Basic Lisp Techniques
David J. Cooper, Jr.
March 19, 2003

ii

0

Copyright c 2003, Franz Inc. and David J. Cooper, Jr.

Foreword1
Computers, and the software applications that power them, permeate every facet of our
daily lives. From groceries to airline reservations to dental appointments, our reliance
on technology is all-encompassing. And, it’s not enough. Every day, our expectations of
technology and software increase:
• smart appliances that can be controlled via the internet
• better search engines that generate information we actually want
• voice-activated laptops
• cars that know exactly where to go
The list is endless. Unfortunately, there is not an endless supply of programmers and
developers to satisfy our insatiable appetites for new features and gadgets. Every day,
hundreds of magazine and on-line articles focus on the time and people resources needed to
support future technological expectations. Further, the days of unlimited funding are over.
Investors want to see results, fast.
Common Lisp (CL) is one of the few languages and development options that can meet
these challenges. Powerful, flexible, changeable on the fly — increasingly, CL is playing
a leading role in areas with complex problem-solving demands. Engineers in the fields of
bioinformatics, scheduling, data mining, document management, B2B, and E-commerce
have all turned to CL to complete their applications on time and within budget. CL,
however, no longer just appropriate for the most complex problems. Applications of modest
complexity, but with demanding needs for fast development cycles and customization, are
also ideal candidates for CL.
Other languages have tried to mimic CL, with limited success. Perl, Python, Java,
C++, C# — they all incorporate some of the features that give Lisp its power, but their
implementations tend to be brittle.
The purpose of this book is to showcase the features that make CL so much better
than these imitators, and to give you a “quick-start” guide for using Common Lisp as a
development environment. If you are an experienced programmer in languages other than
Lisp, this guide gives you all the tools you need to begin writing Lisp applications. If you’ve
had some exposure to Lisp in the past, this guide will help refresh those memories and shed
some new light on CL for you.
1

this Foreword was authored by Franz Inc.

iii

iv
But be careful, Lisp can be addicting! This is why many Fortune 500 companies will
use nothing else on their 24/7, cutting-edge, mission-critical applications. After reading
this book, trying our software, and experiencing a 3 to 10 times increase in productivity,
we believe you will feel the same way.

Contents
1 Introduction
1.1 The Past, Present, and Future of Common Lisp
1.1.1 Lisp Yesterday . . . . . . . . . . . . . .
1.1.2 Lisp Today . . . . . . . . . . . . . . . .
1.1.3 Lisp Tomorrow . . . . . . . . . . . . . .
1.2 Convergence of Hardware and Software . . . .
1.3 The CL Model of Computation . . . . . . . . .

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

1
1
1
1
2
3
3

2 Operating a CL Development Environment
2.1 Installing a CL Environment . . . . . . . . . . . . . . . . . . . . . . . .
2.2 Running CL in a Shell Window . . . . . . . . . . . . . . . . . . . . . . .
2.2.1 Starting CL from a Terminal Window . . . . . . . . . . . . . . .
2.2.2 Stopping CL from a Terminal Window . . . . . . . . . . . . . . .
2.3 Running CL inside a Text Editor . . . . . . . . . . . . . . . . . . . . . .
2.3.1 A Note on Emacs and Text Editors . . . . . . . . . . . . . . . . .
2.3.2 Emacs Terminology . . . . . . . . . . . . . . . . . . . . . . . . .
2.3.3 Starting, Stopping, and Working With CL inside an Emacs Shell
2.4 Running CL as a subprocess of Emacs . . . . . . . . . . . . . . . . . . .
2.4.1 Starting the CL subprocess within Emacs . . . . . . . . . . . . .
2.4.2 Working with CL as an Emacs subprocess . . . . . . . . . . . . .
2.4.3 Compiling and Loading a File from an Emacs buffer . . . . . . .
2.5 Integrated Development Environment . . . . . . . . . . . . . . . . . . .
2.6 The User Init File . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.7 Using CL as a scripting language . . . . . . . . . . . . . . . . . . . . . .
2.8 Debugging in CL . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.8.1 Common debugger commands . . . . . . . . . . . . . . . . . . . .
2.8.2 Interpreted vs Compiled Code . . . . . . . . . . . . . . . . . . .
2.8.3 Use of (break) and C-c to interrupt . . . . . . . . . . . . . . . . .
2.8.4 Profiling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.9 Developing Programs and Applications in CL . . . . . . . . . . . . . . .
2.9.1 A Layered Approach . . . . . . . . . . . . . . . . . . . . . . . . .
2.9.2 Compiling and Loading your Project . . . . . . . . . . . . . . . .
2.9.3 Creating an Application “Fasl” File . . . . . . . . . . . . . . . .
2.9.4 Creating an Image File . . . . . . . . . . . . . . . . . . . . . . .

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

5
5
6
6
6
8
8
8
9
10
10
10
11
12
12
13
14
14
15
15
16
16
16
16
17
17

v

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

vi

CONTENTS
2.9.5
2.9.6

Building Runtime Images . . . . . . . . . . . . . . . . . . . . . . . .
Using an Application Init File . . . . . . . . . . . . . . . . . . . . . .

3 The CL Language
3.1 Overview of CL and its Syntax . . . . . . . . .
3.1.1 Evaluation of Arguments to a Function
3.1.2 Lisp Syntax Simplicity . . . . . . . . . .
3.1.3 Turning Off Evaluation . . . . . . . . .
3.1.4 Fundamental CL Data Types . . . . . .
3.1.5 Functions . . . . . . . . . . . . . . . . .
3.1.6 Global and Local Variables . . . . . . .
3.2 The List as a Data Structure . . . . . . . . . .
3.2.1 Accessing the Elements of a List . . . .
3.2.2 The “Rest” of the Story . . . . . . . . .
3.2.3 The Empty List . . . . . . . . . . . . .
3.2.4 Are You a List? . . . . . . . . . . . . . .
3.2.5 The conditional If . . . . . . . . . . . .
3.2.6 Length of a List . . . . . . . . . . . . .
3.2.7 Member of a List . . . . . . . . . . . . .
3.2.8 Getting Part of a List . . . . . . . . . .
3.2.9 Appending Lists . . . . . . . . . . . . .
3.2.10 Adding Elements to a List . . . . . . . .
3.2.11 Removing Elements from a List . . . . .
3.2.12 Sorting Lists . . . . . . . . . . . . . . .
3.2.13 Treating a List as a Set . . . . . . . . .
3.2.14 Mapping a Function to a List . . . . . .
3.2.15 Property Lists . . . . . . . . . . . . . .
3.3 Control of Execution . . . . . . . . . . . . . . .
3.3.1 If . . . . . . . . . . . . . . . . . . . . . .
3.3.2 When . . . . . . . . . . . . . . . . . . .
3.3.3 Logical Operators . . . . . . . . . . . .
3.3.4 Cond . . . . . . . . . . . . . . . . . . .
3.3.5 Case . . . . . . . . . . . . . . . . . . . .
3.3.6 Iteration . . . . . . . . . . . . . . . . . .
3.4 Functions as Objects . . . . . . . . . . . . . . .
3.4.1 Named Functions . . . . . . . . . . . . .
3.4.2 Functional Arguments . . . . . . . . . .
3.4.3 Anonymous Functions . . . . . . . . . .
3.4.4 Optional Arguments . . . . . . . . . . .
3.4.5 Keyword Arguments . . . . . . . . . . .
3.5 Input, Output, Streams, and Strings . . . . . .
3.5.1 Read . . . . . . . . . . . . . . . . . . . .
3.5.2 Print and Prin1 . . . . . . . . . . . . . .
3.5.3 Princ . . . . . . . . . . . . . . . . . . .
3.5.4 Format . . . . . . . . . . . . . . . . . .

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

18
18
19
19
21
21
22
22
25
25
27
28
29
29
29
30
30
31
32
32
33
33
34
34
35
35
35
36
36
36
37
38
38
39
39
40
40
41
41
42
42
42
43
43

CONTENTS

vii
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

44
44
45
45
46
47
47
48
49
49
49
49
50
50
52
53
54

4 Interfaces
4.1 Interfacing with the Operating System . . . . .
4.2 Foreign Function Interface . . . . . . . . . . . .
4.3 Interfacing with Corba . . . . . . . . . . . . . .
4.4 Custom Socket Connections . . . . . . . . . . .
4.5 Interfacing with Windows (COM, DLL, DDE) .
4.6 Code Generation into Other Languages . . . .
4.7 Multiprocessing . . . . . . . . . . . . . . . . . .
4.7.1 Starting a Background Process . . . . .
4.7.2 Concurrency Control . . . . . . . . . . .
4.8 Database Interfaces . . . . . . . . . . . . . . . .
4.8.1 ODBC . . . . . . . . . . . . . . . . . . .
4.8.2 MySQL . . . . . . . . . . . . . . . . . .
4.9 World Wide Web . . . . . . . . . . . . . . . . .
4.9.1 Server and HTML Generation . . . . . .
4.9.2 Client and HTML Parsing . . . . . . . .
4.10 Regular Expressions . . . . . . . . . . . . . . .
4.11 Email . . . . . . . . . . . . . . . . . . . . . . .
4.11.1 Sending Mail . . . . . . . . . . . . . . .
4.11.2 Retrieving Mail . . . . . . . . . . . . . .

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

55
55
55
56
56
58
58
58
58
59
60
61
61
62
62
63
64
64
64
65

.
.
.
.
.

67
67
67
72
72
73

3.6

3.7

3.8

3.5.5 Pathnames . . . . . . . . . . . . . . .
3.5.6 File Input and Output . . . . . . . . .
Hash Tables, Arrays, Structures, and Classes
3.6.1 Hash Tables . . . . . . . . . . . . . . .
3.6.2 Arrays . . . . . . . . . . . . . . . . . .
3.6.3 Structures . . . . . . . . . . . . . . . .
3.6.4 Classes and Methods . . . . . . . . . .
Packages . . . . . . . . . . . . . . . . . . . . .
3.7.1 Importing and Exporting Symbols . .
3.7.2 The Keyword Package . . . . . . . . .
Common Stumbling Blocks . . . . . . . . . .
3.8.1 Quotes . . . . . . . . . . . . . . . . . .
3.8.2 Function Argument Lists . . . . . . .
3.8.3 Symbols vs. Strings . . . . . . . . . .
3.8.4 Equality . . . . . . . . . . . . . . . . .
3.8.5 Distinguishing Macros from Functions
3.8.6 Operations that cons . . . . . . . . . .

5 Squeakymail
5.1 Overview of Squeakymail
5.2 Fetching and Scoring . . .
5.2.1 Tokenizing . . . .
5.2.2 Scoring . . . . . .
5.3 Browsing and Classifying

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

viii

CONTENTS

A Squeakymail with Genworks’ GDL/GWL
A.1 Toplevel Squeakymail Home Page . . . . . . . . . . . . . . . . . . . . . . . .
A.2 Page for Browsing and Classifying . . . . . . . . . . . . . . . . . . . . . . .

77
77
78

B Bibliography

83

C Emacs Customization
C.1 Lisp Mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
C.2 Making Your Own Keychords . . . . . . . . . . . . . . . . . . . . . . . . . .
C.3 Keyboard Mapping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

85
85
85
86

D Afterword
D.1 About This Book . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
D.2 Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
D.3 About the Author . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

87
87
87
87

Chapter 1

Introduction
1.1
1.1.1

The Past, Present, and Future of Common Lisp
Lisp Yesterday

John McCarthy discovered the basic principles of Lisp in 1958, when he was processing
complex mathematical lists at MIT. Common Lisp (CL) is a high-level computer language,
whose syntax follows a simple list-like structure. The term “Lisp” itself originally stood for
“LISt Processing.” When developing, testing, and running a CL program, at the core is a
modern-day version of the original List Processor which processes (compiles, evaluates, etc.)
the elements of your program. These elements, at the source code level, are represented
as lists. A list, in this context, is just a sequence of items, much like a familiar shopping
list or checklist. Originally the list was pretty much the only data structure supported by
Lisp, but modern-day Common Lisp supports a wide range of flexible and efficient data
structures.
A typical Common Lisp development and runtime environment behaves like a complete
operating system, with multiple threads of execution and the ability to load new code
and redefine objects (functions, etc.) dynamically, i.e. without stopping and restarting the
“machine.”
This “operating system” characteristic also makes CL significantly more flexible than
other popular programming languages. Whereas some other languages, such as shell scripts
or Perl CGI scripts, need to pipe data around, in CL all data manipulations can run
interactively within one single process with a shared memory space.

1.1.2

Lisp Today

The most popular form of Lisp used today, “ANSI Common Lisp,” was initially designed by
Guy Steele in Common Lisp, the Language, 2nd Edition (“CLtL2”) — (see the Bibliography in Appendix B). This version of Lisp became the accepted industry standard. In 1995,
the American National Standards Institute recognized a slightly updated version as ANSI
Common Lisp, the first object-oriented language to receive certification, and ANSI CL remains the only language that meets all of the criteria set forth by the Object Management
Group (OMG) for a complete object-oriented language.
1

2

CHAPTER 1. INTRODUCTION

Paul Graham outlines ANSI CL in his 1996 book ANSI Common Lisp, also listed in
Appendix B.
Official language standardization is important, because it protects developers from becoming saddled with legacy applications as new versions of a language are implemented.
For example, this lack of standardization has been a continuing problem for Perl developers.
Since each implementation of Perl defines the behavior of the language, it is not uncommon
to have applications that require outdated versions of Perl, making it nearly impossible to
use in a mission-critical commercial environment.

1.1.3

Lisp Tomorrow

Many developers once hoped that the software development process of the future would be
more automated through Computer-aided Software Engineering (CASE) tools. Such tools
claim to enable programmers and non-programmers to diagram their applications visually
and automatically generate code. While useful to a certain extent, traditional CASE tools
cannot cover domain-specific details and support all possible kinds of customizations — so
developers inevitably need to hand-code the rest of their applications, breaking the link
between the CASE model and the code. CASE systems fall short of being a true “problemsolving tool.”
The recursive nature of CL, and its natural ability to build applications in layers and
build upon itself — in essence, code which writes code which writes code..., makes CL a
much better software engineering solution. CL programs can generate other CL programs
(usually at compile-time), allowing the developer to define all parts of the application in
one unified high-level language. Using macros and functions, naturally supported by the CL
syntax, the system can convert applications written in the high-level language automatically
into “final” code. Thus, CL is both a programming language and a powerful CASE tool,
and there is no need to break the connection.
Several other features of CL also save significant development time and manpower:
• Automatic Memory Management/Garbage Collection: inherent in CL’s basic architecture.
• Dynamic Typing: In CL, values have types, but variables do not. This means that
you don’t have to declare variables, or placeholders, ahead of time to be of a particular
type; you can simply create them or modify them on the fly1 .
• Dynamic Redefinition: You can add new operations and functionality to the running
CL process without the need for downtime. In fact, a program can be running in one
thread, while parts of the code are redefined in another thread. Moreover, as soon as
the redefinitions are finished, the application will use the new code and is effectively
modified while running. This feature is especially useful for server applications that
cannot afford downtimes – you can patch and repair them while they are running.
• Portability: The “abstract machine” quality of CL makes programs especially portable.
1
While you don’t have to declare variable types, you may declare them. Doing so can help CL’s compiler
to optimize your program further.

1.2. CONVERGENCE OF HARDWARE AND SOFTWARE

3

• Standard Features: CL contains many built-in features and data structures that are
non-standard in other languages. Features such as full-blown symbol tables and package systems, hash tables, list structures, and a comprehensive object system are several “automatic” features which would require significant developer time to recreate
in other languages.

1.2

Convergence of Hardware and Software

The most common criticism of Lisp usually made by programmers unfamiliar with the
language is that Lisp applications are too big and too slow. While this may have been true
20 years ago, it is absolutely not the case today. Today’s inexpensive commodity computers
have more memory than some of the most powerful machines available just five years ago,
and easily meet the computing needs of a typical CL application. Hardware is no longer
the critical factor. Programmers and development time are!
Further, CL programming teams tend to be small, because the inherent features in CL
empower developers to do more. Without excessive effort, a single CL developer can create
and deploy a sophisticated, powerful program in a much shorter period of time than if
another language were used.

1.3

The CL Model of Computation

Programming in CL is distinguished from programming in other languages due to its unique
syntax and development mode. CL allows developers to make changes and test them immediately, in an incremental manner. In other languages, a developer must follow the
“compile-link-run-test” cycle. These extra steps add minutes or hours to each cycle of development, and break the programmer’s thought process. Multiply these figures by days,
weeks, and months — and the potential savings are phenomenal.

4

CHAPTER 1. INTRODUCTION

Chapter 2

Operating a CL Development
Environment
This chapter covers some of the hands-on techniques used when working with a CL development environment, specifically, the Allegro CL R environment from Franz Inc. If you
are completely unfamiliar with any Lisp language, you may wish to reference Chapter 3 as
you read this chapter.
Ideally, while reading this chapter, you should have access to an actual Common Lisp
session and try typing in some of the examples.
When working with Common Lisp, it helps to think of the environment as its own
operating system, sitting on top of whatever operating system you happen to be working
with. In fact, people used to build entire workstations and operating systems in Lisp. The
Symbolics Common Lisp environment still runs as an emulated machine on top of the 64-bit
Compaq/DEC Alpha CPU, with its own Lisp-based operating system.
A complete and current listing of available CL systems can be found on the Association
of Lisp Users website, at
http://www.alu.org/table/systems.htm
These systems follow the same basic principles, and most of what you learn and write on one
system can be applied and ported over to others. The examples in this guide were prepared
using Allegro CL for Linux. Where possible, we will note Allegro CL-specific syntax and
extensions.

2.1

Installing a CL Environment

Installing a CL environment is generally a straightforward process. For Allegro CL on Linux,
this basically involves running a simple shell script which steps you through some licensing
information, then proceeds to unpack the distribution, and sets up the CL executable and
supporting files.
The Windows installation involves somewhat less typing and more mouse clicking.
5

6

2.2

CHAPTER 2. OPERATING A CL DEVELOPMENT ENVIRONMENT

Running CL in a Shell Window

The most primitive and simplest way to run CL is directly from the Unix (Linux) shell
from within a terminal window. Most programmers do not work this way on a daily basis
because other modes of use provide more power and convenience with fewer keystrokes.
This mode of working is sometimes useful for logging into a running CL system on a
remote host (e.g. a CL process acting as a webserver) when using a slow dial-up connection.

2.2.1

Starting CL from a Terminal Window

To start a CL session from a terminal window, simply type
alisp
from the Unix (or DOS) command line. Assuming your shell execution path is set up
properly and CL is installed properly, you will see some introductory information, then will
be presented with a Command Prompt which should look very similar to the following:
CL-USER(1):
CL has entered its read-eval-print loop, and is waiting for you to type something which it
will then read, evaluate to obtain a return-value, and finally print this resulting return-value.
The USER printed before the prompt refers to the current CL package (more on Packages
later), and the number (1) simply keeps track of how many commands have been entered
at the prompt.

