1

Programming in Oberon
A derivative of Programming in Modula-2 (1982)
Niklaus Wirth

Table of Contents
Preface
Part 1
1. Introduction
2. A First Example
3. A Notation to Describe the Syntax of Oberon
4. Representation of Oberon Programs
5. Statements and Expressions
6. Control Structures
6.1 Repetitive Statements
6.2 Conditional Statements
7. Elementary Data Types
7.1 The Type INTEGER
7.2 The Type REAL
7.3 The Type BOOLEAN
7.4 The Type CHAR
7.5 The Type SET
8. Constant and Variable Declarations
9. Arrays
Part 2
10. Procedures
11. The Concept of Locality
12. Parameters
12.1 Variable Parameters
12.2 Value Parameters
12.3 Open Array Parameters
13. Function Procedures
14. Recursion
15. Type Declarations
16. Record Types
17. Dynamic Data Structures and Pointers
18. Procedure Types
Part 3
19. Modules
20. Definitions and Implementations
21. Program Decomposition into Modules
22. The concept of sequence
22.1. About input and output
22.2. Files and Riders
22.3. Texts, Readers and Writers
22.4. Standard Input and Output
Part 4
23. Object-oriented Programming
23.1. The origins of object-oriented programming
23.2. Type extensions and inhomogeneous data structures
23.3. Methods
23.4. Handlers and Messages
Appendix: Syntax, Keywords, Standard functions

2
Preface
This text is an introduction to programming in general, and a guide to programming in the language
Oberon in particular. It is primarily oriented towards people who have already acquired some basic
knowledge of programming and would like to deepen their understanding in a more structured way.
Nevertheless, an introductory chapter is added for the benefit of the beginner, displaying in a concise
form some of the fundamental concepts of computers and of programming them. The text is therefore
also suitable as a self-contained tutorial. The notation used is Oberon, which lends itself well for a
structured approach and leads the student to a working style that has generally become known under the
heading of structured programming.
As a manual for programming in Oberon, the text covers (almost) all facilities of this language. Part 1
covers the basic notions of variable, expression, assignment, conditional and repeated statement, and the
data structure of the indexable array. Together with Part 2, which introduces the important concept of the
procedure or subroutine, it contains essentially the material commonly discussed in introductory
programming courses. Part 3 concerns data types and structures and constitutes the core of an advanced
course on programming. Part 4 introduces the notion of a module, a concept that is fundamental in the
design of large programming systems, and to programming in teams. The two basic notions of files and
texts are introduced in the forms of modules containing the operations performed on files and texts. Texts
are presented as our standard medium for program input and output.
The language Oberon, published in 1988, has a long history. It is a descendant of Algol 60 (1960), Pascal
(1970), and Modula (1979). Algol 60 [1] was designed by an international committee of 13 scientists,
and it was the first language to be specified with a rigorous formalism, a syntax, in machine-independent
form. Algol was largely oriented toward numerical algorithms, as were computers at that time. Pascal [2]
was the result of an enduring debate about widening the range of application of Algol, and it became
widely used, in particular but not only, in education. Modula-2 [3] introduced the concept of modules,
the parts of large systems connected by clearly specified interfaces. The module allows to hide details of
procedures and variables from its users (clients), and it embodies the principle of information hiding. The
language Oberon [4] emerged from the urge to reduce the complexity of programming languages, of
Modula in particular. This effort resulted in a remarkably concise language. The extent of Oberon, the
number of its features and constructs, is smaller even than that of Pascal. Yet it is considerably more
powerful.
The one feature that was added in Oberon was the extensibility of data types (record types). Whereas in
strongly types languages, such as Algol, Pascal, and Modula, every constant, variable, or function has a
fixed type, recognizable from the program text, Oberon allows to define hierarchies of types, and to
determine the actual type of a variable (within the hierarchy) at run-time. This, together with records
containing fields of procedural types, is the stem of object-oriented programming. Such records are then
called objects, and the procedural fields are called methods. Part 4 is added to cover this subject and to
show, that Oberon encourages an object-oriented style of programming with hardly any new language
features.
Zrich, 1. October 2004

N.W.

1. P. Naur, Ed. Revised Report on the Algorithmic Language Algol 60. Comp. J. 5, 349-367 (1962), and
Comm. ACM, 6 (1963) 1 17.
2. N. Wirth. The Programming Language Pascal. Acta Informatica, 1 (1971), 35 63.
3. N. Wirth. Programming in Modula-2. Springer-Verlag, 1982. ISBN 0-387-50150-9
4. N. Wirth. The Programming Language Oberon. Software - Practice and Experience, 18, 7, (July
1988), 671- 690.

Part 1
1. Introduction
Although this manual assumes that its reader is already familiar with the basic notions of computer and
programming, it may be appropriate to start out with the explanation of some concepts and their
terminology. We recognize that - with rare exceptions - programs are written - more appropriately:
designed - with the purpose of being interpreted by a computer. The computer then performs a process,
i.e. a sequence of actions, according to the specifications given by that program. The process is also
called a computation.
The program itself is a text. Since it specifies a usually fairly complex process, and must do so with
utmost precision and care for all details, the meaning of this text must be specified very precisely. Such
precision requires an exact formalism. This formalism has become known as a language. We adopt this
name, although a language is normally spoken and much less precisely defined. Our purpose here is to
learn the formalism or language Oberon. It has a long tradition among programming languages. Its
ancestor was Algol 60, followed by Pascal and Modula-2.
A program usually specifies a process that causes its interpreter, i.e. the computer, to read data (the socalled input) from some sources and to vary its subsequent actions according to the accepted data. This
implies that a program does not only specify a (single) process, but an entire - usually unbounded - class
of computations. We have to ensure that these processes act according to the given specifications (or
should we say expectations?) in all cases of this class. Whereas we could verify that this specification is
met in the case of a single computation, this is impossible in the general case, because the class of all
permitted processes is much too large. The conscientious programmer ensures the correctness of his
program by careful design and analysis. Careful design is the essence of professional programming.
The task of designing a program is further complicated by the fact that the program not only must
describe an entire class of computations, but often should also be interpreted (executed) by different
interpreters (computers). At earlier times, this required the manual transcription of the program from its
source form into different computer codes, taking into account their various characteristics and
limitations. The difficulties have been drastically reduced, albeit not eliminated, by the creation of high
level languages with formal definitions and the construction of automatic translators converting the
program into the codes of the various computers.
In principle, the formal language should be defined in an abstract, perhaps axiomatic fashion without
reference to an actual computer or interpretation mechanism. If this were achieved, the programmer
would have to understand the formal language only. However, such generality is costly and often
restrictive, and in many cases the programmer should still know the principal characteristics of his
computer(s). Nevertheless, the qualified programmer will make as little reference to specific computer
characteristics as possible and rely exclusively on the rules of the formal language in order to keep his
program general and portable. The language Oberon assists in this task by confining computer
dependencies to specific objects, by allowing to encapsulate them in specific, small parts of a program
text.
From the foregoing it follows that a translation process lies between the program's formulation and its
interpretation. This process is called a compilation, because it condenses the program's source text into a
cryptic computer code. The quality of this compilation may be crucial to the efficiency of the program's
ultimate interpretation. We stress the fact that there may be many compilers for a given language (even
for the same computer). Some may be more efficient than others. We recognize that efficiency is a
characteristic of implementations rather than the language. It therefore is important to distinguish
between the concepts of language and implementation.
We summarize:
- A program is a piece of text.
- The program specifies computations or processes.
- A process is performed by an interpreter, usually a computer, interpreting (executing) the program.
- The meaning of the program is specified by a formalism called programming language.
- A program specifies a class of computations, the input data acting as parameter of each individual
process.

4
-

Prior to its execution, a program text is translated into computer code by a compiler. This process is
called a compilation.

Program design includes ensuring that all members of this class of computations act according to
specification. This is done by careful analytic verification and by selective empirical testing of
characteristic cases.
Programs should refrain from making reference to characteristics of specific interpreters (computers)
whenever possible. Only the lack of such reference ensures that their meaning can be derived from rules
of the language.
A compiler is a program translating programs from their source form to specific computer codes.
Programs need to be compiled before they are executed. Programming in the wider sense not only
includes the formulation of the program, but also the concrete preparation of the text, its compilation,
correction of errors, so-called debugging, and the planning of tests. The modern programmer uses many
tools for these tasks, including text editors, compilers, and debuggers. He also has to be familiar with the
environment of these components. We shall not describe these aspects, but concentrate on the language
Oberon.

2. A First Example
Let us follow the steps of development of a simple program and thereby explain some of the fundamental
concepts of programming and of the basic facilities of Oberon. The task shall be, given two natural
numbers x and y, to compute their greatest common divisor (gcd). The mathematical knowledge needed
for this problem is the following:
1. if x equals y, x (or y) is the desired result
2. the gcd of two numbers remains unchanged, if we replace the larger number by the difference of the
numbers, i.e. subtract the smaller number form the larger one.
Expressed in mathematical terms, these rules take the form
1.
2.

gcd(x, x) = x
if x > y, gcd(x, y) = gcd(x-y, y)

The basic recipe, the so-called algorithm, is then the following: Change the numbers x and y according
to rule 2 such that their difference decreases. Repeat this until they are equal. Rule 2 guarantees that the
changes are such that gcd(x,y) always remains the same, and rule 1 guarantees that we finally find the
result.
Now we must put these recommendations into terms of Oberon. A first attempt leads to the following
sketch. Note that the symbol # means "unequal".
WHILE x # y DO
"apply rule 2, reducing the difference and maintaining x > 0 and y > 0"
END
The sentence within quotes is plain English. The second version refines the first version by replacing the
English by formal terms:
WHILE x # y DO
IF x > y THEN x := x-y ELSE y := y-x END
END
This piece of text is not yet a complete program, but it shows already the essential characteristic of a
structured programming language. Version 1 is a statement, and this statement contains another,
subordinate statement (within quotes). In version 2 this is elaborated, and yet further subordinate
statements emerge (expressing the replacement of a value x by another value x-y). This hierarchy of
statements expresses the underlying structure of the algorithm. It becomes explicit due to the structure of
the language, allowing the nesting of components of a program. It is therefore important to know the
language's structure (syntax) in full detail. Textually we express nesting or subordination by appropriate
indentation. Although this is not required by the rules of the language, it helps in the understanding of a
text very considerably.
Reflecting an algorithm's inherent structure by the textual structure of the program is a key idea of
structured programming. It is virtually impossible to recognise the meaning of a program when its

5
structure is removed, such as done by a compiler when producing computer code. And we should keep in
mind that a program is worthless, unless it exists in some form in which a human can understand it and
gain confidence in its design.
We now proceed towards the goal of producing a complete program from the above fragment. We realize
that we need to specify an action that assigns initial values to the variables x and y, as well as an action
that makes the result visible. For this purpose we should actually know about a computer's facilities to
communicate with its user. Since we do not wish to refer to a specific machinery, and particularly not in
such a frequent and important case as the generation of output, we introduce abstractions of such
communication facilities. These facilities are not directly part of the language, but are procedures
declared in some (library) modules, to which all programs have access. We postulate read and write
procedures as shown in the following example, and will assume that data are thus read from a keyboard
and written on a display.
Texts.Scan(S); x := S.i;
Texts.Scan(S); y := S.i;
WHILE x # y DO
IF x > y THEN x := x-y ELSE y := y-x END
END;
Texts.WriteInt(W, x, 6)
The procedure Scan scans an input text and reads a (non-negative) integer (S.i). The procedure WriteInt
outputs an integer as specified by its first parameter (x). The second parameter (6) indicates the number
of digits available for the representation of this value in the output text. Both procedures are taken from
the imported (library) module Texts, and they use a locally declared scanner S and a globally declared
writer W, connecting to an input source (the text following the command) and output sink (the Log text)
imported from the environment Oberon. Details are explained in Chapter 22 of this tutorial.
In the next and final version we complete our text such that it becomes a genuine Oberon text:
PROCEDURE Gcd*;
VAR x, y: INTEGER; S: Texts.Scanner;
BEGIN Texts.OpenScanner(S, Oberon.Par.text, Oberon.Par.pos);
Texts.Scan(S); x := S.i; Texts.WriteString(W, " x ="); Texts.WriteInt(W, x, 6);
Texts.Scan(S); y := S.i; Texts.WriteString(W, " y ="); Texts.WriteInt(W, y, 6);
WHILE x # y DO
IF x > y THEN x := x-y ELSE y := y-x END
END;
Texts.WriteString(W, " gcd ="); Texts.WriteInt(W, x, 6); Texts.WriteLn(W);
Texts.Append(Oberon.Log, W.buf)
END Gcd.
The essential additions in this step are declarations. In Oberon, all names of objects occurring in a
program, such as variables and constants, have to be declared. A declaration introduces the object's
identifier (name), specifies the kind of the object (whether it is a variable, a constant, or something else)
and indicates general, invariant properties, such as the type of a variable or the value of a constant.
The entire piece of text is called a procedure, given a name, and has the following format (see also Ch.
10):
PROCEDURE name;
<declarations>
BEGIN
<statements>
END name
In this and the following examples, a particular property of the Oberon system becomes apparent.
Whereas in older languages, the notions of program and executable unit were identical, we distinguish
carefully between them in Oberon terminology: What used to be called a program is now called a
module, a system unit typically incorporating variables and executable statements resident in a system
(see also Ch. 19). Compilers accept a module as a compilable unit of text. An executable unit of program
we call a procedure. In the Oberon system, any parameterless procedure can be used as a command, if its
name is marked with an asterisk. It can be activated by clicking on its name visible anywhere on the

6
display. A module may contain one or several commands. As a consequence, the preceding and
following examples of programs appear in the form of procedures (commands), and we will always
assume that they are embedded in a module importing two service modules Texts and Oberon and
containing the declaration of a writer W for output:
MODULE Pattern;
IMPORT Texts, Oberon;
VAR W: Texts.Writer;
PROCEDURE command*; (*such as Gcd*)
BEGIN (*using W for output*)
Texts.Append(Oberon.Log, W.buf)
END command;
BEGIN Texts,OpenWriter(W)
END Pattern.
A few more comments concerning our Gcd example are in order. As already mentioned, the procedures
WriteLn, WriteString, WriteInt and Scan are not part of the language Oberon itself. They are defined in a
module called Texts which is presumed to be available. The often encountered modules Texts, Files and
Oberon will be defined and explained in Chapter 22 of this book. Here we merely point out that they
need to be imported in order to be known in a program. This is done by including the names of these
modules in the import list in the importing modules heading.
The procedure WriteString outputs a string, i.e. a sequence of characters (enclosed in quotes). The
procedure WriteLn inserts a line break in the output text. For further explanations we refer to chapters
22.3 and 22.4 of this book:
And this concludes the discussion of our first example. It has been kept quite informal. This is admissible
because the goal was to explain an existing program. However, programming is designing, creating new
programs. For this purpose, only a precise, formal description of our tool is adequate. In the next chapter,
we introduce a formalism for the precise description of correct, "legal" program texts. This formalism
makes it possible to determine in a rigorous manner whether a written text meets the language's rules.

3. A Notation to Describe the Syntax of Oberon

A formal language is an infinite set of sequences of symbols. The members of this set are called
sentences, and in the case of a programming language these sentences are programs. The symbols are
taken from a finite set called the vocabulary. Since the set of programs is infinite, it cannot be
enumerated, but is instead defined by rules for their composition. Sequences of symbols that are
composed according to these rules are said to be syntactically correct programs; the set of rules is the
syntax of the language.
Programs in a formal language then correspond to grammatically correct sentences of spoken languages.
Every sentence has a structure and consists of distinct parts, such as subject, object, and predicate.
Similarly, a program consists of parts, called syntactic entities, such as statements, expressions, or
declarations. If a construct A consists of B followed by C, i.e. the concatenation BC, then we call B and
C syntactic factors and describe A by the syntactic formula
A = BC.
If, on the other hand, an A consists of a B or, alternatively, of a C, we call B and C syntactic terms and
express A as
A = B | C.
Parentheses may be used to group terms and factors. It is noteworthy that here A, B, and C denote
syntactic entities of the formal language to be described, whereas the symbols =, | , parentheses, and the
period are symbols of the meta-notation describing syntax. The latter are called meta-symbols, and the
meta-notation introduced here is called Extended Backus-Naur Formalism (EBNF).
In addition to concatenation and choice, EBNF also allows to express option and repetition. If a construct
A may be either a B or nothing (empty), this is expressed as
A = [B].

7
and if an A consists of the concatenation of any number of Bs (including none), this is denoted by
A = {B}.
This is all there is to EBNF! A few examples show how sets of sentences are defined by EBNF
formulas:
(A|B)(C|D)
A[B]C
A{BA}
{A|B}C

AC AD BC BD
ABC AC
A ABA ABABA ABABABA ...
C AC BC AAC ABC BBC BAC ...

Evidently, EBNF is itself a formal language. If it suits its purpose, it must at least be able to describe
itself! In the following definition of EBNF in EBNF, we use the following names for entities:
statement:
expression:
term:
factor:

a syntactic equation
a list of alternative terms
a concatenation of factors
a single syntactic entity or a parenthesized expression

The formal definition of EBNF is now given as follows:

syntax =
statement =
expression =
term =
factor =

{statement}.
identifier "=" expression ".".
term {"|" term}.
factor {factor}.
identifier | string | "(" expression ")" | "[" expression "]" | "{" expression "}".

Identifiers denote syntactic entities; strings are sequences of symbols taken from the defined language's
vocabulary. For the denotation of identifiers we adopt the widely used conventions for programming
languages, namely:
An identifier consists of a sequence of letters and digits, where the first character must be a letter. A
string consists of any sequence of characters enclosed by quote marks (or apostrophes).
A formal statement of these rules in terms of EBNF is given in the subsequent chapter.

4. Representation of Oberon Programs

The preceding chapter has introduced a formalism, by which the structures of well-formed programs will
subsequently be defined. It defines, however, merely the way in which programs are composed as
sequences of symbols, in contrast to sequences of characters. This "shortcoming" is quite intentional: the
representation of symbols (and thereby programs) in terms of characters is considered too much
dependent on individual implementations for the general level of abstraction appropriate for a language
definition. The creation of an intermediate level of representation by symbol sequences provides a useful
decoupling between language and ultimate program representation. The latter depends on the available
character set. As a consequence, we need to postulate a set of rules governing the representation of
symbols as character sequences. The symbols of the Oberon vocabulary are divided into the following
classes:
identifiers, numbers, strings, operators and delimiters, and comments.
The rules governing their representation in terms of the standard ISO character set are the following:
1. Identifiers are sequences of letters and digits. The first character must be a letter. Capital and lowercase letters are considered as distinct.
identifier = letter {letter|digit}.
Examples of well-formed identifiers are
Alice likely jump

BlackBird

SR71

Examples of words which are no identifiers are

sound proof
sound-proof

(blank space is not allowed)

(neither is a hyphen)

8
2N
Miller's

(first character must be a letter)

(no apostrophe allowed)

Sometimes an identifier has to be qualified by another identifier; this is expressed by prefixing i with j
and a period (j.i); the combined identifier is called a qualified identifier (abbreviated as qualident). Its
syntax is
qualident = {identifier "."} identifier.
2. Numbers are either integers or real numbers. The former are denoted by sequences of digits. Numbers
must not include any spaces. Real numbers contain a decimal point and a fractional part. In addition, a
scale factor may be appended. It is specified by the letter E and an integer which is possibly preceded by
a sign. The E is pronounced as "times 10 to the power of".
Examples of well-formed numbers are
1981

3.25

5.1E3

4.0E-10

Examples of character sequences that are not recognized as numbers are

1,5
1'000'000
3.5En

no comma may appear

neither may apostrophs
no letters allowed (except the E)

The exact rules for forming numbers are given by the following syntax:
number = integer | real.
integer = digit {digit}.
real = digit {digit} "." {digit} [ScaleFactor].
ScaleFactor = "E" ["+"|"-"] digit {digit}.
Note: Integers are taken as octal numbers, if followed by the letter B, or as hexadecimal numbers if
followed by the letter H.
3. Strings are sequences of any characters enclosed in quote marks. In order that the closing quote is
recognized unambiguously, the string itself evidently cannot contain a quote mark. To allow strings with
quote marks, a string may be enclosed within apostrophes instead of quote marks. In this case, however,
the string must not contain apostrophes.
string = '"' {character} '"' | "'" {character} "'".
Examples of strings are
"no comment"
"Buck's Corner"
'he said "do not fret", and fired a shot'
4. Operators and delimiters are either special characters or reserved words. These latter are written in
capital letters and must not be used as identifiers. Hence is it advantageous to memorize this short list of
words. The operators and delimiters composed of special characters are
+
*
/
:=
&
~
=
#
<
>
<=
>=
()
[]
{}

addition, set union

subtraction, set difference
multiplication, set intersection
division, symmetric set difference
assignment
logical AND
logical NOT
equal
unequal
less than
greater than
less than or equal
greater than or equal
parentheses
index brackets
set braces

9
(* *)
comment brackets
^
dereferencing operator
, . ; : .. | punctuation symbols
The reserved words are enumerated in the following list; their meaning will be explained throughout the
subsequent chapters:
ARRAY
CONST
ELSE
IF
MODULE
PROCEDURE
SET
UNTIL

BEGIN
DEFINITION
ELSIF
IMPORT
OF
RECORD
THEN
VAR

BY
DIV
END
IN
OR
REPEAT
TO
WHILE

CASE
DO
FOR
MOD
POINTER
RETURN
TYPE

It is customary to separate consecutive symbols by a space , i.e. one or several blanks. However, this is
mandatory only in those cases where the lack of such a space would merge the two symbols into one. For
example, in "IF x = y THEN" spaces are necessary in front of x and after y, but could be omitted around
the equal sign.
5. Comments may be inserted between any two symbols. They are arbitrary sequences of characters
enclosed in the comment brackets (* and *). Comments are skipped by compilers and serve as additional
information to the human reader. They may also serve to signal instructions (options) to the compiler.

