PHYS2020 - Computational Physics, based on the C programming language

http://www.phys.unsw.edu.au/~mcba/phys2020

Copyright (c) Michael Ashley 2004.

Quick links: assignment questions, preliminary marks.


New: sample final exam, extra notes on numerical modelling of data, etc. A program showing linear fitting to data.

Example mid-session questions Solutions to Q11, 12, and 13.

Zipfile containing all the on-line notes (last updated: 13 Jun 2004) (you can download this to your computer, and browse the contents off-line).

Student feedback April 2004, and my responses.

These notes are a work in progress, and will be updated regularly. The complete notes will be handed out in hardcopy form in week 14.

The course syllabus, assessment details, etc are available on-line.

Please have a look at the preliminary marks to date, and make sure that there is nothing missing. Contact Michael Ashley ASAP if there is.

IMPORTANT: if you are finding the course difficult, see me (Michael Ashley, Room 129 Old Main Building, phone 9385 5465, e-mail m.ashley@unsw.edu.au) for assistance. If there are concepts you are having difficulty with, please let me know as soon as possible so that I can cover them in lectures and provide additional notes on the web.

Also, to gain maximum benefit from the course, you should TURN UP TO LECTURES! It is NOT sufficient to simply download these notes from the web. The lectures contain additional value. I also spend considerable time answering questions from the students, covering exam questions, giving hints on the assignments, and reviewing the answers to past assignments.

The lectures

Lectures often have an "aside" where I talk for 10-15 minutes on some interesting (to me, hopefully to you as well) non-examinable topic related to the use of computers in physics. Some of these asides are mentioned below.

In 2004 the lectures were as follows:

Week  Date   Topics
----------------------------------------------------------------------------
1 Mar Introduction to the various ways that physicists use
computers. Overview of computer languages, including C, C++, C#,
FORTRAN, perl, python, Java, IDL, Mathematica, Maple.
5 Mar GNU/Linux, bit/bytes/words, binary/octal/hexadecimal, I/O redirection
 8 Mar Assignments, how they work, what is expected
12 Mar cygwin, online manual pages, more on assignments
15 Mar floating point, operations, AND, OR, XOR
19 Mar operations, assignment operators (e.g., +=)
22 Mar integer arithmetic, arrays, functions
 26 Mar (Aside: first results from MASS) for loops, while loops, printf,
static/automatic variables, scope of variables.
5 Apr No lecture.
19 Apr (Aside: seeing from Dome C), random numbers - why they are useful -
how to generate them, structures, using the gdb debugger.
23 Apr The Mandelbrot set, complex numbers, using Imagemagick.
26 Apr Public holiday.
30 Apr Solutions to the example mid-session exam.
3 May Mid-session exam.
7 May Quasi-random numbers, the Sobel sequence, Gnuplot,
sorting numbers, qsort.
10 May Answers to assignments 5 & 6. Bug in original sobel.c.
The effect of array-out-of-bounds errors. John Horton
Conway's Game of Life. Life32.
14 May Answers to the mid-session exam questions.
17 May 1D cellula automata.
  21 May Command-line arguments.
24 May Gravity.
28 May Gravity; Numerical integration; midpoint method; 4th order Runge-Kutta.
31 May Statistical analysis of data; moments of distributions; means,
variance, standard-deviations, average deviation; skew, kurtosis;
rounding errors when calculating the variance.
4 Jun Answer to assignment 9; API - application programming interface;
dynamic allocation of memory at run-time - malloc; more on pointers.
7 Jun Median; chi-square minimisation; least-square; linear
fitting; polynomial fitting; generalised fitting.
11 Jun Notes handed out; GNU readline; exam discussion.

In 2003 the lectures were as follows:

Week  Date   Topics
----------------------------------------------------------------------------
1 3 Mar Introduction to the various ways that physicists use
computers. Overview of computer languages, including C, C++, C#,
FORTRAN, perl, python, Java, IDL, Mathematica, Maple.
7 Mar hello.c, compiling, make, printf.
2 10 Mar Introduction to PHYS2020 assignments, C data types: integers,
floating point, chars, character strings. C operators: unary and
binary.
14 Mar for, while, loops. The circle program.
3 17 Mar (Aside: the AASTINO experiment). Bytes, words, floating point
representation.
21 Mar Mandelbrot program. Floating-point "for" loops. Infinite loops.
The math library, libm.a.
4 24 Mar Arrays.
28 Mar Pointers. GNU debugger.
5 31 Mar (Aside: gamma ray bursts). GRB030329. Image analysis.
4 Apr scanf, finding the min/max of an array.
6 7 Apr GNU readline, assert, random numbers,
11 Apr Random numbers, sorting, qsort.
7 14 Apr Gnuplot and tutorial with Anthony Frith.
8 28 Apr Numerical integration, 2-body gravitation, C structures,
example questions for the mid-session exam.
2 May Mid-session exam.
9 5 May (Aside: the evaporation of globular clusters). Numerical
integration, 2-body gravitation.
9 May Numerical integration, 2-body gravitation,
Euler and midpoint approximations.
10 12 May 1D cellular automata, Stephen Wolfram's A New Kind of Science,
command-line arguments.
16 May Mid-term exam solutions Q1-Q3.
11 19 May (Aside: NOAA weather satellite imagery). Mid-term exam
solutions Q4-Q9.
23 May Mid-term exam solutions Q10. 2D cellular automata,
Conway's Game of Life.
12 26 May Life32, different rules for 2D cellular automata,
(Aside: Turing machines, artificial intelligence),
speeding-up our version of Conway's Game of Life,
solution to Assignment 9.
30 May (Aside: Tierra artificial life program by Thomas Ray),
finding gliders automatically in the Game of Life, the roots of a
quadratic equation: subtle issues with rounding errors and
handling special cases.
13 2 Jun Subtleties in calculating standard deviations; average deviations;
the switch statement; allocating memory.
6 Jun Simulations of gas particles, xgas, writing our own code to
track the motion of particles hitting walls.
14 9 Jun Public holiday

Partial index

Why learn C?

  1. C is the most widely used programming language.
  2. The basic ideas in C are common to many other programming languages.
  3. Even though FORTRAN is still in heavy use by physicists, it is on its way out.
  4. If you want to get a computer-related job outside the field of Physics, it is helpful to know C.
  5. Many scientific instruments are programmed in C (e.g., CCD cameras and special purpose interface cards invariably come with a C library interfaces).
  6. C is closer to assembly language, so you can have finer control over what the computer is doing, and thereby make faster programs.
  7. There is a free C compiler available (GNU C, gcc), that is of very high quality and that has been ported to numerous machines.
  8. UNIX is written in C, so it is easier to interface with UNIX-like operating systems (such as GNU/Linux) if you write in C.
  9. Once you have a working knowledge of C, you will find perl easy to learn, and perl is a very useful language to know, but that is another story...

Some problems with C

A brief history of C

C evolved from a language called B, written by Ken Thompson at Bell Labs in 1970. Ken used B to write one of the first implementations of UNIX. B in turn was a decendant of the language BCPL (developed at Cambridge (UK) in 1967), with most of its instructions removed.

In fact, so many instructions were removed in going from BCPL to B, that Dennis Ritchie of Bell Labs put some back in (in 1972), and called the language C.

The famous book The C Programming Language was written by Kernighan and Ritchie in 1978, and has remained the definitive reference book on C.

The original C was still too limiting, and not standardized, and so in 1983 an ANSI committee was established to formalise the language definition.

By 1993 the ANSI standard became well accepted and almost universally supported by compilers. The GNU C compiler, gcc, has become, in many ways, the standard implementation of C.

What about C++ and C#?

C++ is a superset of C, incorporating many object-oriented programming ideas. However, it is not particularly widely used. Java is a better choice in most cases.

C# is a Java-like language developed by Microsoft. Enough said.

Interesting books

Other on-line resources

For coverage of typical problems encountered with C, have a look at the Comp.lang.c FAQ.

For a rather good tutorial introduction to C, check out this document by Gordon Dodrill.

The simplest possible C program

Here it is.

Hello world

Before reading further, have a look at the generic Hello World, program written in C.

But before we get too excited and start to write really useful programs, there are a few fundamental concepts that you will need to understand first.

Computing fundamentals

Bits, bytes, and words

Bit

A bit is the smallest unit of information. It can be thought of as the answer to a yes/no question. A bit has the value 1 or 0 (which can be equated with yes/no, true/false).

Byte

A byte consists of eight bits, ordered from the most significant bit (MSB) on the left, to the least significant bit (LSB) on the right. E.g., 011011000 is a byte. A byte can be interpretted as a base-2 number.

There are 256 (i.e., 2 to the power 8) possible combinations of  8 bits, ranging from 00000000 to 11111111. These 256 combinations could, in principle, be used to represent any 256 things (e.g., we could define 00000000 to be "apple" and 00000001 to be 3.14159265359).

In practice there are three commonly used interpretations of a byte:

Unsigned byte

An unsigned byte can represent integers from zero to 255.

Signed byte

A byte can also be used to represent integers from -128 to +127. Such a byte is called a signed byte, and the MSB is used to indicate the sign (0 for positive, 1 for negative), -128 is represented as 10000000, -1 is 11111111, 0 is 0, and +127 is 01111111.

Non-examinable: To find the bit pattern for a negative number, you first write down the representation of the number without the minus sign, then invert all the bits, then add one. e.g., -1 is 00000001 inverted (which is 11111110), plus one (which gives 11111111). This representation is called "2's complement", and is used because (1) it makes binary arithmetic easier to implement in the computer hardware, and (2) it avoids the waste of a bit pattern in the case of 1000000, which you might think was -0.

A byte as an ASCII character

The American Standard Code for Information Interchange (ASCII) associates the various possible bit patterns in a byte with characters (i..e., things like 'A', 'B', 'C'...'Z', 'a', 'b', 'c', ...'z', '*', '&', '@', etc).  For example, 'A' is stored as 65 (decimal). The ASCII table allows you to translate a simple text document into a list of bytes (which you might then store in a file, see below). For a complete list of the ASCII table, type "man ascii".

Words

In order to represent numbers outside the range of an unsigned or signed byte, multiple bytes can be grouped together to form words of various lengths. Words usually contain a power of two bytes, e.g., 2, 4, 8, 16, and so on. They may either be signed or unsigned. For example, an unsigned 2-byte word can represent numbers from 0 to 65536 (2 to the power 16), a signed 2-byte word ranges from -32768 to + 32767. Words are sometimes called "short words", "words", and "long words", for the 2, 4, and 8-byte variations, but this usage is not uniform.

32-bit computer, 64-bit computers

Most PCs are 32-bit computers, i.e., they process 32-bits (4 bytes) of data in one operation. In a few years, 64-bit computer will become very common.

