Programming with

lcc-win32

by Jacob Navia and Q Software Solutions GmbH

© 2000-2003 Jacob Navia and Q Software Solutions GmbH. This document is part of the
lcc-win32 documentation. Distribution in any form is explicitly not allowed.
Chapter 1 Introduction to C 1

1.1 Organization of C programs 2

1.2 Hello 3
1.2.1 An overview of the compilation process 5
1.2.2 Summary of section Hello. 6
1.3 An overview of the standard libraries 7
1.4 Windows specific headers 8
1.5 Passing arguments to a program 8
1.5.1 Iteration constructs 10
1.5.1.1 for 11
1.5.1.2 while 11
1.5.1.3 do 11
1.5.2 Basic types 11
1.5.3 Summary. 12
1.6 Declarations and definitions 13
1.6.1 Variable declaration 13
1.6.2 Function declaration 15
1.6.3 Function definitions 16
1.6.4 Variable definition 16
1.6.5 Statement syntax 16
1.7 Errors and warnings 17
1.8 Reading from a file 18
1.9 Commentaries 23
1.9.1 Standard comments 24
1.9.1.1 Describing a function 24
1.9.1.2 Describing a file 26
1.10 An overview of the whole language 26
1.10.1 Statements 28
1.10.2 Declarations 32
1.10.3 Pre-processor 34
1.10.4 Windows specific defined symbols 35
1.10.5 Control-flow 36
1.10.6 Windows specific syntax 36
1.10.7 A closer view 37
1.10.7.1 Identifiers. 37
 1.10.7.2 Constants. 37
1.10.7.3 Array syntax. 38
1.10.7.4 Function call syntax 38
1.10.7.5 Functions with variable number of arguments. 39
1.10.7.6 Assignment. 39
1.10.7.7 Postfix 40
1.10.7.8 Subtraction. 40
1.10.7.9 Conditional operator. 40
1.10.7.10 struct. 41
1.10.7.11 Unions. 41
1.10.7.12 typedef. 41
1.10.7.13 register. 41
1.10.7.14 sizeof. 41
1.10.7.15 enum. 41
1.10.7.16 Prototypes. 41
1.10.7.17 variable length array. 42
1.10.7.18 const. 42
1.10.7.19 unsigned. 42
1.10.7.20 bit fields 42
1.10.7.21 stdcall. 43
1.10.7.22 break and continue statements 43
1.10.7.23 Null statements 44
1.10.7.24 Comments 44
1.10.7.25 Switch statement. 44
1.10.7.26 inline 45
1.10.7.27 Logical operators 46
1.10.7.28 Bitwise operators 46
1.10.7.29 Address-of operator 47
1.10.7.30 Indirection 47
1.10.7.31 Precedence of the different operators. 49
1.11 Simple programs 51
1.11.1 Problem 1: strchr 51
1.11.2 Problem 2: strlen 51
1.11.3 Problem 3: ispowerOfTwo 52
1.11.4 Problem 4: strlwr 53
1.11.5 Problem 5: paste 54
1.12 Using arrays and sorting 58
1.12.1 Summary of Arrays and sorting 67
1.13 Pointers and references 67
1.14 Structures and unions 70
1.14.1 Structures 70
1.14.2 Structure size 73
 1.14.3 Defining new types 74
1.14.4 Unions 75
1.15 Using structures 77
1.15.1 Fine points of structure use 79
1.16 Identifier scope and linkage 80
1.17 Top-down analysis 82
1.18 Extending a program 85
1.19 Improving the design 90
1.20 Path handling 91
1.20.1 Normal handling 91
1.20.2 Security considerations 94
1.21 Traditional string representation in C 97
1.22 Memory management and memory layout 99
1.22.1 Functions for memory allocation 101
1.22.2 Memory layout under windows 101
1.23 Counting words 102
1.23.1 The organization of the table 104
1.23.2 Memory organization 105
1.23.3 Displaying the results 107
1.23.4 Code review 108
1.24 A closer look at the pre-processor 109
1.25 Time and Date functions 111
1.26 Using structures (continued) 114
1.26.1 Lists 114
1.26.2 Hash tables 117
1.27 The preprocessor in detail 118
1.27.1 Preprocessor commands 118
1.27.1.1 Preprocessor macros 119
1.27.1.2 Conditional compilation 120
1.27.1.3 The pragma directive 120
1.27.1.4 The ## operator 121
1.27.2 Things to watch when using the preprocessor 121
1.28 Pitfalls of the C language 122
1.28.1 Defining a variable in a header file 122
1.28.2 Confusing = and == 122
 1.28.3 Forgetting to close a comment 122
1.28.4 Easily changed block scope. 122
1.28.5 Using the ++ or -- more than once in an expression. 123
1.28.6 Unexpected Operator Precedence 123
1.28.7 Extra Semi-colon in Macros 124

Chapter 2 Windows Programming 125

2.1 Introduction 125

2.2 WinMain 128
2.3 Resources 131
2.4 The dialog box procedure 135
2.5 A more advanced dialog box procedure 138
2.6 User interface considerations 140
2.7 Libraries 143
2.8 Dynamically linked libraries (DLLs) 149
2.9 Using a DLL 152
2.10 A more formal approach. 155
2.10.1 New syntax 155
2.10.2 Event oriented programming 156
2.11 A more advanced window 157
2.12 A more complex example: a "clone" of spy.exe 163
2.12.1 Creating the child windows 163
2.12.2 Moving and resizing the child windows 164
2.12.3 Starting the scanning. 164
2.12.4 Building the window tree. 165
2.12.5 Scanning the window tree 165
2.12.6 Review 166
2.12.7 Filling the status bar 168
2.12.8 Auxiliary procedures 169
2.13 Numerical calculations in C. 172
2.14 Filling the blanks 177
2.15 Using the graphical code generator 186
2.16 Understanding the wizard generated sample code 190
 2.16.1 Making a new menu or modifying the given menu. 190
2.16.2 Adding a dialog box. 190
2.16.3 Drawing the window 191
2.16.4 Initializing or cleaning up 191
2.16.5 Getting mouse input. 191
2.16.6 Getting keyboard input 191
2.16.7 Handling moving/resizing 191
2.16.8 Getting notified when the mouse passes over your window 192
2.16.9 Creating additional controls in your window without using a dialog box
192
2.17 Customizing controls 193
2.17.1 Processing the WM_CTLCOLORXXX message 193
2.17.2 Using the WM_DRAWITEM message 195
2.18 Etc. 199
2.18.1 Clipboard 200
2.18.2 Communications. 200
2.18.3 Files 200
2.18.4 File systems 201
2.18.5 Graphics 201
2.18.6 Handles and Objects 201
2.18.7 Inter-Process Communications 202
2.18.8 Mail 202
2.18.9 Multimedia 203
2.18.10 Network 203
2.18.11 Hooks 203
2.18.12 Registry 203
2.18.13 Shell Programming 204
2.18.14 Services 204
2.18.15 Windows 204
2.19 Advanced C programming with lcc-win32 206
2.19.1 Operator overloading 206
2.19.2 References 206
2.19.3 Overloaded functions 207
2.19.4 Default arguments 207
2.20 Advanced windows techniques 207
2.20.1 Memory mapped files 207
 2.20.2 Letting the user browse for a folder: using the shell 210
2.21 Some Coding Tips 213
2.21.1 Determining which version of Windows the program is running 214
2.21.2 Translating the value returned by GetLastError() into a readable string
214
2.21.3 Clearing the screen in text mode 214
2.21.4 Getting a pointer to the stack 215
2.21.5 Disabling the screen saver from a program 215
2.21.6 Drawing a gradient background 216
2.21.7 Capturing and printing the contents of an entire window 216
2.21.8 Centering a dialog box in the screen 219
2.21.9 Determining the number of visible items in a list box 219
2.21.10 Starting a non-modal dialog box 220
2.21.11 Propagating environment variables to the parent environment 220
2.21.12 Restarting the shell under program control 220
2.21.13 Translating client coordinates to screen coordinates 221
2.21.14 Passing an argument to a dialog box procedure 221
2.21.15 Calling printf from a windows application 221
2.21.16 Enabling or disabling a button or control in a dialog box. 221
2.21.17 Making a window class available for all applications in the system. 222
2.21.18 Accessing the disk drive directly without using a file system 222
2.21.19 Enumerating registry subkeys 223
2.21.20 Retrieving the Last-Write Time 224
2.21.21 Setting the System Time 225
2.21.22 Changing a File Time to the Current Time 225
2.21.23 Displaying the amount of disk space for each drive 225
2.22 FAQ 226
2.22.1 How do I create a progress report with a Cancel button? 226
2.22.2 How do I show in the screen a print preview? 228
2.22.3 How do I change the color of an edit field? 229
2.22.4 How do I draw a transparent bitmap? 229
2.22.5 How do I draw a gradient background? 232
2.22.6 How do I calculate print margins? 233
2.22.7 How do I calculate the bounding rectangle of a string of text? 233
2.22.8 How do I close an open menu? 234
2.22.9 How do I center a dialog box in the screen? 234
 2.22.10 How do I create non-rectangular windows? 235
2.22.11 How do I implement a non-blinking caret? 235
2.22.12 How do I create a title window (splash screen)? 235
2.22.13 How do I append text to an edit control? 239
2.22.14 How do I spawn a process with redirected stdin and stdout? 240
2.22.15 How to modify the width of the list of a combo box 240
2.22.16 How do I modify environment variables permanently? 242
2.22.17 How do I translate between dialog units and pixels? 243
2.22.18 How do I translate between client coordinates to screen coordinates? 243
2.22.19 When should I use critical sections and when is a mutex more appropi-
ate? 243
2.22.20 Why is my call to CreateFile failing when I use conin$ or conout$? 244
2.23 Overview of lcc-win32’s documentation 245
2.24 Bibliography 245
2.25 Newsgroups 246
2.26 Appendix. Code listings 250
2.26.1 The window tree. 250
2.26.2 Counting chars: countchars.c 255
 1

Chapter

Introduction to C
This tutorial to the C language supposes you have the lcc-win32 compiler system installed.
You will need a compiler anyway, and lcc-win32 is free for you to use, so please (if you
haven’t done that yet) download it and install it before continuing. http://www.q-software-
solutions.com
What the C language concerns, this is not a full-fledged introduction to all of C. There are
other, better books that do that (see the bibliography at the end of this book). Even if I try to
explain things from ground up, there isn’t here a description of all the features of the language.
Note too, that this is not just documentation or a reference manual. Functions in the standard
library are explained, of course, but no exhaustive documentation of any of them is provided
in this tutorial.1
But before we start, just a quick answer to the question: why learn C?
C has been widely criticized, and many people are quick to show its problems and drawbacks.
But as languages come and go, C stands untouched. The code of lcc-win32 has software that
was written many years ago, by many people, among others by Dennis Ritchie, the creator of
the language itself2. The answer to this question is very simple: if you write software that is
going to stay for some time, do not learn “the language of the day”; learn C.
C doesn’t impose you any point of view. It is not object oriented, but you can do object ori-
ented programming in C if you wish.3 It is not a functional language but you can do functional
programming4 with it if you feel like. Most LISP interpreters and Scheme interpreters/compil-
ers are written in C. You can do list processing in C, surely not so easily like in lisp, but you
can do it. It has all essential features of a general purpose programming language like recur-
sion, procedures as first class data types, and many others that this tutorial will show you.
Many people feel that C lacks the simplicity of Java, or the sophistication of C++ with its tem-
plates and other goodies. True. C is a simple language, without any frills. But it is precisely
this lack of features that makes C adapted as a first time introduction into a complex high-level
language that allows you fine control over what your program is doing without any hidden

1. For an overview of the lcc-win32 documentation see "How to find more information"
2. Dennis Ritchie wrote the pre-processor of the lcc-win32 system.
3. Objective C generates C, as does Eiffel and several other object-oriented languages. C, precisely
because of this lack of a programming model is adapted to express all of them. Even C++ started as a pre-
processor for the C compiler.
4. See the “Illinois FP” language implementations in C, and many other functional programming
languages that are coded in C.
2 C Tutorial

features. The compiler will not do anything else than what you told it to do. The language
remains transparent, even if some features from Java like the garbage collection are incorpo-
rated into the implementation of C you are going to use.5
As languages come and go, C remains. It was at the heart of the UNIX operating system devel-
opment in the seventies6, it was at the heart of the microcomputer revolution in the eighties,
and as C++, Delphi, Java, and many others came and faded, C remained, true to its own
nature.

1.1 Organization of C programs

A program in C is written in one or several text files called source modules. Each of those
modules is composed of functions, i.e. smaller pieces of code that accomplish some task7, and
data, i.e. variables or tables that are initialized before the program starts. There is a special
function called main that is where the execution of the program begins.8 In C, the organization
of code in files has semantic meaning. The main source file given as an argument to the com-
piler defines a compilation unit.9
A unit can import common definitions using the #include preprocessor directive, or just by
declaring some identifier as extern.10
C supports the separate compilation model, i.e. you can split the program in several indepen-
dent units that are compiled separately, and then linked with the link editor to build the final
program. Normally each module is written in a separate text file that contains functions or data
declarations. Interfaces between modules are written in “header files” that describe types or
functions visible to several modules of the program. Those files have a “.h” extension, and
they come in two flavors: system-wide, furnished with lcc-win32, and private, specific to the
application you are building.
A function has a parameter list, a body, and possibly a return value.11 The body can contain
declarations for local variables, i.e. variables activated when execution reaches the function
body.

5. Lisp and scheme, two list oriented languages featured automatic garbage collection since several
decades. APL and other interpreters offered this feature too. Lcc-win32 offers you the garbage collector
developed by Hans Boehm.
6. And today, the linux kernel is written entirely in C as most operating systems.
7. There is no distinction between functions and procedures in C. A procedure is a function of return
type void.
8. Actually, the startup code calls main. When main returns, the startup code regains control and ends
the program. This is explained in more detail in the technical documentation.
9. Any program, in any computer in any language has two main types of memory at the start:
The code of the program, i.e. the sequence of machine instructions that the program will execute. This
section has an “entry point”, the above mentioned “main” procedure in C, or other procedure that is
used as the entry point
The static data of the program, i.e. the string literals, the numbers or tables that are known when the
program starts. This data area con be further divided into an initialized data section, or just empty,
reserved space that is initialized by the operating system to zero when the program is loaded.
10. There is no way to import selectively some identifiers from another included file. Either you import
all of it or none.
 Hello 3

The body is a series of expressions separated by semicolons. Each statement can be an arith-
metic operation, an assignment, a function call, or a compound statement, i.e. a statement that
contains another set of statements.

1.2 Hello
To give you an idea of the flavor of C we use the famous example given already by the authors
of the language12. We build here a program that when invoked will put in the screen the mes-
sage “hello”.
#include <stdio.h> (1)
int main(void) (2)
{ (3)
printf("Hello\n"); (4)
return 0; (5)
} (6)

1) Using a feature of the compiler called ‘pre-processor’, you can textually include a whole
file of C source with the “#include” directive. In this example we include from the standard
includes of the compiler the “stdio.h” header file.13
2) We define a function called “main” that returns an integer as its result, and receives no
arguments (void). 14

3) The body of the function is a list of statements enclosed by curly braces.

4) We call the standard function “printf” that formats its arguments in a character string that is
displayed in the screen. A function call in C is written like this:function-name ‘(‘
argument-list ‘)’. In this case the function name is “printf”, and its argument
list is the character string “Hello\n”15. Character strings are enclosed in double quotes.
They are represented in C as an array of characters finished by a zero byte.

5) The return statement indicates that control should be returned (hence its name) to the
calling function. Optionally, it is possible to specify a return result, in this case the integer
zero.

6) The closing brace finishes the function scope.

11. In C, only one return value is possible. A function, however can return several return values if it
modifies its environment.
12. This example is a classic, and appears already in the tutorial of the C language published by B. W.
Kernighan in 1974, four years before the book “The C programming language” was published. Their example
would still compile today, albeit with some warnings:
main() { printf(“Hello world\n”); }
13. The name of the include file is enclosed within a <> pair. This indicates the compiler that it should
look for this include file in the standard include directory, and not in the current directory. If you want to
include a header file in another directory or in the compilation directory, use the double quotes to enclose the
name of the file, like #include “myfile.h”
14. This is one of the two possible definitions of the “main” function. Later we will see the other one.
15. Character strings can contain sequences of characters that denote graphical characters like new line
(\n) tab (\t), backspace (\b), or others. In this example, the character string is finished by the new line
character \n.
4 C Tutorial

Programs in C are defined in text files that normally have the .c extension. You can create
those text files with any editor that you want, but lcc-win32 proposes a specialized editor for
this task called “Wedit”. This program allows you to enter the program text easily, since it is
adapted to the task of displaying C source text.
To make this program then, we start Wedit and enter the text of that program above.16

Once this
is done, you can compile, and link-edit your program by just clicking in the compile menu or
pressing F9.17
To run the program, you use the “execute” option in the compile menu (Ctrl+F5), or you open
a command shell and type the program’s name. Let’s do it the hard way first.

16. You start wedit by double clicking in its icon, or, if you haven’t an icon for it by going to the “Start”
menu, run, and then type the whole path to the executable. For instance, if you installed lcc-win32 in c:\lcc,
wedit will be in c:\lcc\bin\wedit.exe
17. If this doesn’t work or you receive warnings, you have an installation problem (unless you made a
typing mistake). Or maybe I have a bug. When writing mail to me do not send messages like: “It doesn’t
work”. Those messages are a nuisance since I can’t possibly know what is wrong if you do not tell me exactly
what is happening. Wedit doesn’t start? Wedit crashes? The computer freezes? The sky has a black color?
Keep in mind that in order to help you I have to reproduce the problem in my setup. This is impossible
without a detailed report that allows me to see what goes wrong.
Wedit will make a default project for you, when you click the “compile” button. This can go wrong if
there is not enough space in the disk to compile, or the installation of lcc-win32 went wrong and Wedit can’t
find the compiler executable, or many other reasons. If you see an error message please do not panic, and try
to correct the error the message is pointing you to.
A common failure happens when you install an older version of Wedit in a directory that has spaces in it.
Even if there is an explicit warning that you should NOT install it there, most people are used to just press
return at those warnings without reading them. Then, lcc-win32 doesn’t work and they complain to me. I have
improved this in later versions, but still problems can arise.
 Hello 5

The first thing we need to know is the name of the program we want to start. This is easy; we
ask Wedit about it using the “Executable stats” option in the “Utils” menu. We get the follow-
ing display.

We see at the first line of the bottom panel, that the program executable is called:
h:\lcc\projects\hello.exe
We open a command shell window, and type the command:
C:\>h:\lcc\projects\lcc1\hello.exe
Hello
C:\>
Our program displays the character string “Hello” and then a new line, as we wanted. If we
erase the \n of the character string, press F9 again to recompile and link, the display will be:
C:\>h:\lcc\projects\lcc1\hello.exe
Hello
C:\>
But how did we know that we have to call “printf” to display a string?
Because the documentation of the library told us so… The first thing a beginner to C must do
is to get an overview of the libraries provided already with the system so that he/she doesn’t
waste time rewriting programs that can be already used without any extra effort. Printf is one
of those, but are several thousands of pre-built functions of all types and for all tastes. We
present an overview of them in the next section.

1.2.1 An overview of the compilation process

When you press F9 in the editor, a complex sequence of events, all of them invisible to you,
produce an executable file. Here is a short description of this, so that at least you know what’s
happening behind the scene.
Wedit calls the C compiler proper. This program is called lcc.exe and is in the installation
directory of lcc, in the bin directory. For instance, if you installed lcc in c:\lcc, the compiler
will be in c:\lcc\bin.
This program will read your source file, and produce another file called object file,18 that has
the same name as the source file but a .obj extension. C supports the separate compilation
6 C Tutorial

model, i.e. you can compile several source modules producing several object files, and rely in
the link-editor lcclnk.exe to build the executable.
Lcclnk.exe is the link-editor, or linker for short. This program reads different object files,
library files and maybe other files, and produces either an executable file or a dynamically
loaded library, a DLL.
When compiling your hello.c file then, the compiler produced a “hello.obj” file, and from that,
the linker produced a hello.exe executable file. The linker uses several files that are stored in
the \lcc\lib directory to bind the executable to the system DLLs, used by all programs:
kernel32.dll, crtdll.dll, and many others.
The workings of the lcc compiler are described in more detail in the technical documentation.
Here we just tell you the main steps.
• The source file is first pre-processed. The #include directives are resolved, and the text
of the included files is inserted into the source file.19
• The front end of the compiler proper processes the resulting text. Its task is to generate a
series of intermediate code statements.20 The code generator that emits assembler
instructions from it processes these.21
• Eventually the compiler produces an object file with the .obj extension. This file is
passed then (possibly with other object files) to the linker lcclnk that builds the
executable.
Organizing all those steps and typing all those command lines can be boring. To easy this, the
IDE will do all of this with the F9 function key.

1.2.2 Summary of section Hello.

• The #include pre-processor directive allow us to include standard headers22 in our
programs to use their definitions.
• C programs begin in a function called “main”
• A function is a set of C statement that performs a single task.
• Functions can receive arguments and can return a result.

18. This has nothing to do with object oriented programming of course!

19. The result of this process can be seen if you call the compiler with the –E flag. For instance, to see
what is the result of pre-processing the hello.c file you call the compiler in a command shell window with the
command line: lcc –E hello.c.
20. Again, you can see the intermediate code of lcc by calling the compiler with lcc –z hello.c.
This will produce an intermediate language file called hello.lil that contains the intermediate language
statements.
21. Assembly code can be generated with the lcc-S hello.c command, and the generated assembly
file will be called hello.asm.
22. We will see that #include applies to user defined include headers too.
 An overview of the standard libraries 7

1.3 An overview of the standard libraries

The documentation of lcc-win32, specifically the user manual, has the exhaustive list of all
libraries provided by the system. The on-line help system has a detailed description of each
one of those. It would be a waste to repeat all that detail here, so only a broad outline will be
given:23

Header Purpose
stdio.h Standard input and output. Here you will find the definitions of the FILE structure,
used for all input and output, the definitions of all functions that deal with file opening,
closing, etc.
The famous printf function is defined here too, together with sprintf, fprintf, and all the
related family of functions.

math.h Mathematical functions. sin,cos,atan, log, exp, etc. We have here

trigonometry (sin, cos, tan, atan, etc).
rounding (ceil, floor)
logarithms (log, exp, log10, etc)
square and cubic roots (sqrt, cbrt)
Useful constants like pi, e, etc.

stdlib.h Standard library functions. We have here:

abort (abnormal termination of the program)
exit (normal termination)
atoi, itoa (text to integer conversion)
malloc,calloc,free (dynamic memory module)
rand, srand (random numbers)
putenv, getenv (environment management)
qsort (sorting)
strtod, strtol (conversion from string to double/long)
sleep (suspend execution for a certain period of time)

stddef.h This file defines macros and types that are of general use in a program. NULL, offsetof,
ptrdiff_t, size_t, and several others.

string.h String handling. Here are defined all functions that deal with the standard represen-
tation of strings as used in C. See “Traditional string representation in C” on page 97.

ctype.h character classification (isalpha, islower, isdigit)

stdarg.h Functions with variable number of arguments are described here.
complex.h Functions for complex numbers
time.h Time related functions

23. In the user’s manual there is an exhaustive list of the entire set of header files distributed with lcc-
win32. Please look there for an in-depth view.
8 C Tutorial

1.4 Windows specific headers

windows.h All windows definitions. Creating a window, opening a window, this is an extensive
header file, makes approx half a megabyte of definitions. Note that under lcc-win32,
several headers like winbase.h of other distributions are concentrated in a single file.

winsock.h Network (tcpip)

shellapi.h Windows Shell

1.5 Passing arguments to a program

We can’t modify the behavior of our hello program with arguments. We have no way to pass it
another character string for instance, that it should use instead of the hard-wired “hello\n”. We
can’t even tell it to stop putting a trailing new line character.
Programs normally receive arguments from their environment. A very old but still quite effec-
tive method is to pass a command line to the program, i.e. a series of character strings that the
program can use.
Let’s see how arguments are passed to a program.24
#include <stdio.h> (1)
int main(int argc,char *argv[]) (2)
{
int count; (3)

for (count=0;count < argc;count++) {(4)

printf( (5)
"Argument %d = %s\n",
count,
argv[count]);
} (6)
return 0;
}

1) We include again stdio.h

2) We use a longer definition of the “main” function as before. This one is as standard as the
previous one, but allows us to pass parameters to the program. There are two arguments:

int argc This is an integer that in C is known as “int”. It contains the number of arguments
passed to the program plus one.

24. Here we will only describe the standard way of passing arguments as specified by the ANSI C
standard, the one lcc-win32 uses. Under the Windows operating system, there is an alternative entry point,
called WinMain, and its arguments are different than those described here. See the Windows programming
section later in this tutorial.
 Passing arguments to a program 9

char *argv[] This is an array of pointers to characters25 containing the actual arguments
given. For example, if we call our program from the command line with the arguments
“foo” and “bar”, the argv[ ] array will contain:
argv[0] The name of the program that is running.
argv[1] The first argument, i.e. “foo”.
argv[2] The second argument, i.e. “bar”.
We use a memory location for an integer variable that will hold the current argument to be
printed. This is a local variable, i.e. a variable that can only be used within the enclosing
scope, in this case, the scope of the function “main”.26

3) We use the “for” construct, i.e. an iteration. The “for” statement has the following structure:
• Initialization. Things to be done before the loop starts. In this example, we set the
counter to zero. We do this using the assign statement of C: the “=” sign. The general
form of this statement is
• variable “=” value
• Test. Things to be tested at each iteration, to determine when the loop will end. In this
case we test if the count is still smaller than the number of arguments passed to the
program, the integer argc.
• Increment. Things to be updated at each iteration. In this case we add 1 to the counter
with the post-increment instruction: counter++. This is just a shorthand for writing
counter = counter + 1.
• Note that we start at zero, and we stop when the counter is equal to the upper value of
the loop. Remember that in C, array indexes for an array of size n elements always start
at zero and run until n-1.27

4) We use again printf to print something in the screen. This time, we pass to printf the
following arguments:
"Argument %d = ‘%s’\n"
count
argv[count]

25. This means that you receive the machine address of the start of an integer array where are stored the
start addresses of character strings containing the actual arguments. In the first position, for example, we will
find an integer that contains the start position in RAM of a sequence of characters containing the name of the
program. We will see this in more detail when we handle pointers later on.
26. Local variables are declared (as any other variables) with:
<type> identifier;
For instance
int a;
double b;
char c;
Arrays are declared in the same fashion, but followed by their size in square brackets:
int a[23];
double b[45];
char c[890];
10 C Tutorial

Printf will scan its first argument. It distinguishes directives (introduced with a per-cent
sign %), from normal text that is outputted without any modification. We have in the
character string passed two directives a %d and a %s.
The first one, a %d means that printf will introduce at this position, the character
representation of a number that should also be passed as an argument. Since the next
argument after the string is the integer “count”, its value will be displayed at this point.
The second one, a %s means that a character string should be introduced at this point.
Since the next argument is argv[count], the character string at the position “count” in the
argv[ ] array will be passed to printf that will display it at this point.

5) We finish the scope of the for statement with a closing brace. This means, the iteration
definition ends here.
Now we are ready to run this program. Suppose that we have entered the text of the program in
the file “args.c”. We do the following:
h:\lcc\projects\args> lcc args.c
h:\lcc\projects\args> lcclnk args.obj
We first compile the text file to an object file using the lcc compiler. Then, we link the result-
ing object file to obtain an executable using the linker lcclnk. Now, we can invoke the program
just by typing its name:28
h:\lcc\projects\args> args
Argument 0 = args
We have given no arguments, so only argv[0] is displayed, the name of the program, in this
case “args”. Note that if we write:
h:\lcc\projects\args> args.exe
Argument 0 = args.exe
We can even write:
h:\lcc\projects\args> h:\lcc\projects\args.exe
Argument 0 = h:\lcc\projects\args.exe
But that wasn’t the objective of the program. More interesting is to write:
h:\lcc\projects\args> args foo bar zzz
Argument 0 = args
Argument 1 = foo
Argument 2 = bar
Argument 3 = zzz
The program receives 3 arguments, so argc will have a value of 4. Since our variable count
will run from 0 to argc-1, we will display 4 arguments: the zeroth, the first, the second, etc.

1.5.1 Iteration constructs

We introduced informally the “for” construct above, but a more general introduction to loops
is necessary to understand the code that will follow.

27. An error that happens very often to beginners is to start the loop at 1 and run it until its value is
smaller or equal to the upper value. If you do NOT use the loop variable for indexing an array this will work,
of course, since the number of iterations is the same, but any access to arrays using the loop index (a common
case) will make the program access invalid memory at the end of the loop.
28. The detailed description of what happens when we start a program, what happens when we compile,
how the compiler works, etc, are in the technical documentation of lcc-win32. With newer versions you can
use the compilation driver ‘lc.exe’, that will call the linker automatically.
 Passing arguments to a program 11

There are three iteration constructs in C: “for”, “do”, and “while”.

1.5.1.1 for
The “for” construct has

1. An initialization part, i.e. code that will be always executed before the loop begins,

2. A test part, i.e. code that will be executed at the start of each iteration to determine if
the loop has reached the end or not, and

3. An increment part, i.e. code that will be executed at the end of each iteration.
Normally, the loop counters are incremented (or decremented) here.
The general form is then:
for(init ; test ; increment) {
statement block
}

1.5.1.2 while
The “while” construct is much more simple. It consists of a single test that determines if the
loop body should be executed or not. There is no initialization part, nor increment part.
The general form is:
while (test) {
statement block
}
Any “for” loop can be transformed into a “while” loop by just doing:
init
while (test) {
statement block
increment
}

1.5.1.3 do
The “do” construct is a kind of inverted while. The body of the loop will always be executed
at least once. At the end of each iteration the test is performed. The general form is:
do {
statement block
} while (test);
Using the “break” keyword can stop any loop. This keyword provokes an exit of the block of
the loop and execution continues right afterwards.
The “continue” keyword can be used within any loop construct to provoke a jump to the end
of the statement block. The loop continues normally, only the statements between the continue
keyword and the end of the loop are ignored.

1.5.2 Basic types

The implementation of the C language by the lcc-win32 compiler has the following types built
in:29All this types are part of the standard ANSI C language. With the exception of the
_Complex type they should appear in most C implementations and they do appear in all win-
dows compilers.30
12 C Tutorial

Type Size (bytes) Description

_Bool1 1 Logical type, can be either zero or one.

char 1 Character or small integer type. Comes in two flavors:

signed or unsigned.

short 2 Integer stored in 16 bits. Signed or unsigned.

int 4 Integer stored in 32 bits. Signed or unsigned.

long 4 Identical to int

long long 8 Integer stored in 64 bits. Signed or unsigned.

float 4 Floating-point single precision. (Approx 7 digits)

double 8 Floating-point double precision. (Approx. 15 digits)

long double 12 Floating point extended precision (Approx 20 digits)

float _Complex 8 Complex number. Each _Complex is composed of two

double _Complex 16 parts: real and imaginary part. Include <complex.h> when
using them.
long double _Complex 24

1. The actual type of the Boolean type should be “bool”, but in the standard it was specified that this
type wouldn't be made the standard name for now, for compatibility reasons with already running
code. If you want to use bool, you should include the header “stdbool.h”.

These are the basic types of lcc-win32. Other types for numbers exist. To use them you should
include the corresponding header file, they are not “built in” into the compiler. They are built

Type Header Size (bytes) Description

qfloat qfloat.h 56 350 bits floating point
bignum bignum.h variable Extended precision number

using a property of the compiler that allows you to define your own kind of numbers and their
operations. This is called operator overloading and will be explained further down.

1.5.3 Summary.
² Functions receive arguments and return a result in the general case. The type of the result
is declared before its name in a function declaration or definition.
² The “main” function can receive arguments from its calling environment.
² We have to declare the type of each identifier to the compiler before we can use it.

29. In most compilers the char/short/int/long types are present but their sizes can change from machine to
machine. Some embedded systems compilers do not support floating point. Many compilers do not
implement the recent types _Bool, long long, or long double. Within the windows environment how-
ever, the char/short/int/long/float/double types are identical to this ones in all windows compilers I
know of.
30. Microsoft Visual C implements "long double" as double, and calls the long long type "__int64".
 Declarations and definitions 13

1.6 Declarations and definitions

It is very important to understand exactly the difference between a declaration and a definition
in C.
A declaration introduces an identifier to the compiler. It says in essence: this identifier is a xxx
and its definition will come later. An example of a declaration is
extern double sqrt(double);
With this declaration, we introduce to the compiler the identifier sqrt, telling it that it is a func-
tion that takes a double precision argument and returns a double precision result. Nothing
more. No storage is allocated for this declaration, besides the storage allocated within the
compiler internal tables.31
A definition tells the compiler to allocate storage for the identifier. For instance, when we
defined the function main above, storage for the code generated by the compiler was created,
and an entry in the program’s symbol table was done. In the same way, when we wrote:
int count;
above, the compiler made space in the local variables area of the function to hold an integer.
And now the central point: You can declare a variable many times in your program, but there
must be only one place where you define it. Note that a definition is also a declaration, because
when you define some variable, automatically the compiler knows what it is, of course. For
instance if you write:
double balance;
even if the compiler has never seen the identifier balance before, after this definition it knows
it is a double precision number.32

1.6.1 Variable declaration

A variable is declared with
<type> <identifier> ;
like
int a;
double d;
long long h;
All those are definitions of variables. If you just want to declare a variable, without allocating
any storage, because that variable is defined elsewhere you add the keyword extern:
extern int a;
extern double d;
extern long long d;
Optionally, you can define an identifier, and assign it a value that is the result of some calcula-
tion:

31. Note that if the function so declared is never used, absolutely no storage will be used. A declaration
doesn’t use any space in the compiled program, unless what is declared is effectively used. If that is the case,
the compiler emits a record for the linker telling it that this object is defined elsewhere.
32. Note that when you do not provide for a declaration, and use this feature: definition is a declaration;
you can only use the defined object after it is defined. A declaration placed at the beginning of the program
module or in a header file frees you from this constraint. You can start using the identifier immediately, even
if its definition comes much later, or even in another module.
14 C Tutorial

double fn(double f) {
double d = sqrt(f);
// more statements
}
Note that initializing a value with a value unknown at compile time is only possible within a
function scope. Outside a function you can still write:
int a = 7;
or
int a = (1024*1024)/16;
but the values you assign must be compile time constants, i.e. values that the compiler can fig-
ure out when doing its job.
Pointers are declared using an asterisk:
int *a;

This means that a will contain the machine address of some unspecified integer. 33
You can save some typing by declaring several identifiers of the same type in the same decla-
ration like this:
int a,b=7,*c,h;
Note that c is a pointer to an integer, since it has an asterisk at its left side. This notation is
somehow confusing, and forgetting an asterisk is quite common. Use this multiple declara-
tions when all declared identifiers are of the same type and put pointers in separate lines.
The syntax of C declarations has been criticized for being quite obscure. This is true; there is
no point in negating an evident weakness. In his book “Deep C secrets”34 Peter van der Lin-
den writes a simple algorithm to read them. He proposes (chapter 3) the following:
The Precedence Rule for Understanding C Declarations.
Rule 1: Declarations are read by starting with the name and then reading in precedence order.
Rule 2: The precedence, from high to low, is:
2.A : Parentheses grouping together parts of a declaration
2.B: The postfix operators:
2.B.1: Parentheses ( ) indicating a function prototype, and
2.B.2: Square brackets [ ] indicating an array.
2.B.3: The prefix operator: the asterisk denoting "pointer to".
Rule 3: If a const and/or volatile keyword is next to a type specifier e.g. int, long, etc.) it
applies to the type specifier. Otherwise the const and/or volatile keyword applies to the pointer
asterisk on its immediate left.
Using those rules, we can even understand a thing like:
char * const *(*next)(int a, int b);
We start with the variable name, in this case “next”. This is the name of the thing being
declared. We see it is in a parenthesized expression with an asterisk, so we conclude that “next
is a pointer to…” well, something. We go outside the parentheses and we see an asterisk at the

33. Machine addresses are just integers, of course. For instance, if you have a machine with 128MB of
memory, you have 134 217 728 memory locations. They could be numbered from zero up, but Windows uses
a more sophisticated numbering schema called “Virtual memory”.
34. Deep C secrets. Peter van der Linden ISBN 0-13-177429-8
 Declarations and definitions 15

left, and a function prototype at the right. Using rule 2.B.1 we continue with the prototype.
“next is a pointer to a function with two arguments”. We then process the asterisk: “next is a
pointer to a function with two arguments returning a pointer to…” Finally we add the char *
const, to get
“next” is a pointer to a function with two arguments returning a pointer to a constant pointer to
char.
Now let’s see this:
char (*j)[20];
Again, we start with “j is a pointer to”. At the right is an expression in brackets, so we apply
2.B.2 to get “j is a pointer to an array of 20”. Yes what? We continue at the left and see ”char”.
Done. “j” is a pointer to an array of 20 chars. Note that we use the declaration in the same form
without the identifier when making a cast:
j = (char (*)[20]) malloc(sizeof(*j));
We see in bold and enclosed in parentheses (a cast) the same as in the declaration but without
the identifier j.

1.6.2 Function declaration

A declaration of a function specifies:
• The return type of the function
• Its name
• The types of each argument
The general form is:
<type> <Name>(<type of arg 1>, ... <type of arg N> ) ;
double sqrt(double) ;
Note that an identifier can be added to the declaration but its presence is optional. We can
write:
double sqrt(double x);
if we want to, but the “x” is not required and will be ignored by the compiler.
Functions can have a variable number of arguments. The function “printf” is an example of a
function that takes several arguments. We declare those functions like this:
int printf(char *, ...);

The ellipsis means “some more arguments”.35

Why are function declarations important?
When I started programming in C, prototypes for functions didn’t exist. So you could define a
function like this:
int fn(int a)
{
return a+8;
}
and in another module write:
fn(7,9);

35. The interface for using functions with a variable number of arguments is described in the standard
header file “stdarg.h”. See “Functions with variable number of arguments.” on page 39.
16 C Tutorial

without any problems.

Well, without any problems at compile time of course. The program crashed or returned non-
sense results. When you had a big system of many modules written by several people, the
probability that an error like this existed in the program was almost 100%. It is impossible to
avoid mistakes like this. You can avoid them most of the time, but it is impossible to avoid
them always.
Function prototypes introduced compile time checking of all function calls. There wasn’t any-
more this dreaded problem that took us so many debugging hours with the primitive debug-
gers of that time. In the C++ language, the compiler will abort compilation if a function is used
without prototypes. I have thought many times to introduce that into lcc-win32, because ignor-
ing the function prototype is always an error. But, for compatibility reasons I haven’t done it
yet.36

1.6.3 Function definitions

Function definitions look very similar to function declarations, with the difference that instead
of just a semi colon, we have a block of statements enclosed in curly braces, as we saw in the
function “main” above. Another difference is that here we have to specify the name of each
argument given, these identifiers aren’t optional any more: they are needed to be able to refer
to them within the body of the function. Here is a rather trivial example:
int addOne(int input)
{
return input+1;
}

1.6.4 Variable definition

A variable is defined when the compiler allocates space for it. For instance, at the global level,
space will be allocated by the compiler when it sees a line like this:
int a;
or
int a = 67;
In the first case the compiler allocates sizeof(int) bytes in the non-initialized variables section
of the program. In the second case, it allocates the same amount of space but writes 67 into it,
and adds it to the initialized variables section.

1.6.5 Statement syntax

In C, the enclosing expressions of control statements like if, or while, must be enclosed in
parentheses. In many languages that is not necessary and people write:
if a < b run();
in C, the if statement requires a parentheses
if (a<b) run();

36. There is a strong commitment, from the part of the compiler writers, to maintain the code that was
written in the language, and to avoid destroying programs that are working. When the standards committee
proposed the prototypes, all C code wasn’t using them yet, so a transition period was set up. Compilers would
accept the old declarations without prototypes and just emit a warning. Some people say that this period
should be over by now (it is more than 10 years that we have prototypes already), but still, new compilers like
lcc-win32 are supporting old style declarations.
 Errors and warnings 17

The assignment in C is an expression, i.e. it can appear within a more complicated expression:
if ( (x =z) > 13) z = 0;
This means that the compiler generates code for assigning the value of z to x, then it compares
this value with 13, and if the relationship holds, the program will set z to zero.

1.7 Errors and warnings

It is very rare that we type in a program and that it works at the first try. What happens, for
instance, if we forget to close the main function with the corresponding curly brace? We erase
the curly brace above and we try:
h:\lcc\examples>lcc args.c
Error args.c: 15 syntax error; found `end of input' expecting `}'
1 errors, 0 warnings
Well, this is at least a clear error message. More difficult is the case of forgetting to put the
semi-colon after the declaration of count, in the line 3 in the program above:
D:\lcc\examples>lcc args.c
Error args.c: 6 syntax error; found `for' expecting `;'
Error args.c: 6 skipping `for'
Error args.c: 6 syntax error; found `;' expecting `)'
Warning args.c: 6 Statement has no effect
Error args.c: 6 syntax error; found `)' expecting `;'
Error args.c: 6 illegal statement termination
Error args.c: 6 skipping `)'
6 errors, 1 warnings

D:\lcc\examples>

We see here a chain of errors, provoked by the first. The compiler tries to arrange things by
skipping text, but this produces more errors since the whole “for” construct is not understood.
Error recovering is quite a difficult undertaking, and lcc-win32 isn’t very good at it. So the
best thing is to look at the first error, and in many cases, the rest of the error messages are just
consequences of it.37
Another type of errors can appear when we forget to include the corresponding header file. if
we erase the #include <stdio.h> line in the args program, the display looks like this:
D:\lcc\examples>lcc args.c
Warning args.c: 7 missing prototype for printf
0 errors, 1 warnings
This is a warning. The printf function will be assumed to return an integer, what, in this case,
is a good assumption. We can link the program and the program works. It is surely NOT a
good practice to do this, however, since all argument checking is not done for unknown func-
tions; an error in argument passing will pass undetected and will provoke a much harder type
of error: a run time error.
In general, it is better to get the error as soon as possible. The later it is discovered, the more
difficult it is to find it, and to track its consequences. Do as much as you can to put the C com-
piler in your side, by using always the corresponding header files, to allow it to check every
function call for correctness.

37. You will probably see another display in your computer if you are using a recent version of lcc-
win32. I improved error handling when I was writing this tutorial…
18 C Tutorial

The compiler gives two types of errors, classified according to their severity: a warning, when
the error isn’t so serious that doesn’t allow the compiler to finish its task, and the hard errors,
where the compiler doesn’t generate an executable file and returns an error code to the calling
environment.
We should keep in mind however that warnings are errors too, and try to get rid from them.
The compiler uses a two level “warning level” variable. In the default state, many warnings
aren’t displayed to avoid cluttering the output. They will be displayed however, if you ask
explicitly to raise the warning level, with the option –A. This compiler option will make the
compiler emit all the warnings it would normally suppress. You call the compiler with lcc –
A <filename>, or set the corresponding button in the IDE, in the compiler configuration
tab.
Errors can appear in later stages of course. The linker can discover that you have used a proce-
dure without giving any definition for it in the program, and will stop with an error. Or it can
discover that you have given two different definitions, maybe contradictory to the same identi-
fier. This will provoke a link time error too.
But the most dreaded form of errors are the errors that happen at execution time, i.e. when the
program is running. Most of these errors are difficult to detect (they pass through the compila-
tion and link phases without any warnings…) and provoke the total failure of the software.
The C language is not very “forgiving” what programmer errors concerns. Most of them will
provoke the immediate stop of the program with an exception, or return completely nonsense
results. In this case you need a special tool, a debugger, to find them. Lcc-win32 offers you
such a tool, and you can debug your program by just pressing F5 in the IDE.

Summary:
• Syntax errors (missing semi-colons, or similar) are the easiest of all errors to correct.
• The compiler emits two kinds of diagnostic messages: warnings and errors.
• You can rise the compiler error reporting with the –A option.
• The linker can report errors when an identifier is defined twice or when an identifier is
missing a definition.
• The most difficult errors to catch are run time errors, in the form of traps or incorrect
results.

1.8 Reading from a file

For a beginner, it is very important that the basic libraries for reading and writing to a stream,
and the mathematical functions are well known. Here is an example of a function that will read
a text file, counting the number of characters that appear in the file.
A program is defined by its specifications. In this case, we have a general goal that can be
expressed quickly in one sentence: “Count the number of characters in a file”. Many times, the
specifications aren’t in a written form, and can be even completely ambiguous. What is impor-
tant is that before you embark in a software construction project, at least for you, the specifica-
tions are clear.
#include <stdio.h> (1)
int main(int argc,char *argv[]) (2)
{
int count=0;// chars read (3)
FILE *infile; (4)
int c; (5)
 Reading from a file 19

infile = fopen(argv[1],"r");(6)
c = fgetc(infile); (7)
while (c != EOF) { (8)
count++; (9)
c = fgetc(infile); (10)
}
printf("%d\n",count); (11)
return 0;
}

1) We include the standard header “stdio.h” again. Here is the definition of a FILE structure.

2) The same convention as for the “args” program is used here. The main arguments will not
be explained again.

3) We set at the start, the count of the characters read to zero. Note that we do this in the
declaration of the variable. C allows you to define an expression that will be used to
initialize a variable.38

4) We use the variable “infile” to hold a FILE pointer. Note the declaration for a pointer:
<type> * identifier; the type in this case, is a complex structure (composite type)
called FILE and defined in stdio.h. We do not use any fields of this structure, we just
assign to it, using the functions of the standard library, and so we are not concerned about
the specific layout of it. Note that a pointer is just the machine address of the start of that
structure, not the structure itself. We will discuss pointers extensively later.

5) We use an integer to hold the currently read character.

6) We start the process of reading characters from a file first by opening it. This operation
establishes a link between the data area of your hard disk, and the FILE variable. We pass
to the function fopen an argument list, separated by commas, containing two things: the
name of the file we wish to open, and the mode that we want to open this file, in our
example in read mode. Note that the mode is passed as a character string, i.e. enclosed in
double quotes.

7) Once opened, we can use the fgetc function to get a character from a file. This function
receives as argument the file we want to read from, in this case the variable “infile”, and
returns an integer containing the character read.

8) We use the while statement to loop reading characters from a file. This statement has the
general form: while (condition) { … statements… }. The loop body will be executed for so
long as the condition holds. We test at each iteration of the loop if our character is not the
special constant EOF (End Of File), defined in stdio.h.

9) We increment the counter of the characters. If we arrive here, it means that the character
wasn’t the last one, so we increase the counter.

10) After counting the character we are done with it, and we read into the same variable a new
character again, using the fgetc function.

38. There is another construct in this line, a comment. Commentaries are textual remarks left by the
programmer for the benefit of other human readers, and are ignored by the compiler. We will come back to
commentaries in a more formal manner later.
20 C Tutorial

11) If we arrive here, it means that we have hit EOF, the end of the file. We print our count in
the screen and exit the program returning zero, i.e. all is OK. By convention, a program
returns zero when no errors happened, and an error code, when something happened that
needs to be reported to the calling environment.
Now we are ready to start our program. We compile it, link it, and we call it with:
h:\lcc\examples> countchars countchars.c
288
We have achieved the first step in the development of a program. We have a version of it that
in some circumstances can fulfill the specifications that we received.
But what happens if we just write
h:\lcc\examples> countchars

We get the following box that many of you have already seen several times:39

Why?
Well, let’s look at the logic of our program. We assumed (without any test) that argv[1] will
contain the name of the file that we should count the characters of. But if the user doesn’t sup-
ply this parameter, our program will pass a nonsense argument to fopen, with the obvious
result that the program will fail miserably, making a trap, or exception that the system reports.
We return to the editor, and correct the faulty logic. Added code is in bold.
#include <stdio.h>
#include <stdlib.h> (1)
int main(int argc,char *argv[])
{
int count=0;// chars read
FILE *infile;
int c;

if (argc < 2) { (2)

printf("Usage: countchars <file name>\n");
exit(1); (3)
}
infile = fopen(argv[1],"r");
c = fgetc(infile);
while (c != EOF) {
count++;
c = fgetc(infile);
}
printf("%d\n",count);
return 0;
}

39. This is the display under Windows NT. In other systems like Linux for instance, you will get a “Bus
error” message.
 Reading from a file 21

1) We need to include <stdlib.h> to get the prototype declaration of the exit() function that
ends the program immediately.

2) We use the conditional statement “if” to test for a given condition. The general form of it is:
if (condition) { … statements… } else { … statements… }.

3) We use the exit function to stop the program immediately. This function receives an integer
argument that will be the result of the program. In our case we return the error code 1. The
result of our program will be then, the integer 1.
Now, when we call countchars without passing it an argument, we obtain a nice message:
h:\lcc\examples> countchars
Usage: countchars <file name>
This is MUCH clearer than the incomprehensible message box from the system isn’t it?
Now let’s try the following:
h:\lcc\examples> countchars zzzssqqqqq
And we obtain the dreaded message box again.
Why?
Well, it is very unlikely that a file called “zzzssqqqqq” exists in the current directory. We have
used the function fopen, but we didn’t bother to test if the result of fopen didn’t tell us that the
operation failed, because, for instance, the file doesn’t exist at all!
A quick look at the documentation of fopen (that you can obtain by pressing F1 with the cur-
sor over the “fopen” word in Wedit) will tell us that when fopen returns a NULL pointer (a
zero), it means the open operation failed. We modify again our program, to take into account
this possibility:
#include <stdio.h>
#include <stdlib.h>
int main(int argc,char *argv[])
{
int count=0;// chars read
FILE *infile;
int c;

if (argc < 2) {
printf("Usage: countchars <file name>\n");
exit(1);
}
infile = fopen(argv[1],"r");
if (infile == NULL) {
printf("File %s doesn’t exist\n",argv[1]);
exit(1);
}
c = fgetc(infile);
while (c != EOF) {
count++;
c = fgetc(infile);
}
printf("%d\n",count);
return 0;
}
We try again:
H:\lcc\examples> lcc countchars.c
H:\lcc\examples> lcclnk countchars.obj
22 C Tutorial

H:\lcc\examples> countchars sfsfsfsfs

File sfsfsfsfs doesn't exist
H:\lcc\examples>
Well this error checking works. But let’s look again at the logic of this program.
Suppose we have an empty file. Will our program work?
If we have an empty file, the first fgetc will return EOF. This means the whole while loop will
never be executed and control will pass to our printf statement. Since we took care of initializ-
ing our counter to zero at the start of the program, the program will report correctly the num-
ber of characters in an empty file: zero.
Still it would be interesting to verify that we are getting the right count for a given file. Well
that’s easy. We count the characters with our program, and then we use the DIR directive of
windows to verify that we get the right count.
H:\lcc\examples>countchars countchars.c
466
H:\lcc\examples>dir countchars.c

07/01/00 11:31p 492 countchars.c

1 File(s) 492 bytes
Wow, we are missing 492-466 = 26 chars!
Why?
We read again the specifications of the fopen function. It says that we should use it in read
mode with “r” or in binary mode with “rb”. This means that when we open a file in read mode,
it will translate the sequences of characters \r (return) and \n (new line) into ONE character.
When we open a file to count all characters in it, we should count the return characters too.
This has historical reasons. The C language originated in a system called UNIX, actually, the
whole language was developed to be able to write the UNIX system in a convenient way. In
that system, lines are separated by only ONE character, the new line character.
When the MSDOS system was developed, dozens of years later than UNIX, people decided to
separate the text lines with two characters, the carriage return, and the new line character. This
provoked many problems with software that expected only ONE char as line separator. To
avoid this problem the MSDOS people decided to provide a compatibility option for that case:
fopen would by default open text files in text mode, i.e. would translate sequences of \r\n into
\n, skipping the \r.
Conclusion:
Instead of opening the file with fopen(argv[1], “r“); we use fopen(argv[1],
“rb“);, i.e. we force NO translation. We recompile, relink and we obtain:
H:\lcc\examples> countchars countchars.c
493

H:\lcc\examples> dir countchars.c

07/01/00 11:50p 493 countchars.c

1 File(s) 493 bytes
Yes, 493 bytes instead of 492 before, since we have added a “b” to the arguments of fopen!
Still, we read the docs about file handling, and we try to see if there are no hidden bugs in our
program. After a while, an obvious fact appears: we have opened a file, but we never closed it,
i.e. we never break the connection between the program, and the file it is reading. We correct
this, and at the same time add some commentaries to make the purpose of the program clear.
 Commentaries 23

/*---------------------------------------------------------
Module: H:\LCC\EXAMPLES\countchars.c
Author: Jacob
Project: Tutorial examples
State: Finished
Creation Date: July 2000
Description: This program opens the given file, and
prints the number of characters in it.
----------------------------------------------------------*/
#include <stdio.h>
#include <stdlib.h>
int main(int argc,char *argv[])
{
int count=0;
FILE *infile;
int c;

if (argc < 2) {
printf("Usage: countchars <file name>\n");
exit(1);
}
infile = fopen(argv[1],"rb");
if (infile == NULL) {
printf("File %s doesn't exist\n",argv[1]);
exit(1);
}
c = fgetc(infile);
while (c != EOF) {
count++;
c = fgetc(infile);
}
fclose(infile);
printf("%d\n",count);
return 0;
}
The skeleton of the commentary above is generated automatically by the IDE. Just right-click
somewhere in your file, and choose “edit description”.

Summary:
• A program is defined by its specifications. In this example, counting the number of
characters in a file.
• A first working version of the specification is developed. Essential parts like error
checking are missing, but the program “works” for its essential function.
• Error checking is added, and test cases are built.
• The program is examined for correctness, and the possibility of memory leaks, unclosed
files, etc., is reviewed. Comments are added to make the purpose of the program clear,
and to allow other people know what it does without being forced to read the program
text.

1.9 Commentaries
The writing of commentaries, apparently simple, is, when you want to do it right, quite a diffi-
cult task. Let’s start with the basics.
Commentaries are introduced in two forms:
24 C Tutorial

Two slashes // introduce a commentary that will last until the end of the line. No space
should be present between the first slash and the second one.
A slash and an asterisk /* introduce a commentary that can span several lines and is only ter-
minated by an asterisk and a slash, */. The same rule as above is valid here too: no space
should appear between the slash and the asterisk, and between the asterisk and the slash to be
valid comment delimiters.
Examples:
// This is a one-line commentary. Here /* are ignored anyway.
/* This is a commentary that can span several lines. Note that here the
two slashes // are ignored too */
This is very simple, but the difficulty is not in the syntax of commentaries, of course, but in
their content. There are several rules to keep in mind:
Always keep the commentaries current with the code that they are supposed to comment.
There is nothing more frustrating than to discover that the commentary was actually mislead-
ing you, because it wasn’t updated when the code below changed, and actually instead of help-
ing you to understand the code it contributes further to make it more obscure.
Do not comment what are you doing but why. For instance:
record++; // increment record by one
This comment doesn’t tell anything the C code doesn’t tell us
anyway.
record++; //Pass to next record.
// The boundary tests are done at
// the beginning of the loop above
This comment brings useful information to the reader.
At the beginning of each procedure, try to add a standard comment describing the purpose of
the procedure, inputs/outputs, error handling etc.40
At the beginning of each module try to put a general comment describing what this module
does, the main functions etc.
Note that you yourself will be the first guy to debug the code you write. Commentaries will
help you understand again that hairy stuff you did several months ago, when in a hurry.

1.9.1 Standard comments

The editor of lcc-win32 provides a «Standard comments» feature. There are two types of com-
ments supported: comments that describe a function, and comments that apply to a whole file.
This comments are maintained by the editor, that displays a simple interface for editing them.

1.9.1.1 Describing a function

You place the mouse anywhere within the body of a function and you click the right mouse
button. A context menu appears that offers you to edit the description of the current function.
The interface that appears by choosing this option looks like this:

40. The IDE of lcc-win32 helps you by automatic the construction of those comments. Just press, “edit
description” in the right mouse button menu.
 Commentaries 25

There are several fields that you should fill:

1) Purpose. This should explain what this function does, and how it does it.

2) Inputs: Here you should explain how the interface of this function is designed: the
arguments of the function and global variables used if any.

3) Outputs. Here you should explain the return value of the function, and any globals that are
left modified.

4) Error handling. Here you should explain the error return, and the behavior of the function in
case of an error occurring within its body.
For the description provided in the screen shot above, the editor produces the following out-
put:
/*---------------------------------------------------------------
Procedure: multiple ID:1
Purpose: Compiles a multiple regular expression
Input: Reads input from standard input
Output: Generates a regexp structure
Errors: Several errors are displayed using the "complain"
function
-----------------------------------------------------------------*/
void multiple(void)
{
This comment will be inserted in the interface the next time you ask for the description of the
function.
26 C Tutorial

1.9.1.2 Describing a file

In the same context menu that appears with a right click, you have another menu item that says
«description of file.c», where «file.c» is the name of the current file.
This allows you to describe what the file does. The editor will add automatically the name of
the currently logged on user, most of the time the famous «administrator». The output of the
interface looks like this:

/*-----------------------------------------------------------------
Module: d:\lcc\examples\regexp\try.c
Author: ADMINISTRATOR
Project:
State:
Creation Date:
Description: This module tests the regular expressions
package. It is self-contained and has a main()
function that will open a file given in the
command line that is supposed to conttain
several regular expressions to test. If any
error are discovered, the results are printed
to stdout.
-----------------------------------------------------------------*/
As with the other standard comment, the editor will re-read this comment into the interface.
This features are just an aid to easy the writing of comments, and making them uniform and
structured. As any other feature, you could use another format in another environment. You
could make a simple text file that would be inserted where necessary and the fields would be
tailored to the application you are developing. Such a solution would work in most systems
too, since most editors allow you to insert a file at the insertion point.

1.10 An overview of the whole language

Let’s formalize a bit what we are discussing. Here are some tables that you can use as refer-
ence tables. We have first the words of the language, the statements. Then we have a dictio-
 An overview of the whole language 27

nary of some sentences you can write with those statements, the different declarations and
control-flow constructs. And in the end is the summary of the pre-processor instructions. I
have tried to put everything hoping that I didn’t forget something.
You will find in the left column a more or less formal description of the construct, a short
explanation in the second column, and an example in the third. In the first column, this words
have a special meaning: “id”, meaning an identifier, “type” meaning some arbitrary type and
“expr” meaning some arbitrary C expression.
I have forced a page break here so that you can print these pages separately, when you are
using the system.
28 C Tutorial

1.10.1 Statements

Expression Meaning and value of result Example

The value associated with that identifier. (see “A

identifier id
closer view” on page 37.)

The value defined with this constant (see

“Constants.” on page 37.).

Integer constant. 45 45L 45LL

45.9 45.9f
constant Floating constant
45.9L

character constant ‘A’ L’A’

“Hello“
string literal
L“Hello“

{ constants } Define tables or structure data {1,67}

Integer constants 45

long integer constants 45L

long long (64 bits) integer constant 45LL

octal constant (base 8) introduced with a leading 055 (This is 45 in

integer constants
zero base 8)

0x2d (this is 45 in
Hexadecimal constant introduced with 0x
hexa)

Binary constant introduced with 0b. This is an lcc- 0b101101 (this is 45

win32 extension. in decimal)

45.9 or
double precision constant
4.59e2

45.9f or
floating
Float (single precision) constant 4.59e2f
constants

45.9L or
long double constant
4.59e2L

character
char enclosed in simple quotes ’a’ or ’8’
constant

string literals enclosed in double quotes "a string"

Access the position “index” of the given array.

