Chitika

C Language Tutorial

C Language Tutorial


This section contains a brief introduction to the C language. It is intended
as a tutorial on the language, and aims at getting a reader new to C started as
quickly as possible. It is certainly not intended as a substitute for
any of the numerous textbooks on C.

The best way to learn a new ``human'' language is to speak it right from the
outset, listening and repeating, leaving the intricacies of the grammar for
later. The same applies to computer languages--to learn C, we must start writing
C programs as quickly as possible.

An excellent textbook on C by two well-known and widely respected authors is:

The C Programming Language -- ANSI C
 Brian W. C. Kernighan & Dennis M. Ritchie 
 Prentice Hall, 1988 
Dennis Ritchie designed and implemented the first C compiler on a PDP-11
(a prehistoric machine by today's standards, yet one which had enormous
influence on modern scientific computation). The C language was based on two
(now defunct) languages: BCPL, written by Martin Richards, and B, written by Ken
Thompson in 1970 for the first UNIX system on a PDP-7. The original ``official''
C language was the ``K & R'' C, the nickname coming from the names of the
two authors of the original ``The C Programming Language''. In 1988, the
American National Standards Institute (ANSI) adopted a ``new and improved''
version of C, known today as ``ANSI C''. This is the version described in the
current edition of ``The C Programming Language -- ANSI C''. The ANSI version
contains many revisions to the syntax and the internal workings of the language,
the major ones being improved calling syntax for procedures and standarization
of most (but, unfortunately, not quite all!) system libraries.


1. A First Program

Let's be polite and start by
saluting the world! Type the following program into your favorite editor:

#include < stdio.h> void main() { printf("\nHello World\n"); }
Save the code in the file hello.c, then compile
it by typing:
gcc hello.c

This creates an executable file a.out, which is then
executed simply by typing its name. The result is that the characters ``
Hello World'' are printed out, preceded by an empty line.

A C program contains functions and variables. The functions
specify the tasks to be performed by the program. The ``main'' function
establishes the overall logic of the code. It is normally kept short and calls
different functions to perform the necessary sub-tasks. All C codes must have a
``main'' function.

Our hello.c code calls printf, an output function from
the I/O (input/output) library (defined in the file stdio.h). The
original C language did not have any built-in I/O statements whatsoever. Nor did
it have much arithmetic functionality. The original language was really not
intended for ''scientific'' or ''technical'' computation.. These functions are
now performed by standard libraries, which are now part of ANSI C. The K & R
textbook lists the content of these and other standard libraries in an appendix.

The printf line prints the message ``Hello World'' on
``stdout'' (the output stream corresponding to the X-terminal window
in which you run the code); ``\n'' prints a ``new line'' character,
which brings the cursor onto the next line. By construction, printf
never inserts this character on its own: the following program would produce the
same result:


#include < stdio.h> void main() { printf("\n"); printf("Hello World"); printf("\n"); }
Try leaving out the ``\n'' lines and see what happens.

The first statement ``#include < stdio.h>'' includes a
specification of the C I/O library. All variables in C must be explicitly
defined before use: the ``.h'' files are by convention ``header
files'' which contain definitions of variables and functions necessary for the
functioning of a program, whether it be in a user-written section of code, or as
part of the standard C libaries. The directive ``#include'' tells the
C compiler to insert the contents of the specified file at that point in the
code. The ``< ...>'' notation instructs the compiler to look
for the file in certain ``standard'' system directories.

The void preceeding ``main'' indicates that main is of
``void'' type--that is, it has no type associated with it, meaning that
it cannot return a result on execution.

The ``;'' denotes the end of a statement. Blocks of statements are put in
braces {...}, as in the definition of functions. All C statements are defined in
free format, i.e., with no specified layout or column assignment. Whitespace
(tabs or spaces) is never significant, except inside quotes as part of a
character string. The following program would produce exactly the same result as
our earlier example:

#include < stdio.h>
void main(){printf("\nHello World\n");}

The reasons for arranging your programs in lines and indenting to show
structure should be obvious!


2. Let's Compute

The following program,
sine.c, computes a table of the sine function for angles between 0
and 360 degrees.

/************************/ /* Table of */ /* Sine Function */ /************************/ /* Michel Vallieres */ /* Written: Winter 1995 */ #include < stdio.h> #include < math.h> void main() { int angle_degree; double angle_radian, pi, value; /* Print a header */ printf ("\nCompute a table of the sine function\n\n"); /* obtain pi once for all */ /* or just use pi = M_PI, where M_PI is defined in math.h */ pi = 4.0*atan(1.0); printf ( " Value of PI = %f \n\n", pi ); printf ( " angle Sine \n" ); angle_degree=0; /* initial angle value */ /* scan over angle */ while ( angle_degree <= 360 ) /* loop until angle_degree > 360 */ { angle_radian = pi * angle_degree/180.0 ; value = sin(angle_radian); printf ( " %3d %f \n ", angle_degree, value ); angle_degree = angle_degree + 10; /* increment the loop index */ } }
The code starts with a series of comments indicating its the purpose, as
well as its author. It is considered good programming style to identify and
document your work (although, sadly, most people only do this as an
afterthought). Comments can be written anywhere in the code: any characters
between /* and */ are ignored by the compiler and can be used to make the code
easier to understand. The use of variable names that are meaningful within the
context of the problem is also a good idea.

The #include statements now also include the header file for the
standard mathematics library math.h. This statement is needed to
define the calls to the trigonometric functions atan and
sin. Note also that the compilation
must include the mathematics library explicitly by typing

gcc sine.c -lm

Variable names are arbitrary (with some compiler-defined maximum length,
typically 32 characters). C uses the following standard variable types:

int    -> integer variable
short  -> short integer
long   -> long integer
float  -> single precision real (floating point) variable
double -> double precision real (floating point) variable
char   -> character variable (single byte)
The compilers checks for consistency in the types of all variables used in
any code. This feature is intended to prevent mistakes, in particular in
mistyping variable names. Calculations done in the math library routines are
usually done in double precision arithmetic (64 bits on most workstations). The
actual number of bytes used in the internal storage of these data types depends
on the machine being used.

