at the
interpreter prompt), it shows us the custom docstring for our silly attribute:
Help on class Silly in module __main__:
class Silly(builtins.object)
| Data descriptors defined here:
|
| __dict__
|
dictionary for instance variables (if defined)
|
| __weakref__
|
list of weak references to the object (if defined)
|
| silly
|
This is a silly property
Once again, everything is working as we planned. In practice, properties are

normally only defined with the first two parameters; the getter and setter functions.
The docstring is defined as a normal docstring on the getter and copied into the
property, while the deletion function is left empty because object attributes are
rarely deleted. If a coder does try to delete one that doesn't have a deletion function
specified, however, it will raise an exception, so if there is any chance of a legitimate
reason to delete our property, we should supply that function.
[ 133 ]
Decorators: another way to create properties
(If you've never used Python decorators before, you might want to skip this section
and come back to it after we've discussed the decorator pattern in Chapter 8.)
Decorators were introduced in Python 2.4 as a way to modify functions dynamically
by passing them as arguments to other functions, which eventually return a new
function. We won't be covering decorators in-depth at this time, but the basic syntax
is easy to grasp. If you've never used them before, you can still follow along.
Applying a decorator can be as simple as prefixing the function name with an @
symbol, and placing the result just before the definition of the function that is being
decorated. The property function itself can be used with decorator syntax to turn a
get function into a property:
class Foo:
@property
def foo(self):
return "bar"
This applies property as a decorator, and is equivalent to applying it as

foo = property(foo). The main difference, from a readability perspective,
is that we get to mark the foo function as a property at the top of the method,
instead of after it is defined, where it can be easily overlooked.
Going one step further, we can specify a setter function for the new property
as follows:
class Foo:
@property
def foo(self):
return self._foo
@foo.setter
def foo(self, value):
self._foo = value
This syntax looks a little odd. First we decorate the foo method as a getter. Then we
decorate a new method with exactly the same name with the setter attribute of the
original decorated foo method! Remember, the property function returns an object;
this object is automatically set up to have a setter attribute, and this attribute can
be applied as a decorator to other functions. Using the same name for the get and set
methods is not required, but it does help group the multiple methods that create one
property together.
[ 134 ]
Chapter 5
We can, of course, also specify a deletion function with @foo.deleter. We cannot

specify a docstring using property decorators, so we need to rely on the property
copying the docstring from the initial getter method.
Here's our previous Silly class rewritten to use property as a decorator:
class Silly:
@property
def silly(self):
"This is a silly property"
print("You are getting silly")
return self._silly
@silly.setter
def silly(self, value):
print("You are making silly {}".format(value))
self._silly = value
@silly.deleter
def silly(self):
print("Whoah, you killed silly!")
del self._silly
This class operates exactly the same as our earlier version, including the help text.
You can use whichever syntax you feel is more readable and elegant.
When should we use properties?
With the property keyword smearing the division between behavior and data, it can
be confusing to know which one to choose. The example use case we saw earlier is
one of the most common uses of properties: we have some data on a class that we
later want to add behavior to. There are also other factors to take into account when
deciding to use a property.
Technically, in Python, data, properties, and methods are all attributes on a class. The
fact that a method is callable does not distinguish it from other types of attributes;
indeed, we'll see in Chapter 7 that it is possible to create normal objects that are
callable, and also that functions and methods are themselves normal objects.
The fact that methods are just callable attributes, and properties are just customizable
attributes can help us in our decision. Methods should only represent actions; things
that can be done to or performed by the object. When you call a method, even with
only one argument, it should do something. Methods are generally verbs.
[ 135 ]
That leaves us to decide between standard data attributes and properties. In

general, always use a standard attribute until you need to control access to that
property in some way. In either case, your attribute should be a noun. The only
difference between an attribute and a property is that we can invoke custom actions
automatically when a property is retrieved, set, or deleted.
Let's try a more realistic example. A common need for custom behavior is caching
a value that is difficult to calculate or expensive to look up (requiring, for example,
a network request or database query). The goal is to store the value locally to avoid
repeated calls to the expensive calculation.
We can do this with a custom getter on the property. The first time the value is
retrieved, we perform the lookup or calculation. Then we could locally cache the
value as a private attribute on our object (or in dedicated caching software), and the
next time the value is requested, we return the stored data. Here's how we might
cache a webpage:
from urllib.request import urlopen
class WebPage:
def __init__(self, url):
self.url = url
self._content = None
@property
def content(self):
if not self._content:
print("Retrieving New Page...")
self._content = urlopen(self.url).read()
return self._content
We can test this code to see that the page is only retrieved once:
>>> import time
>>> webpage = WebPage("http://ccphillips.net/")
>>> now = time.time()
>>> content1 = webpage.content
Retrieving New Page...
>>> time.time() - now
22.43316888809204
>>> now = time.time()
>>> content2 = webpage.content
>>> time.time() - now
1.9266459941864014
>>> content2 == content1
True
[ 136 ]
Chapter 5
On my awful satellite connection it takes twenty seconds the first time I load the
content, but the second time, I get the result in two seconds (which is really just the
amount of time it took to type the lines into the interpreter).
Custom getters are also useful for attributes that need to be calculated on the fly,
based on other object attributes. For example, we might want to calculate the average
for a list of integers:
class AverageList(list):
@property
def average(self):
return sum(self) / len(self)
This very simple class inherits from list, so we get list-like behavior for free. We
just add a property to the class, and presto, our list can have an average:
>>> a = AverageList([1,2,3,4])
>>> a.average
2.5
Of course, we could have made this a method instead, but then we should call it
calculate_average(), since methods represent actions. But a property called
average is more suitable; both easier to type, and easier to read.
Custom setters are useful for validation, as we've already seen, but they can also be
used to proxy a value to another location. For example, we could add a content setter
to the WebPage class that automatically logs into our web server and uploads a new
page whenever the value is set.
Managing objects
We've been focused on objects and their attributes and methods. Now, we'll take a
look at designing higher-level objects; the kinds of objects that manage other objects.
The objects that tie everything together.
The difference between these objects and most of the examples we've seen so far
is that our examples tend to represent concrete ideas. Management objects are
more like office managers; they don't do the actual "visible" work out on the floor,
but without them, there would be no communication between departments and
nobody would know what they are supposed to do. Analogously, the attributes
on a management class tend to refer to other objects that do the "visible" work; the
behaviors on such a class delegate to those other classes at the right time, and pass
messages between them.
[ 137 ]
As an example, we'll write a program that does a find and replace action for text files
stored in a compressed ZIP file. We'll need objects to represent the ZIP file and each
individual text file (luckily, we don't have to write these classes, they're available in
the Python Standard Library). The manager object will be responsible for ensuring
three steps occur in order:
1. Unzipping the compressed file.
2. Performing the find and replace action.
3. Zipping up the new files.
The class is initialized with the .zip filename and search and replace strings. We
create a temporary directory to store the unzipped files in, so that the folder stays
clean. We also add a useful helper method for internal use that helps identify an
individual filename inside that directory:
import
import
import
import
sys
os
shutil
zipfile
class ZipReplace:
def __init__(self, filename, search_string,
replace_string):
self.search_string = search_string
self.replace_string = replace_string
self.temp_directory = "unzipped-{}".format(
filename)
def _full_filename(self, filename):
return os.path.join(self.temp_directory, filename)
Then we create an overall "manager" method for each of the three steps. This method
delegates responsibility to other methods. Obviously, we could do all three steps
in one method, or indeed, in one script without ever creating an object. There are
several advantages to separating the three steps:
Readability: The code for each step is in a self-contained unit that is easy to
read and understand. The method names describe what the method does,
and no additional documentation is required to understand what is going on.
Extensibility: If a subclass wanted to use compressed TAR files instead of

ZIP files, it could override the zip and unzip methods without having to
duplicate the find_replace method.
[ 138 ]
Chapter 5
Partitioning: An external class could create an instance of this class and call
the find and replace method directly on some folder without having to zip
the content.
The delegation method is the first in the code below; the rest of the methods are
included for completeness:
def zip_find_replace(self):
self.unzip_files()
self.find_replace()
self.zip_files()
def unzip_files(self):
os.mkdir(self.temp_directory)
zip = zipfile.ZipFile(self.filename)
try:
zip.extractall(self.temp_directory)
finally:
zip.close()
def find_replace(self):
for filename in os.listdir(self.temp_directory):
with open(self._full_filename(filename)) as file:
contents = file.read()
contents = contents.replace(
self.search_string, self.replace_string)
with open(
self._full_filename(filename), "w") as file:
file.write(contents)
def zip_files(self):
file = zipfile.ZipFile(self.filename, 'w')
file.write(
self._full_filename(filename), filename)
shutil.rmtree(self.temp_directory)
if __name__ == "__main__":
ZipReplace(*sys.argv[1:4]).zip_find_replace()
For brevity, the code for zipping and unzipping files is sparsely documented. Our
current focus is on object-oriented design; if you are interested in the inner details
of the zipfile module, refer to the documentation in the standard library, either
online at http://docs.python.org/library/zipfile.html or by typing import
zipfile ; help(zipfile) into your interactive interpreter. Note that this example
only searches the top-level files in a ZIP file; if there are any folders in the unzipped
content, they will not be scanned, nor will any files inside those folders.
[ 139 ]
The last two lines in the code allow us to run the example from the command line by
passing the zip filename, search string, and replace string as arguments:
python zipsearch.py hello.zip hello hi
Of course, this object does not have to be created from the command line; it could be
imported from another module (to perform batch ZIP file processing) or accessed as
part of a GUI interface or even a higher-level management object that knows what to
do with ZIP files (for example to retrieve them from an FTP server or back them up
to an external disk).
As programs become more and more complex, the objects being modeled
become less and less like physical objects. Properties are other abstract objects
and methods are actions that change the state of those abstract objects. But at the
heart of every object, no matter how complex, is a set of concrete properties and
well-defined behaviors.
Removing duplicate code
Often the code in management style classes such as ZipReplace is quite generic
and can be applied in many different ways. It is possible to use either composition
or inheritance to help keep this code in one place, thus eliminating duplicate code.
Before we look at any examples of this, let's discuss a tiny bit of theory. Specifically:
why is duplicate code a bad thing?
There are several reasons, but they all boil down to readability and maintainability.
When we're writing a new piece of code that is similar to an earlier piece, the easiest
thing to do is copy the old code and change whatever needs to change (variable
names, logic, comments) to make it work in the new location. Alternatively, if we're
writing new code that seems similar, but not identical to code elsewhere in the
project, the easiest thing to do is write fresh code with similar behavior, rather
than figure out how to extract the overlapping functionality.
But as soon as someone has to read and understand the code and they come across
duplicate blocks, they are faced with a dilemma. Code that might have made sense
suddenly has to be understood. How is one section different from the other? How
are they the same? Under what conditions is one section called? When do we call the
other? You might argue that you're the only one reading your code, but if you don't
touch that code for eight months it will be as incomprehensible to you as to a fresh
coder. When we're trying to read two similar pieces of code, we have to understand
why they're different, as well as how they're different. This wastes the reader's time;
code should always be written to be readable first.
[ 140 ]
Chapter 5
I once had to try to understand someone's code that had three identical
copies of the same three hundred lines of very poorly written code. I
had been working with the code for a month before I realized that the
three "identical" versions were actually performing slightly different tax
calculations. Some of the subtle differences were intentional, but there
were also obvious areas where someone had updated a calculation in
one function without updating the other two. The number of subtle,
incomprehensible bugs in the code could not be counted.
Reading such duplicate code can be tiresome, but code maintenance is an even
greater torment. As the preceding story suggests, keeping two similar pieces of
code up to date can be a nightmare. We have to remember to update both sections
whenever we update one of them, and we have to remember how the multiple
sections differ so we can modify our changes when we are editing each of them. If
we forget to update both sections, we will end up with extremely annoying bugs that
usually manifest themselves as, "but I fixed that already, why is it still happening?"
The result is that people who are reading or maintaining our code have to spend
astronomical amounts of time understanding and testing it compared to if we
had written the code in a non-repetitive manner in the first place. It's even more
frustrating when we are the ones doing the maintenance. The time we save by
copy-pasting existing code is lost the very first time we have to maintain it. Code is
both read and maintained many more times and much more often than it is written.
Comprehensible code should always be paramount.
This is why programmers, especially Python programmers (who tend to value
elegant code more than average), follow what is known as the Don't Repeat
Yourself, or DRY principle. DRY code is maintainable code. My advice to beginning
programmers is to never use the copy and paste feature of their editor. To
intermediate programmers, I suggest they think thrice before they hit Ctrl + C.
But what should we do instead of code duplication? The simplest solution is often
to move the code into a function that accepts parameters to account for whatever
sections are different. This isn't a terribly object-oriented solution, but it is frequently
sufficient. For example, if we have two pieces of code that unzip a ZIP file into two
different directories, we can easily write a function that accepts a parameter for the
directory to which it should be unzipped instead. This may make the function itself
slightly more difficult to read, but a good function name and docstring can easily
make up for that, and any code that invokes the function will be easier to read.
That's certainly enough theory! The moral of the story is: always make the effort to
refactor your code to be easier to read instead of writing bad code that is only easier
to write.
[ 141 ]
In practice
Let's explore two ways we can reuse existing code. After writing our code to replace
strings in a ZIP file full of text files, we are later contracted to scale all the images in a
ZIP file to 640x480. Looks like we could use a very similar paradigm to what we used
in ZipReplace. The first impulse, obviously, would be to save a copy of that file and
change the find_replace method to scale_image or something similar. But, that's
just not cool. What if someday we want to change the unzip and zip methods to also
open TAR files? Or maybe we want to use a guaranteed unique directory name for
temporary files. In either case, we'd have to change it in two different places!
We'll start by demonstrating an inheritance-based solution to this problem. First
we'll modify our original ZipReplace class into a superclass for processing generic
ZIP files:
import os
import shutil
import zipfile
class ZipProcessor:
def __init__(self, zipname):
self.zipname = zipname
zipname[:-4])
def process_zip(self):
self.unzip_files()
self.process_files()
self.zip_files()
zip = zipfile.ZipFile(self.zipname)
try:
finally:
zip.close()
file = zipfile.ZipFile(self.zipname, 'w')
file.write(self._full_filename(
filename), filename)
[ 142 ]
Chapter 5
We changed the filename property to zipfile to avoid confusion with the

filename local variables inside the various methods. This helps make the code more
readable even though it isn't actually a change in design. We also dropped the two
parameters to __init__ (search_string and replace_string) that were specific to
ZipReplace. Then we renamed the zip_find_replace method to process_zip and
made it call an (as yet undefined) process_files method instead of find_replace;
these name changes help demonstrate the more generalized nature of our new class.
Notice that we have removed the find_replace method altogether; that code is
specific to ZipReplace and has no business here.
This new ZipProcessor class doesn't actually define a process_files method; so if
we ran it directly, it would raise an exception. Since it actually isn't meant to be run
directly, we also removed the main call at the bottom of the original script.
Now, before we move on to our image processing app, let's fix up our original
zipsearch to make use of this parent class:
from zip_processor import ZipProcessor

import sys
import os
class ZipReplace(ZipProcessor):
def __init__(self, filename, search_string,
replace_string):
super().__init__(filename)
def process_files(self):
'''perform a search and replace on all files
in the temporary directory'''
with open(self._full_filename(filename)) as file:
with open(
self._full_filename(filename), "w") as file:
if __name__ == "__main__":
ZipReplace(*sys.argv[1:4]).process_zip()
[ 143 ]
This code is a bit shorter than the original version, since it inherits its ZIP processing
abilities from the parent class. We first import the base class we just wrote and make
ZipReplace extend that class. Then we use super() to initialize the parent class.
The find_replace method is still here, but we renamed it to process_files so
the parent class can call it. Because this name isn't as descriptive as the old one, we
added a docstring to describe what it is doing.
Now, that was quite a bit of work, considering that all we have now is a program
that is functionally no different from the one we started with! But having done that
work, it is now much easier for us to write other classes that operate on files in a
ZIP archive, such as our photo scaler. Further, if we ever want to improve the zip
functionality, we can do it for all classes by changing only the one ZipProcessor
base class. Maintenance will be much more effective.
See how simple it is, now to create a photo scaling class that takes advantage of the
ZipProcessor functionality. (Note: this class requires the third-party pygame library
to be installed. You can download it from http://www.pygame.org/.)
import os
import sys
from pygame import image
from pygame.transform import scale
class ScaleZip(ZipProcessor):
def process_files(self):
'''Scale each image in the directory to 640x480'''
im = image.load(self._full_filename(filename))
scaled = scale(im, (640,480))
image.save(scaled, self._full_filename(filename))
if __name__ == "__main__":
ScaleZip(*sys.argv[1:4]).process_zip()
All that work we did earlier paid off! Look how simple this class is! All we do is open
each file (assuming that it is an image; it will unceremoniously crash if the file cannot
be opened), scale it, and save it back. The ZipProcessor takes care of the zipping
and unzipping without any extra work on our part.
[ 144 ]
Chapter 5
Or we can use composition
Now, let's try solving the same problem using a composition-based solution. Even
though we're completely changing paradigms, from inheritance to composition, we
only have to make a minor modification to our ZipProcessor class:
import os
import shutil
import zipfile
class ZipProcessor:
def __init__(self, zipname, processor):
self.zipname = zipname
zipname[:-4])
self.processor = processor
def process_zip(self):
self.unzip_files()
self.processor.process(self)
self.zip_files()
zip = zipfile.ZipFile(self.zipname)
try:
finally:
zip.close()
file = zipfile.ZipFile(self.zipname, 'w')
file.write(self._full_filename(filename), filename)
[ 145 ]
All we did was change the initializer to accept a processor object. The process_zip
function now calls a method on that processor object; the method called accepts a
reference to the ZipProcessor itself. Now we can change our ZipReplace class to be
a suitable processor object that no longer uses inheritance:
import sys
import os
class ZipReplace:
def __init__(self, search_string,
replace_string):
def process(self, zipprocessor):
'''perform a search and replace on all files in the
temporary directory'''
for filename in os.listdir(
zipprocessor.temp_directory):
with open(
zipprocessor._full_filename(filename)) as file:
with open(zipprocessor._full_filename(
filename), "w") as file:
if __name__ == "__main__":
zipreplace = ZipReplace(*sys.argv[2:4])
ZipProcessor(sys.argv[1], zipreplace).process_zip()
We didn't actually change much here; the class no longer inherits from
ZipProcessor, and when we process the files, we accept a zipprocessor object that
gives us the function to calculate __full_filename. In the bottom two lines, when
we run from the command line, we first construct a ZipReplace object. This is then
passed into the ZipProcessor constructor so the two objects can communicate.
This design is a terrific separation of interests. Now we have a ZipProcessor that
can accept any object that has a process method to do the actual processing. Further,
we have a ZipReplace that can be passed to any method, function, or object that
wants to call its process function; it is no longer tied to the zip processing code
through an inheritance relationship; it could now be applied with equal ease to a
local or network filesystem, for example, or to a different kind of compressed file
such as a RAR archive.
[ 146 ]
Chapter 5
Any inheritance relationship can be modeled as a composition relationship (change

the "is a" to a "has a parent") instead, but that does not mean it always should be.
And the reverse is not true, most composition relationships cannot be (properly)
modeled as inheritance.
Case study
For this case study, we'll try to delve further into the question, "when should I choose
an object versus a built-in type?" We'll be modeling a Document class that might be
used in a text editor or word processor. What objects, functions, or properties should
it have?
We might start with a str for the Document contents, but strings aren't mutable. A
mutable object is one that can be changed; but a str is immutable, we can't insert
a character into it or remove one without creating a brand new string object. That's
leaving a lot of str objects for Python's garbage collector to clean up behind us.
So, instead of a string, we'll use a list of characters, which we can modify at will. In
addition, a Document would need to know the current cursor position within the list,
and should also store a filename for the document.
Now, what methods should it have? There are a lot of things we might want to do
to a text document, including inserting and deleting characters, cut, copy, paste, and
saving or closing the document. It looks like there are copious amounts of both data
and behavior, so it makes sense to put all this stuff into its own Document class.
The question is, should this class be composed of a bunch of basic Python objects
such as str filenames, int cursor positions, and a list of characters? Or should
some or all of those things be specially defined objects in their own right? What
about individual lines and characters, do they need to have classes of their own?
We'll answer these questions as we go, but let's just design the simplest possible
Document class first and see what it can do:
class Document:
def __init__(self):
self.characters = []
self.cursor = 0
self.filename = ''
def insert(self, character):
self.characters.insert(self.cursor, character)
self.cursor += 1
def delete(self):
[ 147 ]

del self.characters[self.cursor]
def save(self):
f = open(self.filename, 'w')
f.write(''.join(self.characters))
f.close()
def forward(self):
self.cursor += 1
def back(self):
self.cursor -= 1
This simple class allows us full control over editing a basic document. Have a look at
it in action:
>>> doc = Document()
>>> doc.filename = "test_document"
>>> doc.insert('h')
>>> doc.insert('e')
>>> doc.insert('l')
>>> doc.insert('l')
>>> doc.insert('o')
>>> "".join(doc.characters)
'hello'
>>> doc.back()
>>> doc.delete()
>>> doc.insert('p')
>>> "".join(doc.characters)
'hellp'
Looks like it's working. We could connect a keyboard's letter and arrow keys to these
methods and the document would track everything just fine.
But what if we want to connect more than just arrow keys. What if we want to
connect the Home and End keys as well? We could add more methods to the
Document class that search forward or backwards for newline characters (in Python,
a newline character, or \n represents the end of one line and the beginning of a new
one) in the string and jump to them, but if we did that for every possible movement
action (move by words, move by sentences, Page Up, Page Down, end of line,
beginning of whitespace, and more), the class would be huge. Maybe it would be
better to put those methods on a separate object. What we can do is turn the cursor
[ 148 ]
Chapter 5
attribute into an object that is aware of its position and can manipulate that position.
We can move the forward and back methods to that class, and add a couple more for
the Home and End keys:
class Cursor:
def __init__(self, document):
self.document = document
self.position = 0
def forward(self):
self.position += 1
def back(self):
self.position -= 1
def home(self):
while self.document.characters[
self.position-1] != '\n':
self.position -= 1
if self.position == 0:
# Got to beginning of file before newline
break
def end(self):
while self.position < len(self.document.characters
) and self.document.characters[
self.position] != '\n':
self.position += 1
This class takes the document as an initialization parameter so the methods have
access to the contents of the document's character list. It then provides simple
methods for moving backwards and forwards, as before, and for moving to the
home and end positions.
This code is not very safe. You can very easily move past the ending
position, and if you try to go home on an empty file it will crash. These
examples are kept short to make them readable, that doesn't mean
they are defensive! You can improve the error checking of this code as
an exercise; it might be a great opportunity to expand your exception
handling skills.
[ 149 ]
The Document class itself is hardly changed, except for removing the two methods
that were moved to the Cursor class:
class Document:
def __init__(self):
self.characters = []
self.cursor = Cursor(self)
self.filename = ''
self.characters.insert(self.cursor.position,
character)
self.cursor.forward()
def delete(self):
del self.characters[self.cursor.position]
def save(self):
f = open(self.filename, 'w')
f.write(''.join(self.characters))
f.close()
We simply updated anything that accessed the old cursor integer to use the new
object instead. We can test that the home method is really moving to the newline
character.
>>> d = Document()
>>> d.insert('h')
>>> d.insert('e')
>>> d.insert('l')
>>> d.insert('l')
>>> d.insert('o')
>>> d.insert('\n')
>>> d.insert('w')
>>> d.insert('o')
>>> d.insert('r')
>>> d.insert('l')
>>> d.insert('d')
>>> d.cursor.home()
>>> d.insert("*")
>>> print("".join(d.characters))
hello
*world
[ 150 ]
Chapter 5
Now, since we've been using that string join function a lot (to concatenate the
characters so we can see the actual document contents), we can add a property
to the Document class to give us the complete string:
@property
def string(self):
return "".join(self.characters)
This makes our testing a little simpler:

>>> print(d.string)
hello
world
This framework is easy enough to extend to create a complete text editor document.
Now, let's make it work for rich text; text that can have bold, underlined, or italic
characters. There are two ways we could process this; the first is to insert "fake"
characters into our character list that act like instructions such as "bold characters
until you find a stop bold character". The second is to add information to each
character indicating what formatting it should have. While the former method
is probably more common, we'll implement the latter solution. To do that, we're
obviously going to need a class for characters. This class will have an attribute
representing the character, as well as three boolean attributes representing
whether it is bold, italic, or underlined.
Hmm, Wait! Is this character class going to have any methods? If not, maybe we
should use one of the many Python data structures instead; a tuple or named tuple
would probably be sufficient. Are there any actions that we would want to do to, or
invoke on a character?
Well, clearly, we might want to do things with characters, such as delete or copy
them, but those are things that need to be handled at the Document level, since they
are really modifying the list of characters. Are there things that need to be done to
individual characters?
Actually, now that we're thinking about what a Character actually is... what is
it? Would it be safe to say that a Character is a string? Maybe we should use an
inheritance relationship here? Then we can take advantage of the numerous methods
that str instances come with.
What sorts of methods are we talking about? There's startswith, strip, find,
lower, and many more. Most of these methods expect to be working on strings that
contain more than one character. In contrast, if Character were to subclass str,
we'd probably be wise to override __init__ to raise an exception if a multi-character
string were supplied. Since all those methods we'd get for free wouldn't really apply
to our Character class, it turns out we shouldn't use inheritance, after all.
[ 151 ]
This leaves us at our first question; should Character even be a class? There is a
very important special method on the object class that we can take advantage of
to represent our characters. This method, called __str__ (two underscores, like
__init__), is used in string manipulation functions like print and the str
constructor to convert any class to a string. The default implementation does
some boring stuff like printing the name of the module and class and its address
in memory. But if we override it, we can make it print whatever we like. For our
implementation, we can make it prefix characters with special characters to represent
whether they are bold, italic, or underlined. So we will create a class to represent a
character, and here it is:
class Character:
def __init__(self, character,
bold=False, italic=False, underline=False):
assert len(character) == 1
self.character = character
self.bold = bold
self.italic = italic
self.underline = underline
def __str__(self):
bold = "*" if self.bold else ''
italic = "/" if self.italic else ''
underline = "_" if self.underline else ''
return bold + italic + underline + self.character
This class allows us to create characters and prefix them with a special character
when the str() function is applied to them. Nothing too exciting there. We only
have to make a few minor modifications to the Document and Cursor classes to work
with this class. In the Document class, we add these two lines at the beginning of the
insert method:
if not hasattr(character, 'character'):
character = Character(character)
This is a rather strange bit of code. Its basic purpose is to check if the character being
passed in is a Character or a str. If it is a string, it is wrapped in a Character class
so all objects in the list are Character objects. However, it is entirely possible that
someone using our code would want to use a class that is neither Character nor
string, using duck typing. If the object has a character attribute, we assume it is a
"Character-like" object. But if it does not, we assume it is a "str-like" object and
wrap it in a Character. This helps the program take advantage of duck typing as
well as polymorphism; as long as an object has a character attribute, it can be used
in the Document. This could be very useful, for example, if we wanted to make
a programmer's editor with syntax highlighting: we'd need extra data on the
character, such as what type of token the character belongs to.
[ 152 ]
Chapter 5
In addition, we need to modify the string property on Document to accept the new
Character values. All we need to do is call str() on each character before we join it:
@property
def string(self):
return "".join((str(c) for c in self.characters))
This code uses a generator expression, which we'll discuss in the next chapter. It's
simply a shortcut to perform a specific action on all the objects in a sequence.
Finally we also need to check Character.character, instead of just the string
character we were storing before, in the home and end functions when we're
looking to see if it matches a newline.
def home(self):
while self.document.characters[
self.position-1].character != '\n':
self.position -= 1
if self.position == 0:
# Got to beginning of file before newline
break
def end(self):
while self.position < len(
self.document.characters) and \
self.document.characters[
self.position
].character != '\n':
self.position += 1
This completes the formatting of characters. We can test it to see that it works:
>>> d = Document()
>>> d.insert('h')
>>> d.insert('e')
>>> d.insert(Character('l', bold=True))
>>> d.insert(Character('l', bold=True))
>>> d.insert('o')
>>> d.insert('\n')
>>> d.insert(Character('w', italic=True))
>>> d.insert(Character('o', italic=True))
>>> d.insert(Character('r', underline=True))
>>> d.insert('l')
>>> d.insert('d')
>>> print(d.string)
[ 153 ]

he*l*lo
/w/o_rld
>>> d.cursor.home()
>>> d.delete()
>>> d.insert('W')
>>> print(d.string)
he*l*lo
W/o_rld
>>> d.characters[0].underline = True
>>> print(d.string)
_he*l*lo
W/o_rld
>>>
As expected, whenever we print the string, each bold character is preceded by a *,

each italic character by a /, and each underlined character by a _. All our functions
seem to work, and we can modify characters in the list after the fact. We have a
working rich text document object that could be plugged into a user interface and
hooked up with a keyboard for input and a screen for output. Naturally, we'd want
to display real bold, italic, and underlined characters on the screen, instead of using
our __str__ method, but it was sufficient for the basic testing we demanded of it.
Exercises
We've looked at various ways that objects, data, and methods can interact with each
other in an object-oriented Python program. As usual, your first thoughts should
be how you can apply these principles to your own work. Do you have any messy
scripts lying around that could be rewritten using an object-oriented manager? Look
through some of your old code and look for methods that are not actions. If the name
isn't a verb, try rewriting it as a property.
Think about code you've written in any language. Does it break the DRY principle?
Is there any duplicate code? Did you copy and paste code? Did you write two
versions of similar pieces of code because you didn't feel like understanding the
original code? Go back over some of your recent code now and see if you can refactor
the duplicate code using inheritance or composition. Try to pick a project you're still
interested in maintaining; not code so old that you never want to touch it again. It
helps keep your interest up when you do the improvements!
[ 154 ]
Chapter 5
Now, look back over some of the examples we saw in this chapter. Start with the
cached webpage example which uses a property to cache the retrieved data. An
obvious problem with this example is that the cache is never refreshed. Add a
timeout to the getter for the property, and only return the cached page if the page
has been requested before the timeout has expired. You can use the time module
(time.time() - an_old_time returns the number of seconds that have elapsed
since an_old_time) to determine whether the cache has expired.
Now look at the composition and inheritance based versions of ZipProcessor.
We wrote an inheritance-based ScaleZipper, but didn't port it to the composite
ZipProcessor. Try writing the composite ScaleZipper and compare the two pieces
of code. Which version do you find easier to use? Which is more elegant? What
is easier to read? These are subjective questions; the answer varies for each of us.
Knowing the answer, however, is important; if you find you prefer inheritance over
composition, you have to pay attention that you don't overuse inheritance in your
daily coding. If you prefer composition, make sure you don't miss opportunities to
create an elegant inheritance-based solution.
Finally, add some error handlers to the various classes we created in the case study.
They should ensure single characters are entered, that you don't try to move the
cursor past the end or beginning of the file, that you don't delete a character that
doesn't exist, and that you don't save a file without a filename. Try to think of as
many edge cases as you can, and account for them. Consider different ways to
handle them; should you raise an exception when the user tries to move past the
end of the file, or just stay on the last character?
Pay attention, in your daily coding, to the copy and paste commands. Every time you
use them in your editor, consider whether it would be a good idea to improve your
program's organization so that you only have one version of the code you are about
to copy.
[ 155 ]
Summary
In this chapter, we focused on identifying objects, especially objects that are not
immediately apparent; objects that manage and control. In particular, we covered:
Why objects should have both data and behavior
How properties blur the distinction between data and behavior
The DRY principle and the follies of duplicate code
Inheritance and composition for reducing code duplication
In the next chapter we'll cover several of the built-in Python data structures and
objects, focusing on their object-oriented properties and how they can be extended
or adapted.
[ 156 ]
Python Data Structures

In our examples so far, we've already seen many of the built-in Python data
structures in action. You've probably covered many of them in introductory books
or tutorials. In this chapter, we'll be discussing the object-oriented features of these
data structures, when they should be used instead of a regular class, and when they
should not be used. In particular we'll be covering:
Tuples
Dictionaries
Lists and sets
How and why to extend built-in objects
Empty objects
Let's start with the most basic Python built-in, one that we've seen many times
already, the one that we've extended in every class we have created: the object.
Technically, we can instantiate an object without writing a subclass:
>>> o = object()
>>> o.x = 5
AttributeError: 'object' object has no attribute 'x'
Unfortunately, as you can see, it's not possible to set any attributes on an object that
was instantiated directly. This isn't because the Python developers wanted to force
us to write our own classes, or anything so sinister. No, they simply wanted to save
memory; a lot of memory. When Python allows an object to have arbitrary attributes,
it takes a certain amount of system memory to keep track of what attributes each
object has, for storing both the attribute name and its value. Even if no attributes
are stored, a certain amount of memory is allocated for potential new attributes.
Given the dozens, hundreds, or thousands of objects (every class extends object) in a
typical Python program, this small amount of memory would quickly become a large
amount of memory. So Python disables arbitrary properties on object, and several
other built-ins, by default.
It is possible to restrict arbitrary properties on our own classes using slots. Slots are
beyond the scope of this book, but you now have a search term if you are looking for
more information. In normal use, there isn't much benefit to using slots, but if you're
writing an object that will be duplicated thousands of times throughout the system,
they can help save memory, just as they do for object.
It is, however, trivial to create an empty object class of our own; we saw it in our
earliest example:
class MyObject:
pass
And, as we've also seen, it's possible to set attributes on such classes:
>>> m = MyObject()
>>> m.x = "hello"
>>> m.x
'hello'
Clearly, if we wanted to group properties together, we could store them in an

empty object like this. But we don't normally want to do this. There are other builtins designed for storing data, and we'll be looking at them very soon. As has been
stressed throughout this book, classes and objects should only be used when you
want to specify both data and behaviors. The main reason to write an empty class is
when we want to quickly block something out and we know we're going to have to
come back and add behavior to it later. It is much easier to adapt behaviors to a class
than it is to replace a data structure with an object and change all references to it. It
is important to decide from the outset if the data is just data, or if it is an object in
disguise. Once that design decision is made, the rest of the design typically falls
into place.
[ 158 ]
Chapter 6
Tuples are objects that can store a specific number of other objects in order. They are
immutable, so we can't add, remove, or replace objects on the fly. This may seem
like a massive restriction, but the truth is, if you need to modify a tuple, you're using
the wrong data type (a list would be more suitable). The primary benefit of tuples'
immutability is that we can use them as keys in dictionaries, and in other locations
where an object requires a hash value.
Tuples are used to store data; behavior cannot be stored in a tuple. If we require
behavior to manipulate a tuple, we have to pass the tuple into a function (or method
on another object) that performs the action.
Tuples should generally store values that are somehow different from each other.
For example, we would not put three stock symbols in a tuple, but we might create a
tuple of stock symbol, current price, high, and low for the day. The primary purpose
of a tuple is to aggregate different pieces of data together into one container. Indeed,
a tuple can be the easiest tool to replace the "object with no data" idiom from the
previous section.
We can create a tuple by separating the values with a comma. Usually tuples
are wrapped in parentheses to make them easy to read and to group them from
other parts of an expression, but this is not always mandatory. The following two
assignments are identical (they record a stock, the current price, the high, and the
low for a rather profitable company):
>>> stock = "GOOG", 613.30, 625.86, 610.50
>>> stock2 = ("GOOG", 613.30, 625.86, 610.50)
If we're grouping a tuple inside of some other object, such as a function call, list
comprehension, or generator, the parentheses are required. Otherwise it would
be impossible for the interpreter to know if it is a tuple or another parameter. For
example, the following function accepts a tuple and a date, and returns a tuple of
the date and the middle value between the stock's high and low value:
import datetime
def middle(stock, date):
symbol, current, high, low = stock
return (((high + low) /2), date)
mid_value, date = middle(("GOOG", 613.30, 625.86, 610.50),
datetime.date(2010, 1, 6))
[ 159 ]
The tuple is created directly inside the function call by separating the values with
commas and enclosing the entire tuple in parenthesis. The tuple is followed by a
comma to separate it from the second argument.
This example also illustrates tuple unpacking. The first line inside the function
unpacks the stock parameter into four different variables. The tuple has to be exactly
the same length as the number of variables, or it will raise an exception. We can also
see an example of tuple unpacking on the last line, where the tuple returned inside
the function is unpacked into two values, mid_value, and date. Granted, this is a
strange thing to do, since we supplied the date to the function in the first place, but it
gave us a chance to see unpacking at work.
Unpacking is a very useful feature in Python. We can group variables together to
make storing and passing them around simpler, but the moment we need to access
all of them, we can unpack them into separate variables. Of course, sometimes we
only need access to one of the variables in the tuple. We can use the same syntax that
we have used for other sequence types (lists and strings for example) to access an
individual value:
>>> stock = "GOOG", 613.30, 625.86, 610.50
>>> high = stock[2]
>>> high
625.86
We can even use slice notation to extract larger pieces of tuples:

>>> stock[1:3]
(613.3, 625.86)
These examples, while illustrating how useful tuples can be, also demonstrate one of
their major disadvantages: readability. How does someone reading this code know
what is in the second position of a specific tuple? They can guess, from the name
of the variable we assigned it to, that it is a high of some sort, but if we had just
accessed the tuple value in a calculation without assigning it, there would be no such
indication. They would have to paw through the code to find where the tuple was
declared before they could discover what it does. Accessing tuple members directly
is fine in some circumstances, but don't make a habit of it. Such so-called "magic
numbers" (numbers that seem to come out of thin air with no apparent meaning
within the code) are the source of many coding errors and hours of frustrated
debugging. Try to use tuples only when you know that all the values are going to
be useful at once and it's normally going to be unpacked when it is accessed. If you
have to access a member directly or using a slice, and the purpose of that value is not
immediately obvious, at least include a comment explaining where it came from.
[ 160 ]
Chapter 6
Named tuples
So, what do we do when we want to group values together, but know we're
frequently going to need to access them individually? Well, we could use an empty
object, as discussed in the previous section (but that is rarely useful unless we
anticipate adding behavior later), or we could use a dictionary (most useful if we
don't know exactly how many or which specific data will be stored), as we'll cover in
the next section.
If, however, we do not need to add behavior to the object, and we know in advance
what attributes we need to store, we can use a named tuple. Named tuples are tuples
with attitude. Named tuples are objects without behavior. Named tuples are a great
way to group data together, especially read-only data.
Constructing a named tuple takes a bit more work than a normal tuple. First we have
to import namedtuple, as it is not in the namespace by default. Then we describe the
named tuple by giving it a name and outlining its attributes. This returns a class-like
object that we can instantiate with the required values as many times as we want:
from collections import namedtuple
Stock = namedtuple("Stock", "symbol current high low")
stock = Stock("GOOG", 613.30, high=625.86, low=610.50)
The namedtuple constructor accepts two arguments. The first is an identifier for the
named tuple. The second is a string of space-separated attributes that the named
tuple can have. The first attribute should be listed, followed by a space, then the
second attribute, then another space, and so on. The result is an object that can be
used to instantiate other objects. This new object can be called just like a normal
class. The constructor must have exactly the right number of arguments; these can be
passed in order, or as keyword arguments, but all attributes must be specified. We
can create as many instances of this "class" as we like, with different values for each.
The resulting named tuple can then be packed and unpacked like a normal tuple, but
we can also access individual attributes on it as if it were a class:
>>> stock.high
625.86
>>> symbol, current, high, low = stock
>>> current
613.3
Remember that constructing named tuples is a two-step process. First, use

collections.namedtuple to create a class, and then create instances
of that class.
[ 161 ]
Named tuples are perfect for many "data only" representations, but they are not ideal
for all situations. Like tuples and strings, named tuples are immutable, so we cannot
modify an attribute once it has been set. For example, the current value of our stock
has gone down since we started this discussion, but we can't set the new value:
>>> stock.current = 609.27
AttributeError: can't set attribute
If we need to be able to change stored data, a dictionary may be what we

need instead.
Dictionaries
Dictionaries are incredibly useful objects that allow us to map objects directly to
other objects. An empty object with attributes to it is a sort of dictionary; the names
of the properties map to the property values. This is actually closer to the truth than
it sounds; internally, objects normally represent attributes as a dictionary, where the
values are properties or methods on the objects. Even the attributes on a module are
stored, internally, in a dictionary.
Dictionaries are extremely efficient at looking up a value, given a specific lookup
object that maps to that value. They should always be used when you want to
find one object based on another object. The object that is being stored is called the
value; the object that is being used as an index is called the key. We've already seen
dictionary syntax in some of our previous examples, but for completeness, we'll go
over it again. Dictionaries can be created either using the dict() constructor, or
using the {} syntax shortcut. In practice the latter format is almost always used. We
can pre-populate a dictionary by separating the keys from the values using a colon,
and separating the key value pairs using a comma.
For example, in a stock application, we would most often want to look up prices by
the stock symbol. We can create a dictionary that uses stock symbols as keys, and
tuples of current, high, and low as values like this:
stocks = {"GOOG": (613.30, 625.86, 610.50),
"MSFT": (30.25, 30.70, 30.19)}
[ 162 ]
Chapter 6
As we've seen in previous examples, we can then look up values in the dictionary by
requesting a key inside square brackets. If the key is not in the dictionary, it will raise
an exception:
>>> stocks["GOOG"]
(613.3, 625.86, 610.5)
>>> stocks["RIM"]
KeyError: 'RIM'
We can, of course, catch the KeyError and handle it. But we have other options.
Remember, dictionaries are objects, even if their primary purpose is to hold other
objects. As such, they have several behaviors associated with them. One of the most
useful of these methods is the get method; it accepts a key as the first parameter and
an optional default value if the key doesn't exist:
>>> print(stocks.get("RIM"))
None
>>> stocks.get("RIM", "NOT FOUND")
'NOT FOUND'
For even more control, we can use the setdefault method. If the key is in the
dictionary, this method behaves just like get; it returns the value for that key.
Otherwise, if the key is not in the dictionary, it will not only return the default
value we supply in the method call (just like get does), it will also set the key to
the same value. Another way to think of it is that setdefault sets a value in the
dictionary only if that value has not previously been set. Then it returns the value
in the dictionary, either the one that was already there, or the newly provided
default value.
>>> stocks.setdefault("GOOG", "INVALID")
(613.3, 625.86, 610.5)
>>> stocks.setdefault("RIM", (67.38, 68.48, 67.28))
(67.38, 68.48, 67.28)
>>> stocks["RIM"]
(67.38, 68.48, 67.28)
The GOOG stock was already in the dictionary, so when we tried to setdefault it to
an invalid value, it just returned the value already in the dictionary. RIM was not in
the dictionary, so setdefault returned the default value and set the new value in
the dictionary for us. We then check that the new stock is, indeed, in the dictionary.
[ 163 ]
Three other very useful dictionary methods are keys(), values(), and items(). The
first two return an iterator over all the keys and all the values in the dictionary. We
can use these like lists or in for loops if we want to process all the keys or values.
The items() method is probably the most useful; it returns an iterator over tuples
of (key, value) pairs for every item in the dictionary. This works great with tuple
unpacking in a for loop to loop over associated keys and values. This example
does just that to print each stock in the dictionary with its current value:
>>> for stock, values in stocks.items():
...
print("{} last value is {}".format(stock, values[0]))
...
GOOG last value is 613.3
RIM last value is 67.38
MSFT last value is 30.25
Each key/value tuple is unpacked into two variables named stock and values (we
could use any variable names we wanted, but these both seem appropriate) and then
printed in a formatted string.
Notice that the stocks do not show up in the same order in which they were inserted.
Dictionaries, due to the efficient algorithm (known as hashing) that is used to make
key lookup so fast, are inherently unsorted.
So, there are numerous ways to retrieve data from a dictionary once it has been
instantiated; we can use square brackets as index syntax, the get method, the
setdefault method, or iterate over the items method, among others.
Finally, as you probably already know, we can set a value in a dictionary using the
same indexing syntax we use to retrieve a value:
>>> stocks["GOOG"] = (597.63, 610.00, 596.28)
>>> stocks['GOOG']
(597.63, 610.0, 596.28)
Google's price is lower today, so I've updated the tuple value in the dictionary. We
can use this index syntax to set a value for any key, regardless of whether the key is
in the dictionary. If it is in the dictionary, the old value will be replaced with the new
one; otherwise, a new key/value pair will be created.
[ 164 ]
Chapter 6
We've been using strings as dictionary keys, so far but we aren't limited to string
keys. It is common to use strings as keys, especially when we're storing data in a
dictionary to gather it together (instead of using an object with named properties).
But we can also use tuples, numbers, or even objects we've defined ourselves as
dictionary keys. We can even use different types of keys in a single dictionary:
random_keys = {}
random_keys["astring"] = "somestring"
random_keys[5] = "aninteger"
random_keys[25.2] = "floats work too"
random_keys[("abc", 123)] = "so do tuples"
class AnObject:
def __init__(self, avalue):
self.avalue = avalue
my_object = AnObject(14)
random_keys[my_object] = "We can even store objects"
my_object.avalue = 12
try:
random_keys[[1,2,3]] = "we can't store lists though"
except:
print("unable to store list\n")
for key, value in random_keys.items():
print("{} has value {}".format(key, value))
This code shows several different types of keys we can supply to a dictionary. It also
shows one type of object that cannot be used. We've already used lists extensively,
and we'll be seeing many more details of them in the next section. Because lists can
change at any time (by adding or removing items, for example), they cannot hash
to a specific value. Objects that are hashable basically have a defined algorithm
that converts the object into an integer value for rapid lookup. This hash is what is
actually used to look up values in a dictionary. Strings map to integers based on the
characters in the string, while tuples combine hashes of the items inside the tuple,
for example. Any two objects that are somehow considered equal (like strings with
the same characters or tuples with the same values) should have the same hash
value, and the hash value for an object should never ever change. Lists, however, can
have their contents changed, which would change their hash value (two lists should
only be equal if their contents are the same). Because of this, they can't be used as
dictionary keys. For the same reason, dictionaries cannot be used as keys into
other dictionaries.
[ 165 ]
In contrast, there are no limits on the types of objects that can be used as dictionary
values. We can use a string key that maps to a list value, for example, or we can have
a nested dictionary as a value in another dictionary.
When should we use dictionaries?
Dictionaries are extremely versatile and have numerous uses. There are two major
ways that dictionaries can be used. The first is dictionaries where all the keys
represent different instances of similar objects; for example, our stock dictionary.
This is an indexing system. We use the stock symbol as an index to the values. The
values could even have been complicated self-defined objects that made buy and sell
decisions or set a stop-loss, rather than simple tuples.
The second design is dictionaries where each key represents some aspect of a single
object; in this case, we'd probably use a separate dictionary for each object, and
they'd all have similar (though often not identical) sets of keys. This latter situation
may have overlapped with named tuples; named tuples should typically be used
when we know exactly what attributes the data must store, and we know that all
pieces of the data must be supplied at once (when the item is constructed). Further,
named tuples are only useful when we know this data will not change. For other
instances, if we need to build the dictionary keys over a period of time, or each
dictionary may have slightly (or widely) varying sets of keys, or we need to
change values frequently, a dictionary is much better.
Using defaultdict
We've seen how to use setdefault to set a default value if a key doesn't exist, but
this can get a bit monotonous if we need to set a default value every time we look up
a value. For example, if we're writing code that counts the number of times a letter
occurs in a given sentence, we could do this:
def letter_frequency(sentence):
frequencies = {}
for letter in sentence:
frequency = frequencies.setdefault(letter, 0)
frequencies[letter] = frequency + 1
return frequencies
[ 166 ]
Chapter 6
Every time we access the dictionary, we need to check that it has a value already,
and, if not, set it to zero. When something like this needs to be done every time
an empty key is requested, we can use a different version of the dictionary,
called defaultdict:
from collections import defaultdict
frequencies = defaultdict(int)
frequencies[letter] += 1
return frequencies
This code looks like it couldn't possibly work. The defaultdict class accepts a
function in its constructor; whenever a key is accessed that is not already in the
dictionary, it calls that function, with no parameters, to create a default value.
In this case, the function it calls is int, which is actually the constructor for an integer
object. Normally integers are created simply by typing an integer number into our
code, and if we do create one using the int constructor, we pass it the item we want
to create (for example, to convert a string of digits into an integer). But if we call int
without any arguments, it returns, conveniently, the number zero. In this code, if the
letter doesn't exist in the defaultdict, the number zero is returned when we access
it. Then we add one to this number to indicate we've found another instance of that
letter, and the next time we find one, we increment the value again.
The defaultdict class is useful for creating containers. If we want to create a
dictionary of stock prices for the past 30 days, we could use a stock symbol as the
key, and store the prices in a list; the first time we access the stock price, we would
want it to create an empty list. Simply pass list into the defaultdict constructor,
and it will be called every time an empty key is accessed. We can do similar things
with sets or even empty dictionaries if we want to associate one with a key.
Of course, we can also write our own functions and pass them into the defaultdict
constructor. Suppose we want to create a defaultdict where each new element
contains a tuple of the number of items inserted into the dictionary at that time and
an empty list to hold other things. Nobody knows why we would want to create
such an object, but let's have a look:
num_items = 0
def tuple_counter():
global num_items
num_items += 1
return (num_items, [])
d = defaultdict(tuple_counter)
[ 167 ]
When we run this code, we can access empty keys and insert into the list all in
one statement:
>>> d = defaultdict(tuple_counter)
>>> d['a'][1].append("hello")
>>> d['b'][1].append('world')
>>> d
defaultdict(<function tuple_counter at 0x82f2c6c>,
{'a': (1, ['hello']), 'b': (2, ['world'])})
When we print the dict at the end, we see that the counter really was working.
This example, while succinctly demonstrating how to create our own
function for defaultdict, is not actually very good code; using a global
variable means that if we created four different defaultdict segments
that each used a tuple_counter, it would count the number of entries
in all dictionaries, rather than having a different count for each one.
It would be better to create a class and pass a method on that class to
defaultdict.
Lists
Lists are the least object-oriented of Python's data structures. While lists are,
themselves, objects, there is a lot of syntax in Python to make using them as painless
as possible. Unlike many other object-oriented languages, lists in Python are simply
available. We don't need to import them and rarely need to call methods on them.
We can loop over a list without explicitly requesting an iterator object, and we can
construct a list (like a dictionary) with custom syntax. Further, list comprehensions
and generator expressions turn them into a veritable Swiss-army knife of
functionality.
We won't go into too much detail of the syntax; you've seen it in introductory
tutorials across the web and previous examples in this book. You can't code Python
very long without learning how to use lists! Instead, we'll be covering when lists
should be used, and their nature as objects. If you don't know how to create or
append to a list, how to retrieve items from a list, or what "slice notation" is,
I direct you to the official Python tutorial, post-haste. It can be found online at:
http://docs.python.org/py3k/tutorial/
[ 168 ]
Chapter 6
Lists, in Python, should normally be used when we want to store several instances
of the "same" type of object; lists of strings or lists of numbers; most often, lists of
objects we've defined ourselves. Lists should always be used when we want to store
items in some kind of order. Often, this is the order in which they were inserted, but
they can also be sorted by some criteria.
As we saw in the case study from the previous chapter, lists are also very useful
when we need to modify the contents: insert to or delete from an arbitrary location
of the list, or update a value within the list.
Like dictionaries, Python lists use an extremely efficient and well-tuned internal data
structure so we can worry about what we're storing, rather than how we're storing
it. Many object-oriented languages provide different data structures for queues,
stacks, linked lists, and array-based lists. Python does provide special instances of
some of these classes, if optimizing access to huge sets of data is required. Normally,
however, the list data structure can serve all these purposes at once, and the coder
has complete control over how they access it.
Don't use lists for collecting different attributes of individual items. We do not
want, for example, a list of the properties a particular shape has. Tuples, named
tuples, dictionaries, and objects would all be more suitable for this purpose. In some
languages, they might create a list in which each alternate item is a different type;
for example, they might write ['a', 1, 'b', 3] for our letter frequency list. They'd
have to use a strange loop that accesses two elements in the list at once, or a modulus
operator to determine which position was being accessed.
Don't do this in Python. We can group related items together using a dictionary,
as we did in the previous section (if sort order doesn't matter), or using a list of
tuples. Here's a rather convoluted example that demonstrates how we could do the
frequency example using a list. It is much more complicated than the dictionary
examples, and illustrates how much of an effect choosing the right (or wrong) data
structure can have on the readability of our code.
import string
CHARACTERS = list(string.ascii_letters) + [" "]
frequencies = [(c, 0) for c in CHARACTERS]
index = CHARACTERS.index(letter)
frequencies[index] = (letter,frequencies[index][1]+1)
return frequencies
[ 169 ]
This code starts with a list of possible characters. The string.ascii_letters

