C Programming - Data Structures and Algorithms

C Programming:
Data Structures and Algorithms
An introduction to elementary
programming concepts in C
Jack Straub, Instructor
Version 2.07 DRAFT
C Programming: Data Structures and Algorithms, Version 2.07 DRAFT
C Programming: Data Structures and Algorithms

Version 2.07 DRAFT
Copyright © 1996 through 2006 by Jack Straub
ii 06/19/06
Table of Contents
COURSE OVERVIEW ........................................................................................ IX
1. BASICS....................................................................................................13
1.1 Objectives...................................................................................................................................... 13
1.2 Typedef.......................................................................................................................................... 13
1.2.1 Typedef and Portability ............................................................................................................. 13
1.2.2 Typedef and Structures.............................................................................................................. 14
1.2.3 Typedef and Functions .............................................................................................................. 14
1.3 Pointers and Arrays ..................................................................................................................... 16
1.4 Dynamic Memory Allocation ...................................................................................................... 17
1.5 The Core Module.......................................................................................................................... 17

1.5.1 The Private Header File............................................................................................................. 18
1.5.2 The Principal Source File .......................................................................................................... 18
1.5.3 The Public Header File.............................................................................................................. 19
1.6 Activity .......................................................................................................................................... 21
2. DOUBLY LINKED LISTS.........................................................................23

2.1 Objectives...................................................................................................................................... 23
2.2 Overview ....................................................................................................................................... 23
2.3 Definitions ..................................................................................................................................... 24

2.3.1 Enqueuable Item........................................................................................................................ 25
2.3.2 Anchor....................................................................................................................................... 26
2.3.3 Doubly Linked List ................................................................................................................... 26
2.3.4 Methods..................................................................................................................................... 26
2.3.5 ENQ_create_list: Create a New Doubly linked List.................................................................. 27
2.3.6 ENQ_create_item: Create a New Enqueuable Item .................................................................. 28
2.3.7 ENQ_is_item_enqed: Test Whether an Item is Enqueued ........................................................ 29
2.3.8 ENQ_is_list_empty: Test Whether a List is Empty................................................................... 29
2.3.9 ENQ_add_head: Add an Item to the Head of a List .................................................................. 29
2.3.10 ENQ_add_tail: Add an Item to the Tail of a List.................................................................. 30
2.3.11 ENQ_add_after: Add an Item After a Previously Enqueued Item........................................ 30
2.3.12 ENQ_add_before: Add an Item Before a Previously Enqueued Item .................................. 30
2.3.13 ENQ_deq_item: Dequeue an Item from a List ..................................................................... 31
2.3.14 ENQ_deq_head: Dequeue the Item at the Head of a List ..................................................... 31
2.3.15 ENQ_deq_tail: Dequeue the Item at the Tail of a List ......................................................... 32
2.3.16 ENQ_GET_HEAD: Get the Address of the Item at the Head of a List................................ 32
2.3.17 ENQ_GET_TAIL: Get the Address of the Item at the Tail of a List.................................... 32
2.3.18 ENQ_GET_NEXT: Get the Address of the Item After a Given Item .................................. 33
2.3.19 ENQ_GET_PREV: Get the Address of the Item Before a Given Item ................................ 33
2.3.20 ENQ_GET_LIST_NAME: Get the Name of a List.............................................................. 33
2.3.21 ENQ_GET_ITEM_NAME: Get the Name of an Item ......................................................... 34
2.3.22 ENQ_destroy_item: Destroy an Item ................................................................................... 34
iii 06/19/06
2.3.23 ENQ_empty_list: Empty a List............................................................................................. 35
2.3.24 ENQ_destroy_list: Destroy a List......................................................................................... 35
2.4 Case Study .................................................................................................................................... 35
2.5 Activity .......................................................................................................................................... 39
3. SORTING.................................................................................................41
3.1 Objectives...................................................................................................................................... 41
3.2 Overview ....................................................................................................................................... 41
3.3 Bubble Sort ................................................................................................................................... 41
3.4 Select Sort ..................................................................................................................................... 42
3.5 Mergesort...................................................................................................................................... 42
3.6 A Mergesort Implementation in C.............................................................................................. 43

3.6.1 The Mergesort Function’s Footprint.......................................................................................... 43
3.6.2 The Pointer Arithmetic Problem ............................................................................................... 43
3.6.3 The Assignment Problem .......................................................................................................... 44
3.6.4 The Comparison Problem.......................................................................................................... 45
3.6.5 The Temporary Array................................................................................................................ 46
4. MODULES ...............................................................................................47
4.1 Objectives...................................................................................................................................... 47
4.2 Overview ....................................................................................................................................... 47
4.3 C Source Module Components.................................................................................................... 47

4.3.1 Public Data ................................................................................................................................ 47
4.3.2 Private Data ............................................................................................................................... 48
4.3.3 Local Data ................................................................................................................................. 49
4.4 Review: Scope ............................................................................................................................... 49
4.5 A Bit about Header Files ............................................................................................................. 49
4.6 Module Conventions .................................................................................................................... 49
5. ABSTRACT DATA TYPES......................................................................51

5.1 Objectives...................................................................................................................................... 51
5.2 Overview ....................................................................................................................................... 51
5.3 Exception Handling...................................................................................................................... 52
5.4 Classes of ADTs ............................................................................................................................ 54

5.4.1 The Complex Number ADT...................................................................................................... 54
iv 06/19/06
5.4.2 The List ADT ............................................................................................................................ 55
5.4.3 Implementation Choices............................................................................................................ 60
6. STACKS ..................................................................................................69
6.1 Objectives...................................................................................................................................... 69
6.2 Overview ....................................................................................................................................... 69
6.3 Stacks And Recursion .................................................................................................................. 72
6.4 A Minimal Stack Module............................................................................................................. 76

6.4.1 STK Module Public Declarations.............................................................................................. 76
6.4.2 STK_create_stack: Create a Stack............................................................................................. 76
6.4.3 STK_push_item: Push an Item onto a Stack ............................................................................. 77
6.4.4 STK_pop_item: Pop an Item off a Stack................................................................................... 77
6.4.5 STK_peek_item: Get the Top Item of a Stack.......................................................................... 77
6.4.6 STK_is_stack_empty: Determine If a Stack is Empty ............................................................. 78
6.4.7 STK_is_stack_full: Determine If a Stack is Full ...................................................................... 78
6.4.8 STK_clear_stack ....................................................................................................................... 78
6.4.9 STK_destroy_stack: Destroy a Stack ....................................................................................... 79
6.4.10 Simple Stack Example.......................................................................................................... 79
6.4.11 Implementation Details......................................................................................................... 80
6.5 A More Robust Stack Module..................................................................................................... 82

6.5.1 Stack Marks............................................................................................................................... 82
6.5.2 Segmented Stacks...................................................................................................................... 84
7. PRIORITY QUEUES ................................................................................87

7.1 Objectives...................................................................................................................................... 87
7.2 Overview ....................................................................................................................................... 87
7.3 Queues ........................................................................................................................................... 88

7.3.1 QUE_create_queue.................................................................................................................... 90
7.3.2 QUE_create_item ...................................................................................................................... 91
7.3.3 QUE_clear_queue ..................................................................................................................... 91
7.3.4 Other QUE Module Methods .................................................................................................... 92
7.3.5 QUE Module Sample Program.................................................................................................. 93
7.4 Simple Priority Queues................................................................................................................ 93

7.4.1 PRQ_create_priority_queue ...................................................................................................... 95
7.4.2 PRQ_create_item....................................................................................................................... 96
7.4.3 PRQ_is_queue_empty ............................................................................................................... 96
7.4.4 PRQ_add_item .......................................................................................................................... 97
7.4.5 PRQ_remove_item .................................................................................................................... 97
7.4.6 PRQ_GET_DATA .................................................................................................................... 97
7.4.7 PRQ_GET_PRIORITY ............................................................................................................. 97
7.4.8 PRQ_destroy_item .................................................................................................................... 98
7.4.9 PRQ_empty_queue.................................................................................................................... 98
7.4.10 PRQ_destroy_queue ............................................................................................................. 98
7.4.11 Priority Queue Example ....................................................................................................... 99
7.4.12 Simple Priority Queue Module Implementation................................................................. 102
v 06/19/06
7.5 A More Robust Priority Queue Implementation..................................................................... 104
8. THE SYSTEM LIFE CYCLE ..................................................................107

8.1 Objectives.................................................................................................................................... 107
8.2 Overview ..................................................................................................................................... 107

8.2.1 Specification Phase.................................................................................................................. 107
8.2.2 Design Phase ........................................................................................................................... 108
8.2.3 Implementation Phase ............................................................................................................. 108
8.2.4 Acceptance Testing Phase ....................................................................................................... 108
8.2.5 Maintenance ............................................................................................................................ 109
8.3 Testing ......................................................................................................................................... 109

8.3.1 Testing at the System Specification Level............................................................................... 109
8.3.2 Testing at the Design Level..................................................................................................... 109
8.3.3 Testing at the Implementation Level ....................................................................................... 110
8.3.4 Testing at the Acceptance Testing Level................................................................................. 111
8.3.5 Testing at the Maintenance Level............................................................................................ 111
9. BINARY TREES ....................................................................................113

9.1 Objectives.................................................................................................................................... 113
9.2 Overview ..................................................................................................................................... 113
9.3 Binary Tree Representation ...................................................................................................... 115

9.3.1 Contiguous Array Representation ........................................................................................... 115
9.3.2 Dynamically Linked Representation ....................................................................................... 116
9.4 A Minimal Binary Tree Implementation ................................................................................. 116

9.4.1 Public Declarations.................................................................................................................. 117
9.4.2 Private Declarations ................................................................................................................ 117
9.4.3 BTREE_create_tree................................................................................................................. 118
9.4.4 BTREE_add_root .................................................................................................................... 118
9.4.5 BTREE_add_left ..................................................................................................................... 119
9.4.6 BTREE_add_right ................................................................................................................... 120
9.4.7 BTREE_get_root ..................................................................................................................... 120
9.4.8 BTREE_get_data, BTREE_get_left, BTREE_get_right ......................................................... 120
9.4.9 BTREE_is_empty.................................................................................................................... 121
9.4.10 BTREE_is_leaf................................................................................................................... 121
9.4.11 BTREE_traverse_tree ......................................................................................................... 122
9.4.12 BTREE_destroy_tree, BTREE_destroy_subtree ................................................................ 122
9.5 Using a Binary Tree as an Index............................................................................................... 124
9.6 Using a Binary Tree as an Index – Demonstration.................................................................. 127
9.7 Traversing a Binary Tree .......................................................................................................... 130

9.7.1 Inorder Traversal ..................................................................................................................... 131
9.7.2 Preorder Traversal ................................................................................................................... 131
9.7.3 Postorder Traversal.................................................................................................................. 132
vi 06/19/06
10. N-ARY TREES.......................................................................................135

10.1 Objectives.................................................................................................................................... 135
10.2 Overview ..................................................................................................................................... 135
10.3 A Small N-ary Tree Implementation ........................................................................................ 136

10.3.1 Public Data Types............................................................................................................... 137
10.3.2 Private Declarations............................................................................................................ 137
10.3.3 NTREE_create_tree............................................................................................................ 137
10.3.4 NTREE_add_root ............................................................................................................... 137
10.3.5 NTREE_add_child.............................................................................................................. 138
10.3.6 NTREE_add_sib: Add a Sibling to a Node ........................................................................ 138
10.3.7 NTREE_get_root ................................................................................................................ 139
10.3.8 NTREE_has_child .............................................................................................................. 139
10.3.9 NTREE_has_sib ................................................................................................................. 139
10.3.10 NTREE_get_data, NTREE_get_child, NTREE_get_sib .................................................... 140
10.3.11 NTREE_destroy_tree.......................................................................................................... 140
10.4 Directories................................................................................................................................... 140

10.4.1 A Simple Directory Module ............................................................................................... 143
10.4.2 Public Data Types............................................................................................................... 143
10.4.3 CDIR_create_dir................................................................................................................. 143
10.4.4 CDIR_add_child................................................................................................................. 143
10.4.5 CDIR_add_property ........................................................................................................... 144
10.4.6 CDIR_get_node .................................................................................................................. 144
10.4.7 CDIR_get_property ............................................................................................................ 145
10.4.8 CDIR_destroy_dir .............................................................................................................. 145
10.4.9 Implementation Structure ................................................................................................... 146
10.4.10 CDIR_create_dir Implementation....................................................................................... 146
10.4.11 CDIR_add_child Implementation....................................................................................... 147
10.4.12 CDIR_add_property ........................................................................................................... 148
10.4.13 CDIR_get_node Implementation........................................................................................ 148
10.4.14 CDIR_get_property Implementation .................................................................................. 149
10.4.15 CDIR_destroy_dir Implementation .................................................................................... 149
10.4.16 Directories Discussion Wrap-up......................................................................................... 150
PRACTICE FINAL EXAMINATION ..................................................................151

Sample Questions ..................................................................................................................................... 151
Answers ..................................................................................................................................................... 155
QUIZZES ..........................................................................................................159
Quiz 1......................................................................................................................................................... 159
Quiz 2......................................................................................................................................................... 160
Quiz 3......................................................................................................................................................... 161
Quiz 4......................................................................................................................................................... 162
Quiz 5......................................................................................................................................................... 163
vii 06/19/06
Quiz 6......................................................................................................................................................... 164
Quiz 7......................................................................................................................................................... 165
Quiz 8......................................................................................................................................................... 166
viii 06/19/06
Course Overview
C Programming: Data Structures and Algorithms is a ten week course, consisting of
three hours per week lecture, plus assigned reading, weekly quizzes and five homework
projects. This is primarily a class in the C programming language, and introduces the
student to data structure design and implementation.
Objectives
Upon successful completion of this course, you will have demonstrated the following
skills:
• The ability to write C-language code according to a project specification;
• The ability to develop C-language modules following standard industry practices
and conventions; and
• The ability to properly design data structures and the algorithms to transform
them.
In order to receive credit for this course, you must meet the following criteria:
• Achieve 80% attendance in class;
• Achieve a total of 70% on the final examination; and
• Satisfactorily complete all projects.
Instructor
Jack Straub
425 888 9119 (9:00 a.m. to 3:00 p.m. Monday through Friday)
jstraub@centurytel.net
http://faculty.washington.edu/jstraub/
Text Books
Required
No text book is required. However it is strongly recommended that you acquire one of the
data structures text books listed below; at least one of your projects will require you to do
your own research on a data structure not covered in class.
Recommended
C A Reference Manual, Fifth Edition by Samuel P. Harbison, and Guy L. Steele Jr.,
Prentice Hall, 2002
C Primer Plus, Fourth Edition by Stephen Prata, Sams Publishing, 2002
Recommended Data Structures Textbooks
Data Structures and Program Design in C, Second Edition by Robert Kruse et al.;
Prentice Hall, 1997
Fundamentals of Data Structures in C by Ellis Horowitz, Sartaj Sahni and Susan
Anderson-Freed; W. H. Freeman, 1992
Algorithms in C, Third Edition Parts 1 - 4 by Robert Sedgewick; Addison-Wesley, 1998
Introduction ix 06/19/06
Course Outline
Week Topics Assigned Reading Work Due
1 Basic Skills, Core Kruse Chapters 1 and 2

Module Horowitz Chapter 1
Sedgewick Chapters 1 and 2
2 Doubly Linked Lists Kruse Chapter 5, through 5.2 Quiz 1,

Horowitz Chapter 4 Project 1
Sedgewick Chapter 3, through 3.5
3 Sorting Kruse Chapter 7, through 7.7 Quiz 2

Horowitz Chapter 7, through 7.6
Sedgewick Chapter 6 through 6.5,
Chapter 8
4 Modules, Abstract Kruse Section 4.8 Quiz 3,

Data Types Horowitz Section 1.3 Project 2
Sedgewick Section 4.1
5 Stacks Kruse Sections 3.1 and 3.2 Quiz 4

Horowitz Section 3.1
Sedgewick Sections 4.2 through 4.5
6 Priority Queues Kruse Sections 7.8 through 7.10 Quiz 5,

Horowitz Section 3.2 Project 3
Sedgewick Section 4.6,
Chapter 9 through 9.1
7 System Life Cycle Kruse Chapter 9, through 9.4 Quiz 6

Horowitz Section 1.1,
Chapter 5, through 5.3
Sedgewick Chapter 5, through 5.4
8 Binary/N-ary Trees Kruse Chapter 10, through 10.2 Quiz 7,

Horowitz Chapter 5, remainder Project 4
Sedgewick Chapter 5, remainder
9 Final Exam Kruse Chapter 10, remainder Quiz 8,

Horowitz Chapter 10, through 10.5 Project 5
Sedgewick Chapter 16
10 Wrap up
Introduction x 06/19/06
Recommended Reading
Topics Assigned Reading
Week 1 Basic Skills, Core Module H&S Sections 5.10, 5.4.1, 5.8
Prata pp. 569 - 571, 475 - 481
Week 2 Doubly Linked Lists H&S Section 5.6, through 5.6.4

Prata pp. 668 – 676
Week 3 Sorting H&S Sections 20.5; Chapter 19

Prata pp. 650 – 654
Week 4 Modules, Abstract Data Types Prata pp. 676 – 698
Week 5 Stacks H&S Sections 7.4.4, 7.5.8
Week 6 Priority Queues Kruse Chapter 11

Prata pp. 689 – 711
Week 7 System Life Cycle Kruse Chapter 12
Week 8 Binary/N-ary Trees Prata 711 – 736
Week 9 Final Exam H&S Chapter 4
Week 10 Wrap up
Introduction xi 06/19/06
1. Basics
We will begin this section by reviewing C skills that are particularly important to the
study of data structures and algorithms, including typedefs, pointers and arrays, and
dynamic memory allocation. Next we will define a set of utilities that will be useful in
implementing the data structures that we will discuss in the remainder of the course. By
the end of this section you will be ready to complete your first project.
1.1 Objectives
At the conclusion of this section, and with the successful completion of your first project,
you will have demonstrated the ability to:
• Use typedef to declare the basic types used to represent a data structure;
• Use dynamic memory allocation to create the components of a data structure; and
• Implement core utilities to serve as the foundation of a software development
project.
1.2 Typedef
In most C projects, the typedef statement is used to create equivalence names for other C
types, particularly for structures and pointers, but potentially for any type. Using typedef
equivalence names is a good way to hide implementation details. It also makes your code
more readable, and improves the overall portability of your product.
1.2.1 Typedef and Portability

Typedef is frequently used to improve code portability. To cite a simple example,
suppose you had a need to declare a data structure that was guaranteed to occupy the
same amount of memory on every platform. If this structure had integer fields you might
be tempted to declare them to be either short or long, believing that these types would
always translate to two- or four-byte integers, respectively. Unfortunately ANSI C makes
no such guarantee. Specifically, if you declare a field to be type long, it will break when
you try to port your code to an Alpha running Digital UNIX, where an int is four bytes,
and a long is eight bytes. To avoid this problem you can use typedef in conjunction with
conditional compilation directives to declare integer equivalence types; on most
platforms the equivalence type for a four-byte field will be long, but under Alpha/Digital
UNIX it will be int:
#if defined( ALPHA ) && defined ( DIGITAL_UNIX )
typedef int BYTE4_t;
#else
typedef long BYTE4_t;
#endif
For the record, a variable or field that needed to be exactly four bytes will then be
declared like this:
BYTE4_t field_name;
Section 1: Basics 13 6/19/06

1.2.2 Typedef and Structures

To create a structure variable suitable for describing a street address, and to pass the
variable to a routine that could print it, you might use code like that shown in Figure 1-1,
which explicitly declares all structure components using the struct keyword. Normally,
however, you would declare equivalence names for the address structure using typedef,
and then declare structure variables and parameters using the equivalence names, as
shown in Figure 1-2. Note that, in this example, one typedef statement was used to create
two equivalence names: ADDRESS_t, which is equivalent to struct address_s, and
ADDRESS_p_t, which is equivalent to struct address_s*. This is a very common
technique in modern C usage.
struct address_s
{
char *street;
char *city;
char *region;
char *country;
char *postal_code;
};
static void print_address(

struct address_s *address_info
);
static void print_an_address( void )

{
struct address_s address;
address.street = “1823 23rd Ave NE”;

address.city = “Seattle”;
address.region = “WA”;
address.postal_code = “98023”;
print_address( &address );
}
Figure 1-1: Traditional Structure Declarations
1.2.3 Typedef and Functions

Recall that the type of a function is a combination of the function’s return type, and the
number and type of each of the function’s parameters. Typedef can be used to declare an
equivalence name for a function type, and, subsequently, the equivalence name can be
used to declare a prototype for a function of that type. One use of this is to reduce
repetition when declaring many functions of the same type. For example, if you are
implementing a standard sort algorithm, you will need to define a prototype for a
compare function; specifically, a function used to determine whether one object is greater
than another. A prototype for such a function would traditionally be declared like this:

/* Returns 1 if item1 is greater than *
* or equal to item2; 0 otherwise. */
static int is_greater_equal(
const void *item1,
const void *item2
);
typedef struct address_s

{
char *street;
char *city;
char *region;
char *country;
char *postal_code;
} ADDRESS_t, *ADDRESS_p_t;
static void print_address(

ADDRESS_p_t address_info
);
static void print_an_address( void )

{
ADDRESS_t address;
address.street = “1823 23rd Ave NE”;

address.city = “Seattle”;
address.region = “WA”;
address.postal_code = “98023”;
print_address( &address );
}
Figure 1-2: Structure Declarations using typedef
Instead, however, we might consider declaring an equivalence type for a compare

function, then declaring each compare function prototype using the equivalence type. We
could do that this way:
typedef int COMPARE_PROC_t( const void *, const void * );
static COMPARE_PROC_t is_greater_equal;

This technique can also be used to drastically simplify compound declarations. For
example, what if we had to declare a field in a structure to be a pointer to a compare
function? This is a frequent occurrence, and traditionally would be done this way:
typedef sort_data_s
{
int *sort_array;
int (*test_proc)( const void *item1, const void *item2 );
} SORT_DATA_t, *SORT_DATA_p_t;
This structure declaration becomes much easier to write (and to read) if you use the
compare function equivalence type from the previous example, and extend it to embrace
type pointer to compare function, as shown in Figure 1-3.
Possibly the ultimate example of using typedef to simplify function-related declarations
is to rewrite the prototype for the ANSI function signal. Signal is a function with two

parameters; the first is type int, and the second is pointer to function with one parameter
of type int that returns void; the return value of signal is pointer to function with one
parameter of type int that returns void. The prototype for signal is usually declared like
this:
void (*signal(
int sig,
void (*func)(int)))(int);
This prototype is much easier to decipher if we just declare and use an equivalence for
type pointer to function with one parameter of type int that returns void:
typedef void SIG_PROC_t( int );
typedef SIG_PROC_t *SIG_PROC_p_t;
SIG_PROC_p_t signal(
int sig,
SIG_PROC_p_t func
);
typedef COMPARE_PROC_t *COMPARE_PROC_p_t;
typedef struct sort_data_s

{
int *sort_array;
COMPARE_PROC_p_t test_proc;
typedef COMPARE_PROC_t *COMPARE_PROC_p_t;
typedef struct sort_data_s

{
int *sort_array;
COMPARE_PROC_p_t test_proc;
Figure 1-3: Typedef and Function Types
1.3 Pointers and Arrays

In C, pointers and arrays are very closely related. With only two exceptions, the name of
an array is equivalent to a pointer to the first element of the array. In Figure 1-4, the first
parameter of the sort_dates function is declared to be pointer to date structure; the actual
call to sort_dates passes the name of an array of date_structures as the corresponding
argument, and the C compiler obligingly converts the name to a pointer. Since a C
subroutine can’t determine the length, or cardinality, of an array passed as an argument,
the second argument to sort_dates specifies the length of the array.
The two exceptions are that an array name cannot be an lvalue (it cannot appear on the
left side of the equal sign in an assignment); and that C does not treat the name as a
pointer when used as the argument of the sizeof operator. If you use the name of an array
as an argument of sizeof, the size of the entire array is computed. This allows us to create
a very convenient macro which I have called CARD (short for cardinality); this takes the
size of the array divided by the size of the first element of the array, which yields the total

number of elements in the array. As illustrated in Figure 1-5, this macro allows us to
dynamically determine the size of an array.
1.4 Dynamic Memory Allocation

Dynamic memory allocation in C is performed using one of the standard library functions
malloc, calloc or realloc (or a cover for one of these routines). Dynamically allocated
memory must eventually be freed by calling free. If allocated memory is not properly
freed a memory leak results, which could result in a program malfunction.
typedef struct date_s

{
short year;
char month;
char day;
} DATE_t, *DATE_p_t;
static void sort_dates( DATE_p_t dates, int num_dates );
DATE_t dates[4] = { {1066, 3, 27},

{1941, 12, 1},
{1492, 10, 12},
{1815, 10, 14}
};
. . .
sort_dates( dates, 4 );
Figure 1-4: Pointers and Arrays
The ability to dynamically allocate memory accounts for much of the power of C. It also
accounts for much of the complexity of many C programs, and, subsequently, is the
source of many of their problems. One of the biggest problems associated with
dynamically allocated memory comes from trying to deal with allocation failure. Many
organizations effectively short-circuit this problem by writing cover routines for the
dynamic memory allocation routines that do not return when an error occurs; instead,
they abort the program. In Figure 1-6 we see a cover for malloc; writing cover routines
for calloc and realloc is left as an exercise for the student.
#define CARD( arr ) (sizeof((arr))/sizeof(*(arr)))

. . .
sort_dates( dates, CARD( dates ) );
Figure 1-5: Arrays and sizeof
1.5 The Core Module

I am going to leave a formal definition of the term module for later. For now, let’s just
say that a module is a collection of related facilities, including functions, macros, and
data type declarations. A module consists of a minimum of three files, a private header
file, a public header file and a principal source file. In this section we will develop the
core module, which will contain a collection of facilities to assist us in writing code
throughout the remainder of the course. The name of the module will be CDA (short for
C: Data Structures and Algorithms) and will consist of the following three files:
• cdap.h (the private header file),
• cda.h (the public header file) and
• cda.c (the principal source file)
#include <stdlib.h>
void *PROJ_malloc( size_t size )

{
void *mem = malloc( size );
if ( mem == NULL && size > 0 )

abort();
return mem;
}
Figure 1-6: Cover routine for malloc
1.5.1 The Private Header File

The concept of a private header file will occupy much of our time later in the course. For
our first module it must be present (in this course every module must have a private
header file), but otherwise has little significance. It consists entirely of an include
sandwich (recall the every header file must have an include sandwich), and a statement to
include the public header file. It is shown in its entirety in Figure 1-7.
#ifndef CDAP_H /* begin include sandwich */

#define CDAP_H /* second line of include sandwich */
#include <cda.h> /* include public header file */
#endif /* end include sandwich */
Figure 1-7: CDA Module Private Header File cdap.h
1.5.2 The Principal Source File

The principal source file for the CDA module will contain cover routines for malloc,
calloc, realloc and free. The cover routine for malloc resembles closely the cover that we
saw in Section 1.4, and is shown below:
void *CDA_malloc( size_t size )
{
void *mem = malloc( size );
if ( mem == NULL && size > 0 )

abort();
return mem;
}
The cover routines for calloc and realloc will be called CDA_calloc and CDA_realloc,
respectively, and are left as an exercise for the student. The cover routine for free is
called CDA_free, and is shown in Figure 1-8. Outside of the cover routines in cda.c,
none of the code that we write in this class will ever make direct use of the standard
memory allocation routines. Instead, they will make use of the CDA covers.
1.5.3 The Public Header File

The CDA module public header file consists of three parts, as discussed below.
void CDA_free( void *mem )

{
if ( mem != NULL )
free( mem );
}
Figure 1-8: CDA_free
Part 1: Common or Convenient Macros

This part of the public header file consists of the declaration of macros for true
and false to help make our code more readable, and macros to encapsulate
frequent operations to help make our code more reliable. The declarations are
shown in Figure 1-9, and discussed in the following paragraphs.
#define CDA_TRUE (1)

#define CDA_FALSE (0)
#define CDA_ASSERT( exp ) (assert( (exp) ))

#define CDA_CARD( arr ) (sizeof((arr))/sizeof(*(arr)))
#define CDA_NEW( type ) ((type *)CDA_malloc( sizeof(type) ))
#define CDA_NEW_STR( str ) \
(strcpy( (char *)CDA_malloc( strlen( (str) ) + 1 ), (str) ))
#define CDA_NEW_STR_IF( str ) \
((str) == NULL ? NULL : CDA_NEW_STR( (str) ))
Figure 1-9: CDA Module Public Macros
• CDA_TRUE and CDA_FALSE merely help to make our code more

readable by giving us symbolic names for Boolean values.
• CDA_ASSERT, for now, is a simple cover for the standard library assert
macro. On many projects it is common to use alternative assert macros
that provide more detailed feedback then the standard assert macro. We
are not going to do that now; we do, however, want to leave open the
possibility of doing it later. So we will write a macro to use in place of the
usual assert and use it in all of our projects; later, if we decide to write a
new version, all we will have to do is change cda.h and recompile our
code in order to take advantage of it. Here is an example of its use:

Instead of:
assert( size <= MAXIMUM_SIZE );
Use:
CDA_ASSERT( size <= MAXIMUM_SIZE );
• CDA_CARD is simply a generic implementation of the CARD macro that
we saw in Section 0. For example:
int inx = 0;
int iarr[10];
for ( inx = 0 ; inx < CDA_CARD( iarr ) ; ++inx )
iarr[inx] = -1;
• CDA_NEW is a macro to encapsulate what we call a new operation; that is,
the allocation of memory to serve as a data structure instance, or a
component of a data structure instance. For example:
Instead of:
ADDRESS_p_t address = CDA_malloc( sizeof(ADDRESS_t) );
Use:
ADDRESS_p_t address = CDA_NEW( ADDRESS_t );
• CDA_NEW_STR and CDA_NEW_STR_IF encapsulate the operations
needed to make a copy of a string. This is an important activity, because
what we call a string in C is merely the address of an array of type char. If
you try to store such an address in a data structure you leave yourself open
to problems because that memory typically belongs to someone else, and it
can be modified or deallocated at any time; so, before storing a string in a
data structure, you must be certain that it occupies memory that you
control. CDA_NEW_STR unconditionally makes a copy of a string;
CDA_NEW_STR_IF evaluates to NULL if the target string is NULL, and
to CDA_NEW_STR if the string is non-NULL. Here is an example of their
use:
Instead of:
if ( string == NULL )
copy = NULL;
else
{
copy = CDA_malloc( strlen( string ) + 1 );
strcpy( copy, string );
}
Use:
copy = CDA_NEW_STR_IF( string );
Part 2: Common or Convenient Type Declarations
This part of the public header file consists of the declaration of a Boolean type to
help make our code more readable, plus declarations of types to represent integers
of a specific size to help make our code more portable. The representative integer
types will be:

• signed and unsigned eight-bit integer: CDA_INT8_t and CDA_UINT8_t

• signed and unsigned sixteen-bit integer: CDA_INT16_t and
CDA_UINT16_t
• signed and unsigned thirty two-bit integer: CDA_INT32_t and
CDA_UINT32_t
The first three such type declarations are shown in ; the remaining declarations
are left as an exercise.
typedef int CDA_BOOL_t;

typedef signed char CDA_INT8_t;
typedef unsigned char CDA_UINT8_t;
Figure 1-10: CDA Module Public Type Declarations
Part 3: Prototypes for the Public Functions

This part of the public header file consists of the prototypes for the functions in
cda.c.
1.6 Activity
Interactively develop a module to encapsulate a timer.

