Find file
Fetching contributors…
Cannot retrieve contributors at this time
3138 lines (2396 sloc) 98.6 KB

cxxforth: A Simple Forth Implementation in C++

by Kristopher Johnson

This is free and unencumbered software released into the public domain.

Anyone is free to copy, modify, publish, use, compile, sell, or distribute this software, either in source code form or as a compiled binary, for any purpose, commercial or non-commercial, and by any means.

In jurisdictions that recognize copyright laws, the author or authors of this software dedicate any and all copyright interest in the software to the public domain. We make this dedication for the benefit of the public at large and to the detriment of our heirs and successors. We intend this dedication to be an overt act of relinquishment in perpetuity of all present and future rights to this software under copyright law.


cxxforth is a simple implementation of Forth in C++.

There are many examples of Forth implementations available on the Internet, but most of them are written in assembly language or low-level C, with a focus in maximizing efficiency and demonstrating traditional Forth implementation techniques. This Forth is different: My goal is to use modern C++ to create a Forth implementation that is easy to understand, easy to port, and easy to extend. I'm not going to talk about register assignments or addressing modes or opcodes or the trade-offs between indirect threaded code, direct threaded code, subroutine threaded code, and token threaded code. I'm just going to build a working Forth system in a couple thousand lines of C++ and Forth.

An inspiration for this implementation is Richard W.M. Jones's JONESFORTH. JONESFORTH is a Forth implementation written as a very readable tutorial, and I am adopting its style for our higher-level implementation. This Forth kernel is written as a C++ file with large comment blocks, and there is a utility, cpp2md, that takes that C++ file and converts it to a Markdown-format document with nicely formatted commentary sections between the C++ code blocks.

As in other Forth systems, the basic design of this Forth is to create a small kernel in native code (C++, in this case), and then implement the rest of the system with Forth code. The kernel has to provide the basic primitives needed for memory access, arithmetic, logical operations, and operating system access. With those primitives, we can then write Forth code to extend the system.

I am writing C++ conforming to the C++14 standard. If your C++ compiler does not support C++14 yet, you may need to make some modifications to the code to get it built.

The Forth words provided by cxxforth are based on those in the ANS Forth draft standard and Forth 2012 standard. I don't claim conformance to any standard, but you can use these standards as a crude form of documentation for the Forth words that are implemented here. cxxforth implements many of the words from the standards' core word sets, and a smattering of words from other standard word sets.

In addition to words from the standards, cxxforth provides a few non-standard words. Each of these is marked with "Not a standard word" in accompanying comments.

While this Forth can be seen as a toy implementation, I do want it to be usable for real-world applications. Forth was originally designed to be something simple you could build yourself and extend and customize as needed to solve your problem. I hope people can use cxxforth like that.

It is assumed that the reader has some familiarity with C++ and Forth. You may want to first read the JONESFORTH source or the source of some other Forth implementation to get the basic gist of how Forth is usually implemented.

Building cxxforth

Building the cxxforth executable and other targets is easiest if you are on a UNIX-ish system that has make, cmake, and Clang or GCC. If you have those components, you can probably build cxxforth by just entering these commands:

cd wherever_your_files_are/cxxforth

If successful, the cxxforth executable will be built in the wherever_your_files_are/cxxforth/build/ subdirectory.

If you don't have one of those components, or if 'make' doesn't work, then it's not too hard to build it manually. You will need to create a file called cxxforthconfig.h, which can be empty, then you need to invoke your C++ compiler on the cxxforth.cpp source file, enabling whatever options might be needed for C++14 compatibility and to link with the necessary C++ and system libraries. For example, on a Linux system with gcc, you should be able to build it by entering these commands:

cd wherever_your_files_are/cxxforth
touch cxxforthconfig.h
g++ -std=c++14 -o cxxforth cxxforth.cpp

Running cxxforth

Once the cxxforth executable is built, you can run it like any other command-line utility.

If you run it without any additional arguments, it will display a welcome message and then aloow you to enter Forth commands. Enter "bye" to exit.

If there are any additional arguments, cxxforth will load and interpret those files. For example, the cxxforth/tests directory contains a file hello.fs that defines a word hello. So, if you are in the cxxforth directory and you enter this:

build/cxxforth tests/hello.fs

Then cxxforth will load that file, and you can enter hello to execute the word that was loaded from hello.fs.

The Code

I start by including headers from the C++ Standard Library. I also include cxxforth.h, which declares exported functions and includes the cxxforthconfig.h file produced by the CMake build.

A macro CXXFORTH_DISABLE_FILE_ACCESS can be defined to prevent cxxforth from defining words for opening, reading, and writing files. You may want to do this on a platform that does not support file access, or if you don't need those words and want a smaller executable.

#include "cxxforth.h"

#include <algorithm>
#include <cctype>
#include <chrono>
#include <cstdlib>
#include <cstring>
#include <ctime>
#include <iomanip>
#include <iostream>
#include <list>
#include <stdexcept>
#include <string>
#include <thread>

#include <cstdio>
#include <fstream>

using std::cerr;
using std::cout;
using std::endl;
using std::exception;
using std::ptrdiff_t;
using std::runtime_error;
using std::size_t;
using std::string;

GNU Readline Support

cxxforth can use the GNU Readline library for user input if it is available.

The CMake build will automatically detect whether the library is available, and if so define CXXFORTH_USE_READLINE.

However, even if the GNU Readline library is available, you may not want to link your executable with it due to its GPL licensing terms. You can pass -DCXXFORTH_DISABLE_READLINE=ON to cmake to prevent it from searching for the library.

#include "readline/readline.h"
#include "readline/history.h"

Configuration Constants

I have a few macros to define the size of the Forth data space, the maximum numbers of cells on the data and return stacks, and the maximum number of word definitions in the dictionary.

These macros are usually defined in the cxxforthconfig.h header that is generated by CMake and included by cxxforth.h, but I provide default values in case they have not been defined.

#define CXXFORTH_VERSION "1.2.0"

#define CXXFORTH_DATASPACE_SIZE (16 * 1024 * sizeof(Cell))



Data Types

Next I define some types.

A Cell is the basic Forth type. I define the Cell type using the C++ uintptr_t type to ensure it is large enough to hold an address. This generally means that it will be a 32-bit value on 32-bit platforms and a 64-bit value on 64-bit platforms. (If you want to build a 32-bit Forth on a 64-bit platform with clang or gcc, you can pass the -m32 flag to the compiler and linker.)

I won't be providing any of the double-cell operations that traditional Forths provide. Double-cell operations were important in the days of 8-bit and 16-bit Forths, but with cells of 32 bits or more, many applications have no need for them.

I'm also not dealing with floating-point values. Floating-point support would be useful, but I'm trying to keep this simple.

Forth doesn't require type declarations; a cell can be used as an address, an unsigned integer, a signed integer, or a variety of other uses. However, in C++ we will have to be explicit about types to perform the operations we want to perform. So I define a few additional types to represent the ways that a Cell can be used, and a few macros to cast between types without littering the code with a lot of reinterpret_cast<...>(...) expressions.

namespace {

using Cell  = uintptr_t;      // unsigned value
using SCell = intptr_t;       // signed value

using Char  = unsigned char;
using SChar = signed char;

using CAddr = Char*;          // Any address
using AAddr = Cell*;          // Cell-aligned address

#define CELL(x)    reinterpret_cast<Cell>(x)
#define CADDR(x)   reinterpret_cast<Char*>(x)
#define AADDR(x)   reinterpret_cast<AAddr>(x)
#define CHARPTR(x) reinterpret_cast<char*>(x)
#define SIZE_T(x)  static_cast<size_t>(x)

constexpr auto CellSize = sizeof(Cell);

Boolean Constants

Here I define constants for Forth true and false Boolean flag values.

Note that the Forth convention is that a true flag is a cell with all bits set, unlike the C++ convention of using 1 or any other non-zero value to mean true, so we need to be sure to use these constants for all Forth words that return a Boolean flag.

constexpr Cell False = 0;
constexpr Cell True  = ~False;

The Definition Struct

My first big departure from traditional Forth implementations is how I will store the word definitions for the Forth dictionary. Traditional Forths intersperse the word names in the shared data space along with code and data, using a linked list to navigate through them. I am going to use a std::list of Definition structs, outside of the data space.

Use of std::list has these benefits:

  • The Definition structures won't use data space. The C++ library will take care of allocating heap space as needed.
  • The Definition structures won't move around in memory after being added to the list. (In contrast, use of std::vector or other C++ collections might move elements as they are modified.)

One of the members of Definition is a C++ std::string to hold the name. I won't need to worry about managing the memory for that variable-length field. The name field will be empty for a :NONAME definition.

A Definition also has a code field that points to the native code associated with the word, a does field pointing to associated Forth instructions, a parameter field that points to associated data-space elements, and some bit flags to keep track of whether the word is IMMEDIATE and/or HIDDEN. We will explore the use of these fields later when I talk about the inner and outer interpreters.

Definition has a static field executingWord that contains the address of the Definition that was most recently executed. This can be used by Code functions to refer to their definitions.

Finally, Definition has a few member functions for executing the code and for accessing the hidden and immediate flags.

using Code = void(*)();

struct Definition {
    Code   code      = nullptr;
    AAddr  does      = nullptr;
    AAddr  parameter = nullptr;
    Cell   flags     = 0;
    string name;

    static constexpr Cell FlagHidden    = (1 << 1);
    static constexpr Cell FlagImmediate = (1 << 2);

    static const Definition* executingWord;

    void execute() const {
        auto saved = executingWord;
        executingWord = this;
        executingWord = saved;

    bool isHidden() const    { return (flags & FlagHidden) != 0; }

    void toggleHidden()      { flags ^= FlagHidden; }

    bool isImmediate() const { return (flags & FlagImmediate) != 0; }

    void toggleImmediate()   { flags ^= FlagImmediate; }

    bool isFindable() const  { return !name.empty() && !isHidden(); }

I will use a pointer to a Definition as the Forth execution token (XT).

using Xt = Definition*;

#define XT(x) reinterpret_cast<Xt>(x)

Global Variables

With the types defined, next I define global variables, starting with the Forth data space and the data and return stacks.