attribute provides a string of all the letters, lower and upper case, in order. We
convert this to a list, and then use list concatenation (the plus operator causes two
lists to be merged into one) to add one more character, the space. These are the
available characters in our frequency list (the code would break if we tried to add a
letter that wasn't in the list, but an exception handler could solve this).
The first line inside the function uses a list comprehension to turn the CHARACTERS
list into a list of tuples. List comprehensions are an important, non-object-oriented
tool in Python; we'll be covering them in detail in the next chapter.
Then we loop over each of the characters in the sentence. We first look up the index
of the character in the CHARACTERS list, which we know has the same index in our
frequencies list, since we just created the second list from the first. We then update
that index in the frequencies list by creating a new tuple, discarding the original one.
Aside from the garbage collection and memory waste concerns, this is rather difficult
to read!
The resulting code works, but is not nearly so elegant as the dictionary. The code has
two advantages over the earlier dictionary example, however. The list stores zero
frequencies for characters not in the sentence, and when we receive the list, it comes
in sorted order. The output shows the difference:
>>> letter_frequency("the quick brown fox jumps over the lazy dog")
[('a', 1), ('b', 1), ('c', 1), ('d', 1), ('e', 3), ('f', 1), ('g', 1),
('h', 2), ('i', 1), ('j', 1), ('k', 1), ('l', 1), ('m', 1), ('n', 1),
('o', 4), ('p', 1), ('q', 1), ('r', 2), ('s', 1), ('t', 2), ('u', 2),
('v', 1), ('w', 1), ('x', 1), ('y', 1), ('z', 1), ('A', 0), ('B', 0),
('C', 0), ('D', 0), ('E', 0), ('F', 0), ('G', 0), ('H', 0), ('I', 0),
('J', 0), ('K', 0), ('L', 0), ('M', 0), ('N', 0), ('O', 0), ('P', 0),
('Q', 0), ('R', 0), ('S', 0), ('T', 0), ('U', 0), ('V', 0), ('W', 0),
('X', 0), ('Y', 0), ('Z', 0), (' ', 8)]
The dictionary version could be adapted to provide these advantages by

pre-populating the dictionary with zero values for all available characters, and
by sorting the keys on the returned dictionary whenever we need them in order.
Like dictionaries, lists are objects too, and they have several methods that can be
invoked upon them. The most common is append(element), which adds an element
to the list. Similarly, insert(index, element) inserts an item at a specific position.
The count(element) method tells us how many times an element appears in the
list, and index()as we saw in the previous exampletells us the index of an
item in the list. The reverse() method does exactly what it says: turning the list
around. The sort() method is also obvious, but it has some fairly complicated
object-oriented behaviors, which we'll cover now.
[ 170 ]
Chapter 6
Sorting lists
Without any parameters, sort will generally do the expected thing. If it's a list of
strings, it will place them in alphabetical order. This operation is case sensitive, so all
capital letters will be sorted before lower case letters, that is Z comes before a. If it is a
list of numbers, they will be sorted in numerical order. If a list of tuples is provided,
the list is sorted by the first element in each tuple. If a mixture of unsortable items is
supplied, the sort will raise a TypeError exception.
If we want to place objects we define ourselves into a list and make those objects
sortable, we have to do a bit more work. The special method __lt__, which stands
for "less than", should be defined on the class to make instances of that class
comparable. The sort method on list will access this method on each object to
determine where it goes in the list. This method should return True if our class is
somehow less than the passed parameter, and False otherwise. Here's a rather silly
class that can be sorted based on either a string or a number:
class WeirdSortee:
def __init__(self, string, number, sort_num):
self.string = string
self.number = number
self.sort_num = sort_num
def __lt__(self, object):
if self.sort_num:
return self.number < object.number
return self.string < object.string
def __repr__(self):
return"{}:{}".format(self.string, self.number)
The __repr__ method makes it easy to see the two values when we print a list. This
__lt__ implementation compares the object to another instance of the same class (or
any duck typed object that has string, number, and sort_num attributes; it will fail
if those attributes are missing). The following output illustrates this class in action,
when it comes to sorting:
>>> a = WeirdSortee('a', 4, True)
>>> b = WeirdSortee('b', 3, True)
>>> c = WeirdSortee('c', 2, True)
>>> d = WeirdSortee('d', 1, True)
>>> l = [a,b,c,d]
>>> l
[ 171 ]

[a:4, b:3, c:2, d:1]
>>> l.sort()
>>> l
[d:1, c:2, b:3, a:4]
>>> for i in l:
...
i.sort_num = False
...
>>> l.sort()
>>> l
[a:4, b:3, c:2, d:1]
The first time we call sort, it sorts by numbers, because sort_num is True on all the
objects being compared. The second time, it sorts by letters. The __lt__ method is
the only one we need to implement to enable sorting. Technically, however, if it is
implemented, the class should normally also implement the similar __gt__, __eq__,
__ne__, __ge__, and __le__ methods, so that all of the <, >, ==, !=, >=, and <=
operators also work properly.
The sort method can also take an optional key argument. This argument is a
function that can transform each object in a list into an object that can be somehow
compared. This is useful if we have a tuple of values and want to sort on the second
item in the tuple rather than the first (which is the default for sorting tuples):
>>> x = [(1,'c'), (2,'a'), (3, 'b')]
>>> x.sort()
>>> x
[(1, 'c'), (2, 'a'), (3, 'b')]
>>> x.sort(key=lambda i: i[1])
>>> x
[(2, 'a'), (3, 'b'), (1, 'c')]
The lambda keyword in the command line creates a function that takes a tuple as
input and uses sequence lookups to return the item with index 1 (that is the second
item in the tuple).
As another example, we can also use the key parameter to make a sort case
insensitive. To do this, we simply need to compare the all lowercase versions
of strings, so we can pass the built-in str.lower function as the key function:
>>> l = ["hello", "HELP", "Helo"]
>>> l.sort()
>>> l
[ 172 ]
Chapter 6
['HELP', 'Helo', 'hello']
>>> l.sort(key=str.lower)
>>> l
['hello', 'Helo', 'HELP']
Remember, even though lower is a method on string objects, it is also a function that
can accept a single argument, self. In other words, str.lower(item) is equivalent
to item.lower(). When we pass this function as a key, it performs the comparison
on lowercase values instead of doing the default case-sensitive comparison.
Sets
Lists are extremely versatile tools that suit most container object applications.
But they are not useful when we want to ensure objects in the list are unique. For
example, a song library may contain many songs by the same artist. If we want to
sort through the library and create a list of all the artists, we would have to check
the list to see if we've added the artist already before we add them again.
This is where sets come in. Sets come from mathematics, where they represent an
unordered group of (usually) unique numbers. We can add a number to a set five
times, but it will show up in the set only once.
In Python, sets can hold any hashable object, not just numbers. Hashable objects
are the same objects that can be used as keys in dictionaries, so again, lists and
dictionaries are out. Like mathematical sets, they can store only one copy of each
object. So if we're trying to create a list of song artists, we can create a set of string
names and simply add them to the set. This example starts with a list of (song, artist)
tuples and creates a set of the artists:
song_library = [("Phantom Of The Opera", "Sarah Brightman"),
("Knocking On Heaven's Door", "Guns N' Roses"),
("Captain Nemo", "Sarah Brightman"),
("Patterns In The Ivy", "Opeth"),
("November Rain", "Guns N' Roses"),
("Beautiful", "Sarah Brightman"),
("Mal's Song", "Vixy and Tony")]
artists = set()
for song, artist in song_library:
artists.add(artist)
print(artists)
[ 173 ]
There is no built-in syntax for an empty set as there is for lists and dictionaries; we
create a set using the set() constructor. However, we can use the curly braces of
dictionary syntax to create a set, so long as the set contains values. If we use colons
to separate pairs of values, it's a dictionary, as in {'key': 'value', 'key2':
'value2'}. If we just separate values with commas, it's a set, as in {'value',
'value2'}. Items can be added individually to the set using its add method.
If we run this script, we see that the set works as advertised:
{'Sarah Brightman', "Guns N' Roses", 'Vixy and Tony', 'Opeth'}
If you're paying attention to the output, you'll notice that the items are not printed
in the order they were added to the sets. Sets, like dictionaries, are unordered. They
both use an underlying hash-based data structure for efficiency. Because they are
unordered, sets cannot have items looked up by index. The primary purpose of a set
is to divide the world into two groups: "things that are in the set", and, "things that
are not in the set". It is easy to check if an item is in the set or to loop over the items
in a set, but if we want to sort or order them, we'll have to convert the set to a list.
This output shows all three of these activities:
>>> "Opeth" in artists
True
>>> for artist in artists:
...
print("{} plays good music".format(artist))
...
Sarah Brightman plays good music
Guns N' Roses plays good music
Vixy and Tony play good music
Opeth plays good music
>>> alphabetical = list(artists)
>>> alphabetical.sort()
>>> alphabetical
["Guns N' Roses", 'Opeth', 'Sarah Brightman', 'Vixy and Tony']
While the primary feature of a set is uniqueness, that is not its primary purpose.
Sets are most useful when two or more of them are used in combination. Most of the
methods on the set type operate on other sets, allowing us to efficiently combine or
compare the items in two or more sets. These methods have strange names if you're
not familiar with mathematical sets, since they use the same terminology used in
mathematics. We'll start with three methods that return the same result regardless
of which is the calling set and which is the called set.
[ 174 ]
Chapter 6
The union method is the most common and easiest to understand. It takes a second
set as a parameter and returns a new set that contains all elements that are in either
of the two sets; if an element is in both original sets, it will, of course, only show up
once in the new set. Union is like a logical or operation, indeed, the | operator can be
used to get the same effect, if you don't like calling methods.
Conversely, the intersection method accepts a second set and returns a new
set that contains only those elements that are in both sets. It is like a logical and
operation, and can also be referenced using the & operator.
Finally, the symmetric_difference method tells us what's left; it is the set of objects
that are in one set or the other, but not both. The following example illustrates these
methods by comparing some artists from my song library to those in my sister's:
my_artists = {"Sarah Brightman", "Guns N' Roses",
"Opeth", "Vixy and Tony"}
auburns_artists = {"Nickelback", "Guns N' Roses",
"Savage Garden"}
print("All: {}".format(my_artists.union(auburns_artists)))
print("Both: {}".format(auburns_artists.intersection(my_artists)))
print("Either but not both: {}".format(
my_artists.symmetric_difference(auburns_artists)))
If we run this code, we see that these three methods do what the print statements
suggest they will do:
All: {'Sarah Brightman', "Guns N' Roses", 'Vixy and Tony',
'Savage Garden', 'Opeth', 'Nickelback'}
Both: {"Guns N' Roses"}
Either but not both: {'Savage Garden', 'Opeth', 'Nickelback',
'Sarah Brightman', 'Vixy and Tony'}
These methods all return the same result regardless of which set calls the other. We
can say my_artists.union(auburns_artists) or auburns_artists.union(my_
artists) and get the same result. There are also methods that return different
results depending on who is the caller and who is the argument.
[ 175 ]
These methods include issubset and issuperset, which are the inverse of each
other. Both return a boolean. The issubset method returns True, if all of the items
in the calling set are also in the set passed as an argument. The issuperset method
returns True, if all of the items in the argument are also in the calling set. Thus
s.issubset(t) and t.issuperset(s) are identical. They will both return True if t
contains all the elements in s.
Finally, the difference method returns all the elements that are in the calling set,
but not in the set passed as an argument; this is like half a symmetric_difference.
The difference method can also be represented by the - operator. The following code
illustrates these methods in action:
my_artists = {"Sarah Brightman", "Guns N' Roses",
"Opeth", "Vixy and Tony"}
bands = {"Guns N' Roses", "Opeth"}
print("my_artists is to bands:")
print("issuperset: {}".format(my_artists.issuperset(bands)))
print("issubset: {}".format(my_artists.issubset(bands)))
print("difference: {}".format(my_artists.difference(bands)))
print("*"*20)
print("bands is to my_artists:")
print("issuperset: {}".format(bands.issuperset(my_artists)))
print("issubset: {}".format(bands.issubset(my_artists)))
print("difference: {}".format(bands.difference(my_artists)))
This code simply prints out the response of each method when called from one set on
the other. Running it gives us the following output:
my_artists is to bands:
issuperset: True
issubset: False
difference: {'Sarah Brightman', 'Vixy and Tony'}
********************
bands is to my_artists:
issuperset: False
issubset: True
difference: set()
The difference method, in the second case, returns an empty set, since there are no
items in bands that are not in my_artists.
[ 176 ]
Chapter 6
The union, intersection, and difference methods can all take multiple sets as
arguments; they will return, as we might expect, the set that is created when the
operation is called on all the parameters.
So the methods on sets clearly suggest that sets are meant to operate on other sets,
and that they are not just containers. If we have data coming in from two different
sources and need to quickly combine them in some way, to determine where the data
overlaps, or is different, we can use set operations to efficiently compare them. Or
if we have data incoming that may contain duplicates of data that has already been
processed, we can use sets to compare the two and process only the new data.
Extending built-ins
We discussed briefly in Chapter 3 how built-in data types can be extended using
inheritance. Now, we'll go into more detail as to when we would want to do that.
When we have a built-in container object that we want to add functionality to, we
have two options. We can either create a new object, which holds that container as
an attribute (composition), or we can subclass the built-in object and add or adapt
methods on it to do what we want (inheritance).
Composition is usually the best alternative if all we want to do is use the container
to store some objects using that container's features. That way, it's easy to pass that
data structure into other methods and they will know how to interact with it. But we
need to use inheritance if we want to change the way the container actually works.
For example, if we want to ensure every item in a list is a string with exactly five
characters, we need to extend list and override the append() method to raise an
exception for invalid input. We'd also have to override __setitem__(self, index,
value), a special method on lists that is called whenever we use the x[index] =
"value" syntax.
That's right, all that special non-object-oriented looking syntax we've been looking
at for accessing lists, dictionary keys, looping over containers, and similar tasks is
actually "syntactic sugar" that maps to an object-oriented paradigm underneath.
We might ask the Python designers why they did this, when common perception
suggests that object-oriented programming is always better. That question is easy
to answer. In the following hypothetical examples, which is easier to read, as a
programmer? Which requires less typing?:
c = a + b
c = a.add(b)
l[0] = 5
l.setitem(0, 5)
[ 177 ]

d[key] = value
d.setitem(key, value)
for x in alist:
#do something with x
it = alist.iterator()
while it.has_next():
x = it.next()
#do something with x
The highlighted sections show what object-oriented code might look like (in practice,
these methods actually exist as special double-underscore methods on associated
objects). Python programmers agree that the non-object-oriented syntax is easier to
read and to write. Non-Python programmers say that syntax like this means Python
is not object-oriented. That, however, is hogwash. All of the above Python syntaxes
map to object-oriented methods underneath the hood. These methods have special
names (with double-underscores before and after) to remind us that there is a better
syntax out there. However, we now have the means to override these behaviors. For
example, we can make a special integer that always returns 0 when we add two of
them together:
class SillyInt(int):
def __add__(self, num):
return 0
This is a very strange thing to do, granted, but it illustrates perfectly the objectoriented principles in action. And now we have an argument when people tell us
Python isn't truly object-oriented. It's just object-oriented that has been made easy to
work with. Check out the above class in action:
>>> a = SillyInt(1)
>>> b = SillyInt(2)
>>> a + b
0
The awesome thing about the __add__ method is that we can add it to any class we
write, and if we use the + operator on instances of that class, it will be called. This is
how string, tuple, and list concatenation works.
This is true of all the special methods. If we want to use x in myobj syntax, we can
override __contains__. If we want to use myobj[i] = value syntax, we implement
__setitem__ and if we want to use something = myobj[i], we implement
__getitem__.
[ 178 ]
Chapter 6
There are thirty-three of these special methods on the list class. We can use the dir
function to see all of them:
>>> dir(list)
['__add__', '__class__', '__contains__', '__delattr__','__delitem__', '__
doc__', '__eq__', '__format__', '__ge__', '__getattribute__', '__getitem_
_', '__gt__', '__hash__', '__iadd__', '__imul__', '__init__', '__iter__',
'__le__', '__len__', '__lt__', '__mul__', '__ne__', '__new__', '__reduce_
_', '__reduce_ex__', '__repr__', '__reversed__', '__rmul__', '__setattr__
', '__setitem__', '__sizeof__', '__str__', '__subclasshook__', 'append',
'count', 'extend', 'index', 'insert', 'pop', 'remove', 'reverse', 'sort'
Further, if we want any additional information on how any of these methods works,
we can use the help function:
>>> help(list.__add__)
Help on wrapper_descriptor:
__add__(...)
x.__add__(y) <==> x+y
The plus operator on lists concatenates two lists. We don't have room to discuss all
of the available special functions in this book, but you are now able to explore all this
functionality with dir and help. The official online Python reference (http://docs.
python.org/) has plenty of useful information as well. Focus, especially, on the
abstract base classes discussed in the collections module.
So to get back to the earlier point about when we would want to use composition
versus inheritance: if we need to somehow change any of the methods on the class,
including the special methods we definitely need to use inheritance. If we used
composition, we could write methods that do the validation or alterations and ask
the caller to use those methods, but there is nothing stopping them from accessing
the property directly (no private members, remember?). They could insert an item
into our list that does not have five characters, and that might confuse other methods
in the list.
Often, the need to extend a built-in data type is an indication that we're using the
wrong sort of data type. It is not always the case, but if you're suddenly looking to
extend a built-in, carefully consider whether or not a different data structure would
be more suitable.
[ 179 ]
As a last example, let's consider what it takes to create a dictionary that remembers
the order in which keys were inserted. One way (likely not the best way) to do this
is to keep an ordered list of keys that is stored in a specially derived subclass of
dict. Then we can override the methods keys, values, __iter__, and items to
return everything in order. Of course, we'll also have to override __setitem__ and
setdefault to keep our list up to date. There are likely to be a few other methods
in the output of dir(dict) that need overriding to keep the list and dictionary
consistent (clear and __delitem__ come to mind, to track when items are
removed), but we won't worry about them for this example.
So we'll be extending dict and adding a list of ordered keys. Trivial enough, but
where do we create the actual list? We could include it in the __init__ method,
which would work just fine, but we have no guarantees that any subclass will call
that initializer. Remember the __new__ method we discussed in Chapter 2? I said it
was generally only useful in very special cases. This is one of those special cases.
We know __new__ will be called exactly once, and we can create a list on the new
instance that will always be available to our class. With that in mind, here is our
entire sorted dictionary:
from collections import KeysView, ItemsView, ValuesView
class DictSorted(dict):
def __new__(*args, **kwargs):
new_dict = dict.__new__(*args, **kwargs)
new_dict.ordered_keys = []
return new_dict
def __setitem__(self, key, value):
'''self[key] = value syntax'''
if key not in self.ordered_keys:
self.ordered_keys.append(key)
super().__setitem__(key, value)
def setdefault(self, key, value):
if key not in self.ordered_keys:
self.ordered_keys.append(key)
return super().setdefault(key, value)
def keys(self):
return KeysView(self)
def values(self):
return ValuesView(self)
def items(self):
[ 180 ]
Chapter 6
return ItemsView(self)
def __iter__(self):
'''for x in self syntax'''
return self.ordered_keys.__iter__()
The __new__ method simply creates a new dictionary and then puts an empty list
on that object. We don't override __init__, as the default implementation works
(actually, this is only true if we initialize an empty DictSorted object, which is
standard behavior. If we want to support other variations of the dict constructor,
which accept dictionaries or lists of tuples, we'd need to fix __init__ to also update
our ordered_keys). The two methods for setting items are very similar; they both
update the list of keys, but only if the item hasn't been added before. We don't want
duplicates in the list, but we can't use a set here; it's unordered!
The keys, items, and values methods all return views onto the dictionary. The
collections library provides three read-only View objects onto the dictionary; they use
the __iter__ method to loop over the keys, and then use __getitem__ (which we
didn't need to override) to retrieve the values. So we only need to define our custom
__iter__ method to make these three views work. You would think the superclass
would do to create these views properly using polymorphism, but if we don't
override these three methods, they don't return properly ordered views.
Finally, the __iter__ method is the really special one; it ensures that if we loop
over the dictionary's keys (using for...in syntax), it will return the values in the
correct order. It simply does this by returning the __iter__ of the ordered_keys
list, which returns the same iterator object that would be used if we used for...in on
the list instead. Since ordered_keys is a list of all available keys (due to the way we
overrode other methods), this is the correct iterator object for the dictionary as well.
Let's look at a few of these methods in action, compared to a normal dictionary:
>>> ds = DictSorted()
>>> d = {}
>>> ds['a'] = 1
>>> ds['b'] = 2
>>> ds.setdefault('c', 3)
3
>>> d['a'] = 1
>>> d['b'] = 2
>>> d.setdefault('c', 3)
3
>>> for k,v in ds.items():
[ 181 ]

...
print(k,v)
...
a 1
b 2
c 3
>>> for k,v in d.items():
...
print(k,v)
...
a 1
c 3
b 2
Ah, our dictionary is sorted and the normal dictionary is not. Hurray!
If you wanted to use this class in production, you'd have to override
several other methods to ensure the keys are up-to-date in all cases.
However, you don't need to do this; the functionality this class provides
is already available in Python, using the OrderedDict object in the
collections module. Try importing the class from collections, and
use help(OrderedDict) to find out more about it.
Case study
To tie everything together, we'll be writing a simple link collector, which will visit
a website and collect every link on every page it finds in that site. Before we start,
though, we'll need some test data to work with. Simply write some HTML files
to work with that contain links to each other and to other sites on the internet,
something like this:
<html>
<body>
<a href="contact.html">Contact us</a>
<a href="blog.html">Blog</a>
<a href="esme.html">My Dog</a>
<a href="/hobbies.html">Some hobbies</a>
<a href="/contact.html">Contact AGAIN</a>
<a href="http://www.archlinux.org/">Favorite OS</a>
</body>
</html>
[ 182 ]
Chapter 6
Name one of the files index.html so it shows up first when pages are served. Make
sure the other files exist, and keep things complicated so there is lots of linking
between them. The examples for this chapter include a directory called case_study_
serve (one of the lamest personal websites in existence!) if you'd rather not set them
up yourself.
Now, start a simple web server by entering the directory containing all these files
and run the following command:
python3 -m http.server
This will start a server running on port 8000; you can see the pages you made by
visiting http://localhost:8000/ in your web browser.
I doubt anyone can get a website up and running with less work! Never
let it be said, "you can't do that easily with Python".
The goal will be to pass our collector the base URL for the site (in this case:
http://localhost:8000/), and have it create a list containing every unique link
on the site. We'll need to take into account three types of URLs (links to external
sites, which start with http://, absolute internal links, which start with a /, and
relative links, for everything else). We also need to be aware that pages may link to
each other in a loop; we need to be sure we don't process the same page multiple
times, or it may never end. With all this uniqueness going on, it sounds like we're
going to need some sets.
Before we get into that, let's start with the basics. What code do we need to connect
to a page and parse all the links from that page?:
from urllib.parse import urlparse
import re
import sys
LINK_REGEX = re.compile(
"<a [^>]*href=['\"]([^'\"]+)['\"][^>]*>")
class LinkCollector:
self.url = "http://" + urlparse(url).netloc
def collect_links(self, path="/"):
full_url = self.url + path
page = str(urlopen(full_url).read())
links = LINK_REGEX.findall(page)
[ 183 ]

print(links)
if __name__ == "__main__":
LinkCollector(sys.argv[1]).collect_links()
This is a short piece of code, considering what it's doing. It connects to the server
in the argument passed on the command line, downloads the page, and extracts all
the links on that page. The __init__ method uses the urlparse function to extract
just the hostname from the url; so even if we pass in http://localhost:8000/
some/page.html it will still operate on the top-level of the host, http://
localhost:8000/. This makes sense, because we want to collect all the links
on the site.
The collect_links method connects to and downloads the specified page from
the server, and uses a regular expression to find all the links in the page. Regular
expressions are an extremely powerful string processing tool. Unfortunately, they
have a steep learning curve; if you haven't used them before, I strongly recommend
studying any of the entire books or websites on the topic. If you don't think they're
worth knowing, try writing the above code without them.
The example above stops in the middle of the collect_links method to print the
value of links. This is a common way to test a program as we're writing it: stop
and output the value to ensure it is the value we expect. Here's what it outputs for
our example:
['contact.html', 'blog.html', 'esme.html', '/hobbies.html',
'/contact.html', 'http://www.archlinux.org/']
So now we have a collection of all the links in the first page. What should we do
with it? We can't just pop the links into a set to remove duplicates, because links
may be relative or absolute. For example contact.html and /contact.html point
to the same page. No, the first thing we should do is normalize all the links to
their full URL, including hostname and relative path. We can do this by adding
a normalize_url method to our object:
def normalize_url(self, path, link):
if link.startswith("http://"):
return link
elif link.startswith("/"):
return self.url + link
else:
return self.url + path.rpartition('/'
)[0] + '/' + link
[ 184 ]
Chapter 6
This method converts all URLs to complete URLs with protocol and hostname. Now
the two contact pages have the same value and we can store them in a set. We'll have
to modify __init__ to create the set, and collect_links to put all the links into it.
Then we'll have to visit all the non-external links and collect them too. But wait a
minute, if we do that, how do we keep from revisiting a link when we encounter the
same page twice? It looks like we're actually going to need two sets, a set of collected
links, and a set of visited links. This suggests that we were wise to choose a set to
represent our data; we know that sets are most useful when we're manipulating
more than one of them. Let's set these up:
self.url = "http://+" + urlparse(url).netloc
self.collected_links = set()
self.visited_links = set()
self.visited_links.add(full_url)
links = {self.normalize_url(path, link
) for link in links}
self.collected_links = links.union(
self.collected_links)
unvisited_links = links.difference(
self.visited_links)
print(links, self.visited_links,
self.collected_links, unvisited_links)
The line that creates the normalized list of links uses a set comprehension, no
different from a list comprehension, except that it creates a set out of the values.
We'll be covering these babies in detail in the next chapter. Once again, the method
stops to print out the current values, so we can verify we don't have our sets
confused, and that difference really was the method we wanted to call to
collect unvisited_links. We can then add a few lines of code that loop over
all the unvisited links and adds them to the collection as well:
for link in unvisited_links:
if link.startswith(self.url):
self.collect_links(urlparse(link).path)
[ 185 ]
The if statement ensures that we are only collecting links from the one website; we
don't want to go off and collect all the links from all the pages on the internet (unless
we're Google or the Internet Archive!). If we modify the main code at the bottom of
the program to output the collected links, we can see it seems to have collected
them all:
if __name__ == "__main__":
collector = LinkCollector(sys.argv[1])
collector.collect_links()
for link in collector.collected_links:
print(link)
It displays all the links we've collected, and only once; even though many of the
pages in my example linked to each other multiple times:
$ python3 link_collector.py http://localhost:8000
http://localhost:8000/
http://en.wikipedia.org/wiki/Cavalier_King_Charles_Spaniel
http://masterhelenwu.com
http://archlinux.me/dusty/
http://localhost:8000/blog.html
http://ccphillips.net/
http://localhost:8000/contact.html
http://localhost:8000/taichi.html
http://www.archlinux.org/
http://localhost:8000/esme.html
http://localhost:8000/hobbies.html
Even though it collected links to external pages, it didn't go off collecting links from
any of the external pages we linked to. This is a great little program if we want to
collect all the links in a site. But it doesn't give me all the information I might need
to build a site map; it tells me which pages I have, but it doesn't tell me which pages
link to other pages. If we want to do that instead, we're going to have to make some
modifications.
[ 186 ]
Chapter 6
The first thing we should do is look at our data structures. The set of collected links
doesn't work anymore; we want to know which links were linked to from which
pages. The first thing we could do, then, is turn that set into a dictionary of sets for
each page we visit. The dictionary keys will represent the exact same data that is
currently in the set. The values will be sets of all the links on that page. Here are
the changes:
from urllib.parse import urlparse
import re
import sys
LINK_REGEX = re.compile(
"<a [^>]*href=['\"]([^'\"]+)['\"][^>]*>")
self.url = "http://%s" % urlparse(url).netloc
self.collected_links = {}
self.visited_links = set()
self.visited_links.add(full_url)
links = {self.normalize_url(path, link
) for link in links}
self.collected_links[full_url] = links
for link in links:
self.collected_links.setdefault(link, set())
unvisited_links = links.difference(
self.visited_links)
for link in unvisited_links:
if link.startswith(self.url):
self.collect_links(urlparse(link).path)
def normalize_url(self, path, link):
if link.startswith("http://"):
return link
elif link.startswith("/"):
return self.url + link
else:
return self.url + path.rpartition('/'
)[0] + '/' + link
[ 187 ]

if __name__ == "__main__":
collector = LinkCollector(sys.argv[1])
collector.collect_links()
for link, item in collector.collected_links.items():
print("{}: {}".format(link, item))
It is a surprisingly small change; the line that originally created a union of two sets
has been replaced with three lines that update the dictionary. The first of these
simply tells the dictionary what the collected links for that page are. The second
creates an empty set for any items in the dictionary that have not already been added
to the dictionary, using setdefault. The result is a dictionary that contains all the
links as its keys, mapped to sets of links for all the internal links, and empty sets for
the external links.
Exercises
The best way to learn how to choose the correct data structure is to do it wrong a few
times. Take some code you've recently written, or write some new code that uses a
list. Try rewriting it using some different data structures. Which ones make more
sense? Which don't? Which have the most elegant code?
Try this with a few different pairs of data structures. You can look at examples
you've done for previous chapter exercises. Are there objects with methods where
you could have used a namedtuple or a dict instead? Attempt both and see. Are
there dictionaries that could have been sets because you don't really access the
values? Do you have lists that check for duplicates? Would a set suffice? Or maybe
several sets?
If you want some specific examples to work with, try adapting the link collector to
also save the title used for each link. Perhaps you can generate a site map in HTML
that lists all the pages on the site, and contains a list of links to other pages, named
with the same link titles.
Have you written any container objects recently that you could improve by
inheriting a built-in and overriding some of the "special" double-underscore
methods? You may have to do some research (using dir and help, or the Python
library reference) to find out which methods need overriding. Are you sure
inheritance is the correct tool to apply; could a composition-based solution be more
effective? Try both (if it's possible) before you decide. Try to find different situations
where each method is better than the other.
[ 188 ]
Chapter 6
If you were familiar with the various Python data structures and their uses before
you started this chapter, you may have been bored. But if that is the case, there's a
good chance you use data structures too much! Look at some of your old code and
rewrite it to use more self-made objects. Carefully consider the alternatives, and try
them all out; which one makes for the most readable and maintainable system?
Always critically evaluate your code and design decisions. Make a habit of reviewing
old code and take note if your understanding of "good design" has changed since
you've written it. Software design has a large aesthetic component, and like painters,
we all have to find the style that suits us best.
Summary
We've covered several built-in data structures and attempted to understand how to
choose one for specific applications. Sometimes the best thing we can do is create a
new class of objects, but often, one of the built-ins provides exactly what we need.
When it doesn't, we can always use inheritance or composition to adapt them to our
needs. In particular, we covered:
Dictionaries and default dictionaries
Lists and sets
Overriding special variables on built-ins
In the next chapter we'll discuss how to integrate the object-oriented and not-soobject-oriented aspects of Python. Along the way, we'll discover that it's more
object-oriented than it looks at first sight!
[ 189 ]
Python Object-oriented
Shortcuts
Now let's look at some aspects of Python that appear more reminiscent of
structural or functional programming than object-oriented programming. Although
object-oriented programming is the most popular kid on the block these days, the old
paradigms still offer useful tools. Most of these tools are really syntactic sugar over
an underlying object-oriented implementation; we can think of them as a further
abstraction layer built on top of the (already abstracted) object-oriented paradigm.
In this chapter we'll be covering:
Built-in functions that take care of common tasks in one call
List, set, and dictionary comprehensions
Generators
Functions as objects
Python built-in functions
There are numerous functions in Python that perform a task or calculate a result
on certain objects without being methods on the class. Their purpose is to abstract
common calculations that apply to many types of classes. This is applied duck
typing; these functions accept objects with certain attributes or methods that satisfy
a given interface, and are able to perform generic tasks on the object.
Python Object-oriented Shortcuts
Len
The simplest example is the len() function. This function counts the number of
items in some kind of container object such as a dictionary or list. For example:
>>> len([1,2,3,4])
4
Why don't these objects have a length property instead of having to call a function
on them? Technically, they do. Most objects that len() will apply to have a method
called __len__() that returns the same value. So len(myobj) seems to call
myobj.__len__().
Why should we use the function instead of the method? Obviously the method is a
special method with double-underscores suggesting that we shouldn't call it directly.
There must be an explanation for this. The Python developers don't make such
design decisions lightly.
The main reason is efficiency. When we call __len__ on an object, the object has
to look the method up in its namespace, and, if the special __getattribute__
method (which is called every time an attribute or method on an object is accessed)
is defined on that object, it has to be called as well. Further __getattribute__ for
that particular method may have been written to do something nasty like refusing
to give us access to special methods such as __len__! The len function doesn't
encounter any of this. It actually calls the __len__ function on the underlying class,
so len(myobj) maps to MyObj.__len__(myobj).
Another reason is maintainability. In the future, the Python developers may want to
change len() so that it can calculate the length of objects that don't have a __len__,
for example by counting the number of items returned in an iterator. They'll only
have to change one function instead of countless __len__ methods across the board.
Reversed
The reversed() function takes any sequence as input, and returns a copy of that
sequence in reverse order. It is normally used in for loops when we want to loop
over items from back to front.
Similar to len, reversed calls the __reversed__() function on the class for the
parameter. If that method does not exist, reversed builds the reversed sequence
itself using calls to __len__ and __getitem__. We only need to override
__reversed__ if we want to somehow customize or optimize the process:
[ 192 ]
Chapter 7
normal_list=[1,2,3,4,5]
class CustomSequence():
def __len__(self):
return 5
def __getitem__(self, index):
return "x{0}".format(index)
class FunkyBackwards(CustomSequence):
def __reversed__(self):
return "BACKWARDS!"
for seq in normal_list, CustomSequence(), FunkyBackwards():
print("\n{}: ".format(seq.__class__.__name__), end="")
for item in reversed(seq):
print(item, end=", ")
The for loops at the end print the reversed versions of a normal list, and instances
of the two custom sequences. The output shows that reversed works on all three of
them, but has very different results when we define __reversed__ ourselves:
list: 5, 4, 3, 2, 1,
CustomSequence: x4, x3, x2, x1, x0,
FunkyBackwards: B, A, C, K, W, A, R, D, S, !,
Note: the above two classes aren't very good sequences, as they don't
define a proper version of __iter__ so a forward for loop over them
will never end.
Enumerate
Sometimes when we're looping over an iterable object in a for loop, we want access
to the index (the current position in the list) of the current item being processed.
The for loop doesn't provide us with indexes, but the enumerate function gives us
something better: it creates a list of tuples, where the first object in each tuple is the
index and the second is the original item.
[ 193 ]
This is useful if we want to use index numbers directly. Consider some simple code
that outputs all the lines in a file with line numbers:
import sys
filename = sys.argv[1]
with open(filename) as file:
for index, line in enumerate(file):
print("{0}: {1}".format(index+1, line), end='')
Running this code on itself as the input file shows how it works:
1: import sys
2: filename = sys.argv[1]
3:
4: with open(filename) as file:
5:
for index, line in enumerate(file):
6:
print("{0}: {1}".format(index+1, line), end='')
The enumerate function returns a list of tuples, our for loop splits each tuple into
two values, and the print statement formats them together. It adds one to the index
for each line number, since enumerate, like all sequences is zero based.
Zip
The zip function is one of the least object-oriented functions in Python's collection.
It takes two or more sequences and creates a new sequence of tuples. Each tuple
contains one element from each list.
This is easily explained by an example; let's look at parsing a text file. Text data is
often stored in tab-delimited format, with a "header" row as the first line in the file,
and each line below it describing data for a unique record. A simple contact list in
tab-delimited format might look like this:
first
john
jane
david
last email
smith [email protected]
doan [email protected]
neilson
[email protected]
A simple parser for this file can use zip to create lists of tuples that map headers to
values. These lists can be used to create a dictionary, a much easier object to work
with in Python than a file!
import sys
[ 194 ]
Chapter 7
contacts = []
header = file.readline().strip().split('\t')
for line in file:
line = line.strip().split('\t')
contact_map = zip(header, line)
contacts.append(dict(contact_map))
for contact in contacts:
print("email: {email} -- {last}, {first}".format(
**contact))
What's actually happening here? First we open the file, whose name is provided
on the command line, and read the first line. We strip the trailing newline, and
split what's left into a list of three elements. We pass '\t' into the strip method to
indicate that the string should be split at tab characters. The resulting header list
looks like ["first", "last", "email"].
Next, we loop over the remaining lines in the file (after the header). We split each
line into three elements. Then, we use zip to create a sequence of tuples for each
line. The first sequence would look like [("first", "john"), ("last", "smith"),
("email", "[email protected]")].
Pay attention to what zip is doing. The first list contains headers; the second contains
values. The zip function created a tuple of header/value pairs for each matchup.
The dict constructor takes the list of tuples, and maps the first element to a key and
the second to a value to create a dictionary. The result is added to a list.
At this point, we are free to use dictionaries to do all sorts of contact-related
activities. For testing, we simply loop over the contacts and output them in a
different format. The format line, as usual, takes variable arguments and keyword
arguments. The use of **contact automatically converts the dictionary to a bunch
of keyword arguments (we'll understand this syntax before the end of the chapter)
Here's the output:
email: [email protected] -- smith, john
email: [email protected] -- doan, jane
email: [email protected] -- neilson, david
If we provide zip with lists of different lengths, it will stop at the end of the shortest
list. There aren't many useful applications of this feature, but zip will not raise an
exception if that is the case. We can always check the list lengths and add empty
values to the shorter list, if necessary.
[ 195 ]
The zip function is actually the inverse of itself. It can take multiple sequences and
combine them into a single sequence of tuples. Because tuples are also sequences,
we can "unzip" a zipped list of tuples by zipping it again. Huh? Have a look at
this example:
>>> list_one = ['a', 'b', 'c']
>>> list_two = [1, 2, 3]
>>> zipped = zip(list_one, list_two)
>>> zipped = list(zipped)
>>> zipped
[('a', 1), ('b', 2), ('c', 3)]
>>> unzipped = zip(*zipped)
>>> list(unzipped)
[('a', 'b', 'c'), (1, 2, 3)]
First we zip the two lists and convert the result into a list of tuples. We can then use
parameter unpacking to pass these individual sequences as arguments to the zip
function. zip matches the first value in each tuple into one sequence and the second
value into a second sequence; the result is the same two sequences we started with!
Other functions
Another key function is sorted(), which takes an iterable as input, and returns a
list of the items in sorted order. It is very similar to the sort() method on lists, the
difference being that it works on all iterables, not just lists.
Like list.sort, sorted accepts a key argument that allows us to provide a function
to return a sort value for each input. It can also accept a reverse argument.
Three more functions that operate on sequences are min, max, and sum. These each
take a sequence as input, and return the minimum or maximum value, or the sum of
all values in the sequence. Naturally, sum only works if all values in the sequence are
numbers. The max and min functions use the same kind of comparison mechanism as
sorted and list.sort, and allow us to define a similar key function. For example,
the following code uses enumerate, max, and min to return the indices of the values
in a list with the maximum and minimum value:
def min_max_indexes(seq):
minimum = min(enumerate(seq), key=lambda s: s[1])
maximum = max(enumerate(seq), key=lambda s: s[1])
return minimum[0], maximum[0]
[ 196 ]
Chapter 7
The enumerate call converts the sequence into (index, item) tuples. The lambda
function passed in as a key tells the function to search the second item in each
tuple (the original item). The minimum and maximum variables are then set to the
appropriate tuples returned by enumerate. The return statement takes the first
value (the index from enumerate) of each tuple and returns the pair. The following
interactive session shows how the returned values are, indeed, the indices of the
minimum and maximum values:
>>> alist = [5,0,1,4,6,3]
>>> min_max_indexes(alist)
(1, 4)
>>> alist[1], alist[4]
(0, 6)
We've only touched on a few of the more important Python built-in functions. There
are numerous others in the standard library, including:
all and any, which accept an iterable and returns True if all, or any, of the
items evaluate to true (that is a non-empty string or list, a non-zero number,

an object that is not None, or the literal True).
eval, exec, and compile, which execute string as code inside the interpreter.
hasattr, getattr, setattr, and delattr, which allow attributes on an

object to be manipulated as string names.
And many more! See the interpreter help documentation for each of the
functions listed in dir(__builtins__).
Comprehensions
We've already seen a lot of Python's for loop. It allows us to loop over any object
that supports the iterable protocol and do something specific with each of the
elements in turn.
Supporting the iterable protocol simply means an object has an __iter__ method
that returns another object that supports the iterator protocol. Supporting the iterator
protocol is a fancy way of saying it has a __next__ method that either returns the
next object in the sequence, or raises a StopIteration exception when all objects
have been returned.
[ 197 ]
As you can see, the for statement, in spite of not looking terribly object-oriented,
is actually a shortcut to some extremely object-oriented designs. Keep this in mind
as we discuss comprehensions, as they, too, appear to be the polar opposite of an
object-oriented tool. Yet, they use the same iteration protocol as for loops. They're
just another kind of shortcut.
List comprehensions
List comprehensions are one of the most powerful tools in Python, so people tend
to think of them as advanced. They're not. Indeed, I've taken the liberty of littering
previous examples with comprehensions and assuming you'd understand them.
While it's true that advanced programmers use comprehensions a lot, it's not because
they're advanced, it's because they're trivial, and handle some of the most common
operations in programming.
Let's have a look at one of those common operations, namely, converting a list of
items into a list of related items. Specifically, let's assume we just read a list of strings
from a file, and now we want to convert it to a list of integers; we know every item in
the list is an integer, and we want to do some activity (say, calculate an average) on
those numbers. Here's one simple way to approach it:
input_strings = ['1', '5', '28', '131', '3']
output_integers = []
for num in input_strings:
output_integers.append(int(num))
This works fine, it's only three lines of code. If you aren't used to comprehensions,
you may not even think it looks ugly! Now, look at the same code using a list
comprehension:
input_strings = ['1', '5', '28', '131', '3']
output_integers = [int(num) for num in input_strings]
We're down to one line and we've dropped an append method call. Overall, it's
pretty easy to tell what's going on, even if you're not used to comprehension syntax.
The square brackets show we're creating a list. Inside this list is a for loop that loops
over each item in the input sequence. The only thing that may be confusing is what's
happening between the list's opening brace and the start of the for loop. Whatever
happens here is applied to each of the items in the input list. The item in question is
referenced by the num variable from the loop. So it's converting each such item to
an int.
[ 198 ]
Chapter 7
That's all there is to a basic list comprehension. They are not so advanced, after all!
Comprehensions are highly optimized; list comprehensions are far faster than for
loops when we are looping over a huge number of items. If readability alone isn't a
convincing reason to use them as much as possible, then speed should be.
Converting one list of items into a related list isn't the only thing we can do with a
list comprehension. We can also choose to exclude certain values by adding an if
statement inside the comprehension. Have a look:
output_ints = [int(n) for n in input_strings if len(n) < 3]
I shortened the name of the variable from num to n and the result variable to output_
ints so it would still fit on one line. Other than that, all that's different between this
example and the previous one is the if len(n) < 3 part. This extra code excludes any
strings with more than two characters. The if statement is applied before the int
function, so it's testing the length of a string. Since our input strings are all integers
at heart, it refers to any number over ninety-nine. Now that is all there is to list
comprehensions! We use them to map input values to output values, applying a
filter along the way to exclude any values that don't meet a specific condition.
Any iterable can be the input to a list comprehension; anything we can wrap in
a for loop can also be placed inside a comprehension. For example, text files are
iterable; each call to __next__ on the file's iterator will return one line of the file.
The contact file example we used earlier (to try out the zip function) can use a list
comprehension instead:
import sys
header = file.readline().strip().split('\t')
contacts = [
dict(
zip(header, line.strip().split('\t'))
) for line in file
]
This time, I've added some whitespace to make it more readable (list comprehensions
don't have to fit on one line). This example is doing the same thing as the previous
version: creating a list of dictionaries from the zipped header and split lines for each
line in the file.
[ 199 ]
Er, what? Don't worry if that code or explanation doesn't make sense; it's a bit
confusing. One little list comprehension is doing a pile of work here, and the code
is hard to understand, read, and ultimately, maintain. This example shows that list
comprehensions aren't always the best solution; most programmers would agree that
the earlier for loop is more readable than this version. Remember: the tools we are
provided with should not be abused! Always pick the right tool for the job, and that
job is writing maintainable code.
Set and dictionary comprehensions
Comprehensions aren't restricted to lists. We can use a similar syntax with braces to
create sets and dictionaries as well. Let's start with sets. One way to create a set is to
wrap a list comprehension in the set() constructor, which converts it to a set. But
why waste memory on an intermediate list that gets discarded when we can create a
set directly?
Here's an example that uses a named tuple to model author/title/genre triads, and
then retrieves a set of all the authors that write in a specific genre:
from collections import namedtuple
Book = namedtuple("Book", "author title genre")
books = [
Book("Pratchett", "Nightwatch", "fantasy"),
Book("Pratchett", "Thief Of Time", "fantasy"),
Book("Le Guin", "The Dispossessed", "scifi"),
Book("Le Guin", "A Wizard Of Earthsea", "fantasy"),
Book("Turner", "The Thief", "fantasy"),
Book("Phillips", "Preston Diamond", "western"),
Book("Phillips", "Twice Upon A Time", "scifi"),
]
fantasy_authors = {
b.author for b in books if b.genre == 'fantasy'}
That set comprehension sure is short in comparison to the set up required! If we'd
used a list comprehension, of course, Terry Pratchett would have been listed twice.
As it is, the nature of sets removes the duplicates and we end up with:
>>> fantasy_authors
{'Turner', 'Pratchett', 'Le Guin'}
[ 200 ]
Chapter 7
We can introduce a colon to create a dictionary comprehension. This converts a

sequence into a dictionary using key : value pairs. For example, it may be useful to
quickly look up the author or genre in a dictionary if we know the title. We can use a
dictionary comprehension to map titles to book objects:
fantasy_titles = {
b.title: b for b in books if b.genre == 'fantasy'}
Now we have a dictionary and can look up books by title using the normal syntax.
In summary, comprehensions are not advanced Python, and they aren't "nonobject-oriented" tools that should be avoided. They are simply a more concise and
optimized syntax for creating a list, set, or dictionary from an existing sequence.
Generator expressions
Sometimes we want to process a new sequence without placing a new list, set,
or dictionary into system memory. If we're just looping over items one at a time,
and don't actually care about having a final container object created, creating that
container is a waste of memory. When processing one item at a time, we only need
the current object stored in memory at any one moment. But when we create a
container, all the objects have to be stored in that container before we start
processing them.
For example, consider a program that processes log files. A very simple log might
contain information in this format:
Jan 26, 2010 11:25:25 DEBUG

Jan 26, 2010 11:25:36 INFO

Jan 26, 2010 11:25:46 WARNING
serious.
Jan 26, 2010 11:25:52 WARNING
Jan 26, 2010 11:25:59 INFO

Jan 26, 2010 11:26:13 DEBUG

you want to figure something out.
Jan 26, 2010 11:26:32 INFO

but helpful.
Jan 26, 2010 11:26:40 WARNING
Jan 26, 2010 11:26:54 WARNING
This is a debugging message.