2. Doubly Linked Lists

In this section we will examine one of the most common and versatile data structures in
data processing: the doubly linked list. In designing the VAX, Digital Equipment
Corporation engineers felt that this type of list was so important that they designed
machine-level instructions for manipulating them.
In addition to learning about doubly linked lists, in this lesson you will begin to learn
how to formally define data structures, and to encapsulate data structure functionality in
modules.
2.1 Objectives
At the conclusion of this section, and with the successful completion of your second
project, you will have demonstrated the ability to:
• Formally define the enqueuable item and doubly linked list structures; and
• Implement a circular, doubly linked list data structure.
2.2 Overview
There are several ways to implement a doubly linked list. The common feature of all such
implementations is that a doubly linked list consists of zero or more enqueuable items.
An enqueuable item consists of, at a minimum, two pointers; when an enqueuable item is
enqueued, one pointer provides a forward link to the next item in the list (if any) and the
other a backward link to the previous item in the list (if any). Figure 2-1 portrays three
enqueuable items in a list.

Figure 2-1: Enqueuable Items
The specific implementation that we will consider in this course is a doubly linked list
that is anchored by a queue head, and that is circular. The anchor consists of, at a
minimum, a forward/backward link pair. The forward link points to the first item in the
list, and the backward link points to the last item in the list. The list is circular because
the forward link of the last item in the list, and the backward link of the first item in the
list point to the anchor. This arrangement is illustrated in Figure 2-2.
Section 2: Doubly Linked Lists 23 06/19/06

Anchor
Additional Additional
Data Data
Figure 2-2 Anchored, Circular List
Before we go on, a quick word about the picture shown in Figure 2-2. This is the
standard way of representing a doubly linked list, but it sometimes leads to confusion
among students. In particular, it is easy to interpret the drawing as if the backward link of
an item held the address of the backward link of the previous item. This is not so; the
forward and backward links of an item always contain the base addresses of the structures
it links to. A slightly more realistic representation of a linked list can be seen in Figure
2-3; however, this representation is harder to draw and harder to read, so the style of
representation used in Figure 2-2 is the one that we will use for the remainder of this
course.
Anchor
Additional Additional
Data Data
Figure 2-3 Another List Representation
2.3 Definitions
In this section we will discuss formal definitions for the elements of our implementation
of a linked list. Specifically, we need to define exactly what we mean by a doubly linked
list, an anchor and an enqueuable item; we also have to strictly define the states that these
items may occupy.

2.3.1 Enqueuable Item

In our implementation, an enqueuable item is a struct consisting of two pointers of type
pointer to enqueuable item, followed by a name and zero or more bytes of application
data. The two pointers, called flink for forward link and blink for backward link, must
occupy the first two fields of the struct, and the name (type char*) must occupy the third.
These first three fields are the static part of the item. In C, we can’t declare a data type
representing a variable amount of data, but we can create a data type to represent the
static portion as follows:
typedef struct enq_item_s
{
struct enq_item_s *flink;
struct enq_item_s *blink;
char *name;
} ENQ_ITEM_t, *ENQ_ITEM_p_t;
Enqueuable items that contain additional data can be declared by creating a type whose
first member is a field of type ENQ_ITEM_t. An example is shown in Figure 2-4.
typedef struct address_s

{
ENQ_ITEM_t item;
char name[31];
char address1[31];
char address2[31];
char city[31];
char state[31];
char postal_code[21];
} ADDRESS_t, *ADDRESS_p_t;
Figure 2-4 Enqueuable Item with Application Data
Now that we know what an enqueuable item looks like, we have to define the states that
such an item can occupy. In this case there are only two: an enqueuable item may be
enqueued or unenqueued. We define an enqueued item as one whose flink and blink point
to other enqueuable items; when an item is unenqueued, its flink and blink point to itself.
This is illustrated in Figure 2-5.
Figure 2-5 Enqueuable Item in the Unenqueued State

2.3.2 Anchor
An anchor is a struct consisting of a flink, a blink and a name; in other words, it is an
enqueuable item with no application data. The declaration of the anchor type is shown
below:
typedef ENQ_ITEM_t ENQ_ANCHOR_t, *ENQ_ANCHOR_p_t;
2.3.3 Doubly Linked List

We define a doubly linked list as an anchor plus zero or more enqueuable items; a list is
always identified by the address of its anchor.
Note: For the remainder of this course, unqualified use of the word list will be
synonymous with doubly linked list. Unqualified reference to the anchor of a list
typically will be interchangeable with a reference to the list itself; for example,
“pass the list to subroutine X” is interchangeable with “pass the address of the
anchor of the list to subroutine X.”
A doubly linked list may occupy two states: empty and nonempty. A nonempty list
contains at least one item; an empty list contains none. When a list is in a nonempty state,
the flink and blink of the anchor point to the first and last items in the list, respectively;
the blink of the first item and the flink of the last item each point to the anchor. When the
list is empty, the flink and blink of the anchor each point to the anchor. In a nonempty
list, the first item in the list is referred to as the list head or just head; the last item in the
list is referred to as the list tail, or just tail. This is illustrated in Figure 2-6.
Now that we know how the structure of a doubly linked list is defined, we need to define
the operations that may be performed on a list, and the protocols required to perform
those operations. Operations, together with the protocols for performing them are called
methods, and will be discussed in the next section.
2.3.4 Methods
In this class we will define the following operations that may be performed on a doubly
linked list, and the elements that belong to a doubly linked list:
• Create a new doubly linked list
• Create a new enqueuable item
• Test whether an item is enqueued
• Test whether a list is empty
• Add an item to the head of a list
• Add an item to the tail of a list
• Add an item after a previously enqueued item
• Add an item before a previously enqueued item
• Dequeue an item from a list
• Dequeue the item at the head of a list

Figure 2-6 List States
• Dequeue the item at the tail of a list

• Get the item at the head of a list (without dequeing it)
• Get the item at the tail of a list
• Given an item, get the following item
• Given an item, get the previous item
• Get the name of a list
• Get the name of an item
• Destroy an item
• Empty a list
• Destroy a list
In the following sections, we will discuss some the details of the above methods.
Determining the remaining details, and actually implementing the methods will constitute
your second project.
2.3.5 ENQ_create_list: Create a New Doubly linked List

This method will dynamically allocate the space for a new list anchor, and return its
address to the caller. The caller is responsible for ultimately freeing the memory by
calling ENQ_destroy_list when the list is no longer needed.
Synopsis:
ENQ_ANCHOR_p_t ENQ_create_list( const char *name );
Where:
name -> the name of the list

Returns:
The address of the list
Notes:
1. The caller is responsible for freeing the memory occupied
by the list by calling ENQ_destroy_list.
2. The list name is copied into a private buffer which is
freed when the list is destroyed.
The implementation of this method is straightforward; the only trick is to remember to
initialize the queue to an empty state after creating it:
ENQ_ANCHOR_p_t ENQ_create_list( const char *name )
{
ENQ_ANCHOR_p_t list = CDA_NEW( ENQ_ANCHOR_t );
list->flink = list;
list->blink = list;
list->name = CDA_NEW_STR_IF( name );
return list;
}
2.3.6 ENQ_create_item: Create a New Enqueuable Item

Creating a new item is essentially the same as creating a new list; allocate space for the
item, then initialize it to the unenqueued state. The only difference is that allowance must
be made for the extra application data space. To this end an extra argument must be
passed to indicate the entire amount of space to be allocated; note that this includes the
space required for the static portion of the item. We’re also going to add an assertion
to verify that the requested size is at least as great as required. When the item is no longer
needed, the caller is responsible for freeing the memory it occupies by calling
ENQ_destroy_item.
Synopsis:
ENQ_ITEM_p_t ENQ_create_item( const char *name,
size_t size
);
Where:
name -> the name of the item
size == size of item required
Returns:
The address of the item
Notes:
by the item by calling ENQ_destroy_item.
2. The item name is copied into a private buffer which is
freed when the item is destroyed.
Here is the implementation of this method:
ENQ_ITEM_p_t ENQ_create_item( const char *name, size_t size )
{
ENQ_ITEM_p_t item = (ENQ_ITEM_p_t)CDA_malloc( size );
CDA_ASSERT( size >= sizeof(ENQ_ITEM_t) );

item->flink = item;

item->blink = item;
item->name = CDA_NEW_STR_IF( name );
return item;
}
2.3.7 ENQ_is_item_enqed: Test Whether an Item is Enqueued

Determining whether an item is enqueued merely requires knowing the definition of the
possible states for an item, as discussed above.
Synopsis:
CDA_BOOL_t ENQ_is_item_enqed( ENQ_ITEM_p_t item );
Where:
item -> item to test
Returns:
CDA_TRUE if the item is enqueued, CDA_FALSE otherwise
Notes:
None
Here’s the method implementation:
CDA_BOOL_t ENQ_is_item_enqed( ENQ_ITEM_p_t item )
{
CDA_BOOL_t rcode =
(item->flink == item ? CDA_FALSE : CDA_TRUE);
return rcode;
}
2.3.8 ENQ_is_list_empty: Test Whether a List is Empty

As in the previous section, determining whether a list is empty merely requires knowing
the definition of the possible states for a list.
Synopsis:
CDA_BOOL_t ENQ_is_list_empty( ENQ_ANCHOR_p_t list );
Where:
list -> list to test
Returns:
CDA_TRUE if the list is empty, CDA_FALSE otherwise
Notes:
None
The implementation of this method is left to the student.
2.3.9 ENQ_add_head: Add an Item to the Head of a List

This method inserts an item at the front of a list.
Synopsis:
ENQ_ITEM_p_t ENQ_add_head( ENQ_ANCHOR_p_t list,
ENQ_ITEM_p_t item
);
Where:
list -> list in which to enqueue
item -> item to enqueue

Returns:
address of enqueued item
Notes:
None
The implementation of this method is left to the student. You should use an assertion to
verify that the item is not already enqueued before performing the operation:
CDA_ASSERT( !ENQ_is_item_enqed( item ) );
2.3.10 ENQ_add_tail: Add an Item to the Tail of a List

This operation is nearly identical to ENQ_add_head.
Synopsis:
ENQ_ITEM_p_t ENQ_add_tail( ENQ_ANCHOR_p_t list,
ENQ_ITEM_p_t item
);
Where:
list -> list in which to enqueue
Returns:
Notes:
None
verify that the item is not already enqueued before performing the operation.
2.3.11 ENQ_add_after: Add an Item After a Previously Enqueued Item

Synopsis:
ENQ_ITEM_p_t ENQ_add_after( ENQ_ITEM_p_t item,
ENQ_ITEM_p_t after
);
Where:
after -> item after which to enqueue
Returns:
Notes:
None
verify that the item to enqueue is not already enqueued.
2.3.12 ENQ_add_before: Add an Item Before a Previously Enqueued Item

Synopsis:
ENQ_ITEM_p_t ENQ_add_before( ENQ_ITEM_p_t item,
ENQ_ITEM_p_t before
);
Where:
before -> item before which to enqueue

Returns:
Notes:
None
verify that the item to enqueue is not already enqueued.
2.3.13 ENQ_deq_item: Dequeue an Item from a List

This method will remove an item from a list and return its address. Note that, because of
our careful definition of an item in an unenqueued state, it is not necessary to make sure
that the item is enqueued before dequeing it; the chosen algorithm works equally well on
both enqueued and unenqueued items.
Synopsis:
ENQ_ITEM_p_t ENQ_deq_item( ENQ_ITEM_p_t item );
Where:
item -> item to dequeue
Returns:
address of dequeued item
Notes:
None
The only trick to performing this operation is to make sure that, once dequeued, the item
is set to an unenqueued state.
ENQ_ITEM_p_t ENQ_deq_item( ENQ_ITEM_p_t item )
{
item->blink->flink = item->flink;
item->flink->blink = item->blink;
item->flink = item;
item->blink = item;
return item;
}
2.3.14 ENQ_deq_head: Dequeue the Item at the Head of a List

This method removes the item from the head of a list and returns its address.
Synopsis:
ENQ_ITEM_p_t ENQ_deq_head( ENQ_ANCHOR_p_t list );
Where:
list -> list from which to dequeue
Returns:
If queue is nonempty, the address of the dequeued item;
Otherwise the address of the list
Notes:
None
The implementation of this method is left to the student. Take careful note of the
documentation for the return value, and remember to initialize the dequeued item after
dequeing it.

2.3.15 ENQ_deq_tail: Dequeue the Item at the Tail of a List

This method removes the item from the tail of a list and returns its address.
Synopsis:
ENQ_ITEM_p_t ENQ_deq_tail( ENQ_ANCHOR_p_t list );
Where:
list -> list from which to dequeue
Returns:
If queue is nonempty, the address of the dequeued item;
Notes:
None
The implementation of this method is left to the student. Take careful note of the
documentation of the return value, and remember to initialize the dequeued item after
dequeing it.
2.3.16 ENQ_GET_HEAD: Get the Address of the Item at the Head of a List
This method returns the address of the item at the head of a list without dequeing it. It is
so straightforward that many list implementations don’t bother with it. In this class,
however, we will concentrate on providing all operations on a data structure from within
the module that owns the data structure. We will, in this case, follow a middle ground;
rather than implement the method as a procedure, we will make it a macro.
Note: Some students are confused by the synopsis, below, thinking that it
shows the prototype of a function, contradicting the above statement that this
will be implemented as a macro. This is not true. This is merely a standard
why of summarizing how a method is used, whether its implementation is as a
function, or a function-like macro. See, for example, the documentation for
getc macro in Chapter 15 of Harbison & Steele.
Synopsis:
ENQ_ITEM_p_t ENQ_GET_HEAD( ENQ_ANCHOR_p_t list );
Where:
list -> list to interrogate
Returns:
If queue is nonempty, the address of the first list item;
Notes:
None
Here’s the implementation of the method.
#define ENQ_GET_HEAD( list ) ((list)->flink)
2.3.17 ENQ_GET_TAIL: Get the Address of the Item at the Tail of a List
This method, also a macro, is nearly identical to ENQ_GET_HEAD.
Synopsis:
ENQ_ITEM_p_t ENQ_GET_TAIL( ENQ_ANCHOR_p_t list );
Where:

Returns:
If queue is nonempty, the address of the last list item;
Notes:
None
The implementation of this method is left to the student. Like ENQ_GET_HEAD it
should be implemented as a macro.
2.3.18 ENQ_GET_NEXT: Get the Address of the Item After a Given Item
Given an item, this method, implemented as a macro, returns the address of the next item
in the list without dequeing it.
Synopsis:
ENQ_ITEM_p_t ENQ_GET_NEXT( ENQ_ITEM_p_t item );
Where:
item -> item to interrogate
Returns:
If there is a next item, the address of the next item;
Otherwise the address of the list that item belongs to
Notes:
None
The implementation of this method is left to the student. It should be implemented as a
macro.
2.3.19 ENQ_GET_PREV: Get the Address of the Item Before a Given Item
Given an item, this macro returns the address of the previous item in the list without
dequeing it.
Synopsis:
ENQ_ITEM_p_t ENQ_GET_PREV( ENQ_ITEM_p_t item );
Where:
Returns:
If there is a previous item, the address of the previous item;
Otherwise the address of the list that item belongs to
Notes:
None
macro.
2.3.20 ENQ_GET_LIST_NAME: Get the Name of a List

This may seem like a trivial operation, but later on if we decide to change the
implementation we’ll be glad we implemented this as a method. We’ll make it a macro.
Notice that the return type is a pointer to a constant string; the caller may NOT modify it.
Synopsis:
const char *ENQ_GET_LIST_NAME( ENQ_ANCHOR_p_t list );
Where:

Returns:
The name of the list
Notes:
The string representing the list name belongs to the
implementation; the caller may not modify it.
Here’s the implementation:
#define ENQ_GET_LIST_NAME( list ) \
((const char *)((list)->name))
2.3.21 ENQ_GET_ITEM_NAME: Get the Name of an Item

This method is nearly identical to ENQ_GET_LIST_NAME. It will be a macro that
returns the name of an item as type (const char *).
Synopsis:
const char *ENQ_GET_ITEM_NAME( ENQ_ITEM_p_t item );
Where:
Returns:
The name of the item
Notes:
The string representing the item name belongs to the
implementation; the caller may not modify it.
macro.
2.3.22 ENQ_destroy_item: Destroy an Item

This method will free the memory associated with an item. If the item is enqueued, it will
be dequeued before freeing. Note that the method explicitly return NULL.
Synopsis:
ENQ_ITEM_p_t ENQ_destroy_item( ENQ_ITEM_p_t item );
Where:
item -> item to destroy
Returns:
NULL
Notes:
The item to destroy may be enqueued or unenqueued. If
enqueued, it will be dequeued prior to destruction.
There are two things to remember as part of the implementation. First, don’t forget to free
the item name. Second, thanks to the way that we chose to define an enqueuable item,
there is no need to test whether an item is enqueued before we dequeue it.
ENQ_ITEM_p_t ENQ_destroy_item( ENQ_ITEM_p_t item )
{
ENQ_deq_item( item );
CDA_free( item->name );
CDA_free( item );
return NULL;
}

2.3.23 ENQ_empty_list: Empty a List

This method will remove all items from a list, and destroy them, leaving the list empty.
Synopsis:
ENQ_ANCHOR_p_t ENQ_empty_list( ENQ_ANCHOR_p_t list );
Where:
list -> list to empty
Returns:
The address of the list
Notes:
All items enqueued in the list will be destroyed.
Here is the implementation:
ENQ_ANCHOR_p_t ENQ_empty_list( ENQ_ANCHOR_p_t list )
{
while ( !ENQ_is_list_empty( list ) )
ENQ_destroy_item( list->flink );
return list;
}
2.3.24 ENQ_destroy_list: Destroy a List

This method will empty a list, then free the memory occupied by the list anchor. Note
that it explicitly returns NULL.
Synopsis:
ENQ_ANCHOR_p_t ENQ_destroy_list( ENQ_ANCHOR_p_t list );
Where:
list -> list to destroy
Returns:
NULL
Notes:
All items enqueued in the list will be destroyed.
The implementation of this method is left to the student. Remember to empty the list and
free the list name before freeing the anchor.
2.4 Case Study

Alice has joined a team of programmers working on an accounting system for a
restaurant. She has been given the job of writing a module that will temporarily
accumulate the total receipts and tips for the restaurant’s waiters and waitresses, and then
print out the results. The major requirements she was given are listed below:
1. The module must have an initialization method which will be called once, and which
should initialize the module’s data structures to an empty state.
2. The module must have a shutdown method which will be called once, and which will
free any resources allocated for the module.
3. The module must have a method that will accept the name of an employee, the
amount of a single check, and the amount of the tip associated with the check. For
any employee there are likely to be many checks, hence many calls to this method.

4. The module must have a print method that will print, alphabetically, the name of each
employee, their total receipts and tips, and the average amount of each check and tip.
Here is an example of the output of the print method:
Brenda
Total receipts: 328.99 (Average: 82.25)
Total tips: 62.00 (Average: 15.50)
Tom
Terry
Alice has decided that she will implement the accumulation method by allocating a
bucket for each employee as the employee is “discovered,” that is, when the
accumulation method has been passed the name of an employee for whom a bucket has
not already been allocated. Each time a new receipt is received for a known employee,
the amount of the receipt will be accumulated in the existing bucket. She has also decided
that she will store each bucket in a list of some kind, in alphabetical order. Here is the
pseudocode for the accumulation method, which Alice has decided to call addReceipt:
addReceipt( name, check, tip )
for each bucket in list
if bucket-name == name
add check to bucket
add tip to bucket
increment # receipts in bucket
else if bucket-name > name
allocate new-bucket
add check to new-bucket
add tip to new-bucket
set # receipts in new-bucket = 1
add new-bucket to list before bucket
else
next bucket
if no bucket in list satisfies the above:
allocate new-bucket
add check to new-bucket
add tip to new-bucket
set # receipts in new-bucket = 1
add new-bucket to end of list
Alice has decided to create and maintain her list using the ENQ module. That means that
her remaining three methods will be implemented rather simply, as follows:
1. The initialization method will create the list.
2. The print method will traverse the list from front to back, printing out the
employee information found in each bucket.
3. The shutdown method will destroy the list.
In keeping with local project standards, Alice now has to select a name for her module,
and create public and private header files for it. She has chosen the name TIPS. Her
public and private header files (with comments removed to save space) follow:
/* TIPS Private Header File */
#ifndef TIPSP_H
#define TIPSP_H
#include <tips.h>
#endif /* #ifndef TIPSP_H */
/* TIPS Public Header File */

#ifndef TIPS_H
#define TIPS_H
void TIPS_addReceipt(
const char *waitress,
double check,
double tip
);
void TIPS_close(
void
);
void TIPS_init(
void
);
void TIPS_printReceipts(
void
);
#endif /* #ifndef TIPS_H */
Alice has only one decision left: how to implement her “bucket.” She knows that this will
be a data structure with fields for accumulating checks, tips, and check count. Since she
has decided to implement her list via the ENQ module, she knows that her bucket will
have to be an enqueuable item, as defined by the ENQ module; that is, it will have to
have as its first member an ENQ_ITEM_t structure. Alice’s implementation of the TIPS
source file is shown below.
#include <cda.h>
#include <enq.h>
#include <tipsp.h>
#define NOT_FOUND (0)

#define FOUND_EXACT (1)
#define FOUND_GREATER (2)
typedef struct receipts_s

{
ENQ_ITEM_t item;
double checkTotal;
double tipTotal;
int numReceipts;
} RECEIPTS_t, *RECEIPTS_p_t;
static const char *tipListName = "Tip Queue";
static ENQ_ANCHOR_p_t anchor = NULL;

void TIPS_init( void )

{
CDA_ASSERT( anchor == NULL );
anchor = ENQ_create_list( tipListName );
}
void TIPS_addReceipt( const char *waitperson, double check, double tip )

{
RECEIPTS_p_t receipts = NULL;
RECEIPTS_p_t bucket = NULL;
int result = NOT_FOUND;
int compare = 0;
CDA_ASSERT( anchor != NULL );
receipts = (RECEIPTS_p_t)ENQ_GET_HEAD( anchor );

while ( (result == NOT_FOUND) && ((ENQ_ANCHOR_p_t)receipts != anchor) )
{
compare = strcmp( waitperson, ENQ_GET_ITEM_NAME( (ENQ_ITEM_p_t)receipts ));
if ( compare == 0 )
result = FOUND_EXACT;
else if ( compare < 0 )
result = FOUND_GREATER;
else
receipts = (RECEIPTS_p_t)ENQ_GET_NEXT( (ENQ_ITEM_p_t)receipts );
}
switch ( result )
{
case FOUND_EXACT:
receipts->checkTotal += check;
receipts->tipTotal += tip;
++receipts->numReceipts;
break;
case FOUND_GREATER:
bucket = (RECEIPTS_p_t)ENQ_create_item( waitperson, sizeof(RECEIPTS_t) );
bucket->checkTotal = check;
bucket->tipTotal = tip;
bucket->numReceipts = 1;
ENQ_add_before( (ENQ_ITEM_p_t)bucket, (ENQ_ITEM_p_t)receipts );
break;
case NOT_FOUND:
bucket = (RECEIPTS_p_t)ENQ_create_item( waitperson, sizeof(RECEIPTS_t) );
bucket->checkTotal = check;
bucket->tipTotal = tip;
bucket->numReceipts = 1;
ENQ_add_tail( anchor, (ENQ_ITEM_p_t)bucket );
break;
default:
CDA_ASSERT( CDA_FALSE );
break;
}
}
void TIPS_printReceipts( void )

{
RECEIPTS_p_t receipts = NULL;
receipts = (RECEIPTS_p_t)ENQ_GET_HEAD( anchor );

while ( receipts != (RECEIPTS_p_t)anchor )
{
printf( "%s\n", ENQ_GET_ITEM_NAME( (ENQ_ITEM_p_t)receipts ) );
printf( "Total receipts: %.2f (Average: %.2f)\n",
receipts->checkTotal,
receipts->checkTotal / receipts->numReceipts

);
printf( "Total tips: %.2f (Average: %.2f)\n",
receipts->tipTotal,
receipts->tipTotal / receipts->numReceipts
);
printf( "\n" );
receipts = (RECEIPTS_p_t)ENQ_GET_NEXT( (ENQ_ITEM_p_t)receipts );
}
}
void TIPS_close( void )

{
ENQ_destroy_list( anchor );
anchor = NULL;
}
2.5 Activity
Discuss the concept of subclassing, and ways of implementing it in C.

