############### ############ ######   ######
                 ##           ## ##        ##  ##  ## ##  ##
                 ######   ###### #######   ##   ##  ###  ##
                     ##   ##        ##    ##     ##     ##
                     ##   ##     #######   ##   ##  ###  ##
                     ##   ##     ##        ##  ##  ## ##  ##
                     #######     ############ ######   ######

                       A  MINIMAL  PROCEDURAL  LANGUAGE

                                  RELEASE 7

                            Nils M Holm, 1996-2019

        1. Introduction

        T3X is a small, portable, procedural, block-structured,
        recursive, almost typeless, and to some degree object oriented
        language. Its syntax is derived from Pascal and BCPL and its
        object model is based solely on classes, objects, messages, and
        full encapsulation. It is a tool for creating reusable packages
        rather than data types. The structured approach to programming
        is well-understood, provides a sufficient degree of abstraction,
        and can easily be translated into native machine code at the
        same time. The object model eases the development of general and
        reusable code. T3X is an imperative language. This means that
        a program consists of a set of instructions which tell the
        computer in what way to manipulate the data defined by the
        program. An instruction is also called a statement. In
        structured programming languages, there are four fundamental
        ways of formulating statements:

        - Assignments
        - Sequences
        - Branches
        - Iterations

        The assignment is a fundamental property of imperative
        languages. It is used to move data from one location to another
        by assigning values to variables. In a sequence - which is
        basically a list of statements - the statements are processed
        from the top towards the bottom of the list.  Each statement is
        guaranteed to be completely processed before the next one is
        interpreted. A branch is a statement which is executed only if
        an associated condition applies. Iteration is the repetition of
        a statement depending on a condition. In a block-structured
        language, statements may be grouped in statement blocks or
        compound statements. Each block may have its own local data
        which cannot be affected by statements contained in other

        An additional layer of abstraction is added to an imperative,
        block-structured language by providing user-defined procedures
        or functions (in this document, these terms will be used
        synonymously). A procedure is a statement or a set of statements
        which is bound to a symbolic name. A procedure can be executed
        by coding a call to that procedure. Most languages provide a
        mechanism to transport data to a procedure and return a value to
        the calling program. Some languages (like BCPL and Pascal) make
        a distinction between procedures and functions, others (like C)
        do not. In languages which make a distinction between procedures
        and functions, only functions may return values. In T3X, all
        procedures return values, but the caller is free to ignore them.
        Therefore, procedures and functions are basically the same.

        Another level of abstraction is provided by adding an object
        model to the language. The object model of T3X consists solely

        - Classes
        - Objects
        - Messages

        Classes are used to encapsulate code and data of a program. A
        class may contain any number of data objects and procedures.
        Only public procedures (so-called methods) may be called by
        procedures (or methods) outside of the class. Objects are used
        to instantiate classes. Each instance of a class has its own
        private data area. Hence the same class may be used as a
        template for the creation of multiple independent objects.
        Messages are used to activate methods of specific objects. T3X
        does not provide inheritance or different levels of protection
        (like public data or 'friend' relationships). All class-level
        data objects are fully encapsulated in objects and only
        accessible via methods.

        T3X is an almost typeless language. There exist two different
        types, so-called atomic variables, which may hold small data
        objects, like characters, numbers and references to other data
        objects, and vectors, which are used to store logically
        connected groups of small data objects. In addition, there are
        constants, templates for defining structured data objects and
        classes, and different types of procedure declarations. The T3X
        compiler does not allow some combinations of operators which do
        not make sense (like assigning a value to a procedure or sending
        a message to a vector). Consequently, T3X's type checking is
        much more strict than, for example, BCPL's, but much less
        restrictive than Pascal's or even C's. Weakly typed and typeless
        languages have been exposed to a lot of criticism in the past,
        because they are considered 'insecure', but the degree of
        simplicity and flexibility which is bought by 'sacrificing' this
        bit of security is immense.

        The type checking mechanisms of the T3X language are limited to
        the detection of

        - wrong argument counts in procedure calls
        - assignments to non-variables
        - calls of non-procedures
        - instantiations of non-classes
        - sending messages to non-objects
        - sending non-messages to objects
        - dependencies on non-classes

        BTW: during the development of an early version of T3X, a severe
        error occurred in the compiler. After tracking it down, it
        turned out that was limited to the (type-safe) ANSI C version of
        the translator and did not affect the T3X version. Of course,
        this was coincidence, but to some degree it contradicts the
        proposition that typeless languages are per less insecure than
        typed languages.

        1.1 The History of T3X

        The first version of T3X was created in the middle of the
        1990's. Its primary design goals were

        1. Simplicity and straight-forward syntax and semantics
        2. Very high portability
        3. A small (64KB text + 64KB data) and simple implementation,
           preferably in written in itself
        4. Suitable for both interpretation and native code generation

        One might think that there must have been quite a few languages
        providing these features, but obviously my search did not lead
        to any satisfactory result, or T3X would not have been invented.
        The language which came closest to these requirements was BCPL.
        The typeless approach, which has been very consistently
        implemented in this language, leads to clear, simple, and
        flexible semantics. The language is portable, its implementation
        is small and can easily be done in BCPL itself. The compiler
        provided by Martin Richards, the inventor of BCPL, generates
        code which is aimed at interpretation (for the purpose of
        porting the compiler), but may be translated into native code as

        Unfortunately, the syntax of BCPL is rather hard to to parse by
        a recursive descent parser (RD) and some precedence rules are
        chosen in a way that the creator of T3X found hard to grasp.  An
        RD parser had always been a prerequisite for the language,
        because they are very easy to implement. Of course, Richard's
        BCPL compiler is small, elegant and easy to understand, even
        though it does use syntax trees and a bottom up parsing
        technique. However, RD parsers are still simpler.

        Even if BCPL did not match the requirements exactly, it came
        pretty close, and studying the language and compiler sources has
        influenced the design of T3X a lot. Without BCPL, T3X would not
        be the language it is today.

        The most important thing when designing a programming language
        might be to define its main purpose. The design goal of T3X was
        to create a portable, simple, and easy to understand notation
        for the description of algorithms. T3X was never aimed at
        industrial software development. Its purpose is to support the
        programmer in the process of reasoning about problems. It should
        be a productivity tool in the sense that it provides a
        playground for new ideas and allows the creator of these ideas
        to share it with others using a formal notation. Such a
        notation, of course, has to be clear, simple, easy to learn, and
        it would be a great advantage, if a compiler for this notation
        would be available in many different environments.

        Naturally, my interpretation of productivity is not exactly the
        same as in the rather profit-oriented 'real world' and the
        design of the T3X language reflects this intention well. T3X is
        not suitable for writing large scale application programs.
        Originally, it was is more a notation than a programming
        language. Because it is simple and straight-forward, it does not
        force its user to pay too much attention to the language itself.
        Instead, it provides some very basic building-stones which may
        be used to construct a formal solution for a given problem.

        So T3X provides only a very basic set of building-stones, but it
        turns out that this set is suitable to solve a variety of
        different problems in a convenient way - including, for example,
        the creation of a compiler and runtime environment for the T3X
        language itself.

        2. A Tour through the T3X Language

        T3X is an almost typeless, block-structured, procedural, object
        oriented programming language. Programs, classes, procedures,
        statements, and expressions form a hierarchy: Programs consist
        of classes, procedures, and statements, classes contain
        procedures and statements, procedures usually contain
        statements, and statements mostly contain expressions. Variables
        may be atoms (ordinal) or vectors (one-dimensional arrays).
        Since there are no different types, composed data types - called
        structures - are basically equal to vectors. Constants may be
        used to represent frequently used or tunable values.

        This chapter is written in bottom-up order, so that the building
        stones of larger entities already have been explained when the
        entities themselves are discussed.

        2.1 The Input Alphabet

        The T3X compiler expects its input in the form of an ASCII file
        (a sequence of octets where the least significant seven bits of
        each octet contain the ASCII code of one character and the high
        bit is set to zero). The following characters will be treated as
        white space (The C-style 0x-notation is used to represent
        hexa-decimal numbers):

        - blank (0x20)
        - horizontal tab (0x09)
        - line feed (0x0A)
        - carriage return (0x0D)
        - form feed (0x0C)

        White space characters delimit tokens, but will otherwise be
        ignored by the compiler.

        Valid input characters are the upper and lower case alphabetic
        characters A-Z, a-z, the decimal digits 0-9, and the following
        special characters:

        " # % & ( ) * + , - . / : ; < = > @ [ \ ] ^ _ | ~

        Characters which are not contained in this alphabet may only
        occur in string literals, character literals, and comments.
        Otherwise they will cause an error during program compilation.


        A comment may be introduced at almost any point in a T3X program
        using an exclamation point (!). It extends up to, but not
        including, the end of the current line. Therefore, a comment is
        treated the same way as a single white space character, and

        WH! this is a comment
        ILE(1) ;

        is equal to

        WH ILE(1) ;

        and not to

        WHILE(1) ;

        Therefore, comments may not occur inside of a single token, but
        only between two tokens. This is particularly valid for string
        literals and character literals which are single tokens as well.
        A ! character inside one of these literals is treated as an
        ordinary character.

        2.3 Naming Conventions

        Symbolic names may contain alphabetic characters, the underscore
        character (_), and decimal digits, where the first character
        must be alphabetic or an underscore. Upper case characters will
        be folded to lower case. Therefore, the names

        abc abC aBc aBC Abc AbC ABc ABC

        would all refer to the same symbol. The T3X compiler always uses
        all characters contained in two symbols to distinguish them, so




        are guaranteed to be different. The maximum length of symbol
        names may be limited by other factors, though, like the maximum
        length of a token.

        2.4 Data Declarations

        T3X is not a totally typeless language. It is called a
        'typeless' language anyway, because even BCPL (which pushes the
        typeless concept quite to its limit) has at least two types
        (variables and MANIFEST constants) which require different
        handling at compile time. In T3X, the following types exist:

        - Atomic variables
        - Constants
        - Vectors
        - Structures
        - Procedures
        - Methods
        - Objects
        - Classes

        Vectors and structures are basically the same and there is no
        big difference between methods and procedures, either. Atomic
        variables are used to hold small numeric values or single ASCII
        characters (which are represented by numbers) or pointers to
        other objects. Constants are used to provide symbolic names for
        immutable numeric values. Vectors are sequences of atomic
        variables. A structure is a set of constants which is used to
        give names to specific members of a vector. Procedures process
        parameters and return values just like mathematical functions.
        Since T3X is an imperative language, they usually have effects,
        too. A method is a procedure which is used to query or alter the
        state of an object. An object is an instance of a class.
        Classes will be discussed in detail in the section about object
        oriented programming.

        2.4.1 Atomic Variables

        Each (atomic) T3X variable allocates exactly one machine word.
        When talking about variables in the remainder of this document,
        the attribute atomic is implied. Vectors will be implicitly
        referred to as vectors or arrays.

        Variables are defined using a VAR statement. Any number of names
        may be defined in a single statement:

        VAR x_coord, y_coord, depth;

        Although, it is recommended to define only logically connected
        variables in a single statement.

        All types of values may be stored in a variable: numeric values,
        pointers to strings, pointers to vectors, pointers to
        structures, pointers to objects, or single characters. The range
        of numeric values which may be stored in a variable actually
        depends upon the implementation. The Tcode engine uses only 16
        bits to represent a 'cell' or a machine word - independently
        from the underlying platform. Therefore, programs which use
        values not in the range -32767...32767 should be considered
        to be machine-dependent. The T3X compiler will not allow the
        use of numeric literals outside of this range.

        When a variable is placed in an expression (frequently also
        called a 'right-hand side' (RHS) value, it evaluates to its
        value. When it is placed on the left-hand side of an assignment,
        however, it evaluates to its address (which will be dereferenced
        immediately by the following assignment operator, though). The
        assignment statement

        X := 5;

        would change the values stored in the variable X to 5.

        2.4.2 Constants

        Constants are variables that exist only at compile time
        (so-called 'compile time variables'). Instead of an
        automatically assigned address, they are initialized with an
        explicitly specified value when they are declared. Since they
        are compile time entities, the values of constants may not
        change at run time. Any number of constants may be declared in a
        single CONST statement:

        CONST READ = 1, WRITE = 2, RDWR = READ | WRITE;

        Each constant name must be followed by an equal sign (=) and a
        constant expression that evaluates to the value of the constant.
        Constant expressions will be explained in a later section.

        Constants may occur only in RHS expressions, where they evaluate
        to their values.

        2.4.3 Vectors

        Vectors are compile time variables, too. When they are declared,
        they will be initialized with the address of an array of
        subsequent machine words, the so-called vector members or vector
        elements. The address of a vector is equal to the address of its
        first member. Any number of vectors may be defined in a single
        VAR statement. Declarations of vectors and atomic variables may
        be mixed in one and the same statement:

        VAR RingBuffer[SIZE], Head, Tail;

        Vector declarations differ from atomic variable declarations by
        the trailing square brackets containing a constant expression
        which specifies the size of the vector in machine words. The
        first member of a vector has an index value 0 and the last one
        has an index of vectorsize-1 (SIZE-1in the above example). The
        size of a vector may range from 1 to 32767 elements.

        Since vector addresses are stored in compile time variables,
        they may not change at run time. It is legal to change the
        values of vector members, though. When occurring in RHS
        expressions, vector names evaluate to the addresses of their
        associated arrays.

        Single members of a vector may be addressed using the subscript
        operator []. The expression


        for example, evaluates to the fifth member of the vector V
        (given that the first member of the vector is actually referred
        to as the zeroth member). Subscripted vectors may occur on the
        left sides of expressions, as well. The assignment

        v[i] := 99;

        would change the I'th member of V to 99. Like atomic variables,
        the members of vectors may be used to store any data type, even
        pointers to vectors. See the description of the []-operator for
        details about nested vectors.

        A special case of the vector is the byte vector. Like 'ordinary'
        vectors (vector of machine words), they are declared in VAR

        VAR Input::256, Output::256;

        The only difference between a vector and a byte vector is the
        computation of the required size. The size value after the
        ::-operator specifies the number of characters required. The
        amount of memory actually allocated depends on the size of a
        machine word on the target machine, which is returned by the
        core class procedure T3X.BPW() (the method BPW of the class
        T3X). For all Tcode programs, T3X.BPW()=2 applies. The size of a
        byte vector is computed using the following formula (T being an
        instance of the T3X class):

        vectorsize + T.BPW() - 1

        It allocates enough space for at least VECTORSIZE characters.
        No further type information is associated with vectors.
        Therefore, it is valid to access byte vector members using []
        and word vector members using ::. The former is discouraged,
        though, because the actual size of a vector might depend on a
        specific implementation and alignment errors may occur at

        A byte vector may not be larger than 32767 bytes (16384 machine
        words on the Tcode machine).

        2.4.4 Structures

        A structure is a composed data object. Only one structure may be
        defined in a single STRUCT statement:

        STRUCT POINT = PT_X, PT_Y;

        Such a statement does not actually create a new data object, but
        only the 'layout' of a structure. For example, to create an
        actual POINT data object, an additional VAR statement is

        VAR point_a[POINT], point_b[POINT];

        This statement creates two POINT entities, point_a and point_b.
        The members of such structures can be addressed using the
        subscript operator: point_a[PT_X] and point_a[PT_Y].

        Structures do not really have an own type. As the declaration
        and member access syntax already suggests, they are ordinary
        arrays and the member names are constants. In fact, the

        STRUCT S = A, B, C;

        is perfectly equal to

        CONST S = 3, A = 0, B = 1, C = 2;

        The STRUCT statement only defines symbolic names for accessing
        vector members with a fixed position and known meaning. The
        structure name is another constant which holds the number of
        constants used to name the members (and therefore the size of
        the entire structure in machine words).

        2.5 Factors

        This section describes the most basic elements of each T3X
        program, the factors which may be used inside of expressions.

        There are many different kinds of factors: symbols, numeric
        literals, character literals, string literals, tables, procedure
        calls, messages, and class constants. A factor may only occur in
        expressions and a single factor is the minimum form of an
        expression. Factors may be prefixed by unary operators and they
        may be combined using binary or ternary operators. Basically,
        all sorts of factors are exchangeable: where one of them may
        occur, all others are allowed. The only exception is the symbol
        which has some additional properties which make it special. For
        example, symbols may be subscripted and it is possible to
        compute their addresses. These operations are limited to
        symbols. All other operations may be applied to any kind of
        factor, even if it makes little sense, like the multiplication
        of two strings (which will yield a highly environment-dependent

        "Hello" * "World"

        The evaluation of a symbol depends on its type. Variables and
        constants evaluate to their values, vectors and objects evaluate
        to their addresses. Class names evaluate to instance sizes.
        Structure names and structure member names are treated in the
        same way as constants.

        Class constants are public constants which are defined inside of
        classes. To include a class constant in an expression, it must
        be prefixed with the name of the defining class and a dot:


        Like 'ordinary' constants, class constants evaluate to their

        Numeric literals are written in decimal, hexa-decimal, or binary
        notation and represent their own values. A percent sign may be
        used to negate a number:

        %123 = -123

        The difference between %123 and -123 is that %123 is a factor
        while -123 is an expression ('minus' applied to a numeric
        factor). In fact, the percent sign has little meaning in T3X
        today, since the compiler accepts ordinary minus prefixes in
        constant expression contexts, too. In early T3X versions,
        constant expressions were limited to single factors and
        therefore, the percent sign was required to define negative
        constant values. The % prefix is kept for compatibility. An
        optimizing compiler might turn -N into %N, if N is a constant
        numeric factor.

        Hexa-decimal notation may be used to represent a numeric value
        when prefixing the literal with the strings '0x' or '0X' (null,
        X). No space is allowed between the prefix and the hexa-decimal
        digits. The number 4095, for example, can be written as '0xfff'
        or '0xFFF'. The characters 'A' through 'F' (alternatively
        'a'...'f') are used to represent the hexa-decimal digits with
        the values '10' to '15'. No difference is made between upper and
        lower case characters. The literals

        0x1f 0X1f 0x1F 0X1F

        all express the decimal value 31. The percent prefix may be
        combined with hexa-decimal factors as well, e.g. %0xf.

        Numbers may be expressed in binary notation by prefixing the
        literal with the strings '0b' or '0B' (null, B). No space is
        allowed between the prefix and the binary digits. The number
        165, for example, could be written as '0b10101010'.

        Note: The literals 0x8000 and 0b1000000000000000 should not be
        used to express the (decimal) value -32768. This value is not
        defined in T3X. Since 0x8000 is useful to mask the most
        significant bit of a pattern, though, the compiler will allow
        its use, but it will not allow the notation -32768.

        Character literals are single characters or escape sequences
        enclosed by single quote characters, like

        'a' '0' '\s' ''' '\'' '\\'

        A character literal evaluates to the ASCII code of the enclosed
        character. An escape sequence may be used to include certain
        unprintable or special characters. The backslash character is
        used to introduce such a sequence. The '\' itself and the
        following character will be removed and replaced with the
        associated special character. Note that no escape sequence is
        required to represent an apostrophe: '''.  Besides most C-style
        sequences, the following translations will be performed:
        \e->ESC, \q->", and \s->space. The latter has been included to
        improve readability. Unlike C, T3X accepts uppercase sequences
        as well: \s and \S both evaluate to space. The escape character
        may be used to escape itself. Thereby, it loses its special
        meaning and '\\' evaluates to a single literal backslash. A
        summary of all escape sequences is listed in the quick reference

        String literals are sequences of characters delimited by double
        quotes ("):

        "Hello, World!\n"

        Each character either represents itself or is part of an escape
        sequence as described above. Each character is stored in a
        single byte. String literals are terminated with a NUL
        character, so N+1 bytes are required to store a string of the
        length N (but sizes are always allocated in machine words).

        Since a string is an array of subsequent bytes, the ::-operator
        may be used to access its individual characters.

        At runtime, each string literal evaluates to the address of its
        first character.

        A more general form of a literal vector is the table. A table is
        a static initialized vector and a generalization of BCPL-style
        TABLEs.  Syntactically, it is a list of table members delimited
        by square brackets:

        [ 7, "MOD", @modulo ]

        Each table member occupies exactly one machine word. A string,
        for example, is included as a pointer, while the string literal
        itself is placed outside of the table. Therefore, table members
        can be accessed using the subscript operator []: if

        x = [ 77,88,99 ]



        evaluates to 99. The square bracket notation was chosen for
        delimiting tables because of the strong connection between
        vectors and the subscript operator.

        The type of each table member may be any out of the following:

        - Constant expressions
        - Strings
        - Addresses of global data objects
        - Addresses of procedures (including methods)
        - Tables (packed or unpacked)
        - Embedded expressions

        Constant expressions include everything which has a value that
        can be computed at compile time (like numeric literals). The
        inclusion of strings has been explained above. Addresses of
        global variables and procedures are represented by their symbol
        name prefixed with the address operator @.

        What makes tables particularly flexible is the possibility to
        nest them:

        [ [ 2, 9, 4 ],
          [ 7, 5, 3 ],
          [ 6, 1, 8 ] ]

        Like strings, embedded tables are stored outside of the
        surrounding table and included as pointers. If, for example, the
        above table is assigned to the symbol V, the following
        equations hold:

        v[0] = [ 2, 9, 4 ]
        v[1] = [ 7, 5, 3 ]
        v[2] = [ 6, 1, 8 ]

        Since the result of applying a subscript operator to a table
        containing tables is another table (vector), the subscript
        operator may be applied one more time, and consequently,


        would result in 5:

        v       = [ [2,4,9], [7,5,3], [6,1,8] ]
        v[1]    = [7,5,3]
        v[1][1] = 5

        (Remember that the first element of a vector has an index of 0.)

        A table that contains at least one non-constant expression is
        called a 'dynamic table'. Non-constant expressions must be put
        in parentheses when they are contained in a table:

        v := [ "a * b = ", (a*b) ];

        When there are multiple subsequent dynamic values in a table,
        a single pair of parentheses enclosing the entire set is

        [ "sums", (a+b), (a+c), (b+c) ]

        can be written more conveniently as

        [ "sums", (a+b, a+c, b+c) ]

        Embedded (non-constant) expressions are computed freshly each
        time the flow of the program passes the table in which they are
        contained. Therefore, the values of table members computed by
        embedded expressions may be different each time the table is
        evaluated. This is why such a table is called 'dynamic'. The
        parentheses show the compiler that an expression is non-constant
        and make it generate additional code to fill in the value of the
        expression whenever the table is evaluated. Therefore, static
        (constant) expressions should never be parenthesized in tables,
        because doing so would result in inefficient code. For example,

        v := [ "5 * 7 = ", (5*7) ];

        works, but computes 5*7 each time the table is evaluated. Even
        if the optimizing compiler folds 5*7 to 35, the value would
        still have to be stored in the table each time it is passed.

        On the other hand, including dynamic expressions in a table
        without any parentheses will lead to an error:

        v := [ "a * b = ", a*b ];

        will not work unless both A and B are constant.

        Tables may be prefixed with the keyword PACKED. Packed tables
        may only contain byte-sized values. Therefore, their members are
        limited to constant expressions with bit patterns where only the
        least significant 8 bits contain values other than 0. Expressed
        in numbers, this is the range from -128 to 255.

        A string may be considered to be a special case of the packed
        table.  Consequently, each string may be written as a packed
        table as well. For example,


        is equal to

        PACKED [ 'T', '3', 'X', 0 ]

        Note the trailing zero in the vector literal! Both, strings and
        packed tables will be padded with zeros up to the next word

        The maximum number of members per table may be limited, but at
        least 128 elements per table must be allowed by any T3X
        implementation. The elements contained in nested tables do not
        count, but the entire embedded table counts as a single member.
        The same limit may exist for packed tables and string literals.

        Procedure calls are represented by a procedure name followed by
        zero or more comma-separated arguments, enclosed by parentheses:

        find(text, "word", 0, TEXT_SIZE);

        Each argument may be any valid expression. When a procedure P
        expects zero arguments, the parentheses must still be supplied:
        P(). A procedure call evaluates to the return value of the
        called procedure.

        In T3X, only procedures may be called. Calls to absolute
        addresses and computed calls - like in BCPL - are not allowed.
        There is a mechanism to perform indirect calls, though: the CALL
        operator. More detailed information on procedure calls and the
        procedure call operators can be found in later sections.

        A message is used to activate a method of a class. It is sent to
        an instance of its class, also known as an object. The message
        syntax is equal to a procedure call prefixed with the name of
        the instance to which the message shall be sent, separated from
        it by a dot:

        t.write(T3X.SYSOUT, "Hello, World!\n", 14);

        Details about messages can be found in the chapter on object
        oriented programming.

        2.5.1 Signed and Unsigned Values

        Numeric entities usually carry a sign in T3X. This means that a
        part of a bit pattern representing a number is reserved to
        indicate the number's sign, positive or negative. On
        two's-complement machines, the most significant bit (high bit)
        contains the sign flag. If this bit is set, the number is
        negative and otherwise it is positive. Therefore, the numeric
        range on the Tcode machine includes the values -32767 to 32767
        (in bit patterns 0xffff to 0x7fff; remember that -32768 is

        Under some circumstances, it is desirable to interpret a number
        as an unsigned entity instead - for example when comparing
        pointers. In this document, a leading dot is used to indicate
        an unsigned number. In T3X itself, no such notation exists, but
        some operators may be modified with a leading dot to turn them
        into 'unsigned operators'. Unsigned operators treat the sign bit
        as a part of the value. Therefore, the domain of these operators
        is {0 ... 65535} rather than {-32767...32767}.

        Note that not all unsigned values may be expressed in the form
        of decimal numeric literals in T3X.

        Since the modified operators operate on raw 'bit patterns', -1
        and 65535 represent the same value to them on two's-complement
        machines. To avoid confusion, signed and unsigned operators
        should only be applied to the following ranges:

        Range             Operators
        --------------    -----------------------------
        -32767...32767    signed:    *  /  <  >  <=  >=
        0...65535         unsigned: .* ./ .< .> .<= .>=

        2.6 Expressions

        In expressions, operators may be used to modify or combine
        factors in various ways. Most operators may be applied to any
        kind of operand (even if the resulting operation may not
        result in any meaningful value).

        There are different kinds of operators and, like procedures,
        they are classified by the number of their arguments, which are
        called 'operands' in this context. There are unary (prefix)
        operators, binary (infix) operators, ternary (also infix)
        operators, and then there is one variadic operator.

        Operators may also be classified by their precedences. The
        higher the precedence of an operator is, the stronger it binds
        its operands. For example, the term operators (product,
        quotient, remainder) bind stronger than the sum operators (sum,
        difference). Therefore,

        a * b + c * d

        is equal to

        (a * b) + (c * d)

        Like in math expressions, parentheses may be used to override
        the default grouping. The precedence rules are simple in T3X:

        1. Postfix operators bind strongest.
        2. Unary operators bind stronger than binary operators.
        2. Binary operators bind stronger than the conditional operator.
        4. The ascending order of precedence for binary operators is as
           follows: disjunction (1), conjunction (2), equation (3),
           ordering relation (4), bit (5), sum (6), and term (7)

        Although postfix operators are also 'binary', they are called
        'postfix operators' in this text, while 'binary operators'
        means binary infix operators.

        The precedence rules in 4. are similar to the rules used in the
        evaluation of common math expressions.

        Another property of an operator is its associativity. An
        operator associates to the left, when a sequence of identical
        operations is evaluated from the left to the right and
        associates to the right, when such sequences are evaluated
        from the right to the left.

        Associativity   Expression    Meaning
        -------------   -----------   -------------
        left            A op B op C   (A op B) op C
        right           A op B op C   A op (B op C)

        In T3X, all binary operations with the sole exception of :: are
        left-associative. The byte operator is right-associative.