Binary, octal, decimal, hexadecimal

Integers can be written in various bases. The common bases used with computers are binary (base-2), octal (base-8), decimal (base-10), and hexadecimal (base-16). Octal and hexadecimal are often useful when doing low-level programming, since they encode a binary number in a smaller amount of typing than using binary notation.

The following table compares the various numbering systems (note that you can't directly specify a binary constant in the C programming language); octal constants are distinguished by a leading zero; hexadecimal constants by a leading "0x"):

 Decimal Binary Octal Hexadecimal
--------------------------------
0 0000 000 0x0
1 0001 001 0x1
2 0010 002 0x2
3 0011 003 0x3
4 0100 004 0x4
5 0101 005 0x5
6 0110 006 0x6
7 0111 007 0x7
8 1000 010 0x8
9 1001 011 0x9
10 1010 012 0xa
11 1011 013 0xb
12 1100 014 0xc
13 1101 015 0xd
14 1110 016 0xe
15 1111 017 0xf

Conversion between binary, octal, and hexadecimal is fairly trivial using the above table. For example, to convert the binary number 11100101010 to octal, break the number into groups of three bits starting from the least significant bit (i.e., 11 100 101 010), then convert each three-bit group to octal (i.e., 3452) and add a leading 0 to tell C it is octal (i.e., 03452). To convert the number to hexadecimal, break it into groups of four bits (i.e., 111 0010 1010), convert each group (i.e., 72a) and add a leading "0x" (i.e., 0x72a).

To convert between octal and hexadecimal, it is probably easiest to go via binary.

To convert from binary to decimal, note that n-th bit from the least-significant is worth 2 to the power n. So,

    11100101010 
\\\\\\\\\\\_ 0 x 1 = 0
\\\\\\\\\\_ 1 x 2 = 2
\\\\\\\\\_ 0 x 4 = 0
\\\\\\\\_ 1 x 8 = 8
\\\\\\\_ 0 x 16 = 0
\\\\\\_ 1 x 32 = 32
\\\\\_ 0 x 64 = 0
\\\\_ 0 x 128 = 0
\\\_ 1 x 256 = 256
\\_ 1 x 512 = 512
\_ 1 x 1024 = 1024

Total = 1832

Floating-point numbers

Scientific computation often requires the use of floating-point numbers, e.g., 1.23. There is an IEEE standard that describes how floating-point numbers can be represented as a group of bytes. The basic idea is to allocate a certain number of bits to the exponent (base-2), and the rest to the mantissa (which has been normalised to be between 0.5 and 1). Both the exponent and mantissa have sign bits.

Floating-point numbers typically come in 4, 8, or 16 byte sizes. In C these are usually called "float", "double", and "long double" respectively.

For any given number of bytes allocated to store a floating-point number, there is clearly a tradeoff between the range of the exponent, and the precision of the mantissa: the more bits you use for the exponent, the fewer are left over for the mantissa. We will see later how this compromise has been chosen by the IEEE for the various byte sizes.

Note that there are some redundant bit patterns in the representation of floating-point numbers: e.g.,  0.0  with any exponent is still zero. One of these bit patterns is used as a flag to indicate that a calculation produced an illegal result. This bit pattern is called "NaN" which stands for "not a number". E.g, if you take the square root of a negative number, you will receive the result NaN. Any operation on a NaN will continue to produce a NaN (e.g., if you subtract two NaNs you will still have a NaN), this is useful to propagate such error conditions in your programs so that you are aware when they affect the final result of a calculation.

GNU/Linux fundamentals

The operating system

The operating system of a computer is, somewhat loosely speaking, the program that is always operating in the background on the computer. It handles the interaction with the hardware, and running other programs for the user(s) of the computer. The operating system that we will be using is called GNU/Linux. Other examples of operating systems are Windows XP, OpenBSD, and FreeBSD.

The shell

The shell is an interactive program that allows the user to interface with the operating system. This is the program that provides the command-line prompt (it also helps you run programs, allows you to use the arrow keys on your keyboard to edit the command-line, and a bunch of other useful things). The shell that we will be using is called bash, it is the standard shell that is used with GNU/Linux. Other examples of shells are ksh, tcsh, and zsh.

Files

A file consists of two components:

File data

The data in a file is simply a list of bytes. The list may be empty (i.e., the file may have no data). The list of bytes might be used for a variety of different purposes, for example, it may be a C program, an e-mail message, a photograph, a song, a movie. In principle there is no way to distinguish the purpose of a file's data by looking at it. In practice, you can usually make a pretty good guess. To list the data in a file, you can, e.g., use "cat myfile".

File metadata

The metadata associated with a file consists of miscellaneous additional information that the operating system keeps with the file. For example: the name of the file, the number of bytes in the file, the time at which the file was created, which user owns the file, and the file permissions (i.e., the list of users who read and/or write the file, and whether the file is executable (i.e., is a program that can be run)). The term "metadata" means "data about data". To list the metadata associated with a file, you can, e.g., use "stat myfile".

File names and file extensions

When you create a file, you give it a name. The name is a list of  bytes. In principle the name could be bizarre, such as "  _1  `", but in practice you should choose simple, lowercase, names without spaces, e.g., "thesis.txt", "photo.jpg", "assignment0001.c". Note the convention of suffixing a file name with an extension that gives a clue as what the file is being used for. So, C programs invariably end with the ".c" extension.

The GNU/Linux filesystem

Directories (folders)

To assist with organising groups of files, GNU/Linux has the concept of a directory (or folder). A directory is actually just a file (this adheres to the UNIX philosophy of making things simple and logical). A directory contains a list of the file names that are "in" the directory. Directories can contain directories. This leads to a hierarchical filesystem, where, in order to fully specify the location of a file, you have to specify the list of directories to follow to reach the file. By convention a slash character '/' is used to separate the directory names when giving the fully qualified filename. "/" itself is known as the root directory - there are no directories above the root directory in the filesystem. "/thesis.txt" would then refer to file "thesis.txt" in the root directory. "/home/fred/thesis/chapter4.txt" would be the file "chapter4.txt" in the directory "thesis", in the directory "fred", in the directory "home", in the root directory.

The Windows operating system uses letters such as "C:", "D:", etc to specify different filesystems (which might be in on different hard disks). GNU/Linux places these filesystems within the "/" hierarchy, i.e., the floppy disk might be accessible as "/mnt/floppy", the CD-ROM might be "/mnt/cdrom".

The current working directory

GNU/Linux has the concept of the current working directory. You can specify a file in the current working directory by just giving the file name. For convenience, "./" refers to the current working directory, and "../" refers to the directory above the current working directory. The fully-qualified filename of a file in the current working directory can be written as, e.g., "./thesis.txt".

Programs

To get a computer to do something for you, you create and run a program. You will be doing this many times during this course. To create a program, you type in the program's source code, and then use, e.g, the C compiler (another program, which someone else wrote) to turn the source code into a program. The program itself is just a file. The file's metadata tells the operating system that the file is executable (i.e., that it is a program able to be run by the computer). To run the program, you just type its fully qualified filename.

GNU/Linux commands

GNU/Linux provides you with many hundreds of programs, often called commands, but they are really exactly like the programs that you can create yourself (actually, some of the commands are a subset of the bash shell, but this is a detail).  The commands range from listing the contents of a directory ("ls"), listing the contents of a file ("cat"), the C compiler ("gcc"), to the GNU image manipulation program ("gimp").

To run a program, you simply type its filename. In the case of your own programs, you should use the fully qualified filename (e.g., "./a.out"). In the case of  programs provided by GNU/Linux, the shell has a list of default directories where it searches for programs, so you normally only need to type the filename (without any directories). E.g., the "ls" program to list directories can be run by simply typing "ls", or "/bin/ls", or, if your current working directory happened to be "/bin", you could type "./ls".

Input and output

In general, for a program to do useful work, it needs to be provided with input of some sort, and to produce output.

Input to a program can take many forms: e.g., typing at the keyboard, a file containing data, movements of a mouse, or something more exotic such as audio input from a sound card, or video from a camera.

Output from a program can also take many forms, e.g., text appearing on your computer screen, a graph on your screen, a file of results, or something more exotic such as audio output, or movement of a robotic arm.

GNU/Linux provides the following input and output methods for all programs (I am covering all of the possibilities here, but you are only expected to really understand the first two at this stage):

Standard input

Standard input (or STDIN) is a stream of bytes provided to the program. The program can read from the stream, and can determine when there is no more data. Standard input comes from one of three sources depending on how you run your program. For example, suppose that your program is called "a.out" (the default name assumed by the C compiler), then


./a.out standard input comes from the keyboard
./a.out < file standard input comes from a file called "file"
command | ./a.out standard input comes from the standard output of "command"

Standard output

Standard output (or STDOUT) is a stream of bytes produced by a program. Standard output goes to one of three locations depending on how you run your program. E.g.,

./a.out standard output appears on the computer screen as text
./a.out > file standard output is sent to the file called "file"
./a.out | command standard output is sent to the standard input of "command"

Standard error

Standard error (or STDERR) is a stream of bytes produced by a program. Usually it contains error messages (e.g., 'divide by zero') that are not normally considered as part of the program output. Standard error can be sent to one of three locations. E.g.,

./a.out standard error appears on the computer
screen as text, intermingled with the standard output
./a.out 2> file standard error is sent to the file called "file"
./a.out 2>&1 | command standard error (and standard output) is sent to
the standard input of "command"

Command-line arguments

Anything you type after the program name, and before any special shell characters such as '>' and '|', is made available to your program as a command-line argument. These can be a useful source of additional input to a program. E.g.,

./a.out 23.45 albatross 7

In the above example, "23.45", "albatross", and "7" are three command-line arguments.We will see later how you can read them from within a C program.

Exit status

Another form of output that is available to all programs is an integer exit status. By convention, this is used to indicate whether the program ran successfully (in which case the exit status is zero) or if an error occurred (in which case the exit status is nonzero and its value can be decoded to determine what the error was). The exit status of a program is stored in the shell variable called "$?" (don't worry if you don't understand this at this stage).