3. Sorting
In this section we will discuss sorting algorithms in general, and three sorting algorithms
in detail: selection sort, bubble sort and mergesort.
3.1 Objectives
At the conclusion of this section, and with the successful completion of your third
• Define the differences between the selection sort, bubble sort and mergesort
sorting algorithms; and
• Implement a traditional mergesort algorithm.
3.2 Overview
The effort to bring order to the vast amounts of data that computers collect is reflective of
humankind’s eons long effort to organize and catalog information. The two main reasons
to sort data are:
• To prepare organized reports; and
• To pre-process data to reduce the time and/or complexity of a second pass
analysis process.
The main problem with sorts is that they tend to be time consuming. There have been
many clever optimized sorting algorithms developed, but they tend to introduce so much
coding complexity that they are hard to understand and implement. It would be nice if
these optimized algorithms could be packaged as general utilities and used by developers
without having to understand their details, and sometimes they are; but there are two
reasons why this isn’t always practical:
• The most efficient optimizations usually take into account detailed knowledge of
the data being sorted. For example, sorting the results of a chemical analysis
might take into account expectations about the distribution of data based on
previous experience.
• Some knowledge of the format of the data structures being sorted is required by
the implementation. For example, the excellent implementation of quick sort in
the C Standard Library function qsort requires that data be organized in an array,
therefore it cannot be used to sort a linked list.
We will now discuss the details of several common sorting techniques. As a project you
will implement the mergesort algorithm as a mechanism to sort generalized arrays.
3.3 Bubble Sort

The bubble sort algorithm moves sequentially through a list of data structures dividing it
into a sorted part, and an unsorted part. A bubble sort starts at end of the list, compares
adjacent elements, and swaps them if the right-hand element (the element later in the list)
is less than the left-hand element (the element with earlier in the list). With successive
Section 3: Sorting 41 06/19/06

passes through the list, each member percolates or bubbles to its ordered position at the
beginning of the list. The pseudocode looks like this:
numElements = number of structures to be sorted
for ( inx = 0 ; inx < numElements - 1 ; ++inx )
for ( jnx = numElements - 1 ; jnx != inx ; --jnx )
if ( element( jnx ) < element( jnx - 1 ) )
swap( element( jnx ), element( jnx - 1 ) )
3.4 Select Sort

A selection sort looks a lot like a bubble sort. It moves iteratively through a list of data
structures, dividing the list into a sorted and an unsorted part. The pseudocode looks like
this:
numElements = number of structures to be sorted
for ( inx = 0 ; inx < numElements - 1 ; ++inx )
least = inx
for ( jnx = inx + 1 ; jnx < numElements ; ++jnx )
if ( element( least ) > element( jnx ) )
least = jnx
swap( element( least ), element( inx ) )
The big difference between a select sort and a bubble sort is the efficiency it introduces
by reducing the number of swaps that must be performed on each pass through the list.
As you can see, instead of performing a swap each time it determines that an element is
out of order, it merely keeps track of the position of the smallest element it has found so
far; then, at the end of each iteration, it performs exactly one swap, installing the smallest
element in the correct position in the list.
3.5 Mergesort
The main idea behind the mergesort algorithm is to recursively divide a data structure in
half, sort each half independently, and then merge the results. Since the data structure
needs to be split, this algorithm works best with well-organized, contiguous structures
such as arrays. To mergesort an array, divide the array in half, and independently sort the
two halves. When each half has been sorted, merge the results into a temporary buffer,
then copy the contents of the buffer back to the original array. Here is the pseudocode for
sorting an array:
mergesort( array, numElements )
if ( numElements > 1 )
lowHalf = numElements / 2
highHalf = numElements - lowHalf
array2 = array + lowHalf
mergesort( array, lowHalf )
mergesort( array2, highHalf )
inx = jnx = knx = 0

while ( inx < lowHalf && jnx < highHalf )
if ( array[inx] < array2[jnx] )
tempArray[knx++] = array[inx++]
else
tempArray[knx++] = array2[jnx++]

while ( inx < lowHalf )
while ( jnx < highHalf )
tempArray[knx++] = array2[jnx++]
for ( inx = 0 ; inx < numElements ; ++inx )

array[inx] = tempArray[inx]
3.6 A Mergesort Implementation in C

We will now discuss the process of converting the above mergesort pseudocode into an
actual implementation in C. Completing the process will be the subject of your next
project. The mergesort algorithm will be encapsulated in a single function, and our first
job will be to decide on the function’s return type, and the type of the function’s
parameters, then we will have four technical problems to overcome.
3.6.1 The Mergesort Function’s Footprint

We’re not yet ready to say what all the function’s parameters are, but we can say that the
return type will be void, and the first two parameters will be a pointer to the array, and the
number of elements in the array. Like qsort, we want to be able to sort an array of any
type, so we will want our array pointer to be type void*; and in keeping with common
practice we will make the number-of-elements parameter type size_t. Here is what we
have so far:
void mergesort( void *array,
size_t num_elements,
(other parameters)
);
3.6.2 The Pointer Arithmetic Problem

This line of pseudocode illustrates the first we have to overcome:
The problem with the above code is that it requires the compiler to do pointer arithmetic;
and, in order to do pointer arithmetic the compiler must know the size of the element
referenced by the pointer. And since the input to our mergesort function can be an array
of any data type, there is no way for the compiler to intrinsically determine the size of an
element in the array. Specifically, recall from your past classes that you cannot do pointer
arithmetic with a void pointer. To solve this problem we will have to do the pointer
arithmetic ourselves; this requires us to know the size of an element in the array, and the
only way to know that is if the caller tells us, so we will have to add another parameter to
the mergesort function’s parameter list:
size_t element_size,
(other parameters)
);
Now that we know the size of an element in the array, consider this:
• The C data type used to represent a single byte is char; and
• If we have a pointer to the first byte of one element in the array, and we increase
the value of the pointer by the size of an element, we will have a new pointer to
the next element of the array.
Now we can solve pointer arithmetic problems such as array2 = array + lowHalf with
code like this:
typedef unsigned char BYTE_t;
(other parameters)
)
{
if( num_elements > 1 )
{
size_t lowHalf = num_elements / 2;
size_t highHalf = num_elements - lowHalf;
BYTE_t *array1 = array;
BYTE_t *array2 = array1 + lowHalf * element_size;
. . .
3.6.3 The Assignment Problem

This line of pseudocode illustrates the second problem to be overcome:
First, recall that indexing into an array requires pointer arithmetic, which, as we discussed
above, is a problem for the compiler because it doesn’t intrinsically know the size of an
element in one of our arrays. Second, note that performing an assignment like that
suggested by the above pseudocode requires copying the all the bytes from one element
of one array into another array; and, since the compiler doesn’t know how many bytes
that is, it can’t perform the copy for us. The solution is to use the elementSize parameter
to do the following:
• Locate the address of the element in the source array that we want to copy;
• Locate the address of the element in the target array that we want to copy to; and
• Use memcpy to perform the copy.
Here, for example, is one possible way to translate the second while loop in the mergesort
pseudocode into C code:
while ( inx < (int)lowHalf )

{
memcpy( tempArray + knx * element_size,
array1 + inx * element_size,
element_size
);
++inx;
++knx;
}

3.6.4 The Comparison Problem

The next problem we want to talk about concerns the comparison portion of the algorithm
as represented by this line of pseudocode:
You should already recognize that we have yet another pointer arithmetic problem here.
But, in addition, we have a problem of interpretation. Specifically: how can our
implementation interpret the data in the arrays in a way that allows us to determine which
element is less than the other? The answer is: we can’t. The only person who can tell us
how to interpret the data is the programmer who called our mergesort function to begin
with. We solve the problem the same way the ANSI library function qsort does: we ask
the caller to pass the address of a function that knows how to interpret the data in the
array, then we call the function to perform the comparison. The comparison function that
the user writes will take as arguments the addresses of two elements which mergesort will
pass as type void*. The comparison function will cast the pointer to a type appropriate to
the type of the array, perform the comparison, and return one of the following values:
• -1 if the first element is less than the second;
• 1 if the first element is greater than the second; and
• 0 if the two elements are equal.
Here is what the final function declaration looks like after adding the comparison
function parameter:
typedef int CMP_PROC_t( const void*, const void * );
typedef CMP_PROC_t *CMP_PROC_p_t;
typedef unsigned char BYTE_t;
CMP_PROC_p_t cmp_proc
);
And here is one possible way to implement the first while loop from the mergesort
pseudocode:

while ( inx < (int)lowHalf && jnx < (int)highHalf )
{
if ( cmp_proc( array1 + inx * element_size,
array2 + jnx * element_size
) < 0
)
{
memcpy( tempArray + knx * element_size,
array1 + inx * element_size,
element_size
);
inx++;
knx++;
}
else
{
/* to be completed by the student */
}
}
3.6.5 The Temporary Array

The last problem we need to solve is the presence of the temporary array: where did it
come from? We’ll have to allocate it dynamically; and, of course, free it before exiting.
And we’ll use the cover routines from our CDA module:
tempArray = CDA_malloc( num_elements * element_size );
. . .
CDA_free( temp_array );

4. Modules
In this section we will discuss how projects are modularized. We will concentrate on the
way in which individual programs are organized into source code modules, including
public, private and local data structures and methods. You will be introduced to common
industry conventions for naming and controlling the modules and their constituent parts.
4.1 Objectives
• Organize a program into modules;
• Apply naming and access control conventions to modules; and
• Identify the difference between public, private and local data structures and
methods.
4.2 Overview
The successful completion of a large, complex project is almost always accomplished by
breaking the project into smaller, more easily digested pieces. This process, which
Douglas Hoffstadter so aptly described as “chunking,” is formally called modularization,
and some of the pieces that result from the process are often referred to as modules.
The process of modularization begins at a very high level. Figure 4-1 shows how a
theoretical General Administration System is broken into three executable processes for
encapsulating general ledger, accounts payable/receivable and inventory control
functionality. The general ledger process is broken into chunks of functionality
representing general ledger utilities (GL), user interface (UI), database (DB) and error
(ERR) processing. The error chunk is further broken into sets of source files for
performing signal processing, stack dumping and error reporting. The chunking process
continues from there, as the source files declare data structures, subroutines, sub-
subroutines, etc.
For our purposes, we will define a C Source Module as beginning at the next to last level
of the tree, with GL, UI, DB and ERR each representing a separate module. In the next
section we will discuss the components of the C source module.
4.3 C Source Module Components

A C source module (from now on we’ll just say module most of the time) consists of
three classes of data: public data for use outside the module, private data for use inside
the module and local data for using within a source file component of a module. These
three classes of data are discussed below.
4.3.1 Public Data

A module is typically intended as a building block for an application. This means that
there must be some way for other modules to pass data in, and get results out. To
accomplish this, the module publishes a set of public data declarations and methods; that
Section 4: Modules 47 06/19/06
is, declarations and methods that can be used or called from outside the module. These
public declarations and methods are often called an application programmer’s interface,
or API for short.
General
Administration
General Payables/ Inventory

Ledger Receivables Control
GL UI DB ERR
SIG DMP RPT
Figure 4-1 Chunking
The mechanism for publishing public data is the module’s public header file. A module
typically only has one public header file, which contains prototypes for all public
methods, and declarations of data structures, data types and macros required to interact
with those methods. To use an example from the C Standard Library, we might say that
the string handling module is represented by the public header file string.h.
It is very important to do a thorough job of defining a public API before beginning
implementation. Other programmers, sometimes hundreds of them, working outside your
module are going to be using your API, and if you decide to change it in mid-
development you will have an impact on each of them.
4.3.2 Private Data

Not all modules have private data. Those that do are constructed from multiple source
files, or are prepared for the eventuality of adding additional source files. When this is the
case, there is often a need to share data declarations among the module’s source files,
declarations that you don’t wish to share with the rest of the project. Likewise, there may
be a need for one source file to make a call to a function in another source file, which
should not be available to the project at large. Declarations and methods shared within,
but not without a module’s source files are called private.
Private declarations and methods are desirable; they represent a form of data hiding.
When data is hidden from users of your API you can change it without affecting anyone
but the other programmers working on your module. So, for example, if you discover a
clever way to optimize a private data structure after beginning development you can
deploy it with relatively little impact on the overall project.
Private data declarations and methods are published via the module’s private header file.
On a properly managed project, it is understood that no source file in a module ever
includes another module’s private header file (except for friends of the module, but that’s
a topic we won’t be covering).

4.3.3 Local Data

Within a single source file there are almost always subroutines that exist for use only by
other functions in the file. Likewise, a source file will often declare data types and global
state variables solely for local use. These data and subroutines are referred to as local or
static. To force functions and global variables to be local, you must explicitly declare
them to be static; other declarations, including typedefs and macros, are local by default.
Local declarations should never be placed in a header file; local declarations, including
prototypes of local functions, should always be placed at the top of the source file that
uses them.
4.4 Review: Scope

This is a good time to stop and review the concept of scope for an identifier. The scope of
an identifier represents the extent of its visibility. A function or global variable with
external scope can be seen throughout the entire program; functions and global variables
have external scope by default. Functions and global variables declared static have their
scope restricted to a single source file. Macros, typedef names and other declarations,
such as enum and struct, are always restricted in scope to a single source file. Variables
and prototypes declared inside a compound statement are limited in scope to that
compound statement.
4.5 A Bit about Header Files

Before we go one we should also have a brief review of header files.
Header files are often referred to as include files, but this is misleading. There are files
that may be included that are not header files; other .c files, for example.
By convention, the name of a header file always ends in “.h,” and a header file never
includes defining declarations (that is, declarations that result in memory allocations).
These declarations are valid in a header file because they require no memory allocation:
typedef int CDA_BOOL_t;
int abs( int );
extern int errNum;
But these declarations do require memory allocation, and are not allowed in a header file:
int errNum = 0;
int abs( int val )

{
return val < 0 ? -val : val;
}
Finally, every header file should contain an include sandwich.
4.6 Module Conventions

Clearly established conventions for writing modules can help to better organize the data
on a project, and to prevent conflicts and bugs. These are some of the problems that a
simple set of conventions can help to avoid:

• Which module does printFractalData belong to?

• What’s the name of the header file that declares ADDRESS_DATA_t?
• I need to look at the source file for dumpStack; what file is it in?
• Is dumpCurrRegisters a private or public method?
• I named one of my public functions processErr and so did George over in the data
base group; which one of us has to change?
• I declared a prototype for a function in the UI module with two arguments, and last
week they changed it to need three arguments, and I was up all might figuring that
out!
The following conventions are similar to those adopted by many large, successful
projects. You are expected to follow them in the course of preparing your project source
code.
1. Every module is assigned a short, unique name. For example, ENQ for the doubly
linked list module, and SRT for the general sorting module.
2. Every module will name its public header file using the module name, followed by .h.
For example, enq.h and srt.h will be the public header files for the ENQ and SRT
modules, respectively.
3. Every module will name its private header file using the module name followed by
p.h. For example, enqp.h and srtp.h.
4. Every module will name its principal source file using the module name. If the
module requires additional source files, they will be named using the module name
followed by an underscore as a prefix. For example, the ERR module may contain
source files err.c and err_dump.c.
5. The name of a public identifier declared by a module will always begin with the
module name in upper case followed by a single underscore. For example,
ENQ_ITEM_t and ENQ_add_head.
6. The name of a private identifier declared by a module will always begin with the
module name in upper case followed by two underscores. For example,
ERR__DATA_p_t and ERR__default_abt_handler.
7. The name of a local identifier will always begin with a lower-case character, and will
NOT identify the module to which it belongs.
8. A source file will never explicitly declare public or private data; the only valid means
to obtain such a declaration is to include the public or private header file that
publishes it.
9. Local data declarations will never appear in a header file; they always appear in the
source file that requires them.
10. A source file within a module will always include the module’s public and private
header files. Specifically in this class, the module’s private header file will include
the public header file, and a source file within the module will include just the
module’s private header file. This requires all modules to have a private header file
even if one isn’t strictly needed.

5. Abstract Data Types

In this section we will define the term abstract data type or ADT, and show you how it
relates to our concept of a module. We will close the section with a definition of the list
ADT.
5.1 Objectives
• Define the term abstract data type;
• Describe the list ADT; and
• Create a module that implements an abstract data type.
5.2 Overview
An abstract data type is a set of values and associated operations that may be performed
on those values. The classic example of an abstract data type is the set of integers, and
associated operations that may be performed on integers, such as addition, subtraction,
multiplication, etc. Another example of the abstract data type is the array, in conjunction
with the array operator, []. In C, operations may be implemented using built-in
operators or methods. Table 1 shows, in part, how the operations for the abstract data
type int are implemented.
Operation Built-in Method

Operator
Addition +
Subtraction -
Multiplication *
Division /
Modulus %
Increment ++
Decrement --
Absolute abs()
Value
nth Power pow()
Section 5: Abstract Data Types 51 06/19/06

nth Root pow()
Table 1 The Integer ADT
An operation performed on one or more valid members of the set of values must yield
another member of the set; in the set int, for example, 3 + 5 yields 8. When an operation
is attempted on a member that is not a member of the set; or when the result of an
operation falls outside of the set; an exception is said to occur. For example, in the
context of the int ADT, the attempt to compute the square root of -1 leads to an
exception. When an exception occurs, the abstract data type implementation should
handle the exception. Exception handling is discussed in Section 5.3, below.
To implement an abstract data type in C that is not provided by the compiler or standard
library, the programmer:
1. Carefully defines the set of values that are to be associated with the ADT;
2. Fully defines the operations that may be performed within the context of the
ADT;
3. Declares a set of data structures to represent the ADT’s legal values; and
4. Implements individual methods to perform the defined operations.
As we have defined it in this class, a module is an excellent vehicle for representing an
ADT; two examples are our implementation of stacks and hash tables.
5.3 Exception Handling

An exception occurs when an abstract data type’s method is handed invalid input, or
when the operation performed by the method yields invalid data. When an exception
occurs, the method must handle the exception. The three most common ways of handling
an exception are:
• Ignore the exception.
This is what C (usually) does when an attempt is made to access an element
outside an array, or when two large integers are multiplied together yielding an
integer overflow.
• Return an error value.
This is what the C standard library function fgets does when you try to read past
the end of a file.
• Throw the exception.
This is what C does when you attempt to divide by zero; it’s also what our cover
routines for the malloc family of functions do when an attempt to allocate
memory fails.
Ignoring an exception can be dangerous. However it’s not always a bad strategy,
particularly when a method is buried deep in an implementation, and relies on code that
you wrote to pass it only valid data. Performing validation tests in this case can be
inefficient, and can increase the complexity of your code, leading to the possibility that
the validation will actually introduce a flaw. A good compromise is often to employ
assertions, which can be used to validate input during testing, and deactivated in

production. This is how the ENQ module’s add-head method deals with an attempt to add
an item that is already enqueued.
Returning an error value is often the best way to handle an exception, but has its
drawbacks. Many times an error return is so rare, and so devastating that it is hard to test;
the classic example of this is an error return from malloc. Often an inexperienced (or
careless) programmer has neglected to check for an error return. In this case, the program
failure, when it finally occurs, may be many statements after the actual flaw, and can be
very difficult to debug. In cases such as these, the method that detects the exception may
be better off throwing it.
Throwing the exception involves raising a signal. One way to do this is to call the
standard library routine abort, which, in most implementations, raises SIGABRT. Any
signal can be raised by calling the standard library function raise. In many ways throwing
an exception in a method is the best way to handle rare errors; if a user of the method
really needs to know that an error occurred, he or she can define a signal handler to trap
the signal. Figure 5-1 shows an example of a routine that locks a system resource, and
then calls a method that may throw an exception; if the exception is thrown, the user traps
it, unlocks the system resource, and then re-raises the signal. This is also called catching
the exception.
#include <signal.h>
typedef void SIG_PROC_t( int sig );
typedef SIG_PROC_t *SIG_PROC_p_t;
static SIG_PROC_t trap_sigint;
static SIG_PROC_p_t old_sigint = SIG_DFL;
. . .
if ( (old_sigint = signal( SIGINT, trap_sigint )) == SIG_ERR )
abort();
SYS_lock_res( SYS_RESOURCE_OPEN );
queue_id = SYS_open_queue( “PROCESS_STREAM” );
SYS_unlock_res( SYS_RESOURCE_OPEN );
if ( signal( SIGINT, old_sigint ) == SIG_ERR )
abort();
. . .
static void trap_sigint( int sig )

{
SYS_unlock_res( SYS_RESOURCE_OPEN );
if ( signal( sig, old_sigint ) == SIG_ERR )
abort();
raise( sig );
}
Figure 5-1 Catching an Exception
A more sophisticated routine might attempt to recover from the thrown exception by
saving the system state using setjmp, then executing a longjmp to restore the system
state. This mechanism should only be used when recovery from an exception is
absolutely crucial, and the programmer knows exactly how to use it.

5.4 Classes of ADTs

Your textbook divides abstract data types into three classes:
1. Atomic Data Types
These are data types that are usually considered indivisible, for example the type
int. In C, most atomic data types are represented by the built-in data types.
2. Fixed-Aggregate Data Types
These are data types composed of components of a fixed size. The classic
example of a fixed-aggregate data type in C is the complex number data type (see
below).
3. Variable-Aggregate Data Types
These data types are composed of components of varying size. The most common
example in C is the array data type. Our doubly linked lists and hash tables are
also variable-aggregate data types.
5.4.1 The Complex Number ADT

As an example of a fixed-aggregate data type, consider the complex numbers. In
mathematics, a complex number can be expressed as the sum of its real and imaginary
parts. A typical representation of a complex number may be defined as:
a + ib
where i is the square root of -1. The addition of two complex numbers, (a1 + ib1) +
(a2 + ib2) is defined as:
(a1 + a2) + i(b1 + b2)
To implement this functionality in C, we might declare the data type and method shown
in Figure 5-2.
typedef struct cpx_num_s
{
double real;
double imaginary;
} CPX_NUM_t, *CPX_NUM_p_t;
CPX_NUM_p_t CPX_compute_sum( CPX_NUM_p_t cpx1, CPX_NUM_p_t cpx2 )

{
CPX_NUM_p_t cpx_sum = CDA_NEW( CPX_NUM_t );
cpx_sum->real = cpx1->real + cpx2->real;

cpx_sum->imaginary = cpx1->imaginary + cpx2->imaginary;
return cpx_sum;
}
Figure 5-2 Complex Number ADT Addition
Taken in conjunction with the declaration of CPX_NUM_t and CPX_NUM_p_t, above,

the complex number abstract data type could be implemented using the set of methods
shown in Figure 5-3.

CPX_NUM_p_t
CPX_compute_diff( CPX_NUM_p_t cpx1, CPX_NUM_p_t cpx2 );
/* returns (cpx1 - cpx 2) */
CPX_NUM_p_t
CPX_compute_neg( CPX_NUM_p_t cpx )
/* returns -cpx */
CPX_NUM_p_t
CPX_compute_prod( CPX_NUM_p_t cpx1, CPX_NUM_p_t cpx2 )
/* returns (cpx1 * cpx 2) */
CPX_NUM_p_t
CPX_compute_sum( CPX_NUM_p_t cpx1, CPX_NUM_p_t cpx2 );
/* returns (cpx1 + cpx2) */
CPX_NUM_p_t
CPX_compute_quot( CPX_NUM_p_t cpx1, CPX_NUM_p_t cpx2 );
/* returns (cpx1 / cpx 2) */
Figure 5-3 Complex Number ADT Methods
5.4.2 The List ADT

As an example of an ADT consider the definition of the list ADT provided by your
textbook. To paraphrase:
• A list is a sequence of some type T.
Kruse defines the following operations that may be performed on a list:
• Create the list;
• Determine whether the list is empty;
• Determine whether the list is full;
• Find the size of the list;
• Add a new entry to the end of the list;
• Traverse the list (performing some operation at each node); and
• Clear the list
To the list of operations, I am going to add the following:
• Destroy the list.
Note that the definition of the ADT says nothing about how the list is to be implemented.
I can imagine implementing the list as an array, which would be relatively efficient, or as
a linked list utilizing our ENQ module, which would be less efficient but more flexible.
Let’s go ahead and design the public interface for a list ADT, then give a couple of
examples of alternative implementation strategies.
First, in accordance with our module naming standards, we need a name for the module; I
have chosen LIST. That means our module will consist (at a minimum) of list.c, list.h and
listp.h. When the user creates a list, she will tell us the size of an entry in the list (note
how different this is from our ENQ module, where entries can be different sizes; in a list,

the entries are all the same type, hence the same size). She will also give us a hint about
the maximum size of the list. Why is this called a “hint”? Because the implementation
may choose to force the maximum size to a higher, more convenient number, or to ignore
it altogether. In this specific example, if we implement the list as an array we will need
the maximum size, but if we use the ENQ module we will ignore it. Once the list is
successfully created and initialized we will return to the user an ID that she can
subsequently use to access the list. We want this ID to be an opaque type, that is,
something that identifies the list, but does not allow the user access to the internals. The
ID will most likely be a pointer to a control structure that is declared in our private header
file. Often in C we use the type void* for such opaque data types; doing so, however,
results in weak typing; that’s because your compiler will allow you to silently assign to a
void pointer almost any other pointer type. In order to achieve strong typing we are going
to assume that the control structure is type struct list__control_s, and then we can declare
the public list ID as follows:
typedef struct list__control_s *LIST_ID_t;
Note that, from the perspective of the public API this is what, in C, we refer to as an
incomplete declaration. From examining the public header file we can conclude that a list
ID is a pointer to a particular kind of struct, but we have no idea what members are
contained in the struct. Since we don’t know what the members are we can’t access them.
To be truly opaque the user shouldn’t even assume that a list ID is a pointer type; and, in
fact, we should be free to change our minds next week and make the ID an int. So to be
really flexible, we should provide the user with a value that she can use to initialize a list
ID that isn’t yet assigned to a list. We will call this value LIST_NULL_ID, and declare it
like this:
#define LIST_NULL_ID (NULL)
With these two public declarations, a user can declare and initialize a list ID variable this
way:
LIST_ID_t list = LIST_NULL_ID;
What if we do change our minds next week and decide to make a list ID an int? Then we
will just change LIST_NULL_ID to something appropriate, such as –1. The user will
have to recompile her code, but will not have to change any of it.
Traversal and destruction of a list pose special problems because they require
manipulation or disposal of data owned by the user. To state the problem a little
differently:
• When we traverse the list we will touch each entry and do something with the
data; but do what?
• When we destroy the list, the data in each entry may need to be disposed of; but
how?
Only the user can answer these questions. So here’s what we are going to do: when the
user calls the traversal method she will pass the address of a function, called a traversal
proc, which knows how to manipulate the data at each entry. And when she calls the
destroy method she will call pass the address of a function, called a destroy proc that
knows how to dispose of the data in an entry. Such functions are called callbacks because
the implementation uses them to call back to the user’s code in order accomplish some
part of its operation. We will declare the types of these functions as follows:
typedef void LIST_DESTROY_PROC_t( void * );
typedef LIST_DESTROY_PROC_t *LIST_DESTROY_PROC_p_t;
typedef void LIST_TRAVERSAL_PROC_t( void * );

typedef LIST_TRAVERSAL_PROC_t *LIST_TRAVERSAL_PROC_p_t;
Keeping the above declarations in mind, we can define the public interface as follows.
5.4.2.1 LIST_create_list
This method will create an empty list and return to the user a value that identifies the list.
Synopsis:
LIST_ID_t LIST_create_list( size_t max_list_size,
size_t entry_size,
const char *name
)
Where:
max_list_size == a hint about the maximum size of the list
entry_size == the size of an entry in the list
name -> the name of the list
Returns:
The list ID
Exceptions:
Throws SIGABRT if the list can’t be created.
Notes:
by the list by calling LIST_destroy_list.
4. Following creation, the list is guaranteed to hold at least
max_list_size entries; it may be able to hold more. See
also LIST_add_entry.
5.4.2.2 LIST_add_entry
This method will add an entry to the end of the list.
Synopsis:
LIST_ID_t LIST_add_entry( LIST_ID_t list, const void *data )
Where:
list == ID of a previously created list
data -> data to be appended to list tail
Returns:
data
Exceptions:
Throws SIGABRT if the new entry can’t be created.
Notes:
1. The data argument must point to a block of memory equal in
size to the entry size as specified in LIST_create_list. A
new entry is created for the list and the data is COPIED
INTO IT.

5.4.2.3 LIST_traverse_list
This method will traverse the list in order, calling the user’s traversal proc at each node.
Synopsis:
LIST_ID_t LIST_traverse_list(
LIST_ID_t list,
LIST_TRAVERSAL_PROC_p_t traversal_proc
)
Where:
traversal_proc -> function to call for each node
Returns:
list
Exceptions:
None
Notes:
1. For consistency with other modules and methods, the
traversal proc may be NULL.
5.4.2.4 LIST_is_list_empty
This method returns a Boolean value indicating whether a list is empty.
Synopsis:
CDA_BOOL_t LIST_is_list_empty( LIST_ID_t list )
Where:
Returns:
CDA_TRUE if list is empty, CDA_FALSE otherwise
Exceptions:
None
Notes:
None
5.4.2.5 LIST_is_list_full
This method returns a Boolean value indicating whether a list is full.
Synopsis:
CDA_BOOL_t LIST_is_list_full( LIST_ID_t list )
Where:
Returns:
CDA_TRUE if list is full, CDA_FALSE otherwise
Exceptions:
None
Notes:
None
5.4.2.6 LIST_get_list_size
This method returns the number of elements in the list.