5. Statements and Expressions

The specification of an action is called a statement. Statements can be interpreted (executed), and that
interpretation (execution) has an effect. The effect is a transformation of the state of the computation, the
state being represented by the collective values of the program's variables. The most elementary action is
the assignment of a value to a variable. The assignment statement has the form
assignment = designator ":=" expression.
and its corresponding action consists of three parts in this sequence:
1. Evaluate the designator designating a variable.
2. Evaluate the expression yielding a value.
3. Replace the value of the variable identified in 1. by the value obtained in 2.
Simple examples of assignments are
i := 1

x := y+z

Here i obtains the value 1, and x the sum of y and z, and the previous values are lost. Evidently, every
variable in an expression must previously have been assigned a value. Observe that the following pairs of
statements, executed in sequence, do not have the same effect:
i := i+1; j := 2*i
j := 2*i; i := i+1
Assuming the initial value i = 0, the first pair yields i = 1, j = 2, whereas the second pair yields j = 0. If
we wish to exchange the values of variables i and j, the statement sequence
i := j; j := i
will not have the desired effect. We must introduce a temporary value holder, say k, and specify the three
consecutive assignments
k := i; i := j; j := k
An expression is in general composed of several operands and operators. Its evaluation consists of
applying the operators to the operands in a prescribed sequence, in general taking the operators from left
to right. The operands may be constants, or variables, or functions. (The latter will be explained in a later
chapter.) The identification of a variable will, in general, again require the evaluation of a designator.
Here we shall confine our presentation to simple variables designated by an identifier. Arithmetic

10
expressions (there exist other expressions too) involve numbers, numeric variables, and arithmetic
operators. These include the basic operations of addition (+), subtraction (-), multiplication (*), and
division. They will be discussed in detail in the chapter on basic data types. Here it may suffice to
mention that the slash (/) is reserved for dividing real numbers, and that for integers we use the operator
DIV which truncates the quotient.
An expression consists of consecutive terms. The expression
T0 + T1 + ... + Tn
is equivalent to
((T0 + T1) + ...) + Tn
and is syntactically defined by the rules
SimpleExpression = ["+"|"-"] term {AddOperator term}.
AddOperator =
"+" | "-" | "OR".
Note: for the time being, the reader may consider the syntactic entities expression and SimpleExpression
as equivalent. Their difference and the operators OR, AND, and NOT will be explained in the chapter on
the data type BOOLEAN.
Each term similarly consists of factors. The term
F0 * F1 * ... * Fn
is equivalent to
((F0 * F1) * ...) * Fn
and is syntactically defined by the rules
term = factor {MulOperator factor}.
MulOperator = "*" | "/" | "DIV" | "MOD" | "&".
Each factor is either a constant, a variable, a function, or an expression itself enclosed by parentheses.
Examples of (arithmetic) expressions are
2*3+4*5
15 DIV 4 * 4
15 DIV (4*4)
2+3*4-5
6.25 / 1.25 + 1.5

= (2*3)+(4*5)
= (15 DIV 4)*4
= 15 DIV 16
= 2+(3*4)-5
= 5.0 + 1.5

= 26
= 12
=0
= 9
= 6.5

The syntax of factors, implying that a factor may itself be an expression, is evidently recursive. The
general form of designators will be explained later; here it suffices to know that an identifier denoting a
variable or a constant is a designator.
factor = number | string | set | designator [ActualParameters] | "(" expression ")" | "~" factor.
The rules governing expressions are actually quite simple, and complicated expressions are rarely used.
Nevertheless, we must point out a few basic rules that are well worth remembering.
1. Every variable in an expression must previously have been assigned a value.
2. Two operators must never be written side by side. For instance a * -b is illegal and must be written as
a*(-b).
3. The multiplication sign must never be omitted when a multiplication is required. For example, 2n is
illegal and must be written as 2*n.
4. MulOperators are binding more strongly than AddOperators.
5. When in doubt about evaluation rules (i.e. precedence of operators), use additional parentheses to
clarify. For example, a + b * c may just as well be written as a+(b*c).
The assignment is but one of the possible forms of statements. Other forms will be introduced in the
following chapters. We enumerate these forms by the following syntactic definition

11
statement = [ assignment | ProcedureCall |
WhileStatement | RepeatStatement | ForStatement |
IfStatement | ReturnStatement ].
Several of these forms are structured statements, i.e. some of their components may be statements again.
Hence, the definition of statement is, like that of expressions, recursive.
The most basic structure is the sequence. A computation is a sequence of actions, where each action is
specified by a statement, and is executed after the preceding action is completed. This strict sequentiality
in time is an essential assumption of sequential programming. If a statement S1 follows S0, then we
indicate this sequentiality by a semicolon
S0; S1
This statement separator (not terminator) indicates that the action specified by S0 is to be followed
immediately by the action corresponding to S1. A sequence of statements is syntactically defined as
StatementSequence = statement {";" statement}.
The syntax of statements implies that a statement may consist of no symbols at all. In this case, the
statement is said to be empty and evidently denotes the null action. This curiosity among statements has a
definite reason: it allows semicolons to be inserted at places where they are actually superfluous, such as
at the end of a statement sequence.

6. Control Structures
It is a prime characteristic of computers that individual actions can be selected, repeated, or performed
conditionally depending on some previously computed results. Hence the sequence of actions performed
is not always identical with the sequence of their corresponding statements. The sequence of actions is
determined by control structures indicating repetition, selection, or conditional execution of given
statements.
6.1 Repetitive Statements
The most common situation is the repetition of a statement or statement sequence under control of a
condition: the repetition continues as long as the condition is met. This is expressed by the while
statement. Its syntax is
WhileStatement = "WHILE" expression "DO" StatementSequence "END".
and its corresponding action is
1.
2.

Evaluate the condition which takes the form of an expression yielding the value TRUE or FALSE,
If the value is TRUE, execute the statement sequence and then repeat with step 1; if the value is
FALSE, terminate.

The expression is of type BOOLEAN. This will be further discussed in the chapter on data types. Here it
suffices to know that a simple comparison is a Boolean expression. An example was given in the
introductory example, where repetition terminates when the two comparands have become equal. Further
examples involving while statements are:
1. Initially, let q = 0 and r = x; then count the number of times y can be subtracted from x, i.e. compute
the quotient q = x DIV y, and remainder r = x MOD y, if x and y are natural numbers.
WHILE r >= y DO r := r-y; q := q+1 END
2. Initially, let z = 1 and i = k; then multiply z k times by x, i.e. compute z = x^k, if z and k are natural
numbers:
WHILE i > 0 DO z := z*x; i := i-1 END
When dealing with repetitions, it is important to remember the following points:
1. During each repetition, progress must be made towards meeting the goal, namely "getting closer" to
satisfying the termination condition. An obvious corollary is that the condition must be somehow
affected from within the repeated computation. The following statements are either incorrect or
dependent on some critical precondition as stated.

12
WHILE i > 0 DO
k := 2*k
(*i is not changed*)
END
WHILE i # 0 DO
i := i-2
(*i must be even and positive*)
END
WHILE n # i DO
n := n*i; i := i+1
END
2. If the condition is not satisfied initially, the statement is vacuous, i.e. no action is performed.
3. In order to obtain a grasp of the effect of the repetition, we need to establish a relationship that is
stable, called an invariant. In the division example above, this is the equation q*y + r = x holding each
time the repetition is started. In the exponentiation example it is z * xi = xk which, together with the
termination condition i = 0 yields the desired result z = xk.
4. The repetition of identical computations should be avoided (although a computer has infinite patience
and will not complain). A simple rule is to avoid expressions within repetitive statements, in which no
variable changes its value. For example, the statement
WHILE i < 3*N DO tab[i] := x + y*z + z*i; i := i+1 END
should be formulated more effectively as
n := 3*N; u := x + y*z;
WHILE i < n DO tab[i] := u + z*i; i := i+1 END
In addition to the while statement, there is the repeat statement, syntactically defined as
RepeatStatement = "REPEAT" StatementSequence "UNTIL" expression.
The essential difference to the while statement is that the termination condition is checked each time after
(instead of before) execution of the statement sequence. As a result, the sequence is executed at least
once. An advantage is that the condition may involve variables that are undefined when the repetition is
started.
REPEAT i := i+5; j := j+7; k := i DIV j UNTIL k > 23
REPEAT r := r-y; q := q+1 UNTIL r < y
6.2 Conditional Statements
The conditional statement, also called if statement, has the form
IfStatement = "IF" expression "THEN" StatementSequence
{"ELSIF" expression "THEN" StatementSequence}
["ELSE" StatementSequence]
"END".
The following example illustrates its general form.
IF R1 THEN S1
ELSIF R2 THEN S2
ELSIF R3 THEN S3
ELSE S4
END
The meaning is obvious from the wording. However, it must be remembered that the expressions R1 ...
R3 are evaluated one after the other, and that as soon as one yields the value TRUE, its corresponding
statement sequence is executed, whereafter the IF statement is considered as completed. No further
conditions are tested. Examples are:
IF x = 0 THEN s := 0
ELSIF x < 0 THEN s := -1

13
ELSE s := 1
END
IF ODD(k) THEN z := z*x END
IF k > 10 THEN k := k-10; d := 1 ELSE d := 0 END
The constructs discussed so far enable us to develop a few simple, complete programs as subsequently
described. The first example is an extension of our introductory example computing the greatest common
divisor gcd of two natural numbers x and y. The extension consists in the addition of two variables u and
v, and statements which lead to the computation of the least common multiple (lcm) of x and y. The lcm
and the gcd are related by the equation
lcm(x,y) * gcd(x,y) = x*y
As explained earlier, this and following examples will be formulated as command procedures and are
assumed to be embedded in module Pattern defined in Chap. 2 which imports modules Texts and
Oberon.
PROCEDURE gcdlcm*;
VAR x, y, u, v: INTEGER; S: Texts.Scanner;
BEGIN Texts.OpenScanner(S, Oberon.Par.text, Oberon.Par.pos);
Texts.Scan(S); x := S.i; Texts.WriteString(W, " x ="); Texts.WriteInt(W, x, 6);
Texts.Scan(S); y := S.i; Texts.WriteString(W, " y ="); Texts.WriteInt(W, y, 6);
u := x; v := y;
WHILE x # y DO
(*gcd(x,y) = gcd(x0,y0), x*v + y*u = 2*x0*y0*)
IF x > y THEN
x := x - y; u := u + v
ELSE
y := y - x; v := v + u
END
END ;
Texts.WriteInt(W, x, 6); Texts.WriteInt(W, (u+v) DIV 2, 6); Texts.WriteLn(W);
Texts.Append(Oberon.Log, W.buf)
END gcdlcm.
This example again shows the nesting of control structures. The repetition expressed by a while
statement includes a conditional structure expressed by an if statement, which in turn includes two
statement sequences, each consisting of two assignments. This hierarchical structure is made transparent
by appropriate indentation of the "inner" parts.
Another example demonstrating a hierarchical structure computes the real number x raised to the power
i, where i is a non-negative integer.
PROCEDURE Power*;
VAR i: INTEGER; x, z: REAL; S: Texts.Scanner;
BEGIN Texts.OpenScanner(S, Oberon.Par.text, Oberon.Par.pos); Texts.Scan(S);
WHILE S.class = Texts.Real DO
x := S.x; Texts.WriteString(W, " x ="); Texts.WriteReal(W, x, 16);
Texts.Scan(S); i := S.i; Texts.WriteString(W, " i ="); Texts.WriteInt(W, i, 4);
z := 1.0;
WHILE i > 0 DO
(* z * x^i = x0^i0 *)
z := z*x; i := i-1
END ;
Texts.WriteReal(W, z, 16); Texts.WriteLn(W); Texts.Scan(S)
END ;
Texts.Append(Oberon.Log, W.buf)
END Power.
Here we subject the computation to yet another repetition: Each time a result has been computed, another
value pair x, i is requested. This outermost repetition is controlled by the relation S.class = Texts.Real,

14
which indicates whether a real number x had actually been read. As an example, activation of the
command
M.Power 5.0 2

2.0 5

3.0 3 ~

will generate the results 25.0, 32.0, 27.0.

The straight-forward computation of a power by repeated multiplication is, although obviously correct,
not the most economical. We now present a more sophisticated and more efficient solution. It is based on
the following consideration: The goal of the repetition is to reach the value i = 0. This is done by
successively reducing i, while maintaining the invariant z * xi = x0i0, where x0 and i0 denote the initial
values of x and i. A faster algorithm therefore must rely on decreasing i in larger steps. The solution
given here halves i. But this is only possible, if i is even. Hence, if i is odd, it is first decremented by 1.
Of course, each change of i must be accompanied by a corrective action on z in order to maintain the
invariant. A detail: the subtraction of 1 from i is not expressed by an explicit statement, because it is
performed implicitly by the subsequent division by 2. Two further details are noteworthy: The function
ODD(i) is TRUE, if i is an odd number, FALSE otherwise. x and z denote real values, as opposed to
integer values. Hence, they can represent fractions, too.
PROCEDURE Power*;
VAR i: INTEGER; x, z: REAL; S: Texts.Scanner;
BEGIN Texts.OpenScanner(S, Oberon.Par.text, Oberon.Par.pos); Texts.Scan(S);
WHILE S.class = Texts.Real DO
x := S.x; Texts.WriteString(W, " x ="); Texts.WriteReal(W, x, 16);
Texts.Scan(S); i := S.i; Texts.WriteString(W, " i ="); Texts.WriteInt(W, i, 4);
z := 1.0;
WHILE i > 0 DO
(* z * x^i = x0^i0 *)
IF ODD(i) THEN z := z*x END ;
x := x*x; i := i DIV 2
END ;
Texts.WriteReal(W, z, 16); Texts.WriteLn(W); Texts.Scan(S)
END ;
Texts.Append(Oberon.Log, W.buf)
END Power.
The next sample command has a structure that is almost identical to the preceding program. It computes
the logarithm of a real number x whose value lies between 1 and 2. The invariant in conjunction with the
termination condition (b = 0) implies the desired result sum = log2(x).
PROCEDURE Log2*;
VAR x, a, b, sum: REAL; S: Texts.Scanner;
BEGIN Texts.OpenScanner(S, Oberon.Par.text, Oberon.Par.pos); Texts.Scan(S);
WHILE S.class = Texts.Real DO
x := S.x; Texts.WriteReal(W, x, 15); (*1.0 <= x < 2.0*)
a := x; b := 1.0; sum := 0.0;
REPEAT
(*log2(x) = sum + b*log2(a)*)
a := a*a; b := 0.5*b;
IF a >= 2.0 THEN
sum := sum + b; a := 0.5*a
END
UNTIL b < 1.0E-7;
Texts.WriteReal(W, sum, 16); Texts.WriteLn(W); Texts.Scan(S)
END ;
Texts.Append(Oberon.Log, W.buf)
END Log2.
Normally, routines for computing standard mathematical functions need not be programmed in detail,
because they are available from a collection of modules similar to those for input and output. Such a
collection is, somewhat inappropriately, called a library. In the following example, again exhibiting the
use of a repetitive statement, we use routines for computing the cosine and the exponential function from

15
a library called Math and generate a table of values for a damped oscillation. Typically, the available
standard routines include the sin, cos, exp, ln (logarithm), sqrt (square root), and the arctan functions.
PROCEDURE Oscillation*;
CONST dx = 0.19634953; (*pi/16*)
VAR i, n: INTEGER;
x, y, r: REAL; S: Texts.Scanner;
BEGIN Texts.OpenScanner(S, Oberon.Par.text, Oberon.Par.pos);
Texts.Scan(S); n := S.i; Texts.WriteInt(W, n, 6);
Texts.Scan(S); r := S.x; Texts.WriteReal(W, r, 15); Texts.WriteLn(W);
i := 0; x := 0.0;
REPEAT x := x + dx; i := i+1;
y := Math.exp(-r*x) * Math.cos(x);
Texts.WriteReal(W, x, 15); Texts.WriteReal(W, y, 15); Texts.WriteLn(W)
UNTIL i >= n;
Texts.Append(Oberon.Log, W.buf)
END Oscillation.

7. Elementary Data Types

We have previously stated that all variables need be declared. This means that their names are introduced
in the heading of the program. In addition to introducing the name (and thereby enabling a compiler to
detect and indicate misspelled identifiers), declarations have the purpose of associating a data type with
each variable. This data type represents information about the variable which is permanent in contrast,
for example, to its value. This information may again aid in detecting inconsistencies in an erroneous
program, inconsistencies that are detectable by mere inspection of the program text without requiring its
interpretation.
The type of a variable determines its set of possible values and the operations which may be applied to it.
Each variable has a single type that may be deduced from its declaration. Each operator requires
operands of a specific type and produces a result of a specific type. It is therefore visible from the
program text, whether or not a given operator is applicable to a given variable.
Data types may be declared in the program. Such constructed types are usually based on composition of
basic types. There exist a number of most frequently used, elementary types that are basic to the language
and need not be declared. They are called standard types and will be introduced in this chapter, although
some have already appeared in previous examples.
In fact, not only variables have a type, but so do constants, functions, operands and (results of) operators.
In the case of a constant, the type is usually deducible from the constant's notation, otherwise from its
explicit declaration.
First we describe the standard data types of Oberon, and thereafter elaborate on the form of declarations
of variables and constants. Further kinds of data types and declarations are deferred to later chapters.
7.1 The Type INTEGER
This type represents the whole numbers, and any value of type INTEGER is therefore an integer.
Operators applicable to integers include the basic arithmetic operations
+
*
DIV
MOD

addition
subtraction
multiplication
division
modulus (remainder)

Integer division is denoted by DIV. If we define the integer quotient and the modulus of x and y by
q = x DIV y,

r = x MOD y

then q and r are related by the equation x = q*y + r and by the constraint 0 <= r < y. For example
15 DIV 4 = 3
-15 DIV 4 = -4

15 MOD 4 = 3
-15 MOD 4 = 1

15 = 3*4 + 3
-15 = (-4)*4 + 1

16
Sign inversion is denoted by the monadic minus sign. Furthermore, there exist the operators ABS(x) and
ODD(x), the former yielding the absolute value of x, the latter the Boolean result "x is odd".
Every computer will restrict the set of values of type INTEGER to a finite set of integers, usually the
interval -2N-1 ... 2N-1-1, where N is a small integer, often 16 or 32, depending on the number of bits a
computer uses to represent an integer. If an arithmetic operation produces a result that lies outside that
interval, then overflow is said to have occured. The computer will give an appropriate indication and,
usually, terminate the computation. The programmer should ensure that overflows will not result during
execution of the program.
7.2 The Type REAL
Values of type REAL are real numbers. The available operators are again the basic arithmetic operations
and ABS. Division is denoted by / (instead of DIV). Constants of type REAL are characterized by having
a decimal point and possibly a decimal scale factor. Examples of real number denotations are
1.5 1.50 1.5E2 2.34E-2 0.0
The scale factor consists of the capital letter E followed by an integer. It means that the preceding real
number is to be multiplied by 10 raised to the scale factor. Hence
1.5E2 = 150.0,

2.34E-2 = 0.0234

The important point to remember is that real values are internally represented as pairs consisting of a
fractional number and a scale factor. This is called floating-point representation. Of course, both parts
consist of a finite number of digits. As a consequence, the representation is inherently inexact, and
computations involving real values are inexact because each operation may be subject to truncation or
rounding.
The following program makes that inherent imprecision of computations involving operands of type
REAL apparent. It computes the harmonic function
H(n) = 1 + 1/2 + 1/3 + ... + 1/n
in two different ways, namely once by summing terms from left to right, once from right to left.
According to the rules of arithmetic, the two sums ought to be equal. However, if values are truncated (or
even rounded), the sums will differ for sufficiently large n. The correct way is evidently to start with the
small terms.
PROCEDURE Harmonic*;
VAR i, n: INTEGER;
x, d, s1, s2: REAL; S: Texts.Scanner;
BEGIN Texts.OpenScanner(S, Oberon.Par.text, Oberon.Par.pos); Texts.Scan(S);
WHILE S.class = Texts.Int DO
n := S.i; Texts.WriteString(W, "n = "); Texts.WriteInt(W, n, 6);
s1 := 0.0; d := 0.0; i := 0;
REPEAT
d := d + 1.0; INC(i);
s1 := s1 + 1.0/d;
UNTIL i >= n;
Texts.WriteReal(W, s1, 16);
s2 := 0.0;
REPEAT
s2 := s2 + 1.0/d;
d := d - 1.0; DEC(i)
UNTIL i = 0;
Texts.WriteReal(W, s2, 16); Texts.WriteLn(W); Texts.Scan(S)
END ;
Texts.Append(Oberon.Log, W.buf)
END Harmonic.
The major reason for strictly distinguishing between real numbers and integers lies in the different
representation used internally. Hence, also the arithmetic operations are implemented by instructions