This is an information method.
This is a warning. It could be
Another warning sent.
Here's some information.
Debug messages are only useful if
Information is usually harmless,
Warnings should be heeded.
Watch for warnings.
Log files for popular web servers, databases, or e-mail servers can contain many
gigabytes of data. If we want to process each line in the log, we don't want to use a
list comprehension on those lines; it would create a list containing every line in the
file. This probably wouldn't fit in memory and could bring the computer to its knees,
depending on the operating system.
[ 201 ]
If we used a for loop on the log file, we could process one line at a time before
reading the next one into memory. Wouldn't be nice if we could use comprehension
syntax to get the same effect?
This is where generator expressions come in. They use the same syntax as
comprehensions, but they don't create a final container object. To create a
generator expression, wrap the comprehension in () instead of [] or {}.
The following code parses a log file in the previously presented format, and outputs
a new log file that contains only the WARNING lines:
import sys
inname = sys.argv[1]
outname = sys.argv[2]
with open(inname) as infile:
with open(outname, "w") as outfile:
warnings = (l for l in infile if 'WARNING' in l)
for l in warnings:
outfile.write(l)
This program takes the two filenames on the command line, uses a generator
expression to filter out the warnings (in this case, it uses the if syntax, and leaves
the line unmodified), and then outputs the warnings to another file. If we run it on
our sample file, the output looks like this:
Jan 26, 2010
serious.
Jan 26, 2010
Jan 26, 2010
Jan 26, 2010
11:25:46 WARNING
This is a warning. It could be
11:25:52 WARNING
11:26:40 WARNING
11:26:54 WARNING
Another warning sent.

Warnings should be heeded.
Watch for warnings.
Of course, with such a short input file, we could have safely used a list
comprehension, but if the file is millions of lines long, the generator expression
will have a huge impact on both memory and speed.
Generator expressions can also be useful inside function calls. For example,
we can call sum, min, or max on a generator expression instead of a list, since these
functions process one object at a time. We're only interested in the result, not any
intermediate container.
In general, a generator expression should be used whenever possible. If we don't
actually need a list, set, or dictionary, but simply need to filter or convert items in a
sequence, a generator expression will be most efficient. If we need to know the length
of a list, or sort the result, remove duplicates, or create a dictionary, we'll have to use
the comprehension syntax.
[ 202 ]
Chapter 7
Generators
Generator expressions are actually a sort of comprehension too; they compress the
more advanced (this time it really is more advanced!) generator syntax into one line.
The greater generator syntax looks even less object-oriented than anything we've
seen, but we'll discover that once again, it is a simple syntax shortcut to create a kind
of object.
Let's take the log file example a little further. If we want to delete the WARNING
column from our output file (since it's redundant; this file contains only warnings),
we have several options, at various levels of readability. We can do it with a
generator expression:
import sys
inname, outname = sys.argv[1:3]
warnings = (l.replace('\tWARNING', '')
for l in infile if 'WARNING' in l)
for l in warnings:
outfile.write(l)
That's perfectly readable, though I wouldn't want to make the expression any more
complicated than that. We could also do it with a normal for loop:
import sys
for l in infile:
if 'WARNING' in l:
outfile.write(l.replace('\tWARNING', ''))
That's maintainable, but so many levels of indent in so few lines is kind of ugly. Now
let's consider a truly object-oriented solution, without any shortcuts:
import sys
class WarningFilter:
def __init__(self, insequence):
self.insequence = insequence
def __iter__(self):
[ 203 ]

return self
def __next__(self):
l = self.insequence.readline()
while l and 'WARNING' not in l:
l = self.insequence.readline()
if not l:
raise StopIteration
return l.replace('\tWARNING', '')
filter = WarningFilter(infile)
for l in filter:
outfile.write(l)
No doubt about it: that is ugly and difficult to read. What is happening here? Well,
we created an object that takes a file object as input, and then provides a __next__
method to allow it to work as an iterator in for loops. That method reads lines from
the file, discarding them if they are not WARNING lines. When it encounters a WARNING
line, it returns it, and the for loop will call __next__ again to get the next line. When
we run out of lines, we raise StopIteration to tell the loop we're finished. It's pretty
ugly compared to the other examples, but it's also powerful; now that we have a
class in our hands, we can do whatever we want to it.
With that background behind us, we finally get to see generators in action. This next
example does exactly the same thing as the previous one: it creates an object that
allows us to loop over the input:
import sys
def warnings_filter(insequence):
for l in insequence:
if 'WARNING' in l:
yield l.replace('\tWARNING', '')
filter = warnings_filter(infile)
for l in filter:
outfile.write(l)
Ok, that's pretty readable, maybe... at least it's short. But what on earth is going on
here, it doesn't make sense. And what is yield, anyway?
[ 204 ]
Chapter 7
Last question first: yield is the key to generators. When Python sees yield in a
function, it takes that function and wraps it up in an object not unlike the one in our
previous example. Think of the yield statement as similar to the return statement;
it exits the function and returns a line. Unlike return, when the function is called
again, it will start where it left off; on the line after the yield statement. In this
example, there is no line after the yield statement, so it jumps to the next iteration of
the for loop. Since the yield statement is inside an if statement, it only yields lines
that contain WARNING.
While it looks like that function is simply looping over the lines, it is really creating
an object; a generator object:
>>> print(warnings_filter([]))
<generator object warnings_filter at 0xb728c6bc>
I passed an empty list into the function to act as an iterator. All the function does
is create and return a generator object. That object has __iter__ and __next__
methods on it, much like the one we created in the previous example. Whenever
__next__ is called, the generator runs the function until it finds a yield statement. It
then returns the value from yield, and the next time __next__ is called, it picks up
where it left off.
This use of generators isn't that advanced, but if you don't realize the function is
creating an object, it can seem magical. We can even have multiple calls to yield in
a single function; it will simply pick up at the most recent yield and continue to the
next one.
There is even more to generators than what we have covered. We can send values
back into generators when calling yield, turning them into a dark art called
coroutines. While technically objects, coroutines encourage us to think very
differently from the object-oriented principles we've been discussing, and are beyond
the scope of this book. Do a search if you are interested in learning more about them.
One prominent feature of many object-oriented programming languages is a tool

called method overloading. Method overloading simply refers to having multiple
methods with the same name that accept different sets of arguments. In statically
typed languages, this is useful if we want to have a method that accepts either an
integer or a string, for example. In non-object-oriented languages we might need two
functions called add_s and add_i to accommodate such situations. In statically typed
object-oriented languages, we'd need two methods, both called add, one that accepts
strings, and one that accepts integers.
[ 205 ]
In Python, we only need one method, which accepts any type of object. It may have
to do some testing on the object type (for example, if it is a string, convert it to an
integer), but only one method is required.
However, method overloading is also useful when we want a method with the
same name to accept different numbers or sets of arguments. For example, an e-mail
message method might come in two versions, one of which accepts an argument for
the from e-mail address. The other method might look up a default from address
instead. Python doesn't permit multiple methods with the same name, but it does
provide a different, equally flexible, interface.
We've seen some of the possible ways to send arguments to methods and functions
in previous examples, but now we'll cover all the details. The simplest function
accepts no arguments. We probably don't need an example, but here's one for
completeness:
def no_args():
pass
and here's how it's called:

no_args()
A function that does accept arguments will provide the names of those arguments in
a comma-separated list. Only the name of each argument needs to be supplied.
When calling the function, these positional arguments must be specified in order,
and none can be missed or skipped. This is the most common way we've specified
arguments in our previous examples:
def mandatory_args(x, y, z):
pass
and to call it:

mandatory_args("a string", a_variable, 5)
Any type of object can be passed as an argument: an object, a container, a primitive,

even functions and classes. The above call shows a hard-coded string, an unknown
variable, and an integer passed into the function.
[ 206 ]
Chapter 7
Default arguments
If we want to make an argument optional, rather than creating a second method with
a different set of arguments, we can specify a default value in a single method, using
an equals sign. If the calling code does not supply this argument, it will be assigned
a default value. However, the calling code can still choose to override the default by
passing in a different value. Often, a default value of None, or an empty string or list
is suitable.
Here's a function definition with default arguments:
def default_arguments(x, y, z, a="Some String", b=False):
pass
The first three arguments are still mandatory and must be passed by the calling code.
The last two parameters have default arguments supplied.
There are several ways we can call this function. We can supply all arguments in
order, as though all the arguments were positional arguments.
kwargs("a string", variable, 8, "", True)
Or we can supply just the mandatory arguments in order, leaving the keyword
arguments to be assigned their default values:
kwargs("a longer string", some_variable, 14)
We can also use the equals sign syntax when calling a function to provide values in a
different order or to skip default values that we aren't interested in. For example, we
can skip the first keyword arguments and supply the second one:
kwargs("a string", variable, 14, b=True)
Surprisingly, we can even use the equals sign syntax to mix up the order of
positional arguments, so long as all of them are supplied.
>>> kwargs(y=1,z=2,x=3,a="hi")
3 1 2 hi False
With so many options, it may seem hard to pick one, but if you think of the
positional arguments as an ordered list, and keyword arguments as sort of like a
dictionary, you'll find that the correct layout tends to fall into place. If you need to
require the caller to specify an argument, make it mandatory; if you have a sensible
default, then make it a keyword argument. Choosing how to call the method
normally takes care of itself, depending on which values need to be supplied,
and which can be left at their defaults.
[ 207 ]
One thing to take note of with keyword arguments is that anything we provide as a
default argument is evaluated when the function is first interpreted, not when it is
called. This means we can't have dynamically generated default values. For example,
the following code won't behave quite as expected:
number = 5
def funky_function(number=number):
print(number)
number=6
funky_function(8)
funky_function()
print(number)
If we run this code, it outputs the number 8, first, but then it outputs the number 5
for the call with no arguments. We had set the variable to the number 6, as evidenced
by the last line of output, but when the function is called, the number 5 is printed; the
default value was calculated when the function was defined, not when it was called.
This is tricky with empty containers. For example, it is common to ask calling code
to supply a list that our function is going to manipulate, but the list is optional. We'd
like to make an empty list as a default argument. We can't do this; it will create only
one list, when the code is first constructed:
>>> def hello(b=[]):
...
b.append('a')
...
print(b)
...
>>> hello()
['a']
>>> hello()
['a', 'a']
Whoops, that's not quite what we expected! The usual way to get around this is to
make the default value None, and then use the idiom if argument is None: arg = []
inside the method. Pay close attention!
Variable argument lists
Default values alone do not allow us all the flexible benefits of method overloading.
The thing that makes Python really slick is the ability to write methods that accept
an arbitrary number of positional or keyword arguments without explicitly naming
them. We can also pass arbitrary lists and dictionaries into such functions.
[ 208 ]
Chapter 7
For example, a function to accept a link or list of links and download the web pages
could use such variadic arguments, or varargs. Instead of accepting a single value
that is expected to be a list of links, we can accept an arbitrary number of arguments,
where each argument is a different link. We do this by specifying the * operator in
the function definition:
def get_pages(*links):
for link in links:
#download the link with urllib
print(link)
The *links says "I'll accept any number of arguments and put them all in a list
of strings named links". If we supply only one argument, it'll be a list with one
element, if we supply no arguments, it'll be an empty list. Thus, all these function
calls are valid:
get_pages()
get_pages('http://www.archlinux.org')
get_pages('http://www.archlinux.org',
'http://ccphillips.net/')
We can also accept arbitrary keyword arguments. These arrive into the function as
a dictionary. They are specified with two asterisks (as in **kwargs) in the function
declaration. This tool is commonly used in configuration setups. The following class
allows us to specify a set of options with default values:
class Options:
default_options = {
'port': 21,
'host': 'localhost',
'username': None,
'password': None,
'debug': False,
}
def __init__(self, **kwargs):
self.options = dict(Options.default_options)
self.options.update(kwargs)
def __getitem__(self, key):
return self.options[key]
[ 209 ]
All the interesting stuff in this class happens in the __init__ method. We have a
dictionary of default options and values at the class level. The first thing the __init__
method does is make a copy of this dictionary. We do that instead of modifying the
dictionary directly in case we instantiate two separate sets of options. (Remember,
class level variables are shared between instances of the class.) Then, __init__ uses
the update method on the new dictionary to change any non-default values to those
supplied as keyword arguments. The __getitem__ method simply allows us to use
the new class using indexing syntax. Here's a session demonstrating the class
in action:
>>> options = Options(username="dusty", password="drowssap",
debug=True)
>>> options['debug']
True
>>> options['port']
21
>>> options['username']
'dusty'
We're able to access our options instance using dictionary indexing syntax,
and the dictionary includes both default values and the ones we set using
keyword arguments.
The keyword argument syntax can be dangerous, as it may break the "explicit
is better than implicit" rule. In the above example, it's possible to pass arbitrary
keyword arguments to the Options initializer to represent options that don't exist in
the default dictionary. This may not be a bad thing, depending on the purpose of the
class, but it makes it hard for someone using the class to discover what valid options
are available. It also makes it easy to enter a confusing typo ("Debug" instead of
"debug", for example) that adds two options where only one should have existed.
The above example is not that bad if we instruct the user of the class to only pass
default options (we could even add some code to enforce this rule). The options are
documented in the class definition so it'll be easy to look them up.
Keyword arguments are also very useful when we need to accept arbitrary
arguments to pass to a second function, but we don't know what those arguments
will be. We saw this in action in Chapter 3, when we were building support for
multiple inheritance.
[ 210 ]
Chapter 7
We can, of course, combine the variable argument and variable keyword argument
syntax in one function call, and we can use normal positional and default arguments
as well. The following example is somewhat contrived, but demonstrates the four
types in action:
import shutil
import os.path
def augmented_move(target_folder, *filenames,
verbose=False, **specific):
'''Move all filenames into the target_folder, allowing
specific treatment of certain files.'''
def print_verbose(message, filename):
'''print the message only if verbose is enabled'''
if verbose:
print(message.format(filename))
for filename in filenames:
target_path = os.path.join(target_folder, filename)
if filename in specific:
if specific[filename] == 'ignore':
print_verbose("Ignoring {0}", filename)
elif specific[filename] == 'copy':
print_verbose("Copying {0}", filename)
shutil.copyfile(filename, target_path)
else:
print_verbose("Moving {0}", filename)
shutil.move(filename, target_path)
This example will process an arbitrary list of files. The first argument is a target
folder, and the default behavior is to move all remaining non-keyword argument
files into that folder. Then there is a keyword-only argument, verbose, which tells
us whether to print information on each file processed. Finally, we can supply a
dictionary containing actions to perform on specific filenames; the default behavior
is to move the file, but if a valid string action has been specified in the keyword
arguments, it can be ignored or copied instead. Notice the ordering of the parameters
in the function; first the positional argument is specified, then the *filenames list,
then any specific keyword-only arguments, and finally, a **specific dictionary to
hold remaining keyword arguments.
We create an inner helper function, print_verbose, which will print messages only
if the verbose key has been set. This function keeps code readable by encapsulating
this functionality into a single location.
[ 211 ]
In common cases, this function would likely be called as:

>>> augmented_move("move_here", "one", "two")
This command would move the files one and two into the move_here directory,
assuming they exist (There's no error checking or exception handling in the function,
so it would fail spectacularly if the files or target directory didn't exist). The move
would occur without any output, since verbose is False by default.
If we want to see the output, we can call it with:
>>> augmented_move("move_here", "three", verbose=True)
Moving three
This moves one file, named three, and tells us what it's doing. Notice that it is
impossible to specify verbose as a positional argument in this example; we must
pass a keyword argument. Otherwise Python would think it was another filename in
the *filenames list.
If we want to copy or ignore some of the files in the list, instead of moving them, we
can pass additional keyword arguments:
>>> augmented_move("move_here", "four", "five", "six",
four="copy", five="ignore")
This will move the sixth file and copy the fourth, but won't display any output, since
we didn't specify verbose. Of course, we can do that, too, and keyword arguments
can be supplied in any order:
>>> augmented_move("move_here", "seven", "eight", "nine",
seven="copy", verbose=True, eight="ignore")
Copying seven
Ignoring eight
Moving nine
Unpacking arguments
There's one more nifty trick involving variable arguments and keyword arguments.
We've used it in some of our previous examples, but it's never too late for an
explanation. Given a list or dictionary of values, we can pass those values into a
function as if they were normal positional or keyword arguments. Have a look
at this code:
def show_args(arg1, arg2, arg3="THREE"):
print(arg1, arg2, arg3)
[ 212 ]
Chapter 7
some_args = range(3)
more_args = {
"arg1": "ONE",
"arg2": "TWO"}
print("Unpacking a sequence:", end=" ")
show_args(*some_args)
print("Unpacking a dict:", end=" ")
show_args(**more_args)
Here's what it looks like when we run it:

Unpacking a sequence: 0 1 2
Unpacking a dict: ONE TWO THREE
The function accepts three arguments, one of which has a default value. But when
we have a list of three arguments, we can use the * operator inside a function call to
unpack it into the three arguments. If we have a dictionary of arguments, we can use
the ** syntax to unpack it as a collection of keyword arguments.
This is most often useful when mapping information that has been collected from
user input or from an outside source (an internet page, a text file) to a function or
method call.
Remember our earlier example that used headers and lines in a text file to create a
list of dictionaries with contact information? Instead of just adding the dictionaries
to a list, we could use keyword unpacking to pass the arguments to the __init__
method on a specially built Contact object that accepts the same set of arguments.
See if you can adapt the example to make this work.
Functions are objects too
Programming languages that over-emphasize object-oriented principles tend to

frown on functions that are not methods. In such languages, you're expected to
create an object to sort of wrap the single method involved. There are numerous
situations where we'd like to pass around a small object that is simply called to
perform an action. This is most frequently done in event-driven programming, such
as graphical toolkits or asynchronous servers; we'll see some design patterns that use
it in the next two chapters.
[ 213 ]
In Python, we don't need to wrap such methods in an object, because functions

already are objects! We can set attributes on functions (though this isn't a common
activity), and we can pass them around to be called at a later date. They even
have a few special properties that can be accessed directly. Here's yet another
contrived example:
def my_function():
print("The Function Was Called")
my_function.description = "A silly function"
def second_function():
print("The second was called")
second_function.description = "A sillier function."
def another_function(function):
print("The description:", end=" ")
print(function.description)
print("The name:", end=" ")
print(function.__name__)
print("The class:", end=" ")
print(function.__class__)
print("Now I'll call the function passed in")
function()
another_function(my_function)
another_function(second_function)
If we run this code, we can see that we were able to pass two different functions into
our third function, and get different output for each one:
The
The
The
Now
The
The
The
The
Now
The
description: A silly function

name: my_function
class: <class 'function'>
I'll call the function passed in
Function Was Called
description: A sillier function.
name: second_function
class: <class 'function'>
I'll call the function passed in
second was called
We set an attribute on the function, named description (not very good

descriptions, admittedly). We were also able to see the function's __name__ attribute,
and to access its class, demonstrating that the function really is an object with
attributes. Then we called the function by using the callable syntax (the parentheses).
[ 214 ]
Chapter 7
The fact that functions are top-level objects is most often used to pass them around to
be executed at a later date, for example, when a certain condition has been satisfied.
Let's build an event-driven timer that does just this:
import datetime
import time
class TimedEvent:
def __init__(self, endtime, callback):
self.endtime = endtime
self.callback = callback
def ready(self):
return self.endtime <= datetime.datetime.now()
class Timer:
def __init__(self):
self.events = []
def call_after(self, delay, callback):
end_time = datetime.datetime.now() + \
datetime.timedelta(seconds=delay)
self.events.append(TimedEvent(end_time, callback))
def run(self):
while True:
ready_events = (e for e in self.events if e.ready())
for event in ready_events:
event.callback(self)
self.events.remove(event)
time.sleep(0.5)
In production, this code should definitely have extra documentation using

docstrings! The call_after method should at least mention that the delay is in
seconds and that the callback function should accept one argument: the timer
doing the calling.
We have two classes here. The TimedEvent class is not really meant to be accessed
by other classes; all it does is store an endtime and callback. We could even use a
tuple or namedtuple here, but as it is convenient to give the object a behavior that
tells us whether or not the event is ready to run, we use a class instead.
[ 215 ]
The Timer class simply stores a list of upcoming events. It has a call_after method
to add a new event. This method accepts a delay parameter representing the
number of seconds to wait before executing the callback, and the callback itself:
a function to be executed at the correct time. This callback function should accept
one argument.
The run method is very simple; it uses a generator expression to filter out any events
whose time has come, and executes them in order. The timer loop then continues
indefinitely, so it has to be interrupted with a keyboard interrupt (Ctrl + C or Ctrl +
Break). We sleep for half a second after each iteration so as to not grind the system to
a halt.
The important things to note here are the lines that touch callback functions.
The function is passed around like any other object and the timer never knows or
cares what the original name of the function is or where it was defined. When it's
time to call the function, the timer simply applies the parenthesis syntax to the
stored variable.
Here's a set of callbacks that test the timer:
from timer import Timer
import datetime
def format_time(message, *args):
now = datetime.datetime.now().strftime("%I:%M:%S")
print(message.format(*args, now=now))
def one(timer):
format_time("{now}: Called One")
def two(timer):
format_time("{now}: Called Two")
def three(timer):
format_time("{now}: Called Three")
class Repeater:
def __init__(self):
self.count = 0
def repeater(self, timer):
format_time("{now}: repeat {0}", self.count)
timer.call_after(5, self.repeater)
timer = Timer()
timer.call_after(1, one)
[ 216 ]
Chapter 7
timer.call_after(2, one)
timer.call_after(2, two)
timer.call_after(4, two)
timer.call_after(3, three)
timer.call_after(6, three)
repeater = Repeater()
timer.call_after(5, repeater.repeater)
format_time("{now}: Starting")
timer.run()
This example allows us to see how multiple callbacks interact with the timer. The
first function is the format_time function. It uses the string format method to
add the current time to the message, and illustrates variable arguments in action.
The format_time method will accept any number of positional arguments, using
variable argument syntax, which are then forwarded as positional arguments to
the string's format method. After that we create three simple callback methods that
simply output the current time and a short message telling us which callback has
been fired.
The Repeater class demonstrates that methods can be used as callbacks too, since
they are really just functions. It also shows why the timer argument to the callback
functions is useful: we can add a new timed event to the timer from inside a
presently running callback.
Then we simply create a timer and add several events to it that are called after
different amounts of time. Then we start the timer running; the output shows that
events are run in the expected order:
02:53:35:
02:53:36:
02:53:37:
02:53:37:
02:53:38:
02:53:39:
02:53:40:
02:53:41:
02:53:45:
02:53:50:
02:53:55:
02:54:00:
Starting
Called One
Called One
Called Two
Called Three
Called Two
repeat 0
Called Three
repeat 1
repeat 2
repeat 3
repeat 4
[ 217 ]
Using functions as attributes
One of the interesting effects of functions being objects is that they can be set as
callable attributes on other objects. It is possible to add or change a function to an
instantiated object:
class A:
def print(self):
print("my class is A")
def fake_print():
print("my class is not A")
a = A()
a.print()
a.print = fake_print
a.print()
This code creates a very simple class with a print method that doesn't tell us
anything we don't know. Then we create a new function that tells us something
we don't believe.
When we call print on an instance of the A class, it behaves as expected. If we
then set the print method to point at a new function, it tells us something different:
my class is A
my class is not A
It is also possible to replace methods on classes, instead of objects, although in that

case we have to add the self argument to the parameter list. This will change the
method for all instances of that object, even ones that have already been instantiated.
Obviously, replacing methods like this can be very dangerous and confusing to
maintain. Somebody reading the code will see that a method has been called, and
look up that method on the original class. But the method on the original class is not
the one that was called. Figuring out what really happened can become a very tricky
debugging session.
It does have its uses though. Often, replacing or adding methods at run time (called
monkey-patching) is used in automated testing. If testing a client-server application,
we may not want to actually connect to the server when testing the client; that may
result in accidental transfers of funds or embarrassing test e-mails being sent to real
people. Instead, we can set up our test code to replace some of the key methods on
the object that sends requests to the server, so it only records that the methods have
been called.
[ 218 ]
Chapter 7
Monkey-patching can also be used to fix bugs or add features in third-party code that
we are interacting with and does not behave quite the way we need it to. It should,
however, be applied sparingly, it's almost always a "messy hack". Sometimes,
though, it is the only way to adapt an existing library to suit our needs.
Callable objects
Since functions are just objects that happen to respond to the call syntax, we start to
wonder if it's possible to write objects that can be called yet aren't real functions. Yes,
of course!
Any object can be turned into a callable, as easily as giving it a __call__ method
that accepts the required arguments. Let's make our Repeater class from the timer
example a little easier to use by making it a callable:
class Repeater:
def __init__(self):
self.count = 0
def __call__(self, timer):
format_time("{now}: repeat {0}", self.count)
self.count += 1
timer.call_after(5, self)
timer = Timer()
timer.call_after(5, Repeater())
format_time("{now}: Starting")
timer.run()
This example isn't much different from the earlier class; all we did was change the
name of the repeater function to __call__ and pass the object itself as a callable.
Note that when we make the call_after call, we pass the argument Repeater().
Those two parentheses are creating a new instance of the class, they are not explicitly
calling the class. That happens later, inside the timer. If we want to execute the
__call__ method on a newly instantiated object, we'd use a rather odd syntax:
Repeater()(). The first set of parentheses constructs the object; the second set
executes the __call__ method.
[ 219 ]
Case study
To tie together some of the principles presented in this chapter, let's build a mailing
list manager. The manager will keep track of e-mail addresses categorized into
named groups. When it's time to send a message, we can pick a group and send the
message to all e-mail addresses assigned to that group.
Now, before we start working on this project, we ought to have a safe way to test it,
without sending e-mails to a bunch of real people. Luckily, Python has our back here;
like the test HTTP server, it has a built in Simple Mail Transfer Protocol (SMTP)
server that we can instruct to capture any messages we send without actually
sending them. We can run the server with the following command:
python -m smtpd -n -c DebuggingServer localhost:1025
Running this command at a command prompt will start an SMTP server running on
port 1025 on the local machine. But we've instructed it to use the DebuggingServer
class (it comes with the built-in SMTP module), which, instead of sending mails
to the intended recipients, simply prints them on the terminal screen as it receives
them. Neat, eh?
Now, before writing our mailing list, let's write some code that actually sends mail.
Of course, Python supports this in the standard library too, but it's a bit of an odd
interface, so we'll write a new function to wrap it all cleanly:
import smtplib
from email.mime.text import MIMEText
def send_email(subject, message, from_addr, *to_addrs,
host="localhost", port=1025, **headers):
email = MIMEText(message)
email['Subject'] = subject
email['From'] = from_addr
for header, value in headers.items():
email[header] = value
sender = smtplib.SMTP(host, port)
for addr in to_addrs:
del email['To']
email['To'] = addr
sender.sendmail(from_addr, addr, email.as_string())
sender.quit()
[ 220 ]
Chapter 7
We won't cover the code inside this method too thoroughly; the documentation in
the standard library can give you all the information you need to use the smtplib
and email modules effectively.
We've used both variable argument and keyword argument syntax in the function
call; any unknown arguments are mapped to extra addresses to send to; any extra
keyword arguments are mapped to e-mail headers.
The headers passed into the function represent auxiliary headers that can be
attached to a method. Such headers might include Reply-To, Return-Path,
or X-pretty-much-anything. Can you see a problem here?
Any valid identifier in Python cannot include the - character. In general, that
character represents subtraction. So it's not possible to call a function with
Reply-To = [email protected]. Perhaps we were too eager to use keyword
arguments because they are a new tool we just learned this chapter?
We'll have to change the argument to a normal dictionary; this will work because
any string can be used as a key in a dictionary. By default, we'd want this dictionary
to be empty, but we can't make the default parameter an empty dictionary. No,
we'll have to make the default argument None, and then set up the dictionary
at the beginning of the method:
def send_email(subject, message, from_addr, *to_addrs,
host="localhost", port=1025, headers=None):
headers = {} if headers is None else headers
If we have our debugging SMTP server running in one terminal, we can test this
code in a Python interpreter:
>>> send_email("A model subject", "The message contents",
"[email protected]", "[email protected]", "[email protected]")
Then if we check the output from the debugging SMTP server, we get the following:
---------- MESSAGE FOLLOWS ---------Content-Type: text/plain; charset="us-ascii"
MIME-Version: 1.0
Content-Transfer-Encoding: 7bit
Subject: A model subject
From: [email protected]
To: [email protected]
X-Peer: 127.0.0.1
The message contents
[ 221 ]

------------ END MESSAGE --------------------- MESSAGE FOLLOWS ---------Content-Type: text/plain; charset="us-ascii"
MIME-Version: 1.0
Content-Transfer-Encoding: 7bit
Subject: A model subject
From: [email protected]
To: [email protected]
X-Peer: 127.0.0.1
The message contents
------------ END MESSAGE ------------
Excellent, it has "sent" our e-mail to the two correct addresses with subject and
message contents included.
Now that we can send messages, let's work on the e-mail group management system.
We'll need an object that somehow matches e-mail addresses with the groups they
are in. Since this is a many-to-many relationship (any one e-mail address can be in
multiple groups, any one group can be associated with multiple e-mail addresses),
none of the data structures we've studied seem quite ideal. We could try a dictionary
of group-names matched to a list of associated e-mail addresses, but that would
duplicate e-mail addresses. We could also try a dictionary of e-mail addresses
matched to groups, resulting in a duplication of groups. Neither seems optimal. Let's
try this latter version, even though intuition tells me the groups to e-mail address
solution would be more straightforward.
Since the values in our dictionary will always be collections of unique e-mail
addresses, we should probably store them in a set. We can use defaultdict to
ensure there is always a set available for each key:
class MailingList:
'''Manage groups of e-mail addresses for sending e-mails.'''
def __init__(self):
self.email_map = defaultdict(set)
def add_to_group(self, email, group):
self.email_map[email].add(group)
Now let's add a method that allows us to collect all the e-mail addresses in one or
more groups. We can use a set comprehension to take care of this easily:
def emails_in_groups(self, *groups):
groups = set(groups)
return {e for (e, g) in self.email_map.items()
if g & groups}
[ 222 ]
Chapter 7
Ok, that set comprehension needs explaining, doesn't it? First look at what we're
iterating over: self.email_map.items(). That method, of course, returns a tuple
of key-value pairs for each item in the dictionary. The values are sets of strings
representing the groups. We split these into two variables named e and g, short for
e-mail and groups. We only return the key (the e-mail address) for each item though,
since the desired output is a set of e-mail addresses.
The only thing left that may not make sense is the condition clause. This clause
simply intersects the groups set with the set of groups associated with the e-mails.
If the result is non-empty, the e-mail gets added, otherwise, it is discarded. The g &
groups syntax is a shortcut for g.intersection(groups); the set class does this by
implementing the special method __and__ to call intersection.
Now, with these building blocks, we can trivially add a method to our MailingList
class that sends e-mail to specific groups:
def send_mailing(self, subject, message, from_addr,
*groups, **kwargs):
emails = self.emails_in_groups(*groups)
send_email(subject, message, from_addr,
*emails, **kwargs)
This function stresses on variable arguments. As input, it takes a list of groups as

variable arguments, and optional keyword arguments as a dictionary. It doesn't
care about the keyword arguments at all; it simply passes those arguments on to
the send_email function we defined earlier. It then gets the list of e-mails for the
specified groups, and passes those as variable arguments into send_email.
The program can be tested by ensuring the SMTP debugging server is running in one
command prompt, and, in a second prompt, load the code using:
>>> python -i mailing_list.py
Create a MailingList object with:

>>> m = MailingList()
Then create a few fake e-mail addresses and groups, along the lines of:
>>> m.add_to_group("[email protected]", "friends")
>>> m.add_to_group("[email protected]", "friends")
>>> m.add_to_group("[email protected]", "family")
>>> m.add_to_group("[email protected]", "professional")
[ 223 ]
Finally, use a command like this to send e-mails to specific groups:

>>> m.send_mailing("A Party",
"Friends and family only: a party", "[email protected]", "friends",
"family", headers={"Reply-To": "[email protected]"})
E-mails to each of the addresses in the specified groups should show up in the
console on the SMTP server.
Exercises
If you don't use comprehensions in your daily coding very often, the first thing you
should do is search through some existing code and find some for loops. See if any
of them can be trivially converted to a generator expression or a list, set, or dictionary
comprehension.
Test the claim that list comprehensions are faster than for loops. This can be done
with the built-in timeit module. Use the help documentation for the timeit.
timeit function to find out how to use it. Basically, write two functions that do the
same thing, one using a list comprehension, and one using a for loop. Pass each
function into timeit.timeit, and compare the results. If you're feeling adventurous,
compare generators and generator expressions as well. Testing code using timeit
can become addictive, so bear in mind that code does not need to be hyper-fast
unless it's being executed an immense number of times, such as on a huge input list
or log file.
Try writing the case study using groups as dictionary keys and lists of e-mail
addresses as the values. You'll likely be surprised at how little needs to be changed.
If you're interested, try reworking it to accept first and last names as well as e-mail
addresses. Then allow customizing e-mail messages (use str.format) to have the
person's first or last name in each message.
Except the send_mailing method itself, the MailingList object is really quite
generic. Consider what needs to be done to make it perform any generic activity on
each e-mail address, instead of just sending mail. Hint: callback functions will be
very useful.
Play around with generator functions. Start with basic iterators that require multiple
values (mathematical sequences are canonical examples; the Fibonacci sequence is
overused if you can't think of anything better). Try some more advanced generators
that do things like take multiple input lists and somehow yield values that merge
them. Generators can also be used on files; can you write a simple generator that
shows those lines that are identical in two files?
[ 224 ]
Chapter 7
Summary
We covered several very different topics in this chapter. Each represented an

important non-object-oriented feature that is popular in Python. Just because
we can use object-oriented principles does not always mean we should!
However, we also saw that "the Python way" often just provides a shortcut to
traditional object-oriented syntax. Knowing the object-oriented principles
underlying these tools allows us to use them effectively in our own classes.
We covered:
Built-in functions
Comprehensions and generators
Function arguments, variable arguments, and keyword arguments
Callback functions and callable objects
In the next chapter, we're going to study design patterns; building blocks that
object-oriented programmers use to create maintainable applications. In many cases,
we'll see that, as in this chapter, Python provides syntax for popular design patterns
that we can use instead.
[ 225 ]
Python Design Patterns I

We've covered the basic building blocks of object-oriented programming. Now, we'll
look at secondary structures that can be built from those blocks. These higher-level
structures, called design patterns, can help organize complex systems. In the next
two chapters, we'll be covering:
What design patterns are
Numerous specific patterns
A canonical implementation of each pattern in Python
Python syntax to replace certain patterns
Design patterns
When engineers and architects decide to build a bridge, or a tower, or a building,

they follow certain principles to ensure structural integrity. There are various
possible designs for bridges (suspension or cantilever for example), but if the
engineer doesn't use one of the standard designs, and doesn't have a brilliant
new design, it is likely the bridge he/she designs will collapse.
Design patterns are an attempt to bring this same formal definition for correctly
designed structures to software engineering. There are many different design
patterns to solve different general problems. People who create design patterns first
identify a common problem faced by developers in a wide variety of situations. They
then suggest what might be considered the ideal solution for that problem, in terms
of object-oriented design.
We already have plenty of experience with one of the most common design patterns,
the iterator. Canonically, the iterator pattern is meant to provide a common interface
for looping over the items in a sequence. Separating the looping action from the
sequence that is actually being looped over allows the looping code to be changed
without interfering with either the code doing the looping, or the object being looped
over. For example, two iterators may loop over the items in different directions, or in
a sorted order. Further, it's possible to change the internal structure of an object, but
still allow it to be looped over using a single iterator interface.
In typical design pattern parlance, an iterator is an object with a next() method and
a done() method; the latter returns True if there are no items left in the sequence. In
a programming language without built-in support for iterators, the iterator would be
looped over like this:
while not iterator.done():
item = iterator.next()
# do something with the item
Of course, in Python, the method is named __next__ instead, and we have the much
more readable for item in iterator syntax to actually access the items. Rather than
a done method, it raises StopIteration when completed. The same pattern is being
applied, and it is still based on a design pattern solution, but Python has provided us
with a more readable way to apply and access the pattern.
Knowing a design pattern and choosing to use it in our software does not, however,
guarantee that we are creating a "correct" solution. In 1907, the Qubec Bridge (to
this day, the longest cantilever bridge in the world) collapsed before construction
was completed because the engineers who designed it grossly underestimated the
weight of the steel used to construct it. Similarly, in software development, we may
incorrectly choose or apply a design pattern, and create software that "collapses"
under normal operating situations or when stressed beyond its original design limits.
Any one design pattern proposes a set of objects interacting in a specific way to solve
a general problem. The job of the programmer is to recognize when they are facing a
specific version of that problem, and to adapt the general design in their solution.
In this chapter, we'll be reviewing several common patterns, and how they are
implemented in Python. Python often provides an alternative syntax to make
working with such problems simpler. We will cover both the "traditional" design,
and the Python version for these patterns.
[ 228 ]
Chapter 8
Decorator pattern
The decorator pattern allows us to "wrap" an object that provides core functionality
with other objects that alter that functionality. Any object that uses the decorated
object will interact with it in exactly the same way as if it were undecorated (that is,
the interface of the decorated object is identical to the core object).
There are two primary uses of the decorator pattern:
Enhancing the response of a component that is sending data to a

second component
Supporting multiple optional behaviors
The second option is often a suitable alternative to multiple inheritance. We can

construct a core object, and then create a decorator around that core. Since the
decorator object has the same interface as the core object, we can even wrap the
new object in other decorators. Here's how it looks in UML:
Here, Core, and all the decorators implement a specific Interface. The decorators
maintain a reference to another instance of that Interface, via composition. When
called, the decorator will do some added processing before or after calling its
wrapped interface. The wrapped object may be another decorator, or the core
functionality; multiple decorators may wrap each other, but the object in the
"center" of all those decorators will provide core functionality.
[ 229 ]
Decorator example
Let's look at an example from network programming. We'll be using a TCP socket.
The socket.send() method takes a string of input bytes and outputs them to the
receiving socket at the other end. There are plenty of libraries that accept sockets and
access this function to send data on the stream. Let's create such an object; it will be
an interactive shell that waits for a connection from a client and then prompts the
user for a string response:
import socket
def respond(client):
response = input("Enter a value: ")
client.send(bytes(response, 'utf8'))
client.close()
server = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
server.bind(('localhost',2401))
server.listen(1)
try:
while True:
client, addr = server.accept()
respond(client)
finally:
server.close()
The respond function accepts a socket parameter and prompts for data to be sent
as a reply, then sends it. After that, we construct a server socket and tell it to listen
on port 2401 (I picked the port randomly) on the local computer. When a client
connects, it calls the respond function, which requests data interactively and
responds appropriately. The important thing to notice is that the respond function
only cares about two methods of the socket interface: send and close. To test this,
we can write a very simple client that connects to the same port and outputs the
response before exiting:
import socket
client = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
client.connect(('localhost', 2401))
print("Received: {0}".format(client.recv(1024)))
client.close()
[ 230 ]
Chapter 8
To use these programs:

1. Start the server in one terminal.
2. Open a second terminal window and run the client.
3. At the Enter a value: prompt in the server window, type a value and
press enter.
4. The client will receive what you typed, print it to the console, and exit. Run
the client a second time; the server will prompt for a second value.
Now, looking again at our server code, we see two sections. The respond function
sends data into a socket object. The remaining script is responsible for creating that
socket object. We'll create a pair of decorators that customize the socket behavior
without having to extend or modify the socket itself.
Let's start with a "logging" decorator. This object will simply output any data being
sent to the server's console before it sends it to the client:
class LogSocket:
def __init__(self, socket):
self.socket = socket
def send(self, data):
print("Sending {0} to {1}".format(
data, self.socket.getpeername()[0]))
self.socket.send(data)
def close(self):
self.socket.close()
This class decorates a socket object and presents the send and close interface to
client sockets. A better decorator would also implement (and possibly customize)
all of the socket methods. It should properly implement all of the arguments to
send, (which actually accepts an optional flags argument) as well, but let's keep the
example simple! Whenever send is called on this object, it logs the output to the
screen before sending just like the original socket did.
We only have to change one line in our original code to use this decorator. Instead of
calling respond with the socket, we call it with a decorated socket:
respond(LogSocket(client))
While that's quite simple, we have to ask ourselves why we didn't just extend the
socket class and override the send method. We could call super().send to do the
actual sending, after we logged it. This is a valid design.
[ 231 ]
When faced with a choice between decorators and inheritance, we should only use
decorators if we need to modify the object dynamically, according to some condition.
For example, we may only want to enable the logging decorator if the server is
currently in debugging mode. Decorators also beat out multiple inheritance when we
have multiple optional behaviors. As an example, we can write a second decorator
that compresses data using gzip compression whenever send is called:
import gzip
from io import BytesIO
class GzipSocket:
def __init__(self, socket):
self.socket = socket
def send(self, data):
buf = BytesIO()
zipfile = gzip.GzipFile(fileobj=buf, mode="w")
zipfile.write(data)
zipfile.close()
self.socket.send(buf.getvalue())
def close(self):
self.socket.close()
The send method in this version compresses the incoming data before sending it on
to the client. We don't have room for the example, but it's possible to write a client
that extracts the gzipped content.
Now that we have these two decorators, we can write code that dynamically
switches between them when responding. This example is not complete, but it
illustrates the logic we might follow to mix and match decorators:
client, addr = server.accept()
if log_send:
client = LoggingSocket(client)
if client.getpeername()[0] in compress_hosts:
client = GzipSocket(client)
respond(client)
This code checks a hypothetical configuration variable named log_send. If it's

enabled it wraps the socket in a LoggingSocket decorator. Similarly, it checks if
the client that has connected is in a list of addresses known to accept compressed
content. If so, it wraps the client in a GzipSocket decorator. Notice that none, either,
or both of the decorators may be enabled, depending on the configuration and
connecting client. Try writing this using multiple inheritance and see how confused
you get!
[ 232 ]
Chapter 8
Decorators in Python
The decorator pattern is useful in Python, but there are other options. For example,
we may be able to use monkey-patching, which we discussed in Chapter 7, to get a
similar effect. Single inheritance, where the "optional" calculations are done in one
large method can be an option, and multiple inheritance should not be written off
just because it's not suitable for the specific example seen previously!
In Python, it is very common to use this pattern on functions. As we saw in the
previous chapter, functions are objects too. In fact, this is so common that Python
provides a special syntax to make it easy to apply such decorators to functions.
For example, we can look at the logging example in a more general way. Instead of
logging only send calls on sockets, we may find it helpful to log all calls to certain
functions or methods. The following example implements a decorator that does
just this:
import time
def log_calls(func):
def wrapper(*args, **kwargs):
now = time.time()
print("Calling {0} with {1} and {2}".format(
func.__name__, args, kwargs))
return_value = func(*args, **kwargs)
print("Executed {0} in {1}ms".format(
func.__name__, time.time() - now))
return return_value
return wrapper
def test1(a,b,c):
print("\ttest1 called")
def test2(a,b):
def test3(a,b):
time.sleep(1)
test1 = log_calls(test1)
test1(1,2,3)
test2(4,b=5)
test3(6,7)
[ 233 ]
This decorator function is very similar to the example we explored earlier; in those
cases, the decorator took a socket-like object and created a socket-like object. This
time, our decorator takes a function object and returns a new function object. This
code is comprised of three separate tasks:
A function, log_calls accepts a function.
The function defines (internally) a new function, named wrapper, that does
some extra calculations before calling the original function.
This new function is returned, to replace the original function.
Three sample functions demonstrate the decorator in use. The third one even
includes a sleep call to demonstrate the timing test. We pass each function into the
decorator, which returns a new function. We assign this new function to the original
variable name, effectively replacing the original function with a decorated one.
This syntax allows us to build up decorated function objects dynamically as we did
with the socket example; if we had not replaced the name, we could even have kept
decorated and non-decorated versions for different situations.
Often these decorators are general modifications that are applied permanently to
different functions. In this situation, Python supports a special syntax to apply
the decorator at the time the function is defined. We first saw it in action with the
property decorator. Instead of applying the decorator function after the method
definition, we can use the @decorator syntax to do it all at once:
@log_calls
def test1(a,b,c):
The primary benefit of this syntax is that we can easily see that the function has
been decorated at the time it is defined. If the decorator is applied later, someone
reviewing the code may miss that the function has been altered at all. Answering
a question like, "Why is my program logging function calls to the console?" can
become much more difficult! Obviously, the syntax can only be applied to functions
we define. If we need to decorate functions we didn't write, we have to use the
earlier syntax.
There is more to the decorator syntax than we've seen here. We don't have room
to cover the advanced topics here, so check the Python reference manual or other
tutorials for more information. Decorators can be created as callable objects, not
just functions that return functions. Classes can also be decorated; in that case, the
decorator returns a new class instead of a new function. Finally, decorators can take
arguments to customize them on a per-application basis.
[ 234 ]
Chapter 8
Observer pattern
The observer pattern is useful for state monitoring and event handling situations.
This pattern ensures a single core object can be monitored by an unknown, possibly
expanding, array of "observer" objects. Whenever a value on the core object changes,
it lets all the observer objects know that a change has occurred, by calling an
update() method. Each observer may be responsible for different tasks whenever
the core object changes; the core object doesn't know or care what those tasks are,
and the observers don't typically know or care what other observers are doing. Here
it is in UML:
Observer example
The observer pattern could be useful in a redundant backup system. We can write
a core object that maintains certain values, and then have one or more observers
create serialized copies of that object. These copies might be stored in a database,
on a remote host, or in a local file, for example. Let's implement the core object
using properties:
class Inventory:
def __init__(self):
self.observers = []
self._product = None
self._quantity = 0
def attach(self, observer):
self.observers.append(observer)
@property
def product(self):
return self._product
@product.setter
[ 235 ]

def product(self, value):
self._product = value
self._update_observers()
@property
def quantity(self):
return self._quantity
@quantity.setter
def quantity(self, value):
self._quantity = value
self._update_observers()
def _update_observers(self):
for observer in self.observers:
observer()
This object has two properties that, when set, call the _update_observers method
on itself. All this method does is loop over the available observers and let each one
know that something has changed. In this case, we call the observer object directly;
the object will have to implement __call__ to process the update. This would not be
possible in many object-oriented programming languages, but it's a useful shortcut
in Python that can help make our code more readable.
Now let's implement a simple observer object; this one will just print out some state
to the console:
class ConsoleObserver:
def __init__(self, inventory):
self.inventory = inventory
def __call__(self):
print(self.inventory.product)
print(self.inventory.quantity)
There's nothing terribly exciting here; the observed object is set up in the initializer,
and when the observer is called, we do "something." We can test the observer in an
interactive console:
>>> i = Inventory()
>>> c = ConsoleObserver(i)
>>> i.attach(c)
>>> i.product = "Widget"
Widget
0
>>> i.quantity = 5
Widget
5
[ 236 ]
Chapter 8
After attaching the observer to the inventory object, whenever we change one of the
two observed properties, the observer is called and its action is invoked. We can even
add two different observer instances:
>>> i = Inventory()
>>> c1 = ConsoleObserver(i)
>>> c2 = ConsoleObserver(i)
>>> i.attach(c1)
>>> i.attach(c2)
>>> i.product = "Gadget"
Gadget
0
Gadget
0
This time when we change the product, there are two sets of output, one for each
observer. The key idea here is that we can easily add totally different types of
observers that back up the data in a file, database, or internet application at the
same time.
The observer pattern detaches the code being observed from the code doing the
observing. If we were not using this pattern, we would have had to put code in
each of the properties to handle the different cases that might come up; logging
to the console, updating a database or file, and so on. The code for each of these
tasks would all be mixed in with the observed object. Maintaining it would be a
nightmare, and adding new monitoring functionality at a later date would
be painful.
Strategy pattern
The strategy pattern is a common demonstration of abstraction in object-oriented

programming. The pattern implements different solutions to a single problem
each in a different object. The client code can then choose the most appropriate
implementation dynamically at runtime.
[ 237 ]
Typically, different algorithms have different trade-offs; one might be faster than
another, but uses a lot more memory, while a third algorithm may be most suitable
when multiple CPUs are present or a distributed system is provided. Here is the
strategy pattern in UML:
The User code connecting to the strategy pattern simply needs to know that it is
dealing with the Abstraction interface. The actual implementation chosen will
perform the same task, but in possibly very different ways; either way, the interface
is identical.
Strategy example
The canonical example of the strategy pattern is sort routines; over the years,
numerous algorithms have been invented for sorting a collection of objects; quick
sort, merge sort, and heap sort are all fast sort algorithms with different features,
each useful in its own right, depending on the size and type of inputs, how out of
order they are, and the requirements of the system.
If we have client code that needs to sort a collection, we could pass it to an object
with a sort() method. This object may be a QuickSorter or a MergeSorter object;
but the result will be the same in either case: a sorted list. The strategy used to do the
sorting is abstracted from the calling code, making it modular and replaceable.
Of course, in Python, we typically just call the sorted function or list.sort method
and trust that it will do the sorting in a near-optimal fashion. So we really need to
look at a better example, don't we?
[ 238 ]
Chapter 8
Let's consider a desktop wallpaper manager. For starters, the strategy pattern might
be used to load different image formats (JPEG, GIF, PNG, TIFF) from the hard drive
and then display them. There are libraries that take care of this transparently for
us, however, so let's move a bit higher up the stack. When an image is displayed
on a desktop background, it can be adjusted to the screen size in different ways.
For example, assuming the image is smaller than the screen, it can be tiled across
the screen, centered on it, or scaled to fit. There are other, more complicated,
strategies that can be used as well, such as scaling to the maximum height or width,
combining it with a solid, semi-transparent, or gradient background color, or other
manipulations. While we may want to add these strategies later, let's start with the
basic ones.
Our strategy objects will take two inputs; the image to be displayed, and a tuple
of the width and height of the screen. They will return a new image the size of the
screen, with the image manipulated to fit according to the given strategy.
from pygame import image
from pygame.transform import scale
from pygame import Surface
class TiledStrategy:
def make_background(self, img_file, desktop_size):
in_img = image.load(img_file)
out_img = Surface(desktop_size)
for x in range((out_img.get_width(
) // in_img.get_width()) + 1):
for y in range((out_img.get_height(
) // in_img.get_height()) + 1):
out_img.blit(in_img, (in_img.get_width() * x,
in_img.get_height() * y))
return out_img
class CenteredStrategy:
out_img = Surface(desktop_size)
out_img.fill((0,0,0))
left = (out_img.get_width() - in_img.get_width()) / 2
top = (out_img.get_height() - in_img.get_height()) / 2
out_img.blit(in_img, (left, top))
return out_img
class ScaledStrategy:
return scale(in_img, desktop_size)
[ 239 ]
Here we have three strategies, each using pygame to perform their task. Individual
strategies have a make_background method that accepts the same set of parameters.
Once selected, the appropriate strategy can be called to create a correctly sized
version of the desktop image. TiledStrategy loops over the number of input
images that would fit in the width and height of the image and copies it into each
location, repeatedly. CenteredStrategy figures out how much space needs to be left
on the four edges of the image to center it. ScaledStrategy simply forces the image
to the output size (ignoring aspect ratio).
Consider how switching between these options would be implemented without the
strategy pattern. We'd need to put all the code inside one great big method and use
an awkward if statement to select the expected one. Every time we wanted to add a
new strategy, we'd have to make the method even more ungainly.
Strategy in Python
The above canonical implementation of the strategy pattern, while very common
in most object-oriented libraries, is rarely seen in Python programming. Can you
see why?
That's right: these classes each represent objects that do nothing but provide a single
function. We could just as easily call that function __call__ and make the object
callable directly. Since there is no other data associated with the object, we can
actually create a set of top-level functions and pass them around instead.
Opponents of design pattern philosophy will therefore say, "because Python has
first-class functions, the strategy pattern is unnecessary". In truth, Python's first-class
functions allow us to implement the strategy pattern in a more straightforward way.
Knowing the pattern exists can still help us choose a correct design for our program;
we simply implement it using different syntax. The strategy pattern, or a top-level
function implementation of it, should be used when we need to allow client code or
the end user to select between multiple implementations of the same interface.
State pattern
State pattern is structurally similar to strategy pattern, but its intent and purpose are
very different. The goal of the state pattern is to represent state-transition systems:
systems where it is obvious that an object can be in a specific state, and that certain
activities may drive it to a different state.
[ 240 ]
Chapter 8
To make this work, we need a manager, or context class that provides an interface
for switching states. Internally, this class contains a pointer to the current state; each
state knows what other states it is allowed to be in and will transition to those states
depending on actions invoked upon it.
So we have two types of classes, the context class and multiple state classes. The
context class maintains the current state, and forwards actions to the state classes.
The state classes are typically hidden from any other objects that are calling the
context; it acts like a black box that happens to perform state management internally.
Here's how it looks in UML:
State example
To illustrate the state pattern, let's build an XML parsing tool. The context class will
be the parser itself. It will take a string as input and place the tool in an initial parsing
state. The various parsing states will eat characters, looking for a specific value, and
when that value is found, change to a different state. The goal is to create a tree of
node objects for each tag and its contents. To keep things manageable, we'll parse
only a subset of XML; tags and tag names. We won't be able to handle attributes on
tags. It will parse text content of tags, but won't attempt to parse "mixed" content,
which has tags inside of text. Here is an example "simplified XML" file that we'll be
able to parse:
<book>
<author>Dusty Phillips</author>
<publisher>Packt Publishing</publisher>
<title>Python 3 Object Oriented Programming</title>
<content>
<chapter>
<number>1</number>
<title>Object Oriented Design</title>
</chapter>
[ 241 ]

<chapter>
<number>2</number>
<title>Objects In Python</title>
</chapter>
</content>
</book>
Before we look at the states and the parser, let's consider the output of this program.
We know we want a tree of Node objects, but what does a Node look like? Well,
clearly it'll need to know the name of the tag it is parsing, and since it's a tree, it
should probably maintain a pointer to the parent node and a list of the node's
children in order. Some nodes have a text value, but not all of them. Let's look
at the Node class first:
class Node:
def __init__(self, tag_name, parent=None):
self.parent = parent
self.tag_name = tag_name
self.children = []
self.text=""
def __str__(self):
if self.text:
return self.tag_name + ": " + self.text
else:
return self.tag_name
This class sets default attribute values upon initialization. The __str__ method is
supplied to help visualize the tree structure when we're finished.
Now, looking at the example document, we need to consider what states our parser
can be in. Clearly it's going to start in a state where no nodes have yet been processed.
We'll need a state for processing opening tags and closing tags. And when we're inside
a tag with text contents, we'll have to process that as a separate state, too.
Switching states can be tricky; how do we know if the next node is an opening tag, a
closing tag, or a text node.? We could put a little logic in each state to work that out,
but it actually makes more sense to create a new state whose sole purpose is figuring
out which state we'll be switching to next. If we call this transition state ChildNode,
we end up with the following states:
FirstTag
ChildNode
OpenTag
CloseTag
Text
[ 242 ]
Chapter 8
The FirstTag state will switch to ChildNode, which is responsible for deciding which
of the other three states to switch to; when those states are finished, they'll switch
back to ChildNode. The following state-transition diagram shows the available
state changes:
The states are responsible for taking "what's left of the string", processing as much of
it as they know what to do with, and then telling the parser to take care of the rest of
it. Let's construct the Parser class first:
class Parser:
def __init__(self, parse_string):
self.parse_string = parse_string
self.root = None
self.current_node = None
self.state = FirstTag()
def process(self, remaining_string):
remaining = self.state.process(remaining_string, self)
if remaining:
self.process(remaining)
def start(self):
self.process(self.parse_string)
The initializer sets up a few variables on the class that the individual states will
access. The parse_string is the text that we are trying to parse. The root node is the
"top" node in the XML structure. The current_node is the one that we are currently
adding children to.
[ 243 ]
The important feature of this parser is the process method, which accepts the
remaining string, and passes it off to the current state. The parser (the self
argument) is also passed into the state's process method so that the state can
manipulate it. The state is expected to return the remainder of the unparsed string
when it is finished processing. The parser then recursively calls the process method
on this remaining string to construct the rest of the tree.
Now, let's have a look at the FirstTag state:
class FirstTag:
def process(self, remaining_string, parser):
i_start_tag = remaining_string.find('<')
i_end_tag = remaining_string.find('>')
tag_name = remaining_string[i_start_tag+1:i_end_tag]
root = Node(tag_name)
parser.root = parser.current_node = root
parser.state = ChildNode()
return remaining_string[i_end_tag+1:]
This state finds the index (the i_ stands for index) of the opening and closing angle
brackets on the first tag. You may think this state is unnecessary, since XML requires
that there be no text before an opening tag. However, there may be whitespace that
needs to be consumed; this is why we search for the opening angle bracket instead
of assuming it is the first character in the document. Note that this code is assuming
a valid input file. A proper implementation would be religiously testing for invalid
input, and would attempt to recover or display a very descriptive error message.
The method extracts the name of the tag and assigns it to the root node of the parser.
It also assigns it as the current_node, since that's the one we'll be adding children
to next.
Then comes the important part; the method changes the current state on the parser
object to a ChildNode state. It then returns the remainder of the string (after the
opening tag) to allow it to be processed.
The ChildNode state, which seems quite complicated, turns out to require nothing
but a simple conditional:
class ChildNode:
stripped = remaining_string.strip()
if stripped.startswith("</"):
parser.state = CloseTag()
elif stripped.startswith("<"):
parser.state = OpenTag()
else:
parser.state = TextNode()
return stripped
[ 244 ]
Chapter 8
The strip() call removes whitespace from the string. Then the parser determines
if the next item is an opening or closing tag, or a string of text. Depending on which
possibility occurs, it sets the parser to a particular state, and then tells it to parse the
remainder of the string.
The OpenTag state is similar to the FirstTag state, except that it adds the newly
created node to the previous current_node object's children and sets it as the new
current_node. It places the processor back in the ChildNode state before continuing:
class OpenTag:
node = Node(tag_name, parser.current_node)
parser.current_node.children.append(node)
parser.current_node = node
CloseTag basically does the opposite; it sets the parser's current_node back to the
parent node so any further children in the outside tag can be added to it:
class CloseTag:
assert remaining_string[i_start_tag+1] == "/"
assert tag_name == parser.current_node.tag_name
parser.current_node = parser.current_node.parent
return remaining_string[i_end_tag+1:].strip()
The two assert statements help ensure that the parse strings are consistent. The
if statement at the end of the method simply ensures that the processor terminates
when it is finished. If the parent of a node is None, it means that we are working on
the root node.
Finally, the TextNode state very simply extracts the text before the next close tag and
sets it as a value on the current node:
class TextNode:
text = remaining_string[:i_start_tag]
[ 245 ]

parser.current_node.text = text
return remaining_string[i_start_tag:]
Now we just have to set up the initial state on the parser object we created. The initial
state is a FirstTag object, so just add the following to the __init__ method:
self.state = FirstTag()
To test the class, let's add a main script that opens an file from the command line,
parses it, and prints the nodes:
if __name__ == "__main__":
import sys
with open(sys.argv[1]) as file:
p = Parser(contents)
p.start()
nodes = [p.root]
while nodes:
node = nodes.pop(0)
print(node)
nodes = node.children + nodes
All it does is open the file, load the contents, and parse them. Then it prints each
node and its children in order. The __str__ method we originally added on the
node class takes care of formatting the nodes for printing. If we run the script on
the earlier example, it outputs the tree as follows:
book
author: Dusty Phillips
publisher: Packt Publishing
title: Python 3 Object Oriented Programming
content
chapter
number: 1
title: Object Oriented Design
chapter
number: 2
title: Objects In Python
Comparing this to the original simplified XML document tells us the parser
is working.
[ 246 ]
Chapter 8
State versus strategy
State pattern looks very similar to strategy pattern, indeed the UML diagrams for
the two are identical. The implementation, too, is identical; we could even have
written our states as first-class functions instead of wrapping them in objects,
as was suggested for strategy.
While the two patterns have identical structures, their purposes are very different.
The strategy pattern is used to choose an algorithm at runtime; generally, only one of
those algorithms is going to be chosen for a particular use case. The state pattern, on
the other hand is designed to allow switching between different states dynamically,
as some process evolves. In code, the primary difference is that the strategy pattern
is not typically aware of other strategy objects. In the state pattern, either the state or
the context needs to know which other states that it can switch to.
Singleton pattern
The singleton pattern is one of the most controversial patterns; many have accused
it of being an "anti-pattern"; a pattern that should be avoided, not promoted. In
Python, if someone is using the singleton pattern, they're almost certainly doing
something wrong, probably because they're coming from a more restrictive
programming language.
So why discuss it at all? Singleton is one of the most famous of all design
patterns. It is very useful in overly object-oriented languages, and is a vital
part of object-oriented programming. Finally, the idea behind singleton is
useful, even if we implement that idea in a totally different way in Python.
The basic idea behind the singleton pattern is to allow exactly one instance of a
certain object to exist. Typically, this object is a sort of manager class like those we
discussed in Chapter 5. Such objects often need to be referenced by a wide variety of
other objects, and passing references to the manager object around to the methods
and constructors that need them can make code hard to read.
Instead, when a singleton is used, the separate objects request the single instance
of the manager object from the class, so a reference to it does not need to be passed
around. The UML diagram doesn't fully describe it, but here it is for completeness:
[ 247 ]
In most programming environments, singletons are enforced by making the

constructor private (so no-one can create additional instances of it) and then
providing a static method to retrieve the single instance. This method will create a
new instance the first time it is called, and then return that same instance whenever
it is called again.
Singleton implementation
Python doesn't have private constructors, but for this purpose, it has something
even better. We can use the __new__ class method to ensure that only one instance
is ever created:
class OneOnly:
_singleton = None
def __new__(cls, *args, **kwargs):
if not cls._singleton:
cls._singleton = super(OneOnly, cls
).__new__(cls, *args, **kwargs)
return cls._singleton
When __new__ is called, it normally constructs a new instance of that class. When we
override it, we first check if our singleton instance has been created; if not, we create
it using a super call. Thus, whenever we call the constructor on OneOnly, we always
get the exact same instance:
>>> o1 = OneOnly()
>>> o2 = OneOnly()
>>> o1 == o2
True
>>> o1
<__main__.OneOnly object at 0xb71c008c>
>>> o2
<__main__.OneOnly object at 0xb71c008c>
The two objects are equal and located at the same address; they are the same object.
This particular implementation isn't very transparent, since it's not obvious that a
singleton object has been created. Whenever we call a constructor, we expect a new
instance of that object; in this case we do not get that. Perhaps, good docstrings on
the class could alleviate this problem if we really think we need a singleton.
[ 248 ]
Chapter 8
But we don't need it. Python coders frown on forcing the users of their code into a
specific mindset. We may think only one instance of a class will ever be required,
but other programmers may have different ideas. Singletons can interfere with
distributed computing, parallel programming, and automated testing, for example.
In all those cases it can be very useful to have multiple or alternative instances of a
specific object, even though "normal' operation may never require one.
Module variables can mimic singletons
Normally, in Python, the singleton pattern can be sufficiently mimicked by using

module-level variables. It's not as "safe" as a singleton in that people could reassign
those variables at any time, but like our discussion of private variables in Chapter 2,
this is acceptable in Python. If someone has a valid reason to change those variables,
why should we stop them? It also doesn't stop people from instantiating multiple
instances of the object, but again, if they have a valid reason to do so, why interfere?
Ideally, we should give them a mechanism to get access to the "default singleton"
value, while also allowing them to create other instances if they need them. While,
technically, it is not a singleton at all, it provides the most Pythonic mechanism for
behaving like a singleton.
To use module-level variables instead of a singleton, we simply instantiate an
instance of the class after we've defined it. We can improve our state pattern to use
singletons. Instead of creating a new object every time we change states, we can
create a module-level variable that is always accessible:
class FirstTag:
root = Node(tag_name)
parser.root = parser.current_node = root
parser.state = child_node
class ChildNode:
stripped = remaining_string.strip()
if stripped.startswith("</"):
parser.state = close_tag
elif stripped.startswith("<"):
parser.state = open_tag
else:
[ 249 ]

parser.state = text_node
return stripped
class OpenTag:
node = Node(tag_name, parser.current_node)
parser.current_node.children.append(node)
parser.current_node = node
class TextNode:
text = remaining_string[:i_start_tag]
parser.current_node.text = text
return remaining_string[i_start_tag:]
class CloseTag:
assert remaining_string[i_start_tag+1] == "/"
assert tag_name == parser.current_node.tag_name
parser.current_node = parser.current_node.parent
return remaining_string[i_end_tag+1:].strip()
first_tag = FirstTag()
child_node = ChildNode()
text_node = TextNode()
open_tag = OpenTag()
close_tag = CloseTag()
All we've done is create instances of the various state classes that can be reused.
Notice how we can access these module variables inside the classes, even before
the variables have been defined? This is because the code inside the classes is not
executed until the method is called, and by that point, the entire module will have
been defined.
[ 250 ]
Chapter 8
The difference in this example is that instead of wasting memory creating a bunch of
new instances that must be garbage collected, we are reusing a single state object for
each state. Even if multiple parsers are running at once, only these state classes need
to be used. Of course, if someone wants to create their own instances, they can. So it's
not a true singleton, but convention can strongly suggest that a singleton paradigm
be used. If someone wants to mess with that paradigm... well, they'll have to deal
with the consequences, beneficial or detrimental.
When we originally created the state-based parser, you may have wondered why we
didn't pass the parser object to __init__ on each individual state, instead of passing it
into the process method as we did. The state could then have been referenced as self.
parser. This is a perfectly valid implementation of the state pattern, but it would not
have allowed leveraging the singleton pattern. If the state objects maintain a reference
to the parser, then they cannot be used simultaneously to reference other parsers.
The thing to remember is that these are two very different patterns with different
purposes; the fact that singleton's purpose may be useful for implementing the state
pattern does not, in any way, mean the two patterns are related.
Template pattern
The template pattern is useful for removing duplicate code; it's an implementation
to support the "Don't Repeat Yourself" principle we discussed in Chapter 5. It is
designed for situations where we have several different tasks to accomplish that
have some, but not all, steps in common. The common steps are implemented in a
base class, and the different steps are overridden in subclasses to provide custom
behavior. In some ways, it's like a generalized strategy pattern, except similar
sections of the algorithms are shared using a base class. Here it is in UML format:
[ 251 ]
Template example
Let's create a car sales reporter as an example. We can store records of sales in an
SQLite database table. SQLite is a simple file-based database engine that allows us to
store records using SQL syntax. Python 3 includes SQLite in its standard library, so
there is no setup involved.
We have two common tasks we need to perform:
Select all sales of new vehicles and output them to the screen in a commadelimited format
Output a comma-delimited list of all salespeople with their gross sales and
save it to a file that can be imported to a spreadsheet
These seem like quite different tasks, but they have some common features. In both
cases, we need to perform the following steps:
1. Connect to the database.
2. Construct a query for new vehicles or gross sales.
3. Issue the query.
4. Format the results into a comma-delimited string.
5. Output the data to a file or e-mail.
The query construction and output steps are different for the two tasks, but the other
steps are identical for both. We can use the template pattern to put the common steps
in a base class, and the varying steps in two subclasses.
Before we start, let's create a database and put some sample data in it, using a few
lines of SQL:
import sqlite3
conn = sqlite3.connect("sales.db")
conn.execute("CREATE TABLE Sales (salesperson text, "
"amt currency, year integer, model text, new boolean)")
conn.execute("INSERT INTO Sales values"
" ('Tim', 16000, 2010, 'Honda Fit', 'true')")
" ('Tim', 9000, 2006, 'Ford Focus', 'false')")
" ('Gayle', 8000, 2004, 'Dodge Neon', 'false')")
" ('Gayle', 28000, 2009, 'Ford Mustang', 'true')")
[ 252 ]
Chapter 8
" ('Gayle', 50000, 2010, 'Lincoln Navigator', 'true')")
" ('Don', 20000, 2008, 'Toyota Prius', 'false')")
conn.commit()
conn.close()
Hopefully you can see what's going on here even if you don't know SQL; we've
created a table to hold the data, and used six insert statements to add sales records.
The data is stored in a file named sales.db. Now we have a sample we can work
with in developing our template pattern.
Since we've already outlined the steps that the template has to perform, we can
start by defining the base class that contains the steps. Each step gets its own
method (to make it easy to selectively override any one step), and we have one
more all-encompassing method that calls the steps in turn. Without any method
contents, here's how it might look:
class QueryTemplate:
def connect(self):
pass
def construct_query(self):
pass
def do_query(self):
pass
def format_results(self):
pass
def output_results(self):
pass
def process_format(self):
self.connect()
self.construct_query()
self.do_query()
self.format_results()
self.output_results()
[ 253 ]
The process_format method is the primary method to be called by an outside

client. It ensures each step is executed in order, but it does not care if that step is
implemented in this class or in a subclass. For our examples, we know that three
methods are going to be identical between our two classes:
import sqlite3
class QueryTemplate:
def connect(self):
self.conn = sqlite3.connect("sales.db")
raise NotImplementedError()
def do_query(self):
results = self.conn.execute(self.query)
self.results = results.fetchall()
def format_results(self):
output = []
for row in self.results:
row =[str(i) for i in row]
output.append(", ".join(row))
self.formatted_results = "\n".join(output)
raise NotImplementedError()
To help with implementing subclasses, the two methods that are not specified
raise NotImplementedError. This is a common way to specify abstract interfaces
in Python. The methods could have empty implementations (with pass), or could
even be fully unspecified. Raising NotImplementedError, however, helps the
programmer understand that the class is meant to be subclassed and these methods
overridden; empty methods or methods that simply do not exist are harder to
identify as needing to be implemented and to debug if we forget to implement them.
Now we have a template class that takes care of the boring details, but is flexible
enough to allow the execution and formatting of a wide variety of queries. The
best part is, if we ever want to change our database engine from SQLite to another
database engine (such as py-postgresql), we only have to do it here, in this template
class, and don't have to touch the two (or two hundred) subclasses we've written.
[ 254 ]
Chapter 8
Let's have a look at the concrete classes now:

import datetime
class NewVehiclesQuery(QueryTemplate):
self.query = "select * from Sales where new='true'"
print(self.formatted_results)
class UserGrossQuery(QueryTemplate):
self.query = ("select salesperson, sum(amt) " +
" from Sales group by salesperson")
filename = "gross_sales_{0}".format(
datetime.date.today().strftime("%Y%m%d")
)
with open(filename, 'w') as outfile:
outfile.write(self.formatted_results)
These two classes are actually pretty short, considering what they're doing:
connecting to a database, executing a query, formatting the results, and outputting
them. The superclass takes care of the repetitive work, but lets us easily specify
those steps that vary between tasks. Further, we can also easily change steps that
are provided in the base class. For example, if we wanted to output something other
than a comma-delimited string (for example: an HTML report to be uploaded to a
website), we can still override format_results.
Exercises
While writing this chapter, I discovered that it can be very difficult, and extremely
educational to come up with good examples where specific design patterns should
be used. Instead of going over current or old projects to see where you can apply
these patterns, as I've suggested in previous chapters, think about the patterns
and different situations where they might come up. Try to think outside your own
experiences. If your current projects are in the banking business, consider how you'd
apply these design patterns in a retail or point-of-sale application. If you normally
write web applications, think about using design patterns while writing a compiler.
[ 255 ]
Start with the iterator pattern. We've been looking at iterators all through this book,
including the special comprehension and generator syntaxes. Consider places where
you'd want to implement the iterator pattern from scratch; what objects would you
want to implement __iter__ or __next__ on?
Look at the decorator pattern and come up with some good examples of when to
apply it. Focus on the pattern itself, not the Python syntax we discussed; it's a bit
more general than the actual pattern. The special syntax for decorators is, however,
something you might want to look for places to apply in existing projects too.
What are some good areas to use the observer pattern? Why? Think about not only
how you'd apply the pattern, but how you would implement the same task without
using observer? What do you gain, or lose, by choosing to use it?
Consider the difference between the strategy and state patterns. Implementation-wise,
they look very similar, yet they have different purposes. Can you think of cases where
the patterns could be interchanged? Would it be reasonable to redesign a state-based
system to use strategy instead, or vice-versa? How much different would the design
actually be?
The template pattern is such an obvious application of inheritance to reduce
duplicate code that you may have used it before, without knowing its name. Try to
think of at least half a dozen different scenarios where it would be useful. If you can
do that, you'll be finding places for it in your daily coding all the time.
Summary
In this chapter, we learned that design patterns are useful abstractions that provide
"best practice" solutions for common programming problems. We understood that
design patterns in Python, due to its dynamic nature and built-in syntax, can look
very different from their usual renditions in other languages. We discussed several
patterns in detail, with examples, UML diagrams and a discussion of the differences
between Python and statically typed object-oriented languages. We covered:
What design patterns are

The iterator pattern
The decorator pattern
The observer pattern
The strategy and state patterns
The template pattern
In the next chapter, we'll discuss several more useful design patterns and their
application in Python.
[ 256 ]
Python Design Patterns II

This chapter carries on from the previous chapter by introducing several more
design patterns. Once again, we'll cover the canonical examples as well as any
common alternative implementations in Python. We'll be discussing:
The adapter pattern
The facade pattern
Lazy initialization and the flyweight pattern
The command pattern
The abstract factory pattern
The composition pattern
Adapter pattern
Unlike most of the patterns we reviewed in Chapter 8, the adapter pattern is

designed to interact with existing code. We would not design a brand new set of
objects that implement the adapter pattern. Adapters are used to allow two preexisting objects to work together, even if their interfaces are not compatible. Like
the keyboard adapters that allow USB keyboards to be plugged into PS/2 ports,
an adapter object sits between two different interfaces, translating between them
on the fly. The adapter object's sole purpose is to perform this translating job;
translating may entail a variety of tasks, such as converting arguments to a different
format, rearranging the order of arguments, calling a differently named method, or
supplying default arguments.
In structure, the adapter pattern is similar to a simplified decorator pattern.

Decorators typically provide the same interface that they replace, whereas
adapters map between two different interfaces. Here it is in UML form:
Here, Interface1 is expecting to call a method called make_action(some, arguments).

We already have this perfect Interface2 class that does everything we want (and
to avoid duplication, we don't want to rewrite it!), but it provides a method called
different_action(other, arguments) instead. The Adapter class implements the
make_action interface and maps the arguments to the existing interface.
The advantage here is that the code that maps from one interface to another is all
in one place. The alternative would be to translate it directly in multiple places
whenever we need to access this code.
For example, imagine we have the following pre-existing class, which takes a string
date in the format "YYYY-MM-DD" and calculates a person's age on that day:
class AgeCalculator:
def __init__(self, birthday):
self.year, self.month, self.day = (
int(x) for x in birthday.split('-'))
def calculate_age(self, date):
year, month, day = (
int(x) for x in date.split('-'))
age = year - self.year
if (month,day) < (self.month,self.day):
age -= 1
return age
This is a pretty simple class that does what it's supposed to do. But we have to
wonder what the programmer was thinking, using a specifically formatted string
instead of using Python's incredibly useful built-in datetime library. Most programs
we write are going to be interacting with datetime objects, not strings.
[ 258 ]
Chapter 9
We have several options to address this scenario; we could rewrite the class to accept
datetime objects, which would probably be more accurate anyway. But if this class
has been provided by a third party and we don't know what its internals are, or
we simply aren't allowed to change them, we need to try something else. We could
use the class as it is, and whenever we want to calculate the age on a datetime.
date object, we could call datetime.date.strftime('%Y-%m-%d') to convert it to
the proper format. But that conversion would be happening in a lot of places, and
worse, if we mistyped the %m as %M it would give us the current instead of the entered
month! Imagine if you wrote that in a dozen different places only to have to go back
and change it when you realized your mistake. It's not maintainable code, and it
breaks the DRY principle.
Or, we can write an adapter that allows a normal date to be plugged into a normal
AgeCalculator:
import datetime
class DateAgeAdapter:
def _str_date(self, date):
return date.strftime("%Y-%m-%d")
def __init__(self, birthday):
birthday = self._str_date(birthday)
self.calculator = AgeCalculator(birthday)
def get_age(self, date):
date = self._str_date(date)
return self.calculator.calculate_age(date)
This adapter converts datetime.date and datetime.time (they have the same
interface to strftime) into a string that our original AgeCalculator can use. Now
we can use the original code with our new interface. I changed the method signature
to get_age to demonstrate that the calling interface may also be looking for a
different method name, not just a different type of argument.
Creating a class as an adapter is the usual way to implement this as a pattern, but, as
usual, there are other ways to do it. Inheritance and multiple inheritance can be used
to add functionality to a class. For example, we could add an adapter on the date
class so that it works with the original AgeCalculator:
import datetime
class AgeableDate(datetime.date):
def split(self, char):
return self.year, self.month, self.day
[ 259 ]
It's code like this that makes one wonder if Python should even be legal. All we've
done here is add a split method that takes a single argument (which we ignore) and
returns a tuple of year, month, day. This works flawlessly with our AgeCalculator,
because that code calls strip on a specially formatted string, and strip, in that case
returns a tuple of year, month, day. The AgeCalculator code only cares if strip
exists and returns acceptable values; it doesn't care if we really passed in a string. It
really works:
>>> bd = AgeableDate(1975, 6, 14)
>>> today = AgeableDate.today()
>>> today
AgeableDate(2010, 2, 23)
>>> a = AgeCalculator(bd)
>>> a.calculate_age(today)
34
In this particular instance, such an adapter would be hard to maintain, as we'll soon
forget why we needed to add a strip method to a date class. The method name
is quite ambiguous. That can be the nature of adapters, but if we had created an
adapter explicitly instead of using inheritance, it's more obvious what its purpose is.
Instead of inheritance, you can sometimes also use monkey-patching to add a
method to an existing class. It won't work with the datetime object, as it won't
allow attributes to be added at runtime, but in normal classes, we can just add
a new method that provides the adapted interface that is required by calling code.
It can also be possible to use a function as an adapter; this doesn't really fit the
adapter pattern properly, but often, you can simply pass data into a function
and return it in the proper format for entry into another interface.
Facade pattern
The facade pattern is designed to provide a simple interface to a complex system

of components. The objects in this system may need to be interacted with directly
for complex tasks and interactions. Often, however, there is 'typical' usage for the
system, and these complicated interactions aren't necessary in that common scenario.
The facade pattern allows us to define a new object that wraps this typical usage
of the system. Any code that wants to use the typical functionality can use the
single object's simplified interface. If another project or part of the project finds this
interface is too simple and needs to access more complicated functionality, it is still
able to interact with the system directly. The UML diagram for the facade pattern is
really dependent on the subsystem, but in a cloudy way, it looks like this:
[ 260 ]
Chapter 9
Facade is, in many ways, like adapter. The primary difference is that the facade is
trying to abstract a simpler interface out of a complex system; the adapter is only
trying to map one existing interface to another.
Let's write a simple facade for an e-mail application. The low-level library for
sending e-mail in Python, as we saw in Chapter 7, is quite complicated. The two
libraries for receiving messages are even worse.
It would be nice to have a simple class that allows us to send a single e-mail, and
list the e-mails currently in the inbox on an IMAP or POP3 connection. To keep our
example short, we'll stick with IMAP and SMTP: two totally different subsystems
that happen to deal with e-mail. Our facade will perform only two tasks; sending an
e-mail to a specific address, and checking the inbox on an IMAP connection. It will
make some common assumptions about the connection, such as the host for both
SMTP and IMAP is the same machine, that the username and password for both is
the same, and that they use standard ports. This will cover the case for many e-mail
servers, but if a programmer needs more flexibility, they can always bypass the
facade and access the two subsystems directly.
The class is initialized with the hostname of the e-mail server and a username and
password to log in:
import smtplib
import imaplib
class EmailFacade:
def __init__(self, host, username, password):
self.host = host
self.password = password
[ 261 ]
The send_email method simply formats the e-mail address and message and sends
it using smtplib. This isn't a complicated task, but it requires quite a bit of fiddling
to massage the "natural" input parameters that are passed into the facade to put them
in the correct format for smtplib to send the message:
def send_email(self, to_email, subject, message):
if not "@" in username:
from_email = "{0}@{1}".format(
self.username, self.host)
else:
from_email = self.username
message = ("From: {0}\r\n"
"To: {1}\r\n"
"Subject: {2}\r\n\r\n{3}").format(
from_email,
to_email,
subject,
message)
smtp = smtplib.SMTP(self.host)
smtp.login(self.username, self.password)
smtp.sendmail(from_email, [to_email], message)
The if statement at the beginning of the method is catching whether or not the
username is the entire from e-mail address or just the part on the left side of the @
symbol; different hosts treat the login details differently. Finally, the code to get the
messages currently in the inbox is a ruddy mess; the IMAP protocol is painfully
over-engineered, and the imaplib standard library is only a thin layer over
the protocol:
def get_inbox(self):
mailbox = imaplib.IMAP4(self.host)
mailbox.login(bytes(self.username, 'utf8'),
bytes(self.password, 'utf8'))
mailbox.select()
x, data = mailbox.search(None, 'ALL')
messages = []
for num in data[0].split():
x, message = mailbox.fetch(num, '(RFC822)')
messages.append(message[0][1])
return messages
Now, if we add all this together, we have a simple facade class that can send and
receive messages in a fairly straightforward manner, much simpler than if we had to
interact with the complex libraries directly.
[ 262 ]
Chapter 9
Flyweight pattern
The flyweight pattern is a memory optimization pattern. Novice Python

programmers tend to ignore memory optimization, assuming the built-in
garbage collector will take care of them. This is often perfectly acceptable, but
when developing larger applications with many related objects, paying attention
to memory concerns can have a huge payoff.
In real life, the flyweight pattern is often implemented only after a program
has demonstrated memory problems. It may make sense to design an optimal
configuration from the beginning in some situations, but bear in mind that
premature optimization is the most effective way of ensuring that your
program is too complicated to maintain.
The basic idea behind the flyweight pattern is to ensure that objects that share a
state can use the same memory for that shared state. Think of an inventory system
for car sales. Each individual car has a specific serial number and is a specific color.
But most of the details about that car are the same for all cars of a particular model.
For example, the Honda Fit DX model is a bare-bones car with few features. The
LX model has A/C, tilt, cruise, and power windows and locks. The Sport model
has fancy wheels, a USB charger, and a spoiler. (I recently bought a Fit LX, which I
love; that's why we're looking at a second car sales example in as many chapters!)
Without the flyweight pattern, each individual car object would have to store a
long list of which features it did and did not have. Considering the number of cars
Honda sells in a year, this would add up to a huge amount of wasted memory. Using
the flyweight pattern, we can instead have shared objects for the list of features
associated with a model, and then simply reference that model, along with a serial
number and color, for individual vehicles.
Let's have a look at the UML diagram for the flyweight pattern:
[ 263 ]
Each Flyweight has no specific state; any time it needs to perform an operation on
SpecificState, that state needs to be passed into the Flyweight by the calling code.
Traditionally, the factory that returns a flyweight is a separate object; its purpose
is to return a flyweight for a given key identifying that flyweight. It works like the
singleton pattern we discussed in Chapter 8; if the flyweight exists, we return it;
otherwise we create a new one. In many languages, the factory is implemented,
not as a separate object, but as a static method on the Flyweight class itself.
Both of these options work, but in Python, the flyweight factory is often
implemented using that funky __new__ constructor, similar to what we did with
the singleton pattern. Unlike singleton, which only needs to return one instance of
the class, we need to be able to return different instances depending on the keys.
We could store the items in a dictionary and look them up based on the key. This
solution is problematic, however, because the item will remain in memory so long as
it is in the dictionary. If we sold out of LX model Fits, the Fit flyweight is no longer
necessary, yet it will still be in the dictionary. We could, of course, clean this up
whenever we sell a car, but isn't that what a garbage collector is for?
The Python weakref module comes to our rescue. This module provides a
WeakValueDictionary object, which basically allows us to store items in a dictionary
without the garbage collector caring about them. If a value is in a weak referenced
dictionary and there are no other references to that object stored anywhere (that is,
we sold out of LX models), the garbage collector will eventually clean up for us.
Let's build the factory for our car flyweights first:
import weakref
class CarModel:
_models = weakref.WeakValueDictionary()
def __new__(cls, model_name, *args, **kwargs):
model = cls._models.get(model_name)
if not model:
model = super().__new__(cls)
cls._models[model_name] = model
return model
Basically, whenever we construct a new flyweight with a given name, we first look
up that name in the weak referenced dictionary; if it exists, we return that model,
if not, we create a new one. Either way, we know the __init__ method on the
flyweight will be called every time, regardless of whether it is a new or existing
object. Our __init__ can therefore look like this:
[ 264 ]
Chapter 9
def __init__(self, model_name, air=False, tilt=False,
cruise_control=False, power_locks=False,
alloy_wheels=False, usb_charger=False):
if not hasattr(self, "initted"):
self.model_name = model_name
self.air = air
self.tilt = tilt
self.cruise_control = cruise_control
self.power_locks = power_locks
self.alloy_wheels = alloy_wheels
self.usb_charger = usb_charger
self.initted=True
The if statement ensures that we only initialize the object the first time
__init__ is called. This means we can call the factory later with just the
model name and get the same flyweight object back. However, because the
flyweight will be garbage-collected if no external references to it exist, we have
to be careful not to accidentally create a new flyweight with null values.
Let's add a method to our flyweight that hypothetically looks up a serial number on
a specific model of vehicle, and determines if it has been involved in any accidents.
This method needs access to the car's serial number, which varies from car to car;
it cannot be stored with the flyweight. Therefore, this data must be passed into the
method by the calling code:
def check_serial(self, serial_number):
print("Sorry, we are unable to check "
"the serial number {0} on the {1} "
"at this time".format(
serial_number, self.model_name))
We can define a class that stores the additional information, as well as a reference to
the flyweight:
class Car:
def __init__(self, model, color, serial):
self.model = model
self.color = color
self.serial = serial
def check_serial(self):
return self.model.check_serial(self.serial)
[ 265 ]
We can also keep track of the available models as well as the individual cars on
the lot:
>>> dx = CarModel("FIT DX")
>>> lx = CarModel("FIT LX", air=True, cruise_control=True,
... power_locks=True, tilt=True)
>>> car1 = Car(dx, "blue", "12345")
>>> car2 = Car(dx, "black", "12346")
>>> car3 = Car(lx, "red", "12347")
Now, let's demonstrate the weak referencing at work:

>>> id(lx)
3071620300
>>> del lx
>>> del car3
>>> import gc
>>> gc.collect()
0
>>> lx = CarModel("FIT LX", air=True, cruise_control=True,
... power_locks=True, tilt=True)
>>> id(lx)
3071576140
>>> lx = CarModel("FIT LX")
>>> id(lx)
3071576140
>>> lx.air
True
The id function tells us the unique identifier for an object. When we call it a second
time, after deleting all references to the LX model and forcing garbage collection, we
see that the id has changed. The value in the CarModel __new__ factory dictionary
was deleted and a fresh one created. If we then try to construct a second CarModel
instance, however, it returns the same object (the ids are the same), and, even though
we did not supply any arguments in the second call, the air variable is still set to
True. This means the object was not initialized the second time, just as we designed.
Obviously, using the flyweight pattern can be more complicated than just storing
features on a single car class. When should we choose to use it? The flyweight
pattern is designed for conserving memory; if we have tens or hundreds of
thousands of similar objects, combining similar properties into a flyweight
can have an enormous impact on memory consumption.
[ 266 ]
Chapter 9
Command pattern
The command pattern adds a level of abstraction between actions that must be done,
and the object that invokes those actions, normally at a later time. In the command
pattern, client code creates a Command object that can be executed at a later date.
This object knows about a receiver object that manages its own internal state when
the command is executed on it. The Command object implements a specific interface
(typically it has an execute or do_action method, and also keeps track of any
arguments required to perform the action. Finally, one or more Invoker objects
execute the command at the correct time. Here's the UML diagram:
A common example of command pattern in action is actions on a graphical window.