2.2.2

Stopping CL from a Terminal Window

Now that you have learned how to start CL, you should probably know how to stop it. In
Allegro CL, one easy way to stop the CL process is to type
(excl:exit)
at the Command Prompt. This should return you to the shell. In very rare cases, a stronger
hammer is required to stop the CL process, in which case you can try typing
(excl:exit 0 :no-unwind t)
Try It: Now try starting and stopping CL several times, to get a feel for it. In general, you
should notice that CL starts much more quickly on subsequent invocations. This is because
the entire executable file has already been read into the computer’s memory and does not
have to be read from the disk every time.
A shortcut for the (excl:exit) command is the toplevel command :exit. Toplevel
commands are words which are preceded by a colon [:] that you can type at the CL
Command Prompt as a shorthand way of making CL do something.
Try It: Start the CL environment. Then, using your favorite text editor, create a file
/tmp/hello.lisp and place the following text into it (don’t worry about understanding
the text for now):

2.2. RUNNING CL IN A SHELL WINDOW

7

(in-package :user)
(defun hello ()
(write-string "Hello, World!"))
Save the file to disk. Now, at the CL prompt in the shell where you started the CL
environment, compile and load this file as follows (text after the “USER” prompt represents
what you have to type; everything else is text printed by CL):
CL-USER(3): (compile-file "/tmp/hello")
;;; Compiling file /tmp/hello.lisp
;;; Writing fasl file /tmp/hello.fasl Fasl write complete
#p"/tmp/hello.fasl"
NIL
NIL
CL-USER(4): (load "/tmp/hello")
; Fast loading /tmp/hello.fasl
T
By default, the compile-file command will look for files with the .lisp suffix, and the
load command will load the resulting compiled machine-code (binary) file, which by default
will have the extension .fasl. The extension “.fasl” stands for “FASt Loading” file.
Note that, in general, simply loading an uncompiled Lisp file (using load) will functionally have the same effect as loading a compiled binary (fasl) file. But the fasl file will load
faster, and any functions, objects, etc. defined in it will perform faster, since they will have
been optimized and translated into language which is closer to the actual machine. Unlike
Java “class” files, CL fasl files usually represent native machine-specific code. This means
that compiled CL programs will generally run much faster than compiled Java programs,
but CL programs must be compiled separately for each type of machine on which you want
to run them.
Finally, try executing the newly defined function hello by typing the following at your
command line:
CL-USER(5): (hello)
Hello, World!
"Hello, World!"
You should see the String Hello, World! printed (without double-quotes), then the returned String (with double-quotes) will be printed as well. In the next chapter, we will
learn more about exactly what is going on here. For now the main point to understand is
the idea of compiling and loading definitions from files, and invoking functions by typing
expressions at the CL command line.

8

CHAPTER 2. OPERATING A CL DEVELOPMENT ENVIRONMENT

2.3

Running CL inside a Text Editor

A more powerful way of developing with CL is by running it from within a Unix (or DOS)
shell that is running inside a buffer in a powerful text editor, such as Gnu Emacs1 . One
advantage of this approach is that the editor (in our example, Emacs) keeps a history of
the entire session: both the expressions the programmer has typed, as well as everything
that CL has printed. All these items can be easily reviewed or recalled.

2.3.1

A Note on Emacs and Text Editors

To develop using Common Lisp, you can, of course, use any text editor of your choice,
although Emacs happens to be especially synergistic for working with CL, for reasons we
will discuss later. With a bit of practice, you will find that the combination of Emacs and
Allegro CL provides a more powerful and convenient development environment than any of
today’s flashy “visual programming” environments. If you are used to working with the “vi”
editor, Emacs comes with several “vi” emulation modes which you may want to experiment
with on a transitional basis. If you are used to working with a standard Windows-based
text editor, you will find a smooth transition to Emacs.
This guide provides examples using Gnu Emacs.
To work with CL in an Emacs shell buffer, you should be reasonably familiar with the
text editor. If you are completely new to Gnu Emacs, go through the Emacs Tutorial,
available under the “Help” menu at the top of the Emacs window.

2.3.2

Emacs Terminology

As you will learn in the Emacs Tutorial, when working with Emacs you make frequent use
of key chords. Similar to “chords” on a piano, keychords can allow powerful text processing
with relatively few sequential keystrokes.
They are key combinations involving holding down the Control key and/or the Meta
key, and pressing some other regular key.
See Appendix C for information on which keys represent Control and Meta on your
keyboard, and how you can customize these. To begin, try using the key labeled “Control”
for Control and the keys labeled “Alt” (or the diamond-shaped key on Sun keyboards) for
Meta. If your keyboard has a “Caps Lock” key to the left of the letter “A,” you should
consider remapping this to function as Control, as explained in Appendix C.
This guide uses the following notations to indicate keychords (identical to those referenced in the Emacs Tutorial):
• M-x means to hold down the Meta key, and press (in this case) the “X” key
• C-x means to hold down the Control key, and press (in this case) the “X” key
• C-M-q means to hold down the Control key and the Meta key at the same time , and
press (in this case) the “Q” key
1

Gnu Emacs comes pre-installed with all Linux systems, and for other systems should come as part
of your Allegro CL distribution. In any case, Emacs distributions and documentation are available from
http://www.gnu.org

2.3. RUNNING CL INSIDE A TEXT EDITOR

9

• M-A would mean to hold down the Meta key, and the Shift key (because the “A” is
uppercase), and press (in this case) the “A” key.

2.3.3

Starting, Stopping, and Working With CL inside an Emacs Shell

To start Emacs, type
emacs
or
gnuemacs
then start a Unix (or DOS) shell within emacs by typing M-x shell. Now type
lisp
inside this shell’s window to start a CL session there. To shut down a session, exit CL
exactly as above by typing
(excl:exit)
Exit from the shell by typing
exit
and finally exit from Emacs by typing C-x C-c or M-x kill-emacs. Note that it is good
practice always to exit from CL before exiting from Emacs. Otherwise the CL process may
be left as an “undead” (zombie) process.
When in the CL process, you can move forward and backward in the “stack” (history)
of expressions that you have typed, effectively recalling previous commands with only a few
keystrokes. Use M-p to move backward in the history, and M-n to move forward. You can
also use all of the editor’s text processing commands, for example to cut, copy, and paste
expressions between the command line and other windows and open files (“buffers”) you
may have open.
Try It: try typing the following expressions at the CL prompt. See what return-values
are printed, and try recalling previous expressions (commands) using M-p and M-n:
(list 1 2 3)
(+ 1 2 3)
(> 3 4)
(< 3 4)
Now, of course, you can edit, compile, load, and test your hello.lisp file as we did in
Section 2.2. But this time you are staying within a single environment (the text editor
environment) to achieve all these tasks.

10

2.4

CHAPTER 2. OPERATING A CL DEVELOPMENT ENVIRONMENT

Running CL as a subprocess of Emacs

An even more powerful way of developing with Allegro CL is by running it within Emacs
in conjunction with a special Emacs/Lisp interface provided by Franz Inc. This technique
provides all the benefits of running in an Emacs shell as outlined above, but is more powerful
because the Emacs-CL interface extends Emacs with several special commands. These
commands interact with the CL process, making the editor appear to be “Lisp-aware.” The
combination of Emacs with the Emacs-CL interface results in a development environment
whose utility approaches that of the vintage Symbolics Lisp Machines, which many Lisp
aficionados still consider to be technology advanced well beyond anything produced today.
Complete documentation of the Emacs-CL interface can be viewed by pointing your
web browser at: file:/usr/local/acl62/doc/eli.htm (you may have to modify this
pathname to point to your actual installed location of Allegro CL (acl62)).

2.4.1

Starting the CL subprocess within Emacs

To enable your Emacs to function in the proper manner, place the following expressions in
a file named .emacs in your home directory:
(setq load-path
(cons "/usr/local/acl62/eli" load-path))
(load "fi-site-init")
As above, you may have to modify this pathname to point to your actual installed location
of Allegro CL (acl62). Once you have added these lines to your .emacs file, you should be
able to restart emacs, then issue the command:
M-x fi:common-lisp
to start CL running and automatically establish a network connection between the Emacs
process and the CL process (in theory, the two processes could be running on different
machines, but in practice they usually will run on the same machine). Accept all the
defaults when prompted.
You can also start emacs with the CL subprocess automatically from a terminal window,
by invoking emacs with an optional “function” argument, as follows:
emacs -f fi:common-lisp
This will start Emacs, which itself will immediately launch CL.

2.4.2

Working with CL as an Emacs subprocess

Once you have started CL within Emacs, you will be presented with a special buffer named
*common-lisp* which in many ways is similar to simply running CL within a shell within
Emacs as above. But now you have several more capabilities available to you, as outlined
in Table 2.1.
To summarize, the Lisp-Emacs interface adds capabilities such as the following to your
development environment:

2.4. RUNNING CL AS A SUBPROCESS OF EMACS
Keychord
C-M-x

C-c C-b

M-A
C-c .

Action
Compiles and loads
the function near the
Emacs point
Compiles and loads
the current Emacs
buffer
Shows the Argument
List for a CL function
Finds the definition
of a CL object

11

Emacs-Lisp function
fi:lisp-eval-or-compile-defun

fi:lisp-eval-or-compile-current-buffer

fi:lisp-arglist
fi:lisp-find-definition

Table 2.1: Common Commands of the Emacs-Lisp Interface
• Start and stop the CL process using Emacs commands
• Compile and load files into the running CL process, directly from the open Emacs
buffer you are editing (no need explicitly to save the file, call the compile-file
function, then call the load function, as with the “in-a-shell” techniques above)
• Query the running CL process for information about objects currently defined in it
— for example, querying the CL, directly from Emacs, for the argument List of a
particular function
• Run multiple read-eval-print loops, or listeners, in different threads of CL execution,
as Emacs buffers
• Compile and load specified sections of an open Emacs buffer
• Locate the source code file corresponding to any defined CL object, such as a function
or parameter, and automatically open it in an Emacs buffer
• Perform various debugging actions and other general CL help capabilities, directly
from within the Emacs session
In short, Emacs becomes a remarkably powerful, infinitely customizable, CL integrated
development environment.

2.4.3

Compiling and Loading a File from an Emacs buffer

Try it: Try starting Emacs with a CL subprocess as outlined above. If you have trouble,
consult the URL listed above for complete instructions on running the Lisp-Emacs interface.
Once you have Emacs running with a CL subprocess, you should have a buffer named
*common-lisp*. If your Emacs session is not already visiting this buffer, visit it with the
command C-x b *common-lisp*, or select *common-lisp* from the “Buffers” menu in
Emacs (see Appendix C for instructions on how to set up a “hot key” for this and other
common tasks when working with the Emacs-Lisp interface).

12

CHAPTER 2. OPERATING A CL DEVELOPMENT ENVIRONMENT

Now, split your Emacs frame into two windows, with the command C-x 2. In both
windows you should now see *common-lisp*. Now move the point2 to the other window
with C-x o, and open the hello.lisp file you created for the previous exercise3 . If you
don’t have that file handy, create it now, with the following contents:
(defun hello ()
(write-string "Hello, World!"))

Now, compile and load the contents of this function definition with C-M-x. Finally, switch
to the *common-lisp* window again with C-x o, and try invoking the hello function by
calling it in normal Lisp fashion:
CL-USER(5): (hello)
Hello, World!
"Hello, World!"
You should see the results printed into the same *common-lisp* buffer.
Try it: Now try the following: go back to the hello.lisp buffer, make an edit to
the contents (for example, change the String to "Hello, CL World!"). Now compile the
buffer with this change (Note that you do not necessarily have to save the file). Finally,
return to the *common-lisp* buffer, and try invoking the function again to confirm that
the redefinition has taken effect.
You have now learned the basic essentials for working with the Emacs-Lisp interface.
See Appendix C for more information about Emacs customization and convenient Emacs
commands specific to the Lisp editing mode.

2.5

Integrated Development Environment

Some CL implementations support or integrate with an IDE, or Integrated Development
Environment. Such environments provide visual layout and design capabilities for User
Interfaces and other graphical, visual techniques for application development. For example,
such a system is available for Allegro CL on the Microsoft platforms. The details of these
systems are covered in their own documentation and are beyond the scope of this guide.

2.6

The User Init File

When CL begins execution, it usually looks for an initialization file, or “init” file, to load.
When CL loads this file, it will read and evaluate all the commands contained in it.
The default name and location for this initialization file is
.clinit.cl
2

In technical Emacs parlance, the point is the insertion point for text, and the cursor is the mouse
pointer.
3
Remember that pressing the space bar will automatically complete filenames for you when opening files
in Emacs.

2.7. USING CL AS A SCRIPTING LANGUAGE

13

in the user’s home directory on Unix/Linux, and
c:\clinit.cl
on Windows systems.
For example, the init file can be used to define library functions and variables that the
user employs frequently, or to control other aspects of the CL environment. An example
init file is:
(in-package :user)
(defun load-project ()
(load "~/lisp/part1")
(load "~/lisp/part2"))
This will have the effect of defining a convenient function the user can then invoke to
load the files for the current project. Note the call to the in-package function, which
tells CL which Package to make “current” when loading this file. As we will see in more
detail in Section 3.7, CL Packages allow programs to be separated into different “layers,” or
“modules.” Notice also that in the calls to load we do not specify a file “type” extension,
e.g. .lisp or .fasl. We are making use of the fact that the load function will look first
for a binary file (of type fasl), and if this is not found it will then look for a file of type
cl or lisp. In this example we assume that our Lisp files will already have been compiled
into up-to-date binary fasl files.
In Section 2.9 we will look at some techniques for managing larger projects, and automating both the compiling and loading process for the source files making up a project.
Note: the IDE mentioned in Section 2.5 is also a full-featured and powerful project
system. The IDE is described in its own documentation, and is beyond the scope of this
document.

2.7

Using CL as a scripting language

Although CL is usually used in the interactive mode described above while developing a
program, it can be also used as a scripting language in more of a “batch” mode. For
example, by running CL as:
lisp -e ’(load "~/script")’ < datafile1 > datafile2
it will load all of the function definitions and other expressions in file script.fasl or
script.lisp, will process datafile1 on its Standard Input, and will create datafile2
with anything it writes to its Standard Output. Here is an example of a simple script.lisp
file:
(in-package :user)
(defun process-data ()

14

CHAPTER 2. OPERATING A CL DEVELOPMENT ENVIRONMENT
(let (x j)
;;Read a line of input from datafile1
(setq x (read-line))
;;Remove leading spaces from X
(setq j (dotimes (i (length x))
(when (not (string-equal " " (subseq x i (+ i 1))))
(return i))))
(setq x (subseq x j))
;;Write the result to datafile2
(format t "~A~%" x)))
(process-data)

This file defines a function to process the datafile, then calls this function to cause the
actual processing to occur.
Note the use of the semicolon (“;”), which is a reader macro which tells CL to treat
everything between it and the end of the line as a comment (to be ignored by the CL
Reader).

2.8

Debugging in CL

CL provides one of the most advanced debugging and error-handling capabilities of any
language or environment. The most obvious consequence of this is that your application
will very rarely “crash” in a fatal manner. When an error occurs in a CL program, a break
occurs. CL prints an error message, enters the debugger, and presents the user with one
or more possible restart actions. This is similar to running inside a debugger, or in “debug
mode,” in other languages, but in CL this ability comes as a built-in part of the language.
For example, if we call the + function with an alphabetic Symbol instead of a Number,
we generate the following error:
CL-USER(95): (+ ’r 4)
Error: ‘R’ is not of the expected type ‘NUMBER’
[condition type: TYPE-ERROR]
Restart actions (select using :continue):
0: Return to Top Level (an "abort" restart)
[1] CL-USER(96):
The number [1] at the front of the prompt indicates that we are in the debugger, at Level 1.
The debugger is itself a CL Command Prompt just like the toplevel Command Prompt, but
it has some additional functionality. The next section provides a partial list of commands
available to be entered directly at the debugger’s Command Prompt:

2.8.1

Common debugger commands

• :reset returns to the toplevel (out of any debug levels)

2.8. DEBUGGING IN CL

15

• :pop returns up one level
• :continue continues execution after a break or an error
• :zoom Views the current function call stack4
• :up Moves the currency pointer up one frame in the stack
• :down Moves the currency pointer down one frame in the stack
• :help Describes all debugger commands
Franz Inc.’s Allegro Composer product also offers graphical access to these commands,
as well as additional functionality.

2.8.2

Interpreted vs Compiled Code

CL works with both compiled and interpreted code. When you type expressions at a Command Prompt, or use CL’s load function to load lisp source files directly, you are introducing interpreted code into the system. This code does not undergo any optimization or
translation into a lower-level machine representation — internally to CL it exists in a form
very similar to its source code form. CL must do considerable work every time such code
is evaluated, since it must be translated into lower-level machine instructions on each use.
When you create compiled code, on the other hand, for example by using CL’s compile-file
command in order to produce a fasl file, the code is optimized and translated into an efficient form. Loading the fasl file will replace any previous interpreted definitions with
compiled definitions.
Interpreted code is the best to use for debugging a program, because the code has not
undergone as many transformations, so the debugger can provide the programmer with
more recognizable information.
If you are debugging with compiled code (e.g. after loading a fasl file), and you feel that
not enough information is available from within the debugger, try redefining the function
in interpreted form and running it again. To redefine a function in interpreted form you
can, of course, simply load the lisp source file, rather than the compiled fasl file.

2.8.3

Use of (break) and C-c to interrupt

In order to cause a break in your program interactively, issue the C-c command. When
running inside Emacs, you must use a “double” C-c, because the first C-c will be intercepted
by Emacs. Depending on what your application is doing, it may take a few moments for it
to respond to your break command. Once it responds, you will be given a normal debugger
level, as described above.
4

the call stack is the sequence of functions calls leading up to the error or break.

16

2.8.4

CHAPTER 2. OPERATING A CL DEVELOPMENT ENVIRONMENT

Profiling

CL has built-in functionality for monitoring its own performance. The most basic and
commonly used component is the time Macro. You can “wrap” the time Macro around
any expression. Functionally, time will not affect the behavior of the expression, but it will
print out information about how long the expression took to be evaluated:
CL-USER(2): (time (dotimes (n 10000) (* n 2)))
; cpu time (non-gc) 70 msec user, 0 msec system
; cpu time (gc)
0 msec user, 0 msec system
; cpu time (total) 70 msec user, 0 msec system
; real time 72 msec
It is often interesting to look at the difference in time between interpreted and compiled
code. The example above is interpreted, since we typed it directly at the Command Line,
so CL had no opportunity to optimize the dotimes. In compiled form, the above code
consumes only one millisecond as opposed to 72 milliseconds.
In addition to the basic time Macro, Allegro CL provides a complete Profiler package,
which allows you to monitor exactly what your program is doing, when it is doing it, how
much time it takes, and how much memory it uses.

2.9

Developing Programs and Applications in CL

For developing a serious application in CL, you will want to have a convenient procedure
in place for starting up your development session.

2.9.1

A Layered Approach

The development session consists of your favorite text editor with appropriate customizations, and a running CL process with any standard code loaded into it. While you can
certainly develop serious applications on top of base CL, typically you will work in a layered fashion, on top of some standard tools or libraries, such as an embedded language, a
webserver, etc.
Using one of the techniques outlined below, starting a customized development session
with your standard desired tools should become convenient and transparent.

2.9.2

Compiling and Loading your Project

An individual project will generally consist of a tree of directories and files in your computer’s filesystem. We will refer to this directory tree as the project’s codebase.
Usually, it does not matter in what order the files in the codebase are compiled and
loaded. Typical CL definitions, such as functions, classes, etc., can ordinarily be defined in
any order. However, certain constructs, most notably Packages and Macros, do have order
dependencies. In general, CL Packages and Macros must be defined before they can be
used, or referenced.
To start a development session on your project, you will generally want CL to do the
following:

2.9. DEVELOPING PROGRAMS AND APPLICATIONS IN CL

17

1. Go through your project’s Codebase, finding source files (in the proper order when
necessary, as noted above)
2. For each file found, either load the corresponding binary file (if it is up-to-date), or
compile the source file to create a new binary file and load it (if the source file has
changed since the most recent compilation)
CL does not dictate one standard way of accomplishing this task, and in fact, several
options are available. One option is to use a defsystem package, which will allow you to
prepare a special file, similar to a “make” file, which lists out all your project’s source
files and their required load order. Allegro CL contains its own Defsystem package as
an extension to CL, and the open-source MK:Defsystem is available through Sourceforge
(http://www.sourceforge.net).
Using a Defsystem package requires you to maintain a catalog listing of all your project’s
source files. Especially in the early stages of a project, when you are adding, renaming, and
deleting a lot of files and directories, maintaining this catalog listing by hand can become
tiresome.
For this purpose, you may wish to use a lighter-weight utility for compiling and loading
your files, at least in the early stages of development. For very small projects, you can use
the Allegro CL function excl:compile-file-if-needed, an extension to CL, by writing a
simple function which calls this function repeatedly to compile and load your project’s files.
The Bootstrap package, available at http://gdl.sourceforge.net, provides the function cl-lite, which will traverse a codebase, compiling and loading files, and observing
simple “ordering directive” files which can be placed throughout the codebase to enforce
correct load ordering.
For a new project, the best option is probably to start with a simple but somewhat
manual loading technique, then graduate into more automated techniques as they become
necessary and as you become more familiar with the environment.