17
which are distinct for each type. Oberon postulates that expressions with mixed operands always are of
type REAL.
A real number x is transformed into an integer by the standard function ENTIER(x). It yields the largest
integer not greater than x.
ENTIER(3.1416) = 3

ENTIER(-2.7) = -3

ENTIER(5.0) = 5

7.3 The Type BOOLEAN

A BOOLEAN value is one of the two logical truth values denoted by the standard identifiers TRUE and
FALSE. Boolean variables are usually denoted by identifiers which are adjectives, the value TRUE
implying the presence, FALSE the absence of the indicated property. A set of logical operators is
provided which, together with BOOLEAN variables, form BOOLEAN expressions. These operators are
& (and), OR, and ~ (not). Their results are explained as follows
p&q
p OR q
~p

= "both p and q are TRUE"

= "either p or q or both are TRUE"
= "p is FALSE"

The operators' exact definition, however, is slightly different, although the results are identical:
p&q
p OR q

= IF p THEN q ELSE FALSE

= IF p THEN TRUE ELSE q

This definition implies that the second operand need not be evaluated, if the result is already known from
the evaluation of the first operand. The notable property of this definition of Boolean connectives is that
their result may be well-defined even in cases where the second operand is undefined. As a consequence,
the order of the operands may be significant.
Sometimes it is possible to simplify Boolean expressions by application of simple transformation rules. A
particularly useful rule is de Morgan's law stating the equivalences
~p & ~q = ~(p OR q)
~p OR ~q = ~(p & q)
Relations produce a result of type BOOLEAN, i.e. TRUE if the relation is satisfied, FALSE if not. For
example
7 = 12
7 < 12
1.5 >= 1.6

FALSE
TRUE
FALSE

Relations are syntactically classified as expressions, and the two comparands are so-called
SimpleExpressions (see also chapter on expressions and statements). The result of a comparison is of
type BOOLEAN and can be used in control structures such as if, while, and repeat statements. The
symbol # stands for unequal.
expression = SimpleExpression [relation SimpleExpression].
relation = "=" | "#" | "<" | "<=" | ">" | ">=" | "IN".
It should be noted that, similar to arithmetic operators, there exists a precedence hierarchy among the
Boolean operators. ~ has the highest precedence, then follows &, then OR, and last are the relations. As
with arithmetic expressions, parentheses may be used freely to make the association of operators explicit.
Examples of Boolean expressions are
x = y
(x <= y) & (y < z)
(x > y) OR (y >= z)
~p OR q
Note that a construct such as x < y & z < w is illegal.
The example above draws attention to the rule that the operands of an operator (including relational
operators) must be of the same type. The following relations are therefore illegal:

18
1 = TRUE
5 = 5.0
i+j = p OR q
Also incorrect is e.g. x <= y < z, which must be expanded into (x <= y) & (y < z). However, the
following are correct Boolean expressions:
i+j < k-m
p OR q = (i<j)
A final hint: although "p = TRUE" is legal, it is considered poor style and better expressed as p.
Similarly, replace "p = FALSE" by "~p".
7.4 The Type CHAR
Every computer system communicates with its environment via some input and output devices. They
read, write, or print elements taken from a fixed set of characters. This set constitutes the value range of
the type CHAR. Unfortunately, different brands of computers may use different character sets, which
makes communication between them (i.e. the exchange of programs and data) difficult and often tedious.
However, there exists an internationally standardised set, the ISO set.
The ISO standard defines a set of 128 characters, 33 of which are so-called (non-printing) control
characters. The remaining 95 elements are visible printing characters shown in the following table. The
set is ordered, and each character has a fixed position or ordinal number. For example, A is the 66th
character and has the ordinal number 65. The ISO standard, however, leaves a few places open, and they
can be filled with different characters according to national desires to establish national standards. Most
widely used is the American standard, also called ASCII (American Standard Code for Information
Interchange). Here we tabulate the ASCII set. The ordinal number of a character is obtained by taking its
row number and adding its column number. These numbers are customarily given in hexadecimal form,
and we follow this habit here too. The first two columns contain the control characters; they are
customarily denoted by abbreviations hinting at their intended meaning. However, this meaning is not
inherent in the character code, but only defined by its interpretation. At this point it suffices to remember
that these characters are (usually) not printable.
Table of ASCII characters

0
1
2
3
4
5
6
7
8
9
A
B
C
D
E
F

10

20

30

40

50

60

70

nul
soh
stx
etx
eot
enq
ack
bel
bs
ht
lf
vt
ff
cr
so
si

dle
dc1
dc2
dc3
dc4
nak
syn
etb
can
em
sub
esc
fs
gs
rs
us

!
"
#
$
%
&
'
(
)
*
+
,
.
/

0
1
2
3
4
5
6
7
8
9
:
;
<
=
>
?

@
A
B
C
D
E
F
G
H
I
J
K
L
M
N
O

P
Q
R
S
T
U
V
W
X
Y
Z
[
\
]
^
_

`
a
b
c
d
e
f
g
h
i
j
k
l
m
n
o

p
q
r
s
t
u
v
w
x
y
z
{
|
}
~
del

Constants of type CHAR are denoted by the character enclosed in quote marks or in apostrophes.
Character values can be assigned to variables of type CHAR, but these values cannot be used in
arithmetic operations. Arithmetic operators can be applied, however, to their ordinal numbers obtained
from the transfer function ORD(ch). Inversely the character with the ordinal number n is obtained with
the transfer function CHR(n). These two complementary functions are related by the equations
CHR(ORD(ch)) = ch and

ORD(CHR(n)) = n

for 0 <= n < 128. They permit to compute the numeric value represented by the digit ch as

19
ORD(ch) - ORD("0")
and to compute the digit representing the numeric value n as
CHR(n + ORD("0"))
These two formulas depend on the adjacency of the 10 digits in the ISO character set, where ORD("0") =
48. They are typically used in routines converting sequences of digits into numbers and vice-versa. The
following program piece reads digits and assigns the decimally represented number to a variable x.
x := 0; Texts.Read(R, ch);
WHILE ("0" <= ch) & (ch <= "9") DO
x := 10*x + (ORD(ch) - ORD("0")); Texts.Read(R, ch)
END
Control characters are used for various purposes, mainly to control the functions of devices, but also to
delineate and structure text. An important role of control characters is to specify the end of a line or of a
page of text. There is no universally accepted standard for this purpose. We shall denote the control
character signifying a line end by the identifier EOL (end of line); its actual value depends on the system
used.
In order to be able to denote non-printable characters, Oberon uses their hexadecimal ordinal number
followed by the capital letter X. For example, 0DX is the value of type CHAR denoting the control
character cr (carriage return, line end) with ordinal number 13.
7.5 The Type SET
The values which belong to the type SET are sets of integers between 0 and N-1, where N is a constant
defined by the computer system used. It is usually the computer's wordlength or a small multiple of it,
typically 32. Constants of this type are denoted as sets. Examples are
{2, 3, 5, 7, 11} {0} {8 .. 15} {}
The notation m .. n is a shorthand for m, m+1, ... , n-1, n.
set = [qualident] "{" [element {"," element}] "}".
element = expression [".." expression].
Operations on sets are:
+
*
/

set union
set difference
set intersection
symmetric set difference

Assuming that i denotes an element and u, v denote sets, these operations are defined in terms of set
membership as
1.
2.
3.
4.

i IN (u+v) = (i IN u) OR (i IN v), i.e. the integer i is in the union u+v, if either in u or in v (or in
both).
i IN (u-v) = (i IN u) AND NOT (i IN v), i.e. i is in the difference u-v, if it is in u, but not in v.
i IN (u*v) = (i In u) AND (i IN v), i.e. i is in the intersection u*v, if it is in both u and v.
i IN (u/v) = ((i IN u) # (i IN v)), i.e. i is in the symmetric difference u/v, if it is in either u or in v, but
not in both.

The membership operator IN is regarded as a relational operator. The expression i IN u is of type

BOOLEAN. It is TRUE, if i is a member of the set u. Sets are represented in computer systems as sets of
bits, i.e. by the characteristic function of the set. The i'th bit of u, for example, is 1, if i is a member of u,
0 otherwise. Hence, the set operators are implemented as logical operations applied to the N members of
the set variable. They are therefore very efficient and their execution time is usually even less than that of
an addition of integers.

8. Constant and Variable Declarations

20
It has been mentioned previously that all identifiers used in a program must be declared in the program's
heading, unless they are standard identifiers known in every program (or are imported from some other
module).
If an identifier is to denote a constant value, it must be introduced by a constant declaration which
indicates the value for which the constant identifier stands. A constant declaration has the form
ConstantDeclaration = identifier "=" ConstExpression.
ConstExpression = expression
A ConstExpression is an expression containing constants only. More precisely, its evaluation must be
possible by a mere textual scan without execution of the program A sequence of constant declarations is
preceded by the symbol CONST. Example:
CONST N = 16;
EOL = 0DX;
empty = {};
M = N-1;
Constants with explicit names aid in making a program readable, provided that the constants are given
suggestive names. If, e.g., the identifier N is used instead of its value throughout a program, a change of
that constant can be achieved by changing the program in a single place only, namely in the declaration
of N. This avoids the common mistake that some instances of the constant, spread over the entire
program text, remain undetected and therefore are not updated, leading to inconsistencies.
A variable declaration looks similar to a constant declaration. The place of the constant's value is taken
by the variable's type which, in a sense, can be regarded as the variable's constant property. Instead of an
equal sign, a colon is used.
VariableDeclaration = IdentList ":" type.
IdentList = identifier {"," identifier}.
Variables of the same type may be listed in the same declaration, and a sequence of declarations is
preceded by the symbol VAR. Example:
VAR i, j, k: INTEGER;
x, y, z: REAL;
ch: CHAR

9. The Data Structure Array

So far, we have given each variable an individual name. This is impractical, if many variables are
necessary that are treated in the same way and are of the same type, such as, for example, if a table of
data is to be constructed. In this case, we wish to give the entire set of data a name and to denote
individual elements with an identifying number, a so-called index. The data type is then said to be
structured - more precisely: array structured. In the following example, the variable a consists of N
elements, each being of type INTEGER, and the indices range from 0 to N-1.
VAR a: ARRAY N OF INTEGER
An element is then designated by the array's identifier followed by its selecting index, e.g. a[i], where i is
an expression whose value must lie within the index range specified in the array's declaration.
Syntactically, a[i] is a designator, and the expression i is the selector. If, for example, all N elements of a
are to be given the value 0, this can be expressed conveniently by a repetitive statement, where the index
is given a new value each time.
i := 0;
REPEAT a[i] := 0; INC(i) UNTIL i = N
INC(i) is synonymous with i := i+1. This example illustrates a situation that occurs so frequently that
Oberon provides a special control structure which expresses it more concisely. It is called the for
statement:
FOR i := 0 TO N-1 DO a[i] := 0 END
Its general form is

21
ForStatement =
"FOR" identifier ":=" expression "TO" expression ["BY" ConstExpression] "DO"
StatementSequence
"END".
The expressions before and after the symbol TO define the range through which the so-called control
variable (i) progresses. An optional parameter determines the incrementing (decrementing) value. If it is
omitted, 1 is assumed as default value.
It is recommended that the for statement be used in simple cases only; in particular, no components of the
expressions determining the range must be affected by the repeated statements, and, above all, the control
variable itself must not be changed by the repeated statements. The value of the control variable must be
considered as undefined after the for statement is terminated.
Some additional examples should demonstrate the use of array structures and the for statement. In the
first one, the sum of the N elements of array a is to be computed:
sum := 0;
FOR i := 0 TO N-1 DO sum := a[i] + sum END
In the second example, the minimum value is to be found, and also its index. The repetition's invariant is
min = minimum(a[0], ... , a[i-1]).
min := a[0]; k := 0;
FOR i := 1 TO N-1 DO
IF a[i] < min THEN k := i; min := a[k] END
END
In the third example, we make use of the second to sort the array in ascending order:
FOR i := 0 TO N-2 DO
min := a[i]; k := i;
FOR j := i TO N-1 DO
IF a[j] < min THEN k := j; min := a[k] END
END;
a[k] := a[i]; a[i] := min
END
The for statement is evidently appropriate, if all elements within a given index range are to be processed.
It is inappropriate in most other cases. If, for example, we wish to find the index of an element equal to a
given value x, we have no advance knowledge how many elements have to be inspected. Hence, the use
of a while (or repeat) statement is recommended. This algorithm is known as linear search.
i := 0;
WHILE (i < N) & (a[i] # x) DO INC(i) END
From the negation of the continuation condition, applying de Morgan's law, we infer that upon
termination of the while statement the condition (i = N) OR (a[i] = x) holds. If the latter term is TRUE,
the desired element is found and i is its index; if i = N, no a[i] equals x.
We draw attention to the fact that the termination condition is a composite one, and that it is possible to
simplify this condition by a common technique. Remember, that repetition must terminate, either if the
desired element is found, or if the array's end is reached. The "trick" now consists in marking the end by
a sentinel, namely the value x itself, which will automatically stop the search. All that is required is the
addition of a dummy element a[N] at the array's end, serving as sentinel:
a: ARRAY N+1 OF INTEGER;
a[N] := x; i := 0;
WHILE a[i] # x DO INC(i) END
If upon termination i = N, no original element has the value x, otherwise i is the desired index.
A more challenging problem is the search for a desired element with value x in an array that is ordered,
i.e. a[i-1] <= a[i] for all i = 1 ... N-1. The best technique is the so-called binary search: inspect the
middle element, then apply the same method to either the left or right half. This is expressed by the
following piece of program, assuming N > 0. The repetition's invariant is

22
a[k] < x for all k = 0 ... i-1

and

a[k] > x for all k = j+1 ... N-1

i := 0; j := N-1; found := FALSE;

REPEAT mid := (i+j) DIV 2;
IF x < a[mid] THEN j := mid -1
ELSIF x > a[mid] THEN i := mid+1
ELSE found := TRUE
UNTIL (i > j) OR found
Because each step halves the interval in which x is searched, the number of needed comparisons is only
log2N. A somewhat more efficient version which avoids the composite termination condition is
i := 0; j := N-1;
REPEAT mid := (i+j) DIV 2;
IF x <= a[mid] THEN j := mid-1 END;
IF x >= a[mid] THEN i := mid+1 END
UNTIL i > j;
IF i > j+1 THEN found ELSE not found END
An even more sophisticated version is given below. Its ingenious idea is not to terminate immediately
when the element is found, which is rare compared to the number of unsuccessful comparisons.
i := 0; j := N;
REPEAT mid := (i+j) DIV 2;
IF x < a[mid] THEN j := mid ELSE i := mid+1 END
UNTIL i >= j;
IF j < N) & (a[j] = x) THEN (*found*) END
This ends our list of examples of typical uses of simple arrays.
The elements of an array are all of the same type, but this type may again be an array (in fact, it may be
any structure, as will be seen later). An array of arrays is called a multidimensional array or matrix,
because each index may be considered as spanning a dimension in a Cartesian space. Examples of twodimensional arrays are
a: ARRAY N, N OF REAL
T: ARRAY M, N OF CHAR
These are actually abbreviations of the full forms
a: ARRAY N OF ARRAY N OF REAL
T: ARRAY M OF ARRAY N OF CHAR
The general syntax of an array type is
ArrayType = "ARRAY" ConstExpression {"," ConstExpression } "OF" type.
where the expression denotes the length of the array.
The syntax of designators allows for a similar abbreviation as used in declarations, namely a[i, j] in
place of a[i][j]. The latter form expresses more clearly that j is the selector on the array a[i]. The syntax
for the array element designator is
designator = qualident {"[" ExpList "]"}.
ExpList = expression {"," expression}.
In uses of matrices, the for statement comes to its full bloom, particularly in numeric applications. The
canonical example is the multiplication of two matrices, where each element of the product c = a*b is
defined as
c[i, j] = a[i, 0] * b[0, j] + a[i, 1] * b[1, j] + ... + a[i, K-1] * b[K-1, j]
Given the declarations
a: ARRAY M, K OF REAL;
b: ARRAY K, N OF REAL;
c: ARRAY M, N OF REAL
the multiplication algorithm consists of three nested repetitions as follows:

23
FOR i := 0 TO M-1 DO
FOR j := 01 TO N-1 DO
sum := 0.0;
FOR k := 0 TO K-1 DO
sum := a[i, k]*b[k, j] + sum
END;
c[i,j] := sum
END
END
In a second example we demonstrate the search of a word in a table, so-called table search, where each
word is an array of characters. We assume the table is declared by the T given in the example above, and
that x is given as
x: ARRAY N OF CHAR
Our solution employs the typical linear search
i := 0; found := FALSE;
WHILE ~found & (i < M) DO
found := T[i] = x;
INC(i)
END
If we define equality between two words x and y as x[j] = y[j] for all j = 0 ... N-1, then the "inner"
search can be expressed as
j := 0; equal := TRUE;
WHILE equal & (j < N) DO
equal := T[i, j] = x[j]; INC(j)
END;
found := equal
This solution appears to be somewhat cumbersome. It is transformable into a simpler form, if M > 0 and
N > 0. The complete table search is then expressible as
i := 0;
REPEAT j := 0;
REPEAT b := T[i, j] # x[j]; INC(j)
UNTIL b OR (j = N);
INC(i)
UNTIL ~b OR (i = M)
The result b means "the word x has not been found".
We have now laid enough ground work to develop meaningful, entire programs and shall present three
examples, all of them involving arrays.
In the first example, the goal is to generate a list of powers of 2, each line showing the values 2i, i, and 2i
. This task is quite simple, if the type REAL is used. The program then contains the core
d := 1; f := 1.0;
FOR i := 1 TO N DO
d := 2*d; write(d); (* d = 2^i *)
write(i);
f := f/2.0; write(f) (* f = 2^(-i) *)
END
However, our task shall be to generate exact results with as many decimal digits as needed. For this
reason, we present both the whole number d = 2i and the fraction f = 2-i by arrays of digits, each in the
range 0 ... 9. For f we require N , for d only log10N digits. Note that the doubling of d proceeds from right
to left, the halving of f from left to right. The table of results is shown below.
PROCEDURE PowersOf2*;

24
CONST M = 11; N = 32; (*M ~ N*log(2) *)
VAR i, j, k, exp, c, r, t: INTEGER;
d: ARRAY M OF INTEGER;
f: ARRAY N OF INTEGER;
BEGIN d[0] := 1; k := 1;
FOR exp := 1 TO N-1 DO
(* compute d = 2^exp *) c := 0; (*carry*)
FOR i := 0 TO k-1 DO
t := 2*d[i] + c;
IF t >= 10 THEN d[i] := t - 10; c := 1 ELSE d[i] := t; c := 0 END
END ;
IF c > 0 THEN d[k] := 1; INC(k) END ;
(*output d[k-1] ... d[0]*) i := M;
REPEAT DEC(i); Texts.Write(W, " ") UNTIL i = k;
REPEAT DEC(i); Texts.Write(W, CHR(d[i]+ORD("0"))) UNTIL i = 0;
Texts.WriteInt(W, exp, 4);
(*compute and output f = 2^(-exp) *)
Texts.WriteString(W, " 0."); r := 0; (*remainder*)
FOR j := 1 TO exp-1 DO
r := 10*r + f[j]; f[j] := r DIV 2;
r := r MOD 2; Texts.Write(W, CHR(f[j]+ORD("0")))
END ;
f[exp] := 5; Texts.Write(W, "5"); Texts.WriteLn(W)
END ;
Texts.Append(Oberon.Log, W.buf)
END PowersOf2.
Output of program PowersOf2:
2
4
8
16
32
64
128
256
512
1024
2048
4096
8192
16384
32768
65536
131072
262144
524288
1048576
2097152
4194304
8388608
16777216
33554432
67108864
134217728
268435456
536870912
1073741824
2147483648
4294967296

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32

0.5
0.25
0.125
0.0625
0.03125
0.015625
0.0078125
0.00390625
0.001953125
0.0009765625
0.00048828125
0.000244140625
0.0001220703125
0.00006103515625
0.000030517578125
0.0000152587890625
0.00000762939453125
0.000003814697265625
0.0000019073486328125
0.00000095367431640625
0.000000476837158203125
0.0000002384185791015625
0.00000011920928955078125
0.000000059604644775390625
0.0000000298023223876953125
0.00000001490116119384765625
0.000000007450580596923828125
0.0000000037252902984619140625
0.00000000186264514923095703125
0.000000000931322574615478515625
0.0000000004656612873077392578125
0.00000000023283064365386962890625

Our second example is similar in nature. Its task is to compute the fractions d = 1/i exactly. The difficulty
lies, of course, in the representation of those fractions that are infinite sequences of digits, e.g. 1/3 =
0.333.... Fortunately, all fractions have a repeating period, and a reasonable and useful solution is to mark
the beginning of the period and to terminate at its end. How do we find the beginning and the end of the
period? Let us first consider the algorithm for computing the digits of the fraction.