For each of these arrays, there are constants that point to the end of the array, so I can easily test whether I need to report an overflow.


constexpr CAddr dataSpaceLimit = &dataSpace[CXXFORTH_DATASPACE_SIZE];
constexpr AAddr dStackLimit    = &dStack[CXXFORTH_DSTACK_COUNT];
constexpr AAddr rStackLimit    = &rStack[CXXFORTH_RSTACK_COUNT];

The Forth dictionary is a list of Definitions. The most recent definition is at the back of the list.

std::list<Definition> definitions;

For each of the global arrays, I need a pointer to the current location.

For the data space, I have the dataPointer, which corresponds to Forth's HERE.

For each of the stacks, I need a pointer to the element at the top of the stack. The stacks grow upward. When a stack is empty, the associated pointer points to an address below the actual bottom of the array, so I will need to avoid dereferencing these pointers under those circumstances.

CAddr dataPointer = nullptr;
AAddr dTop        = nullptr;
AAddr rTop        = nullptr;

The inner-definition interpreter needs a pointer to the next instruction to be executed. This will be explained below in the Inner Interpreter section.

Xt* nextInstruction = nullptr;

I have to define the static executingWord member declared in Definition.

const Definition* Definition::executingWord = nullptr;

There are a few special words whose XTs I will use frequently when compiling or executing. Rather than looking them up in the dictionary as needed, I'll cache their values during initialization.

Xt doLiteralXt       = nullptr;
Xt setDoesXt         = nullptr;
Xt exitXt            = nullptr;
Xt endOfDefinitionXt = nullptr;

I need a flag to track whether we are in interpreting or compiling state. This corresponds to Forth's STATE variable.

Cell isCompiling = False;

I provide a variable that controls the numeric base used for conversion between numbers and text. This corresponds to the Forth BASE variable.

Whenever using C++ stream output operators, I will need to ensure the stream's numeric output base matches numericBase. To make this easy, I'll define a macro SETBASE() that calls the std::setbase I/O manipulator and use it whenever writing numeric data using the stream operators.

Cell numericBase = 10;

#define SETBASE() std::setbase(static_cast<int>(numericBase))

The input buffer is a std::string. This makes it easy to use the C++ I/O facilities, and frees me from the need to allocate a statically sized buffer that could overflow. I also have a current offset into this buffer, corresponding to the Forth >IN variable.

string sourceBuffer;
Cell sourceOffset = 0;

I need a buffer to store the result of the Forth WORD word. As with the input buffer, I use a string so I don't need to worry about memory management.

Note that while this is a std:string, its format is not a typical strings. The buffer returned by WORD has the word length as its first character. That is, it is a Forth counted string.

string wordBuffer;

I need a buffer to store the result of the Forth PARSE word. Unlike WORD, this is a "normal" string and I won't need to store the count at the beginning of this buffer.

string parseBuffer;

I store the argc and argv values passed to main() so I can make them available for use by the Forth program by our non-standard #ARGS and ARG Forth words, defined later.

size_t commandLineArgCount = 0;
const char** commandLineArgVector = nullptr;

I need a variable to store the result of the last call to SYSTEM, which the user can retrieve by using $?.

int systemResult = 0;

Stack Primitives

I will be doing a lot of pushing and popping values to and from our data and return stacks, so in lieu of sprinkling pointer arithmetic throughout our code, I'll define a few simple functions to handle those operations. I expect the compiler to expand calls to these functions inline, so this isn't inefficient.

Above I defined the global variables dTop and rTop to point to the top of the data stack and return stack. I will use the expressions *dTop and *rTop when accessing the top-of-stack values. I will also use expressions like *(dTop - 1) and *(dTop - 2) to reference the items beneath the top of stack.

When I need to both read and remove a top-of-stack value, my convention will be to put both operations on the same line, like this:

Cell x = *dTop; pop();

A more idiomatic C++ way to write this might be Cell x = *(dTop--);, but I think that's less clear.

// Make the data stack empty.
void resetDStack() {
    dTop = dStack - 1;

// Make the return stack empty.
void resetRStack() {
    rTop = rStack - 1;

// Return the depth of the data stack.
ptrdiff_t dStackDepth() {
    return dTop - dStack + 1;

// Return the depth of the return stack.
ptrdiff_t rStackDepth() {
    return rTop - rStack + 1;

// Push cell onto data stack.
void push(Cell x) {
    *(++dTop) = x;

// Pop cell from data stack.
void pop() {

// Push cell onto return stack.
void rpush(Cell x) {
    *(++rTop) = x;

// Pop cell from return stack.
void rpop() {


Forth provides the ABORT and ABORT" words, which interrupt execution and return control to the main QUIT loop. I will implement this functionality using a C++ exception to return control to the top-level interpreter.

The C++ functions abort() and abortMessage() defined here are the first primitive functions that will be exposed as Forth words. For each such word, I will spell out the Forth name of the primitive in all-caps, and provide a Forth comment showing the stack effects. For words described in the standards, I will generally not provide any more information, but for words that are not standard words, I'll provide a brief description.

class AbortException: public runtime_error {
    AbortException(const string& msg): runtime_error(msg) {}
    AbortException(const char* msg): runtime_error(msg) {}

// ABORT ( i*x -- ) ( R: j*x -- )
void abort() {
    throw AbortException("");

// ABORT-MESSAGE ( i*x c-addr u -- ) ( R: j*x -- )
// Not a standard word.
// Same semantics as the standard ABORT", but takes a string address and length
// instead of parsing the message string.
void abortMessage() {
    auto count = SIZE_T(*dTop); pop();
    auto caddr = CHARPTR(*dTop); pop();
    string message(caddr, count);
    throw AbortException(message);

Runtime Safety Checks

Old-school Forths are implemented by super-programmers who never make coding mistakes and so don't want the overhead of bounds-checking or other nanny hand-holding. However, I'm just a dumb C++ programmer, and I'd like some help to catch mistakes.

To that end, I have a set of macros and functions that verify that I have the expected number of arguments available on the stacks, that I'm not going to run off the end of an array, that I'm not going to try to divide by zero, and so on.

You can define the macro CXXFORTH_SKIP_RUNTIME_CHECKS to generate an executable that doesn't include these checks, so when you have a fully debugged Forth application you can run it on that optimized executable for improved performance.

When the CXXFORTH_SKIP_RUNTIME_CHECKS macro is not defined, these macros will check conditions and throw an AbortException if the assertions fail. I won't go into the details of these macros here. Later we will see them used in the definitions of our primitive Forth words.


#define RUNTIME_NO_OP()                      do { } while (0)
#define RUNTIME_ERROR(msg)                   RUNTIME_NO_OP()
#define RUNTIME_ERROR_IF(cond, msg)          RUNTIME_NO_OP()
#define REQUIRE_DSTACK_DEPTH(n, name)        RUNTIME_NO_OP()
#define REQUIRE_RSTACK_DEPTH(n, name)        RUNTIME_NO_OP()
#define REQUIRE_ALIGNED(addr, name)          RUNTIME_NO_OP()
#define REQUIRE_VALID_HERE(name)             RUNTIME_NO_OP()


#define RUNTIME_ERROR(msg)                   do { throw AbortException(msg); } while (0)
#define RUNTIME_ERROR_IF(cond, msg)          do { if (cond) RUNTIME_ERROR(msg); } while (0)
#define REQUIRE_DSTACK_DEPTH(n, name)        requireDStackDepth(n, name)
#define REQUIRE_DSTACK_AVAILABLE(n, name)    requireDStackAvailable(n, name)
#define REQUIRE_RSTACK_DEPTH(n, name)        requireRStackDepth(n, name)
#define REQUIRE_RSTACK_AVAILABLE(n, name)    requireRStackAvailable(n, name)
#define REQUIRE_ALIGNED(addr, name)          checkAligned(addr, name)
#define REQUIRE_VALID_HERE(name)             checkValidHere(name)
#define REQUIRE_DATASPACE_AVAILABLE(n, name) requireDataSpaceAvailable(n, name)

template<typename T>
void checkAligned(T addr, const char* name) {
    RUNTIME_ERROR_IF((CELL(addr) % CellSize) != 0,
                     string(name) + ": unaligned address");

void requireDStackDepth(size_t n, const char* name) {
    RUNTIME_ERROR_IF(dStackDepth() < static_cast<ptrdiff_t>(n),
                     string(name) + ": stack underflow");

void requireDStackAvailable(size_t n, const char* name) {
    RUNTIME_ERROR_IF((dTop + n) >= dStackLimit,
                     string(name) + ": stack overflow");

void requireRStackDepth(size_t n, const char* name) {
    RUNTIME_ERROR_IF(rStackDepth() < ptrdiff_t(n),
                     string(name) + ": return stack underflow");

void requireRStackAvailable(size_t n, const char* name) {
    RUNTIME_ERROR_IF((rTop + n) >= rStackLimit,
                     string(name) + ": return stack overflow");

void checkValidHere(const char* name) {
    RUNTIME_ERROR_IF(dataPointer < dataSpace || dataSpaceLimit <= dataPointer,
                     string(name) + ": HERE outside data space");

void requireDataSpaceAvailable(size_t n, const char* name) {
    RUNTIME_ERROR_IF((dataPointer + n) >= dataSpaceLimit,
                     string(name) + ": data space overflow");


Forth Primitives

Now I will start defining the primitive operations that are exposed as Forth words. You can think of these as the opcodes of a virtual Forth processor. Once I have the primitive operations defined, I can then write definitions in Forth that use these primitives to build more-complex words.