        In the remainder of this section, all available operators
        will be explained. The appearance is ordered by descending

        2.6.1 Procedure Calls and Subscripts

        The operators (), [], and :: are the only postfix operators.
        They are always applied to primary factors in the form of
        symbols, e.g.:


        The ()-operator is a variadic operator. Given the procedure call

        P(a1, ..., aN)

        its arity is N+1 (P plus N arguments). The meaning of the
        operator is the application of the procedure P to the (optional)
        arguments a1 through aN. If P does not have any formal
        arguments, its syntax is P().

        The value of the operation depends on the semantics of P. See
        the description of the RETURN statement for further details.

        () may only be applied to symbols of the type 'procedure'. The
        procedure must have been declared before its first application,
        either by using a procedure declaration or a forward
        declaration. The number of arguments to a procedure call will be
        checked against the arity of the called procedure. If the
        numbers do not match, an error will be signalled.

        Each argument in a procedure call may be any valid expression
        itself, which includes, of course, procedure calls. Given the
        binary function P2, the following expression is perfectly valid:

        P2( P2(1, 2), P2( 3, P2(4, 5) ) )

        An indirect procedure call may be performed using the CALL
        operator. The expression

        CALL PP(a1, ..., aN)

        evaluates to the result of the application of PP to a1...aN, but
        in this case PP is a 'procedure pointer' instead of an actual
        procedure. A procedure pointer is an ordinary variable which has
        been assigned the address of a procedure by using the address
        operator '@':

        PP := @P;

        In indirect procedure calls, no type checking (as described
        above) will be performed. If PP is the name of a procedure
        instead of a variable, the keyword CALL will be ignored.

        In direct and indirect calls, the calling convention is 'call
        by value'. This means that all arguments of the call will be
        evaluated before control is transferred to the called procedure,
        so that the value of each parametric expression will be
        transported to the procedure.

        Note: Since vectors evaluate to their addresses, passing a
        vector by value will actually pass a reference to the vector.
        Therefore, vectors are always passed by reference: Instead of
        passing the entire vector, only the address of its first member
        is transported to the called procedure. In the procedure the
        address will be stored in an (atomic) parameter variable. Since
        parameters are always atomic and therefore evaluate to their
        values and vectors evaluate to their addresses, both the actual
        vector and the parameter will reference the same memory location
        and subsequently, the parameter may be used in the same way as
        the original vector.

        Procedure parameters are guaranteed to be evaluated in the order
        of occurrence (from the left to the right). For example, given
        the expression

        P( Q(), R() );

        the programmer may rely on the fact that Q() will be called
        before R().

        The subscript operator [] may be applied to vectors as well as
        to atomic variables. The subscript in

        symbol [subscript]

        may be any valid expression. If A is a vector, the subscript


        evaluates to the B'th member of A. If A is an atomic variable,
        the operation evaluates to the B'th member of the vector pointed
        to by A. This means that both subscripts in the following
        example would evaluate to the same value:

        var v[100], pv;
        var a1, a2;

        pv := v;
        a1 := v[25];
        a2 := pv[25];

        Since there is no nesting limit for vectors, any number of
        subscript operators may follow a single symbol. Assuming that V5
        holds a vector containing five levels of nested vectors, the


        could be used to access single elements at the deepest nesting
        level. Chains of subscripts evaluate from the left to the right.

        The 'byte subscript' operator :: differs from the ordinary
        (word) subscript operator in several ways. First, it addresses
        bytes in (byte) vectors and second, it is right-associative.
        The expression


        evaluates to the B'th byte of the vector A. Therefore, :: is
        mostly used to access characters in strings. Since the results
        of ::-operations are always limited to byte width, they cannot
        be assumed to return valid addresses. For this reason, byte
        subscripts associate to the right: their result may very well be
        a valid subscript. If the expression

        a :: b :: c

        would evaluate from the left to the right, the result of a::b
        would probably not be a valid address, since it is limited to
        eight bits. In this case, however, the following subscript would
        reference the C'th byte of a non-vector - which would certainly
        not be the desired result. If the expression evaluates from the
        right to the left, though, the subexpression b::c is evaluated
        first and will probably return a valid subscript. This subscript
        is then applied to the vector A.

        Finally, the :: operator differs from [], because it has no
        right-hand side delimiter. Therefore, the right-hand side of ::
        is always a single factor and expressions like


        actually evaluate to


        since :: has the highest precedence. To address the b+c'th byte
        in the array A, the subscript must be but in parentheses:


        2.6.2 Unary Operators

        All unary operators have a high precedence and bind to single
        factors.  Unless explicitly specified using parentheses, they
        never affect subexpressions containing other operators (except
        for postfix operators which have an even higher precedence). The
        suffix operators must bind stronger than the prefix operators,
        because this order leads to much more sensible semantics. For


        means 'negate the result of applying P to A and B' and


        means 'evaluate to the inverse value of the J'th member of V'.

        If the order of precedence would be inversed, the meaning of the
        first example would be 'apply whatever is at the negative
        address of P to A and B' and the second one would mean 'evaluate
        to the J'th member of the vector located at the address
        expressed by the inverse value of V'.

        Alt in all, there are four prefix operators. The minus sign (-)
        (which exists as a binary operator, too) evaluates to the
        negative value of its operand. Like in math expressions, any
        even number of minus signs has no effect. The unary minus sign
        is distinguished from the binary '-' by its context. When the
        sign occurs between two operands, it is binary. If it occurs at
        the place of a factor, it is unary and the factor itself
        follows after the operator.

        The tilde operator (~) results in the value of its operand with
        all bits inverted. Since inverting a bit twice always yields the
        original state, even numbers of ~-operators have no effect,

        The backslash (\) represents the logical NOT (while ~ represents
        the bitwise NOT). This operator evaluates to logical truth (-1),
        if its operand is logically false (0) and vice versa. Only the
        value zero represents logical falsity in T3X and all non-zero
        values represent logical truth. The canonical form of the 'true'
        value is -1. Two (or any other even number of) subsequent
        logical NOT operators may be used to create the normal form of an
        arbitrary truth value.

        The address operator (@) evaluates to the address of its
        operand. Therefore, it may only be applied to symbol names. The
        addresses of constants, structure member names, and classes may
        not be computed using @, because such entities have no
        addresses. Since the subscript operators bind stronger than the
        address operator, @ may be used to compute addresses of vector
        and structure members, and even the addresses of members of
        nested tables:


        computes the address of the J'th member of the embedded vector
        v[i]. Of course, the address operator might be combined with
        byte subsrcipts, as well:


        yields the address of the I'th byte of S.

        It is theoretically possible to compute the address of the I'th
        member of the bytevector S using

        x := s+i;

        and the I'th member of the vector V using

        x := v + i*t.bpw();

        but the expressions

        x := @s::i;
        x := @v[i];

        are both more portable and more comprehensible.

        2.6.3 Term Operators

        The operation A*B evaluates to the product of A and B. If A*B
        does not fit in a machine word, the result is undefined.

        A/B results in the integer part of the quotient of A and B. The
        result is undefined, if B is zero.

        A MOD B evaluates to the difference between A and A./B.*B where
        A./B is an unsigned integer division and .* is an unsigned
        multiplication.  Therefore, A MOD B is the division remainder of
        A./B. Like /, MOD leads to an undefined result, if B=0.

        All term operators respect the signs of both of their operands.
        Two equally signed operands yield a positive result and operands
        with different signs lead to a negative result.

        However, the T3X language also provides some modified operators
        which work on unsigned values. Modified versions of the
        multiplication and division operator exist. Like all modified
        operators, they are prefixed with a dot (.).

        The operation A.*B evaluates to the product of the unsigned
        values .A and .B. A./B results in the integer part of the
        quotient of .A and .B.

        The notation .X is used to denote the unsigned value of X. It
        is not part of the T3X syntax.

        2.6.4 Sum Operators

        A+B evaluates to the sum of A and B and A-B evaluates to their

        2.6.5 Bit Operators

        In T3X, all bit operations have the same precedence. Grouping
        such operations usually requires parentheses. Otherwise
        evaluation is performed from the left to the right.

        The operation A&B results in the bitwise AND of A and B. Each
        bit is the result of computing the logical product of one bit in
        A with the bit at the same position in B.

        A|B yields the result of performing a bitwise OR on A and B.
        Each bit in the result is a logical sum of a bit in A and the
        bit at the same position in B.

        A^B performs a bitwise exclusive OR (XOR). In this case, the
        computation of a single bit is done by combining bits at the
        equal positions using the logical negative equivalence ('not
        equal') operation.

        See the following table for the results of applying logical
        operations to pairs of bits.

        A   B   AND,*   OR,+   XOR,\=
        -   -   -----   ----   ------
        0   0     0      0       0
        0   1     0      1       1
        1   0     0      1       1
        1   1     1      1       0

        A<<B evaluates to the value of A with all bits shifted to the
        left by B positions. This is the same as an unsigned
        multiplication with the B'th power of 2:

        a << b = a .* 2


        2<<3  =  2 .* 2   =  16

        After such an operation, the sign of the result must be
        considered to be undefined. This is not relevant, of course, if
        A is used as a bit field where each bit represents a binary

        A>>B yields the result of shifting the bits of A to the right by
        B positions. This is basically equal to the computation of the

        a ./ 2

        Like in left-shift operations, the sign must be considered
        to be undefined after right-shift operations. The >>-operation
        does clear the most significant bit of its result, though.

        Technically speaking, one might say that the shift operators in
        T3X perform bitwise rather than arithmetic shift operations.

        2.6.6 Relational Operators

        Relational operators are used to compare two operands. The
        relation between the operands is expressed as a truth value: all
        these operators return truth, if their meaning applies to their
        operands and otherwise falsity. The following relational
        operations exist (.X denotes the unsigned value of X):

        Operator    Description
        --------    --------------------------------
        A < B       A is less than B
        A > B       A is greater than B
        A <= B      A is less than or equal to B
        A >= B      A is greater than or equal to B
        A .< B     .A is less than .B
        A .> B     .A is greater than .B
        A .<= B    .A is less than or equal to .B
        A .>= B    .A is greater than or equal to .B
        --------    --------------------------------
        A = B       A is equal to B
        A \= B      A is not equal to B

        Note: the operators expressing equivalence (=, \=) have a lower
        precedence than operators expressing ordering (> , <, >=, <=,
        .<, .>, .<=, .>=). For example,

        A < B = C < D

        is equal to

        (A < B) = (C < D)

        Consequently, the equation sign may be interpreted as 'logical
        equivalence' when used between comparisons: the above expression
        evaluates to true, if either

        (A<B) AND (C<D)


        \(A<B) AND \(C<D)

        applies. Since the inequation operator \= has the same
        precedence as =, it may be used as a negative logical
        equivalence operator (aka an Exclusive OR):

        A<0 \= B<0

        becomes true, if either A or B is negative. If the signs of A
        and B are equal, the expression yields the result 'false'.

        Note, again, that any value may be considered a truth value in
        T3X. Everything but the value zero is interpreted as 'truth',
        and only 0 may be used to express the 'false' value.

        2.6.7 Conjunction and Disjunction

        The operators A/\B and A\/B implement logical conjunction (AND)
        and disjunction (OR). Generally, the expression

        A /\ B

        evaluates to some true value, only if A AND B evaluate to
        'truth'.  Analogously,

        A \/ B

        yields a true result if either A OR B (or both of them) evaluate
        to 'truth'.

        More specifically, /\ and \/ are so-called 'short circuit
        operators'. Since the expression A/\B can lead to a true result
        only if all its operands are true, there is no actual need to
        evaluate B, if A already has yielded 'false'. Therefore, the
        second operand of a conjunction will never be evaluated by a T3X
        program, if the first one already is false. The result will be
        zero in this case. If, on the other hand, the first value is
        true, the result of the entire operation will be the value of
        the second operand. Therefore, the result of

        A /\ B

        can be specified more precisely as

        zero, if A = 0


        B, if A \= 0.

        Similarly, the expression A\/B can never become 'false', if A
        already has been found out to be true. Therefore, no T3X program
        will ever evaluate B in such a case, and the meaning of the

        A \/ B

        can be defined more precisely as

        A, if A \= 0
        B, if A = 0

        Like in mathematical logic, conjunction binds stronger than

        A /\ B \/ C /\ D


        (A /\ B) \/ (C /\ D)

        In chains of equal logical operations, the order of evaluation
        is from the left to the right (as in all binary infix
        operations). This means that chains of conjunctions will be
        evaluated up to the first 'false' operand and chains of
        disjunctions will be processed up to the first 'true' operand.
        In either case, the result of the entire chain is the value of
        the last operand that was evaluated.

        There exists a connection between the logical operators and
        conditional statements: Because of their short circuit nature,
        logical operators may be used to implement flow control inside
        of expressions. The expression

        A /\ B()

        has almost the same meaning as

        IF (A) B();

        The only difference is that the expression yields a value, while
        the statement only has an effect. Likewise, the expression

        A \/ B()

        has the same meaning as

        IF (\A) B();

        when ignoring the value of the expression. The IF-statement will
        be explained in a later section.

        2.6.8 Conditional Expressions

        The ternary conditional operator has the least precedence.
        Therefore, it may be used to combine any kind of expressions
        without using parentheses. The following expression, for
        example, implements the minimum function:

        a < b -> a : b

        Since the operator has three operands, it consists of two parts:
        '->' and ':'. The meaning of the conditional operator is as
        follows: In the expression

        A-> B: C

        the operand A (the condition) is evaluated first. If it
        evaluates to some 'true' value, B will be evaluated and
        otherwise, C will be evaluated. If B is evaluated, C will not be
        evaluated and vice versa. The result of the expression is equal
        to the value of the last evaluated operand.

        Like the logical operators /\ and \/, the conditional operator
        has a connection to conditional statements:

        A-> B(): C()

        is equivalent to

        IE (A) B(); ELSE C();

        except for the fact, of course, that the expression has a value,
        while the statement only an effect. (IE means If/Else and
        introduces a conditional statement with an alternative). The
        IE-statement will be discussed in a later section.

        2.6.9 Constant Expressions

        Constant expressions are used wherever a value must be known at
        compile time. Only a limited set of operators is allowed in
        constant expressions and the order of evaluation is always from
        the left to the right. Only one single unary operator is allowed
        per factor. There are no precedence or associativity rules.


        evaluates to (L+1)*10 and not to L+(1*10) like it would in
        ordinary expressions. The reasons for this decision were (1)
        simplicity of implementation and (2) the fact that most
        conditional expressions contain only a single operator or none
        at all.

        The following operators are recognized inside of constant

        '+' is frequently required to increase the lengths of arrays.
        For example, if a buffer has to be used as a string, an
        additional machine word must be appended to hold the delimiting
        NUL character, e.g.:

        VAR buffer[BUFSIZE + 1];

        The binary - has been added to avoid constructs like

        CONST X = Y + %2;

        which may now be written as

        CONST X = Y - 2;

        '*' can be used to allocate memory for arrays of structures, for


        '|' is useful when creating constant bit maps:

        CONST A = 8, B = 16, AorB = A | B;

        '~' is also useful for creating constant bit maps:

        CONST SignBit = 0x8000, ValueMask = ~SignBit;

        The unary '-', finally, can be used to negate constants. This is
        particularly useful when counting down in FOR loops where the
        step width must be constant:

        FOR (i=99, -1, -STEPWIDTH) p();

        2.6.10 Order of Evaluation

        While the associativity and precedence rules specify which
        operation is to be performed first, the order of evaluation
        determines which factor is to be evaluated first. For example,
        in the expression

        A * B

        A may be evaluated before B or B may be evaluated before A. The
        order of evaluation becomes important, if both A and B have
        effects. If A had the effect of printing 'A' on the terminal
        screen and B would print 'B', the terminal output of above
        expression could be "AB" as well as "BA".

        The order of evaluation is undefined in most operations, but
        there are exceptions: the definitions of the conjunction,
        disjunction, and conditional operators order the evaluation of
        their operands explicitly. Therefore, the order of evaluation
        of an expression like

        A() /\ B()

        is exactly defined. The left-hand side is always evaluated first
        and the right-hand side is only evaluated, if the value of the
        left-hand side is non-zero. Given the effects described above,
        this expression would print "AB", if A() is non-zero and "A", if
        it is zero. It would under no circumstances print "BA".

        The other exception is the order of evaluation of procedure call
        arguments and nested procedure calls. Procedure call arguments
        are guaranteed to be evaluated from the left to the right and
        nested calls are evaluated inside-out. The expression

        P( A(), B() )

        would invariably be evaluated in the order A, B, P. Therefore,
        it is safe, for example, to format a string in a procedure call
        argument in T3X and compute the length of the formatted string
        in a following argument. The statement

        t.write(1, str.format(b, "%S", [(a)]), str.length(b));

        would be guaranteed to print the string in B, as formatted by
        str.format(), correctly.

        Note: A T3X programmer should never rely on any order of
        evaluation not explicitly specified in this subsection! Even if
        precedence rules may suggest a specific order of evaluation, it
        may in fact be different and, even worse, it may change without
        breaking any rules, for instance when turning optimizations on
        or off or using a different compiler version.

        2.7 Statements

        Statements are the basic building stones of T3X programs. While
        expressions just have a value, statements are used to 'tell the
        computer to do something'. This is why T3X is called an
        'imperative language'. Each program is a list of commands which
        is executed in sequence. Each command is also called a statement
        in the terminology of imperative programming.

        There are different kinds of statements: assignments, procedure
        calls, conditional statements, loop statements, branch
        statements, and compound statements. The assignment is an
        essential part of every imperative language. It is frequently
        even used to characterize the imperative approach. Compound
        statements do not have an own meaning, but they are used to
        group statements to form the bodies of loops, conditionals, and
        procedures. All other statement types control the flow of a

        In T3X, all statements have to be terminated with a semicolon.
        This means that a semicolon must follow every statement in a
        program, except for compound statements which are delimited by
        the keywords DO and END. In other procedural languages (like
        BCPL and Pascal), statements are 'separated' rather than
        terminated. In such languages, a delimiter is only necessary, if
        two or more statements are written in sequence - there may not
        be any delimiter after the last statement. The separation rules
        in some languages are rather complex and the saving in
        delimiters is usually not worth the extra expense of having to
        remember these rules. Therefore, the most simple form of
        combining statements has been chosen in T3X:

        Every (non-compound) statement has to be terminated.

        2.7.1 Assignments

        An assignment transfers the value of an expression to a specific
        storage location. For example, the statement

        A := B;

        copies the value of B to A. After the assignment, both variables
        will have the same value. The previous value of A is thereby

        The right-hand side (RHS) of an assignment may be any valid
        expression as described in the previous section. The left-hand
        side is restricted to a subset of expressions that is often
        referred to as LHS values or lvalues (left-hand side values).
        In T3X, each lvalue may be any of the following:

        - atomic variables
        - vector members
        - byte vector members
        - structure members

        where vector members and structure members are essentially the

        Assignments to vector members are not limited to direct members
        of a vector. Addressing elements of multiply nested vectors is
        perfectly valid. The evaluation of variables on left-hand and
        right-hand sides of assignments was explained in detail in the
        section about factors. In short, RHS variables evaluate to their
        values and LHS variables evaluate to their addresses. The
        assignment operator := first evaluates the expression on its
        left side. Then it evaluates the expression to its right and
        stores the result at the address denoted by the LHS.

        A generalization of the evaluation of left-hand sides is the
        following: All but the last reference on a left-hand side of an
        assignment evaluates to its value. Only the last reference
        evaluates to its address. Here are some examples:

        A := B;

        The symbol A references a specific storage location. Since it is
        the only reference in the lvalue, it evaluates to its address.
        In the statement

        A[i] := B;

        A is not the last reference and hence it yields its value (which
        is its address, because  A is a vector). The operation [i]
        references the I'th member of A. Since it is the last reference
        on the LHS, it evaluates to the address of A[i] instead of its
        value. Consequently, the following assignment operator stores B
        at the address of the I'th member of A. The same is valid for
        accessing vector elements at any nesting level. The statement

        A[i1][i2][i3][i4] := B;

        for example, stores B in the i4'th member of A[i1][i2][i3].

        Accessing byte vectors works in the same way:

        A::i := B;

        stores the least significant eight bits of B in the I'th byte of

        Since :: associates to the right, the last evaluated reference
        is the leftmost one in chains of byte operators like

        A::B::i := C;

        Because B::i will be evaluated first in this example, it will
        yield its value. Then, the address of A::(B::i) is computed.
        Since no more references follow after A::, the (least
        significant eight bits of the) value of C will be stored in the
        (B::i)'th byte of A.

        Note: Although the assignment symbol := looks like an operator
        (and is frequently even referred to as such), it may not be used
        inside of expressions, but only to combine expressions. It is a
        command rather than an operator.

        2.7.2 Procedure Calls

        The application of a procedure may form a complete statement:

        fill(a, 'X', 10);

        In this case, the return value of the activated procedure will
        be discarded and only the effects of the procedure will actually
        take effect. The effect of the above statement, for example,
        could be to fill the first 10 characters of the vector A with
        the character 'X'.

        Each procedure - no matter whether it returns a specific value
        or not - may be used in a standalone procedure call. For details
        on procedure calls, see the sections on factors and procedures
        in this manual.

        2.7.3 Conditional Statements

        There are two forms of the conditional statement. The first one
        is the IF statement which is available in most procedural
        languages. Its general syntax is

        IF (expression) statement

        where 'expression' may be any expression and 'statement' may be
        any statement. The IF statement itself does not have to be
        terminated with a semicolon, because its 'body', which is a
        complete statement by itself, already supplies the terminating
        semicolon. The statement which forms the body of the IF
        statement will be executed only, if 'expression' evaluates to a
        'true' (non-zero) value. The following statement turns A into
        its absolute value:

        IF (a < 0) a := -a;

        If A is less than zero, then A will be assigned the value -A,
        thereby changing its sign. Since the body A := -A is executed
        only if, A < 0 applies, this conditional statement always leaves
        a positive value in A. The semicolon in the above example
        belongs to the assignment.

        The second form of the conditional statement is the IE
        statement, which implements a conditional statement with an

        IE (expression) statement-T ELSE statement-F

        Like in IF statements, any valid expression or statement
        may be used in the places of 'expression', 'statement-T',
        and 'statement-F'.

        The meaning of the IE statement is equal to the one of the IF
        statement as long as the expression becomes 'true'. In this
        case, 'statement-T' will be executed. If the expression
        evaluates to 'false', though, 'statement-F' will be executed,
        while an IF statement would not have any effect in this case.

        IF (expr) stmt

        is equal to

        IE (expr) stmt ELSE ;

        IE is an abbreviation for If/Else. In most languages, the IF
        statement may or may not have an alternative. In T3X, there is a
        separate type of statement for each variant. The reason for this
        decision is the 'dangling else' problem, which cannot occur when
        these statement types are separated. If no further information
        is supplied, the following program written in a language which
        allows optional alternatives would be ambiguous:

        IF (condition1)
        IF (condition2) statement1
        ELSE statement2

        The problem is to decide to which IF the ELSE branch belongs: is
        it the alternative of IF (condition1) or IF (condition2)? I.e.,
        does the above mean

        IF (condition1) DO
            IF (condition2) statement1
            ELSE statement2


        IF (condition1) DO
            IF (condition2) statement1
        ELSE statement2

        In fact, most languages will associate the ELSE branch with the
        most recently opened IF statement, and therefore implement the
        second program fragment above. In T3X, such an ambiguity does
        not exist:

        IE (condition1)
            IF (condition2) statement1

        Since the IF statement cannot have an alternative, the ELSE
        branch must belong to IE (condition1).

        2.7.4 Loop Statements

        There are two kinds of loops: 'while' loops and 'for' loops
        which represent two classes of problems: those which are
        computable by algorithms with a known upper limit of iterations
        (FOR-computable or primitive recursive functions) and problems
        which cannot be computed by algorithms with a fixed number of
        iterations (WHILE-computable or general recursive functions).
        Since the FOR-computable functions are a subset of the
        WHILE-computable ones, FOR statements may be considered to be a
        special case of WHILE statements and in fact, it is possible to
        express a FOR loop using WHILE, but not vice versa.

        Note: in fact, WHILE and FOR are completely interchangeable on
        computer systems with bounded memory. On an ideal machine with
        unbounded memory, though, they represent two different classes
        of problems. The Ackermann function, for example, is the
        classical example of a function that is WHILE-computable, but
        not FOR-computable.

        There is a third kind of loop in many other languages, the
        repetitive loop, but it turns out to be a special case of the
        WHILE loop. Repeating loops are not very frequently needed and
        if they are, they can easily be constructed using WHILE, IF and
        LEAVE in T3X.

        The WHILE loop has the following general form:

        WHILE (expression) statement

        where 'expression' may be any expression and 'statement' may be
        any statement. The 'body' consisting of the statement will be
        executed while the test expression in parentheses evaluates to
        some 'true' value.  Hence the name of this loop. If the
        expression becomes 'false' before the statement has been passed
        for the first time, the statement will never be executed.
        However, a loop which tests its exit condition at the end of the
        statement may be constructed using WHILE, IF, LEAVE, and a
        compound statement (which will be explained later in this

        WHILE (-1) DO    ! loop forever
             IF (\condition) LEAVE;

        In this case, statement will be executed at least once, because
        the exit condition -1 is a 'true' constant. In the subsequent IF
        statement, the loop will be left if the condition does not
        apply. LEAVE is used to branch out of a loop. It will also be
        explained later in this chapter.

        The FOR loop exists in two forms: an explicit form and a short
        form.  The explicit form looks as follows:

        FOR (var=start, limit, step) statement

        'Var' is an atomic variable that must be declared earlier in the
        program. Unlike in BCPL, it will not be declared implicitly by
        the FOR statement. 'Start' and 'limit' are expressions and
        'step' is a constant expression. The FOR loop works this way:

        First, 'var' is initialized with the value of 'start'.

        Second, 'var' is compared against 'limit'. If either the

        var < limit /\ step >= 0


        var > limit /\ step < 0

        holds, the statement is executed. Otherwise the loop is left and
        the statement will not be executed.

        Finally, 'step' is added to 'var', and the loop will be repeated
        from the point where the exit condition is checked. Like in a
        WHILE loop, the statement will never be executed, if the exit
        condition already is true when it is checked for the first time.

        The following examples print the numbers from 0 to 9 using the
        procedure 'print' (which is only defined in the first example).
        ('Print' uses some routines of the classes 't3x' and 'string',
        which will be explained in a later chapter.)

        MODULE example(t3x, string);

        OBJECT t[t3x], str[string];

        print(n) DO VAR b::10;
            t.write(T3X.SYSOUT, str.format(b, "%D\n", [(n)]),

        ! --- end of the common part ---

        DO VAR i;
            FOR (i=0, 10, 1) print(i);

        This example counts down from 9 to 0:

        DO VAR i;
            FOR (i=9, -1, -1) print(i);

        Special attention should be paid to the limits of the FOR loops
        in these examples. They always specify the first value which
        will NOT be applied to the statement. Another way to write the
        second example would be the following one, where the FOR loop
        is replaced by a WHILE loop:

        DO VAR i;
            i := 9;
            WHILE (i > -1) DO
                i := i-1;

        The meaning of this program fragment is completely equal to the
        one employing a FOR loop, but the syntax of the FOR statement is
        more compact and expresses the purpose of the statement clearer.

        The step value is optional in FOR statements. In the short form
        of the statement, it is omitted. If only two operands are
        specified in FOR, the step width defaults to one. Therefore, the

        FOR (j=0, 100, 1) print(i);


        FOR (j=0, 100) print(i);

        have exactly the same meaning.

        2.7.5 Branch Statements

        A branch passes control to a specific point in a program.
        Typical destinations for branch commands are the beginnings or
        the ends of loops or the ends of procedures or programs. There
        is no branch command with a freely definable destination like
        Goto in BCPL.

        The LEAVE command causes the immediate termination of the
        innermost WHILE or FOR loop. There are no operands to LEAVE.

        The following code compares the characters in two strings A and
        B. The loop is left at the first position where the strings
        differ, but in any case after 100 steps:

        FOR (i=0, 100) IF (a::i \= b::i) LEAVE;

        The loop is set up for 100 passes and the LEAVE statement makes
        the loop terminate as soon as a mismatch is found.