Often, an action can be invoked by a menu item on the menu bar, a keyboard
shortcut, a toolbar icon, or a context menu. These are all examples of Invoker
objects. The actions that actually occur, such as Exit, Save, or Copy are all command
implementations of CommandInterface. A GUI Window to receive exit, document to
receive save, and ClipboardManager to receive copy commands are all examples of
possible Receivers.
Let's implement a simple command pattern that provides commands for Save and
Exit actions. We'll start with some modest receiver classes:
import sys
class Window:
def exit(self):
sys.exit(0)
class Document:
[ 267 ]

self.contents = "This file cannot be modified"
def save(self):
with open(self.filename, 'w') as file:
file.write(self.contents)
These mock classes model objects that would likely be doing a lot more in a working
environment. The window would need to handle mouse movement and keyboard
events, and the document would need to handle character insertion, deletion, and
selection. But for our purposes these two classes will do what we need.
Now let's define some invoker classes. These will model toolbar, menu, and
keyboard events that can happen; again, they aren't actually hooked up to anything,
but we can see how they are decoupled from the command, receiver, and client code:
class ToolbarButton:
def __init__(self, name, iconname):
self.name = name
self.iconname = iconname
def click(self):
self.command.execute()
class MenuItem:
def __init__(self, menu_name, menuitem_name):
self.menu = menu_name
self.item = menuitem_name
def click(self):
class KeyboardShortcut:
def __init__(self, key, modifier):
self.key = key
self.modifier = modifier
def keypress(self):
Notice how the various action methods each call the execute method on their
respective commands? The commands haven't actually been set on the objects; they
could be passed into the __init__ function, but because they may be changed (for
example, with a customizable keybinding editor), we can set the attributes on the
objects afterwards. Different situations call for different designs, but Python gives
us the flexibility to do what makes the most sense to us, the programmers.
[ 268 ]
Chapter 9
Now, let's hook up the commands:

class SaveCommand:
def execute(self):
self.document.save()
class ExitCommand:
def __init__(self, window):
self.window = window
def execute(self):
self.window.exit()
These commands are extremely straightforward; they demonstrate the basic pattern,
but it is important to note that we can store state and other information with the
command if necessary. For example, if we had a command to insert a character, we
could maintain state for the character currently being inserted.
Now all we have to do is hook up some client and test code to make the commands
work. For basic testing, we can just include this at the end of the script as follows:
window = Window()
document = Document("a_document.txt")
save = SaveCommand(document)
exit = ExitCommand(window)
save_button = ToolbarButton('save', 'save.png')
save_button.command = save
save_keystroke = KeyboardShortcut("s", "ctrl")
save_keystroke.command = save
exit_menu = MenuItem("File", "Exit")
exit_menu.command = exit
First we create two receivers and the two commands. Then we create several of the
available invokers and set the correct command on each of them. To test, we can use
python3 -i filename.py and run code like exit_menu.click(), which will end the
program, or save_keystroke.keystroke(), which will save the fake file.
The above examples, however, do not feel terribly Pythonic, do they? They have a
lot of "boilerplate code" (code that does not accomplish anything, but only provides
structure to the pattern), and the Command classes are all very similar to each other.
Perhaps we could create a generic command object that takes a function as
a callback?
[ 269 ]
Wait, why bother with that, even; maybe we can just use a function or method object
for each command? Instead of an object with an execute() method, we can write a
function and use that as the command, directly. This is a common paradigm for the
command pattern in Python:
import sys
class Window:
def exit(self):
sys.exit(0)
class MenuItem:
def click(self):
self.command()
window = Window()
menu_item = MenuItem()
menu_item.command = window.exit
Now that looks a lot more like Python. At first glance, it looks like we've removed
the command pattern altogether, and we've tightly connected the menu_item and
Window classes. But look closer, there is no tight coupling at all. Any callable can
be set up as the command on the MenuItem, just as before. And the Window.exit
method can be attached to any invoker. Most of the flexibility of the command
pattern has been maintained. We have sacrificed complete decoupling for
readability, but this code is, in my opinion, and that of many Python
programmers, more maintainable than the fully abstracted version.
Of course, since we can add a __call__ method to any object, we aren't restricted to
only functions. The above example is a useful shortcut when the method being called
doesn't have to maintain a state, but in more advanced usage, we can use this code
as well:
class Document:
self.contents = "This file cannot be modified"
def save(self):
with open(self.filename, 'w') as file:
file.write(self.contents)
class KeyboardShortcut:
def keypress(self):
self.command()
[ 270 ]
Chapter 9
class SaveCommand:
def __call__(self):
self.document.save()
document = Document("a_file.txt")
shortcut = KeyboardShortcut()
save_command = SaveCommand(document)
shortcut.command = save_command
Here we have something that looks like the first command pattern, but a bit more
Pythonic. As you can see, changing the invoker to call a callable instead of a
command object with an execute method has not restricted us in any way; in fact, it's
given us more flexibility. We can link to functions directly when that works, yet we
can build a complete callable command object when the situation calls for it.
The command pattern is often extended to support undoable commands. For
example, a text program may wrap each insertion in a separate command with not
only an execute method, but also an undo method that will delete that insertion. A
graphics program may wrap each drawing action (rectangle, line, freehand pixels,
and so on) in a command that has an undo method that resets the pixels to their
original state. In such cases, the decoupling of the command pattern is much more
obviously useful, because each action has to maintain enough of its state to undo that
action at a later date.
Abstract factory pattern
The abstract factory pattern is normally used when we have multiple possible
implementations of a system that depend on some configuration or platform issue.
The calling code requests an object from the abstract factory, not knowing exactly
what class of object will be returned. The underlying implementation returned
may depend on a variety of factors, such as current locale, operating system,
or local configuration.
[ 271 ]
Common examples of the abstract factory pattern at work include code for operating
system independent toolkits, database backends, and country-specific formatters
or calculators. An operating system independent GUI toolkit might use an abstract
factory pattern that returns a set of Winform widgets under Windows, Cocoa
widgets under Mac, GTK widgets under Gnome, and QT widgets under KDE.
Django provides an abstract factory that returns a set of object relational classes
for interacting with a specific database backend (MySQL, PostgreSQL, SQLite, and
others) depending on a configuration setting for the current site. If the application
needs to be deployed in multiple places, each one can use a different database
backend by changing only one configuration variable. Different countries have
different systems for calculating taxes, subtotals, and totals on retail merchandise;
an abstract factory can return a particular tax calculation object.
The UML class diagram for an abstract factory pattern is hard to understand without
a specific example, so let's turn things around and create a concrete example first.
We'll create a set of formatters that depend on a specific locale, and help us format
dates and currencies. There will be an abstract factory class that picks the specific
factory, as well as a couple example concrete factories, one for France and one for the
USA. Each of these will create formatter objects for dates and times, which can be
queried to format a specific value. Here's the diagram:
Comparing that image to the simpler text above it shows that a picture is not always
worth a thousand words, after all, especially considering we haven't even allowed
for factory selection code here.
[ 272 ]
Chapter 9
Of course, in Python, we don't have to implement any interface classes, so we

can discard DateFormatter, CurrencyFormatter, and FormatterFactory. The
formatting classes themselves are pretty simple:
class FranceDateFormatter:
def format_date(self, y, m, d):
y, m, d = (str(x) for x in (y,m,d))
y = '20' + y if len(y) == 2 else y
m = '0' + m if len(m) == 1 else m
d = '0' + d if len(d) == 1 else d
return("{0}/{1}/{2}".format(d,m,y))
class USADateFormatter:
def format_date(self, y, m, d):
y, m, d = (str(x) for x in (y,m,d))
y = '20' + y if len(y) == 2 else y
m = '0' + m if len(m) == 1 else m
d = '0' + d if len(d) == 1 else d
return("{0}-{1}-{2}".format(m,d,y))
class FranceCurrencyFormatter:
def format_currency(self, base, cents):
base, cents = (str(x) for x in (base, cents))
if len(cents) == 0:
cents = '00'
elif len(cents) == 1:
cents = '0' + cents
digits = []
for i,c in enumerate(reversed(base)):
if i and not i % 3:
digits.append(' ')
digits.append(c)
base = ''.join(reversed(digits))
return "{0}{1}".format(base, cents)
class USACurrencyFormatter:
def format_currency(self, base, cents):
base, cents = (str(x) for x in (base, cents))
if len(cents) == 0:
cents = '00'
elif len(cents) == 1:
cents = '0' + cents
[ 273 ]

digits = []
for i,c in enumerate(reversed(base)):
if i and not i % 3:
digits.append(',')
digits.append(c)
base = ''.join(reversed(digits))
return "${0}.{1}".format(base, cents)
These classes use some basic string manipulation to try to turn a variety of possible
inputs (integers, strings of different lengths, and others) into the following formats:
Date
Currency
USA
Mm-dd-yyyy
$14,500.50
France
dd/mm/yyyy
14 50050
There could be more validation on the input in this code, but let's keep it simple and
dumb for this example!
Now that we have the formatters set up, we just need to create the formatter factories:
class USAFormatterFactory:
def create_date_formatter(self):
return USADateFormatter()
def create_currency_formatter(self):
return USACurrencyFormatter()
class FranceFormatterFactory:
def create_date_formatter(self):
return FranceDateFormatter()
def create_currency_formatter(self):
return FranceCurrencyFormatter()
Now we simply need to set up the code that picks the appropriate formatter.
Since this is the kind of thing that only needs to be set up once, we could make it a
singletonexcept singletons aren't very useful in Python. Let's just make the current
formatter a module-level variable instead:
country_code = "US"
factory_map = {
"US": USAFormatterFactory,
"FR": FranceFormatterFactory}
formatter_factory = factory_map.get(country_code)()
[ 274 ]
Chapter 9
In this example, we hardcode the current country code; in practice, it would likely
introspect the locale, operating system, or a configuration file to choose the code. It
uses a dictionary to associate the country codes with factory classes. Then we simply
get the correct class from the dictionary and instantiate it.
It is easy to see what needs to be done when we want to add support for more
countries; just create new formatter classes and the abstract factory itself.
Abstract factories often return a singleton object, but this is not required; in our
code, it's returning a new instance of each formatter every time it's called. There's no
reason the formatters couldn't be stored as instance variables and the same instance
returned for each factory.
Looking back at these examples, we see that, once again, there appears to be a lot
of boilerplate code for factories that just doesn't feel necessary in Python. Often, the
requirements that might call for an abstract factory can be more easily fulfilled by
using a separate module for each factory type (example: USA, France), and then
ensuring that the correct module is being accessed in a factory module. The package
structure for such modules often looks like this:
localize/
__init__.py
backends/

__init__.py

USA.py

France.py

The trick is that __init__.py in the localize package can contain code that
redirects all requests to the correct backend. There is a variety of ways this could
be done. Obviously, we could duplicate each of the methods in the backend
modules and route them through __init__.py as a proxy module. But we can do
it a bit more cleanly as well. If we know that the backend is never going to change
dynamically (that is without a restart), we can simply put some if statements in
__init__.py that check the current country code, and use the usually unacceptable
from .backends.USA import * syntax to import all variables from the appropriate
backend. Or, we could import each of the backends and set a current_backend
variable to point at a specific module:
from .backends import USA, France
if country_code == "US":
current_backend = USA
[ 275 ]
Depending on which solution we choose, our client code would simply have to call
either localize.format_date or localize.current_backend.format_date to get
a date formatted in the current country's locale. The end result is much more Pythonic
than the original abstract factory pattern, and, in typical usage, just as flexible.
Composite pattern
The composite pattern allows complex tree-like structures to be built from simple
components. Composite objects are simply container objects, where the content may
actually be another composite object.
Traditionally, each component in a composite object must be either a leaf node (that
cannot contain other objects) or a composite node. The key is that both composite
and leaf nodes can be treated identically. The UML diagram is very simple:
This simple pattern, however, allows us to create very complex arrangements of

elements, all of which satisfy the interface of the component object. As an example,
here is one such complicated arrangement:
[ 276 ]
Chapter 9
The composite pattern is commonly useful in file/folder-like trees. Regardless of

whether a node in the tree is a normal file or a folder, it is still subject to operations
such as moving, copying, or deleting the node. We can create a component interface
that supports these operations, and then use a composite object to represent folders,
and leaf nodes to represent normal files.
Of course, in Python, once again, we can take advantage of duck typing to implicitly
provide the interface, so we only need to write two classes. Let's define these
interfaces first:
class Folder:
def __init__(self, name):
self.name = name
self.children = {}
def add_child(self, child):
pass
def move(self, new_path):
pass
def copy(self, new_path):
pass
def delete(self):
pass
class File:
def __init__(self, name, contents):
self.name = name
self.contents = contents
pass
pass
def delete(self):
pass
For each folder (composite) object, we maintain a dictionary of children. Often, a

list is sufficient, but in this case, a dictionary will be useful for looking up children
by name. Our paths will be specified as node names separated by the / character,
similar to paths in a UNIX shell.
[ 277 ]
Thinking about the methods involved, we can see that moving or deleting a node
will behave in a similar way, regardless of whether or not it is a file or folder node.
Copying, however, will have to do a recursive copy for folder nodes, where copying
a file node is a trivial operation.
To take advantage of the similar operations, let's extract some of the common
methods into a parent class. Let's take that discarded Component interface and
change it to a base class:
class Component:
self.name = name
new_folder =get_path(new_path)
del self.parent.children[self.name]
new_folder.children[self.name] = self
self.parent = new_folder
def delete(self):
del self.parent.children[self.name]
class Folder(Component):
super().__init__(name)
self.children = {}
pass
pass
class File(Component):
def __init__(self, name, contents):
super().__init__(name)
self.contents = contents
pass
root = Folder('')
def get_path(path):
names = path.split('/')[1:]
node = root
for name in names:
node = node.children[name]
return node
[ 278 ]
Chapter 9
Here we've created the move and delete methods on the Component class. Both of
them access a mysterious parent variable that we haven't set yet. The move method
uses a module-level get_path function that finds a node from a predefined root
node, given a path. All files will be added to to this root node or a child of that node.
For the move method, the target should be a currently existing folder, or we'll get an
error. As with many of the examples in this book, error handling is woefully absent,
to help focus on the principles under consideration.
Let's set up that mysterious parent variable first; this happens, of course, in the
folder's add_child method:
child.parent = self
self.children[child.name] = child
Well, that was simple enough. Let's see if our composite file hierarchy is
working properly:
$ python3 -i 1261_09_18_add_child.py
>>> folder1 = Folder('folder1')
>>> root.add_child(folder1)
>>> root.add_child(folder2)
>>> folder1.add_child(folder11)
>>> file111 = File('file111', 'contents')
>>> folder11.add_child(file111)
>>> file21 = File('file21', 'other contents')
>>> folder2.add_child(file21)
>>> folder2.children
{'file21': <__main__.File object at 0xb7220a4c>}
>>> folder2.move('/folder1/folder11')
{'folder2': <__main__.Folder object at 0xb722080c>, 'file111': <__main__
.File object at 0xb72209ec>}
>>> file21.move('/folder1')
{'file21': <__main__.File object at 0xb7220a4c>, 'folder11': <__main__
.Folder object at 0xb722084c>}
>>>
Yes, we can create folders, add folders to other folders, add files to folders, and move
them around! What more could we ask for in a file hierarchy?
[ 279 ]
Well, we could ask for copying to be implemented, but to conserve space, let's leave
that as an exercise!
The composite pattern is extremely useful for such tree-like structures, including
GUI widget hierarchies, file hierarchies, tree sets, graphs, and HTML DOM. It
can be a useful pattern in Python when implemented according to the traditional
implementation, as the example earlier demonstrates. Sometimes, if only a shallow
tree is being created, we can get away with a list of lists or dictionary of dictionaries
and do not need to implement custom component, leaf, and composite classes as we
did earlier. Other times, we can get away with implementing only one composite
class, and treating leaf and composite objects as one class. Alternatively, Python's
duck typing can make it easy to add other objects to a composite hierarchy, as long
as they have the correct interface.
Exercises
Before diving into exercises for each design pattern, take a moment to implement
the copy method for the File and Folder objects in the previous section. The File
method should be quite trivial; just create a new node with the same name and
contents, and add it to the new parent folder. The copy method on Folder is quite a
bit more complicated, as you first have to duplicate the folder, and then recursively
copy each of it's children to the new location. You can call the copy() method on the
children indiscriminately, regardless of whether each is a file or a folder object. This
will drive home just how powerful the composite pattern can be.
Now, as with the previous chapter, look at the patterns we've discussed, and
consider ideal places where you might implement them. You may want to apply the
adapter pattern to existing code, as it is most often applicable when interfacing with
existing libraries, rather than new code. How can you use an adapter to force two
interfaces to interact with each other correctly?
Can you think of a system complex enough to justify using the facade pattern?
Consider how facades are used in real-life situations, such as the driver-facing
interface of a car, or the control panel in a factory. It is similar in software, except the
users of the facade interface are other programmers, rather than people trained to
use them. Are there complex systems in your latest project that could benefit from
the facade pattern?
It's possible you don't have any huge, memory-consuming code that would benefit
from the flyweight pattern, but can you think of situations where it might be
useful? Anywhere that large amounts of overlapping data need to be processed,
a flyweight is waiting to be used. Would it be useful in the banking industry?
In web applications? At what point does the flyweight pattern make sense?
When is it overkill?
[ 280 ]
Chapter 9
What about the command pattern? Can you think of any common (or better yet,
uncommon) examples of places where the decoupling of action from invocation
would be useful? Look at the programs you use on a daily basis, and imagine how
they are implemented internally. It's likely that many of them use the command
pattern for some purpose or another.
The abstract factory pattern, or the somewhat more Pythonic derivatives we
discussed, can be very useful for creating one-touch-configurable systems. Can
you think of places where such systems are useful?
Finally, consider the composite pattern. There are tree-like structures all around us
in programming; some of them, like our file hierarchy example, are blatant; others
are fairly subtle. What situations might arise where the composite pattern would
be useful? Can you think of places where you can use it in your own code? What
if you adapted the pattern slightly, for example to contain different types of leaf or
composite nodes for different types of objects?
Summary
In this chapter, we went into detail on several more design patterns, covering their
canonical descriptions as well as alternatives for implementing them in Python,
which is often more flexible and versatile than traditional object-oriented languages.
In particular, we covered:
The adapter pattern for matching interfaces
The facade pattern for simplifying complex systems
The flyweight pattern for reducing memory consumption
The command pattern for isolating invokers
The abstract factory pattern for separating implementation
The composition pattern for tree-like structures
In the next chapter, we'll cover some of the common tools for manipulating files,
configuration, and processes.
[ 281 ]
Files and Strings

Let's take a step back from the higher-level patterns now, and cover several Python
constructs we've used in all our examples, but without any details. We'll be looking
at files, IO, and serializing and loading data. Along the way, we'll find out how
complicated Python's excessively simple strings really are. In particular, we'll see:
The complexities of strings, bytes, and byte array
The ins and outs of string formatting
How to open files
Context managers
A few ways to serialize data
Strings
Strings are a basic primitive in Python; we've used them in nearly every example
we've discussed so far. All they do is represent an immutable sequence of characters.
Of course, "character" is a bit of an ambiguous word; can Python strings represent
sequences of accented characters? Chinese characters? What about Greek, Cyrillic,
or Farsi?
In Python 3, the answer is yes. Python strings are all represented in Unicode, a
special character definition that can represent virtually any character in any language
on the planet. This is done seamlessly, for the most part. For now, let's think of
Python 3 strings as an immutable sequence of Unicode characters.
So what can we do with this immutable sequence? We've touched on many of the
ways they can be manipulated in previous examples, but let's quickly cover it all in
one place: a crash course in string theory!
Files and Strings
String manipulation
As you know, strings can be created in Python by wrapping a sequence of characters

in single or double quotes. Multi-line strings can easily be created using three quote
characters, and multiple hard-coded strings can be concatenated together by placing
them side by side (this is useful for placing long strings in function calls without
exceeding a text width limit on your editor). Here are some examples:
a = "hello"
b = 'world'
c = '''a multiple
line string'''
d = """More
multiple"""
e = ("Three " "Strings "
"Together")
That's about all that can be said about creating strings. That last string is
automatically composed into a single string by the interpreter. It is also possible
to concatenate strings using the + operator (as in "hello " + "world").
Strings don't have to be hard-coded, of course. They can also come from various
outside sources such as text files, user input, or the network.
Like other sequences, strings can be iterated over (character by character), sliced, or
concatenated. The syntax is the same as for lists.
The str class has numerous methods on it to make manipulating strings easier. The
dir and help commands in the Python interpreter can tell us how to use all of them;
we'll consider some of the more common ones directly.
Several Boolean convenience methods help us identify whether or not the characters
in a string match a certain pattern. Here is a summary of these methods:
Method
isalpha
Purpose
isdigit
isdecimal
isnumeric
These methods tell us if all characters in the string are numeric digits,
Unicode decimal characters, or have the Unicode numeric value set,
respectively. Be careful with these ones; the period character is not a
decimal, so '45.2'.isdecimal() returns False! The real decimal
character is represented by Unicode value 0660, as in 45.2.
isalnum
Return True if the string contains only alphabet or number characters,

False if there is any punctuation or whitespace.
Return True if all characters in the string are alphabet characters in some
language. If any spaces, punctuation, or numbers appear in the string,
return False.
[ 284 ]
Chapter 10
Method
isspace
Purpose
isupper
islower
Return True if all alphabetic characters in the string are either upper or
lower case, depending on which method is called.
istitle
Return True if the string is in title case. Title case means only the first
character of every word is capitalized, all others are lower case. Be careful
with this one, as it does not strictly fit the grammatical definition of title
formatting. For example the poem "The Glove and the Lions" is a valid
title, even though not all words are capitalized. "The Cremation Of Sam
McGee" is also a valid title, even though there is an uppercase letter in the
middle of the last word.
Return True if the string contains only whitespace characters (space, tab,
newline, and so on). Otherwise, return False.
isidentifier Return True if the string contains a value that can be used as a variable
name in Python. If it has spaces, hyphens, or starts with a number, it will
return False.
isprintable Return True if all characters in the string can be printed on the screen
or a printer. Excludes terminal control characters such as the escape key.
Notably, whitespace characters are printable, even though they are not
visible when printed.
Other methods can be used to determine if a string matches a specific pattern. The
startswith and endswith methods return True if the string starts or ends with
the string of characters passed into the function. The count method tells us how
many times a given substring shows up in the string, while find, index, rfind, and
rindex tell us the position of a given substring within the original string. The two
'r' (for 'right') methods start searching from the end of the string. The find methods
return -1 if the substring can't be found, while index raises a ValueError in this
situation. Have a look at some of these methods in action:
>>> s = "hello world"
>>> s.startswith('hi')
False
>>> s.endswith('ld')
True
>>> s.find('l')
2
>>> s.index('m')
ValueError: substring not found
>>>
[ 285 ]
Files and Strings
Several of the remaining string methods return transformations of the string.

The upper, lower, capitalize, and title methods create new strings with
all alphabetic characters in the given format. The translate method can use a
dictionary to map arbitrary input characters to specified output characters.
For each of these methods, note that the input string remains unmodified; a brand
new str instance is returned instead. If we need to manipulate the resultant string,
we should assign it to a new variable, as in new_value = value.capitalize().
Often, once we've performed the transformation, we don't need the old value
anymore, so a frequent idiom is to assign it to the same variable, as in value =
value.title().
Finally, certain string methods return or operate on lists. The split method accepts
a substring and splits the string into a list of strings wherever that substring occurs.
The partition method splits the string at only the first occurrence of the substring,
and returns a tuple of three values: characters before the substring, the substring
itself, and the characters after the substring.
As the inverse of split, the join method accepts a list of strings, and returns all of
those strings combined together by placing the original string between them. The
replace method accepts two arguments, and returns a string where each instance
of the first argument has been replaced with the second. Here are some of these
methods in action:
>>> s = "hello world, how are you"
>>> s2 = s.split(' ')
>>> s2
['hello', 'world,', 'how', 'are', 'you']
>>> '#'.join(s2)
'hello#world,#how#are#you'
>>> s.replace(' ', '**')
'hello**world,**how**are**you'
>>> s.partition(' ')
('hello', ' ', 'world, how are you')
That's it: a whirlwind tour of the most common methods on the str class! Now
let's look at Python 3's method for composing strings and variables together into
new strings.
[ 286 ]
Chapter 10
String formatting
Python 3 has a versatile string formatting mechanism that allows us to easily

construct strings comprised of hard-coded text and interspersed variables. We've
used it in many previous examples, but it is much more versatile than the simple
formatting specifiers we've used.
Any string can be turned into a format string by calling the format() method on
it. This method returns a new string where specific characters in the input string
have been replaced with values provided in the arguments and keyword arguments
passed into the function. There is no fixed set of arguments required by the format
method; internally, it uses the *args and **kwargs syntax that we discussed
in Chapter 7.
The special characters that are replaced in formated strings are the opening
and closing brace characters: { and }. We can insert pairs of these in a string
and they will be replaced, in order, by any positional arguments passed to the
str.format method:
template = "Hello {}, you are currently {}."
print(template.format('Dusty', 'writing'))
If we run these statements, it replaces the braces with variables, in order:

Hello Dusty, you are currently writing.
This basic syntax is not terribly useful if we want to reuse variables within one string
or decide to use them in a different position. We can place zero-indexed integers
inside the curly braces to tell the formatter which positional variable gets inserted
at a given position in the string. Let's repeat the name:
template = "Hello {0}, you are {1}. Your name is {0}."
If we do use these integer indexes, we have to use them in all the variables. We can't
mix empty braces with positional indexes. For example, this code fails:
template = "Hello {}, you are {}. Your name is {0}."
Running this raises an appropriate exception:

File "1261_10_04_format_some_positions_broken.py", line 2, in
<module>
ValueError: cannot switch from automatic field numbering to manual
field specification
[ 287 ]
Files and Strings
Escaping braces
Brace characters are often useful in strings, aside from formatting. We need a way to
escape them in situations where we want them to be displayed as themselves, rather
than being replaced. This can be done by doubling the braces. For example, we can
use Python to format a basic Java program:
template = """
public class {0} {{
public static void main(String[] args) {{
System.out.println({1});
}}
}}"""
print(template.format("MyClass", "print('hello world')"));
Wherever we see the {{ or }} sequence in the template, that is, the braces enclosing
the Java class and method definition, we know the format method will replace them
with single braces, rather than some argument passed into the format method. Here's
the output:
public class MyClass {
public static void main(String[] args) {
System.out.println("print('hello world')");
}
}
The class name and contents of the output have been replaced with two parameters,
while the double braces have been replaced with single braces, giving us a valid
Java file. Turns out, this is about the simplest possible Python program to print the
simplest possible Java program that can print the simplest possible Python program!
Keyword arguments
If we're formatting complex strings, it can become tedious to remember the order of
the arguments or to update the template if we choose to insert another argument at
the beginning of the list. The format method therefore allows us to specify names
inside the braces instead of numbers. The named variables are then provided to the
format method as keyword arguments instead of positional arguments:
template = """
From: <{from_email}>
To: <{to_email}>
Subject: {subject}
{message}"""
print(template.format(
from_email = "[email protected]",
to_email = "[email protected]",
[ 288 ]
Chapter 10
message = "Here's some mail for you. "
" Hope you enjoy the message!",
subject = "You have mail!"
))
Notice how Python automatically concatenates the two strings we're

passing as a message argument? Very useful for line wrapping, but
be careful not to do this at the top level of a file. You need to be inside
a set of parenthesis (whether grouping a tuple, a function call, or just
explicitly grouping several strings together) for the concatenation to
be enabled.
We can also mix index and keyword arguments (as with all Python function calls, the
keyword arguments must follow the positional ones). We can even mix unlabeled
positional braces with keyword arguments:
print("{} {label} {}".format("x", "y", label="z"))
As expected, this code outputs:

x z y
Container lookups
We aren't restricted to passing simple string variables into the format method. Any
primitive, such as integers or floats can be printed. More interestingly, complex
objects including lists, tuples, dictionaries, and arbitrary objects can be used, and we
can access indexes and variables (but not methods) on those objects from within the
format string.
For example, if our e-mail message had grouped the from and to e-mail addresses
into a tuple, and placed the subject and message in a dictionary, for some reason
(perhaps because that's the input required for an existing send_mail function we
want to use), we can format it like so:
emails = ("[email protected]", "[email protected]")
message = {
'subject': "You Have Mail!",
'message': "Here's some mail for you!"
}
template = """
From: <{0[0]}>
To: <{0[1]}>
Subject: {message[subject]}
{message[message]}"""
print(template.format(emails, message=message))
[ 289 ]
Files and Strings
The variables inside the braces in the template string look a little weird, so let's look
at what they're doing. We have passed one argument as a position-based parameter
and one as a keyword argument. The two e-mail addresses are looked up by 0[x]
where x is either 0 or 1. The initial zero represents, as with other position-based
arguments, the first positional argument passed to format (the emails tuple, in this
case). The square brackets with a number inside are the same kind of index lookup
we see in regular Python code, so 0[0] maps to emails[0], in the emails tuple. The
indexing syntax works with any indexable object, so we see similar behavior when
we access message[subject], except this time we are looking up a string key in a
dictionary. Notice that unlike in Python code, we do not need to put quotes around
the string in the dictionary lookup.
We can even do multiple levels of lookup if we have nested data structures. I would
recommend against doing this too often, as template strings rapidly become difficult
to understand. If we have a dictionary that contains a tuple, we can do this:
emails = ("[email protected]", "[email protected]")
message = {
'emails': emails,
'subject': "You Have Mail!",
'message': "Here's some mail for you!"
}
template = """
From: <{0[emails][0]}>
To: <{0[emails][1]}>
Subject: {0[subject]}
{0[message]}"""
print(template.format(message))
In this example, we access only one variable, at position zero; we look up values
in the dictionary, and for e-mail addresses, we do a second index lookup based
on position. The formatting index system is very flexible, but we have to bear in
mind that the primary goal is to make our code, including templates, as readable
as possible. We need to look at the variables we are passing into a template, and
how we access them inside it. Would we be better off breaking them into separate
positional or keyword arguments to the format method, instead of passing a single
massive object? It depends on what structure the object has to begin with, as well as
the structure and amount of variable content in the format string.
[ 290 ]
Chapter 10
Object lookups
Indexing makes format lookup powerful, but we're not done yet! We can also pass
arbitrary objects as parameters, and use the dot notation to look up attributes on
those objects. Let's change our e-mail message data once again, this time to a class:
class EMail:
def __init__(self,
self.from_addr
self.to_addr =
self.subject =
self.message =
from_addr, to_addr, subject, message):

= from_addr
to_addr
subject
message
email = EMail("[email protected]", "[email protected]",

"You Have Mail!",
"Here's some mail for you!")
template = """
From: <{0.from_addr}>
To: <{0.to_addr}>
Subject: {0.subject}
{0.message}"""
print(template.format(email))
The template in this example is more readable than the previous examples, but
the overhead of creating an e-mail class adds complexity to the Python code. We
wouldn't normally create a class for the express purpose of including the object in a
template. Typically, we'd use this sort of lookup if the object we are trying to format
already exists. This is true of all the examples; if we have a tuple, list, or dictionary,
perhaps we'll pass it into the template directly. Otherwise we'd just create a simple
set of positional and keyword arguments.
Making it look right
It's nice to be able to include variables in template strings, but sometimes the
variables need a bit of coercion to make them look right in the output. For example, if
we are doing calculations with currency, we may end up with a long decimal that we
don't want to show up in our template:
subtotal = 12.32
tax = subtotal * 0.07
total = subtotal + tax
print("Sub: ${0} Tax: ${1} Total: ${total}".format(
subtotal, tax, total=total))
[ 291 ]
Files and Strings
If we run this formatting code, the output doesn't quite look like proper currency:
Sub: $12.32 Tax: $0.8624 Total: $13.1824
Technically, we should never use floating-point numbers in currency

calculations like this; we should construct decimal.Decimal() objects
instead. Floats are dangerous because their calculations are inherently
inaccurate beyond a specific level of precision. But we're looking at
strings, not floats, and currency is a great example for formatting!
To fix the format string above, we can include some additional information inside
the curly braces to adjust the formatting of the parameters. There are tons of things
we can customize, but the basic syntax inside the braces is the same; first we use
whichever of the earlier layouts (positional, keyword, index, attribute access) is
suitable to specify the variable that we want to place in the template string. We
follow this with a colon, and then the specific syntax for the formatting. Here's
an improved version:
subtotal = 12.32
tax = subtotal * 0.07
total = subtotal + tax
print("Sub: ${0:0.2f} Tax: ${1:0.2f} "
"Total: ${total:0.2f}".format(
subtotal, tax, total=total))
The 0.2f format specifier after the colons basically says, from left to right: for values
lower than one, make sure a zero is displayed on the left side of the decimal point;
show two places after the decimal; format the input value as a float.
We can also specify that each number should take up a particular number of
characters on the screen by placing a value before the period in the precision.
This can be useful for outputting tabular data, for example:
orders = [('burger', 2, 5),
('fries', 3.5, 1),
('cola', 1.75, 3)]
print("PRODUCT
QUANTITY
PRICE
SUBTOTAL")
for product, price, quantity in orders:
subtotal = price * quantity
print("{0:10s}{1: ^9d}
${2: <8.2f}${3: >7.2f}".format(
product, quantity, price, subtotal))
[ 292 ]
Chapter 10
Ok, that's a pretty scary looking format string, so let's see how it works before we
break it down into understandable parts:
PRODUCT
burger
fries
cola
QUANTITY
5
1
3
PRICE
$2.00
$3.50
$1.75
SUBTOTAL
$ 10.00
$
3.50
$
5.25
Nifty! So how is this actually happening? We have four variables we are formatting,
in each line in the for loop. The first variable is a string and is formatted with
{0:10s}. The s means it is a string variable, the 10 means it should take up ten
characters. By default, with strings, if the string is shorter than the specified number
of characters, it appends spaces to the right side of the string to make it long enough
(beware, however: if the original string is too long, it won't be truncated!). We can
change this behavior (to fill with other characters or change the alignment in the
format string), as we do for the next value, quantity.
The formatter for the quantity value is {1: ^9d}. The d represents an integer value.
The 9 tells us the value should take up nine characters. But with integers, instead
of spaces, the extra characters are zeros, by default. That looks kind of weird. So
we explicitly specify a space (immediately after the colon) as a padding character.
The carat character ^ tells us that the number should be aligned in the center of this
available padding; this makes the column look a bit more professional. The specifiers
have to be in the right order, although all are optional: Fill first, then align, then the
size, and finally, the type.
We do similar things with the specifiers for price and subtotal. For price we use
{2: <8.2f} and for subtotal, {3: >7.2f}. In both cases, we're specifying a space as
the fill character, but we use the < and > symbols, respectively, to represent that the
numbers should be aligned to the left or right within the minimum space of eight or
seven characters. Further, each float should be formatted to two decimal places.
The 'type' character for different types can affect formatting output as well. We've
seen the s, d, and f types, for strings, integers, and floats. Most of the other format
specifiers are alternative versions of these; for example, o represents octal format
and X represents hexadecimal for integers. The n type specifier can be useful for
formatting integer separators in the current locale's format. For floating-point
numbers, the % type will multiply by 100 and format a float as a percentage.
[ 293 ]
Files and Strings
We've barely scratched the surface of string formatting. These standard formatters
apply to most built-in objects, but it is actually possible for other objects to define
non-standard specifiers. For example, if we pass a datetime object into format,
we can format it using the specifiers used in the datetime.strftime function
as follows:
import datetime
print("{0:%Y-%m-%d %I:%M%p }".format(
datetime.datetime.now()))
It is even possible to write custom formatters for objects we create ourselves, but
that is beyond the scope of this book. Look into overriding the __format__ special
method if you need to do this in your code. The most comprehensive instructions can
be found in PEP 3101 at http://www.python.org/dev/peps/pep-3101/ although
the details are a bit dry. You can find more digestible tutorials using a web search.
Strings are Unicode
At the beginning of this section, we defined strings as collections of immutable

Unicode characters. This actually makes things very complicated at times, because
Unicode isn't really a storage format. If you get a string of bytes from a file or a
socket, for example, they won't be in Unicode. They will, in fact, be the built-in type
bytes. Bytes are immutable sequences of... well, bytes. Bytes are the lowest-level
storage format in computing. They represent 8 bits, and are generally described as
an integer between 0 and 255, or a hexadecimal equivalent between 0 and FF. Bytes
don't specifically represent anything; a sequence of bytes may store characters in an
encoded string, or pixels in an image.
If we print a byte object, any bytes that map to ASCII representations will be printed
as their original character, while non-ASCII bytes (whether they are binary data or
other characters) are printed as hex codes escaped by the \x escape sequence. You
may find it odd that a byte, represented as an integer, can map to an ASCII character.
But ASCII is really just a code where each letter is represented by a different byte
pattern, and therefore, a different integer. The character "a" is represented by the
same byte as the number 97, which is the hexadecimal number 0x61. Specifically,
all of these are an interpretation of the binary pattern 01100001.
Many IO operations only know how to deal with bytes, even if the bytes object
refers to textual data. It is therefore very important to know how to convert
between bytes and Unicode.
[ 294 ]
Chapter 10
The problem is that there are many ways to map bytes to Unicode text. Bytes
are machine-readable values, while text is a human-readable format. Sitting in
between is an encoding that maps a given sequence of bytes to a given sequence
of text characters.
However, there are multiple such encodings (ASCII is only one of them). The same
sequence of bytes can represent completely different text characters when mapped
using different encodings! It is important to decode bytes using the same character
set in which it was encoded, or we will get garbage data. It's not possible to get
text from bytes without knowing how the bytes should be encoded. If we receive
unknown bytes without a specified encoding, the best we can do is guess what
format they are encoded in, and we may be wrong.
Converting bytes to text
If we have an array of bytes from somewhere, we can convert it to Unicode using

the .decode method on the bytes class. This method accepts a string for the name
of the character encoding. There are many such names; common ones for European
languages include ASCII, UTF-8, and latin-1.
The sequence of bytes (in hex) 63 6c 69 63 68 e9 actually represent the characters
of the word clich in the latin-1 encoding. The following example will encode this
sequence of bytes and convert it to a Unicode string using the latin-1 encoding:
characters = b'\x63\x6c\x69\x63\x68\xe9'
print(characters)
print(characters.decode("latin-1"))
The first line creates a bytes object; the b immediately before the string tells us that
we are defining a bytes object instead of a normal Unicode string. Within the string,
each byte is specified usingin this casea hexadecimal number. The \x escapes
within the byte string each say, "the next two characters represent a byte using
hexadecimal digits".
Provided we are using a shell that understands the latin-1 encoding, the two print
statements will output the following strings:
b'clich\xe9'
clich
The first print statement shows that the bytes for ASCII characters are displayed as
those characters. The unknown (unknown to ASCII, that is) character stays in its
escaped hex format. The output includes a b at the beginning of the line to remind
us that it is a bytes representation, not a string.
[ 295 ]
Files and Strings
The next statement decodes the string using latin-1. The decode method returns a
normal (Unicode) string with the correct characters. However, if we had decoded
this same string using the Cyrillic "iso8859-5" encoding, we'd have ended up with
the sequence of characters 'clich'! This is because the \xe9 byte maps to different
characters in the two encodings.
Converting text to bytes
If we need to convert incoming bytes into Unicode, clearly we're also going to have
situations where we convert outgoing Unicode into byte sequences. This is done with
the encode method on the str class, which, in parallel to the decode method, also
requires a character set. The following code creates a Unicode string and encodes it
in a few different character sets:
characters = "clich"
print(characters.encode("UTF-8"))
print(characters.encode("latin-1"))
print(characters.encode("CP437"))
print(characters.encode("ascii"))
The first three encodings create a different set of bytes for the accented character. The
fourth one can't even handle it:
b'clich\xc3\xa9'
b'clich\xe9'
b'clich\x82'
File "1261_10_16_decode_unicode.py", line 5, in <module>
print(characters.encode("ascii"))
UnicodeEncodeError: 'ascii' codec can't encode character '\xe9' in
position 5: ordinal not in range(128)
Do you understand the importance of encoding now? The accented character is

represented in a different way for each encoding; if we use the wrong one when we
are decoding bytes to text, we get the wrong character.
The exception in the last case is not always the desired behavior; there may be
cases where we want the unknown characters to be handled in a different way. The
encode method takes an optional string argument named errors that can define
how such characters should be handled. This string can be one of:
strict
replace
ignore
xmlcharrefreplace
[ 296 ]
Chapter 10
The strict replacement strategy is the default we just saw. When a byte sequence
is encountered that does not have a valid representation in the requested encoding,
an exception is raised. When the replace strategy is used, the character is replaced
with a different character; in ASCII it is a question mark; other encodings may use
different symbols, such as an empty box. The ignore strategy simply discards any
bytes it doesn't understand, while the xmlcharrefreplace strategy creates an
xml entity representing the Unicode character. This can be useful when converting
unknown strings for use in an XML document. Here's how each of the strategies
affects our sample word:
Strategy
replace
"clich".encode("ascii", strategy)
ignore
b'clich'
xmlcharrefreplace
b'cliché'
b'clich?'
It is possible to call the str.encode and bytes.decode methods without passing

an encoding string. The encoding will be set to the default encoding for the current
platform. This will depend on the current operating system and locale or regional
settings; you can look it up using the sys.getdefaultencoding() function. It is
usually a good idea to specify the encoding explicitly, though, since the default
encoding for a platform may change, or the program may one day be extended to
work on text from a wider variety of sources.
If you are encoding text and don't know which encoding to use, it is probably best to
use the UTF-8 encoding. UTF-8 is able to represent any Unicode character. In modern
software, it is a de facto standard encoding to ensure documents in any languageor
even multiple languagescan be exchanged. The various other possible encodings
are useful for legacy documents or in regions that still use different character sets
by default.
The UTF-8 encoding uses one byte to represent ASCII and other common characters,
and up to four bytes for more complex characters. UTF-8 is special because it is
backwards-compatible with ASCII; any ASCII document encoded using UTF-8
will be identical to the original ASCII document.
Mutable byte strings
The bytes type, like str, is immutable. We can use index and slice notation on a
bytes object and search for a particular sequence of bytes, but we can't extend or
modify them. This can be very inconvenient when dealing with IO, as it is often
necessary to buffer incoming or outgoing bytes until they are ready to be sent. For
example, if we are receiving data from a socket, it may take several recv calls before
we have received an entire message.
[ 297 ]
Files and Strings
This is where the bytearray built-in comes in. This type behaves, more or less, like a
list, except it only holds bytes. The constructor for the class can accept a bytes object
to initialize it. The extend method can be used to append another bytes object to
the existing array (for example, when more data comes from a socket or other
IO channel).
Slice notation can be used on bytearray to modify the item inline. For example, this
code constructs a bytearray from a bytes object and then replaces two bytes using:
b = bytearray(b"abcdefgh")
b[4:6] = b"\x15\xa3"
print(b)
Be careful though, if we want to manipulate a single element in the bytearray, it