The printf function can be instructed to print integers, floats
and strings properly. The general syntax is

printf( "format", variables );
where "format" specifies the converstion specification and
variables is a list of quantities to print. Some useful formats are
%.nd integer (optional n = number of columns; if 0, pad with zeroes)
%m.nf float or double (optional m = number of columns,
       n = number of decimal places)
%ns string (optional n = number of columns)
%c     character
\n \t  to introduce new line or tab
\g ring the bell (``beep'') on the terminal





3. Loops

Most real programs contain some construct
that loops within the program, performing repetitive actions on a stream of data
or a region of memory. There are several ways to loop in C. Two of the most
common are the while loop:
while (expression)
 {
 ...block of statements to execute...
 }
and the for loop:
for (expression_1; expression_2; expression_3)
 {
 ...block of statements to execute...
 }
The while loop continues to loop until the conditional
expression becomes false. The condition is tested upon entering the
loop. Any logical construction (see below for a list) can be used in this
context.

The for loop is a special case, and is equivalent to the following
while loop:

expression_1;

 while (expression_2)
 {
 ...block of statements...

 expression_3;
 }
For instance, the following structure is often encountered:
i = initial_i;

 while (i <= i_max)
 {
   ...block of statements...

 i = i + i_increment;
 }
This structure may be rewritten in the easier syntax of the for
loop as:
for (i = initial_i; i <= i_max; i = i + i_increment)
 {
 ...block of statements...
 }
Infinite loops are possible (e.g. for(;;)), but not too good
for your computer budget! C permits you to write an infinite loop, and provides
the break statement to ``breakout '' of the loop. For example,
consider the following (admittedly not-so-clean) re-write of the previous loop:
angle_degree = 0;

 for ( ; ; )
 {
 ...block of statements...

 angle_degree = angle_degree + 10;
 if (angle_degree == 360) break;
 }
The conditional if simply asks whether angle_degree
is equal to 360 or not; if yes, the loop is stopped.


4. Symbolic Constants

You can define constants of
any type by using the #define compiler directive. Its syntax is
simple--for instance
#define ANGLE_MIN 0
#define ANGLE_MAX 360
would define ANGLE_MIN and ANGLE_MAX to the values 0
and 360, respectively. C distinguishes between lowercase and uppercase letters
in variable names. It is customary to use capital letters in defining global
constants.





5. Conditionals

Conditionals are used within
the if and while constructs:
if (conditional_1)
 {
 ...block of statements executed if conditional_1 is true...
 }
 else if (conditional_2)
 {
 ...block of statements executed if conditional_2 is true...
 }
 else
 {
 ...block of statements executed otherwise...
 }
and any variant that derives from it, either by omitting branches or by
including nested conditionals.

Conditionals are logical operations involving comparison of quantities (of
the same type) using the conditional operators:

<  smaller than
 <=  smaller than or equal to
 ==  equal to
 !=  not equal to
 >=  greater than or equal to
 >  greater than
and the boolean operators
&&   and
 ||  or
 !  not

Another conditional use is in the switch construct:

switch (expression)
 {
 case const_expression_1:
 {
     ...block of statements...
         break;
 }
 case const_expression_2:
 {
     ...block of statements...
         break;
 }
 default:
 {
     ...block of statements..
 }
 }
The appropriate block of statements is executed according to the value of
the expression, compared with the constant expressions in the case statement.
The break statements insure that the statements in the cases
following the chosen one will not be executed. If you would want to execute
these statements, then you would leave out the break statements. This
construct is particularly useful in handling input variables.





6. Pointers

The C language allows the programmer
to ``peek and poke'' directly into memory locations. This gives great
flexibility and power to the language, but it also one of the great hurdles that
the beginner must overcome in using the language.

All variables in a program reside in memory; the statements

float x;
 x = 6.5;
request that the compiler reserve 4 bytes of memory (on a 32-bit computer)
for the floating-point variable x, then put the ``value'' 6.5 in it.

Sometimes we want to know where a variable resides in memory. The address
(location in memory) of any variable is obtained by placing the operator
``&'' before its name. Therefore &x is the address
of x. C allows us to go one stage further and define a variable,
called a pointer, that contains the address of (i.e. ``points to'')
other variables. For example:

float x;
 float* px;

 x = 6.5;
 px = &x;
defines px to be a pointer to objects of type float, and sets
it equal to the address of x:

The content of the memory location referenced by a pointer is obtained using
the ``*'' operator (this is called dereferencing the
pointer). Thus, *px refers to the value of x.

C allows us to perform arithmetic operations using pointers, but beware that
the ``unit'' in pointer arithmetic is the size (in bytes) of the object to which
the pointer points. For example, if px is a pointer to a variable
x of type float, then the expression px + 1
refers not to the next bit or byte in memory but to the location of the next
float after x (4 bytes away on most workstations); if
x were of type double, then px + 1 would refer
to a location 8 bytes (the size of a double)away, and so on. Only if
x is of type char will px + 1 actually refer to
the next byte in memory.

Thus, in

char* pc;
 float* px;
 float x;

 x = 6.5;
 px = &x;
 pc = (char*) px;
(the (char*) in the last line is a ``cast'', which converts one
data type to another), px and pc both point to the same
location in memory--the address of x--but px + 1 and
pc + 1 point to different memory locations.

Consider the following simple code.


void main() { float x, y; /* x and y are of float type */ float *fp, *fp2; /* fp and fp2 are pointers to float */ x = 6.5; /* x now contains the value 6.5 */ /* print contents and address of x */ printf("Value of x is %f, address of x %ld\n", x, &x); fp = &x; /* fp now points to location of x */ /* print the contents of fp */ printf("Value in memory location fp is %f\n", *fp); /* change content of memory location */ *fp = 9.2; printf("New value of x is %f = %f \n", *fp, x); /* perform arithmetic */ *fp = *fp + 1.5; printf("Final value of x is %f = %f \n", *fp, x); /* transfer values */ y = *fp; fp2 = fp; printf("Transfered value into y = %f and fp2 = %f \n", y, *fp2); }
Run this code to see the results of these different operations. Note that,
while the value of a pointer (if you print it out with printf) is
typically a large integer, denoting some particular memory location in the
computer, pointers are not integers--they are a completely different
data type.