Indexes start at zero (see “Within the string, the
Array [index ] Table[45]
following abbreviations are recognized:” on
page 37.)
 An overview of the whole language 29

Table[34][23]
This access the
Access the n dimensional array using the indexes i1,
Array[i1][i2]
i2, … in..See “Array syntax.” on page 38. 35th line, 24th
position of
Table

Call the function “fn” and pass it the comma

fn ( args ) separated argument list «args». see “Function call printf(“%d“,5)
syntax” on page 38.

See “Functions with variable number of arguments.”

fn (arg, ...)
on page 39.

Call the function whose machine address is in the

(*fn)(args)
pointer fn.

struct.field Access the member of the structure Customer.Name

Access the member of the structure through a

struct->field Customer->Name
pointer

Assign to the variable1 the value of the right hand

var = value side of the equals sign. See “Assignment.” on a = 45
page 39.

Equivalent to expression = expression + 1.

expression++ Increment expression after using its value. See a = i++
“Postfix” on page 40..

Equivalent to expression = expression – 1.

expression-- Decrement expression after using its value. see a = i--
“Postfix” on page 40.

Equivalent to expression = expression+1. Increment

++expression a = ++I
expression before using its value.

Equivalent to Expression = expression – 1.

--expression a = --i
Decrement expression before using it.

Return the machine address of object. The type of

& object &i
the result is a pointer to object.

Access the contents at the machine address stored in

* pointer *pData
the pointer. .See “Indirection” on page 47.

- expression Subtract expression from zero, i.e. change the sign. -a

Bitwise complement expression. Change all 1 bits to

~ expression ~a
0 and all 0 bits to 1.

Negate expression: if expression is zero, !expression

! expression becomes one, if expression is different than zero, it !a
becomes zero.

Return the size in bytes of expr. .see “sizeof.” on

sizeof(expr) sizeof(a)
page 41.

Change the type of expression to the given type.

(type) expr (int *)a
This is called “cast”.
30 C Tutorial

expr * expr Multiply a*b

expr / expr Divide a/b

expr % expr Divide first by second and return the remainder a%b

expr + expr Add a+b

Subtract expr2 from expr1. .see “Subtraction.” on

expr1 - expr2 a-b
page 40.

expr1 << expr2 Shift left expr1 expr2 bits. a << b

expr1 >> expr2 Shift right expr1 expr2 bits. a >> b

expr1 < expr2 1 if expr1 is smaller than expr2, zero otherwise a < b

1 if expr1 is smaller or equal than expr2, zero

expr1 <= expr2 a <= b
otherwise

1 if expr1 is greater or equal than expr2, zero

expr1 >= expr2 a >= b
otherwise

expr1 > expr2 1 if expr2 is greater than expr2, zero otherwise a > b

expr1 == expr2 1 if expr1 is equal to expr2, zero otherwise a == b

expr1 != expr2 1 if expr1 is different from expr2, zero otherwise a != b

Bitwise AND expr1 with expr2. See “Bitwise

expr1 & expr2 a&8
operators” on page 46.

Bitwise XOR expr1 with expr2. See “Bitwise

expr1 ^ expr2 a^b
operators” on page 46.

Bitwise OR expr1 with expr2. See “Bitwise

expr1 | expr2 a|16
operators” on page 46.

Evaluate expr1. If its result is zero, stop evaluating

a < 5 && a > 0
the whole expression and set the result of the whole
This will be 1 if “a” is
expression to zero. If not, continue evaluating expr2.
expr1 && expr2 between 1 to 4. If a >=
The result of the expression is the logical AND of
5 the second test is not
the results of evaluating each expression. See
performed.
“Logical operators” on page 46.

Evaluate expr1. If the result is one, stop evaluating

the whole expression and set the result of the
a == 5 ||a == 3
expression to 1. If not, continue evaluating expr2.
expr1 || expr2 This will be 1 if either
The result of the expression is the logical OR of the
a is 5 or 3
results of each expression. See “Logical operators”
on page 46.

If expr evaluates to non-zero (true), return val1, a= b ? 2 : 3

expr ? val1:val2 otherwise return val2. a will be 2 if b is true, 3
see “Conditional operator.” on page 40. otherwise

expr *= expr1 Multiply expr by expr1 and store the result in expr a *= 7

expr /= expr1 Divide expr by expr1 and store the result in expr a /= 78
 An overview of the whole language 31

Calculate the remainder of expr % expr1 and store

expr %= expr1 a %= 6
the result in expr

expr += expr1 Add expr1 with expr and store the result in expr a += 6

expr -= expr1 Subtract expr1 from expr and store the result in expr a -= 76

Shift left expr by expr1 bits and store the result in

expr <<= expr1 a <<= 6
expr

Shift right expr by expr1 bits and store the result in

expr >>= expr1 a >>= 7
expr

Bitwise and expr with expr1 and store the result in

expr &= expr1 a &= 32
expr

Bitwise xor expr with expr1 and store the result in

expr ^= expr1 a ^= 64
expr

Bitwise or expr with expr1 and store the result in

expr |= expr1 a |= 128
expr.

a=7,b=8
Evaluate expr, then expr1 and return the result of
expr , expr1 The result of
evaluating the last expression, in this case expr1
this is 8

; Null statement ;

1. Variable can be any value that can be assigned to: an array element or other constructs like *ptr =
5. In technical language this is called an “lvalue”.
32 C Tutorial

1.10.2 Declarations41

Declaration Meaning Example

Identifier will have the specified type within this

scope. In a local scope its value is
type id; int a;
undertemined. In a global scope, its initial value
is zero, at program start.

Identifier will be a pointer to objects of the given

int *pa;
type. You add an asterisk for each level of
type * id; pa will be a pointer to
indirection. A pointer to a pointer needs two
integers
asterisks, etc.

Identifier will be a reference to a single object of int &ra = a;

type & id = expr; the given type. References must be initialized pa will be a reference to an
1
immediately after their declaration. See integer.

Identifier will be an array of expr elements of

the given type. The expression must evaluate to int *ptrArray[56];
type id[expr] a compile time constant or to a constant Array of 56 int
expression that will be evaluated at run time. In pointers.
the later case this is a variable length array.

Define a new type-name for the old type. see typedef unsigned
typedef old new
“typedef.” on page 41. int uint;

Try to store the identifier in a machine register.

The type of identifier will be equivalent to
register id; register int f;
signed integer if not explicitly specified. see
“register.” on page 41.

The definition of the identifier is in another extern int

extern type id;
module. No space is reserved. frequency;

Make the definition of identifier not accessible

static type id static int f;
from other modules.

struct coord {
struct id { Define a compound type composed by the
int x;
declarations enumeration of fields enclosed within curly
int y;
} braces.
};

Within a structure field declaration, declare “id” unsigned n:4

type id:n as a sequence of n bits of type “type”. See “bit n is an unsigned
fields” on page 42. int of 4 bits

union dd {
union id { Reserve storage for the biggest of the declared
double d;
declarations types and store all of them in the same place.
int id[2];
}; see “Unions.” on page 41.
};

enum identifier { Define an enumeration of comma-separated enum color {

enum list identifiers assigning them some integer value. red,green,blue
} see “enum.” on page 41. };
 An overview of the whole language 33

Declare that the given identifier can’t be

const type identifier; changed (assigned to) within this scope. see const int a;
“const.” on page 42.

When applied to integer types do not use the unsigned char a =

unsigned int-type
sign bit. see “unsigned.” on page 42. 178;

Declare that the given object changes in ways volatile int

volatile type identifier
unknown to the implementation. hardware_clock;

Declare the prototype for the given function.

double sqrt(double
type id (args); The arguments are a comma separated list. see
x);
“Prototypes.” on page 41.

Declare a function pointer called “id” with the

type (*id)(args); void (*fn)(int)
given return type and arguments list

id : Declare a label. lab1:

type fn(args) {
...
Definition of a function with return type <type> int add1(int x)
statements
and arguments <args> . { return x+1;}
...
}

Redefine one of the operators like +, * or others

T operator +(T a,
operator opname so that instead of doing the normal operation,
T b) {
(args) { this function is called instead. This is an
... statements
} extension to the C language proposed by lcc-
}
win32 and is NOT in the language standard.

This is a qualifier that applies to functions. If double inline

present, it can be understood by the compiler as overPi(double a)
inline a specification to generate the fastest function {
call possible, generally by means of replicating return a/3.14159;
the function body at each call site. }

1. This is an extension of lcc-win32 and not part of the C standard. It is widely used in other
languages like C++.

41. Lcc-win32 doesn’t yet implement the keyword restrict.

34 C Tutorial

1.10.3 Pre-processor

Double slashes introduce comments up

// commentary to the end of the line.see “Comments” // comment
on page 44.

Slash star introduces a commentary until

/*commentary */ the sequence star slash */ is seen. see /* comment */
“Comments” on page 44.

Replace all appearances of the given

identifier by the corresponding
#define id text #define TAX 6
expression. See “The preprocessor in
detail” on page 118.

Define a macro with n arguments. When #define max(a,b)

#define macro(a,b) used, the arguments are lexically ((a)<(b)?
replaced within the macro. See page 119 (b):(a))1

Erase from the pre-processor tables the

#undef id #undef TAX
given identifier.

Insert the contents of the given file from

#include <header.h> the standard include directory into the #include <stdio.h>
program text at this position.

Insert the contents of the given file from

#include "header.h" #include “foo.h“
the current directory.

If the given identifier is defined (using

#ifdef id #define) include the following lines. #ifdef TAX
Else skip them. See page 120.

#ifndef id The contrary of the above #ifnef TAX

Evaluate expression and if the result is

TRUE, include the following lines. Else
#if (expr) #if (TAX==6)
skip all lines until finding an #else or
#endif

#else the else branch of an #if or #ifdef #else

#elif Abbreviation of #else #if #elif

End an #if or #ifdef preprocessor

#endif #endif
directive statement

If the given identifier is #defined, return

defined (id) #if defined(max)
1, else return 0.

## Token concatenation a##b ab

#line nn Set the line number to nn #line 56

#file "foo.c" Set the file name #file “ff.c”

#error errmsg Show the indicated error to the user #error "undefined cpu"
 An overview of the whole language 35

#pragma instructions Special compiler directives2 #pragma optimize(on)

Special compiler directives. This is a

_Pragma(string) _Pragma("optimize (on)");
C99 feature.

If a \ appears at the end of a line just

before the newline character, the line /\
\ and the following line will be joined by * this is a comment
the preprocessor and the \ character will */
be eliminated.

Replace this token by the current line printf("error line %d\n",

__LINE__
number __LINE__);

Replace this token by the current file printf("error file %s\n",

__FILE__
name __FILE__);

Replace this token by the name of the printf("fn %s\n",__func__

__ func__
current function being compiled. );

STDC Defined as 1 #if STDC

Defined as 1 This allows you to

__LCC__ conditionally include or not code for lcc- #if __LCC__
win32.

1. The parentheses ensure the correct evaluation of the macro.

2. The different pragma directives of lcc-win32 are explained in the user’s manual.

1.10.4 Windows specific defined symbols

WIN32 #defined as 1 #if WIN32

_WIN32 #defined as 1 #if _WIN32

Evaluates to the version of windows you

WINVER #if WINVER == 5
are running

Evaluates to the version of the internet

_WIN32_IE explorer software installed in your #if _WIN32_IE > 0x500
system
36 C Tutorial

1.10.5 Control-flow

if (expression) { block} If the given expression evaluates to something different than zero execute the
else { statements of the following block. Else, execute the statements of the block
block following the else keyword. The else statement is optional. Note that a single
} statement can replace blocks.

while (expression) { If the given expression evaluates to something different than zero, execute the
... statements ... statements in the block, and return to evaluate the controlling expression again.
} Else continue after the block. See “while” on page 11.

do { ... statements ... Execute the statements in the block, and afterwards test if condition is true. If
} while (condition); that is the case, execute the statements again.See “do” on page 11.

Execute unconditionally the expressions in the init statement. Then evaluate the
for(init;test;incr) {
test expression, and if evaluates to true, execute the statements in the block
... statements ...
following the for. At the end of each iteration execute the incr statements and
}
evaluate the test code again. See “for” on page 11.

switch (expression) { Evaluate the given expression. Use the resulting value to test if it matches any of
case int-expr: the integer expressions defined in each of the ‘case’ constructs. If the
statements ... comparison succeeds, execute the statements in sequence beginning with that
break; case statement.
default: If the evaluation of expression produces a value that doesn’t match any of the
statements cases and a “default” case was specified, execute the default case statements in
} sequence. See “Switch statement.” on page 44.

goto label Transfer control unconditionally to the given label.

Within the scope of a for/do/while loop statement, continue with the next
continue iteration of the loop, skipping all statements until the end of the loop.See “break
and continue statements” on page 43.

break Stop the execution of the current do/for/while loop statement.

End the current function and return control to the calling one. The return value of
return expression the function (if any) can be specified in the expression following the return
keyword.

1.10.6 Windows specific syntax

Use the stdcall calling convention for this function or

int _stdcall
_stdcall function pointer: the called function cleans up the
fn(void);
stack. See “stdcall.” on page 43.

__declspec Export this identifier in a DLL to make it visible from int __declspec(
(dllexport) outside. dllexport) fn(int);

#defined as long long for compatibility reasons with

__int64 __int64 big;
Microsoft’s compiler.
 An overview of the whole language 37

1.10.7 A closer view

Let’s explain a bit the terms above. The table gives a compressed view of C. Now let’s see
some of the details.

1.10.7.1 Identifiers.
An identifier is a sequence of non digit characters (including the underscore _, the lowercase
and uppercase Latin letters, and other characters) and digits. Lowercase and uppercase letters
are distinct. An identifier never starts with a digit. There is no specific limit on the maximum
length of an identifier but lcc-win32 will give up at 255 chars.
Identifiers are the vocabulary of your software. When you create them, give a mnemonic that
speaks about the data stored at that location.
Anonymous identifiers (or counters) are usually the one letters ’i’, or c for char, etc. I think the
habit of using i, j, k is quite ancient, maybe inherited from fortran and physics.

1.10.7.2 Constants.
An integer constant begins with a digit, but has no period or exponent part. It may have a pre-
fix that specifies its base and a suffix that specifies its type. A decimal constant begins with a
nonzero digit and consists of a sequence of decimal digits. An octal constant consists of the
prefix 0 optionally followed by a sequence of the digits 0 through 7 only. A hexadecimal con-
stant consists of the prefix 0x or 0X followed by a sequence of the decimal digits and the let-
ters a (or A) through f (or F) with values 10 through 15 respectively. Here are various
examples of integer constants:
1645L (long)

0xF98A (hexa)

2634455LL (long long)

5488UL (unsigned long)

For floating constants, the convention is either to use a decimal point (1230.0) or scientific
notation (in the form of 1.23e3). They can have the suffix ‘F’ (or ‘f’) to mean that they are
float constants, and not double constants as it is implicitly assumed when they have no suffix.
For character string constants, they are enclosed in double quotes. If immediately before the
double quote there is an "L" it means that they are double byte strings. Example:
L"abc"
This means that the compiler will convert this character string into a wide character string and
store the values as double byte character string instead of just ASCII characters.
To include a double quote within a string it must be preceded with a backslash. Example:
"The string \"the string\" is enclosed in quotes"
Note that strings and numbers are completely different data types. Even if a string contains
only digits, it will never be recognized as a number by the compiler: "162" is a string, and to
convert it to a number you must explicitly write code to do the transformation.
Within the string, the following abbreviations are recognized:

Abbreviation Meaning Value

\n New line 10
38 C Tutorial

\r carriage return 12

\b backspace 8

\v vertical tab 11

\t tab 9

\f form feed 12

\e escape 27

\a bell 7

Insert at the current position the Any, since any digit can be
character with the integer value of entered. Example: "ABC\xA" is
\x<hex digits>
the hexadecimal digits. equivalent to "ABC\n"

The same as the \x case above, but

with values entered as 3 octal digits, Any. Example: The string
\<octal number> i.e. numbers in base 8. Note that no "ABC\012" is equivalent to
special character is needed after the "ABC\n"
backslash. The octal digits start
immediately after it.

Character string constants that are too long to write in a single line can be entered in two ways:
char *a = "This is a long string that at the end has a backslash \
that allows it to go on in the next line";
Another way, introduced with C99 is:
char *a = "This is a long string written"
"in two lines";
Note too that character string constants should not be modified by the program. Lcc-win32
stores all character string constants once, even if they appear several times in the program text.
For instance if you write:
char *a = “abc”;
char *b = “abc”;
Both a and b will point to the SAME string, and if either is modified the other will not retain
the original value.

1.10.7.3 Array syntax.

Here are various examples for using arrays.
int a[45]; // Array of 45 elements
a[0] = 23; // Sets first element to 23;
a[a[0]] = 56; // Sets the 24th element to 56
a[23] += 56; // Adds 56 to the 24th element

1.10.7.4 Function call syntax

sqrt( hypo(6.0,9.0) ); // Calls the function hypo with
// two arguments and then calls
// the function sqrt with the
// result of hypo
 An overview of the whole language 39

An argument may be an expression of any object type. In preparing for the call to a function,
the arguments are evaluated, and each parameter is assigned the value of the corresponding
argument.
A function may change the values of its parameters, but these changes cannot affect the values
of the arguments. On the other hand, it is possible to pass a pointer to an object, and the func-
tion may change the value of the object pointed to.
A parameter declared to have array or function type is converted to a parameter with a pointer
type.
The order of evaluation of the actual arguments, and sub expressions within the actual argu-
ments is unspecified. For instance:
fn( g(), h(), m());
Here the order of the calls to the functions g(), h() and m() is unspecified.

1.10.7.5 Functions with variable number of arguments.

To access the unnamed, extra arguments you should include <stdarg.h>. To access the addi-
tional arguments, you should execute the va_start, then, for each argument, you execute a
va_arg. Note that if you have executed the macro va_start, you should always execute
the va_end macro before the function exits. Here is an example that will add any number of
integers passed to it. The first integer passed is the number of integers that follow.
#include <stdarg.h>

int va_add(int numberOfArgs, ...)

{
va_list ap;
int n = numberOfArgs;
int sum = 0;

va_start(ap,numberOfArgs);
while (n--) {
sum += va_arg(ap,int);
}
va_end(ap);
return sum;
}
We would call this function with
va_add(3,987,876,567);
or
va_add(2,456,789);

1.10.7.6 Assignment.
An assignment has the left hand side of the equal’s sign, that must be a value that can be
assigned to, and the right hand side that can be any expression other than void.
int a = 789; // “a” is assigned 789
array[345] = array{123]+897; //An element of an array is assigned
Struct.field = sqrt(b+9.0); // A field of a structure is assigned
p->field = sqrt(b+9.0);
/* A field of a structure is assigned through a pointer. */
Within an assignment there is the concept of “L-value”, i.e. any assignable object. You can’t,
for instance, write:
5 = 8;
40 C Tutorial

The constant 5 can’t be assigned to. It is not an “L-value”, the “L” comes from the left hand
side of the equals sign of course. In the same vein we speak of LHS and RHS as abbreviations
for left hand side and right hand side of the equals sign in an assignment.

1.10.7.7 Postfix
This expressions increment or decrement the expression at their left side returning the old
value. For instance:
array[234] = 678;
a = array[234]++;
In this code fragment, the variable a will get assigned 678 and the array element 234 will have
a value of 679 after the expression is executed. In the code fragment:
array[234] = 678;
a = ++array[234];

The integer a and the array element at the 235th position will both have the value 679.
When applied to pointers, these operators increment or decrement the pointer to point to the
next or previous element. Note that if the size of the object those pointers point to is different
than one, the pointer will be incremented or decremented by a constant different than one too.

1.10.7.8 Subtraction.
When two pointers are subtracted they have to have the same type, and the result is the differ-
ence of the subscripts of the two array elements or, in other words, the number of elements
between both pointers. The size of the result is implementation-defined, and its type (a signed
integer type) is ptrdiff_t defined in the <stddef.h> header.42
When an integer expression is subtracted (or added) to a pointer, it means to increase the
pointer by that number of elements. For instance if the pointer is pointing to the 3rd element of
an array of structures, adding it 2 will provoke to advance the pointer to point to the 5th ele-
ment.

1.10.7.9 Conditional operator.

The first operand of the conditional expression is evaluated first. The second operand is evalu-
ated only if the first compares unequal to 0; the third operand is evaluated only if the first com-
pares equal to 0; the result of the whole expression is the value of the second or third operand
(whichever is evaluated), converted to the type described below.
If both the second and the third operand have an arithmetic type, the result of the expression
has that type. If both are structures, the result is a structure. If both are void, the result is void.
This expressions can be nested.
int a = (c == 66) ? 534 : 698;
the integer a will be assigned 534 if c is equal to 66, 698 otherwise.
struct b *bb = (bstruct == NULL) ? NULL : b->next;
If bstruct is different than NULL, the pointer bb will receive the “next” field of the structure,
otherwise bb will be set to NULL.

42. typedef int ptrdiff_t;

 An overview of the whole language 41

1.10.7.10 struct.
A structure or a union can’t contain another structure that hasn’t been fully specified, but they
can contain a pointer to such a structure since the size of any pointer is always fixed. To build
recursive structures like list you should specify a pointer to the structure, see “Lists” on
page 114.. For a detailed description of this keyword see “Structures” on page 70.

1.10.7.11 Unions.
You can store several values in a single memory location or a group of memory locations with
the proviso that they can’t be accessed at the same time of course. This allows you to reduce
the memory requirements of a structure, or to interpret a sequence of bits in a different fash-
ion. For a detailed discussion see “Unions” on page 75.

1.10.7.12 typedef.
The typedef keyword defines a name that can be used as a synonym for a type or derived type.
In contrast to the struct, union, and enum declarations, typedef declarations doesn’t introduce
new types — it introduces new names for existing types.

1.10.7.13 register.
This keyword is a recommendation to the compiler to use a machine register for storing the
values of this type. The compiler is free to follow or not this directive. The type must be either
an integer type or a pointer. If you use this declaration, note that you aren’t allowed to use the
address-of operator since registers do not have addresses. Lcc-win32 tries to honor your rec-
ommendations, but it is better not to use this declaration and leave the register usage to the
compiler.

1.10.7.14 sizeof.
The result of sizeof is normally a constant integer known when the compiler is running. For
instance sizeof(int) will yield under lcc-win32 the constant 4. In the case of a variable
length array however, the compiler can't know its size on advance, and it will be forced to gen-
erate code that will evaluate the size of the array when the program is running.

1.10.7.15 enum.
An enumeration is a sequence of symbols that are assigned integer values by the compiler. The
symbols so defined are equivalent to integers, and can be used for instance in switch state-
ments. The compiler starts assigning values at zero, but you can change the values using the
equals sign. An enumeration like enum { a,b,c}; will provoke that a will be zero, b will
be 1, and c will be 2. You can change this with enum {a=10,b=25,c=76};

1.10.7.16 Prototypes.
A prototype is a description of the return value and the types of the arguments of a function.
The general form specifies the return value, then the name of the function. Then, enclosed by
parentheses, come a comma-separated list of arguments with their respective types. If the
function doesn’t have any arguments, you should write ‘void’, instead of the argument list. If
the function doesn’t return any value you should specify void as the return type. At each call,
the compiler will check that the type of the actual arguments to the function is a correct one.
The compiler cannot guarantee, however, that the prototypes are consistent across different
compilation units. For instance if in file1.c you declar:
int fn(void);
then, the call
42 C Tutorial

fn();
will be accepted. If you then in file2.c you declare another prototype
void fn(int);
and then you use:
fn(6);
the compiler cannot see this, and the program will be in error, crashing mysteriously at run
time. This kind of errors can be avoided if you always declare the prototypes in a header file
that will be included by all files that use that function. Do not declare prototypes in a source
file if the function is an external one.

1.10.7.17 variable length array.

This arrays are based on the evaluation of an expression that is computed when the program is
running, and not when the program is being compiled. Here is an example of this construct:
int Function(int n)
{
int table[n];
…
}
The array of integers called “table” has n elements. This “n” is passed to the function as an
argument, so its value can’t be known in advance. The compiler generates code to allocate
space for this array when this function is entered.

1.10.7.18 const.
Constant values can’t be modified. The following pair of declarations demonstrates the differ-
ence between a ‘‘variable pointer to a constant value’’ and a ‘‘constant pointer to a variable
value’’.
const int *ptr_to_constant;
int *const constant_ptr;
The contents of any object pointed to by ptr_to_constant shall not be modified through
that pointer, but ptr_to_constant itself may be changed to point to another object. Simi-
larly, the contents of the int pointed to by constant_ptr may be modified, but
constant_ptr itself shall always point to the same location.

1.10.7.19 unsigned.
Integer types (long long, long, int, short and char) have the most significant bit reserved for the
sign bit. This declaration tells the compiler to ignore the sign bit and use the values from zero
the 2n for the values of that type. For instance, a signed short goes from –32767 to 32767, an
unsigned short goes from zero to 65535 (216). See the standard include file <stdint.h> for the
ranges of signed and unsigned integer types.

1.10.7.20 bit fields

A "bit field" is an unsigned or signed integer composed of some number of bits. Lcc-win32
will accept some other type than int for a bit field, but the real type of a bit field will be always
either "int" or "unsigned int".
For example, in the following structure, we have 3 bit fields, with 1, 5, and 7 bits:
struct S {
int a:1;
int b:5;
int c:7;
 An overview of the whole language 43

};
With lcc-win32 the size of this structure will be 4 with no special options. With maximum
packing (-Zp1 option) the size will be two.
When you need to leave some space between adjacent bit fields you can use the notation:
unsigned : n;
For example
struct S {
int a:1;
int b:5;
unsigned:10;
int c:7;
};
Between the bit fields a and b we leave 10 unused bits so that c starts at a 16 bit word bound-
ary.

1.10.7.21 stdcall.
Normally, the compiler generates assembly code that pushes each argument to the stack, exe-
cutes the “call” instruction, and then adds to the stack the size of the pushed arguments to
return the stack pointer to its previous position. The stdcall functions however, return the stack
pointer to its previous position before executing their final return, so this stack adjustment is
not necessary.
The reason for this is a smaller code size, since the many instructions that adjust the stack after
the function call are not needed and replaced by a single instruction at the end of the called
function.
Functions with this type of calling convention will be internally “decorated” by the compiler
by adding the stack size to their name after an “@” sign. For instance a function called fn with
an integer argument will get called fn@4. The purpose of this “decorations” is to force the pre-
vious declaration of a stdcall function so that always we are sure that the correct declarations
was seen, if not, the program doesn’t link.

1.10.7.22 break and continue statements

The break and continue statements are used to break out of a loop or switch, or to continue a
loop at the test site. They can be explained in terms of the goto statement:
while (condition != 0) {
doSomething();
if (condition == 0)
break;
doSomethingElse();
}
is equivalent to:
while (condition != 0) {
doSomething();
if (condition == 0)
goto lab1;
doSomethingElse();
}
lab1:
The continue statement can be represented in a similar way:
while (condition != 0) {
doSomething();
44 C Tutorial

if (condition == 25)
continue;
doSomethingElse();
}
is equivalent to:
restart:
while (condition != 0) {
doSomething();
if (condition == 25)
goto restart;
doSomethingElse();
}
The advantage of avoiding the goto statement is the absence of a label. Note that in the case of
the “for” statement, execution continues with the increment part.
Remember that the continue statement within a switch statement doesn’t mean that execution
will continue the switch but continue the next enclosing for, while, or do statement.

1.10.7.23 Null statements

A null statement is just a semicolon. This is used in two contexts:

1) An empty body of an iterative statement (while, do, or for). For instance you can do:
while (*p++)
; /* search the end of the string */

2) A label should appear just before a closing brace. Since labels must be attached to a
statement, the empty statement does that just fine.

1.10.7.24 Comments
Multi-line comments are introduced with the characters “/” and “*” and finished with the
opposite sequence: “*” followed by “/”. This commentaries can’t be nested. Single line com-
ments are introduced by the sequence “//” and go up to the end of the line. Here are some
examples:
“a//b” Four character string literal
// */ Single line comment, not syntax error
f = g/**//h;Equivalent to f = g/h;
//\
fn();Part of a comment since the last line ended with a “\”

1.10.7.25 Switch statement.

The purpose of this statement is to dispatch to several code portions according to the value in
an integer expression. A simple example is:
enum animal {CAT,DOG,MOUSE};

enum animal pet = GetAnimalFromUser();

switch (pet) {
case CAT:
printf(“This is a cat”);
break;
case DOG:
printf(“This is a dog”);
break;
case MOUSE:
printf(“This is a mouse”);
 An overview of the whole language 45

break;
default:
printf(“Unknown animal”);
break;
}
We define an enumeration of symbols, and call another function, that asks for an animal type
to the user and returns its code. We dispatch then upon the value of the In this case the integer
expression that controls the switch is just an integer, but it could be any expression. Note that
the parentheses around the switch expression are mandatory. The compiler generates code that
evaluates the expression, and a series of jumps (gotos) to go to the corresponding portions of
the switch. Each of those portions is introduced with a “case” keyword that is followed by an
integer constant. Note that no expressions are allowed in cases, only constants that can be
evaluated by the compiler during compilation.
Cases end normally with the keyword “break”, that indicates that this portion of the switch is
finished. Execution continues after the switch. A very important point here is that if you do not
explicitly write the break keyword, execution will continue into the next case. Sometimes this
is what you want, but most often it is not. Beware.
There is a reserved word “default”, that contains the case for all other values that do not appear
explicitly in the switch. It is a good practice to always add this keyword to all switch state-
ments and figure out what to do when the input doesn’t match any of the expected values.
If the input value doesn’t match any of the enumerated cases and there is no default statement,
no code will be executed and execution continues after the switch.
Conceptually, the switch statement above is equivalent to:
if (pet == CAT) {
printf(“This is a cat”);
}
else if (pet == DOG) {
printf(“This is a dog”);
}
else if (pet == MOUSE) {
printf(“This is a mouse”);
} else printf(“Unknown animal”);
Both forms are exactly equivalent, but there are subtle differences:
Switch expressions must be of integer type. The “if” form doesn’t have this limitation.
In the case of a sizeable number of cases, the compiler will optimize the search in a switch
statement to avoid comparisons. This can be quite difficult to do manually with “if”s.
Cases of type other than int, or ranges of values can’t be specified with the switch statement,
contrary to other languages like Pascal, that allow a range here. Switch statements can be
nested to any level (i.e. you can write a whole switch within a case statement), but this makes
the code unreadable and is not recommended.

1.10.7.26 inline
This instructs the compiler to replicate the body of a function at each call site. For instance:
int inline f(int a) { return a+1;}
Then:
int a = f(b)+f(c);
will be equivalent to writing:
int a = (b+1)+(c+1);
46 C Tutorial

Note that this expansion is realized in the lcc-win32 compiler only when optimizations are
ON. In a normal (debug) setting, the “inline” keyword is ignored. You can control this behav-
ior also, by using the command line option “-fno-inline”.

1.10.7.27 Logical operators

A logical expression consists of two boolean expressions (i.e. expressions that are either true
or false) separated by one of the logical operators && (AND) or || (or).
The AND operator evaluates from left to right. If any of the expressions is zero, the evaluation
stops with a FALSE result and the rest of the expressions is not evaluated. The result of several
AND expressions is true if and only if all the expressions evaluate to TRUE.
Example:
1 && 1 && 0 && 1 && 1
Here evaluation stops after the third expression yields false (zero). The fourth and fifth expres-
sions are not evaluated. The result of all the AND expressions is zero.
The OR operator evaluates from left to right. If any of the expressions yields TRUE, evalua-
tion stops with a TRUE result and the rest of the expressions is not evaluated. The result of
several OR expressions is true if and only if one of the expressions evaluates to TRUE.
If we have the expression:
result = expr1 && expr2;
this is equivalent to the following C code:
if (expr1 == 0)
result = 0;
else {
if (expr2 == 0)
result = 0;
else result = 1;
}
In a similar way, we can say that the expression
result = expr1 || expr2;
is equivalent to:
if (expr1 != 0)
result = 1;
else {
if (expr2 != 0)
result = 1;
else result = 0;
}

1.10.7.28 Bitwise operators

The operators & (bitwise AND), ^ (bitwise exclusive or), and | (bitwise or) perform boolean
operations between the bits of their arguments, that must be integers: long long, long, int,
short, or char.
The operation of each of them is as follows:

1) The & (AND) operator yields a 1 bit if both arguments are 1. Otherwise it yields a 0.

2) The ^ (exclusive or) operator yields 1 if one argument is 1 and the other is zero, i.e. it yields
1 if their arguments are different. Otherwise it yields zero
 An overview of the whole language 47

3) The | (or) operator yields 1 if either of its arguments is a 1. Otherwise it yields a zero.
We can use for those operators the following truth table:

a b a&b a^b a|b

0 0 0 0 0
0 1 0 1 1
1 0 0 1 1
1 1 1 0 1

Note that this operators are normal operators, i.e. they evaluate always their operands, unlike
&& or || that use short-circuit evaluation. If we write:
a = 0 && fn(67);
the function call will never be executed. If we write
a = 0&fn(67);
the function call will be executed even if the result is fixed from the start.

1.10.7.29 Address-of operator

The unary operator & yields the machine address of its argument, that must be obviously an
addressable object. For instance if you declare a variable as a “register” variable, you can’t use
this operator to get its address because registers do not live in main memory. In a similar way,
you can’t take the address of a constant like &45 because the number 45 has no address.
The result of the operator & is a pointer with the same type as the type of its argument. If you
take the address of a short variable, the result is of type “pointer to short”. If you take the
address of a double, the result is a pointer to double, etc.
If you take the address of a local variable, the pointer you obtain is valid only until the func-
tion where you did this exits. Afterward, the pointer points to an invalid address and will pro-
duce a machine fault when used, if you are lucky. If you are unlucky the pointer will point to
some random data and you will get strange results, what is much more difficult to find.
In general, the pointer you obtain with this operator is valid only if the storage of the object is
pointing to is not released. If you obtain the address of an object that was allocated using the
standard memory allocator malloc, this pointer will be valid until there is a “free” call that
releases that storage. Obviously if you take the address of a static or global variable the pointer
will be always valid since the storage for those objects is never released.

1.10.7.30 Indirection
The * operator is the contrary of the address-of operator above. It expects a pointer and returns
the object the pointer is pointing to. For instance if you have a pointer pint that points to an
integer, the operation *pint will yield the integer value the pointer pint is pointing to..
The result of this operator is invalid if the pointer it is dereferencing is not valid. In some
cases, dereferencing an invalid pointer will provoke the dreaded window “This program has
performed an invalid operation and will terminate” that windows shows up when a machine
fault is detected. In other cases, you will be unlucky and the dereferencing will yield a non-
sense result.
For instance, this program will crash:
48 C Tutorial

int main(void)
{
char *p;

*p = 0;
return 1;
}
We have never assigned to p an object where it should point to. We are using a dangling
pointer. When we follow this program we obtain:

The debugger tells us that a machine exception has occurred, with the code 0xc0000005. This
means that the CPU has detected an invalid memory reference and has called the exception
mechanism of windows, that notified the debugger of the problem. Note the value of the
pointer in the output window: 0xfffa5a5a.
Lcc-win32 follows the philosophy that the sooner you see an error, the better. When it allo-
cates the stack frame of the function, it will write this value to all memory that has not been
explicitely initialized by the program. When you see this value in the debugger you can be
highly confident that this is an unitialized pointer or variable. Note too that this will not be
done if you turn on optimizations. In that case the pointer will contain whatever was in there
when the compiler allocated the stack frame.
Note that many other compilers do not do this, and some programs run without crashing out of
sheer luck. Since lcc-win32 catches this error, it looks to the useers as if the compiler was
buggy. I have received a lot of complaints because of this.
This kind of problem is one of the most common bugs in C. Fogetting to initialize a pointer is
something that you can never afford to do.
Another error is initializing a pointer within a conditional expression:
char *BuggyFunction(int a)
{
char *result;
 An overview of the whole language 49

if (a > 34) {
result = malloc(a+34);
}
return result;
}
If the argument of this function is less than 35, the pointer returned will be a dangling pointer
since it was never initialized.

1.10.7.31 Precedence of the different operators.

In their book «C, a reference manual», Harbison and Steele propose the following table.

Tokens Operator Class Precedence Associates

names, literals simple tokens primary 16 n/a

a[k] subscripting postfix 16 left-to-right

f(...) function call postfix 16 left-to-right

. (point) direct selection postfix 16 left-to-right

-> indirect selection postfix 16 left-to-right

++ -- increment, decrement postfix 16 left-to-right

(type name){init} compound literal postfix 16 left-to-right

++ -- increment, decrement prefix 15 right-to-left

sizeof size unary 15 right-to-left

~ bitwise not unary 15 right-to-left

! logical not unary 15 right-to-left

- + arithmetic negation, plus unary 15 right-to-left

& address of unary 15 right-to-left

* indirection unary 15 right-to-left

( type name ) casts unary 14 right-to-left

* / % multiplicative binary 13 left-to-right

+ - additive binary 12 left-to-right

<< >> left/right shift binary 11 left-to-right

< > <= >= relational binary 10 left-to-right

== != equal/not equal binary 9 left-to-right

& bitwise and binary 8 left-to-right

^ bitwise xor binary 7 left-to-right

| bitwise or binary 6 left-to-right

&& logical and binary 5 left-to-right

|| logical or binary 4 left-to-right

? : conditional binary 2 right-to-left

50 C Tutorial

= += -= *= /= %= <<= >>= assignment binary 2 right-to-left

&= ^= |=

, sequential evaluation binary 1 left-to-right

 Simple programs 51

1.11 Simple programs

To give you a more concrete example of how C works, here are a few examples of very simple
programs. The idea is to find a self-contained solution for a problem that is simple to state and
understand.

1.11.1 Problem 1: strchr

Find the first occurrence of a given character in a character string. Return a pointer to the char-
acter if found, NULL otherwise.
This problem is solved in the standard library by the strchr function. Let’s write it. The
algorithm is very simple: We examine each character. If it is zero, this is the end of the string,
we are done and we return NULL to indicate that the character is not there. If we find it, we
stop searching and return the pointer to the character position.
char *FindCharInString(char *str, int ch)
{
while (*str != 0 && *str != ch) {
str++;
}
if (*str == ch)
return str;
return NULL;
}
We loop through the characters in the string. We use a while condition requiring that the char-
acter pointed to by our pointer “str” is different than zero and it is different than the character
given. In that case we continue with the next character by incrementing our pointer, i.e. mak-
ing it point to the next char. When the while loop ends, we have either found a character, or we
have arrived at the end of the string. We discriminate between these two cases after the loop.

How can this program fail?

We do not test for NULL. Any NULL pointer passed to this program will provoke a trap. A
way of making this more robust would be to return NULL if we receive a NULL pointer. This
would indicate to the calling function that the character wasn’t found, what is always true if
our pointer doesn’t point anywhere.
A more serious problem happens when our string is missing the zero byte… In that case the
program will blindly loop through memory, until it either finds the byte is looking for, or a
zero byte somewhere. This is a much more serious problem, since if the search ends by finding
a random character somewhere, it will return an invalid pointer to the calling program!
This is really bad news, since the calling program may not use the result immediately. It could
be that the result is stored in a variable, for instance, and then used in another, completely
unrelated section of the program. The program would crash without any hints of what is wrong
and where was the failure.
Note that this implementation of strchr will correctly accept a zero as the character to be
searched. In this case it will return a pointer to the zero byte.

1.11.2 Problem 2: strlen

Return the length of a given string not including the terminating zero.
This is solved by the strlen function. We just count the chars in the string, stopping when
we find a zero byte.
int strlen(char *str)
52 C Tutorial

{
char *p = str;

while (*p != 0) {
p++;
}
return p – str;
}
We copy our pointer into a new one that will loop through the string. We test for a zero byte in
the while condition. Note the expression *p != 0. This means “Fetch the value this pointer
is pointing to (*p), and compare it to zero”. If the comparison is true, then we increment the
pointer to the next byte.43
We return the number of characters between our pointer p and the saved pointer to the start of
the string. This pointer arithmetic is quite handy.

How can this program fail? The same problems apply that we discussed in the previous
example, but in an attenuated form: only a wrong answer is returned, not an outright wrong
pointer. The program will only stop at a zero byte.

1.11.3 Problem 3: ispowerOfTwo

Given a positive number, find out if it is a power of two.
Algorithm: A power of two has only one bit set, in binary representation. We count the bits. If
we find a bit count different than one we return 0, if there is only one bit set we return 1.
Implementation: We test the rightmost bit, and we use the shift operator to shift the bits right,
shifting out the bit that we have tested. For instance, if we have the bit pattern 1 0 0 1, shifting
it right by one gives 0 1 0 0: the rightmost bit has disappeared, and at the left we have a new
bit shifted in, that is always zero.
int ispowerOfTwo(unsigned int n)
{
unsigned int bitcount = 0;

while (n != 0) {
if (n & 1) {
bitcount++;
}
n = n >> 1;
}
if (bitcount == 1)
return 1;
return 0;
}

Our condition here is that n must be different44 than zero, i.e. there must be still some bits to
count to go on. We test the rightmost bit with the binary and operation. The number one has
only one bit set, the rightmost one. By the way, one is a power of two45.

43. The expression (*p != 0) could have been written in the form while (*p), using the implicit
test for a non-zero result in any logical expression. Any expression will be considered true if its value is
anything but zero. It is better, however, to make comparisons explicit.
44. Different than is written in C != instead of ≠ . The symbol ≠ wasn’t included in the primitive
typewriters in use when the C language was designed, and we have kept that approximation. It is consistent
with the usage of ! as logical not, i.e. != would mean not equal.
 Simple programs 53

Note that the return expression could have also been written like this:
return bitcount == 1;
The intention of the program is clearer with the “if” expression.

How can this program fail? The while loop has only one condition: that n is different than
zero, i.e. that n has some bits set. Since we are shifting out the bits, and shifting in always
zero bits since bitcount is unsigned, in a 32 bit machine like a PC this program will stop
after at most 32 iterations. Running mentally some cases (a good exercise) we see that for
an input of zero, we will never enter the loop, bitcount will be zero, and we will return 0,
the correct answer. For an input of 1 we will make only one iteration of the loop. Since 1 &
1 is 1, bitcount will be incremented, and the test will make the routine return 1, the cor-
rect answer. If n is three, we make two passes, and bitcount will be two. This will be
different than 1, and we return zero, the correct answer.
Anh Vu Tran anhvu.tran@ifrance.com made me discover an important bug. If you change
the declaration of “n” from unsigned int to int, without qualification, the above function
will enter an infinite loop if n is negative.
Why?
When shifting signed numbers sign is preserved, so the sign bit will be carried through, pro-
voking that n will become eventually a string of 1 bits, never equal to zero, hence an infinite
loop.

1.11.4 Problem 4: strlwr

Given a string containing upper case and lower case characters, transform it in a string with
only lower case characters. Return a pointer to the start of the given string.46
This is the library function strlwr. In general is not a good idea to replace library functions,
even if they are not part of the standard library (as defined in the C standard) like this one.
We make the transformation in-place, i.e. we transform all the characters of the given string.
This supposes that the user of this program has copied the original string elsewhere if the orig-
inal is needed again.

#include <ctype.h> /* needed for using isupper and tolower */

#include <stdio.h> /* needed for the NULL definition */
char *strTolower(char *str)
{
/* iterates through str */
unsigned char *p = (unsigned char *)str;

if (str == NULL)
return NULL;
while (*p) {
*str = tolower(*p);

45. For a more detailed discussion, see the section Newsgroups at the end of this tutorial.
46. This convention is used in the library function. Actually, it is a quite bad interface, since the return
value doesn’t give any new information to the user, besides the expected side effect of transforming the given
string. A better return value would be the number of changed characters, for instance, that would allow the
caller to know if a transformation was done at all, or the length of the string, or several others. But let’s
implement this function as it is specified in the standard library. Many times, you will see that even if it is
obvious that software must be changed, the consequences of a change are so vast that nobody wants to
assume it, and we get stuck with software “for compatibility reasons”. Here is yet another example.
54 C Tutorial

p++;
}
return str;
}
We include the standard header ctype.h, which contains the definition of several character
classification functions (or macros) like “isupper” that determines if a given character is
upper case, and many others like “isspace”, or “isdigit”. We need to include the stdio.h
header file too, since it contins the definition for NULL.47
The first thing we do is to test if the given pointer is NULL. If it is, we return NULL. Then, we
start our loop that will span the entire string. The construction while(*p) tests if the con-
tents of the character pointer p is different than zero. If this is the case, we transform it into a
lower case one. We increment our pointer to point to the next character, and we restart the
loop. When the loop finishes because we hit the zero byte that terminates the string, we stop
and return the saved position of the start of the string.
Note the cast that transforms str from a char * into an unsigned char *. The reason is that it
could exist a bad implementation of the toupper() function, that would index some table using
a signed char. Characters above 128 would be considered negative integers, what would result
in a table being indexed by a negative offset, with bad consequences, as you may imagine.

How can this program fail?

Since we test for NULL, a NULL pointer can’t provoke a trap. Is this a good idea?
Well this depends. This function will not trap with NULL pointers, but then the error will be
detected later when other operations are done with that pointer anyway. Maybe making a trap
when a NULL pointer is passed to us is not that bad, since it will uncover the error sooner
rather than later. There is a big probability that if the user of our function is calling us to trans-
form the string to lower case, is because he/she wants to use it later in a display, or otherwise.
Avoiding a trap here only means that the trap will appear later, probably making error finding
more difficult.
Writing software means making this type of decisions over and over again.
Obviously this program will fail with any incorrect string, i.e. a string that is missing the final
zero byte. The failure behavior of our program is quite awful: in this case, this program will
start destroying all bytes that happen to be in the range of uppercase characters until it hits a
random zero byte. This means that if you pass a non-zero terminated string to this apparently
harmless routine, you activate a randomly firing machine gun that will start destroying your
program’s data in a random fashion. The absence of a zero byte in a string is fatal for any C
program. In a tutorial this can’t be too strongly emphasized!

1.12 Problem 5: paste

You have got two text files, and you want to merge them in a single file, separated by tabula-
tions. For instance if you have a file1 with this contents:
line 1
line2
and you got another file2 with this contents
line 10
line 11

47. If remembering which headers contain which definitions bothers you (as it bothers me) just use the
<stdheaders.h> header file included with lcc-win32. That file is just an include of all standard header
files.
 Simple programs 55

you want to obtain a file

line1 line 10
line 2 line 11
Note that both file can be the same.
A solution for this could be the following program:
#include <stdio.h>
#include <string.h>
#include <stdlib.h>
/*
We decide arbitrarily that lines longer than 32767 chars will make this program fail.
*/
#define MAXLINELEN 32767
int main(int argc,char *argv[])
{
/*
We need two FILE pointers, and two line buffers to hold each line from each file. We receive
in argc the number of arguments passed + 1, and in the character array argv[] the names of
the two files
*/
FILE *f1,*f2;
char buf1[MAXLINELEN],buf2[MAXLINELEN];

/*
We test immediately if the correct number of arguments has been given. If not, we exit with a
clear error message.
*/
if (argc < 3) {
fprintf(stderr,"Usage: paste file1 file2\n");
exit(1);
}
/*
We open both files, taking care not to open the same file twice. We test with strcmp if they are
equal.
*/
f1 = fopen(argv[1],"r");
if (strcmp(argv[1],argv[2]))
f2 = fopen(argv[2],"r");
else
f2 = f1;
/*
We read line after line of the first file until we reach the end of the first file.
*/
while(fgets(buf1,MAXLINELEN,f1)) {
char *p = strchr(buf1,'\n');
/*
the fgets function leaves a \n in the input. We erase it if it is there. We use for this the strchr
function, that returns the first occurrence of a character in a string and returns a pointer to it.
If it doesn’t it returns NULL, so we test below before using that pointer
*/
if (p)
*p = 0;
/*
We output the first file line, separated from the next with a single tabulation char.
*/
printf("%s\t",buf1);
/*
If there are still lines to be read from file 2, we read them and we print them after doing the
same treatment as above.
56 C Tutorial

*/
if (f2 != f1 && fgets(buf2,MAXLINELEN,f2)) {
p = strchr(buf2,'\n');
if (p)
*p = 0;
printf("%s\n",buf2);
}
/*
If we are duplicating the same file just print the same line again.
*/
else printf("%s\n",buf1);
}
/*
End of the while loop. When we arrive here the first file has been completely scanned. We
close and shut down.
*/
fclose(f1);
if (f1 != f2)
fclose(f2);
return 0;
}
How can this program fail?.
Well, there are obvious bugs in this program. Before reading the answer, try to see if you can
see them. What is important here is that you learn how to spot bugs and that is a matter of log-
ical thinking and a bit of effort.
Solution will be in the next page. But just try to find those bugs yourself.
Before that bug however we see this lines in there:
if (f2 != f1 && fgets(buf2,MAXLINELEN,f2)) {
}
else printf("%s\n",buf1);
If f1 is different from f2 (we have two different files) and file two is shorter than file one, that
if statement will fail after n2 lines, and the else portion will be executed, provoking the dupli-
cation of the contents of the corresponding line of file one.
To test this, we create two test files, file1 and file2. their contents are:
File1:
File 1: line 1
File 1: line 2
File 1: line 3
File 1: line 4
File 1: line 5
File 1: line 6
File 1: line 7
File 1: line 8

File2:
File 2: line 1
File 2: line 2
File 2: line 3
File 2: line 4

We call our paste program with that data and we obtain:

 Simple programs 57

The line five of file one was read and since file two is already finished, we repeat it.
Is this a bug or a feature?
We received vague specifications. Nothing was said about what the program should do with
files of different sizes. This can be declared a feature, but of course is better to be aware of it.
We see that to test hypothesis about the behavior of a program, there is nothing better than test
data, i.e. data that is designed to exercise one part of the program logic.
In many real cases, the logic is surely not so easy to follow as in this example. Building test
data can be done automatically. To build file one and two, this small program will do:
#include <stdio.h>
int main(void)
{
FILE *f = fopen("file1","w");
for (int i =0; i<8;i++)
fprintf(f,"File 1: Line %d\n",i);
fclose(f);
f = fopen("file2","w");
for (int i = 0; i < 5;i++)
fprintf(f,"File 2: Line %d\n",i);
fclose(f);
return 0;
}
This a good example of throw away software, software you write to be executed once. No
error checking, small and simple, so that there is less chance for mistakes.
And now the answer to the other bug above.
58 C Tutorial

One of the first things to notice is that the program tests with strcmp to see if two files are the
same. This means that when the user passes the command line:
paste File1 filE1
our program will believe that they are different when in fact they are not. Windows is not case
sensitive for file names. The right thing to do there is to compare the file names with stricmp,
that ignores the differences between uppercase and lowercase.
But an even greater problem is that we do not test for NULL when opening the files. If any of
the files given in the command line doesn’t exist, the program will crash. Add the necessary
tests before you use it.
Another problem is that we test if we have the right number of arguments (i.e. at least two file
names) but if we have more arguments we simply ignore them. What is the right behavior?
Obviously we could process (and paste) several files at once. Write the necessary changes in
the code above. Note that if you want to do the program really general, you should take into
account the fact that a file could be repeated several times in the input, i.e.
paste file1 file2 file1 file3
Besides, the separator char in our program is now hardwired to the tab character in the code of
the program. Making this an option would allow to replace the tab with a vertical bar, for
instance.
But the problem with such an option is that it supposes that the output will be padded with
blanks for making the vertical bars align. Explain why that option needs a complete rewrite of
our program. What is the hidden assumption above that makes such a change impossible?
Another feature that paste.exe could have, is that column headers are automatically under-
lined. Explain why adding such an option is falling into the featurism that pervades all modern
software. Learn when to stop!

1.13 Using arrays and sorting

Suppose we want to display the frequencies of each letter in a given file. We want to know the
number of ‘a’s, of ‘b’, and so on.
One way to do this is to make an array of 256 integers (one integer for each of the 256 possible
character values) and increment the array using each character as an index into it. When we
see a ‘b’, we get the value of the letter and use that value to increment the corresponding posi-
tion in the array. We can use the same skeleton of the program that we have just built for
counting characters, modifying it slightly.48
#include <stdio.h>
#include <stdlib.h>

int Frequencies[256]; // Array of frequencies

int main(int argc,char *argv[])

{
// Local variables declarations
int count=0;
FILE *infile;
int c;

if (argc < 2) {

48. Yes, code reuse is not only possible in object-oriented programming.

 Using arrays and sorting 59

printf("Usage: countchars <file name>\n");

exit(1);
}
infile = fopen(argv[1],"rb");
if (infile == NULL) {
printf("File %s doesn't exist\n",argv[1]);
exit(1);
}
c = fgetc(infile);
while (c != EOF) {
count++;
Frequencies[c]++;
c = fgetc(infile);
}
fclose(infile);
printf("%d chars in file\n",count);
for (count=0; count<256;count++) {
if (Frequencies[count] != 0) {
printf(“’%3c’ (%4d) = %d\n”, count, count,
Frequencies[count]);
}
}
return 0;
}
We declare an array of 256 integers, numbered from zero to 255. Note that in C the index ori-
gin is always zero.
This array is not enclosed in any scope. Its scope then, is global, i.e. this identifier will be
associated to the integer array for the current translation unit (the current file and its includes)
from the point of its declaration on.
Since we haven’t specified otherwise, this identifier will be exported from the current module
and will be visible from other modules. In another compilation unit we can then declare:
extern int Frequencies[];
and we can access this array. This can be good (it allow us to share data between modules), or
it can be bad (it allows other modules to tamper with private data), it depends on the point of
view and the application. 49
If we wanted to keep this array local to the current compilation unit we would have written:
static int Frequencies[256];
The “static” keyword indicates to the compiler that this identifier should not be made visible
in another module.
The first thing our program does, is to open the file with the name passed as a parameter. This
is done using the fopen library function. If the file exists, and we are able to read from it, the
library function will return a pointer to a FILE structure, defined in stdio.h. If the file can’t be
opened, it returns NULL. We test for this condition right after the fopen call.
We can read characters from a file using the fgetc function. That function updates the current
position, i.e. the position where the next character will be read.

49. Global variables provoke an undocumented coupling between several, apparently unrelated
procedures or modules. Overuse of them is dangerous, and provokes errors that can be difficult to understand
and get rid of. I learned this by experience in long debugging sessions, and now I use global variables more
sparingly.
60 C Tutorial

But let’s come back to our task. We update the array at each character, within the while loop.
We just use the value of the character (that must be an integer from zero to 256 anyway) to
index the array, incrementing the corresponding position. Note that the expression:
Frequencies[count]++
means
Frequencies[count] = Frequencies[count]+1;
i.e.; the integer at that array position is incremented, and not the count variable!
Then at the end of the while loop we display the results. We only display frequencies when
they are different than zero, i.e. at least one character was read at that position. We test this
with the statement:
if (Frequencies[count] != 0) { ... statements ... }
The printf statement is quite complicated. It uses a new directive %c, meaning character, and
then a width argument, i.e. %3c, meaning a width of three output chars. We knew the %d
directive to print a number, but now it is augmented with a width directive too. Width direc-
tives are quite handy when building tables to get the items of the table aligned with each other
in the output.
The first thing we do is to build a test file, to see if our program is working correctly. We build
a test file containing
ABCDEFGHIJK
And we call:
lcc frequencies.c
lcclnk frequencies.obj
frequencies fexample
and we obtain:
D:\lcc\examples>frequencies fexample
13 chars in file

( 10) = 1
( 13) = 1
A ( 65) = 1
B ( 66) = 1
C ( 67) = 1
D ( 68) = 1
E ( 69) = 1
F ( 70) = 1
G ( 71) = 1
H ( 72) = 1
I ( 73) = 1
J ( 74) = 1
K ( 75) = 1
We see that the characters \r (13) and new line (10) disturb our output. We aren’t interested in
those frequencies anyway, so we could just eliminate them when we update our Frequencies
table. We add the test:
if (c >= ' ')
Frequencies[c]++;
i.e. we ignore all characters with value less than space: \r, \n or whatever. Note that we ignore
tabulations too, since their value is 8.
The output is now more readable:
H:\lcc\examples>frequencies fexample
 Using arrays and sorting 61

13 chars in file
A ( 65) = 1
B ( 66) = 1
C ( 67) = 1
D ( 68) = 1
E ( 69) = 1
F ( 70) = 1
G ( 71) = 1
H ( 72) = 1
I ( 73) = 1
J ( 74) = 1
K ( 75) = 1
We test now our program with itself. We call:
frequencies frequencies.c
758 chars in file
I have organized the data in a table to easy the display.

( 32) = 57 ! ( 33) = 2 " ( 34) = 10

# ( 35) = 2 % ( 37) = 5 ' ( 39) = 3

( ( 40) = 18 ) ( 41) = 18 * ( 42) = 2

+ ( 43) = 6 , ( 44) = 7 . ( 46) = 2

/ ( 47) = 2 0 ( 48) = 4 1 ( 49) = 4

2 ( 50) = 3 3 ( 51) = 1 4 ( 52) = 1

5 ( 53) = 2 6 ( 54) = 2 : ( 58) = 1

; ( 59) = 19 < ( 60) = 5 = ( 61) = 11

> ( 62) = 4 A ( 65) = 1 E ( 69) = 2

F ( 70) = 7 I ( 73) = 1 L ( 76) = 3

N ( 78) = 1 O ( 79) = 1 U ( 85) = 2

[ ( 91) = 7 \ ( 92) = 4 ] ( 93) = 7

a ( 97) = 12 b ( 98) = 2 c ( 99) = 33

d ( 100) = 8 e ( 101) = 38 f ( 102) = 23

g ( 103) = 8 h ( 104) = 6 i ( 105) = 43

l ( 108) = 14 m ( 109) = 2 n ( 110) = 43

o ( 111) = 17 p ( 112) = 5 q ( 113) = 5

r ( 114) = 23 s ( 115) = 14 t ( 116) = 29

u ( 117) = 19 v ( 118) = 3 w ( 119) = 1

x ( 120) = 3 y ( 121) = 1 { ( 123) = 6

} ( 125) = 6

What is missing obviously, is to print the table in a sorted way, so that the most frequent char-
acters would be printed first. This would make inspecting the table for the most frequent char-
acter easier. How can we do that in C?
62 C Tutorial

We have in the standard library the function “qsort”, that sorts an array. We study its prototype
first, to see how we should use it:50
void qsort(void *b,size_t n,size_t s,int(*f)(const void *));
Well, this is quite an impressing prototype really. But if we want to learn C, we will have to
read this, as it was normal prose. So let’s begin, from left to right.
The function qsort doesn’t return an explicit result. It is a void function. Its argument list, is
the following:

Argument 1: is a void *. Void *??? What is that? Well, in C you have void, that means none,
and void *, that means this is a pointer that can point to anything, i.e. a pointer to an
untyped value. We still haven’t really introduced pointers, but for the time being just be
happy with this explanation: qsort needs the start of the array that will sort. This array can
be composed of anything, integers, user defined structures, double precision numbers,
whatever. This "whatever" is precisely the “void *”.

Argument 2 is a size_t. This isn’t a known type, so it must be a type defined before in
stdlib.h. By looking at the headers, and following the embedded include directives, we find:
“stdlib.h” includes “stddef.h”, that defines a “typedef” like this:51
typedef unsigned int size_t;
This means that we define here a new type called “size_t”, that will be actually an unsigned
integer. Typedefs allow us to augment the basic type system with our own types. Mmmm
interesting. We will keep this for later use.
In this example, it means that the size_t n, is the number of elements that will be in the
array.