        The LOOP command transfers control to the beginning of the
        innermost loop. Like LEAVE, it has does not have any operands.
        If LOOP is used inside of a FOR loop, it branches to the
        increment part where the value of the index variable is
        modified. In WHILE loops, it branches directly to the point
        where the exit condition is checked.

        The following statement prints the numbers from 1 to 100, but
        skips over those that are multiples of 7:

        FOR (i=1, 101) DO
            IF (i mod 7 = 0) LOOP;

        To return from a procedure, the RETURN statement may be used. It
        has the general forms

        RETURN expression;



        The statement performs a branch to the end of the procedure,
        where local storage is released and the procedure is left. It
        the given expression, if any, and passes its value back to the
        calling procedure. The value received by the calling procedure
        is the value of that expression:

        square(x) RETURN x*x;

        Q() DO VAR y;
            y := square(5);

        In this short example program, 5 is passed as an argument to the
        procedure 'square'. The procedure computes the square of its
        argument and returns it to Q where the result (25) will be stored
        in y. In 'square', the RETURN statement is the last statement in
        the procedure, but in fact RETURN can be used anywhere in a

        When no expression is specified after RETURN, zero will be


        is the short form of

        RETURN 0;

        All of the above branch statements take care of locally
        allocated storage. If local symbols are defined in the bodies
        of loops, for example, LOOP and LEAVE will release this storage
        before branching to their respective destinations. This allows
        the use of these commands in any loop context, even if local
        symbols are present.

        The HALT statement with the general forms

        HALT constant-expression;



        branches to the end of an entire program.

        If necessary, the command cleans up the runtime environment of
        the program. The value of the specified constant expression is
        returned to the calling process. Only the least significant
        eight bits are guaranteed to be returned to the caller.

        The argument of HALT may be omitted. In this case, zero will be
        delivered to the caller.

        2.7.6 Compound Statements

        A compound statement (sometimes also called a block statement or
        statement block) is a group of statements which is treated like
        a single statement under some aspects. For example, a compound
        statement may occur at any place where a simple statement is
        expected. In commands like

        IF (expression) statement

        a compound statement can be used to extend the scope of the
        conditional statement so that it is applied to a group of
        statements instead of a single statement:

        IF (a < '0' \/ a > '9') DO VAR b::3;
            u.printf("Not a valid digit: %C\n", [(a)]);
            RETURN -1;

        In this example, both the u.printf() message and the RETURN
        statement will processed only, if the IF-condition applies. (The
        concept of sending messages will be explained in detail in the
        chapter about object oriented T3X.) The keywords DO and END are
        used to delimit the statement block. There is no terminating
        semicolon after a compound statement. The line

        DO p(); q(); END ;

        would be recognized as a compound statement containing the
        procedure calls P() and Q() followed by an empty statement
        consisting of a single semicolon.

        In T3X, compound statements are ordinary statements and they
        may occur at any place where a statement is expected. Even
        statements like


        are perfectly valid. The use of compound statements in sequences
        like the above becomes clear in the next sections where the
        allocation of local storage in compound statements is explained.

        2.7.7 Local Symbols

        Besides the grouping of commands, compound statements provide a
        mechanism for the definition of local symbols and the allocation
        of dynamic storage. Declaration statements already have been
        explained in a previous section. All data objects which can be
        created globally in T3X may also be declared locally inside of
        compound statements by placing their declarations at the
        beginning of a statement block. Any number of declarations will
        be accepted after the keyword DO.

        The declaration statements themselves do not change when used in
        local contexts. Only the position inside of a statement block
        makes the declared symbols local to that block. The statement

        DO VAR i; FOR (i=0, 10) print(i); END

        for example, applies the procedure 'print' to the numbers from
        0 to 9. The index variable is declared inside of a compound
        statement that also contains the FOR loop generating the
        sequence. The variable I does not exist before the compound
        statement is entered. It will be created automatically at the
        point of its declaration and it will cease to exist at the end
        of the block in which it has been declared. Therefore, variables
        which are local to compound statements are sometimes also called
        'automatic variables'.

        Atomic variables, vectors, structures, constants, and objects
        may all be declared locally. Unlike BCPL, T3X does not support
        nested procedure definitions, though. In the cases of atomic
        variables and vectors, the storage required by the variables is
        allocated when the symbol becomes valid and released when the
        variable is destroyed. Automatic storage will be allocated on
        the runtime stack.

        To illustrate another application of local storage allocation,
        imagine the following situation:

        P() DO
             VAR     big_V[VERY_LARGE_1];
             VAR     big_W[VERY_LARGE_2]; ! Too big


        In this procedure, two tasks requiring large amounts of storage
        shall be run sequentially, but not enough memory for both arrays
        is available. One solution would be the creation of two
        procedures where each one creates local storage for only one of
        the tasks. Another one would be to share the vector, but both
        solutions only work at the cost of readability. T3X provides
        another solution, since the compiler guarantees that local
        storage is allocated exactly at the point of its declaration and
        released immediately at the point of the destruction of the
        associated symbol:

        P() DO
            DO VAR big_V[VERY_LARGE_1];
            END ! big_V gets released here

            DO VAR big_W[VERY_LARGE_2];
            END ! big_W gets released here

        Since compound statements may be nested, naming conflicts may
        occur in many languages, like the following example (in C)

        { int x;
            x = 123;
            { int x;
                x = 456;
                printf("%d\n", x);
            printf("%d\n", x);

        The variable X is defined in the inner and outer block statement
        and assigned two different variables. So the variables has
        different values in the inner and outer block. The inner
        variable X is said to 'shadow' the outer X. The outer instance
        of X becomes invisible inside of the inner block. Therefore, the
        program fragment would print first 456 and then 123 when

        T3X uses stricter scoping rules than most other languages:
        Symbols generally may not be redefined in T3X programs. This
        also applies to global symbols (symbols which are declared at
        the top level, outside of procedure definitions, classes, or
        statement blocks). Hence, shadowing can never happen in T3X. The
        flexibility of local symbols remains, though, since names can be
        reused as soon as a local data object has been destroyed:

        F(x,y) DO VAR i, j;
            ! ...

        G(x,y) DO VAR i, j;     ! The names x,y,i,j are re-used
            ! ...

        As shown in this example, symbol names may be reused in procedure
        definitions (for formal argument names) as well as in subsequent
        compound statements. Since the variables i and j will be
        destroyed at the end of the compound statement forming the body
        of F, they can be reused in G. The same is valid for the
        argument names x and y.

        The following example shows some local and global symbols and
        their scopes.

         +++ VAR     GX, GY;
          |  CLASS A()                                 +++
          |    STRUCT C = R,G,B;                        | C
          |                                       PROC  | L
          |    P(x, y)                             +++  | A
          |      DO VAR x1, x2;               +++   |   | S
          |      END                          ---  ---  | S
          |  END                             BLOCK     ---
        G |  P(x, y) DO VAR x1, y1;                +++
        L |                                         |
        O |      STRUCT  PT = PX, PY;               |
        B |                                  BLOCK  |
        A |      DO VAR i, j;                 +++   | P
        L |          DO VAR x2, y2;      +++   |    | R
          |          END                 ---   |    | O
          |                                    |    | C
          |          DO VAR x2, y2;      +++   |    | E
          |          END                 ---   |    | D
          |      END                          ---   | U
          |                                         | R
          |      DO CONST t=%1, f=~t;         +++   | E
          |          DO VAR x2, y2;      +++   |    |
          |          END                 ---   |    |
          |      END                          ---   |
         --- END                                   ---

        Fig.1 Scopes (example)

        Like all symbols, the global variables GX and GY are valid from
        the point of their declaration, but unlike locally declared
        names, they remain existent up to the end of the program. Their
        scope is the entire program, from the point of their declaration
        to the end of the file. The scopes of all symbols in the example
        are illustrated using vertical bars. Plus signs indicate the
        point where a symbol name becomes valid and its storage is
        allocated, and minus signs mark the point of its destruction.

        Note: the names X2 and Y2, which are used in different scopes,
        denote different variables. A value stored in X1 within the
        first scope, for example, cannot be retrieved in the second or
        the third scope from X1, because the name references different
        locations in different scopes. The variable that is created at
        the beginning of the first scope containing X1 is deleted at the
        end of this scope and the value stored in that variable is lost.
        Assignments to local variables only remain valid between two
        connected +++ and --- indicators.

        All symbols that are defined in a so-called 'class contexts'
        (between the keywords CLASS and the matching scope terminator
        END) are only valid inside of this context. Both, the structure
        C and the procedure P defined in the class A are only valid
        inside of the class context of A. There exists no conflict
        between A.P the method P of A and the procedure P which is
        defined at the top level. Class contexts will be discussed in
        detail in the section on object oriented programming in T3X,
        later in this document.

        2.7.8 Empty Statements

        There are two forms of the 'empty statement' (aka 'null
        statement') in T3X.  The first form is the single semicolon


        and the second one is the empty compound statement:

        DO END

        Both null statements have absolutely no effect. Their only
        purpose is to fill a gap where a statement is required, but
        nothing is to do. They are useful to negate the meanings of
        complex conditions, for example. Instead of negating the
        condition at the cost of making it harder to understand, one
        might turn

        IF (\(complex-condition)) statement


        IE (complex-condition)

        2.8 Procedures

        Each procedure may be considered to be a separate small program.
        It communicates with other procedures using parameters and
        return values and/or through global variables. Each procedure
        has access to all global data objects which have been declared
        before itself. Generally, it is considered good style in
        procedural languages to keep procedures self-contained and use
        global storage as little as possible, but when data has to be
        shared between a big number of different procedures, the use of
        top-level definitions is very common and more efficient.

        The definition of a procedure has only one single form in T3X.
        Since there is no support for nested routines, all procedure
        declarations and definitions must occur at the top level (the
        space between the other global declarations) or in class
        contexts. Public procedures declared in class contexts are
        called methods. They will be explained later.

        The only form of the procedure definition is

        name(a1, ... aN) statement

        where 'name' is the name of the procedure, 'a1', ... 'aN' are
        the names of its formal arguments, and 'statement' is the body
        of the procedure - the part which describes what it does.

        The procedure name may be any valid symbol and it is declared in
        the global context. Therefore, procedure names may not be reused
        ever. One advantage of T3X's strict scoping rules is that
        procedures cannot get shadowed. The arguments 'a1',... 'aN' are
        local to the procedure (not local to the statement forming its
        body). Their names will cease to exist after the statement has
        been accepted. Hence, they may be reused after the procedure
        declaration, but not inside of it. The parentheses around the
        argument list must always be specified, even if the list is

        name() statement

        The number of arguments specified in a procedure declaration
        determines the type of the procedure. The type of a procedure is
        an integer number that specifies the number of its arguments. In
        T3X, the argument counts of all procedure calls will be checked
        against the procedure type. The compiler will not allow calls
        with a wrong number of parameters. This is done because of T3X's
        calling conventions: parameters are passed from the left to the
        right, which places them on the stack in reverse order. Each
        procedure must on receive the correct number of arguments, or
        rhe mapping from local addresses to arguments does not work.
        BCPL and C, on the other hand, pass arguments from the right to
        the left, which allows to compensate for missing or superfluous
        procedure parameters. The only real advantage of this approach,
        however, is the option of defining variadic procedures -
        procedures with a variable number of arguments. Variable
        argument lists can also be realized in T3X, but using a
        different mechanism that will be explained later.

        When a procedure is called, it can receive data through its
        arguments. This works in the following way. Given a procedure

        P(x, a, b, c) RETURN a*x*x + b*x + c;

        and a procedure call

        Q() VAR y; y := P(2, 3, 5, -7); END

        the caller (Q) places the values of the actual arguments 2, 3,
        5, and -7 in a temporary storage location on the runtime stack,
        saves the address of the following operation (in this case the
        assignment) and then transfers control to the procedure P. In P,
        the formal arguments x, a, b, and c reference storage locations
        which exactly match the temporary locations of the values passed
        to the routine, so that X=2, A=3, B=5, and C=-7.

        The procedure P then computes a*x*x+b*x+c and returns the
        resulting value to the caller. Each procedure returns
        automatically when its body has been evaluated completely or
        when an explicit RETURN statement is executed. In the above
        example, both happens at the same time. It is not unusual to
        specify a RETURN statement at the end of a procedure, since only
        RETURN may pass an explicit value back to the caller. Procedures
        that do not return through RETURN have an implicit return value
        of zero. In the example above, though, the value of P is
        explicitly specified. After passing control back to the caller,
        the assignment takes place, and the result of the procedure call
        is stored in Y. Between the procedure return and the subsequent
        assignment, the temporary storage where the actual arguments
        were held is released again.

        The most frequently used form of the procedure has a body
        consisting of a compound statement:

        fib(n) DO VAR f, i, j, k;
            f := 1;
            j := 1;
            FOR (i=1, n) DO
                k := f;
                f := j;
                j := j+k;
            RETURN f;

        Note: The variables declared at the beginning of the procedure

        VAR f, i, j, k;

        belong to the compound statement rather than to the procedure.
        Like in conditional statements and loops, the statement block is
        used to extend the scope of the procedure: not only a single
        statement, but a group of statements forms the body of the

        2.8.1 Recursive Procedures

        It is perfectly safe for a procedure to call itself. Since the
        declaration of a procedure takes place while parsing its head
        (consisting of its name and its argument list), the declaration
        is already in place, when the compiler processes the body.
        Therefore, a procedure may recurse:

        factorial(n) RETURN n = 0-> 1: n * factorial(n-1);

        This small example computes N! or 1 * 2 * ... * N. For the
        trivial case N=0, it simply returns 1. To compute N! where N>0,
        it first computes (N-1)! and then multiplies the result by N.
        To compute the factorial of N-1, it calls itself. Since the
        value of the argument of the recursive call is decremented by
        one at each level of recursion, it will finally reach 0 and the
        procedure will start returning.

        Recursion is safe in T3X, because local variables (which include
        formal arguments) are created freshly each time a declaration is
        passed. Therefore, the symbol N in the above example denotes
        different variables at each level of recursion. The following
        program is a modified factorial procedure that illustrates this

        MODULE visual_fac(util);

        OBJECT  u[util];

        fac(n) DO VAR b::30;
            IE (n = 1) DO
                    u.printf(" 1", 0);
                    RETURN 1;
            ELSE DO
                    u.printf(" %D *", [(n)]);
                    RETURN n * fac(n-1);

        DO VAR b::80;
            u.printf("fac(7) =", 0);
            u.printf(" = %D\n", [(fac(7))]);

        When executed, this program will print

        fac(7) = 7 * 6 * 5 * 4 * 3 * 2 * 1 = 5040

        It is left as an entertaining exercise for the reader to find
        out how the process works.

        Of course, the usual caveats concerning the use of global memory
        and other shared resources in recursive procedures also apply in

        2.8.2 Mutually Recursive Procedures

        Recursive procedures which depend on each other are called
        'mutually recursive'. Mutual recursion introduces the following
        problem: Given the procedures

        A(x) IF (x > 0) B(x-1);

        B(x) IF (x > 0) A(x-1);

        which depend on each other, it does not matter which one is
        declared first - one will always be inaccessible from within the
        other. In the above example, B is undefined in A because it is
        declared after A. When swapping the definitions, A will become
        undefined in B.

        The problem is solved by introducing 'forward declarations',
        which may introduce a procedure before actually declaring it.
        A forward declaration makes a procedure symbol known to the
        compiler, but does not associate any meaning with with it.
        To forward-declare a procedure, the DECL statement is used:

        DECL name(type);

        Like in most declaration statements, any number of
        comma-separated declarations may be included in a single DECL
        statement. 'Name' is the name of a procedure to declare and
        'type' is a constant expression specifying the number of formal
        arguments of that procedure. This value is required to type
        check forward calls to the procedure. The number of formal
        arguments in a subsequent declaration must exactly match the
        type specified in the forward declaration, or a redefinition
        error will be signalled.

        Each names used in a forward declaration may only be re-used in
        one single procedure declaration. Declaring a procedure without
        defining it later is an error, since this may leave forward
        references to the forward-declared procedure unresolved.

        To correct the above program fragment containing the mutually
        recursive procedures A and B, a forward declaration of B has to
        be inserted before the declaration of A:

        DECL B(1);
        A(x) IF (x > 0) B(x-1);
        B(x) IF (x > 0) A(x-1);

        Like procedure definitions, DECL statements are only allowed at
        the top level and in class contexts, but not inside of local

        2.8.3 Variadic Procedures

        All procedures have fixed numbers of arguments in T3X. It is
        possible, however, to pass a variable number of arguments to a
        procedure using a dynamic table. The following simple example
        computes the average of N values stored in the vector V:

        average(n, v) DO VAR i, k;
            k := 0;
            FOR (i=0, n) k := k + v[i];
            RETURN k/n;

        Since vectors are first-class objects in T3X, it is possible to
        inline vectors in procedure applications, thereby forming an
        elegant way of passing a variable number of values to a

        average(5, [ 2, 3, 5, 7, 11 ]);
        average(3, [ (fib(10), fac(5)), 789 ]);

        Another example is illustrated below. The T3X method PRINTF
        of the utility class UTIL is a variadic procedure that
        implements of a subset of of the standard C library procedure
        printf(). The FIB procedure has been defined earlier in this
        chapter. Using these procedures, it it possible to write the
        following program that prints the line

        fib(n) = m

        for each N = 1,... 20 and M = fib(n):

        MODULE print_fib(util);

        OBJECT u[util];

        DO VAR i;
            FOR (i=1, 20)
                u.printf("fib(%D) = %D\n", [(i, fib(i))]);

        UTIL.PRINTF replaces each %D in its first argument with the
        readable representation of one of the value in the table forming
        its second argument. Each time a %D is processed, the procedure
        advances to the next argument. A C version of the program would
        use a variable number of arguments while the T3X version uses a
        dynamic vector to transport a variable number of values.

        BTW: both printf() and UTIL.PRINTF use the number of %-patterns
        to determine the number of arguments passed to it.

        2.9 The T3X Object Model

        Like many popular object oriented languages T3X is a hybrid
        language.  A hybrid language is a language incorporating (at
        least) two different paradigms. T3X uses the object oriented
        approach at a rather abstract level and the procedural approach
        at the lower levels. For example, numbers are no objects in T3X
        and adding numbers is not done by sending messages. In a purely
        object oriented language, the term

        5 + 7

        would be interpreted as send the message '+' with the argument
        '7' to the object '5'. In a procedural language, however, adding
        numbers is done by combining the factors '5' and '7' using the
        '+' operator. Interpreting numbers as objects and expressions
        as messages makes no sense in a procedural language, since
        numbers and operators are not implemented this way. There are
        many hybrid languages employing both the procedural and object
        oriented approach. For example, C++, Java, and Object Pascal fit
        in this category. A well-known member of the family of purely
        object oriented languages would be Smalltalk.

        2.9.1 Object Oriented Programming

        There are various definitions of the object oriented programming
        (OOP) paradigm. The term 'object oriented' has become an umbrella
        term for all kinds of approaches to abstraction and code reuse.

        At the very foundation, an object oriented language encapsulates
        code and data declarations in templates called 'classes'. A
        class contains data declarations, procedures, and public
        procedures ('methods') that form an interface to the data
        contain in the class. Each class can be instantiated by
        declaring an 'object' or 'instance' of that class. Each object
        contains all the data objects declared inside of its class. Note
        that the term 'object' is used to refer to instances of classes
        in this section. Variables and vectors are referred to as 'data
        objects'. Only procedures defined in its class may access the
        data contained in an object. No procedure that is not contained
        the class of an object may ever access data of the object.
        This principle is called encapsulation. It is a fundamental
        property of the OOP paradigm. In T3X, encapsulation cannot be
        broken or bypassed in any way. This property is called 'strict

        Each class may have multiple instances. In this case, each
        object of the class has its own private data area that is
        separate from the data areas of other instanced of the same
        class. This is why classes can be reused. Manipulating the data
        of one object has no effect on other objects of the same class.

        Methods of classes are invoked by sending messages to objects of
        that class. A message is similar to a procedure call. It may
        transport arguments 'into' the object and the method may return
        a value to the caller. Since the method is part of the class of
        the object, it is allowed to access all data objects inside of
        the object. Therefore, methods provide a clean and abstract
        interface to the data of the object. The data structure itself
        is hidden from the user and may change without changing the

        Other OO languages define additional concepts like inheritance,
        protected and public variables, friend relationships, class
        variables, etc, but the T3X object model is limited to

        - Classes
        - Objects
        - Messages

        This is all that is needed to define reusable programs at a high
        level of abstraction.

        2.9.2 Classes

        The general forms of the class declaration are as follows:

        CLASS classname() declarations END

        CLASS classname(required, ...) declarations END

        The context of a class is delimited by the class header
        (consisting of the keyword CLASS, the name of the class, and the
        dependency list in parentheses) and the keyword END. Inside of
        this context, there may be any number of declarations. These
        types of declarations are allowed inside of class contexts:

        - Variables
        - Constants
        - Structures
        - Public constants
        - Public structures
        - Procedures
        - Public procedures (methods)
        - Forward declarations
        - Objects

        Nested classes are not defined in the T3X object model.

        All declarations between CLASS and END are local to the class.
        Therefore, classes add an additional level of scoping between
        the global level and the procedural level. All data objects and
        procedures declared inside of a class are only visible inside of
        that class. The names of entities declared at class level may be
        reused outside of the scope of the class. Hence different
        classes may define data objects, procedures and even methods
        with equal names. The following example illustrates this

        CLASS a()
         VAR  flag;

         PUBLIC flip() flag := \flag;
        ! the scope of class A ends here.

        CLASS b()
         VAR flag;

         PUBLIC flop() flag := \flag;

        This example defines two classes A and B each defining a
        variable named 'flag'. At the end of the scope of A, all

        declarations of A become invisible (encapsulated) and so the
        name 'flag' may be reused in B. Since the procedure 'flip' is
        contained in the same scope as 'flag', it may access the 'flag'
        of A. In the same way, the procedure 'flop' may access the
        'flag' of B. The two variables named 'flag' are different
        entities, though, since they are contained in different classes.

        Like structures, classes are merely templates for data objects.
        They describe the layout of a data structure plus a set of
        methods that may be used to access elements of the structure.
        The size of a class is computed in the same way as the size of a
        structure: it is equal to the sum of the sizes of all class
        members. In expressions, the name of a class is a constant
        evaluating to the size of the class. Classes without any
        instance variables have a size of one machine word.

        The only way to change the state of a class from the outside is
        to send it a message. T3X supports a simplified form of the
        method called a 'class constant'. Class constants may be thought
        of as lightweight methods returning a constant value. They allow
        to export values and structures without having to send a full
        message. OO systems that do not allow to change the state of an
        object without sending a message are said to employ strict
        encapsulation. Strict encapsulation in T3X is illustrated in the
        following figure.

        |                                            |
        |               M e t h o d s                |
        |                                            |
        |              +--------------+              |
        |              |  Variables   |              |
        |              +--------------+              |
        |              |  Constants   |              |
        |              +--------------+              |
        |--------------|  Structures  |--------------|
        |              +--------------+              |
        |              |  Procedures  |              |
        |              +--------------+              |
        |              |  Objects     |              |
        |              +--------------+              |
        |                                            |
        |        C l a s s   C o n s t a n t s       |
        |                                            |

        Fig.2 Strict Encapsulation

        No data objects of a class are visible from the outside. Only
        methods (including class constants) are visible.

        2.9.3 Objects

        Objects are used to instantiate classes. An OBJECT statement is
        to a CLASS declaration what a VAR statement is to a STRUCT
        declaration. While the class defines the layout of an object,
        the OBJECT statement actually creates an object in memory. The
        general form of the object definition is as follows:

        OBJECT an_object[a_class], ... ;

        Any number of objects may be defined in a single OBJECT
        statement. Each class may be instantiated any number of times
        and different classes may be instantiated in the same statement.
        The name of the object to create is specified before the square
        brackets and the class of the object inside of the brackets:

        OBJECT str[string];

        creates an object of the class STRING named STR.

        An object may be a factor in an expression. It evaluates to the
        address of its first member. The notations




        are equivalent.

        When creating multiple instances of a class, only space for the
        data declarations of the class is created. The methods of a
        class belong to the class itself rather than the object. They
        are created when a class is declared.

        The only way to alter the state of an object is to send it a
        message. Therefore, in an ideal OO program the state of each
        object is completely independent from the states of other
        objects, even if they belong to the same class. In a hybrid
        language like T3X, however, the procedures of a class may change
        data objects defined in the global scope. Changing a global
        object from within an object changes the state of all other
        objects of the same class (and maybe even the state of objects
        of other classes). Therefore, this technique is deprecated. Of
        course, there are situations where an object has to change the
        global state, for example when performing input/output
        operations. Classes defining such objects are said to have
        'global effects'.

        2.9.4 Modules and Class Dependencies

        In order to create an instance of a class A inside of a class B,
        the class B must require the class A. In this case, B is said to
        depend on A. The simplest scenario contains two classes which
        are defined inside of the same file:

        CLASS a()
         ! definitions of A

        CLASS b(a)
         OBJECT xa[a];

        Since B instantiates A, it must require A. A class is required
        by including its name in the dependency list of the class header
        of the dependent class. Requiring a class has two effects:

        - it embeds information about the required class in the name
          space of the dependent class

        - it allows procedures of the dependent class to send messages
          to instances of the required class

        Things get a little more complex, if the dependent class and the
        required class are located in different files. Since class names
        are contained in the global scope, they are lost as soon as the
        compiler has finished the translation of the file in which they
        are contained. Hence the required class would be unknown when
        the compiler translates the file containing the dependent class.
        To allow classes to be located in different files, an additional
        level ABOVE the global scope is added. It is called the 'public
        scope' and it persists even when compilation of a file finishes.

        To add a class to the public scope, two steps are necessary.
        First, the file must have a module header which names the file.
        A MODULE statement has the following general form:

        MODULE module_name (required, ...);

        The module name specified in the module header must be the same
        as the actual name of the file containing the module (but not
        including the .t suffix). If, for example, the class A is
        located in a file named tools.t, the module header would look
        like this:

        MODULE tools();

        The parentheses after the module name serve the same purpose as
        in class declarations: they delimit a dependency list. This list
        may be ignored for now. The module header allows the compiler to
        locate the definition of a class, even if multiple classes are
        contained in a single file. If files were named after classes,
        only a single public class could be declared per file.

        The second step required to export a class to the public level
        is to prefix its class header with the keyword PUBLIC:

        PUBLIC CLASS a()

        The following example shows the contents of two files 'file_a'
        (containing the required class A) and 'file_b' (containing the
        dependent class B). When compiling file_a, A is exported to the
        public level. When compiling file_b, A is imported from the
        public level when it is required by B.

        MODULE file_a();                MODULE file_b();
        PUBLIC CLASS a()                CLASS b(a)
         ! definitions                   OBJECT xa[a];
        END                             END

        Another purpose of the MODULE statement is to provide an
        interface between the procedural and the object oriented parts
        of T3X. If only classes could require classes, it would be
        impossible to instantiate a class inside of a T3X program, since
        the main program is a procedure. To instantiate a class in a
        procedural program, it is added to the dependency list of the
        module wishing to instantiate the class. It does not matter
        whether the required class is a public class contained in a
        different module or a 'private' class contained in the same
        module. In the latter case, however, the MODULE statement must
        be located after the class declaration. The following example
        shows a procedural program sending a message to a class.

        CLASS A(t3x)       ! class A depends on the T3X core class
         OBJECT t[t3x];    ! instantiate T3X class

        ! The method M will print some text
        ! using the WRITE method of the T3X class
         PUBLIC m() DO
           t.write(T3X.SYSOUT, "A: received m.\n", 15);

        MODULE main(A);    ! the main module requires A,

        DO OBJECT xa[A];   ! instantiates it,
          xa.m();          ! and sends a message to it

        2.9.5 Methods and Messages

        Messages are used to access the data of an object from the
        outside, which includes altering its state. Procedural programs
        are sets of procedures calling each other. Object oriented
        programs are a set of objects sending messages to each other.
        Sending a message to an object activates a public procedure
        defined in the class of the object. A method definition looks
        like a procedure definition with the keyword PUBLIC attached:

        PUBLIC name(arguments) statement

        Method definitions are only valid in class-level scopes.