7. Arrays

Arrays of any type can be formed in C. The
syntax is simple:
type name[dim];
In C, arrays starts at position 0. The elements of the array occupy
adjacent locations in memory. C treats the name of the array as if it were a
pointer to the first element--this is important in understanding how to do
arithmetic with arrays. Thus, if v is an array, *v is the
same thing as v[0], *(v+1) is the same thing as
v[1], and so on:

Pointer use for an array

Consider the following code, which illustrates the use of pointers:


#define SIZE 3 void main() { float x[SIZE]; float *fp; int i; /* initialize the array x */ /* use a "cast" to force i */ /* into the equivalent float */ for (i = 0; i < SIZE; i++) x[i] = 0.5*(float)i; /* print x */ for (i = 0; i < SIZE; i++) printf(" %d %f \n", i, x[i]); /* make fp point to array x */ fp = x; /* print via pointer arithmetic */ /* members of x are adjacent to */ /* each other in memory */ /* *(fp+i) refers to content of */ /* memory location (fp+i) or x[i] */ for (i = 0; i < SIZE; i++) printf(" %d %f \n", i, *(fp+i)); }
(The expression ``i++'' is C shorthand for ``i = i +
1
''.) Since x[i] means the i-th element of the array
x, and fp = x points to the start of the x
array, then *(fp+i) is the content of the memory address i
locations beyond fp, that is, x[i].





8. Character Arrays

A string constant ,
such as
"I am a string"
is an array of characters. It is represented internally in C by the ASCII
characters in the string, i.e., ``I'', blank, ``a'',
``m'',... for the above string, and terminated by the special null
character ``\0'' so programs can find the end of the string.

String constants are often used in making the output of code intelligible
using printf ;

printf("Hello, world\n");
 printf("The value of a is: %f\n", a);

String constants can be associated with variables. C provides the
char type variable, which can contain one character--1 byte--at a
time. A character string is stored in an array of character type, one ASCII
character per location. Never forget that, since strings are conventionally
terminated by the null character ``\0'', we require one extra storage
location
in the array!

C does not provide any operator which manipulate entire strings at once.
Strings are manipulated either via pointers or via special routines available
from the standard string library string.h. Using character
pointers is relatively easy since the name of an array is a just a pointer to
its first element. Consider the following code:


void main() { char text_1[100], text_2[100], text_3[100]; char *ta, *tb; int i; /* set message to be an arrray */ /* of characters; initialize it */ /* to the constant string "..." */ /* let the compiler decide on */ /* its size by using [] */ char message[] = "Hello, I am a string; what are you?"; printf("Original message: %s\n", message); /* copy the message to text_1 */ /* the hard way */ i=0; while ( (text_1[i] = message[i]) != '\0' ) i++; printf("Text_1: %s\n", text_1); /* use explicit pointer arithmetic */ ta=message; tb=text_2; while ( ( *tb++ = *ta++ ) != '\0' ) ; printf("Text_2: %s\n", text_2); }
The standard ``string'' library contains many useful functions to
manipulate strings; a description of this library can be found in an appendix of
the K & R textbook. Some of the most useful functions are:
char *strcpy(s,ct)     -> copy  ct into s, including ``\0''; return s
char *strncpy(s,ct,n)  -> copy ncharcater of ct into s, return s
char *strncat(s,ct)    -> concatenate ct to end of s; return s
char *strncat(s,ct,n)  -> concatenate n character of ct to end
                        of s, terminate with ``\0''; return s
int strcmp(cs,ct)      -> compare cs and ct; return 0 if cs=ct,
                          <0 if cs0 if cs>ct
char *strchr(cs,c)     -> return pointer to first occurence of c
                         in cs or NULL if not encountered
size_t strlen(cs)      -> return length of cs
(s and t are char*, cs and
ct are const char*, c is an char
converted to type int, and n is an int.)

Consider the following code which uses some of these functions:


#include < string.h> void main() { char line[100], *sub_text; /* initialize string */ strcpy(line,"hello, I am a string;"); printf("Line: %s\n", line); /* add to end of string */ strcat(line," what are you?"); printf("Line: %s\n", line); /* find length of string */ /* strlen brings back */ /* length as type size_t */ printf("Length of line: %d\n", (int)strlen(line)); /* find occurence of substrings */ if ( (sub_text = strchr ( line, 'W' ) )!= NULL ) printf("String starting with \"W\" ->%s\n", sub_text); if ( ( sub_text = strchr ( line, 'w' ) )!= NULL ) printf("String starting with \"w\" ->%s\n", sub_text); if ( ( sub_text = strchr ( sub_text, 'u' ) )!= NULL ) printf("String starting with \"w\" ->%s\n", sub_text); }


9. I/O Capabilities

Character level I/O

C provides (through its libraries) a variety of I/O
routines. At the character level, getchar() reads one character at a
time from stdin, while putchar() writes one character at a
time to stdout. For example, consider

#include < stdio.h> void main() { int i, nc; nc = 0; i = getchar(); while (i != EOF) { nc = nc + 1; i = getchar(); } printf("Number of characters in file = %d\n", nc); }
This program counts the number of characters in the input stream (e.g. in
a file piped into it at execution time). The code reads characters (whatever
they may be) from stdin (the keyboard), uses stdout (the
X-terminal you run from) for output, and writes error messages to
stderr (usually also your X-terminal). These streams are always
defined at run time. EOF is a special return value, defined in
stdio.h, returned by getchar() when it encounters an
end-of-file marker when reading. Its value is computer dependent, but
the C compiler hides this fact from the user by defining the variable EOF. Thus
the program reads characters from stdin and keeps adding to the
counter nc, until it encounters the ``end of file''.

An experienced C programmer would probably code this example as:


#include < stdio.h> void main() { int c, nc = 0; while ( (c = getchar()) != EOF ) nc++; printf("Number of characters in file = %d\n", nc); }
C allows great brevity of expression, usually at the expense of
readability!

The () in the statement (c = getchar()) says to execute
the call to getchar() and assign the result to c before
comparing it to EOF; the brackets are necessary here. Recall that
nc++ (and, in fact, also ++nc) is another way of writing
nc = nc + 1. (The difference between the prefix and postfix notation
is that in ++nc, nc is incremented before it is used,
while in nc++, nc is used before it is incremented. In
this particular example, either would do.) This notation is more compact (not
always an advantage, mind you), and it is often more efficiently coded by the
compiler.

The UNIX command wc counts the characters, words and lines in a
file. The program above can be considered as your own wc. Let's add
a counter for the lines.