Argument 3 is also a size_t. This argument contains the size of each element of the array,
i.e. the number of bytes that each element has. This tells qsort the number of bytes to skip at
each increment or decrement of a position. If we pass to qsort an array of 56 double preci-
sion numbers, this argument will be 8, i.e. the size of a double precision number, and the
preceding argument will be 56, i.e. the number of elements in the array.
Argument 4 is a function: int (*f)(const void *)); Well this is quite hard really.
We are in the first pages of this introduction and we already have to cope with gibberish
like this?
We have to use recursion now. We have again to start reading this from left to right, more or
less. We have a function pointer (f) that points to a function that returns an int, and that
takes as arguments a void *, i.e. a pointer to some unspecified object, that can’t be changed
within that function (const).
This is maybe quite difficult to write, but quite a powerful feature. Functions can be passed
as arguments to other functions in C. They are first class objects that can be used to specify
a function to call.
Why does qsort need this?

50. The prototype is in the header file stdlib.h

51. Finding out where is something defined can be quite a difficult task. The easiest way is to use the
IDE of lcc-win32, right click in the identifier, and choose “goto definition”. If that doesn’t work, you can use
“grep” to search in a set of files.
 Using arrays and sorting 63

Well, since the qsort function is completely general, it needs a helper function, that will tell it
when an element of the array is smaller than the other. Since qsort doesn’t have any a priori
knowledge of the types of the elements of the passed array, it needs a helper function that
returns an integer smaller than zero if the first element is smaller than the next one, zero if the
elements are equal, or bigger than zero if the elements are bigger.
Let’s apply this to a smaller example, so that the usage of qsort is clear before we apply it to
our frequencies problem.
#include <stdlib.h>
#include <string.h> (1)
#include <stdio.h>

int compare(const void *arg1,const void *arg2)(2)

{
/* Compare all of both strings: */ (3)
return stricmp( *( char** ) arg1, * ( char** ) arg2 );
}

int main( int argc, char **argv )

{
/* Eliminate argv[0] from sort: */ (4)
argv++;
argc--;

/* Sort remaining args using qsort */(5)

qsort((void*)argv,(size_t)argc,sizeof(char *),compare);

/* Output sorted list: */

for(int i = 0; i < argc; ++i )(6)
printf( "%s ", argv[i] );
printf( "\n" ); (7)
return 0;
}
The structure of this example is as follows:
We build a program that will sort its arguments and output the sorted result.
To use qsort we define a comparison function that returns an integer, which encodes the rela-
tive lexical ordering of the two arguments passed to it. We use a library function for doing that,
the stricmp52 function, that compares two character strings without caring about case differ-
ences.
But there is quite a lot of new material in this example, and it is worth going through it in
detail.
We include the standard header string.h, to get the definitions of string handling functions like
stricmp.
We define our comparison function with:
int compare(const void *arg1,const void *arg2) { ... }
This means that our compare function will return an int, and that takes two arguments, named
arg1 and arg2, that are pointers to any object (void *). The objects pointed to by arg1, and arg2
will not be changed within this function, i.e. they are “const”.

52. Most compilers do not have the C99 standard implemented. In those compilers you can’t do this and
you will have to declare the loop counter as a normal local variable. Another reason to stick to lcc-win32.
64 C Tutorial

We need to get rid of the void * within our compare function. We know we are going to pass to
this function actually pointers to characters, i.e. machine addresses to the start of character
strings, so we have to transform the arguments into a type we can work with. For doing this we
use a cast. A cast is a transformation of one type to another type at compile time. Its syntax is
like this: (newtype)(expression);. In this example we cast a void * to a char **, a
pointer to a pointer of characters. The whole expression needs quite a lot of reflection to be
analyzed fully. Return here after reading the section about pointers.
Note that our array argv, can be used as a pointer and incremented to skip over the first ele-
ment. This is one of the great weaknesses of the array concept of the C language. Actually,
arrays and pointers to the first member are equivalent. This means that in many situations,
arrays “decay” into pointers to the first element, and loose their “array”ness. That is why you
can do in C things with arrays that would never be allowed in another languages. At the end of
this tutorial we will se how we can overcome this problem, and have arrays that are always
normal arrays that can be passed to functions without losing their soul.
At last we are ready to call our famous qsort function. We use the following call expression:
qsort((void*)argv,(size_t)argc,sizeof(char *),compare);
The first argument of qsort is a void *. Since our array argv is a char **, we transform it into
the required type by using a cast expression: (void *)argv.
The second argument is the number of elements in our array. Since we need a size_t and we
have argc, that is an integer variable, we use again a cast expression to transform our int into a
size_t. Note that typedefs are accepted as casts.
The third argument should be the size of each element of our array. We use the built-in pseudo
function sizeof, which returns the size in bytes of its argument. This is a pseudo function,
because there is no such a function actually. The compiler will replace this expression with an
integer that it calculates from its internal tables. We have here an array of char *, so we just tell
the compiler to write that number in there.
The fourth argument is our comparison function. We just write it like that. No casts are
needed, since we were careful to define our comparison function exactly as qsort expects.
To output the already sorted array we use again a “for” loop. Note that the index of the loop is
declared at the initialization of the “for” construct. This is one of the new specifications of the
C99 language standard, that lcc-win32 follows. You can declare variables at any statement,
and within “for” constructs too. Note that the scope of this integer will be only the scope of the
enclosing “for” block. It can’t be used outside this scope.53
Note that we have written the “for” construct without curly braces. This is allowed, and means
that the “for” construct applies only to the next statement, nothing more. The ...
printf("\n");... is NOT part of the for construct.
Ok, now let’s compile this example and make a few tests to see if we got that right.
h:\lcc\examples> sortargs aaa bbb hhh sss ccc nnn
aaa bbb ccc hhh nnn sss
OK, it seems to work. Now we have acquired some experience with qsort, we can apply our
knowledge to our frequencies example. We use cut and paste in the editor to define a new
compare function that will accept integers instead of char **. We build our new comparison
function like this:

53. The compiler emits a record for the linker, telling it to put there the address of the global, if the
argument is a global variable, or will emit the right instructions to access the address of a local using the
frame pointer. This has been working for a while now.
 Using arrays and sorting 65

int compare( const void *arg1, const void *arg2 )

{
return ( * ( int * ) arg1 - * ( int * ) arg2 );
}
We just return the difference between both numbers. If arg1 is bigger than arg2, this will be a
positive number, if they are equal it will be zero, and if arg1 is smaller than arg2 it will be a
negative number, just as qsort expects.
Right before we display the results then, we add the famous call we have been working so hard
to get to:
qsort(Frequencies,256,sizeof(int),compare);
We pass the Frequencies array, its size, the size of each element, and our comparison function.
Here is the new version of our program, for your convenience. New code is in bold:
#include <stdio.h>
#include <stdlib.h>

int Frequencies[256]; // Array of frequencies

int compare( const void *arg1, const void *arg2 )

{
/* Compare both integers */
return ( * ( int * ) arg1 - * ( int * ) arg2 );
}

int main(int argc,char *argv[])

{
int count=0;
FILE *infile;
int c;

if (argc < 2) {
...
}
infile = fopen(argv[1],"rb");
if (infile == NULL) {
...
}
c = fgetc(infile);
while (c != EOF) {
...
}
fclose(infile);
printf("%d chars in file\n",count);
qsort(Frequencies,256,sizeof(int),compare);
for (count=0; count<256;count++) {
if (Frequencies[count] != 0) {
printf("%3c (%4d) = %d\n",
count,
count,
Frequencies[count]);
}
}
return 0;
}
We compile, link, and then we write
frequencies frequencies.c
957 chars in file
66 C Tutorial

Well, sorting definitely works (you read this display line by line), but we note with dismay that

À ( 192) = 1 Á ( 193) = 1 Â ( 194) = 1

à ( 195) = 1 Ä ( 196) = 1 Å ( 197) = 1

Æ ( 198) = 1 Ç ( 199) = 1 È ( 200) = 1

É ( 201) = 1 Ê ( 202) = 1 Ë ( 203) = 2

Ì ( 204) = 2 Í ( 205) = 2 Î ( 206) = 2

Ï ( 207) = 2 Ð ( 208) = 2 Ñ ( 209) = 3

Ò ( 210) = 3 Ó ( 211) = 3 Ô ( 212) = 3

Õ ( 213) = 3 Ö ( 214) = 3 × ( 215) = 4

Ø ( 216) = 4 Ù ( 217) = 4 Ú ( 218) = 4

Û ( 219) = 5 Ü ( 220) = 5 Ý ( 221) = 5

Þ ( 222) = 5 ß ( 223) = 6 à ( 224) = 6

á ( 225) = 6 â ( 226) = 7 ã ( 227) = 7

ä ( 228) = 7 å ( 229) = 7 æ ( 230) = 7

ç ( 231) = 7 è ( 232) = 8 é ( 233) = 8

ê ( 234) = 10 ë ( 235) = 10 ì ( 236) = 10

í ( 237) = 11 î ( 238) = 11 ï ( 239) = 13

ð ( 240) = 16 ñ ( 241) = 20 ò ( 242) = 21

ó ( 243) = 21 ô ( 244) = 21 õ ( 245) = 24

ö ( 246) = 24 ÷ ( 247) = 25 ø ( 248) = 28

ù ( 249) = 35 ú ( 250) = 38 û ( 251) = 39

ü ( 252) = 46 ý ( 253) = 52 þ ( 254) = 52

ÿ ( 255) = 93

All the character names are wrong!

Why?
Well we have never explicitly stored the name of a character in our integer array; it was
implicitly stored. The sequence of elements in the array corresponded to a character value. But
once we sort the array, this ordering is gone, and we have lost the correspondence between
each array element and the character it was representing.
C offers us many solutions to this problem, but this is taking us too far away from array han-
dling, the subject of this section. We will have to wait until we introduce structures and user
types before we can solve this problem.
 Pointers and references 67

1.13.1 Summary of Arrays and sorting

• Arrays are declared by indicating their size in square brackets, after the identifier
declaration: <type> identifier[SIZE];
• Arrays are equivalent to pointers to their first element.
• Arrays “decay”, i.e. are transformed into pointers, when passed to other functions.
• You can sort an array using the qsort function.

1.14 Pointers and references

Pointers are one of the “hard” subjects of the C language. They are somehow mysterious, quite
difficult for beginners to grasp, and their extensive use within C makes them unavoidable.
Pointers are machine addresses, i.e. they point to data. It is important to have clear this distinc-
tion: pointers are NOT the data they point to, they contain just a machine address where the
data will be found. When you declare a pointer like this:
FILE *infile;
you are declaring: reserve storage for a machine address and not a FILE structure. This
machine address will contain the location where that structure starts.
This allows us to pass a small machine address around, instead of a big FILE structure that has
dozens of fields. The big advantage of pointers is that they are efficient. But, as anything, they
are error prone. There is no free lunch.
The contents of the pointer are undefined until you initialize it. Before you initialize a pointer,
its contents can be anything; it is not possible to know what is in there, until you make an
assignment. A pointer before is initialized is a dangling pointer, i.e. a pointer that points to
nowhere.
A pointer can be initialized by:

1) Assign it a special pointer value called NULL, i.e. empty.

2) Assignment from a function or expression that returns a pointer of the same type. In the
frequencies example we initialize our infile pointer with the function fopen, that returns a
pointer to a FILE.

3) Assignment to a specific address. This happens in programs that need to access certain
machine addresses for instance to use them as input/output for special devices. In those
cases you can initialize a pointer to a specific address. Note that this is not possible under
windows, or Linux, or many operating systems where addresses are virtual addresses. More
of this later.

4) You can assign a pointer to point to some object by taking the address of that object. For
instance:
int integer;
int *pinteger = &integer;
Here we make the pointer “pinteger” point to the int “integer” by taking the address of that
integer, using the “&” operator. This operator yields the machine address of its argument.

5) You can access the data the pointer is pointing to by using the “*” operator. When we want
to access the integer “pinteger” is pointing to, we write:
*pinteger = 7;
68 C Tutorial

This assigns to the “integer” variable indirectly the value 7.

In lcc-win32 pointers can be of two types. We have normal pointers, as we have described
above, and “references”, i.e. compiler maintained pointers, that are very similar to the objects
themselves. 54
References are declared in a similar way as pointers are declared:55
int a = 5; // declares an integer a
int * pa = &a;// declares a pointer to the integer a
int &ra = a;// declares a reference to the integer a
Here we have an integer, that within this scope will be called “a”. Its machine address will be
stored in a pointer to this integer, called “pa”. This pointer will be able to access the data of
“a”, i.e. the value stored at that machine address by using the “*” operator. When we want to
access that data we write:
*pa = 8944;

This means: “store at the address contained in this pointer pa, the value 8944”.
We can also write:
int m = 698 + *pa;

This means: “add to 698 the contents of the integer whose machine address is contained in the
pointer pa and store the result of the addition in the integer m”
We have a “reference” to a, that in this scope will be called “ra”. Any access to this compiler
maintained pointer is done as we would access the object itself, no special syntax is needed.
For instance we can write:
ra = (ra+78) / 79;
Note that with references the “*” operator is not needed. The compiler will do automatically
this for you.
It is obvious that a question arises now: why do we need references? Why can’t we just use the
objects themselves? Why is all this pointer stuff necessary?
Well this is a very good question. Many languages seem to do quite well without ever using
pointers the way C does.
The main reason for these constructs is efficiency. Imagine you have a huge database table,
and you want to pass it to a routine that will extract some information from it. The best way to
pass that data is just to pass the address where it starts, without having to move or make a copy
of the data itself. Passing an address is just passing a 32-bit number, a very small amount of
data. If we would pass the table itself, we would be forced to copy a huge amount of data into
the called function, what would waste machine resources.
The best of all worlds are references. They must always point to some object, there is no such
a thing as an uninitialized reference. Once initialized, they can’t point to anything else but to
the object they were initialized to, i.e. they can’t be made to point to another object, as normal
pointers can. For instance, in the above expressions, the pointer pa is initialized to point to the
integer “a”, but later in the program, you are allowed to make the “pa” pointer point to

54. This has nothing to do with object oriented programming of course. The word object is used here
with its generic meaning.
55. References aren’t part of the C language standard, and are in this sense an extension of lcc-win32.
They are wildly used in another related language (C++), and the implementation of lcc-win32 is compatible
with the implementation of references of that language.
 Pointers and references 69

another, completely unrelated integer. This is not possible with the reference “ra”. It will
always point to the integer “a”.
When passing an argument to a function, if that function expects a reference and you pass it a
reference, the compiler will arrange for you passing only the address of the data pointed to by
the reference.
70 C Tutorial

1.15 Structures and unions

1.15.1 Structures
Structures are a contiguous piece of storage that contains several simple types, grouped as a
single object.56 For instance, if we want to handle the two integer positions defined for each
pixel in the screen we could define the following structure:
struct coordinates {
int x;
int y;
};
Structures are introduced with the keyword “struct” followed by their name. Then we open a
scope with the curly braces, and enumerate the fields that form the structure. Fields are
declared as all other declarations are done. Note that a structure declaration is just that, a dec-
laration, and it reserves no actual storage anywhere.
After declaring a structure, we can use this new type to declare variables or other objects of
this type:
struct coordinate Coords = { 23,78};
Here we have declared a variable called Coords, that is a structure of type coordinate, i.e. hav-
ing two fields of integer type called “x” and “y”. In the same statement we initialize the struc-
ture to a concrete point, the point (23,78). The compiler, when processing this declaration, will
assign to the first field the first number, i.e. to the field “x” will be assigned the value 23, and
to the field “y” will be assigned the number 78.
Note that the data that will initialize the structure is enclosed in curly braces.
Structures can be recursive, i.e. they can contain pointers to themselves. This comes handy to
define structures like lists for instance:
struct list {
struct list *Next;
int Data;
};
Here we have defined a structure that in its first field contains a pointer to the same structure,
and in its second field contains an integer. Please note that we are defining a pointer to an iden-
tical structure, not the structure itself, what is impossible. A structure can’t contain itself.
Double linked list can be defined as follows:
struct dl_list {
struct dl_list *Next;
struct dl_list *Previous;
int Data;
};
This list features two pointers: one forward, to the following element in the list, and one back-
ward, to the previous element of the list.
A special declaration that can only be used in structures is the bit-field declaration. You can
specify in a structure a field with a certain number of bits. That number is given as follows:
struct flags {
unsigned HasBeenProcessed:1;

56. The usage of the #pragma pack construct is explained in lcc-win32 user’s manual. Those
explanations will not be repeated here.
 Structures and unions 71

unsigned HasBeenPrinted:1;
unsigned Pages:5;
};
This structure has three fields. The first, is a bit-field of length 1, i.e. a Boolean value, the sec-
ond is also a bit-field of type Boolean, and the third is an integer of 5 bits. In that integer you
can only store integers from zero to 31, i.e. from zero to 2 to the 5th power, minus one. In this
case, the programmer decides that the number of pages will never exceed 31, so it can be
safely stored in this small amount of memory.
We access the data stored in a structure with the following notation:
<structure-name> ‘.’ field-name
or
<structure-name ‘->’ field-name
We use the second notation when we have a pointer to a structure, not the structure itself.
When we have the structure itself, or a reference variable, we use the point.
Here are some examples of this notation:
void fn(void)
{
coordinate c;
coordinate *pc;
coordinate &rc = c;

c.x = 67; // Assigns the field x

c.y = 78;// Assigns the field y
pc = &c; // We make pc point to c
pc->x = 67; // We change the field x to 67
pc->y = 33;// We change the field y to 33
rc.x = 88;// References use the point notation
}
Structures can contain other structures or types. After we have defined the structure coordinate
above, we can use that structure within the definition of a new one.
struct DataPoint {
struct coordinate coords;
int Data;
};
This structure contains a “coordinate” structure. To access the “x” field of our coordinate in a
DataPoint structure we would write:
struct DataPoint dp;
dp.coords.x = 78;
Structures can be contained in arrays. Here, we declare an array of 25 coordinates:
struct coordinate coordArray[25];

To access the x coordinate from the 4th member of the array we would write:
coordArray[3].x = 89;
Note (again) that in C array indexes start at zero. The fourth element is numbered 3.
Many other structures are possible their number is infinite:
struct customer {
int ID;
char *Name;
char *Address;
double balance;
72 C Tutorial

time_t lastTransaction;
unsigned hasACar:1;
unsigned mailedAlready:1;
};
This is a consecutive amount of storage where:
• an integer contains the ID of the customer,
• a machine address pointing to the start of the character string with the customer name,
• another address pointing to the start of the name of the place where this customer lives,
• a double precision number containing the current balance,
• a time_t (time type) date of last transaction,
• and other bit fields for storing some flags.
struct mailMessage {
MessageID ID;
time_t date;
char *Sender;
char *Subject;
char *Text;
char *Attachements;
};
This one starts with another type containing the message ID, again a time_t to store the date,
then the addresses of some character strings.
The set of functions that use a certain type are the methods that you use for that type, maybe in
combination with other types. There is no implicit “this” in C. Each argument to a function is
explicit, and there is no predominance of anyone.
A customer can send a mailMessage to the company, and certain functions are possible,
that handle mailMessages from customers. Other mailMessages aren’t from custom-
ers, and are handled differently, depending on the concrete application.
Because that’s the point here: an application is a coherent set of types that performs a certain
task with the computer, for instance, sending automated mailings, or invoices, or sensing the
temperature of the system and acting accordingly in a multi-processing robot, or whatever. It
is up to you actually.
Note that in C there is no provision or compiler support for associating methods in the struc-
ture definitions. You can, of course, make structures like this:
struct customer {
int ID;
char *Name;
char *Address;
double balance;
time_t lastTransaction;
unsigned hasACar:1;
unsigned mailedAlready:1;
bool (*UpdateBalance)(struct customer *Customer,
double newBalance);
};
The new field, is a function pointer that contains the address of a function that returns a Bool-
ean result, and takes a customer and a new balance, and should (eventually) update the balance
field, that isn’t directly accessed by the software, other than trough this procedure pointer.
When the program starts, you assign to each structure in the creation procedure for it, the func-
tion DefaultGetBalance() that takes the right arguments and does hopefully the right thing.
 Structures and unions 73

This allows you the flexibility of assigning different functions to a customer for calculating
his/her balance according to data that is known only at runtime. Customers with a long history
of overdraws could be handled differently by the software after all. But this is no longer C, is
the heart of the application.
True, there are other languages that let you specify with greater richness of rules what and how
can be sub classed and inherited. C, allows you to do anything, there are no other rules here,
other the ones you wish to enforce.
You can subclass a structure like this. You can store the current pointer to the procedure some-
where, and put your own procedure instead. When your procedure is called, it can either:
Do some processing before calling the original procedure
Do some processing after the original procedure returns
Do not call the original procedure at all and replace it entirely.
We will show a concrete example of this when we speak about windows sub classing later.
Sub classing allows you to implement dynamic inheritance. This is just an example of the
many ways you can program in C.
But is that flexibility really needed?
Won’t just
bool UpdateBalance(struct customer *pCustomer, double newBalance);
do it too?
Well it depends. Actions of the general procedure could be easy if the algorithm is simple and
not too many special cases are in there. But if not, the former method, even if more compli-
cated at first sight, is essentially simpler because it allows you greater flexibility in small man-
ageable chunks, instead of a monolithic procedure of several hundred lines full of special case
code…
Mixed strategies are possible. You leave for most customers the UpdateBalance field empty
(filled with a NULL pointer), and the global UpdateBalance procedure will use that field to
calculate its results only if there is a procedure there to call. True, this wastes 4 bytes per cus-
tomer in most cases, since the field is mostly empty, but this is a small price to pay, the struc-
ture is probably much bigger anyway.

1.15.2 Structure size

In principle, the size of a structure is the sum of the size of its members. This is, however, just
a very general rule, since it depends a lot on the compilation options valid at the moment of
the structure definition, or in the concrete settings of the structure packing as specified with
the #pragma pack() construct.57
Normally, you should never make any assumptions about the specific size of a structure. Com-
pilers, and lcc-win32 is no exception, try to optimize structure access by aligning members of
the structure at predefined addresses. For instance, if you use the memory manager, pointers
must be aligned at addresses multiples of four, if not, the memory manager doesn’t detect
them and that can have disastrous consequences.

57. Note that putting structure names in typedefs all uppercase is an old habit that somehow belongs to
the way I learned C, but is in no way required by the language. Personally I find those all-uppercase names
clearer as a way of indicating to the reader that a user defined type and not a variable is used, since I have
never used an all-uppercase name for a variable name. Separating these names by upper/lower case improves
the readability of the program, but this is a matter of personal taste.
74 C Tutorial

The best thing to do is to always use the sizeof operator when the structure size needs to be
used somewhere in the code. For instance, if you want to allocate a new piece of memory
managed by the memory manager, you call it with the size of the structure.
GC_malloc(sizeof(struct DataPoint)*67);
This will allocate space for 67 structures of type “DataPoint” (as defined above). Note that we
could have written
GC_malloc(804);
since we have:
struct DataPoint {
struct coordinate coords;
int Data;
};
We can add the sizes:
Two integers of 4 bytes for the coordinate member, makes 8 bytes, plus 4 bytes for the Data
member, makes 12, that multiplies 67 to make 804 bytes.
But this is very risky because of two reasons:
Compiler alignment could change the size of the structure
If you add a new member to the structure, the sizeof() specification will continue to work,
since the compiler will correctly recalculate it each time. If you write the 804 however, when
you add a new member to the structure this number has to be recalculated again, making one
more thing that can go wrong in your program.
In general, it is always better to use compiler-calculated constants like sizeof() instead of hard-
wired numbers.

1.15.3 Defining new types

Structures are then, a way of augmenting the type system by defining new types using already
defined ones. The C language allows you to go one step further in this direction by allowing
you to specify a new type definition or typedef for short.
This syntax for doing this is like this:
typedef <already defined type> <new name>;
For instance, you can specify a new type of integer called “my integer” with:
typedef int my_integer;
Then, you can use this new type in any position where the “int” keyword would be expected.
For instance you can declare:
my_integer i;
instead of:
int i;
This can be used with structures too. For instance, if you want to avoid typing at each time you
use a coordinate struct coordinate a; you can define
typedef struct coordinate COORDINATE;
and now you can just write:
COORDINATE a;

what is shorter, and much clearer.58

 Structures and unions 75

This new name can be used with the sizeof() operator too, and we can write:
GC_malloc(sizeof(COORDINATE));
instead of the old notation. But please keep in mind the following: once you have defined a
typedef, never use the “struct” keyword in front of the typedef keyword, if not, the compiler
will get really confused.

1.15.4 Unions
Unions are similar to structures in that they contain fields. Contrary to structures, unions will
store all their fields in the same place. They have the size of the biggest field in them. Here is
an example:
union intfloat {
int i;
double d;
};
This union has two fields: an integer and a double precision number. The size of an integer is
four in lcc-win32, and the size of a double is eight. The size of this union will be eight bytes,
with the integer and the double precision number starting at the same memory location. The
union can contain either an integer or a double precision number but not the two. If you store
an integer in this union you should access only the integer part, if you store a double, you
should access the double part. Field access syntax is the same as for structures: we use always
the point.
Using the definition above we can write:
int main(void)
{
union intfloat ifl;
union intfloat *pIntfl = &ifl;

pIntfl.i = 2;
pintfl.d = 2.87;
}
First we assign to the integer part of the union an integer, then we assign to the double preci-
sion part a double.
Unions are useful for storing structures that can have several different memory layouts. In
general we have an integer that tells us which kind of data follows, then a union of several
types of data. Suppose the following data structures:
struct fileSource {
char *FileName;
int LastUse;
};

struct networkSource {
int socket;
char *ServerName;
int LastUse;
};

struct windowSource {
WINDOW window;

58. An identifier can also represent a macro or a macro argument, but here we will assume that the pre-
processor already has done its work.
76 C Tutorial

int LastUse;
};
All of this data structures should represent a source of information. We add the following
defines:
#define ISFILE 1
#define ISNETWORK 2
#define ISWINDOW 3
and now we can define a single information source structure:
struct Source {
int type;
union {
struct fileSource file;
struct networkSource network;
struct windowSource window;
} info;
};
We have an integer at the start of our generic “Source” structure that tells us, which of the fol-
lowing possible types is the correct one. Then, we have a union that describes all of our possi-
ble data sources.
We fill the union by first assigning to it the type of the information that follows, an integer that
must be one of the defined constants above. Then we copy to the union the corresponding
structure. Note that we save a lot of wasted space, since all three structures will be stored
beginning at the same location. Since a data source must be one of the structure types we have
defined, we save wasting memory in fields that would never get used.
Another usage of unions is to give a different interpretation of the same data. For instance, an
MMX register in an x86 compatible processor can be viewed as two integers of 32 bits, 4 inte-
gers of 16 bits, or 8 integers of 8 bits. Lcc-win32 describes this fact with a union:
typedef struct _pW {
char high;
char low;
} _packedWord; // 16 bit integer

typedef struct _pDW {

_packedWord high;
_packedWord low;
} _packedDWord; // 32 bit integer of two 16 bit integers

typedef struct _pQW {

_packedDWord high;
_packedDWord low;
} _packedQWord; // 64 bits of two 32 bit structures

typedef union __Union {

_packedQWord packed;
int dwords[2];
short words[4];
char bytes[8];
} _mmxdata; // This is the union of all those types
Union usage is not checked by the compiler, i.e. if you make a mistake and access the wrong
member of the union, this will provoke a trap or another failure at run time. One way of
debugging this kind of problem is to define all unions as structures during development, and
see where you access an invalid member. When the program is fully debugged, you can switch
back to the union usage.
 Using structures 77

1.16 Using structures

Now that we know how we can define structures we can (at last) solve the problem we had
with our character frequencies program.
We define a structure containing the name of the character like this:
typedef struct tagChars {
int CharacterValue;
int Frequency;
} CHARS;
Note that here we define two things in a single statement: we define a structure called “tag-
Chars” with two fields, and we define a typedef CHARS that will be the name of this type.
Within the program, we have to change the following things:
We have to initialize the name field of the array that now will be an array of structures and not
an array of integers.
When each character is read we have to update the frequency field of the corresponding struc-
ture.
When displaying the result, we use the name field instead of our count variable.
Here is the updated program:
#include <stdio.h>
#include <stdlib.h>

typedef struct tagChars {

int CharacterValue;
int Frequency;
} CHARS;

CHARS Frequencies[256]; // Array of frequencies

int compare( const void *arg1, const void *arg2 )

{
CHARS *Arg1 = (CHARS *)arg1;
CHARS *Arg2 = (CHARS *)arg2;
/* Compare both integers */
return ( Arg2->Frequency - Arg1->Frequency );
}

int main(int argc,char *argv[])

{
int count=0;
FILE *infile;
int c;

if (argc < 2) {
printf("Usage: countchars <file name>\n");
exit(1);
}
infile = fopen(argv[1],"rb");
if (infile == NULL) {
printf("File %s doesn't exist\n",argv[1]);
exit(1);
}
for (int i = 0; i<256; i++) {
Frequencies[i].CharacterValue = i;
}
78 C Tutorial

c = fgetc(infile);
while (c != EOF) {
count++;
if (c >= ' ')
Frequencies[c].Frequency++;
c = fgetc(infile);
}
fclose(infile);
printf("%d chars in file\n",count);
qsort(Frequencies,256,sizeof(CHARS),compare);
for (count=0; count<256;count++) {
if (Frequencies[count].Frequency != 0) {
printf("%3c (%4d) = %d\n",
Frequencies[count].CharacterValue,
Frequencies[count].CharacterValue,
Frequencies[count].Frequency);
}
}
return 0;
}
We transformed our integer array Frequencies into a CHARS array with very few changes:
just the declaration. Note that the array is still accessed as a normal array would. By the way, it
is a normal array.
We changed our “compare” function too, obviously, since we are now comparing two CHARS
structures, and not just two integers. We have to cast our arguments into pointers to CHARS,
and I decided that using two temporary variables would be clearer than a complicated expres-
sion that would eliminate those.
The initialization of the CharacterValue field is trivially done in a loop, just before we start
counting chars. We assign to each character an integer from 0 to 256 that’s all.
When we print our results, we use that field to get to the name of the character, since our array
that before qsort was neatly ordered by characters, is now ordered by frequency. As before, we
write the character as a letter with the %c directive, and as a number, with the %d directive.
When we call this program with:
frequencies frequencies.c
we obtain at last:
1311 chars in file

( 32) = 154 e ( 101) = 77 n ( 110) = 60

i ( 105) = 59 r ( 114) = 59 c ( 99) = 52

t ( 116) = 46 u ( 117) = 35 a ( 97) = 34

; ( 59) = 29 o ( 111) = 29 f ( 102) = 27

s ( 115) = 26 ( ( 40) = 25 ) ( 41) = 25

l ( 108) = 20 g ( 103) = 18 F ( 70) = 17

q ( 113) = 16 = ( 61) = 15 C ( 67) = 13

h ( 104) = 12 A ( 65) = 12 d ( 100) = 11

, ( 44) = 11 [ ( 91) = 10 ] ( 93) = 10

* ( 42) = 10 " ( 34) = 10 { ( 123) = 9

 Using structures 79

2 ( 50) = 9 p ( 112) = 9 } ( 125) = 9

1 ( 49) = 8 . ( 46) = 8 y ( 121) = 8

+ ( 43) = 8 S ( 83) = 7 R ( 82) = 7

H ( 72) = 7 > ( 62) = 6 < ( 60) = 6

% ( 37) = 5 m ( 109) = 5 v ( 118) = 5

0 ( 48) = 5 / ( 47) = 4 5 ( 53) = 4

\ ( 92) = 4 V ( 86) = 4 6 ( 54) = 4

- ( 45) = 3 x ( 120) = 3 b ( 98) = 3

' ( 39) = 3 L ( 76) = 3 ! ( 33) = 2

: ( 58) = 2 # ( 35) = 2 U ( 85) = 2

E ( 69) = 2 4 ( 52) = 1 I ( 73) = 1

w ( 119) = 1 O ( 79) = 1 z ( 122) = 1

3 ( 51) = 1 N ( 78) = 1

We see immediately that the most frequent character is the space with a count of 154, followed
by the letter ‘e’ with a count of 77, then ‘n’ with 60, etc.
Strange, where does “z” appear? Ah yes, in sizeof. And that I? Ah in FILE, ok, seems to be
working.

1.16.1 Fine points of structure use

1) When you have a pointer to a structure and you want to access a member of it you should
use the syntax:
pointer->field

2) When you have a structure OBJECT, not a pointer, you should use the syntax:
object.field
Beginners easily confuse this.

3) When you have an array of structures, you index it using the normal array notation syntax,
then use the object or the pointer in the array. If you have an array of pointers to structures
you use:
array[index]->field

4) If you have an array of structures you use:

array[index].field

5) If you are interested in the offset of the field, i.e. the distance in bytes from the beginning of
the structure to the field in question you use the offsetof macro defined in stddef.h:
offsetof(structure or typedef name,member name)
For instance to know the offset of the Frequency field in the structure CHARS above we
would write:
offsetof(CHARS,Frequency)
This would return an integer with the offset in bytes.
80 C Tutorial

Summary: Files are a sequence of bytes. They are central to most programs. Here is a short
overview of the functions that use files:

Name Purpose

fopen Opens a file

fclose Closes a file

fprintf Formatted output to a file

fputc Puts a character in a file

putchar Puts a character to stdout

fputs Puts a string in a file.

fread Reads from a file a specified amount of data into a buffer.

freopen Reassigns a file pointer

fscanf Reads a formatted string from a file

fsetpos Assigns the file pointer (the current position)

Moves the current position relative to the start of the file, to the end of the file, or relative
fseek
to the current position

ftell returns the current position

fwrite Writes a buffer into a file

tmpnam Returns a temporary file name

unlink Erases a file

remove Erases a file

rename Renames a file.

rewind Repositions the file pointer to the beginning of a file.

setbuf Controls file buffering.

ungetc Pushes a character back into a file.

1.17 Identifier scope and linkage

Until now we have used identifiers and scopes without really caring to define precisely the
details. This is unavoidable at the beginning, some things must be left unexplained at first, but
it is better to fill the gaps now.
An identifier in C can denote:59

59. You see the infinite loop here? Tell me: why is this loop never ending? Look at the code again.
 Identifier scope and linkage 81

• an object.
• a function
• a tag or a member of a structure, union or enum
• a typedef
• a label
For each different entity that an identifier designates, the identifier can be used (is visible)
only within a region of a program called its scope. There are four kinds of scopes in C.
The file scope is built from all identifiers declared outside any block or parameter declaration,
it is the outermost scope, where global variables and functions are declared.
A function scope is given only to label identifiers.
The block scope is built from all identifiers that are defined within the block. A block scope
can nest other blocks.
The function prototype scope is the list of parameters of a function. Identifiers declared
within this scope are visible only within it.
Let’s see a concrete example of this:
static int Counter = 780;// file scope
extern void fn(int Counter); // function prototype scope
void function(int newValue, int Counter) // Block scope
{
double d = newValue;
label:
for (int i = 0; i< 10;i++) {
if (i < newValue) {
char msg[45];
int Counter = 78;

sprintf(msg,"i=%d\n",i*Counter); !
}
if (i == 4)
goto label;60
}
}
At the point indicated by the arrow, the poor “Counter” identifier has had a busy life:
• It was bound to an integer object with file scope
• Then it had another incarnation within the function prototype scope
• Then, it was bound to the variables of the function ‘setCounter’ as a parameter
• That definition was again “shadowed” by a new definition in an inner block, as a local
variable.
The value of “Counter” at the arrow is 78. When that scope is finished its value will be the
value of the parameter called Counter, within the function “function”.
When the function definition finishes, the file scope is again the current scope, and “Counter”
reverts to its value of 780.

60. Yes, but then all initializations are done out of their respective contexts. Some people say this is the
wrong way to go, and that each data type should initialize in a separate init procedure. In this concrete
example and in many situations, making a global init procedure is a correct way of building software. Other
contexts may be different of course.
82 C Tutorial

The “linkage” of an identifier refers to the visibility to other modules. Basically, all identifiers
that appear at a global scope (file scope) and refer to some object are visible from other mod-
ules, unless you explicitly declare otherwise by using the “static” keyword.
Problems can appear if you first declare an identifier as static, and later on, you define it as
external. For instance:
static void foo(void);
and several hundred lines below you declare:
void foo(void) {
...
}
Which one should the compiler use? static or not static? That is the question…
Lcc-win32 chooses always non-static, to the contrary of Microsoft’s compiler that chooses
always static. Note that the behavior of the compiler is explicitly left undefined in the stan-
dard, so both behaviors are correct.

1.18 Top-down analysis

The goal of this introduction is not to show you a lot of code, but to tell you how that code is
constructed. A central point in software construction is learning how you decompose a task in
sub-tasks, so that the whole is more modular, and easier to manage and change. Let’s go back
to our frequencies example. We see that the “main” function accomplishes several tasks: it
checks its arguments, opens a file, checks the result, initializes the Frequency array, etc.
This is an example of a monolithic, highly complex function. We could decompose it into
smaller pieces easily, for example by assigning each task to a single procedure.
One of the first things we could do is to put the checking of the input arguments in a single
procedure:
FILE *checkargs(int argc,char *argv[])
{
FILE *infile = NULL;
if (argc < 2) {
printf("Usage: countchars <file name>\n");
}
else {
infile = fopen(argv[1],"rb");
if (infile == NULL) {
printf("File %s doesn't exist\n",argv[1]);
}
}
return infile;
}
We pass the arguments of main to our check procedure, that writes the corresponding error
message if appropriate, and returns either an open FILE *, or NULL, if an error was detected.
The calling function just tests this value, and exits if something failed.
int main(int argc,char *argv[])
{
int count=0;
FILE *infile = checkargs(argc,argv);
int c;

if (infile == NULL)
return 1;
 Top-down analysis 83

for (int i = 0; i<256; i++) {

Frequencies[i].CharacterValue = i;
}
... the rest of "main" ...
}
The details of how the checking of the arguments is done is now away from the main line of
the «main» function. The whole function is now easier to read, and the title of the piece of
code that we hid away tells the reader immediately what is going on behind that function call.
The next step, is the initializing of the Frequencies array. This is common pattern that we find
very often when building software: most programs initialize tables, and do some setup before
actually beginning the computation they want to perform. The best is to collect all those ini-
tializations into a single procedure, so that any new initializations aren’t scattered all around
but concentrated in a single function.61
void Initialize(void)
{
for (int i = 0; i<256; i++) {
Frequencies[i].CharacterValue = i;
}
}
Following our analysis of “main”, we see that the next steps are the processing of the opened
file. We read a character, and we update the Frequencies array. Well, this is a single task that
can be delegated to a function, a function that would need the opened file handle to read from,
and a Frequencies array pointer to be updated.
We develop a ProcessFile function as follows:
int ProcessFile(FILE *infile,CHARS *Frequencies)
{
int count = 0;
int c = fgetc(infile);62
while (c != EOF) {
count++;
if (c >= ' ')
Frequencies[c].Frequency++;
c = fgetc(infile);
}
return count;
}
The interface with the rest of the software for this function looks like this:
count = ProcessFile(infile,Frequencies);
We could have avoided passing the Frequencies array to ProcessFile, since it is a global vari-
able. Its scope is valid when we are defining ProcessFile, and we could use that array directly.

61. We find very often the expression:

while ((c=fgetc(infile)) != EOF) { ... }
instead of the expression above. Both expressions are strictly equivalent, since we first execute the fgetc
function, assigning the result to c, then we compare that result with EOF. In the second, slightly more
complicated, we need a set of parentheses to force execution to execute the fgetc and the assignment first.
There is the danger that c would get assigned the result of the comparison of the fgetc result with EOF instead
of the character itself.
62. The standard files defined by the standard are: stdin, or standard input, to read from the current input
device, the stdout or standard output, and the stderr stream, to show errors. Initially, stdin is bound to the
keyboard, stdout and stderr to the screen.
84 C Tutorial

But there are good reasons to avoid that. Our global array can become a bottleneck, if we
decide later to process more than one file, and store the results of several files, maybe combin-
ing them and adding up their frequencies.
Another reason to explicitly pass the Frequencies array as a parameter is of course clarity.
The Frequencies array is a parameter of this function, since this function modifies it. Pass-
ing it explicitly to a routine that modifies it makes the software clearer, and this is worth the
few cycles the machine needs to push that address in the stack.
When we write software in today’s powerful micro-processors, it is important to get rid of the
frame of mind of twenty years ago, when saving every cycle of machine time was of utmost
importance. Pushing an extra argument, in this case the address of the Frequencies array,
takes 1 cycle. At a speed of 1400-2500 MHz, this cycle isn’t a high price to pay.
Continuing our analysis of our “main” function, we notice that the next task, is displaying the
output of the frequencies array. This is quite a well-defined task, since it takes the array as
input, and should produce the desired display. We define then, a new function DisplayOut-
put() that will do that. Its single parameter is the same Frequencies array.
void DisplayOutput(CHARS *Frequencies)
{
for (int count=0; count<256;count++) {
if (Frequencies[count].Frequency != 0) {
printf("%3c (%4d) = %d\n",
Frequencies[count].CharacterValue,
Frequencies[count].CharacterValue,
Frequencies[count].Frequency);
}
}
}
Let’s look at our “main() function again:
int main(int argc,char *argv[])
{
int count;
FILE *infile = checkargs(argc,argv);

if (infile == NULL)
return 1;
Initialize();
count = ProcessFile(infile,Frequencies);
fclose(infile);
printf("%d chars in file\n",count);
qsort(Frequencies,256,sizeof(CHARS),compare);
DisplayOutput(Frequencies);
}
Note how much clearer our “main” function is now. Instead of a lot of code without any struc-
ture we find a much smaller procedure that is constructed from smaller and easily understand-
able parts.
Now, suppose that we want to handle several files. With this organization, it is straightforward
to arrange for this in a loop. ProcessFile() receives an open FILE and a Frequencies array, both
can be easily changed now. A modular program is easier to modify than a monolithic one!
We see too that the function that process the files leaves them open. We could streamline more
«main» if we got rid of that in the ProcessFile() function, but I find it personally better that the
same function that opens a file closes it too, so that the reader can see if the fopen/fclose calls
match.
 Extending a program 85

1.19 Extending a program

Let’s suppose then, that we want to investigate all frequencies of characters in a directory,
choosing all files that end in a specific extension, for instance *.c. In principle this is easy, we
pass to ProcessFile a different open file each time, and the same Frequencies array.
We develop a GetNextFile function, that using our “*.c” character string, will find the first file
that meets this name specifications (“foo.c” for example), and then will find all the other files
in the same directory that end with a .c extension.
How do we do this?
Well, looking at the documentation, we find that we have two functions that could help us:
findfirst, and findnext. The documentation tells us that findfirst has a prototype like this:
long findfirst( char *spec, struct _finddata_t *fileinfo);
This means that it receives a pointer to a character string, and will fill the fileinfo structure
with information about the file it finds, if it finds it.
We can know if findfirst really found something by looking at the return result. It will be –1 if
there was an error, or a “unique value” otherwise. The documentation tell us too that errno, the
global error variable will be set to ENOENT if there wasn’t any file, or to EINVAL if the file
specification itself contained an error and findfirst couldn’t use it.
Then, there is findnext that looks like this:
int findnext(long handle,struct _finddata_t *fileinfo);
It uses the “unique value” returned by findfirst, and fills the fileinfo structure if it finds another
file that matches the original specification. If it doesn’t find a file it will return –1, as findfirst.
If it does find another file, it will return zero.
We see now that a FILE that was uniquely bound to a name is now a possible ambiguous file
specification, that can contain a _finddata_t (whatever that is) that can be used to read several
FILEs.
We could formalize this within our program like this:
typedef struct tagStream {
char *Name;
struct _finddata_t FindData;
long handle;
FILE *file;
} STREAM;
Our function checkargs() returns a FILE pointer now. It could return a pointer to this
STREAM structure we have just defined, or NULL, if there was an error. Our program then,
would loop asking for the next file, adding to our Frequencies array the new character frequen-
cies found.
The first function to be modified is checkargs. We keep the interface with the calling function
(return NULL on error), but we change the inner workings of it, so that instead of calling
fopen, it calls findfirst.
STREAM *checkargs(int argc,char *argv[])
{
STREAM *infile = NULL;
long findfirstResult;
struct _finddata_t fd;

if (argc < 2) {
printf("Usage: countchars <file name>\n");
}
86 C Tutorial

else {
findfirstResult = findfirst(argv[1],&fd);
if (findfirstResult < 0) {
printf("File %s doesn't exist\n",argv[1]);
return NULL;
}
}
infile = malloc(sizeof(STREAM));
infile->Name = argv[1];
memcpy(&infile->FindData, &fd,
sizeof( struct _finddata_t ));
infile->File = fopen(fd.name,"rb");
infile->handle = findfirstResult;
return infile;
}
We store in the local variable findfirstResult the long returned by findfirst. We test then, if
smaller than zero, i.e. if something went wrong. If findfirst failed, this is equivalent to our
former program when it opened a file and tested for NULL.
But now comes an interesting part. If all went well, we ask the system using the built-in mem-
ory allocator “malloc” for a piece of fresh RAM at least of size STREAM. If this call fails,
there is no more memory left. For the time being (see later) we ignore this possibility.
We want to store in there all the parameters we need to use the findfirst/findnext function pair
with easy, and we want to copy the finddata_t into our own structure, and even put the name of
the stream and a FILE pointer into it. To do that, we need memory, and we ask it to the “mal-
loc” allocator.
Once that done, we fill the new STREAM with the data:
• we set its name using the same pointer as argv[1],
• we copy the fd variable into our newly allocated structure, and
• we set the file pointer of our new structure with fopen, so that we can use the stream to
read characters from it.
Another alternative to using the built-in memory allocator would have been to declare a global
variable, call it CurrentStream that would contain all our data. We could have declared some-
where in the global scope something like:
STREAM CurrentStream;
and use always that variable.
This has several drawbacks however, the bigger of it being that global variables make follow-
ing the program quite difficult. They aren’t documented in function calls, they are always
“passed” implicitly, they can’t be used in a multi-threaded context, etc.
Better is to allocate a new STREAM each time we need one. This implies some memory man-
agement, something we will discuss in-depth later on.
Now, we should modify our ProcessFile function, since we are passing to it a STREAM and
not a FILE. This is easily done like this:
int ProcessFile(STREAM *infile,CHARS *Frequencies)
{
int count = 0;
int c = fgetc(infile->file);
while (c != EOF) {
count++;
if (c >= ' ')
Frequencies[c].Frequency++;
 Extending a program 87

c = fgetc(infile->file);
}
return count;
}
Instead of reading directly from the infile argument, we use the “file” member of it. That’s all.
Note that infile is a pointer, so we use the notation with the arrow, instead of a point to access
the “file” member of the structure.
But there is something wrong with the name of the function. It wrongly implies that we are
processing a FILE instead of a stream. Let’s change it to ProcessStream, and change the name
of the stream argument to instream, to make things clearer:
int ProcessStream(STREAM *instream,CHARS *Frequencies)
{
int count = 0;
int c = fgetc(instream->file);
while (c != EOF) {
count++;
if (c >= ' ')
Frequencies[c].Frequency++;
c = fgetc(instream->file);
}
return count;
}
This looks cleaner.
Now we have to change our “main” function, to make it read all the files that match the given
name.
Our new main procedure looks like this:
int main(int argc,char *argv[])
{
int count=0;
STREAM *infile=checkargs(argc,argv);

if (infile == NULL) {
return(1);
}
Initialize();
do {
count += ProcessStream(infile,Frequencies);
fclose(infile->file);
infile = GetNext(infile);
} while (infile != 0);
printf("%d chars in file\n",count);
qsort(Frequencies,256,sizeof(CHARS),compare);
DisplayOutput(Frequencies);
return 0;
}
We didn’t have to change a lot, thanks to the fact that the complexities of reading and handling
a stream are now hidden in a function, with well-defined parameters. We build a GetNext
function that returns either a valid new stream or NULL, if it fails. It looks like this:
STREAM *GetNext(STREAM *stream)
{
STREAM *result;
struct _finddata_t fd;
long findnextResult = _findnext(stream->handle,&fd);
88 C Tutorial

if (findnextResult < 0)
return NULL;
result = malloc(sizeof(STREAM));
memcpy(result->FindData,
&fd,
sizeof(struct _finddata_t));
result->handle = stream->handle;
result->file = fopen(fd.name,"rb");
return result;
}
In the same manner that we allocate RAM for our first STREAM, we allocate now a new one,
and copy into it our “finddata” handle, and we open the file.
We compile, and we get a compiler warning:
D:\lcc\examples>lcc -g2 freq1.c
Warning freq1.c: 44 missing prototype for memcpy
Warning freq1.c: 94 missing prototype for memcpy
0 errors, 2 warnings
Yes, but where is memcpy defined? We look at the documentation using F1 in Wedit, and we
find out that it needs the <string.h> header file. We recompile and we get:
H:\lcc\examples>lcc freq1.c
Error freq1.c: 95 type error in argument 1 to `memcpy'; found
`struct _finddata_t' expected `pointer to void'
1 errors, 0 warnings
Wow, an error. We look into the offending line, and we see:
memcpy(result->FindData,&fd,sizeof(struct _finddata_t));
Well, we are passing it a structure, and the poor function is expecting a pointer !
This is a serious error. We correct it like this:
memcpy(&result->FindData,&fd,sizeof(struct _finddata_t));
We take the address of the destination structure using the address-of operator “&”. We see that
we would have never known of this error until run-time when our program would have
crashed with no apparent reason; a difficult error to find. Note: always use the right header file
to avoid this kind of errors!
Our program now looks like this:
#include <stdio.h> // We need it for using the FILE structure
#include <stdlib.h>// We need it for using malloc
#include <io.h>// We need it for using findfirst/findnext
#include <string.h>// We need it for memcpy
typedef struct tagChars {
int CharacterValue;// The ASCII value of the character
int Frequency;// How many seen so far
} CHARS;
typedef struct tagStream {
char Name;// Input name with possible “*” or “?” chars in it
struct _finddata_t FindData;
long handle;
FILE *file;// An open file
} STREAM;
CHARS Frequencies[256]; // Array of frequencies
int compare(){} // Skipped, it is the same as above
STREAM *checkargs(int argc,char *argv[])
{
STREAM *infile = NULL;
long findfirstResult;
 Extending a program 89

struct _finddata_t fd;

if (argc < 2) { // Test if enough arguments were passed

printf("Usage: countchars <file name>\n");
}
else {// Call the _findfirst function with the name and info
buffer
findfirstResult = _findfirst(argv[1],&fd);
// Test result of findfirst, and return immediately NULL if
wrong
if (findfirstResult < 0) {
printf("File %s doesn't exist\n",argv[1]);
return NULL;
}
// Ask more memory to the allocator
infile = malloc(sizeof(STREAM));
// Set the name of the new stream
infile->Name = argv[1];
// Note the first argument of this call: it’s the address within the infile structure of the FindData
// member. We take the address with the “&” operator. Since we are using a pointer, we have
// to dereference a pointer, i.e. with “->” and not with the “.”. Note that the “&”’ operator is used
// with the “fd” local variable to obtain a pointer from a structure member.
memcpy(&infile->FindData,&fd,
sizeof(struct _finddata_t));
infile->file = fopen(fd.name,"rb");
infile->handle = findfirstResult;
}
return infile;
}
void Initialize(void) { this is the same as the function before}
int ProcessStream(STREAM *instream,CHARS *Frequencies)
{
int count = 0;
int c = fgetc(instream->file);
while (c != EOF) {
count++;
if (c >= ' ')
Frequencies[c].Frequency++;
c = fgetc(instream->file);
}
return count;
}
void DisplayOutput(CHARS *Frequencies){ this is the same function as before }
STREAM *GetNext(STREAM *stream)
{
STREAM *result;
struct _finddata_t fd;
long findnextResult = _findnext(stream->handle,&fd);

if (findnextResult < 0)
return NULL;
result = malloc(sizeof(STREAM));
memcpy(&result->FindData,&fd,
sizeof(struct _finddata_t));
result->handle = stream->handle;
result->file = fopen(fd.name,"rb");
result->Name = stream->Name;
return result;
}
int main(int argc,char *argv[])
90 C Tutorial

{
int count=0;
STREAM *infile=checkargs(argc,argv);

if (infile == NULL) {
return(1);
}
Initialize();
do {
count += ProcessStream(infile,Frequencies);
fclose(infile->file);
infile = GetNext(infile);
} while (infile != 0);
printf("%d chars in file\n",count);
qsort(Frequencies,256,sizeof(CHARS),compare);
DisplayOutput(Frequencies);
return 0;
}

1.20 Improving the design

There are several remarks that can be done about our program. The first one is that the mem-
ory allocator could very well fail, when there is no more memory available. When the alloca-
tor fails, it returns NULL. Since we never test for this possibility, our program would crash in
low memory conditions. What a shame!
We arrange for this immediately. Instead of using the allocator, we will write a function that
will call the allocator, test the result, and call the exit() routine if there is no more memory left.
Continuing processing without more memory is impossible anyway.
void *xmalloc(unsigned int size)
{
void *result = malloc(size);

if (result == NULL) {
fprintf(sdterr,
No more memory left!\nProcessing stops\n);
exit(1);
}
return result;
}
Note that we keep the same signature, i.e. the same type of result and the same type of argu-
ments as the original function we want to replace. This is function sub classing.
Note too, that we use fprintf instead of printf. Fprintf takes an extra argument, a file where the
output should go. We use the predefined file of standard error, instead of the normal output file
stdout, that printf implicitly takes.
Why?
Because it is possible that the user redirects the output to a file instead of letting the output go
directly to the screen. In that case we would write our error messages to that file, and the user
would not see the error message.63

63. Some people would say that this is not “Standard C”, since the standard doesn’t explicitly allow for
this. But I would like to point out that the standard explicitly states (page 96 of my edition) that: “An
implementation may accept other forms of constant expressions.”. The implementation lcc-win32 then, is free
to accept the above declaration as a constant expression.
 Path handling 91

We change all occurrences of malloc by xmalloc, and this error is gone.

We change too, all other error-reporting functions, to take into account stderr.
But there are other issues. Take for instance our finddata_t structure that we carry around in
each STREAM structure. What’s its use? We do not use it anywhere; just copy it into our
STREAM.
But why we introduced that in the first place?
Well, we didn’t really know much about findfirst, etc, and we thought it could be useful.
So we are stuck with it?
No, not really. Actually, it is very easy to get rid of it. We just change the structure STREAM
like this:
typedef struct tagStream {
char *Name;
long handle;
FILE *file;
} STREAM;
and we take care to erase any references to that member. We eliminate the memcpy calls, and
that’s all. Our program is smaller, uses less memory, and, what is essential, does the same
thing quicker than the older version, since we spare the copying.
It is very important to learn from the beginning that software gains not only with the
lines of code that you write, but also with the lines of code that you eliminate!

1.21 Path handling

1.21.1 Normal handling

But let’s continue with our program. It looks solid, and running it with a few files in the cur-
rent directory works.
Let’s try then:
H:\lcc\examples>freq1 "..\src77\*.c" | more
CRASH!
What’s happening?
Following the program in the debugger, we see that we do not test for NULL, when opening a
file. We correct this both in checkargs and GetNext. We write a function Fopen, using the
same model as xmalloc: if it can’t open a file, it will show an error message in stderr, and exit
the program.
FILE *Fopen(char *name,char *mode)
{
FILE *result = fopen(name,mode);
if (result == NULL) {
fprintf(stderr,
"Impossible to open '%s'\n",name);
exit(1);
}
return result;
}
Ok, we change all fopen() into Fopen() , recompile, and we test again:
H:\lcc\examples>freq1 "..\src77\*.c" | more
92 C Tutorial

Impossible to open ‘Alloc.c’

Well, this looks better, but why doesn’t open Alloc.c?
Well, it seems that the path is not being passed to fopen, so that it tries to open the file in the
current directory, instead of opening it in the directory we specify in the command line.
One way to solve this, would be to change our current directory to the directory specified in
the command line, and then try to open the file. We could do this in checkargs, since it is there
where we open a file for the first time. All other files will work, if we change the current direc-
tory there.
How we could do this?
If the argument contains backslashes, it means there is a path component in it. We could copy
the string up to the last backslash, and then change our current directory to that. For instance,
if we find an argument like “..\src77\*.c”, the path component would be “..\src77\”.
Here is an updated version of checkargs:
STREAM *checkargs(int argc,char *argv[])
{
STREAM *infile = NULL;
long findfirstResult;
struct _finddata_t fd;
char *p;

if (argc < 2) { ... error handling elided ...