Putting it all together

Here are some more complex examples of program input/output:

./a.out 2.1 < data.txt > results.txt
"2.1" is an argument, standard input
comes from "data.txt", standard output
goes to "results.txt"
./a.out > results.txt 2> error.txt
there are no arguments, standard input
is from the keyboard, standard output
to "results.txt" and standard error to
"error.txt"

Command toolkit

Here are some of the basic GNU/Linux commands that you will be using:

cat myfile - concatenate a file (i.e., list the file's data on standard output)

less myfile - like "cat", but allows you to scroll up and down in the file using the page-up, and page-down keys. Exit with 'q'.

rm myfile - remove a file (i.e., delete the file; it is gone for good)

mv myfile newname - renames a file

cp myfile copyOfMyFile - makes a copy of a file, with a different name

cd mydir - change directory

mkdir mynewdir - make a new directory

rmdir mydir - remove the directory (you have to remove the files in it first)

ls - list the files and directories in the current working directory

pwd - print the name of the current working directory

gcc myprog.c - compiles your program

./a.out - runs the program compiled as above

man mytopic - displays the manual pages about "mytopic"

man -k mytopic - searches the manual pages for any information about "mytopic"

gedit myprog.c - starts up a simple GUI editor

logout - logs out of the computer

Fundamentals of C

Character constants, escape sequences, and string constants

A character constant in C is a single character (or an escape sequence such as \n) enclosed in single quotes, e.g., 'A'. It occupies one byte of storage.

The value of a character constant is the numeric value of the character in the computer's character set (e.g., 'A' has the value 65). In 99.99% of cases this is the ASCII character set, but this is not defined by the C standard!

But how do you represent a character such as a single quote itself, or one that doesn't print (such as the ASCII BEL character)? The answer is to use an escape sequence.

For reference, here is a program to print out all the special escape sequences. Have a look at it now.

In addition, you can specify any 8-bit ASCII character using either \ooo or \xhh where `ooo' is an octal number (with from 1 to 3 digits), and 'xhh' is a hexadecimal number (with 1 or 2 digits). For example, \x20 is the ASCII character for SPACE, so ' ' and '\x20' are identical.

String constants are a sequence of zero or more characters, enclosed in double quotes. For example, "test", "", "this is an invalid string" are all valid strings (you can't always believe what a string tells you!). String constants are stored in the computer's memory as a sequence of numbers (usually from the ASCII character set), and are terminated by a null byte (\0). So, "test" would appear in memory as the numbers 116, 110, 115, 116, 0.

A consequence of the null byte terminator is that a C character string can never contain a null byte. In practice, this isn't often a limitation.

We have already used some examples of strings in our programs, e.g, "hello world\n" was a null-terminated character string that we sent to the printf function above.

The above program also shows how to add comments to your C program.

Comments

To make your program more readable, it is good practice to include comments. These do not influence the operation of the program, nor are they printed out when the program runs, they are simply to document things that you think are important.

Comments are delimited by `/*' and `*/', or from '//' to the end of the line. Any text between the first `/*' and the next `*/' is ignored by the compiler, irrespective of the position of line breaks. Note that this means that comments do not nest. Here are some examples of the valid use of comments:

/* This is a comment */
// and so is this
/* here is another one
that spans two lines */
i = /* a big number */ 123456;

Here are some problems that can arise when using comments:

  i = 123456; /* a comment starts here
i = i + 2; this statement is also part of the comment */
/* this is a comment /* and so is this */ but this will generate an error */

The fact that comments don't nest is a real nuisance if you want to comment-out a whole section of code. In this case, it is simplest to use pre-processor directives to do the job. For example:

  i = 123456;
#if 0 // The next two lines are ignored by the compiler.
i = i + 2;
j = i * i;
#endif

Compiling, linking, executing, etc

Source files

C source files contain the C program that you write. The files invariably end with the suffix ".c". You compile them with the C compiler to produce...

Object files

Have a ".o" suffix, and are the result of compiling a source file. They contain the machine language equivalent of the C program, but they are not ready to execute yet, since they need to be linked against libraries of other C object files containing any functions that are used, but not coded for, in the source file (e.g., the printf function). Linking can be done for you by the C compiler, or can be a separate step if you wish. After linking, you have an...

Executable file

This file is ready to run, by typing its name at the command prompt.

Makefiles - the easy way to automate compilation

When developing software, a great deal of time can be save by using the UNIX make utility.

The idea behind make is that you should be able to compile a program simply by typing make (followed by ENTER). For this to work, you have to tell the computer how to build your program, by storing instructions in a Makefile.

While this step takes some effort, you are amply repaid by the time savings later on. make is particularly useful for large programs that consist of numerous source files: make only recompiles the file(s) that need recompiling (it works out which ones to process by looking at the dates of last modification).

Here is an example of a simple Makefile:

test: test.o; gcc test.o -o test
Notes:
  1. A Makefile consists of a series of lines containing targets, dependencies, and shell commands for updating the targets.
  2. In the above example,
  3. Important note: test is a very bad name for a program since it conflicts with the UNIX shell command of the same name, hence if you type "test" it will run the UNIX command rather than your program. To get around this problem, simply use "./test" (if "test" is in your current directory), or rename the program. It is always a good idea to use the leading "./" since it avoids a lot of subtle problems that can be hard to track down.
Makefiles can get much more complicated than this simple example. Here is the next step in complexity, showing the use of a macro definition and a program that depends on multiple files.
OBJS = main.o sub1.o sub2.o 
main: $(OBJS); gcc $(OBJS) -o main
It is well worth learning a little bit about make since it can save a lot of time!

How to compile and run a C program

  1. Create a new directory for the program (not essential, but it helps to be organised).
  2. Use whatever editor you want to generate the program (make sure the program file has a suffix of '.c').
  3. Construct a Makefile for the program (not essential, but worth doing).
  4. Type make to compile the program (if you don't have a Makefile, you will need to type the compiler command yourself).
  5. Run the program by typing its name.
  6. Locate and fix bugs, then go to step 4.
Here is a complete example of the above process for the ``Hello world'' program:
cd
mkdir hello
cd hello
cat > hello.c <<EOF
#include <stdio.h>
int main(void) {
printf("hello world\n");
return 0;
}
EOF
cat > Makefile <<EOF
hello: hello.o; gcc hello.o -o hello
EOF
make
./hello

NOTE: the use of "cat > hello.c <<EOF" above is purely a quick way of entering the source code into the file hello.c. You would normally do this using an editor (such as emacs, vi, pico, nano, nedit, xedit, gedit).


EXERCISE: try doing this yourself with some of the example programs later in these notes.

A program to give information on C data types

Like other programming languages, C has a variety of different data types. The ANSI standard defines the data types that must be supported by a compiler, but it doesn't tell you details such as the range of numbers that each type can represent, or the number of bytes of storage occupied by each type. These details are implementation dependent, and defined in the two system include-files "limits.h" and "float.h". To find out what the limits are, try running the following program with your favourite C compiler.
/* A program to print out various machine-dependent constants */
/* Michael Ashley / UNSW / 03-Mar-2003 */

#include <stdio.h> /* for printf definition */
#include <limits.h> /* for CHAR_MIN, CHAR_MAX, etc */
#include <float.h> /* for FLT_DIG, DBL_DIG, etc */

int main(void) {

printf("char %d bytes %d to %d \n", sizeof(char ), CHAR_MIN, CHAR_MAX );
printf("unsigned char %d bytes %d to %d \n", sizeof(unsigned char ), 0 , UCHAR_MAX );
printf("short %d bytes %hi to %hi \n", sizeof(short ), SHRT_MIN, SHRT_MAX );
printf("unsigned short %d bytes %hu to %hu \n", sizeof(unsigned short), 0 , USHRT_MAX );
printf("int %d bytes %i to %i \n", sizeof(int ), INT_MIN , INT_MAX );
printf("unsigned int %d bytes %u to %u \n", sizeof(unsigned int ), 0 , UINT_MAX );
printf("long %d bytes %li to %li \n", sizeof(long ), LONG_MIN, LONG_MAX );
printf("unsigned long %d bytes %lu to %lu \n", sizeof(unsigned long ), 0 , ULONG_MAX );
printf("float %d bytes %e to %e \n", sizeof(float ), FLT_MIN , FLT_MAX );
printf("double %d bytes %e to %e \n", sizeof(double ), DBL_MIN , DBL_MAX );
printf("precision of float %d digits\n", FLT_DIG);
printf("precision of double %d digits\n", DBL_DIG);
return 0;
}

Notes:

Most of the time you don't need to worry about how many bytes are in each data type, since the limits are usually OK for normal programs. However, a common problem is that the "int" type can be either 2-bytes or 4-bytes in length, depending on the compiler.

Here is the output of the preceeding program when run on a GNU/Linux PC, using gcc:

char           1 bytes -128 to 127
unsigned char 1 bytes 0 to 255
short 2 bytes -32768 to 32767
unsigned short 2 bytes 0 to 65535
int 4 bytes -2147483648 to 2147483647
unsigned int 4 bytes 0 to 4294967295
long 4 bytes -2147483648 to 2147483647
unsigned long 4 bytes 0 to 4294967295
float 4 bytes 1.175494e-38 to 3.402823e+38
double 8 bytes 2.225074e-308 to 1.797693e+308
precision of float 6 digits
precision of double 15 digits

Constants; binary, octal, hexademical, and decimal

We have already seen how to write character constants and strings. Let's now look at other types of constants:

int            123, -1, 2147483647, 040 (octal), 0xab (hexadecimal)
unsigned int 123u, 2147483648, 040U (octal), 0X02 (hexadecimal)
long 123L, 0XFFFFl (hexadecimal)
unsigned long 123ul, 0777UL (octal)
float 1.23F, 3.14e+0f
double 1.23, 2.718281828
long double 1.23L, 9.99E-9L

Note:

It pays to be very careful when specifying numbers, to make sure that you do it correctly, particularly when dealing with issues of precision. For example, have a look at this program:

#include <stdio.h>

int main(void) {
double r;
r = 1.0 + 0.2F;
r = r - 1.2F;
printf("%22.16e\n", r);
return 0;
}

The program gives at output of "-4.4703483581542969e-08", not zero as you might expect.

The lesson to be learnt here is when writing constants, always think carefully about what type you want them to be, and use the suffixes "U", "L", and "F" to be explicit about it. It is not a good idea to rely on the compiler to do what you expect. Don't be surprised if different machines/compilers give different answers if you program sloppily.

Conversion between integers and floating point numbers

In C, as in all computer languages, there are rules that the compiler uses when a program mixes integers and floating point numbers in the same expression.

Let's look at what happens if you assign a floating point number to an integer variable:

#include <stdio.h>

int main(void) {
int i, j;
i = 1.99;
j = -1.99;
printf("i = %d; j = %d\n", i, j);
return 0;
}

This program produces the result "i = 1; j = -1". Note that the floating point numbers have been truncated and not rounded.

When converting integers to floating-point, be aware that a "float" has fewer digits of precision than an "int", even though they both use 4 bytes of storage (on normal PCs). This can result in some strange behaviour, e.g.,

#include <stdio.h>
int main(void) {
unsigned int i;
float f;
i = 4294967295; /* the largest unsigned int */
f = i; /* convert it to a float */
printf("%u %20.13e %20.13e\n", i, f, f - i);
return 0;
}

This program produces the following output when compiled with "gcc":

4294967295 4.2949672960000e+09 1.0000000000000e+00

Rather than rely on automatic type-conversion, you can be explicit about it by using a type-cast operator, e.g.,

f = (float)i;

This converts the number or variable or parethesised expression immediately to its right, to the indicated type. It is a good idea to use type-casting to ensure that you leave nothing to chance.