#include < stdio.h> void main() { int c, nc = 0, nl = 0; while ( (c = getchar()) != EOF ) { nc++; if (c == '\n') nl++; } printf("Number of characters = %d, number of lines = %d\n", nc, nl); }
Can you think of a way to count the number of words in the file?

Higher-Level I/O capabilities

We have already seen that
printf handles formatted output to stdout. The counterpart
statement for reading from stdin is scanf. The syntax
scanf("format string", variables);
resembles that of printf. The format string may contain blanks
or tabs (ignored), ordinary ASCII characters, which must match those in
stdin, and conversion specifications as in printf.

Equivalent statements exist to read from or write to character strings. They
are:

sprintf(string, "format string", variables);
 scanf(string, "format string", variables);
The ``string'' argument is the name of (i.e. a pointer to) the character
array into which you want to write the information.

I/O to and from files

Similar statements also exist for handling I/O to
and from files. The statements are
#include < stdio.h>

 FILE *fp;
          
 fp = fopen(name, mode);
 
 fscanf(fp, "format string", variable list);
 fprintf(fp, "format string", variable list);

 fclose(fp );
The logic here is that the code must
  • define a local ``pointer'' of type FILE (note that the
    uppercase is necessary here), which is defined in < stdio.h>
  • ``open'' the file and associate it with the local pointer via
    fopen
  • perform the I/O operations using fscanf and fprintf
  • disconnect the file from the task with fclose
The
``mode'' argument in the fopen specifies the purpose/positioning in
opening the file: ``r'' for reading, ``w'' for writing,
and ``a'' for appending to the file. Try the following:

#include < stdio.h> void main() { FILE *fp; int i; fp = fopen("foo.dat", "w"); /* open foo.dat for writing */ fprintf(fp, "\nSample Code\n\n"); /* write some info */ for (i = 1; i <= 10 ; i++) fprintf(fp, "i = %d\n", i); fclose(fp); /* close the file */ }
Compile and run this code; then use any editor to read the file
foo.dat.





10. Functions

Functions are easy to use; they
allow complicated programs to be parcelled up into small blocks, each of which
is easier to write, read, and maintain. We have already encountered the function
main and made use of I/O and mathematical routines from the standard
libraries. Now let's look at some other library functions, and how to write and
use our own.

Calling a Function

The call to a function in C simply entails
referencing its name with the appropriate arguments. The C compiler checks for
compatibility between the arguments in the calling sequence and the definition
of the function.

Library functions are generally not available to us in source form. Argument
type checking is accomplished through the use of header files (like
stdio.h) which contain all the necessary information. For example, as
we saw earlier, in order to use the standard mathematical library you must
include math.h via the statement

#include < math.h>
at the top of the file containing your code. The most commonly used header
files are
< stdio.h>  -> defining I/O routines
< ctype.h>  -> defining character manipulation routines
< string.h> -> defining string manipulation routines
< math.h>   -> defining mathematical routines
< stdlib.h> -> defining number conversion, storage allocation
            and similar tasks
< stdarg.h> -> defining libraries to handle routines with variable
            numbers of arguments
< time.h>   -> defining time-manipulation routines
In addition, the following header files exist:
< assert.h> -> defining diagnostic routines
< setjmp.h> -> defining non-local function calls
< signal.h> -> defining signal handlers
< limits.h> -> defining constants of the int type
< float.h>  -> defining constants of the float type
Appendix B in the K & R book describes these libraries in great
detail.

Writing Your Own Functions

A function has the
following layout:
return-type function-name ( argument-list-if-necessary )
{
 ...local-declarations...

 ...statements...

 return return-value;
}
If return-type is omitted, C defaults to int. The
return-value must be of the declared type.

A function may simply perform a task without returning any value, in which
case it has the following layout:

void function-name ( argument-list-if-necessary )
{
 ...local-declarations...

 ...statements...
}

As an example of function calls, consider the following code:


/* include headers of library */ /* defined for all routines */ /* in the file */ #include < stdio.h> #include < string.h> /* prototyping of functions */ /* to allow type checks by */ /* the compiler */ void main() { int n; char string[50]; /* strcpy(a,b) copies string b into a */ /* defined via the stdio.h header */ strcpy(string, "Hello World"); /* call own function */ n = n_char(string); printf("Length of string = %d\n", n); } /* definition of local function n_char */ int n_char(char string[]) { /* local variable in this function */ int n; /* strlen(a) returns the length of */ /* string a */ /* defined via the string.h header */ n = strlen(string); if (n > 50) printf("String is longer than 50 characters\n"); /* return the value of integer n */ return n; }

Arguments are always passed by value in C function calls. This means
that local ``copies'' of the values of the arguments are passed to the routines.
Any change made to the arguments internally in the function are made only to the
local copies of the arguments. In order to change (or define) an argument in the
argument list, this argument must be passed as an address, thereby forcing C to
change the ``real'' argument in the calling routine.

As an example, consider exchanging two numbers between variables. First let's
illustrate what happen if the variables are passed by value:


#include < stdio.h> void exchange(int a, int b); void main() { /* WRONG CODE */ int a, b; a = 5; b = 7; printf("From main: a = %d, b = %d\n", a, b); exchange(a, b); printf("Back in main: "); printf("a = %d, b = %d\n", a, b); } void exchange(int a, int b) { int temp; temp = a; a = b; b = temp; printf(" From function exchange: "); printf("a = %d, b = %d\n", a, b); }
Run this code and observe that a and b are NOT
exchanged! Only the copies of the arguments are exchanged. The RIGHT way to do
this is of course to use pointers:

#include < stdio.h> void exchange ( int *a, int *b ); void main() { /* RIGHT CODE */ int a, b; a = 5; b = 7; printf("From main: a = %d, b = %d\n", a, b); exchange(&a, &b); printf("Back in main: "); printf("a = %d, b = %d\n", a, b); } void exchange ( int *a, int *b ) { int temp; temp = *a; *a = *b; *b = temp; printf(" From function exchange: "); printf("a = %d, b = %d\n", *a, *b); }
The rule of thumb here is that








11. Command-line arguments

It is standard
practice in UNIX for information to be passed from the command line directly
into a program through the use of one or more command-line arguments, or
switches. Switches are typically used to modify the behavior of a
program, or to set the values of some internal parameters. You have already
encountered several of these--for example, the "ls" command lists the
files in your current directory, but when the switch -l is added,
"ls -l" produces a so-called ``long'' listing instead. Similarly,
"ls -l -a" produces a long listing, including ``hidden'' files, the
command "tail -20" prints out the last 20 lines of a file (instead of
the default 10), and so on.