25
Starting out with rem = 1, we repeat multiplying by 10 and dividing the product by i. The integer quotient
is the next digit and the remainder is the new value of rem. This is precisely the conventional method of
division, as illustrated by the following piece of program and the example with i = 7:
1.000000 / 7 = 0.142857
1 0
30
20
60
40
50
1

rem := 1;
REPEAT rem := 10 * rem; nextDigit := rem DIV i; rem := rem MOD i UNTIL ...
We know that the period has ended as soon as a remainder occurs which had been encountered
previously. Hence, our recipe is to remember all remainders and their indices. The latter designate the
place where the period had started. We denote these indices by x and give elements of x initial value 0. In
the above explained division by 7, the values of x are
x[1] = 1, x[2] = 3, x[3] = 2, x[4] = 6, x[5] = 4, x[6] = 5
PROCEDURE Fractions*;
CONST Base = 10; N = 32;
VAR i, j, m, rem: INTEGER;
d: ARRAY N OF INTEGER; (*digits*)
x: ARRAY N+1 OF INTEGER; (*index*)
BEGIN
FOR i := 2 TO N DO
FOR j := 0 TO i-1 DO x[j] := 0 END ;
m := 0; rem := 1;
REPEAT m := m+1; x[rem] := m;
rem := Base * rem; d[m] := rem DIV i; rem := rem MOD i
UNTIL x[rem] # 0;
Texts.WriteInt(W, i, 6); Texts.WriteString(W, " 0.");
FOR j := 1 TO x[rem]-1 DO Texts.Write(W, CHR(d[j]+ORD("0"))) END ;
Texts.Write(W, "'");
FOR j := x[rem] TO m DO Texts.Write(W, CHR(d[j]+ORD("0"))) END ;
Texts.WriteLn(W)
END ;
Texts.Append(Oberon.Log, W.buf)
END Fractions.
Output of program Fractions:
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24

0.5'0
0.'3
0.25'0
0.2'0
0.1'6
0.'142857
0.125'0
0.'1
0.1'0
0.'09
0.08'3
0.'076923
0.0'714285
0.0'6
0.0625'0
0.'0588235294117647
0.0'5
0.'052631578947368421
0.05'0
0.'047619
0.0'45
0.'0434782608695652173913
0.041'6

26
25
26
27
28
29
30
31
32

0.04'0
0.0'384615
0.'037
0.03'571428
0.'0344827586206896551724137931
0.0'3
0.'032258064516129
0.03125'0

Our last example of a program computes a list of prime numbers. It is based on the idea of inspecting the
divisibility of successive integers. The tested integers are obtained by incrementing alternatively by 2 and
4, thereby avoiding multiples of 2 and 3 ab initio. Divisibility needs to be tested for prime divisors only,
which are obtained by storing previously computed results.
PROCEDURE Primes*;
CONST N = 252; M = 16; (*M ~ sqrt(N)*)
LL = 10; (*no. of primes placed on a line*)
VAR i, k, x: INTEGER;
inc, lim, square, L: INTEGER;
prime: BOOLEAN;
P,V: ARRAY M+1 OF INTEGER;
BEGIN L := 0; x := 1; inc := 4; lim := 1; square := 9;
FOR i := 3 TO N DO
(* find next prime number p[i] *)
REPEAT x := x + inc; inc := 6 - inc;
IF square <= x THEN
INC(lim); V[lim] := square; square := P[lim+1] * P[lim+1]
END ;
k := 2; prime := TRUE;
WHILE prime & (k < lim) DO
INC(k);
IF V[k] < x THEN V[k] := V[k] + 2*P[k] END ;
prime := x # V[k]
END
UNTIL prime;
IF i <= M THEN P[i] := x END ;
Texts.WriteInt(W, x, 6); L := L+1;
IF L = LL THEN Texts.WriteLn(W); L := 0 END
END ;
Texts.Append(Oberon.Log, W.buf)
END Primes.
Output of program Primes:
5
41
83
137
191
241
307
367
431
487
563
617
677
751
823
883
967
1031
1093
1171
1237
1303
1409

7
43
89
139
193
251
311
373
433
491
569
619
683
757
827
887
971
1033
1097
1181
1249
1307
1423

11
47
97
149
197
257
313
379
439
499
571
631
691
761
829
907
977
1039
1103
1187
1259
1319
1427

13
53
101
151
199
263
317
383
443
503
577
641
701
769
839
911
983
1049
1109
1193
1277
1321
1429

17
59
103
157
211
269
331
389
449
509
587
643
709
773
853
919
991
1051
1117
1201
1279
1327
1433

19
61
107
163
223
271
337
397
457
521
593
647
719
787
857
929
997
1061
1123
1213
1283
1361
1439

23
67
109
167
227
277
347
401
461
523
599
653
727
797
859
937
1009
1063
1129
1217
1289
1367
1447

29
71
113
173
229
281
349
409
463
541
601
659
733
809
863
941
1013
1069
1151
1223
1291
1373
1451

31
73
127
179
233
283
353
419
467
547
607
661
739
811
877
947
1019
1087
1153
1229
1297
1381
1453

37
79
131
181
239
293
359
421
479
557
613
673
743
821
881
953
1021
1091
1163
1231
1301
1399
1459

27
1471
1543

1481
1549

1483
1553

1487
1559

1489
1567

1493
1571

1499
1579

1511
1583

1523
1597

1531
1601

These examples show that arrays are a fundamental feature used in most programs. There exist hardly
any programs of relevance outside the classroom which do not employ repetitions and arrays (or
analogous data structures).

28

Part 2
10. Procedures
Consider the task of processing a set of data consisting of a header and a sequence of N similar individual
units. It might be generally described as
ReadHeader;
ProcessHeader;
WriteHeader;
FOR i := 1 TO N DO
ReadUnit; ProcessUnit;
Write(i); WriteUnit
END
Clearly, the description of the original task has been made in terms of subtasks, emphasizing the
dominant structure and supressing details. Of course, the subtasks ReadHeader, ProcessHeader, etc.
must now be further described with all the necessary details. Instead of replacing these descriptive
English words with elaborate Oberon programs, we may consider these words as identifiers and define
the details of the subtasks by textually separate pieces of program, called procedures (or subroutines).
These definitions are called procedure declarations, because they define the actions of the procedure and
give it a name. The identifiers in the main program referring to these declarations are said to be
procedure calls, and their action is to invoke the procedure. Syntactically, the procedure call is a
statement.
Procedures play a fundamental role in program design. They aid in displaying the algorithm's structure
and in decomposing a program into logically coherent units. This is particularly important in the case of
complex algorithms, i.e. of long programs. In the above example, it might be considered somewhat
extravagant to declare separate procedures instead of merely substituting the refined program texts for
the identifiers. Nevertheless, the gain in clarity of program structure often recommends the use of explicit
procedures even in such a simple case. But of course procedures become particularly useful, if the same
procedure is to be invoked at several points of the program.
A procedure declaration consists of the symbol PROCEDURE followed by the identifier (together they
form the procedure heading), followed by the symbol BEGIN and the statements for which the procedure
identifier stands and which are therefore called the procedure body. The declaration is terminated by the
symbol END and the repetition of the identifier. The latter enables a compiler to detect mismatched
endings of statements and declarations. The general syntax of procedure declarations will be given later.
A simple example is the following, which computes the sum of a[0] ... a[N-1].
PROCEDURE Add;
BEGIN sum := 0.0;
FOR i := 0 TO N-1 DO sum := a[i] + sum END
END Add
The procedure concept becomes even much more useful due to two additional features that are coupled
with it, namely the concepts of parameters and of locality of names. Parameters make it possible to
invoke the same procedures at different points of the program applying the procedure to different values
and variables as determined at the point of the call. Locality of names and objects is a concept
considerably enhancing the procedure's role in structuring a program and compartmentalizing its parts.
We shall first discuss the concept of locality.
Summarizing, we repeat the following essential points:
1. The procedure aids in exhibiting the inherent structure of a program and in decomposing a
programming task.
2. If a procedure is called from two or more points, it reduces the length of the program and therefore the
programming task and the potential for programming error. A further economic advantage is the
reduction in the size of the compiled code.

11. The Concept of Locality

29
If we inspect the preceding example of procedure Add, we notice that the role of the variable i is strictly
confined to the procedure body. This inherent locality should be expressed explicitly, and can be done by
declaring i inside the procedure declaration. i thereby becomes a local variable.
PROCEDURE Add;
VAR i: INTEGER;
BEGIN sum := 0.0;
FOR i := 0 TO N-1 DO sum := a[i] + sum END
END Add
In some sense, the procedure declaration assumes the form of a separate program. In fact, any declaration
possible in a program, such as constant-, type-, variable-, or procedure declaration, may also occur in a
procedure declaration. This implies that procedure declarations may be nested and are defined
recursively.
It is good programming practice to declare objects local, i.e. to confine the existence of an object to that
procedure in which it has meaning. The procedure, i.e. the section of program text in which a name is
declared, is called its scope. Since declarations can be nested, scopes are also nestable. The possibility of
having objects local to some scope has several consequences. For example, the same name can be used to
denote different objects. In fact, this is a most useful consequence, because a programmer is free to
choose local identifiers without knowledge of those existing in the surrounding scope, as long as they do
not denote objects used in the local scope (in which case he must obviously be aware of them anyhow).
This decoupling of knowledge about different program parts is particularly useful and perhaps even vital
in the case of large programs.
The rules of scope (validity of identifiers) are as follows:
1. The scope of an identifier is the procedure in which its declaration occurs, and all procedures enclosed
by that procedure, subject to rule 2.
2. If an identifier i declared in a procedure P is redeclared in some inner procedure Q enclosed in P, then
procedure Q and all procedures enclosed in Q are excluded from the scope of i declared in P.
3. The standard identifiers of Oberon are considered to be declared in an imaginary scope enclosing the
program.
These rules may also be remembered by the algorithm in which the declaration of a given identifier i is
searched: First, search the declarations of the procedure P in whose body i occurs; if the declaration of i
is not among them, continue the search in the procedure surrounding P; then repeat this same rule until
the declaration is encountered.
VAR a: INTEGER;
PROCEDURE P;
VAR b: INTEGER;
PROCEDURE Q;
VAR b, c: BOOLEAN;
BEGIN
(*here a, b (BOOLEAN), c are visible*)
END Q;
BEGIN
(*here a, b (INTEGER) are visible*)
END P
A consequence of the locality concept and of the rule that a variable does not exist outside its scope is
that its value is lost when its declaring procedure is terminated. This implies that, when the same
procedure is later called again, this value is unknown. The values of local variables are undefined when
the procedure is (re)entered. Hence, if a variable must retain its value between two calls, it must be
declared outside the procedure. The "lifetime" of a variable is the time during which its declaring
procedure is active.
The use of local declarations has three significant advantages:
1. It makes clear that an object is confined to a procedure, usually a small part of the entire program.

30
2. It ensures that inadvertent use of a local object by other parts of the program is detected by the
compiler.
3. It enables the implementation to minimize storage, because a variable's storage is released when the
procedure to which the variable is local is terminated. This storage can then be reused for other
variables.

12. Parameters
Procedures may be given parameters. They are the essential feature that make procedures so useful.
Consider again the previous example of procedure Add. Very likely a program contains several arrays to
which the procedure should be applicable. Redefining it for each such array would be cumbersome,
inelegant, and can be avoided by introducing its operand as a parameter as follows.
PROCEDURE Add(VAR x: Vector);
VAR i: INTEGER;
BEGIN sum := 0;
FOR i := 0 TO N-1 DO sum := x[i] + sum END
END Add
The parameter x is introduced in the parameter list in the procedure heading. It thereby automatically
becomes a local object, in fact is a place-holder for the actual array that is specified in the procedure calls
Add(a); ... ; Add(b)
The arrays a and b are called actual parameters which are substituted for x, which is called formal
parameter. The formal parameter's specification must contain its type. This enables a compiler to check
whether or not an appropriate actual parameter is supplied. We say that a and b, the actual parameters,
must be compatible with the formal parameter x. In the example above, its type is Vector, presumably
declared in the environment of Add as
TYPE Vector = ARRAY N OF REAL;
VAR a, b: Vector
A better version of Add would include not only the array, but also the result sum as a parameter. We shall
later return to this example. But first we need explain that there exist two kinds of formal parameters,
namely variable and value parameters. The former are characterized by the symbol VAR, the latter by its
absence.
We conclude this chapter by giving the syntax of procedure declarations and procedure calls:
ProcedureDeclaration = ProcedureHeading ";" block identifier.
ProcedureHeading = "PROCEDURE" identifier [FormalParameters].
block = {declaration} ["BEGIN" StatementSequence] "END".
FormalParameters = "(" [FPSection {";" FPSection}] ")" [":" qualident].
FPSection = ["VAR"] IdentList ":" FormalType.
FormalType = ["ARRAY" "OF"] qualident.
ProcedureCall = designator [ActualParameters].
ActualParameters = "(" [ExpList] ")".
Apart from the declaration of modules, we now have also encountered all forms of declarations.
declaration = "CONST" {ConstantDeclaration ";"} |
"TYPE" {TypeDeclaration ";"} |
"VAR" {VariableDeclaration ";"} |
ProcedureDeclaration ";".
12.1 Variable Parameters
As its name indicates, the actual parameter corresponding to a formal variable parameter (specified by
the symbol VAR) must be a variable. The formal identifier then stands for that variable.
PROCEDURE exchange(VAR x, y: INTEGER);
VAR z: INTEGER;

31
BEGIN z := x; x := y; y := z
END exchange
The procedure calls
exchange(a, b); exchange(A[i], A[i+1])
then have the effect of the above three assignments, each with appropriate substitutions made upon the
call. The following points should be remembered:
1. Variable parameters may serve to transmit a computed result outside the procedure.
2. The formal parameter acts as a place-holder for the substituted actual parameter.
3. The actual parameter cannot be an expression, and therefore not a constant either, even if no
assignment to its formal correspondent is made.
4. If the actual parameter involves indices, these are evaluated when the formal-actual substitution is
made.
5. The types of corresponding formal and actual parameters must be the same.
12.2 Value Parameters
Value parameters serve to pass a value from the calling side into the procedure and constitute the
predominant case of parameters. The corresponding actual parameter is an expression, of which a
variable or a constant are a particular and simple case. The formal value parameter must be considered as
a local variable of the indicated type. Upon call, the actual expression is evaluated and the result is
assigned to that local variable. As a consequence, the formal parameter may later be assigned new values
without affecting any part of the expression. In a sense, actual expression and formal parameter become
decoupled as soon as the procedure is entered. As an illustration, we formulate the previously shown
program to compute the power z = xi as a procedure.
PROCEDURE ComputePower(VAR z: REAL; x: REAL; i: INTEGER);
BEGIN z := 1.0;
WHILE i > 0 DO
IF ODD(i) THEN z := z*x END;
x := x*x; i := i DIV 2
END
END ComputePower
Possible calls are, for example
ComputePower(u, 2.5, 3)
ComputePower(A[i], B[i], 2)

for u := 2.53
for Ai := Bi2

Concerning value parameters, we must keep in mind that since the formal parameter represents a local
variable, storage is needed for that variable. This may be of concern, if its type is an array of many
elements. In this case, it is recommended to specify a variable parameter, even if this parameter is used
for the import of values only.
Note that in the example above z and x are declared in distinct FPSections, because "VAR z, x: REAL"
would classify x as a variable parameter too, making it impossible to place a general expression in its
corresponding actual parameter place.
12.3 Open Array Parameters
If a formal parameter type denotes an array structure, its corresponding actual parameter must be an array
of the identical type. This implies that it must have elements of identical type and the same bounds of the
index range. Often this restriction is rather severe, and more flexibility is highly desirable. It is provided
by the facility of the so-called open array which requires that the types of the elements of the formal and
actual arrays be the same, but leaves the index range of the formal array open. In this case, arrays of any
size may be substituted as actual parameters. An open array is specified by the element type preceded by
"ARRAY OF". For example, a procedure declared as
PROCEDURE P(s: ARRAY OF CHAR)
allows calls with character arrays of arbitrary length. The actual length is obtained by calling the standard
function LEN(s).

32
13. Function Procedures
So far we have encountered two possibilities to pass a result from a procedure body to its calling place:
the result is either assigned to a non-local variable or to a variable parameter. There exists a third method:
the function procedure. It permits the use of the computed result (as an intermediate value) in an
expression. The function procedure identifier stands for a computation as well as for the computed result.
The procedure declaration is characterized by the indication of the type of the result following the
parameter list. As an example, we rephrase the power computation given above.
PROCEDURE power(x: REAL; i: INTEGER): REAL;
VAR z: REAL;
BEGIN z := 1.0;
WHILE i > 0 DO
IF ODD(i) THEN z := z*x END;
x := x*x; i := i DIV 2
END;
RETURN z
END power
Possible calls are
u := power(2.5, 3)
A[i] := power(B[i], 2)
u := x + power(y, i+1) / power(z, i-1)
The statement which passes the result consists of the symbol RETURN followed by the expression
specifying the result. Return statements may occur at several places in the procedure body and cause
termination of its execution. It is good practice, however, that a single return statement is placed
immediately preceding the closing symbol END.
Return statements may also be used within normal procedures, in which case no expression follows the
RETURN symbol. This facility may serve to signal unusual termination. Such a return statement is
implied in the end of every procedure.
Calls inside an expression are called function designators. Their syntax is the same as that of procedure
calls. However, a parameter list is mandatory, although it may be empty.
ReturnStatement = "RETURN" [expression].
We now revisit the previous example of adding the elements of an array and formulate it as a function
procedure.
PROCEDURE sum(VAR a: Vector; n: INTEGER): REAL;
VAR i: INTEGER; s: REAL;
BEGIN s := 0.0;
FOR i := 0 TO n-1 DO s := a[i] + s END ;
RETURN s
END sum
This procedure, as previously specified, sums the elements a[0] ... a[n-1], where n is given as a value
parameter, and may be different from (but not larger than!) the number of elements N. A more elegant
solution specifies a as an open array, omitting the explicit indication of the array's size.
PROCEDURE sum(VAR x: ARRAY OF REAL): REAL;
VAR i: INTEGER; s: REAL;
BEGIN s := 0.0;
FOR i := 0 TO LEN(x)-1 DO s := x[i] + s END ;
RETURN s
END sum
Obviously, procedures are capable of generating more than one result by making assignments to several
variables. Only one value, however, can be returned as the result of a function. This value, moreover,
cannot be of a structured type. Therefore, the other results must be passed to the caller via VAR
parameter or assignment to variables that are not local to the function procedure. Consider for example

33
the following procedure which computes a primary result denoted as the function's value and a secondary
result used to count the number of times the procedure is called.
PROCEDURE square(x: INTEGER): INTEGER;
BEGIN INC(n); RETURN x*x
END square
There is nothing remarkable about this example as long as the secondary result is used for its indicated
purpose. However, it might be misused as follows:
m := square(m) + n
Here the secondary result occurs as an argument of the expression containing the function designator
itself. The consequence is that e.g. the values
square(m) + n

and n + square(m)

differ, seemingly defying the basic law of commutativity of addition.

Assignments of values from within function procedures to non-local variables are called side-effects. The
programmer should be fully aware of their capability of producing unexpected results when the function
is used inappropriately. We summarize:
1. A function procedure specifies a result which is used at its place of call as an argument of an
expression.
2. The result of a function procedure cannot be structured.
3. If a function procedure generates secondary results, it is said to have side-effects. These must be used
with care. It is advisable to use a regular procedure instead, which passes its results via VAR
parameters.
4. We recommend to choose function identifiers which are nouns rather than verbs. The noun then
denotes the function's result. Boolean functions are appropriately labelled by an adjective. In contrast,
regular procedures should be designated by a verb describing their action.