        A message is sent to an object using the syntax


        Messages may be factors in expressions or standalone statements.
        When used as statements, they must be terminated with a


        The arguments of a method are passed in the same way as the
        arguments of a procedure and, like a procedure, a method returns
        a value. The difference between an 'ordinary' procedure and a
        method is that a method changes the 'instance context' upon
        entry. The instance context points to an instance of the set of
        data objects defined in a class. It is comparable to local
        contexts of procedures: when a procedure is entered, it creates
        a new local scope and when it leaves, it restores the caller's
        context. Unlike a local scope, the instance context is
        persistent, though. Therefore, methods do not create a new
        instance context, but just activate an existing one. The
        caller's context is saved upon entry on the runtime stack and
        restored when the method returns. Since instance contexts are
        persistent, the changes performed by methods are permanent.

        Each object has its own instance context, which is divided into
        the data objects declared in its class. Methods use the instance
        context to access class-level data. By changing the instance
        context upon entry, each object accesses only its own private
        data. The instance context may be thought of as a multiplexer.
        The principle is illustrated in the following figure:

           Class A                      Access V
        +------------+                      |
        |  Method M  |                      |
        |    ...     |                      V
        | Variable V |             +----------------+
        |    ...     |             |    Method M    |
        +------------+             +----------------+
                                 |  Instance Context  |
                                        |      |
                                        |      |
                               ,--------'      '--------,
                               |                        |
                               V                        V
                       +---------------+         +---------------+
                       |   V of X[A]   |         |   V of y[A]   |
                       |      ...      |         |      ...      |
                       +---------------+         +---------------+
                          Object X[A]               Object Y[A]

        Fig.3 Multiplexing Method Applications

        In this figure, class A defines a method M which accesses the
        instance variable V that is also declared in A. The instance of
        V accessed by M depends on the object to which the message is
        sent. Sending X.M(), results in accessing the V of X and sending
        Y.m() results in accessing the V of Y.

        In T3X, the current instance (the currently active instance
        context) can be referred to using the symbol SELF. SELF is a
        pseudo-variable that always refers the object owning the current
        instance context. Therefore, SELF may only be used inside of
        procedures local to classes. Using SELF, an object may send a
        message to itself, as shown in the next example

        CLASS math()
         PUBLIC prod(i, j) DO VAR p;
             p := 1;
             FOR (i=i, j+1) p := p*i;
             RETURN p;

         PUBLIC fac(n) RETURN, n);

        In this example, the method 'fac' of the class 'math' uses the
        method 'prod' of the same class to express the factorial of N by
        sending the message prod(1,n) to itself. Methods may recurse,
        too, since they are basically procedures. Therefore, 'fac' could
        as well be defined in this way:

        PUBLIC fac(n) RETURN n < 1-> 1: n * self.fac(n-1);

        Because objects are basically vectors, they may be passed to
        procedures (or methods) as parameters. By passing an object to a
        procedure, however, the object loses its type information, since
        the pointer to the object is stored in a typeless argument
        variable by the callee. To be able to send messages to such
        objects, the SEND operator is introduced. Its general form is

        SEND(variable, classname, methodname(arguments))

        This operator sends the message methodname(arguments) to the
        object of the class 'classname' pointed to by 'variable'. For
        example, the statement

        y := m.fac(5);

        is equal to

        pm := @m;
        y := SEND(pm, math, fac(5));

        The caveats regarding indirect calls to procedures via CALL also
        apply to the SEND operator. In addition to providing the correct
        number of arguments, the programmer also has to make sure that
        the message is sent to the right type of object.

        2.9.6 Class Constants

        Constants may be public as well:

        PUBLIC CONST symbol = constant_expression;

        Such constants can be accessed from outside the class by sending
        a special form of a message to the class which defines the
        constant. Given the constant MAXLEN of the class STRING:

         PUBLIC CONST MAXLEN = 32767;

        the expression


        would be used to access the value of MAXLEN. So the general
        form of the class constant access is


        The 'classic' way of exporting such a constant would be to
        define a method returning the constant:

        PUBLIC maxlen() RETURN 32767;

        Class constants have the advantage of saving a procedure call.
        They can also also be used in constant expression contexts,
        because their values are known at compile time.

        Structures can be exported in the same way as constants:

        PUBLIC STRUCT structname = member1, ..., memberN;

        Public structures are useful for objects requiring arguments in
        structured form or returning values in this form. For example,
        the SYSTEM.STAT function (the function STAT of the SYSTEM class)
        returns a structure containing information about a specific
        file. In addition, the SYSTEM class provides a public structure
        describing the layout of the structure returned by the STAT
        function. This structure allows programs using the SYSTEM.STAT
        function to decompose the returned information.

        Since class constants cannot be altered, they do not break the
        strict encapsulation principle. Public structures are an
        interface and an implementation at the same time. Since the same
        PUBLIC STRUCT statement is used to define the same structure
        internally and externally, the interface changes automatically
        when the implementation changes.

        2.10 Interface Classes

        An interface class is a class depending on an external object
        file called an extension object. Extension objects may be linked
        against the Virtual Tcode Machine or against native code
        generated by the T3X compiler.

        The declaration of an interface class begins with an ICLASS
        statement. This statement has the following general form:

        ICLASS class_name ("extension_object_name")

        Class_name is the name of the class to declare and
        'extension_object_name' is the name of the object file holding
        the code of the interface class. The name of the object file
        should be specified without any suffixes such as '.o' or '.lib'.

        Like other class contexts, interface class contexts are
        terminated with the keyword END.

        2.10.1 Interface Declarations

        In addition to the declarations allowed in class contexts,
        interface classes may contain so called interface declarations.
        An interface declaration describes an interface procedure
        contained in an extension object. Any number of interface
        procedures may be declared in a single IDECL statement:

        IDECL name(type, call_map), ...;

        'Name' is the name of an interface method to declare and type is
        the number of arguments of that method. 'Call_map' describes the
        types of the parameters passed to the interface procedure. It
        is a bit map where each bit is associated with a procedure
        argument as outlined in the following table.

        Argument   Call Map   Argument   Call Map
        Number     Value      Number     Value
        --------   --------   --------   --------
            1       0x0001        9       0x0100
            2       0x0002       10       0x0200
            3       0x0004       11       0x0400
            4       0x0008       12       0x0800
            5       0x0010       13       0x1000
            6       0x0020       14       0x2000
            7       0x0040       15       0x4000
            8       0x0080       16       0x8000

        Each bit with a value of zero denotes an atomic (numeric)
        argument and each bit that is set to one denotes a vector
        argument (a pointer in the argument list of a corresponding C
        function). For example, the C function _spawn() of the
        extension object system could be defined as follows:

        xcell system__spawn S3(prog, args, wait)
        char  *prog;
        char  **args;
        xcell wait;
             /* code of system__spawn() */

        The S3() macro is used to generate argument lists of three
        arguments. Similar macros exist for declarations of functions
        with up to 7 arguments. They are called S0(), ..., S7(). The
        Sn() macros reverse the supplied argument lists to meet the T3X
        calling conventions and supply a dummy argument to intercept the
        additional instance context parameter that T3X programs pass to
        each method.

        Do use the Sn() and SELF macros. Otherwise, interface procedures
        will not work.

        'Xcell' is a macro expanding to the type of a signed cell
        (integer) of the same size as a pointer.

        All macros discussed here are defined in the 'txx.h' header
        file. The file 'txx.h' also contains some other macros which are
        useful for implementing interface procedures.

        The first and second argument of _spawn() are pointers and so
        the bits 0x0001 and 0x0002 in the call map have to be set. This
        leads to the following interface declaration:

        ICLASS system("system")
         IDECL _spawn(3, 0x0003);

        The call map is required to translate vector addresses to native
        pointer size before passing them to interface procedures. Call
        maps are limited to 16 bits, so interface procedures may not
        have more than 16 arguments.

        NOTE: when passing vectors of pointers to interface procedures
        (like 'char **args' in system__spawn()), the vector of pointers
        must be converted to native pointer size as well. The T3X core
        method T3X.CVALIST performs this operation.

        2.10.2 Calling Interface Procedures

        To the T3X programmer, an interface procedure is a method of an
        interface class. In order to invoke an interface procedure, the
        interface class is instantiated a message is sent to the
        resulting object:

        ICLASS world("world_code")
         IDECL hello(2, 1);

        MODULE test(world);

        OBJECT aworld[world];

             aworld.hello("Hello, world!\n");

        2.10.3 A Sample Interface

        The HELLO method used in the previous subsection could be
        implemented as follows:

        /* world_code.c */
        #include <txx.h>
        #include <string.h>

        xcell world_hello S2(s, n)
        char    *s;
        xcell   n;
            while (n--) {
                write(1, s, strlen(s));
                write(1, "\n", 1);
            return 0;

        The proper arguments to compile this code to an extension
        object depends on the host environment. However, the following
        requirements must be met:

        - the code must be compiled to a relocatable object file

        - the include path of the txx.h file must be specified

        - the LIBRARY macro must be defined, if compiling to an
          extension object

        - the LIBRARY macro must be undefined, if the interface
          procedures in the module will be linked against the Virtual
          Tcode Machine

        To compile the above interface code to an extension object on a
        generic Unix system, a command like this may be used:

        cc -o world_code.o -I /usr/local/include -DLIBRARY \
           -c world_code.c

        When compiling an object to be linked against TXX, -DLIBRARY
        should be omitted.

        2.11 Scoping Rules

        Scoping rules define the contexts in which symbols are valid and
        under which conditions they may be redefined. In T3X, there are
        five different contexts and very strict and simple redefinition

        (1) The public context contains public classes and the public
        entities defined in their contexts. It contains all public
        elements exported by a set of modules. The public context is
        persistent. It is not even destroyed when the compilation of a
        program ends. The public context may contain any number of
        global contexts.

        Technical point: To purge the public symbol table, the storage
        holding the table (usually a file) must be cleared (e.g. by
        deleting the file or truncating it to zero length).

        (2) The global context covers a complete file or module. Its
        declarations are located in the space between procedures and
        class definitions. Each global context may contain any number of
        class contexts or procedure contexts.

        (3) A class context contains all symbols that belong to a
        specific class. They are delimited by a class header (CLASS ...
        or ICLASS ...) and the keyword END. Each class context may
        contain any number of procedure contexts.

        (4) A procedure context is equal to the argument list of a
        procedure plus the statement forming the body of the procedure.
        The argument list is a list of implicit atomic variable
        declarations. No other entities may be defined in this context.
        A single block context (forming the body of the procedure) may
        be embedded in each procedure context.

        (5) A block context begins with the keyword DO and ends with the
        keyword END. Each block context may contain any number of nested
        block contexts. Notice that this definition is recursive, so that
        blocks may be nested to any level. A block context is called
        block^N context, if there are N enclosing blocks.

        The following redefinition rules apply:

        (1) Each public class and the entities defined in its context
        may be redefined once inside of the module containing the
        original definition.

        This happens when a module containing public classes is
        recompiled. In this case, the definitions in the public context
        are silently updated.

        (2) Each symbol used in a forward declaration (DECL statement)
        may be redefined by a one single matching procedure definition.

        (3) Except for forward declarations and public entities, no
        symbol may be redeclared or shadowed ever:

        - Global names may not be reused at class level, procedure
          level, or at block level.

        - Class level names may not be reused at procedure level or at
          block level.

        - Procedure level names may not be reused at block level.

        - Block level names may not be reused in embedded blocks.

        The following figures illustrates the different scopes.

        +-- Public ------------------------------------+
        | +-- Global --------------------------------+ |
        | | +-- Class -----------------------------+ | |
        | | | +-- Procedure ---------------------+ | | |
        | | | | +-- Block^0 -------------------+ | | | |
        | | | | | +--- ... ------------------+ | | | | |
        | | | | | |                          | | | | | |
        | | | | | | +-- Block^N -----------+ | | | | | |
        | | | | | | |                      | | | | | | |
        | | | | | | |                      | | | | | | |
        | | | | | | |                      | | | | | | |
        | | | | | | +----------------------+ | | | | | |
        | | | | | |  ...                     | | | | | |
        | | | | | +--------------------------+ | | | | |
        | | | | |  ...                         | | | | |
        | | | | +------------------------------+ | | | |
        | | | +----------------------------------+ | | |
        | | +--------------------------------------+ | |
        | +------------------------------------------+ |

        Fig.4 Scopes (overview)

        2.11.1 Scoping Conflicts

        At the end of a class context, all names contained in that class
        context as well as the classname itself will be removed from the
        global context. However, the class name will be memorized at a
        different location and may never be reused in the same program.
        This is a workaround for an ugly inconsistency resulting from an
        interference with the module system. Imagine the following

        CLASS A() ... END

        A() DO ... END

        CLASS B(A) ... END

        In this case, the class B would depend upon the procedure A
        which would be a type error. On the other hand, if the name A
        would persist, the following code would be correct:

        CLASS A() ... END

        CLASS B()
         OBJECT XA[A];

        In this case, class B could use class A without being dependent
        on it, which would break class dependencies, because the module
        extension requires that B be dependent on A in this case.
        Therefore, the (admittedly hackish) solution of not permitting
        the reuse of (deleted) class names has been chosen.

        2.12 Type Checking

        In T3X, there are only very few type checking mechanisms to
        detect things like assignments to constants, calls of
        non-procedures, etc. Including the 'class' and 'object' meta
        types, T3X has seven different types on which only specific
        operations are allowed. Type-related semantic checks carried out
        by the compiler intercept the following errors:

        - Assigning values to constants
        - Calling non-procedures
        - Calling procedures with incorrect argument counts
        - Creating objects of non-classes
        - Sending messages to non-objects
        - Sending non-messages to objects
        - Sending messages with incorrect argument counts
        - Creating dependencies on non-classes

        In a truly object oriented language with first class objects,
        however, more extensive type checking would be required. When,
        for example, an object is passed to or returned by a function,
        the passed value becomes a first class value which may be
        assigned to a variable. No type information would be associated
        with such a variable in T3X. Therefore, the compiler could not
        determine the set of messages accepted by this object. This
        problem has been solved by not allowing to send messages to any
        type of data object other than the object, but the SEND operator
        may be used to circumvent this restriction. Like the CALL
        operator, SEND has to be used with special care.

        The following table contains an overview over the operations
        which may be applied to each type or entity:

        Type / Oper.    Asg     Sub     Call    Inst    Recv    Send
        ----   -----    ---     ---     ----    ----    ----    ----
        Constant        no      no      no      no      no      no
        Variable        yes     yes     yes (+) no      yes (*) no
        Vector          no      yes     no      no      no      no
        Procedure       no      no      yes     no      no      no
        Class           no      no      no      yes     yes (#) no
        Object          no      no      no      no      yes     no
        Method          no      no      no      no      no      yes

        Asg = assignment, Sub = subscript, Inst = instantiation,
        Recv = receive messages, Send = send as message

        (+) Using the CALL operator
        (*) Using the SEND operator
        (#) Using class constants

        In addition, any of the above types can be evaluated, yielding a
        value. This is the value assigned to a constant, the content of
        an atomic variable, the size of a class or structure, and the
        address of each other kind of entity (vector, packed vector,
        procedure, method, object).

        2.13 Meta Commands

        Meta commands are used to control the behavior of the compiler.
        No code will be generated for meta commands. Meta commands do
        not even belong to the T3X language itself and different
        implementations may provide different sets of meta commands. The
        commands described in this section should be present in any
        implementation of the language, though.

        All meta commands begin with a hash sign (#) and like all other
        statements, they are terminated with a semicolon. They may occur
        at any place where a statement or a declaration (either local or
        global) is expected, but not inside of statements or
        declarations. The following meta commands exist:

        #CLASSPATH "path";

        This command specifies an alternative path for searching class
        files. Normally, the translator searches class files in the
        current working directory and in some compiled-in paths. When
        bootstrapping the compiler or running it in some other
        non-standard environment, it may be necessary to specify the
        class path explicitly using this command. Only one path may be
        specified. When using multiple #classpath commands, only the
        last one will take effect. This solution is a makeshift and will
        probably be combined with a more flexible technique in the
        future. When a classpath is specified using this meta command,
        it will take precedence over all other (built-in) class paths.


        Turn on emission of debug information like source code line
        numbers and variable names and addresses. When this option is
        turned on, the T3X translator will generate a LINE instruction
        at the beginning of each statement and LSYM, ISYM, or GSYM
        instructions for each local variable, instance variable, or
        global variable. Debug information is intended to be used by a
        source level debugger.

        2.14 The Main Program

        Each program has an initial entry point where execution begins
        at run time. In T3X, the entry point is a compound statement at
        the top level which does not belong to any procedure context.
        This compound statement is mandatory and it must always be the
        last definition in the entire program. Subsequently, the
        smallest valid T3X program is

        DO END

        The main procedure, like any other compound statement, may
        declare its own local symbols. Since it has no name, it cannot
        recurse, though. RETURN may not be used in it, because there is
        no procedure to return to.

        When execution reaches the end of the main procedure, the
        program terminates and delivers a zero return code back to the
        calling process.

        Note that all modules have main procedures. When a program
        consists of multiple modules, their main procedures will be
        called in the order in which the modules are passed to the
        Tcode loader. Hence the main module of a program should always
        be compiled last (or at least be the last module passed to the
        Tcode loader).

        3. The Runtime Environment

        Most of this chapter has been automatically generated from the
        structured document '' which is contained in the T3X
        Release 7 distribution.

        There are three different types of classes: the core class,
        native classes, and interface classes. To the programmer, there
        is no difference between these types, but when implementing
        additional runtime classes, it is important to know the

        The T3X core class is in fact an ordinary interface class,
        but it may contain special code to initialize the runtime
        environment at startup time.

        The native class is the most common type. Such classes are
        written in T3X using the techniques described in the section
        about the object oriented programming and modules. The major
        part of the runtime system is implemented in this way. There is
        no difference between a native class and a user-defined module.
        Programs are linked against native classes by the Tcode linker.

        Interface classes allow to add low-level (LL) functions to a
        program. The LL functions themselves are written in a language
        suitable for systems-level programming. The foreign language
        code is compiled to a relocatable object or library.
        Additionally, a native interface class must be defined to
        describe the functions contained in the object holding the code.

        A runtime class is linked into a program by requiring it either
        at class level or at module level. For example,

        CLASS foo(t3x, iostream)

        would be the header of a class requiring the T3X core class and
        the 'iostream' class, and the statement

        module bar(t3x, char, string);

        would require the core class plus the 'char' and 'string'

        The following runtime classes belong to the T3X environment:

        Name       Type        Description
        --------   ---------   ------------------------------
        t3x        core        basic system routines
        char       native      character manipulation
        iostream   native      buffered I/O-streams
        memory     native      dynamic memory management
        string     native      string manipulation
        system     interface   (mostly) portable system calls
        ttyctl     interface   text terminal control
        xmem       interface   external memory access

        These classes will be explained in detail in the following

        3.1 Meta Information

        3.1.1 Object Names

        The first section of each class description lists a sample
        object declaration of the form

        OBJECT objectname[classname];

        where 'objectname' is used to form messages in the remainder of
        the section.

        Methods are referred to

        - by a sample object name and their method name in descriptions
          of their own class (eg T.OPEN in the description of the
          T3X.CLOSE method)

        - by their classname and their method name in other locations
          (eg T3X.OPEN in the description of the SYSTEM class)

        3.1.2 Argument Descriptions

        Argument descriptions are given in the following format:

        object.method(arg1, ..., argN) ! type1, ..., typeN => typeR


        - 'object' is the name of the sample instance defined in the
           first subsection of each class description

        - 'method' is the name of the method described

        - 'arg1' through 'argN' are symbolic names for the arguments of
          the method

        - 'type1' through 'typeN' are the types of 'arg1' through 'argN'

        - 'typeR' is the type of value returned by the metod

        In this context, 'type' is a rather weak attribute, since T3X is
        a typeless language. In many cases 'type' means a pointer to a
        structure with a specific layout.

        The following types are used in this document:

        Type    Description
        -----   ---------------------------------------------
        Bvec    a vector of bytes
        Char    a byte, often representing an ASCII character
        Ddesc   a directory descriptor
        Fdesc   a file descriptor
        IOS     an I/O-stream
        Num     an integer value (a machine word)
        Str     a null-terminated string
        Vec     a vector

        Return types may depend on the return status of a method. Such
        cases are listed, for example, as

        Fdesc | -1

        meaning 'either a file descriptor or minus 1'. Literal numbers
        in return types represent themselves.

        3.2 T3X -- Core Routines

        3.2.1 T3X Class Usage

        OBJECT T[T3X];

        The T3X class contains a set of procedures which provide access
        to the most common operating system services, like opening,
        reading, writing, and erasing files, copying and comparing
        memory regions, receiving command line arguments, etc. The class
        requires no explicit initialization or shutdown.

        The T3X class does not contain any variables. Therefore, it is
        sufficient to create a single instance per module.

        3.2.2 T.BPW

        T.BPW() ! => Num

        Return the number of 'bytes per word' (BPW) on the host machine.
        When running a Tcode program, this value will always be 2,
        regardless of the host environment. When called by a native
        machine program, the procedure will return the actual machine
        word size of the target machine.

        3.2.3 T.CLOSE

        T.CLOSE(fdesc) ! Fdesc => Num

        Close the file descriptor 'fdesc'. To obtain a valid file
        descriptor, use T.OPEN.

        T.CLOSE returns 0 on success and a negative value in case of an


        3.2.4 T.CVALIST

        T.CVALIST(n, bmap, ilist, olist) ! Num,Num,Vec,Vec => 0

        Convert a Tcode argument list to a native argument list. Since
        the Tcode machine is a 16-bit architecture, argument lists may
        need to be extended before passing them to machine code
        procedures in non-16-bit environments.

        Extending an argument list from 16 to 32 bits (or whatever is
        appropriate on the host system) is done by zero-extending all
        values in the argument vector to the size of a generic pointer
        on the host machine and then adding the offset of the Tcode
        machine's data area. The bitmap 'bmap' specifies the type of
        each argument.

        Argument lists must not be longer than 16 elements (plus a
        trailing null word).

        'N' specifies the number of elements in the argument list
        'ilist'. If a trailing null is required, it must be included in
        this number. 'Ilist' is a vector containing the arguments.
        'Bmap' is a bit field where each bit is associated with an
        argument of 'ilist'. A bit of 'bmap' is set if the associated
        argument is a pointer: if bit #0 is set (bmap & 1), ilist[0] is
        a pointer, if bit #1 is set (bmap & 2), ilist[1] is a pointer,
        etc. 'Olist' will be filled with the extended argument list. It
        must provide up to 18 times the size of a generic pointer in
        bytes (which is usually equal to 18 machine words on the host
        system). It may be allocated using:

        VAR olist[18];

        T.CVALIST may relocate 'olist' if it is not aligned to a native
        machine word boundary. It returns the number of bytes 'olist'
        was moved. This number should be used to compute the new address
        of olist:

        offset := t.cvalist(n, bmap, ilist, olist);
        new_olist := @olist::offset.

        When a negative count is supplied, the effect of T.CVALIST is
        reversed. In this case, each member of 'olist' will be copied
        to 'ilist', thereby truncating it to 16 bits. No pointers may
        be processed in this direction. The 'bmap' argument is ignored
        when converting argument lists this way.

        T.CVALIST is used to prepare argument lists for passing them to
        extension procedures. When a T3X program is run in an
        environment where the size of a pointer is equal to the size of
        a T3X machine word, T.CVALIST simply copies the argument vector.

        SYSTEM.SPAWN uses T.CVALIST internally. Most programs do not
        require its use.

        See also: SYSTEM.SPAWN

        3.2.5 T.GETARG

        T.GETARG(n, buffer, size) ! Num,Str,Num => Num

        Retrieve the 'n'th command line argument and store its first
        'size'-1 characters in 'buffer'. If the length K of the
        requested argument is less than 'size'-1, copy only K
        characters. In either case, append a NUL character to the
        argument string extracted.

        T.GETARG returns the number of characters copied. A return code
        of -1 indicates that a non-existing argument has been requested
        ('n' is too big).

        See also: T.GETENV

        3.2.6 T.GETENV

        T.GETENV(name, buffer, size) ! Str,Str,Num => Num

        Retrieve the value of the environment variable 'name' and store
        up to 'size'-1 characters of its value in 'buffer'. Append a
        NUL character to the text in 'buffer'.

        T.GETENV returns the number of characters copied. A return code
        of -1 indicates that a non-existent variable name has been

        See also: T.GETARG

        3.2.7 T.MEMCOMP

        T.MEMCOMP(r1, r2, len) ! Bvec,Bvec,Num => Num

        Compare up to 'len' bytes of the regions 'r1' and 'r2'. When a
        mismatch is found during the comparison, the procedure returns

        r1::p - r2::p

        where 'p' is the position of the mismatch. When 'len' bytes
        have been compared without encountering a mismatch, zero is


        3.2.8 MEMCOPY

        T.MEMCOPY(dest, src, len) ! Bvec,Bvec,Num => 0

        Copy 'len' bytes from region 'src' to region 'dest'. The
        regions may overlap.


        3.2.9 T.MEMFILL

        T.MEMFILL(region, val, len) ! Bvec,Num,Num => 0

        Fill a region of 'len' bytes with the value of the least
        significant byte of 'val'.


        3.2.10 T.MEMSCAN

        T.MEMSCAN(region, val, len) ! Bvec,Num,Num => Num

        Scan a region of 'len' bytes for 'val'. If the region contains
        'val', return its offset (0...len-1) and otherwise return -1.


        3.2.11 T.NEWLINE

        T.NEWLINE(s) ! Str => Str

        Write a system-dependent newline sequence to the string 's'.
        The sequence will move the cursor to the beginning of a new
        line when sent to terminal screens. The sequence written to 's'
        will never be longer than four characters including the
        terminating NUL character.

        The result of writing T.NEWLINE to a screen may be undefined on
        terminals in 'raw mode'.

        T.NEWLINE returns a pointer to 's'.

        See also: T.WRITE, TTYCTL.MODE

        3.2.12 T.OPEN

        T.OPEN(path, mode) ! Str,Num => Fdesc | -1

        Open the file whose path is specified in 'path' in the given
        'mode'. The exact format of 'path' depends on the operating
        system. The following modes exist:

        Mode constant  ReadOK  WriteOK  Create  Initial Position
        -------------  ------  -------  ------  ----------------
        T3X.OREAD      Yes     No       No      0
        T3X.OWRITE     No      Yes      Yes     0
        T3X.ORDWR      Yes     Yes      No      0
        T3X.OAPPND     Yes     Yes      No      EOF

        When T3X.OWRITE is specified and a file with the given name
        already exists, it will be truncated to zero length.

        T3X.OAPPND is like T3X.ORDWR, but the file pointer will be
        positioned at the end of the file so that T.WRITE will append
        its output to the file.

        T.OPEN returns a file descriptor for accessing 'path' on success
        and -1 in case of an error.

        When a T3X program starts up, there already are some open file
        descriptors which are by default connected to the user's

        Name         Descriptor       Mode
        -----------  ---------------  ----------
        T3X.SYSIN    standard input   read-only
        T3X.SYSOUT   standard output  write-only
        T3X.SYSERR   standard error   write-only

        See also: T.CLOSE, T.READ, T.WRITE, T.SEEK

        3.2.13 T.READ

        T.READ(fdesc, buffer, count) ! Fdesc,Vec,Num => Num

        Read up to 'count' characters from the file descriptor 'fdesc'
        into 'buffer'. Return the number of characters read.

        A return value less than zero indicates a severe error. A
        return value which is less than 'count' usually indicates that
        the end of the input has been reached.