Conceptually, switches behave very much like arguments to functions within C,
and they are passed to a C program from the operating system in precisely the
same way as arguments are passed between functions. Up to now, the
main() statements in our programs have had nothing between the
parentheses. However, UNIX actually makes available to the program (whether the
programmer chooses to use the information or not) two arguments to
main: an array of character strings, conventionally called
argv, and an integer, usually called argc, which
specifies the number of strings in that array. The full statement of the first
line of the program is

main(int argc, char** argv)
(The syntax char** argv declares argv to be a pointer to a
pointer to a character, that is, a pointer to a character array (a character
string)--in other words, an array of character strings. You could also write
this as char* argv[]. Don't worry too much about the details of the
syntax, however--the use of the array will be made clearer below.)

When you run a program, the array argv contains, in order,
all the information on the command line when you entered the command
(strings are delineated by whitespace), including the command itself.
The integer argc gives the total number of strings, and is therefore
equal to equal to the number of arguments plus one. For example, if you
typed

a.out -i 2 -g -x 3 4
the program would receive
argc = 7
 argv[0] = "a.out"
 argv[1] = "-i"
 argv[2] = "2"
 argv[3] = "-g"
 argv[4] = "-x"
 argv[5] = "3"
 argv[6] = "4"
Note that the arguments, even the numeric ones, are all strings
at this point. It is the programmer's job to decode them and decide what to do
with them.

The following program simply prints out its own name and arguments:

#include < stdio.h>

main(int argc, char** argv)
{
 int i;

 printf("argc = %d\n", argc);

 for (i = 0; i < argc; i++)
 printf("argv[%d] = \"%s\"\n", i, argv[i]);
}
UNIX programmers have certain conventions about how to interpret the
argument list. They are by no means mandatory, but it will make your program
easier for others to use and understand if you stick to them. First, switches
and key terms are always preceded by a ``-'' character. This makes them easy to
recognize as you loop through the argument list. Then, depending on the switch,
the next arguments may contain information to be interpreted as integers,
floats, or just kept as character strings. With these conventions, the most
common way to ``parse'' the argument list is with a for loop and a
switch statement, as follows:
#include < stdio.h>
#include < stdlib.h>

main(int argc, char** argv)
{
 /* Set defaults for all parameters: */

 int a_value = 0;
 float b_value = 0.0;
 char* c_value = NULL;
 int d1_value = 0, d2_value = 0;

 int i;

 /* Start at i = 1 to skip the command name. */

 for (i = 1; i < argc; i++) {

 /* Check for a switch (leading "-"). */

 if (argv[i][0] == '-') {

     /* Use the next character to decide what to do. */

     switch (argv[i][1]) {

  case 'a': a_value = atoi(argv[++i]);
    break;

  case 'b': b_value = atof(argv[++i]);
    break;

  case 'c': c_value = argv[++i];
    break;

  case 'd': d1_value = atoi(argv[++i]);
                  d2_value = atoi(argv[++i]);
    break;

     }
 }
 }

 printf("a = %d\n", a_value);
 printf("b = %f\n", b_value);
 if (c_value != NULL) printf("c = \"%s\"\n", c_value);
 printf("d1 = %d, d2 = %d\n", d1_value, d2_value);
}
Note that argv[i][j] means the j-th character of the
i-th character string. The if statement checks for a
leading ``-'' (character 0), then the switch statement allows various
courses of action to be taken depending on the next character in the string
(character 1 here). Note the use of argv[++i] to increase
i before use, allowing us to access the next string in a single
compact statement. The functions atoi and atof are defined
in stdlib.h. They convert from character strings to ints
and doubles, respectively.

A typical command line might be:

a.out -a 3 -b 5.6 -c "I am a string" -d 222 111
(The use of double quotes with -c here makes sure that the
shell treats the entire string, including the spaces, as a single object.)

Arbitrarily complex command lines can be handled in this way. Finally, here's
a simple program showing how to place parsing statements in a separate function
whose purpose is to interpret the command line and set the values of its
arguments:

/********************************/
   /*                         */
   /*   Getting arguments from     */
   /*    */
   /*      the Command Line        */
   /*    */
   /********************************/


     /* Steve McMillan  */
     /* Written: Winter 1995 */


#include < stdio.h>
#include < stdlib.h>

void get_args(int argc, char** argv, int* a_value, float* b_value)
{
 int i;

 /* Start at i = 1 to skip the command name. */

 for (i = 1; i < argc; i++) {

 /* Check for a switch (leading "-"). */

 if (argv[i][0] == '-') {

     /* Use the next character to decide what to do. */

     switch (argv[i][1]) {

  case 'a': *a_value = atoi(argv[++i]);
    break;

  case 'b': *b_value = atof(argv[++i]);
    break;

  default: fprintf(stderr,
    "Unknown switch %s\n", argv[i]);
     }
 }
 }
}

main(int argc, char** argv)
{
 /* Set defaults for all parameters: */

 int a = 0;
 float b = 0.0;

 get_args(argc, argv, &a, &b);

 printf("a = %d\n", a);
 printf("b = %f\n", b);
}





12. Graphical Interfaces: Dialog Boxes

Suppose
you don't want to deal with command line interpretation, but you still want your
program to be able to change the values of certain variables in an interactive
way. You could simply program in a series printf/scanf lines to quiz
the user about their preferences:
.
 .
 .
 printf("Please enter the value of n: ");
 scanf("%d", &n);

 printf("Please enter the value of x: ");
 scanf("%f", &x);

 .
 .
 .

and so on, but this won't work well if your program is to be used as part of
a pipeline (see the UNIX
primer), for example using ther graphics program plot_data,
since the questions and answers will get mixed up with the data stream.