Synopsis:
size_t LIST_is_list_full( LIST_ID_t list )
Where:
Returns:
The size of the list
Exceptions:
None
Notes:
None
5.4.2.7 LIST_clear_list
This method will return a list to its initial, empty state, destroying each node in the
process. If the user specifies a destroy proc, it will be called for each node in the list prior
to destroying the node.
Synopsis:
LIST_ID_t LIST_clear_list( LIST_ID_t list,
LIST_DESTROY_PROC_p_t destroy_proc
)
Where:
destroy_proc -> function to call for each node
Returns:
list
Exceptions:
None
Notes:
1. If not needed, the destroy proc may be NULL
5.4.2.8 LIST_destroy_list
This method will first clear the list (see LIST_clear_list) and then destroy the list itself. If
the user specifies a destroy proc, it will be called for each node in the list prior to
destroying the node.
Synopsis:
LIST_ID_t LIST_destroy_list(
LIST_ID_t list,
)
Where:
destroy_proc -> function to call for each node
Returns:
LIST_NULL_ID
Exceptions:
None

Notes:
If not needed, the destroy proc may be NULL
5.4.3 Implementation Choices

When preparing to implement an ADT you are typically faced with a number of choices.
A comparison of choices will typically reveal some common tradeoffs, such as:
• Efficiency vs. flexibility. Generally (though not always) increased flexibility
comes at the cost of decreased efficiency; and
• Simplicity vs. complexity. Generally (though not always) if you strive to increase
the efficiency and/or flexibility of your code, the code will become more
complex. As code becomes more complex it is more likely to contain flaws and
be more difficult to test and maintain. As a rule you want to keep your code as
simple as possible while providing the flexibility demanded by the requirements
of your current project, and anticipating future requirements. Efficiency is
infrequently a concern; you should only sacrifice simplicity in favor of efficiency
in a few, clearly defined and well-thought-out circumstances.
Hiding the details of your implementation; that is, keeping as much of the module private
as you absolutely can; is the key to successful implementations. Because you can change
the private details of a module without requiring adaptation on the part of the user, you
can easily evolve an implementation to meet changing flexibility and efficiency issues;
you can even provide alternative implementations that provide different levels of tradeoff
between efficiency and flexibility.
Let’s examine two alternative implementations for the list ADT discussed above. One
will implement the list as an array (being relatively efficient) and the other will
implement the list using our ENQ module (being relatively flexible). Keep in mind that
although the two implementations are wildly different, they meet identical functional
requirements.
5.4.3.1 Private Declarations for the Array Implementation

To implement a list as an array we are going to allocate a block of memory suitable for
storage of an array of some maximum number of elements of a fixed size. These two
values correspond to the max_list_size and entry_size arguments passed by the user to the
list create method. When an entry is added to the list we will add it to the next
sequentially available entry of the array. This arrangement is illustrated in Figure 5-4.
In order to facilitate this we are going to need a control structure that can hold a pointer to
the array, the array size, the element size and the next element index. We should also
store a copy of the list name. Thinking ahead a little bit, I know that we are going to want
to do pointer arithmetic to move between entries in the array, so let’s also declare the
type of an entry pointer (we’ll make this equivalent to a char* for reasons that will soon
become clear). Then, since the address of an array is the same as a pointer to the first
element of the array, the member of the control structure that points to the array is type
“pointer to entry.” The complete private header file for our list module when
implemented as an array is shown in Figure 5-5.

Figure 5-4 Implementing a List as an Array
#ifndef LISTP_H
#define LISTP_H
#include <list.h>
#include <stddef.h>
typedef char *LIST__ENTRY_p_t;
typedef struct list__control_s

{
LIST__ENTRY_p_t array;
size_t max_size;
size_t entry_size;
int next;
char *name;
} LIST__CONTROL_t, *LIST__CONTROL_p_t;
#endif
Figure 5-5 Private Declarations for Array List Implementation
Implementing the create method for a list implemented as an array requires us to do the
following:
1. Allocate memory for the control structure;
2. Allocate memory for the array, and store a pointer to it in the control structure;
3. Initialize the remaining members of the control structure; and
4. Return a pointer to the control structure to the caller. Remember that, since the
caller doesn’t include listp.h, there is no way for the caller to use the pointer to
access the control structure.
The complete create method for implementing a list as an array is shown in Figure 5-6.
size_t entry_size,
const char *name
)
{
LIST__CONTROL_p_t list = CDA_NEW( LIST__CONTROL_t );

list->array = CDA_calloc( max_list_size, entry_size );
list->max_size = max_list_size;
list->entry_size = entry_size;
list->next = 0;
return list;
}
Figure 5-6 Create Method for Array List Implementation
5.4.3.2 Private Declarations for the ENQ Implementation

To implement a list using the ENQ module our control structure will simply contain the
address of an ENQ module anchor. To add an entry to the list we will simply create a
new enqueuable item and add it at the tail of the ENQ list. The enqueuable item will have
as its application data a single member that points to a block of memory to hold the user’s
data. This strategy is illustrated in Figure 5-7.
Figure 5-7 Implementing a List using the ENQ Module
Our private declarations will consist of a control structure that contains the address of a
list anchor, plus the size of a user entry. We will also keep a copy of the list name, and
the maximum list size specified by the user (we won’t be using this value, but it doesn’t
hurt to hang onto it, and we may find a use for it in the future). Also, we will need to
declare the type of an item to store in our ENQ list; remember that this will be an
enqueuable item as defined by the ENQ module, and so the type of its first member must
be ENQ_ITEM_t. The complete private header file for implementing a list via the ENQ
module is shown in Figure 5-8.

#ifndef LISTP_H
#define LISTP_H
#include <list.h>
#include <enq.h>
#include <cda.h>
#include <stddef.h>
typedef struct list__control_s

{
ENQ_ANCHOR_p_t anchor;
size_t max_size;
size_t entry_size;
char *name;
} LIST__CONTROL_t, *LIST__CONTROL_p_t;
typedef struct list__entry_s

{
ENQ_ITEM_t item;
void *data;
} LIST__ENTRY_t, *LIST__ENTRY_p_t;
#endif
Figure 5-8 Private Declarations for ENQ List Implementation
Implementing the create method for a list implemented via the ENQ module requires us
to do the following:
1. Allocate memory for the control structure;
2. Create an ENQ list, and store the address of its anchor in the control structure; and
3. Return a pointer to the control structure to the caller. As before, since the caller
doesn’t include listp.h, there is no way for her to use the pointer to access the
control structure.
The complete create method for this implementation option is shown in Figure 5-9.
size_t entry_size,
const char *name
)
{
LIST__CONTROL_p_t list = CDA_NEW( LIST__CONTROL_t );
list->anchor = ENQ_create_list( name );

list->max_size = max_list_size;
list->entry_size = entry_size;
return list;
}
Figure 5-9 Create Method for ENQ List Implementation

5.4.3.3 Efficiency vs. Flexibility: The Add and Get-Size Methods

To demonstrate the issue of efficiency vs. flexibility, lets take a look at the add and get-
size methods for the two different implementations. They are shown in Figure 5-10 and
Figure 5-11.
/* Add method for the array-based implementation */
const void *LIST_add_entry( LIST_ID_t list, const void *data )
{
LIST__ENTRY_p_t nextEntry =
list->array + list->next * list->entry_size;
if ( list->next >= (int)list->max_size )
abort();
memmove( nextEntry, data, list->entry_size );

++list->next;
return data;
}
/* Add method for the ENQ module-based implementation */

const void *LIST_add_entry( LIST_ID_t list, const void *data )
{
LIST__ENTRY_p_t entry =
(LIST__ENTRY_p_t)ENQ_create_item( NULL,
sizeof(LIST__ENTRY_t)
);
entry->data = CDA_malloc( list->entry_size );

memmove( entry->data, data, list->entry_size );
ENQ_add_tail( list->anchor, (ENQ_ITEM_p_t)entry );
return data;
}
Figure 5-10 Alternative List Add Method Implementations
As you can see, the array-based add method is fairly efficient. Using simple pointer
arithmetic we locate the address of the next unused element in the array and copy the
user’s data into it; however it is inflexible because the size of the list cannot exceed the
maximum length of the array, which the user must somehow calculate prior to creating
the list. The ENQ module-based implementation is more flexible because it can
dynamically grow to virtually any length, relieving the user of the need to determine a
maximum length. This flexibility, however, comes at the cost of two extra dynamic
memory allocations, which drastically reduces its efficiency.
The difference in efficiency can be seen even more clearly by examining the get-size
method. The array-based implementation is extremely efficient, simply returning the next
value out of the control structure. The ENQ module-based implementation, however,
must traverse the list counting the elements as it goes.
Note that the array-based implementation can be made more flexible by using realloc to
eliminate its dependence on a maximum size parameter. And the ENQ-module based
implementation can be made more efficient in a couple of ways, for example by
redesigning the LIST__ENTRY_t type to allow the user’s data to reside directly in a

LIST_ENTRY_t variable, thereby eliminating a dynamic memory allocation. However

each of these design changes comes at the cost of greatly increased complexity in the
code.
/* Get-size method for the array-based implementation */
size_t LIST_get_list_size( LIST_ID_t list )
{
size_t rcode = (size_t)list->next;
return rcode;
}
/* Get-size method for the ENQ module-based implementation */

size_t LIST_get_list_size( LIST_ID_t list )
{
size_t rcode = 0;
ENQ_ITEM_p_t item = ENQ_GET_HEAD( list->anchor );
while ( item != list->anchor )

{
++rcode;
item = ENQ_GET_NEXT( item );
}
return rcode;
}
Figure 5-11 Alternative List Get-Size Methods
5.4.3.4 The Traverse and Clear Methods

Earlier when we talked about the traverse and empty methods, we introduced the concept
of the callback function. So for the sake of completeness let’s take a quick look at these
two methods.
The traverse method must sequentially visit each entry in the list and give the user the
opportunity to do something with the data there. For the ENQ module-based
implementation all we do is find the first item, then follow the items’ flinks till we reach
the end. In the array-based implementation, we set a LIST__ENTRY_p_t variable to the
start of the array, and use pointer arithmetic to sequentially find subsequent entries in the
array (this is why made LIST__ENTRY_p_t equivalent to char*; the logic wouldn’t work
with a void*). In either case, each time we walk to a new entry in the list, we take a
pointer to the user’s data and pass it to the user’s traversal proc. The code for both
implementations is shown in Figure 5-12.
To clear a list in the array-based implementation we merely have to set the next indicator
back to 0. To clear an ENQ module-based list we have to destroy each item in the list. In
each case, however, we must first walk to each entry in the list, find the address of the
user’s data, and pass the address to the user’s destroy proc for final disposition. The code
for doing this is shown in Figure 5-13.

/* Traverse method for the array-based implementation */
LIST_ID_t list,
)
{
LIST__ENTRY_p_t entry = list->array;
int inx = 0;
for ( inx = 0 ; inx < list->next ; ++inx )

{
if ( traversal_proc != NULL )
traversal_proc( entry );
entry += list->entry_size;
}
return list;
}
/* Traverse method for the ENQ module implementation */

LIST_ID_t list,
)
{
ENQ_ITEM_p_t item = ENQ_GET_HEAD( list->anchor );
while ( item != list->anchor )

{
LIST__ENTRY_p_t entry = (LIST__ENTRY_p_t)item;
if ( traversal_proc != NULL )
traversal_proc( entry->data );
}
return list;
}
Figure 5-12 Alternative List Traverse Methods

/* Clear method for the array implementation */
)
{
int inx = 0;
for ( inx = 0 ; inx < list->next ; ++inx )

if ( destroy_proc != NULL )
destroy_proc( list->array + inx * list->entry_size );
list->next = 0;
return list;
}
/* Clear method for the ENQ module implementation */

)
{
while ( !ENQ_is_list_empty( list->anchor ) )
{
LIST__ENTRY_p_t entry =
(LIST__ENTRY_p_t)ENQ_deq_head( list->anchor );
destroy_proc( entry->data );
CDA_free( entry->data );
ENQ_destroy_item( (ENQ_ITEM_p_t)entry );
}
return list;
}
Figure 5-13 Alternative List Clear Methods

6. Stacks
One of the most basic data structures in data processing is the stack. Surprisingly simple
in its implementation, it is a highly versatile mechanism, often useful in implementing
recursive and multithreaded algorithms. In this section we will examine a complete stack
implementation. The material that we cover will be used in conjunction with the sorting
algorithms you learned earlier to complete Project 3.
6.1 Objectives
• Compare a stack to a last-in, first-out (lifo) queue;
• Perform a complete, modularized stack implementation; and
• Use a stack to implement recursive algorithms.
6.2 Overview
A stack is often referred to as a last-in, first-out (LIFO) queue; that’s because a stack
grows by adding an item to its tail, and shrinks by removing an item from its tail, so the
last item added to (or pushed onto) the stack is always the first to be removed (or
popped); see Figure 6-1. Although in some implementations you may access data in the
middle of a stack, it is illegal to remove any item but the end-most.
Figure 6-1: Stack Operations: Push and Pop
Section 6: Stacks 69 06/19/06

Stacks are often implemented using a list, with push corresponding to add-tail and pop to
remove-tail. However the traditional implementation, and the most common view of a
stack, is as an array. When a stack is treated as an array there are two alternative
implementations: top-down and bottom-up. As seen in Figure 6-2, the beginning of a top-
down stack is given as the first address following the last element of the array; the
beginning of a bottom-up stack is given as the address of the first element of the array.
Figure 6-2: Two Stack Implementation Strategies
When you push an item onto the stack, the item that you add occupies a location on the
stack. A stack pointer is used to indicate the position in the array that the pushed item
should occupy. In a bottom-up implementation the stack pointer always indicates the first
unoccupied location; to execute a push operation, first store the new item at the location
indicated by the stack pointer, then increment the stack pointer. In a top-down
implementation, the stack pointer always indicates the last occupied location; to execute
a push operation, first decrement the stack pointer, then store the new item at the
indicated location. The push operation is illustrated in Figure 6-3.
Note: It should be clear that bottom-up and top-down stacks are essentially equivalent.
Application programmers tend to prefer bottom-up stacks because they’re more intuitive;
system programmers often prefer top-down stack for historical, and for obscure (and, for
us, irrelevant) technical reasons. For the sake of avoiding confusion the remainder of this
discussion will focus on bottom-up stacks.
A stack is empty when no position on the stack is occupied. When a bottom-up stack is
empty, the stack pointer will indicate the first element of the array; when full, the stack
pointer will indicate the first address after the last element of the array. This is shown in
Figure 6-4. An attempt to push an item onto a full stack results in a stack overflow
condition.
As shown in Figure 6-5, the last item pushed onto the stack can be removed by popping
it off the stack. To pop an item off a bottom-up stack, first decrement the stack pointer,
then remove the item at the indicated location. Note that once an item has been popped

off the stack, the location it formerly occupied is now unoccupied, and you should not
expect to find a predictable value at that location. An application should never try to pop
an item off an empty stack; attempting to do so constitutes an egregious malfunction on
the part of the application.
Figure 6-3: Push Operation
A note about popping, and the values of unoccupied locations: If you experiment with
stacks using a simple test driver you might conclude that the value popped off the stack
continues to occupy the stack at its previous location. However applications that share a
stack among asynchronously executed subroutines will find the values of such locations
unpredictable. Asynchronous or interrupt-driven environments are where the magic of
stacks is most useful; unfortunately they are beyond the scope of this course.
Figure 6-4: Two Stack States
As a first example of using a stack, following is an example of a subroutine to reverse the

characters in a string.

static char *reverse_chars( char *string )
{
char *temp = string;
char stack[50];
char *sptr = stack;
while ( *temp != '\000' )

*sptr++ = *temp++; /* push */
temp = string;
while ( sptr != stack )
*temp++ = *--sptr; /* pop */
return string;
}
Figure 6-5: Pop Operation
6.3 Stacks And Recursion

A single stack can be shared among many instances of a recursive function. A
recursively-invoked function merely begins using the stack at the point that the previous
invocation left it. Suppose we were writing an interpreter that could parse and evaluate
function-like strings like these:
add( a, b ) Evaluate a + b
mul( a, b ) Evaluate a * b
div( a, b ) Evaluate a / b
sum( a, b, c, . . . N ) Compute the sum of the arguments
and( a, b, c, . . . N ) Compute the logical AND of the arguments

from right to left
etc. etc.

Let’s also suppose that our interpreter will also be able to process strings that consist of a
embedded instances of the above, such as this one:
and( 0xff, sum( mul(10, 50), div(24, 3), add( mul(2, 4), 2 ) ) )
Then we could implement our interpreter as a recursive function that called itself each
time it had to evaluate a subexpression. Such a function might look like the function
StackCalc as shown in Figure 6-6.
long StackCalc( const char *expr, const char **end )
{
long rval = 0;
long operator = 0;
long operand = 0;
int count = 0;
CDA_BOOL_t working = CDA_TRUE;
const char *temp = skipWhite( expr );
operator = getOperator( &temp );

while ( working )
{
if ( getOperand( &operand, &temp ) )
{
++count;
push( operand );
}
else
working = CDA_FALSE;
}
switch ( operator )
{
case ADD:
rval = xadd( count );
break;
. . .
}
if ( end != NULL )
*end = skipWhite( temp );
return rval;
}
Figure 6-6 StackCalc Function Entry Point
StackCalc first obtains an operation identifier by parsing

the operator token, such as add, mul, etc. To do this it Before calling getOperator:
add( 4, 12 ) . . .
calls the function getOperator, which takes a input a
pointer to the string being parsed; it’s return value is the temp
operation identifier, and it updates the string to point to After calling getOperator:
the first token after the operator token. Next iteratively add( 4, 12 ) . . .
calls getOperand to obtain each operand. The return temp
value of getOperand is TRUE if an operand was parsed,
and FALSE if the end of the operand list was detected. If getOperand returns TRUE it
also return the operand via the parameter operand. It always updates the string pointer to

point to the next unparsed token in the string. Each time getOperand returns the value of
an operand StackCalc pushes it onto a stack, keeping a count of the number of operands
that have been pushed. When getOperand indicates that the last argument has been parsed
StackCalc calls a function to pop each operand off the stack and perform the evaluation.
The function for performing the sum evaluation is shown in Figure 6-7.
static long xsum( int count )
{
long rcode = 0;
int inx = 0;
for ( inx = 0 ; inx < count ; ++inx )

rcode += pop();
return rcode;
}
Figure 6-7 StackCalc Function to Evaluate sum
The last thing StackCalc does before returning is to set the end parameter to point to the
portion of the string following whatever it has just evaluated. For example, if the string it
evaluated was “add( 2, 4 )” it will leave end point to the end of the string; but if it had just
evaluated the add portion of “sum( add( 1, 2 ), mul( 2, 4 ) )” it will leave end pointing to
mul.
The recursive logic in StackCalc is contained in the function getOperand, which
evaluates the input string as follows:
• Is the next token a right parenthesis? If so the end of the argument list has been
reached.
• Does the next token start with a digit? Is so parse the argument by calling strtol.
• Does the next token start with an alphabetic character? If so parse the
subexpression by recursively calling StackCalc.
Let’s consider what’s going on with the stack while StackCalc is evaluating this string:
and( 0xff, sum( mul(10, 50), div(24, 3), add( mul(2, 4), 2 ) ) )
We will refer to various calls to StackCalc as instances of the function. When the user
first calls StackCalc we enter instance 0. Whenever instance 0 makes a recursive call we
enter instance 1, etc.
1. Instance 0 obtains the identifier for and, and pushes 0xff onto the stack, then calls
instance 1 to evaluate “sum( . . . ).”
2. Instance 1 obtains the identifier for sum and immediately calls instance 2 to
evaluate “mul( . . . ).”
3. Instance 2 obtains the identifier for mul and pushes 10 and 50 onto the stack. At
this point our stack has the state shown in Figure 6-8.
4. Instance 2 pops two operands off the stack, evaluates 10 * 50 and returns the
result to instance 1, which pushes it onto the stack. Then instance 1 again calls
instance 2 to evaluate “div( . . .).”

5. Instance 2 obtains the identifier for div and pushes 24 and 3 onto the stack which
now has the state shown in Figure 6-9.
255 0 owned by instance 0
10 1
owned by instance 2 (1)
50 2
stack pointer 3
..
Figure 6-8 Stack State: Processing mul( 10, 50 )
6. Instance 2 pops two operands off the stack, evaluates 24 / 3 and returns the result
to instance 1 which pushes it onto the stack. Instance 1 calls instance 2 once again
to evaluate “add( . . .).”
7. Instance 2 obtains the identifier for add and immediately calls instance 3 to
evaluate “mul( … ).”
24 2
owned by instance 2 (2)
3 3
stack pointer 4
..
Figure 6-9 Stack State: Processing div( 24, 3 )
8. Instance 3 obtains the identifier for mul and pushes 2 and 4 onto the stack. The
current state of the stack is now shown in Figure 6-10.
9. Instance 3 pops two operands off the stack, evaluating 2 * 4, and returns the result
to instance 2 which pushes it onto the stack. Then instance 2 pushes 2 onto the
stack.
10. Instance 2 pops two operands off the stack, computing 8 + 2, and returns the
result to instance 1which pushes it onto the stack.
11. Instance 1 pops 3 arguments off the stack, evaluating 10 + 8 + 500, and returns
the result to instance 0 which pushes it onto the stack.
12. Instance 0 pops 2 arguments off the stack, evaluating 518 & 0xff, and returns the
result to the user.

500 1
owned by instance 1
8 2
2 3
owned by instance 3
4 4
stack pointer 5
..
Figure 6-10 Stack State: Processing mul( 2, 4 )
6.4 A Minimal Stack Module

A minimal stack module will implement the following methods:
• Create a stack;
• Push an item onto a stack;
• Pop an item off of a stack;
• Peek at a stack (return the value of the last item on the stack without removing it);
• Determine whether a stack is empty;
• Determine whether a stack is full;
• Clear a stack, leaving it empty; and
• Destroy a stack.
We’ll proceed by examining a stack implementation from a functional point of view,
beginning with the public data types and discussing each method in detail; then we’ll
look at an example, and describe the details of the implementation.
6.4.1 STK Module Public Declarations

We have chosen the name STK for our module, so the public declarations will be placed
in a header file named stk.h. In addition to the prototypes for our public methods, stk.h
will contain an incomplete declaration for a stack ID, and a macro declaring a value for a
NULL stack ID (compare these declarations to those for the LIST module which we
discussed earlier). These declarations are shown in Figure 6-11.
#define STK_NULL_ID (NULL)
typedef struct stk__control_s *STK_ID_t;
Figure 6-11 STK Module Pulic Declarations
6.4.2 STK_create_stack: Create a Stack

This method will create a stack of a given size. The stack will be used for storing items
of type (void *).

Synopsis:
STK_ID_t STK_create_stack( size_t size );
Where:
size == size of the stack in items
Exceptions:
Throws SIGABRT on create failure
Returns:
stack ID to be used in all subsequent stack operations
Notes:
None
6.4.3 STK_push_item: Push an Item onto a Stack

This method will push an item of type (void *) onto a stack.
Synopsis:
void STK_push_item( STK_ID_t stack, void *item );
Where:
stack == stack id returned by STK_create_stack
item == item to push
Exceptions:
Throws SIGABRT on stack overflow
Returns:
void
Notes:
None
6.4.4 STK_pop_item: Pop an Item off a Stack

This method will remove the top item from the stack and return it.
Synopsis:
void *STK_pop_item( STK_ID_t stack );
Where:
Exceptions:
See notes
Returns:
Top item of stack
Notes:
This method contains an assertion which will throw SIGABRT if
the user attempts to illegally remove an item from an empty
stack. The assertion is disabled in production code.
6.4.5 STK_peek_item: Get the Top Item of a Stack

This method returns the top item of a stack without removing it.
Synopsis:
void *STK_peek_item( STK_ID_t stack );
Where:
Exceptions:
See notes

Returns:
The value at the top of the stack
Notes:
This method contains and assertion which will throw SIGABRT if
the user attempts to illegally get an item from an empty
stack. The assertion is disabled in production code.
6.4.6 STK_is_stack_empty: Determine If a Stack is Empty

This method indicates whether or not a stack is empty.
Synopsis:
CDA_BOOL_t STK_is_stack_empty( STK_ID_t stack );
Where:
Exceptions:
None
Returns:
CDA_TRUE if stack is empty,
CDA_FALSE otherwise
Notes:
None
6.4.7 STK_is_stack_full: Determine If a Stack is Full

This indicates whether or not a stack is full.
Synopsis:
CDA_BOOL_t STK_is_stack_full( STK_ID_t stack );
Where:
Exceptions:
None
Returns:
CDA_TRUE if stack is full,
CDA_FALSE otherwise
Notes:
None
6.4.8 STK_clear_stack
This method removes all items from a stack, leaving the stack in an empty state.
Synopsis:
void STK_clear_stack( STK_ID_t stack );
Where:
Exceptions:
None
Returns:
void
Notes:
None

6.4.9 STK_destroy_stack: Destroy a Stack

This method destroys a stack, freeing all resources associated with it.
Synopsis:
STK_ID_t STK_destroy_stack( STK_ID_t stack );
Where:
Exceptions:
None
Returns:
STK_NULL_ID
Notes:
None
6.4.10 Simple Stack Example

In this example we will reexamine our mergesort implementation. You will recall that
two of the main problems with our original implementation were its heavy reliance on
dynamic memory allocation, and dealing with pointer arithmetic. This new
implementation will substitute a single stack for the many dynamic memory allocations,
and eliminate the pointer arithmetic problem by dealing solely with arrays of pointers.
The pseudocode for the new mergesort algorithm is found in Figure 6-12. The first
difference you will notice between this implementation and our earlier one is that it is
sorting an array of void pointers, rather than merely using a void pointer to indicate the
start of a generic array. Since the implementation can now assume that the size of each
element of the array is sizeof(void *) there is no longer any reason for the user to pass an
element_size argument. It also means that the user can only sort arrays of pointer, but the
pointers may be of almost any type (specifically, any type except pointer-to-function).
Figure 6-13 shows one example of an array that a user may not sort with the modified
sort routine and two examples that she may.
Another advantage of the new strategy is, since the type of an element in the array is fully
understood by the compiler, we no longer have problems with pointer arithmetic; the
compiler can do the arithmetic quite easily. For example, dividing the array into two parts
is reduced to the following three lines of code:
lowHalf = numElements / 2;
highHalf = numElements – lowHalf;
array2 = array + lowHalf;
The next obvious difference is the logic to make sure the stack is created the first time
mergesort is called. Note that we never destroy the stack; we do this quite intentionally.
In a modern operator system allocating memory exactly once and then hanging on to it
till the program expires is not a memory leak. A drawback of this strategy is that we have
to predetermine the maximum size of an array that we are likely to want sort prior to
writing our code. There are ways to deal with this shortcoming, but to keep the problem
simple for now we will just use a macro, MAX_STACK_SIZE, which defines this size as a
constant.
STK_ID_t stack = STK_NULL_ID;
mergesort( void **array, size_t numElements )

if ( stack == STK_NULL_ID )
stack = STK_create_stack( MAX_STACK_SIZE )
if ( numElements > 1 )
lowHalf = numElements / 2
highHalf = numElements - lowHalf
mergesort( array, lowHalf )
mergesort( array2, highHalf )
inx = jnx = 0
while ( inx < lowHalf && jnx < highHalf )
STK_push_item( stack, array[inx++] )
else
STK_push_item( stack, array2[jnx++] )
while ( inx < lowHalf )

STK_push_item( stack, array[inx++] )
while ( jnx < highHalf )
STK_push_item( stack, array2[jnx++] )
inx = numElements;
while ( inx > 0 )
array[--inx] = STK_pop_item( stack )
Figure 6-12 Stack-Based Implementation of Mergesort
/* An array that may not be sorted with the new mergesort */

int iarr[] = { 5, 1, 7, 9 };
/* Two arrays that may be sorted with the new mergesort */

char *sarr[] = { “dick”, “sally”, “jane”, “spot” };
int *piarr[] = { &iarr[0], &iarr[1], &iarr[2], &iarr[3] };
mergesort( (void**)sarr, CDA_CARD(sarr) ):

mergesort( (void**)piarr, CDA_CARD(piarr) );
Figure 6-13 Calling the New Mergesort Routine
6.4.11 Implementation Details

As illustrated in Figure 6-14, our private header file will consist of a single declaration: a
control structure which contains a pointer to an array of type void* to use as the stack, a
stack pointer and the size of the stack.