        When reading line oriented devices, such as terminals, a return
        value below 'count' may indicate the end of a line. In this
        case, a zero value indicates that the input stream is

        For a summary of standard descriptors (system input and
        output), see T.OPEN.


        3.2.14 T.REMOVE

        T.REMOVE(path) ! Str => Num

        Remove the directory entry specified in 'path'. The exact
        format of 'path' depends on the operating system.

        On systems supporting multiple links (names) for a single file,
        this procedure will only remove the specified link. On such
        systems, other links to the file may still be used to access
        the file. Only when the last link is removed, the file will
        become inaccessible. On other systems, T.REMOVE deletes the
        given file immediately.

        T.REMOVE returns zero, if the directory entry could be deleted
        successfully and a negative value otherwise.

        See also: T.RENAME, SYSTEM.OPENDIR

        3.2.15 T.RENAME

        T.RENAME(old, new) ! Str,Str => Num

        Rename the directory entry whose path is specified in 'old' to
        'new'. 'Old' and 'new' may describe names contained in
        different paths. In this case, the directory entry will be
        moved to the directory specified in 'new'. The old and the new
        name of the directory entry must both reside on the same
        physical device.

        T.RENAME returns zero upon success and a negative value in case
        of an error.

        See also: T.REMOVE

        3.2.16 T.SEEK

        T.SEEK(fdesc, where, origin) ! Fdesc,Num,Num => Num

        Move the file pointer associated with the file descriptor
        'fdesc' to a new position. 'Where' specifies the desired
        position and 'origin' specifies where the motion shall start.
        The following origins are possible:

        Constant      Origin                 Distance
        ------------  ---------------------  --------
        T3X.SEEK_SET  Beginning of the file   +where
        T3X.SEEK_FWD  Current position        +where
        T3X.SEEK_END  End of the file         -where
        T3X.SEEK_BCK  Current position        -where

        T3X.SEEK_SET and T3X.SEEK_FWD move the file pointer forward,
        T3X.SEEK_END and T3X.SEEK_BCK move it backward. In either case,
        'where' is an unsigned value so that offsets may range from 0
        to 65535 bytes.

        T.SEEK returns zero upon success and -1 in case of an error.

        The 'seek' operation may be undefined on sequential access
        devices and pipes.

        See also: T.OPEN, T.CLOSE, T.READ, T.WRITE

        3.2.17 T.WRITE

        T.WRITE(fdesc, buffer, count) ! Fdesc,Vec,Num => Num

        Write 'count' characters from 'buffer' to the file descriptor
        'fdesc'. Return the number of characters actually written.

        A return value which is less than 'count' indicates a severe
        error (such as insufficient space left on a device).

        For a summary of standard descriptors (system input and
        output), see T.OPEN.

        See also: T.OPEN, T.CLOSE, T.READ, SYSTEM.PIPE

        3.3 CHAR -- Character Functions

        3.3.1 CHAR Class Usage


        The CHAR class contains functions for determining character
        types and converting characters. They all operate on ASCII

        This class must be initialized before its use by calling
        CHR.INIT. An explicit shutdown is not required.

        The CHAR class does not contain any variables. Therefore, it is
        sufficient to create a single instance per module.

        3.3.2 CHR.INIT

        CHR.INIT() ! => 0

        Initialize the character class by loading an internal pointer
        with the character type map.

        See also: CHR.MAP

        3.3.3 CHR.ALPHA

        CHR.ALPHA(c) ! Char => Num

        Return TRUE (-1), if 'c' is an alphabetic character (in the
        range 'a'...'z' or 'A'...'Z'). Otherwise return FALSE (0).

        3.3.4 CHR.ASCII

        CHR.ASCII(c) ! Char => Num

        Return TRUE (-1), if 'c' is a valid ASCII value (in the range
        0...127). Otherwise return FALSE (0).

        3.3.5 CHR.CNTRL

        CHR.CNTRL(c) ! Char => Num

        Return TRUE (-1), if 'c' is a control character (in the range
        0...31 or equal to 127). Otherwise return FALSE (0).

        3.3.6 CHR.DIGIT

        CHR.DIGIT(c) ! Char => Num

        Return TRUE (-1), if 'c' is a decimal digit (in the range
        '0'...'9'). Otherwise return FALSE (0).

        3.3.7 CHR.LCASE

        CHR.LCASE(c) ! Char => Char

        If the character 'c' is an upper case character (see
        CHR.UPPER), convert it to lower case and return it. Otherwise,
        return it unchanged.

        See also: CHR.UCASE

        3.3.8 CHR.LOWER

        CHR.LOWER(c) ! Char => Num

        Return TRUE (-1), if 'c' is a lower case letter (in the range
        'a'...'z'). Otherwise return FALSE (0).

        See also: CHR.UPPER

        3.3.9 CHR.MAP

        CHR.MAP() ! => Bvec

        Return the character description map used internally. This map
        is a byte vector of 128 members containing flags for describing
        each ASCII character. It can be used to implement fast
        character class checks. For example,

        IF (chr.lower(c)) ...

        can be written as

        chrmap :=;
        IF (chrmap::c & (CHAR.C_UPPER|CHAR.C_ALPHA) = CHAR.C_ALPHA)

        which saves a procedure call each time a character is tested
        for being lower case.

        The following public constants are defined in the CHAR class
        and can be used for testing character flags:

        Flag          Property
        ------------  --------
        CHAR.C_ALPHA  alphabetic
        CHAR.C_UPPER  upper case
        CHAR.C_DIGIT  decimal digit
        CHAR.C_SPACE  white space
        CHAR.C_CNTRL  control character

        3.3.10 CHR.SPACE

        CHR.SPACE(c) ! Char => Num

        Return TRUE (-1), if 'c' is a space character (HT(9), LF(10),
        VT(11), FF(12), CR(13), space(32)). Otherwise return FALSE (0).

        3.3.11 CHR.UCASE

        CHR.UCASE(c) ! Char => Char

        If the character 'c' is a lower case character (see CHR.LOWER),
        convert it to upper case and return it. Otherwise, return it

        See also: CHR.LCASE

        3.3.12 CHR.UPPER

        CHR.UPPER(c) ! Char => Num

        Return TRUE (-1), if 'c' is a upper case letter (in the range
        'A'...'Z'). Otherwise return FALSE (0).

        See also: CHR.LOWER

        3.3.13 CHR.VALUE

        | CHR.VALUE(c) ! Char => Num

        Return the value of the decimal digit represented by the
        character in 'c' or -1, if the character does not represent a
        decimal digit.

        3.4 IOSTREAM -- I/O-Streams

        3.4.1 IOSTREAM Class Usage


        The IOSTREAM class implements fully buffered I/O streams.

        I/O streams provide a string/character-oriented interface to
        the programmer while performing block-oriented I/O to the file
        or device associated with a stream. This way, they combine the
        speed of block-I/O with the flexibility of character-based I/O.

        This class contains the I/O stream data structure and
        procedures for creating, opening, closing, reading, and writing

        A separate IOSTREAM object must be defined for each stream to
        be used in a program.

        3.4.2 IOS.CLOSE

        IOS.CLOSE() ! => Num

        Shutdown the I/O stream IOS by first flushing its buffer and
        then closing the file associated with the stream. Flushing a
        buffer means to write any pending output (if the stream has been
        written to) and to discard any pending input (if the stream is
        being read from).

        IOS.CLOSE returns zero, if the stream could be closed and
        otherwise -1. After successfully sending CLOSE, the receiving
        stream becomes invalid immediately and should no longer be


        3.4.3 IOS.CREATE

        IOS.CREATE(fd, buffer, len, mode) ! Fdesc,Bvec,Num,Num => 0

        Initialize the iostream IOS with the given parameters. 'Fd'
        is an open file descriptor which will be associated with the
        stream. 'Buffer' will be used for buffering read/write
        operations on the stream. 'Len' specifies the size of 'buffer'
        in bytes. 'Mode' controls the operations allowed on IOS. The
        following flags may be used to build the mode value (by OR'ing
        together their values):

        Mode constant     ReadOK  WriteOK  LF>CRLF  CRLF>LF
        ----------------  ------  -------  -------  -------
        IOSTREAM.FREAD    Yes     No       -        -
        IOSTREAM.FWRITE   No      Yes      -        -
        IOSTREAM.FRDWR    Yes     Yes      -        -
        IOSTREAM.FKILLCR  -       -        -        Yes
        IOSTREAM.FADDCR   -       -        Yes      -
        IOSTREAM.FTRANS   -       -        Yes      Yes

        CRLF>LF denotes that each CR character found in an input stream
        will be silently discarded. This is useful when reading
        DOS-style ASCII text files. LF>CRLF means that a CR character
        will be added before each LF in the output stream. Since
        IOSTREAM.FADDCR has no effect on input and IOSTREAM.FKILLCR has
        no effect on output, IOSTREAM.FTRANS may be used safely on
        input as well as output streams.

        IOS.CREATE merely initializes an IOSTREAM object with the
        supplied parameters. It cannot fail and therefore, it returns
        always 0.

        When using IOS.CREATE to create a stream for accessing standard
        file descriptors (such as T3X.SYSIN and T3X.SYSOUT), these
        streams should never be closed. IOS.FLUSH may be used to
        synchronize them.

        See also: IOS.OPEN, IOS.CLOSE, IOS.FLUSH

        3.4.4 IOS.EOF

        IOS.EOF() ! => Num

        Return a flag indicating whether input has been exhausted on
        the stream IOS. When EOF returns TRUE (-1), no more input can
        be read from IOS. This is the case when the end of the
        associated input file has been reached or when an EOF character
        has been typed on a terminal.


        3.4.5 IOS.FLUSH

        IOS.FLUSH() ! => Num

        Flush the stream IOS and return a value indicating whether
        the operation was successful. Zero means success, -1 means

        Flushing an output stream means to write all pending data to
        the associated file. Flushing an input stream means to discard
        all pending input. The operation performed on a combined
        input/output stream depends on the type of the last operation
        performed before flushing the stream.


        3.4.6 IOS.MOVE

        IOS.MOVE(offset, origin) ! Num,Num => Num

        Move the file pointer of the file descriptor associated with
        IOS to a new position. The position is computed using the
        given 'offset' and 'origin'. 'Offset' is the number of bytes to
        skip and 'origin' specifies where the motion shall begin. The
        following origin values are available:

        Constant           Origin             Direction
        -----------------  -----------------  ---------
        IOSTREAM.SEEK_SET  beginning of file  forward
        IOSTREAM.SEEK_FWD  current position   forward
        IOSTREAM.SEEK_END  end of file        backward
        IOSTREAM.SEEK_BCK  current position   backward

        IOSTREAM.SEEK_SET and IOSTREAM.SEEK_FWD move the file pointer
        forward, IOSTREAM.SEEK_END and IOSTREAM.SEEK_BCK move it
        backward. In either case, 'where' is an unsigned value so that
        offsets may range from 0 to 65535 bytes.

        IOS.MOVE always flushes the stream buffer before changing the
        file pointer.

        It returns zero upon success and -1 in case of an error.

        See also: IOS.FLUSH, T3X.SEEK

        3.4.7 IOS.OPEN

        IOS.OPEN(path, buffer, len, mode) ! Str,Vec,Num,Num => Num

        Open the file specified in 'path' and initialize IOS with the
        resulting file descriptor and the arguments 'buffer', 'len',
        and 'mode'. See IOS.CREATE for details. The exact format of
        'path' depends on the operating system. The following open modes
        ('mode' values) are common:

        Mode constant    ReadOK  WriteOK  Create
        ---------------  -----   -------  ------
        IOSTREAM.FREAD   Yes     No       No
        IOSTREAM.FWRITE  No      Yes      Yes
        IOSTREAM.FRDWR   Yes     Yes      No

        For additional modes, see IOS.CREATE.

        When creating a file, any existing file with the same name will
        be truncated to zero length.

        IOS.OPEN returns zero upon success and -1 in case of an error.


        3.4.8 IOS.RDCH

        IOS.RDCH() ! => Char | -1

        Read a single character from IOS and return it. When the EOF
        condition is true on IOS, return -1 (which cannot be a valid


        3.4.9 IOS.READ

        IOS.READ(buffer, len) ! Vec,Num => Num

        Read up to 'len' characters from IOS into 'buffer'. Return the
        number of characters actually read. A return value less than
        'len' may indicate the end of input or the beginning of a new
        line a on a terminal. A return value of zero always indicates
        the EOF. A value below zero indicates a severe error.


        3.4.10 IOS.READS

        IOS.READS(buffer, len) ! Vec,Num => Num

        Read up to 'len'-1 characters from IOS into 'buffer'. Return
        the number of characters actually read. A return value of zero
        indicates that the EOF has been reached. A value below zero
        indicates a severe error.

        Unlike IOS.READ, IOS.READS stops reading when it encounters a
        line separator (LF) in input. If this happens, the LF character
        will be inserted as the last character into 'buffer'.


        3.4.11 IOS.RESET

        IOS.RESET() ! => 0

        Reset the error flag of the iostream IOS. Resetting the error
        flag is necessary to access a stream after an error has
        occurred (for example, after reading beyond the EOF).

        See also: IOS.EOF, IOS.READ

        3.4.12 IOS.WRCH

        IOS.WRCH(c) ! Char => Char|Num

        Write the character 'c' to the stream IOS. If the character
        could be written, return its ASCII code and otherwise return


        3.4.13 IOS.WRITE

        IOS.WRITE(buffer, len) ! Vec,Num => Num

        Write 'len' characters from 'buffer' to IOS. Return the number
        of characters actually written. A return value less than 'len'
        indicates a severe error (such as no space left on the target


        3.4.14 IOS.WRITES

        IOS.WRITES(str) ! Str => Num

        Write the string 'str' to IOS. Return the number of characters
        actually written. A return value less than the length of 'str'
        indicates a severe error (such as no space left on the target

        See also: IOS.WRITE, IOS.WRCH, IOS.FLUSH

        3.5 MEMORY -- Dynamic Memory Management

        3.5.1 MEMORY Class Usage


        The MEMORY class implements dynamic memory pools. When
        initialized, the address of a static data area is passed to a
        MEMORY object. This area (called a 'pool') will be managed by
        the MEMORY object. Vectors can be allocated from the pool and
        released back to it when they are no longer required.

        A first-match algorithm is used to allocate memory in a pool.
        The algorithm is optimized for sequential allocation. The pool
        is defragmented when releasing memory, but no compaction is
        performed, so fragmentation may still happen.

        MEMORY objects are ineffective when allocating a large number
        of small vectors, since a free list entry has to be created for
        each vector allocated.

        Multiple memory pools may be defined using the MEMORY class,
        but usually one pool per program is most effective.

        3.5.2 MEM.ALLOC

        MEM.ALLOC(size) ! Num => Vec | 0

        Allocate 'size' bytes from the memory pool MEM and return a
        pointer to the allocated vector. If the request could not be
        satisfied due to insufficient memory, return 0.

        Up to 32765 bytes may be allocated in a single request.

        See also: MEM.FREE

        3.5.3 MEM.FREE

        MEM.FREE(vec) ! Vec => 0

        Release the memory occupied by the vector 'vec' to MEM and
        defragment MEM. The space of of 'vec' is be added to the free
        memory pool of MEM.

        'Vec' must be the address of a vector which has been previously
        allocated in MEM. Otherwise, the calling program may be
        terminated with an error message of the form

        MEMORY CLASS: bad block in

        Accessing a freed vector is undefined.

        See also: MEM.ALLOC

        3.5.4 MEM.INIT

        MEM.INIT(pool, size) ! Vec,Num => 0

        Initialize the memory pool MEM and add 'size' bytes to its
        internal freelist. 'Pool' must have a size of at least 'size'
        bytes. 'Size' may not be larger than 32767.

        All previously allocated vectors of MEM will be freed by

        MEM.INIT always returns 0.

        See also: MEM.FREE

        3.5.5 MEM.WALK

        MEM.WALK(vec, sizep, statp) ! Vec,Vec,Vec => Vec

        Traverse the list of vectors in MEM. This list contains both
        allocated and free vectors. Traversing the list works as

        When MEM.WALK is called for the first time, 'vec' must be zero:

        v := mem.walk(0, @p, @s);

        This call will return a pointer to the first vector of MEM.
        The returned vector can be passed to MEM.WALK in a subsequent
        call to retrieve a pointer to the next vector:

        v := mem.walk(v, @p, @s);

        When the 'vec' argument finally points to the last vector in
        MEM, MEM.WALK will return zero.

        The argument 'sizep' is a one-word vector which will be filled
        with the size of the returned vector. 'Statp' is a one-word
        vector which will be filled with the status of the vector
        (1=free, 0=allocated). If either 'statp' or 'sizep' is zero, it
        will be ignored by MEM.WALK.

        3.6 STRING -- String Functions

        3.6.1 STRING Class Usage


        The STRING class contains procedures for manipulating
        NUL-terminated sequences of ASCII characters.

        The STRING class does not contain any data, so a single
        instance per module is sufficient. The class does not require
        any explicit initialization or shutdown.

        3.6.2 STR.COMP

        STR.COMP(a, b) ! Str,Str => Num

        Compare each character in 'a' with the character at the
        corresponding position in 'b'. When a position 'i' is found
        at which the characters of 'a' and 'b' differ, return

        a::i - b::i

        When no mismatch is encountered, return 0. Consequently, the
        return value of STR.COMP can be interpreted as follows:

        Value  Meaning
        -----  -------
        >0     'a' is lexically greater than 'b'
        <0     'a' is lexically less than 'b'
        =0     'a' is equal to 'b'

        See also: STR.FIND, STR.SCAN, STR.RSCAN

        3.6.3 STR.COPY

        STR.COPY(a, b) ! Str,Str => 0

        Copy the string stored at the location 'b' to the location 'a'.
        Return zero.

        See also: STR.FORMAT, STR.NCOPY

        3.6.4 STR.FIND

        STR.FIND(a, b) ! Str,Str => Num

        Find the first occurrence of the string 'b' in the string 'a'.
        Return the offset of the string found, if any. Return -1, if
        'a' does not contain 'b'.

        See also: STR.COMP, STR.SCAN, STR.RSCAN

        3.6.5 STR.FORMAT

        STR.FORMAT(buf, tmpl, list) ! => Str,Str,Vec => Str

        Format the arguments contained in 'list' according to the
        template 'tmpl' and store the resulting string in 'buf'.

        'Tmpl' is a string containing literal characters as well as
        'format descriptors'. A format descriptor is a substring that
        begins with a percent sign and ends with one of the characters
        in {C,D,S,X,%}. When 'tmpl' does not contain any format
        descriptors, it will be copied to 'buf' and 'list' will be
        ignored. When format descriptors exist, each descriptor will be
        used to format one element of 'list'. Instead of the descriptor
        itself, the result of formatting one member of 'list' will be
        inserted into 'buf'.

        A format descriptor has the following syntax:


        ([x] indicates an optional element 'x', {xyz} indicates one
        element out of x, y, and z.)

        - The percent sign '%' starts the descriptor.

        - When a decimal number 'max' is specified after '%', this number
          indicates the minimum field length of the current argument. If
          formatting the current argument yields a result shorter than
          'len', the field will be filled with blanks.

        - ':F' denotes that the character F should be used for filling
          fields instead of blank characters.

        - When 'U' is specified together with 'D', an unsigned numeric
          string will be generated (a signed representation will be
          generated by default).

        - 'L' instructs STR.FORMAT to left-justify the current field,
          'R' instructs it to right-justify it. The default is to
          left-justify strings and to right-justify numbers.

        - The last character specifies the type of the current argument
          in 'list'.

        The following types exist ('i' denotes the index of the current

        Type  Insert list[i] as
        ----  ----------------------------
        C     character
        D     decimal numeric literal
        S     string
        X     hexa-decimal numeric literal

        The string '%%' may be used to include a literal percent sign.

        There must be one member in 'list' for each descriptor in 'tmpl'.


        Template               Argument list  Result
        ---------------------  -------------  ----------------------
        "%D%% of %10:*D = %D"  [10,200,20]    10% of *******200 = 20
        "'%C' = 0X%X = %D"     ['A','A','A']  'A' = 0X41 = 65
        "%:-9LS%:+9RS"         ["ZZZ","YYY"]  ZZZ------++++++YYY

        STR.FORMAT returns the address of 'buf'.


        3.6.6 STR.LENGTH

        STR.LENGTH(a) ! Str => Num

        Return the number of characters contained in 'a' (excluding the
        terminating NUL character).

        3.6.7 STR.NCOPY

        STR.NCOPY(n, a, b) ! Num,Str,Str => 0

        Copy up to 'n'-1 characters of the string stored at the
        location 'b' to the location 'a'. When 'a' is shorter than
        'n'-1 characters, copy the entire string. Always terminate
        'a' with NUL. Return zero. 

        When 'n' = ~0, STR.NCOPY implements STR.COPY.

        See also: STR.COPY, STR.NFORMAT

        3.6.8 STR.NFORMAT

        STR.NFORMAT(n, buf, tmpl, list) ! => Num, Str,Str,Vec => Str

        STR.NFORMAT is like format, but never writes more than 'n'-1
        characters to 'buf'. See STR.FORMAT for details.

        When the given template and arguments would produce a resulting
        string containing more than 'n'-1 characters, STR.NFORMAT will
        return a result that is truncated at the 'n'-1'st character.
        The result will still be NUL-terminated.

        When a format descriptor that uses padding would cause the
        length of 'buf' to be exceeded, first the padding will be
        decreased and then the representation of the described object
        will be truncated. This means that given N=5, the following
        results would be obtained:

        argument        descriptor      result
        --------        ----------      -------
        "123"           "%10LS"         "123  "
        "123"           "%10RS"         "  123"
        "1234567"       "%10LS"         "12345"
        "1234567"       "%10RS"         "12345"

        See also: STR.FORMAT, STR.NCOPY

        3.6.9 STR.NUMTOSTR

        STR.NUMTOSTR(buf, n, radix) ! Str,Num,Num => Str

        Convert a number 'n' to a string representing that number
        with respect to the given 'radix'. The resulting numeric
        string will be stored in 'buf'.

        If 'radix' and 'n' are both negative, a leading minus sign will
        be generated. If 'radix' is positive, an unsigned value of
        N will be assumed.

        'Buf' must provide enough space to hold the resulting literal.

        Valid values for 'radix' range from '2' (binary) to '16' (hexa-
        decimal) and from '-2' (signed binary) to '-16' (signed hexa-

        STR.STRTONUM returns the address of the first character of the
        resulting literal.

        See also: STR.STRTONUM, STR.FORMAT

        3.6.10 STR.PARSE

        STR.PARSE(source, tmpl, list) ! Str,Str,Vec => Num

        Extract patterns described in 'tmpl' from 'source' and store
        the extracted objects in the members of 'list'. Patterns used
        in 'tmpl' are similar to format descriptors used by
        STR.FORMAT. Characters not belonging to patterns are matched
        literally and not stored anywhere.

        STR.PARSE compares each character contained in 'tmpl' with a
        corresponding character in 'source' (like STR.COMP) until it
        finds a '%'-character in 'tmpl'. A percent character indicates
        the beginning of a pattern. Patterns match specific classes of
        characters. Instead of matching the pattern itself, the
        character class described by the pattern is matched.

        Some patterns store the value of the matched substring in an
        element of 'list' and some do not. Each pattern may consist of
        the following parts:


        ([x] indicates an optional element 'x', {xyz} indicates one
        element out of x, y, and z.)

        The special form %[c1...cN] may be used to match any character
        in the range 'c1'...'cN'. The matched characters are not
        stored in this case.

        - If a length 'len' is specified, UP TO 'len' characters will be
          matched. Typically, this option is used together with %S.

        - ':D' instructs STR.PARSE to recognize the character in the
          place of 'D' as a delimiter (default = none).

        The following pattern types exist:

        Type  Store as   Matches
        ----  ---------  ---------------------------------
        C     character  any single character
        D     number     a signed decimal number (+)
        S     string     a string (*)
        W     -          space (any number of '\s' or '\t'
        .     .          characters)
        X     number     a signed hexa-decimal number (+)
        %     -          a percent sign

        (+) These leading prefixes are accepted: {+-%}

        (*) When no length is specified, %S matches the entire rest of
        'source'. ':D' or a length may be specified to match a

        Numbers and characters are stored in 'list[i][0]' (where 'i' is
        the index of the current member of 'list') and strings are
        copied to the location pointed to by 'list[i]'.


        VAR name::50, speed, unit::10;
        STR.PARSE("HAL9000 @ 500 MHz", "%:@S@ %D%W%S",
        [ (name), (@speed), (unit)]);

        will store

        "HAL9000 " in 'name'
        500        in 'speed'
        "MHz"      in 'unit'

        STR.PARSE returns the number of patterns stored.

        See also: STR.FORMAT, STR.COMP

        3.6.11 STR.RSCAN

        STR.RSCAN(s, c) ! Str,Num => Num

        Find the rightmost occurrence of the character 'c' in the
        string 's' and return the offset (position) of the character
        found. If 'c' is not contained in 's', return -1.

        See also: STR.SCAN, STR.FIND, STR.COMP

        3.6.12 STR.SCAN

        STR.SCAN(s, c) ! Str,Num => Num

        Find the first occurrence of the character 'c' in the string
        's' and return the offset (position) of the character found. If
        'c' is not contained in 's', return -1.

        See also: STR.RSCAN, STR.FIND, STR.COMP

        3.6.13 STR.STRTONUM

        STR.STRTONUM(s, radix, lastp) ! Str,Num,Vec => Num

        Compute the value represented by the numeric strung stored in
        the 's'. 'Radix' specifies the base of the literal in 's'. It
        may range from '2' (binary) to '16' (hexa-decimal).

        STR.STRTONUM performs the following steps:

        - Leading TAB ('\t') and space ('\s') characters in 's' are

        - A plus (+) or minus (-,%) sign is recognized.

        - Characters belonging to the specified number class (based
          upon 'radix') are collected and converted to a numeric

        The following characters may represent the digits from 0 to 15:

        When the argument 'lastp' is non-zero, it will be filled with
        the number of characters processed. Consequently, it is the
        offset of the first non-numeric character in 's' when STRTONUM

        STR.STRTONUM returns the computed value.

        No overflow checking is performed.

        See also: STR.NUMTOSTR, STR.PARSE

        3.6.14 STR.XLATE

        STR.XLATE(s, old, new) ! Str,Num,Num => Str

        Replace each occurrence of the character 'old' in the string
        's' with 'new'. Return the address of 's'.

        See also: STR.SCAN, STR.RSCAN

        3.7 TCODE -- Tcode Instruction Set

        3.7.1 TCODE Class Usage



        The TCODE class contains a set of public constants describing
        the instruction set of the Tcode machine. There is no need to
        instantiate this class, since it does not contain any state or
        methods. To access the opcode of a specific Tcode instruction,
        use the class constant notation


        For example, to load the opcode of the JUMP instruction into the
        variable I, use

        I := TCODE.IJUMP;

        The special constant IENDOFSET contains a value which is one
        above the highest value representing a Tcode instructions. To
        check if a variable J contains a valid instruction, the
        following code may be used:

        IF ((J & 0x7F) .>= TCODE.IENDOFSET)
            ; ! Call your illegal instruction handler here

        3.8 UTIL -- Utilities

        3.8.1 UTIL Class Usage

        OBJECT U[UTIL];

        The UTIL (utility) class contains utility procedures for
        miscellaneous tasks. Currently, there are methods for sending
        formatted output to various channels like the system output,
        file descriptors or streams. Using the UTIL class simplifies
        many frequently used code fragments. For example, the code

        DO VAR buffer::256;
            t.write(T3X.SYSOUT, str.format(buffer, "X = %D\N", [(x)]),

        could be replaced with

        u.printf("X = %D\N", [(x)]);

        in modules using the UTIL class.

        3.8.2 UTIL.BUFLEN


        This public constant holds the maximum length of strings
        formatted by the PRINTF, WRITEF, and SWRITEF methods. The
        length returned includes the terminating NUL character.