A convenient alternative is to use a simple graphical interface which
generates a dialog box, offering you the option of varying key
parameters in your program. Our graphics package provides a number of
easy-to-use tools for constructing and using such boxes. The simplest way to set
the integer variable n and the float variable x (i.e. to
perform the same effect as the above lines of code) using a dialog box is as
follows:

/* Simple program to illustrate use of a dialog box */
 
main()
{
 /* Define default values: */

 int n = 0;
 float x = 0.0;

 /* Define contents of dialog window */

 create_int_dialog_entry("n", &n);
 create_float_dialog_entry("x", &x);

 /* Create window with name "Setup" and top-left corner at (0,0) */

 set_up_dialog("Setup", 0, 0);

 /* Display the window and read the results */

 read_dialog_window();

 /* Print out the new values */

 printf("n = %d, x = %f\n", n, x);
}
Compile this program using the alias Cgfx (see the page on compilation)
to link in all necessary libraries.

The two create lines define the entries in the box and the
variables to be associated with them, set_up_dialog names the box and
defines its location. Finally, read_dialog_window pops up a window
and allows you to change the values of the variables. When the program runs, you
will see a box that looks something like this:

Modify the numbers shown, click "OK" (or just hit carriage return), and the
changes are made. That's all there is to it! The great advantage of this
approach is that it operates independently of the flow of data through
stdin/stdout. In principle, you could even control the operation of
every stage in a pipeline of many chained commands, using a separate dialog box
for each.

Loops

Loops

Most real programs contain some construct that loops within the program, performing repetitive actions on a stream of data or a region of memory. There are several ways to loop in C. Two of the most common are the while loop:

while (expression)
{
...block of statements to execute...
}

and the for loop:

for (expression_1; expression_2; expression_3)
{
...block of statements to execute...
}

The while loop continues to loop until the conditional expression becomes false. The condition is tested upon entering the loop. Any logical construction (see below for a list) can be used in this context.

The for loop is a special case, and is equivalent to the following while loop:

expression_1;

while (expression_2)
{
...block of statements...

expression_3;
}

For instance, the following structure is often encountered:

i = initial_i;

while (i <= i_max)
{
...block of statements...

i = i + i_increment;
}

This structure may be rewritten in the easier syntax of the for loop as:

for (i = initial_i; i <= i_max; i = i + i_increment)
{
...block of statements...
}

Infinite loops are possible (e.g. for(;;)), but not too good for your computer budget! C permits you to write an infinite loop, and provides the break statement to ``breakout '' of the loop. For example, consider the following (admittedly not-so-clean) re-write of the previous loop:

angle_degree = 0;

for ( ; ; )
{
...block of statements...

angle_degree = angle_degree + 10;
if (angle_degree == 360) break;
}

The conditional if simply asks whether angle_degree is equal to 360 or not; if yes, the loop is s.

List Of Compilers

Ada compilers
Compiler Author Windows Unix-like Other OSs License type IDE?
GCC GNAT GNU Project Yes Yes No Open
source
GPS, Eclipse
A#
Yes
Yes Open
source

ALGOL compilers

Compiler Author Windows Unix-like Other OSs License type IDE?
ALGOL 60 RHA (Minisystems) Ltd No No DOS, CP/M Free
for personal use
No
ALGOL 68 Genie Marcel van der Veer Yes Yes Various GPL No
Persistant S-algol Paul Cockshott Yes No DOS Copyright only Yes

BASIC compilers

Compiler Author Windows Unix-like Other OSs License type IDE?
BlitzMax Blitz Research Yes Yes No Proprietary Yes
Gambas Benoît
Minisini
No Yes No Open
source
Yes
Chipmunk
Basic
? Yes Yes Yes Freeware Open
DarkBASIC The Game Creators Yes No No Proprietary Yes
FreeBASIC FreeBASIC Development Team Yes Yes DOS Open
source
No
Just BASIC Shoptalk Systems Yes ? ? Freeware Yes
KBasic[1] KBasic Software Yes Yes No Open
source
Yes
Liberty
BASIC
Shoptalk Systems Yes No No Proprietary Yes
NBasic[2] SylvaWare Yes No No Freeware No
PowerBASIC PowerBASIC, Inc. Yes No DOS Proprietary Yes
PureBasic Fantaisie Software Yes Yes Yes Proprietary Yes
Quick Basic Microsoft No No DOS Proprietary Yes
Realbasic Real Software Yes Yes Yes Proprietary Yes
RapidQ William Yu Yes Yes Yes Freeware Yes
Visual
Basic
Microsoft Yes No DOS Proprietary Yes

C# compilers

Compiler Author Windows Unix-like Other OSs License type IDE?
Microsoft
Visual C#
Microsoft Yes No No Proprietary Yes
Microsoft
Visual C# Express
Microsoft Yes No No Freeware Yes
MonoDevelop MonoDevelop Yes Yes Yes[citation
needed
]
Open
source
Yes
SharpDevelop IC#Code Yes No No Open
source
Yes
Portable.NET Dot Gnu
Project
Yes Yes No Open
source
?

C compilers

Compiler Author Windows Unix-like Other OSs License type IDE?
AMPC Axiomatic
Solutions Sdn Bhd
Yes Yes Yes Proprietary Yes
Amsterdam
Compiler Kit
Andrew Tanenbaum
and Ceriel
Jacobs
No Yes Yes BSD No
Clang Low
Level Virtual Machine
Yes Yes Yes BSD No
GCC C GNU Project Yes Yes Yes
IBM mainframe, AmigaOS, VAX/VMS, RTEMS
GPL No
LabWindows/CVI National
Instruments
Yes Yes Yes Proprietary Yes
lcc Chris
Fraser
and David
Hanson
Yes Yes Yes Freeware
(source code available for non-commercial use)
Windows
only
Mark
Williams C
Mark
Williams Company
Yes No Yes Proprietary Yes
Microsoft C Microsoft Yes No No Proprietary Yes
Nwcc Nils
Weller
Yes Yes Yes BSD No
Open64 SGI Google HP Intel Nvidia PathScale Tsinghua
University
and others
Yes Yes Yes GPL No
Pelles C Pelle
Orinius
Yes No No Freeware Yes
Portable C
Compiler
Anders
Magnusson
and others
Yes Yes Yes BSD No
Power
C
Mix
Software
No No Yes Proprietary No
QuickC Microsoft Yes No No Proprietary Yes
SAS/C SAS
Institute
Yes Yes Yes
IBM mainframe, AmigaOS, 68K, 88K, Unix
Proprietary Yes
Tiny C
Compiler
Fabrice
Bellard
Yes Yes No LGPL No
Turbo C Embarcadero No No Yes Proprietary -
V 2.01 freely available
Yes
CCS
C Compiler
CCS,
Inc.
Yes Yes Yes Proprietary Yes
ups
debugger
[3] (includes C
interpreter)
Tom Hughes, Ian Edwards, and others No Yes Yes
Solaris, SunOS
GPL Yes
VBCC Dr.
Volker Barthelmann
Yes Yes Yes Freeware
(source code available, modification not allowed)
No