Operators

C has a rich set of operators (i.e., things like + - * /), which allow you to write complicated expressions quite compactly.

Unary operators

Unary operators are operators that only take one argument. The following list will be confusing when you first see it, so just keep it mind for reference later on.

Some of the operators we have already seen (e.g., "sizeof()"), others are very simple (e.g., +, -), others are really neat (e.g., ~, !), others are useful for adding/subtracting 1 automatically (e.g., ++i, --i, i++, i--), and the rest involve pointers and addressing, which will be covered in detail later.

        sizeof(i)       the number of bytes of storage allocated to i 
+123 positive 123
-123 negative 123
~i one's complement (bitwise complement)
!i logical negation (i.e., 1 if i is zero, 0 otherwise)
*i returns the value stored at the address pointed to by i
&i returns the address in memory of i
++i adds one to i, and returns the new value of i
--i subtracts one from i, and returns the new value of i
i++ adds one to i, and returns the old value of i
i-- subtracts one from i, and returns the old value of i
i[j] array indexing
i (j) calling the function i with argument j
i.j returns member j of structure i
i->j returns member j of structure pointed to by i

Binary operators

Binary operators work on two operands ("binary" here means 2 operands, not in the sense of base-2 arithmetic).

Here is a list. All the usual operators that you would expect are there, with a whole bunch of interesting new ones.

    +          addition
- subtraction
* multiplication
/ division
% remainder (e.g., 2%3 is 2), also called 'modulo'
<< left-shift (e.g., i<<j is i shifted to the left by j bits)
>> right-shift
& bitwise AND
| bitwise OR
^ bitwise exclusive-OR (XOR)
&& logical AND (returns 1 if both operands are non-zero; else 0)
|| logical OR (returns 1 if either operand is non-zero; else 0)
< less than (e.g., i<j returns 1 iff i is less than j)
> greater than
<= less than or equal
>= greater than or equal
== equals
!= does not equal
? conditional operator, explained later...
Note: Here is a table showing the result of applying the logical operators AND, OR, and XOR (exclusive-OR) to all possible combinations of two single-bit numbers:

Left argument
Right argument
AND
OR
XOR
0
0
0
0
0
0
1
0
1
1
1
0
0
1
1
1
1
1
1
0

The following table shows the differences between the bitwise and non-bitwise logical operators when applied to three different randomly chosen 4-digit binary numbers:

Left argument
Right argument
&
|
^
&&
||
0011
0101
0001
0111
0110
1
1
1110
1010
1010
1110
0100
1
1
0000
1111
0000
1111
1111
0
1

Assignment operators

Assignment operators are really just binary operators. The simplest example is "=", which takes the value on the right and places it into the variable on the left. But C provides you with a host of other assignment operators which make life easier for the programmer.
   =      assignment
+= addition assignment
-= subtraction assignment
*= multiplication assignment
/= division assignment
%= remainder/modulus assignment
&= bitwise AND assignment
|= bitwise OR assignment
^= bitwise exclusive OR assignment
<<= left shift assignment
>>= right shift assignment

So, for example, "i += j" is equivalent to "i = i + j". The advantage of the assignment operators is that they can reduce the amount of typing that you have to do, and make the meaning clearer. This is particularly noticeable when, instead of a simple variable such as "i", you have something complicated such as "position[wavelength + correction_factor * 2]";

The thing to the left of the assignment operator has to be something where a result can be stored, and is known as an "lvalue" (i.e., something that can be on the left of an "="). Valid "lvalues" include variables such as "i", and array expressions. It doesn't make sense, however, to use constants or expressions to the left of an equals sign, so these are not "lvalues".

The comma operator

C allows you to put multiple expression in the same statement, separated by a comma. The expressions are evaluated in left-to-right order. The value of the overall expression is then equal to that of the rightmost expression.

For example,

  Example                                  Equivalent to
------- -------------
i = ((j = 2), 3); i = 3; j = 2;
myfunct(i, (j = 2, j + 1), 1); j = 2; myfunct(i, 3, 1);

The comma operator has the lowest precedence, so it is always executed last when evaluating an expression. Note that in the example given comma is used in two distinct ways inside an argument list for a function.

Both the above examples are artificial, and not very useful. The comma operator can be useful when used in "for" and "while" loops as we will see later.

Precedence and associativity of operators

The precedence of an operator gives the order in which operators are applied in expressions: the highest precedence operator is applied first, followed by the next highest, and so on.

The associativity of an operator gives the order in which expressions involving operators of the same precedence are evaluated.

The following table lists all the operators, in order of precedence, with their associativity:

 Operator                                   Associativity
-------- -------------

() [] ->> . left-to-right
- + ++ -- ! ~ * & sizeof (type) right-to-left (note: unary +, -, *, and &)
* / % left-to-right
+ - left-to-right
<< >> left-to-right
< <= > >= left-to-right
== != left-to-right
& left-to-right
^ left-to-right
| left-to-right
&& left-to-right
|| left-to-right
?: right-to-left
= += -= *= /= %= &= ^= |= <<= >>= right-to-left
, left-to-right

Note: the unary operators + (introducing a positive number), - (introducing a negative number), * (pointer dereferencing), and & (address of); appear twice in the above table. The unary forms (on the second line) have higher precedence that the binary forms.

Operators on the same line have the same precedence, and are evaluated in the order given by their associativity.

To specify a different order of evaluation you can use parentheses. In fact, it is often good practice to use parentheses to guard against making mistakes in difficult cases, or to make your meaning clear.

Side effects in evaluating expressions

It is possible to write C expressions that give different answers on different machines, since some aspects of expression-evaluation are not defined by the ANSI standard. This is deliberate since it gives the compiler writers the ability to choose different evaluation orders depending on the underlying machine architecture. You, the programmer, should avoid writing expressions with side effects.

Here are some examples:

myfunc(j, ++j);  /* the arguments may be the same, or differ by one */
array[j] = j++; /* is j incremented before being used as an index? */
i = f1() + f2(); /* the order of evaluation of the two functions
is not defined. If one function affects the
results of the other, then side effects will
result */

Evaluation of logical AND and OR

A useful aspect of C is that it guarantees the order of evaluation of expressions containing the logical AND (&&) and OR (||) operators: it is always left-to-right, and stops when the outcome is known. For example, in the expression "1 || f()", the function "f()" will not be called since the truth of the expression is known regardless of the value returned by "f()".

It is worth keeping this in mind. You can often speed up programs by rearranging logical tests so that the outcome of the test can be determined as soon as possible.

Another good example is an expression such as "i >= 0 && i <n && array[i] == 0". This will guarantee that the index into "array" is within legal bounds (assuming the array has "n" elements).

Incidentally, the "bash" shell uses "&&" and "||" in similar way to C. For example, the shell command line:

cd /home/fred && ./a.out > output && mv output /home/fred/results/

will first change the current working directory to "/home/fred". If (and only if) this succeeds, the program "./a.out" would be run, sending its output to the file "output". If (and only if) the program was successful (i.e., it had an exit code of zero), the output file would be moved to the directory "/home/fred/results/".

What is a C identifier?

An identifier is the name used for a variable, function, data definition, etc.

Rules for identifiers:

Mathematical functions

Prototypes for the math functions are in the system include-file "math.h", so you should put the line

#include <math.h>
in any C source file that calls one of them.

Here is a list of the math functions defined by the ANSI standard:

sin(x)               sine of x
cos(x) cosine of x
tan(x) tan of x
asin(x) arcsine of x, result between -pi/2 and +pi/2
acos(x) arccosine of x, result between 0 and +pi
atan(x) arctan of x, result between -pi/2 and +pi/2
atan2(y,x) arctan of (y/x), result between -pi and +pi
hsin(x) hyperbolic sine of x
hcos(x) hyperbolic cosine of x
htan(x) hyperbolic tan of x
exp(x) exponential function
log(x) natural logarithm
log10(x) logarithm to base 10
pow(x,y) x to the power of y (x**y in FORTRAN)
sqrt(x) the square root of x (x must not be negative)
ceil(x) ceiling; the smallest integer not less than x
floor(x) floor; the largest integer not greater than x
fabs(x) absolute value of x
ldexp(x,n) x times 2**n
frexp(x, int *exp) returns x normalized between 0.5 and 1; the exponent of 2 is in *exp
modf(x, double *ip) returns the fractional part of x; the integral part goes to *ip
fmod(x,y) returns the floating-point remainder of x/y, with the sign of x
In the above table, "x", and "y" are of type "double", and "n" is an "int". All the above functions return "double" results.

C libraries may also include "float" versions of the above. For example, "fsin(x)" takes a float argument and returns a float result.
 

The "for" loop

The basic looping construct in C is the "for" loop.

Here is the syntax of the "for" statement:

   for (initial_expression; loop_condition; loop_expression) statement;

and this is precisely equivalent to the following code fragment:

   initial_expression;
while (loop_condition) {
statement;
loop_expression;
}

An example will, hopefully, clear this up:

   for (i = 0; i < 100; i++) printf("%i\n", i);

which simply prints the first 100 integers. If you want to include more that one statement in the loop, use curly brackets to delimit the body of the loop, e.g.,

   for (i = 0; i < 100; i++) {
j = i * i;
printf("i = %i; j = %i\n", i, j);
}

And here is a more complex example, illustrating nested loops, with an if thrown in for good measure:

ASCII circle printer.

And here is another program illustrating the power of loops: Mandelbrot image generator

How to link with the math library

gcc -o prog prog.c -lm

The "-l" switch stands for "library", which means that the specified library of pre-compiled C routines is searched in order to satisfy any external references from your program "prog.c". The library that is searched in this case is "libm.a" and the path that is used for the search is the default library search path, which includes the directory "/usr/lib", where "libm.a" is found.

To use a library in a directory that is not part of the default library search path, you use the "-L" switch. For example, to search the library "/usr/users/smith/libnano.a", you would use

gcc -o prog prog.c -L/usr/users/smith -lnano

Note: the order of the switches is important. External references are only searched for in libraries to the right of the reference. So, if you have two libraries that call each other, then you need to do something like the following:

gcc -o prog prog.c -L/usr/users/smith -llib1 -llib2 -llib1

Here is a simple example of calling the math library:

#include <stdio.h>
#include <math.h>
int main(void) {
const double pi = 3.14159265358979323846;
double e, d = pi/2;

e = sin(d);
printf("The sine of %f is %f\n", d, e);
return 0;
}

This program produces the result:

The sine of 1.570796 is 1.000000

Aside: the file /usr/include/math.h defines M_PI to be the double precision value of pi. So, use this rather than typing the value of pi yourself.