        3.8.3 U.PRINTF

        U.PRINTF(tmpl, args) ! Str,Vec => Num

        Format the arguments contained in the vector 'args' using

        STR.FORMAT(buffer, tmpl, args)

        where 'buffer' is an internal buffer of the length UTIL.BUFLEN.
        The buffer is then written to the system output device

        U.PRINTF returns the number of characters written using

        3.8.4 U.SWRITEF

        U.SWRITEF(ios, tmpl, args) ! IOS,Str,Vec => Num

        Format the arguments contained in the vector 'args' using

        STR.FORMAT(buffer, tmpl, args)

        where 'buffer' is an internal buffer of the length UTIL.BUFLEN.
        The buffer is then sent to the output stream IOS.

        U.SWRITEF returns the number of characters written using

        3.8.5 U.WRITEF

        U.WRITEF(fd, tmpl, args) ! FDesc,Str,Vec => Num

        Format the arguments contained in the vector 'args' using

        STR.FORMAT(buffer, tmpl, args)

        where 'buffer' is an internal buffer of the length UTIL.BUFLEN.
        The buffer is then written to the file descriptor 'fd'.

        U.WRITEF returns the number of characters written using

        3.9 SYSTEM -- System Interface

        3.9.1 SYSTEM Class Usage


        The SYSTEM class contains a set of procedures which form a
        (more or less) portable interface to the operating system. Most
        procedures have the same names and functions as Unix system
        calls. Some functions may be unavailable on non-Unix systems.

        The SYSTEM class must be initialized using SYS.INIT and shut
        down by calling SYS.FINI.

        SYSTEM does not contain any variables. Therefore, it is
        sufficient to create a single instance per module.

        3.9.2 SYS.INIT

        SYS.INIT() ! => 0

        Initialize the operating system interface.

        See also: SYS.FINI

        3.9.3 SYS.CHDIR

        SYS.CHDIR(path) ! Str => Num

        Change the current working directory to 'path'. The format of
        'path' depends on the operating system.

        SYS.CHDIR returns 0 on success and a negative value in case of
        an error.


        3.9.4 SYS.CLOSEDIR

        SYS.CLOSEDIR(ddesc) ! Ddesc => Num

        Close the directory descriptor 'ddesc'.

        SYS.CLOSEDIR returns 0 on success and a negative value in case
        of an error.


        3.9.5 SYS.DUP

        SYS.DUP(oldfd) ! Fdesc => Fdesc | -1

        Duplicate the file descriptor 'oldfd' and return a new
        descriptor referencing the same file. All operations performed
        on the new descriptor will also affect 'oldfd'.

        SYS.DUP returns a new descriptor on success and a negative
        number in case of an error.


        3.9.6 SYS.DUP2

        SYS.DUP2(oldfd, newfd) ! Fdesc,Fdesc => Num

        Duplicate the file descriptor 'oldfd' and make 'newfd'
        reference the same file. Thereafter, all operations performed
        on one of them will also affect the other. If 'newfd' already
        references a valid descriptor, it will be closed first by
        calling T3X.CLOSE.

        SYS.DUP2 returns zero upon success, and a negative number in
        case of an error.


        3.9.7 SYS.FINI

        SYS.FINI() ! => 0

        Shutdown the operating system interface. After calling
        SYS.FINI, the SYSTEM services become unavailable.

        See also: SYS.INIT

        3.9.8 SYS.FORK

        SYS.FORK() ! => Num

        Duplicate the calling process. The new process -- called the
        child process -- will start running exactly at the point where
        SYS.FORK returns. Each process has an own data segment and an
        own set of file descriptors. Descriptors which where open when
        SYS.FORK was called, will reference the same files in both
        processes, though.

        After the successful creation of the new process, SYS.FORK
        returns zero to the child process and the process ID of the
        child to the parent process.

        In case of an error, SYS.FORK returns -1.


        3.9.9 SYS.GETDIR

        SYS.GETDIR(buf, len) ! Str,Num => Num

        Store the fully qualified path name of the current working
        directory in 'buf'. Do not store more than the first 'len'-1
        characters. Append a trailing NUL character to the path. 'Buf'
        must be at least 'len' characters in length, and it may not be
        smaller than 65 characters.

        If 'len' is less than 65 or the function fails, -1 is returned.
        Upon success, the number of characters stored is returned.


        3.9.10 SYS.KILL

        SYS.KILL(pid, sig) ! Num,Num => Num

        Send a signal to the process with the process ID 'pid'. The
        following constants may be used in the place of 'sig' to
        specify which signal is to be sent to the process:

        Constant        Action
        --------------  -------------------
        SYSTEM.SIGTEST  Test process
        SYSTEM.SIGTERM  Request termination
        SYSTEM.SIGKILL  Force termination

        SYS.KILL returns zero, if the signal could be delivered
        successfully, and a negative number in case of an error.

        Delivering SYSTEM.SIGTEST does not have any effect on the
        process with the given PID. Therefore, it can be used to check
        whether the PID is valid.

        SYSTEM.SIGTERM may be caught by the receiving process to
        initiate a graceful shutdown.

        SYSTEM.SIGKILL terminates the receiving process immediately.

        See also: SYS.FORK, SYS.SPAWN

        3.9.11 SYS.MKDIR

        SYS.MKDIR(path) ! Str => Num

        Create a directory with the name stored in 'path'. The format
        of 'path' depends on the operating system.

        SYS.MKDIR returns zero, if the directory could be created and
        otherwise a negative value.


        3.9.12 SYS.OPENDIR

        SYS.OPENDIR(path) ! Str => Ddesc | -1

        Open the directory specified in 'path'. The exact format of
        'path' depends on the underlying operating system.

        Upon success, SYS.OPENDIR returns a directory descriptor and in
        case of an error, it returns -1.


        3.9.13 SYS.PIPE

        SYS.PIPE(vec) ! Vec => Num

        Create a pipe (a FIFO structure) and fill the vector 'vec' with
        two file descriptors which can be used to access the pipe. Each
        element of 'vec' will be filled with an ordinary file
        descriptor as returned by T3X.OPEN. Therefore, ordinary I/O
        operations like T3X.READ and T3X.WRITE can be used to read and
        write a pipe.

        After the successful creation of a pipe, 'vec[0]' will contain
        the (read-only) output descriptor and 'vec[1]' will contain the
        (write-only) input descriptor of the pipe.

        Data written to 'vec[1]' can be read from 'vec[0]'. Write
        requests will block, if the pipe is full and read requests will
        block, if the pipe is empty. The size of the pipe depends on
        the operating system.

        SYS.PIPE returns zero when a pipe could be created and a
        negative value in case of an error.

        See also: T3X.OPEN, T3X.CLOSE, T3X.READ, T3X.WRITE

        3.9.14 SYS.RDCHK

        SYS.RDCHK(fdesc) ! Fdesc => Num

        Check whether input is available from the file descriptor
        'fdesc' (whether a read operation on 'fdesc' would NOT block).

        SYS.RDCHK returns a non-zero value, if the operation would
        succeed without blocking. If the read request would block, zero
        is returned.

        To check the status of a terminal, use TTYCTL.QUERY instead.

        See also: T3X.READ, TTYCTL.QUERY

        3.9.15 SYS.READDIR

        SYS.READDIR(ddesc, buffer, lim) ! Ddesc,Str,Num => Num

        Read the next directory entry from the directory descriptor
        'ddesc' and fill 'buffer' with the name of that entry. If the
        name is longer than 'lim'-1 characters, truncate it to 'lim'-1
        characters. In any case, terminate the extracted string with a
        NUL character.

        SYS.READDIR returns the length of the extracted string upon
        success, and -1 in case of an error. Reading beyond the end of
        the directory will return -1.


        3.9.16 SYS.RMDIR

        SYS.RMDIR(path) ! Str => Num

        Remove the directory specified in 'path'. The format of 'path'
        depends on the operating system.

        SYS.RMDIR returns zero, if the directory could be removed and
        a negative value otherwise.


        3.9.17 SYS.SPAWN

        SYS.SPAWN(prog, args, mode) ! Str,Vec,Num => Num

        Create a new process by running the program 'prog' with the
        command line options stored in 'args'. The format of the path
        of 'prog' depends on the operating system.

        'Args' is a vector of strings, each one containing one command
        line argument. The last element of the vector must be zero.
        'Mode' controls whether execution of the calling process will be
        suspended until the spawned process exits. The following modes

        Mode name            Meaning
        -------------------  ------------------------------------
        SYSTEM.SPAWN_NOWAIT  Execute the new process concurrently
        SYSTEM.SPAWN_WAIT    Suspend the caller until the child
                             process terminates

        NOTE: some operating systems may restrict the space which can
        be used for passing command line arguments.

        NOTE2: on non-multitasking systems, SYSTEM.SPAWN_NOWAIT may be

        SPAWN returns the exit code of the subprocess when called with
        mode=SYSTEM.SPAWN_WAIT. It returns the process ID (PID) of the
        spawned subprocess when called with mode=SYSTEM.SPAWN_NOWAIT.
        In case of an error, it returns -1.

        See also: SYS.FORK, SYS.WAIT

        3.9.18 SYS.STAT

        SYS.STAT(path, sb) ! Str,Vec => Num

        Retrieve information about the file specified in 'path'. The
        format of 'path' depends on the operating system. The retrieved
        information will be stored in the SYSTEM.STATBUF structure 'sb'
        which has the following format:

        struct STATBUF =
               ST_DEV,     ! device ID
               ST_INO,     ! inode number
               ST_MODE,    ! access bits
               ST_NLINK,   ! number of links
               ST_UID,     ! user ID of owner
               ST_GID,     ! group ID of owner
               ST_RDEV,    ! device type
               ST_SIZE,    ! file size in bytes
               ST_EXT64,   ! file size in 64K blocks
               ST_MTIME,   ! date of last modification (8 bytes)
                           ! Format: CYMDHMSh, see SYS.TIME.
               ST_MT_2,    ! \
               ST_MT_3,    !  > Buffer for ST_MTIME
               ST_MT_4;    ! /

        For the layout of the MTIME structure, see SYS.TIME.

        Depending on the operating system, some fields will be filled
        with more or less meaningful standard values. For example,
        systems not supporting multiple links will fill the ST_NLINK
        field with 1.

        The access field SYSTEM.ST_MODE may have the following flags

        Flag             Description
        ---------------  ----------------------
        SYSTEM.FM_RDOK   file is readable
        SYSTEM.FM_WROK   file is writeable
        SYSTEM.FM_EXOK   file is executable (*)
        SYSTEM.FM_ISDIR  file is a directory

        (*) Files on DOS file systems do not have an executable flag.


        is always zero on DOS systems.

        SYS.STAT returns zero upon success and otherwise a negative


        3.9.19 SYS.TIME

        SYS.TIME(tbuf) ! Bvec => 0

        Fill the buffer 'tbuf' with the current system time. 'Tbuf'
        must provide eight bytes of space which will be filled as

        Field    Value         Range
        -------  ------------  ------
        tbuf::0  year / 100    19...
        tbuf::1  year mod 100  0...99
        tbuf::2  month         1...12
        tbuf::3  day           1...31
        tbuf::4  hour          0...23
        tbuf::5  minute        0...59
        tbuf::6  second        0...59
        tbuf::7  second/100    0...99

        SYS.TIME never fails and always returns 0. It might return an
        incorrect time, though, and on systems without a clock, it may
        fill 'tbuf' with the same values each time it is called.

        3.9.20 SYS.WAIT

        SYS.WAIT(pid) ! => Num

        Wait for a subprocess to terminate and return its exit code.
        'Pid' contains the process ID of the process to wait for. It
        must be a PID obtained from SYS.SPAWN or SYS.FORK.

        WAIT returns the exit code of the process waited for. The exit
        code is the value passed to the HALT command of T3X. When the
        process waited for terminates abnormally, SYS.WAIT returns -1.
        It also returns -1 when an invalid PID is passed to it.

        NOTE: SYS.WAIT does not actually wait for a specific process to
        terminate. Therefore, no more than a single subprocess should be
        active at the same time.

        See also: SYS.SPAWN, SYS.FORK

        3.10 TTYCTL -- Video Terminal Control

        3.10.1 TTYCTL Class Usage


        The TTYCTL class implements a set of routines for controlling
        character-based video terminal screens and reading terminal
        keyboards. Procedures contained in this class include writing
        to the terminal screen, cursor movement, clearing and scrolling
        screen regions, setting display colors (where available), and
        decoding keyboard input.

        NOTICE: It is generally a bad idea to mix writes to the
        terminal screen using

        T3X.WRITE(T3X.SYSOUT, buffer, length)

        (or even to T3X.SYSERR) with the TTYCTL output functions. Doing
        so may interfere with the internal state of the TTYCTL class
        and cause messed up screen output or even undefined behavior.

        The TTYCTL routines must be initialized by calling TTY.INIT and
        shut down by calling TTY.FINI.

        Although the TTYCTL class does have internal state, no more than
        one instance should be created, because that state reflects the
        state of the controlled video terminal.

        3.10.2 TTY.INIT

        TTY.INIT() ! => 0

        Initialize the TTY control structures. This routine must be
        called before any other procedures of this class can be used.
        It performs the following steps. (Depending on the used
        operating system, some of these steps may be skipped.)

        - Open the controlling TTY (/dev/tty) for reading and writing.

        - Determine the user's terminal type (by evaluating the $TERM
          environment variable).

        - Check the terminal's color capability. Assume color, if
          either (1) the terminal's name contains the substring "color"
          or (2) the terminal's name begins with the prefix "xterm".

        - Check the $ANSICOLOR variable which may be set to YES or NO
          to override the color capability detection.

        - Extract some required properties, control strings, and
          key codes from the termcap(5) database.

        TTY.INIT may fail for any of the following reasons:

        - $TERM is not defined
        - The value of $TERM is not a known TTY type
        - The terminal type does not define one or more of the
          following termcap variables:

        In any of the above cases, an explanatory message will be
        printed and the calling program will be terminated.

        See also: TTY.MODE, TTY.RESET

        3.10.3 TTY.CLEAR

        TTY.CLEAR() ! => 0

        Clear the terminal screen using the currently selected color.

        See also: TTY.CLREOL, TTY.COLOR

        3.10.4 TTY.CLREOL

        TTY.CLREOL() ! => 0

        Clear all characters from the cursor position to the end of the
        current line using the currently selected color.

        See also: TTY.CLEAR, TTY.COLOR

        3.10.5 TTY.COLOR

        TTY.COLOR(color) ! Num => 0

        Select new foreground and background colors. 'Color' is created
        by OR'ing together a foreground and a background color value.
        The following values exist (F_ indicates 'foreground' and B_
        indicates 'background'):

        F_YELLOW, F_GREY,
        B_YELLOW, B_GREY

        The special value F_BRIGHT may be OR'ed in to increase the
        intensity of the foreground color. For example,


        selects bright cyan color on blue background.

        On monochrome terminals, only the color values


        should be considered to be defined.

        See also: TTY.COLORS

        3.10.6 TTY.COLORS

        TTY.COLORS() ! => Num

        Return a non-zero value, if the controlled terminal supports

        See also: TTY.COLUMNS, TTY.LINES

        3.10.7 TTY.COLUMNS

        TTY.COLUMNS() ! => Num

        Return the number of columns per line on the screen of the
        controlled terminal.

        See also: TTY.COLORS, TTY.LINES

        3.10.8 TTY.FINI

        TTY.FINI() ! => 0

        Shutdown the TTYCTL class. Restore the state of the TTY to the
        state at initialization time. To restore all original settings,
        TTY.RESET should be called before TTY.FINI.

        See also: TTY.INIT, TTY.MODE

        3.10.9 TTY.LINES

        TTY.LINES() ! => Num

        Return the number of lines on the screen of the controlled

        See also: TTY.COLORS, TTY.COLUMNS

        3.10.10 TTY.MODE

        TTY.MODE(rawflag) ! Num => 0

        Switch the terminal to 'raw mode'. Some terminals (especially
        in the Unix world) must be in 'raw mode' to allow to read
        single characters from them. In non-raw ('cooked') mode,
        reading a TTY device only returns when CR (ENTER,NL) is
        pressed on the terminal's keyboard. To make the read call
        return immediately after reading one single key, the TTY driver
        must be put in raw mode.

        TTY.MODE(1) selects raw mode and TTY.MODE(0) selects the mode
        the controlled TTY was in when TTY.INIT was called.

        These calls may have no effect on other platforms, but when
        switching a TTY driver to raw mode, it should be switched back
        to its original mode before terminating the program. Otherwise,
        the TTY driver may be left in an undesired state and render the
        controlled TTY inaccessible.

        On some systems, cooked mode may not be implemented. In this
        case, TTY.READC will always return after receiving a single key.

        See also: TTY.READC

        3.10.11 TTY.MOVE

        TTY.MOVE(x, y) ! Num,Num => 0

        Move the cursor to the specified location (column 'x', row
        'y'). If the specified coordinates do not exist on the used
        TTY, the result is undefined. Coordinates start at (0,0) in the
        upper/left corner.

        See also: TTY.COLUMNS, TTY.LINES

        3.10.12 TTY.QUERY

        TTY.QUERY() ! => Num

        Check whether there are characters in the keyboard input buffer.
        If there are any characters, TTY.READC would return when called
        in that moment. Otherwise, it would block.

        TTY.QUERY returns -1 if there are characters in the buffer and
        otherwise 0.

        See also: TTY.READC

        3.10.13 TTY.READC

        TTY.READC() ! => Num

        Read a single character from the terminal's keyboard and return
        its key code. For keys generating ASCII characters, the ASCII
        code of the key will be returned. 'Special' keys like arrow
        keys, PREVIOS PAGE, NEXT PAGE, INSERT, DELETE, and the
        programmable function keys return values above 255. The
        following symbols may be used to match special key codes:

        Key code       Label or Keys
        -------------  ----------------------
        TTYCTL.K_HOME  Home
        TTYCTL.K_LEFT  Left arrow
        TTYCTL.K_RGHT  Right arrow
        TTYCTL.K_END   End
        TTYCTL.K_BKSP  Backspace, <--, <X]
        TTYCTL.K_DEL   Del, Delete, Remove
        TTYCTL.K_KILL  Control + Backspace
        TTYCTL.K_INS   Ins, Insert
        TTYCTL.K_CR    CR, Enter, Return, <-'
        TTYCTL.K_UP    Up arrow
        TTYCTL.K_DOWN  Down arrow
        TTYCTL.K_ESC   ESC, Escape
        TTYCTL.K_PREV  Prev, PgUp, PageUp
        TTYCTL.K_NEXT  Next, PgDn, PageDn
        TTYCTL.K_F1    F1
        TTYCTL.K_F2    F2
        TTYCTL.K_F3    F3
        TTYCTL.K_F4    F4
        TTYCTL.K_F5    F5
        TTYCTL.K_F6    F6
        TTYCTL.K_F7    F7
        TTYCTL.K_F8    F8
        TTYCTL.K_F9    F9
        TTYCTL.K_F10   F10

        Some systems require to switch the TTY driver to raw mode (see
        TTY.MODE) before single characters can be received from a


        3.10.14 TTY.RESET

        TTY.RESET() ! => 0

        Reset the terminal to a sane default state. Can be used to reset
        the original color of the terminal before exiting. Note that
        TTY.FINI should still be called after TTY.RESET. TTY.RESET may
        be a null operation on terminals that do not support it.

        See also: TTY.FINI

        3.10.15 TTY.RSCROLL

        TTY.RSCROLL(top, bottom) ! Num,Num => 0

        Scroll the screen region from 'top' to 'bottom' down by one
        line. At the top of the region, a blank line will be inserted
        using the currently selected color. Line numbers start at 0.


        3.10.16 TTY.SCROLL

        TTY.SCROLL(top, bottom) ! Num,Num => 0

        Scroll the screen region from 'top' to 'bottom' up by one line.
        At the bottom of the region, a blank line will be inserted
        using the currently selected color. Line numbers start at 0.


        3.10.17 TTY.WRITEC

        TTY.WRITEC(c) ! Num => Num

        Write the character 'c' to the terminal screen and return its
        ASCII code. The character will be output at the current cursor
        position. Writing a character advances the cursor. When the
        cursor is at the rightmost column when writing a character, the
        cursor position is undefined after the output operation.

        See also: TTY.READC, TTY.WRITES

        3.10.18 TTY.WRITES

        TTY.WRITES(string) ! Str => 0

        Write a string to the terminal screen as if each character of
        the string had been written using TTY.WRITEC. However,
        TTY.WRITES is usually faster than a series of character-based
        TTY.WRITEC messages.

        Writing to the screen does not necessarily wrap around at the
        end of a line. When the given string does not fit in the space
        between the current cursor position and the end of the line, the
        result of the WRITES operation is undefined.

        See also: TTY.WRITEC, TTY.READC

        3.11 XMEM -- External Memory Interface

        3.11.1 XMEM Class Usage

        OBJECT XM[XMEM];

        The XMEM class provides access to external memory blocks. An
        external memory block is a continuous region of memory not
        contained in the T3X data area. XM blocks are byte addressed.
        Bytes in XM blocks can only be read and written using the
        procedures XM.GET, XM.PUT, and friends, as defined by this

        The XMEM class does have state that is implemented in the
        extension object part. Therefore, only one single instance of
        the class may be loaded.

        Since each external memory block must be completely addressable
        using Tcode machine words, their sizes may not exceed 65536

        The XMEM class must be initialized by XM.INIT before its use
        shut down by calling XM.FINI.

        3.11.2 XM.INIT

        XM.INIT() ! => 0

        Initialize the external memory interface.

        See also: XM.FINI

        3.11.3 XM.ALLOC

        XM.ALLOC(len) ! Num => Num | -1

        Allocate a block of external memory of a size of 'len' bytes.
        Upon success, return an identifier which may be used in
        subsequent XM operations. In case of an error (out of memory /
        out of IDs), return -1.

        See also: XM.FREE

        3.11.4 XM.COPY

        XM.COPY(id, dest, source, len) ! => 0 | -1

        Copy 'len' bytes from address 'source' to address 'dest' of the
        external memory block referenced through ID. Neither 'dest' nor
        'source' may exceed X-1-'len' where X is the size of the block
        as specified at allocation time. 'Dest' and 'source' may overlap.

        XM.COPY returns -1, if an invalid ID is passed to it.

        See also: XM.ALLOC, XM.READ, XM.WRITE

        3.11.5 XM.FINI

        XM.FINI() ! => 0

        Shut down the external memory interface.

        See also: XM.INIT

        3.11.6 XM.FREE

        XM.FREE(id) ! Num => 0 | -1

        Release a previously allocated external memory block. 'Id' is
        an identifier returned by XM.ALLOC. GET returns -1, if an
        invalid ID is passed to it.

        See also: XM.ALLOC

        3.11.7 XM.GET

        XM.GET(id, index) ! Num,Num => Num | -1

        Return the byte stored at address 'index' of the external
        memory block referenced by 'id'. 'Index' may not exceed X-1
        where X is the size of the block as specified at allocation
        time. XM.GET returns -1, if an invalid ID is passed to it.

        See also: XM.ALLOC, XM.PUT, XM.READ.

        3.11.8 XM.PUT

        XM.PUT(id, index, value) ! => Num | -1

        Replace the value of the byte stored at address 'index' of the
        external memory block referenced through 'id' with 'value'.
        'Index' may not exceed X-1 where X is the size of the block as
        specified at allocation time. All but the least significant 8
        bits of 'value' will be discarded.

        XM.PUT returns -1, if an invalid ID is passed to it.

        See also: XM.ALLOC, XM.GET, XM.WRITE.

        3.11.9 XM.READ

        XM.READ(id, index, buffer, len) ! => 0 | -1

        Copy 'len' bytes stored at address 'index' of the external
        memory block referenced by 'id' into 'buffer'. 'Index' may not
        exceed X-1-'len' where X is the size of the block as specified
        at allocation time.

        XM.READ returns -1, if an invalid ID is passed to it.

        See also: XM.ALLOC, XM.COPY, XM.GET, XM.WRITE

        3.11.10 XM.WRITE

        XM.WRITE(id, index, buffer, len) ! => 0 | -1

        Copy 'len' bytes from 'buffer' to the address 'index' of the
        external memory block referenced through ID. 'Index' may not
        exceed X-1-'len' where X is the size of the block as specified
        at allocation time.

        XM.WRITE returns -1, if an invalid ID is passed to it.

        See also: XM.ALLOC, XM.COPY, XM.PUT, XM.READ

        4. The Virtual Tcode Machine

        The Tcode machine is the target of the reference implementation
        of T3X. Tcode is suitable for both interpretation and
        transformation to native code. It also provides mechanisms for
        static linking so that multiple Tcode modules can be linked
        together forming one single program. Since version 3, support
        for object oriented programming is built into the virtual Tcode
        machine. This chapter describes Tcode7 and its virtual machine
        in detail.

        4.1 The Architecture

        The Tcode machine is a virtual 16-bit machine basically
        consisting of the following parts:

        +---------------+0xFFFF                 0xFFFF+---------------+
        |               |                             |               |
        |               |       :-- 16 bits --:       |     Stack     |
        |  U n u s e d  |       +-------------+       |      and      |
        |   S p a c e   |   +---|     IP      |       |    Dynamic    |
        |               |   |   +-------------+  +--->|    Storage    |
        |               |   |   |     RR      |  |    |               |
        |- - - - - - - -|   |   +-------------+  | +->|- - - - - - - -|
        |               |   |   |     FP      |--+ |  |    F r e e    |
        |               |   |   +-------------+    |  |               |
        | Tcode Program |<--+   |     SP      |----+  |  M e m o r y  |
        |               |       +-------------+       |               |
        | Instructions  |       |    SELF     |---+   |- - - - - - - -|
        |               |       +-------------+   |   |    Static     |
        |               |          REGISTERS      +-->|     Data      |
        |               |                             |               |
        +---------------+0x0000                 0x0000+---------------+
           CODE ARRAY                                    DATA ARRAY

        Fig.5 The Architecture of the Tcode Machine

        There are two byte-addressable memory regions called the 'code
        array' and the 'data array'. The code array holds the Tcode
        program that is to be executed and the data array is used to
        hold the data used by the program. Each cell in one of the
        arrays is completely addressable using a 16-bit pointer, so
        the maximum size for each array is 65536 bytes.

        Machine words - which are always 2 bytes wide - are stored with
        the least significant byte in the cell with the lower address:
        0x1234 = 0x34 0x12 (little endian byte ordering). However, this
        byte ordering applies only to the way machine words are stored
        in Tcode programs. Native code back ends of the Tcode machine
        may store machine words in any format, so the result of
        accessing the individual bytes of a machine word is undefined.

        The Tcode machine has five 16-bit wide special purpose registers
        which are outlined in the following overview.

        FP, the Frame Pointer.

        The frame pointer always points to the stack frame (aka context)
        of the currently running procedure. FP is implicitly referenced
        by the instructions LDL, LDLV, SAVL, and INCL, which address
        local objects. FP is modified only by HDR, END, MHDR, and ENDM
        instructions. See also: 'calling conventions'.

        IP, the Instruction Pointer.

        This register always points to the instruction that will be
        interpreted next. IP is interpreted as an offset into the code
        array. It cannot be accessed directly and it is changed by
        jump, call, and branch instructions.

        RR, the Return Register.

        This register is used to transport procedure results back to the
        caller. The return register is loaded by the POP instruction and
        saved by CLEAN. See also: 'calling conventions'.

        SELF, the instance context pointer.

        SELF points to the instance context that is currently in effect.
        This is equal to the first byte of the data space of the object
        that is currently receiving a message. Instance contexts are
        static. They are established using MHDR and released using ENDM.
        The SELF register is used by the LDI, LDIV, and INCI
        instructions to compute the addresses of instance variables. See
        also: 'instance contexts'.

        SP, the Stack Pointer.