C++ compilers

Compiler Author Windows Unix-like Other OSs License type IDE?
C++ Builder Embarcadero
(CodeGear)
Yes No No Proprietary Yes
AMD x86 Open64 Compiler Suite AMD No Yes No GPLv2/LGPLv2.1 No
Turbo C++ Explorer Embarcadero
(CodeGear)
Yes No No Freeware Yes
C++
Compiler
Embarcadero
(CodeGear)
Yes No No Freeware No
Ch interpreter SoftIntegration, Inc Yes Yes FreeBSD, Linux, Solaris, HP-UX, AIX, etc Freeware Yes
CINT CERN Yes Yes Yes
Solaris, BeBox, DOS, Convex, etc.
X11/MIT Yes
Borland C++ Borland (CodeGear) Yes No DOS Proprietary Yes
Turbo C++ for DOS Borland (CodeGear) No No DOS Proprietary Yes
Clang Apple Inc. Yes Yes Yes Open
source
No
CodeWarrior Metrowerks Yes Yes Yes ? Yes
Comeau
C/C++
Comeau
Computing
Yes Yes Yes Proprietary No
CoSy compiler development system ACE Associated Compiler Experts Yes Yes No Proprietary No
Digital
Mars
Digital
Mars
Yes No DOS Proprietary No
Djgpp DJ Delorie No No DOS Open
source
Rhide
EDGE
ARM C/C++
Mentor
Graphics
Yes Yes Yes Proprietary Yes
GCC GNU Project Yes Yes Yes Open
source
Xcode, Eclipse, NetBeans
HP
aC++
Hewlett-Packard No Yes No Proprietary No
IAR C/C++ Compilers IAR Systems Yes No No Proprietary Yes
C++ Compiler
Professional Edition
Intel Yes Linux,
Mac OS X
No Proprietary Yes
C++ Compiler
Professional Edition (for non-commercial use)
Intel No Linux No Freeware Yes
KAI C++ Compiler Kuck & Associates, Inc. ? ? ? Proprietary ?
Microtec Mentor
Graphics
Yes Yes Yes Proprietary Yes
MULTI Green
Hills Software
Yes Yes Yes Proprietary Yes
Open Watcom Sybase Yes Yes Yes Open
source
Yes
Open64 HP AMD Tsinghua
University
and others
No Yes No Modified
GPLv2
No
PathScale PathScale and
others
No Yes Yes Proprietary/Open source No
PGI Workstation PGI Yes Yes No Proprietary Yes
ProDev
WorkShop
Silicon
Graphics
No No Yes Proprietary Yes
RealView C/C++ Compiler (armcc) KEIL (ARM Limited) Yes Yes ? Proprietary Yes
Salford C++ Compiler Silverfrost Yes No No Proprietary Yes
SAS/C
C++
SAS
Institute
? ? IBM
mainframe
Proprietary ?
Systems/C,C++ Dignus Yes Yes z/OS/Windows/z/Linux/Linux Proprietary Yes
Solaris
Studio
Oracle No Linux,
Solaris
No Proprietary Yes
Solaris
Studio Express
Oracle No Linux,
Solaris
No Freeware Yes
TenDRA TenDRA
Project
No Yes No Open
source
No
VectorC Codeplay Yes No No Proprietary No
Visual C++ Microsoft Yes No No Proprietary Yes
Visual
C++ Express
Microsoft Yes No No Freeware Yes
VisualAge C++ IBM Yes AIX,
Linux
OS/2 Proprietary Yes
XL C/C++ IBM No Linux (POWER and Cell) and AIX BlueGene, z/OS and z/VM Proprietary Eclipse
Wind
River (Diab) Compiler
Wind River
Systems
Yes Yes Yes Proprietary Yes

Common Lisp
compilers

Compiler Author Target Windows Unix-like Other OSs License type IDE?
Allegro
Common Lisp
Franz, Inc. Native code Yes Yes Yes Proprietary Yes
Armed Bear
Common Lisp

JVM Yes Yes Yes GPL Yes
CLISP
Bytecode Yes Yes Yes GPL No
Clozure CL Clozure Associates Native code Yes Yes No LGPL Yes
CMU Common
Lisp

Native code, Bytecode No Yes No Public Domain Yes
Corman
Common Lisp

Native code Yes No No Proprietary Yes
Embeddable
Common Lisp

Bytecode, C Yes Yes Yes LGPL Yes
GNU Common
Lisp

C Yes Yes No GPL No
LispWorks LispWorks Ltd Native code Yes Yes No Proprietary Yes
Open Genera Symbolics Ivory emulator No Yes No Proprietary Yes
Scieneer
Common Lisp
Scieneer Pty Ltd Native code No Yes No Proprietary No
Steel
Bank Common Lisp

Native code Yes Yes Yes Public Domain Yes

ECMAScript
interpreters

Eiffel
compilers

Compiler Author Windows Unix-like Other OSs License type IDE?
Eiffel Studio Eiffel Software / Community developed (sourceforge) Yes Yes Yes Dual Open source /Proprietary Yes

Fortran compilers

Compiler Author Windows Unix-like Other OSs License type IDE?
Sun Studio Sun
Microsystems, Inc.
No Linux and Solaris No Proprietary No
PGI Workstation The Portland Group Yes Linux only Yes Proprietary No
PathScale Compiler Suite SiCortex No Linux only No Proprietary Yes
Absoft Pro Fortran Absoft Yes Linux and Mac OS X Yes Proprietary Yes
G95 Andy Vaught Yes Yes Yes Open
source
No
gfortran GNU Yes Yes Yes Open
source
No
Intel
Fortran Compiler
Intel Yes Linux and Mac OS X No Proprietary No
Lahey
Fortran
Lahey Computer Systems, Inc. Yes Linux only No Proprietary No
Open64 Google, HP,
Intel, Nvidia, PathScale, Tsinghua
University
and others
Yes Yes Yes Open
source
No
FTN95 Silverfrost Yes No No Proprietary Yes
NAG
Fortran Compiler
Numerical
Algorithms Group
Yes Yes No Proprietary Yes
VL
Fortran
IBM No No z/OS and z/VM Proprietary Eclipse
XL
Fortran
IBM No Linux (POWER and Cell) and AIX BlueGene Proprietary Eclipse