Each of these primitives is a function that takes no arguments and returns no result, other than its effects on the Forth data stack, return stack, and data space. Such a function can be assigned to the code field of a Definition.

When changing the stack, the primitives don't change the stack depth any more than necessary. For example, PICK just replaces the top-of-stack value with a different value, and ROLL uses std::memmove() to rearrange elements rather than individually popping and pushng them.

You can peek ahead to the definePrimitives() function to see how these primitives are added to the Forth dictionary.

Forth Stack Operations

Let's start with some basic Forth stack manipulation words. These differ from the push/pop/rpush/rpop/etc. primitives above in that they are intended to be called from Forth code rather than from the C++ kernel code. So I include runtime checks and use the stacks rather than passing arguments or returning values via C++ call/return mechanisms.

Note that for C++ functions that implement primitive Forth words, I will include the Forth names and stack effects in comments. You can look up the Forth names in the standards to learn what these words are supposed to do.

// DEPTH ( -- +n )
void depth() {

// DROP ( x -- )
void drop() {

// PICK ( xu ... x1 x0 u -- xu ... x1 x0 xu )
void pick() {
    auto index = *dTop;
    REQUIRE_DSTACK_DEPTH(index + 2, "PICK");
    *dTop = *(dTop - index - 1);

// ROLL ( xu xu-1 ... x0 u -- xu-1 ... x0 xu )
void roll() {
    auto n = *dTop; pop();
    if (n > 0) {
        REQUIRE_DSTACK_DEPTH(n + 1, "ROLL");
        auto x = *(dTop - n);
        std::memmove(dTop - n, dTop - n + 1, n * sizeof(Cell));
        *dTop = x;

// >R ( x -- ) ( R:  -- x )
void toR() {
    rpush(*dTop); pop();

// R> ( -- x ) ( R: x -- )
void rFrom() {
    push(*rTop); rpop();

// R@ ( -- x ) ( R: x -- x )
void rFetch() {

// EXIT ( -- ) ( R: nest-sys -- )
void exit() {
    nextInstruction = reinterpret_cast<Xt*>(*rTop); rpop();

Another important set of Forth primitives are those for reading and writing values in data space.

// ! ( x a-addr -- )
void store() {
    auto aaddr = AADDR(*dTop); pop();
    REQUIRE_ALIGNED(aaddr, "!");
    auto x = *dTop; pop();
    *aaddr = x;

// @ ( a-addr -- x )
void fetch() {
    auto aaddr = AADDR(*dTop);
    REQUIRE_ALIGNED(aaddr, "@");
    *dTop = *aaddr;

// c! ( char c-addr -- )
void cstore() {
    auto caddr = CADDR(*dTop); pop();
    auto x = static_cast<Char>(*dTop); pop();
    *caddr = x;

// c@ ( c-addr -- char )
void cfetch() {
    auto caddr = CADDR(*dTop);
    *dTop = static_cast<Cell>(*caddr);

// COUNT ( c-addr1 -- c-addr2 u )
void count() {
    auto caddr = CADDR(*dTop);
    auto count = static_cast<Cell>(*caddr);
    *dTop = CELL(caddr + 1);

Next, I'll define some primitives for accessing and manipulating the data-space pointer, HERE.

template<typename T>
AAddr alignAddress(T addr) {
    auto offset = CELL(addr) % CellSize;
    return (offset == 0) ? AADDR(addr) : AADDR(CADDR(addr) + (CellSize - offset));

void alignDataPointer() {
    dataPointer = CADDR(alignAddress(dataPointer));

// ALIGN ( -- )
void align() {

// ALIGNED ( addr -- a-addr )
void aligned() {
    *dTop = CELL(alignAddress(*dTop));

// HERE ( -- addr )
void here() {

// ALLOT ( n -- )
void allot() {
    dataPointer += *dTop; pop();

// CELLS ( n1 -- n2 )
void cells() {
    *dTop *= CellSize;

// Store a cell to data space.
void data(Cell x) {
    REQUIRE_ALIGNED(dataPointer, ",");
    *(AADDR(dataPointer)) = x;
    dataPointer += CellSize;

// UNUSED ( -- u )
void unused() {
    push(static_cast<Cell>(dataSpaceLimit - dataPointer));

I could implement memory-block words like CMOVE, CMOVE>, FILL, and COMPARE in Forth, but speed is often important for these, so I will make them native primitives.

// CMOVE ( c-addr1 c-addr2 u -- )
void cMove() {
    auto length = SIZE_T(*dTop); pop();
    auto dest = CHARPTR(*dTop); pop();
    auto src = CHARPTR(*dTop); pop();
    std::memcpy(dest, src, length);

// CMOVE> ( c-addr1 c-addr2 u -- )
void cMoveUp() {
    auto length = SIZE_T(*dTop); pop();
    auto dst = CHARPTR(*dTop); pop();
    auto src = CHARPTR(*dTop); pop();
    for (size_t i = 0; i < length; ++i) {
        auto offset = length - i - 1;
        *(src + offset) = *(dst + offset);

// FILL ( c-addr u char -- )
void fill() {
    auto ch = static_cast<Char>(*dTop); pop();
    auto length = SIZE_T(*dTop); pop();
    auto caddr = CHARPTR(*dTop); pop();
    std::fill(caddr, caddr + length, ch);

// COMPARE ( c-addr1 u1 c-addr2 u2 -- n )
void compare() {
    auto len2 = SIZE_T(*dTop); pop();
    auto caddr2 = CHARPTR(*dTop); pop();
    auto len1 = SIZE_T(*dTop); pop();
    auto caddr1 = CHARPTR(*dTop);

    auto minLen = std::min(len1, len2);
    auto cmpResult = std::memcmp(caddr1, caddr2, minLen);

    if (cmpResult < 0) {
        *dTop = static_cast<Cell>(-1);
    else if (cmpResult > 0) {
        *dTop = static_cast<Cell>(1);
    else if (len1 < len2) {
        *dTop = static_cast<Cell>(-1);
    else if (len1 > len2) {
        *dTop = static_cast<Cell>(1);
    else {
        *dTop = 0;

Next I will define I/O primitives.

I keep things simple and portable by using C++ iostream objects.

// KEY ( -- char )
void key() {
    auto ch = static_cast<Cell>(std::cin.get());

// EMIT ( x -- )
void emit() {
    auto cell = *dTop; pop();

// TYPE ( c-addr u -- )
void type() {
    auto length = static_cast<std::streamsize>(*dTop); pop();
    auto caddr = CHARPTR(*dTop); pop();
    cout.write(caddr, length);

// CR ( -- )
void cr() {
    cout << endl;

// . ( n -- )
void dot() {
    cout << SETBASE() << static_cast<SCell>(*dTop);

// U. ( x -- )
void uDot() {
    cout << SETBASE() << *dTop;

// .R ( n1 n2 -- )
void dotR() {
    auto width = static_cast<int>(*dTop); pop();
    auto n = static_cast<SCell>(*dTop); pop();
    cout << SETBASE() << std::setw(width) << n;

// BASE ( -- a-addr )
void base() {

// SOURCE ( -- c-addr u )
void source() {

// >IN ( -- a-addr )
void toIn() {

REFILL reads a line from the user input device. If successful, it puts the result into sourceBuffer, sets sourceOffset to 0, and pushes a TRUE flag onto the stack. If not successful, it pushes a FALSE flag.

This uses GNU Readline if configured to do so. Otherwise it uses the C++ std::getline() function.

// REFILL ( -- flag )
void refill() {

    char* line = readline("");
    if (line) {
        sourceBuffer = line;
        sourceOffset = 0;
        if (*line)
    else {
    if (std::getline(std::cin, sourceBuffer)) {
        sourceOffset = 0;
    else {

ACCEPT is similar to REFILL, but puts the result into a caller-supplied buffer.

// ACCEPT ( c-addr +n1 -- +n2 )
void accept() {

    auto bufferSize = SIZE_T(*dTop); pop();
    auto buffer = CHARPTR(*dTop);

    char* line = readline("");
    if (line) {
        auto lineSize = std::strlen(line);
        auto copySize = std::min(lineSize, bufferSize);
        std::memcpy(buffer, line, copySize);
        *dTop = static_cast<Cell>(copySize);
        if (*line)
    else {
        *dTop = 0;
    string line;
    if (std::getline(std::cin, line)) {
        auto copySize = std::min(line.length(), bufferSize);
        std::memcpy(buffer,, copySize);
        *dTop = static_cast<Cell>(copySize);
    else {
        *dTop = 0;

The text interpreter and other Forth words use WORD to parse a blank-delimited sequence of characters from the input. WORD skips any delimiters at the current input position, then reads characters until it finds the delimiter again. It returns the address of a buffer with the length in the first byte, followed by the characters that made up the word.

In a few places later in the C++ code, you will see the call sequence bl(); word(); count();. This corresponds to the Forth phrase BL WORD COUNT, which is how Forth code typically reads a space-delimited word from the input and get its address and length.

The standards specify that the WORD buffer must contain a space character after the character data, but I'm not going to worry about this obsolescent requirement.