will expect us to pass an integer between 0 and 255 inclusive. This integer represents
a specific bytes pattern. If we try to pass a character or bytes object, it will raise
an exception.
A single byte character can be converted to an integer using the ord (short
for ordinal) function. This function returns the integer representation of a
single character:
b = bytearray(b'abcdef')
b[3] = ord(b'g')
b[4] = 68
print(b)
After constructing the array, we replace the character at index 3 (the fourth character,
as indexing starts at 0, as with lists) with byte 103. This integer was returned by the
ord function and is the ASCII character for the lowercase g. For illustration, we also
replaced the next character up with the byte number 68, which maps to the ASCII
character for the uppercase D.
The bytearray type has methods that allow it to behave like a list (we can append
integer bytes to it, for example), but also like a bytes object; we can use methods
like count and find the same way they would behave on a bytes or str object. The
difference is that bytearray is a mutable type, which can be useful for building up
complex sequences of bytes from a specific input source.
[ 298 ]
Chapter 10
File IO
So far through this book, when our examples touch files, we've operated entirely on
text files. Operating systems, however, actually represent files as a sequence of bytes,
not text.
Because reading bytes and converting the data to text is one of the more common
operations on files, Python wraps the incoming (or outgoing) stream of bytes with
appropriate decode (or encode) calls so we can deal directly with str objects. This
saves us a lot of boilerplate code to be constantly encoding and decoding text.
The open() function is used to open a file. For reading text from a file, we only need
to pass the filename into the function. The file will be opened for reading, and the
bytes will be converted to text using the platform default encoding. As with decode
and encode on bytes and str objects, the open function can accept encoding and
errors arguments to open a text file in a specific character encoding or to choose a
specific replacement strategy for invalid bytes in that encoding. These are normally
supplied to open as keyword arguments. For example, we can use the following code
to read the contents of a text file in ASCII format, converting any unknown bytes
using the replace strategy:
file = open('filename', encoding='ascii', errors='replace')
print(file.read())
file.close()
Of course, we don't always want to read files; often we want to write data to them!
The encoding and errors arguments can also be passed when writing text files. In
addition, to open a file for writing, we need to pass a mode argument as the second
positional argument, with a value of "w":
contents = "an oft-repeated clich"
file = open("filename", "w", encoding="ascii", errors="replace")
file.close()
We could also supply the value "a" as a mode argument, to append to the file, rather
than completely overwriting existing file contents.
These files with their wrappers for converting bytes to text are great, but it'd be
awfully inconvenient if the file we wanted to open was an image, executable, or
other binary file, wouldn't it?
[ 299 ]
Files and Strings
To open a binary file, we simply need to append a 'b' to the mode string. So 'wb'
would open a file for writing bytes, while 'rb' allows us to read them. They will
behave like text files, but without the automatic encoding of text to bytes. When we
read such a file, it will return bytes instead of str, and when we write to it, it will
fail if we try to pass a Unicode object.
Once a file is opened for reading, we can call the read, readline, or readlines
methods to get the contents of the file. The read method returns the entire contents
of the file as an str or bytes object, depending on whether there is a 'b' in the
mode. Be careful not to use this method without arguments on huge files. You don't
want to find out what happens if you try to load that much data into memory!
It is also possible to read a fixed number of bytes from a file; we simply pass an
integer argument to the read method describing how many bytes we want to read.
The next call to read will load the next sequence of bytes, and so on. We can do this
inside a while loop to read the entire file in manageable chunks.
The readline method returns a single line from the file; we can call it repeatedly
to get more lines. The plural readlines method returns a list of all the lines in the
file. Like the read method, it's not safe to use on very large files. These two methods
even work when the file is open in bytes mode, but it only makes sense if we are
parsing text-like data. An image or audio file, for example, will not have newlines in
it (unless the newline byte happened to represent a certain pixel or sound), so
applying readline wouldn't make sense.
For readability and to avoid reading a large file into memory at once, we can also use
a for loop directly on a file object to read each line, one at a time, and process it.
Writing to a file is just as easy; the write method on file objects simply writes a
string (or bytes, for binary data) object to the file; it can be called repeatedly to write
multiple strings, one after the other. The writelines method accepts an iterator
and writes each of the iterated values to the file. It specifically does not turn the
arguments into multiple lines by appending a newline after each one. If each item in
the iterator is expected to be a separate line, they should all have newline characters
at the end. The writelines method is basically a convenience to write the contents
of an iterator without having to explicitly iterate over it using a for loop.
A final important method on file objects is the close method. This method should
be called when we are finished reading or writing the file to ensure any buffered
writes are written to the file, that the file has been properly cleaned up, and that
all resources associated with the file are released back to the operating system.
Technically, this will happen automatically when the script exits, but it's better
to be explicit and clean up after ourselves.
[ 300 ]
Chapter 10
Placing it in context
This need to always close a file can make for some ugly code. Because an exception
may occur during file IO, we ought to wrap all calls to a file in a try...finally
clause, and close the file in finally, regardless of whether IO was successful. This
isn't very Pythonic; there must be a more elegant way to do it.
If we run dir on a file-like object, we see that it has two special methods named
__enter__ and __exit__. These methods turn the file object into what is known as
a context manager. Basically, if we use a special syntax called the with statement,
these methods will be called before and after nested code is executed. On file objects,
the __exit__ method ensures the file is closed, even if an exception is raised. We
no longer have to explicitly manage the closing of the file. Here is what the with
statement looks like in practice:
with open('filename') as file:
for line in file:
print(line, end='')
The open call returns a file object, which has __enter__ and __exit__ methods. The
returned object is assigned to the variable named file by the as clause. We know
the file will be closed when the code returns to the outer indentation level, and that
this will happen even if an exception is raised.
The with statement is used in several places in the standard library where startup or
cleanup code needs to be executed. For example, the urlopen call returns an object
that can be used in a with statement to clean up the socket when we're done. Locks
in the threading module can automatically release the lock when the statement has
been executed.
Most interestingly, because the with statement can apply to any object that has the
appropriate special methods, we can use it in our own frameworks. Keeping with
our string examples, let's create a simple context manager that allows us to construct
a sequence of characters and automatically convert it to a string upon exit:
class StringJoiner(list):
def __enter__(self):
return self
def __exit__(self, type, value, tb):
self.result = "".join(self)
[ 301 ]
Files and Strings
This code simply adds the two special methods required of a context manager to
the list class it inherits from. The __enter__ method performs any required setup
code (in this case, there isn't any) and then returns the object that will be assigned to
the variable after as in the with statement. Often, as we've done here, this is just the
context manager object itself.
The __exit__ method accepts three arguments. In a normal situation, these are all
given a value of None. However, if an exception occurs inside the with block, they
will be set to values related to the type, value, and traceback for the exception. This
allows the __exit__ method to do any cleanup code that may be required even if an
exception occurred. In our example, we simply create a result string by joining the
characters in the string, regardless of whether an exception was thrown.
While this is one of the simplest context managers we could write, and its usefulness
is dubious, it does work with a with statement. Have a look at it in action:
import random, string
with StringJoiner() as joiner:
for i in range(15):
joiner.append(random.choice(string.ascii_letters))
print(joiner.result)
This code simply constructs a string of fifteen random characters. It appends these
to a StringJoiner using the append method it inherited from list. When the with
statement goes out of scope (back to the outer indentation level), the __exit__
method is called, and the result attribute becomes available on the joiner object.
We print this value to see a random string.
Faking files
Sometimes we need code that provides a file-like interface but doesn't actually read
from or write to any real files. For example, we might want to retrieve a string from
a third-party library that only knows how to write to a file. This is an example of the
adapter pattern in action; we need an adapter that converts the file-like interface into
a string-like one.
Two such adapters already exist in the standard library, StringIO and BytesIO.
They behave in much the same way, except that one deals with text characters and
the second deals with bytes data. Both classes are available in the io package.
To emulate a file open for reading, we can supply a string or bytes object to the
constructor. Calls to read or readline will then parse that string as if it was a file.
To emulate a file opened for writing, we simply construct a StringIO or BytesIO
object and call the write or writelines methods. When writing is complete, we can
discover the final contents of the written "file" using the getvalue method. It's really
very simple:
[ 302 ]
Chapter 10
# coding=utf-8
from io import StringIO, BytesIO
source_file = StringIO("an oft-repeated clich")
dest_file = BytesIO()
char = source_file.read(1)
while char:
dest_file.write(char.encode("ascii", "replace"))
char = source_file.read(1)
print(dest_file.getvalue())
This piece of code is, technically, doing nothing more than encoding a str to a
bytes. But it is performing this task using a file-like interface. We first create a source
"file" that contains a string, and a destination "file" to write it to. Then we read one
character at a time from the source, encode it to ASCII using the "replace" error
replacement strategy, and write the resulting byte to the destination file. This code
doesn't know that the object it is calling write on is not a file, nor does it care.
The file interface is common for reading and writing data, even if it's not to a file or
a string. Network IO often uses the same protocol (set of methods) for reading and
writing data to the network, and compression libraries use it to store compressed
data, for example. This is duck typing at work; we can write code that operates on
a file-like object, and it will never need to know if the data actually came from a
compressed file, a string, or the internet.
Storing objects
Nowadays, we take the ability to write data to a file and retrieve it at an arbitrary
later date for granted. As convenient as this is (imagine the state of computing if
we couldn't store anything!), we may often find ourselves converting data we have
stored in a nice object or design pattern in memory into some kind of clunky text or
binary format for storage.
The Python pickle module, allows us to store objects directly in a special object
storage format. It essentially converts an object (and all the objects it holds as
attributes) into a format that can be stored in a file or file-like object or a string of
bytes that we can do whatever we want with.
For basic work, the pickle module has an extremely simple interface. It is comprised
of four basic functions for storing and loading data; two for manipulating file-like
objects, and two for manipulating bytes objects (the latter are just shortcuts to the
file-like interface so we don't have to create a BytesIO file-like object ourselves).
[ 303 ]
Files and Strings
The dump method accepts an object to be written and a file-like object to write the
serialized bytes to. This object must have a write method (or it wouldn't be file-like),
and that method must know how to handle a bytes argument (so a file opened for
text output wouldn't work).
The load method does exactly the opposite; it reads a serialized object from a file-like
object. This object must have the proper file-like read and readline arguments, each
of which must, of course, return bytes. The pickle module will load the object from
these bytes and the load method will return the fully reconstructed object. Here's an
example that stores and then loads some data in a list object:
import pickle
some_data = ["a list", "containing", 5,
"values including another list",
["inner", "list"]]
with open("pickled_list", 'wb') as file:
pickle.dump(some_data, file)
with open("pickled_list", 'rb') as file:
loaded_data = pickle.load(file)
print(loaded_data)
assert loaded_data == some_data
This code works as advertised: the objects are stored in the file and then loaded
from the same file. In each case, we open the file using a with statement so that it is
automatically closed. The file is first opened for writing and then a second time for
reading, depending on whether we are storing or loading data.
The assert statement at the end would raise an error if the newly loaded object is
not equal to the original object. Equality does not imply that they are the same object.
Indeed, if we print the id() of both objects, we would discover they are different.
However, because they are both lists whose contents are equal, the two lists are also
considered equal.
The dumps and loads functions behave much like their file-like counterparts, except
they return or accept bytes instead of file-like objects. The dumps function requires
only one argument; the object to be stored, and it returns a serialized bytes object.
The loads function requires a bytes object and returns the restored object. The 's'
in the method names is short for string; it's a legacy name from older versions of
Python where str objects were used instead of bytes.
[ 304 ]
Chapter 10
Both dump methods accept an optional protocol argument. If we are saving and
loading pickled objects that are only going to be used in Python 3 programs, we
don't need to supply this argument. Unfortunately, if we are storing objects that
may be loaded by older versions of Python, we have to use an older and less
efficient protocol. This should not normally be an issue. Usually, the only program
that would load a pickled object would be the same one that stored it. Pickle is an
unsafe format, so we don't want to be sending them unsecured over the internet to
unknown interpreters.
The argument supplied is an integer version number. The default version is number
3, representing the current highly efficient storage system used by Python 3 pickling.
The number 2 is the older version, which will store an object that can be loaded on
all interpreters back to Python 2.3. As 2.3 is the oldest version of Python that is still
widely used in the wild, version 2 pickling is normally sufficient. Versions 0 and 1
are supported on older interpreters; 0 is an ASCII format, while 1 is a binary format.
As a rule of thumb, then, if you know that the objects you are pickling will only be
loaded by a Python 3 program (for example, only your program will be loading
them), use the default pickling protocol. If they may be loaded by unknown
interpreters, pass a protocol value of 2, unless you really believe they may need
to be loaded by an archaic version of Python.
If we do pass a protocol to dump or dumps, we should use a keyword argument to
specify it: pickle.dumps(my_object, protocol=2). This is not strictly necessary, as
the method only accepts two arguments, but typing out the full keyword argument
reminds readers of our code what the purpose of the number is. Having a random
integer in the method call would be hard to read. Two what? Store two copies of the
object, maybe? Remember, code should always be readable. In Python, less code is
often more readable than longer code, but not always. Be explicit.
It is possible to call dump or load on a single open file more than once. Each call to
dump will store a single object (plus any objects it is composed of or contains), while
a call to load will load and return just one object. So for a single file, each separate
call to dump when storing the object should have an associated call to load when
restoring at a later date.
Customizing pickles
With many common Python objects, pickling "just works". Basic primitives such as
integers, floats, and strings can be pickled, as can any container object, such as lists
or dictionaries, provided the contents of those containers are also picklable. Further,
and importantly, any object can be pickled, so long as all of its attributes are
also picklable.
[ 305 ]
Files and Strings
So what makes an attribute unpicklable? Usually, it has something to do with

time-sensitive attributes that it would not make sense to load in the future. For
example, if we have an open network socket, open file, running thread, or database
connection stored as an attribute on an object, it would not make sense to pickle
these objects; a lot of operating system state would simply be gone when we came
to reload them later. We can't simply pretend a thread or socket connection exists
and make it appear! No, we need to somehow customize how such data is stored
and restored.
Here's a class that loads the contents of a web page every hour to ensure that they
stay up-to-date. It uses the threading.Timer class to schedule the next update:
from threading import Timer
import datetime
class UpdatedURL:
self.url = url
self.contents = ''
self.last_updated = None
self.update()
def update(self):
self.contents = urlopen(self.url).read()
self.last_updated = datetime.datetime.now()
self.schedule()
def schedule(self):
self.timer = Timer(3600, self.update)
self.timer.setDaemon(True)
self.timer.start()
The url, contents, and last_updated are all pickleable, but if we try to pickle an
instance of this class, things go a little nutty on the self.timer instance:
>>> u = UpdatedURL("http://news.yahoo.com/")
>>> import pickle
>>> serialized = pickle.dumps(u)
File "/usr/lib/python3.1/pickle.py", line 1358, in dumps
Pickler(f, protocol, fix_imports=fix_imports).dump(obj)
_pickle.PicklingError: Can't pickle <class 'method'>:
attribute lookup builtins.method failed
[ 306 ]
Chapter 10
That's not a very useful error, but it looks like we're trying to pickle something we
shouldn't be. That would be the Timer; we're storing a reference to self.timer in
the schedule method, and that attribute cannot be serialized.
When pickle tries to serialize an object, it simply tries to store the object's
__dict__ attribute; __dict__ is a dictionary mapping all the attribute names on
the object to their values. Luckily, before checking __dict__, pickle checks to see
if a __getstate__ method exists. If it does, it will store the return value of that
method instead of the __dict__.
Let's add a __getstate__ method to our UpdatedURL class that simply returns a
copy of the __dict__ without a timer:
def __getstate__(self):
new_state = self.__dict__.copy()
if 'timer' in new_state:
del new_state['timer']
return new_state
If we pickle the object now, it will no longer fail. And we can even successfully
restore that object using loads. However, the restored object doesn't have a timer
attribute, so it will not be refreshing the content like it is designed to do. We need
to somehow create a new timer (to replace the missing one) when the object
is unpickled.
As we might expect, there is a complementary __setstate__ method that can be
implemented to customize unpickling. This method accepts a single argument,
which is simply the object returned by __getstate__. If we implement both
methods, __getstate__ is not required to return a dictionary, since __setstate__
will know what to do with whatever object __getstate__ chooses to return. In our
case, we simply want to restore the __dict__, and then create a new timer:
def __setstate__(self, data):
self.__dict__ = data
self.schedule()
The pickle module is very flexible and provides other tools to further customize the
pickling process if you need them. However, these are beyond the scope of this book.
The tools we've covered are sufficient for basic pickling tasks. Objects to be pickled
are normally relatively simple data objects; we would not typically pickle an entire
running program or complicated design pattern, for example.
[ 307 ]
Files and Strings
Serializing web objects
It is not a good idea to load a pickled object from an unknown or untrusted source. It
is possible to inject arbitrary code into a pickled file. This can be used to maliciously
attack a computer via the pickle. Another disadvantage of pickles is that they can
only be loaded by other Python programs, and cannot be easily shared with
other systems.
JavaScript Object Notation (JSON) is a special format for exchanging primitive
data. JSON is a standard format that can be interpreted by a wide array of
heterogeneous client systems. Hence, JSON can be very useful for transmitting data
between completely decoupled systems. Further, JSON does not have any support
for executable code, only data can be serialized; thus it is much more difficult to
inject malicious statements into it.
Because JSON can be easily interpreted by JavaScript engines, it is most often used
for transmitting data from a web server to a JavaScript-capable web browser. If the
web application serving the data is written in Python, it needs a way to convert
internal data into the JSON format.
There is a module to do this, named, as we might expect, json. This module provides
a similar interface to the pickle module, with dump, load, dumps, and loads
functions. The default calls to these functions are nearly identical to those in pickle,
so we won't repeat the details. There are a couple differences; obviously the output
of these calls is valid JSON notation, rather than a pickled object. In addition, the
json functions operate on str objects, rather than bytes. Therefore, when dumping
to or loading from a file, we need to create text files rather than binary ones.
The JSON serializer is not as robust as the pickle module; it can only serialize basic
types such as integers, floats, and strings, and simple containers such as dictionaries
and lists. Each of these has a direct mapping to a JSON representation, but JSON
is unable to represent classes, methods, or functions. It is not possible to transmit
complete objects in this format. Because the receiver of an object we have dumped
to JSON format is normally not a Python object, it would not be able to understand
classes or methods in the same way that Python does, anyway. JSON is a data
notation; objects, as you will recall, are composed of both data and behavior.
If we do have objects for which we want to serialize only the data, we can always
serialize the object's __dict__ attribute. Or we can semi-automate this task by
supplying custom code to create or parse a JSON serializable dictionary from
certain types of objects.
[ 308 ]
Chapter 10
In the json module, both the object storing and loading functions accept optional
arguments to customize the behavior. The dump and dumps methods accept a cls
keyword argument. If passed, this should be a subclass of the JSONEncoder class,
with the default method overridden. This method accepts an object and converts it
to a dictionary that json can digest. If it doesn't know how to process the object, it's
generally good to call the super() method, so that it can take care of serializing
basic types.
The load and loads methods also accept such a cls argument that can be a
subclass of the inverse class, JSONDecoder. However, it is normally sufficient to
pass a function into these methods using the object_hook keyword argument. This
function accepts a dictionary and returns an object; if it doesn't know what to do
with the input dictionary, it can simply return it unmodified.
But that's enough theory, let's look at an example! Imagine we have the following
simple contact class that we want to serialize:
class Contact:
def __init__(self, first, last):
self.first = first
self.last = last
@property
def full_name(self):
return("{} {}".format(self.first, self.last))
We could just serialize the __dict__:

>>> c = Contact("John", "Smith")
>>> json.dumps(c.__dict__)
'{"last": "Smith", "first": "John"}'
But accessing special (double-underscore) attributes in this fashion is kind of crude.

Also, what if the receiving code (perhaps some JavaScript on a web page) wanted
that full_name property to be supplied? Of course, we could construct the
dictionary by hand, but if we need to do a lot of that, it can be useful to create
a custom encoder instead:
import json
class ContactEncoder(json.JSONEncoder):
def default(self, obj):
if isinstance(obj, Contact):
return {'is_contact': True,
'first': obj.first,
'last': obj.last,
'full': obj.full_name}
return super().default(obj)
[ 309 ]
Files and Strings
The default method basically checks to see what kind of object we're trying to
serialize; if it's a contact, we convert it to a dictionary manually, otherwise we let the
parent class handle serialization (by assuming that it is a basic type that json knows
how to handle). Notice that we pass an extra attribute to identify this object as a
contact, since there would be no way to tell upon loading it. This is just a convention;
for a more generic serialization mechanism it might make more sense to store a
string type in the dictionary, or possibly even the full class name, including package
and module. Remember that the format of the dictionary depends on the code at
the receiving end; there has to be an agreement as to how the data is going to
be specified.
We can use this class to encode a contact by passing the class (not an instantiated
object) to the dump or dumps function:
>>> c = Contact("John", "Smith")
>>> json.dumps(c, cls=ContactEncoder)
'{"is_contact": true, "last": "Smith", "full": "John Smith",
"first": "John"}'
For decoding, we can write a function that accepts a dictionary and checks the
existence of the is_contact variable to decide whether to convert it to a contact:
def decode_contact(dic):
if dic.get('is_contact'):
return Contact(dic['first'], dic['last'])
else:
return dic
We can pass this function to the load or loads function using the object_hook
keyword argument:
>>> data = '{"is_contact": true, "last": "smith",
"full": "john smith", "first": "john"}'
>>> c = json.loads(data, object_hook=decode_contact)
>>> c
<__main__.Contact object at 0xa02918c>
>>> c.full_name
'john smith'
Exercises
We've covered a wide variety of topics in this chapter, from strings to files, to object
serialization, and back again. Now it's time to consider how these ideas can be
applied to your own code.
[ 310 ]
Chapter 10
Python strings are very flexible, and Python is an extremely powerful tool for
string-based manipulations. If you don't do a lot of string processing in your daily
work, try designing a tool that is exclusively intended for manipulating strings. Try
to come up with something innovative, but if you're stuck, consider writing a web
log analyzer (how many requests per hour? How many people visit more than five
pages?) or a template tool that replaces certain variable names with the contents of
other files.
Spend a lot of time toying with the string formatting operators. Simply write a bunch
of template strings and objects to pass into the format function, and see what kind of
output you get. Try the exotic formatting operators, like percentage or hexadecimal
notation. Try out the fill and alignment operators, and see how they behave
differently for integers, strings, and floats. Consider writing a class of your own
that has a __format__ method; we didn't discuss this in detail, but explore just how
much you can customize formatting.
Make sure you understand the difference between bytes and str objects. The
distinction is very complicated in older versions of Python (there was no bytes, and
str acted like both bytes and str unless we needed non-ASCII characters in which
case there was a separate unicode object, which was similar to Python 3's str class.
It's even more confusing than it sounds!). It's clearer nowadays; bytes is for binary
data, and str is for character data. The only tricky part is knowing how and when
to convert between the two. For practice, try writing text data to a file opened for
writing bytes (you'll have to encode the text yourself), and then reading from the
same file.
Do some experimenting with bytearray; see how it can act both like a bytes object
and a list or container object at the same time. Try writing a buffer that holds data
in the bytes array until it is a certain length before returning it. You can simulate the
code that puts data into the buffer by using time.sleep calls to ensure data doesn't
arrive too quickly.
If you don't use the with statement in your daily coding, go through all your code
and find all the places you have opened files and forgotten to close them. Pay special
attention to places where you did close the files, but only under the assumption that no
exception was raised. For every one you find, replace the code with a with statement.
Have you ever written code where a custom context manager might be useful? Think
about cases where you have to do arbitrary setup or cleanup around an unknown or
partially known task. That task will be implemented inside the with statement, while
the setup or cleanup code is taken care of in __enter__ and __exit__ methods
on the object. This is a very useful construct, although it's not an overly common
pattern; if you can't think of any places it would suit an old project, try writing one
for a new situation of your devising, just as an exercise.
[ 311 ]
Files and Strings
If you've ever written an adapter to load small amounts of data from a file or
database and convert it to an object, consider using a pickle instead. Pickles are not
efficient for storing massive amounts of data, but they can be useful for loading
configuration or other simple objects. Try coding it multiple ways; using a pickle, a
text file, or a small database. Which is easiest to work with?
Try experimenting with pickling data, then modifying the class that holds the data,
and loading the pickle into the new class. What works? What doesn't? Is there a way
to make drastic changes to a class, such as renaming an attribute or splitting it into
two new attributes and still get the data out of an older pickle? (Hint: try placing a
private pickle version number on each object and update it each time you change the
class; you can then put a migration path in __setstate__.)
If you do any web development at all, especially Web 2.0 applications, do some
experimenting with the JSON serializer. Personally, I prefer to serialize only
standard JSON serializable objects, rather than writing custom encoders or object_
hooks, but the desired effect really depends on the interaction between the frontend
(JavaScript, typically) and backend code. The JSON format isn't very common
outside of JavaScript and web notation, so if you don't do web development you may
want to skip this exercise. Note, however, that JSON is a subset of YAML, so if you
have occasion to generate valid YAML, the JSON serializer can still be useful!
Summary
We've covered string manipulation, file IO, and object serialization. We discussed
how to combine hard-coded strings and program variables into outputtable strings
using the powerful string formatting system, and learned the difference between
binary and textual data. All told, we've seen:
How to use the various str methods
String formatting
bytes versus str
Mutable bytearrays
Files in binary and textual formatters
Context managers and the with statement
Serializing data with pickle and json
In the next chapter, we'll cover one of the most important topics in Python
programming: how to test our code to ensure it's doing what we think it is doing.
[ 312 ]
Testing Object-oriented
Programs
Most skilled Python programmers agree that testing is one of the most important
aspects of software development in Python. Even though this chapter is placed near
the end of the book, it is not an afterthought; everything we have studied so far will
help us when writing tests. We'll be studying:
The importance of unit testing and test-driven development
The standard unittest module
The py.test automated testing suite
Code coverage
Why test?
More and more programmers are learning how important testing their code is. If
you're among them, feel free to skim this section. You'll find the next section, where
we actually learn how to do the tests in Python, more scintillating. If you're not
convinced of the importance of testing, I promise that your code is broken, you just
don't know it. Read on!
Some people argue that testing is more important in Python code because of its
dynamic nature; compiled languages such as Java and C++ are occasionally thought
to be somehow 'safer' because they enforce type checking at compile time. The thing
is, Python tests rarely, if ever, check types. They're checking values. They're making
sure that the right attributes have been set at the right time or that the sequence has
the right length, order, and values. They aren't checking to make absolutely sure that
a list was returned if a tuple or custom container would suffice. These higher-level
things need to be tested in any language. The real reason Python programmers test
more than programmers of other languages is that it is so easy to test in Python!
Testing Object-oriented Programs
But why test? Do we really need to test? What if we didn't test? To answer those
questions, write a program from scratch without any testing at all. Don't run it until
it is completely written, start to finish. Don't pick a large program, just a few classes
interacting. If you can't come up with anything, try a two player tic-tac-toe game; it's
fairly simple if you make both players human players (no artificial intelligence). You
don't even have to try to calculate who the winner is.
Now run your program. And fix all the errors. How many were there? I recorded
eight on my tic-tac-toe implementation, and I'm not sure I caught them all. Did you?
We need to test our code to make sure it works. Running the program, as we just did,
and fixing the errors is one crude form of testing. Most programmers tend to write
a few lines of code and run the program to make sure those lines are doing what
they expect. But changing a few lines of code can affect parts of the program that
the developer hadn't considered as being influenced by the changes. Because they
weren't thinking of this influence, they won't test the other functionality. As we
can see from the the program we just wrote, if they haven't tested it, it's almost
certainly broken.
To handle this, we write automated tests. Written tests are simply programs that
automatically run certain inputs through other programs (or, more often, parts of
other programs, such as one function or class). We can run these test programs in
seconds and cover more possible user inputs than one programmer would think to
test every time they change something.
There are four main reasons to write tests:
To ensure that code is working the way the developer thinks it should
To ensure that code continues working when we make changes
To ensure the developer understood the requirements
To ensure that the code we are writing has a maintainable interface
The first point really doesn't justify the time it takes to write a test; we can simply
test the code directly (perhaps through the interactive interpreter). But as soon as
we have to perform the same sequence of test actions multiple times, it takes less
time to automate those steps once and then run them whenever necessary. It is
a good idea to run tests whenever we have to change code, whether it is during
initial development or maintenance releases. When we have a comprehensive set
of automated tests, we can run them after code changes and know that we didn't
inadvertently break anything that was tested.
[ 314 ]
Chapter 11
However, the most interesting points in the list are the last two. When we write tests
for code, it can actually help us design the API, interface, or pattern that code takes.
Thus, if we, as the programmers, misunderstood the requirements from management
or the client, writing a test can help highlight that misunderstanding. On the other
side, if we're not certain how we want to design a class, we can write a test that
interacts with that class so we have an idea what the most natural way to test it
would be. In fact, it is often beneficial to write the tests before we write the code
we are testing!
Writing tests first is the mantra of test-driven development. Test-driven development

takes the "untested code is broken code" concept one step further and suggests
that only unwritten code should be untested. Do not write any code until you have
written the tests for this code. So the first step is to write a test that proves the code
would work. Obviously, the test is going to fail, since the code hasn't been written.
Then write the code that ensures the test passes. Then write another test for the next
segment of code.
Test-driven development is fun. It allows us to build little puzzles to solve; these are
the tests. Then we implement the code to solve the puzzles. Then we make a more
complicated puzzle, and we write code that solves the new puzzle without unsolving
the previous one.
There are two goals to the test-driven methodology. The first is to ensure that tests
really get written. It's so very easy, after we have written code, to say: "Hmm, it
seems to work, I don't have to write any tests for this. It was just a small change,
nothing could have broken." If the test is already written before we write the code,
we will know exactly when it works (because the test will pass), and we'll know in
the future if it is ever broken by a change we, or someone else has made.
Secondly, writing tests first forces us to consider exactly how the code is going
to be interacted with. It tells us what methods we need objects to have and how
attributes will be accessed. It helps us break up the initial problem into smaller,
testable problems, and then to recombine the tested solutions into larger, also tested
solutions. Writing tests can thus become a part of the design process. Often, if we're
writing a test for a not-fully-specified object, we will discover anomalies in the
design that force us to consider new aspects of the software.
[ 315 ]
As a concrete example, we may be writing code that uses an object-relational

mapper to store object properties in a database. It is common to use an automatically
assigned database id in such objects, and our code might be using this id for various
purposes, say as a key in a dictionary. If we are writing a test for such code, before
we write it, we may realize that our design is faulty because objects do not have
these IDs until they have been saved to the database. If we want to manipulate an
object without saving it in our test, it will highlight this problem before we have
written any code.
Testing makes software better. Writing tests before we release the software makes
it better before the end user sees or purchases the buggy version (I have worked
for companies that thrive on the "the users can test it" philosophy. It's not a healthy
business model!). Writing tests before we write the software makes it better the first
time it is written.
Unit testing
Let's start our exploration with Python's built-in test library. This library provides a
common interface for unit tests. Unit tests are a special kind of automated test that
focuses on testing the least amount of code possible in any one test. Each test tests a
single unit of the total amount of available code.
The Python library for this is called, unsurprisingly: unittest. It provides several
tools for creating and running unit tests, the most important being the TestCase
class. This class provides a set of methods that allow us to compare values, set up
tests, and clean up after running them.
When we want to write a set of unit tests for a specific task, we create a subclass
of TestCase, and write methods that accept no arguments to do the actual testing.
These methods must all start with the string test. If this convention is followed,
they'll automatically be run as part of the test process. Normally, the tests set some
values on an object and then run a method, and use the built-in comparison methods
to ensure that the right results were calculated. Here's a very simple example:
import unittest
class CheckNumbers(unittest.TestCase):
def test_int_float(self):
self.assertEquals(1, 1.0)
if __name__ == "__main__":
unittest.main()
[ 316 ]
Chapter 11
This code simply subclasses the TestCase class and adds a method that calls the
TestCase.assertEquals method. This method will either succeed or raise an
exception, depending on whether the two parameters are equal. If we run this
code, the main function from unittest will give us the following output:
.
-------------------------------------------------------------Ran 1 test in 0.000s
OK
Did you know that floats and integers can compare as equal? Let's add a new test
that fails:
def test_str_float(self):
self.assertEquals(1, "1")
If we run this code, the output is a bit more sinister, since floats and strings can not
be considered equal:
.F
============================================================
FAIL: test_str_float (__main__.CheckNumbers)
-------------------------------------------------------------Traceback (most recent call last):
File "simplest_unittest.py", line 8, in test_str_float
self.assertEquals(1, "1")
AssertionError: 1 != '1'
-------------------------------------------------------------Ran 2 tests in 0.001s
FAILED (failures=1)
The dot on the first line indicates that the first test (the one we wrote before) passed
successfully; the F after it shows that the second test failed. Then at the end, it gives
us some informative output telling us how and where the test failed, along with a
summary of the number of failures.
[ 317 ]
We can have as many test methods on one TestCase class as we like; as long as
the method name begins with test the test runner will execute each one as a
separate test. Each test should be completely independent of other tests. Results or
calculations from a previous test should have no impact on the current test. The key
to writing good unit tests is to keep each test method as short as possible, testing a
small unit of code with each test case. If your code does not seem to naturally break
up into such testable units, it's probably a sign that your design needs rethinking.
Writing tests allows us not only to ensure our code works, but also helps test our
design as well.
Assertion methods
The general layout of a test case is to set certain variables to known values, run one
or more functions, methods, or processes, and then "prove" that the correct results
were returned or calculated by using TestCase assertion methods.
There are a few different assertion methods available to confirm that specific results
have been achieved. We just saw assertEqual, which will cause a test failure if the
two parameters do not pass an equality check. The inverse, assertNotEqual, will
fail if the two parameters do compare as equal. The assertTrue and assertFalse
methods each accept a single expression, and fail if the expression does not pass
an if test. These tests are not checking for the Boolean values True or False; they
return the same value that would be returned if an if statement were used: False,
None, 0, or an empty list, dictionary, or tuple would pass an assertFalse, while
nonzero numbers, containers with values in them, or the value True would pass an
assertTrue.
The assertRaises method accepts an exception class, a callable object (function,
method, or an object with a __call__ method), and arbitrary arguments and
keyword arguments to be passed into the callable. The assertion method will invoke
the function with the supplied arguments, and will fail or raise an error if the method
does not raise the expected exception class.
In addition, each of these methods accepts an optional argument named msg; if
supplied, it will be included in the error message if the assertion fails. This is useful
for clarifying what was expected or explaining where a bug may have occurred to
cause the assertion to fail.
[ 318 ]
Chapter 11
Additional assertion methods in Python 3.1
The unittest library has been extensively updated in Python 3.1. It now contains
several new assertion methods, and allows the assertRaises method to take
advantage of the with statement. The following code only works on Python 3.1
and later. It illustrates the two ways that assertRaises can be called:
import unittest
def average(seq):
return sum(seq) / len(seq)
class TestAverage(unittest.TestCase):
def test_python30_zero(self):
self.assertRaises(ZeroDivisionError,
average,
[])
def test_python31_zero(self):
with self.assertRaises(ZeroDivisionError):
average([])
if __name__ == "__main__":
unittest.main()
The context manager allows us to write the code the way we would normally write
it (by calling functions or executing code directly) rather than having to wrap the
function call in another function call.
In addition, Python 3.1 provides access to several useful new assertion methods:
assertGreater, assertGreaterEqual, assertLess, and assertLessEqual
assertIn ensures that the first of two arguments is an element in the

container object (list, tuple, dictionary, and so on) that is passed as a second
argument. The assertNotIn method does the inverse.
assertIsNone tests that a value is None. Unlike assertFalse, it will not pass
for values of zero, False, or empty container objects; the value must be None.
The assertIsNotNone method does, of course, the opposite.
assertSameElements accepts two container objects as arguments, and
all accept two comparable objects, and ensure that the named inequality
holds.
ensures that they contain the same set of elements, regardless of order.
[ 319 ]
assertSequenceEqual does enforce order. Further, if there's a failure, it will
assertDictEqual, assertSetEqual, assertListEqual, and

assertTupleEqual all do the same thing as assertSequenceEqual, except
show a diff comparing the two lists to see exactly how it failed.
they also ensure that the container objects are the correct type.
assertMultilineEqual accepts two multiline strings and ensures they are

identical. If they are not, a diff is displayed in the error message.
assertRegexpMatches accepts text and a regular expression and confirms
that the text matches the regular expression.
Reducing boilerplate and cleaning up
After writing a few small tests, we often find that we have to do the same setup code
for several related tests. For example, the following simple list subclass has three
methods for simple statistical calculations:
class StatsList(list):
def mean(self):
return sum(self) / len(self)
def median(self):
if len(self) % 2:
return self[int(len(self) / 2)]
else:
idx = int(len(self) / 2)
return (self[idx] + self[idx-1]) / 2
def mode(self):
freqs = defaultdict(int)
for item in self:
freqs[item] += 1
mode_freq = max(freqs.values())
modes = []
for item, value in freqs.items():
if value == mode_freq:
modes.append(item)
return modes
[ 320 ]
Chapter 11
Clearly, we're going to want to test situations with each of these three methods that
have very similar inputs; we'll want to see what happens with empty lists or with
lists containing non-numeric values or with lists containing a normal dataset. We can
use the setUp method on the TestCase class to do initialization for each test. This
method accepts no arguments, and allows us to do arbitrary setup before each test
is run. For example, we can test all three methods on identical lists of integers
as follows:
from stats import StatsList
import unittest
class TestValidInputs(unittest.TestCase):
def setUp(self):
self.stats = StatsList([1,2,2,3,3,4])
def test_mean(self):
self.assertEqual(self.stats.mean(), 2.5)
def test_median(self):
self.assertEqual(self.stats.median(), 2.5)
self.stats.append(4)
self.assertEqual(self.stats.median(), 3)
def test_mode(self):
self.assertEqual(self.stats.mode(), [2,3])
self.stats.remove(2)
self.assertEqual(self.stats.mode(), [3])
if __name__ == "__main__":
unittest.main()
If we run this example, it indicates that all tests pass. Notice first that the setUp
method is never explicitly called inside the three test_* methods. The test suite
does this on our behalf. More importantly notice how test_median alters the list,
by adding an additional 4 to it, yet when test_mode is called, the list has returned
to the values specified in setUp (if it had not, there would be two fours in the list,
and the mode method would have returned three values). This shows that setUp
is called individually before each test, to ensure the test class has a clean slate for
testing. Tests can be executed in any order, and the results of one test do not
depend on results from another.
[ 321 ]
In addition to the setUp method, TestCase offers a no-argument tearDown method,

which can be used for cleaning up after each and every test on the class has run. This
is useful if cleanup entails more than just letting an object be garbage collected. For
example, if we are testing code that does file IO, our tests may create new files as a
side effect of testing; the tearDown method can be used to remove these files and
ensure the system is in the same state it was before the tests ran. Test cases should
never have side effects.
In general, we group test methods into separate TestCase subclasses depending on
what setup code they have in common. Several tests that require the same or similar
setup will be placed in one class, while tests that require unrelated setup go in their
own class.
Organizing and running tests
It doesn't take long for a collection of unit tests to grow very large and unwieldy. It
quickly becomes complicated to load and run all the tests at once. This is a primary
goal of unit testing; it should be trivial to run all tests on our program and get a
quick, "yes or no", answer to the question, "Did my recent changes break any
existing tests?"
It is possible to collect groups of TestCase objects or modules containing tests into
collections called TestSuites, and to load specific tests at specific times. In older
versions of Python, this resulted in a lot of boilerplate code just to load and execute
all the tests on a project. If this much control is needed, the functionality is still
available, but most programmers can use test discovery, which will automatically
find and run tests in the current package or subpackages.
Test discovery is built into Python 3.2 (and the simultaneously developed Python
2.7) and later, but can also be used in Python 3.1 or older versions of Python by
installing the discover module available from http://pypi.python.org/pypi/
discover/.
The discover module basically looks for any modules in the current folder or
subfolders with names that start with the characters test. If it finds any TestCase
or TestSuite objects in these modules, the tests are executed. It's a painless way to
ensure you don't miss running any tests. To use it, simply ensure your test modules
are named test_<something>.py and then run one of the following two commands,
depending on which version of Python you have installed:
Python 3.1 or earlier: python3 -m discover
Python 3.2 or later: python3 -m unittest discover
[ 322 ]
Chapter 11
Ignoring broken tests
Sometimes a test is known to fail, but we don't want the test suite to report a failure
under those conditions. This may be because a broken or unfinished feature has
had tests written, but we aren't currently focusing on improving it. More often, it
happens because a feature is only available on a certain platform, Python version, or
for advanced versions of a specific library. Python provides us with a few decorators
to mark tests as expected to fail or to be skipped under known conditions.
These decorators are:
expectedFailure()
skip(reason)
skipIf(condition, reason)
skipUnless(condition, reason)
These are applied using the Python decorator syntax. The first one accepts no
arguments, and simply tells the test runner not to record the test as a failure even if
it does, in fact, fail. The skip method goes one step further and doesn't even bother
to run the test. It expects a single string argument describing why the test was
skipped. The other two decorators accept two arguments, one a Boolean expression
that indicates whether or not the test should be run, and a similar description. In use,
these three decorators might be applied like this:
import unittest
import sys
class SkipTests(unittest.TestCase):
@unittest.expectedFailure
def test_fails(self):
self.assertEqual(False, True)
@unittest.skip("Test is useless")
def test_skip(self):
@unittest.skipIf(sys.version_info.minor == 1,
"broken on 3.1")
def test_skipif(self):
@unittest.skipUnless(sys.platform.startswith('linux'),
"broken on linux")
def test_skipunless(self):
if __name__ == "__main__":
unittest.main()
[ 323 ]
The first test fails, but it is reported as an expected failure; the second test is never
run. The other two tests may or may not be run depending on the current Python
version and operating system. On my Linux system running Python 3.1 the output
looks like this:
xssF
=============================================================
FAIL: test_skipunless (__main__.SkipTests)
-------------------------------------------------------------Traceback (most recent call last):
File "skipping_tests.py", line 21, in test_skipunless
AssertionError: False != True
-------------------------------------------------------------Ran 4 tests in 0.001s
FAILED (failures=1, skipped=2, expected failures=1)
The x on the first line indicates an expected failure; the two s characters represent
skipped tests, and the F indicates a real failure, since the conditional to skipUnless
was True on my system.
Testing with py.test
The Python unittest module is very verbose and requires a lot of boilerplate code
to set up and initialize tests. It is based on the very popular JUnit testing framework
for Java. It even uses the same method names (you may have noticed they don't
conform to the PEP-8 naming standard, which suggests underscores be used to
separate words in a method name, rather than CamelCase) and test layout.
While this is effective for testing in Java, it's not necessarily the best design for
Python testing.
Because Python programmers like their code to be elegant and simple, other test
frameworks have been developed, outside the standard library. Two of the more
popular ones are py.test and nose. The latter is not yet supported on Python 3, so
we'll focus on py.test here.
Since py.test is not part of the standard library, you'll need to download and install
it yourself; you can get it from the py.test homepage at http://pytest.org/. The
website has comprehensive installation instructions for a variety of interpreters
and platforms.
[ 324 ]
Chapter 11
py.test has a substantially different layout from the unittest module. It doesn't
require test cases to be classes. Instead, it takes advantage of the fact that functions
are objects, and allows any properly named function to behave like a test. Rather
than providing a bunch of custom methods for asserting equality, it simply uses the
assert statement to verify results. This makes tests more readable and maintainable.
When we run py.test, it will start in the current folder and search for any modules
in that folder or subpackages whose names start with the characters test_. If there
are any functions in this module that also start with test, they will be executed as
individual tests.
Further, if there are any classes in the module whose name starts with Test,
any methods on that class that start with test_ will also be executed in the
test environment.
So let's take the simplest possible unittest example we wrote earlier and port it to
py.test:
def test_int_float():
assert 1 == 1.0
For the exact same test, we've written two lines of more readable code, in comparison
to the six lines we used in our first unittest example.
However, we are not precluded from writing class-based tests. Classes can be useful
for grouping related tests together or for tests that need to access related attributes
or methods on the class. This example shows an extended class with a passing and a
failing test; we'll see that the error output is more comprehensive than that provided
by the unittest module:
class TestNumbers:
def test_int_float(self):
assert 1 == 1.0
def test_int_str(self):
assert 1 == "1"
Notice that the class doesn't have to extend any special objects to be picked up as a
test. If we run py.test on this file, the output looks like this:
============== test session starts ==============
python: platform linux2 -- Python 3.1.2 -- pytest-1.2.1
test object 1: class_pytest.py
class_pytest.py .F
[ 325 ]

=================== FAILURES====================
___________ TestNumbers.test_int_str ____________
self = <class_pytest.TestNumbers object at 0x85b4fac>
>
E
def test_int_str(self):
assert 1 == "1"
assert 1 == '1'
class_pytest.py:7: AssertionError
====== 1 failed, 1 passed in 0.10 seconds =======
The output starts with some useful information about the platform and interpreter.
This can be useful for sharing bugs across disparate systems. The third line tells us
the name of the file being tested (if there are multiple test modules picked up, they
will all be displayed) followed by the familiar .F we saw in the unittest module;
the . indicates a passing test, while the F demonstrates a failure.
After all tests have run, the error output for each of them is displayed. It presents a
summary of local variables (there is only one in this example: the self parameter
passed into the function), the source code where the error occurred, and a summary
of the error message. In addition, if an exception other than an AssertionError is
raised, py.test will present us with a complete traceback including source
code references.
By default, py.test suppresses output from print statements if the test is
successful. This is extremely useful for test debugging; if we have a test that is failing,
we can add print statements to the test to check the values of specific variables and
attributes as the test progresses. If the test fails, the values will be output to help with
diagnostics. However, if the test is successful, the output of the print statements will
not be displayed, allowing them to easily be ignored. Most importantly, we don't
have to "clean up" the output by removing print statements; we can leave them in the
tests and if the tests ever fail again, due to changes in our code, the debugging output
will be immediately available.
One way to do setup and cleanup
py.test supports setup and teardown methods similar to those used in unittest,
but it provides even more flexibility. We'll discuss these briefly, since they are
familiar, but they are not used as extensively as in the unittest module, as py.test
provides us with a powerful funcargs facility, which we'll discuss in the next section.
[ 326 ]
Chapter 11
If we are writing class-based tests, we can use two methods called setup_method
and teardown_method in basically the same way that setUp and tearDown are
called in unittest. They are called before and after each method in the class to do
any setup and cleanup duties. There is one difference from the unittest methods
though. Both methods accept an argument: the function object representing the
method being called.
In addition, py.test provides other setup and teardown functions to give us more
control over when setup and cleanup code is executed. The setup_class and
teardown_class methods are expected to be class methods; they accept a single
argument (there is no self argument) representing the class in question.
Finally, we have the setup_module and teardown_module methods, which are run
immediately before and after all tests (in functions or classes) in that module. These
can be useful for "one time" setup, such as creating a socket or database connection
that will be used by all tests in the module. Be careful with this one, as it can
accidentally introduce dependencies between tests if the object being set up
stores state.
That short description probably doesn't do a great job of explaining exactly when
these setup and teardown methods are called, so let's look at an example that tells
us exactly when it happens:
def setup_module(module):
print("setting up MODULE {0}".format(
module.__name__))
def teardown_module(module):
print("tearing down MODULE {0}".format(
module.__name__))
def test_a_function():
print("RUNNING TEST FUNCTION")
class BaseTest:
def setup_class(cls):
print("setting up CLASS {0}".format(
cls.__name__))
def teardown_class(cls):
print("tearing down CLASS {0}\n".format(
cls.__name__))
def setup_method(self, method):
[ 327 ]

print("setting up METHOD {0}".format(
method.__name__))
def teardown_method(self, method):
print("tearing down METHOD {0}".format(
method.__name__))
class TestClass1(BaseTest):
def test_method_1(self):
print("RUNNING METHOD 1-1")
class TestClass2(BaseTest):
The sole purpose of the BaseTest class is to extract four methods that would be
otherwise identical to the test classes and use inheritance to reduce the amount of
duplicate code. So, from the point of view of py.test, the two subclasses have not
only two test methods each, but also two setup and two teardown methods (one at
the class level, one at the method level).
If we run these tests using py.test, the output shows us when the various functions
are called in relation to the tests themselves. We also have to disable the suppression
of output for the print statements to execute; this is done by passing the -s
(or --capture=no) flag to py.test:
py.test setup_teardown.py -s
setup_teardown.py
setting up MODULE setup_teardown
RUNNING TEST FUNCTION
.setting up CLASS TestClass1
setting up METHOD test_method_1
RUNNING METHOD 1-1
.tearing down METHOD test_method_1
RUNNING METHOD 1-2
tearing down CLASS TestClass1
[ 328 ]
Chapter 11
setting up CLASS TestClass2
RUNNING METHOD 2-1
RUNNING METHOD 2-2
tearing down CLASS TestClass2
tearing down MODULE setup_teardown
The setup and teardown methods for the module are executed at the beginning and
end of the session. Then the lone module-level test function we added is run. Next,
the setup method for the first class is executed, followed by the two tests for that
class. The tests, however, are each individually wrapped in separate setup_method
and teardown_method calls. After the methods have been executed, the class
teardown method is called. The same sequence happens for the second class,
before the teardown_module method is finally called, exactly once.
A completely different way to set up variables