14. Recursion
Procedures may not only be called, but can call procedures themselves. Since any procedure that is
visible can be called, a procedure may call itself. This self-reactivation is called recursion. Its use is
appropriate when an algorithm is recursively defined and in particular when applied to a recursively
defined data structure.
Consider as an example the task of listing all possible permutations of n distinct objects a[0] ... a[n-1].
Calling this operation Permute(n), we can formulate its algorithm as follows:
First, keep a[n-1] in its place and generate all permutations of a[0] ... a[n-2] by calling Permute(n-1), then
repeat the same process after having exchanged a[n-1] with a[0]. Continue repeating for all values i = 1
... n-2. This recipe is formulated as a program as follows, using characters as permuted objects.
MODULE Permute;
IMPORT Texts, Oberon;
VAR a: ARRAY 20 OF CHAR;
W: Texts.Writer;
PROCEDURE permute(k: INTEGER);
VAR i: INTEGER; t: CHAR;
BEGIN
IF k = 0 THEN Texts.WriteString(W, a)
ELSE permute(k-1);
FOR i := 0 TO k-1 DO
t := a[i]; a[i] := a[k]; a[k] := t;
permute(k-1);
t := a[i]; a[i] := a[k]; a[k] := t
END
END
END permute;

34
PROCEDURE Go*; (*command*)
VAR n: INTEGER; S: Texts.Scanner;
BEGIN Texts.OpenScanner(S, Oberon.Par.text, Oberon.Par.pos); Texts.Scan(S);
COPY(S.s, a); n := 0;
WHILE a[n] > 0X DO INC(n) END ;
permute(n-1);
Texts.WriteLn(W); Texts.Append(Oberon.Log, W.buf)
END Go;
BEGIN Texts.OpenWriter(W)
END Permute.
The data generated from the command Permute.Go ABC are
ABC BAC CBA BCA ACB CAB
Every chain of recursive calls must terminate at some time, and hence every recursive procedure must
place the recursive call within a conditional statement. In the example given above, the recursion
terminates when the number of the objects to be permuted is 1. (Concerning input and output
conventions, see Ch. 22.3 and 22.4).
The number of possible permutations can easily be derived from the algorithm's recursive definition. We
express this number appropriately as a function np(n). Given n elements, there are n choices for the
element a[n-1], and with each fixed a[n-1] we obtain np(n-1) permutations. Hence the total number
np(n) = n*np(n-1). Evidently np(1) = 1. The computation of np is now expressible as a recursive
function procedure.
PROCEDURE np(n: INTEGER): INTEGER;
BEGIN
IF n <= 1 THEN RETURN 1
ELSE RETURN n * np(n-1)
END
END np
We recognize np as the factorial function, defined as
f(n) = 123 ... n
This formula suggests to program the algorithm using repetition instead of recursion
PROCEDURE np(n: INTEGER): INTEGER;
VAR p: INTEGER;
BEGIN p := 1;
WHILE n > 1 DO p := n*p; DEC(n) END ;
RETURN p
END np
This formulation will compute the result more efficiently than the recursive version. The reason is that
every call requires some "administrative" instructions whose execution costs time. The instructions
representing repetition are less time-consuming. Although the difference may not be too relevant, it is
recommended to employ a repetitive formulation in place of the recursive one, whenever this is easily
possible. It is always possible in principle; however, the repetitive version may complicate and obscure
the algorithm to such a degree that the advantages turn into disadvantages. For example, a repetitive form
of the procedure permute is much less simple and obvious than the one shown above. To illustrate the
usefulness of recursion, two additional examples follow. They typically stem from problems whose
solution is naturally found and explained using recursion.
The first example belongs to the class of algorithms which operate on data whose structure is also
defined recursively. The specific problem consists of converting simple expressions into their
corresponding postfix form, i.e. a form in which the operator follows its operands. An expression shall
here be defined using EBNF as follows:
expression = term {("+"|"-") term}.
term =
factor {("*"|"/") factor}.
factor =
letter | "(" expression")"|"[" expression "]".

35
Denoting terms as T0, T1, and factors as F0, F1, the rules of conversion are
T0 + T1
T0 - T1
F0 * F1
F0 / F1
(E)
[E]

T0 T1 +
T0 T1 F0 F1 *
F0 F1 /
E
E

The following program truthfully mirrors the structure of the syntax of accepted expressions. As the
syntax is recursive, so is the program. This close mirroring is the best guarantee for the program's
correctness. Note also that similarly repetition in the syntax, expressed by the curly brackets, yields
repetition in the program, expressed by while statements. No procedure in this program calls itself
directly. Instead, recursion occurs indirectly by a call of expression on term, which calls factor, which
calls expression. Indirect recursion is obviously much less visible than direct recursion.
This example also illustrates a case of local procedures. Notably, factor is declared local to term, and
term local to expression, following the rule that objects should preferrably be declared local to the scope
in which they are used. This rule may not only be advisable, but even crucial, as demonstrated by the
variables addop (local to term) and mulop (local to factor). If these variables were declared globally, the
program would fail. To find the explanation, we must recall the rule that local variables exist (and are
given storage) during the time in which their procedure is active. An immediate consequence is that in
the case of a recursive call, new incarnations of the local variables are created. Hence, there exist as
many as there are levels of recursion. This also implies that a programmer must ensure that the depth of
recursion never becomes inordinately large.
MODULE Postfix;
IMPORT Texts, Oberon;
VAR ch: CHAR;
W: Texts.Writer; R: Texts.Reader;
PROCEDURE expression;
VAR addop: CHAR;
PROCEDURE term;
VAR mulop: CHAR;
PROCEDURE factor;
BEGIN
IF ch = "(" THEN
Texts.Read(R, ch); expression;
WHILE ch # ")" DO Texts.Read(R, ch) END
ELSIF ch = "[" THEN
Texts.Read(R, ch); expression;
WHILE ch # "]" DO Texts.Read(R, ch) END
ELSE
WHILE (ch < "a") OR (ch > "z") DO Texts.Read(R, ch) END ;
Texts.Write(W, ch)
END ;
Texts.Read(R, ch)
END factor;
BEGIN (*term*) factor;
WHILE (ch = "*") OR (ch = "/") DO
mulop := ch; Texts.Read(R, ch); factor; Texts.Write(W, mulop)
END
END term;
BEGIN (*expression*) term;
WHILE (ch = "+") OR (ch = "-") DO
addop := ch; Texts.Read(R, ch); term; Texts.Write(W, addop)
END
END expression;

36
PROCEDURE Parse*;
BEGIN Texts.OpenReader(R, Oberon.Par.text, Oberon.Par.pos); Texts.Read(R, ch);
WHILE ch # ~ DO
expression; Texts.WriteLn(W); Texts.Append(Oberon.Log, W.buf);
Texts.Read(R, ch)
END
END Parse;
BEGIN Texts.OpenWriter(W);
END Postfix.
A sample of data processed and generated by the command Postfix.Parse is shown below.
a+b
a*b+c
a+b*c
a*(b/[c-d])

ab+
ab*c+
abc*+
abcd-/*

The next program example demonstrating recursion belongs to the class of problems that search for a
solution by trying and testing. A partial "solution" which is once "posted" may, after testing had shown
its invalidity, have to be retracted. This kind of approach is therefore also called backtracking. Recursion
is often very convenient for the formulation of such algorithms.
Our specific example is supposed to find all possible placement of 8 queens on a chess board in such a
fashion that none is checking any other piece, i.e. each row, column, and diagonal must contain at most
one piece. The approach consists of trying to place a queen in column i, starting with i = 0, and,
proceeding to the right, assuming that each column to the left contains a correctly placed queen already.
If no place is free in column i, the next column to its left has to be reconsidered. The information
necessary to deduce whether or not a given square is still free, is represented by the three global variables
called row, d1, d2 such that
row[j] & d1[i+j] & d2[N-1+i-j] = "the square in column i and row j is free"
Recursion occurs directly in procedure TryCol. The auxiliary procedures PlaceQueen and RemoveQueen
could in principle be declared local to TryCol. However, there exists a single chess board only
(represented by row, d1, d2), and these procedures are appropriately considered as belonging to these
global data, and hence not as local to (each incarnation of) TryCol.
MODULE Queens;
IMPORT Texts, Oberon;
CONST N = 8; (*no. of rows and columns*)
VAR x: ARRAY N OF INTEGER;
(*x[i] = j means: a queen is on field j in column i*)
row: ARRAY N OF BOOLEAN;
(*row[i] = "no queen on i-th row"*)
d1: ARRAY 2*N-1 OF BOOLEAN;
(*d1[i] = "no queen on i-th upleft to lowright diagonal"*)
d2: ARRAY 2*N-1 OF BOOLEAN;
(*d2[i] = "no queen on i-th lowleft to upright diagonal"*)
W: Texts.Writer;
PROCEDURE Clear;
VAR i: INTEGER;
BEGIN
FOR i := 0 TO N-1 DO row[i] := TRUE END ;
FOR i := 0 TO 2*(N-1) DO d1[i] := TRUE; d2[i] := TRUE END
END Clear;
PROCEDURE WriteSolution;
VAR i: INTEGER;
BEGIN
FOR i := 0 TO N-1 DO Texts.WriteInt(W, x[i], 4) END ;
Texts.WriteLn(W)

37
END WriteSolution;
PROCEDURE PlaceQueen(i, j: INTEGER);
BEGIN
x[i] := j; row[i] := FALSE; d1[i+j] := FALSE; d2[N-1+i-j] := FALSE
END PlaceQueen;
PROCEDURE RemoveQueen(i, j: INTEGER);
BEGIN
row[i] := TRUE; d1[i+j] := TRUE; d2[N-1+i-j] := TRUE
END RemoveQueen;
PROCEDURE TryCol(i: INTEGER);
VAR j: INTEGER;
BEGIN
FOR j := 0 TO N-1 DO
IF row[j] & d1[i+j] & d2[N-1+i-j] THEN
PlaceQueen(i, j);
IF i < N-1 THEN TryCol(i+1) ELSE WriteSolution END ;
RemoveQueen(i, j)
END
END
END TryCol;
PROCEDURE Find*; (*command*)
BEGIN Clear; TryCol(0); Texts.Append(Oberon.Log, W.buf)
END Find;
BEGIN Texts.OpenWriter(W)
END Queens.

15. Type Declarations

Every variable declaration specifies the variable's type as its constant property. The type can be one of
the standard, primitive types, or it may be of a type declared in the program itself. Type declarations have
the form
TypeDeclaration = identifier "=" type.
They are preceded by the symbol TYPE. Types are classified into unstructured and structured types.
Each type essentially defines the set of values which a variable of this type may assume. A value of an
unstructured type is an atomic unit, whereas a value of structured type has components (elements). For
example, the type INTEGER is unstructured; its elements are atomic. It does not make sense, e.g. to
refer to the third bit of the value 13; the circumstance that a number may "have a third bit", or a second
digit, is a characteristic of its (internal) representation, which intentionally is to remain unknown.
In the following sections we shall show how to declare structured types. We distinguish between various
structuring methods of which we have so far encountered the array only. In addition, there exists the
record type. A facility to introduce structures that vary dynamically during program execution is based
on the concept of pointers and will be discussed in a separate chapter.
type = qualident | ArrayType | RecordType | PointerType | ProcedureType.
Before proceeding to the various kinds of types, we note that in general, if a type T is declared by the
declaration
TYPE T = someType
and a variable t is declared as
VAR t: T
then these two declarations can always be merged into the single declaration
VAR t: someType

38
However, in this case t's type has no explicit name and therefore remains anonymous. Typically, record
types are given explicit names.
The concept of type is important, because it divides a program's set of variables into disjoint classes.
Inadvertent assignments among members of different classes can therefore be detected by a mere
inspection of the program text without executing the program. Given, for example, the declarations
VAR b: BOOLEAN; i: INTEGER; x: REAL
the assignment b := i is impossible, because the types of b and i are incompatible. Two types are said to
be compatible, if they are declared as equal or satisfy certain compatibility rules: If one type T0 is
included in another type T1, then the assignment t1 := t0 is admissible, but not t0 := t1. Type inclusion is
restricted to the standard numerical types. We note that x := i is admisible, whereas i := x is not.
SHORTINT INTEGER LONGINT REAL LONGREAL

16. Record Types

In an array all elements are of the same type. In contrast to the array, the record structure offers the
possibility to declare a collection of elements as a unit even if the elements are of different types. The
following examples are typical cases where the record is the appropriate choice of structuring method. A
date consists of three elements, namely day, month, and year. A description of a person may consist of
the person's names, sex, identification number, and birthdate. This is expressed by the following type
declarations
Date =

RECORD day, month, year: INTEGER END

Person = RECORD
firstName, lastName: ARRAY 32 OF CHAR;
male: BOOLEAN;
idno: INTEGER;
birth: Date
END
The record structure makes it possible to refer either to the entire collection of data or to individual
elements. Elements of a record are also called record fields, and their names are called field identifiers.
This stems from the habit of looking at such data as forms or tables drawn on paper with individual fields
delineated as rectangles and labelled with field names. Similar to arrays, where we denote the i-th
element of an array a by a[i], i.e. by the array identifier followed by an index, we denote the field f of a
record r by r.f , i.e. by the record identifier followed by the field's name. For example, given the
variables
d1, d2 : Date;
p1, p2 : Person;
student: ARRAY 100 OF Person
we can construct, for example, the following variable designators:
d1.day
d2.month
p1.firstName
p1.birth
These examples show that fields may themselves be structured. Similarly, records may be elements of
array or record structures, i.e. there exists the possibility to construct hierarchies of structures. As a
consequence, selectors of elements can be sequenced, as shown by the following examples of
designators. The multi-dimensional array discussed in the chapter on arrays now appears as a particular
case of these structuring hierarchies.
p1.lastName[0]
p2.birth.year
student[23].idno
student[k].firstName[0]

39
At first sight the record may appear as a generalized array, because it relaxes the restriction that all
elements be of the same type. However, in another aspect it is more restrictive than the array: the selector
of the element must be a fixed field identifier, whereas the index selecting an array element may be an
expression, i.e. a result of previous computations.
It is important to note that a record may assume arbitrary combinations of its field's values. Hence, in the
example of the type Date, a value day = 31 may coexist with month = 2, although this is not an actual
date.
The syntax of a record declaration is defined as follows.
RecordType = "RECORD" FieldListSequence "END".
FieldListSequence = FieldList {";" FieldList}.
FieldList = [IdentList ":" type].
and that of a designator is
designator = qualident {selector} .
selector = "." identifier | "[" ExpList "]" | "^".
Note: Oberons concept of record type extension is not discussed in this book.

17. Dynamic Data Structures and Pointers

Array and record structures share the common property that they are static. This implies that variables of
such a structure maintain the same structure during the whole time of their existence. In many
applications, this is an intolerable restriction; they require data which do not only change their value, but
also their composition, size, and structure. Typical examples are lists and trees that grow and shrink
dynamically. Instead of providing list and tree structures, a collection that for some applications would
again not suffice, Oberon offers a basic tool to construct arbitrary structures. This is the pointer type.
Every complex data structure ultimately consists of elements whose structure is static. Pointers, i.e.
values of pointer types, are themselves not structured, but rather are used to establish relationships among
those static elements, usually called nodes. We also say that pointers link elements or point to elements.
Evidently, different pointer variables may point to the same element, hence providing the possibility to
compose arbitrarily complex structures, and at the same time opening many pitfalls for programming
mistakes that are difficult to pinpoint. Operating with pointers indeed requires utmost care.
Pointers in Oberon cannot point to arbitrary variables. The type of variable to which they point must be
specified in the pointer type's declaration, and the pointer type is said to be bound to the referenced
object's type. Example:
TYPE Node = POINTER TO NodeDescriptor
VAR p0, p1: Node
Here Node (and thereby also variables p0 and p1) are bound to the type NodeDescriptor, i.e. they can
point to variables of type NodeDescriptor only. These variables, typically of record type, are not created
by the declaration of p0 and p1. Instead, they are created by a call an allocation procedure. In Oberon, it
is represented by a predefined operator called NEW. The statement NEW(p0) then creates a (record)
variable of type NodeDescriptor and assigns a pointer referring to that variable (i.e. a value of type Node)
to p0. The created variable is said to be dynamically created (allocated); it has no name, is anonymous,
and can be accessed only via a pointer using the dereferencing operator ^. The said variable is denoted by
the designator p0^. If the referenced variable is of record type with, say, fields x and y, then the fileds are
denoted, for example, by p0^.x, p1^.y. Since in this case it is clear that field selection applies to the
referenced record rather than to the pointer, the designators may be abbreviated into p0.x and p1.y.
PointerType = "POINTER" "TO" type.
What really makes pointers such a powerful tool is the circumstance that they may point to variables
which themselves contain pointers. This reminds us of procedures that call procedures and thereby
introduce recursion. In fact, pointers are the tool to implement recursively defined data structures (such
as lists and trees). The nature of the recursive data structure is evident from the declaration of the type of
its elements.

40
Just as every recursion of procedure activation must terminate at some time, so must every recursion in
referencing terminate at some point. The role of the if statement to terminate procedural recursion is here
taken by the special pointer value NIL terminating referencing recursion. NIL points to no object. It is an
obvious consequence that a designator of the form p^ (p^.x, p.x) must never be evaluated, if p = NIL. We
summarize the following essential points.
1. Every pointer type is bound to a type; its values are pointers which point to variables of that type.
2. The referenced variables are anonymous and can be accessed via pointers only.
3. The referenced variables are dynamically created by an allocation procedure which assigns the
variable's pointer to p.
4. The pointer constant NIL belongs to every pointer type and points to no object.
5. The variable referenced by a pointer p is denoted by the designator p^. In order for p^ to be
meaningful, p must not have the value NIL.
6. Field designators of the form p^.x may be abbreviated to p.x.
Lists, also called linear lists or chains, are characterized by consisting of record typed nodes that each
have exactly one element being a pointer to a record of the same type as itself. This implies recursion. A
list pointer declaration then assumes the characteristic form
List = POINTER TO ListDesc
ListDesc =
RECORD
key: INTEGER;
Data: ...
next: List
END
"Data" actually stands for any number of fields representing data pertaining to the listed node. Key is part
of these data; it is mentioned separately here because it is quite common to associate with each element a
unique identifying key, and also because it will be used in subsequent examples of operations on lists.
The essential ingredient here, however, is the field next, so labelled because it evidently is the pointer to
the next element in the list. Direct recursion in data type declarations is not permitted for the obvious
reason that there would be no evident termination. The declaration given above cannot be abbreviated
into
List =
RECORD
key: INTEGER; next: List
END
Assume now that a list is accessible in a program via its first element, denoted by the pointer variable
first: List
The empty list is represented by first = NIL. A non-empty list is most conveniently constructed by
inserting new elements at its front. The following assignments are needed to insert one element (let p be
an auxiliary variable of type List)
NEW(p); (* assign values to p.key and p.data *) p.next := first; first := p
Having constructed a list by repeated insertion of nodes, we may wish to search the list for a node with
key value equal to a given x. We evidently use a repetition; the while statement is appropriate, because
we do not know the number of nodes (and hence repetitions) beforehand. It is wise to include the case of
the empty list!
p := first;
WHILE (p # NIL) & (p.key # x) DO p := p.next ND ;
IF p # NIL THEN found END
We draw attention to the fact that here we make use of the rule that the term b is not evaluated, if in the
expression a & b the factor a is found to be FALSE. If this rule would not hold, the factor p.key # x
might be evaluated with p = NIL, which is illegal.

41
The second frequently encountered dynamic data structure is the tree. It is characterized by its nodes
having n pointer fields each, where n is the degree of the tree. The common and in some sense optimal
case is the binary tree with n = 2. Lists now appear as degenerate trees of degree 1. The respective
declarations are
Tree = POINTER TO TreeNode;
TreeNode =
RECORD
key: INTEGER;
data: ...
left, right: Tree
END
The place of the variable first in the case of lists is taken by a variable to be
root: Tree
with root = NIL denoting the empty tree. Trees are commonly used to represent collections of data in
order of ascending key values, making retrieval very efficient. The following statements represent a
search in an ordered binary tree, whose similarity to the binary search in an ordered array is remarkable.
Again, p is an auxiliary variable (of type Tree).
p := root;
WHILE (p # NIL) & (p.key # x) DO
IF p.key < x THEN p := p.right
ELSE p := p.left
END
END ;
IF p # NIL THEN found END
This example is a repetitive version of the tree search. Next we show the recursive version. It is, in
addition, extended such that a new node is created and inserted at the appropriate place, whenever no
node with key value x exists.
PROCEDURE search(VAR p: Tree; x: INTEGER): Tree;
VAR q: Tree;
BEGIN
IF p # NIL THEN
IF p.key < x THEN
q := search (p.right, x)
ELSIF p.key > x THEN
q := search (p^.left, x)
ELSE
q := p
END
ELSE (*not found, hence insert*)
NEW(q); q.key := x; q.left := NIL; q.right := NIL
END ;
RETURN q
END search
The call search(root, x) now stands for a search of x in the tree represented by the variable root.
And this concludes our examples of operations on lists and trees to illustrate pointer handling. Lists and
trees have nodes which are all of the same type. We draw attention to the fact that the pointer facility
admits the construction of even more general data structures consisting of nodes of various types. Typical
for all these structures is that all nodes are declared as record types. Hence, the record emerges as a
particularly useful data structure in conjunction with pointers.
Creation of nodes is expressed by the standard procedure NEW, which is part of a system's storage
management. We assume that retrieval of storage is performed automatically by a so-called storage
reclamation mechanism, also called garbage collector. It relies on the fact that records that are no longer
referenced by any pointer may be recollected, i.e. their storage can be recycled.

42
18. Procedure Types
So far, we have regarded procedures exclusively as program parts, i.e. as texts that specify actions to be
performed on variables whose values are numbers, logical values, characters, etc. However, we may take
the view that procedures themselves are objects that can be assigned to variables. In this light, a
procedure declaration appears as a special kind of constant declaration, the value of this constant being a
procedure. If we allow variables in addition to constants, it must be possible to declare types whose
values are procedures. These are called procedure types.
A procedure type declaration specifies the number and the types of parameters and, if it is to be a
function procedure, the type of the result. For example, a procedure type with one REAL argument and a
result of the same type is declared by
Function = PROCEDURE (x: REAL): REAL
The mathematical functions of sine, cosine, square root, exponential and logarithm are all of this type.
The general syntax is
ProcedureType = "PROCEDURE" [FormalTypeList].
FormalTypeList =
"(" [["VAR"] FormalType {"," ["VAR"] FormalType}] ")" [":" identifier].
If we now declare a variable f: Function, the assignment f := Math.sin is possible. Subsequently the call
f(x) is equivalent to Math.sin(x).
It is now also possible to declare functions and procedures which themselves have functions and
procedures as parameters. For example, a function procedure to integrate any function over a certain
interval can be expressed in the following way. It adds the values of the given parametric function ofver
the given interval a to b in n steps.
PROCEDURE integral (f: Function; a, b: REAL; n: INTEGER): REAL;
VAR i: INTEGER; sum, dx: REAL;
BEGIN dx := (b-a)/n; sum := 0;
FOR i := 0 TO n-1 DO sum := f(a + i*dx + 0.5*dx) + sum END ;
RETURN sum * dx
END integral
The integrations of, for example, the sine function over the interval 0 to , and that of the exponential
function from 0 to 1 are expressed simply as
integral (Math.sin, 0, 3.14159, 20)
integral (Math.exp, 0, 1, 20)
and they yield 2.0020 and 1.7181 respectively, which approximate the correct values 2*cos(0) and e 1).
To conclude, we emphasize the restriction of Oberon that procedures that are assigned to variables or are
passed as parameters, must not be declared local to any other procedure. Neither can they be standard
procedures (such as ODD, INCL, etc.).