// WORD ( char "<chars>ccc<char>" -- c-addr )
void word() {
    auto delim = static_cast<char>(*dTop);

    wordBuffer.push_back(0);  // First char of buffer is length.

    auto inputSize = sourceBuffer.size();

    // Skip leading delimiters
    while (sourceOffset < inputSize && sourceBuffer[sourceOffset] == delim)

    // Copy characters until we see the delimiter again.
    while (sourceOffset < inputSize && sourceBuffer[sourceOffset] != delim) {

    if (sourceOffset < inputSize) {

    // Update the count at the beginning of the buffer.
    wordBuffer[0] = static_cast<char>(wordBuffer.size() - 1);

    *dTop = CELL(;

PARSE is similar to WORD, but does not skip leading delimiters and provides an address-length result.

// PARSE ( char "ccc<char>" -- c-addr u )
void parse() {

    auto delim = static_cast<char>(*dTop);


    auto inputSize = sourceBuffer.size();

    // Copy characters until we see the delimiter.
    while (sourceOffset < inputSize && sourceBuffer[sourceOffset] != delim) {

    if (sourceOffset == inputSize)
        throw AbortException(string("PARSE: Did not find expected delimiter \'" + string(1, delim) + "\'"));

    // Skip over the delimiter

    *dTop = CELL(;

BL puts a space character on the stack. It is often seen in the phrase BL WORD to parse a space-delimited word, and will be seen later in the Forth definition of SPACE.

// BL ( -- char )
void bl() {
    push(' ');

Next I define arithmetic primitives.

Note that I need to use the SCell type for signed operations, and the Cell type for unsigned operations.

// + ( n1 n2 -- n3 )
void plus() {
    auto n2 = static_cast<SCell>(*dTop); pop();
    auto n1 = static_cast<SCell>(*dTop);
    *dTop = static_cast<Cell>(n1 + n2);

// - ( n1 n2 -- n3 )
void minus() {
    auto n2 = static_cast<SCell>(*dTop); pop();
    auto n1 = static_cast<SCell>(*dTop);
    *dTop = static_cast<Cell>(n1 - n2);

// * ( n1 n2 -- n3 )
void star() {
    auto n2 = static_cast<SCell>(*dTop); pop();
    auto n1 = static_cast<SCell>(*dTop);
    *dTop = static_cast<Cell>(n1 * n2);

// / ( n1 n2 -- n3 )
void slash() {
    auto n2 = static_cast<SCell>(*dTop); pop();
    auto n1 = static_cast<SCell>(*dTop);
    RUNTIME_ERROR_IF(n2 == 0, "/: zero divisor");
    *dTop = static_cast<Cell>(n1 / n2);

// /MOD ( n1 n2 -- n3 n4 )
void slashMod() {
    auto n2 = static_cast<SCell>(*dTop);
    auto n1 = static_cast<SCell>(*(dTop - 1));
    RUNTIME_ERROR_IF(n2 == 0, "/MOD: zero divisor");
    auto result = std::ldiv(n1, n2);
    *(dTop - 1) = static_cast<Cell>(result.rem);
    *dTop = static_cast<Cell>(result.quot);

Next, I define logical and relational primitives.

// AND ( x1 x2 -- x3 )
void bitwiseAnd() {
    auto x2 = *dTop; pop();
    *dTop = *dTop & x2;

// OR ( x1 x2 -- x3 )
void bitwiseOr() {
    auto x2 = *dTop; pop();
    *dTop = *dTop | x2;

// XOR ( x1 x2 -- x3 )
void bitwiseXor() {
    auto x2 = *dTop; pop();
    *dTop = *dTop ^ x2;

// LSHIFT ( x1 u -- x2 )
void lshift() {
    auto n = *dTop; pop();
    *dTop <<= n;

// RSHIFT ( x1 u -- x2 )
void rshift() {
    auto n = *dTop; pop();
    *dTop >>= n;

// = ( x1 x2 -- flag )
void equals() {
    auto n2 = *dTop; pop();
    *dTop = *dTop == n2 ? True : False;

// < ( n1 n2 -- flag )
void lessThan() {
    auto n2 = static_cast<SCell>(*dTop); pop();
    *dTop = static_cast<SCell>(*dTop) < n2 ? True : False;

// U< ( u1 u2 -- flag )
void uLessThan() {
    auto u2 = static_cast<Cell>(*dTop); pop();
    *dTop = static_cast<Cell>(*dTop) < u2 ? True : False;

Now I will define a few primitives that give access to operating-system and environmental data.

// #ARG ( -- n )
// Not a standard word.
// Provide count of command-line arguments.
void argCount() {

// ARG ( n -- c-addr u )
// Not a standard word.
// Provide the Nth command-line argument.
void argAtIndex() {
    auto index = static_cast<size_t>(*dTop);
    RUNTIME_ERROR_IF(index >= commandLineArgCount, "ARG: invalid index");
    auto value = commandLineArgVector[index];
    *dTop = CELL(value);

// BYE ( -- )
void bye() {

// SYSTEM ( c-addr u -- )
// Not a standard word.
// Executes a system command in a subshell.
// See the documentation for the C++ `std::system()` library call for more
// details.
void system() {
    auto length = SIZE_T(*dTop); pop();
    auto caddr = CHARPTR(*dTop); pop();
    string command(caddr, length);
    systemResult = std::system(command.c_str());

// $? ( -- n )
// Not a standard word.
// Returns an implementation-defined status code from the last call to
// `SYSTEM`.  See the documentation for the C++ `std::system()` library call
// for details.
void lastSystemResult() {

// MS ( u -- )
void ms() {
    auto period = *dTop; pop();

// TIME&DATE ( -- +n1 +n2 +n3 +n4 +n5 +n6 )
void timeAndDate () {
    auto t = std::time(0);
    auto tm = std::localtime(&t);
    push(static_cast<Cell>(tm->tm_mon + 1));
    push(static_cast<Cell>(tm->tm_year + 1900));

// UTCTIME&DATE ( -- +n1 +n2 +n3 +n4 +n5 +n6 )
// Not a standard word.
// Like TIME&DATE, but returns UTC rather than local time.
void utcTimeAndDate () {
    auto t = std::time(0);
    auto tm = std::gmtime(&t);
    push(static_cast<Cell>(tm->tm_mon + 1));
    push(static_cast<Cell>(tm->tm_year + 1900));

// .S ( -- )
void dotS() {
    auto depth = dStackDepth();
    cout << SETBASE() << "<" << depth << "> ";
    for (auto i = depth; i > 0; --i) {
        cout << static_cast<SCell>(*(dTop - i + 1)) << " ";

// .RS ( -- )
// Not a standard word.
// Like .S, but prints the contents of the return stack instead of the data
// stack.
void dotRS() {
    auto depth = rStackDepth();
    cout << SETBASE() << "<<" << depth << ">> ";
    for (auto i = depth; i > 0; --i) {
        cout << static_cast<SCell>(*(rTop - i + 1)) << " ";

Inner Interpreter

A Forth system is said to have two interpreters: an "outer interpreter" which reads input and interprets it, and an "inner interpreter" which executes compiled Forth definitions. I will start with the inner interpreter.

There are basically two kinds of words in a Forth system:

  • primitive code: native subroutines that are executed directly by the CPU
  • colon definition: a sequence of Forth words compiled by : (colon), :NONAME, or DOES>.

Every defined word has a code field that points to native code. In the case of primitive words, the code field points to a routine that performs the operation. In the case of a colon definition, the code field points to the doColon() function, which saves the current program state and then starts executing the words that make up the colon definition.

Each colon definition ends with a call to EXIT, which sets up a return to the colon definition that called the current word. In many traditional Forths, the EXIT instruction is implemented as a jump/branch to the next machine-code instruction to be executed. But that's not easy to do in a portable way in C++, so my doColon() just keeps going until it sees an EXIT instruction, then returns to the caller without actually executing it.

In many Forth implementations, the return stack is used to store the address of the next instruction to be invoked upon returning from the routine. But in C++, I can just use a temporary variable to achieve the same thing. So in this Forth, the return stack is really just a secondary stack; it doesn't have anything to do with "returning".

void doColon() {
    auto savedNext = nextInstruction;

    auto defn = Definition::executingWord;

    nextInstruction = reinterpret_cast<Xt*>(defn->does);
    while (*nextInstruction != exitXt) {

    nextInstruction = savedNext;


Now that we know how the inner interpreter works, I can define the words that compile definitions to be executed by that interpreter.

The kernel provides three words that can add a word to the dictionary: CREATE, :, and :NONAME. Each of them constructs a Definition, fills in its name, code, parameter, and does field appropriately, and then adds it to the definitions list.

: and :NONAME then put the interpreter into compilation mode. While in compilation mode, the interpreter will add the XT's of non-immediate words to the current definition. This continues until the ; word ends the definition.

The word DOES> can be used after CREATE to define execution semantics for that word. As with :, this puts the interpreter into compilation state until : is encountered.