One of the most common uses for the various setup and teardown functions is to
ensure certain class or module variables are available with a known value before
each test method is run.
py.test offers a completely different way to do this using what are known as
funcargs, short for function arguments. Funcargs are basically named variables
that are previously set up in a test configuration file. This allows us to separate
configuration from execution of tests, and allows the funcargs to be used across
multiple classes and modules.
To use them, we simply add parameters to our test function. The names of the
parameters are used to look up specific arguments in specially named functions.
For example, if we wanted to test the StatsList class we used earlier, while
demonstrating unittest, we would again want to repeatedly test a list of valid
integers. But we can write our tests like so instead of using setup methods:
def pytest_funcarg__valid_stats(request):
return StatsList([1,2,2,3,3,4])
def test_mean(valid_stats):
assert valid_stats.mean() == 2.5
[ 329 ]

def test_median(valid_stats):
assert valid_stats.median() == 2.5
valid_stats.append(4)
assert valid_stats.median() == 3
def test_mode(valid_stats):
assert valid_stats.mode() == [2,3]
valid_stats.remove(2)
assert valid_stats.mode() == [3]
Each of the three test methods accepts a parameter named valid_stats; this
parameter is created afresh by calling the pytest_funcarg__valid_stats function
defined at the top of the file. It can also be defined in a file called conftest.py if the
funcarg is needed by multiple modules. The conftest.py file is parsed by py.test
to load any "global" test configuration; it is a sort of catchall for customizing the
py.test experience. It's actually normal to put funcargs in this module instead of
your test file, in order to completely separate the configuration from the test code.
As with other py.test features, the name of the factory for returning a funcarg
is important; funcargs are simply functions that are named pytest_funcarg__
<valid_identifier>, where <valid_identifier> is a valid variable name that can
be used as a parameter in a test function. This function accepts a mysterious request
parameter, and returns the object that should be passed as an argument into the
individual test functions. The funcarg is created afresh for each call to an individual
test function; this allows us, for example, to change the list in one test and know that
it will be reset to its original values in the next test.
Funcargs can do a lot more than return simple variables. That request object passed
into the funcarg factory gives us some extremely useful methods and attributes to
modify the funcarg's behavior. The module, cls, and function attributes allow us
to see exactly which test is requesting the funcarg. The config attribute allows us
to check command-line arguments and other configuration data. We don't have
room to go into detail on this topic, but custom command-line arguments can be
used to customize the test experience by running certain tests only if an argument is
passed (useful for slow tests that need to be run less often) or supplying connection
parameters to a database, file, or hardware device.
More interestingly, the request object provides methods that allow us to do
additional cleanup on the funcarg or to reuse it across tests. The former allows us
to use funcargs instead of writing custom teardown functions to clean up open files
or connections, while the latter can help reduce the time it takes to run a test suite if
the setup of a common funcarg is time consuming. This is often used for database
connections, which are slow to create and destroy and do not need to be reinitialized
after each test (although the database still typically needs to be reset to a known state
between tests).
[ 330 ]
Chapter 11
The request.addfinalizer method accepts a callback function that does any

cleanup after each test function that uses a funcarg has been called. This can provide
the equivalent of a teardown method, allowing us to clean up files, close connections,
empty lists or reset queues. For example, the following code tests the os.mkdir
functionality by creating a temporary directory funcarg:
import tempfile
import shutil
import os.path
def pytest_funcarg__temp_dir(request):
dir = tempfile.mkdtemp()
print(dir)
def cleanup():
shutil.rmtree(dir)
request.addfinalizer(cleanup)
return dir
def test_osfiles(temp_dir):
os.mkdir(os.path.join(temp_dir, 'a'))
os.mkdir(os.path.join(temp_dir, 'b'))
dir_contents = os.listdir(temp_dir)
assert len(dir_contents) == 2
assert 'a' in dir_contents
assert 'b' in dir_contents
The funcarg creates a new empty temporary directory for files to be created in.
Then it adds a finalizer call to remove that directory (using shutil.rmtree, which
recursively removes a directory and anything inside it) after the test has completed.
The file system is then left in the same state in which it started.
Then we have the request.cached_setup method, which allows us to create
function argument variables that last longer than one test. This is useful when setting
up an expensive operation that can be reused by multiple tests without the resource
reuse breaking the atomic or unit nature of the tests (so that one test does not rely on
and is not impacted by a previous one). For example, if we want to test the following
echo server, we may want to run only one instance of the server in a separate process
and then have multiple tests connect to that instance.
import socket
s = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
s.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEADDR, 1)
s.bind(('localhost',1028))
[ 331 ]

s.listen(1)
while True:
client, address = s.accept()
data = client.recv(1024)
client.send(data)
client.close()
All this code does is listen on a specific port and wait for input from a client socket.
When it receives input, it just sends the same value back. To test this, we can start the
server in a separate process and cache the result for use in multiple tests. Here's how
the test code might look:
import subprocess
import socket
import time
def pytest_funcarg__echoserver(request):
def setup():
p = subprocess.Popen(
['python3', 'echo_server.py'])
time.sleep(1)
return p
def cleanup(p):
p.terminate()
return request.cached_setup(
setup=setup,
teardown=cleanup,
scope="session")
def pytest_funcarg__clientsocket(request):
s = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
s.connect(('localhost', 1028))
request.addfinalizer(lambda: s.close())
return s
def test_echo(echoserver, clientsocket):
clientsocket.send(b"abc")
assert clientsocket.recv(3) == b'abc'
def test_echo2(echoserver, clientsocket):
clientsocket.send(b"def")
assert clientsocket.recv(3) == b'def'
[ 332 ]
Chapter 11
We've created two funcargs here. The first runs the echo server in a separate process,
and returns the process object. The second instantiates a new socket object for
each test, and closes it when the test has completed, using addfinalizer. The first
funcarg is the one we're currently interested in. It looks much like a traditional unit
test setup and teardown. We create a setup function that accepts no parameters and
returns the correct argument, in this case, a process object that is actually ignored by
the tests, since they only care that the server is running. Then we create a cleanup
function (the name of the function is arbitrary since it's just an object we pass into
another function), which accepts a single argument: the argument returned by
setup. This cleanup code simply terminates the process.
Instead of returning a funcarg directly, the parent function returns the results
of a call to request.cached_setup. It accepts two arguments for the setup and
teardown functions (which we just created), and a scope argument. This last
argument should be one of the three strings "function", "module", or "session"; it
determines just how long the argument will be cached. We set it to "session" in this
example, so it is cached for the duration of the entire py.test run. The process will
not be terminated or restarted until all tests have run. The "module" scope, of course,
caches it only for tests in that module, and the "function" scope treats the object more
like a normal funcarg, in that it is reset after each test function is run.
Test skipping with py.test
As with the unittest module, it is frequently necessary to skip tests in py.test,

for a variety of reasons; the code being tested hasn't been written yet, the test only
runs on certain interpreters or operating systems, or the test is time consuming and
should only be run under certain circumstances.
We can skip tests at any point in our code using the py.test.skip function. It
accepts a single argument: a string describing why it has been skipped. This function
can be called anywhere; if we call it inside a test function, the test will be skipped. If
we call it in a module, all the tests in that module will be skipped. If we call it inside
a funcarg function, all tests that call that funcarg will be skipped.
[ 333 ]
Of course, in all these locations, it is often desirable to skip tests only if certain
conditions are or are not met. Since we can execute the skip function at any place in
Python code, we can execute it inside an if statement. So we may write a test that
looks like this:
import sys
import py.test
def test_simple_skip():
if sys.platform != "fakeos":
py.test.skip("Test works only on fakeOS")
fakeos.do_something_fake()
assert fakeos.did_not_happen
That's some pretty silly code, really. There is no Python platform named fakeos, so
this test will skip on all operating systems. It shows how we can skip conditionally,
and since the if statement can check any valid conditional, we have a lot of power
over when tests are skipped. Often, we check sys.version_info to check the
Python interpreter version, sys.platform to check the operating system, or
some_library.__version__ to check if we have a recent enough version of
a given API.
Since skipping an individual test method or function based on a certain conditional
is one of the most common uses of test skipping, py.test provides a convenience
decorator that allows us to do this in one line. The decorator accepts a single string,
which can contain any executable Python code that evaluates to a Boolean value. For
example, the following test will only run on Python 3 or higher:
import py.test
@py.test.mark.skipif("sys.version_info <= (3,0)")
def test_python3():
assert b"hello".decode() == "hello"
The py.test.mark.xfail decorator behaves similarly, except that it marks a test as

expected to fail, similar to unittest.expectedFailure(). If the test is successful, it
will be recorded as a failure, if it fails, it will be reported as expected behavior. In the
case of xfail, the conditional argument is optional; if it is not supplied, the test will
be marked as expected to fail under all conditions.
[ 334 ]
Chapter 11
py.test extras
py.test is an incredibly powerful library and it can do much much more than
the basics we've discussed here. We haven't even started on its distributed testing
framework (which allows tests to be run across a network of different platforms and
interpreter versions), its numerous built-in or third-party plugins, how incredibly
easy it is to write our own plugins, or the extensive customization and configuration
architecture the framework supplies. You'll have to read the documentation at
http://pytest.org/ for all the juicy details.
However, before we leave for the day, we should discuss some of the more
useful command-line arguments built into py.test. As with most command-line
applications, we can get a list of the available command-line arguments by running
the command py.test --help. However, unlike many programs, the available
command-line options depends on what py.test plugins are installed and whether
we've written any arguments of our own into the conftest.py for the project.
First, we'll look at a couple of arguments that help us with debugging tests. If we
have a large test suite with many tests failing (because we've done invasive code
changes, such as porting the project from Python 2 to Python 3), the py.test output
can quickly get away from us. Passing the -x or --exitfirst command-line
argument to py.test forces the test runner to exit after the first failure. We can then
fix whatever problems are causing that test to fail before running py.test again and
checking out the next failure.
The --pdb argument is similar, except that instead of exiting after a test fails, it drops
to a python debugger shell. If you know how to use the debugger, this feature can
allow you to quickly introspect variables or step through broken code.
py.test also supports an interesting --looponfail or -f argument, although it's
only available if the py.test xdist plugin is installed. This plugin is available from
http://pypi.python.org/pypi/pytest-xdist. If it's installed and we pass the
--looponfail option to py.test, the test suite will automatically rerun itself when
a failing test is encountered. This means that we can wait for a test to fail, then edit
the test and fix the broken code. When we save the file, the test will automatically
run again to tell us if our fix was successful. It's basically like using the --exitfirst
argument repeatedly as we fix one test at a time, but automates the boring
restarting bits!
[ 335 ]
The most important of the py.test arguments is the -k argument, which accepts
a keyword to search for tests. It is used to run specific tests that contain the given
keyword argument in the full name (including package, module, class, and test
name). For example, if we have the following structure:
package: something/
module:
test_something.py

class: TestSomething

method: test_first

method: test_second
We can run py.test -k test_first or even just py.test -k first to run just the
one test_first method. Or, if there are other methods that have that name, we can
run py.test -k TestSomething.test_first or even something.test_something.
TestSomething.test_first. py.test, which will first collect the complete test
name into a dot-separated string, and then checks to see if the string contains the
requested keyword.
How much testing is enough?
We've already established that untested code is broken code. But how can we tell
how well our code is tested? How do we know how much of our code is actually
being tested? The first question is the more important one, but it's hard to answer.
Even if we know we have tested every line of code in our application, we do not
know that we have tested it properly. For example, if we write a stats test that only
checks what happens when we provide a list of integers, it may still fail spectacularly
if used on a list of floats or strings or self-made objects. The onus of designing
complete test suites still lies with the programmer.
The second question, how much of our code is actually being tested, is actually easy
to verify. Code coverage is essentially an estimate of the number of lines of code that
are executed by a program. If we know that number and the number of lines that
are in the program, we can get an estimate of what percentage of the code was really
tested, or covered. If we additionally have an indicator as to which lines were not
tested, we can more easily write new tests to ensure those lines are less broken.
There are two popular tools for coverage testing in Python: figleaf, and
coverage.py. Only coverage.py is Python 3 compatible as I write this, so we'll
focus on it. It can be downloaded from http://nedbatchelder.com/code/
coverage/.
[ 336 ]
Chapter 11
We don't have space to cover all the details of the coverage API, so we'll just look at a
few typical examples. If we have a python script that runs all our unit tests for us (for
example, using unittest.main or a custom test runner or discover), we can use the
command below to perform a coverage analysis:
>>> coverage run coverage_unittest.py
This command will exit normally, but it creates a file named .coverage that holds
the data from the run. We can now use the coverage report command to get an
analysis of code coverage:
>>> coverage report
The output is as follows:

Name
Stmts
Exec Cover
-------------------------------------------------coverage_unittest
7
7
100%
stats
19
6
31%
-------------------------------------------------TOTAL
26
13
50%
This simple report lists the files that were executed (our unit test and a module
it imported). The number of lines of code in each file, and the number that were
executed by the test are also listed. The two numbers are then combined to estimate
the amount of code coverage. If we pass the -m option to the report command, it will
additionally add a column that looks like this:
Missing
----------8-12, 15-23
The ranges of lines listed here identify lines in the stats module that were not
executed during the test run.
The example we just tested code coverage on uses the same stats module we've
been using throughout the chapter, but uses a single test that deliberately misses
testing a lot of code in the file:
import unittest
class TestMean(unittest.TestCase):
def test_mean(self):
self.assertEqual(StatsList([1,2,2,3,3,4]).mean(), 2.5)
if __name__ == "__main__":
unittest.main()
[ 337 ]
This code doesn't test the median or mode functions. These correspond to the line
numbers that the coverage output told us were missing.
The simple report earlier is sufficient, but if we use the command coverage html, we
can get an even fancier interactive html report that we can view in a web browser.
The web page even highlights which lines in the source code were and were not
tested. Here's how it looks:
The green lines were executed by the test, the red ones were not. There are some
other interactive features of these reports if you want to try them out.
We can use the coverage.py module with py.test as well. We simply need to
install the py.test plugin for code coverage, which is available from http://pypi.
python.org/pypi/pytest-coverage/. This adds several command-line options to
py.test, the most useful being --cover-report, which can be set to either html,
report, or annotate (the latter actually modifies the source code to show any lines
that were not covered).
[ 338 ]
Chapter 11
If we ran a coverage report on this section of the chapter, we'd find that we have
not covered most of what there is to know about code coverage! It is possible to
use the coverage API to manage code coverage from our own programs (or test
suites), and coverage.py accepts numerous configuration options that we haven't
touched on. Nor have we discussed the difference between statement coverage and
branch coverage (the latter is much more useful, and the default in recent versions of
coverage.py) or other styles of code coverage.
Bear in mind that while 100% code coverage is a lofty goal that we should all strive
for, 100% coverage is not enough! Just because a statement was tested does not mean
that it was tested properly for all possible input.
Case Study
Let's take a walk through test-driven development by writing a small, tested,

cryptography application. Don't worry, you won't need to understand the
mathematics behind complicated modern encryption algorithms such as Threefish or
RSA. Instead, we'll be implementing a 16th century algorithm known as the Vigenre
cipher. The application simply needs to be able to encode and decode a message,
given an encoding keyword, using this cipher.
First we need to understand how the cipher works if we apply it manually (without
a computer). We start with a table like this:
A
B
C
D
E
F
G
H
I
J
K
L
M
N
O
P
Q
R
S
T
U
V
W
X
Y
Z
B
C
D
E
F
G
H
I
J
K
L
M
N
O
P
Q
R
S
T
U
V
W
X
Y
Z
A
C
D
E
F
G
H
I
J
K
L
M
N
O
P
Q
R
S
T
U
V
W
X
Y
Z
A
B
D
E
F
G
H
I
J
K
L
M
N
O
P
Q
R
S
T
U
V
W
X
Y
Z
A
B
C
E
F
G
H
I
J
K
L
M
N
O
P
Q
R
S
T
U
V
W
X
Y
Z
A
B
C
D
F
G
H
I
J
K
L
M
N
O
P
Q
R
S
T
U
V
W
X
Y
Z
A
B
C
D
E
G
H
I
J
K
L
M
N
O
P
Q
R
S
T
U
V
W
X
Y
Z
A
B
C
D
E
F
H
I
J
K
L
M
N
O
P
Q
R
S
T
U
V
W
X
Y
Z
A
B
C
D
E
F
G
I
J
K
L
M
N
O
P
Q
R
S
T
U
V
W
X
Y
Z
A
B
C
D
E
F
G
H
J
K
L
M
N
O
P
Q
R
S
T
U
V
W
X
Y
Z
A
B
C
D
E
F
G
H
I
K
L
M
N
O
P
Q
R
S
T
U
V
W
X
Y
Z
A
B
C
D
E
F
G
H
I
J
L
M
N
O
P
Q
R
S
T
U
V
W
X
Y
Z
A
B
C
D
E
F
G
H
I
J
K
M
N
O
P
Q
R
S
T
U
V
W
X
Y
Z
A
B
C
D
E
F
G
H
I
J
K
L
N
O
P
Q
R
S
T
U
V
W
X
Y
Z
A
B
C
D
E
F
G
H
I
J
K
L
M
O
P
Q
R
S
T
U
V
W
X
Y
Z
A
B
C
D
E
F
G
H
I
J
K
L
M
N
P
Q
R
S
T
U
V
W
X
Y
Z
A
B
C
D
E
F
G
H
I
J
K
L
M
N
O
Q
R
S
T
U
V
W
X
Y
Z
A
B
C
D
E
F
G
H
I
J
K
L
M
N
O
P
R
S
T
U
V
W
X
Y
Z
A
B
C
D
E
F
G
H
I
J
K
L
M
N
O
P
Q
S
T
U
V
W
X
Y
Z
A
B
C
D
E
F
G
H
I
J
K
L
M
N
O
P
Q
R
T
U
V
W
X
Y
Z
A
B
C
D
E
F
G
H
I
J
K
L
M
N
O
P
Q
R
S
[ 339 ]
U
V
W
X
Y
Z
A
B
C
D
E
F
G
H
I
J
K
L
M
N
O
P
Q
R
S
T
V
W
X
Y
Z
A
B
C
D
E
F
G
H
I
J
K
L
M
N
O
P
Q
R
S
T
U
W
X
Y
Z
A
B
C
D
E
F
G
H
I
J
K
L
M
N
O
P
Q
R
S
T
U
V
X
Y
Z
A
B
C
D
E
F
G
H
I
J
K
L
M
N
O
P
Q
R
S
T
U
V
W
Y
Z
A
B
C
D
E
F
G
H
I
J
K
L
M
N
O
P
Q
R
S
T
U
V
W
X
Z
A
B
C
D
E
F
G
H
I
J
K
L
M
N
O
P
Q
R
S
T
U
V
W
X
Y
Given a keyword TRAIN, we encode the message ENCODED IN PYTHON

as follows:
1. Repeat the keyword and message together such that it is easy to map letters
from one to the other:

E N C O D E D I N P Y T H O N
T R A I N T R A I N T R A I N
2. For each letter in the plaintext, find the row that begins with that letter in
the table.
3. Find the column with the letter associated with the keyword letter for the
chosen plaintext letter.
4. The encoded character is at the intersection of this row and column.
For example, the row starting with E intersects the column starting with T at the
character X. So the first letter in the ciphertext is X. The row starting with N intersects
the column starting with R at the character E, leading to the ciphertext XE. C
intersects A at C, and O intersects I at W. D and N map to Q while E and T map
to X. The full encoded message is XECWQXUIVCRKHWA.
Decoding basically follows the opposite procedure. First find the row with the
character for the shared keyword (the T row) then find the location in that row where
the encoded character (the X) is located. The plaintext character is at the top of the
column for that row (the E).
Implementing it
Our program will need an encode method that takes a keyword and plaintext and
returns the ciphertext, and a decode method that accepts a keyword and ciphertext
and returns the original message.
But before we write those methods, let's follow a test-driven development strategy.
We'll be using py.test for our unit testing. We need an encode method, and we
know what it has to do; let's write a test for that method first:
def test_encode():
cipher = VigenereCipher("TRAIN")
encoded = cipher.encode("ENCODEDINPYTHON")
assert encoded == "XECWQXUIVCRKHWA"
[ 340 ]
Chapter 11
This test, fails, naturally, because we aren't importing a VigenereCipher class

anywhere. Let's create a new module to hold that class.
Let's start with the following VigenereCipher class:
class VigenereCipher:
def __init__(self, keyword):
self.keyword = keyword
def encode(self, plaintext):
return "XECWQXUIVCRKHWA"
If we add a from vigenere_cipher import VigenereCipher line to the top of

our test class and run py.test, the above test will pass! We've finished our first
test-driven development cycle.
Obviously, returning a hardcoded string is not the most sensible implementation of a
cipher class, so let's add a second test:
def test_encode_character():
encoded = cipher.encode("E")
assert encoded == "X"
Ah, that test will fail. Looks like we're going to have to work harder. But I just
thought of something, what if someone tries to encode a string with spaces or
lowercase characters? Before we start implementing the encoding, we better add
some tests for these cases, before we forget them. Expected behavior will be to
remove spaces, and to convert lowercase letters to capitals:
def test_encode_spaces():
encoded = cipher.encoded("ENCODED IN PYTHON")
def test_encode_lowercase():
cipher = VigenereCipher("TRain")
encoded = cipher.encoded("encoded in Python")
If we run the new test suite, we find that the new tests pass (they expect the same
hardcoded string). But they ought to fail later if we forget to account for these cases.
Now we have to think about how to implement our encoding algorithm. Writing
code to use a table like we used in the manual algorithm above is possible, but
seems complicated, considering that each row is just an alphabet rotated by an
offset number of characters.
[ 341 ]
It turns out that we can use modulo arithmetic to combine the characters instead
of doing a table lookup. Given plaintext and keyword characters, if we convert the
two letters to their numerical values (with A being 0 and Z being 25), add them
together, and take the remainder mod 26, we get the ciphertext character! This is a
straightforward calculation, but since it happens on a character-by-character basis,
we should probably put it in its own function. And before we do that, we should
write a test for the new function:
from vigenere_cipher import combine_character
def test_combine_character():
assert combine_character("E", "T") == "X"
assert combine_character("N", "R") == "E"
Now, as usual, we can write the code to make this function work. In all honesty, I
had to run the test several times before I got this function completely correct; first
I returned an integer, and then I forgot to shift the character back up to the normal
ASCII scale from the zero-based scale. Having the test available made it easy to test
and debug these errors, another bonus of test-driven development.
def combine_character(plain, keyword):
plain = plain.upper()
keyword = keyword.upper()
plain_num = ord(plain) - ord('A')
keyword_num = ord(keyword) - ord('A')
return chr(ord('A') + (plain_num + keyword_num) % 26)
I thought we'd be ready to implement our encode function. However, the first
thing we want inside that function is a repeating version of the keyword string that
is as long as the plaintext. Let's implement a function for that first. Or rather, let's
implement the test first!
def test_extend_keyword():
extended = cipher.extend_keyword(16)
assert extended == "TRAINTRAINTRAINT"
Before writing this test, I expected to write extend_keyword as a standalone function

that accepted a keyword and an integer. But as I started drafting the test, I realized
it made more sense to use it as a helper method on the VigenereCipher class. This
shows how test-driven development can help us design more sensible APIs. Here's
the method implementation:
def extend_keyword(self, number):
repeats = number // len(self.keyword) + 1
return (self.keyword * repeats)[:number]
[ 342 ]
Chapter 11
Once again, this took a few runs of the test to get right. I actually used two versions
of the test, one with fifteen and one with sixteen letters to make sure it works if the
integer division has an even number.
Now we're finally ready to write our encode method:
cipher = []
keyword = self.extend_keyword(len(plaintext))
for p,k in zip(plaintext, keyword):
cipher.append(combine_character(p,k))
return "".join(cipher)
That looks correct. Our test suite should pass now, right?
Actually, if we run it, we'll find that two tests are still failing. We totally forgot about
the spaces and lowercase characters! It is a good thing we wrote tests to remind us.
We'll have to add this line at the beginning of the method:
plaintext = plaintext.replace(" ", "").upper()
Now all the tests pass successfully. We're running short on space, so we'll condense
the examples for decoding. Here are a couple tests:
def test_separate_character():
assert separate_character("X", "T") == "E"
assert separate_character("E", "R") == "N"
def test_decode():
decoded = cipher.decode("XECWQXUIVCRKHWA")
assert decoded == "ENCODEDINPYTHON"
Here's the separate_character function:

def separate_character(cypher, keyword):
cypher = cypher.upper()
keyword = keyword.upper()
cypher_num = ord(cypher) - ord('A')
keyword_num = ord(keyword) - ord('A')
return chr(ord('A') + (cypher_num - keyword_num) % 26)
[ 343 ]
And the decode method:

def decode(self, ciphertext):
plain = []
keyword = self.extend_keyword(len(ciphertext))
for p,k in zip(ciphertext, keyword):
plain.append(separate_character(p,k))
return "".join(plain)
These methods have a lot of similarity to those used for encoding. The great thing
about having all these tests written and passing is that we can now go back and
modify our code, knowing it is still safely passing the tests. For example, if we
replace our existing encode and decode methods with these refactored methods,
our tests still pass:
def _code(self, text, combine_func):
text = text.replace(" ", "").upper()
combined = []
keyword = self.extend_keyword(len(text))
for p,k in zip(text, keyword):
combined.append(combine_func(p,k))
return "".join(combined)
return self._code(plaintext, combine_character)
def decode(self, ciphertext):
return self._code(ciphertext, separate_character)
This is the final benefit of test-driven development, and the most important. Once
the tests are written, we can improve our code as much as we like and be confident
that our changes didn't break anything we have been testing for. Further, we know
exactly when our fixes are finished: when the tests all pass.
Of course, our tests may not comprehensively test everything we need them to;
maintenance or code refactoring may still cause undiagnosed bugs that don't show
up in testing. Automated tests are not foolproof. If bugs do occur, however, it is still
possible to follow a test-driven plan; step one is to write a test (or multiple tests) that
duplicates or "proves" that the bug in question is occurring. This will, of course, fail.
Then write the code to make the tests stop failing. If the tests were comprehensive,
the bug will be fixed, and we will know if it ever happens again, as soon as we run
the test suite.
[ 344 ]
Chapter 11
Finally, we can try to determine how well our tests operate on this code. With the
py.test-coverage plugin installed, py.test coverage-report=report tells us
that our test suite has 100% code coverage. This is a great statistic, but we shouldn't
get too cocky about it. Our code hasn't been tested when encoding messages that
have numbers, and its behavior with such inputs is thus undefined. Untested code is
broken code. Untested use cases are undefined scenarios.
Exercises
Practice test-driven development. That is your first exercise. It's easier to do this if
you're starting a new project, but if you have existing code you need to work on,
you can start by writing tests for each new feature you implement. This can become
frustrating as you become more enamored with automated tests. The old, untested
code will start to feel uncomfortable to maintain; you'll start feeling like changes you
make are breaking the code and you have no way of knowing, for lack of tests.
So to get your feet wet with test-driven development, start a fresh project. Once
you've started to appreciate the benefits (you will) and realize that the time spent
writing tests is quickly regained in terms of more maintainable code, you'll want
to start writing tests for existing code. This is when you should start doing it, not
before. Writing tests for code that we "know" works is boring. It is hard to get
interested in the project until you realize just how broken the code we "thought"
was working really is.
Try writing the same set of tests using both the built-in unittest module and
py.test. Which do you prefer? unittest is more similar to test frameworks in other
languages, while py.test is arguably more Pythonic. Both allow us to write objectoriented tests and to test object-oriented programs with ease.
We used py.test in our case study, but we didn't touch on any features that
wouldn't have been easily testable using unittest. Try adapting the tests to use test
skipping or funcargs. Try the various setup and teardown methods, and compare
their use to funcargs; which feels more natural to you?
In our case study, we have a lot of tests that use a similar VigenereCipher object; try
reworking this code to use a funcarg. How many lines of code does it save?
Try running a coverage report on the tests you've written. Did you miss testing
any lines of code? Even if you have 100% coverage, have you tested all the possible
inputs? If you're doing test-driven development, 100% coverage should follow quite
naturally, as you will write a test before the code that satisfies that test. However, if
writing tests for existing code, it is more likely that there will be edge conditions that
go untested.
[ 345 ]
Think carefully about the values that are somehow different: empty lists when you
expect full ones, zero or one or infinity compared to intermediate integers, floats that
don't round to an exact decimal place, strings when you expected numerals, or the
ubiquitous None value when you expected something meaningful. If your tests cover
such edge cases, your code will be in good shape.
Summary
We have finally covered the most important topic in Python programming:

automated testing. After proving its importance and considering best design
principles, we discussed the basic API for two popular Python 3 test frameworks.
In particular we covered:
Unit testing
The unittest module
Assertion methods and code setup/cleanup
The py.test framework
Code coverage
In the next chapter, we'll wrap up our learning with a compendium of

object-oriented frameworks and libraries that work in Python 3.
[ 346 ]
Common Python 3 Libraries

We've covered the principles of object-oriented programming and we've applied
them to Python. We've looked at the ins and outs of object-oriented design, and
the higher-level design patterns that make up good programs. We've seen Python's
tendency to simplify object-oriented solutions. We even know how to test our Python
programs. Yet, are we able to do the common tasks of day-to-day programming?
Yes, we know Python's syntax, and we couldin theorywrite a web framework
or database engine from scratch. Python's true power, however, lies in the work
other people have done before us. In the examples throughout this book, we've
seen many of the Python standard library's modules at work. Yet, we haven't really
covered many of the most common tasks facing Python programmers today. We've
completely bypassed graphical applications and their widgets, input boxes and
buttons: one of the most common interfaces users see today. And we haven't
touched on web backend development: Python's current most prevalent use.
These are complex topics, and we'll see an introduction to each of them here. We'll
be focusing on Python libraries that are available for Python 3 as I write this. Many
popular libraries, such as Django or wxPython, are currently only compatible with
older versions of Python, so they'll be passed over.
In this chapter we'll be covering:
Database libraries and object-relational managers
Point-and-click graphical applications
CherryPy for web applications
Working with XML
Database access
Talking to databases is a very common task in Python, especially in the web

development world. Unfortunately, not many libraries for database access have
been ported to Python 3 in a mature state. We'll be looking at a few of the available
database solutions.
Python comes with built-in support for SQLite 3. We looked at some examples of it
in earlier chapters. SQLite is not suitable for multi-user, multi-threaded access, but
it's perfect for storing configuration or local data. It simply stores all the data in a
single file and allows us to access that data using SQL syntax. All we need to do to
use it is import sqlite3 and read the help file. Here's a short example to get
you started:
import sqlite3
connection = sqlite3.connect("mydb.db")
connection.execute(
"CREATE TABLE IF NOT EXISTS "
"pet (type, breed, gender, name)")
connection.execute("INSERT INTO pet VALUES("
"'dog', 'spaniel', 'female', 'Esme')")
connection.execute("INSERT INTO pet VALUES("
"'cat', 'persian', 'male', 'Oscar')")
results = connection.execute("SELECT breed, name"
" from pet where type='dog'")
for result in results:
print(result[1])
connection.close()
This code first connects to a local file named mydb.db (it creates the file if it
doesn't exist) and runs some SQL query to put a simple table in the database.
Then it queries the same relation and prints one of the results.
Results are returned as iterable sequences of tuples. Each tuple represents a
matching row in the query results. The order of the values in each result tuple is
the same as the order of the values in the query. The name is the second column
in the query (the first is type), so we print result[1] to print the name of the
queried pet.
The Python API SQLite uses conforms to a database API specification known as
DBAPI2. This API is a standard, designed to make it easier for code to interact
with different types of databases in the same way. There are similar APIs for
other databases such as PostgreSQL, MySQL and Oracle, among many others,
but at this time, very few of them are mature on Python 3.
[ 348 ]
Chapter 12
Any database API that follows the DBAPI2 specification will have a connect
function (which may take different arguments for different database connections)
that returns a Connection object. Queries are executed on the connection using
an execute method. Often, additional methods to make querying easier or return
named tuples as results are provided; however, none of this is required by the
DBAPI2 specification.
However, DBAPI2 is fairly low-level and difficult to work with. In objectoriented programming, it is very common to use an Object-Relational Manager,
or ORM to interact with databases. ORMs allow us to use the familiar abstraction
called objects that we've been working with throughout this book, while
connecting their attributes to the relational database paradigm. One of the
most popular ORMs in Python is SQLAlchemy, and it was also the first to be
ported to Python 3.
Introducing SQLAlchemy
SQLAlchemy can be downloaded from http://www.sqlalchemy.org/. Only the

0.6 version and higher is supported on Python 3, and at the time of writing, the only
underlying databases that are supported are SQLite and PostgresSQL.
This isn't a huge deal, as SQLAlchemy provides an abstraction over database APIs,
it is (theoretically) possible to write SQLAlchemy code that works on one database
system and later use the exact same (or only slightly modified) code on another
one. So if you're looking for MySQL support, you could begin to write your code to
use SQLAlchemy with SQLite as a backend first, and port it to MySQL when that
backend is eventually supported.
SQLAlchemy is a very large and robust library; it allows us to do almost anything
imaginable with a database. We'll only be able to touch on the basics in this section.
The idea behind SQLAlchemy, and ORMs in general is to interact with objects that
automatically modify and update database tables in the background. SQLAlchemy
provides multiple ways to map objects to tables; we'll be using the modern
inheritance-based solution. If you need to connect to a legacy database, SQLAlchemy
provides an alternative method that allows arbitrary object classes to be explicitly
mapped to database tables, but we won't have room to cover that here.
[ 349 ]
The first thing we need to do is connect to a database. The sqlalchemy.create_

engine function provides a single point of access for connecting to a database.
It takes a huge number of arguments to customize or tune access. The most
important one is a string URL defining the kind of backend database to be
connected to, the specific database backend to make the connection, the name
of the database, the host the database system is running on, and a username and/or
password to authenticate with. The basic form of the URL resembles a web URL:
driver://user:password@host/dbname.
If we want to use a simple SQLite database, which does not require username,
password, or host; we can simply specify the filename for the database, as we'll
see in the next example.
Then we'll need to create a class that allows objects to store their data in the database,
while optionally supplying behavior as methods on the object. Each instance of the
object will be stored in a separate row in the database, identified by its primary key
(it is usually a good idea to make the primary key a single integer identifier, but
SQLAlchemy does not require this).
Each table in the database is normally represented by a separate class, and special
attributes on each class map to table columns. When we access these attributes on
an object, we get database values, and when we update and save the object, the
database is modified. Here's a simple example for our pets database:
import sqlalchemy as sqa
from sqlalchemy.ext.declarative import declarative_base
Base = declarative_base()
class Pet(Base):
__tablename__ = "pet"
id = sqa.Column(sqa.Integer, primary_key=True)
type = sqa.Column(sqa.String(16))
breed = sqa.Column(sqa.String(32))
gender = sqa.Column(sqa.Enum("male", "female"))
name = sqa.Column(sqa.String(64))
engine = sqa.create_engine('sqlite:///mydata.db')
Base.metadata.create_all(engine)
SQLAlchemy first asks us to set up a Base class by calling a function called

declarative_base. This function returns a class, which we are able to extend in our
declaration. The subclass needs a special attribute named __tablename__ to specify
the name of the table in the database.
[ 350 ]
Chapter 12
This is followed by several column declarations. We add Column objects whose

first argument is a type object (example Integer or String), and subsequent
arguments depend on the type. All of these type objects are provided in the
sqlalchemy package. I generally import this package with the alias sqa to make it
easier to reference the many classes in the package. Some people suggest using from
sqlalchemy import * syntax, so all the objects are available, but as we discussed in
Chapter 2, this can make code very confusing to maintain.
After defining one or more mapped classes that extend the Base object, we connect
to a specific database (in this case, an SQLite file) using the create_engine function.
The Base.metadata.create_all call ensures that all the tables associated with
that Base class exist. It would typically issue some sort of CREATE TABLE call to the
underlying database.
Adding and querying objects
We can create instances of our table objects just like a normal object. The default
constructor on the Base class accepts no arguments. It can often be useful to add an
__init__ method to our subclass that initializes some or all of the variables on the
object. We can also add any other arbitrary methods to the class that we like. Here's
how we might instantiate a new pet object and set some values:
pet = Pet()
pet.id = 1
pet.type = "dog"
pet.breed = "spaniel"
pet.gender = "female"
pet.name = "Esme"
This object can be used like any other Python object, but the object is not yet
connected to the database in any way. Before we can associate the object with a
database table row, we need to create an SQLAlchemy Session object. Sessions are
like staging areas between objects and the database. We can add multiple objects
to the session, as well as use the session to record changes, deletions, and other
database operations. When we're ready for this collection of changes to be saved
to the database, we can commit() them, or, if something goes wrong, we can call
session.rollback() to make all the changes disappear.
Here's how we can add our new pet to the database and save it:
Session = sqa.orm.sessionmaker(bind=engine)
session = Session()
session.add(pet)
session.commit()
[ 351 ]
First we have to get a special Session class by calling the sessionmaker function;
this function needs to know which engine to connect to. Then whenever we want
a session, we instantiate the resulting class. Each session is partially independent
of the others until the changes are committed. Underneath, they basically rely on
database transactions, so similar rules apply, and the rules may vary depending on
the underlying database.
We can also use session objects to query the database. SQLAlchemy queries are
written in a combination of Python functions and raw SQL syntax. We use the
session.query() method to get a Query object. This method accepts arguments
representing the tables or columns to be queried. Then methods on that object can
be cascaded to get a set of results. These methods include:
all(), which returns all items in the table.
first(), which returns the first item.
one(), which returns the only item. If no items or multiple items are found, it
get(primary_key), which accepts a primary key value and returns the

object matching that key.
group_by(), order_by(), and having(), which add the related SQL clauses
to the query.
filter_by(), which uses keyword arguments to query the session.
filter(), which uses more advanced SQL expressions (which we will

discuss shortly) to query.
raises an exception.
The filter_by method allows us to search for items using keyword arguments. For
example, we can say:
session.query(Pet).filter_by(name="Esme").one()
This filter_by argument tries to match a name to a specific string. This returns a
new query object, on which we call the one() method to get a single value (since
there's only one value in our example database, and it matches our criterion, it will
return that result). If we'd called all() instead, it would have returned a list of items
containing, in this case, only one item.
SQL Expression Language
Unlike filter_by, which accepts keyword arguments, the filter method accepts
values in SQLAlchemy's SQL Expression Language. This is a much more powerful
form of querying that applies different operators to column objects. It is an
interesting application of overloading the operator special methods.
[ 352 ]
Chapter 12
For example, if we use session.query(Pet).filter(Pet.name=="Esme") the

expression inside the filter query does NOT do a typical equality comparison
that evaluates to a Boolean value. Instead, it constructs a proper SQL clause that the
filter method will use to query the database. This is done by overriding the __eq__
method on the Pet.name column object. So we need to explicitly state the Pet.
name object for equality comparison. We can't specify name as if it was a keyword
argument; that would cause an error.
SQL Expression Language allows many related operators to be used to construct
queries. Some of the more common ones are:
!= to specify inequality
< for less than comparisons
> for greater than comparisons
<= for less than or equal
>= for greater than or equal
& to combine clauses using an AND query
| to combine clauses using an OR query
~ to negate a query using NOT
The SQLAlchemy Expression Language allows almost any SQL statement to be

constructed using Python, including creating joins, and aggregate clauses, and using
SQL functions. However, we have a lot of topics to cover, so you'll have to look
elsewhere to discover how to use them. Entire books have been written on SQL,
SQLAlchemy, and databases in Python, so this brief introduction can do nothing
more than spark your interest.
Pretty user interfaces
All the examples throughout this book have been run from the command line. This is
great for system administrators, Linux tinkerers, and the grandfather programmers
of our times, but it doesn't allow us to write the sort of modern desktop programs
that everyone is using these days. Indeed, some might argue that even desktop
applications are archaic, and that web applications (which we'll discuss soon)
and mobile apps are more contemporary!
[ 353 ]
The reason we haven't looked at graphical applications is that they invariably rely
on design patterns that offer such a high level of abstraction over lower-level objects
that it's hard to see the objects for the patterns. This isn't terribly useful for learning
about object-oriented programming. But now that we know the ins and outs of
the object-oriented paradigm, we can briefly discover the world of Graphical User
Interfaces, or GUIs, for short.
It would be possible to design graphical interfaces from scratch, interacting with
pixels on the screen to cause visual effects to happen. But nobody does that. Instead,
we use some kind of widget toolkit that provides us with common graphical
elements such as the buttons, textboxes, checkboxes, toolbars, tabbed interfaces,
calendars, and more that we see every day when we look at our desktop computer,
regardless of operating system.
We'll briefly (very briefly, unfortunately) discuss two of these widget toolkits that
run on Python 3. But first, let's discuss a small amount of theory. Graphical programs
invariably use an event-driven architecture. This often means they rely heavily on
the command pattern we discussed in Chapter 9. When interacting with the user, we
never know exactly when they are going to press a key, move the mouse, or click
an object, nor do we know which of these activities they are going to perform at any
time. So we write code to respond to those events only when they occur. This code
should be quick and painless so that the program can get back to waiting for the next
input from the user. This is the world of event-driven programming in a nutshell. It
can be hard to wrap your mind around at first, but it is extremely well suited to the
object-oriented principles we've been discussing all along.
TkInter
The Python standard library comes with a built-in graphical library called tkinter.
It comes preinstalled with Python on most operating systems, although it requires
the TCL/TK interpreter and graphical toolkit to be installed.
The most basic configuration of a TkInter application is to create a Frame object,
add some widget objects to that window, and then let tkinter take over in what is
called a mainloop. This mainloop is responsible for waiting for events to happen and
dispatching them to the code we write. Here's an extremely basic TkInter application,
with nothing displayed on the created window:
import tkinter
class EmptyFrame(tkinter.Frame):
pass
root = tkinter.Tk()
EmptyFrame(master=root).mainloop()
[ 354 ]
Chapter 12
First we create a class that extends the tkinter.Frame class; this is basically just a
container for other widgets. We create a Tk() object to provide a window to hold the
frame, and then call the mainloop to run with this object. If we run this program, it
displays a small, empty window. Not too exciting.
Let's look at an example we can really interact with:
import tkinter
import random
class DiceFrame(tkinter.Frame):
def __init__(self, master):
super().__init__(master)
die = tkinter.Button(self,
text = "Roll!",
command=self.roll)
die.pack()
self.roll_result = tkinter.StringVar()
label = tkinter.Label(self,
textvariable=self.roll_result)
label.pack()
self.pack()
def roll(self):
self.roll_result.set(random.randint(1, 6))
root = tkinter.Tk()
DiceFrame(master=root).mainloop()
There are a few things going on here, almost all of them in an overridden
__init__ method. After initializing the superclass, we create a new Button object,
which represents buttons like 'OK' and 'Cancel' that you have seen so often that
you've probably never thought of them as objects! All TkInter widgets take a parent
widget as their first argument. We pass self here, to make the new button a child of
the frame. We also supply a text argument, representing the string displayed on the
button, and the command argument, which is a function to be called when the button
is clicked. In this case, that function is a method on the same class. Remember that
command pattern? Here it is in action!
Then we call the pack method on our new button, which, in this basic format, simply
sets up a default size and position for the button and window. If we hadn't called
this, the button would not be visible.
[ 355 ]
Later, we use similar code to create and pack a Label object. The label is associated
with a specially constructed StringVar object, provided by TkInter. The neat thing
about this class is that whenever its value is updated to a new string, by calling the
set() method, any widgets that are associated with the object will automatically
have their displays updated to the new value. We use this feature to update the label
to a new random value every time we click the "Roll!" button. When we run this
program, we are presented with a very simple electronic die:
So graphical programming is all about constructing widgets, and connecting

commands to them to be called when certain events occur. The most complicated
part, often, is getting the display to "look right"; that is, to get all the widgets laid
out in an aesthetically pleasing manner that is also easy to use and understand. We
can customize this using the pack method. This method basically allows widgets
to be laid out in either rows or columns. If we need to use columns of rows or rows
of columns, we can pack multiple objects into separate frames (example: frames
containing rows), and then pack those frames into parent frames (example: that
are packed using columns). When packing a widget, we can pass the following
additional arguments to control how the widget will be placed in its parent:
expand: A boolean value to say whether or not to grow the widget beyond its
expected size if the parent window is resized larger. If multiple widgets have
expand set, the extra space is divided between them.
fill: Set to a string value of none, x, y, or both to instruct the widget to fill
anchor: If the widget is not set to fill its space, it can be positioned within
that space. The default, center, will ensure equal spacing on all sides. Other
values can be compass directions such as n, e, s, w, to position the item at the
top, right, bottom, or left of available space, and the values ne, se, sw, and nw
all available space assigned to it in the specified direction.
can be used to position it in one of the four corners.
ipadx and ipady: These integer values provide padding inside the widget
padx and pady: These integer values provide padding between the widget
on either the left and right or top and bottom edges. It has the effect of
increasing the size of the widget.
and the edge of its available space. It has the effect of placing space between
the widget and its neighbors.
[ 356 ]
Chapter 12
side: Use one of left, right, top, or bottom to pack the widgets along a
specific side. Normally, all widgets in a container frame are packed to the
same side; mixing them can have unanticipated effects. If you need more
than a single row or column, you can pack frames inside of other frames.
Here's an example of several of these features in action:

import tkinter
class PackFrame(tkinter.Frame):
button1 = tkinter.Button(self,
text = "expand fill")
button1.pack(expand=True, fill="both", side="left")
text = "anchor ne pady")
button2.pack(anchor="ne", pady=5, side="left")
text = "anchor se padx")
button3.pack(anchor="se", padx=5, side="left")
class TwoPackFrames(tkinter.Frame):
text="ipadx")
button1.pack(ipadx=215)
packFrame1 = PackFrame(self)
packFrame1.pack(side="bottom", anchor="e")
packFrame2 = PackFrame(self)
packFrame2.pack(side="bottom", anchor="w")
self.pack()
root = tkinter.Tk()
TwoPackFrames(master=root).mainloop()
[ 357 ]
This example creates two instances of a PackFrame that contain three buttons each,
packed horizontally (by specifying side="left"). The buttons are each positioned
differently. These frames are then vertically packed into another frame with a very
large button, anchored to the left and right sides of the frames. Here's how the
previous code would render. It's not remotely pretty, but it illustrates most of the
concepts in a single window:
When designing complicated interfaces, packing frames can get monotonous. If

you're attempting to do this, you might want to do some research into TkInter's grid
style of widget layout instead.
We don't have space to discuss grid layout or the wide array of available TkInter
widgets here, but as with SQLAlchemy, hopefully you've had a taste, know what
your options are, and are ready to dive in head-first if you need to build a GUI
application. Graphical interfaces are not more complicated than command-line
ones. They just rely on different design patterns to get their job done.
PyQt
The other major graphical toolkit supported under Python 3 is called PyQt. It is a set
of bindings to the popular cross-platform Qt library, which is available under both
commercial and open source licenses. PyQt can be downloaded from http://www.
riverbankcomputing.co.uk/software/pyqt/download. PyQt is an advanced
library that supports many features not normally considered part of a GUI toolkit.
In some ways, PyQt is a desktop application framework, with extended support for
everything from web browser widgets to databases to multimedia.
[ 358 ]
Chapter 12
However, we only have space here to look at the basics of PyQt as a graphical toolkit.
Let's start, as before, by displaying an empty window:
from PyQt4 import QtGui
app = QtGui.QApplication([])
class EmptyWidget(QtGui.QWidget):
pass
window = EmptyWidget()
window.show()
app.exec_()
This isn't terribly different from the tkinter version. We need to construct the
QApplication object before we create any windows, since constructing it initializes
the Qt internals. We then construct an empty window and call the mainloop on the
app, which is done with a call to exec_ in Qt.
Instead of a dice rolling application, we'll build a rudimentary rock/paper/
scissors window:
from PyQt4 import QtGui
import random
app = QtGui.QApplication([])
choices = ["Rock", "Paper", "Scissors"]
class RockPaperScissorsWidget(QtGui.QWidget):
def __init__(self):
super().__init__()
rock = RPSButton("Rock", self)
paper = RPSButton("Paper", self)
scissors = RPSButton("Scissors", self)
for button in (rock, paper, scissors):
button.resize(100, 100)
rock.move(0,0)
paper.move(0,100)
scissors.move(0,200)
self.response = QtGui.QLabel("", self)
self.response.setGeometry(110, 0, 200, 300)
class RPSButton(QtGui.QPushButton):
def mousePressEvent(self, event):
[ 359 ]

computer_choice = random.choice(choices)
user_choice = self.text()
comp_idx = choices.index(computer_choice)
user_idx = choices.index(user_choice)
message = {
0: 'Tied',
1: 'Computer Wins',
2: 'You Win'}[(comp_idx - user_idx + 3) % 3]
self.parent().response.setText("You chose {0}<br />"