Semicolons in compound statements

The last statement in the body of statements in a "for" loop (or, in fact, in any other compound statement) must be terminated with a semicolon.

For example,

for (i = 0; i < 10; i++) {
x = i * i;
x += 2; /* the semicolon is required here */
} /* do not use a semicolon here */

Variables defined within compound statements

You can create variables that are local to a compound statement by declaring the variables immediately after the leading curly bracket.

Variable storage classes

Variables in C belong to one of two fundamental storage classes: "static" or "automatic".

A static variable is stored at a fixed memory location in the computer, and is created and initialised once when the program is first started. Such a variable maintains its value between calls to the block (a function, or compound statement) in which it is defined.

An automatic variable is created, and initialised, each time the block is entered (if you jump in half-way through a block, the creation still works, but the variable is not initialised). The variable is destroyed when the block is exited.

Variables can be explicitly declared as "static" or "auto" by using these keywords before the data-type definition. If you don't use one of these keywords, the default is "static" for variables defined outside any block, and "auto" for those inside a block.

Actually, there is another storage class: "register". This is like "auto" except that it asks the compiler to try and store the variable in one of the CPU's fast internal registers. In practice, it is usually best not to use the "register" type since compilers are now so smart that they can do a better job of deciding which variables to place in fast storage than you can.

Const - volatile

Variables can be qualified as "const" to indicate that they are really constants that can be initialised, but not altered.

Variables can also be termed "volatile" to indicate that their value may change unexpectedly during the execution of the program (e.g., they may be hardware registers on a PC, able to be altered by external events). By using the "volatile" qualifier, you prevent the compiler from optimising the variable out of loops.

Extern

Variables (and functions) can also be classified as "extern", which means that they are defined external to the current block (or even to the current source file). An "extern" variable must be defined once (and only once) without the "extern" qualifier.

As an example of an "extern" function, all the functions in "libm.a" (the math library) are external to the source file that calls them.

An example showing storage class and variable scope

#include <stdio.h>
int i; /* i is static, and visible to the entire program */
extern j; /* j is static, and visible to the entire program */
static int j; /* k is static, and visible to the routines in this source file */

void func(void) { /* i.e., a function that takes no arguments, and
doesn't return a value */
int m = 1; /* m is automatic, local to this function, and initialised
each time the function is called */
auto int n = 2; /* n is automatic, local to this function, and initialised
each time the function is called */
static int p = 3; /* p is static, local to this function, and initialised
once when the program is first started up */
extern int q; /* q is static, and defined in some external module */

for (i = 0; i < 10; i++) {
int m = 10; /* m is automatic, local to this block, and initialised
each time the block is entered */
printf("m = %i\n", m);
}
}

Initialisation of variables

A variable is initialised by equating it to a constant expression on the line in which it is defined. For example

  int i = 0;

"static" variables are initialised once (to zero if not explicitly initialised), "automatic" variables are initialised when the block in which they are defined is entered (and to an undefined value if not explicitly initialised).

The "constant expression" can contain combinations of any type of constant, and any operator (except assignment, incrementing, decrementing, function call, or the comma operator), including the ability to use the unary "&" operator to find the address of static variables.

Here are some valid examples:

#include <stdio.h>
#include <math.h>
int i = 0;
int j = 2 + 2;
int k = 2 * (3 << 8) / 3;
int m = (int)(&i + 2);
int p = sizeof(float) * 2;
int q = sizeof(p);
float r = (float)(2 * 3);

int main(void) {
printf("i = %i\n", i);
printf("j = %i\n", j);
printf("k = %i\n", k);
printf("m = %i\n", m);
printf("p = %i\n", p);
printf("q = %i\n", q);
printf("r = %f\n", r);
for (r = 0.0; r < 1.0; r += 0.1) {
double s = sin(r);
printf("The sine of %f is %f\n", r, s);
}
return 0;
}

Notes:

If statements

  if (expression)
statement
else if (expression)
statement
else if (expression)
statement
else
statement

Where "statement" is a simple C statement ending in a semicolon, or a compound statement ending in a curly bracket.

Some examples will help:

  if (i == 6) x = 3;

if (i == 6) {
x = 3;
}

if (i) x = 3;

if (i == 2)
x = 3;
else if (i == 1)
if (j)
y = 2;
else /* NOTE: possible ambiguity here, the compiler uses */
y = 3; /* the closest if statement that does not have */
else { /* an else clause */
x = 4;
y = 5;
}

Break and continue

These two statements are used in loop control.

"break" exits the innermost current loop.

"continue" starts the next iteration of the loop.

Infinite loops

Here are three ways of writing an infinite loop, choose whichever one you like:

for (;;;) {
statement;
}

while (1) {
statement;
}

do {
statement;
} while (1);

Infinite loops can be useful. They are normally terminated using a conditional test with a "break" or "return" statement. For example:

while (1) {
statement;
if (logical_condition) break;
}

goto

C does have a "goto" statement, but you normally don't need it. Using "goto" is usually a result of bad programming (although in some circumstances it can be a sensible choice).

The switch statement

The switch statement allows you to choose between multiple paths of execution depending on the value of an expression. It could be replaced by a series of if statements, but switch can make the flow of the program easier to understand.

switch (expression) {
case const-expression: statements case const-expression: statements default: statements }

Beware that control will flow from one case to the next unless you explicitly include a break statement to cause control to skip to the end of the switch statement.

Here is an example of using switch:

switch (i) {
case 0: printf("i had the value zero\n");
break;
case 1:
case 2: printf("i was either one or two\n");
break;
default: printf("i was outside the range 0 to 2\n");
}

Formatted output: printf description

The format string given to the "printf" function may contain both ordinary characters (which are simply printed out) and conversion characters (beginning with a percent symbol, %, these define how the value of an internal variable is to be converted into a character string for output).

Here is the syntax of a conversion specification:

  %{flags: - + space 0 #}{minimum field width}{.}{precision}{length modifier}{conversion character}
 Character Type    Result

d,i int signed decimal integer
o int unsigned octal (no leading zero)
x, X int unsigned hex (no leading 0x or 0X)
u int unsigned decimal integer
c int single character
s char * characters from a string
f double floating point [-]dddd.pppp
e, E double exponential [-]dddd.pppp e[=/-]xx
g, G double floating if the exponent less than -4 or >= precision
else exponential
p void * pointer
n int * the number of characters written so far by printf
is stored into the argument (i.e., not printed)
% print %

Here is an example program to show some of these effects (try running the program yourself):

#include <stdio.h>
int main(void) {
int i = 123;
double f = 3.14159265358979323846;

printf("i = %i\n", i);
printf("i = %o\n", i);
printf("i = %x\n", i);
printf("i = %X\n", i);
printf("i = %+i\n", i);
printf("i = %8i\n", i);
printf("i = %08i\n", i);
printf("i = %+08i\n", i);
printf("f = %f\n", f);
printf("f = %10.3f\n", f);
printf("f = %+10.3f\n", f);
printf("f = %g\n", f);
printf("f = %10.6g\n", f);
printf("f = %10.6e\n", f);

return 0;
}

Notes:

Formatted input: scanf overview

Input in C is similar to output: the same conversion characters are used. The main difference is that you use a routine called "scanf", and you must pass the addresses of the variables you want to change, not their values.

For example:

  scanf("%d", &i);         /* reads an integer into `i' */
scanf("%i", &i); /* reads an integer (or octal, or hex) into `i' */
scanf("%f %i", &f, &i); /* reads a double followed by an integer */

scanf is actually an integer function, which returns the number of input items assigned (or the integer constant EOF (defined in stdio.h) if the end-of-file is reached or an error occurred).

The ampersand character "&" is a unary operator that returns the address of the thing to its right. Recall that a C function can not alter the value of its arguments (but there is nothing stopping it altering the value that is pointed to by one of its arguments!).

scanf details

The scanf function is used for reading a line of input from the standard input stream (usually the users terminal) and extracting from it one or more integers, floating point numbers, character strings, etc.

The first argument to scanf is a format description, which describes how the input line should be decoded. For example, if you want to read an integer, the format would be "%i" (note that this allows you to enter numbers in decimal, hexadecimal or octal (e.g., 123, 0x123, 0123)). If you want to read a floating point number, the format would be "%f". If you want to read an integer followed by a floating point number, the format woud be "%d %f". If the two numbers were separated by a comma on the input line, you would use "%d, %f".

The arguments after the first are pointers to where the data should be stored.

Examples (where i and j are integers, x and y are floats, and fruit and colour are char arrays):

Input data
scanf function call to read the data
123
scanf("%i", &i);
123   0x456
scanf("%i %i", &i, &j);
123.   456.
scanf("%i. %i.", &i, &j);
123.1  456.1
scanf("%f %f", &x, &y);
e=2.71  pi=3.14
scanf("e=%f  pi=%f", &x, &y);
apples   red
scanf("%s %s", fruit, colour);
Guido ate 2 apples
scanf("Guido ate %i %s", &i, fruit);

It is important to check the return value of the scanf function - this is the number of items that were successfully extracted from the input. It may be less than the number you expected, or even zero. If this happens, the input stream pointer is left at the point at which the error occurred, and you can try reading again. scanf can also return the special integer constant EOF to indicate an end-of-file condition.

Variations on scanf


fscanf is very similar to scanf except it takes a FILE argument ("FILE" is a structure that is defined in the <stdio.h> include file - you don't have to worry about what precisely "FILE" is, just follow the examples) to specify which input stream to read from, e.g.,

FILE *fp;
float x, y;

fp = fopen("input.dat", "r");
fscanf(fp, "%f %f", &x, &y);

"sscanf" takes input from a character array rather than an input stream, e.g.,

char input="1.23 4.56";
sscanf(input, "%f %f", &x, &y);

Improving the ease of interaction with your programs - GNU readline

While it is possible to use "scanf" for reading input typed by a person, you can greatly improve the user experience by using the GNU readline package.

The GNU readline package provides some very convenient features such as line editing and history. This means, e.g., that you can use the arrow keys to move within a line, and between previously entered lines. You can also use, e.g., CTRL-A to go to the beginning of the line, CTRL-E to go to the end, and CNTRL-R to search backwards for particular strings. See "man readline" (on a GNU/Linux computer) for a complete description.

Here is an example program showing the use of GNU readline.