// Return reference to the latest Definition.
// Undefined behavior if the definitions list is empty.
Definition& lastDefinition() {
    return definitions.back();

// LATEST ( -- xt )
// Not a standard word.
// Puts the execution token of the most recently CREATEd word on the stack.
void latest() {

// STATE ( -- a-addr )
void state() {

void doCreate() {
    auto defn = Definition::executingWord;
    REQUIRE_DSTACK_AVAILABLE(1, defn->name.c_str());

// CREATE ( "<spaces>name" -- )  Execution: ( -- a-addr )
void create() {

    bl(); word(); count();
    auto length = SIZE_T(*dTop); pop();
    auto caddr = CHARPTR(*dTop); pop();

    RUNTIME_ERROR_IF(length < 1, "CREATE: could not parse name");

    Definition defn;
    defn.code = doCreate;
    defn.parameter = defn.does = AADDR(dataPointer); = string(caddr, length);

// : ( C: "<spaces>name" -- colon-sys )
void colon() {
    isCompiling = true;

    auto& latest = lastDefinition();
    latest.code = doColon;

// :NONAME ( C:  -- colon-sys )  ( S:  -- xt )
void noname() {

    Definition defn;
    defn.code = doColon;
    defn.parameter = defn.does = AADDR(dataPointer);

    isCompiling = true;

void doDoes() {

void setDoes() {
    auto& latest = lastDefinition();
    latest.code = doDoes;
    latest.does = AADDR(nextInstruction) + 1;

// DOES>
void does() {

// (;) ( -- )
// Not a standard word.
// This word is compiled by ; after the EXIT.  It is never executed, but serves
// as a marker for use in debugging.
void endOfDefinition() {
    throw AbortException("(;) should never be executed");

// ; ( C: colon-sys -- )
void semicolon() {
    isCompiling = false;

// IMMEDIATE ( -- )
// Unlike the standard specification, my IMMEDIATE toggles the immediacy bit
// of the most recent definition, rather than always setting it true.
void immediate() { lastDefinition().toggleImmediate(); }

// HIDDEN ( -- )
// Not a standard word.
// Toggles the hidden bit of the most recent definition.
void hidden() { lastDefinition().toggleHidden(); }

Next I'll define a few "special words". They are special in that they are used to implement features of the inner interpreter, and are not generally used by Forth application code. As a signifier of their special nature, the words' names start and end with with parentheses.

// (lit) ( -- x )
// Not a standard word.
// This instruction gets the value of the next cell, puts that on the data
// stack, and then moves the instruction pointer to the next instruction.  It
// is used by `LITERAL` and other Forth words that need to specify a cell value
// to put on the stack during execution.
void doLiteral() {

// (branch) ( -- )
// Not a standard word.
// Used by branching/looping constructs.  Unconditionally adds an offset to
// `next`.  The offset is in the cell following the instruction.
// The offset is in character units, but must be a multiple of the cell size.
void branch() {
    auto offset = reinterpret_cast<SCell>(*nextInstruction);
    nextInstruction += offset / static_cast<SCell>(CellSize);

// (zbranch) ( flag -- )
// Not a standard word.
// Used by branching/looping constructinos.  Adds an offset to `next` if the
// top-of-stack value is zero.  The offset is in the cell following the
// instruction.  If top-of-stack is not zero, then continue to the next
// instruction.
void zbranch() {
    REQUIRE_DSTACK_DEPTH(1, "(zbranch)");
    auto flag = *dTop; pop();
    if (flag == False)


The next section contains words that create elements in the definitions list, look up elements by name, or traverse the list to perform some operation.

// Create a new definition with specified name and code.
void definePrimitive(const char* name, Code code) {

    Definition defn;
    defn.code = code;
    defn.parameter = defn.does = AADDR(dataPointer); = name;

// Determine whether two names are equivalent, using case-insensitive matching.
bool doNamesMatch(CAddr name1, CAddr name2, Cell nameLength) {
    for (size_t i = 0; i < nameLength; ++i) {
        if (std::toupper(name1[i]) != std::toupper(name2[i])) {
            return false;
    return true;

// Find a definition by name.
Xt findDefinition(CAddr nameToFind, Cell nameLength) {
    if (nameLength == 0)
        return nullptr;

    for (auto i = definitions.rbegin(); i != definitions.rend(); ++i) {
        auto& defn = *i;
        if (!defn.isFindable())
        auto& name =;
        if (name.length() == nameLength) {
            auto nameCAddr = CADDR(const_cast<char*>(;
            if (doNamesMatch(nameToFind, nameCAddr, nameLength)) {
                return &defn;
    return nullptr;

// Find a definition by name.
Xt findDefinition(const string& name) {
    return findDefinition(CADDR(const_cast<char*>(, static_cast<Cell>(name.length()));

// FIND ( c-addr -- c-addr 0  |  xt 1  |  xt -1 )
void find() {
    auto caddr = CADDR(*dTop);
    auto length = static_cast<Cell>(*caddr);
    auto name = caddr + 1;
    auto word = findDefinition(name, length);
    if (word == nullptr) {
    else {
        *dTop = CELL(word);
        push(word->isImmediate() ? 1 : Cell(-1));

// EXECUTE ( i*x xt -- j*x )
void execute() {
    auto defn = XT(*dTop); pop();

// >BODY ( xt -- a-addr )
void toBody() {
    auto xt = XT(*dTop);
    *dTop = CELL(xt->parameter);

// XT>NAME ( xt -- c-addr u )
// Not a standard word.
// Gives the name associated with an XT.
void xtToName() {
    auto xt = XT(*dTop);
    auto& name = xt->name;
    *dTop = CELL(;

// WORDS ( -- )
void words() {
    std::for_each(definitions.rbegin(), definitions.rend(), [](auto& defn) {
        if (defn.isFindable()) cout << << " ";


Most Forth systems provide a word SEE that will print out the definition of a word.

My implementation of this word walks through the contents of a definition and tries to "decompile" each cell. If the cell contains the XT of a defined word, then it prints that word's name. Otherwise, it just prints the cell value.

This generally gives a readable view of the word's definition, but it is not exactly equal to the original source text. For example, for this definition:

: add-1-and-2 ( -- )   1 2 + . ;

SEE add-1-and-2 gives this output:

: add-1-and-2 (lit) 1 (lit) 2 + . EXIT ;

It gets even messier when decompiling words that contain branches and string literals, but it works well as a debugging tool when trying to determine why a word is not compiling as expected.

Note that the kernel doesn't know about IF...THEN, or BEGIN...WHILE...THEN, or CONSTANT, or VARIABLE, or various other constructs that would be needed to provide a more accurate representation of the original definition. A better definition of SEE would need to be written in Forth, after those constructs have been defined. But I haven't done that yet.

// Given a cell that might be an XT, search for it in the definitions list.
// Returns a pointer to the definition if found, or nullptr if not.
Xt findXt(Cell x) {
    for (auto i = definitions.rbegin(); i != definitions.rend(); ++i) {
        auto& defn = *i;
        if (&defn == reinterpret_cast<Xt>(x))
            return &defn;
    return nullptr;

/// Display the words that make up a colon or DOES> definition.
void seeDoes(AAddr does) {
    while (XT(*does) != endOfDefinitionXt) {
        auto xt = findXt(*does);
        if (xt)
            cout << " " << xt->name;
            cout << " " << SETBASE() << static_cast<SCell>(*does);
    cout << " ;";

// SEE ( "<spaces>name" -- )
void see() {
    bl(); word(); find();

    auto found = *dTop; pop();
    if (!found) throw AbortException("SEE: undefined word");

    auto defn = XT(*dTop); pop();
    if (defn->code == doColon) {
        cout << ": " << defn->name;
    else if (defn->code == doCreate || defn->code == doDoes) {
        cout << "CREATE " << defn->name << " ( " << CELL(defn->parameter) << " )";
        if (defn->code == doDoes) {
            cout << " DOES>";
    else {
        cout << ": " << defn->name << " <primitive " << SETBASE() << CELL(defn->code) << "> ;";
    if (defn->isImmediate()) cout << " immediate";

Outer Interpreter

The word INTERPRET below implements the outer interpreter. It looks at the sourceBuffer, and repeats the following until it has processed all the characters in the buffer:

  • Parse a space-delimited word.
  • Look up that word in the dictionary.
  • If the word is found:
    • If not in compilation mode, or if the word is an immediate word, then execute it.
    • Otherwise (in compilation mode), compile a call to the word.
  • If the word is not found:
    • Try to parse it as a number.
    • If it is a number:
      • If in compilation mode, then compile it as a literal.
      • Otherwise, put the value on the stack.
    • If it is not a number, then signal an error.

See section 3.4 of the ANS Forth draft standard for a more complete description of the Forth text interpreter.

// Determine whether specified character is a valid numeric digit for current BASE.
bool isValidDigit(Char c) {
    if (numericBase > 10) {
        if (('A' <= c) && (c < ('A' + numericBase - 10)))
            return true;
        if (('a' <= c) && (c < ('a' + numericBase - 10)))
            return true;
    return ('0' <= c) && (c < ('0' + numericBase));

// Return numeric value associated with a character.
Cell digitValue(Char c) {
    if (c >= 'a')
        return c - 'a' + 10;
    else if (c >= 'A')
        return c - 'A' + 10;
        return c - '0';

// >UNUM ( u0 c-addr1 u1 -- u c-addr2 u2 )
// Not a standard word.
// This word is similar to the standard >NUMBER, but provides a single-cell
// result.
void parseUnsignedNumber() {

    auto length = SIZE_T(*dTop);
    auto caddr = CADDR(*(dTop - 1));
    auto value = *(dTop - 2);

    auto i = size_t(0);
    while (i < length) {
        auto c = caddr[i];
        if (isValidDigit(c)) {
            auto n = digitValue(c);
            value = value * numericBase + n;
        else {

    *(dTop - 2) = value;
    *(dTop - 1) = CELL(caddr + i);
    *dTop = length - i;

// >NUM ( n c-addr1 u1 -- n c-addr2 u2 )
// Not a standard word.
// Similar to >UNUM, but looks for a '-' character at the beginning, and
// negates the result if found.
void parseSignedNumber() {

    auto length = SIZE_T(*dTop);
    auto caddr = CHARPTR(*(dTop - 1));

    if (length > 1 && *caddr == '-') {
        *dTop = static_cast<Cell>(length - 1);
        *(dTop - 1) = CELL(caddr + 1);
        *(dTop - 2) = static_cast<Cell>(-static_cast<SCell>(*(dTop - 2)));
    else {

// INTERPRET ( i*x -- j*x )
// Not a standard word.
// Reads words from the input buffer and executes/compiles them.
void interpret() {
    auto inputSize = sourceBuffer.size();
    while (sourceOffset < inputSize) {
        bl(); word(); find();

        auto found = static_cast<int>(*dTop); pop();

        if (found) {
            auto xt = XT(*dTop); pop();
            if (isCompiling && !xt->isImmediate()) {
            else {
        else {
            // find() left the counted string on the stack.
            // Try to parse it as a number.

            auto length = SIZE_T(*dTop); pop();
            auto caddr = CHARPTR(*dTop); pop();

            if (length > 0) {

                auto remainingLength = SIZE_T(*dTop); pop();

                // Note: The parsed number is now on the top of the stack.

                if (remainingLength == 0) {
                    if (isCompiling) {
                        data(*dTop); pop();
                else {
                    throw AbortException(string("unrecognized word: ") + string(caddr, length));
            else {

EVALUATE can be used to invoke INTERPRET with a string as the source buffer.