        The stack pointer points to the object most recently placed on
        the stack. Moving an object onto the stack implicitly decreases
        SP by one machine word. Removing an object increases it by one
        machine word. SP may be explicitly modified using the STACK

        The Tcode machine instructions can be divided into the following
        nine groups:

        - declarations
        - context manipulation
        - stack manipulation
        - arithmetic
        - predicates
        - loading and storing
        - flow control
        - external linkage
        - source level debugging

        Declarations, external linkage, and debug instructions will be
        processed only once (therefore, they may be resolved in a
        preprocessing step). This means that an instruction like

        STR 5 H e l l o

        will not create a new string literal each time it is
        interpreted, but only at the first time. (One might also think
        of this behavior as creating the same object each time a
        declaration executed).

        Arithmetic instructions and predicates expect their arguments on
        the runtime stack and also place their results there. Since
        there are no general purpose registers, most operations are
        performed on stack elements.

        4.2 Calling Conventions

        A procedure must always begin with a HDR instruction, which
        saves the caller's context and creates a new stack frame. It
        must end with an END instruction, which restores the saved stack
        frame and jumps back to the caller.

        A procedure call

        p(a, b, c)

        where a, b and c be global variables, is coded as follows:

        LDG La LDG Lb LDG Lc CALL Lp CLEAN 3

        Each LDG instruction loads the value of a global variable onto
        the stack. 'Lx' denotes the label pointing at the memory
        location 'x'. CALL performs the procedure call which returns
        with its result in the Return Register (RR). 'Lp' denotes the
        label tagging the procedure 'p'. The final CLEAN instruction
        removes the three arguments from the stack and replaces them
        with the value returned in RR so that the top stack element
        finally holds the procedure return value.

        Each procedure may expect the following stack configuration when

            FP+M  Argument #1
            ...   ...
            FP+3  Argument #N-1
            FP+2  Argument #N
            FP+1  Return Address (saved by CALL or CALR)
            FP+0  Old SP (saved by HDR)
            FP-1  Local Variable #1
            FP-2  Local Variable #2
            ...   ...
        SP  FP-J  Local Variable #K
            (free memory starts here)

        Note1: SP and FP-J point to the same address.

        Note2: The arguments are passed to the procedure in reverse
        order with the first argument at the highest address.

        Both arguments and local variables may be accessed using LDL
        instructions. Given the above context,

        LDL -M

        would access the first argument.

        LDL -2

        always loads the value of the last argument, if any. Local
        storage is accessed using positive offsets:

        NUM 25 SAVL 2

        would store 25 in the second local variable.

        Note: negative values in local storage instructions address
        arguments, negative values address local variables!

        4.3 Instance Contexts

        A method is a procedure that is used to access the data of an
        object. Instead of the usual procedure frame as described in
        the previous section, it should contain instructions to build
        destroy a local context and establish and restore the instance
        context, which is held in the SELF register. The MHDR
        instruction, which establishes the instance context, expects the
        new context pointer at FP+2. ENDM is like END, but also restores
        the instance context of the caller. The Tcode of a method looks
        like this:

        CLAB procedure-label
        ... code ...
        CLAB exit-label

        Passing a message M with three arguments to a global object O


        would be coded as follows:

        NUM 1 NUM 2 NUM 3 LDGV Lo CALL Lm CLEAN 4

        Each called method may expect the following stack configuration:

            FP+M  Argument #1
            ...   ...
            FP+4  Argument #N-1
            FP+3  Argument #N
            FP+2  Receiver's Address
            FP+1  Return Address (saved by CALL or CALR)
            FP+0  Old SP (saved by MHDR)
            FP-1  Sender's Address (Old Instance Context, saved by MHDR)
            FP-2  Local Variable #1
            FP-3  Local Variable #2
            ...   ...
        SP  FP-J  Local Variable #K
            ( Free memory starts here )

        4.4 Instruction Cycles

        A cycle is the set of operations which is required to execute
        one single Tcode instruction. Each cycle consists of the
        following steps:

        (1) Load the instruction pointed to by IP. Increment IP by 1.

        (2) If the instruction has an operand (bit #7 is set), load the
        machine word pointed to by IP into an internal register and
        increment IP by 2.

        (3) If the loaded instruction has an opcode which is equal to or
        greater then the opcode of INIT, the instruction has two
        operands.  Load another machine word (pointed to by IP) into
        another internal register and increment IP by 2.

        (4) If the loaded instruction is valid, execute it, otherwise
        signal an error and halt the machine.

        These steps are repeated until the Tcode machine is halted by
        executing HALT.

        If an instruction modifies stack elements, first all its
        operands are removed from the stack, then the operation denoted
        by the instruction is performed, and finally the result is
        placed back on the stack.

        4.5 Startup Conditions

        This section describes the state of the Tcode machine when it
        is started.

        * The contents of the code array and data array are undefined.

        * The content of the Return Register RR is undefined.

        * The content of the class context pointer SELF is undefined.

        * The Instruction Pointer IP is set to zero. (Hence it is a good
          idea to load programs at the bottom of the code array.)

        * The Stack Pointer SP and the Frame Pointer FP are both set to
          zero (or 65536 on non-16-bit hosts) so that decrementing them
          by two will give the address of the highest addressable
          machine word (65534).

        4.6 Symbols Used to Describe Tcode

        Symbol     (bits)  Description
        ---------  ------  --------------------------------------------
        M N        16      signed numeric values
        .N .M      16      unsigned numeric values
        L          16      a label tagging a data word or a procedure
        A          16      an address tagged by a label
        E          16      a label referencing an external procedure
        I          16      a label referencing an interface procedure
        C          8       a character
        X1...XN    16*X    a vector containing N elements of the type X
        memory[X]  16      the content of the X'th machine word in the
                           data array
        memory::X  8       the content of the X'th byte in the data
        S0...SN    16      the N+1 elements most recently pushed onto
                           the stack


        Normally, the most significant bit of each machine word is
        interpreted as a sign flag (1 indicates a negative number). The
        leading dot notation .N indicates that the MSB of N should be
        treated as a part of the value instead of a sign indicator.

        An address is an offset into the code or data array where the
        base (code or data array) is implicitly specified by the
        associated instruction. Addresses are 16 bits wide.

        External labels are used to create a connection between the name
        of an external procedure and a reference to such a procedure.

        Interface labels are used to create a connection between the
        name of an interface procedure and a reference to such a

        S0 denotes the element most recently pushed onto the stack. When
        popping elements from a stack holding N+1 elements, S0 will be
        removed first and SN will be removed last.

        4.7 Declarations

        0x82 CLAB L - Code LABel

        Define a label identified by the value L that tags a subsequent

        0x85 CREF L - Code REFerence

        Define a word-size storage location holding the address of the
        procedure tagged by the label L.

        0x84 DATA N - DATA definition

        Define a word-size storage location containing the value N.

        0x87 VEC N - VECtor declaration

        Define a vector with a length of N machine words and undefined

        0x83 DLAB L - Data LABel

        Define a label identified by the value L that tags a subsequent
        data object.

        0x86 DREF L - Data REFerence

        Define a word-size storage location holding the address of the
        data object tagged by the label L.

        0xCD INIT N L - INITialize

        Originally used to initialize the Tcode environment - hence its
        name. Each program must begin with this instruction. The
        argument N specifies the Tcode version to which the program
        complies. This document describes version 7 of the Tcode
        language. L is a code label tagging the initial entry point of
        the Tcode module containing the instruction.

        0x88 STR N C1 ... CN - define STRing

        Define a vector with a length of (N+2) / 2 machine words
        containing the characters C1 through CN. Each character is
        stored in a separate byte. All trailing bytes of the vector are
        filled with zeros so that a properly terminated string is

        4.8 Context Manipulation

        0x0A END - END procedure

        Remove two elements S0 and S1 from the stack. Restore the
        context of the calling procedure by loading FP with S0 and then
        perform a branch to S1. S1 is a return address which has been
        saved by a CALL or CALR instruction.

        0x0C ENDM - END Method

        First, load SELF with the value previously saved on the stack by
        MHDR, thereby restoring the instance context of the sender. The
        sender's context will be removed from the stack. Then, perform
        END, as described above.

        0x09 HDR - HeaDeR

        Push the context of the calling procedure (FP) and create a
        fresh procedure context by loading FP with SP.

        0x0B MHDR - Method HeaDeR

        First perform HDR, as described above. Then, push the context of
        the sending method or procedure (SELF) and establish a new
        instance context by loading SELF with the machine word pointed
        to by FP+2 (the last argument passed to the answering method).

        4.9 Stack Manipulation

        0x91 CLEAN N - CLEAN up arguments

        Remove N procedure arguments from the stack: SP := SP + N*2, and
        then push the content of RR, the return register, to the stack.

        0x0E DUP - DUPlicate

        Push the current top of the stack (S0), thereby duplicating it.

        0x0D POP

        Pop the top element S0 and load it into the return register RR.

        0x90 STACK N

        Move the stack pointer SP by N machine words: SP := SP - N*2.
        Moving the stack pointer 'down' (N>=1) allocates space on the
        stack, moving it 'up' (N<=-1) deallocates space. STACK is
        primarily used to allocate and release dynamic memory in

        0x0F SWAP

        Exchange the values of S0 and S1.

        4.10 Arithmetic Instructions

        0x1A ADD

        Remove two elements S0 and S1 and push their sum: S1 + S0.

        0x1C BAND - Bitwise AND

        Remove two elements S0 and S1, perform a bitwise AND on them and
        push the result: S1 & S0.

        0x14 BNOT - Bitwise NOT

        Invert each bit of the top element: ~S0.

        0x1D BOR - Bitwise OR

        Remove two elements S0 and S1, perform a bitwise OR on them and
        push the result: S1 | S0.

        0x1F BSHL - Bitwise SHift Left

        Remove two elements S0 and S1, shift the bits of S1 to the left
        by S0 positions and push the result: S1 << S0.

        0x20 BSHR - Bitwise SHift Right

        Remove two elements S0 and S1, shift the bits of S1 to the right
        by S0 positions, clear the most significant byte, and push the
        result: S1 >> S0.

        0x1E BXOR - Bitwise eXclusive OR

        Remove two elements S0 and S1, perform a bitwise XOR on them and
        push the result: S1 ^ S0.

        0x16 DIV - integer DIVide

        Remove two elements S0 and S1, compute the (signed) integer part
        of their quotient and push it: S1 / S0. If S0 = 0, signal a
        fatal error and halt.

        0xCE INCG L N - INCrement Global

        Add the value N to the memory cell tagged by the label L. This
        is exactly the same as LDG L NUM N ADD SAVG L, but much more

        0xCF INCI M N - INCrement Instance variable

        Add the value N to the memory cell whose absolute address is
        SELF + M*2. INCI M N is equal to LDI M NUM N ADD SAVI M, but
        much more efficient.

        0xD0 INCL M N - INCrement Local

        Add the value N to the memory cell whose absolute address is
        FP - M*2. INCL M N is equal to LDL M NUM N ADD SAVL M, but much
        more efficient.

        0x13 LNOT - Logical NOT

        If the top element is equal to zero, replace it with -1 and
        otherwise replace it with zero: S0 = 0-> -1: 0.

        0x19 MOD - MODulo

        Remove two elements S0 and S1, compute their division remainder
        and push it: S1 MOD S0. S1 MOD S0 is defined as S1 - S1./S0.*S0
        If S0 = 0, signal a fatal error and halt.

        0x15 MUL - MULtiply

        Remove two elements S0 and S1, compute their (signed) product
        and push it: S1 * S0. Do not perform any overflow checking.

        0x12 NEG - NEGate

        Negate the top element: -S0.

        0x00 GLUE - GLUE (no operation)

        Rest for a cycle.

        0x1B SUB - SUBtract

        Remove two elements S0 and S1 and push their difference:
        S1 - S0.

        0x18 UDIV - Unsigned integer DIVide

        Remove two elements S0 and S1, compute the integer part of their
        unsigned quotient and push it: S1 ./ S0. If S0 = 0, signal a
        fatal error and halt.

        0x17 UMUL - Unsigned MULtiply

        Remove two elements S0 and S1, compute their unsigned product
        and push it: S1 .* S0. Do not perform any overflow checking.

        4.11 Predicates

        0x21 EQU - EQUal

        Remove two elements S0 and S1. Push true, if they are equal and
        otherwise push false: S1 = S0-> -1: 0.

        0x24 GRTR - GReaTeR than

        Remove two elements S0 and S1. Push true, if S1 is greater than
        S0 and otherwise oush false: S1 > S0-> -1: 0. S0 and S1 are both

        0x26 GTEQ - Greater Than or EQual to

        Remove two elements S0 and S1. Push true, if S1 is greater than
        or equal to S0 and otherwise push false: S1 >= S0-> -1: 0. S0
        and S1 are both signed.

        0x23 LESS - LESS than

        Remove two elements S0 and S1. Push true, if S1 is less than S0
        and otherwise push false: S1 < S0-> -1: 0. S0 and S1 are both

        0x25 LTEQ - Less Than or EQual to

        Remove two elements S0 and S1. Push true, if S1 is less than or
        equal to S0 and otherwise push false: S1 <= S0-> -1: 0. S0 and
        S1 are both signed.

        0x22 NEQU - Not EQUal

        Remove two elements S0 and S1. Push true, if they are not equal
        and otherwise push false: S1 \= S0-> -1: 0.

        0x28 UGRTR - Unsigned GReaTeR than

        Remove two elements S0 and S1. Push true, if .S1 is greater than
        .S0 and otherwise push false: S1 .> S0-> -1: 0.

        0x2A UGTEQ - Unsigned Greater Than or EQual to

        Remove two elements S0 and S1. Push true, if .S1 is greater than
        or equal to .S0 and otherwise push false: S1 .>= S0-> -1: 0.

        0x27 ULESS - Unsigned LESS than

        Remove two elements S0 and S1. Push true, if .S1 is less than
        .S0 and otherwise push false: S1 .< S0-> -1: 0.

        0x29 ULTEQ - Unsigned Less Than or EQual to

        Remove two elements S0 and S1. Push true, if .S1 is less than or
        equal to .S0 and otherwise push false: S1 .<= S0-> -1: 0.

        4.12 Load and Store Instructions

        0x34 DEREF - DEREFerence

        Remove two values S0 and S1, load a machine word from
        memory[S1/2+S0] and push it. A NORM instruction is implied.
        It converts a pointer (S1) and an offset (S0) to a pointer.

        0x35 DREFB - DeREFerence Byte

        Remove two values S0 and S1, load a single byte from
        memory::(S1+S0), and push it. A NORMB instruction is implied. It
        converts a pointer (S1) and an offset (S0) to a pointer.

        0xAB LDG L - LoaD Global

        Push the value stored in the memory cell A tagged by the label
        L: memory[A/2].

        0xAC LDGV L - LoaD Global Vector

        Push the address A tagged by the label L.

        0xAF LDI N - LoaD Instance variable

        Push the value located at memory[(SELF/2)+N]. This instruction
        is used to load the content of an instance variable. N specifies
        the offset of the variable relative to the beginning of the
        object's data area.

        0xB0 LDIV N - LoaD Instance Vector

        Push the address SELF+N*2. This instruction is used to load the
        address of an instance variable. N specifies the offset of the
        variable relative to the beginning of the data area of the
        currently instance context.

        0xAD LDL N - LoaD Local

        Push the value stored at the N'th position 'below' the stack
        frame base. The address of the cell is computed using the
        formula FP - N*2.  Consequently, negative values of N may be
        used to access locations 'above' the frame base.

        0xB1 LDLAB L - LoaD LABel

        Push the address A tagged by the label L. This instruction is
        similar to LDGV, but more general. It may also be used to
        load addresses of procedures.

        0xAE LDLV N - LoaD Local Vector

        Push the absolute address of a local object. N specifies the
        offset of the object relative to the stack frame base. The
        absolute address is computed using the formula FP - N*2.

        0x36 NORM - NORMalize reference

        Remove two elements S0 and S1 and compute the absolute address
        of the S0'th member of the vector pointed to by S1: S1 + S0*2.
        Push the computed address. This instruction converts a pointer
        plus a machine word offset into a pure pointer that references
        the same location.

        0x37 NORMB - NORMalize Byte reference

        Remove two elements S0 and S1 and compute the absolute address
        of the S0'th byte of the vector pointed to by S1 (S0+S1). Push
        the computed address. This instruction converts a pointer plus
        a byte offset into a pure pointer which references the same
        location. This instruction is technically the same as ADD, but
        conceptionally different.

        0xB2 NUM N - load NUMber

        Push the value N.

        0xB8 SAVG L - SAVe Global

        Pop one element and save it in the memory cell A tagged by the
        label L: memory[A/2] := S0.

        0xBA SAVI N - SAVe Instance variable

        Pop one element and save it in the memory cell
        memory[(SELF/2)+N]. This instruction is used to alter the state
        of an instance variable. N specifies the offset of the variable
        relative to the beginning of the current instance context.

        0xB9 SAVL N - SAVe Local

        Pop one element and save it in the storage cell with the address
        FP - N*2. See also LDL.

        0x33 SELF - SELF reference

        Push the content of the SELF register onto the stack.

        0x3C STORB - STORe Byte

        Pop two elements S0 and S1 and store the least significant 8
        bits of S0 in the byte pointed to by S1: memory::S1 := S0.

        0x3B STORE

        Pop two elements S0 and S1 and store the value S0 in the memory
        cell pointed to by S1: memory[S1/2] := S0.

        4.13 Flow Control

        0xBD BRF L - BRanch on False

        Remove the element S0 and branch to the address A tagged by the
        label L, if S0 is false.

        0xBE BRT L - BRanch on True

        Remove the element S0 and branch to the address A tagged by the
        label L, if S0 is true.

        0xC5 CALL L - procedure CALL

        Push the current value of the instruction pointer IP and then
        perform a branch to the address A tagged by the label L.

        0x46 CALR - CALl through Register

        Push the current value of the instruction pointer IP and then
        remove the element S0 from the stack and perform a branch to the
        location to which it points. The destination is implicitly
        located in the code array of the program (branches to the data
        array cannot be done).

        0xC3 DNEXT L - Downward NEXT

        Remove two elements S0 and S1 and branch to the address A tagged
        by the label L, if S1 <= S0.

        The instructions UNEXT and DNEXT have been designed for use in
        counting loops (called FOR-NEXT loops in BASIC). The idea is as
        follows: Before the loop and at the end of the loop, the current
        loop index and the loop limit are both pushed onto the stack.
        UNEXT compares the values and branches out of the loop, if
        index>=limit. DNEXT branches, if index<=limit. Therefore, UNEXT
        is used in upward counting loops and DNEXT is used in countdown

        0xC8 SYS N - SYStem call (OBSOLETE)

        Execute the system procedure associated with the index value N.
        The index value is removed and a call-dependent number of
        arguments is passed to the respective system procedure. The
        system procedure may return a machine word size return value in
        the Return Register. The stack must me cleaned up using CLEAN
        after calling a system procedure. System procedures are
        implemented as methods of the T3X core class. Therefore, they
        should only be invoked by sending a message to an instance of
        the T3X class. The exact semantics of SYS depend on the called
        NOTE: **********************************************************
        The SYS instruction in no longer required in T3X Release 7, and
        implementations of the Tcode Machine are no longer required to
        support it. Compilers should emit ICALL instead.

        0xCA ICALL N - Interface CALL

        Call the interface procedure located at slot N. Slot values are
        dynamically generated using IPROC, IREF, and ICALX. The
        interface procedure may return a machine word size return value
        in the Return Register. The stack must me cleaned up using CLEAN
        after calling an interface procedure. System procedures are
        normally implemented in languages other than T3X (such as C or
        assembly language). They are used to extend the T3X runtime
        environment. The exact semantics of ICALL depend on the called

        0xC4 HALT N

        Instantly halt the Tcode machine. The least significant eight
        bits of N will be delivered as a status code to the process that
        invoked the Tcode Machine program.

        0xC1 JUMP L

        Unconditionally jump to the address A tagged by the label L.

        0xBF NBRF L - Nondestructive BRanch on False

        Branch to the address A tagged by the label L, if S0 is false.
        Do not remove S0.

        0xC0 NBRT L - Nondestructive BRanch on True

        Branch to the address A tagged by the label L, if S0 is true. Do
        not remove S0.

        0xC2 UNEXT L - Upward NEXT

        Remove two elements S0 and S1 and branch to the address A tagged
        by the label L, if S1 >= S0. See also: DNEXT.

        4.14 External Linkage

        0xC7 CALX E - CALl eXternal procedure

        Call a procedure contained in a different module. There must be
        an EXT record defining the external label E in the same module
        and a PUB record with the same name in another module. Both are
        required for the external reference to be resolved.

        0xD5 CMAP N M - Call MAP

        Call maps describe the arguments of interface procedures. The
        operands of CMAP are identical to the parameters of interface
        declarations (see IDECL for details).

        0xD2 EXT E N C1 ... CN - EXTernal reference

        Create an external reference to the symbol represented by the
        characters C1 through CN. E is a so-called external label. Such
        labels are used to reference external symbols in CALX
        instructions. See the section on loading Tcode for details.
        Note: When interpreting a Tcode program, this instruction will
        cause an error (unresolved external reference).

        0xCB ICALX I - Interface CALl of eXternal procedure

        Call the interface procedure described by the interface label I.
        There must be an IREF record defining I in the same module. A
        corresponding IPROC record with the same name as the IREF record
        must be supplied to the Tcode loader to resolve the reference.

        0xC9 ILIB N C1 ... CN - Interface LIBrary

        Name the relocatable object file or library containing the
        interface procedures described by the IPROC records in this
        module. This name can be used for linking the necessary
        libraries when translating Tcode to native code or to make
        sure that a Tcode machine provides the required extensions.

        0xD3 IPROC N M C1 ... CN - Interface PROCedure

        Assign the interface procedure named by the characters C1 though
        CM to the interface slot N. IPROC records are also used to
        resolve ICALX instructions. See the section on resolving
        interface references for details.

        0xD4 IREF I N C1 ... CN - Interface REFerence

        Create an interface reference to the symbol represented by the
        characters C1 through CN. I is a so-called interface label. Such
        labels are used to reference interface procedures in ICALX
        instructions. See the section on resolving interface references
        for details. Note: When interpreting a program containing IREF
        by a Tcode machine, this instruction will lead to an error
        (unresolved interface reference).

        0xD1 PUB L N C1 ... CN - PUBlic reference

        Signal the Tcode linker that the procedure tagged by the label L
        is public and can be referenced externally using the name formed
        by the characters C1 through CN. Again, see the section on
        loading Tcode for details. When interpreting programs using a
        Tcode machine, this instruction type may be ignored safely.

        4.15 Source Level Debugging Support

        0xD6 GSYM L N C1...CN - Global SYMbol

        Name a global symbol. C1 through CN contain the characters of
        the symbol name. L is the ID of the label which marks the named
        symbol. GSYM instructions should be generated for global
        variable names.

        0xD8 ISYM M N C1...CN - Instance SYMbol

        Name an instance variable. C1 through CN contain the characters
        of the symbol name. M is the offset in machine words (into the
        current class context) of the data object named by the symbol.

        0xC6 LINE N - LINE number

        Indicate that the following instructions were created from line
        N of the source code from which the Tcode program was generated.

        0xD7 LSYM M N C1...CN - Local SYMbol

        Name a local symbol. C1 through CN contain the characters of the
        symbol name. M is a signed number holding the position of the
        variable relative to the Frame Pointer.

        4.16 Meta Information

        0x81 HINT N - pass a HINT

        Pass some interesting information to later stages of the
        compiler, like the optimizer or the code generator. The
        information itself is encoded in a single machine word. Hint
        instructions are null operations when executed by the Tcode
        machine. When a program (like a code generator) detects a HINT
        in a context where it does not expect one, it may safely ignore

        4.16.1 Generated Hints

        The following hints are currently generated by Release 7 of the
        the T3X translator

        HDR HINT N

        The number of formal arguments of the procedure being defined.

        Note: This value is particularly useful when translating Tcode
        back into a high level language, since they allow to create a
        procedure header with the proper number of formal arguments
        without having to scan the entire procedure body.

        MHDR HINT N

        The number of formal arguments of the method being defined plus
        one. The new instance context passed to the method is also an
        argument. Therefore, N is one higher than the number of
        arguments defined in the original T3X source program. See also
        the note in description of the HDR instruction.

        VEC M HINT N

        The size of an allocation unit in bytes. If this hint is
        present, the VEC statement will allocate

             M + ((BPW/N)-1)
        M' = ---------------

        machine words (where BPW = the size of bytes per machine word on
        the target platform). Currently, N=2 is emitted for byte
        vectors. Optionally, a value of N=0 may be considered equal to
        N=BPW. On the Tcode machine, M'=M since BPW=2.

        This hint is intended to avoid over-allocation of memory on
        native platforms with large machine word sizes. Byte vector
        sizes are always specified in 16-bit machine words, which would
        lead to 2x over-allocation on 32-bit and 4x over-allocation on
        64-bit machines. This hint allows non-16-but back ends to
        allocate memory for byte vectors more efficiently.

        4.17 Loading Tcode

        The Tcode machine provides a set of instructions for linking
        together modules that were compiled separately to Tcode.
        External references are limited to procedure calls. This means
        that a module can call procedures defined in an external module,
        but it cannot access the data of an external module.

        To call an external procedure, the label that tags the entry
        point of the routine must be declared public (using a PUB
        instruction) in the module containing the called routine. In the
        module of the caller, it must be declared extern (using EXT).
        The 'extern' declaration creates a so-called external label
        which may be used in calls to external procedures (CALX

        The PUB instruction provides a symbolic name for a procedure.
        This symbolic name may be referred to by an EXT instruction in a
        different module. CALX is used to refer to an EXT instruction
        defined in the same module as the EXT record. An external
        reference is resolved in four steps:

        (1) The external label, which is the operand of a CALX
        instruction, is looked up in the external symbol table (a table
        holding EXT records).

        (2) The name contained in the EXT record is looked up in the
        public symbol table (a table holding PUB records).

        (3) The label contained in the matching PUB record replaces the
        external label in the CALX instruction.

        (4) CALX is replaced with CALL.

        The following figure illustrates the principle of external

        +----------------------------+      +----------------------------+
        |                            |      |                            |
        |    ,--> EXT E 4 name >================> PUB L 4 name           |
        |    |                       |      |     CLAB L HDR ... END     |
        |    '-------.-.-----------, |      |                            |
        |            | |           | |      |                            |
        |  CALX E >--' |  CALX E >-' |      |                            |
        |              |             |      |                            |
        |     CALX E >-'             |      |                            |
        |                            |      |                            |
        +----------------------------+      +----------------------------+
               Caller's Module                     Callee's Module

        Fig.6 External References

        Additional Notes

        Since labels are represented by integers in Tcode, label
        collisions will occur when linking two (or more) Tcode modules
        together. Therefore, labels must be renamed in this case: When
        a module A already has been loaded and a module B is to be
        loaded, the highest label ID used in A must be added to each
        (non-external) label in B.

        Two or more EXT records with the same name may exist, because
        the same symbol may be associated with different external labels
        in different modules.

        The existence of two PUB records with the same name is an error
        (redefinition error).

        There must be a matching PUB record for each EXT record.
        Otherwise, an error is signalled (unresolved external).

        4.18 Resolving Interface References

        Interface references are references to procedures which are not
        located in the code area of the Tcode machine. Modules can
        provide interfaces by exporting IPROC records and request
        interfaces using IREF records.  ICALX instructions are used to
        call unresolved interfaces and ICALL instructions are used to
        call resolved interfaces.

        Notice that resolving an interface means to assign a unique slot
        number to the name of an interface procedure. It is in the
        responsibility of the Tcode machine to place the correct
        procedure in this slot. IPROC records are used to assign names
        to slot numbers.

        Since multiple modules may export IPROC records and each module
        starts numbering interfaces at one, the Tcode loader must
        relocate IPROCs by assigning unique slot numbers to them.