43

Part 3
19. Modules
Modules are the most important feature distinguishing Oberon from its ancestor Pascal. We have
already encountered modules, simply because every program is a module. However, most systems are
not only a single module, but rather a set of several modules, each module containing declared objects
such as constants, variables, procedures, and types. Objects declared in a modules M0 can be referenced
in another module M1, if they are explicitly made to be known in M1, i.e. if they are imported by M1.
In the examples of the preceding chapters, we have typically imported procedures for input and output
from modules containing collections of frequently used procedures. These procedures are actually part
of our program, even if we have not programmed them and they are textually disjoint.
The key point is that modules can be kept in a program "library" and are automatically referenced when
a programmer's program is loaded and executed. Hence it is possible to prepare collections of
frequently used operations (such as for input and output), and to avoid reprogramming them each time a
program needs such operations. Modern implementations go even one step further and offer what is
called separate compilation. This signifies that such modules are not stored in the program library in
source form as Oberon texts, but rather in compiled form. Upon program loading, the compiled (main)
module is joined (linked) with the precompiled modules from which it imports objects. In this case, the
compiler must also have access to descriptions of the objects of the previously compiled, imported
modules, when the importing program is compiled. This facility distinguishes separate compilation
from independent compilation as it exists in typical implementations of Fortran, Pascal, and assemblers.
Every subsidiary module may again import objects from other modules. A program therefore
constitutes an entire hierarchy of modules. The main program is said to be at the highest level, those
modules which do not import objects at all being at the lowest level. Usually, a programmer is not even
aware of this hierarchy, because his modules import objects from modules that he has not programmed
himself; therefore he is unaware of their imports and of the module hierarchy below them. In principle,
however, his program is the text written by himself, extended with the texts of the imported modules.
These extensions are usually quite large (even if the direct imports consist of a few output procedures
only). In principle, the indirect imports constitute the entire environment or operating system. In a
single-user computer system, there is virtually no need for any parts that are neither directly nor
indirectly imported by the main program. However, some modules, such as the basic input/output and
file system, may be required by all programs and therefore become de facto resident and may therefore
be regarded as the operating system.
The principal motivation behind the partitioning of a program into modules is - beside the use of
modules provided by other programmers - the establishment of a hierarchy of abstractions. For
example, in the previously encountered cases of imported input/output procedures, we merely wish to
have them available, but do not need to know - or rather do not wish to bother to learn - how these
procedures function in detail. To abstract means to "take away" from the essentials and thereby to
ignore certain details. Each module constitutes an abstraction, if we regard it from "the outside". We
even wish to go one step further: we wish not only to ignore the details of its innards, but to hide them.
The primary reason for this wish is that if the innards are protected from outside access, we can
guarantee their correct functioning, thereby being able to limit the area of error search in the case of a
malfunctioning program.
The second, but not less important reason is to make it possible to change (improve) the innards of
imported modules without having to change (and/or recompile) the importing modules. This effective
decoupling of modules is indispensible for the development of large programs, in particular, if modules
are developed by different people, and if we regard the operating system as the low section of a
program's module hierarchy. Without decoupling, any change or correction in an operating system or in
library modules would become virtually impossible.
A direct consequence of this need for decoupling is the necessity for a textual separation of the
essentials from the details. The essentials of a module are the properties of objects that are importable
into other modules; the details are those parts that are to be hidden and protected. In Oberon, objects to
be visible in other modules are specially marked. The mark is an asterisk following the identifier in its
declaration.

44
An importer of a module needs to know the properties of those objects only which are exported by the
imported module. It is customary and convenient to provide an extract of the module containing those
relevant declarations only. We call it the definition of a module. The definition forms a contract
between the client (importer) and the designer of a module. It is therefore also called the interface of the
module. The implementation part remains the property of the module's designer. As long as the
designer alters details without affecting the definition, he need not report his activity to the clients of his
module. Modules are compiled separately, and are therefore called compilation units.
Concluding this introduction to the module concept, we postulate that adequate implementations
provide full type compatibility checking between objects, independent of whether these objects are
declared in the same or in different modules, i.e. the checking mechanism of the compiler functions also
across module boundaries. However, the programmer must realize that this checking provides no
absolute safeguard against mistakes. After all, it concerns formal, syntactic aspects only; it does not
cover the semantics. It would not detect, for instance, the replacement of the algorithm for the sine
function by that of the cosine function. However, it must not be regarded as the duty of a compiler to
protect programmers against malicious "colleagues".

20. Definitions and Implementations

A definition or interface specifies those properties of a module which are relevant to its clients
(importers, users). It consists of the declarations of the exported identifiers. We will here consider
definitions and implementations as disjoint texts. However, in Oberon there is only one text, namely the
module, in which the exported objects are explicitly marked (by an asterisk). The definition may here
be considered as an extract of the module text.
Variables declared in a definition are global in the sense that they exist during the entire lifetime of the
program, although they are visible and accessible only in those modules (clients) which import the
module of their declaration. In other modules they remain invisible. Such variables are typically state
variables. But in general, export of variables should be used rarely, and if so, preferrably imported
variables should be read only.
In definitions, procedures are specified by their headings only. A heading specifies the procedures
parameters, i.e. their types, and in the case of function procedures the result type. This information
constitutes the procedures signature.
If a type is declared in a definition module, the full details of its declaration are visible in importing
modules. In the case of a record type, the definition specifies only those fields, which are visisble by
clients.
The following simple example exhibits the essential characteristics of modules, although typically
modules are considerably larger pieces of program and contain a longer list of declarations. This
example exports two procedures, put and get, adding data to and fetching data from a buffer which is
hidden. Effectively, the buffer can only be accessed through these two procedures; it is therefore
possible to guarantee the buffer's proper functioning, i.e. sequential access without loss of elements.
DEFINITION Buffer;
VAR nonempty, nonfull: BOOLEAN;
PROCEDURE put(x: INTEGER);
PROCEDURE get(VAR x: INTEGER);
END Buffer.
This definition part contains all the information about the buffer that a client is supposed to know. The
details of its operation, its realization, are contained in the corresponding implementation.
MODULE Buffer; (*cyclic buffer of integers*)
CONST N = 100;
VAR nonempty*, nonfull*: BOOLEAN;
in, out, n: INTEGER;
buf: ARRAY N OF INTEGER;
PROCEDURE put*(x: INTEGER);
BEGIN
IF n < N THEN

45
buf[in] := x; in := (in+1) MOD N;
INC(n); nonfull := n < N; nonempty := TRUE
END
END put;
PROCEDURE get*(VAR x: INTEGER);
BEGIN
IF n > 0 THEN
x := buf[out]; out := (out+1) MOD N;
DEC(n); nonempty := n > 0; nonfull := TRUE
END
END get;
BEGIN (*initialize*) n := 0; in := 0; out := 0;
nonempty := FALSE; nonfull := TRUE
END Buffer.
The definition is the relevant extract from the module losted here, implementing a fifo (first in first out)
queue. This fact, however, is not evident from the definition alone; normally the semantics are
mentioned in the form of a comment or other documentation. Such comments will usually explain what
the module performs, but not how this is achieved. Therefore, different implementations may be
provided for the same definition. The differences may lie in the details of the mechanism; for example,
the buffer might be represented as a linked list instead of an array (allocating buffer portions as needed,
hence not limiting the buffer's size). Or, the differences may even lie in the semantics. The following
program implements a stack (i.e. a lifo queue) instead of a fifo queue, nevertheless fitting the same
definition part. Any change in a module's semantics necessitates corresponding adjustments in the
module's clients, and must therefore be made with utmost care.
MODULE Buffer; (*stack of integers*)
CONST N = 100;
VAR n: INTEGER;
nonempty*, nonfull*: BOOLEAN;
buf: ARRAY N OF INTEGER;
PROCEDURE put*(x: INTEGER);
BEGIN
IF n < N THEN
buf[n] := x; INC(n); nonfull := n < N; nonempty := TRUE
END
END put;
PROCEDURE get*(VAR x: INTEGER);
BEGIN
IF n > 0 THEN
DEC(n); x := buf[n]; nonempty := n > 0; nonfull := TRUE
END
END get;
BEGIN n := 0; nonempty := FALSE; nonfull := TRUE
END Buffer.
Evidently, nonempty is the precondition for get, and nonfull is the precondition of put. This concludes
this introductory example.
The syntax of modules and definitions is
Definition =
"DEFINITION" identifier ";" {import} {declaration}
"END" identifier "." .
declaration = "CONST" {ConstantDeclaration ";"} |
"TYPE" {identifier ["=" type] ";"} |
"VAR" {VariableDeclaration ";" } |
ProcedureHeading ";".

46
Module = "MODULE" identifier ";" [IMPORT IdList} block identifier.
The import list names the modules which are imported. If a client module M1 is a client of M0, i.e.
imports M0, then the objects exported from M0, say a, b, are referenced by qualified identifiers of the
form M.x, for example M0.a, M0.b. This facility permits to import different objects with the same name
from different modules and to avoid conflicts of names. Standard identifiers are automatically imported
into all modules.
The possibility to publicize a module in the form of its definition and at the same time to retain its
operational details hidden in its implementation specification, is particularly convenient for the
establishment of program libraries. Such collections of standard routines belong to every programming
environment. They typically include routines for input and output operations, for file handling, and for
the computation of mathematical functions. Although there exists no rigid standard for Oberon, the
modules Files and Texts can be considered as standard modules available in all implementations of
Oberon (see Ch. 22). Here we present the definition of module Math as a first example, featuring
standard mathematical functions:.
DEFINITION Math;
PROCEDURE sqrt(x: REAL): REAL;
PROCEDURE exp(x: REAL): REAL;
PROCEDURE ln(x: REAL): REAL;
PROCEDURE sin(x: REAL): REAL;
PROCEDURE cos(x: REAL): REAL;
PROCEDURE arctan(x: REAL): REAL;
PROCEDURE real(x: INTEGER): REAL;
PROCEDURE entier(x: REAL): INTEGER;
END Math.

21. Program Decomposition into Modules

The quality of a program has many aspects and is an elusive property. A user of a program may judge it
according to its efficiency, reliability, or convenience of user dialog. Whereas efficiency can be
expressed in terms of numbers, convenience of usage is rather a matter of personal judgement, and all
too often a program's usage is called convenient as long as it is conventional. An engineer of a program
may judge its quality according to its clarity and perspicuity, again rather elusive and subjective
properties. However, if a property cannot be expressed in terms of precise numbers, this is no reason for
classifying it as irrelevant. In fact, program clarity is enormously important, and to demonstrate the
correctness of a program is ultimately a matter of convincing a person that the program is trustworthy.
How can we approach this goal? After all, complicated tasks usually do inherently require complex
algorithms, and this implies a myriad of details. And the details are the jungle in which the devil hides.
The only salvation lies in structure. A program must be decomposed into partitions which can be
considered one at a time without too much regard for the remaining parts. At the lowest level the
elements of the structure are statements, at the next level procedures, and at the highest level modules.
In parallel with program structuring proceeds the structuring of data into arrays, records, etc. at the
lower levels, and through association of variables with procedures and modules at the higher levels. The
essence of programming is finding the right (or at least an appropriate) structure, and the experienced
programmer is the person who has the intuition to find it at the stage of initial conception instead of
during a gradual process of improvements and modifications. Yet, the programmer who has the courage
to restructure when a better solution emerges is still much better off than the one who resigns and
elaborates a program on the basis of a clearly inadequate structure, for this leads to those products that
no one else, and ultimately not even the originator himself can "understand".
Even if there does not exist a recipe to determine the optimal structure of a program, there have
emerged some criteria for guidance in the process of finding good and avoiding bad structure. A basic
rule is that decomposition should be such that the connections between partitions are simple or "thin".
A perhaps oversimplifying criterion for the thickness of a connection - also called interface - between
two parts is the number of items that take part in it. Specifically, the interface of two modules is
sketched in terms of the module's import lists, and a measure for the interface's thickness is the number
of imported items. Hence, we must find a modularization which makes the import lists short. Naturally,
it is difficult to find an optimum, for, the import lists would be shortest, i.e. they would disappear if the
entire program were collapsed into a single module: a clearly undesirable solution.

47
The distinctive property of the module as the largest structuring unit is its capability to hide details and
thereby to establish a new level of abstraction. This property is used in several forms; we can
distinguish between the following cases:
1. The module separates two kinds of data representation and contains the collection of procedures that
perform the data conversion between the two levels. The typical example is a module for conversion
of numbers from their abstract, atomic representation into sequences of decimal digits and vice-versa.
Such modules contain no data of their own, they are typically packages of procedures.
2. The essence of a module is a set of data. It hides the details of the data representation by granting
access to these data through calls of its exported procedures only. An example of this case is a
module which contains a data set storing individual items organized in a way that items are found
quickly. Another is a module whose hidden data set is a disk store; it hides the peculiar details
necessary to operate the disk drive.
3. The module exports a data type and exports its associated operations. Typically, in Oberon such a
module exports one or several types in opaque mode (sometimes these are also called private types).
It thereby hides the details of the data type's structure and also the details of the operations. By hiding
them, it is possible to guarantee the validity of postulated invariant properties of each variable of such
a private type. The difference to modules of class 2 is that here variables of the private types are
declared in the client modules, whereas in class 2 modules the variable is itself hidden. Typical
examples are the queue and stack types, and perhaps the most successful such data abstraction is the
sequential file, also known as a stream.
This classification is not absolute. It cannot be, because all kinds share the common goal of hiding
details. Nevertheless, we can formulate a few rules that serve as guidelines in the design of modules:
1. Keep the number of imported modules small.
2. Keep the number of imported modules in definitions even smaller.
3. Export of variables should be considered as the exception, and imported variables should be treated
as "read-only" objects.
We conclude this chapter with an example that essentially falls into category 3. Let our stated goal be
the design of a cross reference generator for Oberon programs. More precisely, the program's purpose is
to read a text and to generate (1) a listing of the text with added line numbers and (2) a table of all
encountered words (identifiers) in alphabetical order, each with a list of the numbers of the lines in
which the word occurs. Moreover, comments and strings are to be skipped (i.e. their words are not to be
listed), and Oberon key symbols are not to be listed either.
We quickly recognize the task as being divisible into the scanning of the source text (eliminating the
parts that are to be skipped and ignored), and the recording and subsequent tabulating of the words. The
first part is conveniently performed by the main program module, the latter by a subsidiary module
which hides the data set and makes it accessible through two procedures: Insert (i.e. include a word)
and List (i.e. generate the requested table). A third module is used to generate the representation of
numbers as sequences of decimal digits. The three principal modules involved are called XREF,
TableHandler, and Texts.
We begin by presenting the main program XREF that scans the source text. A binary search is used to
recognize key words. The data set is of the type Table, imported from the TableHandler.
DEFINITION TableHandler;
CONST WordLength;
TYPE Word;
PROCEDURE Init (VAR w: Word);
PROCEDURE Insert(VAR s: ARRAY OF CHAR; ln: INTEGER; VAR w: Word);
(*enter pair s, ln in structuire w *)
PROCEDURE List(w: Word)
END TableHandler.
MODULE XREF;
IMPORT Texts, Oberon, TableHandler;
CONST N = 32; (* No. of keywords *)

48
VAR lno: INTEGER; (*current line number*)
Tab: TableHandler.Word;
W: Texts.Writer;
key: ARRAY N, 10 OF CHAR; (*table of key words*)
PROCEDURE heading;
BEGIN INC(lno); Texts.WriteInt(W, lno, 5); Texts.Write(W, " ")
END heading;
PROCEDURE Scan*; (*command*)
VAR beg, end, time: LONGINT;
k, m, l, r: INTEGER;
ch: CHAR;
id: ARRAY TableHandler.WordLength OF CHAR;
T: Texts.Text; R: Texts.Reader;
PROCEDURE copy;
BEGIN Texts.Write(W, ch); Texts.Read(R, ch);
END copy;
BEGIN TableHandler.Init(Tab);
Oberon.GetSelection(T, beg, end, time);
IF time >= 0 THEN
lno := 0; heading;
Texts.OpenReader(R, T, beg); Texts.Read(R, ch);
REPEAT
IF (CAP(ch) >= "A") & (CAP(ch) <= "Z") THEN (*word*)
k := 0;
REPEAT id[k] := ch; INC(k); copy
UNTIL ~(("A" <= CAP(ch)) & (CAP(ch) <= "Z") OR ("a" <= ch) & (ch <= "z"));
id[k] := 0X;
l := 0; r := N; (*binary search for key word*)
REPEAT m := (l+r) DIV 2;
IF key[m] < id THEN l := m+1 ELSE r := m END ;
UNTIL l >= r;
IF (r = N) OR (id # key[r]) THEN TableHandler.Insert(id, lno, Tab) END
ELSIF (ch >= "0") & (ch <= "9") THEN (*number*)
REPEAT copy UNTIL ~("0" <= ch) & (ch <= "9")
ELSIF ch = "(" THEN copy;
IF ch = "*" THEN (*skip comment*)
REPEAT
REPEAT
IF ch = 0DX THEN copy; heading ELSE copy END
UNTIL ch = "*";
copy
UNTIL ch = ")";
copy
END
ELSIF ch = 22X THEN (*string*)
REPEAT copy UNTIL ch = "";
copy
ELSIF ch = 0DX THEN (*end of line*)
copy; heading
ELSE copy
END
UNTIL R.eot;
Texts.WriteLn(W); Texts.Append(Oberon.Log, W.buf);
TableHandler.List(Tab)
END
END Scan;

49
BEGIN Texts.OpenWriter(W);
key[ 0] := "AND";
key[ 1] := "ARRAY";
key[ 2] := "BEGIN";
key[ 3] := "BOOLEAN"; key[ 4] := "BY";
key[ 5] := "CASE";
key[ 7] := "CONST";
key[ 7] := "DIV";
key[ 8] := "DO";
key[ 9] := "ELSE";
key[10] := "ELSIF";
key[11] := "END";
key[12] := "FOR";
key[13] := "IF";
key[14] := "IMPORT";
key[15] := "IN";
key[16] := "MOD";
key[17] := "MODULE";
key[18] := "NOT";
key[19] := "OF";
key[20] := "OR";
key[21] := "POINTER"; key[22] := "PROCEDURE"; key[23] := "RECORD";
key[24] := "REPEAT"; key[25] := "RETURN";
key[26] := "THEN";
key[27] := "TO";
key[28] := "TYPE";
key[29] := "UNTIL";
key[30] := "VAR";
key[31] := "WHILE"
END XREF.
Next we present the table handler. As seen from its definition part, it exports the private type Table and
its operations Insert and List. Notably the structure of the tables, and thereby also the access and search
algorithms remain hidden. The two most likely choices are the organizations of binary trees and of a
hash table. Here we opt for the former. The example is therefore a further illustration of the use of
pointers and dynamic data structures. The module contains a procedure to search and insert a tree
element, and a procedure that traverses the tree for the required tabulation (see also the Chapter on
dynamic data structures). Each tree node is a record with fields for the key, the left and right
descendants, and (the head of) a list of the line numbers. Another form of representation might be
chosen, for example a balanced tree, and such a new implementaion might be provided without clients
being aware of the change, because modules can be compiled separately. This is the key of
modularization and constructing large systems. It requires, however, that the modules interfaces are
wisely chosen.
MODULE TableHandler;
IMPORT Texts, Oberon;
CONST WordLength* = 32;
TYPE Item = POINTER TO RECORD
num: INTEGER; next: Item
END ;
Word* = POINTER TO RECORD
key: ARRAY WordLength OF CHAR;
first: Item;
(*list head*)
left, right: Word
END ;
VAR W: Texts.Writer;
PROCEDURE Init*(VAR w: Word);
BEGIN w := NIL
END Init;
PROCEDURE Insert*(VAR s: ARRAY OF CHAR; lno: INTEGER; VAR w: Word);
(*search node with name s*)
VAR w0: Word; t: Item;
BEGIN
IF w = NIL THEN (*insert new word*)
NEW(w0); COPY(s, w0.key); w0.left := NIL; w0.right := NIL;
NEW(t); t.num := lno; t.next := NIL; w0.first := t; w := w0
ELSIF s < w.key THEN Insert(s, lno, w.left)
ELSIF s > w.key THEN Insert(s, lno, w.right)
ELSE NEW(t); t.num := lno; t.next := w.first; w.first := t
END
END Insert;
PROCEDURE write(w: Word);
VAR t: Item;

50
BEGIN Texts.WriteString(W, w.key); t := w.first;
REPEAT Texts.WriteInt(W, t.num, 5); t := t.next UNTIL t = NIL;
Texts.WriteLn(W)
END write;
PROCEDURE List*(w: Word);
BEGIN
IF w # NIL THEN List(w.left); write(w); List(w.right) END ;
Texts.Append(Oberon.Log, W.buf)
END List;
BEGIN Texts.OpenWriter(W)
END TableHandler.