// EVALUATE ( i*x c-addr u -- j*x )
void evaluate() {

    auto length = static_cast<size_t>(*dTop); pop();
    auto caddr = CHARPTR(*dTop); pop();

    auto savedInput = std::move(sourceBuffer);
    auto savedOffset = sourceOffset;

    sourceBuffer = string(caddr, length);
    sourceOffset = 0;

    sourceBuffer = std::move(savedInput);
    sourceOffset = savedOffset;

QUIT is the top-level outer interpreter loop. It calls REFILL to read a line, INTERPRET to parse and execute that line, then PROMPT and repeat until there is no more input.

There is an exception handler for AbortException that prints an error message, resets the stacks, and continues.

If end-of-input occurs, then it exits the loop and calls CR and BYE.

If QUIT is called from a word called by QUIT, control returns to the top-level loop.

// PROMPT ( -- )
// Not a standard word.
// Displays "ok" prompt if in interpretation mode.
void prompt() {
    if (!isCompiling) {
        cout << "  ok";

// QUIT ( -- )
void quit() {
    static bool alreadyRunning = false;
    if (alreadyRunning)
    alreadyRunning = true;

    isCompiling = false;

    for (;;) {
        try {
            auto refilled = *dTop; pop();
            if (!refilled) // end-of-input

        catch (const AbortException& abortEx) {
            string msg(abortEx.what());
            if (msg.length() > 0) {
                cout << "<<< " << msg << " >>>" << endl;
            isCompiling = false;



File Access Words

One of my goals is to make cxxforth useful for writing simple shell-like scripts and utilities, and so being able to read and write files and execute Forth scripts is a necessity. I am providing a subset of the File-Access and File-Access extension wordsets from the standards.

As with the user input, I will use C++ iostreams to implement the file access words. This means that a Forth fileid is going to be a pointer to a std::fstream instance.

On some platforms, the C++ iostreams library may be unavailable or incomplete, or you may not want the overhead of linking in these functions. In that case, define the macro CXXFORTH_DISABLE_FILE_ACCESS to disable compilation of these words.


#define FILEID(x) reinterpret_cast<std::fstream*>(x)

// R/O ( -- fam )
void readOnly() {

// R/W ( -- fam )
void readWrite() {
    push(static_cast<Cell>(std::ios_base::in | std::ios_base::out));

// W/O ( -- fam )
void writeOnly() {

// BIN ( fam1 -- fam2 )
void bin() {
    *dTop = *dTop | static_cast<Cell>(std::ios_base::binary);

// CREATE-FILE ( c-addr u fam -- fileid ior )
void createFile() {

    auto caddr = CHARPTR(*(dTop - 2));
    auto length = SIZE_T(*(dTop - 1));
    auto fam = static_cast<std::ios_base::openmode>(*dTop); pop();

    string filename(caddr, length);
    auto f = new std::fstream(filename, fam | std::ios_base::trunc);
    if (f->is_open()) {
        *(dTop - 1) = CELL(f);
        *dTop = 0;
    else {
        delete f;
        *(dTop - 1) = 0;
        *dTop = Cell(-1);

// OPEN-FILE ( c-addr u fam -- fileid ior )
void openFile() {

    auto caddr = CHARPTR(*(dTop - 2));
    auto length = SIZE_T(*(dTop - 1));
    auto fam = static_cast<std::ios_base::openmode>(*dTop); pop();

    string filename(caddr, length);
    auto f = new std::fstream(filename, fam);
    if (f->is_open()) {
        *(dTop - 1) = CELL(f);
        *dTop = 0;
    else {
        delete f;
        *(dTop - 1) = 0;
        *dTop = Cell(-1);

// READ-FILE ( c-addr u1 fileid -- u2 ior )
void readFile() {
    auto f = FILEID(*dTop); pop();
    if (f == nullptr) throw AbortException("READ-FILE: not a valid file ID");
    auto length = SIZE_T(*dTop);
    auto caddr = CHARPTR(*(dTop - 1));
    f->read(caddr, static_cast<std::streamsize>(length));
    *dTop = f->bad() ? Cell(-1) : 0;
    *(dTop - 1) = static_cast<Cell>(f->gcount());

// READ-LINE ( c-addr u1 fileid -- u2 flag ior )
void readLine() {
    auto f = FILEID(*dTop);
    if (f == nullptr) throw AbortException("READ-FILE: not a valid file ID");
    if (f->eof()) {
        *dTop = 0;
        *(dTop - 1) = False;
        *(dTop - 2) = 0;
    auto length = SIZE_T(*(dTop - 1));
    auto caddr = CHARPTR(*(dTop - 2));
    f->getline(caddr, static_cast<std::streamsize>(length) + 1);
    if (f->bad()) {
        *dTop = static_cast<Cell>(-1);
        *(dTop - 1) = 0;
        *(dTop - 2) = 0;
    else if (f->eof() && std::strlen(caddr) == 0) {
        *dTop = 0;
        *(dTop - 1) = False;
        *(dTop - 2) = 0;
    else {
        *dTop = 0;
        *(dTop - 1) = True;
        *(dTop - 2) = std::strlen(caddr);

// READ-CHAR ( fileid -- char ior )
// Not a standard word.
// Reads a single character from the specified file.
// On success, ior is 0 and char is the character read.
// On failure, ior is non-zero and char is undefined.
void readChar() {
    auto f = FILEID(*dTop);
    if (f == nullptr) throw AbortException("READ-CHAR: not a valid file ID");
    auto ch = static_cast<unsigned char>(f->get());
    *dTop = static_cast<Cell>(ch);
    if (f->bad()) push(static_cast<Cell>(-1)); else push(0);

// WRITE-FILE ( c-addr u fileid -- ior )
void writeFile() {
    auto f = FILEID(*dTop); pop();
    if (f == nullptr) throw AbortException("WRITE-FILE: not a valid file ID");
    auto length = SIZE_T(*dTop); pop();
    auto caddr = CHARPTR(*dTop);
    f->write(caddr, static_cast<std::streamsize>(length));
    *dTop = f->bad() ? Cell(-1) : 0;

// WRITE-LINE ( c-addr u fileid -- ior )
void writeLine() {
    auto f = FILEID(*dTop); pop();
    if (f == nullptr) throw AbortException("WRITE-FILE: not a valid file ID");
    auto length = SIZE_T(*dTop); pop();
    auto caddr = CHARPTR(*dTop);
    f->write(caddr, static_cast<std::streamsize>(length));
    (*f) << endl;
    *dTop = f->bad() ? Cell(-1) : 0;

// WRITE-CHAR ( char fileid -- ior )
// Not a standard word.
// Writes a single character to the specified file.
void writeChar() {
    auto f = FILEID(*dTop); pop();
    if (f == nullptr) throw AbortException("WRITE-CHAR: not a valid file ID");
    auto ch = static_cast<char>(*dTop);
    *dTop = f->bad() ? Cell(-1) : 0;

// FLUSH-FILE ( fileid -- ior )
void flushFile() {
    auto f = FILEID(*dTop);
    if (f == nullptr) throw AbortException("FLUSH-FILE: not a valid file ID");
    *dTop = f->bad() ? Cell(-1) : 0;

// CLOSE-FILE ( fileid -- ior )
void closeFile() {
    auto f = FILEID(*dTop);
    if (f == nullptr) throw AbortException("CLOSE-FILE: not a valid file ID");
    delete f;
    *dTop = 0;

// DELETE-FILE ( c-addr u -- ior )
void deleteFile() {

    auto caddr = CHARPTR(*(dTop - 1));
    auto length = SIZE_T(*dTop); pop();

    string filename(caddr, length);
    auto result = std::remove(filename.c_str());
    *dTop = static_cast<Cell>(result);

// RENAME-FILE ( c-addr1 u1 c-addr2 u2 -- ior )
void renameFile() {

    auto lengthNew = SIZE_T(*dTop); pop();
    auto caddrNew = CHARPTR(*dTop); pop();
    auto lengthOld = SIZE_T(*dTop); pop();
    auto caddrOld = CHARPTR(*dTop);

    string oldName(caddrOld, lengthOld);
    string newName(caddrNew, lengthNew);
    auto result = std::rename(oldName.c_str(), newName.c_str());

    *dTop = static_cast<Cell>(result);

// INCLUDE-FILE ( i*x fileid -- j*x )
void includeFile() {

    auto f = FILEID(*dTop); pop();
    if (f == nullptr) throw AbortException("INCLUDE-FILE: invalid file ID");

    string line;
    while (std::getline(*f, line)) {


Memory Allocation

By default, cxxforth's data space is 16K cells. This may be enough for moderate needs, but to process large chunks of data it may be insufficent. One way around this is to define CXXFORTH_DATASPACE_SIZE to the size you need, but a better solution might be to allocate and free memory as needed.

ALLOCATE, RESIZE and FREE are Forth wrappers for C++'s std::malloc(), std::realloc(), and std::free().