stack
#ifndef STKP_H
#define STKP_H sptr occupied
#include <stk.h> size occupied

#include <stddef.h>
occupied
typedef struct stk__control_s

{
void **stack;
void **sptr;
size_t size;
} STK__CONTROL_t, *STK__CONTROL_p_t;
#endif
Figure 6-14 STK Module Private Header File
The implementation of all but three of our methods is show below. The implementation
of STK_push_item, STK_pop_item and STK_is_stack_full is left as an exercise.
#include <stkp.h>
#include <stdlib.h>
void STK_clear_stack( STK_ID_t stack )

{
stack->sptr = stack->stack;
}
STK_ID_t STK_create_stack( size_t size )

{
STK__CONTROL_p_t stack = CDA_NEW( STK__CONTROL_t );
stack->stack = CDA_calloc( size, sizeof(void *) );

stack->sptr = stack->stack;
stack->size = size;
return (STK_ID_t)stack;
}
STK_ID_t STK_destroy_stack( STK_ID_t stack )

{
CDA_free( stack->stack );
CDA_free( stack );
return STK_NULL_ID;
}
void *STK_peek_item( STK_ID_t stack )

{
CDA_ASSERT( stack->sptr != stack->stack );
return *(stack->sptr - 1);
}
void *STK_pop_item( STK_ID_t stack )

{
}
void STK_push_item( STK_ID_t stack, void *item )

{
}
CDA_BOOL_t STK_is_stack_empty( STK_ID_t stack )

{
CDA_BOOL_t rcode =
stack->sptr == stack->stack ? CDA_TRUE : CDA_FALSE;
return rcode;
}
CDA_BOOL_t STK_is_stack_full( STK_ID_t stack )

{
}
6.5 A More Robust Stack Module

In addition to the basic functionality described above, additional features are often
available in stack implementations. The most popular ones are these:
• Mark the stack;
• Index into the stack from a mark;
• Clear the stack to a mark;
• Grab stack space; and
• Create a segmented stack.
This additional functionality will be discussed below.
6.5.1 Stack Marks

A stack mark identifies some position within the stack, and permits the user full access to
the occupied portion of the stack. In order to implement marks, we need a new public
declaration: a stack mark type typedef. Our purposes, a stack mark will just be an integer
index into the stack:
typedef int STK_MARK_t, *STK_MARK_p_t;
Next, we need to modify the push method so that it returns a mark representing the
position of the pushed item.
STK_MARK_t STK_push_item( STK_ID_t stack, void *item )
{
STK_MARK_t mark = stack->sptr - stack->stack;
. . .
return mark;
}
Now we can define two new methods: one to interrogate the stack at some offset from a
mark, and one to change the value of a stack location at some offset from a mark. We

will validate, using assertions, that the indexed value is legal. The implementation of
these routines is shown in Figure 6-15 and Figure 6-16.
void *STK_get_item( STK_ID_t stack, STK_MARK_t mark, int offset )
{
CDA_ASSERT( stack->stack + mark >= stack->stack );
CDA_ASSERT( stack->stack + mark < stack->sptr );
CDA_ASSERT( stack->stack + mark + offset >= stack->stack );
CDA_ASSERT( stack->stack + mark + offset < stack->sptr );
return *(stack->stack + mark + offset);

}
Figure 6-15 STK_get_item
void STK_change_item( STK_ID_t stack,

STK_MARK_t mark,
int offset,
void *val
)
{
CDA_ASSERT( stack->stack + mark + offset >= stack->stack );
CDA_ASSERT( stack->stack + mark + offset < stack->sptr );
*(stack->stack + mark + offset) = val;

}
Figure 6-16 STK_change_item
To access an item at a particular location on the stack, the user passes a previously
obtained mark plus an offset. An offset of 0 indicates the marked location itself; an offset
of –1 indicates the item that was pushed immediately before the marked item, and an
offset of 1 indicates the item immediately after the marked item, etc.
The methods to clear a stack and to grab stack space are shown in Figure 6-17 and
Figure 6-18. To clear a stack we simply reset the stack pointer to a marked location; to
grab stack space we simply advance the stack pointer; although we have not explicitly
written any values to the stack locations that we skipped, those locations are now
considered occupied, and may be modified using STK_change_item.
void STK_clear_to_mark( STK_ID_t stack, STK_MARK_t mark )
{
stack->sptr = stack + mark;

}
Figure 6-17 STK_clear_to_mark
STK_MARK_t STK_grab_space( STK_ID_t stack,

int num_slots,
STK_MARK_p_t bottom_mark
)
{
if ( stack->sptr + num_slots >= stack->stack + stack->size )
abort();

if ( bottom_mark != NULL )
*bottom_mark = stack->sptr - stack->stack;
stack->sptr += num_slots;
return stack->sptr - stack->stack - 1;

}
Figure 6-18 STK_grab_space
6.5.2 Segmented Stacks

A segmented stack is implemented via a noncontiguous array. When an application tries
to push an item onto a segmented stack that is full, instead of throwing an exception the
implementation allocates a new stack segment and silently performs the operation. In
order to implement a segmented stack, we’ll need to make a few changes to our data
structure.
flink
blink
name
size
occupied occupied
occupied occupied
occupied occupied
occupied occupied
occupied
occupied
occupied
Figure 6-19 Segmented Stack Layout
First, with a segmented stack, it will be easier to make our stack pointer an integer index
rather than a pointer; that will make it easier to figure out which segment a stack slot falls
in. Second, as suggested by Figure 6-19, each segment in the stack will be an controlled
by an enqueuable item; and rather than keeping a pointer to the stack in the stack control
structure, we will keep a pointer to a list anchor, plus a pointer to the current stack
segment. The new declarations and create method are shown in Figure 6-20 and Figure
6-21. Note that since none of our public API has changed, a segmented implementation
could replace a simple implementation with absolutely no impact on the applications that
use it.
Pushing and popping data on a segmented stack now requires watching out for segment
boundaries, and allocating a new segment when an application tries to push data onto a
full stack. The implementation details are left to the imagination of the reader.
typedef struct stk__stack_seg_s
{

ENQ_ITEM_t item;
void **stack;
} STK__stack_seg_t, *STK__stack_seg_p_t;
typedef struct stk__control_s

{
ENQ_ANCHOR_p_t seg_list;
STK__STACK_SEG_p_t curr_seg;
int sptr;
size_t size;
} STK__CONTROL_t, *STK__CONTROL_p_t;
Figure 6-20 Segmented Stack Private Declarations
STK_ID_t STK_create_stack( size_t size )

{
STK__CONTROL_p_t stack = CDA_NEW( STK__CONTROL_t );
stack->seg_list = ENQ_create_list( “Stack” );

stack->curr_seg = (STK__STACK_SEG_p_t)
ENQ_create_item( sizeof(STK__STACK_SEG_t) );
ENQ_add_tail( stack->seg_list,(ENQ_ITEM_p_t)stack->curr_seg);
stack->curr_seg->stack = CDA_calloc( size, sizeof(void *) );
stack->sptr = 0;
stack->size = size;
return (STK_ID_t)stack;
}
Figure 6-21 Segmented Stack Create Method

7. Priority Queues
In this section we will discuss the queue, a very simple structure that organizes data on a
first-come, first-served basis, and the priority queue, which organizes data according to
an arbitrary designation of importance. Priority queues are a practical means of
implementing schedulers, such as real-time operating system schedulers that execute
processes, or communications schedulers that transmit messages according to their
priority.
In addition, we will continue our discussion of implementation choices that trade-off
speed against data size, and issues of algorithmic complexity. And we will examine two
priority queue implementations: a simple implementation that offers basic functionality
with relatively low complexity and slow execution, and a robust implementation that
offers enhanced functionality and fast execution at the cost of additional complexity and
data size requirements.
7.1 Objectives
At the conclusion of this section, and with the successful completion of your fourth
• Discuss the usefulness and application of priority queues;
• Choose between algorithms that trade-off speed and complexity against data size;
and
• Perform a complete, modularized priority queue implementation.
7.2 Overview
A queue may be visualized as an array to which elements may be added or removed. A
new element is always added to the end of the queue, and elements are always removed
from the front of the queue. Therefore the first element added to the queue is always the
first element removed, so a queue is often referred to a first-in, first-out queue or FIFO.
This arrangement is illustrated in Figure 7-1. A typical application of a queue is an order-
fulfillment system. An online bookseller, for example, takes an order from a customer
over the Internet; orders are added to the end of a queue as they are received. In the
shipping department orders are removed from the front of the queue and fulfilled in the
order that they were received.
A priority queue is a queue in which each element is assigned a priority. A priority is
typically an integer, and higher integers represent higher priorities. All elements having
the same priority are said to belong to the same priority class. As seen in Figure 7-2, a
priority queue may be visualized as an array in which all elements of the same priority
class are grouped together, and higher priority classes are grouped closer to the front of
the array. A new element is added to the tail of its priority class, and elements are always
removed from the front of the queue. A remove operation, therefore, always addresses the
element of the highest priority class that has been enqueued the longest. Imagine
modifying our order fulfillment system to take into account preferred customers. Let’s
say that customers with ten years standing get served before customers with five years
Section 7: Priority Queues 87 06/19/06

standing, and that customers with five years standing get served before new customers.
Then when a customer with five years standing places an order, the order goes in the
queue ahead of orders from new customers, but behind older orders from customers with
five years standing or more.
data
add
data data data data data data
remove
data
Figure 7-1 Visualizing a Queue as an Array
30
data
add
31 30 30 30 25 25 20
data data data data data data data
31 remove
data
Figure 7-2 Visualizing a Priority Queue as an Array
Let’s discuss queues and priority queues a little more detail. We’ll start with queue.
7.3 Queues
As discussed in your textbook, a queue is an ordered sequence of elements on which we
may perform the following operations (note: your textbook defines only nine operations;
I have taken the liberty of adding the destroy operation):
1. Create a queue;
2. Determine if a queue is empty;
3. Determine if a queue is full;
4. Determine the size of a queue;
5. Append a new entry to a queue;

6. Retrieve (without removing) the entry at the front of a queue;
7. Remove the entry at the front of a queue;
8. Clear a queue;
9. Traverse a queue; and
10. Destroy a queue.
As with the list module that we saw earlier, we have a couple of choices when it comes to
the actual implementation of a queue, but we need not make any of those choices in order
to fully specify the behavior of a queue; which is another way of saying that we should be
able to define the public interface of a queue module without specifying the private
details. And to define the public interface first we must choose a module name, then we
must declare the data types to represent:
• The ID of a queue;
• Any callback functions associated with a queue operation; and
• The data that will be used as input and output values for methods such as remove
and append.
The name of our sample queue module will be QUE. Once again employing an
incomplete declaration to achieve both encapsulation and strong typing, a queue ID will
be a pointer to a control structure tag, and we will declare a value to be used as a NULL
queue ID. There will be two callback function types associated with our module, one for
traversing a queue, and one for destroying it. In each case a callback will have a single
argument representing a pointer to the user data associated with an element in the queue
and returning void. That gives us the public declarations shown in Figure 7-3.
#define QUE_NULL_ID (NULL)
typedef struct que__control_s *QUE_ID_t;
typedef void QUE_DESTROY_PROC_t( void *data );

typedef QUE_DESTROY_PROC_t *QUE_DESTROY_PROC_p_t;
typedef void QUE_TRAVERSE_PROC_t( void *data );
typedef QUE_TRAVERSE_PROC_t *QUE_TRAVERSE_PROC_p_t;
Figure 7-3 QUE Module Public Declarations
For input and output values for such functions as append and remove we could keep it
simple and just use a void pointer to represent the user’s data. However, experience
suggests that applications often entwine the use of queues and other kinds of lists. For
example, in a communications application it is not unusual for a process to undergo state
changes such as this one:
1. Wait in a work-in-progress queue while transmission parameters are established;
2. Switch to a priority queue awaiting the availability of resources to perform the
transmission;
3. Switch to a response queue awaiting a response to the transmission;
4. Switch to a garbage collection queue to await destruction or reuse.

To facilitate such state changes it makes sense for a queue element to be able to move
quickly and easily from one queue to another. To that end we will make the design
decision that a value appended to or removed from a queue will always be represented by
an enqueuable item (as defined by our ENQ module), and that such an item will contain a
pointer to the user’s data. To accomplish this, we will need one additional declaration for
a queue item, which is shown in Figure 7-4; we will also define two additional queue
operations:
• Create and return a queue item; and
• Destroy a queue item.
typedef struct que_item_s
{
ENQ_ITEM_t item;
void *data;
} QUE_ITEM_t, *QUE_ITEM_p_t;
Figure 7-4 Another QUE Module Public Declaration
Our decision to make our queue a collection of enqueuable items has pretty much
determined that the heart of our implementation will be a linked list as defined by our
ENQ module, so the private header file for our queue module will consist of the
declaration of a control structure containing a pointer to an anchor. The complete private
header file is shown in Figure 7-5. Next let’s discuss the details of just a couple of the
QUE module methods.
#ifndef QUEP_H
#define QUEP_H
#include <que.h>
#include <enq.h>
typedef struct que__control_s

{
} QUE__CONTROL_t, *QUE__CONTROL_p_t;
#endif
Figure 7-5 QUE Module Private Header File
7.3.1 QUE_create_queue
This method will create a new, empty queue, and return to the user an ID to use in future
operations.
Synopsis:
QUE_ID_t QUE_create_queue( const char *name );
Where:
name == queue name; may be NULL
Exceptions:
Throws SIGABRT if the queue cannot be created
Returns:
Queue ID
Notes:
None


QUE_ID_t QUE_create_queue( const char *name )
{
QUE__CONTROL_p_t qid = CDA_NEW( QUE__CONTROL_t );
qid->anchor = ENQ_create_list( name );

return qid;
}
7.3.2 QUE_create_item
This method will create a queue item containing the user’s data.
Synopsis:
QUE_ITEM_p_t QUE_create_item( const char *name, void *data );
Where:
name == item name; may be NULL
data == user’s data
Exceptions:
Throws SIGABRT if the item cannot be created
Returns:
Address of queue item
Notes:
None
QUE_ITEM_p_t QUE_create_item( const char *name, void *data )
{
QUE_ITEM_p_t item =
(QUE_ITEM_p_t)ENQ_create_item( name, sizeof(QUE_ITEM_t) );
item->data = data;
return item;
}
7.3.3 QUE_clear_queue
This method will destroy all the items in a queue, leaving the queue empty.
Synopsis:
QUE_ID_t QUE_clear_queue( QUE_ID_t qid,
QUE_DESTROY_PROC_p_t destroyProc
);
Where:
qid == ID of queue to clear
destroyProc == address of destroyProc; may be NULL
Exceptions:
None
Returns:
Queue ID
Notes:
If the data contained in the queue items requires cleanup, the
user should pass the address of a clean up function as the
destroy proc argument. If non-NULL, the clean up function will

be called once for each item in the queue, passing the data
from the item.
QUE_ID_t QUE_clear_queue( QUE_ID_t qid,
QUE_DESTROY_PROC_p_t destroyProc
)
{
QUE_ITEM_p_t item = NULL;
while ( !ENQ_is_list_empty( qid->anchor ) )

{
item = (QUE_ITEM_p_t)ENQ_GET_HEAD( qid->anchor );
if ( destroyProc != NULL )
destroyProc( item->data );
QUE_destroy_item( item );
}
return qid;
}
7.3.4 Other QUE Module Methods

Here is a quick synopsis of the other QUE module methods; the implementation is left to
the reader.
QUE_ITEM_p_t QUE_append( QUE_ID_t queue, QUE_ITEM_p_t item )
appends item to queue->anchor using ENQ_add_tail; item is returned.
CDA_BOOL_t QUE_is_queue_empty( QUE_ID_t queue )
tests queue->anchor using ENQ_is_list_empty and returns the result.
CDA_BOOL_t QUE_is_queue_full( QUE_ID_t queue )
always returns CDA_FALSE.
QUE_ITEM_p_t QUE_destroy_item( QUE_ITEM_p_t item )
uses ENQ_destroy_item to destroy item; returns NULL.
QUE_ID_t QUE_destroy_queue( QUE_ID_t qid, QUE_DESTROY_PROC_p_t
destroyProc )
calls QUE_clear_queue to empty the queue, then destroys qid->anchor using
ENQ_destroy_list and frees qid using CDA_free; returns QUE_NULL_ID.
QUE_ITEM_p_t QUE_remove( QUE_ID_t queue )
removes and returns the first item in the queue using ENQ_deq_head; returns
NULL if the queue is empty.
QUE_ITEM_p_t QUE_retrieve( QUE_ID_t queue )
returns, without removing, the first item in the queue using ENQ_GET_HEAD;
returns NULL if the queue is empty.
QUE_ID_t QUE_traverse_queue( QUE_ID_t queue,
QUE_TRAVERSE_PROC_p_t traverse_proc )
traverses the queue using ENQ_GET_HEAD and ENQ_GET_NEXT; for each item
in the queue, calls traverse_proc passing item->data.

7.3.5 QUE Module Sample Program

Figure 7-6 shows a sample program that uses the QUE module. It doesn’t do anything
particularly interesting; it merely demonstrates how the QUE module methods are used.
7.4 Simple Priority Queues

Next we want to define a simple priority queue, and design a module to encapsulate one.
For our purposes, a simple priority queue implementation consists of the following
operations:
1. Create a queue;
2. Create a queue item;
3. Determine if a queue is empty;
4. Add a new item to a queue;
5. Remove the item at the front of a queue;
6. Empty a queue;
7. Destroy a queue item; and
8. Destroy a queue.
In the interest of encapsulation, we will also consider two additional methods, called
accessor methods because they access a field in a structure:
9. Get the priority associated with an item; and
10. Get the data associated with an item.
As with queues, we now have to choose a name for our priority queue module, decide
how to encapsulate the ID of a priority queue, declare the types of any callback functions,
and determine the type of an element that will populate a priority queue. Our module
name will be PRQ, and an ID will once again be an incomplete declaration of a pointer to
a control structure; we will also need to declare a value to use as a NULL ID. We will
have one callback function for use with the empty and destroy methods.

#include <que.h>
#include <stdio.h>
#include <stdlib.h>
static QUE_TRAVERSE_PROC_t traverse;

static QUE_DESTROY_PROC_t destroy;
static const char *itemNames_[] =

{ "hydrogen", "helium", "lithium", "beryllium", "boron",
"carbon", "nitrogen", "oxygen", "fluorine", "neon"
};
int main( int argc, char **argv )

{
int *data = NULL;
QUE_ID_t qid = QUE_NULL_ID;
QUE_ITEM_p_t item = NULL;
int inx = 0;
qid = QUE_create_queue( "test" );

for ( inx = 0 ; inx < CDA_CARD( itemNames_ ) ; ++inx )
{
data = CDA_malloc( sizeof(int) );
*data = inx;
item = QUE_create_item( itemNames_[inx], data );
QUE_append( qid, item );
}
QUE_traverse_queue( qid, traverse );

for ( inx = 0 ; inx < CDA_CARD( itemNames_ ) / 2 ; ++inx )
{
item = QUE_remove( qid );
printf( "Removing %s\n", item->item.name );
CDA_free( item->data );
QUE_destroy_item( item );
}
QUE_destroy_queue( qid, destroy );

return EXIT_SUCCESS;
}
static void traverse( void *data )

{
int *iData = data;
printf( "Traversing %d\n", *iData );
}
static void destroy( void *data )

{
int *iData = data;
printf( "Destroying %d\n", *iData );
CDA_free( iData );
}
Figure 7-6 Sample QUE Module Usage

As with our queue module, a member of the queue will be an enqueuable item within
which user data will be stored as a void*; plus we will need to store an integer value
representing the priority of the queue item. The public declarations for our PRQ module
are shown in Figure 7-7.
#define PRQ_NULL_ID (NULL)
typedef void PRQ_DESTROY_PROC_t( void *data );

typedef PRQ_DESTROY_PROC_t *PRQ_DESTROY_PROC_p_t;
typedef struct prq__control_s *PRQ_ID_t;

typedef struct prq_item_s
{
ENQ_ITEM_t enq_item;
void *data;
CDA_UINT32_t priority;
} PRQ_ITEM_t, *PRQ_ITEM_p_t;
Figure 7-7 PRQ Module Public Declarations
Like the queue module, choosing an enqueuable item as type of element to populate the
priority queue strongly suggests that there will be an ENQ-style list at the heart of our
implementation, and our private declarations are going to strongly resemble the QUE
module private declarations. However, for reasons that will become clear later, we also
want to introduce the idea of a maximum priority. This will be an integer value defining
an upper limit on the priority of an item in a given priority queue. Therefore the control
structure for our PRQ implementation will contain both a pointer to an ENQ anchor, plus
an integer field for storing the maximum priority. The complete private header file for
our simple priority queue implementation is shown in Figure 7-8.
#ifndef PRQP_H
#define PRQP_H
#include <prq.h>
#include <enq.h>
typedef struct prq__control_s

{
CDA_UINT32_t max_priority;
} PRQ__CONTROL_t, *PRQ__CONTROL_p_t;
#endif
Figure 7-8 Simple PRQ Module Private Header File
Now let’s look at the method specification for our simple priority queue. After that we’ll
see an example of using a priority queue, then examine the simple priority queue
implementation.
7.4.1 PRQ_create_priority_queue
This function will create a priority queue, and return the ID of the queue to be used in all
subsequent priority queue operations. The caller names the priority queue, and provides
the maximum priority for the queue as an unsigned integer. The maximum priority is

saved and checked on subsequent operations, but is not otherwise used in the current
implementation (it is for possible future use).
Synopsis:
PRQ_ID_t PRQ_create_queue( const char *name,
CDA_UINT32_t max_priority
);
Where:
name -> queue name
max_priority == maximum priority supported by queue
Returns:
queue id
Exceptions:
Throws SIGABRT if queue can’t be created
Notes:
max_priority is stored and checked on subsequent operations,
but is not otherwise used at this time.
7.4.2 PRQ_create_item
This method will create an item that is a subclass of ENQ_ITEM_t, and that can be added
to a priority queue. The item is created in an unenqueued state.
Synopsis:
PRQ_ITEM_p_t PRQ_create_item( void *value,
CDA_UINT32_t priority
);
Where:
value -> value to be stored in the item
priority == the priority of the item
Returns:
Address of created item
Exceptions:
Throws SIGABRT if item cannot be created
Notes:
None
7.4.3 PRQ_is_queue_empty
This method will determine whether a priority queue is empty.
Synopsis:
CDA_BOOL_t PRQ_is_queue_empty( PRQ_ID_t queue );
Where:
queue == ID of the queue to test
Returns:
CDA_TRUE if the queue is empty, CDA_FALSE otherwise
Exceptions:
None
Notes:
None

7.4.4 PRQ_add_item
This method will add an item to a priority queue.
Synopsis:
PRQ_ITEM_p_t PRQ_add_item( PRQ_ID_t queue,
PRQ_ITEM_p_t item
);
Where:
queue == id of priority queue
item -> item to add
Returns:
Address of enqueued item
Exceptions:
Throws SIGABRT if the item’s priority is higher than the
maximum priority allowed for the queue.
Notes:
None
7.4.5 PRQ_remove_item
This method removes and returns the highest priority item from a priority queue.
Synopsis:
PRQ_ITEM_p_t PRQ_remove_item( PRQ_ID_t queue );
Where:
queue == id of target priority queue
Returns:
If priority queue is non-empty:
The address of the highest priority item in the queue
Otherwise:
NULL
Exceptions:
None
Notes:
None
7.4.6 PRQ_GET_DATA
This is a macro that will return the user data contained in a priority item.
Synopsis:
void *PRQ_GET_DATA( PRQ_ITEM_p_t item );
Where:
item -> item from which to retrieve value
Returns:
The value of item
Exceptions:
None
Notes:
None
7.4.7 PRQ_GET_PRIORITY
This is a macro that will return the priority of a priority item.

Synopsis:
CDA_UINT32_t PRQ_GET_PRIORITY( PRQ_ITEM_p_t item );
Where:
item -> item from which to retrieve priority
Returns:
The priority of item
Exceptions:
None
Notes:
None
7.4.8 PRQ_destroy_item
This method will destroy a priority queue item.
Synopsis:
PRQ_ITEM_p_t PRQ_destroy_item( PRQ_ITEM_p_t item );
Where:
item -> item to destroy;
Returns:
NULL
Exceptions:
None
Notes:
None
7.4.9 PRQ_empty_queue
This method will remove and destroy all items in a priority queue. The caller may
optionally pass the address of a procedure to call prior to destroying the item; if specified,
the value member of each destroyed item will be passed to this procedure.
Synopsis:
PRQ_ID_t PRQ_empty_queue( PRQ_ID_t queue,
PRQ_DESTROY_PROC_p_t destroy_proc
);
Where:
queue == id of queue to empty
destroy_proc -> optional destroy callback procedure
Returns
Queue ID
Exceptions:
None
Notes:
The caller may pass NULL for the destroy_proc parameter.
7.4.10 PRQ_destroy_queue
This method will destroy all items in a priority queue, and then destroy the queue. The
caller may optionally pass the address of a procedure to call prior to destroying each item
in the queue; if specified, the value of each destroyed item will be passed to this
procedure.

Synopsis:
PRQ_ID_t PRQ_destroy_queue( PRQ_ID_t queue,
);
Where:
queue == id of queue to destroy
destroy_proc -> optional destroy callback procedure
Returns
PRQ_NULL_ID
Exceptions:
None
Notes:
The caller may pass NULL for the destroy_proc parameter.
7.4.11 Priority Queue Example

This example depicts a module that maintains a queue of transactions awaiting execution.
The principal functionality is contained in two independent methods,
TRANS_enq_transaction and TRANS_get_transaction. In addition the module has an
initialization method and a shutdown method. These four methods are discussed below,
followed by the code for module.
TRANS_enq_transaction:
This method accepts data representing a transaction to execute, and a
string representing the operation. It must enqueue the transaction for later
execution, presumably when the appropriate system resources become
available. The key idea here is that different types of operations are
assigned different priorities; in particular, add transactions have the
highest priority, modify transactions have a priority just below add
transactions and delete transactions have the lowest priority. The method
dynamically allocates memory in which to store the transaction data, and
an enumerated value specifying the operation. PRQ_create_item is called
specifying the dynamically allocated memory as the data, and a priority
appropriate for the type of operation, and the item it creates is added to the
queue.
TRANS_get_transaction:
This method removes from the queue an item containing a transaction to
be executed. Assuming there is such an item, the method removes from
the item the transaction data and the operation indicator; the transaction
data will be the method’s return data, and the operation is translated into a
string which is returned via the operation parameter. Prior to returning, the
PRQ item is destroyed. If the queue is empty, meaning that there are no
pending transactions to execute, the method returns NULL.
TRANS_init:
This method is to be called once by the application prior to any calls to
TRANS_enq_transaction or TRANS_get_transaction. It performs module
initialization, which in this case simply means creating the transaction
queue.