}
else {
findfirstResult = _findfirst(argv[1],&fd);
if (findfirstResult < 0) {
fprintf(stderr,
"File %s doesn't exist\n",argv[1]);
return NULL;
}
infile = malloc(sizeof(STREAM));
infile->Name = argv[1];
p = strrchr(argv[1],'\\');
if (p) {
*p = 0;
chdir(argv[1]);
*p = '\\';
}
infile->file = fopen(fd.name,"rb");
infile->handle = findfirstResult;
}
return infile;
}
We use the library function strrchr. That function will return a pointer to the last position
where the given character appears in the input string, or NULL, if the given character doesn’t
appear at all. Using that pointer, we replace the backslash with a NULL character. Since a zero
terminates all strings in C, this will effectively cut the string at that position.
Using that path, we call another library function, chdir that does what is name indicates:
changes the current directory to the given one. Its prototype is in <direct.h>.
After changing the current directory, we restore the argument argv[1] to its previous value,
using the same pointer “p”. Note too, that when we enter a backslash in a character constant
(enclosed in single quotes), we have to double it. This is because the backslash is used, as
within strings, for indicating characters like ‘\n’, or others.
 Path handling 93

But this isn’t a good solution. We change the current directory, instead of actually using the
path information. Changing the current directory could have serious consequences in the
working of other functions. If our program would be a part of a bigger software, this solution
would surely provoke more headaches than it solves. So, let’s use our “name” field, that up to
now isn’t being used at all. Instead of passing a name to Fopen, we will pass it a STREAM
structure, and it will be Fopen that will take care of opening the right file. We change it like
this:
FILE *Fopen(STREAM *stream,char *name,char *mode)
{
FILE *result;
char fullname[1024],*p;

p = strrchr(stream->Name,'\\');
if (p == NULL) {
fullname[0] = 0;
}
else {
*p = 0;
strcpy(fullname,stream->Name);
strcat(fullname,"\\");
*p = '\\';
}
strcat(fullname,name);
result = fopen(fullname,mode);
if (result == NULL) {
fprintf(stderr,
"Impossible to open '%s'\n",fullname);
exit(1);
}
return result;
}
We declare a array of characters, with enough characters inside to hold a maximum path, and a
few more. Then, and in the same declaration, we declare a character pointer, p. This pointer
will be set with strrchr. If there isn’t any backslash in the path, we just set the start of our
fullname[ ] to zero. If there is a path, we cut the path component as we did before, and copy
the path component into the fullname variable. The library function strcpy will copy the sec-
ond argument to the first one, including the null character for terminating correctly the string.
We add then a backslash using the strcat function that appends to its first argument the second
string. It does this by copying starting at the terminator for the string, and copying all of its
second argument, including the terminator.
We restore the string, and append to our full path the given name. In our example, we copy
into fullpath the character string “..\src77”, then we add the backslash, and then we add the
rest of the name to build a name like “..\src77\alloc.c”.
This done, we look again into our program. Yes, there are things that could be improved. For
instance, we use the 256 to write the number of elements of the array Frequencies. We could
improve the readability of we devised a macro NELEMNTS, that would make the right calcula-
tions for us.
That macro could be written as follows:
#define NELEMENTS(array) (sizeof(array)/sizeof(array[0]))
This means just that the number of elements in any array, is the size of that array, divided by
the size of each element. Since all elements have the same size, we can take any element to
make the division. Taking array[0] is the best one, since that element is always present.
94 C Tutorial

Now, we can substitute the 256 by NELEMENTS(Frequencies), and even if we change

our program to use Unicode, with 65535 different characters, and each character over two
bytes, our size will remain correct. This construct, like many others, points to one direction:
making the program more flexible and more robust to change.
We still have our 256 in the definition of the array though. We can define the size of the array
using the preprocessor like this:
#define FrequencyArraySize 256
This allows us later by changing this single line, to modify all places where the size of the
array is needed.
Lcc-win32 allows you an alternative way of defining this:
static const int FrequencyArraySize = 256;

This will work just like the pre-processor definition.64

1.21.2 Security considerations

Let’s come back to the code snippet above:
else {
*p = 0;
strcpy(fullname,stream->Name);
strcat(fullname,"\\");
*p = '\\';
}
strcat(fullname,name);
If the length of stream->Name is bigger than the size of the buffer, a security hole appears: the
program will write over the local variables of the function, and later over the return address,
stored in the stack. when our function is active.
We have a stack layout like this:
local variables of the calling function
arguments to the current function
return address
saved frame pointer
local variables: p is at the lowest address, then fullname, then result.
saved registers
stack pointer is here
An overflow when copying the Name field would destroy the saved frame pointer, and maybe
destroy the return address. The copy starts at the lowest address, the start of the fullname
buffer, then goes on to higher addresses.
After the overflow occurs, the function would continue normally until it executes the return
statement. Depending on the type of overflow, the return address would contain characters
from the Name field, that in most cases would lead to a wrong return address and a program
crash.
But that is not always the case. It could be that a user of our program would notice this style of
programming, and give us a file name that, when overflowing in the copy to the temporary

64. Memory allocation problems plague also other languages like C++ that use a similar schema than C.
 Path handling 95

buffer would form a correct return address, that would pass then control to some other routine
that the malicious user prepared for us.
This kinds of exploits can be avoided if we use other functions of the standard library:
else {
*p = 0;
strncpy(fullname,stream->Name,sizeof(fullname));
strncat(fullname,"\\",sizeof(fullname));
*p = '\\';
}
strncat(fullname,name,sizeof(fullname));
Those functions test for overflow conditions and are safer in case of unforeseen input
patterns.
There is a problem with strncpy though. It does NOT terminate the resulting string with a
zero. If we get a Name field with exactly sizeof(fullname) chars, the string will be missing
the trailing zero since strncpy doesn’t add it. One way to cover this possibility is to do:
else {
*p = 0;
fullname[sizeof(fullname)-1] = 0;
strncpy(fullname,stream->Name,sizeof(fullname)-1);
strncat(fullname,"\\",sizeof(fullname)-1);
*p = '\\';
}
strncat(fullname,name,sizeof(fullname)-1);
We finish fullname with a zero, and we copy only up to sizeof(fullname)-1 chars, leaving
our terminating zero intact.
But you should have noted that there is something wrong here. We do initialize the terminating
zero within an else statement. What happens if the execution of the function takes the other
path?
We see the following:
FILE *Fopen(STREAM *stream,char *name,char *mode)
{
FILE *result;
char fullname[1024],*p;

p = strrchr(stream->Name,'\\');
if (p == NULL) {
fullname[0] = 0;
}
else {
*p = 0;
fullname[sizeof(fullname)-1] = 0;
strncpy(fullname,stream->Name,sizeof(fullname)-1);
strncat(fullname,"\\",sizeof(fullname)-1);
*p = '\\';
}
strncat(fullname,name,sizeof(fullname)-1);
result = fopen(fullname,mode);
if (result == NULL) {
fprintf(stderr,
"Impossible to open '%s'\n",fullname);
exit(1);
}
return result;
}
96 C Tutorial

If p is NULL, we will initialize the first char of fullname to zero. Then, execution continues at
the strncat call after the else statement, and we will copy at most sizeof(fullname)-1 chars
into it, overwriting the zero and maybe leaving the fullname character array without the termi-
nating zero if the length of the passed buffer is bigger than the size of fullname. That could
lead to a crash in fopen that surely expects a well formed string.
The solution is to finish the fullname buffer in ALL cases.
FILE *Fopen(STREAM *stream,char *name,char *mode)
{
FILE *result;
char fullname[1024],*p;
p = strrchr(stream->Name,'\\');
fullname[sizeof(fullname)-1] = 0;
if (p == NULL) {
fullname[0] = 0;
}
else {
*p = 0;
strncpy(fullname,stream->Name,sizeof(fullname)-1);
strncat(fullname,"\\",sizeof(fullname)-1);
*p = '\\';
}
strncat(fullname,name,sizeof(fullname)-1);
result = fopen(fullname,mode);
if (result == NULL) {
fprintf(stderr,
"Impossible to open '%s'\n",fullname);
exit(1);
}
return result;
}
Never forget to initialize a variable in BOTH cases of an if statement. Bugs such as this are
very difficult to catch later on.
What will happen if the name of the file is bigger than our buffer? This function will fail. The
fopen call will return NULL since the file name has been truncated, and we will show an error
telling that a truncated file name doesn’t exist. Is this a good behavior?
It depends. For a technical user, a long and truncated file name could be an indicator that the
file name is just too long. Better error reporting would be appropiate if required, for instance at
the start of the function a test could produce a clear message like "Name too long".

Summary: We examined some of the functions that the C library provides for strings and
directories. Strings are ubiquitous in any serious program. We will examine this with more
depth in the next section.
 Traditional string representation in C 97

Working with directories is mandatory if you make any program, even the simplest one.
Here is an overview of the path handling functions as defined in the standard include file
<direct.h>

Function Purpose

getcwd Returns the current directory

chdir Changes the current directory

chdrive Changes the current drive

mkdir Makes a new directory

rmdir Erase a directory if its empty.

diskfree Returns the amount of space available in a disk.

1.22 Traditional string representation in C

In C character strings are represented by a sequence of bytes finished by a trailing zero byte.
For example, if you got:
char *Name = “lcc-win32”;
You will have in memory something like this:

l c c - w i n 3 2 0

108 99 99 45 119 105 110 51 50 0

We will have at each of the position of the string array a byte containing a number: the ASCII
equivalent of a letter. The array will be followed by a zero byte. Zero is not an ASCII charac-
ter, and can’t appear in character strings, so it means that the string finishes there.
This design is quite ancient, and dates to the beginning of the C language. It has several flaws,
as you can immediately see:
• There is no way to know the length of a string besides parsing the whole character array
until a zero byte is found.
• Any error where you forget to assign the last terminated byte, or this byte gets
overwritten will have catastrophic consequences.
• There is no way to enforce indexing checks.
The most frequently used function of this library are:

"strlen" that returns an integer containing the length of the string. Example:
int len = strlen("Some character string");
Note that the length of the string is the number of characters without counting the trailing
zero. The physical length of the string includes this zero byte however, and this has been
(and will be) the source of an infinite number of bugs!

"strcmp" that compares two strings. If the strings are equal it returns zero. If the first is
greater (in the lexicographical sense) than the second it returns a value greater than zero. If
98 C Tutorial

the first string is less than the second it returns some value less than zero. The order for the
strings is based in the ASCII character set.

a == b strcmp(a,b) == 0
a < b strcmp(a,b) < 0
a >= b strcmp(a,b) >= 0

"strcpy" copies one string into another. strcpy(dst,src) copies the src string into
the dst string. This means it will start copying characters from the beginning of the src
location to the dst location until it finds a zero byte in the src string. No checks are ever
done, and it is assumed that the dst string contains sufficient space to hold the src string.
If not, the whole program will be destroyed. One of the most common errors in C program-
ming is forgetting these facts.

"strcat" appends a character string to another. strcat(src,app) will add all the char-
acters of "app" at the end of the "src" string. For instance, if we have the string pointer
that has the characters "lccwin32" as above, and we call the library function str-
cat(str," compiler") we will obtain the following sequence:

l c c w i n 3 2 c o m p i l e r 0

108 99 99 119 105 110 51 50 32 99 111 109 112 105 108 101 114 0

The common operations for strings are defined in the header file <string.h>.

Function Purpose Function Purpose

Get a system error message

(strerror) or prints a user-
strcat Appends strings. strerror
supplied error message
(_strerror).

Find the first occurrence of a

strchr strlen Find the length of a string
character in a string

Find the last occurrence of a

strrchr strncat Append characters of a string.
character in a string

Copy strings up to a maximum

strcmp Compares two strings strncpy
length

Compare strings up to a Scan strings for characters in

strncmp strpbrk
maximum length specified character sets.

Compare strings up to a
strnicmp strspn Find the first substring
maximum length ignoring case

Compare strings using locale-

strcol strstr Find a substring
specific information.

strcpy Copy a string into another stristr Find a string ignoring case.

strcspn Find a substring in a string strtok Find the next token in a string
 Memory management and memory layout 99

strupr Convert string to upper case strdup Duplicate a string. Uses malloc.

strlwr Convert string to lower case strrev Reverse characters in a string

You will find the details in the online documentation.

Besides these functions in the standard C library, the operating system itself provides quite a
few other functions that relate to strings. Besides some relicts of the 16 bit past like lstrcat and
others, we find really useful functions, especially for UNICODE handling.

CharLower CharLowerBuff CharNext CharNextExA

CharPrev CharPrevExA CharToOem CharToOemBuff

CharUpper CharUpperBuff CompareString FoldString

GetStringTypeA GetStringTypeEx GetStringTypeW IsCharAlpha

IsCharAlphaNu- IsCharLower IsCharUpper LoadString

meric

lstrcat lstrcmp lstrcmpi lstrcpy

lstrcpyn lstrlen MultiByteToWideChar OemToChar

OemToCharBuff WideCharToMultiByte wsprintf wvsprintf

1.23 Memory management and memory layout

We have until now ignored the problem of memory management. We ask for more memory
from the system, but we never release it, we are permanently leaking memory. This isn’t a big
problem in these small example applications, but we would surely run into trouble in bigger
undertakings.
Memory is organized in a program in different areas:

1) The initial data area of the program. Here are stored compile time constants like the
character strings we use, the tables we input as immediate program data, the space we
allocate in fixed size arrays, and other items. This area is further divided into initialized
data, and uninitialized data, that the program loader sets to zero before the program starts.
When you write a declaration like int data = 78; the data variable will be stored in
the initialized data area. When you just write at the global level int data; the variable
will be stored in the uninitialized data area, and its value will be zero at program start.

2) The stack. Here is stored the procedure frame, i.e. the arguments and local variables of each
function. This storage is dynamic: it grows and shrinks when procedures are called and
they return. At any moment we have a stack pointer, stored in a machine register, that
contains the machine address of the topmost position of the stack.

3) The heap. Here is the space that we obtain with malloc or equivalent routines. This also a
dynamic data area, it grows when we allocate memory using malloc, and shrinks when
we release the allocated memory with the free() library function.
There is no action needed from your side to manage the initial data area or the stack. The com-
piler takes care of all that.
100 C Tutorial

The program however, manages the heap, i.e. it expects that you keep book exactly and with-
out any errors from each piece of memory you allocate using malloc. This is a very exhausting
undertaking that takes a lot of time and effort to get right. Things can be easy if you always
free the allocated memory before leaving the function where they were allocated, but this is
impossible in general, since there are functions that precisely return newly allocated memory
for other sections of the program to use.
There is no other solution than to keep book in your head of each piece of RAM. Several
errors, all of them fatal, can appear here:
• You allocate memory and forget to free it. This is a memory leak.
• You allocate memory, and you free it, but because of a complicated control flow (many
ifs, whiles and other constructs) you free a piece of memory twice. This corrupts the
whole memory allocation system, and in a few milliseconds all the memory of your
program can be a horrible mess.
• You allocate memory, you free it once, but you forget that you had assigned the memory
pointer to another pointer, or left it in a structure, etc. This is the dangling pointer
problem. A pointer that points to an invalid memory location.
Memory leaks provoke that the RAM space used by the program is always growing, eventu-
ally provoking a crash, if the program runs for enough time for this to become significant. In
short-lived programs, this can have no consequences, and even be declared as a way of mem-
ory management. The lcc compiler for instance, always allocates memory without ever both-
ering to free it, relying upon the windows system to free the memory when the program exits.
Freeing a piece of RAM twice is much more serious than a simple memory leak. It can com-
pletely confuse the malloc() system, and provoke that the next allocated piece of RAM will be
the same as another random piece of memory, a catastrophe in most cases. You write to a vari-
able and without you knowing it, you are writing to another variable at the same time, destroy-
ing all data stored there.
More easy to find, since more or less it always provokes a trap, the dangling pointer problem
can at any moment become the dreaded show stopper bug that crashes the whole program and
makes the user of your program loose all the data he/she was working with.
I would be delighted to tell you how to avoid those bugs, but after more than 10 years working
with the C language, I must confess to you that memory management bugs still plague my pro-
grams, as they plague all other C programmers.65
The basic problem is that the human mind doesn’t work like a machine, and here we are ask-
ing people (i.e. programmers) to be like machines and keep book exactly of all the many small
pieces of RAM a program uses during its lifetime without ever making a mistake.
But there is a solution that I have implemented in lcc-win32. Lcc-win32 comes with an auto-
matic memory manager (also called garbage collector in the literature) written by Hans
Boehm. This automatic memory manager will do what you should do but do not want to do:
take care of all the pieces of RAM for you.
Using the automatic memory manager you just allocate memory with GC_malloc instead of
allocating it with malloc. The signature (i.e. the result type and type of arguments) is the same
as malloc, so by just replacing all malloc by GC_malloc in your program you can benefit of
the automatic memory manager without writing any new line of code.
The memory manager works by inspecting regularly your whole heap and stack address space,
and checking if there is anywhere a reference to the memory it manages. If it doesn’t find any

65. This discussion is based upon the article of Randy Kath, published in MSDN.
 Memory management and memory layout 101

references to a piece of memory it will mark that memory as free and recycle it. It is a very
simple schema, taken to almost perfection by several years of work from the part of the
authors.
To use the memory manager you should add the gc.lib library to your link statement or indi-
cate that library in the IDE in the linker configuration tab.

1.23.1 Functions for memory allocation

malloc Returns a pointer to a newly allocated memory block

free Releases a memory block

calloc Returns a pointer to a newly allocated zero-filled memory block.

realloc Resizes a memory block preserving its contents.

Allocate a memory block in the stack that is automatically destroyed when the function
alloca
where the allocation is requested exits.

_msize Returns the size of a block

_expand Increases the size of a block without moving it.

GC_malloc Allocates a memory block managed by the memory manager.

1.23.2 Memory layout under windows66

A 32 bit address can be used to address up to 4GB of RAM. From this potential address space,
windows reserves for the system 2GB, leaving the other 2GB for each application. These
addresses, of course, are virtual, since not all PCs have 2GB of real RAM installed. To the
windows memory manager, those numbers are just placeholders that are used to find the real
memory address.
Each 32-bit address is divided in three groups, two containing 10 bits, and the third 12 bits.

The translation goes as follows:

The higher order bits (31-21) are used to index a page of memory called the page directory.
Each process contains its own page directory, filled with 1024 numbers of 32 bits each, called
page description entry or PDE for short.
The PDE is used to get the address of another special page, called page table. The second
group of bits (21-12) is used to get the offset in that page table. Once the page frame found, the

66. Note that this is a logical view of this address translation process. The actual implementation is much
more sophisticated, since Windows uses the memory manager of the CPU to speed up things. Please read the
original article to get a more in-depth view, including the mechanism of page protection, the working set, and
many other things.
102 C Tutorial

remaining 12 bits are used to address an individual byte within the page frame. Here is a figure
that visualizes the structure:

We see that a considerable amount of memory is used to… manage memory. To realize the
whole 4GB address space, we would use 4MB of RAM. But this is not as bad as it looks like,
since Windows is smart enough to fill these pages as needed. And anyway, 4MB is not even
0.1% of the total 4GB address space offered by the system.67
Each process has its own page directory. This means that processes are protected from stray
pointers in other programs. A bad pointer can’t address anything outside the process address
space. This is good news, compared to the horrible situation under windows 3.1 or even
MSDOS, where a bad pointer would not only destroy the data of the application where it
belonged, but destroyed data of other applications, making the whole system unstable. But this
means too, that applications can’t share data by sending just pointers around. A pointer is
meaningful only in the application where it was created. Special mechanisms are needed (and
provided by Windows) to allow sharing of data between applications.

1.24 Counting words

There is no introduction to the C language without an example like this:
“Exercise 24.C: Write a program that counts the words in a given file, and reports its result
sorted by word frequency.”
OK. Suppose you got one assignment like that. Suppose also, that we use the C language defi-
nition of a word, i.e. an identifier. A word is a sequence of letters that begins with an under-
score or a letter and continues with the same set of characters or digits..
In principle, the solution could look like this:

1) Open the file and repeat for each character

2) If the character starts a word, scan the word and store it. Each word is stored once. If it is in
the table already, the count is incremented for that word, otherwise it is entered in the table.

3) Sort the table of words by frequency

4) Print the report.

67. Since this is stored in a 32 bit integer, the counter will overflow somewhere in year 2038. I hope I
will be around to celebrate that event…
 Counting words 103

We start with an outline of the “main” procedure. The emphasis when developing a program is
to avoid getting distracted by the details and keep the main line of the program in your head.
We ignore all error checking for the time being.

int main(int argc,char *argv[])

{
FILE *f;
int c;

f = fopen(argv[1],"r"); // open the input file

c = fgetc(f); // Read the first character
while (c != EOF) { // Until the end of file
if (isWordStart(c)) { // Starts a word?
ScanWord(c,f); // Yes. Scan it
}
c = fgetc(f); // Go on with the next character
}
fclose(f); // Done with the file
DoReports(argv[1]); // Print results
return 0; // Return with OK.
}
This would do nicely. We have now just to fill the gaps. Let’s start with the easy ones. A word,
we said, is a sequence of _ [A-Z] [a-z] followed by _ [A-Z][a-z][0-9]. We write
a function that returns 1 if a character is the start of an identifier (word).
int isWordStart(int c)
{
if (c == '_')
return 1;
if (c >= 'a' && c <= 'z')
return 1;
if (c >= 'A' && c <= 'Z')
return 1;
return 0;
}
This function will do its job, but is not really optimal. We leave it like that for the time being.
Remember: optimizations are done later, not when designing the program.
Now, we go to the more difficult task of scanning a word into the computer. The algorithm is
simple: we just keep reading characters until we find a non-word char, that stops our loop. We
use a local array in the stack that will hold until MAXIDLENGTH chars.
#define MAXIDLENGTH 512
int ScanWord(int firstchar,FILE *f)
{
int i = 1, // index for the word buffer
c=0; // Character read
char idbuf[MAXIDLENGTH+1]; // Buffer for the word

idbuf[0] = firstchar; // We have at least one char

c = fgetc(f); // Read the next one
while (isWordStart(c) || (c >= '0' && c <= '9')) {
idbuf[i++] = c; // Store it in the array
if (i >= MAXIDLENGTH) { // Check for overflow!
fprintf(stderr,
"Identifier too long\n");
return 0; // Returning zero will break the loop
// in the calling function
104 C Tutorial

}
c = fgetc(f); // Scan the next char
}
idbuf[i] = 0; // Always zero terminate
EnterWord(idbuf); // Enter into the table
return 1; // OK to go on.
}
We hold the index into our array in the identifier “i”, for index. It starts at one since we receive
already the first character of a word. Note that we test with this index if we are going to over-
flow our local table “idbuf”. We said before that error checking should be abstracted when
designing the program but as any rule, that one has exceptions.
If we were going to leave a lot of obvious errors in many functions around, we would need a
lot of work later on to fix all those errors. Fundamental error checking like a buffer overrun
should always be in our minds from the beginning, so we do it immediately. Note that this test
is a very simple one.

1.24.1 The organization of the table

Now, we have no choice but to start thinking about that “EnterWord” function. All the easy
work is done, we have to figure out now, an efficient organization for our word table. We have
the following requirements:

1) It should provide a fast access to a word to see if a given sequence of characters is there
already.

2) It should not use a lot of memory and be simple to use.

The best choice is the hash table. We use a hash table to hold all the words, and before entering
something into our hash table we look if is in there already. Conceptually, we use the follow-
ing structure:

Our word table is a sequence of lists of words. Each list is longer or shorter, depending on the
hash function that we use and how good our hash function randomizes the input. If we use a
table of 65535 positions (slots) and a good hash algorithm we divide the access time by 65535,
not bad.
To enter something into our table we hash the word into an integer, and we index the slot in the
table. We then compare the word with each one of the words in the list of words at that slot. If
we found it, we do nothing else than increment the count of the word. If we do not find it, we
add the word at the start of that slot.
 Counting words 105

Note that this requires that we define a structure to hold each word and its associated count.
Since all the words are in a linked list, we could use the following structure, borrowing from
the linked list representation discussed above:
typedef struct _WordList {
int Count;
struct _WordList *Next;
char Word[];
} WORDLIST;
We have an integer that holds the number of times this word appears in the text, a pointer to
the next word in the list, and an unspecified number of characters just following that pointer.
This is a variable sized structure, since each word can hold more or less characters. Note that
variable sized structures must have only one “flexible” member and it must be at the end of the
definition.
Our “EnterWord” function can look like this:
void EnterWord(char *word)
{
int h = hash(word); // Get the hash code for this word
WORDLIST *wl = WordTable[h]; // Index the list at that slot
while (wl) { // Go through the list
if (!strcmp(wl->Word,word)) {
wl->Count++; // Word is already in the table.
return; // increment the count and return
}
wl = wl->Next; // Go to the next item in the list
}
// Here we have a new word, since it wasn’t in the table.
// Add it to the table now
wl = NewWordList(word);
wl->Next = WordTable[h];
WordTable[h] = wl;
}
What would be a good hash function for this application?
This is a tutorial, so we keep things simple. Here is a very simple hash function:
int hash(char *word)
{
int h = 0;
while (*word) {
h += *word;
word++;
}
return h & 0xffff;
}
We just add up our characters. If we get a hash value of more than 65535 (the size of our
table), we just take the lower 16 bits of the hash value. Easy isn’t it?
We declare our word table now, like this:
WORDLIST *WordTable[0xffff+1];

1.24.2 Memory organization

Now we write the constructor for our word list structure. It should get more memory from the
system to hold the new structure, and initialize its fields.
WORDLIST *NewWordList(char *word)
{
106 C Tutorial

int len = strlen(word);

WORDLIST *result = more_memory(sizeof(WORDLIST)+len+1);
result->Count = 1;
strcpy(result->Word,word);
return result;
}
We allocate more memory to hold the structure, the characters in the word, and the terminating
zero. Then we copy the characters from the buffer we got, set the count to 1 since we have
seen this word at least once, and return the result. Note that we do not test for failure. We rely
on “more_memory” to stop the program if there isn’t any more memory left, since the pro-
gram can’t go on if we have exhausted the machine resources.
Under windows, the implementation of the standard “malloc” function is very slow. To avoid
calling “malloc” too often, we devise an intermediate structure that will hold chunks of mem-
ory, calling malloc only when each chunk is exhausted.
typedef struct memory {
int used;
int size;
char *memory;
} MEMORY;
Now, we write our memory allocator:
#define MEM_ALLOC_SIZE 0xffff
int memoryused = 0;
void *more_memory(int siz)
{
static MEMORY *mem;
void *result;
if (mem == NULL || mem->used+siz >= mem->size) {
mem = malloc(sizeof(mem)+MEM_ALLOC_SIZE);
if (mem == NULL) {
fprintf(stderr,"No more memory at line %d\n",line);
exit(1);
}
mem->used = 0;
memoryused += MEM_ALLOC_SIZE;
mem->size = MEM_ALLOC_SIZE;
}
result = mem->memory+mem->used;
mem->used += siz;
memset(result,siz,0);
memoryused += siz;
return result;
}

We use a static pointer to a MEMORY structure to hold the location of the current memory
chunk being used. Since it is static it will be initialized to NULL automatically by the compiler
and will keep its value from one call to the next. We test before using it, if the chunk has
enough room for the given memory size we want to allocate or if it is NULL, i.e. this is the
very first word we are entering. If either of those if true, we allocate a new chunk and initialize
its fields.68

68. Note that we allocate MEM_ALLOC_SIZE bytes. If we want to change to more or less bytes, we
just change the #define line and we are done with the change.
 Counting words 107

Otherwise we have some room in our current chunk. We increase our counters and return a
pointer to the position within the “memory” field where this chunk starts. We clean the mem-
ory with zeroes before returning it to the calling function.
Note that we do not keep any trace of the memory we have allocated so it will be impossible to
free it after we use it. This is not so bad because the operating system will free the memory
after this program exists. The downside of this implementation is that we can’t use this pro-
gram within another one that would call our word counting routine. We have a memory leak
“built-in” into our software.
A way out of this is very easy though. We could just convert our mem structures into a linked
list, and free the memory at the end of the program.

1.24.3 Displaying the results

After all this work, we have a program that will compile when run, but is missing the essential
part: showing the results to the user. Let’s fix this.
We need to sort the words by frequency, and display the results. We build a table of pointers to
word-list structures and sort it.
But... to know how big our table should be, we need to know how many words we have
entered. This can be done in two ways: Either count the number of words in the table when
building the report, or count the words as we enter them.
Obviously, the second solution is simpler and requires much less effort. We just declare a glo-
bal integer variable that will hold the number of words entered into the table so far:
int words = 0;

We increment this counter when we enter a new word, i.e. in the function NewWordList. 69
We will need a comparison function for the qsort library function too.
int comparewords(const void *w1,const void *w2)
{
WORDLIST *pw1 = *(WORDLIST **)w1,*pw2 = *(WORDLIST **)w2;

if (pw1->Count == pw2->Count)
return strcmp(pw1->Word,pw2->Word);
return pw1->Count - pw2->Count;
}
Note that we have implemented secondary sort key. If the counts are the same, we sort by
alphabetical order within a same count.
void DoReports(char *filename)
{
int i;
int idx = 0; // Index into the resulting table

// Print file name and number of words

printf("%s: %d different words.\n",filename,words);

69. Global variables like this should be used with care. Overuse of global variables leads to problems
when the application grows, for instance in multi-threaded applications. When you got a lot of global
variables accessed from many points of the program it becomes impossible to use threads becasue the
danger that two threads access the same global variable at a time.
Another problem is that our global is not static, but visible through the whole program. If somewhere else
somebody writes a function called “words” we are doomed. In this case and for this example the glo-
bal variable solution is easier, but not as a general solution.
108 C Tutorial

// allocate the word-list pointer table

WORDLIST **tab = more_memory(words*sizeof(WORDLIST *));

// Go through the entire hash table

for (i=0; i< sizeof(WordTable)/sizeof(WordTable[0]);i++) {
WORDLIST *wl = WordTable[i];

while (wl) {
// look at the list at this slot
tab[idx] = wl;
wl = wl->Next;
idx++;
if (idx >= words && wl) {
fprintf(stderr,"program error\n");
exit(1);
}
}
}
// Sort the table
qsort(tab,words,sizeof(WORDLIST *),comparewords);
// Print the results
for (i=0; i< words;i++) {
printf("%s %5d\n",tab[i]->Word,tab[i]->Count);
}
}

We start by printing the name of the file and the number of different words found. Then, we go
through our hash table, adding a pointer to the word list structure at each non-empty slot.
Note that we test for overflow of the allocated table. Since we increment the counter each time
that we add a word, it would be very surprising that the count didn’t macth with the number of
items in the table. But it is better to verify this.
After filling our table for the qsort call, we call it, and then we just print the results.

1.24.4 Code review

Now that we have a bare skeleton of our program up and running, let’s come back to it with a
critical eye.
For instance look at our “isWordStart” function. We have:
int isWordStart(int c)
{
if (c == '_')
return 1;
if (c >= 'a' && c <= 'z')
return 1;
if (c >= 'A' && c <= 'Z')
return 1;
return 0;
}
A look in the “ctype.h” system header file tells us that for classifying characters we have a lot
of efficient functions. We can reduce all this code to:
int isWordStart(int c)
{
return c == '_' || isalpha(c);
}
 A closer look at the pre-processor 109

The “isalpha” function will return 1 if the character is one of the uppercase or lowercase
alphabetic characters. Always use library functions insted of writing your own. The “isalpha”
function does not make any jumps like we do, but indexes a table of property bits. Much faster.
And what about error checking? Remember, we just open the file given in the command line
without any test of validty. We have to fix this.
Another useful feature would be to be able to report a line number associated with our file,
instead of just an error message that leaves to the user the huge task of finding where is the
offending part of the input file that makes our program crash. This is not very complex. We
just count the new line characters.
The output of our program is far from perfect. It would be better if we justify the columns. To
do that, we have to just count the length of each word and keep a counter to the longest word
we find. Another nice thing to have would be a count of how many words with 1 character we
find, how many with two, etc.
In the Appendix you will find the complete source code containing the answers to this prob-
lem.

1.25 A closer look at the pre-processor

The first phase of the compilation process is the “pre-processing” phase. This consists of scan-
ning in the program text all the preprocessor directives, i.e. lines that begin with a “#” charac-
ter, and executing the instructions found in there before presenting the program text to the
compiler.
We will interest us with just two of those instructions. The first one is the “#define” directive,
that instructs the software to replace a macro by its equivalent. We have two types of macros:
Parameter less. For example:
#define PI 3.1415
Following this instruction, the preprocessor will replace all instances of the identifier PI with
the text “3.11415”.
Macros with arguments. For instance:
#define s2(a,b) ( (a*a + b*b) /2.0)
When the preprocessor finds a sequence like:
s2(x,y)
It will replace it with:
(x*x + y*y)/2.0 )
The problem with that macro is that when the preprocessor finds a statement like:
s2(x+6.0,y-4.8);
It will produce :
(x+6.0*x+6.0 + y+6.0*y+6.0) /2.0 )
What will calculate completely another value:
(7.0*x + 7.0*y + 12.0)/2.0
To avoid this kind of bad surprises, it is better to enclose each argument within parentheses
each time it is used:
#define s2(a,b) (((a)*(a) + (b)*(b))/2.0)
110 C Tutorial

This corrects the above problem but we see immediately that the legibility of the macros suf-
fers… quite complicate to grasp with all those redundant parentheses around.
Another problem arises when you want that the macro resembles exactly a function call and
you have to include a block of statements within the body of the macro, for instance to declare
a temporary variable.
#define s2(x,y) { int temp = x*x+y*y; x=temp+y *(temp+6);}
If you call it like this:
if (x < y) s2(x,y);
else
x = 0;
This will be exapnded to:
if (x < y) { int temp = x*x+y*y; x=temp+y *(temp+6);} ;
else
x = 0;
This will provoke a syntax error.
To avoid this problem, you can use the do... while statement, that consumes the semicolon:
#define s2(x,y) do { int temp = x*x+y*y; x=temp+y *(temp+6);} \
while(0)
Note the \ that continues this long line, and the absence of a semicolon at the end of the macro.
An “#undef” statement can undo the definition of a symbol. For instance
#undef PI
will erase from the pre-processor tables the PI definition above. After that statement the iden-
tifier PI will be ignored by the preprocessor and passed through to the compiler.
The second form of pre-processor instructions that is important to know is the
#if (expression)
… program text …
#else
… program text …
#endif
or the pair
#ifdef (symbol)
#else
#endif
When the preprocessor encounters this kind of directives, it evaluates the expression or looks
up in its tables to see if the symbol is defined. If it is, the “if” part evaluates to true, and the
text until the #else or the #endif is copied to the output being prepared to the compiler. If it is
NOT true, then the preprocessor ignores all text until it finds the #else or the #endif. This
allows you to disable big portions of your program just with a simple expression like:
#if 0
…
#endif
This is useful for allowing/disabling portions of your program according to compile time
parameters. For instance, lcc-win32 defines the macro __LCC__. If you want to code some-
thing only for this compiler, you write:
#ifdef __LCC__
… statements …
#endif
Note that there is no way to decide if the expression:
 Time and Date functions 111

SomeFn(foo);
Is a function call to SomeFn, or is a macro call to SomeFn. The only way to know is to read
the source code. This is widely used. For instance, when you decide to add a parameter to Cre-
ateWindow function, without breaking the millions of lines that call that API with an already
fixed number of parameters you do:
#define CreateWindow(a,b, … ) CreateWindowEx(0,a,b,…)
This means that all calls to CreateWindow API are replaced with a call to another routine that
receives a zero as the new argument’s value.
It is quite instructive to see what the preprocessor produces. You can obtain the output of the
preprocessor by invoking lcc with the –E option. This will create a file with the extension .i
(intermediate file) in the compilation directory. That file contains the output of the preproces-
sor. For instance, if you compile hello.c you will obtain hello.i.

1.26 Time and Date functions

The C library offers a lot of functions for working with dates and time. The first of them is the
time function that returns the number of seconds that have passed since January first 1970, at
midnight. 70
Several structures are defined that hold time information. The most important from them are
the “tm” structure and the “timeb” structure.
struct tm {
int tm_sec;
int tm_min;
int tm_hour;
int tm_mday;
int tm_mon;
int tm_year;
int tm_wday;
int tm_yday;
int tm_isdst;
};
The fields are self-explanatory. The structure “timeb” is defined in the directory include\sys,
as follows:
struct timeb {
time_t time;
unsigned short pad0;
unsigned long lpad0;
unsigned short millitm; // Fraction of a second in ms
unsigned short pad1;
unsigned long lpad1;
// Difference (minutes), moving westward, between UTC and local time
short timezone;
unsigned short pad2;
unsigned long lpad2;
// Nonzero if daylight savings time is currently in effect for the local time zone.
short dstflag;
};
We show here a small program that displays the different time settings.
#include <time.h>

70. This clock will overflow in something like 2.000 years so be prepared for windows 4.000!
112 C Tutorial

#include <stdio.h>
#include <sys/types.h>
#include <sys/timeb.h>
#include <string.h>

void main()
{
char tmpbuf[128], ampm[] = "AM";
time_t ltime;
struct _timeb tstruct;
struct tm *today, *gmt, xmas = { 0, 0, 12, 25, 11, 93 };

/* Display operating system-style date and time. */

strtime( tmpbuf );
printf( "OS time:\t\t\t\t%s\n", tmpbuf );
strdate( tmpbuf );
printf( "OS date:\t\t\t\t%s\n", tmpbuf );

/* Get UNIX-style time and display as number and string. */

time( &ltime );
printf( "Time in seconds since UTC 1/1/70:\t%ld\n", ltime );
printf( "UNIX time and date:\t\t\t%s", ctime( &ltime ) );

/* Display UTC. See note (1) in text */

gmt = gmtime( &ltime );
printf( "Coordinated universal time:\t\t%s", asctime( gmt ) );

/* Convert to time structure and adjust for PM if necessary. */

today = localtime( &ltime );
if( today->tm_hour > 12 ) {
strcpy( ampm, "PM" );
today->tm_hour -= 12;
}
if( today->tm_hour == 0 ) /* Adjust if midnight hour. */
today->tm_hour = 12;

/* See note (2) in text */

printf( "12-hour time:\t\t\t\t%.8s %s\n",
asctime( today ) + 11, ampm );

/* Print additional time information. */

ftime( &tstruct );
printf( "Plus milliseconds:\t\t\t%u\n", tstruct.millitm );
printf( "Zone difference in seconds from UTC:\t%u\n",
tstruct.timezone );
printf( "Time zone name:\t\t\t\t%s\n", _tzname[0] );
printf( "Daylight savings:\t\t\t%s\n", // See note (3) in text
tstruct.dstflag ? "YES" : "NO" );

/* Make time for noon on Christmas, 1993. */

if( mktime( &xmas ) != (time_t)-1 )
printf( "Christmas\t\t\t\t%s\n", asctime( &xmas ) );

/* Use time structure to build a customized time string. */

today = localtime( &ltime );

/* Use strftime to build a customized time string. */

strftime( tmpbuf, 128,
"Today is %A, day %d of the month of %B in the year %Y.\n",
today );
 Time and Date functions 113

printf( tmpbuf );
}
We use this opportunity for introducing new C constructs.
We see the function call gmtime(&ltime);. What does this mean? The function gmtime
requires a pointer to a variable of type time_t. We do not have a pointer, so we make one “on
the fly” by using the “address-of” operator.
The printf statement uses pointer addition to skip the first 11 characters of the result of asc-
time. That function returns a pointer to a character string. To skip characters we just add up a
constant (11) to that pointer, effectively skipping those characters. Since we want to display
the 8 following characters only, we pass a width argument to the %s directive of printf. As you
know, “%s” is a directive that instructs printf to display a character string. Since we give it a
maximum width of 8, only the first 8 chars will be displayed.
We see here the construct (expr) ? val1 : val2; This construct evaluates first the expression, in
this case “tstruct.dstflag”. If the value is different than zero, the return value of the expression
will be the first value, in this case the character string “YES”. If the expression evaluates to
zero, the second value will be chosen, in this case the character string “NO”. The result of this
is passed to printf as an argument.
The Windows system too has a lot of time-related functions. Here is a handy list of the most
important. Note that file times are kept using 64 bits in modern versions of windows, i.e. the
numbers represent the number of 100 nanosecond intervals since January first, 1601. 71

Function Purpose

CompareFileTime Compares two 64-bit file times

DosDateTimeToFileTime Converts MS-DOS date and time values to a 64-bit file time.

FileTimeToDosDateTime Converts a 64-bit file time to MS-DOS date and time values.

Converts a file time based on the Coordinated Universal Time

FileTimeToLocalFileTime
(UTC) to a local file time.

FileTimeToSystemTime Converts a 64-bit file time to system time format

Retrieves the date and time that a file was created, last accessed, and
GetFileTime
last modified.

GetLocalTime Retrieves the current local date and time.

GetSystemTime Retrieves the current system date and time.

Determines whether the system is applying periodic time

GetSystemTimeAdjustment adjustments to its time-of-day clock at each clock interrupt, along
with the value and period of any such adjustments.

Obtains the current system date and time. The information is in

GetSystemTimeAsFileTime
Coordinated Universal Time (UTC) format.

71. I hope this happens to you too…

114 C Tutorial

Retrieves the number of milliseconds that have elapsed since the

GetTickCount
system was started. It is limited to the resolution of the system timer.

Retrieves the current time-zone parameters. These parameters

GetTimeZoneInformation control the translations between Coordinated Universal Time (UTC)
and local time.

Converts a local file time to a file time based on the Coordinated

LocalFileTimeToFileTime
Universal Time (UTC).

Sets the date and time that a file was created, last accessed, or last
SetFileTime
modified.

SetLocalTime Sets the current local time and date.

Sets the current system time and date. The system time is expressed
SetSystemTime
in Coordinated Universal Time (UTC).

Tells the system to enable or disable periodic time adjustments to its

SetSystemTimeAdjustment
time of day clock.

Sets the current time-zone parameters. These parameters control

SetTimeZoneInformation
translations from Coordinated Universal Time (UTC) to local time.

SystemTimeToFileTime Converts a system time to a file time.

Converts a Coordinated Universal Time (UTC) to a specified time

SystemTimeToTzSpecificLocalTime
zone's corresponding local time.

1.27 Using structures (continued)

C allows implementation of any type of structure. Here is a description of some simple ones so
you get an idea of how they can be built and used.

1.27.1 Lists
Lists are members of a more general type of objects called sequences, i.e. objects that have a
natural order. You can go from a given list member to the next element, or to the previous one.
We have several types of lists, the simplest being the single-linked list, where each member
contains a pointer to the next element, or NULL, if there isn’t any. We can implement this
structure in C like this:
typedef struct _list {
struct _list *Next; // Pointer to next element
void *Data;// Pointer to the data element
} LIST;
We can use a fixed anchor as the head of the list, for instance a global variable containing a
pointer to the list start.
LIST *Root;
We define the following function to add an element to the list:
LIST *Append(LIST **pListRoot, void *data)
{
LIST *rvp = *pListRoot;
 Using structures (continued) 115

if (rvp == NULL) { // is the list empty?

// Yes. Allocate memory
*pListRoot = rvp = GC_malloc(sizeof(LIST));
}
else { // find the last element
while (rvp->Next)
rvp = rvp->Next;
// Add an element at the end of the list
rvp->Next = GC_malloc(sizeof(LIST));
rvp = rvp->Next;
}
// initialize the new element
rvp->Next = NULL;
rvp->Data = data;
return rvp;
}
This function receives a pointer to a pointer to the start of the list.
Why?
If the list is empty, it needs to modify the pointer to the start of the list. We would normally
call this function with:
newElement = Append(&Root,data);
Note that loop:
while (rvp->Next)
rvp = rvp->Next;
This means that as long as the Next pointer is not NULL, we position our roving pointer
(hence the name “rvp”) to the next element and repeat the test. We suppose obviously that the
last element of the list contains a NULL “Next” pointer. We ensure that this condition is met
by initializing the rvp->Next field to NULL when we initialize the new element.
To access a list of n elements, we need in average to access n/2 elements.
Other functions are surely necessary. Let’s see how a function that returns the nth member of a
list would look like:
LIST *ListNth(LIST *list, int n)
{
while (list && n-- > 0)
list = list->Next;
return list;
}
Note that this function finds the nth element beginning with the given element, which may or
may not be equal to the root of the list. If there isn’t any nth element, this function returns
NULL.
If this function is given a negative n, it will return the same element that was passed to it.
Given a NULL list pointer it will return NULL.
Other functions are necessary. Let’s look at Insert.
LIST *Insert(LIST *list,LIST *element)
{
LIST *tmp;

if (list == NULL)
return NULL;
if (list == element)
return list;
116 C Tutorial

tmp = list->Next;
list->Next = element;
if (element) {
element->Next = tmp;
}
return list;
}
We test for different error conditions. The first and most obvious is that “list” is NULL. We
just return NULL. If we are asked to insert the same element to itself, i.e. “list” and “element”
are the same object, their addresses are identical, we refuse. This is an error in most cases, but
maybe you would need a circular element list of one element. In that case just eliminate this
test.
Note that Insert(list, NULL); will effectively cut the list at the given element, since
all elements after the given one would be inaccessible.
Many other functions are possible and surely necessary. They are not very difficult to write,
the data structure is quite simple.
Double linked lists have two pointers, hence their name: a Next pointer, and a Previous
pointer, that points to the preceding list element.
Our data structure would look like this:
typedef struct _dlList {
struct _dlList *Next;
struct _dlList *Previous;
void *data;
} DLLIST;
Our “Append” function above would look like: (new material in bold)
LIST *AppendDl(DLLIST **pListRoot, void *data)
{
DLLIST *rvp = *pListRoot;

// is the list empty?

if (rvp == NULL) {
// Yes. Allocate memory
*pListRoot = rvp = GC_malloc(sizeof(DLLIST));
rvp->Previous = NULL;
}
else {
// find the last element
while (rvp->Next)
rvp = rvp->Next;
// Add an element at the end of the list
rvp->Next = GC_malloc(sizeof(DLLIST));
rvp->Next->Previous = rvp;
rvp = rvp->Next;
}
// initialize the new element
rvp->Next = NULL;
rvp->Data = data;
return rvp;
}
The Insert function would need some changes too:
LIST *Insert(LIST *list,LIST *element)
{
LIST *tmp;
 Using structures (continued) 117

if (list == NULL)
return NULL;
if (list == element)
return list;
tmp = list->Next;
list->Next = element;
if (element) {
element->Next = tmp;
element->Previous = list;
if (tmp)
tmp->Previous = element;
}
return list;
}
Note that we can implement a Previous function with single linked lists too. Given a pointer to
the start of the list and an element of it, we can write a Previous function like this:
LIST *Previous(LIST *root, LIST *element)
{
if (root == NULL )
return NULL;
while (root && root->Next != element)
root = root->Next;
return root;
}
Circular lists are useful too. We keep a pointer to a special member of the list to avoid infinite
loops. In general we stop when we arrive at the head of the list. Wedit uses this data structure
to implement a circular double linked list of text lines. In an editor, reaching the previous line
by starting at the first line and searching and searching would be too slow. Wedit needs a dou-
ble linked list, and a circular list makes an operation like wrapping around easier when search-
ing.

1.27.2 Hash tables

A hash table is a table of lists. Each element in a hash table is the head of a list of element that
happen to have the same hash code, or key.
To add an element into a hash table we construct from the data stored in the element a number
that is specific to the data. For instance we can construct a number from character strings by
just adding the characters in the string.
This number is truncated module the number of elements in the table, and used to index the
hash table. We find at that slot the head of a list of strings (or other data) that maps to the same
key modulus the size of the table.
To make things more specific, let’s say we want a hash table of 128 elements, which will store
list of strings that have the same key.
Suppose then, we have the string “abc”. We add the ASCII value of ‘a’ + ‘b’ + ‘c’ and we
obtain 97+98+99 = 294. Since we have only 128 positions in our table, we divide by 128, giv-
ing 2 and a rest of 38. We use the rest, and use the 38th position in our table.
This position should contain a list of character strings that all map to the 38th position. For
instance, the character string “aE”: (97+69 = 166, mod 128 gives 38). Since we keep at each
position a single linked list of strings, we have to search that list to find if the string that is
being added or looked for exists.
A sketch of an implementation of hash tables looks like this:
118 C Tutorial

#define HASHELEMENTS 128

typedef struct hashTable {
int (*hashfn)(char *string);
LIST *Table[HASHELEMENTS];
} HASH_TABLE;
We use a pointer to the hash function so that we can change the hash function easily. We build
a hash table with a function.
HASH_TABLE newHashTable(int (*hashfn)(char *))
{
HASH_TABLE *result = GC_malloc(sizeof(HASH_TABLE));
result->hashfn = hashfn;
return result;
}
To add an element we write:
LIST *HashTableInsert(HASH_TABLE *table, char *str)
{
int h = (table->hashfn)(str);
LIST *slotp = table->Table[h % HASHELEMENTS];

while (slotp) {
if (!strcmp(str,(char *)slotp->data)) {
return slotp;
}
slotp = slotp->Next;
}
return Append(&table->Table[h % HASHELEMENTS],element);
}
All those casts are necessary because we use our generic list implementation with a void
pointer. If we would modify our list definition to use a char * instead, they wouldn’t be neces-
sary.
We first call the hash function that returns an integer. We use that integer to index the table in
our hash table structure, getting the head of a list of strings that have the same hash code. We
go through the list, to ensure that there isn’t already a string with the same contents. If we find
the string we return it. If we do not find it, we append to that list our string
The great advantage of hash tables over lists is that if our hash function is a good one, i.e. one
that returns a smooth spread for the string values, we will in average need only n/128 compar-
isons, n being the number of elements in the table. This is an improvement over two orders of
magnitude over normal lists.

1.28 The preprocessor in detail

The C preprocessor is conceptually a program that reads the text of a C program, performs
some transformations with the text, and passes the resulting text to the compiler proper. In
some C compilers this is actually how it is done. The preprocessor is a separate program,
called very often “cpp”, that reads the C program text and leaves an intermediate file for the C
compiler. This is very slow, and lcc-win32 includes the preprocessing stage but without creat-
ing any intermediate files.

1.28.1 Preprocessor commands

The preprocessor receives its commands with lines beginning with the special character “#”.
This lines can contain:
 The preprocessor in detail 119

1) Macros

2) Conditional compilation instructions

3) Pragma instructions

4) The “##” operator

5) Line instructions

1.28.1.1 Preprocessor macros

The #define command has two forms depending on whether a left parenthesis appears imme-
diately after the name of the macro. The first form, without parenthesis is simply the substitu-
tion of text. An example can be:
#define MAXBUFFERSIZE 8192
This means that whenever the preprocessor find the identifier MAXBUFFERSIZE in the pro-
gram text, it will replace it with the character string “8192”

The second form of the preprocessor macros is the following:

#define add(a,b) ((a)+(b))
This means that when the preprocessor finds the sequence:
int n = add(7,b);
it will replace this with:
int n = ((7)+(b));
Note that the left parenthesis MUST be immediately after the name of the macro without any
white space (blanks, tabs) in between. It is often said that white space doesn’t change the
meaning of a C program but that is not always true, as you can see now. If there is white space
between the macro name and the left parenthesis the preprocessor will consider that this is a
macro with no arguments whose body starts with a left parentheses!
If you want to delete the macro from the preprocessor table you use the
#undef <macro name>
command. This is useful for avoiding name clashes. Remember that when you define a macro,
the name of the macro will take precedence before everything else since the compiler will not
even see the identifier that holds the macro. This can be the origin of strange bugs like:

int fn(int a)
{
// some code
}
If you have the idea of defining a macro like this
#define fn 7987
the definition above will be transformed in

int 7987(int a)
{
}
120 C Tutorial

not exactly what you would expect. This can be avoided by #undefining the macros that you
fear could clash with other identifiers in the program.

1.28.1.2 Conditional compilation

Very often we need to write some kind of code in a special situation, and some other kind in
another situation. For instance we would like to call the function “initUnix()” when we are in a
UNIX environment, and do NOT call that function when in other environments. Besides we
would like to erase all UNIX related instructions when compiling for a windows platform and
all windows related stuff when compiling for Unix.
This is achieved with the preprocessor
#ifdef UNIX
lines for the Unix system
#else
lines for other (non-unix) systems.
#endif
This means:
If the preprocessor symbol “UNIX” is defined, include the first set of lines, else include the
second set of lines.
There are more sophisticated usages with the #elif directive:
#ifdef UNIX
Unix stuff
#elif MACINTOSH
Mac stuff
#elif WIN32
Windows stuff
#else
#error “Unknown system!”
#endif
Note the #error directive. This directive just prints an error message and compilation fails.
The lines that are in the inactive parts of the code will be completely ignored, except (of
course) preprocessor directives that tell the system when to stop. Note too that the flow of the
program becomes much difficult to follow. This feature of the preprocessor can be abused, to
the point that is very difficult to see which code is being actually compiled and which is not.
The IDE of lcc-win32 provides an option for preprocessing a file and showing all inactive
lines in grey color. Go to “Utils” the choose “Show #ifdefs” in the main menu.
Note: You can easily comment out a series of lines of program text when you enclose them in
pairs of
#if 0
// lines to be commented out
#endif

1.28.1.3 The pragma directive

This directive is compiler specific, and means that what follows the #pragma is dependent on
which compiler is running. The pragmas that lcc-wi32 uses are defined in the documentation.
In general pragmas are concerned with implementation specific details, and are an advanced
topic.
 The preprocessor in detail 121

1.28.1.4 The ## operator

This operator allows you to concatenate two tokens:
#define join(a,b) (a##b)
a = join(anne,bob)

When preprocessed this will produce:

a = (annebob)
This is useful for constructing variable names automatically and other more or less obscure
hacks.

1.28.2 Things to watch when using the preprocessor

1) One of the most common errors is to add a semi colon after the macro:
#define add(a,b) a+b;
When expanded, this macro will add a semicolon into the text, with the consequence of a
cascade of syntax errors with apparently no reason.

2) Watch for side effects within macros. A macro invocation is similar to a function call, with
the big difference that the arguments of the function call is evaluated once but in the macro
can be evaluated several times. For instance we have the macro “square”
#define square(a) (a*a)
If we use it like this:
b = square(a++);
After expansion this will be converted into:
b = (a++)*(a++);
and the variable a will be incremented twice.
122 C Tutorial

1.29 Pitfalls of the C language

Look. I am not a religious person. C is not a religion for me, and this means that I see some of
the pitfalls of the language. I write this so that you see them at the start, and they do not bite
you.

1.29.1 Defining a variable in a header file

If you write:
static int foo = 7;
in a header file, each C source file that includes that header will have a different copy of “foo”,
each initialized to 7, but each in a different place. These variables will be totally unrelated,
even if the intention of the programmer is to have a single variable “foo”.
If you omit the static, at least you will get an error at link time, and you will see the bug.

Golden rule: Never define something in a header file. Header files are for declarations only!

1.29.2 Confusing = and ==

If you write
if (a = 6) {
}
you are assigning to “a” the number 6, instead of testing for equality. The “if” branch will be
always taken because the result of that assignment is different from zero.

1.29.3 Forgetting to close a comment

If you write:
a=b; /* this is a bug
c=d; /* c=d will never happen */
The comment in the first line is not terminated. It goes one through the second line and is fin-
ished with the end of the second line. Hence, the assignment of the second line will never be
executed. Wedit helps you avoid this by coloring commentaries in another color as normal
program text.

1.29.4 Easily changed block scope.

Suppose you write the following code:
if (someCondition)
fn1();
else
OtherFn();
This is OK, but if you add some code to debug, for instance, you end up with:
if (someCondition)
fn1();
else
printf(“Calling OtherFn\n”);
OtherFn();
The else is not enclosed in curly braces, so only one statement will be taken. The end result is
that the call to OtherFn is always executed, no matter what.
 Pitfalls of the C language 123

Golden rule: ALWAYS watch out for scopes of “if” or “else” not between curly braces when
adding code.

1.29.5 Using the ++ or -- more than once in an expression.

The ANSI C standard72 specifies that an expression can change the value of a variable only
once within an expression. This means that a statement like:
i++ = ++i;
is invalid, as are invalid this, for instance:
i = i+++++i;
(in clear i++ + ++i)

1.29.6 Unexpected Operator Precedence

The code fragment,
if( chr = getc() != EOF ) {
printf( "The value of chr is %d\n", chr );
}
will always print 1, as long as end-of-file is not detected in getc. The intention was to assign
the value from getc to chr, then to test the value against EOF. The problem occurs in the first
line, which says to call the library function getc. The return value from getc (an integer value
representing a character, or EOF if end-of-file is detected), is compared against EOF, and if
they are not equal (it’s not end-of-file), then 1 is assigned to the object chr. Otherwise, they are
equal and 0 is assigned to chr. The value of chr is, therefore, always 0 or 1.
The correct way to write this code fragment is,
if( (chr = getc()) != EOF ) {
printf( "The value of chr is %d\n", chr );
}
The extra parentheses force the assignment to occur first, and then the comparison for equality
is done.73

72. Doing assignment inside the controlling expression of loop or selection statements is not a good program-
ming practice. These expressions tend to be difficult to read, and problems such as using = instead of == are more
difficult to detect when, in some cases, = is desired.

73. How does it work?

He tests first if the lower 16 bits contain a 1 bit. The number 65535 consists of eight 1s, in binary notation,
since 65535 = 216 – 1.
If the test fails, this means that the least significant bit can’t be in the lower 16 bits. He increases the
counter by 16, and shifts right the number to skip the 8 bits already known to contain zero. If the test
succeeds, this means that there is at least a bit set in the lower 16 bits. Nothing is done, and the program
continues to test the lower 16 bits.
He uses the same reasoning with the 16 bits in x, that now contain either the high or the lower word of x.
255 is 28 – 1. This is applied then recursively. At each step we test 16, then 8, then 4, then 2 and at the end the
last bit.
124 C Tutorial

1.29.7 Extra Semi-colon in Macros

The next code fragment illustrates a common error when using the preprocessor to define con-
stants:
#define MAXVAL 10; // note the semicolon at the end
/* ... */
if( value >= MAXVAL ) break;
The compiler will report an error. The problem is easily spotted when the macro substitution is
performed on the above line. Using the definition for MAXVAL, the substituted version reads,
if( value >= 10; ) break;
The semi-colon (;) in the definition was not treated as an end-of-statement indicator as
expected, but was included in the definition of the macro MAXVAL. The substitution then
results in a semi-colon being placed in the middle of the controlling expression, which yields
the syntax error. Remember: the pre-processor does only a textual substitution of macros.
 Introduction 125

Chapter

Windows Programming
2.1 Introduction
OK, up to now we have built a small program that receives all its input from a file. This is
more or less easy, but a normal program will need some input from the user, input that can’t be
passed through command line arguments, or files. At this point, many introductory texts start
explaining scanf, and other standard functions to get input from a command line interface.
This can be OK, but I think a normal program under windows uses the features of windows.
We will start with the simplest application that uses windows, a dialog box with a single edit
field, that will input a character string, and show it in a message box at exit.
The easiest way to do this is to ask wedit to do it for you. You choose ‘new project’ in the
project menu, give a name and a sources directory, and when the software asks you if it should
generate the application skeleton for you, you answer yes.
You choose a dialog box application, when the main dialog box of the “wizard” appears, since
that is the simplest application that the wizard generates, and will fit our purposes quite well.
But let’s go step by step. First we create a project. The first thing you see is a dialog box, not
very different from the one we are going to build, that asks for a name for the new project. You
enter a name like this:

You press OK, and then we get a more complicated one, that asks quite a lot of questions.
126 C Tutorial

You enter some directory in the second entry field, make sure the “windows executable” at the

bottom is selected, and press ok. Then we get:

You press the “yes” button. This way, we get into the wizard.
 Introduction 127

The first panel of the wizard is quite impressing, with many buttons, etc. Ignore all but the
type of application panel. There, select a “dialog based” application, like this:

You see, the “Dialog based’ check button at the upper left is checked. Then press the OK but-
ton.
Then we get to different dialogs to configure the compiler. You leave everything with the
default values, by pressing Next at each step. At the end, we obtain our desired program. For
windows standards, this is a very small program: 86 lines only, including the commentaries.
We will study this program in an in-depth manner. But note how short this program actually is.
Many people say that windows programs are impossible huge programs, full of fat. This is just
not true!
But first, we press F9 to compile it. Almost immediately, we will obtain:
128 C Tutorial

Dialog.exe built successfully. Well, this is good news!74 Let’s try it. You execute the program
you just built using Ctrl+F5. When we do this, we see our generated program in action:

Just a dialog box, with the famous OK/Cancel buttons, and nothing more. But this is a start.
We close the dialog, either by pressing the “x” button at the top right corner, or just by using
OK or Cancel, they both do the same thing now, since the dialog box is empty.
We come back to the IDE, and we start reading the generated program in more detail. It has
three functions:
• WinMain
• InitializeApp
• DialogFunc
If we ignore the empty function “InitializeApp”, that is just a hook to allow you to setup things
before the dialog box is shown, only two functions need to be understood. Not a very difficult
undertaking, I hope.

2.2 WinMain
Command line programs, those that run in the ill named “msdos window”, use the “main”
function as the entry point. Windows programs, use the WinMain entry point.75
The arguments WinMain receives are a sample of what is waiting for you. They are a mess of
historical accidents that make little sense now. Let’s look at the gory details:
int APIENTRY WinMain(HINSTANCE hinst,
HINSTANCE hinstPrev,
LPSTR lpCmdLine,
int nCmdShow);

This is a function that returns an int, uses the stdcall calling convention76 denoted by APIEN-
TRY, and that receives (from the system) 4 parameters.
hinst, a “HANDLE” to an instance of the program. This will always be 0x400000 in hexadec-
imal, and is never used. But many window functions need it, so better store it away.

74. This has only historical reasons, from the good old days of windows 2.0 or even earlier. You can use
“main” as the entry point, and your program will run as you expect, but traditionally, the entry point is called
WinMain, and we will stick to that for now.
75. A calling convention refers to the way the caller and the called function agrees as to who is going to
adjust the stack after the call. Parameters are passed to functions by pushing them into the system stack.
Normally it is the caller that adjusts the stack after the call returns. With the stdcall calling convention, it is
the called function that does this. It is slightly more efficient, and contributes to keeping the code size small.
76. API means Application Programmer Interface, i.e. entry points into the windows system for use by
the programmers, like you and me.
 WinMain 129

hinstPrev. Again, this is a mysterious “HANDLE” to a previous instance of the program.