"Computer chose {1}<br />"
"{2}".format(user_choice, computer_choice, message))
window = RockPaperScissorsWidget()
window.show()
app.exec_()
This program adds four widgets (three buttons and a results label) to a window.
The buttons are each instances of a custom subclass of QPushButton that we have
designed. The sole purpose of this inheritance is to override the mousePressEvent
method, which is called whenever the user clicks the button. After creating the
buttons, we use absolute positioning functions (resize, move, and setGeometry)
to place the widgets where we want them. Here's how they look:
[ 360 ]
Chapter 12
Inside our overridden mousePressEvent method, we randomly pick a choice for

the computer player, and then calculate the winner. The calculation is done with a
simple modulo mathematical function that evaluates to either 0, 1, or 2, depending
on who won (I had to do some experimenting to get this value right; my math skills
are lacking!). We then use this result to look up a relevant message in a dictionary.
Finally, we display the results on the label by calling the setText method. Each
time a button is pressed, the computer makes a new choice and updates the label.
PyQt supplies numerous widgets, ranging from buttons, checkboxes, and radio
buttons to text entry fields, combo boxes, and sliders to advanced widgets like
calendars, video players, and syntax highlighted text editors. Generally, if you've
seen a widget before, PyQt will supply one. If you need something that's not
available, it is possible to create your own widgets. Like TkInter, PyQt also
supplies advanced layout schemes so that positioning widgets in resizing
windows is painless. (The absolute positioning we used in our example earlier
isn't terribly resizable.)
Choosing a GUI toolkit
PyQt and TkInter are the currently available Python 3 capable GUI toolkits. In
addition, two Python 2 toolkits are extremely popular: PyGTK and wxPython. The
former is in the late stages of being ported and may be available as you read this. The
latter is a very advanced cross-platform system that has the advantage of displaying
native widgets under whichever operating system it is currently running on. This
allows wxPython programs to seamlessly "fit in" with the overall look and feel of the
operating system.
But which should you choose for a given task? It really depends on your personal
preference. You need to work with each one for a while to decide which provides
the programming paradigm you are most comfortable with. It also depends on
your specific needs. If you just want to develop a simple interface for a small script
or program, TkInter is probably your best bet, simply because it comes bundled
with Python and therefore doesn't require any extra work to install or deploy. If
you're planning on developing an intense GUI-centric application with complicated
widgets, you're probably better off using one of the other libraries. Pick one that
supplies the specific widgets you are interested in working with. PyQt is likely the
easiest to learn and work with, but if you have experience in wxWidgets or GTK
from other programming languages, you may find that wxPython or pyGTK would
be a more suitable fit.
[ 361 ]
XML
Many Python programmers consider working with XML to be a right royal pain.
However, XML is extremely popular for a wide range of purposes. Python programs
frequently have to interact with it; both as a consumer that needs to interpret XML
data from another source, and as a producer that needs to create XML data for other
programs or computers to parse.
Python includes three well-documented libraries for interacting with XML
documents in its standard library. Two are based on traditional XML parsing
techniques, while the third is a nice Pythonic interface.
The SAX (Simple API for XML) library is an event-driven system that calls specific
functions when specific objects are encountered in the byte stream: opening and
closing tags, attributes, and contents. It can be unwieldy to work with, but has the
advantage of parsing XML documents "on the fly" without having to load the entire
stream into memory. This is useful for huge documents.
The DOM (Document Object Model) library takes a different approach. It allows
any part of the document to be accessed at any time, and treats the document like
a tree of connected nodes. Each node represents an element, attribute, or text in the
document. Elements can have child elements and each can be accessed randomly.
It allows XML documents to be both read and written as well as modified by
dynamically adding new nodes to the tree.
Both of these tools have their uses, but most common XML tasks in Python can
be parsed, least painfully, using the third built-in library, ElementTree, or a
more advanced library that is based on it, called lxml. Both libraries allow XML
documents to be treated like Python objects, making them easy to read, compose,
interact with, and modify.
ElementTree
The xml.etree.ElementTree package contains several classes and functions

for manipulating XML documents. The most important of these are the Element
and ElementTree classes. An ElementTree essentially represents an entire XML
document in memory; it uses the composite pattern we discussed in Chapter 10 to
construct a tree of Element objects. It has a single pointer to a root node, which
contains relevant child nodes, which may contain more children, and so on.
[ 362 ]
Chapter 12
Any one Element object contains everything between the opening and associated
closing tags in an XML document. It references the tag name, any attributes on the
opening tag, the text inside the element, and a list of child elements (nested opening
and closing tags) recursively. If the text includes interspersed text and tag elements
(as in an HTML document), any text between a closing tag and the following tag
(whether it is a new opening tag or the parent's closing tag) is added to a tail
attribute on the element.
As an example, let's start with a simple HTML document:
<!DOCTYPE html>
<html>
<head>
<title>This is a web page</title>
<link rel="stylesheet" href="styles.css" />
</head>
<body>
<h1>Welcome To My Site</h1>
<article class="simple">
This is <em>my</em> site. It is rather lame.
But it's still <strong>mine</strong>. Mine,
I say.
</article>
</body>
</html>
If you're interested, this document is valid HTML 5 source code. If you're

familiar with XHTML or HTML 4, you'll be relieved to see that they've
made the new version much more readable. There must be some Python
programmers on the standards development team, reminding the world
that readability is important!
Now, the following program will load this HTML document into memory, and
illustrate how the various elements are associated:
from xml.etree.ElementTree import fromstring
with open("html_document.html") as file:
root = fromstring(file.read())
print("ROOT NODE")
print("\ttag:", root.tag)
print("\tnumber of children:", len(root))
[ 363 ]

print("\tchildren:", [c.tag for c in root.getchildren()])
head = root[0]
print("HEAD NODE")
print("\tfirst child:", head[0].tag)
print("\tsecond_child:", head[1].tag)
print("\tlink attributes:", head[1].attrib)
article = root[1][1]
print("ARTICLE NODE")
print("\ttag:", article.tag)
print("\ttext:", article.text)
print("\tfirst child:", article[0].tag, article[0].text)
print("\t\tem's tail:", article[0].tail)
print("\tsecond child:", article[1].tag, article[1].text)
print("\tstrong's tail:", article[1].tail)
If we run this, we can see how child list access, and tag, text, attrib, and tail
attributes work on an element class:
ROOT NODE
tag: html
number of children: 2
children: ['head', 'body']
HEAD NODE
first child: title
second_child: link
link attributes: {'href': 'styles.css', 'rel': 'stylesheet'}
ARTICLE NODE
tag: article
text:
This is
first child: em my

em's tail: site. It is rather lame.
But it's still
second child: strong mine
strong's tail: . Mine,
I say.
The important thing to note is that the child of an element is always another element
with a similar interface. The list lookup supports iteration over a node (as in for
child in element) and slice notation (as in element[1:5]), so it is easy to treat an
element like a normal Python sequence.
[ 364 ]
Chapter 12
One thing to be careful of is that ElementTree is a bit ambiguous when it comes to

checking for children. Do not use the idiom if element to determine if an element
exists or not, because it may return False if the element exists, yet contains no
children. If you want to check if an element contains children, use if len(element).
Conversely, if you want to check if an element exists, use if element is not None.
This short introduction is almost enough if we need to interpret or work with an
element tree we read from a file or was provided across the network. Typically when
we read or receive an XML file, we need to do one of two things:
Parse it node by node and convert it to some other structure
Find specific elements or attributes and look up their value
The first task can be accomplished by recursively iterating over nodes and looking
at their attributes, texts, and tails. The second task usually implies some sort of
searching mechanism. The Element class does provide a few methods to help with
matching elements. There are three of them. They each return a different value,
but all accept a single pattern parameter. This parameter supports a bastardized
version of the XPath selection language. Unfortunately, the entire XPath language is
not supported. Basic features, such as selecting a tag, selecting a tag recursively, and
building a path from the current node work, though:
print('search for child tag:', root.find('head'))
print('search children recursively:', root.findall('.//em'))
print('build path:', root.findtext('./body/article/em'))
If we run this code it outputs:

search for child tag: <Element head at 961f7ac>
search children recursively: [<Element em at 961fb2c>]
build path: my
This example also illustrates the three different search methods, each of which
accepts the same kind of path. The find method returns the first matching element.
The findall method returns a list of matching elements. The findtext method
is a bit different; it finds the first matching sub-element (just like find), but then
returns the text attribute for that element, instead of the element itself. Thus,
e.findtext(path) is identical to e.find(path).text.
[ 365 ]
Constructing XML documents
ElementTree isn't only good for parsing and searching XML documents. It also
provides an intuitive interface for constructing them, using standard list and object
access features. We can use the append function on an element to add a child, or the
helper function, SubElement, which is a little bit less typing. We can set attributes
using dictionary syntax, and text and tails using object attribute access. The
following code constructs a simple HTML document using these features:
from xml.etree.ElementTree import (Element, SubElement,
tostring)
root = Element("html")
head = Element("head")
root.append(head)
title = SubElement(head, "title")
title.text = "my page"
link = SubElement(head, "link")
link.attrib['rel'] = "stylesheet"
link.attrib['href'] = "styles.css"
body = Element("body")
body.text = "This is my website."
root.append(body)
print(tostring(root))
First we create a root element, then append some children to it in order. The
SubElement function does the same thing to append a title to the head. When we
create a link element, we also update the attribute dictionary to set attributes on it.
Finally, we can use the tostring method to convert the element to an XML string,
which looks like this:
<html><head><title>my page</title><link href="styles.css"
rel="stylesheet" /></head><body>This is my website.</body></html>
There's plenty more to ElementTree than we've considered so far, but unlike the
alternative XML libraries, the basics will get you a long ways.
lxml
lxml is an advanced XML parsing library that uses the lightning fast libxml2 library
to do the underlying hard work. It can be downloaded from the lxml website at
http://codespeak.net/lxml/. It is a third-party library and, in the past, has been
difficult to install on some operating systems, although this should not be the case
with the latest releases.
[ 366 ]
Chapter 12
If your needs are basic and can be covered by the ElementTree API we just
discussed, then, by all means, use that. But if you need to parse invalid XML
documents, advanced XPath searching, or CSS selectors, lxml is your tool.
lxml has an interface that is very similar to ElementTree, in fact, for basic usage, lxml
can be used as a drop-in replacement for ElementTree. This will invariably give your
parsing code a speedup, but this compatibility with ElementTree isn't what makes
lxml great. lxml is far more advanced and provides numerous features above and
beyond ElementTree.
The previous example for parsing and searching our XML file only needs one change
to work in lxml; change the import to read from lxml.etree import fromstring
and the code will run unmodified.
My favorite advanced feature of lxml is its support for advanced XPath and CSS
selectors for searching through XML documents. These are far more useful than the
basic ElementTree searches. Here are some examples:
print('xpath attribute:', root.xpath('//link[@href]'))
print('xpath text filter:', root.xpath('//*[contains(em, "my")]'))
print('xpath first child:', root.xpath('/html/body/article/em[1]'))
from lxml.cssselect import CSSSelector
print('css class selector:', CSSSelector('.simple')(root))
print('css tag selector:', CSSSelector('em')(root))
lxml will support any XPath selector that is supported by the underlying libxml2
library. This basically encompasses the entire XPath language definition, although
some of the most exotic selectors may be buggy.
The CSS selectors are very comfortable to anyone used to the jQuery JavaScript
library or similar libraries. CSS selectors compile, internally, to equivalent XPath
selectors before the selection is run. Both the XPath and the CSS selector functions
return a list of all matching elements, similar to the ElementTree findall method.
In addition to these advanced search features, lxml provides:
A parser for badly formed HTML
A unique library for treating elements like objects, so you can access sub-tags
as if they were attributes on objects
A complete XML validation tool that can utilize DTDs, XMLSchema, and
RELAX NG schemas
We don't have space to discuss these, but if you have any advanced or complicated
requirements when it comes to XML or HTML parsing, lxml is invariably the tool
you want to reach for.
[ 367 ]
CherryPy
CherryPy version 3.2 is the first major web application server to be made available
on the Python 3 platform. It can be downloaded from http://cherrypy.org/. It
is not a full-stack framework like the very popular Django, TurboGears, or Zope
libraries. These frameworks provide extra support for data storage, templating,
authentication, and other common web operations. Such features are not impossible
in CherryPy, you're just responsible for finding or implementing them yourself.
CherryPy is a very powerful web server that uses a simple design for building web
applications. Let's jump in head-first with a simple example that serves the HTML
file we developed in the previous section:
import cherrypy
class SimplePage:
@cherrypy.expose
def index(self):
with open("html_document.html") as file:
return file.read()
cherrypy.quickstart(SimplePage())
If we run this program, we can visit http://localhost:8080/ in a web browser

to see the web page in action. All we've done here is create a class to pass to the
quickstart function. That function starts a web server and serves pages from that
class. Any methods of the class we created that have been marked as exposed will be
made available via HTTP at a URL with the same name as the method. Any method
not explicitly marked as exposed can be used internally as a helper method, but
cannot be accessed at any URL.
The method itself simply opens a file and returns the contents of that file. Ultimately,
we've written a web application that serves a single HTML 5 web page.
Of course, having a site that serves only one web page is pretty boring. Let's look at
an example that is just a touch more exciting:
import cherrypy
template = """<!DOCTYPE html>
<html>
<body>
{message}
</body>
</html>"""
[ 368 ]
Chapter 12
class AboutPage:
@cherrypy.expose
def index(self):
return template.format(message="""
I'm not a very interesting person...""")
@cherrypy.expose
def contactPage(self):
print(self)
return template.format(
message="I can't be contacted anywhere.")
class MainPage:
about = AboutPage()
contact = contactPage
@cherrypy.expose
def index(self):
return template.format(message="""
This is the main page.
<a href="/about/">About Me</a>
<a href="/contact/">Contact Me</a>
<a href="/links/">Some Links</a>
""")
@cherrypy.expose
def links(self):
return template.format(
message="No Links Yet")
cherrypy.quickstart(MainPage())
This example shows three ways that pages can be added to a site. The obvious one is
to add an exposed method, such as the links method above, to the class. But we can
add exposed objects in other ways too:
By defining a separate function and including the attribute in the class

definition as we did with contactPage
By defining a separate class and including an instance of it in the class

definition, as we did with aboutPage
By adding the exposed method to the object after the class has been
instantiated using code such as app.some_page = AnExposedClass()
[ 369 ]
You've probably figured out already that the index method is a special method. It
doesn't map to the /index URL; instead, it is the method called if no path is added
after the ending slash.
We can also accept HTML form arguments. Let's create a real contact page:
import cherrypy
class ContactPage:
@cherrypy.expose
def index(self, message=None):
if message:
print("The user submitted:\n{0}".format(
message))
return "Thank you!"
return """<form>
<textarea name="message"></textarea>
<input type="submit" />
</form>"""
cherrypy.quickstart(ContactPage())
This page displays a different result depending on the presence of a message

variable in the keyword arguments. If no such argument is supplied, the visitor is
presented with a form to enter a message in. If the argument is supplied, the value
of the message is printed to the console (normally we'd do something useful with
the value, such as e-mailing it somewhere or storing it in a database or file for later
retrieval). Then a thank you message is returned to the client.
So how did that message parameter get set? Basically, any named inputs in a form
(in this case, the message textarea) are mapped to keyword arguments when the
page is submitted. It's that simple!
A full web stack?
As we discussed, CherryPy is just a web application server; it is not a web

framework. It provides a complete web server and the basic features to map HTTP
requests to code that must be executed when those requests are made. It also
provides, with a bit of configuration, complete SSL support, the ability to set and
retrieve cookies, caching support, HTTP authentication, and sessions. However,
it is missing two key features that many other frameworks supply: templating
and data storage.
[ 370 ]
Chapter 12
Many websites use databases for data storage, but CherryPy does not supply
this ability. Do we really need it to? We really just need database connectivity; it
doesn't have to be built into the web framework. Indeed, why don't we just use
SQLAlchemy, which we discussed earlier in the chapter? In fact, this is what the
TurboGears framework uses for its database access.
This then, still leaves us to solve the templating problem, another framework feature
that CherryPy lacks. Templating is the process of taking static strings or files, and
replacing certain substrings in those files with new strings, based on some kind of
context. The str.format function we covered in Chapter 10 is a basic example of
templating. It allows us to replace modifiers with variables passed into the function.
Indeed, this was the template method we used in the example of a simple CherryPy
application earlier.
Most template languages go beyond this ability to allow things like conditionals
(including data in the template only if a certain condition is met, such as two
variables being equal, or a user being logged in), and looping (including data in a
template repeatedly, such as creating a table or unordered list containing multiple
items from a Python list). Some even go so far as to allow arbitrary Python code to be
executed within the template.
There are a myriad opinions on what a template language should be, which is
why, for Python 2, there have been an immeasurable number of different template
languages devised. This diversity hasn't spread to Python 3 yet, but one of the most
powerful templating languages, Jinja2 is already available on the Python 3 platform.
It can be downloaded from http://jinja.pocoo.org/.
As a sort of case study, let's take these three toolsCherryPy, SQLAlchemy, and
Jinjaand create a quick and dirty blogging engine! We'll start with the SQLAlchemy
models; these define the data that will be stored in the database:
import datetime
import sqlalchemy as sqa
from sqlalchemy.ext.declarative import declarative_base
Base = declarative_base()
class Article(Base):
__tablename__ = "article"
rowid = sqa.Column(sqa.Integer, primary_key=True)
title = sqa.Column(sqa.String)
message = sqa.Column(sqa.String)
pub_date = sqa.Column(sqa.DateTime)
def __init__(self, title, message):
[ 371 ]

self.title = title
self.message = message
self.pub_date=datetime.datetime.now()
class Comment(Base):
__tablename__ = "comment"
rowid = sqa.Column(sqa.Integer, primary_key=True)
article_id = sqa.Column(sqa.Integer,
sqa.ForeignKey('article.rowid'))
article = sqa.orm.relationship(Article, backref="comments")
name = sqa.Column(sqa.String)
message = sqa.Column(sqa.String)
def __init__(self, article_id, name, message):
self.article_id = article_id
self.name = name
self.message = message
engine = sqa.create_engine('sqlite:///blog.db')
Base.metadata.create_all(engine)
Session = sqa.orm.sessionmaker(bind=engine)
We create two models with some fields. The two models are associated with a
ForeignKey relationship on the Comment class.
The rowid field is a special one; in SQLite, every model is automatically given a
unique integer rowid. We don't have to do anything to populate this number, it's
simply available from the database. This wouldn't work with PostgreSQL or another
engine; we'd have to set up a sequence or autoincrement field instead.
We add an __init__ method to each class to make it easier to construct new
instances. Then we associate the engine, create the tables, and create a Session class
to interact with the database later.
Jinja Templating
Now, we can set up Jinja to serve some templates from a folder for us:
import jinja2
templates = jinja2.Environment(loader=jinja2.FileSystemLoader(
'blog_templates'))
[ 372 ]
Chapter 12
Well that was easy. This gives us a templates variable that we can use to load
templates based on filename from the given folder. Before we create the CherryPy
app server, let's have a look at the templates. Let's scrutinize the simple template for
adding a blog article first:
{% extends "base.html" %}
{% block title %}New Entry{% endblock %}
{% block content %}
<form method="POST" action="/process_add/">
Title: <input name="title" type="text" size="40" /><br />
<textarea name="message" rows="10" cols="40">
</textarea><br />
<input type="submit" value="Publish" />
</form>
{% endblock %}
This sort of resembles normal HTML, but all those {% things are new. That's Jinja
markup (it's also very similar to Django markup, if you have used or are interested in
using Django's template system) for a template tag. Template tags are instructions to
the templating system to do something special here. There are two types of template
tags in use: extends, and block. The extends tag essentially tells the template
system to start with the base.html, but replace any named blocks with the named
blocks in this file. And that's what the block and endblock tags are; named blocks to
override whatever is specified in the parent template, base.html. This may be a bit
clearer if we know what base.html looks like:
<!DOCTYPE html>
<html>
<head><title>{% block title %}{% endblock %}</title></head>
<body>
<h1>My Blog</h1>
<ul>
<li><a href="/">Main</a></li>
<li><a href="/add/">Add Entry</a></li>
</ul>
<hr />
{% block content %}
{% endblock %}
<body>
<html>
This looks even more like a normal HTML page; it shows where the two named
blocks should go in the context of a larger page.
[ 373 ]
Extending base.html in other templates allows us to ignore the parts of every page
that stay the same. Further, if we want to add a link to the menu or otherwise modify
the overall site, we only have to do it in this single template file.
The other template, index.html is substantially more complex:
{% extends "base.html" %}
{% block title %}My Blog{% endblock %}
{% block content %}
{% for article in articles %}
<h2>{{article.title}}</h2>
<em>{{article.pub_date.strftime('%b %d %Y')}}</em>
<p>{{article.message}}</p>
<div style="margin-left: 6em">
<h3>Comments</h3>
{% for comment in article.comments %}
<em>{{comment.name}} wrote:</em>
<p>
{{comment.message}}
</p>
{% endfor %}
{% include "comment_form.html" %}
</div>
<hr />
{% endfor %}
{% endblock %}
It includes the same extends and block tags as the earlier template. In addition,
it introduces us to the for template tag, which loops over all the articles (or all the
comments in an article) and renders slightly different HTML for each of them. It also
renders a bunch of variables using the {{<variable_name>}} syntax. The variable
names are passed into the template from our CherryPy application or are assigned
within the context, as is done inside the for loops.
The rendering of the pub_date variable on the article is particularly interesting, as
the item is a datetime.datetime object, and we can see that Jinja allows us to call
the strftime method on this object directly.
[ 374 ]
Chapter 12
Finally, the include tag allows us to render part of the template in a separate file,
comment_form.html, which looks like this:
<form method="POST"
action="/process_comment/{{article.rowid}}/">
Name: <input name="name" type="text" size="30" /><br />
<textarea name="message" rows="5" cols="30">
</textarea><br />
<input type="submit" value="Comment" />
</form>
That's basic Jinja syntax in a nutshell; there's a lot more that can be done with it, of
course, but these basics are enough to get you interested. They're also enough for our
simple blog engine!
The CherryPy blog web application
In the interest of understanding how web applications are designed, note that I
didn't write those templates before I wrote the CherryPy application we're about to
see. Instead, I developed iteratively, creating both the code and templates to add an
article, followed by the code and templates to display an article, and finally, setting
up the comment system. I grouped all the resulting templates together in the above
section so we could focus on Jinja template syntax. Now, let's focus on CherryPy and
how those templates are called!
First, here's our blog engine with the index method:
import cherrypy
class Blog:
@cherrypy.expose
def index(self):
session = Session()
articles = session.query(Article).all()
template = templates.get_template("index.html")
content = template.render(articles=articles)
session.close()
return content
cherrypy.quickstart(Blog())
Here's where we start to see our three puzzle pieces merging together. CherryPy,
of course, is serving the page. Jinja is creating the page using our templates. And
SQLAlchemy is giving Jinja the data it needs to display
[ 375 ]
First we construct a session and use it to search for all the available articles. We
then get a template by name; this comes from the templates object we set up earlier
in the module. Then we render the template, passing one keyword argument into it.
Keyword arguments map to variables inside the template context; the template we
defined earlier will loop over the articles we passed in this function. Then we return
the rendered content to let CherryPy display it.
The code to display the form for adding a new article is even simpler; we just render
the template, since it doesn't need any variables:
@cherrypy.expose
def add(self):
template = templates.get_template("add.html")
return template.render()
You may have noticed in our templates that the forms for adding articles and
comments have action attributes pointing to process_add and process_comment
URLs. The process_add URL simply constructs a new article from the form
parameters (title and name), which come to us from CherryPy as keyword
arguments. Then it raises an exception to redirect the client to the main view,
which will display the new article:
@cherrypy.expose
def process_add(self, title=None, message=None):
session = Session()
article = Article(title, message)
session.add(article)
session.commit()
session.close()
raise cherrypy.HTTPRedirect("/")
The process_comment method is much the same, except it also accepts a positional
argument. Positional arguments come between forward slash characters in the URL,
so the following method signature would actually map to /process_comment/3/ if
an article_id of 3 is passed:
@cherrypy.expose
def process_comment(self, article_id, name=None,
message=None):
session = Session()
comment = Comment(article_id, name, message)
session.add(comment)
session.commit()
session.close()
raise cherrypy.HTTPRedirect("/")
[ 376 ]
Chapter 12
And there we have it, a complete, simple blog engine with absolutely no
authentication and which will fill up with spam in a matter of minutes.
But it works! And we wrote it all using Python 3 objects.
Exercises
We covered a wide variety of material in this chapter, but we didn't cover anything
in a lot of detail. These exercises, then, will be all about extra reading. All of the tools
we discussed in this chapter have terrific documentation on their websites, including
tutorials, API references, and specific examples. If any of the topics we discussed are
of special interest to you, review the documentation for those libraries. Try them out,
see how far you can push them.
Acquire the knowledge you need to write a complicated GUI program, then write
it. Do it in both TkInter and PyQt, and decide which toolkit you prefer. Find out
what database backends SQLAlchemy currently supports under Python 3, as this
number will grow quickly from the two available at the time of writing. Similarly,
do research into available web frameworks, and see how they compare to CherryPy.
Is Jinja the best tool for templating available? Try some other products and see what
you think of them. Or if you have some free time and would like a challenge, take
some of the string manipulation knowledge we gained in Chapter 10 and write your
own templating engine!
Try out ElementTree and lxml and see if you can uncover their similarities and
differences. What if you were to merge lxml into the web stack we created above and
use it to create the HTML documents instead of rendering them using templates?
(hint: This is a bad idea. But try it anyway!)
There are hundreds of other Python libraries and APIs available out there, and
more and more of them are becoming available for Python 3 every day. If you have
a specific problem you need to solve, the chances are that there is a support library
available that will help you solve it. We touched on some popular ones in the
examples throughout this book and most especially in this chapter. But there is so
much more out there. For example, we didn't discuss scientific or display packages at
all. Do some research into available libraries, find out what's available. You'll never
know when it might be useful!
[ 377 ]
Summary
The variety of topics related in this chapter is quite staggering. We started with
databases, graduated to Graphical User Interfaces, diverged into a discussion of
XML, and ultimately built a small web application. The goal was to introduce
popular, available libraries for major real-world tasks. The number of libraries
available for Python 3 is steadily growing as more and more developers choose to
support the cleaner syntax used in this version of the Python language. We have
seen an overview of:
SQLAlchemy for databases
TkInter for graphical interfaces
PyQt for different graphical interfaces
ElementTree for XML parsing
lxml for better XML parsing
CherryPy for web applications
Jinja for string templating in web applications
Thus ends our tour through the world of Object-oriented Programming in Python.
I sincerely hope you enjoyed the ride, and are excited to test your new skills on
innovative new programming problems. Thanks for your attention, and please
watch your step as you leave the vehicle.
[ 378 ]
Index
Symbols
__call__ method 270
__getattribute__ method 192
__iter__ method 181
__new__ class method 248
__new__ method 181
__next__ method 197
__repr__ method 171
__str__ method 242
_update_observers method 236
A
absolute imports 46
abstract factory pattern
about 271
examples 272
implementing 273-275
UML class diagram 272
abstraction 16
access control 50, 51
adapter pattern
about 257, 259
benefits 258
structure 258
UML diagram 258
add_child method 279
add_point method 127
Agent class 90
aggregation 19
all() method 352
API SQLite 348
append() method 170
append method 98
arguments, for controlling widget

anchor 356
expand 356
fill 356
ipadx 356
ipady 356
padx 356
pady 356
side 357
assertDictEqual method 320
assertEqual method 318
assertFalse method 318
assert function 38
assertGreaterEqual method 319
assertGreater method 319
assertIn method 319
assertion methods, Python 3.1
about 319
assertDictEqual 320
assertGreater 319
assertGreaterEqual 319
assertIn 319
assertIsNone 319
assertIsNotNone 319
assertLess 319
assertLessEqual 319
assertListEqual 320
assertMultilineEqual 320
assertNotIn 319
assertRegexpMatches 320
assertSameElements 319
assertSequenceEqual 320
assertSetEqual 320
assertTupleEqual 320
assertion methods, unit testing

about 318
assertEqual 318
assertFalse 318
assertNotEqual 318
assertRaises 318
assertTrue 318
assertIsNone method 319
assertIsNotNone method 319
assertLessEqual method 319
assertLess method 319
assertListEqual method 320
assertMultilineEqual method 320
assertNotEqual method 318
assertNotIn method 319
assertRaises method 318
assertRegexpMatches method 320
assertSameElements method 319
assertSequenceEqual method 320
assertSetEqual method 320
assertTrue method 318
assertTupleEqual method 320
association 9
attributes 11, 12
AudioFile object 78
authentication 113
authentication and authorization system,
case study
about 112
designing 112-122
authorization 113
B
Base class 350
BaseException 95
basic inheritance
about 63
built-ins, extending 66
example 63-65
overriding 67, 68
super() 68
behaviors
about 7
adding to class data, properties used
129-131
boolean convenience methods

isalnum 284
isalpha 284
isdecimal 284
isdigit 284
isidentifier 285
islower 285
isnumeric 284
isprintable 285
isspace 285
istitle 285
isupper 285
bottom-up design 61
built-in functions, Python
about 191
all function 197
any function 197
compile function 197
delattr function 197
enumerate function 193
eval function 197
exec function 197
getattr function 197
hasattr function 197
len() function 192
reversed() function 192
setattr function 197
sorted() function 196
true function 197
zip function 194
bytearray type 298
BytesIO 302
bytes mode 300
C
calculate_distance method 38
call_after method 215
call_me method 73
callable attributes 218
callable objects 219
callback functions
about 213-217
canonical example, strategy pattern 238, 240
capitalize method 286
[ 380 ]
case study
authentication and authorization system
112
cryptography application, writing 339
Document class, modeling 147
mailing list manager 220
notebook application 53-61
simple link collector 182
simple real estate application 80
CHARACTERS list 170
CherryPy
about 368
example 368
features 368-371
CherryPy blog web application
about 375
designing 375-377
ChildNode state 244
cipher
working 339
class 9
class diagram
about 9
association 9
multiplicity 11
cleanup function 333
close method 300
code coverage 336
collect_links method 184
command-line arguments, py.test
--exitfirst argument 335
--looponfail argument 335
--pdb argument 335
-f argument 335
-k argument 336
py.test --help 335
py.test output 335
to py.test forces 335
command object 267
command pattern
about 267
example 267
UML diagram, 267
Component class 279
composite pattern
about 276
UML diagram 276
composition
about 17
example 17, 18
composition relationship 20
comprehensions
about 197
dictionary comprehensions 200, 201
generator expressions 201
list comprehensions 198
set comprehensions 200, 201
config attribute 330
connect function 349
connection object 349
constructor 39
Contact class 65
ContactList class 66
count() method 170
count method 285
coverage.py 336
coverage report command 337
coverage testing
about 336
coverage.py 336
figleaf 336
create_engine function 351
cryptography application, case study
cipher, working 339
writing 339, 340
CurrencyFormatter class 273
current_node 243
Cursor class 150
D
data 7
database access 348
database module 46
database object 47
DateFormatter class 273
datetime objects 258
DBAPI2 348
DebuggingServer class 220
declarative_base 350
[ 381 ]
decorator pattern
about 229
in Python 233, 234
in UML 229
network programming example 230-232
uses 229
decorator pattern, in Python
about 233
logging example 233, 234
decorators
about 134
applying 134
default arguments 207
defaultdict
using 166, 167
defaultdict constructor 167
default method 310
delay parameter 216
delegation method 139
design patterns
about 227, 228
abstract factory pattern 271
adapter pattern 257
command pattern 267
composite pattern 276
decorator pattern 229
facade pattern 260
flyweight pattern 263
observer pattern 235
singleton pattern 247
state pattern 240
strategy pattern 237
template pattern 251
diamond problem, multiple inheritance
about 71
example 71-75
dict() constructor 162
dict.update method 85
dict class 67
dictionaries
about 162
defaultdict, using 166, 167
stock application example 162-165
uses 166
dictionary comprehensions 200, 201
DictSorted object 181
difference method 176
discover module 322

distance method 126
docstrings 41
Document class, case study
modeling 147-154
DOM (Document Object Model) library 362
done() method 228
dot notation 35
duck typing 22
dumps function 304
E
Element object 363
ElementTree
about 362, 365
example 363, 364
XML documents, constructing 366
encapsulation 15
endswith method 285
enumerate function 193, 194
exceptional circumstances 109
exception hierarchy
about 106, 108
custom exceptions, defining 108
KeyboardInterrupt exception 107
SystemExit 107
exceptions
about 95
case study 112
handling 101-105
raising 95-100
using, for exceptional circumstances
109-[111
execute() method 270, 349
F
facade pattern
about 260, 262
structure 261
UML diagram 260, 261
figleaf 336
file IO
about 299, 300
files, faking 302, 303
placing, in context 301, 302
[ 382 ]
filename property 143

filter() method 352
filter_by method 352
find_replace method 138
findall method 367
find method 285, 365
findtext method 365
first() method 352
FirstTag state 245
Flyweight class 264
flyweight pattern
about 263
UML diagram 264
foo function 134
foo method 134
format() method 287
format_string function 50
format_time function 217
format method 217
FormatterFactory class 273
Friend class 71
fruit farming example 12
full_name property 309
funcargs 329
G
generator expressions 201, 202
generator syntax 203-205
get(primary_key) method 352
get_path function 279
get_valid_input method 91
get method 163
group_by() method 352
GUI toolkit
PyGTK 361
selecting 361
wxPython 361
H
hashable object 173
having() method 352
help function 179
House class 86
I
imaplib standard library 262
information hiding 15
inheritance
about 20
abstraction 22
basic inheritance 63
example 21, 22
multiple inheritance 23, 68
inheritance diagram 71
initialize_database function 48
insert() method 170
intersection method 175
Invoker objects 267
isalnum method 284
isalpha method 284
isdecimal method 284
isdigit method 284
isidentifier method 285
islower method 285
isnumeric method 284
isprintable method 285
isspace method 285
issubset method 176
issuperset method 176
istitle method 285
isupper method 285
iterative development model 8
iterator pattern 228
J
JavaScript Object Notation (JSON) 308
Jinja Templating 372-374
join function 151
JSONEncoder class 309
JSON serializer 308
K
KeyboardInterrupt exception 107
L
lambda function 197
len() function 192
links method 369
[ 383 ]
list.sort method 238

list comprehensions 198-200
lists
about 168-170
sorting 171, 172
loads function 304
lower method 286
lxm 366
lxml
features 366, 367
M
mailing list manager, case study
about 220
building 220-223
make_background method 240
menu_item class 270
message variable 370
method overloading 205
Method Resolution Order 72
methods 13
mixin 68
modules
about 43, 44
absolute imports 46
importing 45
organizing 45
relative imports 47-50
monkey-patch 69
mousePressEvent method 361
move method 279
multiple inheritance
about 23, 68
arguments lists, formatting 75, 77
diamond problem 71
example 70
mixin 68
options 69
working 70
multiplicity 11
mutable byte strings 297, 298
N
name attribute 131
namedtuple constructor 161
named tuples
about 161
creating 161
network programming example, decorator
pattern 230
new card catalog program, case study
about 24-30
UML diagram 26
UML sequence diagram 27
next() method 228
no_return function 100
Node objects 242
normalize_url method 184
notebook application, case study 53-61
NotImplementedError 254
O
object
about 7, 9
attributes 11, 12
behaviors 13
class diagram 9
classes relationship, describing 9
data 11
identifying 125-128
kinds 9
managing 137
storing 303
object-oriented 7
object-oriented Analysis (OOA) 8
object-oriented Design (OOD) 8
Object-Relational Managers(ORMs)
about 349
SQLAlchemy 349
object management
about 137
composition-based solution 145, 146
delegation method 139
duplicate code, removing 140, 141
example 138
existing code, reusing 142-144
extensibility 138
files, unzipping 139
files, zipping 139
partitioning 139
readability 138
[ 384 ]
object relations
association 9
composition 17
inheritance 20
objects, storing
about 303, 305
pickles, customizing 305, 307
web objects, serializing 308-310
observer pattern
about 235
example 235-237
in UML 235
one() method 352
open() function 299
OpenTag state 245
order_by() method 352
OrderedDict object 182
order method 65
P
p.reset() method 36
package 45
pack method 355
parameters 13
parent variable 279
Parser class 243
partition method 286
pattern parameter 365
paypal module 47
perimeter function 126
phone attribute 68
pickle module 303
plain_text attribute 51
play() method 78
Point class
creating 35
reset method 36
polymorphism
about 22, 78, 79
inheritance, using 78, 79
print method 218
process_comment method 376
process_files method 143
process_format method 254
process_zip function 146
process method 244
processor object 146

Product class 46
products module 46
prompt_init static method 85
properties
creating, decorators used 134, 135
using 135-137
property constructor 132
property keyword 131
property object
about 132
working 132, 133
public interface
creating 14-17
Purchase class 87
py.test
about 324
command-line arguments 335
running 325, 326
py.test.skip function 333
pygame library 144
PyGTK 361
PyQt
about 358
download link 358
features 359-361
Python
access control 50, 51
built-in data structures 157
built-in functions 191
case study 24
coverage testing 336
database access 348
default arguments 207
design patterns 227
docstrings 41
duck typing 22
objects, identifying 125
pickle module 303
strings 283
test, need for 313
test driven development 315
unit testing 316
unittest module 324
unpacking arguments 212, 213
variable argument lists 208-212
[ 385 ]
Python classes
attributes, adding 35
creating 33, 34
implementing 35-38
object, initializing 38-40
Python data structures
about 157
built-ins, extending 177-181
dictionaries 162
lists 168
named tuples 161
objects 157, 158
sets 173
tuples 159, 160
tuples, creating 159
PYTHONPATH 46
Q
Query object 352
quickstart function 368
R
readline method 300
read method 300
relative imports 47-50
Rental class 87
Repeater class 217
replace method 286
request.addfinalizer method 331
request.cached_setup method 331
request object 330
reset method 36
reverse() method 170
reversed() function 192, 193
root node 243
run method 216
S
SAX (Simple API for XML) library 362
self parameter 36
send_email method 262
send_mail function 47, 289
session.query() method 352
sessionmaker function 352
set() constructor 174
set() method 356

set_name() method 130
set comprehensions 200, 201
setdefault method 163
sets 173-176
setter attribute 134
setText method 361
setup function 333
setUp method 322
Silly class 135
simple link collector, case study
writing 182-188
simple real estate application, case study
about 80
designing 80-91
singleton pattern
about 247
implementing 248
mimicking, module variables used 249, 250
skip method 323
smtplib module 69
sort() method 170
sorted() function 196, 238
split method 260
SQLAlchemy
about 349
features 349, 350
objects, adding 351, 352
objects, querying 351, 352
SQL Expression Language 352
sqlalchemy.create_engine function 350
SQLAlchemy Expression Language 353
SQL Expression Language 352
SQLite 348
SQLite 3 348
startswith method 285
state pattern
about 240
differences, with strategy pattern 247
example 241-245
in UML 241
str() function 152
str.format method 287
str.lower function 172
strategy pattern
about 237
canonical example 238, 240
[ 386 ]
in Python 240
in UML 238
strategy pattern, in Python 240
str class 284
string.ascii_letters attribute 170
string formatting
about 287
brace characters, escaping 288
container lookups 289, 290
keyword arguments 288, 289
object lookups 291
variables, making look right 291-293
StringIO 302
string manipulation
about 284
examples 284
strings
about 283
mutable byte strings 297
string formatting 287
string manipulation 284
Unicode characters 294
StringVar object 356
strip() 245
subclass 64
Subclass object 73
SubElement function 366
super() 68
superclass 64
supplier class 65
symmetric_difference method 175
SyntaxError exception 96
sys.exit function 107
SystemExit exception 107
T
tearDown method 322
template pattern
about 251
example 252-255
in UML 251
test
need for 313, 314
TestCase 322
test driven development 315
testing, with py.test

about 324, 325
setup_class method 327
setup_method 327
setup_module method 327
teardown_class method 327
teardown_method 327
teardown_module method 327
test, skipping 333, 334
unittest example 325, 326
variables, setting up 329-333
TestSuites 322
TextNode state 245
TimedEvent class 215
Timer class 216
title method 286
Tk() object 355
TkInter
about 354
basic configuration 354
features 354-358
tkinter.Frame class 355
top-down design 61
translate method 286
tuples
about 159
creating 159
functions 159
unpacking 160
uses 160
TurboGears framework 371
U
undo method 271
Unicode
about 294
bytes, converting to text 295
text, converting to bytes 296, 297
Unified Modeling Language (UML) 9, 10
union method 175-177
unit testing
about 316, 317
assertion methods 318
broken tests, ignoring 323, 324
situations, testing 321, 322
[ 387 ]
tests, organizing 322

tests, running 322
unittest library 319
unittest module 324
unit tests 316
unpacking arguments 212, 213
unzip method 138
update() method 235
upper method 286
urlparse function 184
Window.exit method 270

Window class 270
writelines method 300
write method 300
wxPython 361
values 13
varargs 209
variable argument lists 208-212
zipfile module 139

zip function 194, 195
zip method 138
zipprocessor object 146
ZipReplace class 142
X
XML 362
XML parsing techniques 362
weakref module 264

WeakValueDictionary object 264
web objects
serializing 308-310
WebPage class 137
[ 388 ]
Thank you for buying
Python 3 Object Oriented Programming

About Packt Publishing
Packt, pronounced 'packed', published its first book "Mastering phpMyAdmin for Effective
MySQL Management" in April 2004 and subsequently continued to specialize in publishing
highly focused books on specific technologies and solutions.
Our books and publications share the experiences of your fellow IT professionals in adapting
and customizing today's systems, applications, and frameworks. Our solution based books
give you the knowledge and power to customize the software and technologies you're using
to get the job done. Packt books are more specific and less general than the IT books you have
seen in the past. Our unique business model allows us to bring you more focused information,
giving you more of what you need to know, and less of what you don't.
Packt is a modern, yet unique publishing company, which focuses on producing quality,
cutting-edge books for communities of developers, administrators, and newbies alike. For
more information, please visit our website: www.packtpub.com.
About Packt Open Source
In 2010, Packt launched two new brands, Packt Open Source and Packt Enterprise, in order to
continue its focus on specialization. This book is part of the Packt Open Source brand, home
to books published on software built around Open Source licences, and offering information
to anybody from advanced developers to budding web designers. The Open Source brand
also runs Packt's Open Source Royalty Scheme, by which Packt gives a royalty to each Open
Source project about whose software a book is sold.
Writing for Packt
We welcome all inquiries from people who are interested in authoring. Book proposals
should be sent to [email protected]. If your book idea is still at an early stage and you
would like to discuss it first before writing a formal book proposal, contact us; one of our
commissioning editors will get in touch with you.
We're not just looking for published authors; if you have strong technical skills but no writing
experience, our experienced editors can help you develop a writing career, or simply get some
additional reward for your expertise.
Expert Python Programming

ISBN: 978-1-847194-94-7
Paperback: 372 pages
Best practices for designing, coding, and distributing

your Python software
1.
Learn Python development best practices from

an expert, with detailed coverage of naming
and coding conventions
2.
Apply object-oriented principles, design

patterns, and advanced syntax tricks
3.
Manage your code with distributed version

control
4.
Profile and optimize your code
Python Testing: Beginners Guide

ISBN: 978-1-847198-84-6
An easy and convenient approach to testing your

powerful Python projects
1.
Covers everything you need to test your code

in Python
2.
Easiest and enjoyable approach to learn

Python testing
3.
Write, execute, and understand the result of

tests in the unit test framework
4.
Packed with step-by-step examples and

clear explanations
Please check www.PacktPub.com for information on our titles
Spring Python 1.1

ISBN: 978-1-849510-66-0
Create powerful and versatile Spring Python

applications using pragmatic libraries and
useful abstractions
1.
Maximize the use of Spring features in

Python and develop impressive Spring
Python applications
2.
Explore the versatility of Spring Python

by integrating it with frameworks, libraries,
and tools
3.
Discover the non-intrusive Spring way of

wiring together Python components
4.
Packed with hands-on-examples, case studies,

and clear explanations for better understanding
Matplotlib for Python Developers

ISBN: 978-1-847197-90-0
Build remarkable publication-quality plots the

easy way
1.
Create high quality 2D plots by using

Matplotlib productively
2.
Incremental introduction to Matplotlib, from

the ground up to advanced levels
3.
Embed Matplotlib in GTK+, Qt, and wxWidgets

applications as well as web sites to utilize them
in Python applications
4.
Deploy Matplotlib in web applications and

expose it on the Web using popular web
frameworks such as Pylons and Django
Please check www.PacktPub.com for information on our titles

Python 3 Orientado A Objeto

Uploaded by

Copyright:

Available Formats

Python 3 Orientado A Objeto

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Python 3 Orientado A Objeto

Uploaded by

Copyright:

Available Formats

Python 3 Object Oriented

Harness the power of Python 3 objects

Python 3 Object Oriented Programming

First published: July 2010

Production Reference: 1160710

Published by Packt Publishing Ltd.

Cover Image by Asher Wishkerman ([email protected] )

Editorial Team Leader

About the Author

About the Reviewers

Dan McGee is a software developer currently living in Chicago, Illinois. He has

Lawrence Oluyede is a 26 years old software development expert in Python and

Inheritance provides abstraction

Chapter 2: Objects in Python

Who can access my data?

Chapter 3: When Objects are Alike

Chapter 4: Expecting the Unexpected

Chapter 5: When to Use Object-oriented Programming

Chapter 6: Python Data Structures

Treat objects as objects

When should we use dictionaries?

Chapter 7: Python Object-oriented Shortcuts

Chapter 8: Python Design Patterns I

Python built-in functions

Chapter 9: Python Design Patterns II

Chapter 10: Files and Strings

Strings are Unicode

Mutable byte strings

Converting bytes to text

Chapter 11: Testing Object-oriented Programs

Additional assertion methods in Python 3.1

Reducing boilerplate and cleaning up

Chapter 12: Common Python 3 Libraries

Pretty user interfaces

Constructing XML documents

A full web stack?

What this book covers

Chapter 1, Object-oriented Design covers important object-oriented concepts. It deals

What you need for this book

Python version 3.0 or higher

Who this book is for

If you're new to object-oriented programming techniques, or if you have basic

A block of code is set as follows:

Any command-line input or output is written as follows:

Tips and tricks appear like this.

You can contact us at [email protected] if you are having a problem with

What object-oriented means

The difference between object-oriented design and object-oriented

The basic principles of object-oriented design

Basic Unified Modeling Language and when it isn't evil

review our history

Object-oriented Design (OOD) is the process of converting such requirements into

Objects and classes

However, some programmers think UML is a waste of time. Citing iterative

Specifying attributes and behaviors

We now have a grasp on some basic object-oriented terminology. Objects are

Data describes objects

Behaviors are actions

Adding models and methods to individual objects allows us to create a system of