2.9.3

Creating an Application “Fasl” File

Fasl files in Allegro CL and most other CL systems have the convenient property that they
can simply be concatenated, e.g. with the Unix/Linux cat command, to produce a single
fasl file. Loading this single file will have the same effect as loading all the individual files
which went into it, in the same order as they were concatenated.
In certain situations, this can be a convenient mechanism for loading and/or distributing
your project. For example, you can prepare a concatenated fasl file, email it to a colleague,
and the colleague needs only to load this file. Thus, there is no chance of confusion about
how (e.g. in what order) to load individual files.

2.9.4

Creating an Image File

Loading Lisp source files or compiled binary files into memory at the start of each session
can become slow and awkward for large programs.
An alternative approach is to build an image file. This entails starting a CL process,
loading all application files into the running CL process, then creating a new “CL Image.”
In Allegro CL, this is done by invoking the function:

18

CHAPTER 2. OPERATING A CL DEVELOPMENT ENVIRONMENT

excl:dumplisp
The Image File is typically named with a .dxl extension. Once you have created an Image
File, you can start a session simply by starting CL with that Image File, as follows:
lisp -I <image-file-name>.dxl
You must then compile/load any files that have changed since the Image File was created.
Starting CL with the Image File has the same effect as starting a base CL and loading all
your application files, but is simpler and much faster.

2.9.5

Building Runtime Images

CL Image files built with excl:dumplisp contain the entire CL environment, and may
also contain information which is specific to a certain machine or certain site. Allegro CL
also provides a mechanism for building runtime images, which contain only the information needed to run an end-user application. These Runtime Images can be packaged and
distributed just as with any stand-alone software application.
Preparing runtime images for different platforms (e.g. Sun Solaris, Linux, Windows,
etc.) is usually a simple matter of running a compilation and building a runtime image for
the target platform — application source code typically remains identical among different
platforms. See your platform-specific documentation for details about creating Runtime
Images.

2.9.6

Using an Application Init File

In some cases, for a finished production application, you will want the CL process to execute certain commands upon startup (for example, loading some data from a database).
However, you do not want to depend on the end-user having the appropriate .clinit.cl
in his or her home directory. For such cases, you can specify a different location for the init
file, which can be installed along with the application itself. One way to accomplish this is
with the -q command-line argument to the lisp command. This argument will tell CL to
look in the current directory, rather than in the default location, for the init file.
To use this technique, you would set up an “application home” directory, place the
application’s .clinit.cl file in this directory, and start the application with a production
startup script. This script would first change into the “application home” directory, then
start CL with your application’s image file, or start a Runtime Image. Such a startup script
might look like this:
cd $MYAPP_HOME
lisp -I image.dxl -q
In this example, both the application’s init file and the application’s image file would be
housed in the directory named by the environment variable $MYAPP HOME.

Chapter 3

The CL Language
If you are new to the CL language, we recommend that you supplement this chapter with
other resources. See the Bibliography in Appendix B for some suggestions. The Bibliography also lists two interactive online CL tutorials from Tulane University and Texas A&M,
which you may wish to visit.
In the meantime, this chapter will provide a condensed overview of the language. Please
note, however, that this book is intended as a summary, and will not delve into some of the
more subtle and powerful techniques possible with Common Lisp.

3.1

Overview of CL and its Syntax

The first thing you should observe about CL (and most languages in the Lisp family) is
that it uses a generalized prefix notation.
One of the most frequent actions in a CL program, or at the toplevel read-eval-print
loop, is to call a function. This is most often done by writing an expression which names
the function, followed by its arguments. Here is an example:
(+ 2 2)
This expression consists of the function named by the symbol “+,” followed by the arguments
2 and another 2. As you may have guessed, when this expression is evaluated it will return
the value 4.
Try it: Try typing this expression at your command prompt, and see the return-value
being printed on the console.
What is it that is actually happening here? When CL is asked to evaluate an expression (as in the toplevel read-eval-print loop), it evaluates the expression according to the
following rules:
1. If the expression is a number (i.e. looks like a number), it simply evaluates to itself (a
number):
CL-USER(8): 99
99
19

20

CHAPTER 3. THE CL LANGUAGE
2. If the expression looks like a string (i.e. is surrounded by double-quotes), it also simply
evaluates to itself:
CL-USER(9): "Be braver -- you can’t cross a chasm in two small jumps."
"Be braver -- you can’t cross a chasm in two small jumps."
3. If the expression looks like a literal symbol, it will simply evaluate to that symbol
(more on this in Section 3.1.4):
CL-USER(12): ’my-symbol
MY-SYMBOL
4. If the expression looks like a list (i.e. is surrounded by parentheses), CL assumes that
the first element in this list is a symbol which names a function or a macro, and
the rest of the elements in the list represent the arguments to the function or macro.
(We will discuss functions first, macros later). A function can take zero or more
arguments, and can return zero or more return-values. Often a function only returns
one return-value:
CL-USER(14): (expt 2 5)
32

Try it: Try typing the following functional expressions at your command prompt, and
convince yourself that the printed return-values make sense:

(+ 2 2)
(+ 2)
2
(+)
(+ (+ 2 2) (+ 3 3))
(+ (+ 2 2))
(sys:user-name)
(user-homedir-pathname)
(get-universal-time) ;; returns number of seconds since January 1, 1900

3.1. OVERVIEW OF CL AND ITS SYNTAX

21

(search "Dr." "I am Dr. Strangelove")
"I am Dr. Strangelove"
(subseq "I am Dr. Strangelove"
(search "Dr." "I am Dr. Strangelove"))

3.1.1

Evaluation of Arguments to a Function

Note that the arguments to a function can themselves be any kind of the above expressions.
They are evaluated in order, from left to right, and finally they are passed to the function
for it to evaluate. This kind of nesting can be arbitrarily deep. Do not be concerned about
getting lost in an ocean of parentheses — most serious text editors can handle deeply nested
parentheses with ease, and moreover will automatically indent your expressions so that you,
as the programmer, never have to concern yourself with matching parentheses.

3.1.2

Lisp Syntax Simplicity

One of the nicest things about Lisp is the simplicity and consistency of its syntax. What
we have covered so far pertaining to the evaluation of arguments to functions is just about
everything you need to know about the syntax. The rules for macros are very similar, with
the primary difference being that all the arguments of a macro are not necessarily evaluated
at the time the macro is called — they can be transformed into something else first.
This simple consistent syntax is a welcome relief from other widely used languages such
as C, C++, Java, Perl, and Python. These languages claim to use a “more natural” “infix”
syntax, but in reality their syntax is a confused mixture of prefix, infix, and postfix. They
use infix only for the simplest arithmetic operations such as
2 + 2
and
3 * 13
and use postfix in certain cases such as
i++
and
char*
But mostly they actually use a prefix syntax much the same as Lisp, as in:
split(@array, ":");
So, if you have been programming in these other languages, you have been using prefix
notation all along, but may not have noticed it. If you look at it this way, the prefix
notation of Lisp will seem much less alien to you.

22

3.1.3

CHAPTER 3. THE CL LANGUAGE

Turning Off Evaluation

Sometimes you want to specify an expression at the toplevel read-eval-print loop, or inside
a program, but you do not want CL to evaluate this expression. You want just the literal
expression. CL provides the special operator quote for this purpose. For example,

(quote a)
will return the literal symbol A. Note that, by default, the CL reader will convert all symbol
names to uppercase at the time the symbol is read into the system. Note also that symbols,
if evaluated “as is,” (i.e. without the quote), will by default behave as variables, and CL
will attempt to retrieve an associated value. More on this later.
Here is another example of using quote to return a literal expression:

(quote (+ 1 2))
Unlike evaluating a list expression “as is,” this will return the literal list (+ 1 2). This list
may now be accessed as a normal list data structure1 .
Common Lisp defines the abbreviation ’ as a shorthand way to “wrap” an expression
in a call to quote. So the previous examples could equivalently be written as:

’a
’(+ 1 2)

3.1.4

Fundamental CL Data Types

Common Lisp natively supports many data types common to other languages, such as
numbers, strings, and arrays. Native to CL is also a set of types which you may not have
come across in other languages, such as lists, symbols, and hash tables. In this overview we
will touch on numbers, strings, symbols, and lists. Later in the book we will provide more
detail on some CL-specific data types.
Regarding data types, CL follows a paradigm called dynamic typing. Basically this
means that values have type, but variables do not necessarily have type, and typically
variables are not “pre-declared” to be of a particular type.
Numbers
Numbers in CL form a hierarchy of types, which includes Integers, Ratios, Floating Point,
and Complex numbers. For many purposes you only need to think of a value as a “number,”
without getting any more specific than that. Most arithmetic operations, such as +, -, *,
1
It should be noted, however, that when a quoted literal list is created like this, it is now technically a
constant — you should not modify it with destructive operators.

3.1. OVERVIEW OF CL AND ITS SYNTAX

23

/, etc, will automaticaly do any necessary type coercion on their arguments and will return
a number of the appropriate type.
CL supports a full range of floating-point decimal numbers, as well as true Ratios,
which means that 1/3 is a true one-third, not 0.333333333 rounded off at some arbitrary
precision.
As we have seen, numbers in CL are a native data type which simply evaluate to themselves when entered at the toplevel or included in an expression.
Strings
Strings are actually a specialized kind of Array, namely a one-dimensional array (vector)
made up of characters. These characters can be letters, numbers, or punctuation, and in
some cases can include characters from international character sets (e.g. Unicode) such as
Chinese Hanzi or Japanese Kanji. The string delimiter in CL is the double-quote character
(").
As we have seen, strings in CL are a native data type which simply evaluate to themselves
when included in an expression.
Symbols
Symbols are such an important data structure in CL that people sometimes refer to CL
as a “Symbolic Computing Language.” Symbols are a type of CL object which provides
your program with a built-in mechanism to store and retrieve values and functions, as well
as being useful in their own right. A symbol is most often known by its name (actually a
string), but in fact there is much more to a symbol than its name. In addition to the name,
symbols also contain a function slot, a value slot, and an open-ended Property-list slot in
which you can store an arbitrary number of named properties.
For a named function such as +, the function-slot of the symbol + contains the actual
function itself. The value-slot of a symbol can contain any value, allowing the symbol to
act as a global variable, or Parameter. And the Property-list, or plist slot, can contain an
arbitrary amount of information.
This separation of the symbol data structure into function, value, and plist slots is one
obvious distinction between Common Lisp and most other Lisp dialects. Most other dialects
allow only one (1) “thing” to be stored in the symbol data structure, other than its name
(e.g. either a function or a value, but not both at the same time). Because Common Lisp
does not impose this restriction, it is not necessary to contrive names, for example for your
variables, to avoid conflicting with existing “reserved words” in the system. For example,
“list” is the name of a built-in function in CL. But you may freely use “list” as a variable
as well. There is no need to contrive arbitrary abbreviations such as “lst.”
How symbols are evaluated depends on where they occur in an expression. As we have
seen, if a symbol appears first in a list expression, as with the + in (+ 2 2), the symbol is
evaluated for its function slot. If the first element of an expression is a symbol which indeed
does contain a function in its function slot, then any symbol which appears anywhere else
(i.e. in the rest) of the expression is taken as a variable, and it is evaluated for its global or
local value, depending on its scope. More on variables and scope later.

24

CHAPTER 3. THE CL LANGUAGE

As noted in Section 3.1.3, if you want a literal symbol itself, one way to achieve this is
to “quote” the symbol name:
’a
Another way is for the symbol to appear within a quoted list expression:
’(a b c)
’(a (b c) d)
Note that the quote (’) applies across everything in the list expression, including any subexpressions.
Lists
Lisp takes its name from its strong support for the list data structure. The list concept
is important to CL for more than this reason alone — lists are important because all CL
programs themselves are lists! Having the list as a native data structure, as well as the form
of all programs, means that it is especially straightforward for CL programs to compute
and generate other CL programs. Likewise, CL programs can read and manipulate other
CL programs in a natural manner. This cannot be said of most other languages, and is one
of the primary distinguishing characteristics of CL.
Textually, a list is defined as zero or more elements surrounded by parentheses. The
elements can be objects of any valid CL data types, such as numbers, strings, symbols, lists,
or other kinds of objects. As we have seen, you must quote a literal list to evaluate it or
CL will assume you are calling a function.
Now look at the following list:
(defun hello () (write-string "Hello, World!"))
This list also happens to be a valid CL program (function definition, in this case). Don’t
worry about analyzing the function right now, but do take a few moments to convince
yourself that it meets the requirements for a list. What are the types of the elements in
this list?
In addition to using the quote (’) to produce a literal list, another way to produce a
list is to call the function list. The function list takes any number of arguments, and
returns a list made up from the result of evaluating each argument (as with all functions,
the arguments to the list function get evaluated, from left to right, before being passed
into the function). For example,
(list ’a ’b (+ 2 2))
will return the list (A B 4). The two quoted symbols evaluate to symbols, and the function
call (+ 2 2) evaluates to the number 4.

3.1. OVERVIEW OF CL AND ITS SYNTAX

3.1.5

25

Functions

Functions form the basic building block of CL. Here we will give a brief overview of how to
define a function; later we will go into more detail on what a function actually is.
A common way to define named functions in CL is with the macro defun, which stands
for DEFinition of a FUNction. Defun takes as its arguments a symbol, an argument list,
and a body:
(defun my-first-lisp-function ()
(list ’hello ’world))
Because defun is a macro, rather than a function, it does not follow the hard-and-fast rule
that all its arguments are evaluated as expressions — specifically, the symbol which names
the function does not have to be quoted, nor does the argument list. These are taken as a
literal symbol and a literal list, respectively.
Once the function has been defined with defun, you can call it just as you would call
any other function, by wrapping it in parentheses together with its arguments:
CL-USER(56): (my-first-lisp-function)
(HELLO WORLD)
CL-USER(57): (defun square(x)
(* x x))
SQUARE
CL-USER(58): (square 4)
16
Declaring the types of the arguments to a function is not required.

3.1.6

Global and Local Variables

Global Variables
Global variables in CL are usually also known as special variables. They can be established
by calling defparameter or defvar from the read-eval-print loop, or from within a file that
you compile and load into CL:
(defparameter *todays-temp* 90)
(defvar *humidity* 70)
The surrounding asterisks on these symbol names are part of the symbol names themselves,
and have no significance to CL. This is simply a long-standing naming convention for global
variables (i.e. parameters), to make them easy to pick out with the human eye.
Defvar and defparameter differ in one important way: if defvar is evaluated with
a symbol which already has a global value, it will not overwrite it with a new value.
Defparameter, on the other hand, will overwrite it with a new value:

26

CHAPTER 3. THE CL LANGUAGE

CL-USER(4): *todays-temp*
90
CL-USER(5): *todays-humidity*
70
CL-USER(6): (defparameter *todays-temp* 100)
*TODAYS-TEMP*
CL-USER(7): (defvar *todays-humidity* 50)
*TODAYS-HUMIDITY*
CL-USER(8): *todays-temp*
100
CL-USER(9): *todays-humidity*
70
*todays-humidity* did not change because we used defvar, and it already had a previous
value.
The value for any Parameter can always be changed from the toplevel with setq:
CL-USER(11): (setq *todays-humidity* 30)
30
CL-USER(12): *todays-humidity*
30
Although setq will work with new variables, stylistically it should only be used with alreadyestablished variables.
During application development, it often makes sense to use defvar rather than defparameter
for variables whose values might change during the running and testing of the program. This
way, you will not unintentionally reset the value of a parameter if you compile and load the
source file where it is defined.
Local Variables
New local variables can be introduced with the macro let:
CL-USER(13): (let ((a 20)
(b 30))
(+ a b))
50
As seen in the above example, let takes an assignment section and a body. Let is a macro,
rather than a function2 , so it does not follow the hard-and-fast rule that all its arguments
are evaluated. Specifically, the assignment section
2

See Section 3.8.5 for a discussion of how to recognize macros and functions

3.2. THE LIST AS A DATA STRUCTURE

27

((a 20)
(b 30))
is not evaluated (actually, the macro has the effect of transforming this into something else
before the CL evaluator even sees it). Because this assignment section is not evaluated by
the let macro, it does not have to be quoted, like a list which is an argument to a function
would have to be. As you can see, the assignment section is a list of lists, where each
internal list is a pair whose first is a symbol, and whose second is a value (which will be
evaluated). These symbols (a and b) are the local variables, and they are assigned to the
respective values.
The body consists of any number of expressions which come after the assignment section
and before the closing parenthesis of the let statement. The expressions in this body are
evaluated normally, and of course any expression can refer to the value of any of the local
variables simply by referring directly to its symbol. If the body consists of more than one
expression, the final return-value of the body is the return-value of its last expression.
New Local variables in CL are said to have lexical scope, which means that they are
only accessible in the code which is textually contained within the body of the let. The
term lexical is derived from the fact that the behavior of these variables can be determined
simply by reading the text of the source code, and is not affected by what happens during
the program’s execution.
Dynamic scope happens when you basically mix the concept of global parameters and
local let variables. That is, if you use the name of a previously established parameter
inside the assignment section of a let, like this:
(let ((*todays-humidity* 50))
(do-something))
the parameter will have the specified value not only textually within the the body of the
let, but also in any functions which may get called from within the body of the let. The
global value of the parameter in any other contexts will not be affected. Because this scoping
behavior is determined by the runtime “dynamic” execution of the program, we refer to it
as dynamic scope.
Dynamic scope is often used for changing the value of a particular global parameter
only within a particular tree of function calls. Using Dynamic scope, you can accomplish
this without affecting other places in the program which may be referring to the parameter.
Also, you do not have to remember to have your code “reset” the parameter back to a
default global value, since it will automatically “bounce” back to its normal global value.
Dynamic scoping capability is especially useful in a multithreaded CL, i.e., a CL process
which can have many (virtually) simultaneous threads of execution. A parameter can take
on a dynamically scoped value in one thread, without affecting the value of the parameter
in any of the other concurrently running threads.

3.2

The List as a Data Structure

In this section we will present some of the fundamental native CL operators for manipulating
lists as data structures. These include operators for doing things such as:

28

CHAPTER 3. THE CL LANGUAGE
1. finding the length of a list
2. accessing particular members of a list
3. appending multiple lists together to make a new list
4. extracting elements from a list to make a new list

3.2.1

Accessing the Elements of a List

Common Lisp defines the accessor functions first through tenth as a means of accessing
the first ten elements in a list:
CL-USER(5): (first ’(a b c))
A
CL-USER(6): (second ’(a b c))
B
CL-USER(7): (third ’(a b c))
C
For accessing elements in an arbitrary position in the list, you can use the function nth,
which takes an integer and a list as its two arguments:
CL-USER(8): (nth 0 ’(a b c))
A
CL-USER(9): (nth 1 ’(a b c))
B
CL-USER(10): (nth 2 ’(a b c))
C
CL-USER(11): (nth 12 ’(a b c d e f g h i j k l m n o p))
M
Note that nth starts its indexing at zero (0), so (nth 0 ...) is equivalent to (first ...),
and (nth 1 ...) is equivalent to (second ...), etc.

3.2. THE LIST AS A DATA STRUCTURE

3.2.2

29

The “Rest” of the Story

A very common operation in CL is to perform some operation on the first of a list, then
perform the same operation on each first of the rest of the list, repeating the procedure
until the end of the list, i.e. the Empty list, is reached. The function rest is very helpful
in such cases:
CL-USER(59): (rest ’(a b c))
(B C)
As you can see from this example, rest returns a list consisting of all but the first of its
argument.

3.2.3

The Empty List

The symbol NIL is defined by Common Lisp to be equivalent to the Empty List, (). NIL
also has the interesting property that its value is itself, which means that it will always
evaluate to itself, whether or not it is quoted. So NIL, ’NIL, and () all evaluate to the
Empty List, whose default printed representation is NIL:
CL-USER(14): nil
NIL
CL-USER(15): ’nil
NIL
CL-USER(16): ()
NIL
The function null can be used to test whether or not something is the Empty List:
CL-USER(17): (null ’(a b c))
NIL
CL-USER(18): (null nil)
T
CL-USER(19): (null ())
T
As you may have deduced from the above examples, the symbol T is CL’s default representation for “true.” As with NIL, T evaluates to itself.

3.2.4

Are You a List?

To test whether or not a particular object is a list, CL provides the function listp. Like
many other functions which end with the letter “p,” this function takes a single argument
and checks whether it meets certain criteria (in this case, whether it qualifies as a list).
These predicate functions ending in the letter “p” will always return either T or NIL:

30

CHAPTER 3. THE CL LANGUAGE

CL-USER(20): (listp ’(pontiac cadillac chevrolet))
T
CL-USER(21): (listp 99)
NIL
CL-USER(22): (listp nil)
T
Note that (listp nil) returns T, since NIL is indeed a list (albeit the empty one).