Again, an unused parameter, that will always contain zero, maintained there for compatibil-
ity reasons with older software.

lpCmdLine. This one is important. It is actually a pointer to a character string that contains the
command line arguments passed to the program. Note that to the contrary of “main”, there
isn’t an array of character pointers, but just a single character string containing all the com-
mand line.

nCmdShow. This one contains an integer that tells you if the program was called with the
instruction that should remain hidden, or should appear normally, or other instructions that
you should use when creating your main window. We will ignore it for now.
OK OK, now that we know what those strange parameters are used (or not used) for, we can
see what this function does.
int APIENTRY WinMain(HINSTANCE hinst,
HINSTANCE hinstPrev,
LPSTR lpCmdLine,
int nCmdShow)
{
WNDCLASS wc; // A structure of type WNDCLASS

memset(&wc,0,sizeof(wc)); // We set it to zero

wc.lpfnWndProc = DefDlgProc; // Procedure to call for handling
messages
wc.cbWndExtra = DLGWINDOWEXTRA;
wc.hInstance = hinst;
wc.hCursor = LoadCursor(NULL, IDC_ARROW);
wc.hbrBackground = (HBRUSH) (COLOR_WINDOW + 1);
wc.lpszClassName = "dialog";
RegisterClass(&wc);

return DialogBox(hinst,
MAKEINTRESOURCE(IDD_MAINDIALOG),
NULL,
(DLGPROC) DialogFunc);
}
We see that the main job of this function is filling the structure wc, a WNDCLASS structure
with data, and then calling the API77 DialogBox. What is it doing?
We need to register a class in the window system. The windows system is object oriented,
since it is derived from the original model of the window and desktop system developed at
Xerox, a system based in SmallTalk, an object oriented language. Note that all windows sys-
tems now in use, maybe with the exception of the X-Window system, are derived from that
original model. The Macintosh copied it from Xerox, and some people say that Microsoft cop-
ied it from the Macintosh. In any case, the concept of a class is central to windows.
A class of windows is a set of window objects that share a common procedure. When some
messages or events that concern this window are detected by the system, a message is sent to
the window procedure of the concerned window. For instance, when you move the mouse over
the surface of a window, the system sends a message called WM_MOUSEMOVE to the win-
dows procedure, informing it of the event.

77. PLEASE never do this if you use the garbage collector!

130 C Tutorial

There are quite a lot of messages, and it would be horrible to be forced to reply to all of them
in all the windows you create. Fortunately, you do not have to. You just treat the messages that
interest you, and pass all the others to the default windows procedure.
There are several types of default procedures, for MDI windows we have MDIDefWindow-
Proc, for normal windows we have DefWindowProc, and for dialog boxes, our case here, we
have the DefDlgProc procedure.
When creating a class of windows, it is our job to tell windows which procedure should call
when something for this window comes up, so we use the class registration structure to inform
it that we want that all messages be passed to the default dialog procedure and we do not want
to bother to treat any of them. We do this with:
wc.lpfnWndProc = DefDlgProc;
As we saw with the qsort example, functions are first class objects in C, and can be passed
around easily. We pass the address of the function to call to windows just by setting this field
of our structure.
This is the most important thing, conceptually, that we do here. Of course there is some other
stuff. Some people like to store data in their windows78. We tell windows that it should reserve
some space, in this case the DLGWINDOWEXTRA constant, that in win.h is #defined as 30. We
put in this structure too, for obscure historical reasons, the hinst handle that we received in
WinMain. We tell the system that the cursor that this window uses is the system cursor, i.e. an
arrow. We do this with the API LoadCursor that returns a handle for the cursor we want. The
brush that will be used to paint this window will be white, and the class name is the character
string “dialog”.
And finally, we just call the RegisterClass API with a pointer to our structure. Windows does
its thing and returns.
The last statement of WinMain, is worth some explanation. Now we have a registered class,
and we call the DialogBox API, with the following parameters:
DialogBox(hinst,
MAKEINTRESOURCE(IDD_MAINDIALOG),
NULL, (DLGPROC) DialogFunc);
The hinst parameter, that many APIs still want, is the one we received from the system as a
parameter to WinMain. Then, we use the MAKEINTRESOURCE macro, to trick the compiler
into making a special pointer from a small integer, IDD_MAINDIALOG that in the header file
generated by the wizard is defined as 100. That header file is called dialogres.h, and is quite
small. We will come to it later.
What is this MAKEINTRESOURCE macro?
Again, history, history. In the prototype of the DialogBox API, the second parameter is actu-
ally a char pointer. In the days of Windows 2.0 however, in the cramped space of MSDOS
with its 640K memory limit, passing a real character string was out of the question, and it was
decided (to save space) that instead of passing the name of the dialog box resource as a real
name, it should be passed as a small integer, in a pointer. The pointer should be a 32 bit pointer
with its upper 16 bits set to zero, and its lower 16 bits indicating a small constant that would be
searched in the resource data area as the “name” of the dialog box template to load.
Because we need to load a template, i.e. a series of instructions to a built-in interpreter that
will create all the necessary small windows that make our dialog box. As you have seen, dia-
log boxes can be quite complicated, full of edit windows to enter data, buttons, trees, what

78. When the IDE asks you if you want to open it as a resource say NO. We want to look at the text of
that file this time.
 Resources 131

have you. It would be incredible tedious to write all the dozens of calls to the CreateWin-
dow API, passing it all the coords of the windows to create, the styles, etc.
To spare you this Herculean task, the designers of the windows system decided that a small
language should be developed, together with a compiler that takes statements in that language
and produce a binary file called resource file.
This resource files are bound to the executable, and loaded by the system from there automat-
ically when using the DialogBox primitive. Among other things then, that procedure needs to
know which dialog template should load to interpret it, and it is this parameter that we pass
with the MAKEINTRESOURCE macro.
Ok, that handles (at least I hope) the second parameter of the DialogBox API. Let’s go on,
because there are still two parameters to go!
The third one is NULL. Actually, it should be the parent window of this dialog box. Normally,
dialog boxes are written within an application, and they have here the window handle of their
parent window. But we are building a stand-alone dialog box, so we left this parameter empty,
i.e. we pass NULL.
The last parameter, is the DialogFunc function that is defined several lines below. The
DefDlgProc needs a procedure to call when something important happens in the dialog box: a
button has been pushed, an edit field receives input, etc.
Ok, this closes the call of the DialogBox API, and we are done with WinMain. It will return
the result of the DialogBox function. We will see later how to set that result within our Dialog
box procedure.

2.3 Resources
We mentioned before, that there is a compiler for a small resource language that describes our
dialog boxes. Let’s look at that with a little bit more detail before we go to our dialog proce-
dure.
Open that file that should be called dialog.rc if you gave the project the “dialog” name79, and
look at this lines:
IDD_MAINDIALOG DIALOG 7, 20, 195, 86 (1)
STYLE DS_MODALFRAME|WS_POPUP|WS_VISIBLE|WS_CAPTION|WS_SYSMENU
(2)
CAPTION "dialog" (3)
FONT 8, "Helv" (4)
BEGIN
DEFPUSHBUTTON "OK", IDOK, 149, 6, 40, 14 (5)
PUSHBUTTON "Cancel", IDCANCEL, 149, 23, 40, 14 (6)
END
We see that all those statements concern the dialog box, its appearance, the position of its child
windows, etc. Let’s go statement by statement:

1) We find here the same identifier IDD_MAINDIALOG, and then the DIALOG statement,
together with some coordinates. Those coordinates are expressed in Dialog Units, not in
pixels. The motivation behind this, is to make dialog boxes that will look similar at all
resolutions and with different screen sizes. The units are based somehow in the size of the

79. You will be prompted for a header file, where are stored the definitions for things like
IDD_MAINDIALOG. Choose the one generated by the wizard. Its name is <project name>res.h, i.e. for a
project named “test” we would have “testres.h”.
132 C Tutorial

system font, and there are APIs to get from those units into pixels, and from pixels into
those units.

2) The STYLE statement tells the interpreter which things should be done when creating the
window. We will see later when we create a real window and not a dialog box window, that
there can be quite a lot of them. In this case the style indicates the appearance
(DS_MODALFRAME), that this window is visible, has a caption, and a system menu.

3) The CAPTION statement indicates just what character string will be shown in the caption.

4) In a similar way, the FONT statement tells the system to use Helv

5) The following statements enumerate the controls of the dialog box, and their descriptions
are enclosed in a BEGIN/END block. We have two of them, a push button that is the
default push button, and a normal pushbutton

6) the Cancel button. Both of them have a certain text associated with them, a set of coords as
all controls, and an ID, that in the case of the OK button is the predefined symbol IDOK,
with the numerical value of 1, and in the case of the Cancel button IDCANCEL (numerical
value 2).
To convert this set of instruction in this language into a binary resource file that windows can
interpret, we use a compiler called a resource compiler. Microsoft’s one is called rc, Borland’s
one is called "brc", and lcc-win32’s one is called lrc. All of them take this resource language
with some minor extensions depending on the compiler, and produce a binary resource file for
the run time interpreter of windows.
The resource compiler of lcc-win32 is explained in detail in the technical documentation, and
we will not repeat that stuff again here. For our purposes it is enough to know that it is compat-
ible with the other ones.
The binary resource files generated by the resource compiler are passed to the linker that con-
verts them into resource specifications to be included in the executable.
Note that actually you do not need to know this language, because the IDE has a resource edi-
tor that can be used to build graphically using drag and drop the dialog box. But the emphasis
here is to introduce you to the system so that you know not only what button should you push,
but why you should push that button too.
But we wanted originally to make a dialog box containing an edit field. We are far away from
our objective yet.
 Resources 133

We come back to Wedit, after closing our text file “dialog.rc”, and we go to the “Design”
menu bar and press “Open/new”. 80The resource editor opens up, and we see the following
display:

Near the top left border you see a button like this:

If you click in that button, the controls palette opens:

80. Why should introductory texts be clear?

Why not make obfuscated sentences?
There is even an obfuscated C contest. Who writes the most incomprehensible program? Quite a
challenge! With the documentation you can do even better: you write docs that are so incomprehensible that
nobody reads them!
134 C Tutorial

Controls palette The whole operation of the editor is quite simple: The smaller
window represents all the controls that you can put in a dialog
box: entry fields, buttons, checkboxes, and several others. You
select one button with the mouse, and drag it to your dialog box.
There you drop it at the right position. To add an entry field then,
we just push the edit field icon, the third one from the left in the
upper row of icons, and drag it to our dialog box in the main
wedit window. After doing this, it will look like this:

Our entry field becomes the selected item, hence the red handles
around it. After resizing if necessary, we need to enter its identi-
fier, i.e. the symbolic name that we will use in our program to
refer to it.

We will refer then in our program to this entry field with the name
IDENTRYFIELD, maybe not a very cute name, but at least better
than some bare number. The editor will write a
#define IDENTRYFIELD 101
in the generated header file.

The number 101 is an arbitrary constant, chosen by the editor. We resize the dialog box a bit (I
like dialogs that aren’t bigger than what they should be), and we press the “test” button.
We see a display like this:

We can enter text in the entry field, and pushing Cancel or OK will finish the test mode and
return us to the dialog box editor.
 The dialog box procedure 135

OK, seems to be working. We save, and close the dialog box editor. We come back to our dia-
log procedure, where we will use this new entry field to get some text from the user.

2.4 The dialog box procedure

static BOOL CALLBACK DialogFunc(HWND hwndDlg, UINT msg, WPARAM wParam,
LPARAM lParam)
{
switch (msg) {
case WM_INITDIALOG:
InitializeApp(hwndDlg,wParam,lParam);
return TRUE;
case WM_COMMAND:
switch (LOWORD(wParam)) {
case IDOK:
EndDialog(hwndDlg,1);
return 1;
case IDCANCEL:
EndDialog(hwndDlg,0);
return 1;
}
break;
case WM_CLOSE:
EndDialog(hwndDlg,0);
return TRUE;
}
return FALSE;
}
A dialog box procedure is called by the system. It has then, a fixed argument interface, and
should return a predefined value. It receives from the system the handle of the dialog box win-
dow, the message, and two extra parameters.
Normally these procedures are a big switch statement that handles the messages the program
is interested in. The return value should be TRUE if the dialog box procedure has handled the
message passed to it, FALSE otherwise.
The general form of a switch statement is very simple: you have the switch expression that
should evaluate to an integer and then you have several “cases” that should evaluate to com-
pile time constants.
All those names in the switch statement above are just integers that have been given a sym-
bolic name in windows.h using the preprocessor #define directive. A “break” keyword sepa-
rates one “case” from the next one.
Note that in C a case statement can finish without a break keyword.
In that case (!) execution continues with the code of the next case. In any case, of course, a
return statement finishes execution of that case, since control is immediately passed to the
calling function.81
In this procedure we are interested in only three messages, hence we have only three “cases”
in our switch:

1) WM_INITDIALOG. This message is sent after the window of the dialog box has been
created, but before the dialog is visible in the screen. Here is done the initialization of the

81. You put the cursor under the WM_INITDIALOG identifier and press F1.
136 C Tutorial

dialog box data structures, or other things. The wizard inserts here a call to a procedure for
handling this message.

2) WM_COMMAND. This message is sent when one of the controls (or child windows if you
want to be exact) has something to notify to the dialog: a button has been pressed, a check
box has been pressed, data has been entered in an entry field, etc. Since we can have several
controls, we use again a switch statement to differentiate between them. Switch statements
can be nested of course.

3) WM_CLOSE. This message arrives when the user has pressed the “close” button in the
system menu, or has typed the Alt+F4 keyboard shortcut to close the dialog.
Now, the whole purpose of this exercise is to input a character string. The text is entered by the
user in our entry field. It is important, from a user’s perspective, that when the dialog box is
displayed, the cursor is at the beginning of the entry field. It could be annoying to click each
time in the entry field to start editing the text. We take care of this by forcing the focus to the
entry field.
Under windows, there is always a single window that has the focus, i.e. receives all the input
from the keyboard and the mouse. We can force a window to have the focus using the SetFo-
cus API.
static int InitializeApp(HWND hDlg,WPARAM wParam, LPARAM lParam)
{
SetFocus(GetDlgItem(hDlg,IDENTRYFIELD));
return 1;
}
We add this call in the procedure InitializeApp. We test, and… it doesn’t work. We still have
to click in the edit field to start using it. Why?
Because, when we read the documentation of the WM_INITDIALOG message82 it says:
WM_INITDIALOG
hwndFocus = (HWND) wParam; // handle of control to receive focus
lInitParam = lParam; // initialization parameter
Parameters
hwndFocus
Value of wParam. Identifies the control to receive the default keyboard focus. Windows
assigns the default keyboard focus only if the dialog box procedure returns TRUE.
Well, that is it! We have to return FALSE, and our SetFocus API will set the focus to the con-
trol we want. We change that, and… it works! Another bug is gone.83
Note that the SetFocus API wants a window handle. To get to the window handle of a control
in the dialog box we use its ID that we took care of defining in the dialog editor. Basically we
give to the SetFocus function the result of calling the API GetDlgItem. This is nice, since we
actually need only one window handle, the window handle of the dialog box that windows
gives to us, to get all other window handles of interest.
Now comes a more difficult problem. When the user presses the OK button, we want to get the
text that he/she entered. How do we do that?

82. This example shows you how to get rid of those problems, and the kind of problems you will
encounter when programming under Windows. The only solution in most cases is a detailed reading of the
documentation. Fortunately, Windows comes with a clear documentation that solves most problems.
83. All the notifications messages from edit fields begin with the EN_ prefix, meaning Edit field
Notification.
 The dialog box procedure 137

We have two problems in one: the first is to decide when we want to get the text, and the other
is how to get that text.
For the first one the answer is clear. We want to read the text only when the user presses the
OK button. If the Cancel button is pressed, or the window is closed, we surely aren’t interested
in the text, if any. We will read the text then when we handle the message that the OK button
window sends to us when is pressed. We change our dialog procedure like this:
case WM_COMMAND:
switch (LOWORD(wParam)) {
case IDOK:
ReadText(hwndDlg);
EndDialog(hwndDlg,1);
return 1;
case IDCANCEL:
EndDialog(hwndDlg,0);
return 1;
}
break;
We add a call to a function that will get the text into a buffer. That function looks like this:
static char buffer[1024];
int ReadText(HWND hwnd)
{
memset(buffer,0,sizeof(buffer));
if (GetDlgItemText(hwnd,
IDENTRYFIELD,
buffer,
sizeof(buffer))) {
return 1;
}
return 0;
}
We define a buffer that will not be visible from other modules, hence static. We set a fixed
buffer with a reasonable amount of storage.
Our function cleans the buffer before using it, and then calls one of the workhorses of the dia-
log procedures: the API GetDlgItemText. This versatile procedure will put in the designated
buffer, the text in a control window, in this case the text in the entry field. We again indicate to
the API which control we are interested in by using its numerical ID. Note that GetDlgItem-
Text returns the number of characters read from the control. If there isn’t anything (the user
pressed OK without entering any text), GetDlgItemText returns zero.
The first time that we do that; we will surely will want to verify that the text we are getting is
the one we entered. To do this, we use the API MessageBox that puts up a message in the
screen without forcing us to register a window class, define yet another window procedure,
etc.
We add then to our window procedure, the following lines:
case IDOK:
if (ReadText(hwndDlg)) {
MessageBox(hwndDlg,buffer,
"text entered",MB_OK);
EndDialog(hwndDlg,1);
}
return 1;
MessageBox takes a parent window handle, in this case the handle of the dialog box proce-
dure, a buffer of text, a title of the message box window, and several predefined integer con-
138 C Tutorial

stants, that indicate which buttons it should show. We want just one button called OK, so we
pass that constant.
Note too, that if the user entered no text, we do NOT call the EndDialog API, so the dialog
box will refuse to close, even if we press the OK button. We force the user to enter some text
before closing the dialog. Since we haven’t changed anything in the logic for the Cancel but-
ton, the dialog box will still close when the user presses those buttons. Only the behavior of
the OK button will change.
The EndDialog API takes two parameters: the dialog box window handle that it should
destroy, and a second integer parameter. The dialog box will return these values as the result of
the DialogBox call from WinMain remember?
Since WinMain returns itself this value as its result, the value returned by the DialogBox will
be the return value of the program.

2.5 A more advanced dialog box procedure

Doing nothing and not closing the dialog box when the user presses OK is not a good inter-
face. You expect a dialog box to go away when you press OK don’t you?
A user interface like this makes for an unexpected behavior. Besides, if we put ourselves in the
user’s shoes, how can he/she find out what is wrong? The software doesn’t explain anything,
doesn’t tell the user what to do to correct the situation, it just silently ignores the input. This is
the worst behavior we could imagine.
Well, there are two solutions for this. We can disable the OK button so that this problem
doesn’t appear at all, or we could put up a message using our MessageBox API informing the
user that a text must be entered.
Let’s see how we would implement the first solution.
To be really clear, the OK button should start disabled, but become active immediately after
the user has typed some text. If, during editing, the user erases all text that has been entered,
the OK button should revert to its inactive state.
We can do this by processing the messages that our edit field sends to us. Edit fields are very
sophisticated controls, actually a full-blown mini-editor in a small window. Each time the user
types anything in it, the edit field procedure sends us a WM_COMMAND message, informing
us of each event.
We change our dialog procedure as follows:
case WM_COMMAND:
switch (LOWORD(wParam)) {
case IDOK:
// suppressed, stays the same
case IDCANCEL:
EndDialog(hwndDlg,0);
return 1;
case IDENTRYFIELD:
switch (HIWORD(wParam)) {
case EN_CHANGE:
if (GetDlgItemText( hwndDlg, IDENTRYFIELD,
buffer,sizeof(buffer))) {
EnableWindow(
GetDlgItem(hwndDlg,IDOK),1);
}
else
EnableWindow(
 A more advanced dialog box procedure 139

GetDlgItem(hwndDlg,IDOK),0);
break;
}
break;
}
break;
We add a new case for this message. But we see immediately that this nested switch state-
ments are getting out of hand. We have to split this into a function that will handle this mes-
sage. We change again our dialog box procedure as follows:
case IDCANCEL:
EndDialog(hwndDlg,0);
return 1;
case IDENTRYFIELD:
return
EntryFieldMessages(hwndDlg,wParam);
This is much clearer. We put the code for handling the entry field messages in its own proce-
dure, “EntryFieldMessages”. Its code is:
int EntryFieldMessages(HWND hDlg,WPARAM wParam)
{
HWND hIdOk = GetDlgItem(hDlg,IDOK);

switch (HIWORD(wParam)) {
case EN_CHANGE:
if (GetDlgItemText(hDlg,IDENTRYFIELD,
buffer, sizeof(buffer))) {
// There is some text in the entry field. Enable the IDOK button.
EnableWindow(hIdOk,1);
}
else // no text, disable the IDOK button
EnableWindow(hIdOk,0);
break;
}
return 1;
}
Let’s look at this more in detail. Our switch statement uses the HIWORD of the first message
parameter. This message carries different information in the upper 16 bits (the HIWORD) than
in the lower 16 bits (LOWORD). In the lower part of wParam we find the ID of the control
that sent the message, in this case IDENTRYFIELD, and in the higher 16 bits we find which
sub-message of WM_COMMAND the control is sending to us, in this case EN_CHANGE84,
i.e. a change in the text of the edit field.
There are many other notifications this small window is sending to us. When the user leaves
the edit field and sets the focus with the mouse somewhere else we are notified, etc. But all of
those notifications follow the same pattern: they are sub-messages of WM_COMMAND, and
their code is sent in the upper 16 bits of the wParam message parameter.
Continuing the analysis of EntryFieldMessages, we just use our friend GetDlgItemText to get
the length of the text in the edit field. If there is some text, we enable the IDOK button with the
API EnableWindow. If there is NO text we disable the IDOK button with the same API. Since
we are getting those notifications each time the user types a character, the reaction of our
IDOK button will be immediate.

84. I remember calling the software support when installing some hardware: many of the options of the
installation software were disabled but there was no way of knowing why.
140 C Tutorial

But we have still one more thing to do, before we get this working. We have to modify our Ini-
tializeApp procedure to start the dialog box with IDOK disabled, since at the start there is no
text in the entry field.
static int InitializeApp(HWND hDlg,WPARAM wParam, LPARAM lParam)
{
SetFocus(GetDlgItem(hDlg,IDENTRYFIELD));
// Disable the IDOK button at the start.
EnableWindow(GetDlgItem(hDlg,IDOK),0);
return 1;
}
We recompile, and it works. The OK button starts disabled (grayed), and when we type the
first character it becomes active, just as we wanted. When we select all the text in the entry
field, and then erase it, we observe that the button reverts to the inactive state.

2.6 User interface considerations

There was another way of informing the user that text must be entered: a MessageBox call,
telling him/her precisely what is wrong. This alternative, making something explicit with a
message, or implicit, like the solution we implemented above appears very often in windows
programming, and it is very difficult to give a general solution to it. It depends a lot of course,
upon the application and its general style. But personally, I prefer explicit error messages
rather than implicit ones. When you receive an error message, you know exactly what is
wrong and you can take easily steps to correct it. When you see a menu item disabled, it is
surely NOT evident what the hell is happening and why the software is disabling those
options.85
But there are other user-interface considerations in our dialog box to take into account too.
One of them is more or less evident when you see how small the letters in the edit field are.
Dialog boxes use a default font that shows very thin and small characters. It would be much
better if we would change that font to a bigger one.
In the initialization procedure, we set the font of the edit field to a predefined font. Windows
comes with several predefined items, ready for you to use without much work. One of them is
the System font that comes in two flavors: monospaced, and proportional. We use the mono-
spaced one. Our initialization procedure then, looks now like this:
static int InitializeApp(HWND hDlg,WPARAM wParam, LPARAM lParam)
{
HFONT font;

font = GetStockObject(ANSI_FIXED_FONT);
SendDlgItemMessage(hDlg,IDENTRYFIELD,
WM_SETFONT,(WPARAM)font,0);
EnableWindow(GetDlgItem(hDlg,IDOK),0);
SetFocus(GetDlgItem(hDlg,IDENTRYFIELD));
return 1;
}
A HFONT is a font “handle”, i.e. an integer that represents a font for windows. We get that
integer using the GetStockObject API. This function receives an integer code indicating which
object we are interested in and returns it. There are several types of object we can get from it:
fonts, brushes, pens, etc.86

85. Now is a good time to read the documentation for that API. It will not be repeated here.
 User interface considerations 141

Yet another point missing in our dialog box is a correct title, or prompt. The title of our dialog
is now just “dialog”. This tells the user nothing at all. A friendlier interface would tell the user
what data the software is expecting from him/her. We could change the title of the dialog to a
more meaningful string.
The program calling our dialog procedure could give this string as a parameter to the dialog
box. Dialog boxes can receive parameters, as any other procedure. They receive them in the
parameters passed to the WM_INITDIALOG message.
A closer look to the documentation of the WM_INITDIALOG message tell us that the lParam
message parameter contains for the WM_INITDIALOG 32 bits of data passed in the last
parameter of an API called DialogBoxParam.
We have to modify our calling sequence to the dialog, and instead of using DialogBox we use
the DialogBoxParam API. Looking into our program, we see that the DialogBox API was
called in our WinMain function (see above). We should modify this call then, but a new prob-
lem appears: where does WinMain know which string to pass to the DialogBoxParam API?
Well, we could decide that this string would be the parameters passed to WinMain in the lpC-
mdLine argument. This is the most flexible way. We modify then the call to DialogBox like
follows:
return DialogBoxParam (hinst,
MAKEINTRESOURCE(IDD_MAINDIALOG),
NULL, (DLGPROC) DialogFunc,
(int)lpCmdLine);
Since our dialog box is now constructed with DialogBoxParam, we receive in the lParam mes-
sage parameter the same pointer that we gave to the DialogBoxParam API. Now, we have just
to set that text in the caption of our dialog box and it’s done. We do that (again) in our initial-
ization procedure by adding:
SetWindowText(hDlg, (char *)lParam);
The SetWindowText API sets the caption text of a window, if that window has a caption bar of
course. To test this, we have to tell Wedit to pass a command line argument to the program
when it calls the debugger or executes the program with Ctrl+F5. We do this by selecting the
“debugger” tab in the configuration of wedit:

86. We introduced that to see if we were really getting a string. Since now we are returning that data to
the calling program, that message should disappear.
142 C Tutorial

The debugger tab is in the upper left corner. When we select it, we arrive at the following tab:

Note the first line “Command line arguments to pass to program”. There, we write the string
that we want shown in the dialog box.
When now we press Ctrl+F5, we see our dialog box like this:

Nice, we can pass our dialog box a “prompt” string. This makes our dialog box more useful as
a general input routine. Remember that the objective of this series of sections was to introduce
you a general routine to input a character string from the user. We are getting nearer.
Still, there is one more consideration that we haven’t solved yet. We have a buffer of a limited
length, i.e. 1024 characters. We would like to limit the text that the user can enter in the dialog
box so that we avoid overflowing our buffer. We can do this with the message
EM_SETLIMITTEXT. We have to send this message to the control when we start the dialog
box, so that the limit will be effective before the user has an occasion of overflowing it. We
add then
SendDlgItemMessage(hDlg,IDENTRYFIELD,
EM_SETLIMITTEXT,512,0);
 Libraries 143

2.7 Libraries
What we would like is a way of using this dialog box in our applications of course. How could
we do that?
One way would be to call it as an independent program. We could use the facilities for calling
a program within the windows system, and pass our prompt in the command line parameters.
This would work, but the problem of getting the string from the user would be quite compli-
cated to solve. Programs can only return an error code, and in some situations this error code
can only be from zero to 255… We can’t pass pointers just like that from one program to
another; we can’t just pass a pointer to a character string as the result of the program.
Why?
Because windows, as other systems like linux, Solaris, and UNIX in general, uses a protected
virtual address schema. The machine addresses that the program uses are virtual, as if the pro-
gram was the only one running in the machine. It is the operating system and the CPU that
does the translation of those virtual addresses into real RAM locations in the machine you are
using. This means the addresses of each program aren’t meaningful for another program. We
can pass special pointers (shared memory) within windows, but that’s too advanced stuff for
an introductory text, sorry.
But there are many other ways of solving this problem without costly interface development.
What do we want? Let’s get that clear first, worrying about implementation details later.
int GetString(char *prompt, char *buffer,int bufferlen);
This routine would return either true or false, depending if the user has pressed OK or can-
celled the operation. If the user pressed OK, we would find the string that was entered in the
buffer that we pass to GetString. To avoid any overflow problems, we would pass the length of
the character string buffer, so that the GetString routine stops input when we reach that limit.
The C language supports code reuse. You can compile code that is useful in many situations
and build libraries of routines that can be reused over and over again. The advantages are
many:
• The code has to be written once and debugged once.
• The size of our program stays small, since we call a routine instead of repeating the code
all over the place.
Function calls are the mechanism of code reuse since a function can be used in many situa-
tions. Libraries are just a collection of routines that are linked either directly or indirectly with
the main program.
From the standpoint of the user of the library, not the one who is building it, the usage is quite
simple:

1) You have to include the header file of the library to make the definition available to the
compiler.

2) You use the functions or data

3) You add the library to the set of files that the linker uses to build your program.
This simplicity of usage makes libraries a good way to reuse and share the code between dif-
ferent applications.
Under windows we have two types of libraries:
144 C Tutorial

Static libraries. These libraries are built in files that normally have the .lib extension and are
linked with the program directly, i.e. they are passed to the linker as arguments. The linker
takes the code of the needed functions from the library and copies the code into the pro-
gram.

Dynamic libraries. These aren’t copied into the main program, but are resolved at load time by
the program loader. When you double-click a program’s icon you activate a system pro-
gram of windows called program loader that goes to the disk, finds the code for the execut-
able you have associated with the icon, and loads it from disk into RAM. When doing this,
the loader finds if the program needs any dynamic libraries, that normally have the .DLL
extension, and reads their code too, linking it dynamically to the program being loaded.
Which one should we use in this application?
Static or not static? That is the question!
Our program needs a class registration before it can call the DialogBoxParam API. If we use
the static library approach, we would have to require that the user of the library calls some ini-
tialization routine before, to allow us to register our class with windows.
But this would complicate the interface. We introduce with this requirement yet another thing
that can go wrong with the program, yet another thing to remember.
A way out of this dilemma would be to take care of doing the registration automatically. We
could setup an integer variable that would start as zero. Before calling our DialogBoxParam
procedure we would test the value of this variable.
If it is zero it means that our class wasn’t registered. We would register our class and set this
variable to one, so that the next call finds a value different than zero and skips the class regis-
tration code.
 Libraries 145

We have to tell the IDE to produce a static library now, instead of a normal executable file. We
do this by going into the linker configuration tab, and checking the library radio-button, like
this:

You see the “Static library” radio-button checked. The name of the library is the name of the
project with the .lib extension.
Now we are ready to change our WinMain. We change our WinMain function to be the Get-
String procedure, like this:
static char buffer[1024];

static int classRegistered;

int APIENTRY GetString(char *prompt, char *destbuffer,int bufferlen)

{
WNDCLASS wc;
int result;
HANDLE hinst;

hinst = GetModuleHandle(NULL);
if (classRegistered == 0) {
memset(&wc,0,sizeof(wc));
wc.lpfnWndProc = DefDlgProc;
wc.cbWndExtra = DLGWINDOWEXTRA;
wc.hInstance = hinst;
wc.hCursor = LoadCursor(NULL, IDC_ARROW);
wc.hbrBackground = (HBRUSH) (COLOR_WINDOW + 1);
wc.lpszClassName = "dialog";
RegisterClass(&wc);
classRegistered = 1;
}
146 C Tutorial

result = DialogBoxParam(hinst,
MAKEINTRESOURCE(IDD_MAINDIALOG),
NULL,
(DLGPROC) DialogFunc,
(int)prompt);

if (result == 1) {
strncpy(destbuffer,buffer,bufferlen-1);
destbuffer[bufferlen-1] = 0;
}
return result;
}
We have several things to explain here.

1) We move the declaration of our static buffer that was before further down, to the beginning
of the file, so that we can use this buffer in the GetString procedure to copy its contents into
the destination buffer.

2) We declare our flag for testing if the class has been registered as a static int, i.e. an integer
visible only in this module. We do not need to initialize it to zero, since the C language
guarantees that all non-explicitly initialized static variables will be set to zero when the
program starts.

3) We modify the declarations of local variables in the GetString procedure, adding a result
integer variable, and a HANDLE that will hold the instance of the current module. Before,
we received this as a parameter in the arguments of WinMain, but now we have to get it by
some other means. The solution is to call the GetModuleHandle API, to get this. We
indicate it that we want the handle of the currently running executable by passing it a
NULL parameter.

4) We test then our global flag classRegistered. If it is zero, we haven’t registered the class,
and we do it now. Afterwards, we set the variable to one, so that this piece of code will not
be executed again.

5) We call our DialogBox procedure just like we did before, but now we assign its result to an
integer variable and we test if the result is one (i.e. the user pressed OK). If that is the case,
we copy the string from the temporary buffer to the destination buffer. Note that we use the
strncpy function. This standard library function takes an extra parameter, a maximum
length to copy. We do not want to overflow the destination buffer under any circumstances,
so we only copy a maximum of bufferlen characters minus one, to account for the
terminating zero of the string. We ensure afterwards that the string is zero terminated, and
we return the result.
The rest of the program remains the same, so it is not shown. It is important to remember to
get rid of that MessageBox call however!87
We compile the library, and we want to test it, but… we need a test program. A library is not
an executable by itself.
Besides this, we need a header file that the user of the library will use to get the prototype of
our function. Its contents are very simple:
int GetString(char *prompt, char *destBuffer,int bufferlen);
and that’s all.

87. Well, this is not great deal; we have just to answer YES when it proposes to create the skeleton.
 Libraries 147

Now, we have to define a new project that will contain the test code. We do that in the same
way as we have created the other projects: we choose the ‘Create project’ option in the
‘Project’ menu bar, and we name it appropriately “testdialog”.
We do NOT specify a windows application. Since our library should be independent whether
it is called from a windows program or a console application, we should test that now.

Now, when creating the project, we ask the wizard to create a console application.88 We leave
everything by default, but when we arrive at the linker settings dialog, we add our dialog.lib to
the libraries entry field, like this:
Another issue to remember, is the following:
We need a resource file. Since in our library there are no resources, we have to add those
resources to our test program. This is easy to do in this case: we just add the dialog.rc resource
file to the project. The interface for the users of the library however is terrible. All programs
that use our library will be forced to include somehow the resource for the dialog box! Includ-
ing a resource in another resource file is difficult, to say the least.
Wow, this looks like a showstopper actually.
OK. This presents an unexpected and serious trouble for our library project, but we will not
leave things at midway. We finish our test program by changing our “main” function, like this:
extern int APIENTRY GetString(char *prompt,char *buf,int len);

int main(void)
{
char buffer[1024];
if (GetString("Enter a string",buffer,sizeof(buffer))) {
printf("String is %s\n",buffer);
}
else printf("User cancelled!\n");
return 0;
}

88. It could be argued that this could be done with DLLs too: the linker should export all externally
visible symbols. In practice is better only to export symbols that should be visible. This avoids name clashes.
148 C Tutorial

When we start this program from the command line, we see:

This isn’t that bad, for a start. We have a sophisticated line editor, complete with arrow inter-
face to move around in the text, clipboard support built in, delete and backspace keys already
taken care of, etc. If we would have to write ourselves an equivalent program, it would cost us
probably days of development. To develop the clipboard interface already is quite a challenge.
But we are confronted to the problem of resources. Our static library idea was a dead end. We
have to find something else.

Summary: The C language supports the concept of code reuse in the form of libraries. The
static libraries are combined with the main application at link time (statically). They can’t
contain resources.
 Dynamically linked libraries (DLLs) 149

2.8 Dynamically linked libraries (DLLs)

A dynamically linked library is just like a static library: it contains a set of useful functions
that can be called from other software. As with normal .lib libraries, there is no main function.
Unlike static libraries however, they have several features that make them a very interesting
alternative to static libraries:
• When they are loaded, the loader calls a function of the library to allow load time
initializations. This allows us to register our class, for instance, or do other things.
• When the program that loads them starts or ends a new thread the system arranges for
calling the same function. This allows us to take special actions when this event occurs.
We do not need this feature for our application here, but other software do.
• When the library is unloaded, either because the program explicitly does it or because
simply the program is finished, we get notified. Here we can reverse the actions we
performed when the library was loaded: we can, for instance, unregister our window
class.
• DLLs can contain resources. This solves the problem of forcing the user of the library to
link a bunch of resources to his/her program.
DLLs need to specify which functions are going to be exported, i.e. made visible to the outside
world. With static libraries this is not really necessary since the librarian will write all the sym-
bols with the external type to the library symbol table automatically.89
We can declare that a symbol will be exported using two methods:

1) We can put in the declaration of the symbol the __declspec(dllexport) mark.

2) We can write the name of the symbol in a special file called definitions file (with the .def
extension) and pass this file to the linker.
Which method you use is a matter of taste. Writing __declspec(dllexport) in the source code is
quite ugly, and may be non-portable to other systems where the conventions for dynamically
linked code may be completely different. A definitions file spares us to hard wire that syntax
in the source code.
The definitions file has its drawbacks too however. We need yet another file to maintain,
another small thing that can go wrong.
For our example we will use the __declspec(dllexport) syntax since we have only one function
to export.
We return to our library project, and reopen it. We go again to the linker configuration tab, that
now is called “librarian” since we are building a static library, and we check the radio-button
corresponding to a DLL project. We answer yes when Wedit says whether it should rebuild the
makefile and there we are. Now we have to make the modifications to our small library.
We have to define a function that will be called when the library is loaded. Traditionally, the
name of this function has been LibMain since the days of Windows 3.0 or even earlier. We
stick to it and define the following function:
int WINAPI LibMain(HINSTANCE hDLLInst, DWORD Reason, LPVOID Reserved)
{
switch (Reason)
{

89. You can change this by pressing the corresponding button in the linker configuration tab, or by
giving the argument –nounderscores to the linker, when building the DLL.
150 C Tutorial

case DLL_PROCESS_ATTACH:
hinst = hDLLInst;
DoRegisterClass();
break;
case DLL_PROCESS_DETACH:
UnregisterClass("dialog",hDLLInst);
break;
case DLL_THREAD_ATTACH:
break;
case DLL_THREAD_DETACH:
break;
}
return TRUE;
}
This function, like our dialog function or many other functions under windows, is a callback
function, i.e. a function that is called by the operating system, not directly from our code.
Because of this fact, its interface, the arguments it receives, and the result it returns is fixed.
The operating system will always pass the predefined arguments to it, and expect a well-
defined result.
The arguments that we receive are the following:

1) We receive a HANDLE to the instance of the DLL. Note that we have to pass to several
functions this handle later on, so we will store it away in a global variable.

2) We receive a DWORD (an unsigned long) that contains a numerical code telling us the
reason why we are being called. Each code means a different situation in the life cycle of
the DLL. We have a code telling us that the we were just loaded
(DLL_PROCESS_ATTACH), another to inform us that we are going to be unloaded
(DLL_PROCESS_DETACH), another to inform us that a new thread of execution has been
started (DLL_THREAD_ATTACH) and another to tell us that a thread has finished
(DLL_THREAD_DETACH).

3) The third argument is reserved by the system for future use. It is always zero.
The result of LibMain should be either TRUE, the DLL has been correctly initialized and
loading of the program can continue, or zero meaning a fatal error happened, and the DLL is
unable to load itself.
Note that we return always TRUE, even if our registration failed.
Why?
If our registration failed, this module will not work. The rest of the software could go on run-
ning however, and it would be too drastic to stop the functioning of the whole software
because of a small failure in a routine that could be maybe optional.
Why the registration of our class could fail?
One of the obvious reasons is that the class is already registered, i.e. that our calling program
has already loaded the DLL, and it is loading it again. Since we do not unregister our class,
this would provoke that we try to register the class a second time.
For the time being, we need to handle only the event when the DLL is loaded or unloaded. We
do two things when we receive the DLL_PROCESS_ATACH message: we store away in our
global variable the instance of the DLL, and then we register our string dialog box class. We
could have just done it in the LibMain function, but is clearer to put that code in its own rou-
tine. We write then:
static void DoRegisterClass(void)
 Dynamically linked libraries (DLLs) 151

{
WNDCLASS wc;

memset(&wc,0,sizeof(wc));
wc.lpfnWndProc = DefDlgProc;
wc.cbWndExtra = DLGWINDOWEXTRA;
wc.hInstance = hinst;
wc.hCursor = LoadCursor(NULL, IDC_ARROW);
wc.hbrBackground = (HBRUSH) (COLOR_WINDOW + 1);
wc.lpszClassName = "dialog";
RegisterClass(&wc);
}
You see that the code is the same as the code we had originally in WinMain, then in our Get-
String procedure, etc.
To finish LibMain, we handle the message DLL_PROCESS_DETACH unregistering the class
we registered before.
With this, our GetString procedure is simplified: We do not need to test our flag to see if the
class has been registered any more. We can be certain that it was.
int APIENTRY __declspec(dllexport)
GetString(char *prompt, char *destbuffer,int bufferlen)
{
int result;

result = DialogBoxParam(hinst,
MAKEINTRESOURCE(IDD_MAINDIALOG),
NULL,
(DLGPROC) DialogFunc,
(int)prompt);

if (result == 1) {
strncpy(destbuffer,buffer,bufferlen-1);
destbuffer[bufferlen-1] = 0;
}
return result;
}
We compile and link our DLL by pressing F9. Note that when we are building a DLL, lcc-
win32 will generate three files and not only one as with a normal executable.

1) We obtain of course a dialog.dll file that contains the DLL.

2) We obtain an import library that allows other programs to be linked with this DLL. The
name will be dialog.lib, but this is not a normal library. It is just a library with almost no
code containing stubs that indicate the program loader that a specific DLL is used by the
program.

3) We obtain a text file called dialog.exp that contains in text form the names that are exported
from the DLL. If, for any reason we wanted to regenerate the import library for the DLL,
we could use this file together with the buildlib utility of lcc-win32 to recreate the import
library. This can be important if you want to modify the names of the exported functions,
establish synonyms for some functions or other advanced stuff.
152 C Tutorial

2.9 Using a DLL

To use our newly developed DLL we just plug-in the older test program we had done for our
static library. The interface is the same; nothing has changed from the user code of our library.
The only thing that we must do differently is the link step, since now we do not need to add the
resource file to our program.
Wedit leaves our DLL in the lcc directory under the project main directory. We just recompile
our testdialog.c program that we had before. Here is the code for testdialog.c again:
extern int APIENTRY GetString(char *prompt,char *buffer,int len);

int main(int argc,char *argv[])

{
char buffer[1024];

if (GetString("Enter a string",
buffer,sizeof(buffer))) {
printf("String is %s\n",buffer);
}
else printf("User cancelled\n");
return 0;
}
We compile this in the directory of the project, without bothering to create a project. Suppose
that our project is in
h:\lcc\projects\dialog
and the dll is in
h:\lcc\projects\dialog\lcc
We compile with the command:
lcc testdialog.c
then we link with the command:
lcclnk testdialog.obj lcc\dialog.lib
Perfect! We now type the name of the program to test our dll.
testdialog
but instead of the execution of the program we see a dialog box like this:

The system tells us that it can’t find the dll.

Well, if you reflect about this, this is quite normal. A DLL must be linked when the execution
of the program starts. The system will search in the start directory of the program, and in all
directories contained in the PATH environment variable. If it doesn’t find a “dialog.dll” any-
where it will tell the user that it can’t execute the program because a missing DLL, that’s all.
The solution is to copy the DLL into the current directory, or copy the DLL in one of the direc-
tories in your PATH variable. Another solution of course is to go to the directory where the
DLL is, and start execution of the program there.
 Using a DLL 153

This dependency on the DLL is quite disturbing. All programs that use the DLL in this fashion
would need to have the DLL in their startup directory to be able to work at all.
A way to get rid of this is to avoid linking with the DLL import library. Yes you will say, but
how will we use the DLL?
DLLs can be loaded into the program’s address space with the API LoadLibrary. This API will
do what the program loader does when loading a program that contains a reference to a DLL.
If the load succeeds, the API will return us a handle to the library, if not, it will return us an
INVALID_HANDLE as defined in windows.h.
After loading the DLL, we can get the address of any exported function within that DLL just
by calling another windows API: GetProcAddress. This API receives a valid DLL handle, and
a character string containing the name of the function to find, and will return an address that
we can store in a function pointer.
Let’s do this.
#include <windows.h> (1)
#include <stdio.h> (2)
int (APIENTRY *pfnGetString)(char *,char *,int); (3)

int main(int argc,char *argv[])

{
char buffer[1024];
HANDLE dllHandle = LoadLibrary( (4)
"h:\\lcc\\examples\\dialog\\lcc\\dialog.dll");

if (dllHandle == INVALID_HANDLE_VALUE) { (5)

fprintf(stderr,"Impossible to load the dll\n");
exit(0);
}
pfnGetString = (int (APIENTRY *)(char *,char *,int))
GetProcAddress(dllHandle,"_GetString@12"); (6)
if (pfnGetString == NULL) {
fprintf(stderr,
"Impossible to find the procedure GetString\n");
exit(1);
}
if (pfnGetString(
"Enter a string",buffer,sizeof(buffer))) {
printf("String is %s\n",buffer);
}
else printf("User cancelled\n");
return 0;
}
We go step by step:
We need to include windows.h for getting the prototypes of LoadLibrary, and GetProcAd-
dress, besides some constants like INVALID_HANDLE_VALUE.
stdio.h is necessary too, since we use fprintf
This is a function pointer called pfnGetString, that points to a function that returns an int and
takes a char *, another char * and an int as arguments. If you do not like this syntax please bear
with me. Even Dennis Ritchie says this syntax isn’t really the best one.
We store in our dllHandle, the result of calling LoadLibrary with an absolute path so it will
always work, at least in this machine. Note that the backslashes must be repeated within a
character string.
154 C Tutorial

We test if the handle received is a correct one. If not we put up some message and we exit the
program.
We are now ready to assign our function pointer. We must cast the return value of GetProcAd-
dress to a function like the one we want. The first part of the statement is just a cast, using the
same construction that the declaration of the function pointer before, with the exception that
we do not include the identifier of course, since this is a cast. But the arguments to GetPro-
cAddress are weird. We do not pass really the name of the function GetString, but a
name _GetString@12. Where does this name come from?
The rest of the program stays the same.
To understand where this weird name comes from, we have to keep in mind the following
facts:
lcc-win32 like many other C compilers, adds always an underscore to the names it generates
for the linker.90
Since our function is declared as _stdcall, windows conventions require that we add to the
name of the function the character ‘@’ followed by the size of the function arguments. Since
our function takes a char pointer (size 4) another char pointer, and an integer (size 4 too), we
have 12 bytes of procedure arguments, hence the 12. Note that all types smaller than an integer
will be automatically be promoted to integers when passing them to a function to keep the
stack always aligned, so that we shouldn’t just take the size of the arguments to make the addi-
tion. All of this can become really complicated if we have structures that are pushed by value
into the stack, or other goodies.
The best thing would be that our DLL would export _GetString@12 as GetString. PERIOD.
Well, this is exactly where our dialog.def file comes handy. Here is a dialog.def that will solve
our problem.
LIBRARY dialog
EXPORTS
_GetString@12=GetString
We have in the first line just the name of the DLL in a LIBRARY statement, and in the second
line two names. The first one is the name as exported by the compiler, and the second one is
the name as it should be visible from outside. By default, both are the same, but now we can
separate them. With these instructions, the linker will put in the export table of the DLL the
character string “GetString”, instead of the compiler-generated name.91
Once this technical problems solved, we see that our interface is much more flexible now. We
could just return FALSE from our interface function if the DLL wasn’t there, and thus disable
some parts of the software, but we wouldn’t be tied to any DLL. If the DLL isn’t found, the
functionality it should fulfill can’t be present but nothing more, no catastrophes.

90. In the documentation of windows, you will find out that in the .def file you can write other kinds of
statements. None of them are supported by lcc-win32 and are silently ignored since they are redundant with
the command line arguments passed to the linker. Do not write anything else in the .def file besides the names
of the exports.
91. The Macintosh works in the same manner, albeit with a much more primitive system.
 A more formal approach. 155

2.10 A more formal approach.

2.10.1 New syntax

Now that we know how to make a DLL, we should be able to understand more of C, so let’s
come back again to the syntax and take a closer look.
We have extensively used the switch statement above. This statement allows you to select sev-
eral cases from the possible ones and act accordingly. It is composed from a switch expression
that should evaluate to an integer result, and several pieces of code, associated to an integer
value. The compiler generates code to evaluate the switch expression, and jump accordingly to
the corresponding case.
In C, a case will continue with the next one, unless you finish it with a break expression, the
same expression that is used to break from a loop.
switch (expr) {
case 123:
a = p*6;
case 251:
a = p*9;
}
In this example, if expr is equal to 123, the expression a = p*6; will be executed, then the
expression a = p*9; Yes, I know this is not evident, but it’s like that, and zillion lines of
code are already written that use this feature. Cases can be easily merged:
switch (expr) {
case 123:
case 834:
case –45:
a = p*6;
break;
case 251:
a = p*9;
}
Note that here a default case is missing. This is not a good practice, even if you do not want to
do anything in the default case. More readable is:
switch (expr) {
case 123:
case 834:
case –45:
a = p*6;
case 251:
a = p*9;
break;
default:
break;
}
Here we see that you do NOT intend to do anything in the default case.
Another problem area is function pointer casting, what leads straight into gibberish-looking
code.
int (APIENTRY *pfn)(char *,char *,int);
Better is to typedef such constructs with:
typedef int (APIENTRY *pfn)(char *,char *,int);
Now we declare our function pointer just with
156 C Tutorial

pfn pfnGetString;
and we can cast the result of GetProcAddress easily with:
pfnGetString = (pfn)GetProcAddress( ... );
Hiding the gibberish in a typedef is quite useful in this situations, and much more readable
later.

2.10.2 Event oriented programming

In another level, what those programs show us is how to do event-oriented programs, i.e. pro-
grams designed to react to a series of events furnished by the event stream.
Under windows, each program should make this event-stream pump turn, by writing some-
where:
while (GetMessage()) {
ProcessMessage();
}

we will describe the exact syntax later.92 This message-pump is hidden now from view under
the DefDlgProc procedure, but it is the source of all the messages passed to the dialog proce-
dure that we defined.
A windows program is designed to react to those messages, or events. It will process them in
sequence, until it decides to stop the message pump ending the program.
The general structure is then:
• Initialize the application, register the window classes, etc.
• Start the message pump
• Process events until a predefined event (generally the closing of the main window)
provokes the stop of the message pump.
• Cleanup
This was the standard way of windows programming until the C++ “wizards” decided that this
message pump was too ugly to be shown to programmers and hid it behind an “application
object”. Later, several years later, things changed again and the ATL environment made the
famous “message pump” visible again. 93
Texts started appearing explaining the surprised programmers what was WinMain and that
“message pump” that was again the non-plus-ultra of modernity.94 Luckily for people pro-

92. Application frameworks like MFC introduce an additional simplifying layer between you and the
operating system. Much has been said and written about them, and here I will not discuss this in much more
detail. Suffice to note that the purpose of lcc-win32 is to let you be always in control of what is going on. You
can do here anything, contrary to a framework, where you can only do what the framework provides for, and
nothing else.
True, an application framework can simplify the coding, and many people use them. It would be feasible
to build such a framework with lcc-win32, but … I will leave this problem “as an exercise to the reader”…
93. In the data processing field, we have been always recycling very old ideas as “new, just improved”.
Object oriented programming was an idea that came from the old days of Simula in the beginning of the
seventies but was “rediscovered” or rather “reinvented” in the late 80s. Garbage collection was standard in
lisp systems in the seventies, and now has been discovered again by Mr. Bill Gates, in the next century, with
his proposal for the C# language.
94. Remember the basics of Boolean logic: a bit ANDed with another will be one only if both bits are 1.
A bit Ored with another with return 1 only if one or both of them is 1.
 A more advanced window 157

gramming in C, all those fads were invisible. Since C programming emphasizes low-level
knowledge of what is going on, the “message pump” has always been there and we knew
about it.
Messages are sent to window objects that are members of a class of similar objects or window
class. The procedure SendMessage can also send messages to a window object, as if they
would have been sent by the system itself. We send, for instance, the message
EM_SETLIMITTEXT to an entry field to set the maximum number of characters that the
entry field will process.
The system relies on the window procedure passing all messages that it doesn’t handle to the
default message procedure for that class of objects, to maintain a coherent view of the desktop
and to avoid unnecessary code duplication at each window.
We are now ready to start our next project: building a real window, not just a simple dialog.

2.11 A more advanced window

We will create the next example with the wizard too. It is a handy way of avoiding writing a
lot of boilerplate code. But it is obvious that we will need to understand every bit of what the
wizard generates. We create a new project, as usual with the “project” then “Create” menu
option, and we give it the name of ”winexample”. The wizard shows us a lot of dialogs with
many options that will be explained later, so just stick with the defaults. Choose a single win-
dow application:

After pressing the “next” button till the dialog boxes disappear, we press F9, compile the
whole, and we run it. We see a single white window, with a status bar at the bottom, a sum-
mary menu, and nothing else. Well, this is the skeleton of a windows application. Let’s see it
in more detail.
We start as always in the same place. We go to the WinMain function, and we try to figure out
what is it doing. Here it is:
int WINAPI WinMain(HINSTANCE hInstance, HINSTANCE hPrevInstance, LPSTR
lpCmdLine, INT nCmdShow)
{
MSG msg;
HANDLE hAccelTable;
hInst = hInstance;
if (!InitApplication())
return 0;
hAccelTable = LoadAccelerators(hInst,MAKEINTRESOURCE(IDACCEL));
if ((hwndMain = CreatewinexampleWndClassWnd()) == (HWND)0)
return 0;
CreateSBar(hwndMain,"Ready",1);
ShowWindow(hwndMain,SW_SHOW);
while (GetMessage(&msg,NULL,0,0)) {
if (!TranslateAccelerator(msg.hwnd,hAccelTable,&msg)) {
TranslateMessage(&msg);
DispatchMessage(&msg);
}
}
return msg.wParam;
158 C Tutorial

}
We have the same schema that we saw before, but this time with some variations. We start the
application (registering the window class, etc.), we load the keyboard accelerators, we create
the window, the status bar, we show our window, and then we enter the message loop until we
receive a WM_QUIT, that breaks it. We return the value of the “wParam” parameter of the last
message received (WM_QUIT of course).
Simple isn’t it?
Now let’s look at it in more detail.
The “InitApplication” procedure initializes the WNDCLASS structure with a little more care
now, since we are not using our DefDialogProc any more, there are a lot of things we have
to do ourselves. Mostly, that procedure uses the standard settings:
static BOOL InitApplication(void)
{
WNDCLASS wc;

memset(&wc,0,sizeof(WNDCLASS));
// The window style
wc.style = CS_HREDRAW|CS_VREDRAW |CS_DBLCLKS ;
wc.lpfnWndProc = (WNDPROC)MainWndProc;
wc.hInstance = hInst;
// The color of the background
wc.hbrBackground = (HBRUSH)(COLOR_WINDOW+1);
wc.lpszClassName = "winexampleWndClass";
// The menu for this class
wc.lpszMenuName = MAKEINTRESOURCE(IDMAINMENU);
// default cursor shape: an arrow.
wc.hCursor = LoadCursor(NULL,IDC_ARROW);
// default icon
wc.hIcon = LoadIcon(NULL,IDI_APPLICATION);
if (!RegisterClass(&wc))
return 0;
// ---TODO--- Call module specific initialization routines here

return 1;
}
The style of the window used is a combination of integer constants like CS_HREDRAW, and
others, combined using the OR operator, the vertical bar. What does this mean?
This is a standard way of using bits in C. If you go to the definition of CS_HREDRAW (right-
click in that name and choose the “Goto definition” option), you will see that the value is 2.
Other constants like CS_DBLCLKS have a value of 8. All those numbers are a power of two.
Well, a power of two by definition will always have a single bit set. All other bits will be zero.
If you OR those numbers with each other, you will obtain a number that has the bits set that
correspond to the different values used. In our case this statement is equivalent to:
wc.style = 2 | 1 | 8;
8 ored with 1 is 1 0 0 1, ored with 2 is 1 0 1 1, what is equal to 11 in decimal notation.
This is a very common way of using flags in C. Now, if you want to know if this window is a
window that answers to double-clicks, you just have to query the corresponding bit in the style
integer to get your answer. You do this with the following construct:
if (wc.style & CS_DBLCLKS) {

}
 A more advanced window 159

We test in this if expression, if the value of the “style” integer ANDed with 8 is different than
zero. Since CS_DBLCLKS is a power of two, this AND operation will return the value of that
single bit.95 Note too that 1 is a power of two since 2 to the power of zero is one.
We will return to this at the end of this section.
Coming back to our initialization procedure, there are some new things, besides this style set-
ting. But this is just a matter of reading the windows documentation. No big deal. There are
many introductory books that augment their volume just with the easy way of including a lot
of windows documentation in their text. Here we will make an exception.
But what is important however is that you know how to look in the documentation! Suppose
you want to know what the hell is that CS_DBLCLKS constant, and what does it exactly mean.
You press F1 in that identifier and nothing. It is not in the index.
Well, this constant appears in the context of RegisterClass API. When we look at the docu-
mentation of RegisterClass, we find a pointer to the doc of the WNDCLASS structure. Going
there, we find in the description of the style field, all the CS_* constants, neatly explained.
Note that not all is in the index. You have to have a feeling of where to look. Lcc-win32 comes
with a help file of reasonable size to be downloaded with a standard modem. It is 13MB com-
pressed, and it has the essentials. A more detailed documentation complete with the latest stuff
is in the Software Development Kit (SDK) furnished by Microsoft. It is available at their Web
site, and it has a much more sophisticated help engine.
After initializing the window class, the WinMain function loads the accelerators for the appli-
cation. This table is just a series of keyboard shortcuts that make easy to access the different
menu items without using the mouse and leaving the keyboard. In the resource editor you can

95. You may wonder what that variable “msg” stands for. It is a structure of type MSG, that is defined in
windows.h as follows:
typedef struct tagMSG
{ HWND hwnd;
UINT message;
WPARAM wParam;
LPARAM lParam;
DWORD time;
POINT pt; } MSG, *PMSG, *NPMSG, *LPMSG;
Note that in C you can append as many names to a pointer to a structure as you like, and in windows this
is used a lot. The reason is an historical one. In windows 16 bits there were several types of pointers: near (16
bit pointers), far (long pointers of 32 bits) and generally the near pointers were prefixed with NP, the 32 bit
ones with LP. This was retained for compatibility reasons until today, even if there are only 32 bit pointers
now.
This structure contains then, the following fields:
• hwnd: The handle of the specific window to which the message is directed.
• message: A 16-bit value identifying the message.
• wparam: A 32-bit value identifying the first message parameter. Its meaning depends on the
message being sent.
• lParam: A 32-bit value identifying the second message parameter.
• time: A 32-bit value identifying the time when the event that provoked this message happened.
• pt: This is a POINT structure containing the coordinates of the mouse in the instant the event
happened that provoked this message.
160 C Tutorial

edit them, add/delete/change, etc. To do this you start the resource editor and you press the
“dir” button in the upper right. You will see the following display.