22. The Concept of a Sequence

22.1. About input and output
The usefulness and the success of high-level programming languages rests on the principle of
abstraction, the hiding of details that pertain to the computer which is used to execute the program
rather than to the algorithm expressed by the program. The domain that has most persistently relied on
abstraction is that of input and output operations. This is not surprising, because input and output
inherently involve the activation of devices that are peripheral to the computer, and whose structure,
function, and operation differ strongly among various kinds and brands. Many programming languages
have typically incorporated statements for reading and writing data in sequential form without reference
to specific devices or storage media. Such an abstraction has many advantages, but there exist always
applications where some property of some device is to be utilized that is poorly, if at all, available
through the standard statements of the language. Also, generality is usually costly, and consequently
operations that are conveniently implemented for some devices may be inefficient if applied to other
devices. Hence, there also exists a genuine desire to make transparent the properties of some devices for
applications that require their efficient utilization. Simplification and generalization by suppression of
details is then in direct conflict with the desire for transparency for efficient use.
In Oberon this intrinsic dilemma has been resolved - or rather circumvented - by not including any
statements for input and output at all. This extreme approach was made acceptable because of two
facilities. First, there exists the module structure allowing the construction of a hierarchy of (library)
modules representing various levels of abstraction. Second, Oberon permits the expression of computer
specific operations, such as communication with peripheral interfaces. These operations are typically
contained in modules at the lowest level of this hierarchy, and are therefore counted among the socalled low-level facilities. A program wishing to ignore the details of device handling imports
procedures from the standard modules at the higher levels of this hierarchy. When desiring utmost
efficiency or requiring access to specific properties of specific devices, one either uses low-level
modules, so-called device drivers, or uses the primitives themselves. In the latter case, the programmer
pays the price of intransportability, for his programs refer directly to particulars of either his computer
or its operating system.
It is impossible to present in this context any operations of devices at the low levels of the module
hierarchy, because there exists a wide variety of such devices. We restrict the following material to the
presentation of two typical modules used in performing conventional input and output operations.
Module Texts we have already encountered in examples in preceding chapters. The main concept to be
presented is the file, data considered as sequences of elements of the same type. For this purpose, we
present the definition of module Files.
We must generally distinguish between legible and illegible input and output. Legible input and output
serves to communicate between the computer and its user. Mostly the data elements are characters i.e.
values of type CHAR; the exception is graphical input and output. Legible data are input through
keyboards, scanners, etc. Legible output is generated by displays and printers. Illegible input and output
is made from and to so-called peripheral storage media, such as disks and tapes, but also may originate
from sensors - e.g. in laboratories or drawing offices - and to devices that are controlled by computers,
such as plotters, factory assembly lines, traffic signals, and networks. Data for illegible input and output
can be of any type, and need not be of type CHAR.

51
The vast majority of input and output operations of both the legible and illegible variety is appropriately
considered as sequential. Their data are of a structure that does not exist as a basic data structuring
method in Oberon, such as the array or the record. Sequences have the following characteristics:
1. All elements are of the same type.
2. The number of elements is not known a priori. The sequence is therefore (a simple case of) a dynamic
structure. The number of elements is called the length of the sequence.
3. A sequence can be modified only by appending elements at its end. Appending an element is called
writing.
4. Only a single element of a stream is visible (accessible) at any one time. Accessing this element is
called reading, and the sequence is read by advancing from one element to the next.
In passing we state that the sequence as described above is perhaps the most successful case of data
abstraction encountered. It is certainly more widely used in actual practice than the often quoted
examples of stacks and queues. The language Pascal had included it among its basic data structuring
methods along with arrays, records, and sets.
Before proceeding with the postulation and explanation of modules for handling sequences, which we
will call files and texts, we wish to point out an important separation of function performed for legible
input and output. On the one hand, there is the actual transport of data to and from the computer. This
involves the activation and sensing of the state of the peripheral device, be it a keyboard, a display, or a
printer. On the other hand, there is the function of transforming the representation of data. If, for
example, the value of an expression of type INTEGER is to be transmitted to a display, the computerinternal representation must be transformed into the decimal representation as a sequence of digits. The
display device then translates the character representation (usually consisting of 8 bits for each
character) into a pattern of visible dots or lines. However, the former translation can be considered as
device independent, and is therefore a prime candidate for separation from device specific operations. It
can be performed by the same routines without regard whether the sequence is to be stored on a disk or
to be made visible on a display.
A third class of functions that can well be separated from physical data transfer and from conversion of
data representation, pertains to devices associated with more than a single sequence, the primary
example being a disk store. We refer to the operations of allocating storage space and associating names
with individual texts or files. Considering that texts and files are dynamic structures, storage allocation
is of considerable complexity. The naming of individual files and in particular the management of
directories (in order to quickly locate individual files) is another task requiring an elaborate mechanism.
Both storage allocation and directory management are the tasks of individual service modules. There
seem to exist as many ways to manage these tasks as there exist operating systems. And this is precisely
where diversity transcends the many levels of input and output operations. We therefore strictly adhere
to and confine ourselves to the abstract notion of a sequence.
22.2. Files and Riders
Elements of a sequence cannot be identified by an index, as in the case of arrays, nor by a field name, as
in the case of a record. Elements are instead accessed one by one, advancing strictly sequentially. This
notion itself implies the postulation of an access machanism. In the Oberon module Files, we call it
Rider. A rider is a data structure, an object, which can be placed on a file at a given position, and then
be used to write or read single elements. Hence, module Files defines not only one, but two data types:
File and Rider. The former stands for the data, the latter for the operations performed. It now follows,
that several riders may be positioned on the same file, and be moved independently. Not the file has a
reading or writing position, but each rider. Normally, however, one rider only is used for a file.
DEFINITION Files;
TYPE File;
Rider = RECORD eof: BOOLEAN; res: LONGINT END ;
PROCEDURE Old (name: ARRAY OF CHAR): File;
PROCEDURE New (name: ARRAY OF CHAR): File;
PROCEDURE Register (f: File);
PROCEDURE Close (f: File);
PROCEDURE Purge (f: File);

52
PROCEDURE Length (f: File): LONGINT;
PROCEDURE GetDate (f: File; VAR t, d: LONGINT);
PROCEDURE Set (VAR r: Rider; f: File; pos: LONGINT);
PROCEDURE Pos (VAR r: Rider): LONGINT;
PROCEDURE Base (VAR r: Rider): File;
PROCEDURE Read (VAR r: Rider; VAR ch: CHAR);
PROCEDURE ReadInt (VAR R: Rider; VAR x: INTEGER);
PROCEDURE ReadLInt (VAR R: Rider; VAR x: LONGINT);
PROCEDURE ReadReal (VAR R: Rider; VAR x: REAL);
PROCEDURE ReadLReal (VAR R: Rider; VAR x: LONGREAL);
PROCEDURE ReadString (VAR R: Rider; VAR x: ARRAY OF CHAR);
PROCEDURE Write (VAR r: Rider; ch: CHAR);
PROCEDURE WriteInt (VAR R: Rider; x: INTEGER);
PROCEDURE WriteLInt (VAR R: Rider; x: LONGINT);
PROCEDURE WriteReal (VAR R: Rider; x: REAL);
PROCEDURE WriteLReal (VAR R: Rider; x: LONGREAL);
PROCEDURE WriteString (VAR R: Rider; x: ARRAY OF CHAR);
PROCEDURE Delete (name: ARRAY OF CHAR; VAR res: INTEGER);
PROCEDURE Rename (old, new: ARRAY OF CHAR; VAR res: INTEGER);
END Files.
The procedures are listed in four groups. The first group operates on files, the second and the third on
riders for reading and writing respectively, and the fourth on the file directory only. We assume that
files are stored on a persistent medium, such as a magnetic disk or flash-rom, and that all files are listed
in a directory with names.
In the first group, procedure Old yields the file listed in the directory by the specified name. New
generates a new (empty) file with specified name. Actual registration in the directory is performed by
procedure Register, typically after the entire file had been generated. Close terminates the writing of a
file (flushing buffers), and it is implied in Register. GetDate yields the date and time of a files creation.
In the second group, Set places a rider on a file at a specified position (between 0 and the length of the
file). Pos indicates the riders current position, and Base the file on which it is placed. The Read
procedures read a specified type of data element (consisting of one or several bytes) without any
transformation of representation. They advance the rider by the appropriate amount. A string is assumed
to be terminated by a 0X character.
In the third group, procedures operate similarly for generating (writing) a file. Note that the rider must
not necessarily be positioned at the end of the file, although this is the normal, prevalent case.
We are now in a position to show, how files are typically written and read in Oberon. Let us assume the
declarations
f: Files.File; r: Files.Rider
First, the empty file is created, then a rider is associated with it, then we assume it is generated
sequentially (here in a whyile-loop), and finally (and optionally) it is registered in the directory.
f := Files.New(MyFile); Files.Set(r, f, 0);
WHILE more DO compute(next); Files.Write(r, next) END ;
Files.Register(f)
The file can thereafter be read by the following pattern. First, the file with specified name is associated
with variable f. Then a rider is placed at the start of the file, and then advanced by reading.
f := Files.Old(MyFile); Files.Set(r, f, 0);
WHILE ~r.eof DO Files.Read(f, next); process(next) END
Note that the rider indicates that the end of the file had been reached after the first unsuccessful attempt
of reading.
22.3. Texts, Readers and Writers

53
Texts, basically, are sequences of characters. In the Oberon System, however, texts have some
additional properties. As they can be displayed and printed, they must carry properties which determine
their style and appearance. In particular, a font is specified. This is a style, the patterns by which
characters are represented. Subsequences of charcaters in a text may have their individual fonts.
Furthermore, Oberon texts specify color and vertical offset (allowing for negative offset for
subscripting, and positive offset for superscripting). Texts are typically subjected to complicated editing
operations, which require a flexible, internal data representation. Therefore Texts in Oberon differ
substantiall from Files in many respect. However, in their essence they are also sequences, and the basic
operations strongly resemble those of files.
Our special interest in texts is justified by the fact, that humans communicate with computers mostly
via texts, be they typed on a keyboard or displayed on a screen. As a consequence, reading and writing
of texts usually includes a change of data representation. For example, integers will have to be changed
into sequences of decimal digits on output, and the reading of integers requires the reading of a
sequence of digits and the computation of the represented value. These conversions are included in the
read/write procedures of module Texts, which we have used in many examples in preceding chapters.
The following is its definition, which evidently resembles that of Files. In place of the single type
Rider we find the two types Reader and Writer.
DEFINITION Texts;
IMPORT Files, Fonts;
CONST replace = 0; insert = 1; delete = 2;
Inval = 0; Name = 1; String = 2; Int = 3; Real = 4; LongReal = 5; Char = 6;
TYPE Text = POINTER TO RECORD
len: LONGINT
END ;
Buffer;
Reader = RECORD
eot: BOOLEAN;
fnt: Fonts.Font;
col, voff: SHORTINT
END ;
Scanner = RECORD (Reader)
nextCh: CHAR;
line, class: INTEGER;
i: LONGINT;
x: REAL;
y: LONGREAL;
c: CHAR;
len: SHORTINT;
s: ARRAY 32 OF CHAR;
END ;
Writer = RECORD
buf: Buffer;
fnt: Fonts.Font;
col, voff: SHORTINT;
END ;
PROCEDURE Open(t: Text; name: ARRAY OF CHAR);
PROCEDURE Delete(t: Text; beg, end: LONGINT);
PROCEDURE Insert(t: Text; pos: LONGINT; b: Buffer);
PROCEDURE Append(t: Text; b: Buffer);
PROCEDURE ChangeLooks
(T: Text; beg, end: LONGINT; sel: SET; fnt: Fonts.Font; col, voff: SHORTINT);
PROCEDURE OpenReader (VAR r: Reader; t: Text; pos: LONGINT);
PROCEDURE Read (VAR r: Reader; VAR ch: CHAR);
PROCEDURE Pos (VAR r: Reader): LONGINT;

54
PROCEDURE OpenScanner (VAR S: Scanner; t: Text; pos: LONGINT);
PROCEDURE Scan (VAR S: Scanner);
PROCEDURE OpenWriter (VAR w: Writer);
PROCEDURE SetFont (VAR w: Writer; fnt: Fonts.Font);
PROCEDURE SetColor (VAR w: Writer; col: SHORTINT);
PROCEDURE SetOffset (VAR w: Writer; voff: SHORTINT);
PROCEDURE Write (VAR w: Writer; ch: CHAR);
PROCEDURE WriteLn (VAR w: Writer);
PROCEDURE WriteString (VAR w: Writer; s: ARRAY OF CHAR);
PROCEDURE WriteInt (VAR w: Writer; x, n: LONGINT);
PROCEDURE WriteHex (VAR w: Writer; x: LONGINT);
PROCEDURE WriteReal (VAR w: Writer; x: REAL; n: INTEGER);
PROCEDURE WriteRealFix (VAR w: Writer; x: REAL; n, k: INTEGER);
PROCEDURE WriteLongReal (VAR w: Writer; x: LONGREAL; n: INTEGER);
PROCEDURE WriteDate (VAR w: Writer; t, d: LONGINT);
PROCEDURE Load (VAR r: Files.Rider; t: Text);
PROCEDURE Store (VAR r: Files.Rider; t: Text)
END Texts.
Procedure OpenWriter uses as defaults a standard font, black color, and zero offset.
Simple, sequential reading and writing of a text now follow the patterns of reading and writing a file.
Let us assume the following declarations of variables:
T: Texts.Text;
R: Texts.Reader;
W: Texts.Writer;
NEW(T); Texts.Open(T, MyName); Texts.OpenReader(R, T, 0); Texts.Read(R, ch);
WHILE ~R.eot DO process(ch); Texts.Read(R, ch) END
NEW(T); Texts.Open(T, MyName); Texts.OpenWriter(W); generate(ch)
WHILE more DO Texts.Write(W, ch); generate(ch) END ;
Texts.Append(T, W.buf)
Note that the Oberon paradigm is to write a text, or a piece of text, first into a buffer, and thereafter
insert or append it to a text. This is done for reasons of efficiency, because the possibly needed
rendering of the text, for example on a display, can be done once upon insertion of the buffered piece of
text rather than after generating each character.
Very often one wants to read a text consisting of a sequence of items which are not all of the same type,
such as numbers, strings, names, etc. When using procedures reading items of a fixed type, the
programmer must know the exact sequence in which the various items will appear in the text. Typically
one does not know, and even if a specific order is specified, mistakes may occur. So one should like to
have available a reading mechanism that reads items one at a time, but lets the program determine what
type of item was read, and take further steps accordingly. This facility is provided in the Oberon text
module by the machanism called scanning. In place of a Reader we use a Scanner. Procedure Scan(S)
then reads the next item. Its kind can be determined by inspecting the field S.class, and its value
accordingly is given by S.i in the case an integer was read, S.x if a real number was read, S.s if a name
or a string was read. This scheme has proven to be most useful because of its flexibility. It defines the
following syntax for texts to be scanned. Blanks and line ends serve to separate consecutive items.
item
name
integer
hexdig
sign
real
longreal

=
=
=
=
=
=
=

name | integer | real | longreal | string | SpecialChar.

letter {letter | digit}.
[sign] digit {digit} | digit {digit | hexdig} H.
A | B | C | D | E | F.
[-].
[sign] digit {digit} . digit {digit} [E [sign] digit {digit}].
[sign] digit {digit} . digit {digit} D [sign] digit {digit}.

55
A string is any sequence of any characters (except quotes) enclosed in quotes. Special characters are are
all characters except letters, digits, and the quote mark.
22.4. Standard Input and Output
In order to simplify the description of input and output of texts in simple cases, the Oberon system
introduces some conventions and global variables. We call this source of input and sink of output
standard input and output.
Standard output sink is the text Log defined as global variable in module Oberon. Given a global writer
W, the text, previously written by Write procedures into the writers buffer (W.buf) is sent to the
displayed Log text by the statement
Texts.Append(Oberon.Log, W.buf)
The standard input is often assumed to be the text following the command which invoked execution of
a given procedure. This input text is identified by the variable Par in module Oberon. It is considered
as parameter of the invoked command, and it is typically read by the scanning mechanism described
above. The necessary statements are:
Texts.OpenScanner(S, Oberon.Par.text, Oberon.Par.pos); Texts.Scan(S)
The first item thus read may be the first item of the desired text, or it may be the name of the text file to
be scanned, or something else, according to the specification of the individual command. The following
conventions have established themselves for the designation of input texts. Assume that the command
(procedure) name is followed by
- an identifier (possibly qualified). Then this is the name of the input text (input file),
- an asterisk (*). Then the marked viewer (window) contains the input text,
- an @ symbol. Then the most recent text selection is the beginning of the input text,
- an arrow (^). Then the most recent selection denotes the file name of the input text.
These conventions are expressed by the following function procedure yielding the designated input text:
PROCEDURE This*(): Texts.Text;
VAR beg, end, time: LONGINT;
S: Texts.Scanner; T: Texts.Text; v: Viewers.Viewer;
BEGIN Texts.OpenScanner(S, Oberon.Par.text, Oberon.Par.pos); Texts.Scan(S);
IF S.class = Texts.Char THEN
IF S.c = "*" THEN (*input in marked viewer*)
v := Oberon.MarkedViewer();
IF (v.dsc # NIL) & (v.dsc.next IS TextFrames.Frame) THEN
T := v.dsc.next(TextFrames.Frame).text; beg := 0
END
ELSIF S.c = "@" THEN (*input starts at selection*)
Oberon.GetSelection(T, beg, end, time);
IF time < 0 THEN T := NIL END
ELSIF S.c = ^ THEN (*selection is the file name*)
Oberon.GetSelection(T, beg, end, time);
IF time >= 0 THEN (*there is a selection*)
Texts.OpenScanner(S, Oberon.Par.text, Oberon.Par.pos); Texts.Scan(S);
IF S.class = Texts.Name THEN (*input is named file*)
NEW(T); Texts.Open(T, S.s); pos := 0
END
END
END
ELSIF S.class = Texts.Name THEN (*input is named file*)
NEW(T); Texts.Open(T, S.s); pos := 0
END ;
RETURN T
END This;

56
Let us assume that this function procedure is defined in module Oberon, together with the global
variable pos, indicating the position in the text where input is to start. The following program for
copying a text may serve as a pattern for selecting the input according to the conventions postulated
above.
MODULE ProgramPattern;
IMPORT Texts, Oberon;
VAR W: Texts.Writer; (*global writer*)
PROCEDURE copy(VAR R: Texts.Reader);
VAR ch: CHAR;
BEGIN Texts.Read(R, ch);
WHILE ~R.eot DO Texts.Write(W, ch); Texts.Read(R, ch) END
END copy
PROCEDURE SomeCommand*;
VAR R: Texts.Reader; (*local reader*)
BEGIN Texts.OpenReader(R, Oberon.This.text, Oberon.Par.pos);
copy(R, W); Texts.Append(Oberon.Log, W.buf)
END SomeCommand;
BEGIN Texts.OpenWriter(W)
END ProgramPattern.
Module Oberon can be considered as the core of the Oberon system housing a small number of global
resources. These include the previously encountered output Log, and the record Par, specifying the
viewer, frame, and text in which the activated command lies, and hence providing access to its input
parameters. Here we show only an excerpt of the definition of module Oberon:
TYPE ParList = POINTER TO RECORD
vwr: Viewers.Viewer;
frame: Display.Frame;
text: Texts.Text;
pos: LONGINT;
END ;
VAR Log: Texts.Text;
Par: ParList;
PROCEDURE Time (): LONGINT;
PROCEDURE AllocateUserViewer (DX: INTEGER; VAR X, Y: INTEGER);
PROCEDURE AllocateSystemViewer (DX: INTEGER; VAR X, Y: INTEGER);
(*provide coordinates X and Y for a new viewer to be allocated*)
PROCEDURE GetSelection (VAR text: Texts.Text; VAR beg, end, time: LONGINT);
To conclude this introduction to Oberon conventions about input and output, we show how a new
viewer (window) is opened, and how the output text is directed into this text viewer. Optimal
positioning of the new viewer is achieved by procedure Oberon.AllocateViewer, with the specified set
of standard, frequently needed commands in the title bar.
PROCEDURE OpenViewer(T: Texts.Text);
VAR V: MenuViewers.Viewer; X, Y: INTEGER;
BEGIN Oberon.AllocateUserViewer(Oberon.Mouse.X, X, Y);
V := MenuViewers.New(TextFrames.NewMenu(
"Name", "System.Close System.Copy System.Grow Edit.Search Edit.Store"),
TextFrames.NewText(T, 0), TextFrames.menuH, X, Y)
END OpenViewer;