3.2.5

The conditional If

Before continuing with a number of other basic list functions, we will cover the macro if,
which allows simple conditionals. If takes three arguments, a test-form, a then-form, and
an else-form. When an if form is evaluated, it first evaluates its test-form. If the form
returns non-NIL, it will evaluate the then-form, else it will evaluate the else-form. The
nesting of multiple if expressions is possible, but not advised; later we will cover other
constructs which are more appropriate for such cases.
Here are some simple examples using the if operator:
CL-USER(2): (if (> 3 4)
"no"
"yes")
"yes"

CL-USER(3): (if (listp ’("Chicago" "Detroit" "Toronto"))
"it is a list"
"it ain’t a list")
"it is a list"
CL-USER(4): (if (listp 99)
"it is a list"
"it ain’t a list")
"it ain’t a list"

3.2.6

Length of a List

Normally you can use the function length to get the number of elements in a list as an
integer:
CL-USER(5): (length ’(gm ford chrysler volkswagen))
4

3.2. THE LIST AS A DATA STRUCTURE

31

CL-USER(6): (length nil)
0
The length function, as with most of Common Lisp, can itself be implemented in CL. Here
is a simplified version of an our-length function which illustrates how length might be
implemented:
(defun our-length (list)
(if (null list)
0
(+ (our-length (rest list)) 1)))
Note that the function uses the symbol list to name its argument, which is perfectly valid
as we discussed in section 3.1.4.
In English, this function says basically: “If the list is empty, its length is zero. Otherwise
its length is one greater than the length of its rest.” As with many functions which operate
on lists, this recursive definition is a natural way to express the length operation.

3.2.7

Member of a List

The member function will help you to determine whether a particular item is an element of
a particular list. Like many similar functions, member uses eql to test for equality, which is
one of the most basic equality functions in CL (see Section 3.8.4 for a discussion of equality
in CL). Eql basically means the two objects must be the same symbol, integer, or actual
object (i.e. the same address in memory).
Member takes two arguments: an item and a list. If the item is not in the list, it returns
NIL. Otherwise, it returns the rest of the list, starting from the found element:
CL-USER(7): (member ’dallas ’(boston san-francisco portland))
NIL
CL-USER(8): (member ’san-francisco ’(boston san-francisco portland))
(SAN-FRANCISCO PORTLAND)
As with length, we could define member using a function definition which is very close to
the English description of what the function is supposed to do3 :
(defun our-member (elem list)
(if (null list)
nil
(if (eql elem (first list))
list
(our-member elem (rest list)))))
3
The use of the “nested” if statement in this example, while functionally correct, is a violation of good
CL style. Later we will learn how to avoid nested if statements.

32

CHAPTER 3. THE CL LANGUAGE

In English, you might read this function to say “If the list is empty, return NIL. Otherwise,
if the desired item is the same as the first of the list, return the entire list. Otherwise, do
the same thing on the rest of the list.”
Note that, for the purposes of any kind of logical operators, returning any non-NIL value
is “as good as” returning T. So the possible return-values of the member function are as good
as returning T or NIL as far as any logical operators are concerned.

3.2.8

Getting Part of a List

Subseq is a common function to use for returning a portion of a list (or actually any type
of Sequence). Subseq takes at least two arguments, a list and an integer indicating the
position to start from. It also takes an optional third argument, an integer indicating the
position to stop. Note that the position indicated by this third argument is not included in
the returned sub-list:
CL-USER(9): (subseq ’(a b c d) 1 3)
(B C)
CL-USER(10): (subseq ’(a b c d) 1 2)
(B)
CL-USER(11): (subseq ’(a b c d) 1)
(B C D)
Note also that the optional third argument in effect defaults to the length of the list.

3.2.9

Appending Lists

The function append takes any number of lists, and returns a new list which results from
appending them together. Like many CL functions, append does not side-effect, that is, it
simply returns a new list as a return-value, but does not modify its arguments in any way:
CL-USER(6): (setq my-slides ’(introduction welcome lists functions))
(INTRODUCTION WELCOME LISTS FUNCTIONS)
CL-USER(7): (append my-slides ’(numbers))
(INTRODUCTION WELCOME LISTS FUNCTIONS NUMBERS)
CL-USER(8): my-slides
(INTRODUCTION WELCOME LISTS FUNCTIONS)
CL-USER(9): (setq my-slides (append my-slides ’(numbers)))
(INTRODUCTION WELCOME LISTS FUNCTIONS NUMBERS)
CL-USER(10): my-slides
(INTRODUCTION WELCOME LISTS FUNCTIONS NUMBERS)

3.2. THE LIST AS A DATA STRUCTURE

33

Note that the simple call to append does not affect the variable my-slides. If we wish
to modify the value of this variable, however, one way to do this is by using setq in
combination with the call to append. Note also the use of setq directly at the toplevel with
a new variable name. For testing and “hacking” at the command prompt, this is acceptable,
but in a finished program all such global variables should really be formally declared with
defparameter or defvar.

3.2.10

Adding Elements to a List

To add a single element to the front of a list, you can use the function cons:
CL-USER(13): (cons ’a ’(b c d))
(A B C D)
Cons is actually the low-level primitive function upon which many of the other list-constructing
functions are built. As you may have guessed, “cons” stands for “CONStruct,” as in “construct a list.” When you read or hear people talking about a CL program doing a lot of
“consing,” they are referring to the program’s behavior of building a lot of list structures,
many of which are transitory and will need to be “freed up” by the automatic memory
management subsystem, or “garbage collector.”
As with append, cons is non-destructive, meaning it does no side-effecting (modification)
to its arguments.

3.2.11

Removing Elements from a List

The function remove takes two arguments, any item and a list, and returns a new list with
all occurences of the item taken out of it:
CL-USER(15): (setq data ’(1 2 3 1000 4))
(1 2 3 1000 4)
CL-USER(16): (remove 1000 data)
(1 2 3 4)
CL-USER(17): data
(1 2 3 1000 4)
CL-USER(18): (setq data (remove 1000 data))
(1 2 3 4)
CL-USER(19): data
(1 2 3 4)
Like append and cons, remove is non-destructive and so does not modify its arguments. As
before, one way to achieve modification with variables is to use setq.

34

3.2.12

CHAPTER 3. THE CL LANGUAGE

Sorting Lists

The function sort will sort any sequence (including, of course, a list), based on comparing
the elements of the sequence using any applicable comparison function, or predicate function.
Because of efficiency reasons, the designers of CL made the decision to allow sort to modify
(“recycle” the memory in) its sequence argument — so you must always “catch” the result
of the sorting by doing something explicitly with the return-value:
CL-USER(20): (setq data ’(1 3 5 7 2 4 6))
(1 3 5 7 2 4 6)
CL-USER(21): (setq data (sort data #’<))
(1 2 3 4 5 6 7)
CL-USER(22): (setq data (sort data #’>))
(7 6 5 4 3 2 1)
Notice that the comparison function must be a function which can take two arguments —
any representative two elements in the given sequence — and compare them to give a T or
NIL result. The “#’” notation is a shorthand way of retrieving the actual function object
associated with a symbol or expression (in this case, the < or > symbol). We will cover more
on this in Section 3.4.
In the above examples, we simply reset the value of the variable data to the result of
the sort, which is a common thing to do. If one wanted to retain the original pre-sorted
sequence, a “safe” version of sort could be defined as follows:
(defun safe-sort (list predicate)
(let ((new-list (copy-list list)))
(sort new-list predicate)))

3.2.13

Treating a List as a Set

The functions union, intersection, and set-difference take two lists and compute the
corresponding set operation. Because mathematical sets have no notion of ordering, the
order of the results returned by these functions is purely arbitrary, so you should never
depend on the results of these set operations being in any particular order:
CL-USER(23): (union ’(1 2 3) ’(2 3 4))
(1 2 3 4)
CL-USER(25): (intersection ’(1 2 3) ’(2 3 4))
(3 2)
CL-USER(26): (set-difference ’(1 2 3 4) ’(2 3))
(4 1)

3.3. CONTROL OF EXECUTION

3.2.14

35

Mapping a Function to a List

If you have one list, and desire another list of the same length, there is a good chance that
you can use one of the mapping functions. Mapcar is the most common of such functions.
Mapcar takes a function and one or more lists, and maps the function across each element
of the list, producing a new resulting list. The term car comes from the original way in
Lisp of referring to the first of a list (“contents of address register”). Therefore mapcar
takes its function and applies it to the first of each successive rest of the list:
CL-USER(29): (defun twice (num)
(* num 2))
TWICE
CL-USER(30): (mapcar #’twice ’(1 2 3 4))
(2 4 6 8)
“Lambda” (unnamed) functions are used very frequently with mapcar and similar mapping
functions. More on this in Section 3.4.3.

3.2.15

Property Lists

Property lists (“plists”) provide a simple yet powerful way to handle keyword-value pairs.
A plist is simply a list, with an even number of elements, where each pair of elements
represents a named value. Here is an example of a plist:

(:michigan "Lansing" :illinois "Springfield"
:pennsylvania "Harrisburg")
In this plist, the keys are keyword symbols, and the values are strings. The keys in a plist
are very often keyword symbols. keyword symbols are symbols whose names are preceded by
a colon (:), and which are generally used just for matching the symbol itself (i.e. typically
they are not used for their symbol-value or symbol-function).
For accessing members of a plist, CL provides the function getf, which takes a plist
and a Key:
CL-USER(34): (getf ’(:michigan "Lansing"
:illinois "Springfield"
:pennsylvania "Harrisburg")
:illinois)
"Springfield"

3.3

Control of Execution

“Control of execution” in CL boils down to “control of evaluation.” Using some standard
and common macros, you control which forms get evaluated, among certain choices.

36

CHAPTER 3. THE CL LANGUAGE

(if (eql status :all-systems-go)
(progn (broadcast-countdown) (flash-colored-lights)
(play-eerie-music)(launch-rocket))
(progn (broadcast-apology) (shut-down-rocket)))
Figure 3.1: Progn used with If

3.3.1

If

Perhaps the most basic operator for imposing control is the if operator, which we have
already introduced briefly in Section 3.2.5. The If operator takes exactly three arguments,
each of which is an expression which may be evaluated. The first is a test-form, the second
is a then-form, and the third is an else-form (which may be left out for a default of NIL).
When CL evaluates the if form, it will first evaluate the test-form. Depending on whether
the return-value of the test-form is non-NIL or NIL, either the then-form or the else-form
will be evaluated.
If you want to group multiple expressions together to occur as part of the then-form or
the else-form, you must wrap them in an enclosing block. Progn is commonly used for this
purpose. Progn will accept any number of forms as arguments, and will evaluate each of
them in turn, finally returning the return-value of the last of them (see Figure 3.1).

3.3.2

When

In some ways, when is similar to if, but you should use when in cases where you know that
the else-clause is simply NIL. This turns out to be a common situation. Another advantage
of using when in these cases is that there is no need for progn to group multiple forms —
when simply takes a Test-form then any number of “Then-forms:”
CL-USER(36): (when (eql database-connection :active)
(read-record-from-database)
(chew-on-data-from-database)
(calculate-final-result))
999
CL-USER(37): (when (> 3 4)
(princ "I don’t think so..."))
NIL
(Note that the database functions above are examples only and are not pre-defined in CL).

3.3.3

Logical Operators

The Logical Operators and and or evaluate one or more expressions given as arguments,
depending on the return-values of the expressions. As we covered in Section 3.2.3, T is

3.3. CONTROL OF EXECUTION

37

CL’s default representation for “True,” NIL (equivalent to the Empty List) is CL’s default
representation for “False,” and any non-NIL value is “as good as” T as far as “Truth” is
concerned:
CL-USER(38): (and (listp ’(chicago detroit boston new-york))
(listp 3.1415))
NIL
CL-USER(39): (or (listp ’(chicago detroit boston new-york))
(listp 3.1415))
T
In fact what is happening here is that the operator and evaluates its arguments (expressions)
one at a time, from left to right, returning NIL as soon as it finds one which returns NIL —
otherwise it will return the value of its last expression. Or evaluates its arguments until it
finds one which returns non-NIL, and returning that non-NIL return-value if it finds one.
Otherwise it will end up returning NIL.
Not takes a single expression as an argument and negates it — that is, if the expression returns NIL, not will return T, and if the argument returns non-NIL, not will return
NIL. Logically, not behaves identically with the function null — but semantically, null
carries the meaning of “testing for an Empty List,” while not carries the meaning of logical
negation:
CL-USER(45): (not NIL)
T
CL-USER(46): (not (listp 3.1415))
T

3.3.4

Cond

Because the use of “nested” if statements is not appropriate, CL provides the macro cond to
accomodate this. Use cond in situations when in other languages you might use a repeated
“If...then...else if...else if...else if...” Cond takes as arguments any number of test-expression
forms. Each of these forms consists of a list whose first is an expression to be evaluated,
and whose rest is any number of expressions which will be evaluated if the first form
evaluates to non-NIL. As soon as a non-NIL form is found, its corresponding expressions
are evaluated and the entire cond is finished (i.e. there is no need for an explicit “break”
statement or anything of that nature):
CL-USER(49): (let ((n 4))
(cond ((> n 10) "It’s a lot")
((= n 10) "It’s kind of a lot")
((< n 10) "It’s not a lot")))
"It’s not a lot"

38

CHAPTER 3. THE CL LANGUAGE

Because T as an expression simply evaluates to itself, a convenient way to specify a “default”
or “catch-all” clause in a cond is to use T as the final conditional expression:
CL-USER(50): (let ((n 4))
(cond ((> n 10) "It’s a lot")
((= n 10) "It’s kind of a lot")
(t "It’s not a lot")))
"It’s not a lot"

3.3.5

Case

If you want to make a decision based on comparing a variable against a known set of
constant values, case is often more concise than cond:
CL-USER(51): (let ((color :red))
(case color
(:blue "Blue is okay")
(:red "Red is actually her favorite color")
(:green "Are you nuts?")))
"Red is actually her favorite color"
Note that case uses eql to do its matching, so it will only work with constants which are
symbols (including keyword symbols), integers, etc. It will not work with strings.
CL also provides the macros ecase and ccase, which work just like case but provide
some automatic error-handling for you in cases when the given value does not match any
of the keys.
The symbol otherwise has special meaning to the case macro — it indicates a “default,”
or “catch-all” case:

CL-USER(52): (let ((color :orange))
(case color
(:blue "Blue is okay")
(:red "Red is actually her favorite color")
(:green "Are you nuts?")
(otherwise "We have never heard of that!")))
"We have never heard of that!"

3.3.6

Iteration

In addition to its innate abilities to do recursion and mapping, CL provides several operators for doing traditional iteration (“looping”). These include macros such as do, dolist,
dotimes, and the “everything-including-the-kitchen-sink” iterating macro, loop. Dolist
and dotimes are two of the most commonly used ones, so we will cover them here:
CL-USER(53): (let ((result 0))
(dolist (elem ’(4 5 6 7) result)

3.4. FUNCTIONS AS OBJECTS

39

(setq result (+ result elem))))
22
As seen above, dolist takes an initialization-list and a body. The Initialization-list consists
of an iterator-variable, a list-form, and an optional return-value-form (which defaults to
NIL). The body will be evaluated once with the iterator-variable set to each element in the
list which is returned by the list-form. When the list has been exhausted, the return-valueform will be evaluated and this value will be returned from the entire dolist. In other
words, the number of iterations of a dolist is driven by the length of the list returned by
its list-form.
Dotimes is similar in many ways to dolist, but instead of a list-form, it takes an
expression which should evaluate to an integer. The body is evaluated once with the iteratorvariable set to each integer starting from zero (0) up to one less than the value of the
Integer-form:
CL-USER(54): (let ((result 1))
(dotimes (n 10 result)
(setq result (+ result n))))
46
In situations when you want the program to exit early from an iteration, you can usually
accomplish this by calling return explicitly.

3.4
3.4.1

Functions as Objects
Named Functions

When working with CL, it is important to understand that a function is actually a kind
of object, separate from any symbol or symbol-name with which it might be associated.
You can access the actual function-object in a symbol’s function-slot by calling the function
symbol-function on a symbol, like this:
(symbol-function ’+)
#<Function +>
or this:
(symbol-function ’list)
#<Function LIST>
As you can see, the printed representation of a named function-object contains the function’s
symbol-name enclosed in pointy brackets, preceded by a pound sign.
You can explicitly call a function-object with the function funcall :
CL-USER(10): (funcall (symbol-function ’+) 1 2)
3

40

CHAPTER 3. THE CL LANGUAGE

CL-USER(11): (setq my-function (symbol-function ’+))
#<Function +>
CL-USER(12): (funcall my-function 1 2)
3
A shorthand way of writing symbol-function is to write #’ before the name of a function:
CL-USER(10): (funcall #’+ 1 2)
3
CL-USER(11): (setq my-function #’+)
#<Function +>
CL-USER(12): (funcall my-function 1 2)
3

3.4.2

Functional Arguments

Functional arguments are arguments to functions which themselves are function-objects.
We have seen this already with mapcar, sort, and funcall:
CL-USER(13): (defun add-3 (num)
(+ num 3))
ADD-3
CL-USER(14): (symbol-function ’add-3)
#<Interpreted Function ADD-3>
CL-USER(15): (mapcar #’add-3 ’(2 3 4))
(5 6 7)
CL-USER(17): (funcall #’add-3 2)
5
The idea of passing functions around to other functions is sometimes referred to as higher-order
functions. CL provides natural support for this concept.

3.4.3

Anonymous Functions

Sometimes a function will be used so seldom that it is hardly worth going to the extent
of creating Named function with defun. For these cases, CL provides the concept of the
lambda, or “anonymous” (unnamed) function. To use a lambda function, you place the
entire function definition in the position where you normally would put just the symbol
which names the function, identified with the special symbol lambda:

3.4. FUNCTIONS AS OBJECTS

41

CL-USER(19): (mapcar #’(lambda(num) (+ num 3))
’(2 3 4))
(5 6 7)
CL-USER(20): (sort ’((4 "Buffy") (2 "Keiko") (1 "Judy") (3 "Aruna"))
#’(lambda(x y)
(< (first x) (first y))))
((1 "Judy") (2 "Keiko") (3 "Aruna") (4 "Buffy"))

3.4.4

Optional Arguments

Optional arguments to a function must be specified after any required arguments, and are
identified with the symbol &optional in the argument list. Each optional argument can be
either a symbol or a list containing a symbol and an expression returning a default value.
In the case of a symbol, the default value is taken to be NIL:
CL-USER(23): (defun greeting-list (&optional (username "Jake"))
(list ’hello ’there username))
GREETING-LIST
CL-USER(24): (greeting-list)
(HELLO THERE "Jake")
CL-USER(25): (greeting-list "Joe")
(HELLO THERE "Joe")

3.4.5

Keyword Arguments

Keyword arguments to a function are basically named optional arguments. They are defined
much the same as optional arguments:
(defun greeting-list (&key (username "Jake")
(greeting "how are you?"))
(list ’hello ’there username greeting))
But when you call a function with keyword arguments, you “splice in” what amounts to a
plist containing keyword symbols for the names of the arguments along with their values:
CL-USER(27): (greeting-list :greeting "how have you been?")
(HELLO THERE "Jake" "how have you been?")
CL-USER(28): (greeting-list :greeting "how have you been?" :username "Joe")
(HELLO THERE "Joe" "how have you been?")
Note that with keyword arguments, you use a normal symbol (i.e. without the preceding
colon) in the definition of the function, but a keyword symbol as the “tag” when calling
the function. There is a logical reason behind this, which you will understand soon.

42

CHAPTER 3. THE CL LANGUAGE

3.5

Input, Output, Streams, and Strings

Input and Output in CL is done through objects called streams. A stream is a source
or destination for pieces of data which generally is connected to some physical source or
destination, such as a terminal console window, a file on the computer’s hard disk, or a web
browser. Each thread of execution in CL defines global parameters which represent some
standard Streams:
• *standard-input* is the default data source.
• *standard-output* is the default data destination.
• *terminal-io* is bound to the console (command prompt) window, and by default
this is identical to both *standard-input* and *standard-output*.