You click in the “Accelerator” tree tab, and you will see the following:

We have here the accelerator called IDM_EXIT that has the value of 300. This is just Ctrl+Q
for quit. The key value is 81, the ASCII value of the letter ‘q’, with a flag indicating that the
control key must be pressed, to avoid quitting just when the user happens to press the letter q
in the keyboard!
Double-clicking in the selected line leads us to yet another dialog:

Here you can change the accelerator as you want. The flags are explained in the documenta-
tion for the resource editor.
But this was just another digression, we were speaking about WinMain and that statement:
LoadAccelerators… Well, let’s go back to that piece of code again.
 A more advanced window 161

After loading the accelerators, the status bar is created at the bottom of the window, and then,
at last, we show the window. Note that the window is initially hidden, and it will be shown
only when all things have been created and are ready to be shown. This avoids screen flicker-
ing, and saves execution time. It would be wasteful to redraw the window before we created
the status bar, since we would have to do that again after it is created.
We will speak about status bar later, since that is not crucial now. What really is important is
the message loop that begins right after we call ShowWindow.
while (GetMessage(&msg,NULL,0,0)) {
if (!TranslateAccelerator(msg.hwnd,hAccelTable,&msg)) {
TranslateMessage(&msg);
DispatchMessage(&msg);
}
}
This loop calls the API GetMessage. If that returns TRUE, we call the API to translate the
accelerators. This API will convert a sequence of keystrokes into a WM_COMMAND mes-
sage, as it was sent by a menu, if it finds a correspondence between the keystrokes and the
accelerator table that we loaded just a few lines above. If TranslateAccelerator doesn’t find
any correspondence, we go on by calling TranslateMessage, that looks for sequences of key
pressing and key up messages, and does the dirty work of debouncing the keyboard, handling
repeat sequences, etc. At last, we dispatch our message to the procedure indicated in the win-
dow class.
And that is it. We loop and loop and loop, until we eventually get a WM_QUIT, that provokes
that GetMessage returns FALSE, and the while loop is finished.96
Wow. Finished?
We have hardly started. What is interesting now, is that we have a skeleton to play with. We
will show in the next sections how we add things like a dialog box, etc.

Summary: Windows programming looks intimidating at first. But it is just the looks.Before
we go on, however, as promised, let’s look in more details to how flags are set/unset in C.
Flags are integers where each bit is assigned a predefined meaning. Usually with a pre-
processor define statement, powers of two are assigned a symbolic meaning like in the case of
CS_DBLCLKS above. In a 32-bit integer we can stuff 32 of those. We test those flags with:

if (flag & CONSTANT) {

}

we set them with:

flag |= CONSTANT;

we unset them with:

flag &= ~CONSTANT;

This last statement needs further explanations. We use the AND operator with the complement
of the constant. Since those constants have only one bit set, the complement of the constant is

96. Never forget this: local variables do NOT retain their value from one call of the function to the next
one!
162 C Tutorial

an integer with all the bits turned into ones except the bit that we want to unset. We AND that
with our flag integer: since all bits but the one we want to set to zero are one, we effectively
turn off that bit only, leaving all others untouched.
 A more complex example: a "clone" of spy.exe 163

2.12 A more complex example: a "clone" of spy.exe

What can we do with the empty window that Wedit generates?
Let's do a more difficult problem: We want to find out all the windows that are opened at a
given time in the system. We will display those windows in a tree control, since the child win-
dows give naturally a tree structure. When the user clicks in a window label, the program
should display some information about the window in the status bar.
We generate a skeleton with Wedit, as described above. We create a new project, and generate
a simple, single window application.

2.12.1 Creating the child windows

OK. Now we come back to the task at hand. The first thing to do is to create the tree control
window. A good place to do these kinds of window creations is to use the opportunity the sys-
tem gives to us, when it sends the WM_CREATE message to the main window. We go to the
procedure for the main window, called MainWndProc, and we add the WM_CREATE case to
the switch of messages:
LRESULT CALLBACK MainWndProc(HWND hwnd,UINT msg,WPARAM wParam,LPARAM
lParam)
{
static HWND hwndTree;

switch (msg) {
case WM_CREATE:
hwndTree = CreateTree(hwnd,IDTREEWINDOW);
break;
This is "top down" design. We hide the details of the tree window creation in a function that
returns the window handle of the tree control. We save that handle in a static variable. We
declare it as static, so that we can retrieve its value at any call of MainWndProc.97
Our CreateTree function, looks like this:
static HWND _stdcall CreateTree(HWND hWnd,int ID)
{
return
CreateWindowEx(WS_EX_CLIENTEDGE,
WC_TREEVIEW,"",
WS_VISIBLE|WS_CHILD|WS_BORDER|TVS_HASLINES|
TVS_HASBUTTONS|TVS_DISABLEDRAGDROP,
0,0,0,0,
hWnd,(HMENU)ID,hInst,NULL);
}
This function receives a handle to the parent window, and the numeric ID that the tree control
should have. We call the window creation procedure with a series of parameters that are well
described in the documentation. We use the value of the hInst global as the instance, since
the code generated by Wedit conveniently leaves that variable as a program global for us to
use.
Note that we give the initial dimensions of the control as zero width and zero height. This is
not so catastrophic as it seems, since we are relying in the fact that after the creation message,
the main window procedure will receive a WM_SIZE message, and we will handle the sizing

97. In page 218. Published by St Martin’s Press. 1990. ISBN 0-312-06179-X (pbk)
164 C Tutorial

of the tree control there. This has the advantage that it will work when the user resizes the
main window too.

2.12.2 Moving and resizing the child windows

We add code to the WM_SIZE message that Wedit already had there to handle the resizing of
the status bar at the bottom of the window.
LRESULT CALLBACK MainWndProc(HWND hwnd,UINT msg,WPARAM wParam,LPARAM
lParam)
{
static HWND hwndTree;
RECT rc,rcStatus;

switch (msg) {
case WM_CREATE:
hwndTree = CreateTree(hwnd,IDTREEWINDOW);
break;
case WM_SIZE:
SendMessage(hWndStatusbar,msg,wParam,lParam);
InitializeStatusBar(hWndStatusbar,1);
GetClientRect(hwnd,&rc);
GetWindowRect(hWndStatusbar,&rcStatus);
rc.bottom -= rcStatus.bottom-rcStatus.top;
MoveWindow(hwndTree,0,0,rc.right,rc.bottom,1);
break;
We ask windows the current size of the main window with GetClientRect. This proce-
dure will fill the rectangle passed to it with the width and height of the client area, i.e. not con-
sidering the borders, title, menu, or other parts of the window. It will give us just the size of the
drawing surface.
We have a status bar at the bottom, and the area of the status bar must be subtracted from the
total area. We query this time using the GetWindowRect function, since we are interested in
the whole surface of the status bar window, not only in the size of its drawing surface. We sub-
tract the height of the window from the height that should have the tree control, and then we
move it to the correct position, i.e. filling the whole drawing surface of the window. And we
are done with drawing.

2.12.3 Starting the scanning.

Now we pass to the actual task of our program. We want to fill the tree control with a descrip-
tion of all the windows in the system. A convenient way to do this is to change the "New"
menu item into "Scan", and start scanning for windows when the user chooses this item.
To do this, we add an entry into the MainWndProc_OnCommand function:
void MainWndProc_OnCommand(HWND hwnd,
int id, HWND hwndCtl, UINT codeNotify)
{
switch(id) {
case IDM_NEW:
BuildTree(hwnd);
break;
case IDM_EXIT:
PostMessage(hwnd,WM_CLOSE,0,0);
break;
}
}
 A more complex example: a "clone" of spy.exe 165

Simple isn't it? We just call "BuildTree" and we are done.

2.12.4 Building the window tree.

We start with the desktop window, we add it to the tree, and then we call a procedure that will
enumerate all child windows of a given window. We have two directions to follow: the child
windows of a given window, and the sibling windows of a given window. This is true for the
desktop window too.
Let's look at the code of "BuildTree":
int BuildTree(HWND parent)
{
HWND Start = GetDesktopWindow();
HWND hTree = GetDlgItem(parent,IDTREEWINDOW);
TV_INSERTSTRUCT TreeCtrlItem;
HTREEITEM hNewNode;

memset(&TreeCtrlItem,0,sizeof(TreeCtrlItem));
TreeCtrlItem.hParent = TVI_ROOT;
TreeCtrlItem.hInsertAfter = TVI_LAST;
TreeCtrlItem.item.mask = TVIF_TEXT | TVIF_PARAM;
TreeCtrlItem.item.pszText = "Desktop";
hNewNode = TreeView_InsertItem(hTree,&TreeCtrlItem);
Start = GetWindow(Start,GW_CHILD);
Scan(hTree,hNewNode,Start);
return 1;
}
We start at the start, and we ask windows to give us the window handle of the desktop window.
We will need the tree window handle too, so we use "GetDlgItem" with the parent window
of the tree control, and it's ID. This works, even if the parent window is a normal window, and
not a dialog window.
We go on by filling our TV_INSERTSTRUCT with the right values. This is a common inter-
face for many window functions. Instead of passing n parameters, we just fill a structure and
pass a pointer to it to the system. Of course, it is always a good idea to clean the memory space
with zeroes before using it, so we zero it with the "memset" function. Then we fill the fields
we need. We say that this item is the root item, that the insertion should happen after the last
item, that the item will contain the text "Desktop", and that we want to reserve place for a
pointer in the item itself (TVIF_PARAM). Having done that, we use the macro for inserting an
item into the tree.
The root item created, we should then scan the siblings and child windows of the desktop.
Since the desktop is the root of all windows it has no siblings, so we start at its first child. The
GetWindow function, gives us a handle to it.

2.12.5 Scanning the window tree

We call our "Scan" function with the handle of the tree control, the handle to the just inserted
item, and the window handle of the first child that we just obtained.
The "Scan" function looks like this:
void Scan(HWND hTree,HTREEITEM hTreeParent,HWND Start)
{
HWND hwnd = Start,hwnd1;
TV_INSERTSTRUCT TreeCtrlItem;
HTREEITEM htiNewNode;
char bufTxt[256],bufClassName[256],Output[1024];
166 C Tutorial

while (hwnd != NULL) {

SendMessage(hwnd,WM_GETTEXT,250,(LPARAM) bufTxt);
GetClassName(hwnd,bufClassName,250);
wsprintf(Output,"\"%s\" %s",bufTxt,bufClassName);
memset(&TreeCtrlItem,0,sizeof(TreeCtrlItem));
TreeCtrlItem.hParent = hTreeParent;
TreeCtrlItem.hInsertAfter = TVI_LAST;
TreeCtrlItem.item.mask = TVIF_TEXT | TVIF_PARAM;
TreeCtrlItem.item.pszText = (LPSTR) Output;
TreeCtrlItem.item.lParam = (LPARAM) hwnd;
htiNewNode =
TreeView_InsertItem ( hTree,&TreeCtrlItem);
if((hwnd1=GetWindow(hwnd,GW_CHILD))!=NULL)
Scan(hTree,htiNewNode,hwnd1);
hwnd=GetWindow(hwnd,GW_HWNDNEXT);
}
}
We loop through all sibling windows, calling ourselves recursively with the child windows.
In our loop we do:
We get the text of the window, to show it in our tree. We do this by sending the WM_GETTEXT
message to the window.
We get the class name of the window.
We format the text (enclosed in quotes) and the class name in a buffer.
We start filling the TV_INSERTSTRUCT. These steps are very similar to what we did for the
desktop window.
After inserting our node in the tree, we ask if this window has child windows. If it has, we call
Scan recursively with the new node and the new child window.
Then we ask if this window has sibling windows. If it has, the main loop will go on since
GetWindow will give us a non-null window handle. If it hasn't we are done and we exit.

2.12.6 Review
Let's look at our "BuildTree" function again and ask us:
How could this fail?
We notice immediately several things.
We always add items to the tree at the end, but we never cleanup the tree control. This means
that after a few times the user has clicked in the menu, we will have several times all the win-
dows of the system in our tree. All nodes should be deleted when we start.
The tree control will redraw itself several times when we add items. This is unnecessary and
produces a disturbing blinking in the display. We should hold the window without any redraw-
ing until all changes are done and then redraw once at the end.
We modify the "BuildTree" procedure as follows:
int BuildTree(HWND parent)
{
HWND Start = GetDesktopWindow();
HWND hTree = GetDlgItem(parent,IDTREEWINDOW);
TV_INSERTSTRUCT TreeCtrlItem;
HTREEITEM hNewNode;

SendMessage(hTree,WM_SETREDRAW,0,0);
 A more complex example: a "clone" of spy.exe 167

TreeView_DeleteAllItems(hTree);
memset(&TreeCtrlItem,0,sizeof(TreeCtrlItem));
TreeCtrlItem.hParent = TVI_ROOT;
TreeCtrlItem.hInsertAfter = TVI_LAST;
TreeCtrlItem.item.mask = TVIF_TEXT | TVIF_PARAM;
TreeCtrlItem.item.pszText = "Desktop";
hNewNode = TreeView_InsertItem(hTree,&TreeCtrlItem);
Start = GetWindow(Start,GW_CHILD);
Scan(hTree,hNewNode,Start);
TreeView_Expand(hTree,hNewNode,TVE_EXPAND);
SendMessage(hTree,WM_SETREDRAW,1,0);
return 1;

}
We enclose all our drawing to the control within two calls to the SendMessage function, that
tell essentially the tree control not to redraw anything. The third parameter (i.e. the wParam of
the message) is a Boolean flag that indicates whether redrawing should be on or off. This
solves the second problem.
After setting the redraw flag to off, we send a command to the control to erase all items it may
have. This solves our first problem.
Here is the output of the program after we press the "Scan" menu item.

A lot of code is necessary to make this work, but thankfully it is not our code but window's.
The window resizes, redraws, etc., without any code from us.
168 C Tutorial

2.12.7 Filling the status bar

Our task consisted in drawing the tree, but also of displaying some useful information about a
window in the status bar when the user clicks on a tree item.
First, we have to figure out how we can get notified when the user clicks in an item.
The tree control (as many other controls) sends notifications through its WM_NOTIFY mes-
sage. We add a snippet of code to our MainWndProc procedure:
case WM_CREATE:
hwndTree = CreateTree(hwnd,IDTREEWINDOW);
break;
case WM_NOTIFY:
return HandleWmNotify(hwnd,wParam,lParam);
The function HandleWmNotify looks as follows:
LRESULT HandleWmNotify(HWND hwnd, WPARAM wParam, LPARAM lParam)
{
NMHDR *nmhdr;
TV_HITTESTINFO testInfo;
HWND hTree = GetDlgItem(hwnd,IDTREEWINDOW);
HTREEITEM hti;
HWND hwndStart;

nmhdr = (NMHDR *)lParam;

switch (nmhdr->code) {
case NM_CLICK:
memset(&testInfo,0,sizeof(TV_HITTESTINFO));
GetCursorPos(&testInfo.pt);
MapWindowPoints(HWND_DESKTOP,hTree,&testInfo.pt,1);
hti = TreeView_HitTest(hTree,&testInfo);
if (hti == (HTREEITEM)0) break;
hwndStart = GetTreeItemInfo(hTree,hti);
SetTextInStatusBar(hwnd,hwndStart);
break;
}
return DefWindowProc(hwnd,WM_NOTIFY,wParam,lParam);
}
We just handle the NM_CLICK special case of all the possible notifications that this very com-
plex control can send. We use the NMHDR part of the message information that is passed to us
with this message in the lParam message parameter.
Our purpose here is to first know if the user has clicked in an item, or somewhere in the back-
ground of the tree control. We should only answer when there is actually an item under the
coordinates where the user has clicked. The algorithm then, is like this:
Get the mouse position. Since windows has just sent a click message, the speed of current
machines is largely enough to be sure that the mouse hasn't moved at all between the time that
windows sent the message and the time we process it. Besides, when the user is clicking it is
surely not moving the mouse at super-sonic speeds.
Map the coordinates we received into the coordinates of the tree window.
Ask the tree control if there is an item under this coordinates.
If there is none we stop
Now, we have a tree item. We need to know which window is associated with this item, so that
we can query the window for more information. Since we have left in each item the window
handle it is displaying, we retrieve this information. We hide the details of how we do this in a
subroutine "GetTreeItemInfo", that returns us the window handle.
 A more complex example: a "clone" of spy.exe 169

Using that window handle we call another function that will display the info in the status bar.
We pass all messages to the default window procedure. this is a non-intrusive approach. The
tree control could use our notifications for something. We just need to do an action when this
event happens, but we want to disturb as little as possible the whole environment.

2.12.8 Auxiliary procedures

To retrieve our window handle from a tree item, we do the following:
static HWND GetTreeItemInfo(HWND hwndTree,HTREEITEM hti)
{
TV_ITEM tvi;

memset(&tvi,0,sizeof(TV_ITEM));
tvi.mask = TVIF_PARAM;
tvi.hItem = hti;
TreeView_GetItem(hwndTree,&tvi);
return (HWND) tvi.lParam;
}
As you can see, it is just a matter of filling a structure and querying the control for the item. we
are interested only in the PARAM part of the item.
More complicated is the procedure for querying the window for information. Here is a simple
approach:
void SetTextInStatusBar(HWND hParent,HWND hwnd)
{
RECT rc;
HANDLE pid;
char info[4096],*pProcessName;

GetWindowRect(hwnd,&rc);
GetWindowThreadProcessId(hwnd,&pid);
pProcessName = PrintProcessNameAndID((ULONG)pid);
wsprintf(info,
"Handle: 0x%x %s, left %d, top %d, right %d, bottom %d,
height %d, width %d, Process: %s",
hwnd,
IsWindowVisible(hwnd)? "Visible" : "Hidden",
rc.left,rc.top,rc.right,rc.bottom,
rc.bottom-rc.top,
rc.right-rc.left,
pProcessName);
UpdateStatusBar(info, 0, 0);
}
The algorithm here is as follows:
Query the window rectangle (in screen coordinates).
We get the process ID associated with this window
We call a subroutine for putting the name of the process executable file given its process ID.
We format everything into a buffer
We call UpdateStatusBar, generated by wedit, with this character string we have built.
The procedure for finding the executable name beginning with a process ID is quite advanced,
and here we just give it like that.
static char * PrintProcessNameAndID( DWORD processID )
{
170 C Tutorial

static char szProcessName[MAX_PATH];

HMODULE hMod;
DWORD cbNeeded;
HANDLE hProcess = OpenProcess( PROCESS_QUERY_INFORMATION |
PROCESS_VM_READ,
FALSE, processID );
szProcessName[0] = 0;
if ( hProcess ) {
if ( EnumProcessModules( hProcess, &hMod, sizeof(hMod),
&cbNeeded) ) {
GetModuleBaseName( hProcess, hMod, szProcessName,
sizeof(szProcessName) );
}
CloseHandle( hProcess );
}
return szProcessName;
}
Note that you should add the library PSAPI.LIB to the linker command line. You should do
this in the linker tab in the configuration of wedit:
 A more complex example: a "clone" of spy.exe 171

And now we are done. Each time you click in an item window, the program will display the
associated information in the status bar:

Summary: There are many things that could be improved in this small program. For instance,
it could be useful to have a right mouse menu, or a dialog box with much more information
etc. This is just a blueprint to get you started however.
The whole code for this program is in the appendix 4.
172 C Tutorial

2.13 Numerical calculations in C.

Well, we have a beautiful window with nothing in it. Blank. It would look better if we would
draw something in it isn’t it? By the way, this is an introduction to C, not to windows…
What can we draw?
Let’s draw a galaxy. In his wonderful book “Computers Pattern Chaos and Beauty”, Clifford
A. Pickover98 writes:
We will approximate a galaxy viewed from above with logarithmic spirals. They are eas-
ily programmable in a computer, representing their stars with just dots. One arm is 180
degrees out of phase with the other. To obtain a picture of a galactic distribution of dots,
simply plot dots at (r,θ ) according to:
r1 = e[θ tan φ]

r2 = e[(π+θ) tan φ]

where r1 and r2 correspond to the intertwined spiral arms. The curvature of the galactic
arms is controlled by φ which should be about 0.2 radians for realistic results. In addi-
tion, 0 < θ < 1000 radians. For greater realism, a small amount of random jitter may be
added to the final points.
He is kind enough to provide us with a formal description of the program in some computer
language similar to BASIC. Here it is:
Algorithm: How to produce a galaxy.

Notes: The program produces a double logarithmic spiral. The purpose of the random number
generator is to add jitter to the distribution of stars.

Variables:
in = curvature of galactic arm (try in = 2)
maxit = maximum iteration number
scale = radial multiplicative scale factor
cut = radial cutoff
f = final cutoff

Code:
loop1: Do i = 0 to maxit;
theta = float(i)/50;
r = scale*exp(theta*tan(in));
if r > cut then leave loop1;
x = r * cos(theta)+50;
y = r * sin(theta)+50;
call rand(randx);
call rand(randy);
PlotDotAt(x+f*randx,y+f*randy);
end
loop2: Do i = 0 to maxit;

98. The resource editor has several editors specialized for each kind of resource. You get a dialog box
editor, a menu editor, a string table editor, an accelerators editor, and an image editor. Each one is called
automatically when clicking in a resource from the menu, obtained with the dir button.
 Numerical calculations in C. 173

theta = float(i)/50;
theta2 = (float(i)/50)-3.14;
r = scale*exp(theta2*tan(in));
if r > cut then leave loop2;
x = r * cos(theta)+50;
y = r*sin(theta)+50;
call rand(randx);
call rand(randy);
PlotDotAt(x+f*randx,y+f*randy);
end
This are quite clear specs. Much clearer than other “specs” you will find in your future career
as programmer… So let’s translate this into C. We can start with the following function:
void DrawGalaxy(HDC hDC,double in,
int maxit,double scale,
double cut, double f)
{
double theta, theta2, r, x, y, randx, randy;
for (int i = 0; i <= maxit; i++) { (1)
theta = ((double)i)/CENTER; (2)
r = scale*exp(theta*tan(in)); (3)
if (r > cut) break; (4)
x = r * cos(theta)+CENTER; (5)
y = r * sin(theta)+CENTER; (6)
randx = (double)rand() / (double)RAND_MAX; (7)
randy = (double)rand() / (double)RAND_MAX; (8)
PlotDotAt(hDC,x+f*randx,y+f*randy,RGB(0,0,0));
}
for (int i = 0; i <= maxit; i++) {
theta = ((double)i)/CENTER;
theta2 = ( ((double)i)/CENTER) –3.14;
r = scale * exp(theta2*tan(in)); (9)
if (r > cut) break;
x = r*cos(theta)+CENTER;
y = r*sin(theta)+CENTER;
randx = (double)rand() / (double) RAND_MAX;
randy = (double)rand() / (double) RAND_MAX;
PlotDotAt(hDC,x+f*randx,y+f*randy,RGB(255,0,0)); (10)
}
}
We translate both loops into two for statements. The exit from those loops before they are fin-
ished is done with the help of a break statement. This avoids the necessity of naming loops
when we want to break out from them, what could be quite fastidious in the long term…
I suppose that in the language the author is using, loops go until the variable is equal to the
number of iterations. Maybe this should be replaced by a strictly smaller than… but I do not
think a point more will do any difference.
Note the cast of i (double)i. Note too that I always write 50.0 instead of 50 to avoid
unnecessary conversions from the integer 50 to the floating-point number 50.0. This cast is not
necessary at all, and is there just for “documentation” purposes. All integers when used in a
double precision expression will be automatically converted to double precision by the com-
piler, even if there is no cast.
The functions exp and tan are declared in math.h. Note that it is imperative to include math.h
when you compile this. If you don’t, those functions will be assumed to be external functions
that return an int, the default. this will make the compiler generate code to read an integer
instead of reading a double, what will result in completely nonsensical results.
174 C Tutorial

A break statement “breaks” the loop.

This statement means
r = (r*cos(theta)) + 5 and NOT
r = r * (cos(theta)+CENTER;
In line 8 we use the rand() function. This function will return a random number between
zero and RAND_MAX. The value of RAND_MAX is defined in stdlib.h. If we want to obtain a
random number between zero and 1, we just divide the result of rand() by RAND_MAX. Note
that the random number generator must be initialized by the program before calling rand() for
the first time. We do this in WinMain by calling
srand((unsigned)time(NULL));
This seeds the random number generator with the current time.
We are missing several pieces. First of all, note that CENTER is
#define CENTER 400
because with my screen proportions in my machine this is convenient. Note that this shouldn’t
be a #define but actually a calculated variable. Windows allows us to query the horizontal and
vertical screen dimensions, but… for a simple drawing of a spiral a #define will do.
The function PlotPixelAt looks like this:
void PlotDotAt(HDC hdc,double x,double y,COLORREF rgb)
{
SetPixel(hdc,(int)x,(int)y,rgb);
}
The first argument is an “HDC”, an opaque pointer that points to a “device context”, not fur-
ther described in the windows documentation. We will speak about opaque data structures
later. A COLORREF is a triple of red, green, and blue values between zero (black) and 255
(white) that describe the colors of the point. We use a simple schema for debugging purposes:
we paint the first arm black (0,0,0) and the second red (255,0,0).
In event oriented programming, the question is “which event will provoke the execution of
this code”?
Windows sends the message WM_PAINT to a window when it needs repainting, either
because its has been created and it is blank, or it has been resized, or when another window
moved and uncovered a portion of the window. We go to out MainWndProc function and add
code to handle the message. We add:
case WM_PAINT:
dopaint(hwnd);
break;
We handle the paint message in its own function. This avoids an excessive growth of the
MainWndProc function. Here it is:
void dopaint(HWND hwnd)
{
PAINTSTRUCT ps;
HDC hDC;
hDC = BeginPaint(hwnd,&ps);
DrawGalaxy(hDC,3.0,20000,2500.0,4000.0,18.1);
EndPaint(hwnd,&ps);
}
We call the windows API BeginPaint, passing it the address of a PAINTSTRUCT, a struc-
ture filled by that function that contains more information about the region that is to be
painted, etc. We do not use it the information in it, because for simplicity we will repaint the
 Numerical calculations in C. 175

whole window each time we receive the message, even if we could do better and just repaint
the rectangle that windows passes to us in that parameter. Then, we call the code to draw our
galaxy, and inform windows that we are done with painting.
Well, this finishes the coding. We need to add the
#include <math.h>
#include <time.h>
at the beginning of the file, since we use functions of the math library and the time() function
to seed the srand() function.
We compile and we obtain:

It would look better, if we make a better background, and draw more realistic arms, but for a
start this is enough.
176 C Tutorial

There are many functions for drawing under windows of course. Here is a table that provides
short descriptions of the most useful ones:

Function Purpose

AngleArc Draws a line segment and an arc.

Draws an elliptical arc using the currently selected pen.

Arc
You specify the bounding rectangle for the arc.

ArcTo is similar to the Arc function, except that the

ArcTo
current position is updated.

Returns the current arc direction for the specified device

GetArcDirection
context. Arc and rectangle functions use the arc direction.

Draws a line from the current position up to, but not

LineTo
including, the specified point.

Updates the current position to the specified point and

MoveToEx
optionally returns the previous position.

PolyBezier Draws one or more Bézier curves.

PolyBezierTo Same as PolyBézier but updates the current position.

PolyDraw Draws a set of line segments and Bézier curves.

Draws a series of line segments by connecting the points

PolyLine
in the specified array.

Updates current position after doing the same as

PolyLineTo
PolyLine.

PolyPolyLine Draws multiple series of connected line segments.

Sets the drawing direction to be used for arc and rectangle

SetArcDirection
functions.

There are many other functions for setting color, working with rectangles, drawing text
(TextOut), etc. Explaining all that is not the point here, and you are invited to read the docu-
mentation.

Summary: C converts integer and other numbers to double precision when used in a double
precision expression. This will be done too when an argument is passed to a function.
When the function expects a double and you pass it an int or even a char, it will be con-
verted to double precision by the compiler.
All functions that return a double result must declare their prototype to the compiler so that
the right code will be generated for them. An unprototyped function returning a double will
surely result in incorrect results!
Opaque data structures are hidden from client code (code that uses them) by providing just
a void pointer to them. This way, the client code is not bound to the internal form of the
 Filling the blanks 177

structure and the designers of the system can modify it without affecting any code that uses
them. Most of the windows data structures are used this way: an opaque “HANDLE” is
given that discloses none of the internals of the object it is pointing to.

2.14 Filling the blanks

Input goes through dialog boxes under windows. They are ubiquitous; so let’s start to fill our
skeleton with some flesh. Let’s suppose, for the sake of the example that we want to develop a
simple text editor. It should read some text, draw it in the screen, and provide some utilities
like search/replace, etc.
First, we edit our menu, and add an “edit” item. We open the directory window, and select the
menu:

We arrive at the menu editor99. If we open each branch of the tree in its left side, it looks like
this:

99. This type of interface requires an action from the part of the user to indicate when it is finished
modifying the name and desires to “apply” the changes. Another possibility would be that the resource editor
applies the changes letter by letter as the user types them in, as some other editors do. This has the advantage
of being simpler to use, but the disadvantage of being harder to program and debug. As always, an the
appearance of the user interface is not only dictated by the user comfort, but also by the programming effort
necessary to implement it. You will see this shortly when you are faced with similar decisions.
178 C Tutorial

We have at the left side the tree representing our menu. Each submenu is a branch, and the
items in the branch; the leaves are the items of the submenu. We select the “File” submenu and
press the “insert” key. We obtain a display like this:

A new item is inserted after the currently selected one. The name is “Popup”, and the first item
is “New item”. We can edit those texts in the window at the right: We can change the symbolic
name, and set/unset several options. When we are finished, we press “Apply” to write our
changes to the resource. 100
OK, we change the default names to the traditional “Edit” and “Search”, to obtain this display:

We will name the new item IDM_SEARCH. I am used to name all those constants starting with
IDM_ from ID Menu, to separate them in my mind from IDD_ (ID Dialog).
We can now start drawing the “Search” dialog. Just a simple one: a text to search, and some
buttons to indicating case sensitivity, etc. We close the menu editor, and we start a new dialog.

100. A debugger is a program that starts another program, the “program to be debugged” or “debuggee”,
and can execute it under the control of the user, that directs the controlled execution. All C development
systems offer some debugger, and lcc-win32 is no exception. The debugger is described in more detail in the
user’s manual, and it will not be described here. Suffice to note that you start it with F5 (or Debugger in the
compiler menu), you can single step at the same level with F4 and trace with F8. The debugger shows you in
yellow the line the program will execute next, and marks breakpoints with a special symbol at the left. Other
debuggers may differ from this of course, but the basic operations of all of them are quite similar. Note that
lcc-win32 is binary compatible with the debugger of Microsoft: you can debug your programs using that
debugger too.
To be able to use the debugger you need to compile with the g2 flag on. That flag is normally set by
default. It directs the compiler to generate information for the debugger, to enable it to show source lines and
variable values. The compiler generates a whole description of each module and the structures it uses called
“debug information”. This information is processed by the linker and written to the executable file. If you turn
the debugging flag off the debugger will not work. The best approach is to leave this flag on at all times.
Obviously the executable size will be bigger, since the information uses up space on disk. If you do not want
it, you can instruct the linker to ignore it at link time. In this way, just switching that linker flag on again will
allow you to debug the program.
The debug information generated by lcc-win32 uses the NB09 standard as published by Microsoft and
Intel. This means that the programs compiled with lcc-win32 can be debugged using another debugger that
understands how to use this standard.
 Filling the blanks 179

In the “Resources” submenu, we find a “New” item, with several options in it. We choose the
“dialog” option.
The dialog editor builds an empty dialog and we are shown the following parameters dialog:

Even if this quite overwhelming, we are only interested in two things: the title of the dialog
and the symbolic identifier. We leave all other things in their default state. We name the dialog
IDD_SEARCH, and we give it the title “Text search”. After editing it looks like this:

We press the OK button, and we do what we did with the dialog in the DLL, our first example.
The finished dialog should look roughly like this:

An edit field, and two push button for OK and Cancel. The edit field should receive the ID
IDTEXT.
Now comes the interesting part. How to connect all this?
We have to first handle the WM_COMMAND message, so that our main window handles the
menu message when this menu item is pressed. We go to our window procedure MainWnd-
Proc. Here it is:
180 C Tutorial

LRESULT CALLBACK MainWndProc(HWND hwnd,UINT msg,WPARAM wParam,LPARAM

lParam)
{
switch (msg) {
case WM_SIZE:
SendMessage(hWndStatusbar,msg,wParam,lParam);
InitializeStatusBar(hWndStatusbar,1);
break;
case WM_MENUSELECT:
return MsgMenuSelect(hwnd,msg,wParam,lParam);
case WM_COMMAND:
HANDLE_WM_COMMAND(hwnd,wParam,lParam,MainWndProc_OnCommand);
break;
case WM_DESTROY:
PostQuitMessage(0);
break;
default:
return DefWindowProc(hwnd,msg,wParam,lParam);
}
return 0;
}
We can see that it handles already quite a few messages. In order,
We see that when the main window is resized, it resizes its status bar automatically.
When the user is going through the items of our menu, this window receives the
WM_MENUSELECT message from the system. We show the appropriate text with the explana-
tions of the actions the menu item in the status bar.
When a command (from the menu or from a child window) is received, the parameters are
passed to a macro defined in windowsx.h that breaks up the wParam and lParam parameters
into their respective parts, and passes those to the MainWndProc_OnCommand function.
When this window is destroyed, we post the quit message.
The function for handling the commands looks like this:
void MainWndProc_OnCommand(HWND hwnd,
int id, HWND hwndCtl, UINT codeNotify)
{
switch(id) {
// ---TODO--- Add new menu commands here
case IDM_EXIT:
PostMessage(hwnd,WM_CLOSE,0,0);
break;
}
}
We find a comment as to where we should add our new command. We gave our menu item
“Search” the symbolic ID of IDM_SEARCH. We modify this procedure like this:
void MainWndProc_OnCommand(HWND hwnd, int id, HWND hwndCtl, UINT
codeNotify)
{
switch(id) {
// ---TODO--- Add new menu commands here
case IDM_SEARCH:
{
char text[1024];
if (CallDialog(IDD_SEARCH,SearchDlgProc,
(LPARAM)text))
DoSearchText(text);
}
 Filling the blanks 181

break;
case IDM_EXIT:
PostMessage(hwnd, WM_CLOSE,0,0);
break;
}
}
When we receive the menu message then, we call our dialog. Since probably we will make
several dialogs in our text editor, it is better to encapsulate the difficulties of calling it within
an own procedure: CallDialog. This procedure receives the numeric identifier of the dialog
resource, the function that will handle the messages for the dialog, and an extra parameter for
the dialog, where it should put the results. We assume that the dialog will return TRUE if the
user pressed OK, FALSE if the user pressed the Cancel button.
If the user pressed OK, we search the text within the text that the editor has loaded in the func-
tion DoSearch.
How will our function CallDialog look like?
Here it is:
int CallDialog(int id,DLGPROC proc,LPARAM parameter)
{
int r = DialogBoxParam(hInst,MAKEINTRESOURCE(id),
hwndMain, proc, parameter);
return r;
}
We could have returned the value of the DialogBoxParam API immediately but I like storing
function return values in local variables. You never know what can happen, and those values
are easily read in the debugger.
We have to write a dialog function, much like the one we wrote for our string DLL above. We
write a rough skeleton, and leave the details for later:
BOOL CALLBACK SearchDlgProc(HWND hwnd, UINT msg, WPARAM wParam, LPARAM
lParam)
{
switch (msg) {
case WM_INITDIALOG:
return TRUE;
case WM_CLOSE:
EndDialog(hwnd,0);
break;
}
return FALSE;
}
This does nothing, it doesn’t even put the text in the received parameter, but what we are inter-
ested in here, is to first ensure the dialog box shows itself. Later we will refine it. I develop
software like this, as you may have noticed: I try to get a working model first, a first approxi-
mation. Then I add more things to the basic design. Here we aren’t so concerned about design
anyway, since all this procedures are very similar to each other.
The other procedure that we need, DoSearchText, is handled similarly:
int DoSearchText(char *txt)
{
MessageBox(NULL,"Text to search:",txt, MB_OK );
return 1;
}
We just show the text to search. Not very interesting but…
182 C Tutorial

We compile, link, the main application window appears, we select the “Search” menu, and…
we see:

What’s wrong?????
Well, we have inverted the parameters to the MessageBox procedure, but that’s surely not
the point. Why is the dammed dialog box not showing?
101
Well, here we need a debugger. We need to know what is happening when we call the dia-
log box. We press F5, and we start a debugging session. The debugger stops at WinMain.

Now, wait a minute, our window procedure that receives the message from the system is
called indirectly from Windows, we can’t just follow the program blindly. If we did that, we
would end up in the main loop, wasting our time.

101.Why didn’t we use the DLL to ask for the string to search? Mostly because I wanted to give you an
overview of the whole process. A good exercise would be to change the program to use the DLL.
Which changes would be necessary? How would you link?
 Filling the blanks 183

No, we have to set a breakpoint there. We set a breakpoint when we call the dialog using the
F2 accelerator. We see that Wedit sets a sign at the left to indicate us that there is a breakpoint
there. Then we press F5 again to start running the program.

Our program starts running, we go to the menu, select the search item, and Wedit springs into
view. We hit the breakpoint. Well that means at least that we have correctly done things until
here: the message is being received. We enter into the CallDialog procedure using the F8
accelerator. We step, and after going through the DialogBoxParam procedure we see no dialog
and the return result is –1. The debugger display looks like this:
We see the current line highlighted in yellow, and in the lower part we see the values of some
variables. Some are relevant some are not. Luckily the debugger picks up r as the first one. Its
value is –1.
184 C Tutorial

Why –1?

A quick look at the doc of DialogBoxParam tells us “If the function fails, the return value
is -1.”
Ahh, how clear. Yes of course, it failed. But why?
Mystery. There are no error codes other than just general failure. What could be wrong?
Normally, this –1 means that the resource indicated by the integer code could not be found. I
have learned that the hard way, and I am writing this tutorial for you so that you learn it the
easy way. The most common cause of this is that you forgot to correctly give a symbolic name
to your dialog.
We close the debugger, and return to the resource editor. There we open the dialog box proper-
ties dialog box (by double clicking in the dialog box title bar) and we see… that we forgot to
change the name of the dialog to IDM_SEARCH!!! We correct that and we try again.
 Filling the blanks 185

OK, this looks better. The dialog is showing.

The rest is quite trivial most of the wok was done when building the DLL. Actually, the dialog
box is exactly the same.102

102. Newer versions of Wedit check for this. Older ones aren’t so sophisticated so please take care.
186 C Tutorial

2.15 Using the graphical code generator

As we saw before, writing all that code for a simple dialog is quite a lot of work. It is impor-
tant that you know what is going on, however. But now we can see how we can make Wedit
generate the code for us.
The code generator is not enabled by default. You have to do it by choosing the “Output” item
in the resources menu. This leads to an impressing looking box like this:

This dialog shows you the files that will be generated or used by Wedit. Those that will be
generated have a box at the right, to enable or disable their generation. The others are used if
available, but only for reading. The last item is an additional path to look for bitmaps, icons
and other stuff that goes into resources.
You notice that the first item is disabled. You enable it, and type the full path of a file where
you want Wedit to write the dialog procedures. Notice that by default, the name of this file is
<name of the project>.c. This could provoke that your winexample.c that we worked so hard
to write, could be overwritten.103 Choose another name like “dialogs.c” for instance.

103. Look in the weditreslib folder. The file commmsg.c contains the default procedure and all the
machinery necessary for the run time of the framework.
 Using the graphical code generator 187

Now, when you save your work, all the dialog procedures will be written for you. But before,
we have to tell the framework several things.
The first will be the prefix to use for all the procedures that will be used in this dialog box. We
define this in the main properties dialog box obtained when you double click in the dialog title.

“Dlg400” is an automatic generated name, not very convincing. We can put in there a more
meaningful name like “DlgSearch” for instance. We will see shortly where this name is used.
What we want to do now is to specify to the editor where are the events that interest us. For
each of those events we will specify a callback procedure that the code generated by the editor
will call when the event arrives. Basically all frameworks, no matter how sophisticated, boil
down to that: a quick way of specifying where are the events that you want to attach some
code to.
The events we can choose from are in the properties dialog boxes, associated with each ele-
ment of the dialog box. You have the general settings in the properties window associated with
the dialog box (that appears when you double click in the title), and you have the buttons prop-
erties that appear when you right click in a button.
Those dialog boxes have generally a standard or «Properties» part that allows you to change
things like the element’s text, or other simple properties, and a part that is visible only when
you have selected the C code generation. That part is normally labeled “events” and appears at
the right. That label tells us which kind of events the framework supports: window messages.
There are many events that can possibly happen of course, but the framework handles only
those.

We see at the right tab a typical “messages” part: We have some buttons to choose the mes-
sages we want to handle, the function name prefix we want for all callback procedures for this
element, and other things like the font we want to use when the elements are displayed.
We see again that “Dlg400”… but this allows us to explain how those names are generated
actually. The names of the generated functions are built from the prefix of the dialog box, then
the prefix for the element, and then a fixed part that corresponds to the event name. We edit
the prefix again, following this convention.

The “Selected” message is on, so the framework will generate a call to a function called Dlg-
SearchOnOkSeelected().Happily for us, we do not have to type those names our-
selves.
Without changing anything else we close the button properties and save our work. We open
the c source file generated by the editor.
188 C Tutorial

We obtain the following text:

/* Wedit Res Info */
#ifndef __windows_h
#include <windows.h>
#endif
#include "winexampleres.h"

BOOL APIENTRY DlgSearch(HWND hwnd,UINT msg,WPARAM wParam,LPARAM lParam)

{
static WEDITDLGPARAMS WeditDlgParams;

switch(msg)
{
case WM_INITDIALOG:
SetWindowLong(hwnd,DWL_USER,
(DWORD)&WeditDlgParams);
DlgSearchInit(hwnd,wParam,lParam);
/* store the input arguments if any */
SetProp(hwnd,"InputArgument",(HANDLE)lParam);
break;
case WM_COMMAND:
switch (LOWORD(wParam))
{
case IDOK:
DlgSearchOnOKSelected(hwnd);
break;
}
break;
}
return(HandleDefaultMessages(hwnd,msg,wParam,lParam));
}
We have here a normal callback procedure for our dialog. It handles two messages:
WM_INITDIALOG and WM_COMMAND. The callback procedures are in bold type. There are
two of them: The initialization callback called “DlgSearchInit”, and the one we
attached to the OK button above, “DlgSearchOnOkSelected”.
There are more things in there, but for the time being we are interested in just those ones,
because they have to be written by you!
What you want to do when the dialog box has been created but it is not yet visible?
This is the purpose of the first callback. In our text search dialog we could write in the edit
field the last searched word, for instance, to avoid retyping. Or we could fill a combo box with
all the words the user has searched since the application started, or whatever. Important is that
you remember that in that function all initializations for this dialog box should be done,
including memory allocation, populating list boxes, checking check buttons or other chores.
The second callback will be called when the user presses the OK button. What do you want to
do when this event happens? In our example we would of course start the search, or setup a set
of results so that the procedure that called us will know the text to search for, and possibly
other data.
Different controls will react to different events. You may want to handle some of them. For
instance you may want to handle the event when the user presses a key in an edit field, to
check input. You can use the framework to generate the code that will trap that event for you,
and concentrate in a procedure that handles that event.
How would you do this?
 Using the graphical code generator 189

You open the properties dialog of the edit control and check the “Update” message. This will
prompt the editor to generate a call to a function of yours that will handle the event. The
details are described in the documentation of the resource editor and will not be repeated here.
What is important to retain are the general principles at work here. The rest is a matter of read-
ing the docs, to find out which events you can handle for each object in the dialog, and writing
the called functions with the help of the windows documentation.
But what happens if there is an event that is not foreseen by the framework? With most frame-
works this is quite a problem, happily not here. You have just to check the button “All” in the
dialog properties and the framework will generate a call to a default procedure (named <pre-
fix>Default) at each message. There you can handle all events that the operating system sends.
I have tried to keep the framework open so that unforeseen events can be still treated correctly.
Another way to customize the framework is to modify the default procedure provided in the
library weditres.lib. The source of that procedure is distributed in the source distributions of
lcc-win32104 and is relatively easy to modify.

104. To find that easily just press F12 and click in its name in the function list.
190 C Tutorial

2.16 Understanding the wizard generated sample code

OK. Go to new, make a new project, and use all the default options for an application with a
single window and a status bar. Let’s see how you modify that to make it the start of your pro-
gram.

2.16.1 Making a new menu or modifying the given menu.

Add your item with the resource editor, and give it an identifier, normally written in uppercase
like: IDMENU_ASK_PARAMETERS, or similar. This is surely not the text the user will see,
but a symbolic name for an integer constant, that windows will send to the program when the
user presses this option. We can then, continue our beloved programming traditions.
Once that is done, and your menu displays correctly, go to the MainWndProc function.105
There, you will see a big switch with different code to handle the events we are interested in.
The menu sends a command event, called WM_COMMAND. There you see that this event will be
handled by the HANDLE_COMMAND macro. It is just a shorthand to break down the 64 bits
that windows sends us in smaller pieces, disassembling the message information into its con-
stituent parts. This macro ends calling the MainWndProc_OnCommand function that is a bit
higher in the text. There, you will find a switch with the comment:
//---TODO--- Insert new commands here.
Well, do exactly that, and add (as a new case in the switch) your new identifier
IDMENU_ASK_PARAMETERS. There you can do whatever you want to do when that menu
item is clicked.

2.16.2 Adding a dialog box.

Draw the dialog box with controls and all that in the resource editor, and then open it as a
result of a menu option, for instance. You would use DialogBox, that handy primitive
explained in detail in the docs to fire up your dialog.106 You have to write the procedure to
handle the dialog box’s messages first. You can start the dialog box in a non-modal mode with
the CreateDialog API.
To make this a bit more explicit, let’s imagine you have defined your dialog under the name of
IDD_ASK_PARAMS in the resource editor.107 You add a menu item corresponding to the dia-
log in the menu editor, one that will return IDM_PARAMETERS, say. You add then in the func-
tion MainWndProc_OnCommand code like this:
case IDM_PARAMETERS:
r = DialogBox(hInst,
MAKEINTRESOURCE( IDD_ASK_PARAMS),
ghwndMain,ParamsDlgProc);
break;

105. Go to “help”, then click in Win32 API, get to the index and write the name of that function.
106. Again, this is the #defined identifier of the dialog, not the dialog’s title!
107. The registry has been criticized because it represents a single point of failure for the whole system.
That is obviously true, but it provides as a redeeming value, a standard way of storing and retrieving
configuration data and options. It allows your application to use the same interface for storing this data,
instead of having to devise a schema of files for each application. The software is greatly simplified by this,
even if it is risky, as a general principle.
 Understanding the wizard generated sample code 191

You give to that API the instance handle of the application, the numerical ID of the dialog
enclosed in the MAKEINTRESOURCE macro, the handle of the parent window, and the
name of the procedure that handles the messages for the dialog.

2.16.3 Drawing the window

You have to answer to the WM_PAINT message. See the documentation for a longer descrip-
tion. This will provoke drawing when the window needs repainting only. You can force a
redraw if you use the InvalidateRect API.
You add code like this:
case WM_PAINT:
PAINTSTRUCT ps;
HDC hDC = BeginPaint(hwnd,&ps);
// Code for painting using the HDC goes here
EndPaint(hwnd,&ps);
break;
You use the API BeginPaint to inform windows that you are going to update the window. Win-
dows gives you information about the invalid rectangles of the window in the PAINTSTRUCT
area. You pass to windows the address of such an area, and the handle to your window. The
result of BeginPaint is an HDC, a Handle to a Device Context, that is required by most
drawing primitives like TextOut, LineTo, etc. When you are finished, you call EndPaint,
to inform windows that this window is updated.
To draw text you use TextOut, or DrawText. Note that under windows there is no auto-
matic scrolling. You have to program that yourself or use a multi-line edit control.

2.16.4 Initializing or cleaning up

You can write your initialization code when handling the WM_CREATE message. This mes-
sage is sent only once, when the window is created. To cleanup, you can rely on the
WM_CLOSE message, or better, the WM_DESTROY message. Those will be sent when the
window is closed/destroyed. Note that you are not forced to close the window when you
receive the WM_CLOSE message. Even if this is not recommended, you can handle this mes-
sage and avoid passing it to the DefWndProc procedure. In this case the window is not
destroyed. Another thing is when you receive the WM_DESTROY message. There, you are just
being informed that your window is going to be destroyed anyway.

2.16.5 Getting mouse input.

You can handle the WM_LBUTTONDOWN, or WM_RBUTTONDOWN messages. To follow the
mouse you handle the WM_MOUSEMOVE messages. In the information passed with those mes-
sage parameters you have the exact position of the mouse in pixel coordinates.

2.16.6 Getting keyboard input

Handle the WM_KEYDOWN message or the WM_CHAR message. Windows allows for a fine-
grained control of the keyboard, with specific messages when a key is pressed, released, repeat
counts, and all the keyboard has to offer.

2.16.7 Handling moving/resizing

You get WM_MOVE when your window has been moved, WM_SIZE when your window has
been resized. In the parameters of those messages you find the new positions or new size of
192 C Tutorial

your window. You may want to handle those events when you have child windows that you
should move with your main window, or other measures to take, depending on your applica-
tion.

2.16.8 Getting notified when the mouse passes over your

window
You get the WM_MOUSEMOVE message from windows. In the wParam parameter you get a
flag telling you if the control key is pressed, and other parameters. In the lParam parameter
you get the coordinates of the cursor.

2.16.9 Creating additional controls in your window without

using a dialog box
You use the CreateWindow API with a predefined window class. You pass it as the parent-
window parameter the handle of the window where you want to put the control on. For
instance if you want to create somewhere in your window an edit field, you would write:
hEditWnd = CreateWindow("EDIT", // window class is “edit”
NULL,
WS_CHILD | WS_VISIBLE |
ES_MULTILINE |
WS_VSCROLL | WS_HSCROLL |
ES_AUTOHSCROLL | ES_AUTOVSCROLL,
0,
0,
(Rect.right - Rect.left), //Assume a rectangle structure
(Rect.bottom - Rect.top), // contains this coordinates
hParentWnd,
(HMENU) 14376,
hInst, // Application instance
NULL);
 Customizing controls 193

2.17 Customizing controls

Under windows you have a set of building blocks at your disposal that is, even it is quite rich,
fixed. A plain listbox will look like a listbox and there is no obvious way to modify the
appearence of it, not even for doing such simple things like writing the text in red, or painting
the background in brown.
The reason for this is simple: there are infinetly many ways of customizing a control, and if
Windows would provide an API for each one, the windows API would not fit in a computer no
matter what.
The solution provided by the system is a general interface that gives you enough elements to
be able to do whatever you want to do. You can:

1) customize the appearence of a control by making some of its parts “owner draw”, i.e.
drawed by yourself.

2) Process some special windows messages that allow you to change the font, the background
color, and other characteristics by modifying the settings of a passed device context (HDC).
In both cases the interface is quite simple: Windows sends a message to the procedure of the
parent window of the control, to ask it to draw a specific part of the control. Some (or all) of
the control behaviour is retained, but the appearence can be changed as you like.

2.17.1 Processing the WM_CTLCOLORXXX message

We can start with a simple example that draws a listbox in a special way. Let’s use the second
method first, since it is easier.
Suppose we want to display the text in a listbox in red color. In the dialog procedure which
owns the list box (or in the windows procedure if it is a list box created independently) you
receive the WM_CTLCOLORLISTBOX message. In the wparam parameter you receive an
HDC that you can customize using SetTextColor to draw the text with the color you wish. To
inform windows that you have processed this message, you should return a handle to a brush
that will be used to paint the background of the listbox. Notice that windows does not destroy
this object. You should destroy it when the dialog box finishes.
Creating and destroying the brush that you should return is cumbersome. If you are not inter-
ested in modifying the background, you can use the API GetStockObject() returning the han-
dle of the white brush like this:
return GetStockObject(WHITE_BRUSH);
Objects created with GetStockObejct do not need to be destroyed, so you spare yourself the
job of creating and destroying the brush. The code would look like this:
BOOL WINAPI MyDlgProc(HWND hwnd, UINT msg,WPARAM wParam,LPARAM lParam)
{
HDC hDC;
COLORREF oldcolor;
switch (msg) {
...
case WM_CTLCOLORLISTBOX:
hDC = (HDC)wParam;
color = RGB(192,0,0); // red color
SetTextColor(hDC,color);
return GetStockObject(WHITE_BRUSH);
...
}
}
194 C Tutorial

If you want to return another brush than white, you should handle the WM_INITDIALOG
message to create the brush, and the WM_NCDESTROY message to destroy the brush you
created to avoid memory leaks. In this case, the above example should be augmented with:
BOOL WINAPI MyDlgProc(HWND hwnd, UINT msg,WPARAM wParam,LPARAM lParam)
{
HDC hDC;
COLORREF oldcolor;
static HBRUSH hBrush;
switch (msg) {
case WM_INITDIALOG:
hBrush = CreateSolidBrush(BK_COLOR);
// other initializations go here
...
case WM_CTLCOLORLISTBOX:
hDC = (HDC)wParam;
color = RGB(192,0,0); // red color
SetTextColor(hDC,color);
return hBrush;
...
case WM_NCDESTROY:
DeleteObject(hBrush);
break;
}
}
We create the brush at the intialization of the list box, we use it when we receive the
WM_CTLCOLORLISTBOX message, and we destroy it when we receive the
WM_NCDESTROY message.
This technique can be used not only with list boxes but with many other controls. We have:

Message Control

WM_CTLCOLORBTN Button
WM_CTLCOLOREDIT Edit field
WM_CTLCOLORDLG Dialog box
WM_CTLCOLORLISTBOX List box
WM_CTLCOLORSCROLLBAR Scroll bar
WM_CTLCOLORSTATIC Static controls (text, etc)
WM_CTLCOLORLISTBOX + A combo box sends both messages
WM_CTLCOLOREDIT to the parent window

All this messages have the same parameters as in the list box example and the procedure is
exactly the same.
The WM_CTLCOLORSTATIC is interesting because it allows you to add text in any font and
color combination to a window or dialog by just calling SelectObject with a special font or
font size using the passed HDC of this message.
 Customizing controls 195

2.17.2 Using the WM_DRAWITEM message

Suppose we want to draw a list box that draws an image at the left side of the text in each line.
In this case, just setting some properties of the HDC we receive with
WM_CTLCOLORLISTBOX will not cut it. We need to do the entire drawing of each line of
the list box.
We can do this by setting the style of the list box to owner draw (the resource editor will do
this for you) and we need to process the WM_DRAWITEM message. Windows sends us a
pointer to a DRAWITEMSTRUCT that contains all the necessary information to do the draw-
ing in the LPARAM parameter. This structure is defined as follows:

Member Description
unsigned int CtlType This member can be one of the following values.
ODT_BUTTON button
ODT_COMBOBOX combo box
ODT_LISTBOX list box
ODT_LISTVIEW List-view control
ODT_MENU Owner-drawn menu item
ODT_STATIC static control
ODT_TAB Tab control

unsigned int CtlID Specifies the identifier of the combo box, list box, button, or static con-
trol. This member is not used for a menu item.

unsigned int itemID Specifies the menu item identifier for a menu item or the index of the
item in a list box or combo box. For an empty list box or combo box,
this member can be–1. This allows the application to draw only the
focus rectangle at the coordinates specified by the rcItem member even
though there are no items in the control

unsigned int itemAction Specifies the drawing action that windows expect us to perform. This
action is indicated with the following flags:
ODA_DRAWENTIRE The entire control needs to be drawn.
ODA_FOCUS The control has lost or gained the keyboard focus. The
itemState member should be checked to determine whether the control
has the focus.
ODA_SELECT The selection status has changed. The itemState mem-
ber should be checked to determine the new selection state.

HWND hwndItem Window handle of the control

RECT rcItem Rectangle where we should do our drawing

unsigned long *itemData Specifies the application-defined value associated with the menu item.
196 C Tutorial

Member Description
unsigned int itemState State of the item after the current drawing action takes place. This can
be a combination of the following values:
ODS_CHECKED This bit is used only in a menu.
ODS_COMBOBOXEDIT The drawing takes place in the edit control
of an owner-drawn combo box.
ODS_DEFAULT The item is the default item.
ODS_DISABLED The item is to be drawn as disabled.
ODS_FOCUS The item has the keyboard focus.
ODS_GRAYED This bit is used only in a menu.
ODS_HOTLIGHT The item is being hot-tracked, that is, the item will
be highlighted when the mouse is on the item.
ODS_INACTIVE The item is inactive and the window associated with
the menu is inactive.
ODS_NOACCEL The control is drawn without the keyboard accelera-
tor cues.
ODS_NOFOCUSRECT The control is drawn without focus indicator
cues.
ODS_SELECTED The menu item's status is selected.

With all this information, we can do whatever we want inside the rcItem rectangle. Here is the
example for our problem of drawing a bitmap in each line of the combo box.
First, in our dialog procedure we handle the message WM_DRAWITEM, of course:
BOOL WINAPI MyDlgProc(HWND hwnd, UINT msg,WPARAM wParam,LPARAM lParam)
{
HDC hDC;
COLORREF oldcolor;
switch (msg) {
case WM_INITDIALOG:
LoadListboxBitmaps(hwnd);
// other initializations can follow
...
case WM_DRAWITEM:
DrawListboxLine(hwnd,(LPDRAWITEMSTRUCT)lParam);
return 1; // We processed this message
...
case WM_NCDESTROY:
DestroyListboxBitmaps(hwnd);
// other cleanup tasks can go here
}
}
Our procedure to draw each line of the dialog box can look like this:
long DrawListboxLine(HWND hwndDlg,LPDRAWITEMSTRUCT lpdis)
{
HBITMAP hbmpPicture,hbmpOld;
int y,state;
HDC hdcMem;
TEXTMETRIC tm;
RECT rcBitmap;
COLORREF oldcolor;
char tchBuffer[200];

// If there are no list box items, skip this message.

if (lpdis->itemID == -1) return 0;
 Customizing controls 197

switch (lpdis->itemAction) {
case ODA_SELECT:
case ODA_DRAWENTIRE:
// Windows DOES NOT erase the background of owner draw list boxes.
// Erase the background to start with a coherent state.
FillRect(lpdis->hDC,&lpdis->rcItem,
GetStockObject (WHITE_BRUSH) );

// We get the handle of the bitmap to draw from the item-data slot in each
// list box line. The initialization procedure sets this
hbmpPicture =(HBITMAP)SendMessage(lpdis->hwndItem,
LB_GETITEMDATA, lpdis->itemID, (LPARAM) 0);

// We select the image in a memory DC, then we select it. This is a

// standard technique for copying a bitmap into the screen
hdcMem = CreateCompatibleDC(lpdis->hDC);
hbmpOld = SelectObject(hdcMem, hbmpPicture);
BitBlt(lpdis->hDC,
lpdis->rcItem.left, lpdis->rcItem.top,
lpdis->rcItem.right - lpdis->rcItem.left,
lpdis->rcItem.bottom - lpdis->rcItem.top,
hdcMem, 0, 0, SRCCOPY);

// We are done with our memory DC. Clean up.

SelectObject(hdcMem, hbmpOld);
DeleteDC(hdcMem);

// Now, we retrieve the text for this line.

SendMessage(lpdis->hwndItem, LB_GETTEXT,
lpdis->itemID, (LPARAM) tchBuffer);

// How big is the text of this item?

GetTextMetrics(lpdis->hDC, &tm);

// Vertically center the text

y = (lpdis->rcItem.bottom+lpdis->rcItem.top-tm.tmHeight)/2;

// Send the text to the screen 6 pixels after the bitmap

TextOut(lpdis->hDC,XBITMAP+6,y,tchBuffer,strlen(tchBuffer));

// Is the item selected?

if (lpdis->itemState & ODS_SELECTED) {
// Yes. Invert the whole rectangle (image+text) to signal the selection
BitBlt(lpdis->hDC,
lpdis->rcItem.left, lpdis->rcItem.top,
lpdis->rcItem.right - lpdis->rcItem.left,
lpdis->rcItem.bottom - lpdis->rcItem.top,
lpdis->hDC, 0, 0, PATINVERT);
}
break;
// Do nothing if we gain or lose the focus. This is a laxist view but... it works!
case ODA_FOCUS:
break;
}
return TRUE;
}
And after all this work, here it is:
198 C Tutorial
 Etc. 199

2.18 Etc.
Lcc-win32 gives you access to all this:

Clipboard Just that. A common repository for shared data. Quite a few
formats are available, for images, sound, text, etc.