TRANS_shutdown:
This method is to be called by the application to terminate transaction
processing. The queue is destroyed. Note that many items contain memory
that was dynamically allocated by TRANS_enq_transaction, and now
requires freeing. This is accomplished by passing to PRQ_destroy_queue
the address of a destroy proc to perform the cleanup. (Note that the destroy
proc in the example code contains a call to printf that you wouldn’t
normally see in production code; the logic is there to help make the
example a little more meaningful to the reader.)
#include <stdio.h>
#include <cda.h>
#include <prq.h>
#include <trans.h>
#define DEL_PRI (0)

#define MOD_PRI (1)
#define ADD_PRI (2)
#define MAX_PRI (ADD_PRI)
typedef enum operation_e

{
del,
mod,
add
} OPERATION_e_t;
typedef struct queue_data_s

{
OPERATION_e_t operation;
void *data;
} QUEUE_DATA_t, *QUEUE_DATA_p_t;
static PRQ_DESTROY_PROC_t destroyProc;
static PRQ_ID_t queue_ = PRQ_NULL_ID;
void TRANS_init( void )

{
CDA_ASSERT( queue_ == PRQ_NULL_ID );
queue_ = PRQ_create_queue( "TRANSACTION QUEUE", MAX_PRI );
}
void TRANS_shutdown( void )

{
if ( queue_ != PRQ_NULL_ID )
queue_ = PRQ_destroy_queue( queue_, destroyProc );
}
void TRANS_enq_transaction( void *data, const char *operation )

{
CDA_UINT32_t pri = 0;
QUEUE_DATA_p_t qdata = CDA_NEW( QUEUE_DATA_t );
PRQ_ITEM_p_t item = NULL;

qdata->data = data;
if ( strcmp( operation, "delete" ) == 0 )
{
pri = DEL_PRI;
qdata->operation = del;
}
else if ( strcmp( operation, "modify" ) == 0 )
{
pri = MOD_PRI;
qdata->operation = mod;
}
else if ( strcmp( operation, "add" ) == 0 )
{
pri = ADD_PRI;
qdata->operation = add;
}
else
abort();
item = PRQ_create_item( qdata, pri );

PRQ_add_item( queue_, item );
}
void *TRANS_get_transaction( const char **operation )

{
QUEUE_DATA_p_t qdata = NULL;
void *data = NULL;
PRQ_ITEM_p_t item = PRQ_remove_item( queue_ );
if ( item != NULL )
{
qdata = PRQ_GET_DATA( item );
data = qdata->data;
switch ( qdata->operation )
{
case del:
*operation = "delete";
break;
case mod:
*operation = "modify";
break;
case add:
*operation = "add";
break;
default:
assert( CDA_FALSE );
break;
}
CDA_free( qdata );
PRQ_destroy_item( item );
}
return data;
}
static void destroyProc( void *data )

{
QUEUE_DATA_p_t qdata = data;
printf( "Destroying transaction type %d\n", qdata->operation
);
CDA_free( qdata );
}
7.4.12 Simple Priority Queue Module Implementation

The code for the simple implementation is shown below, except for the details of the
create method, which will be an exercise for the reader. Note that, thanks to careful
selection of the mechanism for implementing the enabling doubly linked module, the
code is quite straightforward.
#include <stdlib.h>
#include <cda.h>
#include <enq.h>
#include <prqp.h>
PRQ_ID_t PRQ_create_queue( const char *name,

CDA_UINT32_t max_priority
)
{
. . .
}
PRQ_ITEM_p_t PRQ_create_item( void *data, CDA_UINT32_t priority )

{
item = (PRQ_ITEM_p_t)ENQ_create_item( NULL,

sizeof(PRQ_ITEM_t)
);
item->data = data;
item->priority = priority;
return item;
}
PRQ_ITEM_p_t PRQ_add_item( PRQ_ID_t queue, PRQ_ITEM_p_t item )

{
CDA_BOOL_t found = CDA_FALSE;
PRQ_ITEM_p_t temp = NULL;
ENQ_ANCHOR_p_t anchor = queue->anchor;
if ( item->priority > queue->max_priority )

abort();
temp = (PRQ_ITEM_p_t)ENQ_GET_HEAD( anchor );

while ( !found && temp != NULL )
if ( temp == (PRQ_ITEM_p_t)anchor )
temp = NULL;
else if ( temp->priority < item->priority )
found = CDA_TRUE;
else
temp = (PRQ_ITEM_p_t)ENQ_GET_NEXT( (ENQ_ITEM_p_t)temp );

if ( temp != NULL )
ENQ_add_before( (ENQ_ITEM_p_t)item, (ENQ_ITEM_p_t)temp );
else
ENQ_add_tail( anchor, (ENQ_ITEM_p_t)item );
return item;
}
PRQ_ITEM_p_t PRQ_remove_item( PRQ_ID_t queue )

{
ENQ_ITEM_p_t item = ENQ_deq_head( queue->anchor );
if ( item == queue->anchor )
item = NULL;
return (PRQ_ITEM_p_t)item;
}
PRQ_ITEM_p_t PRQ_destroy_item( PRQ_ITEM_p_t item )

{
ENQ_destroy_item( (ENQ_ITEM_p_t)item );
return NULL;
}
CDA_BOOL_t PRQ_is_queue_empty( PRQ_ID_t queue )

{
CDA_BOOL_t rcode = ENQ_is_list_empty( queue->anchor );
return rcode;
}
PRQ_ID_t PRQ_empty_queue( PRQ_ID_t queue,

)
{
while ( !ENQ_is_list_empty( anchor ) )

{
item = (PRQ_ITEM_p_t)ENQ_GET_HEAD( anchor );
destroy_proc( item->data );
ENQ_destroy_item( (ENQ_ITEM_p_t)item );
}
return queue;
}
PRQ_ID_t PRQ_destroy_queue( PRQ_ID_t queue,

)
{
PRQ_empty_queue( queue, destroy_proc );
ENQ_destroy_list( queue->anchor );
CDA_free( queue );

return PRQ_NULL_ID;
}
7.5 A More Robust Priority Queue Implementation

The simple priority queue implementation discussed above is suitable for many
applications. But many other applications require additional functionality. Consider the
following three scenarios, in which an operating system utilizes a priority queue to
execute processes (that is, programs) of various priorities:
Scenario 1: A priority 25 process has control of the CPU when it is preempted by
a priority 30 process; this means that the priority 25 process must give up control
of the CPU so that the priority 30 process can be run in its place. The priority 25
process will be returned to the priority queue to wait at the head of its priority
class until such time as the priority 30 process is finished. Our simple priority
queue implementation is unsuitable for this because it can only add items to the
tail of a priority class.
Scenario 2: A few priority 15 processes are dominating control of the CPU,
meanwhile several priority 14 processes are languishing in the priority queue. To
circumvent this possibly damaging situation the operating system will
occasionally ignore the priority 15 processes, and grant control to the first waiting
priority 14 process; to do this, it must be able to remove the item at the head of
priority class 14.
Scenario 3: The operating system allows an operator to cancel a process waiting
in the priority queue. To do this it must be able to remove an item from the
priority queue regardless of its position.
A more robust priority queue implementation might offer these additional public
methods:
• Dequeue an arbitrary item from the queue;
• Enqueue an item at the head of its priority class; and
• Dequeue an item from the head of its priority class.
Implementing these additional methods require no changes to the public API that we
defined for the simple implementation presented above. Some new methods have to be
added to the API, and some changes need to be made to the module internals; they are the
subjects of your next project. In addition, we want to consider an optimization of the
implementation.
The scenarios cited above require the intervention of a part of an operating system called
the scheduler because it schedules processes for execution. The scheduler is a critical
component that runs virtually every time a process is executed; a slow scheduler will
have a negative impact on nearly every task that your computer performs. It is therefore
in our interest to make sure that the scheduler runs as quickly as possible. Consider the
scheduler’s interaction with the priority queue; from the point of view of the simple
implementation that we examined in the last section, what is the most time consuming
operation that it has to perform? That would be adding an item to the queue, because it

has to search the entire linked list to locate the end of the target priority class. If we keep
this same architecture for the robust implementation we will have the same problem with
locating the head of a priority class in order to execute the enqueue-priority-class-head
and dequeue-priority-class-head operations. So to optimize the operation of our robust
implementation we will borrow from the design of the VAX/VMS scheduler, originally
designed by Digital Equipment Corporation, which eliminates searching for the head and
tail of a priority class.
flink flink flink flink

blink blink blink blink
name name name name
data data data
pri = 0 pri = 0 pri = 0
flink flink
blink blink
name name
data
pri = 1
flink
blink
name
flink flink flink

blink blink blink
name name name
data data
0 pri = 3 pri = 3
1
2
3
Figure 7-9 PRQ Optimized Structure
Refer to Figure 7-9. This depicts an implementation strategy that uses not one linked list,
but an array of linked lists. The array contains one pointer-to-anchor for every priority
class, and each list contains only items from a single priority class. Now to add an item to
the head or tail of a priority class, we can use the priority as an index into the array to
find the target list; once we have the target list, we merely add the item using
ENQ_add_tail (for the normal add method) or ENQ_add_head (for the new add-priority-
class-head method).
Note that our new implementation now runs faster, but at the expense of occupying
additional space, and some added complexity of implementation. In the simple
implementation we had to create one control structure, plus one list. In the robust
implementation we will have to allocate a control structure, then allocate an array of
pointers-to-anchors, then create a list for each element of the array. And to remove an
item from the queue, we must first locate the non-empty queue associated with the
highest priority class and then remove its head.

8. The System Life Cycle

In this section we will examine the process of system development, called the system life
cycle. This describes how a system is conceived, designed, constructed and maintained.
We will also take a close look at the testing activities that are needed at each step of the
system life cycle in order to ensure the success of the project.
8.1 Objectives
At the conclusion of this section you will be able to:
• Describe the components of the system life cycle; and
• Describe how testing fits into the system life cycle.
8.2 Overview
The steps in which a data processing system should be constructed are called the system
life cycle. There are many different ways of viewing this process, and many different
disciplines that define it, but they all contain these five essential activities:
1. Specification
2. Design
3. Implementation
4. Acceptance Testing
5. Maintenance
The process is said to be iterative. A problem found during the design step could lead to a
reevaluation of the specification. A problem found during implementation could lead to a
change in the system design, which could in turn lead to changes in the system
specification.
Note that step four is dedicated to testing. This does not refer to the kind of testing that
you, as a developer, perform in the course of writing your code. This is a special kind of
activity that concentrates on proving the requirements of the system as established in the
specification phase. In fact, as we’ll see, testing activities are not confined to any one
phase of the process, but take place at every step along the way.
The sections below will present an overview of each phase of the system life cycle. Then
we’ll go back and examine each phase a second time, concentrating on the testing
activities that take place for that phase.
8.2.1 Specification Phase

This phase of the system life cycle will define and bound the problem to be solved, or
process to be executed. No coding is done at this time. Only the functional requirements
of the system are discussed: What is the system trying to accomplish? What data are to be
captured? Once the data are captured, what questions will they answer, and what
decisions will they help us to make?
Section 8: The System Life Cycle 107 06/19/06

During this step, a great deal of attention is paid to the external specification (in some
disciplines, this is broken into a separate step). The external specification defines the
characteristics of human interaction with the system. It defines the format of the screens
(or graphical user interface) that will be used for data entry, and the reports that will be
used to summarize data. It also defines the criteria that will determine whether or not the
final implementation meets the requirements of the specification.
It is crucial that the system specification clearly delineates the bounds and characteristics
of the system, and that it represents a concise agreement or contract between the system
developers and system users. An incomplete or poorly documented specification
invariably leads to a frustrating and endless implementation.
8.2.2 Design Phase

System design determines what data processing tools will be used or created to meet the
system specification criteria. For example, hardware, programming languages and data
base packages are selected. The project is broken into programs, the interaction between
programs is defined, and the programs are broken into high-level modules. Rarely (and in
poorly managed projects) is any coding performed during this step. In object-oriented
systems, the major system objects are modeled; that is, a definition of the data contained
in an object, and the interface to the object is determined.
Also during this step, project management documents the coding, documentation and
implementation standards that system construction must follow. A code management
system and quality assurance tools are selected, and guidelines for their use are
documented in the implementation standard. Project personnel are designated to be
responsible for enforcing conformance to the standards.
8.2.3 Implementation Phase

It is during implementation that actual construction of the system begins. The high level
modules identified during system design are broken into submodules, and utilities
modules are defined. Following the established coding standard, functions within
modules are coded according to system design and assembled into individual programs.
8.2.4 Acceptance Testing Phase

As we’ll see shortly, testing can be viewed from a variety of perspectives. In this context,
we refer to the functional testing that determines whether the requirements of the system
are met. Tests are tied directly to that portion of the system specification that documents
the requirements that the system is intended to satisfy. When problems are uncovered,
they are reported to the development organization for resolution.
It is important during this phase of the process that testing is carefully conducted and
problems are carefully monitored. On all projects, test cases must be carefully
documented, and agreed to by both the developers and the users. A test case management
system can be extremely helpful in documenting test cases, and establishing
dependencies between test cases. On all projects but the smallest, a defect tracking
system should be used to track flaws and flaw resolution.

It is best if the implementation organization is not responsible for this phase of the system
life cycle. If the end user is not sufficiently technical, or otherwise unable to do the
testing, a third party should be employed.
8.2.5 Maintenance
This part of the system life cycle entails fixing flaws that are discovered in the system
after deployment. Normally, fixes to the software aren’t distributed before the next
scheduled release of the system. Occasionally, fixes, or field releases, are dispatched
immediately. Field releases are expensive and dangerous; if a thorough job is done during
the specification and testing phases of the system life cycle, they will not be necessary.
Note that maintenance activity often takes place in parallel with development of a new
system release. In this case, changes made by maintenance must also be integrated with
the new system code.
8.3 Testing
The concept of testing goes far beyond the acceptance testing described in Section 8.2.4.
Testing begins with the start of the system specification, and embraces every stage of the
system life cycle, as discussed below.
8.3.1 Testing at the System Specification Level

Testing at the specification level mainly involves high-level documentation of the
acceptance testing procedure, and the needs of the test team at the time of acceptance
testing. A member of the test team should participate in all requirements definition
activities. At this stage, the test team is responsible for pointing out when requirements
conflict, or when requirements are essentially untestable. For example, a requirement that
a system run without failing for five years is unrealistic, because it would take a
minimum of five years to determine whether the requirement is met.
Immediately upon completion of the system specification, the test team should begin
documenting specific test cases to be executed during acceptance testing.
8.3.2 Testing at the Design Level

In many ways, the design phase is to the implementation team what the specification
phase is to the design team. Testing during this phase mainly entails documenting the
acceptance criteria for the programs and modules that are its output. As a requirement of
a program or module is tentatively defined, the question should be asked, Is the
requirement testable? If the answer is no, the requirement should be modified.
Occasionally at this time a prototype representing some small portion of the system may
be developed in order to prove that a part of the design is feasible. Such a prototype is
sometimes called a proof of concept prototype.
Upon completion of the design phase, the designers should have a high-level plan for
determining whether the implementation conforms to the design, and should begin
developing specific test cases for proving this.

Also during this time the test team should develop both high-level and detailed plans for
conducting acceptance testing. Often acceptance testing will require specialized tools that
must be either purchased or developed. If possible, the testers should coordinate the
acquisition and use of such tools with the developers. The developers may find that these
same tools are helpful in their own testing efforts, and the testers may request that the
developers design their code in such a way as to make deployment of the tools more
effective.
8.3.3 Testing at the Implementation Level

Testing during this phase is often called unit testing. This is a very broad term that can
apply to any of the following types of testing.
Algorithm Testing
This kind of testing entails proving that an algorithm is sound prior to coding
it in a function or module. The proof may be mathematical in nature, or it may
involve coding the algorithm into a test driver that will exercise it in detail.
Function Testing
This kind of testing will prove that an individual function correctly validates
its input, and produces the correct output. Sometimes a function is tested at
the same time a module is tested; a symbolic debugger can be a valuable aid,
if this is the case. Sometimes the function will first be coded into a test driver,
which will exercise it independently of the module.
Module Testing
This kind of testing proves that a module can correctly do its job. Exercising
the program in which the module resides may test the module, but it can be
more valuable to code the module into a test driver, first.
Program Testing
This kind of testing proves that a program correctly validates its input, and
produces the expected output.
In three items out of the above list, we discussed using test drivers to perform testing.
This is also called bench testing. One of the advantages of bench testing is that it isolates
a coding unit before testing it. To use an automotive example, suppose your car won’t
start, and you suspect the starter motor. It would be hard to test your theory while the
starter was still in the car, because the alternator, battery, cables, or a variety of other
things might be the source of the problem. So you take the starter out of the car, put in on
a bench, and hook it up to a battery and cables that you know are working correctly. The
battery and cables on the bench are directly analogous to a test driver.
Another advantage of bench testing is that it can more thoroughly be used to test a system
component. Say you want to test a function’s response to all possible illegal input. This
would be difficult to do within the module and program that owns the function, because
they are designed to always provide the function with valid data. A test driver, on the
other hand, can be specially designed to feed the function any possible data, including
invalid input.

In order to be able to do bench testing, the component that you want to test has to be
easily removed from the program, module or function within which it resides. It is always
a good idea to keep bench testing in mind as you design your code. In fact, it can help to
keep this in mind during the system design phase, and sometimes even during the system
specification phase.
Finally, an important part of the developer’s job is to design and carefully document test
cases that can be used to prove the proper execution of each module, and to assemble
these test cases into a module test plan. The module test plan can be used throughout
development to verify the module; later, the maintenance group will use it to verify
changes that they make to your module.
8.3.4 Testing at the Acceptance Testing Level

The activities that we discussed in Section 8.2.4 can be broken down into three types of
testing:
Functional testing is the execution specific test cases to prove a system
requirement.
Retesting is functional testing performed to prove a system requirement that was
previously found to be unmet.
Regression testing takes place after any change is made to the system. It is
functional testing that proves that requirements that were met prior to the change
continue to be met.
Regression testing may be the most difficult of the three activities. A regression is a flaw
that is introduced when another flaw is corrected. The regression might occur in the
neighborhood of the flaw that was fixed, but it often occurs in a separate module, or even
a separate program (the possibility of introducing a regression outside of the module is
greatly reduced by doing a good job designing modules with hidden implementations,
and well defined APIs). When the development organization fixes a flaw uncovered by
the test team, the test team must re-execute the test cases associated with the flaw. In
addition, they must subject large portions of the system to regression testing. Ideally,
every test case that had formerly passed will be re-executed. Since this is not always be
feasible, the test team will often choose not to re-execute test cases judged not to be at all
related to the flaw; for example, if the flaw that was fixed was a spelling error in a screen
associated with the inventory data base, the test team might choose not to re-execute test
cases against reports from the receivables data base.
8.3.5 Testing at the Maintenance Level

Testing at the maintenance level is a combination of testing strategies at the
implementation and acceptance testing levels. A maintenance programmer who fixes a
flaw will typically unit test the solution, then probably regression test the module it
resides in. The maintenance group as a whole will (if they’re doing their job) regression
test the entire system on a weekly or monthly basis.
In order for the maintenance group to be able to do their job well, the development group
must do a careful, thorough job of documenting their module test plans during
implementation. Regression testing tools can be an invaluable aid in automating

regression tests, and improving the overall quality of the system. If specific regression
testing tools are to be used, they should be identified during the design phase; test cases
devised during implementation should be designed as much as possible to work with the
selected tools.

9. Binary Trees
The next data structure we’re going to examine is the tree. A tree is a versatile
mechanism for solving many problems, including parsing and fast indexing. In this
section we will define in detail the implementation of a special kind of tree, the binary
tree. In the next section, we will learn how to use a binary tree to create a more general
kind of structure called an n-ary tree.
9.1 Objectives
• Provide a formal definition of a binary tree;
• Construct a binary tree as an array or as a linked structure; and
• Use a binary tree as an index.
9.2 Overview
Conceptually, a binary tree is a structure that consists of zero or more linked nodes. In a
minimal implementation, a node consists of exactly three elements: a data pointer, a left
pointer and a right pointer. This concept is illustrated in Figure 9-1.
data
left
right
data data
left left
right right
data data data data

left left left left
right right right right
data data data data data

left left left left left
right right right right right
Figure 9-1 A Binary Tree
With respect to a binary tree we can define the following components and concepts (refer
to Figure 9-2 and Figure 9-3):
Node:
The basic component of a binary tree, comprising a left pointer, a right pointer
and data.
Root Node:
The first node in the tree.
Section 9: Binary Trees 113 06/19/06

Child:
For a given node, another node that depends from the original node’s left or
right pointer. The child depending from the left pointer is referred to as the left
child and the child depending from the right pointer is referred to as the right
child. Any node in the tree may have zero, one or two children.
Root Node
data Node A Level 1
left
right
data Node B data Node C Level 2

left left
Left child of B
right right
Parent of E
Leaf Node
data Node D data Node E Level 3

left left
right right
Right child of B
Figure 9-2 Binary Tree Concepts
Leaf Node:
A node with zero children.
Parent:
The node from which a given node depends. The root node of a binary tree
has no parent; every other node has exactly one parent.
Empty Tree:
For our purposes, a binary tree will be called empty if it has no root node.
(Note that in some implementations an empty tree might be a binary tree with
only a root node.)
Level:
A relative position in the binary tree. The root node occupies level 1 the
children of the root occupy level 2 and so on. In general, the children of a
node that occupies level n themselves occupy level n + 1. The maximum
number of nodes that can occupy level N is 2(N– 1).
Depth:
The maximum number of levels in a binary tree. This value determines the
maximum capacity of a binary tree. If a binary tree has a depth of N, it can
contain a maximum of 2N – 1 nodes.
Distance Between Two Nodes:
The distance between two nodes is the number of levels that must be crossed
to traverse from one to the other. In Figure 9-2, the distance between nodes A
and E is 2.

Balanced Binary Tree:

A binary tree in which the distance between the root node and any leaf differs
by no more than one.
data data
left left
right right
data data data data

left left left left
data data data data data

left left left left left
right right right right right
data data data data

left left left left
Balanced Binary Tree Unbalanced Binary Tree

Figure 9-3 Balanced Binary Trees
Subtree:
Within a binary tree, a subtree is any node plus all of its descendants. The top
node is called the root of the subtree. In this way a binary tree is said to be
recursively defined.
9.3 Binary Tree Representation

A binary tree is most often represented in one of two ways:
• As a contiguous array
• As a dynamically linked structure
These two representations will be examined, below.
9.3.1 Contiguous Array Representation

A binary tree can be represented as an array only if the maximum depth of the tree is
known in advance. The size of the array is then the maximum number of nodes in the
array. The root node occupies the array element at index 0; the children of the root
occupy the elements at indices 1 and 2. In general, the left and right children of the node
at index n occupy array elements 2 * n + 1 and 2 * n + 2, respectively. The parent of a
node at index m occupies the array element at index (m - 1) / 2. Figure 9-4 illustrates
how a binary tree of depth 4 can be represented by an array.

9.3.2 Dynamically Linked Representation

Representing a binary tree as an array has some important applications, particularly in the
area of persistent storage. However for large binary trees it has two disadvantages:
• If the binary tree is not full, as is usually the case, the array will contain a lot of
wasted space.
• Once the tree becomes full, it cannot be extended without allocating a new array.
It is often preferable to dynamically allocate each node as it is required, and to keep
pointers to the node’s children within the node. This arrangement creates a tree more like
that shown above, in Figure 9-1.
A
B C
D E F G
H I J K L M N O
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14
A B C D E F G H I J K L M N O
#define MAX_NODES( depth ) ((int)pow( 2, (depth) ))

#define LEFT( inx ) (2 * (inx) + 1)
#define RIGHT( inx ) (2 * (inx) + 2)
#define PARENT( inx ) ((inx - 1) / 2)
Figure 9-4 Representing a Binary Tree as an Array
9.4 A Minimal Binary Tree Implementation

Next we would like to examine a binary tree module implemented using a linked
representation. Each node in the tree will be allocated as needed, and will consist of
pointers for user data, left child and right child. The following operations will be defined:
• Create a binary tree
• Add a root node to a tree
• Add a left child to a node
• Add a right child to a node
• Get a tree’s root node
• Get the data associated with a node
• Get a node’s left child

• Get a node’s right child

• Inquire whether a tree is empty
• Inquire whether a node is a leaf
• Traverse a binary tree
• Destroy a subtree
• Destroy a binary tree
9.4.1 Public Declarations

Our implementation will, of course, be modular. The name of our module will be
BTREE. The two principal data types declared in our public header file will be
incomplete data types that serve to identify a tree, and a node in a tree. There are two
macros that can be used to represent a NULL tree ID, and a NULL node ID. And there
are two callback proc declarations, one for destroying a tree or subtree, the other for
traversing, or visiting a tree. There is also a declaration for a traverse order type that will
be discussed later. All the public data type declarations are shown in Figure 9-5.
#define BTREE_NULL_ID (NULL)
#define BTREE_NULL_NODE_ID (NULL)
typedef void BTREE_DESTROY_PROC_t( void *data );

typedef BTREE_DESTROY_PROC_t *BTREE_DESTROY_PROC_p_t;
typedef void BTREE_VISIT_PROC_t( void *data );

typedef BTREE_VISIT_PROC_t *BTREE_VISIT_PROC_p_t;
typedef struct btree__control_s *BTREE_ID_t;

typedef struct btree__node_s *BTREE_NODE_ID_t;
typedef enum btree_traverse_order_e

{
BTREE_PREORDER,
BTREE_INORDER,
BTREE_POSTORDER
} BTREE_TRAVERSE_ORDER_e_t;
Figure 9-5 BTREE Module Public Data Types
9.4.2 Private Declarations

As shown in Figure 9-6, our implementation will declare two private data structures: a
control structure and a node structure. The address of the control structure will be the ID
of the tree; the address of a node will be that node’s ID. The control structure contains
nothing more than a pointer to the root node. A node structure contains pointers for its
left and right children, and for the user’s data. We have also chosen to add pointers for
the node’s parent, and a pointer back to control structure in which the node resides;
strictly speaking these two extra pointers are not required, however experience shows that
they may be helpful in managing the implementation.
#ifndef BTREEP_H
#define BTREEP_H

#include <btree.h>
typedef struct btree__node_s *BTREE__NODE_p_t;

typedef struct btree__control_s *BTREE__CONTROL_p_t;
typedef struct btree__node_s

{
void *data;
BTREE__CONTROL_p_t tree;
BTREE__NODE_p_t parent;
BTREE__NODE_p_t left;
BTREE__NODE_p_t right;
} BTREE__NODE_t;
typedef struct btree__control_s

{
BTREE__NODE_p_t root;
} BTREE__CONTROL_t;
#endif
Figure 9-6 BTREE Private Header File
9.4.3 BTREE_create_tree
This method will create an empty binary tree and return its ID.
Synopsis:
BTREE_ID_t BTREE_create_tree( void );
Returns:
Binary tree ID
Exceptions:
Throws SIGABRT if tree cannot be created
Notes:
None
Here is the code for BTREE_create_tree:
BTREE_ID_t BTREE_create_tree( void )
{
BTREE__CONTROL_p_t tree = CDA_NEW( BTREE__CONTROL_t );
tree->root = NULL;
return tree;
}
9.4.4 BTREE_add_root
This method will add the root node to a tree; the tree must not already have a root node.
Synopsis:
BTREE_NODE_ID_t
BTREE_add_root( BTREE_ID_t tree, void *data );
Where:
tree == tree to which to add
data -> user data to be stored with the node

Exceptions:
Throws SIGABRT if node can’t be created
Notes:
Tree must be empty when this routine is called
Here is the code for this method:
BTREE_NODE_ID_t BTREE_add_root( BTREE_ID_t tree, void *data )
{
BTREE__NODE_p_t node = CDA_NEW( BTREE__NODE_t );
CDA_ASSERT( tree->root == NULL );
node->data = data;
node->tree = tree;
node->parent = NULL;
node->left = NULL;
node->right = NULL;
tree->root = node;
return node;
}
9.4.5 BTREE_add_left
This method will add a left child to a node; the node must not already have a left child.
Synopsis:
BTREE_NODE_ID_t
BTREE_add_left( BTREE_NODE_ID_t node, void *data );
Where:
node == id of node to which to add
data -> user data to be associated with this node
Returns:
ID of new node
Exceptions:
Throws SIGABRT if node cannot be created
Notes:
The node must not already have a left child
BTREE_NODE_ID_t BTREE_add_left( BTREE_NODE_ID_t node, void *data )
{
BTREE__NODE_p_t left = CDA_NEW( BTREE__NODE_t );
CDA_ASSERT( node->left == NULL );
left->data = data;
left->tree = node->tree;
left->parent = node;
left->left = NULL;
left->right = NULL;
node->left = left;
return left;
}

9.4.6 BTREE_add_right
This method will add a right child to a node; the node must not already have a right child.
Synopsis:
BTREE_NODE_ID_t
BTREE_add_right( BTREE_NODE_ID_t node, void *data )
Where:
node == id of node to which to add
data -> user data to be associated with this node
Returns:
ID of new node
Exceptions:
Throws SIGABRT if node cannot be created
Notes:
The node must not already have a right child
The code for this method is left as an exercise to the student.
9.4.7 BTREE_get_root
This method returns the root node of a tree.
Synopsis:
BTREE_NODE_ID_t BTREE_get_root( BTREE_ID_t tree );
Where:
tree == ID of tree to interrogate
Returns:
ID of root node; BTREE_NODE_NULL if tree is empty
Exceptions:
None
Notes:
None
BTREE_NODE_ID_t BTREE_get_root( BTREE_ID_t tree )
{
return tree->root;
}
9.4.8 BTREE_get_data, BTREE_get_left, BTREE_get_right

These three methods obtain the data, and left and right children, respectively, associated
with a node.
Synopsis:
void *BTREE_get_data( BTREE_NODE_ID_t node );
BTREE_NODE_ID_t BTREE_get_left( BTREE_NODE_ID_t node );
BTREE_NODE_ID_t BTREE_get_right( BTREE_NODE_ID_t node );
Where:
node == id of node to interrogate
Returns:
The data, left or right child associated with the node;
BTREE_get_left and BTREE_get_right return BTREE_NULL_NODE
if there is no relevant child.