3.5.1

Read

Read is the standard mechanism for reading data items into CL. Read takes a single optional
argument for its stream, and reads a single data item from that stream. The default for the
stream is *standard-input*. If you simply type (read) at the command prompt, CL will
wait for you to enter some valid Lisp item, and will return this item. Read simply reads, it
does not evaluate, so there is no need to quote the data being read. You can set a variable
to the result of a call to read:
(setq myvar (read))
The function read-line is similar to read, but will read characters until it encounters
a new line or end-of-file, and will return these characters in the form of a string. Read-line
is appropriate for reading data from files which are in a “line-based” format rather than
formatted as Lisp expressions.

3.5.2

Print and Prin1

Print and prin1 each take exactly one CL object and output its printed representation to
a stream, which defaults to *standard-output*. Print puts a space and a newline after
the item (effectively separating individual items with whitespace), and Prin1 outputs the
item with no extra whitespace:
CL-USER(29): (print ’hello)
HELLO
HELLO
In the above example, you see the symbol HELLO appearing twice: the first time is the
output actually being printed on the console, and the second is the normal return-value of
the call to print being printed on the console by the read-eval-print loop. Print and Prin1
both print their output readably, meaning that CL’s read function will be able to read the
data back in. For example, the double-quotes of a string will be included in the output:

3.5. INPUT, OUTPUT, STREAMS, AND STRINGS

43

CL-USER(30): (print "hello")
"hello"
"hello"

3.5.3

Princ

Princ, as distinct from print and prin1, prints its output in more of a human-readable
format, which means for example that the double-quotes of a string will be stripped upon
output:
CL-USER(31): (princ "hello")
hello
"hello"
in the above example, as before, we see the output twice: once when it is actually printed
to the console, and again when the return-value of the call to princ is printed by the
read-eval-print loop.

3.5.4

Format

Format is a powerful generalization of prin1, princ, and other basic printing functions, and
can be used for almost all output. Format takes a Stream, a control-string, and optionally
any number of additional arguments to fill in placeholders in the control-string:
CL-USER(33): (format t "Hello There ~a" ’bob)
Hello There BOB
NIL
In the above example, t is used as a shorthand way of specifying *standard-output* as the
stream, and ~a is a control-string placeholder which processes its argument as if by princ.
If you specify either an actual stream or T as the second argument to format, it will print
by side-effect and return NIL as its return-value. However, if you specify NIL as the second
argument, format does not output to any stream at all by side-effect, but instead returns
a string containing what would have been output to a stream if one had been provided:
CL-USER(34): (format nil "Hello There ~a" ’bob)
"Hello There BOB"
In addition to ~a, there is a wealth of other format directives for an extensive choice of
output, such as outputting items as if by prin1, outputting lists with a certain character
after all but the last item, outputting numbers in many different formats including Roman
numerals and English words, etc.

44

CHAPTER 3. THE CL LANGUAGE

3.5.5

Pathnames

In order to name directories and files on a computer filesystem in an operating-system
independent manner, CL provides the pathname system. CL pathnames do not contain
system-specific pathname strings such as the forward-slash of Unix/Linux filenames or the
backward-slash of MSDOS/Windows filenames. A CL pathname is basically a structure
which supports a constructor function, make-pathname, and several accessor functions, such
as pathname-name, pathname-directory, and pathname-type. The pathname structure
contains six slots:
• Directory
• Name
• Type
• Device
• Version
• Host
On standard Unix filesystems, only the directory, name, and type slots are relevant. On
MSDOS/Windows filesystems, the device slot is also relevant. On the Symbolics platform,
which has a more advanced filesystem, the version and host slots also have relevance.
The constructor function, make-pathname, takes keyword arguments corresponding to
the six possible slots of the pathname.
The following is an example of creating a pathname which would appear on Unix as
/tmp/try.txt:
CL-USER(15): (defparameter *my-path*
(make-pathname :directory (list :absolute "tmp")
:name "try"
:type "txt"))
*MY-PATH*
CL-USER(16): *my-path*
#p"/tmp/readme.txt"
As seen in the above example, the :directory slot is technically a list, whose first is
a keyword symbol (either :absolute or :relative), and whose rest is the individual
directory components represented as strings.

3.5.6

File Input and Output

Doing file input and output in CL is essentially a matter of combining the concepts of
streams and pathnames. You must first open a file, which entails associating a stream
with the file. CL does provide the open function, which directly opens a file and associates a stream to it. For a number of reasons, however, generally you will use a macro

3.6. HASH TABLES, ARRAYS, STRUCTURES, AND CLASSES

45

called with-open-file which not only opens the file, but automatically closes it when
you are done with it, and performs additional cleanup and error-handling details for you.
With-open-file takes a specification-list and a body. The specification-list consists of a
stream-variable (a symbol) and a pathname, followed by several optional keyword arguments which specify, for example, whether the file is for input, output, or both. Here is
an example of opening a file and reading one item from it (note that the file denoted by
*mypath* must exist, or an error will be signalled. You can use the function probe-file
to test for a file’s existence.):
(with-open-file (input-stream *my-path*)
(read input-stream))
No extra keyword arguments are required in the specification-list in this example, since
“read-only” mode is the default for opening a file. In order to write to a file, however,
additional keyword arguments must be given:
(with-open-file (output-stream *my-path*
:direction :output
:if-exists :supersede
:if-does-not-exist :create)
(format output-stream "Hello There"))

3.6

Hash Tables, Arrays, Structures, and Classes

CL provides several other built-in data structures to meet a wide variety of needs. Arrays
and hash tables support the storage of values in a manner similar to lists and plists, but
with much faster look-up speed for large data sets. Structures provide another plist-like
construct, but more efficient and structured.
Classes extend the idea of structures to provide a full object-oriented programming
paradigm supporting multiple inheritance, and methods which can dispatch based on any
number of specialized arguments (“multimethods”). This collection of features in CL is
called the Common Lisp Object System, or CLOS.

3.6.1

Hash Tables

A hash table is similar to what is known in some other languages as an associative array. It
is comparable to a single-dimensional array which supports keys which are any arbitrary CL
object, which will match using eql, e.g. symbols or integers. Hash tables support virtually
constant-time lookup of data, which means that the time it takes to look up an item in a
hash table will remain stable, even as the number of entries in the hash table increases.
To work with hash tables, CL provides the constructor function make-hash-table and
the accessor function gethash. In order to set entries in a hash table, use the setf operator
in conjunction with the gethash accessor function. Setf is a generalization of setq which
can be used with places resulting from a function call, in addition to exclusively with
symbols as with setq. Here is an example of creating a hash table, putting a value in it,
and retrieving the value from it:

46

CHAPTER 3. THE CL LANGUAGE

CL-USER(35): (setq os-ratings (make-hash-table))
#<EQL hash-table with 0 entries @ #x209cf822>
CL-USER(36): (setf (gethash :windows os-ratings) :no-comment)
:NO-COMMENT
CL-USER(37): (gethash :windows os-ratings)
:NO-COMMENT
T
Note that the gethash function returns two values. This is the first example we have
seen of a function which returns more than one value. The first return-value is the value
corresponding to the found entry in the hash table, if it exists. The second return-value
is a Boolean value (i.e. T or NIL), indicating whether the value was actually found in the
hash table. If the value is not found, the first return-value will be NIL and the second
return-value will also be NIL. This second value is necessary to distinguish from cases when
the entry is found in the hash table, but its value just happens to be NIL.
There are several ways to access multiple return-values from a function, e.g. by using
multiple-value-bind and multiple-value-list.

3.6.2

Arrays

Arrays are data structures which can hold single- or multi-dimensional “grids” full of values,
which are indexed by one or more integers. Like hash tables, arrays support very fast access
of the data based on the integer indices. You create arrays using the constructor function
make-array, and refer to elements of the Array using the accessor function aref. As with
hash tables, you can set individual items in the array using setf:
CL-USER(38): (setq my-array (make-array (list 3 3)))
#2A((NIL NIL NIL) (NIL NIL NIL) (NIL NIL NIL))
CL-USER(39): (setf (aref my-array 0 0) "Dave")
"Dave"
CL-USER(40): (setf (aref my-array 0 1) "John")
"John"
CL-USER(41): (aref my-array 0 0)
"Dave"
CL-USER(42): (aref my-array 0 1)
"John"
The make-array function also supports a myriad of optional keyword arguments for initializing the array at the time of creation.

3.6. HASH TABLES, ARRAYS, STRUCTURES, AND CLASSES

3.6.3

47

Structures

Structures support creating your own data structures with named slots, similar to, for example, the pathname data structure which we have seen. Structures are defined with the macro
defstruct, which supports many options for setting up the instance constructor functions,
slots, and accessor functions of the structure. For a complete treatment of structures, we
refer you to one of the CL references or to the on-line documentation for defstruct.

3.6.4

Classes and Methods

Classes and methods form the core of the Common Lisp Object System (CLOS), and add
full-blown object-oriented capabilities to Common Lisp.
A complete treatment of CLOS is beyond the scope of this book, but we will provide
a brief overview. Classes are basically “blueprints” or “prototypes” for objects. objects,
in this context, are a kind of structure for grouping together both data and computational
functionality. Methods are very similar to functions, but they can be specialized to apply
to particular combinations of object types.
CLOS is a generic function object-oriented system, which means that methods exist
independently from classes (i.e. methods do not “belong to” classes). Methods can take
different combinations of class instance types as arguments, however, and can be specialized
based upon the particular combinations of types of their arguments. Simple CL datatypes,
such as strings and numbers, are also considered as classes.
Here is an example of defining a class and a method, making an instance of the class,
and calling the method using the instance as the argument:
CL-USER(43): (defclass

box () ((length :initform 10)
(width :initform 10)
(height :initform 10)))

#<STANDARD-CLASS BOX>
CL-USER(44): (defmethod volume ((self box))
(* (slot-value self ’length)
(slot-value self ’width)
(slot-value self ’height)))
#<STANDARD-METHOD VOLUME (BOX)>
CL-USER(45): (setq mybox (make-instance ’box))
#<BOX @ #x209d135a>
CL-USER(46): (volume mybox)
1000
As seen in the above example, the class definition takes at least three arguments: a symbol
identifying the name of the class, a mixin-list which names superclasses (none, in this
example), and a list of slots for the class, with default values specified by the :initform
keyword argument.

48

CHAPTER 3. THE CL LANGUAGE

The method definition is very much like a function definition, but its arguments (single
argument in the case of this example) are specialized to a particular class type. When we
call the method, we pass an instance of the class as an argument, using a normal argument
list. This is unlike Java or C++, where a method is considered to be “of” a particular class
instance, and only additional arguments are passed in an argument list.
CL is more consistent in this regard.
Because normal CL data types are also considered to be classes, we can define methods
which specialize on them as well:
CL-USER(47): (defmethod add ((value1 string) (value2 string))
(concatenate ’string value1 value2))
#<STANDARD-METHOD ADD (STRING STRING)>
CL-USER(48): (defmethod add ((value1 number) (value2 number))
(+ value1 value2))
#<STANDARD-METHOD ADD (NUMBER NUMBER)>
CL-USER(49): (add "hello " "there")
"hello there"
CL-USER(50): (add 3 4)
7

3.7

Packages

Typically, symbols in CL exist within a particular package. You can think of packages as
“area codes” for symbols. They are namespaces used within CL to help avoid name clashes.
Package names are generally represented by keyword symbols (i.e. symbols whose names
are preceded by a colon). ANSI CL has a “flat” package system, which means that packages
do not “contain” other packages in a hierarchical fashion, but you can achieve essentially
the same effect by “layering” packages. Actually, hierarchical packages have been added to
Allegro CL as an extension to the ANSI Standard, and may be added to the ANSI standard
at some point. However, the real use of hierarchical packages is to avoid name clashes with
package names themselves. You can do this in any case simply by separating the components
of a package name with a delimiter, for example, a dot: :com.genworks.books.
In a given CL session, CL always has a notion of the current package which is stored
in the (dynamic) variable *package*. It is important always to be aware of what the
current package is, because if you are in a different package than you think you are, you
can get very surprising results. For example, functions that are defined in :package-1 will
not necessarily be visible in :package-2, so if you attempt to run those functions from
:package-2, CL will think they are undefined.
Normally the CL command prompt prints the name of the current package. You can
move to a different package, e.g. foo, with the toplevel comand :package :foo.
Small programs can be developed in the :user package, but in general, a large application should be developed in its own package.

3.8. COMMON STUMBLING BLOCKS

3.7.1

49

Importing and Exporting Symbols

Packages can export symbols to make them accessible from other packages, and import
symbols to make them appear to be local to the package.
If a symbol is exported from one package but not imported into the current package, it
is still valid to refer to it. In order to refer to symbols in other packages, qualify the symbol
with the package name and a single colon: package-2:foo.
If a symbol is not exported from the outside package, you can still refer to it, but this
would represent a style violation, as signified by the fact that you have to use a double-colon:
package-2::bar.

3.7.2

The Keyword Package

We have seen many occurences of symbols whose names are preceded by a colon (:). These
symbols actually reside within the :keyword package, and the colon is just a shorthand way
of writing and printing them. Keyword symbols are generally used for enumerated values
and to name (keyword) function arguments. They are “immune” to package (i.e. there is
no possibility of confusion about which package they are in), which makes them especially
convenient for these purposes.

3.8

Common Stumbling Blocks

This section discusses common errors and stumbling blocks that new CL users often encounter.

3.8.1

Quotes

One common pitfall is not understanding when to quote expressions. Quoting a single
symbol simply returns the symbol, whereas without the quotes, the symbol is treated as a
variable:
CL-USER(6): (setq x 1)
1
CL-USER(7): x
1
CL-USER(8): ’x
X
Quoting a list returns the list, whereas without the quotes, the list is treated as a function
(or macro) call:
CL-USER(10): ’(+ 1 2)
(+ 1 2)

50

CHAPTER 3. THE CL LANGUAGE

CL-USER(11): (first ’(+ 1 2))
+
CL-USER(12): (+ 1 2)
3
CL-USER(13): (first (+ 1 2))
Error: Attempt to take the car of 3 which is not listp.

3.8.2

Function Argument Lists

When you define a function with defun, the arguments go inside their own list:
CL-USER(19): (defun square (num)
(* num num))
SQUARE
But when you call the function, the argument list comes spliced in directly after the function
name:
CL-USER(20): (square 4)
16
A common mistake is to put the argument list into its own list when calling the function,
like this:
CL-USER(21) (square (4))
Error: Funcall of 4 which is a non-function.
If you are used to programming in Perl or Java, you will likely do this occasionally while
you are getting used to CL syntax.

3.8.3

Symbols vs. Strings

Another point of confusion is the difference between symbols and strings. The symbol AAA
and the string "AAA" are treated in completely different ways by CL. Symbols reside in
packages; strings do not. Moreover, all references to a given symbol in a given package
refer to the same actual address in memory — a given symbol in a given package is only
allocated once. In contrast, CL might allocate new memory whenever the user defines a
string, so multiple copies of the same string of characters could occur multiple times in
memory. Consider this example:
CL-USER(17): (setq a1 ’aaa)
AAA
CL-USER(18): (setq a2 ’aaa)
AAA

3.8. COMMON STUMBLING BLOCKS

51

CL-USER(19): (eql a1 a2)
T

EQL compares actual pointers or integers — A1 and A2 both point to the same symbol
in the same part of memory.
CL-USER(20): (setq b1 "bbb")
"bbb"
CL-USER(21): (setq b2 "bbb")
"bbb"
CL-USER(22): (eql b1 b2)
NIL
CL-USER(23): (string-equal b1 b2)
T
Again, EQL compares pointers, but here B1 and B2 point to different memory addresses,
although those addresses happen to contain the same string. Therefore the comparison by
string-equal, which compares strings character-by-character, rather than the pointers to
those strings, returns T.
Another difference between symbols and strings is that CL provides a number of operations for manipulating strings that do not work for symbols:
CL-USER(25): (setq c "Jane dates only Lisp programmers")
"Jane dates only Lisp programmers"
CL-USER(26): (char c 3)
#\e
CL-USER(27): (char ’xyz 1)
Error: ‘XYZ’ is not of the expected type ‘STRING’
CL-USER(29): (subseq c 0 4)
"jane"
CL-USER(30): (subseq c (position #\space c))
" dates only lisp programmers"
CL-USER(31): (subseq c 11 (length c))
"only lisp programmers"

52

CHAPTER 3. THE CL LANGUAGE

(defun nearly-equal? (num1 num2 &optional (tolerance *zero-epsilon*))
(< (abs (- num1 num2)) tolerance))
Figure 3.2: Function for Floating-point Comparison

3.8.4

Equality

CL has several different predicates for testing equality. This is demanded because of the
many different object types in CL, which dictate different notions of equality.
A simple rule of thumb is:
• Use eql to compare symbols, pointers (such as pointers to lists), and integers
• Use = to compare most numeric values
• Use string-equal for case-insensitive string comparison
• Use string= for case-sensitive string comparison
• Use equal to compare other types of objects (such as lists and single characters)
• For comparing floating-point numbers which may differ by a very small margin, but
should still be considered equal, the safest technique is to take the absolute value
of their difference, and compare this with an “epsilon” value within a tolerance of
zero. In Figure 3.2 is a simple function which does this, which assumes that a global
parameter is defined, *zero-epsilon*, as a numeric value very close to zero:
Examples:
CL-USER(33): (setq x :foo)
:FOO
CL-USER(34): (eql x :foo)
T
CL-USER(35): (= 100 (* 20 5))
T
CL-USER(36): (string-equal "xyz" "XYZ")
T
CL-USER(37): (setq abc ’(a b c))
(A B C)
CL-USER(38): (equal abc (cons ’a ’(b c)))
T

3.8. COMMON STUMBLING BLOCKS

53

CL-USER(39): (equal #\a (char "abc" 0))
T
CL-USER(40): (eql abc (cons ’a ’(b c)))
NIL
The last expression returns NIL because, although the two lists have the same contents, they
reside in different places in memory, and therefore the pointers to those lists are different.
Note that several of the functions which by default use eql for matching will take an
optional keyword argument test which can override this default:
CL-USER(54): (member "blue" ’("red" "blue" "green" "yellow")
:test #’string-equal)
("blue" "green" "yellow")
Recall from Section 3.2.7 that member returns the rest of the list starting at the found
element.

3.8.5

Distinguishing Macros from Functions

Since macros4 and functions both occur as the first element in an expression, at first
glance it may appear difficult to distinguish them. In practice, this rarely becomes an issue.
For beginners in the language, the best approach is to make a default assumption that
something is a function unless you recognize it as a macro. Most macros will be readily
identifiable because one of the following will apply:
• It is a familiar core part of CL, such as if, cond, let, setq, setf, etc.
• Its name begins with “def,” such as with defun, defparameter, etc.
• Its name begins with “with-,” as in with-open-file, with-output-to-string, etc.
• Its name begins with “do,” for example dolist and dotimes.
Beyond these general rules of thumb, if you find yourself in doubt about whether something is a function or a macro, you can:
1. Check with on-line reference documentation or in a CL reference book (see the Bibliography in Appendix B)
2. Use the symbol-function function to ask your CL session, like this:
CL-USER(24): (symbol-function ’list)
#<Function LIST>
CL-USER(25): (symbol-function ’with-open-file)
#<macro WITH-OPEN-FILE @ #x2000b07a>
4
Special Operators are one other category of operator in CL, which internally are implemented differently
from macros, but which for our purposes we can consider like macros.

54

3.8.6

CHAPTER 3. THE CL LANGUAGE

Operations that cons

Because CL programs do not have to free and allocate memory explicitly, they can be
written much more quickly. As an application evolves, some operations which “cons” can
be replaced by operations which “recycle” memory, reducing the amount of work which the
garbage collector (CL’s automatic memory management subsystem) has to do.
In order to write programs which conserve on garbage collecting, one technique is simply to avoid the unnecessary use of operations which cons. This ability will come with
experience.
Another technique is to use destructive operations, which are best used only after one
has gained more experience working with CL. Destructive operations are for the most part
beyond the scope of this book.

Chapter 4

Interfaces
One of the strong points of modern commercially-supported CL environments is their flexibility in working with other environments. In this chapter we present a sampling of some
interfaces and packages which come in handy for developing real-world CL applications and
connecting them to the outside environment. Some of the packages covered in this chapter
are commercial extensions to Common Lisp as provided with Franz Inc’s Allegro CL, and
some of them are available as open-source offerings.