Assert as a tool for improving program reliability


By now you will be aware that it is very easy to write buggy programs in C, and it may not always be obvious at runtime that there is a problem. Checking for all possible error conditions in a program can be tedious. A useful tool to improve program reliability is the "assert" construct defined in the system include-file "assert.h". An example will clarify this:
#include <assert.h>
#include <stdio.h>

int main(void) {
int i;

printf("enter a positive integer > ");
assert(1 == scanf("%i", &i));
assert(i > 0);
printf("you successfully entered %i\n", i);
  return 0;
}
If the expression inside parentheses after "assert" is true, the program continues as normal, if the expression is false (i.e, zero), then the program aborts with an error message.

This may sound like a trivial concept, but by using "assert" extensively, you will be surprised at how your programming improves.

It is particularly useful to use "assert" to check for conditions that should never arise in your program, e.g., suppose you write a function that should only ever be called with an argument between -1.0 and 1.0, then, use "assert" in the function to check this, e.g.,
double myAsin(double x) {
assert(fabs(x) < 1.0);
...
}
Also, you can use assert to check for unexpected conditions such as failure to open a file, or failure to allocate memory, e.g.,
assert(NULL != (fp = fopen("data.dat", "r")));
assert(NULL != (p = malloc(1024));

Generating random numbers

Random numbers are used in all sorts of ways in mathematics and physics. They are a fundamental tool that is often useful when conducting numerical simulations of physical systems with a computer.

Writing a computer program to generate random numbers requires understanding some fairly subtle concepts.

There are several different algorithms for generating a new random number given an initial random number. One of the simplest (and best) algorithms for generating random integers is a multiplicative congruential algorithm: you simply multiply the current random number generator by a constant, and then take the modulus. The choice of the constant and the modulus is critical. A good choice is 16807 and 2147483647 (see Numerical Recipes in C, Press et al, 2nd Ed, page 278).

But how do you generate the initial random number, the so-called "seed", for the algorithm?

One technique is to use the time (in milli or micro-seconds since the last integral second). This is easy to program, but has a few disadvantages, e.g, there are only 1000 or 1000000 different seeds, and these numbers are not necessarily equally likely to occur (e.g., the operating system might tend to start programs at particular times - see here for more details).

Another technique on GNU/Linux systems is to read from the file /dev/random. /dev/random contains random numbers generated from "random" events such as the arrival time of internet packets, the movement of the mouse, the timing of user keystrokes, etc. If you "od -h /dev/random" ("od -h" prints out the file as hexadecimal characters) you will find that it will eventually stop, waiting for new events to generate randomness - move the mouse and it will continue. An alternative is /dev/urandom, which will continuously generate random numbers, using an algorithm if events are too infrequent.

If you require very high quality random numbers, then you should read about the subject, perhaps using Numerical Recipies in C as a starting point.

The C libraries provide the functions "srandom" and "random" to generate random integers between 0 and RAND_MAX (defined in "stdlib.h") inclusive. Use the on-line manual pages to learn about "srandom" and "random" (i.e., type "man srandom" at the prompt). Here is an example program showing their usage, and seeding them with the time in microseconds.

Quasi-random numbers

For some applications, you don't want truly random numbers, but would prefer numbers that cover an n-dimensional space more unformly. For example, let's take a more detailed look at Monte-Carlo integration.

When using random numbers, the error involved in Monte-Carlo integration goes down as 1/sqrt(N), where N is the number of points. This means, e.g., if you use 10 times as many points, the error in your estimation of the integral will be reduced by a factor of about 1/sqrt(10), or 0.32. Can we do better than this? It turns out that a simple uniform grid of points has an error that goes as 1/N, which is about 3 times better than the random case. However, a uniform grid has the undesirable property that you have to decide on the grid spacing in advance, and you can't simply add more points at a later time.

What we need is a quasi-random number generator that chooses numbers that tend to fill the gaps in the sequence. There are several possibilities, and we will explore one designed by a Russian mathematician by the name of Sobel in the week 7 assignment.


Sorting

A frequent requirement is to sort quantities (numbers, character strings, or more complicated structures) into some order (ascending, descending, or something more complex). The C library provides the function "qsort", an implementation of the famous quicksort algorithm. It is usually best to use "qsort" than write your own sorting algorithm.

Here are two example programs showing the use of qsort.

Gnuplot

Here are Anthony Frith's notes on Gnuplot.

Numerical integration

Here is an example of using numerical integration to follow the motion of a planet around the Sun:

Writing fast programs

Reasons for worrying about how fast a program is include:

Hints on improving speed:

2D cellular automata - Conway's Game of Life

In 1970, the mathematician John Horton Conway published the description of a simple 2D cellular automata which he called the Game of Life. Cells in a 2D grid were either "alive" or "dead", and their state in the next generation was determined by their current state and the number of nearest neighbours (from 0 to 8, inclusive). The rules were chosen to simulate some properties of biological systems (e.g., a cell would die of "overcrowding" if too many of its neighbours were alive, and die of "lack of support" if too few of its neighbours were alive).

Here is a "simple" example of how to program the Game of Life on a computer.

If we are interested in following the evolution of the game for many thousands of generations, it is advantageous to think of ways of speeding up our simple implementation. The first thing to try is to turn on full optimisation during the compilation, using the "-O4" switch. The next step is to make the critical functions "inline" (see here for the code). We can also think about the algorithm, and realise that a lot of our calculation of the number of nearest neighbours was involved in handling the special case of being on the boundary; we can re-write our program to separate out the boundary case, thereby allowing a simplified (faster) function to do most of the neighbour calculations, at the expense of increased complexity in handling the evolution from one generation to the next. Finally, we can try to use the fact that much of the Life "universe" tends to be sparsely populated, allowing us to produce a second neighbourhood-calculating function for the special case that the preceeding cell had no nearest neighbours. These are by no means the only speed-ups that are possible, but they give you some flavour of what is possible.

The following table gives the time taken to follow one million generations using a 850MHz Pentium III computer and the various tricks from the last paragraph.

    Time      Technique
[seconds]
--------------------------------------------------
330 Original program, no optimisation
205 Original program, -O4 optimisation
113 Inline function calls 45 Boundary case handled separately 37 Sparse population handled separately

So, we have managed almost a factor of ten improvement in speed over our original attempt.

To show why this might be useful, let's now alter our program to automatically identify glider patterns (those that can self-propagate). We will choose a brute-force technique where we randomly seed the centre of our rectangular "universe" with alive/dead cells, and then evolve the system until one of the boundary cells is hit. When this happens we stop the evolution and print a 21x21 grid around the cell, hopefully capturing a glider in full flight! Here is the code, and you can see that it is a straightforward variation on the earlier program. It typically finds a glider within 2 seconds on a 850MHz Pentium III computer.

The on-line manual pages

Use "man -k keyword" to search for information about "keyword". For example, suppose you want to know something about random number generators: "man -k random" on a GNU/Linux system will return output containing lines like these:

ppmspread            (1)  - displace a portable pixmap's pixels by a random amount
rand (3) - random number generator
RAND_bytes (3ssl) - generate random data
random (3) - random number generator
random (4) - kernel random number source devices
RAND_pseudo_bytes [RAND_bytes] (3ssl) - generate random data
rand [sslrand] (1ssl) - generate pseudo-random bytes
rand [sslrand] (3ssl) - pseudo-random number generator
seed48 [drand48] (3) - generate uniformly distributed pseudo-random numbers

The number in parentheses is called the section number (e.g., section 3 is reserved for C functions, section 4 for low-level kernel interfaces), and you need to specify this if there are entries in multiple volumes. So, e.g., to read about the kernel random number source devices, you would type "man 4 random".

The GNU C debugger "gdb"

When debugging a program, a common technique is to insert "print" statements to display the values of variables. A much more powerful technique is to use the GNU C debugger, "gdb". Other C compilers will have debuggers as well.

Firstly, you must compile your program with the debug switch (-g). It is also a good idea to disable optimisation with -O0.

gcc -g -O0 main.c

Then run your program using gdb:

gdb a.out

The following are some of the more common commands that are available:

help                       - obtain help
help {arg} - obtain help on topic {arg}
list - list 10 lines around current position
list {line} - list 10 lines around line number {line}
list {beg},{end} - list lines from line number {beg} to {end}
run arg1 arg2 ... - begin execution of the program, with the arguments
break line - suspend execution at the given line number
step - steps (and executes) one line in your program
print {exp} ... - print the value of the given expressions
printf "string", exp, ... - print expressions using format string(C)
quit - quit gdb
<ENTER> - repeats the last command (very useful with "step")

Arrays

In mathematics and physics you often need to group numbers together into arrays, i.e., 1, 2, or higher-dimensional structures.

Arrays are declared in C as follows (for example):

int counts[100];
float temperature[1024];
double mat[2][3];

In this example, "count" is an array that can hold 100 integers, "temperature" is an array that can hold 1024 floats, and "mat" is a 2x3 array (or matrix) of doubles.

Note carefully that the first element of an array in C is element number 0 (rather than 1 as in FORTRAN). While this may be confusing to die-hard FORTRAN programmers, it is really a more natural choice. A side-effect of this choice is that the last element in an array has an index that is one less that the declared size of the array. This is a source of some confusion, and something to watch out for.

So, "counts[0]" is the first element of the array "counts", defined above. And "counts[99]" is the last element. There are 100 elements in total.

Initialising arrays

To initialise an array to have particular numerical values, specify the initial values in a list within curly brackets. For example:

int primes[100] =
{2, 3, 5, 7, 11, 13, 17, 19, 23, 29, 31, 37, 41, 43,
47, 53, 59, 61, 67, 71, 73, 79, 83, 89, 97, 101, 103, 107,
109, 113, 127, 131, 137, 139, 149, 151, 157, 163, 167, 173, 179, 181,
191, 193, 197, 199, 211, 223, 227, 229, 233, 239, 241, 251, 257, 263,
269, 271, 277, 281, 283, 293, 307, 311, 313, 317, 331, 337, 347, 349,
353, 359, 367, 373, 379, 383, 389, 397, 401, 409, 419, 421, 431, 433,
439, 443, 449, 457, 461, 463, 467, 479, 487, 491, 499, 503, 509, 521,
523, 541};

float temp[1024] = {5.0F, 2.3F};

double trouble[] = {1.0, 2.0, 3.0};

In this example, "primes" is initialised with the values of the first 100 primes (trust me!), and the first two elements (temp[0] and temp[1]) of "temp" are initialised to 5.0F and 2.3F respectively. The remaining elements of "temp" are set to 0.0F. Note that we use the trailing "F" on these numbers to indicate that they are floats, not doubles.

The array "trouble" in the above example contains three double numbers. Note that its length is not explicitly declared, C is smart enough to calculate the length for you.

Static arrays that are not explicitly initialised are set to zero. Automatic arrays that are not explicitly initialised, have undefined values.