Exceptions:
None
Notes:
None
Here is the code for these methods:
void *BTREE_get_data( BTREE_NODE_ID_t node )
{
return node->data;
}
BTREE_NODE_ID_t BTREE_get_left( BTREE_NODE_ID_t node )

{
return node->left;
}
BTREE_NODE_ID_t BTREE_get_right( BTREE_NODE_ID_t node )

{
return node->right;
}
9.4.9 BTREE_is_empty
This method returns true if a tree is empty. Recall that, according to our definition, a tree
is empty if it has no root node.
Synopsis:
CDA_BOOL_t BTREE_is_empty( BTREE_ID_t tree );
Where:
tree == id of tree to test
Returns:
CDA_TRUE if tree is empty,
CDA_FALSE otherwise
Exceptions:
None
Notes:
None
9.4.10 BTREE_is_leaf
This method returns true if a node is a leaf. Recall that, according to our definition, a
node is a leaf if it has no children.
Synopsis:
CDA_BOOL_t BTREE_is_leaf( BTREE_NODE_ID_t node );
Where:
node == id of node to test
Returns:
CDA_TRUE if node is a leaf,
CDA_FALSE otherwise
Exceptions:
None

Notes:
None
9.4.11 BTREE_traverse_tree
A binary tree is traversed by examining, or visiting, every node in the tree in some order.
In our implementation, there are three possible orders of traversal:
• Inorder traversal
• Preorder traversal
• Postorder traversal
A fourth possible traversal, level traversal, is discussed in your textbook. The details of
binary tree traversal will be discussed later. Recall that BTREE_VISIT_PROC_p_t is
declared as follows:
typedef void BTREE_VISIT_PROC_t( void *data );
typedef BTREE_VISIT_PROC_t *BTREE_VISIT_PROC_p_t;
Synopsis:
BTREE_ID_t
BTREE_traverse_tree( BTREE_ID_t tree,
BTREE_TRAVERSE_ORDER_e_t order,
BTREE_VISIT_PROC_p_t visit_proc
);
Where:
tree == id of tree to traverse
order == order in which to traverse tree
visit_proc -> user proc to call each time a node is visited
Returns:
tree
Exceptions:
None
Notes:
1. Order may be BTREE_INORDER, BTREE_PREORDER or
BTREE_POSTORDER.
2. Each time visit_proc is called the data associated with the
node is passed.
9.4.12 BTREE_destroy_tree, BTREE_destroy_subtree

These methods will destroy a tree or subtree, respectively. Each node in the tree or
subtree is visited and freed. Prior to destroying a node, the caller is given the opportunity
to free resources associated with a node’s data via an optional destroy procedure. Recall
that BTREE_DESTROY_PROC_p_t is declared as follows:
typedef void BTREE_DESTROY_PROC_t( void *data );
typedef BTREE_DESTROY_PROC_t *BTREE_DESTROY_PROC_p_t;
Synopsis:
BTREE_NODE_ID_t
BTREE_destroy_subtree( BTREE_NODE_ID_t node,
BTREE_DESTROY_PROC_p_t destroy_proc
);

BTREE_ID_t
BTREE_destroy_tree( BTREE_ID_t tree,
);
Where:
tree == id of tree to destroy
node == root of subtree to destroy
destroy_proc -> proc to call prior to node destruction
Returns:
BTREE_destroy_subtree returns BTREE_NULL_NODE_ID
BREE_destroy_tree returns BTREE_NULL_ID
Exceptions:
None
Notes:
NULL may be passed in place of destroy_proc
The method to destroy a subtree is a recursive procedure that destroys all left children,
followed by all right children, followed by the node itself. Note that the parent of the
node must be updated, and a special check must be made to determine whether the root of
the tree has been destroyed. Here is the code for the method:
BTREE_NODE_ID_t
BTREE_destroy_subtree( BTREE_NODE_ID_t node,
)
{
if( node->left != NULL )
BTREE_destroy_subtree( node->left, destroy_proc );
if( node->right != NULL )

BTREE_destroy_subtree( node->right, destroy_proc );
if( destroy_proc != NULL )

destroy_proc( node->data );
if( node == node->tree->root )

node->tree->root = NULL;
else if ( node == node->parent->left )
node->parent->left = NULL;
else if ( node == node->parent->right )
node->parent->right = NULL;
else
CDA_free( node );
return BTREE_NULL_NODE_ID;
}
The method to destroy a tree merely destroys the subtree represented by the root, and
then frees the control structure:
BTREE_ID_t
BTREE_destroy_tree( BTREE_ID_t tree,
)
{

BTREE_destroy_subtree( tree->root, destroy_proc );
CDA_free( tree );
return BTREE_NULL_ID;
}
9.5 Using a Binary Tree as an Index

data
left
{
right key = abbey
position = 2
}
0 1 2 3
Data washington lincoln abbey shmabbey
Array (other data) (other data) (other data) (other data)
Figure 9-7 Indexing an Array with a Binary Tree
One of the more straightforward uses of a binary tree is as a sorted index to a data
structure. Suppose we have a list of records stored in an array, and each record is
identified by a unique name called a key. An example would be an array of student
records keyed by student name. Each record can be quickly found by using its array
index, but you typically want to access the record for a particular name. To be able to
quickly look up a record by name, we could create a lookup index for the array in the
form of a binary tree. Each node in the binary tree would contain the name of a record in
the array, plus the array index of the record. This arrangement is illustrated in Figure 9-7.
typedef struct data_s
{
char *key;
int position;
} DATA_t, *DATA_p_t;
char *recs[] = { . . . }
if ( num-recs > 0 )
tree = BTREE_create_tree()
data = CDA_NEW( DATA_t )
data->key = CDA_NEW_STR( recs[0] )
data->position = 0
root = BTREE_add_root( tree, data )
for ( inx = 1 ; inx < num-recs ; ++inx )
data = CDA_NEW( DATA_t )
data->key = CDA_NEW_STR( recs[inx] )

data->position = inx
add( data, root )
Figure 9-8 Pseudocode for Start of Index Creation
Here is how we go about building the lookup index (refer to Figure 9-8). First we declare
a key record, DATA_t, capable of holding a key and an array index. Now for each record
in the array, allocate and add a new key record to a binary tree. The key record for array
index 0 will be stored in the root of the tree. Additional key records will be added to the
tree by descending through the tree comparing the key to be added to the key stored in an
existing node. Each time the new key is found to be less than an existing key, descend to
the left; if the new key is greater than the existing key descend to the right. When you
attempt to descend and find that the node you are descending from does not have the
target child you’re done descending, and add the key as the target child.
add( DATA_p_t data, BTREE_NODE_t node )
node_data = BTREE_get_data( node )
strc = strcmp( data->key, node_data->key )
CDA_ASSERT( strc != 0 )
if ( strc < 0 )
next_node = BTREE_get_left( node )
if ( next_node == BTREE_NULL_NODE )
BTREE_add_left( node, data )
else
add( data, next_node )
else
next_node = BTREE_get_right( node )
BTREE_add_right( node, data )
else
add( data, next_node )
Figure 9-9 Index Creation Add Function Pseudocode
Figure 9-9 shows the pseudocode for a recursive function to accomplish the add
operation. Figure 9-10 shows the result of using the recursive function to create an index
using the following array of keys:
static const char *test_data[] =
{ "washington", "lincoln", "abbey", "shmabbey",
"gettysburg", "alabama", "zulu", "yorkshire",
"xerxes", "wonderland", "tiparary", "sayville",
"montana", "eratosthenes", "pythagoras", "aristotle",
"kant"
};

WA
LI ZU
AB SH YO
GE SA TI XE
AL KA MO WO
ER PY
AR
Figure 9-10 A Binary Tree Index
Now suppose you have a key, and you want to search the binary tree for it in order to
determine the array position that the associated record occupies. You would follow a
procedure similar to the one used to create the tree. Starting with the root, compare the
key to the key stored in the node; if they are equal you’re done, otherwise recursively
branch left or right until you find the key you’re looking for, or you reach a leaf; if you
reach a leaf without finding the target key, it means the key isn’t in the binary tree. The
maximum number of comparisons that you have to perform is equal to the depth of the
tree. The pseudocode for the search function is shown in Figure 9-11. Section 9.6
contains sample code that demonstrates the creation and validation of an index; the
results of the validation are shown in Figure 9-13.
Using a binary tree as an index can be a valuable tool, but before doing so you must
analyze your data carefully. The index will be most efficient if keys tend to occur
randomly; if they tend to occur in order, you will wind up with a very inefficient index
like that shown in Figure 9-12. In cases where you must make certain your index is
efficient you will have to resort to balancing your tree; this subject is discussed in your
textbook.

root = BTREE_get_root( tree )
inx = get_position( key, root )
. . .
/* Return array position of key, or -1 if key not found */
int get_position( char *key, BTREE_NODE_t node )
data = BTREE_get_data( node )
strc = strcmp( key, data->key )
if ( strc == 0 )
rcode = data->position
else if ( strc < 0 )
next_node = BTREE_get_left( node )
rcode = -1
else
rcode = get_position( key, next_node )
else if ( strc > 0 )
next_node = BTREE_get_right( node )
if ( next_node = BTREE_NULL_NODE )
rcode = -1
else
rcode = get_position( key, next_node )
return rcode
Figure 9-11 Pseudocode for Searching a Binary Tree Index
Alabama
Alaska
Arizona
California
Colorado
Connecticut
Delaware
Figure 9-12 An Inefficient Binary Tree Index
9.6 Using a Binary Tree as an Index – Demonstration

#include <stdio.h>
#include <string.h>
#include <btree.h>

typedef struct data_s

{
char *key;
int position;
} DATA_t, *DATA_p_t;
static void create_tree( const char **recs, int num_recs );

static void add( DATA_p_t data, BTREE_NODE_ID_t node );
static int get_from_btree( const char *key );
static int get_pos( const char *key, BTREE_NODE_ID_t node );
static const char *test_data[] =

{ "washington", "lincoln", "abbey", "shmabbey",
"gettysburg", "alabama", "zulu", "yorkshire",
"xerxes", "wonderland", "tiparary", "sayville",
"montana", "eratosthenes", "pythagoras", "aristotle",
"kant"
};
static BTREE_ID_t tree = BTREE_NULL_ID;
int main( int argc, char **argv )

{
int inx = 0;
int jnx = 0;
create_tree( test_data, CDA_CARD( test_data ) );

for ( inx = 0 ; inx < CDA_CARD( test_data ) ; ++inx )
{
jnx = get_from_btree( test_data[inx] );
if ( inx != jnx )
printf( "%3d !!!", jnx );
printf( "%3d, %s\n", inx, test_data[inx] );
}
return EXIT_SUCCESS;
}
static void create_tree( const char **recs, int num_recs )

{
DATA_p_t data = CDA_NEW( DATA_t );
BTREE_NODE_ID_t root = BTREE_NULL_NODE_ID;
int inx = 0;
if ( num_recs > 0 )
{
data->key = CDA_NEW_STR( recs[0] );
data->position = 0;
tree = BTREE_create_tree();
root = BTREE_add_root( tree, data );
for ( inx = 1 ; inx < num_recs ; ++inx )

{
data = CDA_NEW( DATA_t );
data->key = CDA_NEW_STR( recs[inx] );
data->position = inx;

add( data, root );
}
}
}
static void add( DATA_p_t data, BTREE_NODE_ID_t node )

{
DATA_p_t node_data = BTREE_get_data( node );
BTREE_NODE_ID_t next_node = BTREE_NULL_NODE_ID;
int strc = 0;
strc = strcmp( data->key, node_data->key );

CDA_ASSERT( strc != 0 );
if ( strc < 0 )
{
next_node = BTREE_get_left( node );
if ( next_node == BTREE_NULL_NODE_ID )
BTREE_add_left( node, data );
else
add( data, next_node );
}
else
{
next_node = BTREE_get_right( node );
BTREE_add_right( node, data );
else
add( data, next_node );
}
}
static int get_from_btree( const char *key )

{
BTREE_NODE_ID_t root = BTREE_get_root( tree );
int rcode = 0;
rcode = get_pos( key, root );

return rcode;
}
static int get_pos( const char *key, BTREE_NODE_ID_t node )

{
DATA_p_t node_data = BTREE_get_data( node );
BTREE_NODE_ID_t next_node = BTREE_NULL_NODE_ID;
int strc = 0;
int rcode = -1;
strc = strcmp( key, node_data->key );

if ( strc == 0 )
rcode = node_data->position;
else if ( strc < 0 )
{
next_node = BTREE_get_left( node );
;
else

rcode = get_pos( key, next_node );
}
else
{
next_node = BTREE_get_right( node );
;
else
rcode = get_pos( key, next_node );
}
return rcode;
}
0, washington
1, lincoln
2, abbey
3, shmabbey
4, gettysburg
5, alabama
6, zulu
7, yorkshire
8, xerxes
9, wonderland
10, tiparary
11, sayville
12, montana
13, eratosthenes
14, pythagoras
15, aristotle
16, kant
Figure 9-13 Index Validation Results
9.7 Traversing a Binary Tree

As mentioned earlier, you traverse a binary tree by visiting each node in the tree in a
designated order; in our implementation, the three possible orders are inorder, preorder
and postorder. Our implementation has one public method, BTREE_traverse_tree that
handles all three cases; it looks like this:
BTREE_ID_t
BTREE_traverse_tree( BTREE_ID_t tree,
BTREE_TRAVERSE_ORDER_e_t order,
BTREE_VISIT_PROC_p_t visit_proc
)
{
switch ( order )
{
case BTREE_PREORDER:
traverse_preorder( tree->root, visit_proc );
break;
case BTREE_INORDER:
traverse_inorder( tree->root, visit_proc );
break;
case BTREE_POSTORDER:

traverse_postorder( tree->root, visit_proc );
break;
default:
break;
}
return tree;
}
Each of the three subroutines will be a recursive procedure that traverses the tree in the
indicated order, and calls visit_proc each time a node is visited, passing the node’s data.
The cases are discussed individually, below.
9.7.1 Inorder Traversal

Inorder traversal involves following a tree to the left until a null child is reached, at which
time the parent is visited. Traversal continues recursively with the left branch of the
parent’s right child. The pseudocode for this procedure is shown in Figure 9-14.
void inorder( BTREE__NODE_p_t node )
if ( node != NULL )
inorder( node->left )
visit( node )
inorder( node->right )
Figure 9-14 Inorder Traversal Pseudocode
Here is a code sample that uses inorder traversal to traverse the binary tree index created
in Section 9.6. The results of the traversal are shown in Figure 9-15.
static BTREE_VISIT_PROC_t visit_proc;
static void traverse_index_inorder( BTREE_t tree )
{
BTREE_traverse_tree( tree, BTREE_INORDER, visit_proc );
}
static void visit_proc( void *visit_data )

{
DATA_p_t data = visit_data;
printf( "%3d, %s\n", data->position, data->key );

}
9.7.2 Preorder Traversal

In a preorder traversal, a node is visited, and then all nodes along the node’s left branch
are recursively visited. When a null node is encountered, the nodes along the left branch
of the parent’s right child are recursively visited. The pseudocode for a preorder traversal
is shown in Figure 9-16.
2, abbey
5, alabama
15, aristotle
13, eratosthenes
4, gettysburg

16, kant
1, lincoln
12, montana
14, pythagoras
11, sayville
3, shmabbey
10, tiparary
0, washington
9, wonderland
8, xerxes
7, yorkshire
6, zulu
Figure 9-15 Inorder Traversal Results
void preorder( BTREE__NODE_p_t node )

if ( node != NULL )
visit( node )
preorder( node->left )
preorder( node->right )
Figure 9-16 Preorder Traversal Pseudocode
If we substitute a preorder traversal for the inorder traversal in the example in Section
9.7.1 we will get the results shown in Figure 9-17. Of course, this isn’t a very good
example of the use of a preorder traversal. A better example would be the use of binary
trees to parse prefix expressions; this subject is discussed in your textbook.
0, washington
1, lincoln
2, abbey
4, gettysburg
5, alabama
13, eratosthenes
15, aristotle
16, kant
3, shmabbey
11, sayville
12, montana
14, pythagoras
10, tiparary
6, zulu
7, yorkshire
8, xerxes
9, wonderland
Figure 9-17 Preorder Traversal Results
9.7.3 Postorder Traversal

In a postorder traversal, all the children of a node are visited before the node is visited.
The pseudocode for the postorder traversal procedure is shown in Figure 9-19.
void postorder( BTREE__NODE_p_t node )
if ( node != NULL )
postorder( node->left )
postorder( node->right )
visit( node )
Figure 9-18 Postorder Traversal Pseudocode

If we substitute a postorder traversal for the inorder traversal in the example in Section
9.7.1 we will get the results shown in Figure 9-17. Once again, this isn’t a very good
example of the use of this type of traversal. A better example would be the use of binary
trees to parse postfix expressions as discussed in your textbook.
15, aristotle
13, eratosthenes
5, alabama
16, kant
4, gettysburg
2, abbey
14, pythagoras
12, montana
11, sayville
10, tiparary
3, shmabbey
1, lincoln
9, wonderland
8, xerxes
7, yorkshire
6, zulu
0, washington
Figure 9-19 Postorder Traversal Results

10. N-ary Trees

In this section we will discuss the n-ary tree structure. Like a binary tree, an n-ary tree is
a collection of linked nodes beginning with a root; however, where each node in a binary
tree can have up to two children, each node in an n-ary tree can have an unlimited
number of children. Interestingly this would seem to suggest that an n-ary tree has a more
extensive structure than a binary tree, however n-ary trees are typically constructed using
binary trees.
10.1 Objectives
• Define an n-ary tree, and explain the relationships between nodes in an n-ary tree;
• Perform a modularized n-ary tree implementation using a binary tree; and
• Use an n-ary tree to construct a directory.
10.2 Overview
Figure 10-1 shows one way to view an n-ary tree, suggesting that each node contains a
link to each of zero or more children. Note that not every node has to have the same
number of children. One way to build a tree like this would be to keep an array of child
pointers in each node. However this approach has one big disadvantage: arrays are static
entities. Each node would have a maximum number of children, and no matter how many
children a node actually has, it would be forced to allocate an array big enough to hold
the maximum.
Figure 10-1 N-ary Tree: Multilink View
Another approach to constructing an n-ary tree is to use linked lists to link all the children
of a node. The most popular is approach is to store an n-ary tree as a binary tree using the
left child, right sibling method. This method is illustrated in Figure 10-2. It shows that
Section 10: N-ary Trees 135 06/19/06

the right child of node N in the binary tree is treated as a sibling of node N in the n-ary
tree. The right child of node N in the binary tree is referred to as node N’s nearest sibling,
and the right child of any sibling of N is node N’s sibling. The left child of node M in the
binary tree is referred to as node M’s nearest child, and any sibling of a child of M is also
a child of M. In considering the relationships between nodes in an N-ary tree we have one
situation that is, at least at first, anti-intuitive. Suppose that, in a binary tree, B is the right
child of A. Then in the corresponding n-ary tree, B is a sibling of A, but A is not a sibling
of B.
A
B C D
E F G
Figure 10-2 N-ary Tree: Binary Tree View
10.3 A Small N-ary Tree Implementation

Let’s look at a modular n-ary tree implementation. It will be similar to our binary tree
implementation, and will incorporate methods to perform the following operations:
• Create an n-ary tree.
• Add a root node to an n-ary tree.
• Add a child to a node.
• Add a sibling to a node.
• Get the root node of an n-ary tree.
• Determine whether a node has a child.
• Determine whether a node has a sibling.
• Get the data associated with a node.
• Get the nearest child of a node.
• Get the nearest sibling of a node.
• Destroy an n-ary tree.

10.3.1 Public Data Types

Our module name will be NTREE, and our public declarations will rely heavily on the
declarations of our BTREE module, which makes sense because our n-ary trees will be
binary trees; they are shown in Figure 10-3. We will declare incomplete types to
represent the ID of a tree and a node, and we will use macros to define values to represent
a NULL tree and a NULL node. We will also need a destroy callback type.
#include <btree.h>
#define NTREE_NULL_ID (BTREE_NULL_ID)

#define NTREE_NULL_NODE_ID (BTREE_NULL_NODE_ID)
typedef BTREE_DESTROY_PROC_t NTREE_DESTROY_PROC_t;

typedef BTREE_DESTROY_PROC_p_t NTREE_DESTROY_PROC_p_t;
typedef BTREE_ID_t NTREE_ID_t;

typedef BTREE_NODE_ID_t NTREE_NODE_ID_t;
Figure 10-3 NTREE Module Public Declarations
10.3.2 Private Declarations

Our implementation has no private declarations of consequence; everything is borrowed
from the BTREE module. The complete private header file is shown in Figure 10-4.
#ifndef NTREEP_H
#define NTREEP_H
#include <ntree.h>
#endif
Figure 10-4 NTREE Module Private Header File
10.3.3 NTREE_create_tree
The description of NTREE_create_tree is identical to BTREE_create_tree. The code for
this method is shown below.
NTREE_ID_t NTREE_create_tree( void )
{
BTREE_ID_t tree = BTREE_create_tree();
return tree;
}
10.3.4 NTREE_add_root
The description of NTREE_add_root is identical to that of BTREE_add_root. The code is
shown below.
NTREE_NODE_ID_t NTREE_add_root( NTREE_ID_t tree, void *data )
{
BTREE_NODE_ID_t node = BTREE_NULL_NODE_ID;
CDA_ASSERT( BTREE_is_empty( tree ) );
node = BTREE_add_root( tree, data );
return node;
}

10.3.5 NTREE_add_child
This method will add a child to the list of a node’s children; it does so by locating the last
child in the list, and adding a sibling.
Synopsis:
NTREE_NODE_ID_t
NTREE_add_child( NTREE_NODE_ID_t node, void *data );
Where:
node == node to which to add
data == data to store at new node
Returns:
ID of new node
Exceptions:
Notes:
None
NTREE_NODE_ID_t
NTREE_add_child( NTREE_NODE_ID_t node, void *data )
{
NTREE_NODE_ID_t child = NTREE_NULL_NODE_ID;
if( !NTREE_has_child( node ) )

child = BTREE_add_left( node, data );
else
child = NTREE_add_sib( NTREE_get_child( node ), data );
return child;
}
10.3.6 NTREE_add_sib: Add a Sibling to a Node

This method will add a sibling to a node’s list of children.
Synopsis:
NTREE_NODE_ID_t
NTREE_add_sib( NTREE_NODE_ID_t node, void *data );
Where:
node == node to which to add
data == data to store at new node
Returns:
ID of new node
Exceptions:
Notes:
None
NTREE_NODE_ID_t NTREE_add_sib( NTREE_NODE_ID_t node, void *data )
{
NTREE_NODE_ID_t last = node;
NTREE_NODE_ID_t sib = NTREE_NULL_NODE_ID;

while( NTREE_has_sib( last ) )
last = NTREE_get_sib( last );
sib = BTREE_add_right( last, data );
return sib;
}
10.3.7 NTREE_get_root
This method is identical to BTREE_get_root. The code consists of a simple call to
BTREE_get_root.
10.3.8 NTREE_has_child
This method will determine whether a node has a child, which is the same as determining
if it has a left child in its binary tree.
Synopsis:
CDA_BOOL_t NTREE_has_child( NTREE_NODE_ID_t node );
Where:
node == node to interrogate
Returns:
CDA_TRUE if node has a child,
CDA_FALSE otherwise
Exceptions:
None
Notes:
None
CDA_BOOL_t NTREE_has_child( NTREE_NODE_ID_t node )
{
CDA_BOOL_t rcode = CDA_TRUE;
BTREE_NODE_ID_t child = BTREE_get_left( node );
if( child == BTREE_NULL_NODE_ID )

rcode = CDA_FALSE;
return rcode;
}
10.3.9 NTREE_has_sib
This method will determine whether a node has a sibling, which is the same as
determining if it has a right child in its binary tree.
Synopsis:
CDA_BOOL_t NTREE_has_sib( NTREE_NODE_ID_t node );
Where:
node == node to interrogate
Returns:
CDA_TRUE if node has a sibling,
CDA_FALSE otherwise
Exceptions:
None

Notes:
None
CDA_BOOL_t NTREE_has_sib( NTREE_NODE_ID_t node )
{
CDA_BOOL_t rcode = CDA_TRUE;
BTREE_NODE_ID_t sib = BTREE_get_right( node );
if ( sib == BTREE_NULL_NODE_ID )
rcode = CDA_FALSE;
return rcode;
}
10.3.10 NTREE_get_data, NTREE_get_child, NTREE_get_sib

These methods return the data, nearest child and nearest sibling of a node. The
descriptions of these three routines are identical to the descriptions of BTREE_get_data,
BTREE_get_left and BTREE_get_right, respectively. The code for each method consists
of a simple call to its analogous BTREE method.
10.3.11 NTREE_destroy_tree
The description of this routine is identical to the description of BTREE_destroy_tree. The
code is shown below.
NTREE_ID_t
NTREE_destroy_tree( NTREE_ID_t tree,
NTREE_DESTROY_PROC_p_t destroy_proc
)
{
BTREE_destroy_tree( tree, destroy_proc );
return NTREE_NULL_ID;
}
10.4 Directories
N-ary trees are used in a variety of applications, including those for parsing expressions
consisting of multiple operands (such as conditional expressions in C, which require
evaluation of three operands). But one of the most popular uses is the creation of
directories.
A directory is a hierarchical arrangement of nodes, some or all of which can contain other
nodes. One familiar example is the directory structure on your computer’s hard drive.
Figure 10-5 shows part of the layout of the directory structure on the C drive on my
Windows NT workstation.

\
bin cygnus home prog files
bash.exe bash.exe .bashrc .aliases
directx msoffice
cygwin
etc H-i586 share cygn.bat
termcap
bin include lib
libbfd.a libfl.a
limits.h stdlib.h
ar.exe as.exe
Figure 10-5 A Disk Directory Structure
The root directory, represented by the backslash (\) is a container that can hold other
directories and “regular” files, among them the bin, Cygnus, home and program files
subdirectories. The bin subdirectory contains two non-container files; the Cygnus
subdirectory contains the cygwin subdirectory, which in turn contains three additional
subdirectories plus the regular file cygnus.bat. As the figure suggests, an n-ary tree is a
good data structure for representing this organization. The root node represents the root
directory, and children of the root represent all the files and subdirectories contained by
the root. Children of the n-ary node cygwin represent files and subdirectories contained in
the cygwin subdirectory, and so forth. As an n-ary tree, the directory structure is
extensible to any practical limit.
More generally we can say that a directory element is a node that has a name, a list of
properties and may or may not contain other directory elements. A property is a
distinguishing characteristic of a node; it has a name and a value. A directory is any
collection of directory elements beginning with a root node that must be a container.
Within a directory, every node is represented by a distinguished name; a distinguished
name, or DN, is the concatenation of the node’s name with the names of 0 or more of its
ancestors. A fully qualified distinguished name, or FQDN, is the DN of a node beginning
with the root, and every node must have a unique FQDN. Any DN that is not fully
qualified has a context relative to some other node in the tree. Consider the node ar.exe in
the above illustration. Its name is ar.exe, and some of properties are its size (184,320
bytes) its creation date (December 25, 1999) and its type (application). It has a DN
relative to cygwin of H-i586\bin\ar.exe, and an FQDN of \cygnus\cygwin\H-
i586\bin\ar.exe.

root
sprockets widgets gadgets
personnel finance mfging
people people personnel finance distrib
design machine qc
people people people
john paul george
tom dick harry
dick jane sally

Figure 10-6 Organizational Directory
In Figure 10-6 we see an example of another popular type of directory, the

organizational directory. Children of the root represent different organizations, in the
case of our example possibly different companies. Each organization is broken down into
one or more organizational units, which in turn are broken down into groups that contain
either subgroups or members. Organization sprockets has three organizational units,
personnel, finance and manufacturing. Personnel has one group named people which has
three members, dick, jane and sally. Note that two nodes within the hierarchy are both
named dick, however their FQDNs are unique: \sprockets\personnel\people\dick and
\sprockets\finance\people\dick. Each node in the directory has a list of properties. For
example, \sprockets might have a list of properties that include the following:
• ceo = “George F. Snyde”
• address1 = “121 Buena Vista Drive”
• city = “San Jose”
\sprockets\personnel might have a list of properties that include these:
• director = “A. C. Calhoun”
• location = “Downtown”
And \sprockets\personnel\people\spot might have, among other properties, this one:
• fullname = “Spot D. Barker”

10.4.1 A Simple Directory Module

Let’s take a look at a very simple (and incomplete) module for building and interrogating
a directory. Our module name will be CDIR (as in C Language API For Directory
Access), and we’ll consider the following operations:
• Create a new directory
• Add a child to a directory element
• Add a property to a directory element
• Get a directory node
• Interrogate a property in a directory element
• Destroy a directory
To keep our example simple we will assume that every directory element can be a
container. We will also define a property to be a name/value pair, where both the name
and the value are NULL-terminated character strings.
10.4.2 Public Data Types

We’re going to need an incomplete data type to represent a directory ID, and another to
represent a node in a directory tree; and we’ll need types to represent both a NULL
directory ID and a NULL node ID. As you might expect, our implementation will
ultimately be based on the NTREE module, so our public types will be based on
declarations from the NTREE module. They are shown in Figure 10-7.
#define CDIR_NULL_ID (NTREE_NULL_ID)
#define CDIR_NULL_NODE_ID (NTREE_NULL_NODE_ID)
typedef NTREE_ID_t CDIR_ID_t;

typedef NTREE_NODE_ID_t CDIR_NODE_ID_t;
Figure 10-7 CDIR Module Public Data Types
10.4.3 CDIR_create_dir
This method will create a new directory with an empty root node.
Synopsis:
CDIR_ID_t CDIR_create_dir( const char *name );
Where:
name == name of the directory
Returns:
ID of directory
Exceptions:
Throws SIGABRT if directory can’t be created
Notes:
When the directory is no longer needed, the user must destroy
it by calling CDIR_destroy_dir.
10.4.4 CDIR_add_child
This method will add a child to some node in the directory. The name of the new node
may be specified as a relative DN or as an FQDN.