// ALLOCATE ( u -- a-addr ior )
void memAllocate() {
    auto size = SIZE_T(*dTop);
    auto p = std::malloc(size);
    if (p) {
        *dTop = CELL(p);
    else {
        *dTop = 0;

// RESIZE ( a-addr1 u -- a-addr2 ior )
void memResize() {
    auto size = SIZE_T(*dTop);
    auto addr = AADDR(*(dTop - 1));
    auto p = std::realloc(addr, size);
    if (p) {
        *dTop = 0;
        *(dTop - 1) = CELL(p);
    else {
        *dTop = static_cast<Cell>(-1);
        *(dTop - 1) = 0;

// FREE ( a-addr -- ior )
void memFree() {
    auto addr = AADDR(*dTop);
    *dTop = 0;


In initializeDefinitions(), I set up the initial contents of the dictionary. This is the Forth kernel that Forth code can use to implement the rest of a working system.

void definePrimitives() {
    static struct {
        const char* name;
        Code code;
    } immediateCodeWords[] = {
        // name             code
        // ------------------------------
        {";",               semicolon},
        {"does>",           does},
        {"immediate",       immediate},
    for (auto& w: immediateCodeWords) {
        definePrimitive(, w.code);

    static struct {
        const char* name;
        Code code;
    } codeWords[] = {
        // name           code
        // ------------------------------
        {"!",               store},
        {"#args",           argCount},
        {"$?",              lastSystemResult},
        {"(;)",             endOfDefinition},
        {"(branch)",        branch},
        {"(does)",          setDoes},
        {"(lit)",           doLiteral},
        {"(zbranch)",       zbranch},
        {"*",               star},
        {"+",               plus},
        {"-",               minus},
        {".",               dot},
        {".r",              dotR},
        {".rs",             dotRS},
        {".s",              dotS},
        {"/",               slash},
        {"/mod",            slashMod},
        {":",               colon},
        {":noname",         noname},
        {"<",               lessThan},
        {"=",               equals},
        {">body",           toBody},
        {">in",             toIn},
        {">num",            parseSignedNumber},
        {">r",              toR},
        {">unum",           parseUnsignedNumber},
        {"@",               fetch},
        {"abort",           abort},
        {"abort-message",   abortMessage},
        {"accept",          accept},
        {"align",           align},
        {"aligned",         aligned},
        {"allocate",        memAllocate},
        {"allot",           allot},
        {"and",             bitwiseAnd},
        {"arg",             argAtIndex},
        {"base",            base},
        {"bl",              bl},
        {"bye",             bye},
        {"c!",              cstore},
        {"c@",              cfetch},
        {"cells",           cells},
        {"cmove",           cMove},
        {"cmove>",          cMoveUp},
        {"compare",         compare},
        {"count",           count},
        {"cr",              cr},
        {"create",          create},
        {"depth",           depth},
        {"drop",            drop},
        {"emit",            emit},
        {"evaluate",        evaluate},
        {"execute",         execute},
        {"exit",            exit},
        {"fill",            fill},
        {"find",            find},
        {"free",            memFree},
        {"here",            here},
        {"hidden",          hidden},
        {"interpret",       interpret},
        {"key",             key},
        {"latest",          latest},
        {"lshift",          lshift},
        {"ms",              ms},
        {"or",              bitwiseOr},
        {"parse",           parse},
        {"pick",            pick},
        {"prompt",          prompt},
        {"quit",            quit},
        {"r>",              rFrom},
        {"r@",              rFetch},
        {"refill",          refill},
        {"resize",          memResize},
        {"roll",            roll},
        {"rshift",          rshift},
        {"see",             see},
        {"source",          source},
        {"state",           state},
        {"system",          system},
        {"time&date",       timeAndDate},
        {"type",            type},
        {"u.",              uDot},
        {"u<",              uLessThan},
        {"unused",          unused},
        {"utctime&date",    utcTimeAndDate},
        {"word",            word},
        {"words",           words},
        {"xor",             bitwiseXor},
        {"xt>name",         xtToName},
        {"bin",             bin},
        {"close-file",      closeFile},
        {"create-file",     createFile},
        {"delete-file",     deleteFile},
        {"flush-file",      flushFile},
        {"include-file",    includeFile},
        {"open-file",       openFile},
        {"r/o",             readOnly},
        {"r/w",             readWrite},
        {"read-char",       readChar},
        {"read-file",       readFile},
        {"read-line",       readLine},
        {"rename-file",     renameFile},
        {"w/o",             writeOnly},
        {"write-char",      writeChar},
        {"write-file",      writeFile},
        {"write-line",      writeLine},
    for (auto& w: codeWords) {
        definePrimitive(, w.code);

    doLiteralXt = findDefinition("(lit)");
    if (doLiteralXt == nullptr) throw runtime_error("Can't find (lit) in kernel dictionary");

    setDoesXt = findDefinition("(does)");
    if (setDoesXt == nullptr) throw runtime_error("Can't find (does) in kernel dictionary");

    exitXt = findDefinition("exit");
    if (exitXt == nullptr) throw runtime_error("Can't find EXIT in kernel dictionary");

    endOfDefinitionXt = findDefinition("(;)");
    if (endOfDefinitionXt == nullptr) throw runtime_error("Can't find (;) in kernel dictionary");

The Forth Part

With our C++ kernel defined, now I can define the remainder of the system using Forth. To do this, I will create an array of Forth text lines to be evaluated when cxxforth initializes itself.

In this section, I won't go into the details of every word defined. In most cases, referring to the standards will be enough to understand what the word is supposed to do and the definition will be easy to understand. But I will provide commentary for a few complicated definitions.

Writing Forth definitions as C++ strings is a little awkward in that I have to escape every " and backslash with a backslash.

static const char* forthDefinitions[] = {

I'll start by defining the remaining basic stack operations. PICK and ROLL are the basis for many of them.

Note that while I'm not implementing any of the Forth double-cell arithmetic operations, double-cell stack operations are still useful.

    ": dup     0 pick ;",
    ": over    1 pick ;",
    ": swap    1 roll ;",
    ": rot     2 roll ;",
    ": nip     swap drop ;",
    ": tuck    swap over ;",
    ": 2drop   drop drop ;",
    ": 2dup    over over ;",
    ": 2over   3 pick 3 pick ;",
    ": 2swap   3 roll 3 roll ;",
    ": 2>r     swap >r >r ;",
    ": 2r>     r> r> swap ;",
    ": 2r@     r> r> 2dup >r >r swap ;",

FALSE and TRUE are useful constants.

    ": false   0 ;",
    ": true    -1 ;",

] enters compilation mode.

[ exits compilation mode.

    ": ]   true state ! ;",
    ": [   false state ! ; immediate",

Forth has a few words for incrementing/decrementing the top-of-stack value.

    ": 1+   1 + ;",
    ": 1-   1 - ;",

    ": cell+   1 cells + ;",
    ": char+   1+ ;",
    ": chars   ;",

+! ( n|u a-addr -- ) adds a value to a cell in memory.

    ": +!   dup >r @ + r> ! ;",

NEGATE and INVERT can be implemented in terms of other primitives.

    ": negate   0 swap - ;",
    ": invert   true xor ;",

, ( x -- ) places a cell value in dataspace.

C, ( char -- ) places a character value in dataspace.

    ": ,    here  1 cells allot  ! ;",
    ": c,   here  1 chars allot  c! ;",

ERASE fills a region with zeros.

    ": erase  0 fill ;",

We have a few extended relational operators based upon the kernel's relational operators. In a lower-level Forth system, these might have a one-to-one mapping to CPU opcodes, but in this system, they are just abbreviations.

    ": >     swap < ;",
    ": u>    swap u< ;",
    ": <>    = invert ;",
    ": 0<    0 < ;",
    ": 0>    0 > ;",
    ": 0=    0 = ;",
    ": 0<>   0= invert ;",

2* and 2/ multiply or divide a value by 2 by shifting the bits left or right.

    ": 2*   1 lshift ;",
    ": 2/   1 rshift ;",

A Forth variable is just a named location in dataspace. I will use CREATE and reserve a cell.

    ": variable   create 0 , ;",
    ": ?          @ . ;",

A Forth constant is similar to a variable in that it is a value stored in dataspace, but using the name automatically puts the value on the stack. I can implement this using CREATE...DOES>.

    ": constant    create ,    does>  @ ;",
    ": 2constant   create , ,  does>  dup cell+ @ swap @ ;",

/CELL is not a standard word, but it is useful to be able to get the size of a cell without using 1 CELLS.

    "1 cells   constant /cell",

DECIMAL and HEX switch the numeric base to 10 or 16, respectively.

    ": decimal   10 base ! ;",
    ": hex       16 base ! ;",

' gets the next word from the input stream and looks up its execution token.

    ": '   bl word find drop ;",

The word LITERAL takes a cell from the stack at compile time, and at runtime will put that value onto the stack. I implement this by compiling a call to (lit) word followed by the value.

Because I will be using (lit) in other word definitions, I'll create a constant '(lit) containing its XT.

    "' (lit)     constant '(lit)",
    ": literal   '(lit) , , ; immediate",

['] is like ', but is an immediate compiling word that causes the XT to beR put on the stack at runtime.

    ": [']   ' '(lit) , , ; immediate",

RECURSE compiles a call to the word currently being defined.

    ": recurse     latest , ; immediate",

CHAR gets the next character and puts its ASCII value on the stack.

[CHAR] is like CHAR, but is an immediate compiling word.

    ": char     bl word char+ c@ ;",
    ": [char]   char '(lit) , , ; immediate",

Control Structures

See the [Control StructuresjonesforthControlStructures section of jonesforth.f for an explanation of how these words work.