Multidimensional arrays

Multidimensional arrays are declared and referenced in C by using multiple sets of square brackets. For example,

int table[2][3][4];

"table" is a 2x3x4 array, with the rightmost array subscript changing most rapidly as you move through memory (the first element is "table[0][0][0]", the next element is "table[0][0][1]", and the last element is "table[1][2][3]").

When writing programs that use large arrays (say, more than a few megabytes), you should be very careful to ensure that array references in your program are as close as possible to being consecutive (otherwise you may get severe swapping problems, i.e., the memory is swapped to hard disk, resulting in more than a factor of 1000 slow-down).

Initialisation of multidimensional arrays

This is best demonstrated with an example:

int mat[3][4] = {
{ 0, 1, 2, 23},
{ 3, 4, 5, 24},
{ 6, 7, 8, 25}
}

Operations on arrays

If you wish to, e.g., add two arrays together, you have to do it element-by-element yourself using a "for" loop. For example:

int a0[10], a1[10], a2[10];

for (int i = 0; i < 10; i++) {
a0[i] = a1[i] + a2[i];
}
Even something as simple as setting one array to have the same value as another has to be done with a "for" loop. E.g. :
int a0[10], a1[10];

for (int i = 0; i < 10; i++) {
a0[i] = a1[i];
}

Character arrays

Arrays can be of any type. In C, you manipulate character strings as arrays of characters, and operations on the strings (such as concatenation, searching) are done by calling special libary functions (e.g., strcat, strcmp).

Note that when calling a string-manipulation function, the end of the string is taken as the position of the first NUL character (which has a numerical value of zero) in the string.

Character arrays can be initialised in the following ways:

char str[] = {'a', 'b', 'c'};
char prompt[] = "please enter a number";

In the example, "str" has length 3 bytes, and "prompt" has length 22 bytes, which is one more than the number of characters in "please enter a number", the extra byte is used to store a NUL character (zero) as an indication of the end of the string. "str" does not having a trailing NUL.

Pointers

If there is one thing that sets C apart from many other languages, it is the pervasive use of pointers.

To understand pointers, it helps to have a mental image of the way in which the memory of a computer is organised. e.g., imagine that your computer has 128MB of memory; then you could label each one of these 128 million bytes of memory with a number: byte 0 would be the first byte in memory, byte 1 the second, and so on up to 128 million. These numbers are referred to as the  addresses of the memory cells (note: I'm actually simplifying the true situation; in a modern computer each separate process has it's own virtual address-space, and there are several layers of indirection involved in going from the addresses that C uses, to the actual physical piece of memory on the computer motherboard). A pointer is then simply a variable containing a memory address. This is a very natural concept when you think in terms of the hardware of the computer, which is a viewpoint that C encourages.

Suppose that "i" is declared as "int i". Then the address in memory at which "i" is stored is given by "&i", where "&" is the unary address operator. One says that "&i" is a pointer to "i". On a 32-bit computer such as a typical PC, a pointer is 4 bytes in length (therefore allowing up to 4GB of memory to be addressed; this is an uncomfortably small number).

Pointers are declared by specifying the type of thing they point at. For example,

int *p;

defines "p" as a pointer to an int (so, therefore, "*p" is an int, hence the form of the declaration).

Note carefully, then by declaring "p" as in the above example, the compiler simply allocates space for the pointer (4 bytes for 32-bit computers, 8 bytes for 64-bit computers), not for the variable that the pointer points to! This is a very important point, and is often overlooked. Before using "*p" in an expression, you have to ensure that "p" is set to point to a valid int. This "int" must have had space allocated for it, either statically, automatically, or by dynamically allocating memory at run-time (e.g., using the "malloc" function, see later).

Here is a rather convoluted example showing some of the uses of pointers:

#include <stdio.h>
int main(void) {
int m = 0, n = 1, k = 2;
int *p;
char msg[] = "hello";
char *cp;
/* Line 7 */
p = &m; /* Line 8: p now points to m */
*p = 1; /* Line 9: m now equals 1 */
k = *p; /* Line 10: k now equals 1 */
cp = msg; /* Line 11: cp points to the first character of msg */
*cp = 'H'; /* Line 12: change the case of the 'h' in msg */
cp = &msg[4]; /* Line 13: cp points to the 'o' */
*cp = 'O'; /* Line 14: change its case */

printf("m = %d, n = %d, k = %d\nmsg = \"%s\"\n", m, n, k, msg);
return 0;
}

By using the GNU debugger, gdb, we can identify the actual memory locations that the C compiler uses to store the variable in the above program. It is then possible to draw up a table (see below) showing the detailed operation of the program. You should follow this example very carefully so that you understand what is going on.

Variable Starting      Value after the specified line has executed
location 7 8 9 10 11 12 13 14
------------------------------------------------------------------

m 0xbffff98c 0 0 1 1 1 1 1 1
n 0xbffff988 1 1 1 1 1 1 1 1
k 0xbffff984 2 2 2 1 1 1 1 1
p 0xbffff980 ? 98c 98c 98c 98c 98c 98c 98c
msg 0xbffff970 hello hello hello hello hello Hello Hello HellO
cp 0xbffff96c ? ? ? ? 970 970 976 976

The memory addresses are shown in hexadecimal, and only the rightmost three digits are shown (e.g., 98c) for convenience in displaying the table. Memory addresses are always given in units of bytes. The above table shows, for example, that the variable "m" occupies the memory addresses 0xbffff98c, 0xbffff98d, 0xbffff98e, and 0xbffff98f. You will note that 16 bytes are reserved for "msg", even though "msg" only needs 6 bytes (5 for the characters in "hello" and one for the trailing null byte). The C compiler won't guarantee any particular placement in memory, and may leave gaps, particularly to align numbers with 4-byte boundaries (i.e., so that their memory addresses are divisable by 4).

Note the very important point that the name of an array ("msg" in the above example), if used without an index, is considered to be a pointer to the first element of the array. In fact, an array name followed by an index is exactly equivalent to a pointer followed by an offset. For example,

#include <stdio.h>

int main(void) {
char msg[] = "hello world";
char *cp;

cp = msg;

cp[0] = 'H';
*(msg+6) = 'W';

printf("%s\n", msg);
printf("%s\n", &msg[0]);
printf("%s\n", cp);
printf("%s\n", &cp[0]);
return 0;
}

Note, however, that "cp" is a variable, and can be changed, whereas "msg" is a constant, and is not an lvalue (i.e., it can not be used on the left of an equals sign).

Pointer arithmetic

You can add and subtract numbers to/from pointers, but the results may surprise you. For example, if "p" is a pointer to an "int" (i.e., "int *p"), then "p++" actually adds 4 to "p"! The logic behind this is that adding one to "p" should make it point to the next integer, which has a memory address that is 4 bytes greater.

Pointers used as arguments to functions

We have already seen that C functions can not alter their arguments. The simple reason for this is that when you call a function, the compiler passes copies of the values of the arguments to the function. For example, if you have a function as follow:

void myfunc(int n) {
n = 6;
}

and you call it with

int i = 0;
myfunc(i);
Then "myfunc" is passed the number 0. "myfunc" knows nothing whatsoever about where this number came from. The zero is stored in the local variable n. "myfunc" changed n to 6, but this doesn't have any effect on i in the main program.

In order to change the value of a variable in the main program, you need to give the function a pointer to the variable. In other words, you tell the function whereabouts in memory the variable is. With this crucial bit of information, the function can go ahead and alter the value of the variable. For example,
void myfunc(int *n) {
*n = 6;
}

and you call it with

int i = 0;
myfunc(&i);

Now, the value of i in the main program will be altered by the call to "myfunc".

Here is an example of a function that swaps the values of its two arguments, which must be pointers to integers:

void swap_args(int *pi, int *pj) {
int temp;

temp = *pi;
*pi = *pj;
*pj = temp;
}

If all the asterisks were left out of this routine, then it would still compile OK, but its only effect would be to swap local copies of "pi" and "pj".

Pointers and arrays

Consider an array declared as follows:
int a[10];

Then, as we have seen, a[0] is the first element of the array, and a[9] is the last. The address of the first element is "&a[0]", and by convention this can also be written simply as "a".

Structures

Any programming language worth its salt has to allow you to manipulate more complex data types than simply ints, floats, and arrays. You need to have structures of some sort. This is best illustrated by example:

To define a structure, use the following syntax:

struct time {
int hour;
int minute;
float second;
};

This defines a new data type, called "time", which contains 3 fields (stored in consecutive locations in the computer's memory).

Logically, it makes sense to associate the various fields of "time" together, since they are part of one physical entity: the time.

To declare a variable of type "time", you could do the following:

struct time t1[10], t2 = {12, 0, 0.0F};

This defines "t1" to be an array of 10 "time" structures. The variable "t2" is also a "time" structure, and is initialised to 12 hours, 0 minutes, 0.0 seconds.

For an example program using structures, see the solution to Q12 of the sample mid-session test here.

To refer to an individual element of a structure, you use the following syntax:

t1[0].hour = 12;
t1[0].minutes = 0;
t1[0].second = t2.second;
t1[1] = t2;

Alternatively, and somewhat confusingly, if "p", say, is a pointer to a structure, then you can access the elements of the structure using notation like "p->hour". For example:

struct time *p;             // p is a pointer to a time structure
struct time t = {1,2,3.0}; // t is a time structure
p = &t; // p now points to t
p->hour = 6; // this statement is exactly equivalent to...
t.hour = 6; // ...this one

Structures can contain any type, including arrays and other structures. A common use is to create a linked list by having a structure contain a pointer to a structure of the same type, for example,

struct person {
char *name[80];
char *address[256];
struct person *next_person;
};

where "person.next_person" is actually a pointer to a "person" struct. Sounds complicated, but it is simple when you get the hang of it.

Structures are the first step towards object-oriented programming (OOP), where most of the programming effort is devoted to defining the best representation of the data, and the operations which can be performed on the data.

In physics, a structure can be used to represent the state of a physical system. For example, here is a structure that might be used to represent the state of a particle in three-dimensional space:
struct particle {
double mass;
double position[3];
double velocity[3];
};

We could then define an array of 1000 particles as simply as:
struct particle p[1000];

You can take this one step further and define your own type (like "int", "double") by doing:
typedef struct {
double mass;
double position[3];
double velocity[3];
} particle;
In which case you can define the array of 1000 particles as:
particle p[1000];

String functions

The standard C libraries include a bunch of functions for manipulating strings (i.e., arrays of chars).