4.1

Interfacing with the Operating System

One of the most basic means of interacting with the outside world is simply to invoke
commands at the operating system level, in similar fashion to Perl’s “system” command.
In Allegro CL, one way to accomplish this is with the function excl.osi:command-output1 .
This command takes a string which represents a command (possibly with arguments) to be
executed through an operating system shell, such as the “C” shell on Unix:
(excl.osi:command-output "ls")
When invoked in this fashion, this function returns the the Standard Output from the shell
command as a string. Alternatively, you can specify that the output go to another location,
such as a file, or a variable defined inside the CL session.
When using excl.osi:command-output, you can also specify whether CL should wait
for the shell command to complete, or should simply go about its business and let the shell
command complete asynchronously.

4.2

Foreign Function Interface

Allegro CL’s foreign function interface allows you to load compiled libraries and object
code from C, C++, Fortran, and other languages, and call functions defined in this code
1

You must have a fully patched ACL 6.2 or later ACL version to access the excl.osi functionality (if your
machine can access the internet, you can bring your ACL installation to the current patch level with the
function sys:update-allegro). Previous ACL versions have a function excl:run-shell-command which
operates similarly to excl.osi:command-output.

55

56

CHAPTER 4. INTERFACES

as if they were defined natively in CL. This technique requires a bit more setup work than
using simple shell commands with excl.osi:command-output, but it will provide better
performance and more transparent operation once it is set up.

4.3

Interfacing with Corba

Corba, or the Common Object Request Broker Architecture, is an industry-standard mechanism for connecting applications written in different object-oriented languages.
To use Corba, you define a standard interface for your application in a language called
Interface Definition Language (Corba IDL), and each application must compile the standard
interface and implement either a servant or a client for the interface. In this manner, servants
and clients can communicate through a common protocol, regardless of what language they
are written in.
In order to use Corba with CL requires a Corba IDL compiler implemented in CL,
as well as a run-time ORB, or Object Request Broker, to handle the actual client/server
communications. Several such Corba IDL compilers and ORBs are available for CL. An
example of such a system is Orblink, which is available as an optional package for Allegro
CL. Orblink consists of a complete Corba IDL compiler, and a CL-based ORB, which runs
as a CL thread rather than as a separate operating system process.
Using Orblink to integrate a CL application to a Corba client/server environment is a
straightforward process. To start, you simply obtain the Corba IDL files for the desired
interface, and compile them in CL with the command corba:idl. This will automatically
define a set of classes in CL corresponding to all the classes, methods, etc. which make up
the interface.
To use CL as a client, you can now simply make instances of the desired classes, and use
them to call the desired methods. The Orblink ORB will take care of communicating these
method calls across the network, where they will be invoked in the actual process where the
corresponding servant is implemented (which could be on a completely different machine
written in a completely different language).
To use CL as a servant, you must implement the relevant interfaces. This consists of
defining classes in CL which inherit from the “stub” servant classes automatically defined by
the compilation of the Corba IDL, then defining methods which operate on these classes,
and which perform as advertised in the interface. Typically, a Corba interface will only
expose a small piece of an entire application, so implementing the servant classes would
usually represent a small chore relative to the application itself.

4.4

Custom Socket Connections

Just about every CL implementation provides a mechanism to program directly with network sockets, in order to implement “listeners” and network client applications. As a
practical example, Figure 4.1 shows a Telnet server, written by John Foderaro of Franz
Inc., in 22 lines of code in Allegro CL. This Telnet server will allow a running CL process to
accept Telnet logins on port 4000 on the machine on which it is running, and will present

4.4. CUSTOM SOCKET CONNECTIONS

(defun start-telnet (&optional (port 4000))
(let ((passive (socket:make-socket :connect :passive
:local-host "127.1"
:local-port port
:reuse-address t)))
(mp:process-run-function
"telnet-listener"
#’(lambda (pass)
(let ((count 0))
(loop
(let ((con (socket:accept-connection pass)))
(mp:process-run-function
(format nil "tel~d" (incf count))
#’(lambda (con)
(unwind-protect
(tpl::start-interactive-top-level
con
#’tpl::top-level-read-eval-print-loop
nil)
(ignore-errors (close con :abort t))))
con)))))
passive)))
Figure 4.1: Telnet Server in 22 Lines

57

58

CHAPTER 4. INTERFACES

the client with a normal CL Command Prompt (for security, it will by default only accept
connections from the local machine).

4.5

Interfacing with Windows (COM, DLL, DDE)

Most Commercial CL implementations for the Microsoft Windows platform will allow CL
applications to use various Microsoft-specific means for integrating with the Microsoft platform. For example, an Allegro CL application can be set up to run as a COM server, or
compiled into a DLL to be used by other applications.

4.6

Code Generation into Other Languages

CL excels at reading and generating CL code. It is also a powerful tool for analyzing and
generating code in other languages. An example of this can be found in Franz Inc’s htmlgen
product, which provides a convenient CL macro to generate HTML for a web page.
Another example is MSC’s AutoSim product, which uses a CL program to model an
automobile, then generates optimized C programs for solving specialized systems of linear
equations relating to vehicle dynamics.
CL also played a crucial role in preventing the potential Y2K disaster by automatically
analyzing and repairing millions of lines of COBOL!

4.7

Multiprocessing

Multiprocessing allows your application to carry on separate tasks virtually simultaneously.
Virtually, because on a single-processor machine, only one operation can physically be happening at any one time. However, there are situations when you want your program to
appear to be executing separate execution threads, or processes, simultaneously. Web serving is a good example. As described in Section 4.9, you may be running your CL application
as a webserver. This means that multiple users, on different machines in different parts of
your company or even around the world, may be sending requests to your application in
the form of web browser (client) HTTP commands. If one user’s request takes an especially
long time to process, you probably do not want all your other users to have to wait for a
response while that one request is being processed.
Multiprocessing addresses such problems by allowing your CL application to continue
servicing other requests, while “simultaneously” completing that long computation.
Allegro Multiprocessing is covered in the document multiprocessing.htm which ships
with all editions of Allegro CL.

4.7.1

Starting a Background Process

Starting another thread of execution in Allegro CL can be as simple as calling the function
mp:process-run-function. This function spawns a separate thread of execution, while
the execution of code from whence you call it continues. Here is a simple example which
you can run from the command line:

4.7. MULTIPROCESSING

59

(defvar *count* 0)
(mp:process-run-function
(list :name "counting to a million in background"
:priority -10)
#’(lambda() (dotimes (n 1000000) (setq *count* n))))

This starts a thread named “counting to a million in background” with a relative priority
of -10. Notice that after you call this function, your toplevel prompt returns immediately.
The toplevel Read-Eval-Print loop is continuing on its merry way, while your new process is
running in the background, counting to a million (actually, 999999). While the background
process is running, you can sample the value of count at any time:

CL-USER(16):
424120
CL-USER(17):
593783
CL-USER(18):
679401
CL-USER(19):
765012

*count*
*count*
*count*
*count*

Note that on a fast processor, you may need to increase the count value in this example to
be able to take samples before the code completes.

4.7.2

Concurrency Control

Sometimes you want to prevent two threads from modifying a value in memory at the same
time. For this purpose, you can use process locks. A process lock is an object which can be
seized by at most one running process at a time. Using process locks, you can ensure that
multiple threads of execution will interact as you expect.
In our previous counting example, let’s say you want to prevent another thread from
modifying the count variable until the count is complete. First, you would establish a
process lock specific for this purpose:
CL-USER(20): (setq count-lock (mp:make-process-lock))
#<MULTIPROCESSING:PROCESS-LOCK @ #x5a02fba>

Now, run the background function within the context of seizing this lock:

60

CHAPTER 4. INTERFACES

CL-USER(21): (mp:process-run-function
(list :name "counting to a million in background"
:priority -10)
#’(lambda()
(mp:with-process-lock (count-lock)
(dotimes (n 1000000) (setq count n)))))
#<MULTIPROCESSING:PROCESS counting to a million in background @ #x5a0332a>
You now can adopt a policy that any piece of code wishing to modify the value of count must
first seize the count-lock. This will ensure that nothing will interfere with the counting
process. If you do the following while the counting process is running:
CL-USER(22): (mp:with-process-lock (count-lock) (setq count 10))
you will notice that execution will block until the counting process is finished. This is
because the count-lock is already seized, and execution is waiting for this lock to be
released before proceeding.
It is usually a good idea to make your process locks as fine-grained as possible, to avoid
making processes wait unecessarily.
From your toplevel prompt, you can monitor which processes are running in the current
CL image with the toplevel :proc command:
CL-USER(28): :proc
P Bix Dis
Sec dSec
*
7
1
2
2.0
*
1
8
28
0.1
*
3
0
0
0.0
*
4
0
0
0.0
*
6
0
3
0.0

Priority
-10
0
0
0
0

State
runnable
runnable
waiting
inactive
waiting

Process Name, Whostate, Arrest
counting to a million in background
Initial Lisp Listener
Connect to Emacs daemon, waiting for input
Run Bar Process
Editor Server, waiting for input

See the multiprocessing.htm for details on each column in this report, but essentially
this is showing the state of each process currently in the system. This example shows that
Allegro CL uses separate threads for maintaining the connection to the Emacs text editor.
As we will see in Section 4.9, the AllegroServe webserver automatically creates and
manages separate threads to deal with simultaneous web connections.

4.8

Database Interfaces

CL in general, and Allegro CL specifically, provides a myriad of mechanisms for connecting
to databases. Most of these mechanisms differ only slightly in their operation, so you can
generally switch among the different mechanisms without major rework in your application.
The two mechanisms we will cover here are both included with the Enterprise and Platinum
editions of Allegro CL.

4.8. DATABASE INTERFACES

4.8.1

61

ODBC

ODBC, or Open Database Connectivity, is a standard means of connecting to a relational
database based on SQL (Structured Query Language, the industry standard Data Definition
and Data Manipulation language for relational databases). Virtually all popular database
systems can be connected to using ODBC. On Windows these facilities are usually free
of charge; on Unix they may or may not be free. In any case, ODBC is a good solution
to consider if you want to connect your application to several different relational database
systems without writing and maintaining separate pieces of code specific to each.
ODBC uses the concept of data sources, each with a name. There are operating system
dependent ways to set up data sources on your machine to point to actual physical databases,
which may be housed on the local machine or remote. Assuming we have a data source
named “login” with a table named “USER,” connecting to and querying the database can
be as simple as the following:
CL-USER(40): (setq handle (dbi:connect :data-source-name "login"))
#<DBI::ODBC-DB "DSN=login;DB=login;...[truncated]" @ #x4b2c3b2>
CL-USER(41): (dbi:sql "select * from USER" :db handle)
(("1" "dcooper8" "guessme" "Dave" NIL "Cooper" NIL NIL NIL NIL ...)
("7" "awolven" NIL NIL NIL NIL NIL NIL NIL NIL ...))
("GEN_ID" "LOGIN" "PASSWORD" "FIRST_NAME" "MIDDLE_NAME" "LAST_NAME"
"COMPANY" "ADDRESS_1" "ADDRESS_2" "CITY" ...)
As with most CL database interfaces, the ODBC interface returns all SQL queries with
multiple values: The first (and main) return value is a list of lists, corresponding to the
rows in the returned query. The second return value is a list of field names, corresponding
to the values in each of the first lists.
Because ODBC by definition must retain portability among database systems, it is
generally unable to take full advantage of database-specific features and performance optimizations.

4.8.2

MySQL

MySQL is a popular relational database which is available both in commercially supported
and free, open-source form. While you can connect to a MySQL database using ODBC,
Allegro CL also provides a direct connection to MySQL, which performs faster and is easier
to set up than the ODBC connection. From the application level, its use is simple enough
that an application using the MySQL Direct interface could be converted to use ODBC
without major rework, should that become necessary. Therefore, if you know your Allegro
CL application will be using MySQL for the immediate future, we recommend using the
MySQL Direct interface rather than ODBC.
Here is the same example from above, using the MySQL Direct interface instead of the
ODBC one:
CL-USER(50): (setq handle (dbi.mysql:connect

:user "dcooper8"
:password "guessme"

62

CHAPTER 4. INTERFACES

:database "login"))
#<DBI.MYSQL:MYSQL connected to localhost/3306 @ #x5bd114a>
CL-USER(51): (dbi.mysql:sql "select * from USER" :db handle)
((1 "dcooper8" "bhaga<<" "Dave" #:|null| "Cooper" #:|null| ...)
(7 "awolven" #:|null| #:|null| #:|null| #:|null| #:|null| ...))
("GEN_ID" "LOGIN" "PASSWORD" "FIRST_NAME" "MIDDLE_NAME"
"LAST_NAME" "COMPANY" "ADDRESS_1" "ADDRESS_2" "CITY" ...)

A few differences from ODBC to note:
• The functions are in the dbi.mysql package rather than the dbi package.
• Since there is no need to set up separate data sources for MySQL Direct, we connect
to a MySQL instance on the local machine directly by specifying a user, password,
and database name.
• NULL values in the return query are represented with the special symbol #:|null|
rather than with NIL as in ODBC.

4.9
4.9.1

World Wide Web
Server and HTML Generation

The dynamic and multi-threaded nature of AllegroCL makes it a natural fit for powering
webserver applications. Franz provides a popular open-source webserver called AllegroServe
which is designed just for this purpose.
With a standard Allegro CL 6.2 installation you can load AllegroServe and access its
exported symbols as follows:
(require :aserve)
(use-package :net.aserve)
(use-package :net.html.generator)
(use-package :net.aserve.client)
With AllegroServe, you can map website addresses to actual functions which will run in
response to a request. Here is a simple example:

(defun hello (req ent)
(with-http-response (req ent)
(with-http-body (req ent)
(html "Hello World"))))
(net.aserve:publish :path "/hello.html"
:function #’hello)

4.9. WORLD WIDE WEB

63

Now, whenever a web visitor goes to the /hello.html URI on our server, she will receive
in response a page with the simple text “Hello World.” More dynamic examples would of
course be more interesting, and we present some in the next chapter.
Note the use of the html operator in the previous example. This is a macro from the
HTMLgen package which ships with Allegroserve. HTMLgen allows convenient generation
of dynamic, well-structured HTML from a lisp-like source format. Please see the AllegroServe and HTMLgen documentation at http://opensource.franz.com/aserve/index.html
for details on both of these.

4.9.2

Client and HTML Parsing

Where there is a server there must be a client. In the World Wide Web, we typically think
of clients as being interactive web browsers such as Netscape or Internet Explorer. However,
web clients can also be autonomous programs. This is how “agents” and “crawlers” work –
they are actually programmatic web browsers which continually send requests to webservers
and parse the results, gathering information.
Allegroserve includes a web client for such purposes. Using it can be as simple as a
call to net.aserve.client:do-http-request. For example, to retrieve the URI from the
above example programmatically:
CL-USER(1): (do-http-request "localhost:9000/hello.html")
"Hello World"
200
((:TRANSFER-ENCODING . "chunked") (:SERVER . "AllegroServe/1.2.25")
(:DATE . "Tue, 04 Mar 2003 21:36:20 GMT") (:CONNECTION . "Close"))
#<URI http://localhost:9000/hello.html>

As you can see, the do-http-request function returns four values:
1. the actual contents of the response, as a string
2. the response code (e.g. 200 for a normal response, 404 for a “Page Not Found,” etc.)
3. an assoc-list containing the HTTP headers
4. a URI object, an internal CL object containing information about the URI
For more complex client requests which return structured HTML, an HTML parser is
available. For example, you might create a news agent by sending an HTTP request to a
news site, then parsing the HTML to create convenient list structures from which you can
elicit the text of the news stories. Please see http://opensource.franz.com/xmlutils/index.html
for details on this parser.

64

4.10

CHAPTER 4. INTERFACES

Regular Expressions

Many applications have to work with strings of characters. Doing this in an efficient manner
can be a key to the efficiency and performance of your entire application. Allegro CL
provides a Regular Expression module for this purpose. The Regular Expression module
allows you to search, replace, and manipulate patterns in character strings in an efficient
and optimized manner.
As a simple example, let us replace one character in a string with another. Assume we
have some field names as variables in CL, which can legally contain dash (“-”) characters.
Say we want to use these names as a basis for field names in an SQL database, where dashes
are not allowed as part of a name. We will replace all dashes with underscores:
(defun replace-dashes (word)
(excl:replace-regexp word "-" "_"))
CL-USER(59): (replace-dashes "top-o-the-morning")
"top_o_the_morning"
Please see the document regexp.htm, which ships with all editions of Allegro CL, for a full
Regular Expression reference.

4.11

Email

Email was the first “killer app” of the Internet, and arguably continues in that role to this
day. Despite threats from viruses and unsolicited commercial email, many people in the
modern world still depend on email for their daily communication. CL provides an excellent
environment for automating email processing. Not coincidentally, in the next chapter we
present an example of using CL to perform text classification on email, an effective means
for filtering unwanted and unsolicited commercial bulk email.

4.11.1

Sending Mail

Franz provides a convenient open-source package for sending messages to an SMTP (Simple
Mail Transfer Protocol) server.
With a standard Allegro CL 6.2 installation you can load this package and access its
exported symbols as follows:
(require :smtp)
(use-package :net.post-office)
This package provides the send-letter function. This function takes the name of an
SMTP server along with the recipient, the text of an email, and other optional arguments,
and actually delivers the mail to the specified SMTP host. Here is an example:
(send-letter "smtp.myisp.com" "dcooper@genworks.com" "joe@bigcompany.com"
"See you at the game tonight." :subject "Tonight’s Game"
That’s all it takes to have your application shoot off an email.

4.11. EMAIL

4.11.2

65

Retrieving Mail

The two main Internet standards for retrieving mail are POP and IMAP. Although IMAP
is newer and more feature-filled than POP, many ISPs still support only POP, so we will
cover the POP interface here. Franz provides a single file, imap.cl which contains client
interfaces both for POP and IMAP.
With a standard Allegro CL 6.2 installation you can load this package and access its
exported symbols as follows:
(require :imap)
(use-package :net.post-office)
Here is an example of using the POP client interface. First, you must create a connection
handle to the POP server:
CL-USER(4): (setq mailbox (net.post-office:make-pop-connection "pop.myisp.com"
:user "joe"
:password "guessme"))
#<NET.POST-OFFICE::POP-MAILBOX @ #x5dcc33a>
Now we can get a count of messages in the Inbox:
GDL-USER(5): (net.post-office:mailbox-message-count mailbox)
1
and fetch the one message:
GDL-USER(9): (net.post-office:fetch-letter mailbox 1)
"Content-type: text/plain; charset=US-ASCII
MIME-Version: 1.0 ...
..."
Finally, we can delete this message and close our connection:
GDL-USER(10): (net.post-office:delete-letter mailbox 1)
NIL
GDL-USER(12): (net.post-office:close-connection mailbox)
T
Please see the documentation at http://opensource.franz.com/postoffice/index.html
for complete coverage of the POP and IMAP facilities.
In the next chapter, we will combine most of what we have learned in this chapter to
put together a web-based program for browsing and classifying email. Because we make
use of pre-built libraries and interfaces such as the ones described in this chapter, we can
create a functioning multi-user application with less than 500 lines of code.

66

CHAPTER 4. INTERFACES

Chapter 5

Squeakymail
This chapter will present a sample application in CL, called “Squeakymail,” for filtering
and browsing email. Squeakymail demonstrates the use of many of Allegro CL’s interfaces
and special features covered in the previous chapter, in an application of under 500 lines.
Please note that Squeakymail is an open-source application and is still evolving, so you
should obtain the latest code if you wish to work with it. Please contact Genworks at
info@genworks.com for information on obtaining the current Squeakymail codebase.

5.1

Overview of Squeakymail

Squeakymail is a multi-user, server-based application which performs two main tasks:
1. It fetches mail for each user from one or more remote POP (Post Office Protocol)
accounts, analyses it, scores it for its probability of being “spam” (junk email), and
stores it in a local queue for final classification by the user. This fetching goes on
perpetually, as a background thread running in the CL image.
2. It provides a simple web-based interface for each user to be able to view the incoming
mail headers and bodies, and confirm or override the pre-scoring on each message.
Once classified, the individual tokens (words) from each message are added to inmemory spam probability tables, so that the scoring accuracy can become better over
time.
There are some obvious extensions to Squeakymail which could be done, for example
providing an option to auto-classify messages whose spam probability exceeds a certain
threshhold, avoiding the need for the user even to see those messages.