        IPROC records and corresponding IREF records may be located in
        the same module. In any case, though, the IPROC record must
        precede the IREF record and the IREF record must precede any
        ICALX instructions referencing the IREF record. In this
        section, IPROC and IREF are assumed to be in a different

        IPROC records provide a symbolic name for an interface
        procedure. This symbolic name may be referred to by IREF records.
        ICALX instructions are used to refer to IREF records defined in
        the same module. An external reference is resolved in four

        (1) The that which is the operand of a ICALX instruction is
        looked up in a table holding IREF records.

        (2) The name contained in the matching IREF record is looked up
        in another table which holds IPROC records.

        (3) The label contained in the matching IPROC record replaces
        the external label in the ICALX instruction.

        (4) ICALX is replaced with ICALL.

        The following figure illustrates the principle of interface

               Caller's Module                     Callee's Module
        +----------------------------+      +----------------------------+
        |                            |      |                            |
        |    ,--> IREF I 4 name >===============> IPROC L 4 name         |
        |    |                       |      |          \/                |
        |    '-------.-.-----------, |      |          ||                |
        |            | |           | |      |          ||                |
        | ICALX I >--' | ICALX I >-' |      |          ||                |
        |              |             |      |          ||                |
        |    ICALX I >-'             |      |          ||                |
        |                            |      |          ||                |
        +----------------------------+      +----------||----------------+
               Tcode Machine or external object        ||
        |                                              ||                |
        |      Interface procedure slot #0             ||                |
        |      ...                                     ||                |
        |      Interface procedure slot #L <============'                |
        |      ...                                                       |
        |                                                                |

        Fig.7 Interface References

        The existence of two IPROC records with the same name is an
        error (redefinition error).

        5. Formal Definitions

        5.1 Syntax Description Language

        The syntax of the T3X language is described using a BNF-style
        format similar to the one accepted by the YACC parser generator.
        A more detailed description follows.

        The T3X grammar is described as a set of rules of the following

             | Pattern-2
             | ...
             | Pattern-N

        It reads 'Name may also be written as Pattern-1 OR Pattern-2
        OR ...  OR Pattern-N'. Each pattern may consist of names of
        rules or terminal symbols. Each terminal symbol is enclosed in
        apostrophes, like '=', 'CONST', or '0x'. An apostrophe may be
        included in a terminal (symbol) by doubling it. Consequently, a
        terminal represented by an apostrophe is written ''''.

        Here is an example: The rule

        BinaryDigit: '0' | '1' ;

        is read 'A BinaryDigit may be represented by either the string
        '0' or the string '1'. A (recursive) rule to define
        arbitrary-length binary numbers based upon BinaryDigit would
        look like this:

             | BinaryDigit BinaryNumber

        In this case, a BinaryNumber would be either a single
        BinaryDigit or a BinaryDigit followed by another BinaryNumber
        (and therefore more BinaryDigits).

        The ampersand symbol (&) is used to indicate that no white space
        is allowed between the elements of a pattern. While the above
        rule would match the string

        1 0 1 1 0 1 0 1

        a rule containing the concatenation symbol would match the


        Such a rule would be written as:

             | BinaryDigit & BinaryNumber

        Ellipses (...) are used to represent obvious parts of sequences
        in patterns. For example, the following two patterns are equal:



        A special rule named <character>, which is not defined inside
        of the formal grammar, is used to refer to an arbitrary ASCII

        Note that T3X is case-insensitive. Hence all alphabetic string
        in the following grammar match any combination of upper and
        lower case letters. E.g. 'WHILE' would match 'while', 'While',
        'WHILE', 'wHiLe', and all other combinations.

        5.2 The Formal Syntax

               DeclList CompoundStmt

             | Declaration DeclList

               'VAR' VarDeclList ';'
             | 'CONST' ConstDeclList ';'
             | 'DECL' ForwardDeclList ';'
             | 'STRUCT' Symbol '=' StructMemList ';'
             | ProcDecl
             | ClassDecl
             | 'PUBLIC' ClassDecl
             | 'OBJECT' ObjDeclList ';'
             | 'MODULE' Symbol '(' ModList ')' ';'
             | 'MODULE' Symbol '(' ')' ';'

             | VarDeclList ',' VarDecl

             | Symbol '[' ConstValue ']'
             | Symbol '::' ConstValue

               Symbol '=' ConstValue
             | Symbol '=' ConstValue ',' ConstDeclList

             | Symbol ',' ModList

             | Symbol ',' StructMemList

               'CLASS' Symbol '(' ModList ')' InstDeclList 'END'
             | 'CLASS' Symbol '(' ')' InstDeclList 'END'
             | 'ICLASS' Symbol '(' String ')' IClassInstDeclList 'END'

             | IClassInstDecl IClassInstDeclList

             | InstDecl InstDeclList

             | InstDecl

               'VAR' VarDeclList ';'
             | 'CONST' ConstDeclList ';'
             | 'DECL' ForwardDeclList ';'
             | 'STRUCT' Symbol '=' StructMemList ';'
             | ProcDecl
             | 'OBJECT' ObjDeclList ';'
             | 'PUBLIC' ProcDecl
             | 'PUBLIC' 'CONST' ConstDeclList ';'
             | 'PUBLIC' 'STRUCT' Symbol '=' StructMemList ';'

               'IDECL' InfDeclList

             | InfDecl ',' InfDeclList

               Symbol '(' ConstValue ',' ConstValue ')'

               Symbol '[' Symbol ']'
             | Symbol '[' Symbol ']' ',' ObjDeclList

             | ForwardDecl ',' ForwardDeclList

               Symbol '(' ConstValue ')'

               Symbol '(' ArgumentList ')' Statement
             | Symbol '(' ')' Statement

             | Symbol ',' ArgumentList

             | Symbol ':=' Expression ';'
             | Symbol Subscripts ':=' Expression ';'
             | ProcedureCall
             | 'CALL' ProcedureCall ';'
             | Symbol '.' ProcedureCall ';'
             | 'SEND' '(' Symbol ',' Symbol ',' ProcedureCall ')' ';'
             | 'IF' '(' Expression ')' Statement
             | 'IE' '(' Expression ')' Statement 'ELSE' Statement
             | 'WHILE' '(' Expression ')' Statement
             | 'FOR' '(' Symbol '=' Expression ',' Expression ')'
             | 'FOR' '(' Symbol '=' Expression ',' Expression ','
                         ConstValue ')'
             | 'LEAVE' ';'
             | 'LOOP' ';'
             | 'RETURN' ';'
             | 'RETURN' Expression ';'
             | 'HALT' ';'
             | 'HALT' ConstValue ';'
             | ';'

               'DO' 'END'
             | 'DO' LocalDeclList 'END'
             | 'DO' StatementList 'END'
             | 'DO' LocalDeclList StatementList 'END'

             | LocalDecl LocalDeclList

               'VAR' VarDeclList ';'
             | 'CONST' ConstDeclList ';'
             | 'STRUCT' Symbol '=' StructMemList ';'
             | 'OBJECT' ObjDeclList ';'

             | Statement StatementList

             | Disjunction '->' Expression ':' Expression

             | Disjunction '\/' Conjunction

             | Conjunction '/\' Equation

             | Equation '=' Relation
             | Equation '\=' Relation

             | Relation '<' BitOperation
             | Relation '>' BitOperation
             | Relation '<=' BitOperation
             | Relation '>=' BitOperation
             | Relation '.<' BitOperation
             | Relation '.>' BitOperation
             | Relation '.<=' BitOperation
             | Relation '.>=' BitOperation

             | BitOperation '&' Sum
             | BitOperation '|' Sum
             | BitOperation '^' Sum
             | BitOperation '<<' Sum
             | BitOperation '>>' Sum

             | Sum '+' Term
             | Sum '-' Term

             | Term '*' Factor
             | Term '/' Factor
             | Term '.*' Factor
             | Term './' Factor
             | Term 'MOD' Factor

             | String
             | Table
             | 'PACKED' PackedTable
             | Symbol
             | Symbol Subscripts
             | Symbol '.' Symbol
             | ProcedureCall
             | 'CALL' ProcedureCall
             | Symbol '.' ProcedureCall
             | 'SEND' '(' Symbol ',' Symbol ',' ProcedureCall ')'
             | '@' Symbol
             | '-' Factor
             | '\' Factor
             | '~' Factor
             | '(' Expression ')'

               '[' Expression ']'
             | '::' Factor
             | '[' Expression ']' Subscripts

               '[' MemberList ']'

             | TableMember ',' MemberList

             | String
             | Table
             | 'PACKED' PackedTable
             | '@' Symbol
             | '(' Expression ')'

               '[' PackedTableMembers ']'

             | PackedTableMember ',' PackedTableMembers

             | Number

               Symbol '(' ')'
             | Symbol '(' ExprList ')'

             | Expression ',' ExprList

             | ConstValue '*' SimpleConst
             | ConstValue '+' SimpleConst
             | ConstValue '-' SimpleConst
             | ConstValue '|' SimpleConst

             | Number
             | '-' SimpleConst
             | '~' SimpleConst

             | '%' & DecimalNumber
             | '0X' & HexNumber
             | '0B' & BinaryNumber
             | '%' & '0X' & HexNumber
             | '%' & '0B' & BinaryNumber
             | '''' & AnyChar & ''''

             | BinaryDigit & BinaryNumber

             | DecimalDigit & DecimalNumber

             | HexDigit & HexNumber

             | Letter & SymbolChars

             | Letter
             | DecimalDigit
             | Letter & SymbolChars
             | DecimalDigit & SymbolChars





               '"' & StringChars & '"'

             | AnyChar & StringChars

             | '\' & <character>

        6. Quick Reference

        6.1 Language Overview

        6.1.1 Declarations

        Statement                        Description
        -------------------------------  ---------------------------
        VAR name, ... ;                  Declare atomic variables
        VAR name[cexpr], ... ;           Declare vectors
        VAR name::cexpr, ... ;           Declare byte vectors
        VAR name[structname], ...;       Declare structured vectors
        CONST name = cexpr, ... ;        Declare constants
        PUBLIC CONST ...                 Declare class constants
        STRUCT sname = m1, ... mN ;      Declare structure
        PUBLIC STRUCT ...                Declare class constants
        CLASS cname(required, ...)       Declare class
        PUBLIC CLASS ...                 Declare public class
        OBJECT name[cname], ... ;        Declare instance
        DECL pname(cexpr), ... ;         Forward-declare procedures
        pname(a1, ..., aN) stmt          Declare procedure
        PUBLIC pname(a1, ..., aN) stmt   Declare method
        ICLASS cname("object-name") ...  Declare interface class
        IDECL iname(args, cmap);         Declare interface procedure
        MODULE mname(required, ...);     Declare module

        6.1.2 Statements

        Statement                         Description
        ------------------------------    ---------------------------
        lvalue := expr;                   Assignment
        IF (expr) stmt                    Conditional statement
        IE (expr) stmt-T ELSE stmt-F      IF with alternative
        WHILE (expr) stmt                 Unbounded loop
        FOR (var=start, limit, cexpr)     Counting loop
        FOR (var=start, limit) stmt       Counting loop, step = 1
        LEAVE;                            Leave innermost WHILE/FOR
        LOOP;                             Restart innermost WHILE/FOR
        RETURN expr;                      Exit procedure, return value
        RETURN;                           Exit, return 0
        HALT cexpr;                       Halt program, return status
        HALT;                             Halt with 0
        pname(a1, ..., aN);               Procedure call
        CALL ptr(a1, ..., aN);            Indirect procedure call
        oname.mname(a1, ..., aN)          Send message
        SEND(ptr, cname, mname(a1, ..., aN));
                                          Send indirect message
        DO decls ... stmts ... END        Compound statement
        ;                                 Empty statement

        6.1.3 Operators

        Operator    Prec Assoc Description
        ----------- ---- ----- ----------------------------------
        (expr)      0    -     Grouping
        p(a1, ...)  0    L     Procedure call
        v[expr]     0    L     Array subscript
        v::expr     0    R     Byte subscript
        @lvalue     1    -     Address of lvalue
        ~X          2    -     Bitwise complement
        \X          2    -     Logical complement
        -X          2    -     Negation
        X * Y       3    L     Product
        X / Y       3    L     Quotient (integer part)
        X MOD Y     3    L     Division remainder
        X .* Y      3    L     Product (unsigned)
        X ./ Y      3    L     Quotient (unsigned)
        X + Y       4    L     Sum
        X - Y       4    L     Difference
        X & Y       5    L     Bitwise product (AND)
        X | Y       5    L     Bitwise sum (OR)
        X ^ Y       5    L     Bitwise not-equal operation (XOR)
        X << Y      5    L     Bitwise left shift
        X >> Y      5    L     Bitwise right shift
        X < Y       6    L     Less than
        X <= Y      6    L     Less/equal
        X > Y       6    L     Greater than
        X >= Y      6    L     Greater/equal
        X .< Y      6    L     Less than (unsigned)
        X .<= Y     6    L     Less/equal (unsigned)
        X .> Y      6    L     Greater than (unsigned)
        X .>= Y     6    L     Greater/equal (unsigned)
        X = Y       7    L     Equal to
        X \= Y      7    L     Not equal to
        X /\ Y      8    L     Short circuit logical AND
        X \/ Y      9    L     Short circuit logical OR
        X -> Y : Z 10    L     Conditional expression

        6.1.4 Meta Commands

        Meta command          Description
        ------------------    --------------------------------------
        #CLASSPATH "path";    Alternative location for locating
                              class files.
        #DEBUG;               Emit debug information

        6.2 Runtime Support Routines

        6.2.1 T3X

        Procedure                    Description
        ---------------------------  -----------------------------
        T.BPW()                      Bytes per machine word
        T.CLOSE(fd)                  Close file
        T.CVALIST(n, bmap, in, out)  Convert argument list
        T.GETARG(n, buf, max)        Copy command line argument
        T.GETENV(name, buf, max)     Copy environment variable
        T.MEMCOMP(r1, r2, len)       Compare memory regions
        T.MEMCOPY(dest, src, len)    Copy memory regions
        T.MEMFILL(dest, char, len)   Fill memory regions
        T.MEMSCAN(src, char, lim)    Search for bytes in memory
        T.NEWLINE(buf)               Retrieve newline sequence
        T.OPEN(path, mode)           Open file
        T3X.OAPPND                   Append mode
        T3X.ORDWR                    Read/write mode
        T3X.OREAD                    Read-only mode
        T3X.OWRITE                   Write-only mode
        T.READ(fd, buf, len)         Read file
        T.REMOVE(path)               Delete file
        T.RENAME(old, new)           Rename file
        T.SEEK(fd, pos, org)         Move file pointer
        T3X.SEEK_BCK                 Origin: relative, go back
        T3X.SEEK_END                 Origin: end, go back
        T3X.SEEK_FWD                 Origin: relative, go forward
        T3X.SEEK_SET                 Origin: beginning, go forward
        T.WRITE(fd, buf, len)        Write file

        6.2.2 Char

        Procedure        Description
        ---------------  ------------------------------
        CHR.INIT()       Initialize CHAR class
        CHAR.C_ALPHA     property: alphabetic
        CHAR.C_CNTRL     property: control character
        CHAR.C_DIGIT     property: decimal digit
        CHAR.C_SPACE     property: white space
        CHAR.C_UPPER     property: upper case character
        CHR.ALPHA(char)  Alphabetic character test
        CHR.ASCII(char)  ASCII character test
        CHR.CNTRL(char)  Control character test
        CHR.DIGIT(char)  Numeric character test
        CHR.LCASE(char)  Convert to lower case
        CHR.LOWER(char)  Lower case character test
        CHR.MAP()        Return property map
        CHR.SPACE(char)  White space character test
        CHR.UCASE(char)  Convert to upper case
        CHR.UPPER(char)  Upper case character test

        6.2.3 IOStream

        Procedure                       Description
        ------------------------------  -------------------------------
        IOS.CLOSE()                     Close stream
        IOS.CREATE(fd, buf, len, mode)  Create stream
        IOS.EOF()                       EOF test
        IOS.FLUSH()                     Write buffered data
        IOS.MOVE(offset, origin)        Move stream pointer
        IOSTREAM.SEEK_BCK               Current position, move back
        IOSTREAM.SEEK_END               End of file, move back
        IOSTREAM.SEEK_FWD               Current position, move forward
        IOSTREAM.SEEK_SET               Beginning of file, move forward
        IOS.OPEN(path, buf, len, mode)  Open file
        IOSTREAM.FADDCR                 Add CR before LF on output
        IOSTREAM.FKILLCR                Remove CR from input
        IOSTREAM.FRDWR                  Read/write mode
        IOSTREAM.FREAD                  Read-only mode
        IOSTREAM.FTRANS                 FADDCR and FKILLCR
        IOSTREAM.FWRITE                 Write-only mode
        IOS.RDCH()                      Read single character
        IOS.READ(buf, len)              Read data
        IOS.RESET()                     Reset error flag
        IOS.READS(buf, len)             Read single line
        IOS.WRCH(char)                  Write single character
        IOS.WRITE(buf, len)             Write data
        IOS.WRITES(str)                 Write string

        6.2.4 Memory

        Procedure                      Description
        -----------------------------  ---------------
        MEM.INIT(pool, size)           Initialize pool
        MEM.WALK(block, sizep, statp)  Walk block list
        MEM.ALLOC(size)                Allocate block
        MEM.FREE(block)                Free block

        6.2.5 String

        Procedure                    Description
        ---------------------------  -------------------------
        STR.COMP(s1, s2)             Compare strings
        STR.COPY(dest, src)          Copy string
        STR.FIND(str, pat)           Find substring
        STR.FORMAT(buf, tmpl, list)  Format string
        STR.LENGTH(str)              Length of string
        STRING.MAXLEN                Maximum string length
        STR.NUMTOSTR(buf, n, radix)  Convert number to string
        STR.PARSE(str, tmpl, list)   Extract fields
        STR.RSCAN(str, char)         Find rightmost character
        STR.SCAN(str, char)          Find character
        STR.STRTONUM(s, radix, lp)   Convert string to number
        STR.XLATE(str, old, new)     Replace characters

        6.2.6 Tcode

        Procedure                  Description
        ------------------------   --------------------------------
        TCODE.Iinstruction Tcode   Instruction constants, see 6.3.2

        6.2.7 Util

        Procedure                   Description
        --------------------------  --------------------------
        UTIL.BUFLEN                 Maximum string length
        U.PRINTF(tmpl, args)        Format and write to SYSOUT
        U.SWRITEF(ios, tmpl, args)  Format and write to stream
        U.WRITEF(fd, tmpl, args)    Format and write to file

        6.2.8 System

        Procedure                     Description
        ----------------------------  ----------------------------------
        SYS.INIT()                    Initialize system interface
        SYS.FINI()                    Shut down system interface
        SYS.CHDIR(path)               Change current working directory
        SYS.CLOSEDIR(dirfd)           Close directory
        SYS.DUP(fd)                   Duplicate file descriptor
        SYS.DUP2(old, new)            Duplicate FD to existing FD
        SYS.FORK()                    Duplicate the calling process
        SYS.KILL(pid, sig)            Send signal to process
        SYSTEM.SIGKILL                Signal: kill process
        SYSTEM.SIGTERM                Signal: terminate process
        SYSTEM.SIGTEST                Signal: test process id
        SYS.MKDIR(path)               Create new directory
        SYS.OPENDIR(path)             Open directory
        SYS.PIPE(fdvec)               Create pipe
        SYS.RDCHK(fd)                 Check FD for pending input
        SYS.READDIR(dirfd, buf, max)  Read directory entry
        SYS.RMDIR(path)               Remove directory
        SYS.SPAWN(prog, args, mode)   Spawn program 
        SYSTEM.SPAWN_NOWAIT           Mode: create background process
        SYSTEM.SPAWN_WAIT             Mode: wait for process termination
        SYS.STAT(path, sb)            Retrieve file statistics
        SYSTEM.STATBUF                File statistics buffer (SB)
        SYSTEM.ST_DEV                 SB: device ID
        SYSTEM.ST_EXT                 SB: size of file / 64K
        SYSTEM.ST_GID                 SB: group ID of owner
        SYSTEM.ST_INO                 SB: inode number
        SYSTEM.ST_MODE                SB: permission flags
        SYSTEM.ST_MTIME               SB: modification time
        SYSTEM.ST_MT_2                SB: 8-byte MTIME buffer
        SYSTEM.ST_MT_3                SB: 8-byte MTIME buffer
        SYSTEM.ST_MT_4                SB: 8-byte MTIME buffer
        SYSTEM.ST_NLINK               SB: number of links
        SYSTEM.ST_RDEV                SB: device type
        SYSTEM.ST_SIZE                SB: size mod 64K
        SYSTEM.ST_UID                 SB: user ID of owner
        SYS.TIME(tbuf)                Get system time
        SYS.WAIT(pid)                 Wait for subprocess termination

        6.2.9 TTYCtl

        Procedure               Description
        ---------------------   -----------------------------------
        TTY.INIT()              Initialize TTY interface
        TTY.FINI()              Shut down TTY interface
        TTY.CLEAR()             Clear terminal screen
        TTY.CLREOL()            Clear to end of line
        TTY.COLOR(color)        Select color
        TTYCTL.B_BLACK          \
        TTYCTL.B_BLUE            |
        TTYCTL.B_CYAN            |
        TTYCTL.B_GREEN           | Background colors
        TTYCTL.B_GREY            |
        TTYCTL.B_MAGENTA         |
        TTYCTL.B_RED             |
        TTYCTL.B_YELLOW         /
        TTYCTL.F_BLACK          \
        TTYCTL.F_BLUE            |
        TTYCTL.F_CYAN            |
        TTYCTL.F_GREEN           | Foreground colors
        TTYCTL.F_GREY            |
        TTYCTL.F_MAGENTA         |
        TTYCTL.F_RED             |
        TTYCTL.F_YELLOW         /
        TTYCTL.F_BRIGHT         Foreground intensity flag
        TTY.COLORS()            Can the TTY do color?
        TTY.COLUMNS()           Number of columns on TTY screen
        TTY.MODE(raw)           Select raw or cooked mode
        TTY.MOVE(x, y)          Move cursor
        TTY.LINES()             Number of lines on TTY screen
        TTY.QUERY()             Check for pending keyboard input
        TTY.READC()             Read single character from keyboard
        TTYCTL.K_BKSP           Key code: Backspace          
        TTYCTL.K_CR             Key code: Enter / CR / Return
        TTYCTL.K_DEL            Key code: Delete             
        TTYCTL.K_DOWN           Key code: Down arrow         
        TTYCTL.K_END            Key code: End                
        TTYCTL.K_ESC            Key code: Escape             
        TTYCTL.K_F1             Key code: F1                 
        TTYCTL.K_F2             Key code: F2                 
        TTYCTL.K_F3             Key code: F3                 
        TTYCTL.K_F4             Key code: F4                 
        TTYCTL.K_F5             Key code: F5                 
        TTYCTL.K_F6             Key code: F6                 
        TTYCTL.K_F7             Key code: F7                 
        TTYCTL.K_F8             Key code: F8                 
        TTYCTL.K_F9             Key code: F9                 
        TTYCTL.K_F10            Key code: F10                
        TTYCTL.K_HOME           Key code: Home               
        TTYCTL.K_INS            Key code: Insert             
        TTYCTL.K_KILL           Key code: Kill, Erase line   
        TTYCTL.K_LEFT           Key code: Left arrow         
        TTYCTL.K_NEXT           Key code: Next, Page down    
        TTYCTL.K_PGDN           Key code: Next, Page down    
        TTYCTL.K_PGUP           Key code: Prev, Page up      
        TTYCTL.K_PREV           Key code: Prev, Page up      
        TTYCTL.K_RIGHT          Key code: Right arrow        
        TTYCTL.K_UP             Key code: Up arrow           
        TTY.RSCROLL(top, bot)   Scroll region down
        TTY.SCROLL(top, bot)    Scroll region up
        TTY.WRITEC(char)        Write single character to screen
        TTY.WRITES(string)      Write string to screen

        6.2.10 XMem

        Procedure                      Description
        -----------------------------  ------------------------------
        XM.INIT()                      Initialize XMEM interface
        XM.FINI()                      Shut down XMEM interface
        XM.ALLOC(size)                 Allocate external memory block
        XM.FREE(id)                    Release allocated block
        XM.GET(id, index)              Read single byte from block
        XM.PUT(id, index, value)       Write single byte from block
        XM.READ(id, index, buf, len)   Read region from block
        XM.WRITE(id, index, buf, len)  Write region to block

        6.3 Miscellanea

        6.3.1 Escape Sequences

        Escape    ASCII
        Sequence  Code   Name  Description
        --------  -----  ----  --------------------------
        \a \A     0x07   BEL   Ring terminal bell
        \b \B     0x08   BS    Backspace
        \e \E     0x1B   ESC   Introduce control sequence
        \f \F     0x12   FF    Form feed
        \n \N     0x0A   LF    Line feed
        \q \Q \"  0x22   -     Literal quote character
        \r \R     0x0D   CR    Carriage return
        \t \T     0x09   HT    Horizontal tabulator
        \v \V     0x0B   VT    Vertical tabulator
        \\        0x5C   -     Literal backslash

        6.3.2 The Tcode Instruction Set

        0x 00/80 10/90  20/A0 30/B0  40/C0  50/D0
        00 GLUE  STACK* BSHR  LDIV*  NBRT*  INCL**
        01 HINT* CLEAN* EQU   LDLAB* JUMP*  PUB**S
        02 CLAB* NEG    NEQU  NUM*   UNEXT* EXT**S
        05 CREF* MUL    LTEQ  DREFB  CALL*  CMAP**
        06 DREF* DIV    GTEQ  NORM   CALR   GSYM**S
        08 STR*S UDIV   UGRTR SAVG*  SYS*   ISYM**S
        09 HDR   MOD    ULTEQ SAVL*  ILIB*S
        0A END   ADD    UGTEQ SAVI*  ICALL*
        0B MHDR  SUB    LDG*  STORE  ICALX*
        0D POP   BOR    LDL*  BRF*   INIT**
        0E DUP   BXOR   LDLV* BRT*   INCG**
        0F SWAP  BSHL   LDI*  NBRF*  INCI**

        * instruction has an argument.
        ** instruction has two arguments.
        S instruction has a string argument.
        In any of these cases, add 0x80 to the opcode.

        6.3.3 ASCII Table

        hex 00  10  20  30  40  50  60   70
             0  16  32  48  64  80  96  112 dec
        00  NUL DLE      0   @   P   '   p   00
        01  SOH DC1  !   1   A   Q   a   q   01
        02  STX DC2  "   2   B   R   b   r   02
        03  ETX DC3  #   3   C   S   c   s   03
        04  EOT DC4  $   4   D   T   d   t   04
        05  ENQ NAK  %   5   E   U   e   u   05
        06  ACK SYN  &   6   F   V   f   v   06
        07  BEL ETB  '   7   G   W   g   w   07
        08  BS  CAN  (   8   H   X   h   x   08
        09  HT  EM   )   9   I   Y   i   y   09
        0A  LF  SUB  *   :   J   Z   j   z   10
        0B  VT  ESC  +   ;   K   [   k   {   11
        0C  FF  FS   ,   <   L   \   l   |   12
        0D  CR  GS   -   =   M   ]   m   }   13
        0E  SO  RS   .   >   N   ^   n   ~   14
        0F  SI  US   /   ?   O   _   o  DEL  15

        7. License

        This manual is part of the T3X compiler package which is
        distributed under the following terms.

        T3X -- A Compiler for the Minimal Procedural Language T3X
        Nils M Holm, 1996-2019

        Redistribution and use in source and binary forms, with or
        without modification, are permitted provided that the following
        conditions are met:

        (1) Redistributions of source code must retain the above
            copyright notice, this list of conditions and the following

        (1) Redistributions in binary form must reproduce the above
            copyright notice, this list of conditions and the following
            disclaimer in the documentation and/or other materials
            provided with the distribution.

        SUCH DAMAGE.

contact  |  privacy