Note that before using any of these functions, you should include the line "#include <string.h>" in your program. This include-file defines prototypes for the following functions, as well as defining special types such as "size_t", which are operating-system dependent.

char *strcat(s1, s2)            Concatenates s2 onto s1. null-terminates s1.
Returns s1.

char *strchr(s, c) Searches string s for character c. Returns a
pointer to the first occurence, or null pointer
if not.

int strcmp(s1, s2) Compares strings s1 and s2. Returns an integer
that is less than 0 is s1 < s2; zero if s1 = s2;
and greater than zero is s1 > s2.

char *strcpy (s1, s2) Copies s2 to s1, returning s1.

size_t strlen(s) Returns the number of characters in s, excluding
the terminating null.

char *strncat(s1, s2, n) Concatenates s2 onto s1, stopping after `n'
characters or the end of s2, whichever occurs
first. Returns s1.

int strncmp(s1, s2, n) Like strcmp, except at most `n' characters are
compared.

char *strncpy(s1, s2) Like strcpy, except at most `n' characters are
copied.

char *strrchr(s, c) Like strchr, except searches from the end of
the string.

int strstr(s1, s2) Searches for the string s2 in s1. Returns a
pointer if found, otherwise the null-pointer.

size_r strspn(s1, s2) Returns the number of consecutive characters
in s1, starting at the beginning of the string,
that are contained within s2.

size_r strcspn(s1, s2) Returns the number of consecutive characters
in s1, starting at the beginning of the string,
that are not contained within s2.

char *strpbrk(s1, s2) Returns a pointer to the first character in s1
that is present in s2 (or NULL if none).

char *strerror(n) Returns a pointer to a string describing the
error code `n'.

char *strtok(s1, s2) Searches s1 for tokens delimited by characters
from s1. The first time it is called with a
non-null s1, it writes a NULL into s1 at the
first position of a character from s2, and
returns a pointer to the beginning of s1. When
strtok is then called with a null s1, it finds
the next token in s1, starting at one position
beyond the previous null.

File operations

<stdio> defines a number of functions that are used for accessing files. Before using a file, you have to declare a pointer to it, so that it can be referred to with the functions. You do this with, for example,

FILE *in_file;
FILE *out_file;

where "FILE" is a type defined in <stdio> (it is usually a complicated structure of some sort). To open a file, you do, for example:

in_file  = fopen("input_file.dat",  "r");
out_file = fopen("output_file.dat", "w");

Note the use of "r" to indicate read access, and "w" to indicate write access. The following modes are available:

"r"   read
"w" write (destroys any existing file with the same name)
"rb" read a binary file
"wb" write a binary file (overwriting any existing file)
"r+" opens an existing file for random read access, and
writing to its end
"w+" opens a new file for random read access, and writing to
its end (destroys any existing file with the same name)

Once the file is opened, you can use the following functions to read/write it:

int fread(void *ptr, int size, int nelm, FILE *fp) Reads nelm elements
of data, each size bytes long, storing them
at the location pointed to by ptr. Returns
the number of elements read.

int fwrite(void *ptr, int size, int nelm, FILE *fp) Writes nelm elements
of data, each size bytes long, from the
location pointed to by ptr. Returns
the number of elements written.

int getc(FILE *fp) Returns the next character from `fp', or
EOF on error or end-of-file.

int putc(int c, FILE *fp) Write the character c to the file `fp',
returning the character written, or EOF on
error.

int fscanf(FILE *fp, char *format, ...) Like scanf, except input is taken
from the file fp.

int fprintf(FILE *fp, char *format, ...) Like printf, except output is written
to the file fp.

char *fgets(char *line, int n, FILE *fp) Gets the next line of input from the
file fp, up to `n-1' characters in length.
The newline character is included at the end
of the string, and a null is appended. Returns
`line' if successful, else NULL if end-of-file
or other error.

int fputs(char *line, FILE *fp) Outputs the string `line' to the file fp.
Returns zero is successful, EOF if not.

When you have finished with a file, you should close it with 'fclose':

int fclose(FILE *fp)            Closes the file 'fp', after flushing any
buffers. This function is automatically called
for any open files by the operating
system at the end of a program.

It can also be useful to flush any buffers associated with a file, to guarantee that the characters that you have written have actually been sent to the file:

int fflush(FILE *fp)            Flushes any buffers associated with the 
file 'fp'.

To check the status of a file, the following functions can be called:

int feof(FILE *fp)              Returns non-zero when an end-of-file is read.

int ferror(FILE *fp) Returns non-zero when an error has occurred,
unless cleared by clearerr.

void clearerr(FILE *fp) Resets the error and end-of-file statuses.

int fileno(FILE *fp) Returns the integer file descriptor associated
with the file (useful for low-level I/O).

Note that the above "functions" are actually defined as pre-processor macros in C. This is quite a common thing to do.

Multiple precision arithmetic

As an example of how to program in C, let's explore the topic of multiple precision arithmetic. All the hard work will be done by GNU's Multiple Precision Arithmetic Library (GNU MP), written by Torbjorn Granlund.

Multiple precision arithmetic allows you to perform calculations to a greater precision than the host computer will normally allow. The penalty you pay is that the operations are slower, and that you have to call subroutines to do all the calculations.

GNU MP can perform operations on integers and rational numbers. It uses preprocessor macros (defined in gmp.h) to define special data types for storing these numbers. MP_INT is an integer, and MP_RAT is a rational number. However, since a multiple precision number may occupy an arbitrarily large amount of memory, it is not sufficient to allocate memory for each number at compile time. GNU MP copes with this problem by dynamically allocating more memory when necessary.

To begin with you need to initialise each variable that you are going to use. When you are finished using a variable, you should free the space it uses by calling a special function.

  MP_INT x, y;
mpz_init(&x);
mpz_init(&y);

/* operations on x and y */

mpz_clear(&x);
mpz_clear(&y);

Let's now try a full program. For example, calculating the square root of two to about 200 decimal places:

#include <gmp.h>
#include <stdio.h>

int main(void) {
char two[450], guess[225];
int i;
MP_INT c, x, temp, diff;

two[0] = '2';
for (i = 1; i < sizeof(two)-1; i++) {
two[i] = '0';
}
two[i] = 0;
mpz_init_set_str(&c, two, 10);

guess[0] = '1';
for (i = 1; i < sizeof(guess)-1; i++) {
guess[i] = '0';
}
guess[i] = 0;
mpz_init_set_str(&x, guess, 10);

mpz_init(&temp);
mpz_init(&diff);

do {
mpz_div(&temp, &c, &x);
mpz_sub(&diff, &x, &temp);
mpz_abs(&diff, &diff);
mpz_add(&x, &temp, &x);
mpz_div_ui(&x, &x, 2U);
} while (mpz_cmp_ui (&diff, 10U) > 0);

printf("the square root of two is approximately ");
mpz_out_str(stdout, 10, &x);
printf("\n");
return 0;
}

To compile, link, and run the program, use

gcc -o two two.c -lgmp
./two

The preprocessor, cpp


Any line in your C source code that starts with the "#" (hash) chararacter is a preprocessor directive. Here are some examples of what is possible:

#define PI 3.1415926535
#define SQR(a) (sqrarg=(a),sqrarg*sqrarg)

#include "filename" /* from the current directory */
#include <filename> /* from the system directories (as modified by the "-I" switch) */

#define DEBUG
/* defines the symbol DEBUG */
#ifdef DEBUG
/* code here is compiled if DEBUG is defined */
#elif defined UNIX
/* code here is compiled if DEBUG is not defined, and UNIX is defined */
#else
/* code here is compiled if neither DEBUG or UNIX are defined */
#endif

#if 0
/* code here is never compiled */
#endif

Command-line arguments, and returning a status to the shell

Up until now, our "main" function has always been declared as "int main(void)". We have seen that the integer return value is available to the UNIX "shell" in the "$?" environment variable, and this can be used to alter the execution of scripts. It turns out that "main" can actually take arguments, and these are traditionally given as "int argc, char *argv[]", where argc is one plus the number of command-line arguments provided to the program, and argv is an array of character strings containing the arguments. argv[0] is usually the name of the program itself. An example will hopefully clarify this:

#include <stdio.h>

int main(int argc, char *argv[]) {
printf("this program is called '%s'\n", argv[0]);
if (argc == 1) {
printf("it was called without any arguments\n");
} else {
int i;
printf("it was called with %d arguments\n", argc - 1);
for (i = 1; i < argc; i++) {
printf("argument number %d was <%s>\n", i, argv[i]);
}
}
exit(argc);
}

echo $?

Allocating memory at run-time - malloc

The situation often arises that you don't know the size of a data structure in your program at compile-time. For example, suppose you are writing a program to read an 2D image from a file into the computer's memory. If the size of the image can vary at run-time, it would be nice not to have to pre-define the size of the memory array in your program. C allows you to do this through the concept of allocating and deallocating memory.

Allocating memory means requesting the computer to give you access to a contiguous block of memory. E.g., suppose you need a megabyte; you ask the computer "can I use a megabyte of memory, and if so, can you please tell me where the memory is?". You ask the question by calling the function "malloc", which takes a single argument: the number of bytes of memory that you want. malloc will attempt to allocate the memory, and if it is successful it will return a pointer to the first byte of the memory block (if malloc was unsuccessful, it returns NULL; this is a rare enough occurence for reasonable memory requests that you should handle it with "assert").

Deallocating memory means telling the computer "I have now finished using this block of memory that you previously allocated for me with malloc, so you can now have the memory back again for some other purpose". You deallocate memory by calling the "free" function.

Some important points to remember when using malloc:

Here is an example to get you started:

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

/* Generates a two dimensional matrix, and fills it randomly with
zeroes and ones. */

int main(void) {
int xdim, ydim;
int i, j;
int *p, *q;

// Obtain the dimensions of the matrix from the user.

printf("x dimension of matrix? > ");
scanf("%d", &xdim);
printf("y dimension of matrix? > ");
scanf("%d", &ydim);

// malloc the memory needed for the matrix. Note the use of "sizeof(int)"
// which allows the program to run on machines with a variety of word sizes.

assert(NULL != (p = (int *)malloc(xdim * ydim * sizeof(int))));

// Fill the matrix with random 0s and 1s.

for (i = 0; i < xdim * ydim; i++) {
if (random() > RAND_MAX/2) {
*(p+i) = 1;
} else {
*(p+i) = 0;
}
}

// Now print the matrix out, showing the use of pointer arithmetic.

for (i = 0; i < ydim; i++) {
q = p + i * xdim;
for (j = 0; j < xdim; j++) {
printf("%1d", *(q++));
}
printf("\n");
}

// Free the memory we had allocated.

free((void *)p);

// And remove all information about where the memory used to be.
// This is overkill in this example, since the program immediately
// terminates, but it is good practice, in order to avoid
// accidentally using free'ed memory.

p = q = NULL;
return 0;
}
C programming course, School of Physics, UNSW / Michael Ashley / mcba@newt.phys.unsw.edu.au