One word we have here that is not described in JONESFORTH is AHEAD, which is essentially equivalent to FALSE IF. It is the start of an unconditional forward jump. It is useful for words, like SLITERAL below, that need to store data while compiling. Such words can use AHEAD, then use words like ,, C,, or ALLOT to put data into the dictionary at the current compilation point, then use THEN so that at runtime the inner interpreter will jump over that data.

    ": if       ['] (zbranch) , here 0 , ; immediate",
    ": then     dup  here swap -  swap ! ; immediate",
    ": else     ['] (branch) , here 0 ,  swap dup here swap -  swap ! ; immediate",
    ": ahead    ['] (branch) , here 0 , ; immediate",

    ": begin    here ; immediate",
    ": again    ['] (branch) ,  here - , ; immediate",
    ": until    ['] (zbranch) ,  here - , ; immediate",
    ": while    ['] (zbranch) ,  here swap  0 , ; immediate",
    ": repeat   ['] (branch) ,  here - ,  dup  here swap -  swap ! ; immediate",

Here are some common Forth words I can define now that I have control structures.

    ": ?dup       dup if dup then ;",

    ": abs        dup 0 < if negate then ;",

    ": max        2dup < if swap then drop ;",
    ": min        2dup > if swap then drop ;",

    ": space      bl emit ;",
    ": spaces     begin  dup 0> while  space 1-  repeat  drop ;",

I wish I could explain Forth's POSTPONE, but I can't, so you will just have to Google it.

    ": postpone   bl word find  1 = if , else  '(lit) , ,  ['] , ,  then ; immediate",

A Forth VALUE is just like a constant in that it puts a value on the stack when invoked. However, the stored value can be modified with TO.

VALUE is, in fact, exactly the same as CONSTANT in this Forth. And so you could use TO to change the value of a constant, but that's against the rules.

    ": value    constant ;",

    ": value!   >body ! ;",

    ": to       state @ if",
    "               postpone ['] postpone value!",
    "           else",
    "               ' value!",
    "           then ; immediate",

DEFER and IS are similar to VALUE and TO, except that the value is an execution token, and when the created word is used it invokes that xt. IS can be used to change the execution token. In C++ terms, you can think of this as a pointer to a function pointer.

DEFER and IS are not ANS Forth standard words, but are in common use, and are described formally at

    ": defer       create ['] abort ,",
    "              does> @ execute ;",

    ": defer@      >body @ ;",
    ": defer!      >body ! ;",

    ": is          state @ if",
    "                  postpone ['] postpone defer!",
    "              else",
    "                  ' defer!",
    "              then ; immediate",

    ": action-of   state @ if",
    "                  postpone ['] postpone defer@",
    "              else",
    "                  ' defer@",
    "              then ; immediate",


S" ( "ccc<quote>" -- caddr u )

This word parses input until it finds a " (double quote) and then puts the resulting string's address and length on the stack. It works in both compilation and interpretation mode.

In interpretation mode, it just returns the address and length of the string in the input buffer.

In compilation mode, I have to copy the string somewhere where it can be found at execution time. The word SLITERAL implements this. It compiles a forward-branch instruction, then copies the string's characters into the current definition between the branch and its target instruction, then at the branch target location I use a couple of LITERALs to put the address and length of the word in the definition onto the stack.

." ( "ccc<quote>" -- )

This word prints the given string. We can implement it in terms of S" and TYPE.

.( ( "ccc<quote>" -- )

This is like .", but is an immediate word. It can be used to display output during the compilation of words.

    ": sliteral",                      // ( c-addr len )
    "    postpone ahead",              // ( c-addr len orig )
    "    2dup swap >r >r",             // ( c-addr len orig ) ( R: len orig )
    "    cell+ swap",                  // copy into the first byte after the offset
    "    dup allot  cmove align",      // allocate dataspace and copy string into it
    "    r> dup postpone then",        // resolve the branch
    "    cell+ postpone literal",      // compile literal for address
    "    r> postpone literal",         // compile literal for length
    "; immediate",

    ": s\"   [char] \" parse",
    "        state @ if postpone sliteral then ; immediate",

    ": .\"   postpone s\" postpone type ; immediate",

    ": .(    [char] ) parse type ; immediate",

/STRING ( c-addr1 u1 n1 -- c-addr2 u2 ) adjusts a character string by adding to the address and subtracting from the length.

    ": /string   dup >r - swap r> + swap ;",

ABORT" checks whether a result is non-zero, and if so, it throws an exception that will be caught by QUIT, which will print the given message and then continue the interpreter loop.

    ": (abort\")   rot if abort-message then 2drop ;",
    ": abort\"     postpone s\" postpone (abort\") ; immediate",

INCLUDED is the word for reading additional source files. For example, you can include the file tests/hello.fs and then run its hello word by doing the following:

s" tests/hello.fs" INCLUDED

INCLUDE is a simpler variation, used like this:

INCLUDE tests/hello.fs

INCLUDE is not part of the AND standard, but is in Forth 2012.


    ": included",
    "    r/o open-file abort\" included: unable to open file\"",
    "    dup include-file",
    "    close-file abort\" included: unable to close file\" ;",

    ": include   bl word count included ;",



There is a good reason that none of the Forth defintions above have had any stack diagrams or other comments: our Forth doesn't support comments yet. I have to define words to implement comments.

I will support two standard kinds of Forth comments:

  • If \ (backslash) appears on a line, the rest of the line is ignored.
  • Text between ( and ) on a single line is ignored.

Also, I will support #! as a synonym for \, so that we can start a UNIX shell script with something like this:

#! /usr/local/bin/cxxforth

Note that a space is required after the \, (, or #! that starts a comment. They are blank-delimited words just like every other Forth word.

To-Do: ( should support multi-line comments.

    ": \\   source nip >in ! ; immediate",
    ": #!   postpone \\ ; immediate",
    ": (    [char] ) parse 2drop ; immediate",

ABOUT is not a standard word. It just prints licensing and credit information.

.DQUOT is also not a standard word. It prints a double-quote (") character.

    ": .dquot   [char] \" emit ;",

    ": about",
    "      cr",
    "      .\" cxxforth " CXXFORTH_VERSION "\" cr",
    "      .\" by Kristopher Johnson\" cr",
    "      cr",
    "      .\" This is free and unencumbered software released into the public domain.\" cr",
    "      cr",
    "      .\" Anyone is free to copy, modify, publish, use, compile, sell, or distribute this\" cr",
    "      .\" software, either in source code form or as a compiled binary, for any purpose,\" cr",
    "      .\" commercial or non-commercial, and by any means.\" cr",
    "      cr",
    "      .\" In jurisdictions that recognize copyright laws, the author or authors of this\" cr",
    "      .\" software dedicate any and all copyright interest in the software to the public\" cr",
    "      .\" domain. We make this dedication for the benefit of the public at large and to\" cr",
    "      .\" the detriment of our heirs and successors. We intend this dedication to be an\" cr",
    "      .\" overt act of relinquishment in perpetuity of all present and future rights to\" cr",
    "      .\" this software under copyright law.\" cr",
    "      cr",
    "      .\" THE SOFTWARE IS PROVIDED \" .dquot .\" AS IS\" .dquot .\"  WITHOUT WARRANTY OF ANY KIND, EXPRESS OR\" cr",
    "      cr",
    "      .\" For more, visit <>.\" cr ;",

The C++ main() function will look for the Forth word MAIN and execute it.

The MAIN word calls PROCESS-ARGS, which is not a standard word. It looks at the number of command-line arguments. If there are no arguments other than the executable path, then it prints the WELCOME message. If there are arguments, then it attempts to call INCLUDED on each of them.

If you want to write your own custom startup code, MAIN is the place to put it.

    ": welcome",
    "    .\" cxxforth " CXXFORTH_VERSION "\" cr",
    "    .\" Type \" .dquot .\" about\" .dquot .\"  for more information.  \"",
    "    .\" Type \" .dquot .\" bye\" .dquot .\"  to exit.\" cr ;",

    ": process-args",
    "    #args 1 = if welcome exit then",
    "    1 begin",
    "        dup #args <",
    "    while",
    "        dup arg included cr",
    "        1+",
    "    repeat",
    "    drop ;",

    ": main   process-args quit ;",

That is the end of the built-in Forth definitions.

With the forthDefinitions array filled, all I need to do is call EVALUATE on each line to load them into the system.

void defineForthWords() {
    static size_t lineCount = sizeof(forthDefinitions) / sizeof(forthDefinitions[0]);
    for (size_t i = 0; i < lineCount; ++i) {
        auto line = forthDefinitions[i];
        auto length = std::strlen(line);

void initializeDefinitions() {

} // end anonymous namespace

const char* cxxforth_version = CXXFORTH_VERSION;

extern "C" void cxxforth_reset() {

    std::memset(dStack, 0, sizeof(dStack));
    dTop = dStack - 1;

    std::memset(rStack, 0, sizeof(rStack));
    rTop = rStack - 1;

    std::memset(dataSpace, 0, sizeof(dataSpace));
    dataPointer = dataSpace;


extern "C" int cxxforth_main(int argc, const char** argv) {
    try {
        commandLineArgCount = static_cast<size_t>(argc);
        commandLineArgVector = argv;


        auto mainXt = findDefinition("MAIN");
        if (!mainXt)
            throw runtime_error("MAIN not defined");

        return 0;
    catch (const exception& ex) {
        cerr << "cxxforth: " << ex.what() << endl;
        return -1;

Finally we have our main(). If there are no command-line arguments, it prints a banner and help message. Then it calls cxxforth_main().

You can define the macro CXXFORTH_NO_MAIN to inhibit generation of main(). This is useful for incorporating cxxforth.cpp into another application or library.


int main(int argc, const char** argv) {
    return cxxforth_main(argc, argv);