Communications Read and write from COM ports.

Consoles and text

The “msdos” window improved.
mode support
Debug Help Why not? Write a debugger. Any one can do it. It is quite
interesting as a program.

Device I/O Manage all kind of cards with DeviceIOControl.

Dynamically linked Yes, I know. It is hot in DLL Hell. But at least you get separate
libraries (DLLs) modules, using binary interfaces that can be replaced one by one.
This can lead to confusion, but it is inherent in the method.

Files The disk is spinning anyway. Use it!

File Systems Journaling file systems, NTFS, FAT32. As you like it.

Graphics Windows are graphical objects. The GDI machinery allows you to
draw simple objects with lines or regions, but you can go to higher
dimensions with DirectX or OpenGl.

Handles and Objects Objects that the system manages (windows, files, threads, and
many others) are described by a numerical identifier. A handle to
the object.

Hooks Install yourself in the middle of the message queue, and hear what
is being passed around: you receive the messages before any other
window receives them.

Inter-Process Com- Client/Server, and many other forms of organizing applications are
munications available. You have all the primitives to do any kind of
architecture. Synchronization, pipes, mailslots, you name it.

Mail Send/receive mail messages using the Messaging API.

Multimedia Sound, video, input devices.

Network Yes, TCP/IP. Send data through the wire; develop your own
protocol on top of the basic stuff. You have all the tools in here.

Virtual memory Use those megabytes. They are there anyway. Build huge tables of
data. Use virtual memory, reserve contiguous address space, etc.

Registry. A global database for configuration options.1

200 C Tutorial

Services Run processes in the background, without being bothered by a

console, window, or any other visible interface.

Shell programming Manage files, folders, shortcuts, and the desktop.

Windows Yes, Windows is about windows. You get a huge variety of

graphical objects, from the simple check box to sophisticated, tree-
displaying stuff. An enormous variety of things that wait for you,
ready to be used.
1. Media Control Interface

2.18.1 Clipboard
The data in the clipboard is tagged with a specific format code. To initiate the data transfer to
or from the clipboard you use OpenClipboard, GetClipboardData allows you to read
it, SetClipboardData to write it, etc. You implement this way the famous Cut, Copy and
Paste commands that are ubiquitous in most windows applications. Predefined data formats
exist for images (CF_BITMAP, CF_TIFF), sound (CF_WAVE, CF_RIFF), text
(CF_TEXT), pen data (CF_PENDATA) and several others.

2.18.2 Communications.
You use the same primitives that you use for files to open a communications port. Here is the
code to open COM1 for instance:
HANDLE hComm;
char *gszPort = “COM1”;
hComm = CreateFile( gszPort,
GENERIC_READ | GENERIC_WRITE,
0,
0,
OPEN_EXISTING,
FILE_FLAG_OVERLAPPED,
0);
You use that handle to call ReadFile and WriteFile APIs. Communications events are
handled by SetCommMask, that defines the events that you want to be informed about (break,
clear-to-send, ring, rxchar, and others). You can change the baud rate managing the device
control block (SetCommState), etc. As with network interfaces, serial line programming is
a black art.

2.18.3 Files
Besides the classical functions we have discussed in the examples, Windows offers you more
detailed file control for accessing file properties, using asynchronous file input or output, for
managing directories, controlling file access, locking, etc. In a nutshell, you open a file with
CreateFile, read from it with ReadFile, write to it with WriteFile, close the connection to it
with CloseHandle, and access its properties with GetFileAttributes. Compared with the simple
functions of the standard library those functions are more difficult to use, since they require
more parameters, but they allow you a much finer control.
 Etc. 201

2.18.4 File systems

These days files are taken for granted. File systems not. Modern windows file systems allow
you to track file operations and access their journal data. You can encrypt data, and at last
under windows 2000 Unix’s mount operation is recognized. You can establish symbolic links
for files, i.e, consider a file as a pointer to another one. This pointer is dereferenced by the file
system when accessing the link.
Windows supports several file systems:

1) NTFS.NTFS is the preferred file system on Windows. It was designed to address the
requirements of high-performance file servers and server networks as well as desktop
computers, and in doing so, address many of the limitations of the earlier FAT16 and
FAT32 file systems.

2) FAT32 The File Allocation Table (FAT) file system organizes data on fixed disks and
floppy disks. The main advantage of FAT volumes is that they are accessible by MS-DOS,
Windows, and OS/2 systems. FAT is the only file system currently supported for floppy
disks and other removable media. FAT32 is the most recently defined FAT-based file
system format, and it's included with Windows 95 OSR2, Windows 98, and Windows
Millennium Edition. FAT32 uses 32-bit cluster identifiers but reserves the high 4 bits, so in
effect it has 28-bit cluster identifiers.

3) UDF file system. The implementation is compliant with ISO 13346 and supports UDF
versions 1.02 and 1.5. OSTA (Optical Storage Technology Association) defined UDF in
1995 as a format to replace CDFS for magneto-optical storage media, mainly DVD-ROM.
UDF is included in the DVD specification and is more flexible than CDFS.

2.18.5 Graphics
GDI is the lowest level, the basic machinery for drawing. It provides you:
• Bitmap support
• Brush support for painting polygons.
• Clipping that allows you to draw within the context of your window without worrying
that you could overwrite something in your neighbor’s window. Filled shapes, polygons
ellipses, pie rectangle, lines and curves.
• Color management, palettes etc.
• Coordinate spaces, and transforms.
• Text primitives for text layout, fonts, captions and others.
• Printing
But higher levels in such a vast field like graphics are surely possible. Lcc-win32 offers the
standard jpeg library of Intel Corp to read/write and display jpeg files. Under windows you
can do OpenGl, an imaging system by Silicon Graphics, or use DirectX, developed by
Microsoft.

2.18.6 Handles and Objects

An object is implemented by the system with a standard header and object-specific attributes.
Since all objects have the same structure, there is a single object manager that maintains all
objects. Object attributes include the name (so that objects can be referenced by name), secu-
rity descriptors to rule access to the information stored in those objects, and others, for
202 C Tutorial

instance properties that allow the system to enforce quotas. The system object manager allows
mapping of handles from one process to another (the DuplicateHandle function) and is
responsible for cleanup when the object handle is closed.

2.18.7 Inter-Process Communications

You can use the following primitives:
• Atoms. An atom table is a system-defined table that stores strings and corresponding
identifiers. An application places a string in an atom table and receives a 16-bit integer,
called an atom that can be used to access the string. The system maintains a global atom
table that can be used to send information to/from one process to another: instead of
sending a string, the processes send the atom id.
• Clipboard. This is the natural way to do inter-process communications under windows:
Copy and Paste.
• Mailslots. A mailslot is a pseudofile; it resides in memory, and you use standard Win32
file functions to access it. Unlike disk files, however, mailslots are temporary. When all
handles to a mailslot are closed, the mailslot and all the data it contains are deleted. A
mailslot server is a process that creates and owns a mailslot. A mailslot client is a
process that writes a message to a mailslot. Any process that has the name of a mailslot
can put a message there. Mailslots can broadcast messages within a domain. If several
processes in a domain each create a mailslot using the same name, the participating
processes receive every message that is addressed to that mailslot and sent to the
domain. Because one process can control both a server mailslot handle and the client
handle retrieved when the mailslot is opened for a write operation, applications can
easily implement a simple message-passing facility within a domain.
• Pipes. Conceptually, a pipe has two ends. A one-way pipe allows the process at one end
to write to the pipe, and allows the process at the other end to read from the pipe. A two-
way (or duplex) pipe allows a process to read and write from its end of the pipe.
• Memory mapped files can be used as a global shared memory buffer.

2.18.8 Mail
The Messaging API (MAPI) allows you to program your messaging application or to include
this functionality into your application in a vendor-independent way so that you can change
the underlying message system without changing your program.
 Etc. 203

2.18.9 Multimedia

Audio. You can use Mixers, MIDI, and waveform audio using MCI.108DirectSound offers a
more advanced sound interface.

Input devices. You can use the joystick, precise timers, and multimedia file input/output.

Video. Use AVI files to store video sequences, or to capture video information using a simple,
message-based interface.

2.18.10 Network
Windows Sockets provides you will all necessary functions to establish connections over a
TCP/IP network. The TCPIP subsystem even supports other protocols than TCPIP itself. But
whole books have been written about this, so here I will only point you to the one I used when
writing network programs: Ralph Davis “Windows NT Network programming”, from Addi-
son Wesley.

2.18.11 Hooks
A hook is a mechanism by which a function can intercept events (messages, mouse actions,
keystrokes) before they reach an application. The function can act on events and, in some
cases, modify or discard them. This filter functions receive events, for example, a filter func-
tion might want to receive all keyboard or mouse events. For Windows to call a filter function,
the filter function must be installed—that is, attached—to an entry point into the operating
system, a hook (for example, to a keyboard hook). If a hook has more than one filter function
attached, Windows maintains a chain of those, so several applications can maintain several
hooks simultaneously, each passing (or not) its result to the others in the chain.

2.18.12 Registry
The registry stores data in a hierarchically structured tree. Each node in the tree is called a key.
Each key can contain both sub keys and values. Sometimes, the presence of a key is all the
data that an application requires; other times, an application opens a key and uses the values
associated with the key. A key can have any number of values, and the values can be in any
form. Registry values can be any of the following types:

108. ULARGE_INTEGER is defined in the windows headers like this:

typedef union _ULARGE_INTEGER {
struct {DWORD LowPart; DWORD HighPart;};
long long QuadPart;
} ULARGE_INTEGER,*PULARGE_INTEGER;
The union has two members: a anonymous one with two 32 bit integers, and another with a long long
integer, i.e. 64 bits. We can access the 64-bit integer’s low and high part as 32 bit numbers. This is useful for
functions returning two results in a 64 bit number.
204 C Tutorial

• Binary data
• 32 bit numbers
• Null terminated strings
• Arrays of null terminated strings. The array ends with two null bytes.
• Expandable null terminated strings. These strings contain variables like %PATH% that
are expanded when accessed.

2.18.13 Shell Programming

Windows provides users with access to a wide variety of objects necessary for running appli-
cations and managing the operating system. The most numerous and familiar of these objects
are the folders and files, but there are also a number of virtual objects that allow the user to do
tasks such as sending files to remote printers or accessing the Recycle Bin. The shell organizes
these objects into a hierarchical namespace, and provides users and applications with a consis-
tent and efficient way to access and manage objects.

2.18.14 Services
A service application conforms to the interface rules of the Service Control Manager (SCM).
A user through the Services control panel applet can start it automatically at system boot, or by
an application that uses the service functions. Services can execute even when no user is
logged on to the system

2.18.15 Windows
Here is a short overview of the types of controls available to you.

Control Description

Edit Single or multi line text editor.

Checkbox For a set of multiple choices

Listbox For displaying lists

Combobox A list + an edit control

Group boxes, static text, filled rectangles. Used for labels,

Static
grouping and separation.

Push buttons Used to start an action

Radio buttons Used for choosing one among several possible choices.

Scroll bars Used for scrolling a view.

Animation
controls
Display AVI files

Date and Time Used to input dates

Headers Allow the user to resize a column in a table

 Etc. 205

List view Used to display images and text.

Used to make a scrollable region that contains other controls.

Pager
You scroll the controls into view with the pager.

Used in lengthy operations to give a graphic idea of how much

Progress bar
time is still needed to finish the operation.

Used to pack several dialog boxes into the same place, avoiding
Property Sheets
user confusion by displaying fewer items at the same time.

Allows editing text with different typefaces (bold, italic) with

Richedit different fonts, in different colors… The basic building block to
build a text processor.

Status bars Used to display status information at the bottom of a window

Tab controls The building block to make property sheets.

Toolbar controls A series of buttons to accelerate application tasks.

Show explanations about an item currently under the mouse in a

Tooltips
pop-up window.

Trackbars Analogical volume controls, for instance.

Tree view Displays hierarchical trees.

206 C Tutorial

2.19 Advanced C programming with lcc-win32

2.19.1 Operator overloading

When you write:
int a=6,b=8;
int c = a+b;
you are actually calling a specific intrinsic routine of the compiler to perform the addition of
two integers. Conceptually, it is like if you were doing:
int a=6,b=8;
int c = operator+(a,b);
This “operator+” function is inlined by the compiler. The compiler knows about this operation
(and several others), and generates the necessary assembly instructions to perform it at run
time.
Lcc-win32 allows you to define functions written by you, to take the place of the built-in oper-
ators. For instance you can define a structure complex, to store complex numbers. Lcc-win32
allows you to write:
COMPLEX operator+(COMPLEX A, COMPLEX B)
{
… Code for complex number addition goes here
}
This means that whenever the compiler sees “a+b” and “a” is a COMPLEX and “b” is a COM-
PLEX, it will generate a call to the previously defined overloaded operator, instead of com-
plaining about a “syntax error”.
This is called in “tech-speak” operator overloading. There are several rules for writing those
functions and using this feature. All of them explained in-depth in the user’s manual. This
short notice is just a pointer, to show you what is possible.
The implementation of this feature is compatible with the C++ language that offers a similar
facility.

2.19.2 References
References are a special kind of pointers that are always dereferenced when used. When you
declare a reference, you must declare immediately the object they point to. There are no
invalid references since they can’t be assigned. Once a reference is declared and initialized,
you can’t reassign them to another object.
They are safer pointers than normal pointers, since they are guaranteed correct, unless the
object they point to is destroyed, of course. References are initialized with the construct:
int a;
int &pa = a;
The “pa” variable is a reference to an integer (an “int &”), and it is immediately initialized to
point to the integer “a”. Note that you do not have to take the address of “a”, but just put its
name. The compiler takes the address.
Again, here is a short pointer only. A complete description is found in the user manual.
 Advanced windows techniques 207

2.19.3 Overloaded functions

You can declare a function that receives several types of arguments, i.e. a generic function by
using the "overloaded" keyword. Suppose a function that receives as arguments a qfloat or a
double.
int overloaded docalcs(qfloat *pq) { …}

int overloaded docalcs(double *pd) { … }

This function can receive either a "qfloat" number or a double number as input. The compiler
notices the type of argument passed in the call and arranges for calling the right function.
Notice that you define two internally different functions, and that the decision of which one
will be called will be done according to the type of arguments in the call.
It is not possible to declare a function overloaded after a call to this function is already gener-
ated. The following code will NOT work:
docals(2.3);
int overloaded docals(double *pd);
Here the compiler will signal an error.

2.19.4 Default arguments

Default arguments allow you to simplify the interface of a function by providing arguments
that if not present will assume a default value.
For instance:
int fn(int a,int b = 78);
This declares a function called «frn», that takes one argument and optionally a second. When
the second is not given, the compiler will arrange for it being 78. For instance
fn(2);
is equivalent to
fn(2,78);

2.20 Advanced windows techniques

Windows is not only drawing of course. It has come a long way since windows 3.0, and is now
a sophisticated operating system. You can do things like memory-mapped files for instance,
that formerly were possible only under UNIX. Yes, “mmap” exists now under windows, and it
is very useful.

2.20.1 Memory mapped files

Memory mapped files allow you to see a disk file as a section of RAM. The difference
between the disk and the RAM disappears. You can just seek to a desired position by incre-
menting a pointer, as you would do if you had read the whole file into RAM, but more effi-
ciently. It is the operating system that takes care of accessing the disk when needed. When you
close the connection, the operating system handles the clean up.
Here is a small utility that copies a source disk file to a destination using memory-mapped
files.
int main (int argc, char **argv)
{
int fResult = FAILURE;
208 C Tutorial

ULARGE_INTEGER liSrcFileSize, // See109

liBytesRemaining,
liMapSize,
liOffset;
HANDLE hSrcFile = INVALID_HANDLE_VALUE,
hDstFile = INVALID_HANDLE_VALUE,
hSrcMap = 0,
hDstMap = 0;

BYTE * pSrc = 0,
* pDst = 0;

char * pszSrcFileName = 0,
* pszDstFileName = 0;

if (argc != 3) // test if two arguments are given in the command line

{
printf("usage: copyfile <srcfile> <dstfile>\n");
return (FAILURE);
}

pszSrcFileName = argv[argc-2]; // Src is second to last argument

pszDstFileName = argv[argc-1]; // Dst is the last argument
/* We open the source file for reading only, and make it available for other processes
even if we are copying it. We demand to the operating system to ensure that the file
exists already */
hSrcFile = CreateFile (pszSrcFileName,
GENERIC_READ, //Source file is opened for reading only
FILE_SHARE_READ,// Shareable
0, OPEN_EXISTING, 0, 0);
if (INVALID_HANDLE_VALUE == hSrcFile)
{
printf("couldn't open %s\n", pszSrcFileName);
goto DONE;
}
/* We open the destination file for reading and writing with no other access allowed.
We demand the operating system to create the file if it doesn’t exist.
*/
hDstFile = CreateFile (pszDstFileName,
GENERIC_READ|GENERIC_WRITE, 0,
0, CREATE_ALWAYS, 0, 0);
if (INVALID_HANDLE_VALUE == hDstFile)
{
printf("couldn't create %s\n", pszSrcFileName);
goto DONE;
}
/*
We need to query the OS for the size of this file. We will need this information later when we
create the file-mapping object. Note that we receive a 64 bit number splitted in two. We receive a
32-bit integer containing the result’s lower 32 bits, and we pass to the function the address
where it should put the remaining bits! Well, if you find this interface strange (why not return a 64
bit integer?) please do not complain to me. Note too the strange form of the error checking
afterwards: we check for a return value of all bits set to one, and check the result of the
GetLastError() API.
*/
SetLastError(0);
liSrcFileSize.LowPart = GetFileSize(hSrcFile,

109. This is not really necessary. Passing a NULL value works too.
 Advanced windows techniques 209

&liSrcFileSize.HighPart)
if (0xFFFFFFFF == liSrcFileSize.LowPart &&
GetLastError() != NO_ERROR){
printf("couldn't get size of source file\n");
goto DONE;
}
/*
Special case: If the source file is zero bytes, we don't map it because there's no need to and
anyway CreateFileMapping cannot map a zero-length file. But since we've created the
destination, we've successfully "copied" the source.
*/
if (0 == liSrcFileSize.QuadPart)
{
fResult = SUCCESS;
goto DONE;
}
/*
Map the source file into memory. We receive from the OS a HANDLE that corresponds to the
opened mapping object.
*/
hSrcMap = CreateFileMapping (hSrcFile,
0, PAGE_READONLY, 0, 0, 0);
if (!hSrcMap){
printf("couldn't map source file\n");
goto DONE;
}
/*
Now we create a file mapping for the destination file using the size parameters we got above.
*/
hDstMap = CreateFileMapping (hDstFile, 0,
PAGE_READWRITE,
liSrcFileSize.HighPart,
liSrcFileSize.LowPart, 0);
if (!hDstMap)
{
DEBUG_PRINT("couldn't map destination file\n");
goto DONE;
}
/*
Now that we have the source and destination mapping objects, we build two map views of the
source and destination files, and do the file copy.
To minimize the amount of memory consumed for large files and make it possible to copy files
that couldn't be mapped into our virtual address space entirely (those over 2GB), we limit the
source and destination views to the smaller of the file size or a specified maximum view size
(MAX_VIEW_SIZE--which is 96K).
If the size of file is smaller than the max view size, we'll just map and copy it. Otherwise, we'll
map a portion of the file, copy it, then map the next portion, copy it, etc. until the entire file is
copied.
MAP_SIZE is 32 bits because MapViewOfFile requires a 32-bit value for the size of the view.
This makes sense because a Win32 process's address space is 4GB, of which only 2GB (2^31)
bytes may be used by the process. However, for the sake of making 64-bit arithmetic work below
for file offsets, we need to make sure that all 64 bits of limpest are initialized correctly.
*/
liBytesRemaining.QuadPart = liSrcFileSize.QuadPart;
/* This assignment sets all 64 bits to this value */
liMapSize.QuadPart = MAX_VIEW_SIZE;

do {
/* Now we start our copying loop. The “min” macro returns the smaller of two numbers. */
liMapSize.QuadPart = min(liBytesRemaining.QuadPart,
210 C Tutorial

liMapSize.QuadPart)

liOffset.QuadPart = liSrcFileSize.QuadPart –
liBytesRemaining.QuadPart;

pSrc = (BYTE *)MapViewOfFile(hSrcMap, FILE_MAP_READ,

liOffset.HighPart,
liOffset.LowPart, liMapSize.LowPart);
pDst = (BYTE *)MapViewOfFile(hDstMap,FILE_MAP_WRITE,
liOffset.HighPart,
liOffset.LowPart, liMapSize.LowPart);
/* We use memcpy to do the actual copying */
memcpy(pDst, pSrc, liMapSize.LowPart);

UnmapViewOfFile (pSrc);
UnmapViewOfFile (pDst);

liBytesRemaining.QuadPart -= liMapSize.QuadPart;
}
while (liBytesRemaining.QuadPart > 0);

fResult = SUCCESS;
DONE:
/* We are done, Note the error treatment of this function. We use gotos to reach the end of the
function, and here we cleanup everything. */
if (hDstMap) CloseHandle (hDstMap);

if(hDstFile!=INVALID_HANDLE_VALUE) CloseHandle(hDstFile);

if (hSrcMap) CloseHandle(hSrcMap);

if (hSrcFile != INVALID_HANDLE_VALUE)
CloseHandle (hSrcFile);

if (fResult != SUCCESS)
{
printf("copying %s to %s failed.\n",
pszSrcFileName, pszDstFileName);
DeleteFile (pszDstFileName);
}
return (fResult);
}

Summary: To get a pointer to a memory mapped file do the following:

• Open the file with CreateFile
• Create a mapping object with CreateFileMapping using the handle you receive from
CreateFile.
• Map a portion (or all) of the file contents, i.e. create a view of it, with MapViewOfFile.

2.20.2 Letting the user browse for a folder: using the shell
A common task in many programming situations is to let the user find a folder (directory) in
the file system hierarchy. When you want to search for certain item, for instance, or when you
need to allow the user to change the current directory of your application. The windows shell
offers a ready-made set of functions for you, and the resulting function is quite short. Let’s
first see its specification, i.e. what do we want as an interface to this function.
 Advanced windows techniques 211

Required is a character string where the result will be written to, and a title for the user inter-
face element. The result should be 1 if the path contains a valid directory, 0 if there was any
error, the user cancelled, whatever.
To be clear about the specifications let’s look at this example:
int main(void)
{
char path[MAX_PATH];
if (BrowseDir("Choose a directory",path)) {
printf("You have choosen %s\n",path);
}
else printf("action cancelled\n");
return 0;
}
How do we write “BrowseDir” in windows?
Here it is:
#include <shlobj.h>
#include <stdio.h>

int BrowseDir(unsigned char *Title,char *result)

{
LPMALLOC pMalloc; (1)
BROWSEINFO browseInfo;(2)
LPITEMIDLIST ItemIDList;(3)
int r = 0; (4)

if (S_OK != SHGetMalloc(&pMalloc))(5)
return 0;
memset(&browseInfo,0,sizeof(BROWSEINFO));(6)
browseInfo.hwndOwner = GetActiveWindow();(7)
browseInfo.pszDisplayName = result;(8)
browseInfo.lpszTitle = Title;(9)
browseInfo.ulFlags = BIF_NEWDIALOGSTYLE;(10)
ItemIDList = SHBrowseForFolder(&browseInfo);(11)
if (ItemIDList != NULL) {
*result = 0;
if (SHGetPathFromIDList(ItemIDList,result))(12)
{
if (result[0]) r = 1;(13)
pMalloc->lpVtbl->Free(pMalloc,ItemIDList);(14)
}
}
pMalloc->lpVtbl->Release(pMalloc);(15)
return r;
}
Small isn’t it?
Let’s see the gory details.
We need a local variable that will hold a pointer to a shell defined function that will allocate
and release memory. The shell returns us a result that needs memory to exist. We need to free
that memory, and we have to take care of using the same function that the shell uses to allocate
memory. This pointer to an interface (the malloc interface) will be in our local variable pMal-
loc.
The shell needs information about the environment, and some pointers to put the results of the
interaction with the user. We will see more of this when we fill this structure below.
212 C Tutorial

The shell uses a lot of stuff, and we have to take care to avoid filling our brain with unending
details. What is an ITEMLIST? Actually I haven’t even bothered to read the docs about it,
since the only use I found is to pass it around to other shell functions.
The result of the function is initialized to zero, i.e. we will set this result to 1 only and only if
there is a valid path in the buffer.
OK. Here we start. The first thing to do then, is to get the pointer to the shell allocator. If any-
thing goes wrong there, there is absolutely nothing we can do and the best thing is to return
immediately with a FALSE result.
We have to clean up the structure (note that this is a local variable, so its contents are as yet
undefined). We use the primitive memset and set all members of the structure to zero. This is a
common idiom in C: clear all memory before using and assign to it a known value. Since the
default value of many structure members is zero, this easies the initialization of the structure
since we do not have to explicitly set them to NULL or zero.
We start the filling of the relevant members. We need to put the owner window handle in the
hwndOwner member. We get the handle of the current active window using a call to a win-
dows API.
The shell needs place to store the display name of the chosen path. Note that this is not what
we want (the full path) but just the last item in the path, i.e. the last directory. Why this is so?
Because the user could choose a path like “My documents”, and in this case we could detect
that the user has chosen a “standard well known name” for the directory. But we do not use
this feature, and just give the shell some place in our… result variable. Since we overwrite this
data later, this is harmless and avoids a new variable.110
We set the lpszTitle member to the value we passed in. This is a text that will be displayed at
the top of the user interface window that appears, and should remain the user what kind of
folder he/she is looking for.
As flags we just pass BFID_USENEWUI, meaning that we want the shell to use the improved
user interface, with drag and drop, new folder, the possibility of deleting folders, whatever.
And we are done with the filling of the structure! We just have to call our famous
SHBrowseForFolder, and assign the result to ItemIdList. Here is an image of the user
interface display that appears in a windows 2000 machine; in other versions of windows it will
look different. The user interface is quite sophisticated, and it is all at our disposal without
writing any code (well almost!). What is better; even if we had spent some months developing

110. Paragraph 6.5.2: “Expressions”: << Between the previous and the next sequence point an object
shall have its stored value modified at most once by the evaluation of an expression. >>. Then, the standards
adds in a footnote: “This paragraph renders undefined statement expressions such as:
i = ++i + 1;
a[i++] = i;
 Some Coding Tips 213

a similar thing, we would have to maintain it, test it, etc. Note that you can transparently
browse the network, give a symbolic path like “My computer” or other goodies.

If the call worked, i.e. if the user interface returns a valid pointer and not NULL, we should
translate the meaningless stuff we receive into a real path, so we call SHGetPathFro-
mIDList, to do exactly that. We pass it a pointer to the result character string that we receive
as second argument.
If that function returns OK, we verify that the returned path is not empty, and if it is, we set the
result of this function to TRUE.
Now we have to clean-up. We use the COM interface pointer we received from SHGetMal-
loc, and use its only member (lpVtbl) to get into the Free function pointer member. There are
other function pointers in that structure for sure, but we do not use them in this application. We
pass to that Free function a pointer to the interface it gave to us, and then the object to free.
When we are done with a COM interface we have to remember to call a function to release it,
passing it again a pointer to the returned object. We are done now, we return the result and
exit.
How can this program fail?
There are only three API calls here, and all of them are tested for failure. In principle there is
no way this can fail, although it could fail for other reasons: it could provoke a memory leak
(for instance if we had forgotten the call to the Release method at the end), or it could use
resources that are never released (the list of identifiers that we obtain from the system, etc.

2.21 Some Coding Tips

Here are some useful receipts for solving specific windows problems.
214 C Tutorial

2.21.1 Determining which version of Windows the program is

running
BOOL InWinNT() //test for NT
{
OSVERSIONINFO osv;
GetVersionEx(&osv);
osv.dwOSVersionInfoSize=sizeof(osv);
return osv.dwPlatformId==VER_PLATFORM_WIN32_NT;
}

2.21.2 Translating the value returned by GetLastError() into a

readable string
BOOL GetFormattedError(LPTSTR dest,int size)
{
DWORD dwLastError=GetLastError();
if(!dwLastError)
return 0;
BYTE width=0;
DWORD flags;
flags = FORMAT_MESSAGE_MAX_WIDTH_MASK &width;
flags |= FORMAT_MESSAGE_FROM_SYSTEM;
flags |= FORMAT_MESSAGE_IGNORE_INSERTS;
return 0 != FormatMessage(flags,
NULL,
dwLastError,
MAKELANGID(LANG_NEUTRAL, SUBLANG_DEFAULT),
dest,
size,
NULL);
}

2.21.3 Clearing the screen in text mode

The following code will clear the screen in text mode.
#include <windows.h>

/* Standard error macro for reporting API errors */

#define PERR(bSuccess, api){if(!(bSuccess)) \
printf("%s:Error %d from %s \
on line %d\n", __FILE__, GetLastError(), api, __LINE__);}

void cls( HANDLE hConsole )

{
COORD coordScreen = { 0, 0 }; /* Home the cursor here */
BOOL bSuccess;
DWORD cCharsWritten;
CONSOLE_SCREEN_BUFFER_INFO csbi; /* to get buffer info */
DWORD dwConSize; /* number of character cells in the current buffer */
/* get the number of character cells in the current buffer */

bSuccess = GetConsoleScreenBufferInfo( hConsole, &csbi );

PERR( bSuccess, "GetConsoleScreenBufferInfo" );
dwConSize = csbi.dwSize.X * csbi.dwSize.Y;

/* fill the entire screen with blanks */

bSuccess = FillConsoleOutputCharacter( hConsole, (TCHAR) ' ',

dwConSize, coordScreen, &cCharsWritten );
 Some Coding Tips 215

PERR( bSuccess, "FillConsoleOutputCharacter" );

/* get the current text attribute */

bSuccess = GetConsoleScreenBufferInfo( hConsole, &csbi );

PERR( bSuccess, "ConsoleScreenBufferInfo" );

/* now set the buffer's attributes accordingly */

bSuccess = FillConsoleOutputAttribute(hConsole, csbi.wAttributes,

dwConSize, coordScreen, &cCharsWritten );
PERR( bSuccess, "FillConsoleOutputAttribute" );
/* put the cursor at (0, 0) */

bSuccess = SetConsoleCursorPosition( hConsole, coordScreen );

PERR( bSuccess, "SetConsoleCursorPosition" );
return;
}
This function can be called like this:
cls(GetStdHandle(STD_OUTPUT_HANDLE));
The library TCCONIO.LIB contains many other functions for text manipulation using the
console interface. The corresponding header file is TCCONIO.H, which is automatically
included when you include conio.h
This library was contributed by Daniel Guerrero (daguer@geocities.com)

2.21.4 Getting a pointer to the stack

To get a pointer to the stack, use the following code:
far MyFunction()
{
int x;
int far *y = &x;
}
NOTE: This pointer will not be valid when the function is exited, since the stack contents will
change.

2.21.5 Disabling the screen saver from a program

Under Windows NT, you can disable the screen saver from your application code. To detect if
the screen saver is enabled, use this:
SystemParametersInfo( SPI_GETSCREENSAVEACTIVE,
0,
pvParam,
0
);
On return, the parameter pvParam will point to TRUE if the screen saver setting is enabled in
the system control panel applet and FALSE if the screen saver setting is not enabled.
To disable the screen saver setting, call SystemParametersInfo() with this:
SystemParametersInfo( SPI_SETSCREENSAVEACTIVE,
FALSE,
0,
SPIF_SENDWININICHANGE
);
216 C Tutorial

2.21.6 Drawing a gradient background

You can draw a smooth gradient background using the following code:
void DrawBackgroundPattern(HWND hWnd)
{
HDC hDC = GetDC(hWnd); // Get the DC for the window.
RECT rectFill; // Rectangle for filling band.
RECT rectClient; // Rectangle for entire client area.
float fStep; // How large is each band?
HBRUSH hBrush;
int iOnBand; // Loop index

// How large is the area you need to fill?

GetClientRect(hWnd, &rectClient);

// Determine how large each band should be in order to cover the

// client with 256 bands (one for every color intensity level).
fStep = (float)rectClient.bottom / 256.0f;

// Start filling bands

for (iOnBand = 0; iOnBand < 256; iOnBand++) {

// Set the location of the current band.

SetRect(&rectFill,
0, // Upper left X
(int)(iOnBand * fStep), // Upper left Y
rectClient.right+1, // Lower right X
(int)((iOnBand+1) * fStep)); // Lower right Y

// Create a brush with the appropriate color for this band.

hBrush = CreateSolidBrush(RGB(0, 0, (255 – iOnBand)));

// Fill the rectangle.

FillRect(hDC, &rectFill, hBrush);

// Get rid of the brush you created.

DeleteObject(hBrush);
};

// Give back the DC.

ReleaseDC(hWnd, hDC);
}

2.21.7 Capturing and printing the contents of an entire window

//
// Return a HDC for the default printer.
//
HDC GetPrinterDC(void)
{
PRINTDLG pdlg;
memset(&pdlg, 0, sizeof(PRINTDLG));
pdlg.lStructSize = sizeof(PRINTDLG);
pdlg.Flags = PD_RETURNDEFAULT | PD_RETURNDC;
PrintDlg(&pdlg);
return pdlg.hDC;
}
//
// Create a copy of the current system palette.
 Some Coding Tips 217

//
HPALETTE GetSystemPalette()
{
HDC hDC;
HPALETTE hPal;
HANDLE hLogPal;
LPLOGPALETTE lpLogPal;
// Get a DC for the desktop.
hDC = GetDC(NULL);
// Check to see if you are a running in a palette-based
// video mode.
if (!(GetDeviceCaps(hDC, RASTERCAPS) & RC_PALETTE)) {
ReleaseDC(NULL, hDC);
return NULL;
}
// Allocate memory for the palette.
lpLogPal = GlobalAlloc(GPTR, sizeof(LOGPALETTE) + 256 *
sizeof(PALETTEENTRY));
if (!hLogPal)
return NULL;
// Initialize.
lpLogPal->palVersion = 0x300;
lpLogPal->palNumEntries = 256;
// Copy the current system palette into the logical palette.
GetSystemPaletteEntries(hDC, 0, 256,
(LPPALETTEENTRY) (lpLogPal->palPalEntry));
// Create the palette.
hPal = CreatePalette(lpLogPal);
// Clean up.
GlobalFree(lpLogPal);
ReleaseDC(NULL, hDC);
return hPal;
}
//
// Create a 24-bit-per-pixel surface.
//
HBITMAP Create24BPPDIBSection(HDC hDC, int iWidth, int iHeight)
{
BITMAPINFO bmi;
HBITMAP hbm;
LPBYTE pBits;
// Initialize to 0s.
ZeroMemory(&bmi, sizeof(bmi));
// Initialize the header.
bmi.bmiHeader.biSize = sizeof(BITMAPINFOHEADER);
bmi.bmiHeader.biWidth = iWidth;
bmi.bmiHeader.biHeight = iHeight;
bmi.bmiHeader.biPlanes = 1;
bmi.bmiHeader.biBitCount = 24;
bmi.bmiHeader.biCompression = BI_RGB; // Create the surface.
hbm = CreateDIBSection(hDC,&bmi, DIB_RGB_COLORS,&pBits,NULL, 0);
return (hbm);
}
//
// Print the entire contents (including the non-client area) of
// the specified window to the default printer.
BOOL PrintWindowToDC(HWND hWnd)
{
HBITMAP hbm;
HDC hdcPrinter;
218 C Tutorial

HDC hdcMemory;
HDC hdcWindow;
int iWidth;
int iHeight;
DOCINFO di;
RECT rc;
DIBSECTION ds;
HPALETTE hPal;
// Do we have a valid window?
if (!IsWindow(hWnd))
return FALSE;
// Get a HDC for the default printer.
hdcPrinter = GetPrinterDC();
if (!hdcPrinter)
return FALSE;
// Get the HDC for the entire window.
hdcWindow = GetWindowDC(hWnd);
// Get the rectangle bounding the window.
GetWindowRect(hWnd, &rc);
// Adjust coordinates to client area.
OffsetRect(&rc, -rc.left, -rc.top);
// Get the resolution of the printer device.
iWidth = GetDeviceCaps(hdcPrinter, HORZRES);
iHeight = GetDeviceCaps(hdcPrinter, VERTRES);
// Create the intermediate drawing surface at window resolution.
hbm = Create24BPPDIBSection(hdcWindow, rc.right, rc.bottom);
if (!hbm) {
DeleteDC(hdcPrinter);
ReleaseDC(hWnd, hdcWindow);
return FALSE;
}
// Prepare the surface for drawing.
hdcMemory = CreateCompatibleDC(hdcWindow);
SelectObject(hdcMemory, hbm);
// Get the current system palette.
hPal = GetSystemPalette(); // If a palette was returned.
if (hPal) { // Apply the palette to the source DC.
SelectPalette(hdcWindow, hPal, FALSE);
RealizePalette(hdcWindow);
// Apply the palette to the destination DC.
SelectPalette(hdcMemory, hPal, FALSE);
RealizePalette(hdcMemory);
}
// Copy the window contents to the memory surface.
BitBlt(hdcMemory, 0, 0, rc.right, rc.bottom,
hdcWindow, 0, 0, SRCCOPY);
// Prepare the DOCINFO.
ZeroMemory(&di, sizeof(di));
di.cbSize = sizeof(di);
di.lpszDocName = "Window Contents"; // Initialize the print job.
if (StartDoc(hdcPrinter, &di) > 0) {// Prepare to send a page.
if (StartPage(hdcPrinter) > 0) {
// Retrieve the information describing the surface.
GetObject(hbm, sizeof(DIBSECTION), &ds);
// Print the contents of the surface.
StretchDIBits(hdcPrinter,
0, 0, iWidth, iHeight,
0, 0, rc.right, rc.bottom, ds.dsBm.bmBits,
(LPBITMAPINFO) & ds.dsBmih, DIB_RGB_COLORS,
SRCCOPY);
 Some Coding Tips 219

// Let the driver know the page is done.

EndPage(hdcPrinter);
}
// Let the driver know the document is done.
EndDoc(hdcPrinter);
}
// Clean up the objects you created.
DeleteDC(hdcPrinter);
DeleteDC(hdcMemory);
ReleaseDC(hWnd, hdcWindow);
DeleteObject(hbm);
if (hPal)
DeleteObject(hPal);
}

2.21.8 Centering a dialog box in the screen

Use the following code:
{
RECT rc;
GetWindowRect(hDlg, &rc);
SetWindowPos(hDlg, NULL,
((GetSystemMetrics(SM_CXSCREEN) - (rc.right - rc.left)) / 2),
((GetSystemMetrics(SM_CYSCREEN) - (rc.bottom - rc.top)) / 2),
0, 0, SWP_NOSIZE | SWP_NOACTIVATE);
}

2.21.9 Determining the number of visible items in a list box

In a list box, if the number of lines is greater than the number of lines in the list box, some of
them will be hidden. In addition, it could be that the list box is an owner draw list box, making
the height of each line a variable quantity. Here is a code snippet that will handle all cases,
even when all of the items of the list box are visible, and some white space is left at the bot-
tom.
The basic idea is to subtract each line height from the total height of the client area of the list
box.
int ntop, nCount, nRectheight, nVisibleItems;
RECT rc, itemrect;
// First, get the index of the first visible item.
ntop = SendMessage(hwndList, LB_GETTOPINDEX, 0, 0);
// Then get the number of items in the list box.
nCount = SendMessage(hwndList, LB_GETCOUNT, 0, 0);
// Get the list box rectangle.
GetClientRect(hwndList, &rc);
// Get the height of the list box’s client area.
nRectheight = rc.bottom – rc.top;
// This counter will hold the number of visible items.
nVisibleItems = 0;
// Loop until the bottom of the list box. or the last item has been reached.
while ((nRectheight > 0) && (ntop < nCount))
{
// Get current line’s rectangle.
SendMessage(hwndList, LB_GETITEMRECT,
ntop,(DWORD)(&itemrect));
// Subtract the current line height.
nRectheight = nRectheight – (itemrect.bottom – itemrect.top);
nVisibleItems++; // Increase item count.
220 C Tutorial

Ntop++; // Move to the next line.

}

2.21.10 Starting a non-modal dialog box

Non-modal dialog boxes behave as independent top level windows. They can be started
using the CreateDialog function.
HWND CreateDialog(
HINSTANCE hInstance,// handle to application instance
LPCTSTR lpTemplate,// Identifies dialog box template name.
HWND hWndParent,// Handle to owner window.
DLGPROC lpDialogFunc // Pointer to dialog box procedure.
);
Non-modal dialog boxes should be used with care, since they are equivalent to starting another
thread of execution. In addition, you should never forget that the user can restart the same
sequence of events that led to that dialog box, causing a second dialog box to appear.

2.21.11 Propagating environment variables to the parent

environment
Under Windows, there is no ‘export’ directive as in Unix systems. To propagate the value of
an environment variable to the rest of the system, use the following registry key:
HKEY_CURRENT_USER \
Environment
You can modify system environment variables by editing the following registry key:
HKEY_LOCAL_MACHINE \
SYSTEM \
CurrentControlSet \
Control \
Session Manager \
Environment
Note that any environment variable that needs to be expanded (for example, when you use
%SYSTEM%) must be stored in the registry as a REG_EXPAND_SZ registry value. Any val-
ues of type REG_SZ will not be expanded when read from the registry.
The problem with this method, however, is that changes will be effective only after the next
logoff, probably not exactly what you want.
To effect these changes without having to log off, broadcast a WM_SETTINGCHANGE mes-
sage to all windows in the system, so that any applicable applications (such as Program Man-
ager, Task Manager, Control Panel, etc.) can perform an update.
SendMessageTimeout(HWND_BROADCAST, WM_SETTINGCHANGE, 0,
(LPARAM) "Environment", SMTO_ABORTIFHUNG,
5000, &dwReturnValue);
In theory, this will work, but there is a problem with Windows 95. Under that system, there is
no other way to set those variables than rebooting the system!

2.21.12 Restarting the shell under program control

In many cases it can be necessary to restart the shell. To do this, find the window of the
Explorer, send it a quit message, and restart it. The following code snippet will work:
HWND hwndShell = FindWindow("Progman", NULL);
PostMessage(hwndShell, WM_QUIT, 0, 0L);
WinExec("Explorer.exe",SW_SHOW);
 Some Coding Tips 221

2.21.13 Translating client coordinates to screen coordinates

To determine the screen coordinates for the client area of a window, call the ClientToScreen
function to translate the client coordinates returned by GetClientRect into screen coordinates.
The following code demonstrates how to use the two functions together:
RECT rMyRect;
GetClientRect(hwnd, (LPRECT)&rMyRect);
ClientToScreen(hwnd, (LPPOINT)&rMyRect.left);
ClientToScreen(hwnd, (LPPOINT)&rMyRect.right);

2.21.14 Passing an argument to a dialog box procedure

You can pass a void * to a dialog box procedure by calling:
result = DialogBoxParam(hInst,// Instance of the application.
MAKEINTRESOURCE(id), // The resource ID or dialog box name.
GetActiveWindow(), // The parent window.
Dlgfn, // The dialog box procedure.
(DWORD) "Hello"); // The arguments.
In your dialog box procedure (here DlgFn), you will find those arguments in the lParam
parameter of the WM_INITDIALOG message.

2.21.15 Calling printf from a windows application

Windows applications do not have a console, i.e., the ‘DOS’ window. To access the console
from a Windows application, create one, initialize stdout, and use it as you would normally
use it from a native console application.
#include <windows.h>
#include <stdio.h>
#include <fcntl.h>
int main(void)
{
int hCrt;
FILE *hf;

AllocConsole();
hCrt = _open_osfhandle((long) GetStdHandle (
STD_OUTPUT_HANDLE),_O_TEXT );
hf = fdopen( hCrt, "w" );
*stdout = *hf;
setvbuf( stdout, NULL, _IONBF, 0 );
printf("Hello world\n");
return 0;
}

2.21.16 Enabling or disabling a button or control in a dialog box.

You should first get the window handle of the control using
hwndControl = GetDlgItem(hwndDlg,IDBUTTON);
Using that window handle, call EnableWindow.
222 C Tutorial

2.21.17 Making a window class available for all applications in

the system.
Under 32-bit Windows and Windows NT, a style of CS_GLOBALCLASS indicates that the
class is available to every DLL in the process, not every application and DLL in the system, as
it does in Windows 3.1.
To have a class registered for every process in the system under Windows NT:
• Register the class in a DLL.
• Use a style of CS_GLOBALCLASS.
• List the DLL in the following registry key.
HKEY_LOCAL_MACHINE\SOFTWARE\Microsoft\Windows NT\CurrentVersion\
Windows\AppInit_DLLs
This will force the DLL to be loaded into every process in the system, thereby registering the
class in each and every process.
NOTE: This technique does not work under Windows 95.

2.21.18 Accessing the disk drive directly without using a file

system
To open a physical hard drive for direct disk access (raw I/O) in a Win32-based application,
use a device name of the form
\\.\PhysicalDriveN
where N is 0, 1, 2, and so forth, representing each of the physical drives in the system.
To open a logical drive, direct access is of the form
\\.\X:
where X: is a hard-drive partition letter, floppy disk drive, or CD-ROM drive.
You can open a physical or logical drive using the CreateFile() application programming inter-
face (API) with these device names provided that you have the appropriate access rights to the
drive (that is, you must be an administrator). You must use both the CreateFile()
FILE_SHARE_READ and FILE_SHARE_WRITE flags to gain access to the drive.
Once the logical or physical drive has been opened, you can then perform direct I/O to the data
on the entire drive. When performing direct disk I/O, you must seek, read, and write in multi-
ples of sector sizes of the device and on sector boundaries. Call DeviceIoControl() using
IOCTL_DISK_GET_DRIVE_GEOMETRY to get the bytes per sector, number of sectors,
sectors per track, and so forth, so that you can compute the size of the buffer that you will
need.
Note that a Win32-based application cannot open a file by using internal Windows NT object
names; for example, attempting to open a CD-ROM drive by opening
\Device\CdRom0
does not work because this is not a valid Win32 device name. You can use the QueryDosDe-
vice() API to get a list of all valid Win32 device names and see the mapping between a par-
ticular Win32 device name and an internal Windows NT object name. An application running
at a sufficient privilege level can define, redefine, or delete Win32 device mappings by calling
the DefineDosDevice() API.
 Some Coding Tips 223

2.21.19 Enumerating registry subkeys

The following example demonstrates the use of the RegQueryInfoKey, RegEnumKeyEx, and
RegEnumValue functions. The hKey parameter passed to each function is a handle to an open
key. This key must be opened before the function call and closed afterward.
// QueryKey - Enumerates the subkeys of key, and the associated
// values, then copies the information about the keys and values
// into a pair of edit controls and list boxes.
// hDlg - Dialog box that contains the edit controls and list boxes.
// hKey - Key whose subkeys and values are to be enumerated.
void QueryKey(HWND hDlg, HANDLE hKey)
{
CHAR achKey[MAX_PATH];
CHAR achClass[MAX_PATH] = ""; // buffer for class name
DWORD cchClassName = MAX_PATH; // size of class string
DWORD cSubKeys; // number of subkeys
DWORD cbMaxSubKey; // longest subkey size
DWORD cchMaxClass; // longest class string
DWORD cValues; // number of values for key
DWORD cchMaxValue; // longest value name
DWORD cbMaxValueData; // longest value data
DWORD cbSecurityDescriptor; // size of security descriptor
FILETIME ftLastWriteTime; // last write time

DWORD i, j;
DWORD retCode, retValue;

CHAR achValue[MAX_VALUE_NAME];
DWORD cchValue = MAX_VALUE_NAME;
CHAR achBuff[80];

// Get the class name and the value count.

RegQueryInfoKey(hKey, // key handle
achClass, // buffer for class name
&cchClassName, // size of class string
NULL, // reserved
&cSubKeys, // number of subkeys
&cbMaxSubKey, // longest subkey size
&cchMaxClass, // longest class string
&cValues, // number of values for this key
&cchMaxValue, // longest value name
&cbMaxValueData, // longest value data
&cbSecurityDescriptor, // security descriptor
&ftLastWriteTime); // last write time

SetDlgItemText(hDlg, IDE_CLASS, achClass);

SetDlgItemInt(hDlg, IDE_CVALUES, cValues, FALSE);

SendMessage(GetDlgItem(hDlg, IDL_LISTBOX),
LB_ADDSTRING, 0, (LONG) "..");

// Enumerate the child keys, until RegEnumKeyEx fails. Then

// get the name of each child key and copy it into the list box.

SetCursor(LoadCursor(NULL, IDC_WAIT));
for (i = 0, retCode = ERROR_SUCCESS;
retCode == ERROR_SUCCESS; i++)
{
retCode = RegEnumKeyEx(hKey,
i,
224 C Tutorial

achKey,
MAX_PATH,
NULL,
NULL,
NULL,
&ftLastWriteTime);
if (retCode == (DWORD)ERROR_SUCCESS)
{
SendMessage(GetDlgItem(hDlg, IDL_LISTBOX),
LB_ADDSTRING, 0, (LONG) achKey);
}
}
SetCursor(LoadCursor (NULL, IDC_ARROW));

// Enumerate the key values.

SetCursor(LoadCursor(NULL, IDC_WAIT));

if (cValues)
{
for (j = 0, retValue = ERROR_SUCCESS;
j < cValues; j++)
{
cchValue = MAX_VALUE_NAME;
achValue[0] = '\0';
retValue = RegEnumValue(hKey, j, achValue,
&cchValue,
NULL,
NULL, // &dwType,
NULL, // &bData,
NULL); // &bcData

if (retValue == (DWORD) ERROR_SUCCESS )

{
achBuff[0] = '\0';

// Add each value to a list box.

if (!lstrlen(achValue))
lstrcpy(achValue, "<NO NAME>");
wsprintf(achBuff, "%d) %s ", j, achValue);
SendMessage(GetDlgItem(hDlg,IDL_LISTBOX2),
LB_ADDSTRING, 0, (LONG) achBuff);
}
}

SetCursor(LoadCursor(NULL, IDC_ARROW));
}

2.21.20 Retrieving the Last-Write Time

The following example uses the GetFileTime function to retrieve the last-write time for a file.
It converts the time to local time based on the current time-zone settings, and creates a date
and time string that can be shown to the user.
BOOL GetLastWriteTime(HANDLE hFile, LPSTR String)
{
FILETIME ftCreate, ftAccess, ftWrite;
SYSTEMTIME stUTC, stLocal;

// Retrieve the file times for the file.

if (!GetFileTime(hFile, &ftCreate, &ftAccess, &ftWrite))
 Some Coding Tips 225

return FALSE;

// Convert the last-write time to local time.

FileTimeToSystemTime(&ftWrite, &stUTC);
SystemTimeToTzSpecificLocalTime(NULL, &stUTC, &stLocal);

// Build a string showing the date and time.

sprintf(String, "%02d/%02d/%d %02d:%02d",
stLocal.wDay, stLocal.wMonth, stLocal.wYear,
stLocal.wHour, stLocal.wMinute);

return TRUE;
}

2.21.21 Setting the System Time

The following example sets the system time using the SetSystemTime function.
BOOL SetNewTime(WORD hour, WORD minutes)
{
SYSTEMTIME st;
char *pc;

GetSystemTime(&st); // gets current time

st.wHour = hour; // adjusts hours
st.wMinute = minutes; // and minutes
if (!SetSystemTime(&st)) // sets system time
return FALSE;
return TRUE;
}

2.21.22 Changing a File Time to the Current Time

The following example sets the last-write time for a file to the current system time using the
SetFileTime function.
BOOL SetFileToCurrentTime(HANDLE hFile)
{
FILETIME ft;
SYSTEMTIME st;
BOOL f;

GetSystemTime(&st); // gets current time

SystemTimeToFileTime(&st, &ft); // converts to file time format
f = SetFileTime(hFile, // sets last-write time for file
(LPFILETIME) NULL, (LPFILETIME) NULL, &ft);

return f;
}

2.22 Displaying the amount of disk space for each drive

#include <windows.h>
#include <stdio.h>
// This program will loop through all drives in the system and will print the
// capacity of each drive, the number of bytes free/used, and the percentage free.
// At the end of the loop it will print the totals.
int main(void)
{
long long BytesAvailable,TotalBytesAvailable=0;
long long capacity,TotalCapacity=0;
226 C Tutorial

long long userFree,TotalUserFree=0;

long long used,TotalUsed=0;
long double percent;
int counter = 'C'; // Start with C drive ignoring floppies
char diskname[512];

strcpy(diskname,"C:\\");
printf("%-6s %15s %15s %15s %6s\n","Drive", "Capacity",
"Available","Used","Free");
while (counter != (1+'Z')) {
diskname[0] = counter;
if (GetDiskFreeSpaceEx(diskname, &BytesAvailable,
&capacity,&userFree)) {
percent = 100.0L*(((long double)BytesAvailable)/(long
double)capacity);
used = capacity-BytesAvailable;
/*
printf formats:
%-6s format string in 6 position, left justified (negative width)
%15'lld format 64 bit in 15 positions separating digits in groups (')
%6.2Lf format long double (Lf) in 6 positions with 2 decimals
*/
printf("%-6s %15'lld %15'lld %15'lld %6.2Lf%%\n",
diskname,capacity,BytesAvailable,used,percent);
TotalBytesAvailable+=BytesAvailable;
TotalCapacity+=capacity;
TotalUsed+=used;
}
counter++;
}
// Now print the totals
percent = 100.0L*(((long double)TotalBytesAvailable)/(long
double)TotalCapacity);
printf("\n%-6s %15'lld %15'lld %15'lld %6.2Lf%%\n", "Total:",
TotalCapacity,TotalBytesAvailable,TotalUsed,percent);
return 0;
}
The output of this program can look like this:
Drive Capacity Available Used Free
C:\ 6,292,303,872 2,365,452,288 3,926,851,584 37.59%
D:\ 10,487,197,696 3,563,794,432 6,923,403,264 33.98%
E:\ 31,453,437,952 17,499,627,520 13,953,810,432 55.64%
F:\ 15,726,731,264 10,327,638,016 5,399,093,248 65.67%
H:\ 569,366,528 0 569,366,528 0.00%
I:\ 31,790,673,920 27,672,530,944 4,118,142,976 87.05%

Total: 96,319,711,232 61,429,043,200 34,890,668,032 63.78%

Since it doesn’t check if a drive is a CD drive, drive H: is listed, and it has obviously 0
bytes available since it is read only.

2.23 FAQ

2.23.1 How do I create a progress report with a Cancel button?

Here is the answer from the MSDN Knowledge base, article Q76415
 FAQ 227

The following article describes the mechanisms necessary to implement a progress or activity
indicator with a Cancel Operation option, to be used for lengthy and CPU-intensive subtasks
of an application. Examples of operations using this include: copying multiple files, directory
searches, or printing large files.
The progress dialog box is implemented in two steps:
• Initialize the dialog box before starting lengthy or CPU intensive subtask.
• After each unit of the subtask is complete, call ProgressYield() to determine if the user
has canceled the operation and to update the progress or activity indicator.
This is the description of the progress dialog procedure. The procedure uses a global variable
(Cancel) to inform the CPU-intensive subtask that the user has indicated a desire to terminate
the subtask.
WORD Cancel = FALSE; /* This must be global to all modules */
/* which call ProgressYield() */
BOOL FAR PASCAL ProgressDlgProc(hDlg, message, wParam, lParam)
HWND hDlg;
unsigned message;
WORD wParam;
DWORD lParam;
{
switch (message)
{
/* Use other messages to update the progress or activity */
/* indicator. */
.
case WM_COMMAND:
switch (wParam)
{
case ID_CANCEL: /* ID_CANCEL = 2 */
Cancel = TRUE;

default:
return FALSE;
}
.
.
default:
return FALSE;
}

}
The following describes the ProgressYield procedure, which should be called after each unit
of the CPU-intensive subtask is completed. The ProgressYield procedure uses the IsDialog-
Message function (described in the "Microsoft Windows Software Development Kit Refer-
ence Volume 1"). IsDialogMessage will convert keyboard messages into selection commands
for the corresponding dialog box.
void ProgressYield(HWND hwnd)
{
MSG msg;

/* Remove all available messages for any window that belong */

/* to the current application. */
while (PeekMessage(&msg, NULL, 0, 0, PM_REMOVE))
{
/* Translate and Dispatch the given message if the window */
/* handle is null or the given message is not for the */
228 C Tutorial

/* modeless dialog box hwnd. */

if (!hwnd || !IsDialogMessage(hwnd, &msg))
{
TranslateMessage(&msg);
DispatchMessage(&msg);
}
}
}
The following describes how to incorporate the progress dialog as part of an application's sub-
task that is CPU-intensive. The PROGRESS_DLG resource should contain a button with an
ID of 2, because this is the wParam of the WM_COMMAND that will be sent when the user
presses the ESC key. The button should also have the BS_DEFPUSHBUTTON style so that
the ENTER key will also result in the termination of the CPU-intensive subtask.
FARPROC lpProgressProc;
HWND hwndProgress; /* This needs to be global if */
/* accessed by other modules. */
.
.
/* Initialize before starting CPU-intensive work. */
lpProgressProc = MakeProcInstance(ProgressDlgProc,
hInst); /* Current instance. */
hwndProgress = CreateDialog(hInst, /* Current instance. */
"PROGRESS_DLG", /* Resource. */
hwndParent, /* Parent handle. */
lpProgressProc);/* Instance address. */
ShowWindow(hwndProgress);
.
.
/* Start CPU intensive work here. */
.
.
/* Before or after each unit of work, the application */
/* should do the following: */
ProgressYield(hwndProgress);
if (Cancel == TRUE)
break; /* Terminate CPU-intensive work immediately. */
.
.
/* End CPU-intensive work here. */
.
.
DestroyWindow(hwndProgress);
FreeProcInstance(lpProgressProc);
.

2.23.2 How do I show in the screen a print preview?

If a screen font is available that exactly matches (or at least very closely corresponds to) the
chosen printer font, then the process is very straightforward and consists of seven steps:
 FAQ 229

• Retrieve a Device Context (DC) or an Information Context (IC) for the printer.
• Call EnumFontFamilies() to obtain a LOGFONT structure for the selected printer font.
The nFontType parameter to the EnumFontFamilies() callback function specifies if a
given font is a device font.
• Get a DC for the screen.
• Convert the lfHeight and lfWidth members of the LOGFONT structure from printer
resolution units to screen resolution units. If a mapping mode other than MM_TEXT is
used, round-off error may occur.
• Call CreateFontIndirect() with the LOGFONT structure.
• Call SelectObject(). GDI will select the appropriate screen font to match the printer font.
• Release the printer device context or information context and the screen device context.
If a screen font that corresponds to the selected printer font is not available, the process is
more difficult. It is possible to modify the character placement on the screen to match the
printer font to show justification, line breaks, and page layout. However, visual similarity
between the printer fonts and screen fonts depends on a number of factors, including the num-
ber and variety of screen fonts available, the selected printer font, and how the printer driver
describes the font. For example, if the printer has a serifed Roman- style font, one of the GDI
serifed Roman-style fonts will appear to be very similar to the printer font. However, if the
printer has a decorative Old English-style font, no corresponding screen font will typically be
available. The closest available match would not be very similar.
To have a screen font that matches the character placement of a printer font, do the following:
• Perform the preceding seven steps to retrieve an appropriate screen font.
• Get the character width from the TEXTMETRIC structure returned by the EnumFonts
function in step 2 above. Use this information to calculate the page position of each
character to be printed in the printer font.
• Allocate a block of memory and specify the spacing between characters. Make sure that
this information is in screen resolution units.
• Specify the address of the memory block as the lpDx parameter to ExtTextOut(). GDI
will space the characters as listed in the array.