57

Part 4
23. Object-oriented Programming
23.1. The origins of object-oriented programming
The term object-oriented programming (OOP) originated around 1975. It expressed a shift in the
paradigm of programming. In the conventional, procedural programming style the algorithm stands in
the foreground of the programmers concerns, the algorithm operating on a set of data. In the objectoriented view the data stand in the foreground, the data on which algorithms are applied, and which are
grouped into what is now called an object.
Programming is a notoriously difficult task, and the rapidly increasing power of modern computers made
the situation worse. The tasks to be modeled and solved became more and more complicated. Their
complexity often surpassed the capabilities of the human mind. In this situation, it is not surprising that
the demand for better tools, for a new discipline, even for panaceas became urgent and ardent. The new
paradigm of object-oriented programming was therefore welcomed, as their proponents raised hopes for
an easier approach to complex problems. Partly, these hopes were exaggerated, partly the essence of
OOP was driven to unwise extremes for the sake of unity of doctrine. For example, even simple integers
were to be regarded as objects in the new sense. What was this new sense?
Before we try to answer this question, we should point out that the notion, but not the term OOP was
introduced at least 10 years earlier. In 1965 the language Simula was introduced as a variant of Algol
adapted to and extended for the needs of simulation of systems governed by discrete events. (What later
became objects were then called elements). A well-known example of such a system was a supermarket.
Here customers, merchandise, cashiers, personnel, shelves, carts, were all modeled as objects, and they
formed classes. For modeling such systems, this was the obvious approach. Only ten years later, under
the leadership of Smalltalk, was the new viewpoint extended to other applications. Of foremost interest
were operating systems, where resource pools, processors, windows, texts and devices became regarded
as autonomous agents. Graphics editors were another suitable application. They are used to handle
various figures such as lines, circles, rectangles, ellipses, text captions, each forming their own class.
In these examples, the objects constitute almost autonomous entities, in the simulated models as well as
the operating systems. They not only feature attributes and properties common to all members of their
class, but also a common behavior represented by associated procedures. These are typically activated by
a higher power such as a scheduler in the case of simulation, or a human being in the case of an
operating system or a graphics editor. Both data and procedures together constitute and characterize
objects.
The designers of Smalltalk clearly wished to present not only a new language, but also a new paradigm,
a new approach to programming. To be effective and convincing in this endeavor, they provided also a
new terminology to underscore a different quality of programming. In this effort, record-structured
variables became objects, their associated procedures became methods, a data type became a class, and
calling a procedure is now termed sending a message to an object. This is denoted as
object.method (message)
In Simula, the ancestor of OOP, there was not a set of methods to be invoked by messages, but rather a
single coroutine representing the algorithmic behavior of the simulated object, which was regarded as
continuously alive. We may now consider the parts between breakpoints as methods, although the
analogy is somewhat lacking. For example, a customer of the supermarket would be characterized by
first entering the scene, picking up a cart, wandering along the shelves, picking up merchandise, passing
a cashier, and finally returning the cart and leaving the scene. Objects (customers) would emerge (be
allocated), proceed, and leave (be deallocated). Hence, typically a dynamic data structure, a linked list, is
used to represent the set of currently involved objects.
It now seems that OOP merely provides (or dictates?) a different view of familiar programming
concepts, but does not require any additional features in a language, except perhaps some notational
conventions. The same old contents in a new form? Subsequently, we will investigate this question.
23.2. Type extensions and inhomogeneous data structures

58
Let us consider a simple, rudimentary line drawing editor. It is typical for many applications in so far as
it operates on a set of objects. In this case, the objects are straight lines, rectangles, circles, ellipses, text
captions, and others. These objects are typically represented as a linked list, to which it is easy to add
and delete new instances.
However, using a statically typed language, we immediately encounter a difficulty. Declaring each of the
figures by its own record type, we are prevented from forming a single list. Declaring a separate list for
each type of figure appears as artificial and cumbersome. On the other hand, integrating all figure types
into a single declaration appears as equally contorted and cumbersome: First, a discrimination field is
necessary, subsequently giving rise to various IF statements. Second, a single declaration will contain
record fields that pertain to only a few, but not to all figure types. Whatever one chooses to do, the
solution remains unsatisfactory.
We therefore introduce a new feature in the language Oberon. It is called type extension. Basically, in
our example, we declare a base type called Figure, and then a derived type for each individual kind of
figure, here called Line, Rectangle, Circle, etc. The key is that each instance of these subtypes is also
considered as of its base type Figure. Such declarations of derived types have the form:
RecordType = RECORD ( TypeIdentifier ) FieldListSequence "END".
The type identifier specifies the base type of the new, derived record type, which thereby becomes a
subtype. Evidently, it now becomes possible to define entire hierarchies of types. Consider our example:
TYPE Figure = POINTER TO FigureDesc;
FigureDesc = RECORD (*this is the base type*)
x, y, w, h: INTEGER; (*coordinates, width and height of object*)
next: Figure (*next in list of figures*)
END ;
RectangleDesc = RECORD (FigureDesc) lw: INTEGER END ;
CircleDesc = RECORD (FigureDesc) radius: INTEGER END ;
EllipseDesc = RECORD (FigureDesc) a, b: INTEGER END ;
CaptionDesc = RECORD (FigureDesc) cap: ARRAY 32 OF INTEGER END ;
LineDesc = RECORD (FigureDesc) END
Accordingly, a descriptor of a rectangle consists of the fields x, y, w, h, and lw (line width), the one for a
circle of the fields x, y, w, h, radius, the one for an ellipse of x, y, w, h, a, b, and the one for a text
caption of x, y, w, h, cap. The derived types retain all fields of the base type. In object-oriented
terminology, they are said to inherit the properties of the base. We will not adopt this anthropomorphic
term.
We call the derived types extended types, because they extend the base type with additional properties.
In object-oriented terminology, the base type is called an abstract type, on which the concrete types, the
extended types, rest. All of them are compatible also with their base type. Therefore, it is possible to link
them into a single list through the base types field next, and therefore to build data structures that
contain elements of different types, that is, inhomogeneous structures. For many applications, this is an
important and necessary requirement.
When, for example, traversing such an inhomogeneous list, it may be necessary to determine the type of
each encountered object. It is only known that every element in the list is of type Figure, but now we
wish to determine the subtype of an individual element. This is possible through the new facility of the
type test, which is classified as a Boolean expression, and has the form (syntax)
expression = variable IS identifier.
Assume a variable p of type Figure, then the traversal of a list and the discriminated application of a
drawing procedure can be expressed as follows:
WHILE p # NIL DO
IF p IS Line THEN DrawLine(p)
ELSIF p IS Circle THEN DrawCircle(p)

59
END ;
p := p.next
END
We will discover in the next section that there is actually a better, shorter way of expressing this action.
The type test is a necessary feature, because Oberon has deviated from the dogma of strictly static data
typing. Note, however, that the actual type T of a variable declared to be of base type T0 can only be a
subtype of T0. A Figure can be a Line, Circle, etc., but not, for example, an integer. Type tests are
therefore not needed for most data accesses. Furthermore, type tests can be implemented very efficiently.
In addition to the type test, Oberon also offers the construct of type guard, which has the form of a
selector in variable designators.
designator = qualident {"." ident | "[" ExpList "]" | "(" qualident ")" | "^" }.
Consider the example of the designator
p(Circle).radius
The simple designator p.radius would not be acceptable, because p is of type Figure, which does not
feature a field radius. With the type guard, the programmer can ascertain that in this case p is also of
type Circle, in which case the field radius is indeed applicable. Whereas p is of base type Figure,
p(Circle) is of type Circle. Of course, it must be assumed that the programmer has made sure that his
claim is correct, and the system will have to check whether this is indeed so (at run time). The type guard
in a sense resembles an array index, where a system must check whether the actual index lies within the
specified array bounds.
23.3. Methods
It has already been remarked that objects are characterized not only by their attributes (data), but also by
their behavior, by their procedures. Just as attributes may differ among various subtypes, so may their
behavior. Therefore, it is sensible to associate the procedures directly with individual objects or at least
their subtypes. In strongly object-oriented languages this is achieved by letting class declarations specify
associated procedure declarations. In Oberon, we use what is available and refrain from introducing new
features. We simply add to a records data fields other fields that represent procedures, that is, are of
procedure types. We redefine our example of the type Figure accordingly. Let us assume the two
operations of drawing and marking figures.
FigureDesc = RECORD
x, y, w, h: INTEGER;
draw: PROCEDURE (f: Figure);
mark: PROCEDURE (f: Figure; on: BOOLEAN);
next: Figure (*next in list of figures*)
END ;
Whereas in strictly object-oriented languages methods are automatically associated with every object
upon its creation, in Oberon this must be achieved by explicit assignments. Such initial assignments are
called installations (of procedures). For example:
VAR c: Circle;
NEW(c); c.x := .. ; c.draw := DrawCircle; c.mark := MarkCircle;
The call of a method, i.e. of an object-bound procedure, is expressed as in the following example:
c.mark(c, TRUE)
The association of type-specific procedures to every object has the substantial advantage that subclassdiscriminating conditional statements can be avoided. For example, the above call automatically
activates the appropriate MarkCircle procedure without having to execute a sequence of if statements
distinguishing between lines, circles, rectangles etc. This contributes to efficiency. The drawing of all
objects in a list is now expressed quite simply as
WHILE p # NIL DO p.draw(p); p := p.next END
and in each call automatically the appropriate drawing procedure is activated.

60
23.4. Handlers and Messages
In Part 3 of this text, the importance of proper decomposition (modularization) of systems was
emphasized. A module restricts the visibility of declarations and incorporates the concept of information
hiding. In object-oriented languages a type definition, i.e. a class declaration, assumes the role of a
module, also in the sense of unit of separate compilation. In Oberon, we prefer to keep the constructs
expressing information hiding, and those for type and procedure definition separated and independent.
Hence, every module may contain one or several type definitions, and perhaps none at all.
Nevertheless, considering a data type and its operators as a logical unit, and to hide details of
implementation, are old and proven techniques manifest in the concept of the abstract data type.
Particularly for complex objects it is highly desirable that the various subtypes can be defined and
implemented by separate teams of programmers, and this perhaps years after the definition of the basis
was completed and had been published or distributed. It is desirable to be able to define every derived
type in its own module. This demand also holds for base types, with operators applying to the set of
objects (in contrast to individual objects) defined in the base module, as well as for those defining the
derived type in client modules. Among such operations we mention the so-called broadcast, the traversal
of the objects data structure (list, tree) with application of a method to every element. Declarations of
such broadcasts are thus inherently confined to the base module.
Now it may happen that a subtype introduces its specific operations, including some that are not shared
by other subtypes. Obviously, it is desirable that also these operations can be broadcast with the
respective procedure defined in the base module, even if the new operators were introduced years after
the base had been established. The base, however, cannot be changed nor accessed, because in the
meantime it might have been distributed to many customers. How can this dilemma be resolved?
The solution lies in replacing the set of methods by a single procedure, which discriminates between the
various methods. Such a single procedure we call a handler. It has only two parameters: The object to
which it is applied, and an identification of the operation with its actual parameters. The second
parameter has the form of a record, and we call it a message. For such a message type to be capable of
denoting any operation, even those to be introduced in the future, the same feature is used as for adding
new types derived from old ones: Type extension. Let us explain this solution by our example of the
graphics editor.
First, the declaration of the base type now becomes
FigureDesc = RECORD
x, y, w, h: INTEGER; (*coordinates, width and height of object*)
handle: Handler;
next: Figure (*next in list of figures*)
END ;
with the procedure type Handler being defined as
Handler = PROCEDURE (f: Figure; VAR msg: Message)
and the new type Message by the (extensible) null record
Message = RECORD END
If certain operations applying to all objects are already known at the time of defining the base type
(which is usually the case), the respective message (sub)types are directly declared, as for example:
DrawMsg = RECORD (Message) END ;
MarkMsg = RECORD (Message) on: BOOLEAN END ;
MoveMsg = RECORD (Message) dx, dy: INTEGER END ;
As an aside, we note that the record field next need not be exported from the base module, thus keeping
the structure of the set of objects (list, tree) and its operations, such as traversal, hidden and flexible. A
broadcast can now be specified as follows:
PROCEDURE Broadcast(VAR msg: Message);
VAR f: Figure;
BEGIN f := root; (*root is a global variable in the base module*)
WHILE f # NIL DO f.handle(f, msg); f := f.next END

61
END Broadcast
Whenever a concrete figure type is introduced, this will typically be done in a new client module
importing the base module Figures. In addition to the new subtype, for example Rectangle, its associated
handler is declared following the pattern shown below. This handler is installed in the field handle of
every object of type Rectangle.
PROCEDURE Handle(f: Figure; VAR msg: Message);
VAR r: Rectangle;
BEGIN r := f(Rectangle);
IF msg IS DrawMsg THEN (*draw rectangle r*)
ELSIF msg IS MarkMsg THEN MarkRectangle(r, msg(MarkMsg).on)
ELSIF msg IS MoveMsg THEN
INC(r.x, msg(MoveMsg).x); INC(r.y, msg(MoveMsg).y)
ELSIF
END
END Handle
The statement for moving a single object f by displacements dX and dY is now somewhat complicated:
VAR msg: MoveMsg;
msg.dx := dX; msg.dy := dY; f.handle(f, msg)
However, we must keep in mind that mostly messages are sent to objects indirectly, that is, when the
recipient is not known by the sender. A good example is the broadcast:
msg.dx := dX; msg.dy := dY; Broadcast(msg)
The particular flexibility of this technique using handlers and messages sometimes identified as
polymorphism as well as its extensibility stems from the fact that unknown messages are simply
ignored. They fall through the cascade of if elsif statements in the handlers of objects to which the
message does not apply.
Let us, for example, add a new module, say Blinkers, to our editor system. It contains the declaration of
the derived type Blinker (BlinkerDesc). Let it require a new method, causing the object to blink in a
certain rhythm. We represent it by the (derived) message BlinkMsg. Then the call of the handler of any
other object in the global object list, through which a broadcast will propagate, will have no effect,
because the case
ELSIF msg IS BlinkMsg THEN
does not exist in the handler. The technique using handlers guarantees optimal, unlimited extensibility of
existing systems by merely adding new modules on top of an existing hierarchy.
We summarise the technique as follows:
1. An object is represented by a pointer to a record. The record features a single, procedural field called
handle. The procedure assigned to handle is called a handler. It features two parameters. The first
designates the object to which the handler is applied. The second identifies the operation to be selected.
It is a VAR parameter with record structure, and it is called a message.
2. The handler defines the behavior of an object. It is called with a message which is typically of an
extension of type Message specific to the object type on hand.
3. The message parameter specifies the action to be taken by its (sub)type. Its record fields constitute the
parameters of the action. Their number and types are specific to the particular message type and action.
4. The handler consists of a single if elsif cascade. Type tests discriminate between the various
message subtypes and thereby actions.
5. Assigning the handler to an object is called installation.
6. An action is initiated by first setting up the message and then sending the message to the object.
7. Messages may be broadcast to all objects of a heterogeneous data structure without knowledge of the
nature of the structure. Inapplicable messages will simply be ignored.

62
This concludes our brief introduction to the object-oriented paradigm of programming. We realize that
almost no language features had to be added to Oberon to support it. Apart from the already present
facilities of records and of procedural types, only the notion of type extension is both necessary and
crucial. It allows to construct hierarchies of types and to build inhomogeneous data structures. As a
consequence of abandoning the rule of strictly static typing, the introduction of dynamic type tests
became necessary. The further facility of the type guard is merely one of convenience.
The procedural style of programming, which is most appropriate in many applications, and the objectoriented style can co-exist in Oberon quite happily. If one chooses to program in the conventional,
procedural style, one can ignore the object-oriented facilities, and they do not get in ones way.

63

Appendix
1. The Syntax of Oberon
ident = letter {letter | digit}.
number = integer | real.
integer = digit {digit} | digit {hexDigit} "H" .
real = digit {digit} "." {digit} [ScaleFactor].
ScaleFactor = ("E" | "D") ["+" | "-"] digit {digit}.
hexDigit = digit | "A" | "B" | "C" | "D" | "E" | "F".
digit = "0" | "1" | "2" | "3" | "4" | "5" | "6" | "7" | "8" | "9".
CharConstant = """ character """ | digit {hexDigit} "X".
string = """ {character} """ .
qualident = [ident "."] ident.
identdef = ident ["*"].
TypeDeclaration = identdef "=" type.
type = qualident | ArrayType | RecordType | PointerType | ProcedureType.
ArrayType = ARRAY length {"," length} OF type.
length = ConstExpression.
RecordType = RECORD ["(" BaseType ")"] FieldListSequence END.
BaseType = qualident.
FieldListSequence = FieldList {";" FieldList}.
FieldList = [IdentList ":" type].
IdentList = identdef {"," identdef}.
PointerType = POINTER TO type.
ProcedureType = PROCEDURE [FormalParameters].
VariableDeclaration = IdentList ":" type.
designator = qualident {"." ident | "[" ExpList "]" | "(" qualident ")" | "^" }.
ExpList = expression {"," expression}.
expression = SimpleExpression [relation SimpleExpression].
relation = "=" | "#" | "<" | "<=" | ">" | ">=" | IN | IS.
SimpleExpression = ["+"|"-"] term {AddOperator term}.
AddOperator = "+" | "-" | OR .
term = factor {MulOperator factor}.
MulOperator = "*" | "/" | DIV | MOD | "&" .
factor = number | CharConstant | string | NIL | set |
designator [ActualParameters] | "(" expression ")" | "~" factor.
set = "{" [element {"," element}] "}".
element = expression [".." expression].
ActualParameters = "(" [ExpList] ")" .
statement = [assignment | ProcedureCall |
IfStatement | CaseStatement | WhileStatement | RepeatStatement |
LoopStatement | WithStatement | EXIT | RETURN [expression] ].
assignment = designator ":=" expression.
ProcedureCall = designator [ActualParameters].
IfStatement = IF expression THEN StatementSequence
{ELSIF expression THEN StatementSequence}
[ELSE StatementSequence]
END.
CaseStatement = CASE expression OF case {"|" case} [ELSE StatementSequence] END.
Case = [CaseLabelList ":" StatementSequence].
CaseLabelList = CaseLabels {"," CaseLabels}.
CaseLabels = ConstExpression [".." ConstExpression].
WhileStatement = WHILE expression DO StatementSequence END.
LoopStatement = LOOP StatementSequence END.
WithStatement = WITH qualident ":" qualident DO StatementSequence END .

64
ProcedureDeclaration = ProcedureHeading ";" ProcedureBody ident.
ProcedureHeading = PROCEDURE identdef [FormalParameters].
ProcedureBody = DeclarationSequence [BEGIN StatementSequence] END.
ForwardDeclaration = PROCEDURE "^" identdef [FormalParameters].
FormalParameters = "(" [FPSection {";" FPSection}] ")" [":" qualident].
FPSection = [VAR] ident {"," ident} ":" FormalType.
FormalType = {ARRAY OF} qualident.
DeclarationSequence = {CONST {ConstantDeclaration ";"} |
TYPE {TypeDeclaration ";"} | VAR {VariableDeclaration ";"}}
{ProcedureDeclaration ";" | ForwardDeclaration ";"}.
Module = MODULE ident ";" [ImportList] DeclarationSequence
[BEGIN StatementSequence] END ident "." .
ImportList = IMPORT import {"," import} ";" .
Import = ident [":=" ident].

2. Symbols and Keywords

+
*
/
~
&
.
,
;
|
(
[
{

:=
^
=
#
<
>
<=
>=
..
:
)
]
}

ARRAY
BEGIN
CASE
CONST
DIV
DO
ELSE
ELSIF
END
EXIT
IF
IMPORT
IN

IS
LOOP
MOD
MODULE
NIL
OF
OR
POINTER
PROCEDURE
RECORD
REPEAT
RETURN
THEN

TO
TYPE
UNTIL
VAR
WHILE
WITH

3. Standard Data Types

CHAR, BOOLEAN, SHORTINT
INTEGER
LONGINT, REAL, SET
LONGREAL

(8 bits)
(16 or 32 bits)
(32 bits)
(64 bits)

4. Standard Functions and Procedures

Name

Argument type

Result type

ABS(x)
ODD(x)
CAP(x)
ASH(x, n)
LEN(v, n)
LEN(v)
ORD(x)
CHR(x)
SHORT(x)

numeric
integer type
CHAR
integer types
v: array type
v: array type
CHAR
INTEGER
LONGINT
INTEGER
LONGREAL
SHORTINT

type of x
BOOLEAN
CHAR
LONGINT
LONGINT
LONGINT
INTEGER
CHAR
INTEGER
SHORTINT
REAL
INTEGER

LONG(x)

absolute value
x MOD 2 = 1
corresponding capital letter
x 2n (arithmetic shift)
length of v in dimension n
length of v in dimension 0
ordinal number of character x
character with ordinal number x
identity (truncation possible!)

identity

65

ENTIER(x)
INC(v, n)
INC(v)
DEC(v, n)
DEC(v)
INCL(v, n)
EXCL(v, n)
COPY(x, v)
NEW(v)
HALT(x)

INTEGER
LONGINT
REAL
LONGREAL
real type
LONGINT
integer types
integer types
integer types
integer types
v: SET; n: integer type
v: SET; n: integer type
x: character arry, string
pointer type
integer constant

largest integer not graeter x

v := v + n
v := v + 1
v := v n
v := v -1
v := v + {n}
v := v - {n}
v := x
allocate v^
terminate computation