5.2

Fetching and Scoring

One part of our application will run on a background thread, periodically polling users’
remote POP3 email sources for new messages. Specifically, the mail fetching thread will
perform the following for each remote POP3 account of each local user:
67

68

CHAPTER 5. SQUEAKYMAIL

(defun fetch-mails ()
(mp:process-run-function "fetching mail in the background"
#’(lambda()
(do ()()
(let ((start (get-universal-time)))
(with-db-rows ((login) :db *mysql-logins* :table "USER")
(fetch-mail login))
(sleep (max 0 (- *mail-fetching-interval*
(- (get-universal-time) start)))))))))
Figure 5.1: Starting a Background Thread
1. Grab a process lock for the current local user, to prevent other threads from simultaneously modifying that user’s data (files or in-memory).
2. Create a POP3 connection to the specified POP3 server.
3. Open the local user’s local Incoming Mail file for appending.
4. Fetch the current Message Count (i.e., number of messages) from the POP server
5. Iterate through each numbered message, doing the following:
(a) Fetch the entire contents of the message into memory.
(b) Parse the message into its individual header elements and its body.
(c) Compute individual tokens for each header and for the body.
(d) Compute a spam probability “score” based on these tokens.
(e) Write an entry into the local user’s Incoming Mail file, containing the headers,
body, list of tokens, and spam probability score
(f) Finally, mark the message for deletion from the remote POP3 server.
Figures 5.1 and 5.2 contain the basic code for this part of the program. The code in
Figure 5.1 starts the background thread with a call to mp:process-run-function, giving
a name to the process and a lambda expression (anonymous function object) to be run by
the process. That function enters into an infinite iteration (with do). On each iteration of
this infinite loop, it scans through the table of local login accounts using the with-db-rows
macro from the dbi.mysql package. For each row, the lexical variable login is set to the
respective user’s login name, and the fetch-mail function from Figure 5.2 is invoked for
that user. Finally, the process goes into a Sleep mode until the next scheduled interval, at
which point it will wake up and perform yet another iteration of its infinite loop.
The code in Figure 5.2 fetches and scores mail for an individual local user. It first
establishes a process lock for that particular user for the context of the rest of the body of
the function. The lock object itself comes from a call to the function user-process-lock
(Figure 5.3), which retrieves from a hash table a lock specific to the local user, or creates a

5.2. FETCHING AND SCORING

69

(defun fetch-mail (user)
(mp:with-process-lock ((user-process-lock user))
(with-db-rows ((host login password) :table "POP_ACCOUNT"
:db *mysql-local*
:where (=text local_login user))
(format t "Fetching mail from ~a for local user ~a~%" host user)
(with-open-pop-box (mb host login password)
(with-open-file
(out (string-append (format nil "/var/mail/incoming/~a" user))
:direction :output :if-exists :append :if-does-not-exist :create)
(let ((message-count (mailbox-message-count mb)))
(dotimes (n message-count)
(let* ((message (fetch-letter mb (1+ n))))
(multiple-value-bind (headers body)
(parse-mail-header message)
(let ((tokens (tokenize headers body)))
(let ((*package* *token-package*))
(multiple-value-bind (score scored-tokens)
(score-tokens tokens user)
(print (list :headers headers
:body body
:tokens tokens
:score score
:scored-tokens scored-tokens) out)))))))
(delete-letter mb (1+ n)))))))))

Figure 5.2: Fetching and Scoring Remote Mail for a Local User

(defun user-process-lock (user)
(or (gethash user *mailbox-locks-hash*)
(setf (gethash user *mailbox-locks-hash*)
(mp:make-process-lock :name (format nil "fetching mail for ~a" user)))))

Figure 5.3: Retrieving or Creating a Process Lock

70

CHAPTER 5. SQUEAKYMAIL

(defmacro with-open-pop-box
((handle server user password &optional (timeout 10)) &body body)
‘(let ((,handle (make-pop-connection ,server :user ,user :password ,password
:timeout ,timeout)))
(unwind-protect (progn ,@body)
(when ,handle (close-connection ,handle)))))

Figure 5.4: A macro to open and ensure clean closing of a POP3 connection

new lock if one does not yet exist. Running the rest of the function body within this process
lock will ensure that no other threads of execution will interfere with this user’s data during
the fetching operation. For example, another part of our application will provide a web
interface for users to browse and classify their mail interactively. This classification also
modifies the user’s Incoming Mail file, usually by deleting one or more messages from it.
By running our mail fetching inside a user-specific process lock, we eliminate any chance of
these different update operations happening simultaneously. Doing so would likely corrupt
the Incoming Mail file, resulting in lost messages or worse.
This process locking does mean, however, that interactive users might experience a slight
delay when trying to visit their Mail Classification web page, if they happens to do this
while a fetch cycle their her mail is in progress. This is because the interactive process also
requires that same user-specific lock, and will go into “wait” mode until the fetching process
completes and releases the lock.
The fetch-mail function next prints a status message to the console indicating the
local user and remote host for the current fetching operation.
Next it enters into the code written by the with-open-pop-box macro (Figure 5.4),
which opens a connection to the specified POP3 account, assigns the name mb to this
connection, and finally guarantees that the connection will be closed when the code within
with-open-pop-box either completes or results in any kind of error condition. Now the
fetch-mail function enters another context, with-open-file, which gives us a stream
named out for writing to the user’s Incoming Mail file.
Now we query the mb POP connection for the number of messages, and bind this result to the variable message-count. This tells us how many times we have to fetch a
message, which we do inside the code of a dolist, using the fetch-letter function from
the net.post-office package. Note that we add one (1) to the dotimes variable n when
fetching the letter, since POP3 message numbers begin with 1, rather than with 0 as is the
case for dolist iterator variables.
Now we parse the message into headers and body, using multiple-value-bind, since
the parse-mail-header function returns the headers and body as two return values.
Next we tokenize the message (described in Section 5.2.1), compute a score based on the
tokens (Section 5.2.2), and write all the relevant information to the user’s Incoming Mail
file in the convenient form of a plist.
Finally, we mark the message for deletion from the remote POP server, with the
delete-letter function, also from the :net.post-office package.

5.2. FETCHING AND SCORING

(defun tokenize (headers body)
(setq body (excl:replace-regexp
(excl:replace-regexp body "<!-.[^>]*->" "") "<![^>]*>" ""))
(append (tokenize-headers headers)
(mapcan #’(lambda(item) (if (stringp item)
(tokenize-string item)
(list item)))
(flatten (parse-html body)))))
(defun tokenize-headers (headers)
(mapcan #’(lambda(cons)
(let ((header (first cons))
(string (rest cons)))
(let ((strings (excl:split-regexp "\b+" string)))
(mapcar #’(lambda(string)
(intern (string-append header "*" string)
*token-package*)) strings)))) headers))
(defun tokenize-string (string)
(let ((words (excl:split-regexp "\b+" string)) result)
(dolist (word words (nreverse result))
(when (and (<= (length word) *max-token-length*)
(tokenizable? word))
(push (intern word *token-package*) result)))))

Figure 5.5: Tokenizing using Regular Expression facility

71

72

CHAPTER 5. SQUEAKYMAIL

5.2.1

Tokenizing

Tokenizing, the process of breaking up a text into individual words or units, is at the heart
of our spam-filtering strategy. Our current version of tokenizing makes heavy use of Allegro
CL’s Regular Expression API, a set of functions for manipulating text. The three functions
listed in Figure 5.5 comprise our current tokenization system. The first function, tokenize,
first uses excl:replace-regexp to remove all HTML and XML comments from the body
(inserting random HTML comments is a common trick used by spammers to confound
filters). The function then returns the list of tokenized headers appended to the list of
tokenized words from the message body. To produce the list of tokenized words, we first
use the Franz HTML parser to pull out individual HTML tags and strings (this will have
no effect if the message does not have HTML to begin with). Since the parser returns a
nested list of tags, we use flatten to convert it into a flat list. Then we pass each string
into the tokenize-string function.
Tokenize-headers accepts a list of headers, each of which is a dotted list whose first
is the name of the header, and whose rest is a string with the value of the header. We
split the value of the header into individual words by using excl:split-regexp to split
on whitespace. For each word thus obtained, we prepend it with the name of the header.
This distinguishes header tokens from body tokens, since a given token in a Subject line,
for example, can have a very different spam probability from the same token in the body
of the message. Thus, a message with:
Subject: I Love You
will generate tokens Subject*I, Subject*Love, and Subject*You.
The tokenize-string function takes a string and essentially splits it on whitespace to
produce individual words. We reject words longer than *max-token-length*, and check
that the word is “tokenizable,” meaning that it will not confuse the CL reader when reading
the token as a symbol. Currently the only known character sequence which will confuse
the CL reader is a sequence of more than one dot, e.g. “...” so these are rejected as well.
All the strings are converted to symbols and interned into a certain CL package we have
created for this purpose. Converting the strings into symbols makes it much faster to look
up the tokens in hash tables later.

5.2.2

Scoring

Once we have a message tokenized, the actual score computation is fairly straightforward.
We call the function score-tokens with the list of tokens extracted from the message. This
function uses Paul Graham’s1 “Plan for Spam” technique to score the messages. It assumes
the existence of a hash table mapping tokens to their individual spam probabilities. In
Section 5.3 we will see how these hash tables are created and maintained, on a per-user
basis.
The scoring function first binds ht to the user-specific probabilities table using the function restore-probabilities, which retrieves the probability table from memory, restores
it from a saved file, or creates a new one if it does not yet exist. Next it looks up each
1

http://www.paulgraham.com/spam.html

5.3. BROWSING AND CLASSIFYING

73

(defun score-tokens (tokens user)
(let ((ht (restore-probabilities user)))
(let* ((scored-tokens
(sort (mapcar #’(lambda(token)
(list token (or (gethash token ht)
*unknown-probability*)))
tokens)
#’> :key #’(lambda(token) (abs (- (second token) 0.5)))))
(probs (mapcar #’second
(subseq scored-tokens 0
(min *number-of-relevant-tokens*
(length scored-tokens))))))
(values
(let ((prod (apply #’* probs)))
(/ prod (+ prod (apply #’* (mapcar #’(lambda (x) (- 1 x)) probs)))))
scored-tokens))))

Figure 5.6: Scoring a message based on its tokens
token in the probabilities table, creating a list pairing each token with its corresponding
spam probability. It then sorts this result based on the distance of the probability from the
neutral 0.5.
The actual probability values, probs, is then computed by taking the top
*number-of-relevant-tokens* (default 15) values from this list. The combined probability is computed using Bayes’ formula as described in Paul Graham’s Plan for Spam
paper.
A possible improvement to the scoring, mentioned by Paul Graham, would be not to
allow duplicate tokens (or limit the number of duplicate tokens) in computing the top values.

5.3

Browsing and Classifying

Our application’s “left brain” will present an interface to users, allowing them to “train” the
mail classifier. The interface covered here is somewhat bare-bones; we have also prepared
a more feature-full interface using Genworks’ GDL/GWL system. The code for that is
included in Appendix A.
The interface presented here is missing some key features. The most obvious one is a
security mechanism to prevent someone from accessing another user’s mail (GWL provides
a built-in mechanism for login and password security; such a system is left as an exercise for those who wish to implement something like Squeakymail in Common Lisp and
AllegroServe).
Our interface will present a dynamically created web page which will show the headers
from the user’s incoming mail, along with the score of each and resulting spam/nonspam
status. Using a radio button control for each mail, the user will be able to correct any
misclassifications.

74

CHAPTER 5. SQUEAKYMAIL

(defun classify (req ent)
(let ((query (request-query req)))
(let ((keys (mapcar #’first query))
(values (mapcar #’rest query)))
(let ((user (rest (assoc "user" query :test #’string-equal)))
(spam-indices (let ((ht (make-hash-table :values nil)))
(mapc #’(lambda(key value)
(when (and (search "spamp-" key) (string-equal value "SPAM"))
(setf (gethash (read-from-string (subseq key 6)) ht) t))) keys values) ht))
(nonspam-indices (let ((ht (make-hash-table :values nil)))
(mapc #’(lambda(key value)
(when (and (search "spamp-" key) (string-equal value "NON-SPAM"))
(setf (gethash (read-from-string (subseq key 6)) ht) t))) keys values) ht)))
(let ((messages (classify-and-write-files user spam-indices nonspam-indices)))
(mp:process-run-function (list :name (format nil "Squeakymail: Recomputing probabilities for user ~a in the background" user)
:priority -5 :quantum 1) #’(lambda()(sleep 5)(compute-probabilities user)))
(with-http-response (req ent) (with-http-body (req ent) (classify-form messages user))))))))
(publish :path "/classify" :function #’classify)

Figure 5.7: Main Function to produce Classification Web Page
The main classify function is shown in Figure 5.7, and is published (mapped to a
URL) also in Figure 5.7. The classify function, as with all Allegroserve http response
functions, takes req and ent arguments. The req argument comes from the browser, and
contains the values from any submitted html forms. These values are bound to the query
variable. When the classify function is called for the first time, it contains no form data.
On subsequent invocations, however, it might contain form data indicating which messages
are spam and nonspam. Later we will see how to create the form to submit these values.
Once we extract the form keys (field names) and values from the query information, we
compute two hash tables: spam-indices and nonspam-indices. These contain the numerical indices representing the user’s selection of radio buttons on each message. Next we call
the classify-and-write-files function with the spam-indices and nonspam-indices.
The internals of this function will be described soon; basically it moves messages from the
user’s incoming mail file, and returns any remaining (unclassified) messages along with any
new messages which have arrived in the user’s incoming mail file in the meantime.
Then we create a temporary file as incoming-buffer-pathname to contain the unclassified messages, spams-pathname to contain our spam corpus, nonspams-pathname to contain
our nonspam corpus, and popqueue for appending to the user’s actual mail file.
Now, using a cond expression, we distribute each message to the appropriate file.
The classify function then spawns a low-priority background process to recompute
the spam token probabilities for the user.
Finally, it generates the actual web page with messages and classification form, inside a
with-http-response and with-http-body, by calling the classify-form function.
Figure 5.8 contains the code for the classify-and-write-files function. This function
takes a list of spam-indices and nonspam-indices, and moves messages around appropriately.
This is all done within the context of the user-specific process lock, to prevent two user
requests from rearranging files simultaneously which might result in corrupted files.
The classify-and-write-files function first populates a list of messages, bound to
the let variable messages by opening and reading from the user’s incoming mail file. This
is done with the read-incoming-messages function, listed in Figure 5.9.
The actual web page comes from the classify-form function, listed in Figure 5.10. This
form is created with the html macro, which allows us to write a “lispified” html template

5.3. BROWSING AND CLASSIFYING

75

(defun classify-and-write-files (user spam-indices nonspam-indices)
(let (remaining-messages)
(mp:with-process-lock ((user-process-lock user))
(let (*read-eval*)
(let ((messages (read-incoming-messages user)))
(when messages
(let ((incoming-buffer-pathname (sys:make-temp-file-name))(count -1)
(spams-pathname
(progn (ensure-directories-exist (string-append "/var/mail/" user ".data/corpi/"))
(string-append "/var/mail/" user ".data/corpi/spams")))
(nonspams-pathname (string-append "/var/mail/" user ".data/corpi/nonspams")))
(with-open-file (spams spams-pathname
:direction :output :if-exists :append :if-does-not-exist :create)
(with-open-file (nonspams nonspams-pathname
:direction :output :if-exists :append :if-does-not-exist :create)
(with-open-file (popqueue (string-append "/var/mail/" user)
:direction :output :if-exists :append :if-does-not-exist :create)
(with-open-file (incoming-buffer incoming-buffer-pathname
:direction :output :if-exists :supersede :if-does-not-exist :create)
(dolist (message messages)
(let ((index (getf message :index)))
(cond ((gethash index spam-indices)
(record-tokens (mapcar #’(lambda(token)(intern token *token-package*))
(getf message :tokens)) :spam user)
(print message spams))
((gethash index nonspam-indices)
(record-tokens (mapcar #’(lambda(token) (intern token *token-package*))
(getf message :tokens)) :nonspam user)
(print message nonspams)
(pop-format message popqueue))
(t
(setf (getf message :index) (incf count))
(push message remaining-messages)
(print message incoming-buffer)))))))))
(sys:copy-file incoming-buffer-pathname (string-append "/var/mail/incoming/" user)
:overwrite t)
(delete-file incoming-buffer-pathname))))))
(nreverse remaining-messages)))

Figure 5.8: Function to distribute messages to the appropriate files
with embedded Lisp code which will be evaluated. In this manner, we create the html
page including a form whose action (response URL) invokes the same classify function as
described above. We then group the messages, based on their pre-computed score, into a
list of expected spams and expected nonspams. Finally we present the headers from these
messages to the user, each with its radio button pre-selected appropriately if we have a
sufficient confidence level of the spam/nonspam probability of the message.
When the user presses the “Submit” button, the form is submitted and the same
classify function will respond, distributing messages into the appropriate files. This process is repeated indefinitely as long as the user continues pressing the “Submit” button.

76

CHAPTER 5. SQUEAKYMAIL

(defun read-incoming-messages (user)
(fetch-mail user)
(let (result (*read-eval* nil) (count -1)
(incoming-pathname (string-append "/var/mail/incoming/" user)))
(when (probe-file incoming-pathname)
(with-open-file (in incoming-pathname)
(do ((message (read in nil nil) (read in nil nil)))
((null message) (nreverse result))
(setf (getf message :index) (incf count))
(push message result))))))

Figure 5.9: Reading incoming messages

(defun classify-form (messages user)
(html
(:html
(:head (:title "Classify Incoming Mail for " (:princ user)))
(:body (:h2 (:center "Classify Incoming Mail for " (:princ user)))
((:form :action "/classify" :method :post)
(:p ((:input :type :submit :value "Classify!")))
(:center
((:input :type :hidden :name :user :value user))
((:table :bgcolor :black :cellpadding 1 :cellspacing 1)
(let ((grouped-messages (list nil nil)))
(mapc #’(lambda(message)
(if (> (getf message :score) 0.8)
(push message (second grouped-messages))
(push message (first grouped-messages)))) messages)
(setq grouped-messages (append (nreverse (first grouped-messages))
(nreverse (second grouped-messages))))
(when grouped-messages
(html ((:tr :bgcolor :yellow) (:th "Bad") (:th "Good") (:th "From") (:th "Subject") (:th "Date") (:th "Score"))))
(dolist (message grouped-messages)
(let ((headers (getf message :headers))
(score (getf message :score)))
(let ((from (rest (assoc "from" headers :test #’string-equal)))
(subject (rest (assoc "subject" headers :test #’string-equal)))
(date (rest (assoc "date" headers :test #’string-equal))))
(html ((:tr :bgcolor (if (> score 0.8) :pink "#aaffaa"))
((:td :bgcolor :red)
((:input :type :radio :name (format nil "spamp-~a" (getf message :index))
:value :spam :if* (> score 0.8) :checked :checked)))
((:td :bgcolor :green)
((:input :type :radio :name (format nil "spamp-~a" (getf message :index))
:value :non-spam :if* (< score 0.5) :checked :checked)))
(:td (:small (:princ (string-append (subseq subject 0 (min (length subject) *max-subject-length*))
(if (> (length subject) *max-subject-length*) "..." "")))))
(:td (:small (:princ (short-date date))))
(:td (:princ score))))))))))
(:p ((:input :type :submit :value "Classify!"))))))))

Figure 5.10: Generating the actual web page

Appendix A

Squeakymail with Genworks’
GDL/GWL
Genworks International produces and sells a Knowledge Base system based on Allegro CL
called General-purpose Declarative Language. Bundled with GDL is also GWL, Generative
Web Language, an integrated web application server which runs in conjunction with GDL
and AllegroServe.
GDL and GWL can also generate and manipulate 3D geometric entities, giving you
additional power in a full-featured toolset for creating engineering and business applications.
For an introduction to GDL/GWL, please see the GDL Tutorial, available in PDF
format from:
http://www.genworks.com/papers/gdl-tutorial.pdf
In this chapter we present an alternative classifying/browsing interface for Squeakymail,
based on GDL/GWL. This version of the interface contains some significant features not
found in the version found in Chapter 5:
• Secure user login, provided by a separate GWL module
• Ability to preview the message bodies in a linked page
• Convenient interfaces for a user to add, update, and delete remote POP accounts and
explicit message filters.
• Facility for “plugging in” a user’s Squeakymail access into a larger web-based application framework, consisting of for example, calendaring and personal task management
and planning apps.