Haskell compilers

Compiler Author Windows Unix-like Other OSs License type Actively maintained?
HBC Lennart
Augustsson
Thomas
Johnsson
? Yes No Open
source
No
GHC GHC Yes Yes No Open
source
Yes
NHC Niklas Röjemo Yes Yes No Open
source
No
YHC YHC Yes Yes No Open
source
No
JHC John Meacham Yes Yes No Open
source
Yes

Pascal compilers

Compiler Author Windows Unix-like Other OSs License type IDE?
CodeGear Delphi Embarcadero
(CodeGear)
Yes No No Proprietary Yes
Free Pascal FreePascal.org Yes Yes Yes Open
source
Lazarus
GCC
Pascal
GNU Project Yes Yes Yes Open
source
No
Kylix Borland (CodeGear) No Yes
(Linux)
No Proprietary Yes
Object
Pascal for Windows
Borland (CodeGear) Yes
(3.x)
No No Proprietary Yes
Turbo
Pascal
CodeGear (Borland) No No Yes Freeware Yes
Vector
Pascal
Glasgow
University
Yes Yes No OpenSource No
Scheme

compilers

Compiler Author Target Windows Unix-like Other OSs License type IDE?
Gambit Marc Feeley C Yes Yes Yes LGPL No
Chez Scheme Kent Dybvig native ? Yes ? Proprietary No

Smalltalk compilers

Compiler Author Windows Unix-like Other OSs License type IDE?
Smalltalk
MT
ObjectConnect Yes No No Proprietary Yes

CIL compilers

Compiler Author Windows Unix-like Other OSs License type IDE?
Microsoft
Visual Studio
Microsoft Yes No No Proprietary Yes
mono Mono Yes Yes No Open
source
Yes
Portable.NET Dot Gnu
Project
Yes Yes No Open
source
?
ILDJIT Simone Campanoni No Yes No GPLv2 No

Open source
compilers

Production quality, open source compilers.

Research compilers

Research compilers are mostly not robust or complete enough to handle real,
large applications. They are used mostly for fast prototyping new language
features and new optimizations in research areas.

  • MILEPOST GCC:
    popular interactive plugin-based open-source research compiler that combines
    the strength of the production quality stable GCC that
    supports more than 30 families of architectures, multiple languages and can
    compile real, large applications including Linux, and the flexibility of the
    common Interactive
    Compilation Interface
    that transforms production compilers into
    interactive research toolsets. It is the first production compiler that
    features interactive plugin framework and machine learning engine to be able
    to adapt to any architecture automatically and predict profitable
    optimizations. It has been originally developed during 2006-2009 by the MILEPOST
    consortium
    including IBM, INRIA, University of Edinburgh, ARC and CAPS
    Entreprise. Since 2009, MILEPOST GCC is a part of the community-driven Collective
    Tuning Initiative
    to improve the quality and reproducibility of the
    research on code and architecture optimization, develop collaborative research
    infrastructure with unified interfaces and common
    optimization repository
    and boost innovation.
  • Programming Without Coding Technology (PWCT)
    A specialist innovative technology wherein the programmer need not write code
    but can visually specify every functional aspect of the program similar to
    flowcharts and algorithms. PWCT include (Mahmoud Programming Language, RPWI
    Environment & DoubleS [Super Server] Programming Paradigm).PWCT is
    free-open source. PWCT uses Interaction by presenting a GUI between a Human
    language and a Programming language so doing anything require knowing
    Procedure instead of being Declarative.
  • Open64: one of the most
    popular research compilers today, many branches exist. Here is a list of
    research papers from the CGO
    2009
    . (Open64 merges the open source changes from the PathScale compiler
    mentioned.)
  • Interactive
    Compilation Interface
    - a plugin system with high-level API to transform
    production-quality compilers such as GCC into
    powerful and stable research infrastructure while avoiding developing new
    research compilers from scratch.
  • SUIF:
    inactive since 2001
    • MACHINE SUIF a branch focusing on machine-specific analyses
      and optimizations
  • ROSE compiler
    framework
    : an open source compiler framework to generate source-to-source
    analyzers and translators, developed at Lawrence
    Livermore National Laboratory

  • PIPS:
    a source-to-source compiler framework with a Fortran 77, Fortran 95 and C
    front-end, focussing on advanced analyses and transformations.
  • Polaris compiler for Fortran
    • Cetus for C/C++, successor of Polaris compiler
  • OpenIMPACT Compiler
  • Phoenix
    optimization and analysis framework by Microsoft

  • Very Portable Optimizer (VPO) from the University of Virginia

  • COINS compiler infrastructure
  • Trimaran for research in instruction-level parallelism
  • Parafrase-2 Inactive. It is a source-to-source
    vectorizing/parallelizing compiler, with Fortran and C front-ends.
    • The PARADIGM compiler. Derived from Parafrase-2, it is a
      source-to-source research compiler for distributed-memory multicomputers for
      Fortran 77 and HPF.
  • MLton
    standard ML compiler (SML compiler)

  • Jikes Research Virtual machine(RVM): a research compiler for
    Java
  • Soot: a Java Optimization framework
  • The Scale compiler
  • HotpathVM:
    a Java virtual machine using a trace-based just-in-time
    compiler

  • ILDJIT:
    a compilation framework that targets the CIL bytecode that includes both
    static and dynamic compilers. ILDJIT provides a plugin-based framework for
    static, as well as dynamic tasks like code translations, code analysis, code
    optimizations, runtime instrumentation and memory management. Its plugin-based
    framework allows users to easily customize execution both at installation
    time, as well as at run-time (by dynamically loading and unloading plugins
    without perturbing execution). ILDJIT thus enables efficient co-design
    research at the architectural-boundary. Moreover, its multi-threaded design
    allows novel introspection of parallel compilation strategies to reduce
    overheads and dynamically optimize running code on today's x86 multi-core
    systems.

See also

External links