Synopsis:
CDIR_NODE_ID_t CDIR_add_child( CDIR_ID_t cdir,
CDIR_NODE_ID_t base,
const char **names,
size_t num_names
);
Where:
cdir == the (possibly NULL) ID of the target directory
base == the (possibly NULL) ID of a node in the directory
names == an array of strings representing the
distinguished name of the child to create
num_names == size of the names array
Returns:
CDIR_NULL_NODE_ID if names is an invalid DN, otherwise
the ID of the new node
Exceptions:
Throws SIGABRT if node can’t be created due to resource
allocation problems (e.g. dynamic memory failure)
Notes:
1. The distinguished name of the target node is the
concatenation of the strings in the names array.
2. The parent of the new node must already exist; that is, the
node whose DN is formed by concatenating the first
num_names – 1 strings from the names array. If the parent
node does not already exist, the operation will fail and
CDIR_NULL_NODE_ID will be returned.
3. If base is CDIR_NULL_NODE_ID then cdir must be non-NULL and
the names array is understood to define an FQDN. Otherwise
cdir is ignored, and the names array is understood to
define a DN relative to base.
10.4.5 CDIR_add_property
This method will add a property to a node.
Synopsis:
const char *CDIR_add_property( CDIR_NODE_ID_t node,
const char *name,
const char *value
);
Where:
node == node to which to add property
name == name of the property to add
value == value of the property to add
Returns:
value
Exceptions:
Throws SIGABRT if property can’t be created
Notes:
None
10.4.6 CDIR_get_node
This method will return the ID of a node in the directory. The name of the target node
may be a relative DN, or an FQDN
Synopsis:
CDIR_NODE_ID_t CDIR_get_node( CDIR_ID_t cdir,
const char **names,
size_t num_names
);
Where:
cdir == the (possibly NULL) ID of the target directory
base == the (possibly NULL) ID of a node in the directory
names == an array of strings representing the
distinguished name of the child to create
num_names == size of the names array
Returns:
CDIR_NULL_NODE_ID if the target node could not be located,
otherwise the ID of the target node
Exceptions:
None
Notes:
1. The distinguished name of the target node is the
concatenation of the strings in the names array.
2. If base is CDIR_NULL_NODE_ID then cdir must be non-NULL and
the names array is understood to define an FQDN. Otherwise
cdir is ignored, and the names array is understood to
define a DN relative to base.
10.4.7 CDIR_get_property
This method will return the value of a property in a node.
Synopsis:
CDA_BOOL_t CDIR_get_property( CDIR_NODE_ID_t node,
const char *name,
const char **value
);
Where:
node == node in which to find property
name == name of the property to find
value -> variable to which to return property value
Returns:
CDA_TRUE if property could be found, CDA_FALSE otherwise
Exceptions:
None
Notes:
None
10.4.8 CDIR_destroy_dir
This method will destroy a directory.
Synopsis:
CDIR_ID_t CDIR_destroy_dir( CDIR_ID_t cdir );
Where:
cdir == directory ID
Returns:
CDIR_NULL_ID

Exceptions:
None
Notes:
None
10.4.9 Implementation Structure

Our implementation will be based on an n-ary tree as defined by our NTREE module. A
directory node will simply be an NTREE node. The list of properties associated with a
node will be maintained in the data portion of the NTREE node; specifically, the data
portion of a node will be an ENQ anchor, where the name of the anchor will be the name
of the node. Each time we add a property we will add an item to the tail of the list; the
name of the item will be the name of the property, and the value of the property will be
stored in the data portion of the item itself. This strategy is illustrated in Figure 10-8.
ENQ anchor
NTREE node property
data flink flink flink

left blink blink blink
right name name name
121
George
Buena
F.
Vista
Snyde
sprockets Drive
address1 ceo
Figure 10-8 CDIR Implementation Strategy
Since our basic CDIR types will just be based on NTREE types the only thing we have to
declare in our private header file will be the structure to represent a property. This will be
an enqueuable item with a single field for the user data. This field is declared to be type
char, but when we create an item we will allocate enough extra space for an entire value,
and this field will contain the first byte. The complete private header file is shown in
Figure 10-9.
10.4.10 CDIR_create_dir Implementation

The CDIR create method just has to create an n-ary tree, and add a root to it. Like all of
our CDIR nodes, the root will need an anchor for a list of properties. Note that the root
node never has a name provided by the user; it is always referred to in some generic way,
such as root, /, \ or dot (.). So we’ll use the name of the root’s anchor to store the name of
the directory. Here is the implementation for this method.

CDIR_ID_t CDIR_create_dir( const char *name )
{
CDIR_ID_t cdir = NTREE_create_tree();
ENQ_ANCHOR_p_t anchor = ENQ_create_list( name );
NTREE_add_root( cdir, anchor );
return cdir;
}
#ifndef CDIRP_H
#define CDIRP_H
#include <cdir.h>
#include <enq.h>
typedef struct cdir__property_s

{
ENQ_ITEM_t item;
char value;
} CDIR__PROPERTY_t, *CDIR__PROPERTY_p_t;
#endif
Figure 10-9 CDIR Module Private Header File
10.4.11 CDIR_add_child Implementation

Adding a node to a directory is our most complex operation, because the DN must be
formulated from the names array and interpreted as either a relative DN or FQDN (much
of this complexity is embedded in CDIR_get_node). Since the entire names array defines
the DN of the new node, the names array up through the next-to-last element must be the
DN of the parent node. Here is the implementation.
CDIR_NODE_ID_t CDIR_add_child( CDIR_ID_t cdir,
const char **names,
size_t num_names
)
{
CDIR_NODE_ID_t node = CDIR_NULL_NODE_ID;
CDIR_NODE_ID_t temp = CDIR_NULL_NODE_ID;
CDA_ASSERT( num_names > 0 );

temp = CDIR_get_node( cdir, base, names, num_names - 1 );
if ( temp != CDIR_NULL_NODE_ID )
{
ENQ_ANCHOR_p_t anchor =
ENQ_create_list( names[num_names - 1] );
node = NTREE_add_child( temp, anchor );
}
return node;
}

10.4.12 CDIR_add_property
This method will add a new property to a node’s property list. Since the user passes in the
node, all we have to do is obtain the anchor of the property list, create a new item
containing the property value, and add it to the tail of the list. Here is the code.
const char *CDIR_add_property( CDIR_NODE_ID_t node,
const char *name,
const char *value
)
{
ENQ_ANCHOR_p_t anchor = NTREE_get_data( node );
size_t size =
sizeof(CDIR__PROPERTY_t) + strlen( value );
CDIR__PROPERTY_p_t prop =
(CDIR__PROPERTY_p_t)ENQ_create_item( name, size );
strcpy( &prop->value, value );

ENQ_add_tail( anchor, (ENQ_ITEM_p_t)prop );
return value;
}
10.4.13 CDIR_get_node Implementation

As did CDIR_add_child, this method will have to formulate a DN out the names array
passed by the user, and then interpret the DN as either relative or fully qualified. Note
that an array of length zero must refer to either the node passed by the user (for a relative
DN) or to the root node (for an FQDN). Here is the implementation.
CDIR_NODE_ID_t CDIR_get_node( CDIR_ID_t cdir,
const char **names,
size_t num_names
)
{
CDIR_NODE_ID_t node = base;
ENQ_ANCHOR_p_t anchor = NULL;
int inx = 0;
if ( node == CDIR_NULL_NODE_ID )
node = NTREE_get_root( cdir );
for ( inx = 0 ;
(inx < (int)num_names) && (node != CDIR_NULL_NODE_ID) ;
++inx
)
{
CDA_BOOL_t found = CDA_FALSE;
node = NTREE_get_child( node );
while ( !found && (node != CDIR_NULL_NODE_ID) )
{
anchor = NTREE_get_data( node );
if ( strcmp( names[inx], ENQ_GET_LIST_NAME( anchor )
) == 0
)
found = CDA_TRUE;

else
node = NTREE_get_sib( node );
}
}
return node;
}
10.4.14 CDIR_get_property Implementation

For this method all we have to do is obtain the property list from a node (which is passed
by the caller) and traverse it until we find an item with the name of the target property.
Once we find the item, we pass the address of the value back to the caller via the value
parameter. If we reach the end of the property list without locating the target property
then the property doesn’t exist, in which case we return NULL for the value of the
property, and CDA_FALSE for the function’s return value.
CDA_BOOL_t CDIR_get_property( CDIR_NODE_ID_t node,
const char *name,
const char **value
)
{
CDA_BOOL_t rcode = CDA_FALSE;
ENQ_ANCHOR_p_t anchor = NTREE_get_data( node );
ENQ_ITEM_p_t item = ENQ_GET_HEAD( anchor );
char *propv = NULL;
while ( !rcode && (item != anchor) )

if ( strcmp( ENQ_GET_ITEM_NAME( item ), name ) == 0 )
{
rcode = CDA_TRUE;
propv = &((CDIR__PROPERTY_p_t)item)->value;
}
else
*value = propv;
return rcode;
}
10.4.15 CDIR_destroy_dir Implementation

The implementation for this method is fairly simple: we just destroy the n-ary tree that
encapsulates our directory. The one wrinkle is that each node in the directory contains a
list of properties that have to be destroyed, but we easily handle that by using the NTREE
destroy method’s destroy-proc feature. Note that we don’t have to supply our users (that
is, the users of the CDIR module) with a destroy proc because none of the data in the
directory actually belongs to the user; it all belongs to the CDIR implementation. Here is
the code.
static NTREE_DESTROY_PROC_t destroyProc;
CDIR_ID_t CDIR_destroy_dir( CDIR_ID_t cdir )
{
NTREE_destroy_tree( cdir, destroyProc );
return CDIR_NULL_ID;

}
static void destroyProc( void *data )

{
ENQ_destroy_list( data );
}
10.4.16 Directories Discussion Wrap-up

The discussion of directories, above, merely touched the surface of this complex topic.
And our definition of a directory module is incomplete; for example, we should have
methods to change or delete a property, and a method to delete a node (along with all of
its children). Adding methods to change and delete a property is straightforward and is
left to the student. Adding a method to delete a node is a little tougher. Deleting a node
from our directory will entail adding an NTREE method to delete an NTREE node; this
in turn means adding a BTREE method to delete a BTREE node, which requires studying
up on the subject of binary tree pruning. But our purpose here was not to do a complete
directory implementation; it was to introduce directories, and to show how a directory
can be built using an n-ary tree. If you want to tackle the job of adding a delete-node
method, your textbook has all the details you need on pruning a binary tree. And to learn
more about directories and directory APIs, you should browse the Netscape DevEdge
documentation for Directory Server and LDAP, at
http://developer.netscape.com/docs/index.html.

Practice Final Examination

This lesson will prepare you for the final examination. The final examination will consist
of 20 questions worth five percentage points each; partial credit may apply to some of the
questions. To receive credit for the course, you must achieve a score of 70% on the final.
To prepare for the final, review your notes, assigned readings, and quizzes, and then
answer the following sample questions. During the exam itself, you may use your books
and notes. You will have two hours to complete the examination.
Sample Questions
1. Use CDA_malloc or CDA_calloc to allocate enough memory for an array
consisting of NELEMENTS elements. Each element should be type
CDA_INT32. Use a for loop to initialize each element of the array to –1.
Suggestion: If you think you know the answer, bench test it; code
your solution into a test driver and see if it really works like you
think it does.
2. The X Window System makes frequent use of callback routines much like the
destroy callbacks we wrote when designing some of our modules. Each such
callback has return type of void, and takes three parameters, each of which is type
void*. Use typedef statements to create the names XT_CBPROC_t, equivalent to
the type of a callback function, and XT_CBPROC_p_t, equivalent to type pointer-
to-callback function.
3. Complete the following subroutine, which traverses a linked list in search of an
item with a given name. If found, the item is dequeued and returned, otherwise,
NULL is returned.
ENQ_ITEM_p_t ENQ_deq_named_item( ENQ_ANCHOR_p_t anchor,
const char *name
)
{
ENQ_ITEM_p_t item = NULL;
return item;
}
think it does.
4. List two contexts in which the name of an array is not equivalent to the address of
an array?
5. In our ENQ module, what is the definition of a list in the empty state?
6. According to our module naming standards, to which module does the function
UI__dump_parms belong, and where is its prototype published?
Practice Final Examination 151 06/19/06
7. In the context of our HASH module, what is the definition of a collision?

8. In the context of our HASH module, complete the following function, which will
create an index out of the string of bytes pointed to by string. The length of string
is given by length, and the size of the table is given by table_size.
static size_t keyHash( const CDA_UINT8_t *string,
size_t length,
size_t tableSize
)
{
size_t index = 0;
return index;
}
think it does.
9. Write the function QUE_remove as discussed in your notes. Assume that the
queue is implemented via the ENQ module (also as discussed in your notes).
think it does.
10. Discuss the testing activities that occur during the design phase of the system life
cycle.
11. What is a regression test?
12. According to your notes, what are the two major categories of complexity.
13. Describe the order in which a stack pointer is incremented and dereferenced in a
push operation. What mechanism would you use to prevent stack overflow in a
push operation?
14. Complete the following subroutine. It will count the number of items stored in a
priority queue with the given priority; the queue is maintained as a single doubly
linked list (just like our "simple" priority queue implementation). Refer to your
notes for the declarations of PRQ_ID_t, etc.
size_t PRQ_get_class_len( PRQ_ID_t queue, int class )
{
int result=0;
return result;
}

think it does.
15. Assume that a binary tree is stored as an array, where the root node is stored in
index 0 of the array.
a) At what index will you find the parent of the node stored at index 197?
b) At what index will you find the right child of the node stored at index 233?
16. Refer to the accompanying figure. Is the illustrated binary tree balanced? Why or
why not?
17. List three ways to traverse a binary tree. Which method of traversal would you
use to print the keys of an index in alphabetical order?
18. Refer to the accompanying figure, an n-ary tree implemented via a binary tree,
and answer the following questions.
a) Which node is the parent of node F?
b) Is node G a sibling of node I?
c) How many children does node E have?

19. (Note: this topic is not currently covered in your notes, and will not be
represented on the final examination.) Examine the following source code, and
answer the questions that follow.
#define FALSE (0)
typedef int BOOL;
static BOOL inited = FALSE;
static void print_err( char *err )

{
int facility_id = 42;
if ( inited )
printf( "%d: err\n", facility_id, err );
}
a) Which components in the code represent variable space requirements?
b) Which components in the code represent fixed space requirements?

Answers
1.
CDA_INT32 *array = CDA_calloc( NELEMENTS, sizeof(CDA_INT32) );
int inx = 0;
for ( inx = 0 ; inx < NELEMENTS ; ++inx )

array[inx] = -1;
2.
typedef void XT_CBPROC_t( void *, void *, void * );
typedef XT_CBPROC_t *XT_CBPROC_p_t;
3.
ENQ_ITEM_p_t ENQ_deq_named_item( ENQ_ANCHOR_p_t anchor,
const char *name
)
{
ENQ_ITEM_p_t rval = NULL;
ENQ_ITEM_p_t item = ENQ_GET_HEAD( anchor );
while ( (item != anchor) && (rval == NULL) )

if ( strcmp( item->name, name ) == 0 )
rval = item;
else
item = item->flink;
return rval;
}
4. When it is used as the operand of the sizeof operator; a pointer may be used as an
lvalue, but the name of an array may not.
5. The flink and blink of the anchor point to the anchor.
6. The function belongs to the UI module. Since UI is followed by a double
underscore, it must be a private function, so its prototype is published in uip.h.
7. A collision occurs when two different keys hash to the same element of the hash
table's array.
8.
static size_t keyHash( const CDA_UINT8_t *string,
size_t length,
size_t tableSize
)
{
size_t index = 0;
size_t inx = 0;
const char *temp = string;
for ( inx = 0 ; inx < length ; ++inx )

index += *temp++;
index %= tableSize;
return index;
}

9.
QUE_ITEM_p_t QUE_remove( QUE_ID_t queue )
{
ENQ_ITEM_p_t item = ENQ_deq_head( anchor );
if ( item == anchor )
item = NULL;
return (QUE_ITEM_p_t)item;
}
10. Acceptance criteria for programs and modules are documented. Requirements are
certified as testable. A proof of concept prototype may be built. A high-level test
plan for verifying that the implementation conforms to the design is created.
11. A regression is a new flaw that is created when another flaw is fixed. A
regression test attempts to find regressions in a system that has been changed.
12. The two major categories of complexity are space complexity and time
complexity.
13. In a bottom-up stack implementation, first the stack pointer is dereferenced,
adding an item to the stack, then the stack pointer is incremented, making it point
to the next free item on the stack.
To prevent a stack overflow, use a conditional statement that throws SIGABRT if
a push operation would cause an overflow:
if ( stack is full )
abort();
push
Note that an assertion would be inappropriate, because assertions are disabled in
production code.
14.
size_t PRQ_get_class_len( PRQ_ID_t queue, CDA_UINT32_t class )
{
int result = 0;
item = (PRQ_ITEM_p_t)ENQ_GET_HEAD( anchor );

while ( (ENQ_ITEM_p_t)item != anchor &&
item->priority == class
)
item = (PRQ_ITEM_p_t)ENQ_GET_NEXT( &item->enq_item );
while ( (ENQ_ITEM_p_t)item != anchor &&

item->priority == class
)
{
++result;
item = (PRQ_ITEM_p_t)ENQ_GET_NEXT(&item->enq_item );
}
return result;
}

15. a) (197 - 1) / 2 = 98
b) 2 * 233 + 2 = 468
16. No. The distances between the root and each leaf sometimes vary by more than 1.
17. Three methods to traverse a binary tree are preorder, postorder and inorder. To
print the keys of an index in alphabetical order use an inorder traversal.
18. a) B
b) No; siblinghood is a one-way relationship, so I is a sibling of G, but G is not a
sibling of I.
c) None
19. a) Variable space requirements: The function's parameter, err; the function's
automatic variable, facility_id; and the parameters passed to printf.
b) Fixed space requirements: The static variable inited and the instructions
comprising the function print_err.

Quizzes
Quiz 1
Due Week 2
1. Write the typedefs for the following data types:
CDA_BOOL_t: type int.
OBJECT_t: a structure whose first two members, named flink and blink, are type
“pointer to OBJECT_t”; and whose third member, named object_id, is type (char *).
OBJECT_p_t: type pointer to OBJECT_t.
XQT_PROC_t: a function which returns CDA_BOOL_t, and takes a single argument

which is type pointer to OBJECT_t.
XQT_PROC_p_t: type pointer to XQT_PROC_t.
Use XQT_PROC_t to declare a static function named thread_proc.

2. Use CDA_malloc to allocate an array of 100 elements of type char*. Use a for loop
to initialize each element of the array to NULL.
3. Write cover routines for calloc and realloc like the one that we wrote for malloc in
class. Call the covers CDA_calloc and CDA_realloc, respectively.
Quizzes Page 159 6/19/2006

Quiz 2
Due Week 3
1. Review the definition for an item in the unenqueued state, and an anchor in the empty
state. Now review the implementation of ENQ_deq_item. One step at a time, show
what will happen if you pass an unenqueued item to ENQ_deq_item. One step at a
time, show what will happen if you pass an anchor in the empty state to
ENQ_deq_item.
2. Complete the following subroutine which will traverse one of our linked lists,
printing out the name of every item in the list:
void printNames( ENQ_ANCHOR_p_t list )
{
ENQ_ITEM_p_t item = NULL;
for ( item = ENQ_GET_HEAD( list ) ;
. . . ;
. . .
)
. . .
}
3. Write a function that will test whether or not your linked list function ENQ_deq_tail
will work correctly if passed the anchor of an empty list.

Quiz 3
Due Week 4
1. Briefly discuss the difference between a selection sort and a bubble sort.
2. Examine the following code:
int arr[5] = { 30, 20, 50, 70, 10 };
int *parr = &arr[4];
int inx = 0;
inx = *parr++;
a) Is the above code legal?
b) After executing the above code, what will be the value of inx?
c) After executing the above code, where will parr be pointing?
3. Examine the following subroutine:

static int storage[100];
static int *pstore = storage;
store( int number )

{
*pstore++ = number;
}
The routine store is to be called repeatedly. If it is called more than 100 times
it will malfunction; i.e., it will try to store an integer in a location outside the
array store. Add an if statement to store that will call abort if the operation
would be illegal.

Quiz 4
Due Week 5
1. Do a module’s private functions have external scope?
2. A static global variable has scope that extends to what?
3. Where will the private declarations for the module PRT be published?
4. Where should you place the declarations used only by the source file prt.c?
5. Write the minimum amount of code required to implement the private header file
for the INT module
6. In the context of an abstract data type, give an example of an exception.
7. What are three ways to handle an exception?

Quiz 5
Due Week 6
1. Write the code for STK_pop_item as discussed in your notes, in Section 6.4.4.
2. Complete the following function. This function will push strings passed by a
caller onto a stack until the caller passes NULL, then it will pop the strings off the
stack and print them.
#define MAX_STACK_SIZE (100)
static STK_ID__t stack = STK_NULL_ID;
void stringStuff( char *string )
{
if ( stack == STK_NULL_ID )
{
. . .
}
if ( string != NULL )
. . .
else
{
/* Note: as you pop strings off the stack, how will you
* know when you’re done?
*/
. . .
}
}

Quiz 6
Due Week 7
1. Complete routine PRQ_create_queue for the simple priority queue
implementation as discussed in your notes.
2. Revise the control structure for our PRQ module so that it can handle the
optimized implementation described in your notes.
3. Rewrite the PRQ create method to work with the optimized implementation
described in your notes.
4. A PRQ queue contains one priority 10 item, one priority 5 item and one priority 0
item. If PRQ_remove_item is called three times, in what order will the items be
returned?
5. Write a function that can be used to test whether your PRQ_remove_item method
works if passed the ID of an empty queue.

Quiz 7
Due Week 8
1. As defined in your notes, what are the phases of the system life cycle?
2. In what phases of the system life cycle is the module test plan used?
3. Describe the difference between unit testing and acceptance testing.

Quiz 8
Due Week 9
1. When is a binary tree balanced?
2. Complete routine BTREE_is_empty as described in Section 8.3.9 of your notes.
3. Complete routine BTREE_is_leaf as described in Section 8.3.10 of your notes.
4. Design and implement a method to delete a property from a node in the CDIR
module as discussed in the N-ary tree section of your notes.


C Programming - Data Structures and Algorithms

Uploaded by

C Programming - Data Structures and Algorithms

Uploaded by

C Programming:

Data Structures and Algorithms

C Programming: Data Structures and Algorithms

COURSE OVERVIEW ........................................................................................ IX

1.3 Pointers and Arrays ..................................................................................................................... 16

1.4 Dynamic Memory Allocation ...................................................................................................... 17

1.5 The Core Module.......................................................................................................................... 17

1.6 Activity .......................................................................................................................................... 21

2. DOUBLY LINKED LISTS.........................................................................23

2.2 Overview ....................................................................................................................................... 23

2.3 Definitions ..................................................................................................................................... 24

2.4 Case Study .................................................................................................................................... 35

2.5 Activity .......................................................................................................................................... 39

3.2 Overview ....................................................................................................................................... 41

3.3 Bubble Sort ................................................................................................................................... 41

3.4 Select Sort ..................................................................................................................................... 42

3.6 A Mergesort Implementation in C.............................................................................................. 43

4.2 Overview ....................................................................................................................................... 47

4.3 C Source Module Components.................................................................................................... 47

4.4 Review: Scope ............................................................................................................................... 49

4.5 A Bit about Header Files ............................................................................................................. 49

4.6 Module Conventions .................................................................................................................... 49

5. ABSTRACT DATA TYPES......................................................................51

5.2 Overview ....................................................................................................................................... 51

5.3 Exception Handling...................................................................................................................... 52

5.4 Classes of ADTs ............................................................................................................................ 54

6.2 Overview ....................................................................................................................................... 69

6.3 Stacks And Recursion .................................................................................................................. 72

6.4 A Minimal Stack Module............................................................................................................. 76

6.5 A More Robust Stack Module..................................................................................................... 82

7. PRIORITY QUEUES ................................................................................87

7.2 Overview ....................................................................................................................................... 87

7.3 Queues ........................................................................................................................................... 88

7.4 Simple Priority Queues................................................................................................................ 93

8. THE SYSTEM LIFE CYCLE ..................................................................107

8.2 Overview ..................................................................................................................................... 107

8.3 Testing ......................................................................................................................................... 109

9. BINARY TREES ....................................................................................113

9.2 Overview ..................................................................................................................................... 113

9.3 Binary Tree Representation ...................................................................................................... 115

9.4 A Minimal Binary Tree Implementation ................................................................................. 116

9.5 Using a Binary Tree as an Index............................................................................................... 124

9.6 Using a Binary Tree as an Index – Demonstration.................................................................. 127

9.7 Traversing a Binary Tree .......................................................................................................... 130

10. N-ARY TREES.......................................................................................135

10.2 Overview ..................................................................................................................................... 135

10.3 A Small N-ary Tree Implementation ........................................................................................ 136

10.4 Directories................................................................................................................................... 140

PRACTICE FINAL EXAMINATION ..................................................................151

Answers ..................................................................................................................................................... 155

Quiz 2......................................................................................................................................................... 160

Quiz 3......................................................................................................................................................... 161

Quiz 4......................................................................................................................................................... 162

Quiz 5......................................................................................................................................................... 163

Quiz 7......................................................................................................................................................... 165

Quiz 8......................................................................................................................................................... 166

1 Basic Skills, Core Kruse Chapters 1 and 2

2 Doubly Linked Lists Kruse Chapter 5, through 5.2 Quiz 1,

3 Sorting Kruse Chapter 7, through 7.7 Quiz 2

4 Modules, Abstract Kruse Section 4.8 Quiz 3,

5 Stacks Kruse Sections 3.1 and 3.2 Quiz 4

6 Priority Queues Kruse Sections 7.8 through 7.10 Quiz 5,

7 System Life Cycle Kruse Chapter 9, through 9.4 Quiz 6

8 Binary/N-ary Trees Kruse Chapter 10, through 10.2 Quiz 7,

9 Final Exam Kruse Chapter 10, remainder Quiz 8,

Week 2 Doubly Linked Lists H&S Section 5.6, through 5.6.4

Week 3 Sorting H&S Sections 20.5; Chapter 19

Week 4 Modules, Abstract Data Types Prata pp. 676 – 698

Week 5 Stacks H&S Sections 7.4.4, 7.5.8

Week 6 Priority Queues Kruse Chapter 11

Week 7 System Life Cycle Kruse Chapter 12

Week 8 Binary/N-ary Trees Prata 711 – 736

Week 9 Final Exam H&S Chapter 4

1.2.1 Typedef and Portability

Section 1: Basics 13 6/19/06

1.2.2 Typedef and Structures