A.1

Toplevel Squeakymail Home Page

The code in Figure A.1 shows the object definition corresponding to the Squeakymail application home page. Note the customer-row being passed in as an input-slot. This customerrow is a database record, provided by a separate GWL module, containing information on
an authenticated user.
77

78

APPENDIX A. SQUEAKYMAIL WITH GENWORKS’ GDL/GWL

The child objects classify, pop-accounts, and filters are to be presented as hyperlinks to specific pages within the main Squeakymail page. Classify allows is a page
allowing the user to browse and classify mail. Pop-accounts allows the user to manage
remote POP accounts. Finally, filters allows the user to manage explicit filters, which
allow messages to bypass the standard Baysian filtering mechanism.
The view defined in Figure A.2 defines a main-sheet output function, which generates
the actual HTML for the page.

A.2

Page for Browsing and Classifying

Figure A.3 builds up a sequence of current messages, each potentially to be displayed on
its own sub-page, and the view defined in A.4 defines a main-sheet output function which
processes incoming mail, then calls the classify-form output function to emit the actual
HTML for the classification form.
The message object defined in Figure A.5 is a simple message taking inputs of from,
subject, to, body, score, and scored-tokens. Each message can potentially be rendered
in HTML using the main-sheet output function defined in the view in Figure A.6.

A.2. PAGE FOR BROWSING AND CLASSIFYING

(define-object assembly (base-html-sheet)
:input-slots
(customer-row)
:computed-slots
((local-login (the customer-row login)))
:objects
((classify :type ’classify
:strings-for-display "Check and Classify"
:pass-down (local-login))
(pop-accounts :type ’html-sql-table
:sql-server *local-connection*
:table-name ’pop-account
:fixed-fields (list :local-login (the customer-row login))
:strings-for-display "Pop Accounts"
:columns-and-names
‘(("host" "Mail Server")
("login" "Username")
("password" "Password")))
(filters :type ’html-sql-table
:sql-server *local-connection*
:table-name ’filters
:strings-for-display "Filters"
:fixed-fields (list :local-login (the customer-row login))
:columns-and-names
‘(("source" "Source"
:choices (:explicit ("Subject" "To" "From" "Body")))
("expression" "Expression" :size 45)
("destination" "Put In"
:choices (:explicit ("Spam" "Nonspam")))))))

Figure A.1: Object for Squeakymail Home Page

79

80

APPENDIX A. SQUEAKYMAIL WITH GENWORKS’ GDL/GWL

(define-view (html-format assembly)()
:output-functions
((main-sheet
()
(html (:html (:head
(:title "Squeakymail - Enjoy a squeaky-clean email experience"))
(:body (when *developing?* (the write-development-links))
(when (the return-object) (the write-back-link))
(:center (:h2 "Welcome to Squeakymail")
(:i "Enjoy a Squeaky-Clean Email Experience"))
(:p (the classify (write-self-link)))
(:p (the filters (write-self-link)))
(:p (the pop-accounts (write-self-link)))
(the write-standard-footer)))))))

Figure A.2: Output Function for Squeakymail Home Page

(define-object classify (base-html-sheet)
:input-slots
(local-login)
:computed-slots
((message-list

(progn (fetch-mail (the local-login))
(read-incoming-messages (the local-login)))))

:objects
((messages :type ’message
:sequence (:size (length (the message-list)))
:data (nth (the-child index) (the message-list))
:headers (getf (the-child :data) :headers)
:scored-tokens (getf (the-child :data) :scored-tokens)
:from (or (rest (assoc "from" (the-child headers)
:test #’string-equal)) "")
:subject (or (rest (assoc "subject" (the-child headers)
:test #’string-equal)) "")
:to (or (rest (assoc "to" (the-child headers)
:test #’string-equal)) "")
:date (or (rest (assoc "date" (the-child headers)
:test #’string-equal)) "")
:body (getf (the-child :data) :body)
:score (getf (the-child :data) :score))))

Figure A.3: Object for Squeakymail Classification Page

A.2. PAGE FOR BROWSING AND CLASSIFYING

(define-view (html-format classify)()
:output-functions
((main-sheet
()
(let ((query-plist (the query-plist)))
(classify-and-write-files query-plist (the local-login)))
(the (restore-attribute-defaults! (list :message-list :query-plist)))
(write-the (classify-form (list-elements (the messages)) (the local-login))))
(classify-form
(messages user)
(html
(:html
(:head (:title (if messages (html "Classify Incoming Mail for ")
(html "No Mail for ")) (:princ user)))
(:body (:p (when gwl:*developing?* (the write-development-links)))
(:h2 (:center (if messages (html "Classify Incoming Mail for ")
(html "No Mail for ")) (:princ user)))
((:form :action "/answer" :method :post)
(:p (the write-back-link))
(:p ((:input :type :submit :value (if messages "Classify!" "Check Again!"))))
(:center
(the (:write-form-fields))
((:table :bgcolor :black :cellpadding 1 :cellspacing 1)
(let ((grouped-messages (list nil nil nil)))
(mapc #’(lambda(message)
(let ((score (the-object message score)))
(cond ((null score) (push message (first grouped-messages)))
((> score 0.8) (push message (third grouped-messages)))
(t (push message (second grouped-messages)))))) messages)
(setq grouped-messages (append (nreverse (first grouped-messages))
(nreverse (second grouped-messages))
(nreverse (third grouped-messages))))
(when grouped-messages
(html ((:tr :bgcolor :yellow) (:th "Bad") (:th "Good") (:th "From")
(:th "Subject") (:th "Date") (:th "Score"))))
(dolist (message grouped-messages)
(let ((from (the-object message from))
(subject (the-object message subject))
(date (the-object message date))
(score (the-object message score)))
(html ((:tr :bgcolor (cond ((null score) (gethash :sky-summer *color-table*))
((> (the-object message score) 0.8) :pink) (t "#aaffaa")))
((:td :bgcolor :red)
((:input :type :radio :name (format nil "spamp-~a" (the-object message index))
:value :spam :if* (and score (> (the-object message score) 0.8)) :checked :checked)))
((:td :bgcolor :green)
((:input :type :radio :name (format nil "spamp-~a" (the-object message index))
:value :non-spam :if* (and score (< (the-object message score) 0.5)) :checked :checked)))
(:td (:small (:princ-safe (excl:replace-regexp from """ ""))))
(:td (:small (the-object message
(write-self-link
:display-string
(with-output-to-string(*html-stream*)
(html
(:princ-safe
(string-append (subseq subject 0 (min (length subject)
*max-subject-length*))
(if (> (length subject) *max-subject-length*)
"..." "")))))))))
(:td (:small (:princ-safe (short-date date))))
(:td (:princ score)))))))))
(when messages (html (:p ((:input :type :submit :value "Classify!"))))))))))))

Figure A.4: Output Function for Classification Page

(define-object message (base-html-sheet)
:input-slots
(from subject to body score scored-tokens))
Figure A.5: Object for a Single Message

81

82

APPENDIX A. SQUEAKYMAIL WITH GENWORKS’ GDL/GWL

(define-view (html-format message)()
:output-functions
((main-sheet
()
(html
(:html
(:head
(:title (:princ (the subject)))
(:body
(:p (the write-back-link))
(:p (:table (:tr (:td "From") (:td (:princ-safe (the from))))
(:tr (:td "To") (:td (:princ-safe (the to))))
(:tr (:td "Subject") (:td (:princ-safe (the subject))))
(:tr (:td "Score") (:td (:princ-safe (the score))))
(:tr ((:td :colspan 2) (:pre (:princ (the body)))))))
(when *show-tokens?*
(html (:p ((:table :bgcolor :black)
(dolist (pair (subseq (the scored-tokens) 0
(min (length (the scored-tokens)) 50)))
(html ((:tr :bgcolor (if (> (second pair) 0.5)
:pink "#aaffaa"))
(:td (:princ-safe (first pair)))
(:td (format *html-stream* "~2,7f"
(second pair)))))))))))))))))

Figure A.6: Output Function for Displaying a Message

Appendix B

Bibliography
This Bibliography should also be considered as a Recommended Reading list.
• ANSI Common Lisp, by Paul Graham. Prentice-Hall, Inc, Englewood Cliffs, New
Jersey. 1996.
• Common Lisp, the Language, 2nd Edition, by Steele, Guy L., Jr., with Scott E. Fahlman,
Richard P. Gabriel, David A. Moon, Daniel L. Weinreb, Daniel G. Bobrow, Linda
G. DeMichiel, Sonya E. Keene, Gregor Kiczales, Crispin Perdue, Kent M. Pitman,
Richard C. Waters, and Jon L. White. Digital Press, Bedford, MA. 1990.
• ANSI CL Hyperspec, by Kent M. Pitman. Available at
http://www.xanalys.com
• Learning Gnu Emacs, by Eric Raymond. O’Reilly and Associates. 1996.
• Association of Lisp Users website, many other resources and references, at http://www.alu.org
• Usenet Newsgroup comp.lang.lisp
• Successful Lisp, by David Lamkin, available at
http://psg.com/~dlamkins/left/sl/sl.html
• Franz Website (Commercial CL Vendor), at
http://www.franz.com
• Xanalys Lispworks Website (Commercial CL Vendor), at
http://www.lispworks.com
• Symbolics Website (Commercial CL Vendor), at
83

84

APPENDIX B. BIBLIOGRAPHY
http://www.symbolics.com
• Digitool Website (Macintosh Common Lisp Vendor), at
http://www.digitool.com
• Corman Lisp Website (Commercial CL Vendor), at
http://www.corman.net
• Genworks International Website (CL-based tools, services, and pointers to other resources) at
http://www.genworks.com
• Knowledge Technologies International Website, at
http://www.ktiworld.com
• Tulane Lisp Tutorial, available at
http://www.cs.tulane.edu/www/Villamil/lisp/lisp1.html
• Texas A&M Lisp Tutorial, available at
http://grimpeur.tamu.edu/colin/lp/

Appendix C

Emacs Customization
C.1

Lisp Mode

When you edit a Lisp file in Emacs, Emacs should automatically put you into Lisp Mode or
Common Lisp Mode. You will see this in the status bar near the bottom of the screen. Lisp
Mode gives you many convenient commands for working with List expressions, otherwise
known as Symbolic Expressions, or s-expressions, or sexp’s for short.
For a complete overview of Lisp Mode, issue the command M-x describe-mode in
Emacs. Table C.1 describes a sampling of the commands available in Lisp Mode1 .

C.2

Making Your Own Keychords

You may wish to automate some commands which do not already have keychords associated
with them. The usual way to do this is to invoke some global-set-key commands in the
.emacs file (Emacs’ initialization file) in your home directory. Here are some example entries
you could put in this file, along with comments describing what they do:

1

Some of these commands are actually available in Fundamental mode as well

Keychord
C-M-space
M-w
C-y
C-M-k
C-M-f
C-M-b
M-/

Action
Mark sexp starting at point
Copy marked expression
Paste copied expression
Kill the sexp starting at point
Move point forward by one sexp
Move point backward by one sexp
Dynamically expand current
word (cycles through choices on
repeat)

Emacs Function
mark-sexp
kill-ring-save
yank
kill-sexp
forward-sexp
backward-sexp
dabbrev-expand

Table C.1: Common Commands of the Emacs-Lisp Interface
85

86

APPENDIX C. EMACS CUSTOMIZATION

;; Make C-x & switch to the *common-lisp* buffer
(global-set-key "\C-x&" ’(lambda() (interactive)
(switch-to-buffer "*common-lisp*")))
;; Make function key 5 (F5) start or visit an OS shell.
(global-set-key [f5] ’(lambda() (interactive) (shell)))
;; Make sure M-p functions to scroll through previous commands.
(global-set-key "\M-p" ’fi:pop-input)
;; Make sure M-n functions to scroll through next commands.
(global-set-key "\M-n" ’fi:push-input)
;; Enable the hilighting of selected text to show up.
(transient-mark-mode 1)

C.3

Keyboard Mapping

If you plan to spend significant time working with your keyboard, consider investing some
time in setting it up so you can use it effectively. As a minimum, you should make sure
your Control and Meta keys are set up so that you can get at them without moving your
hands away from the “home keys” on the keyboard. This means that:
• The Control key should be to the left of the “A” key, assessible with the little finger
of your left hand.
• The Meta keys should be on each side of the space bar, accessible by tucking your
right or left thumb under the palm of your hand.
If you are working on an XWindow terminal (e.g. a Unix or Linux machine, or a PC
running an X Server such as Hummingbird eXceed) you can customize your keyboard with
the xmodmap Unix command. To use this command, prepare a file called .Xmodmap with
your desired customizations, then invoke this file with the command
xmodmap .Xmodmap
Here is an example .Xmodmap file for a PC keyboard, which makes the Caps Lock key
function as Control, and ensures that both Alt keys will function as Meta:
remove Lock = Caps_Lock
keysym Caps_Lock = Control_L
add Control = Control_L
keysym Alt_L = Meta_L Alt_L
For PCs running Microsoft Windows, the Windows Control Panel should contain options
for customizing the keyboard in a similar manner.

Appendix D

Afterword
D.1

About This Book

This book was typeset using a Lisp-based markup language. For the printed output, the
Lisp-based markup language generated LATEX code and was typeset using LATEX by Leslie
Lamport, which is built atop TEX by Donald Knuth.

D.2

Acknowledgements

I have received inspiration and instruction from many enabling me to write this book.
I would like to thank, in no particular order, John McCarthy, Erik Naggum of Naggum
Software and comp.lang.lisp, John Foderaro of Franz Inc., Professor John Laird who taught
me CL in college; John Mallery and Paul Graham, who taught me that CL is the platform
to be using for server applications, James Russell, Andrew Wolven, I would especially like
to thank Lisa Fettner, Dr. Sheng-Chuan Wu, and others at Franz Inc for their eagle-eyed
proofreading, Peter Karp from Stanford Research Institute for creating the original outline,
and Hannu Koivisto for helping to improve the PDF output. Also thank you to Michael
Drumheller of the Boeing Co. for providing feedback on the first edition. “Thank you” to
my wife Kai, for putting up with my antics while I was writing this (and in general). I
would also like to thank Fritz Kunze of Franz Inc. for telling me to “go write a book.”
Any remaining errors in this version are of course my own.

D.3

About the Author

David J. Cooper has been Chief Engineer with Genworks International since November
1997. His principal activity is working with large manufacturing companies, helping them
to solve engineering and manufacturing problems, mostly with CL-based tools. He has also
led instruction courses at General Motors, Visteon Automotive, Raytheon Aircraft, Ford
Motor Company, Knowledge Technologies International, and others. Prior to his career
with Genworks, Mr. Cooper worked with Ford Motor Company, mostly on deployment of
CL-based systems in the field of mechanical engineering and manufacturing. He also has
software development experience in the CAD and database industries. He received his M.S.
87

88

APPENDIX D. AFTERWORD

degree in Computer Science from the University of Michigan in 1993, and holds a B.S.
in Computer Science and a B.A. in German from Michigan as well. His current interests
include marketing and promoting Genworks’ GDL/GWL Knowledge Base product offering.
He is available for consultation and product inquiries through Genworks International at
http://www.genworks.com, david.cooper@genworks.com, +1 (248) 910-0912.

Index
Allegro CL, 5
American National Standards Institute, 1
and, 36
anonymous functions, 40
ANSI Common Lisp, 1
append, 32
aref, 46
argument lists
function, 50
arguments
optional, 41
to a function, 21
arrays, 46
character, 23
Association of Lisp Users, 5
associative array, 45
Automatic Memory Management, 2

code generation, 58
command-output, 55
Common Lisp, 1
ANSI, 1
Common Lisp Object System, 47
Common Lisp, the Language, 2nd Edition, 1
comparison function, for sorting, 34
Computer-aided Software Engineering, 2
concurrency control, 59
cond, 37
cons, 33
consing
avoidance, 54
control string for format, 43
corba, 56
data mining, iii
data types, CL, 22
Defclass, 47
Defmethod, 47
defparameter, 33
defstruct, 47
defvar, 33
destructive operators, 54
do, 38
Doctor Strangelove, 21
document management, iii
dolist, 38
dotimes, 38
Dynamic Redefinition, 2
dynamic scope, 27
Dynamic Typing, 2, 22

B2B, iii
batch mode, running CL in, 13
bioinformatics, iii
C, 21
C++, iii, 21
C#, iii
car, 35
CASE, 2
case, 38
ccase, 38
CGI scripts, 1
CL, 1
data types, 22
CL environment
Installing, 5
CL Packages, 13
Classes, 47
CLOS, 47
CLtL2, 1

E-commerce, iii
ecase, 38
elements of a list
accessing, 28
89

90
email, 64
Empty List, The, 29
epsilon
zero, 52
eql, 50
equal, 50
equality, 50
evaluation
turning off, 22
evaluation rules, 19
excl.osi:command-output, 55
expression evaluation, 19
Floating Point, 22
format, 43
format directives, 43
function
calling, 19
Functional arguments, 40
functions
anonymous, 40
predicate, 29
functions as Objects, 39
funding
unlimited
over, iii
garbage collector, 54
generic function, 47
global variables, 25
Gnu Emacs, 8
Graham
Paul, 2
Guy Steele, 1
hash tables, 45
if, 30
infix, 21
input
using read, 42
Integers, 22
interface
operating system, 55
to other languages, 55
with Windows, 58

INDEX
interfaces, 55
to databases, 60
intersection, 34
iteration, 38
Java, iii, 21
John McCarthy, 1
lambda, 40
Let, 26
lexical scope, 27
Linux, 5
Lisp, 1
Common, 1
ANSI, 1
list, 1
adding single elements to, 33
appending, 32
part of
getting, 32
Property, 35
removing elements from, 33
sorting, 34
treating as a Set, 34
lists, 24
logical operators, 36
loop, 38
macro, 25
make-array, 46
make-instance, 47
make-pathname, 44
mapcar, 35
mapping, 35
McCarthy
John, 1
member, 31
methods, 47
specialization of, 47
Mixin
synonym for Superclass, 47
multiprocessing, 58
multithreading
dynamic scope useful for, 27
MySQL, 61

INDEX
NIL, 29
not, 37
null, 29, 37
Numbers, 22
Object Management Group, 1
objects, 47
OMG, 1
open, 44
operating system, Lisp-based, 1
operating systems
Linux, 5
Lisp-based, 5
Symbolics
filesystem, 44
Unix
filesystem, 44
operators
logical, 36
Optional arguments, 41
or, 36
Package, 13
package, 48
Packages
CL, 13
pathname-directory, 44
pathname-name, 44
pathname-type, 44
pathnames, 44
Paul Graham, 1
people resources
needed to support expectations, iii
Perl, iii, 1, 2, 21
plist, 35
of symbol, 23
postfix, 21
predicate, 34
predicate functions, 29
prefix, 19, 21
prin1, 42
princ, 43
print, 42
Property list, 35
Python, iii, 21

91
quoting, 49
Ratios, 22
read, 42
read-eval-print, 19
regular expressions, 64
remove, 33
rest of a list, 29
return, 39
rules
expression evaluation, 19
scheduling, iii
scope
dynamic, 27
lexical, 27
variable, 23
set operations, 34
set-difference, 34
setf, 45, 46
setq, 26, 33
shared memory space, 1
sockets, 56
sort, 34
special variables, 25
Steele
Guy, 1
Strangelove
Doctor, 21
Stream, 43
Streams, 42
string-equal, 50
strings, 23, 50
structures, 47
subseq, 32
Superclass
synonym for Mixin, 47
symbol, 20
Symbolics Common Lisp, 5
Symbols, 23
symbols, 50
exporting, 49
importing, 49
T
as default representation for Truth, 29

92
tables
hash, 45
test-expression forms
for cond, 37
threads, 1
toplevel, 19
turning off evaluation, 22
Typing
Dynamic, 22
union, 34
Unix, 44
variables
global, 25
special, 25
webserver, 62
With-open-file, 45

INDEX