2.23.3 How do I change the color of an edit field?

See page 193.

2.23.4 How do I draw a transparent bitmap?

A portion of a graphic image that does not change the contents of the screen is termed "trans-
parent." The DrawIcon function can create an image that contains transparent portions. It is
also possible to obtain this functionality using the BitBlt function; however, there are some
additional steps involved.
The first step is to obtain the contents of the area where the bitmap will be drawn and to store
this background image in a memory display context (DC). Mask out the area of the back-
ground that corresponds to the nontransparent portion of the image bitmap and mask out all
transparent pixels from the image bitmap. Use the XOR raster operation to merge the image
bitmap into the background bitmap. Finally, use the BitBlt function to move the merged image
to the destination DC.
230 C Tutorial

The following nine steps describe a process used to draw transparent bitmaps:
• Create a DC to hold the image bitmap.
• Select the image bitmap into the DC.
• Create a memory DC to hold the final image. This is the destination DC.
• Copy the portion of the screen that will be covered by the image into the destination DC.
• Create an "AND mask" that contains the mask of the colors to draw (the nontransparent
portions of the image). To do this, perform the following three steps:
• Set the background color of the image DC to the color that will be transparent in the
image.
• Create a monochrome DC.
• BitBlt the image into the monochrome DC. This will create an AND mask of the bitmap
by setting pixels that match the background color to white (1), and setting all other
pixels to black (0).
• Use BitBlt with the SRCAND raster operation code to copy the AND mask onto the
destination DC. Use BitBlt with the SRCAND raster operation code to copy the inverse
of the AND mask onto the image DC. Use BitBlt with the SRCPAINT raster operation
code to copy the image DC onto the destination DC. Use BitBlt to copy the contents of
the destination DC to the appropriate portion of the screen.
The following code is a function that demonstrates the preceding steps:
void DrawTransparentBitmap(HDC hdc, HBITMAP hBitmap, short xStart,
short yStart, COLORREF cTransparentColor)
{
BITMAP bm;
COLORREF cColor;
HBITMAP bmAndBack, bmAndObject, bmAndMem, bmSave;
HBITMAP bmBackOld, bmObjectOld, bmMemOld, bmSaveOld;
HDC hdcMem, hdcBack, hdcObject, hdcTemp, hdcSave;
POINT ptSize;

hdcTemp = CreateCompatibleDC(hdc);
SelectObject(hdcTemp, hBitmap); // Select the bitmap

GetObject(hBitmap, sizeof(BITMAP), (LPSTR)&bm);

ptSize.x = bm.bmWidth; // Get width of bitmap
ptSize.y = bm.bmHeight; // Get height of bitmap
DPtoLP(hdcTemp, &ptSize, 1); // Convert from device

// to logical points

// Create some DCs to hold temporary data.

hdcBack = CreateCompatibleDC(hdc);
hdcObject = CreateCompatibleDC(hdc);
hdcMem = CreateCompatibleDC(hdc);
hdcSave = CreateCompatibleDC(hdc);

// Create a bitmap for each DC. DCs are required for a number of
// GDI functions.

// Monochrome DC
bmAndBack = CreateBitmap(ptSize.x, ptSize.y, 1, 1, NULL);

// Monochrome DC
 FAQ 231

bmAndObject = CreateBitmap(ptSize.x, ptSize.y, 1, 1, NULL);

bmAndMem = CreateCompatibleBitmap(hdc, ptSize.x, ptSize.y);

bmSave = CreateCompatibleBitmap(hdc, ptSize.x, ptSize.y);

// Each DC must select a bitmap object to store pixel data.

bmBackOld = SelectObject(hdcBack, bmAndBack);
bmObjectOld = SelectObject(hdcObject, bmAndObject);
bmMemOld = SelectObject(hdcMem, bmAndMem);
bmSaveOld = SelectObject(hdcSave, bmSave);

// Set proper mapping mode.

SetMapMode(hdcTemp, GetMapMode(hdc));

// Save the bitmap sent here, because it will be overwritten.

BitBlt(hdcSave, 0, 0, ptSize.x, ptSize.y, hdcTemp, 0, 0, SRCCOPY);

// Set the background color of the source DC to the color. contained in the parts of
// the bitmap that should be transparent
cColor = SetBkColor(hdcTemp, cTransparentColor);

// Create the object mask for the bitmap by performing a BitBlt

// from the source bitmap to a monochrome bitmap.
BitBlt(hdcObject, 0, 0, ptSize.x, ptSize.y, hdcTemp, 0, 0,SRCCOPY);

// Set the background color of the source DC back to the original color.
SetBkColor(hdcTemp, cColor);

// Create the inverse of the object mask.

BitBlt(hdcBack, 0, 0, ptSize.x, ptSize.y, hdcObject, 0,
0,NOTSRCCOPY);

// Copy the background of the main DC to the destination.

BitBlt(hdcMem, 0, 0, ptSize.x, ptSize.y, hdc, xStart,
yStart,SRCCOPY);

// Mask out the places where the bitmap will be placed.

BitBlt(hdcMem, 0, 0, ptSize.x, ptSize.y, hdcObject, 0, 0, SRCAND);

// Mask out the transparent colored pixels on the bitmap.

BitBlt(hdcTemp, 0, 0, ptSize.x, ptSize.y, hdcBack, 0, 0, SRCAND);

// XOR the bitmap with the background on the destination DC.

BitBlt(hdcMem, 0, 0, ptSize.x, ptSize.y, hdcTemp, 0, 0, SRCPAINT);

// Copy the destination to the screen.

BitBlt(hdc, xStart, yStart, ptSize.x, ptSize.y, hdcMem, 0,
0,SRCCOPY);

// Place the original bitmap back into the bitmap sent here.
BitBlt(hdcTemp, 0, 0, ptSize.x, ptSize.y, hdcSave, 0, 0, SRCCOPY);

// Delete the memory bitmaps.

DeleteObject(SelectObject(hdcBack, bmBackOld));
DeleteObject(SelectObject(hdcObject, bmObjectOld));
DeleteObject(SelectObject(hdcMem, bmMemOld));
DeleteObject(SelectObject(hdcSave, bmSaveOld));

// Delete the memory DCs.

DeleteDC(hdcMem);
232 C Tutorial

DeleteDC(hdcBack);
DeleteDC(hdcObject);
DeleteDC(hdcSave);
DeleteDC(hdcTemp);
}
The following is an example of how the DrawTransparentBitmap function might be called:
DrawTransparentBitmap(hdc, // The destination DC.

hBitmap, // The bitmap to be drawn.

xPos, // X coordinate.
yPos, // Y coordinate.
0x00FFFFFF); // The color for transparent
// pixels (white, in this
// example).

2.23.5 How do I draw a gradient background?

/* DrawBackgroundPattern() *
Purpose: This function draws a gradient pattern that transitions between blue and black.
This is similar to the background used in Microsoft setup programs. */
void DrawBackgroundPattern(HWND hWnd)
{
HDC hDC = GetDC(hWnd); // Get the DC for the window
RECT rectFill; // Rectangle for filling band
RECT rectClient; // Rectangle for entire client area
float fStep; // How large is each band?
HBRUSH hBrush;
int iOnBand; // Loop index

// How large is the area you need to fill?

GetClientRect(hWnd, &rectClient);

// Determine how large each band should be in order to cover the client with 256
// bands (one for every color intensity level)
fStep = (float)rectClient.bottom / 256.0f;

// Start filling bands

for (iOnBand = 0; iOnBand < 256; iOnBand++) {
// Set the location of the current band
SetRect(&rectFill,
0, // Upper left X
(int)(iOnBand * fStep), // Upper left Y
rectClient.right+1, // Lower right X
(int)((iOnBand+1) * fStep)); // Lower right Y

// Create a brush with the appropriate color for this band

hBrush = CreateSolidBrush(RGB(0, 0, (255 - iOnBand)));
// Fill the rectangle
FillRect(hDC, &rectFill, hBrush);

// Get rid of the brush you created

DeleteObject(hBrush);
}
// Give back the DC
ReleaseDC(hWnd, hDC);
}
 FAQ 233

2.23.6 How do I calculate print margins?

An application can determine printer margins as follows:
Calculate the left and top margins
• Determine the upper left corner of the printable area calling GetDeviceCaps() with the
PHYSICALOFFSETX and PHYSICALOFFSETY indices. For example:
// Init our pt struct in case escape not supported
pt.x = 0; pt.y = 0;
// Locate the upper left corner of the printable area
pt.x = GetDeviceCaps(hPrnDC, PHYSICALOFFSETX);
pt.y = GetDeviceCaps(hPrnDC, PHYSICALOFFSETY);
Determine the number of pixels required to yield the desired margin (x and y offsets) by
calling GetDeviceCaps() using the LOGPIXELSX and LOGPIXELSY flags.
/* Figure out how much you need to offset output to produce the left and top margins for the
output in the printer. Note the use of the "max" macro. It is possible that you are asking for
margins that are not possible on this printer. For example, the HP LaserJet has a 0.25"
unprintable area so we cannot get margins of 0.1". */
xOffset = max (0, GetDeviceCaps (hPrnDC, LOGPIXELSX) *
nInchesWeWant - pt.x);
yOffset = max (0, GetDeviceCaps (hPrnDC, LOGPIXELSY) *
nInchesWeWant - pt.y);
/* When doing all the output, you can either offset it by the above values or call SetViewportOrg()
to set the point (0,0) at the margin offset you calculated.*/
SetViewportOrg (hPrnDC, xOffset, yOffset); //all other output here
• Calculate the bottom and right margins. Obtain the total size of the physical page
(including printable and unprintable areas) calling GetDeviceCaps() with the
PHYSICALWIDTH and PHYSICALHEIGHT indices in Windows NT.
• Determine the number of pixels required to yield the desired right and bottom margins
by calling GetDeviceCaps using the LOGPIXELSX and LOGPIXELSY flags.
• Calculate the size of the printable area with GetDeviceCaps() using the HORZRES and
VERTRES flags. The following code fragment illustrates steps a through c:
// Get the size of the printable area
pt.x = GetDeviceCaps(hPrnDC, PHYSICALWIDTH);
pt.y = GetDeviceCaps(hPrnDC, PHYSICALHEIGHT);

xOffsetOfRightMargin = xOffset +
GetDeviceCaps (hPrnDC, HORZRES) -
pt.x -
GetDeviceCaps (hPrnDC, LOGPIXELSX) *
wInchesWeWant;

yOffsetOfBottomMargin = yOffset +
GetDeviceCaps (hPrnDC, VERTRES) -
pt.y -
GetDeviceCaps (hPrnDC, LOGPIXELSY) *
wInchesWeWant;
NOTE: Now, you can clip all output to the rectangle bounded by xOffset, yOffset, xOffsetOf-
RightMargin, and yOffsetOfBottomMargin.

2.23.7 How do I calculate the bounding rectangle of a string of

text?
/* Get a bounding rectangle for a string of text output to a specified coordinate in a DC using the
currently selected font NOTE: The reference DC must have a True Type font selected */
234 C Tutorial

BOOL GetTextBoundingRect(HDC hDC, // Reference DC

int x, // X-Coordinate
int y, // Y-Coordinate
LPSTR lpStr, // The text string to evaluate
DWORD dwLen, // The length of the string
LPRECT lprc) // Holds bounding rectangle
{
LPPOINT lpPoints;
LPBYTE lpTypes;
int i, iNumPts;

// Draw the text into a path

BeginPath(hDC);
i = SetBkMode(hDC, TRANSPARENT);
TextOut(hDC, x, y, lpStr, dwLen);
SetBkMode(hDC, i);
EndPath(hDC);

// How many points are in the path

iNumPts = GetPath(hDC, NULL, NULL, 0);
if (iNumPts == -1) return FALSE;

// Allocate room for the points

lpPoints = (LPPOINT)GlobalAlloc(GPTR, sizeof(POINT) * iNumPts);
if (!lpPoints) return FALSE;

// Allocate room for the point types

lpTypes = GlobalAlloc(GPTR, iNumPts);
if (!lpTypes) {
GlobalFree(lpPoints);
return FALSE;
}

// Get the points and types from the current path

iNumPts = GetPath(hDC, lpPoints, lpTypes, iNumPts);

// More error checking

if (iNumPts == -1) {
GlobalFree(lpTypes);
GlobalFree(lpPoints);
return FALSE;
}

// Intialize the rectangle

SetRect(lprc, 0xFFFFF, 0xFFFFF, 0, 0);
}

2.23.8 How do I close an open menu?

To cancel an active menu send a WM_CANCELMODE message to the window that owns it.
When you send the WM_CANCELMODE message to the owner window of an active menu,
the window exits the menu mode. You can use this technique to force the abandonment of
active TrackPopupMenu or TrackPopupMenuEx.

2.23.9 How do I center a dialog box in the screen?

To center a dialog box on the screen before it is visible, add the following lines to the process-
ing of the WM_INITDIALOG message:
 FAQ 235

{
RECT rc;

GetWindowRect(hDlg, &rc);

SetWindowPos(hDlg, NULL,
((GetSystemMetrics(SM_CXSCREEN) - (rc.right - rc.left)) / 2),
((GetSystemMetrics(SM_CYSCREEN) - (rc.bottom - rc.top)) / 2),
0, 0, SWP_NOSIZE | SWP_NOACTIVATE);
}
This code centers the dialog horizontally and vertically.

2.23.10 How do I create non-rectangular windows?

In previous versions of Windows and Windows NT, it was possible to create only rectangular
windows. To simulate a non-rectangular window required a lot of work on the application
developer's part. Besides handling all drawing for the window, the application was required to
perform hit-testing and force underlying windows to repaint as necessary to refresh the "trans-
parent" portions of the window.
Windows 95 and Windows NT version 3.51 greatly simplify this by providing the SetWin-
dowRgn function. An application can now create a region with any desired shape and use Set-
WindowRgn to set this as the clipping region for the window. Subsequent painting and mouse
messages are limited to this region, and Windows automatically updates underlying windows
that show through the non-rectangular window. The application need only paint the window as
desired.
For more information on using SetWindowRgn, see the Win32 API documentation.

2.23.11 How do I implement a non-blinking caret?

Although Windows is designed to blink the caret at a specified interval, a timer function and
SetCaretBlinkTime() can be used to prevent Windows from turning the caret off by following
these three steps:
• Call SetCaretBlinkTime(10000), which instructs Windows to blink the caret every
10,000 milliseconds (10 seconds). This results in a "round-trip" time of 20 seconds to go
from OFF to ON and back to OFF (or vice versa).
• Create a timer, using SetTimer(), specifying a timer procedure and a 5,000 millisecond
interval between timer ticks.
• In the timer procedure, call SetCaretBlinkTime(10000). This resets the timer in
Windows that controls the caret blink.
When an application implements this procedure, Windows never removes the caret from the
screen, and the caret does not blink.

2.23.12 How do I create a title window (splash screen)?

A "splash" screen (title screen) can be used by an application to display important information
about a program. The splash screen can also be used as an activity indicator during application
startup.
A splash screen is usually used when it is known that an application will take more than a sec-
ond or two to display its first UI elements. The splash screen gives the user visual feedback
that the application is starting. If a splash screen is not used, the user may assume that the start
command was not run properly, and the user may try to start the application again.
236 C Tutorial

To display a splash window, create an overlapped window. In the WndProc for this window,
intercept the WM_NCCALCSIZE message and return NULL (instead of the DefWindowProc
function), which prevents Microsoft Windows from redefining the window's maximum
default full-screen client area size to accommodate borders and caption bar.
This window can be created before the main application initializes. The main application can
send messages to this window to display its initialization information or a bitmap.
The following code shows one way to implement a splash screen window:
#define SZ_INIT TEXT("Initializing application...")
#define SZ_LOAD TEXT("Loading resources... ")
#define SZ_CLOSE TEXT("Closing splash window...")
#define SZ_SPLASH TEXT("Splash window")

#define ID_TIMER_CLOSE0x1111
#define ID_TIMER_INIT0x1112
#define ID_TIMER_LOAD0x1113
#define ID_TIMER_DONE0x1114

ATOM MyRegisterClass(HINSTANCE hInstance);

BOOL InitInstance(HINSTANCE hInstance, int nCmdShow);
LRESULT CALLBACK WndProc(HWND hWnd, UINT message, WPARAM wParam, LPARAM
lParam);

HINSTANCE hInst = NULL;

TCHAR SplashWndClass[28];
int APIENTRY WinMain(HINSTANCE hInstance,
HINSTANCE hPrevInstance,
LPSTR lpCmdLine,
int nCmdShow)
{
MSG msg;

lstrcpy(SplashWndClass,TEXT("SplashWindow"));
MyRegisterClass(hInstance);
// Perform application initialization:
if (!InitInstance (hInstance, nCmdShow))
{
return FALSE;
}

// Main message loop:

while (GetMessage(&msg, NULL, 0, 0))
{
TranslateMessage(&msg);
DispatchMessage(&msg);
}

return msg.wParam;
}

//
// FUNCTION: MyRegisterClass()
//
// PURPOSE: Registers the window class.
//
// COMMENTS:
//
// This function and its use is only necessary if you want
// this code to be compatible with Win32 systems prior to the
 FAQ 237

// 'RegisterClassEx' function that was added to Windows 95. It is

// important to call this function so that the application will
// get 'well formed' small icons associated with it.

ATOM MyRegisterClass(HINSTANCE hInstance)

{
WNDCLASSEX wcex;

wcex.cbSize = sizeof(WNDCLASSEX);
wcex.style = CS_HREDRAW | CS_VREDRAW;
wcex.lpfnWndProc = (WNDPROC)WndProc;
wcex.cbClsExtra = 0;
wcex.cbWndExtra = 0;
wcex.hInstance = hInstance;
wcex.hIcon = NULL;
wcex.hCursor = LoadCursor(NULL, IDC_ARROW);
wcex.hbrBackground = (HBRUSH)(COLOR_WINDOW+1);
wcex.lpszMenuName = NULL;
wcex.lpszClassName = SplashWndClass;
wcex.hIconSm = NULL;

return RegisterClassEx(&wcex);
}

//
// FUNCTION: InitInstance(HANDLE, int)
//
// PURPOSE: Saves instance handle and creates main window
//
// COMMENTS:
//
// In this function, we save the instance handle in a global
// variable and create and display the main program window.
//
BOOL InitInstance(HINSTANCE hInstance, int nCmdShow)
{
HWND hWnd;
RECT rect;
int splashwidth = 350;
int splashheight = 350;

hInst = hInstance; // Store instance handle in this global variable

SystemParametersInfo(SPI_GETWORKAREA, 0, (LPVOID) &rect, 0);

hWnd = CreateWindowEx(WS_EX_TOOLWINDOW,
SplashWndClass,
NULL,
WS_OVERLAPPED,
(rect.right - rect.left - splashwidth)/2,
(rect.bottom - rect.top - splashheight)/2,
splashwidth,
splashheight,
NULL,
NULL,
hInstance,
NULL);
if (!hWnd)
{
return FALSE;
238 C Tutorial

ShowWindow(hWnd, nCmdShow);
UpdateWindow(hWnd);

return TRUE;
}

//
// FUNCTION: WndProc(HWND, unsigned, WORD, LONG)
//
// PURPOSE: Processes messages for the main window.
//
// WM_COMMAND- Process the application menu
// WM_PAINT- Paint the main window
// WM_DESTROY- Post a quit message and return
//
//
LRESULT CALLBACK WndProc(HWND hWnd, UINT message, WPARAM wParam, LPARAM
lParam)
{
switch (message)
{
case WM_NCCALCSIZE: //CAPTURE THIS MESSAGE AND RETURN NULL
return NULL;
case WM_CREATE:
SetTimer(hWnd, ID_TIMER_INIT, 1000, NULL);
SetTimer(hWnd, ID_TIMER_LOAD, 2000, NULL);
SetTimer(hWnd, ID_TIMER_DONE, 4000, NULL);
SetTimer(hWnd, ID_TIMER_CLOSE, 5000, NULL);
break;
case WM_PAINT:
{
PAINTSTRUCT ps = { 0 };
RECT rc = { 0 };
HDC hDC = BeginPaint(hWnd, &ps);
GetClientRect(hWnd, &rc);
InflateRect(&rc, -2,-2);
Rectangle(hDC, rc.left, rc.top, rc.right, rc.bottom);
InflateRect(&rc, -15,-15);
HFONT hFont = CreateFont(-35,-35, 0, 0,0,0,0,
0,0,0,0,0,0,TEXT("Arial"));
HFONT hOldFont = (HFONT) SelectObject(hDC, hFont);
DrawText(hDC, SZ_SPLASH, lstrlen(SZ_SPLASH),
&rc, DT_WORDBREAK);
SelectObject(hDC, hOldFont);
EndPaint(hWnd, &ps);
}
break;
case WM_DESTROY:
PostQuitMessage(0);
break;
case WM_TIMER:
{
HDC hDC = GetDC(hWnd);
RECT rc = { 0 };
GetClientRect(hWnd, &rc);
KillTimer(hWnd, wParam);
switch (wParam)
{
 FAQ 239

case ID_TIMER_CLOSE:
DestroyWindow(hWnd);
break;
case ID_TIMER_INIT:
TextOut(hDC, rc.right-200, rc.bottom-20,
SZ_INIT, lstrlen(SZ_INIT));
break;
case ID_TIMER_LOAD:
TextOut(hDC, rc.right-200, rc.bottom-20,
SZ_LOAD, lstrlen(SZ_LOAD));
break;
case ID_TIMER_DONE:
TextOut(hDC, rc.right-200, rc.bottom-20,
SZ_CLOSE, lstrlen(SZ_CLOSE));
break;
}
ReleaseDC(hWnd, hDC);
}
break;
default:
return DefWindowProc(hWnd, message, wParam, lParam);
}
return 0;
}

2.23.13 How do I append text to an edit control?

Windows-based applications often use edit controls to display text. These applications some-
times need to append text to the end of an edit control instead of replacing the existing text.
There are two different ways to do this in Windows:
Use the EM_SETSEL and EM_REPLACESEL messages.
Use the EM_SETSEL message with the clipboard functions to append text to the edit control's
buffer.
The EM_SETSEL message can be used to place a selected range of text in a Windows edit
control. If the starting and ending positions of the range are set to the same position, no selec-
tion is made and a caret can be placed at that position. To place a caret at the end of the text in
a Windows edit control and set the focus to the edit control, do the following:
HWND hEdit = GetDlgItem (hDlg, ID_EDIT);
int ndx = GetWindowTextLength (hEdit);
SetFocus (hEdit);
SendMessage (hEdit, EM_SETSEL, (WPARAM)ndx, (LPARAM)ndx);
Once the caret is placed at end in the edit control, you can use the EM_REPLACESEL to
append text to the edit control. An application sends an EM_REPLACESEL message to
replace the current selection in an edit control with the text specified by the lpszReplace
(lParam) parameter. Because there is no current selection, the replacement text is inserted at
the current caret location. This example sets the selection to the end of the edit control and
inserts the text in the buffer:
SendMessage (hEdit, EM_SETSEL, (WPARAM)ndx, (LPARAM)ndx);
SendMessage (hEdit, EM_REPLACESEL, 0, (LPARAM) ((LPSTR)
szBuffer));
Another way to insert text into an edit control is to use the Windows clipboard. If the applica-
tion has the clipboard open or finds it convenient to open the clipboard, and copies the text
into the clipboard, then it can send the WM_PASTE message to the edit control to append text.
Of course, any data that was in the clipboard will be lost.
240 C Tutorial

Before sending the WM_PASTE message, the caret must be placed at the end of the edit con-
trol text using the EM_SETSEL message. Below is "pseudo" code that shows how to imple-
ment this method:
OpenClipBoard () ;
EmptyClipBoard() ;
SetClipBoardData() ;

SendMessage (hEdit, EM_SETSEL, (WPARAM)ndx, (LPARAM)ndx);

SendMessage (hEdit, WM_PASTE, 0, 0L);
This "pseudo" code appends text to the end of the edit control. Note that the data in the clip-
board must be in CF_TEXT format.

2.23.14 How do I spawn a process with redirected stdin and

stdout?
The CreateProcess() API through the STARTUPINFO structure enables you to redirect the
standard handles of a child console based process. If the dwFlags member is set to
STARTF_USESTDHANDLES, then the following STARTUPINFO members specify the
standard handles of the child console based process:

HANDLE hStdInput - Standard input handle of the child process.

HANDLE hStdOutput - Standard output handle of the child process.

HANDLE hStdError - Standard error handle of the child process.

You can set these handles to either a pipe handle, file handle, or any handle that can do syn-
chronous reads and writes through the ReadFile() and WriteFile() API. The handles must be
inheritable and the CreateProcess() API must specify that inheritable handles are to be inher-
ited by the child process by specifying TRUE in the bInheritHandles parameter. If the parent
process only wishes to redirect one or two standard handles, specifying GetStdHandle() for the
specific handles causes the child to create the standard handle as it normally would without
redirection. For example, if the parent process only needs to redirect the standard output and
error of the child process, then the hStdInput member of the STARTUPINFO structure is filled
as follows:
hStdInput = GetStdHandle(STD_INPUT_HANDLE);
NOTE: Child processes that use such C run-time functions as printf() and fprintf() can behave
poorly when redirected. The C run-time functions maintain separate IO buffers. When redi-
rected, these buffers might not be flushed immediately after each IO call. As a result, the out-
put to the redirection pipe of a printf() call or the input from a getch() call is not flushed
immediately and delays, sometimes-infinite delays occur. This problem is avoided if the child
process flushes the IO buffers after each call to a C run-time IO function. Only the child pro-
cess can flush its C run-time IO buffers. A process can flush its C run-time IO buffers by call-
ing the fflush() function.
NOTE: Windows 95 and Windows 98 require an extra step when you redirect the standard
handles of certain child processes.

2.23.15 How to modify the width of the list of a combo box

The combo box in Windows is actually a combination of two or more controls; that's why it's
called a "combo" box.
 FAQ 241

To make the combo box list wider or narrower, you need the handle of the list box control
within the combo box. This task is difficult because the list box is actually a child of the desk-
top window (for CBS_DROPDOWN and CBS_DROPDOWNLIST styles). If it were a child
of the ComboBox control, dropping down the list box would clip it to the parent, and it
wouldn't display.
A combo box receives WM_CTLCOLOR messages for its component controls when they
need to be painted. This allows the combo box to specify a color for these controls. The
HIWORD of the lParam in this message is the type of the control. In case of the combo box,
Windows sends it a WM_CTLCOLOR message with the HIWORD set to
CTLCOLOR_LISTBOX when the list box control needs to be painted. The LOWORD of the
lParam contains the handle of the list box control.
Once you obtain the handle to the list box control window, you can resize the control by using
the MoveWindow API.
The following code sample demonstrates how to do this. This sample assumes that you have
placed the combo box control in a dialog box.
LRESULT CALLBACK NewComboProc (HWND hWnd, UINT message, WPARAM
wParam, LPARAM lParam ); // prototype for the combo box subclass proc
HANDLE hInst; // Current app instance.
BOOL bFirst; // A flag.

// Dialog procedure for the dialog containing the combo box.

BOOL CALLBACK DialogProc(HWND hDlg, UINT message, WPARAM
wParam, LPARAM lParam)
{
FARPROC lpfnNewComboProc;
switch (message) {
case WM_INITDIALOG:
bFirst = TRUE; // Set flag here - see below for usage.
// Subclass the combo box.
lpfnOldComboProc = (FARPROC ) SetWindowLong (
GetDlgItem ( hDlg, IDC_COMBO1 ),
GWL_WNDPROC,
(LONG)NewComboProc );
break;
case WM_DESTROY:
(FARPROC ) SetWindowLong (GetDlgItem ( hDlg,
IDC_COMBO1 ),
GWL_WNDPROC,
(LONG)lpfnOldComboProc );
break;
default:
break;
}

return FALSE;

} // End dialog proc.

// Combobox subclass proc.

LRESULT CALLBACK NewComboProc (HWND hWnd, UINT message, WPARAM
wParam, LPARAM lParam );
{
static HWND hwndList;
static RECT rectList;

if ( WM_CTLCOLORLISTBOX == message ) // 32 bits has new message.

242 C Tutorial

{
// Is this message for the list box control in the combo?
// Do only the very first time, get the list
// box handle and the list box rectangle.
// Note the use of GetWindowRect, as the parent
// of the list box is the desktop window

if ( bFirst ) {
hwndList = (HWND) lParam ; // HWND is 32 bits.
GetWindowRect ( hwndList, &rectList );
bFirst = FALSE;
}
// Resize listbox window cx by 50 (use your size here).
MoveWindow ( hwndList, rectList.left, rectList.top,
( rectList.right - rectList.left + 50 ),
rectList.bottom - rectList.top, TRUE );
}
// Call original combo box procedure to handle other combo messages.
return CallWindowProc ( lpfnOldComboProc, hWnd, message,
wParam, lParam );
}

2.23.16 How do I modify environment variables permanently?

You can modify user environment variables by editing the following Registry key:
HKEY_CURRENT_USER\Environment
You can modify system environment variables by editing the following Registry key:
HKEY_LOCAL_MACHINE \
SYSTEM \
CurrentControlSet \
Control \
Session Manager \
Environment
Note that any environment variable that needs to be expanded (for example, when you use
%SYSTEM%) must be stored in the registry as a REG_EXPAND_SZ registry value. Any val-
ues of type REG_SZ will not be expanded when read from the registry.
Note that RegEdit.exe does not have a way to add REG_EXPAND_SZ. Use RegEdt32.exe
when editing these values manually.
However, note that modifications to the environment variables do not result in immediate
change. For example, if you start another Command Prompt after making the changes, the
environment variables will reflect the previous (not the current) values. The changes do not
take effect until you log off and then log back on.
To effect these changes without having to log off, broadcast a WM_SETTINGCHANGE mes-
sage to all windows in the system, so that any interested applications (such as Program Man-
ager, Task Manager, Control Panel, and so forth) can perform an update.
For example, on Windows NT, the following code fragment should propagate the changes to
the environment variables used in the Command Prompt:
SendMessageTimeout(HWND_BROADCAST, WM_SETTINGCHANGE, 0,
(LPARAM) "Environment", SMTO_ABORTIFHUNG,
5000, &dwReturnValue);
None of the applications that ship with Windows 95, including Program Manager and the
shell, respond to this message. Thus, while this article can technically be implemented on
Windows 95, there is no effect except to notify third-party applications. The only method of
 FAQ 243

changing global environment variables on Windows 95 is to modify the autoexec.bat file and
reboot.

2.23.17 How do I translate between dialog units and pixels?

When a application dynamically adds a child window to a dialog box, it may be necessary to
align the new control with other controls that were defined in the dialog box's resource tem-
plate in the RC file. Because the dialog box template defines the size and position of a control
in dialog-box units rather than in screen units (pixels), the application must translate dialog-
box units to screen units to align the new child window.
An application can use the following two methods to translate dialog- box units to screen
units:
The MapDialogRect function provides the easier method. This function converts dialog-box
units to screen units automatically.
For more details on this method, please see the documentation for the MapDialogRect func-
tion in the Microsoft Windows Software Development Kit (SDK).
Use the GetDialogBaseUnits function to retrieve the size of the dialog base units in pixels. A
dialog unit in the x direction is one-fourth of the width that GetDialogBaseUnits returns. A
dialog unit in the y direction is one-eighth of the height that the function returns.
For more details on this method, see the documentation for the GetDialogBaseUnits function
in the Windows documentation of lcc-win32.

2.23.18 How do I translate between client coordinates to screen

coordinates?
To determine the screen coordinates for the client area of a window, call the ClientToScreen
function to translate the client coordinates returned by GetClientRect into screen coordinates.
The following code demonstrates how to use the two functions together:
RECT rMyRect;

GetClientRect(hwnd, (LPRECT)&rMyRect);
ClientToScreen(hwnd, (LPPOINT)&rMyRect.left);
ClientToScreen(hwnd, (LPPOINT)&rMyRect.right);

2.23.19 When should I use critical sections and when is a mutex

more appropiate?
Critical sections and mutexes provide synchronization that is very similar, except that critical
sections can be used only by the threads of a single process. There are two areas to consider
when choosing which method to use within a single process:
Speed. The Synchronization overview says the following about critical sections:
• Critical section objects provide a slightly faster, more efficient mechanism for mutual-
exclusion synchronization.
• Critical sections use a processor-specific test and set instruction to determine mutual
exclusion.
Deadlock. The Synchronization overview says the following about mutexes:
244 C Tutorial

• If a thread terminates without releasing its ownership of a mutex object, the mutex is
considered to be abandoned. A waiting thread can acquire ownership of an abandoned
mutex, but the wait function's return value indicates that the mutex is abandoned.
• WaitForSingleObject() will return WAIT_ABANDONED for a mutex that has been
abandoned. However, the resource that the mutex is protecting is left in an unknown
state.
• There is no way to tell whether a critical section has been abandoned.

2.23.20 Why is my call to CreateFile failing when I use conin$ or

conout$?
If you attempt to open a console input or output handle by calling the CreateFile() function
with the special CONIN$ or CONOUT$ filenames, this call will return
INVALID_HANDLE_VALUE if you do not use the proper sharing attributes for the fdw-
ShareMode parameter in your CreateFile() call. Be sure to use FILE_SHARE_READ when
opening "CONIN$" and FILE_SHARE_WRITE when opening "CONOUT$".
 Overview of lcc-win32’s documentation 245

2.24 Overview of lcc-win32’s documentation

The documentation of lcc-win32 comes in four files:

² manual.exe. This file contains the following documentation:
1.Tutorial.pdf. This document.
2.Manual.pdf. This is the user’s manual, where you will find information about how
the system is used, command line options, menu descriptions, how to use the
debugger, etc. It explains how to build a project, how to setup the compiler, each
compiler option, all that with all the details.
3.Lcc-win32.doc. This is a technical description for interested users that may want to
know how the system is built, how the programs that build it were designed, the
options I had when writing them, etc.
Other documentation is available as .hlp files shipped with the main file lccwin32.exe.
Specifically:
4.c-library.chm. Here are the standard library functions documentation, and the
online-help for the IDE.
The documentation of the windows API is distributed in a relatively large file called
win32hlp.exe. This is absolutely essential, unless you know it by heart… When installed, this
file will become
5.Win32.hlp. Here you will find the documentation of the windows API.
That file is not complete however. More documentation for the new features of Win32 can be
found in the win32apidoc.exe file, also in the lcc distribution site. When installed, that file
will install:
• Shelldoc.doc. This documents the windows shell, its interfaces, function definitions,
etc.
• Wininet.doc. This documents the TCP/IP subsystem for network programming.
• CommonControls.doc. This explains the new controls added to windows after 1995.

Note that Wedit will detect if the documentation is installed, and will allow you to see the doc-
umentation of any function just with pressing the F1 key. This is a nice feature, especially for
beginners. Install a full version if you aren’t an expert. A version without the documentation it
is a pain, since you have to go fishing for that information each time you want to call an API,
not a very exciting perspective.

2.25 Bibliography
Here are some books about C.

«C Unleashed»
Richard Heathfield, Lawrence Kirby et al.
Heavy duty book full of interesting stuff like structures, matrix arithmetic, genetic algorithms
and many more. The basics are covered too, with lists, queues, double linked lists, stacks, etc.
«Algorithms in C»
Robert Sedgewick.
246 C Tutorial

I have only the part 5, graph algorithms. For that part (that covers DAGs and many others) I
can say that this is a no-nonsense book, full of useful algorithms. The code is clear and well
presented.
«C a reference manual»
(Fifth edition) Samuel P Harbison Guy L Steele Jr.
If you are a professional that wants to get all the C langauge described in great detail this book
is for you. It covers the whole grammar and the standard library with each part of it described
in detail.
«The C programming language»
Brian W Kernighan, Dennis Ritchie. (second edition)
This was the first book about C that I got, and it is still a good read. With many exercises, it is
this tutorial in a better rendering...
«A retargetable C compiler: design and implementation»
Chris Fraser and Dave Hanson
This book got me started in this adventure. It is a book about compiler construction and not
really about the C language but if you are interested in knowing how lcc-win32 works this is
surely the place to start.
Safer C
Les Hatton
As we have seen in the section «Pitfalls of the C language», C is quite ridden with problems.
This book address how to avoid this problems and design and develop you work to avoid get-
ting bitten by them.

2.26 Newsgroups
Internet newsgroups are a great way of sharing information. There is an lcc newsgroup
comp.compilers.lcc. Here is a question that appeared in another interesting newsgroup:
comp.std.c that shows an interesting discussion:
From: serin_d@my-deja.com
I need to write an algorithm in C, to:

1) determine the most significant set bit

2) determine the least significant set bit
in a byte/int/long whatever. I am looking for an efficient algorithm that
does not involve iterating through every bit position individually.
Cheers
Serin
Many people answered, and the discussion about which algorithm to use was a very informa-
tive one.
From: pornin@bolet.ens.fr (Thomas Pornin)
Organization: Ecole Normale Superieure, Paris
Notwithstanding the problem of knowing the exact size of a type in standard
C (you would have better luck with unsigned types, by the way), use a
dichotomy:
/*
* Returns the least significant bit in the 32-bit value stored in x
 Newsgroups 247

* (return value from 0 to 31; 32 if no bit is set)

*/
int least_significant_set_bit(unsigned long x)111
{
int t = 0;
if (x & 65535UL == 0) { t += 16; x >>= 16; }
if (x & 255UL == 0) { t += 8; x >>= 8; }
if (x & 15UL == 0) { t += 4; x >>= 4; }
if (x & 3UL == 0) { t += 2; x >>= 2; }
if (x & 1UL == 0) { t += 1; x >>= 1; }
if (!x) t ++;
return t;
}
For completeness: comp.std.c is about the C standard. Your question would
be better addressed in comp.lang.c.112
Another participant posted a different version of this algorithm. Here it is:
From: "Douglas A. Gwyn" <DAGwyn@null.net>
There is no Standard C function for this (traditionally called "find first
one bit"). Followup has been set accordingly. To get that thread started
off, here is a scheme that you might consider:
For example assume a 64-bit word:
Set bit location accumulator to 0 and total mask to 0xFFFFFFFFFFFFFFFF.
Mask word with 0xFFFFFFFF00000000 to see if first one bit is in the left
half of the word; if so, add 32 to the bit location accumulator and update
total mask by ANDing with this mask, else update total mask by ANDing with
the complement (~) of this mask. (This first mask update step can be simpli-
fied by omitting the initialozations and just storing 32 or 0 and the appro-
priate mask.)If the whole word is 0, return a failure-to-find indication.
Mask with total mask and 0xFFFF0000FFFF0000 to see if " is in left half of
whatever half was just determined; if so add 16 to accumulator and update
total mask by ANDing with this mask, else update total mask by ANDing with
the complement (~) of this mask.
Mask with total mask and 0xFF00FF00FF00FF00 to see if " " " add 8 " "
....
Mask with 0xAAAAAAAAAAAAAAAA to see if is in odd # bit position; if so add
1 (last mask update is not necessary). The above can be done in a compact
loop, but since you're worried about efficiency the loop should be com-
pletely unrolled and the parenthesized optimizations made. Accumulator now
contains bit location (counting from right starting with 0). If you per-
formed the final mask update, the total mask is now the isolated first one
bit.
Of course, the last one bit can be found in a similar fashion.
Now that the general idea is exposed, try to find optimizations. For exam-
ple, instead of masking the original word with total mask and new mask each
time, update the original word by masking it with the new contribution to
the total mask and don't maintain a total mask variable at all.
--
comp.lang.c.moderated - moderation address: clcm@plethora.net
This is surely an improvement over the first algorithm, since the shifts are gone.

111. I have to disagree with Thomas. comp.lang.c is quite boring, full of empty discussions very often.
There are good discussions of course, but there is a lot more noise.
112. Yes, moderated groups have less noise.
248 C Tutorial

Another answer was the following:

From: Francis Glassborow <francis.glassborow@ntlworld.com>
Subject: Re: Most significant bit algorithm
Before giving any guidelines to a solution, note that this was the wrong
place to ask, you should have posted to comp.lang.c.moderated.113
The answers are likely to be different for the different size types. In
addition, some possible 'solutions' are subject knowing the endianess of
your system.
For an 8-bit byte, consider masking in the following order:
bits 7 & 0
bits 6 & 1
bits 5 & 2
bits 4 & 3
The first of these to be non-zero tells you that either one or both of the
bits you want to locate have been determined. Overall that is likely to pro-
vide little advantage over mask each bit and test.
For other types, as long as you know their layout in bytes (unsigned char)
use a union to map the value to an array of unsigned char. Now test the
individual chars against zero. The highest and lowest non-zero byte can now
be tested for the requisite bit. However if you really have a need for max-
imum efficiency and can sacrifice portability to this end, consider writing
an assembler code routine.114
Another contribution was:
From: Conor O'Neill <conor.oneill@aethos.co.uk>
Organization: Speaking for myself
Least significant bit is fairly easy:
unsigned int least_significant_bit(unsigned int x)115
{
return x & ~(x - 1);
}
A little bit of playing with examples written out in binary should convince
you that it is correct. It even works with x == 0 (i.e., it returns 0). I
don't know if there is a similar algorithm for the most significant bit.
I don’t either. But it is surely correct I tested it.
Another answer was:
From: "Peter L. Montgomery" <Peter-Lawrence.Montgomery@cwi.nl>
Organization: CWI, Amsterdam
In article dkeisen@best.com () writes:
>Mathew Hendry <math@vissci.com> wrote:
>>serin_d@my-deja.com wrote:
>>: I need to write an algorithm in C, to [...] determine the least
>In 15 years of programming, I've never had to do anything
>like this. Not once.
>Sure looks to me like y'all did someone's homework for him.

113. See the lcc-win32 solution at the end of the discussion.

114. The lcc compiler needs to know if a given number is a power of two. It uses this construct too.
115.
 Newsgroups 249

I write number theoretic codes. Locating the rightmost and leftmost signif-
icant bits in a nonzero word are two of the 20 or so important primitives
needed for more complicated algorithms. If a programming language supports
“AND” and “OR” on bits, it should support these primitives too. Alas, few
do.
For example, the left-to-right binary method of exponentiation starts with
the most significant bit of a number and proceeds downward.
A more complicated example is a binary GCD (greatest common divisor).
Assume uint64 is an unsigned 64-bit type:
uint64 GCD(uint64 x, uint64 y){
int nzero;
uint64 x1, y1;
if (x == 0 || y == 0) return x | y;
nzero = trailing_zero_count(x | y); /* Shared power of 2 */
x1 = x >> trailing_zero_count(x);
y1 = y >> trailing_zero_count(y);
while (x1 != y1) { /* Both are odd */
if (x1 > y1) {
x1 = (x1 - y1) >> trailing_zero_count(x1 - y1);
} else {
y1 = (y1 - x1) >> trailing_zero_count(y1 - x1);
}
}
return x1 << nzero;
}
This short program has five references to a function locating the least sig-
nificant bit. No wonder that processors such as the Intel Pentium and Alpha
21264 have hardware instructions to locate the rightmost and leftmost sig-
nificant bits in a word. Optimizing these is much more than homework.
Now, you surely are wondering, what is the solution in lcc-win32?
Easy:

#include <intrinsics.h>
int getmsb(unsigned int n) /* returns position of most significant bit
*/
{
return _bsf(n);
}
int getlsb(unsigned int n) /* returns position of least significant bit
*/
{
return _bsr(n);
}
The functions _bsf (bit scan forward) and _bsr (bit scan reverse) are intrinsics. Those func-
tions will be mapped by the compiler directly into assembly, in this case to the machine
instructions BSF and BSR that all Intel compatible PCs have. This is extremely efficient, since
they will just issue:
bsr eax,eax
The number to be searched will be placed in the eax register by the compiler, and the machine
will leave the result in it. This is conceptually the same as calling a function, but takes just a
few cycles, depending on the bit pattern of the number.
Of course, instead of writing a cover fuinction you can use the intrinsics themselves to avoid
the overhead of a function call and a return.
250 C Tutorial

2.27 Appendix. Code listings

2.27.1 The window tree.

WindowTree.c
#include <windows.h>
#include <windowsx.h>
#include <commctrl.h>
#include <string.h>
#include <psapi.h>
#include "windowtreeres.h"
#define IDTREEWINDOW 10545
HINSTANCE hInst;// Instance handle
HWND hwndMain;//Main window handle

LRESULT CALLBACK MainWndProc(HWND hwnd,UINT msg,WPARAM wParam,LPARAM

lParam);

// Global Variables for the status bar control.

HWND hWndStatusbar;

static char * PrintProcessNameAndID( DWORD processID )

{
static char szProcessName[MAX_PATH];
HMODULE hMod;
DWORD cbNeeded;

HANDLE hProcess = OpenProcess( PROCESS_QUERY_INFORMATION |

PROCESS_VM_READ,
FALSE, processID );
szProcessName[0] = 0;
if ( hProcess ) {
if ( EnumProcessModules( hProcess, &hMod, sizeof(hMod),
&cbNeeded) ) {
GetModuleBaseName( hProcess, hMod, szProcessName,
sizeof(szProcessName) );
}
CloseHandle( hProcess );
}
return szProcessName;
}

void UpdateStatusBar(LPSTR lpszStatusString, WORD partNumber, WORD

displayFlags)
{
SendMessage(hWndStatusbar,
SB_SETTEXT,
partNumber | displayFlags,
(LPARAM)lpszStatusString);
}

void InitializeStatusBar(HWND hwndParent,int nrOfParts)

{
const int cSpaceInBetween = 8;
int ptArray[40];
RECT rect;

GetClientRect(hwndParent, &rect);
ptArray[nrOfParts-1] = rect.right;
 Appendix. Code listings 251

SendMessage(hWndStatusbar,
SB_SETPARTS,
nrOfParts,
(LPARAM)(LPINT)ptArray);
}

static BOOL CreateSBar(HWND hwndParent,char *initialText,int nrOfParts)

{
hWndStatusbar = CreateStatusWindow(WS_CHILD | WS_VISIBLE |
WS_BORDER|SBARS_SIZEGRIP,
initialText,
hwndParent,
IDM_STATUSBAR);
if(hWndStatusbar)
{
InitializeStatusBar(hwndParent,nrOfParts);
return TRUE;
}

return FALSE;
}

static BOOL InitApplication(void)

{
WNDCLASS wc;

memset(&wc,0,sizeof(WNDCLASS));
wc.style = CS_HREDRAW|CS_VREDRAW |CS_DBLCLKS ;
wc.lpfnWndProc = (WNDPROC)MainWndProc;
wc.hInstance = hInst;
wc.hbrBackground = (HBRUSH)(COLOR_WINDOW+1);
wc.lpszClassName = "windowtreeWndClass";
wc.lpszMenuName = MAKEINTRESOURCE(IDMAINMENU);
wc.hCursor = LoadCursor(NULL,IDC_ARROW);
wc.hIcon = LoadIcon(NULL,IDI_APPLICATION);
if (!RegisterClass(&wc))
return 0;
/*@@0<-@@*/
// ---TODO--- Call module specific initialization routines here

return 1;
}

HWND CreatewindowtreeWndClassWnd(void)
{
return CreateWindow("windowtreeWndClass","windowtree",

WS_MINIMIZEBOX|WS_VISIBLE|WS_CLIPSIBLINGS|WS_CLIPCHILDREN|WS_MAXIMIZEB
OX|WS_CAPTION|WS_BORDER|WS_SYSMENU|WS_THICKFRAME,
CW_USEDEFAULT,0,CW_USEDEFAULT,0,
NULL,
NULL,
hInst,
NULL);
}

void Scan(HWND hTree,HTREEITEM hTreeParent,HWND Start)

{
HWND hwnd = Start,hwnd1;
252 C Tutorial

TV_INSERTSTRUCT TreeCtrlItem;
HTREEITEM htiNewNode;

while (hwnd != NULL) {

char bufTxt[256],bufClassName[256],Output[1024];
SendMessage(hwnd,WM_GETTEXT,250,(LPARAM) bufTxt);
GetClassName(hwnd,bufClassName,250);
wsprintf(Output,"\"%s\" %s",bufTxt,bufClassName);
memset(&TreeCtrlItem,0,sizeof(TreeCtrlItem));
TreeCtrlItem.hParent = hTreeParent;
TreeCtrlItem.hInsertAfter = TVI_LAST;
TreeCtrlItem.item.mask = TVIF_TEXT | TVIF_PARAM;
TreeCtrlItem.item.pszText = (LPSTR) Output;
TreeCtrlItem.item.lParam = (LPARAM) hwnd;
htiNewNode = TreeView_InsertItem(hTree,&TreeCtrlItem);

if((hwnd1=GetWindow(hwnd,GW_CHILD))!=NULL)
Scan(hTree,htiNewNode,hwnd1);
hwnd=GetWindow(hwnd,GW_HWNDNEXT);

}
}

int BuildTree(HWND parent)

{
HWND Start = GetDesktopWindow();
HWND hTree = GetDlgItem(parent,IDTREEWINDOW);
TV_INSERTSTRUCT TreeCtrlItem;
HTREEITEM hNewNode;

SendMessage(hTree,WM_SETREDRAW,0,0);
TreeView_DeleteAllItems(hTree);
Start = GetWindow(Start,GW_CHILD);
memset(&TreeCtrlItem,0,sizeof(TreeCtrlItem));
TreeCtrlItem.hParent = TVI_ROOT;
TreeCtrlItem.hInsertAfter = TVI_LAST;
TreeCtrlItem.item.mask = TVIF_TEXT | TVIF_PARAM;
TreeCtrlItem.item.pszText = "Desktop";
hNewNode = TreeView_InsertItem(hTree,&TreeCtrlItem);
Scan(hTree,hNewNode,Start);
TreeView_Expand(hTree,hNewNode,TVE_EXPAND);
SendMessage(hTree,WM_SETREDRAW,1,0);
return 1;

}
void MainWndProc_OnCommand(HWND hwnd, int id, HWND hwndCtl, UINT
codeNotify)
{
switch(id) {
case IDM_NEW:
BuildTree(hwnd);
break;
case IDM_EXIT:
PostMessage(hwnd,WM_CLOSE,0,0);
break;
}
}
static HWND _stdcall CreateTree(HWND hWnd,int ID)
{
return CreateWindowEx(WS_EX_CLIENTEDGE,WC_TREEVIEW,"",
 Appendix. Code listings 253

WS_VISIBLE|WS_CHILD|WS_BORDER|TVS_HASLINES|
TVS_HASBUTTONS|TVS_DISABLEDRAGDROP,
0,0,0,0,
hWnd,(HMENU)ID,hInst,NULL);
}

void SetTextInStatusBar(HWND hParent,HWND hwnd)

{
RECT rc;
HANDLE pid,thread;
char info[4096];

GetWindowRect(hwnd,&rc);
thread = (HANDLE)GetWindowThreadProcessId(hwnd,&pid);
wsprintf(info,
"Handle: 0x%x %s, left %d, top %d, right %d, bottom %d, height
%d, width %d, Process: %s",
hwnd,
IsWindowVisible(hwnd)? "Visible" : "Hidden",
rc.left,rc.top,rc.right,rc.bottom,
rc.bottom-rc.top,rc.right-
rc.left,PrintProcessNameAndID((ULONG)pid));
UpdateStatusBar(info, 0, 0);
}
static HWND GetTreeItemInfo(HWND hwndTree,HTREEITEM hti)
{
TV_ITEM tvi;

memset(&tvi,0,sizeof(TV_ITEM));
tvi.mask = TVIF_PARAM;
tvi.hItem = hti;
TreeView_GetItem(hwndTree,&tvi);
return (HWND) tvi.lParam;
}

LRESULT HandleWmNotify(HWND hwnd, WPARAM wParam, LPARAM lParam)

{
NMHDR *nmhdr;
TV_HITTESTINFO testInfo;
HWND hTree = GetDlgItem(hwnd,IDTREEWINDOW);
HTREEITEM hti;
HWND hwndStart;

nmhdr = (NMHDR *)lParam;

switch (nmhdr->code) {
case NM_CLICK:
memset(&testInfo,0,sizeof(TV_HITTESTINFO));
GetCursorPos(&testInfo.pt);
MapWindowPoints(HWND_DESKTOP,hTree,&testInfo.pt,1);
hti = TreeView_HitTest(hTree,&testInfo);
if (hti == (HTREEITEM)0) break;
hwndStart = GetTreeItemInfo(hTree,hti);
SetTextInStatusBar(hwnd,hwndStart);
break;
}
return DefWindowProc(hwnd,WM_NOTIFY,wParam,lParam);
}

LRESULT CALLBACK MainWndProc(HWND hwnd,UINT msg,WPARAM wParam,LPARAM

lParam)
254 C Tutorial

{
static HWND hwndTree;
RECT rc,rcStatus;

switch (msg) {
case WM_CREATE:
hwndTree = CreateTree(hwnd,IDTREEWINDOW);
break;
case WM_NOTIFY:
return HandleWmNotify(hwnd,wParam,lParam);
case WM_SIZE:
SendMessage(hWndStatusbar,msg,wParam,lParam);
InitializeStatusBar(hWndStatusbar,1);
GetClientRect(hwnd,&rc);
GetWindowRect(hWndStatusbar,&rcStatus);
rc.bottom -= rcStatus.bottom-rcStatus.top;
MoveWindow(hwndTree,0,0,rc.right,rc.bottom,1);
break;
case WM_COMMAND:
HANDLE_WM_COMMAND(hwnd,wParam,lParam,MainWndProc_OnCommand);
break;
case WM_DESTROY:
PostQuitMessage(0);
break;
default:
return DefWindowProc(hwnd,msg,wParam,lParam);
}
return 0;
}

int WINAPI WinMain(HINSTANCE hInstance, HINSTANCE hPrevInstance, LPSTR

lpCmdLine, INT nCmdShow)
{
MSG msg;
HANDLE hAccelTable;

hInst = hInstance;
if (!InitApplication())
return 0;
hAccelTable = LoadAccelerators(hInst,MAKEINTRESOURCE(IDACCEL));
if ((hwndMain = CreatewindowtreeWndClassWnd()) == (HWND)0)
return 0;
CreateSBar(hwndMain,"",1);
ShowWindow(hwndMain,SW_SHOW);
PostMessage(hwndMain,WM_COMMAND,IDM_NEW,0);
while (GetMessage(&msg,NULL,0,0)) {
if (!TranslateAccelerator(msg.hwnd,hAccelTable,&msg)) {
TranslateMessage(&msg);
DispatchMessage(&msg);
}
}
return msg.wParam;
}

WindowTreeRes.h
#defineIDACCEL100
#defineIDM_NEW200
#defineIDM_OPEN210
#defineIDM_SAVE220
#defineIDM_SAVEAS230
 Appendix. Code listings 255

#defineIDM_CLOSE240
#defineIDM_PRINT250
#defineIDM_PAGESETUP260
#defineIDM_EXIT300
#defineIDM_ABOUT500
#defineIDMAINMENU600
#defineIDAPPLICON710
#defineIDAPPLCURSOR810
#defineIDS_FILEMENU2000
#defineIDS_HELPMENU2010
#defineIDS_SYSMENU2030
#defineIDM_STATUSBAR3000
Compiling from the command line:
lc windowtree.c psapi.lib
stricmp is called strcasecmp in some UNIX systems.

2.27.2 Counting chars: countchars.c

#include <stdio.h>
#include <string.h>
#include <stdlib.h>
#include <ctype.h>

int isWordStart(int c);

int line = 1;
int words = 0;
int maxwordlength = 0;

typedef struct _WordList {

int Count;
int Line;
struct _WordList *Next;
char Word[];
} WORDLIST;

WORDLIST *WordTable[0xffff+1];

typedef struct memory {

int used;
int size;
char memory[];
} MEMORY;

#define MEM_ALLOC_SIZE 0xffff

int memoryused = 0;
void *more_memory(int siz)
{
static MEMORY *mem;
void *result;
if (mem == NULL || mem->used+siz >= mem->size) {
mem = malloc(sizeof(mem)+MEM_ALLOC_SIZE);
if (mem == NULL) {
fprintf(stderr,"No more memory at line %d\n",line);
exit(1);
}
mem->used = 0;
memoryused += MEM_ALLOC_SIZE;
mem->size = MEM_ALLOC_SIZE;
}
256 C Tutorial

result = mem->memory+mem->used;
mem->used += siz;
memset(result,siz,0);
memoryused += siz;
return result;
}

WORDLIST *NewWordList(char *word)

{
int len = strlen(word);
WORDLIST *result = more_memory(sizeof(WORDLIST)+len+1);
result->Count = 1;
result->Line = line;
strcpy(result->Word,word);
words++;
if (len > maxwordlength)
maxwordlength = len;
return result;
}

int hash(char *word)

{
int h = 0;
while (*word) {
h += *word;
word++;
}
return h & 0xffff;
}

void EnterWord(char *word)

{
int h = hash(word);
WORDLIST *wl = WordTable[h];
while (wl) {
if (!strcmp(wl->Word,word)) {
wl->Count++;
return;
}
wl = wl->Next;
}
wl = NewWordList(word);
wl->Next = WordTable[h];
WordTable[h] = wl;
}

#define MAXIDLENGTH 512

int ScanWord(int firstchar,FILE *f)
{
int i = 1,c=0;
char idbuf[MAXIDLENGTH+1];

idbuf[0] = firstchar;
c = fgetc(f);
while (isWordStart(c) || isdigit(c)) {
idbuf[i++] = c;
if (i >= MAXIDLENGTH) {
fprintf(stderr,"Identifier too long at line %d\n",line);
return 0;
}
 Appendix. Code listings 257

c = fgetc(f);
}
if (c == '\n')
line++;
idbuf[i] = 0;
EnterWord(idbuf);
return 1;
}

int isWordStart(int c)
{
if (c == '_')
return 1;
if (c >= 'a' && c <= 'z')
return 1;
if (c >= 'A' && c <= 'Z')
return 1;
return 0;
}

int comparewords(const void *w1,const void *w2)

{
WORDLIST *pw1 = *(WORDLIST **)w1,*pw2 = *(WORDLIST **)w2;

if (pw1->Count == pw2->Count)
return strcmp(pw1->Word,pw2->Word);
return pw1->Count - pw2->Count;
}

void DoReports(char *filename)

{
int i,lastcount,identical;
int idx = 0;
xprintf("%s: %d lines %d different words. Memory used %'d\n",
filename,line,words,memoryused);
WORDLIST **tab = more_memory(words*sizeof(WORDLIST *));
for (i=0; i< sizeof(WordTable)/sizeof(WordTable[0]);i++) {
WORDLIST *wl = WordTable[i];

while (wl) {
tab[idx] = wl;
wl = wl->Next;
idx++;
if (idx >= words && wl) {
fprintf(stderr,"program error\n");
exit(1);
}
}
}
qsort(tab,words,sizeof(WORDLIST *),comparewords);
identical = 1;
lastcount = -1;
for (i=0; i< words;i++) {
if (lastcount != tab[i]->Count) {
if (identical > 2)
printf("\t\t\t-----------%d lines with count %d\n",
identical,lastcount);
else printf("\n");
identical = 1;
258 C Tutorial

}
else identical++;
lastcount = tab[i]->Count;
printf("%*s %5d %6d\n",-maxwordlength,tab[i]->Word,
tab[i]->Count,tab[i]->Line);
}
}

int main(int argc,char *argv[])

{
FILE *f;
int c;

if (argc < 2) {
fprintf(stderr,"Usage: countwords <file name>\n");
return 1;
}
f = fopen(argv[1],"r");
if (f == NULL)
return 1;
c = fgetc(f);
while (c != EOF) {
if (isWordStart(c)) {
if (!ScanWord(c,f))
return 1;
}
else if (c == '\n')
line++;
c = fgetc(f);
}
fclose(f);
DoReports(argv[1]);
return 0;
}