Skip to content
Switch branches/tags

Latest commit


Git stats


Failed to load latest commit information.
Latest commit message
Commit time

% FORTH(1) % Richard Howe % November 2016


forth - a forth interpreter


forth [-s file] [-e string] [-l file] [-m size] [-VthvLSnx] [-] [files]


A Forth interpreter built around a library, libforth, that implements a complete Forth interpreter.

For a much better, albeit 16-bit only, Forth that is also very well documented and (slightly more) standards compliant, please visit the more up to date Embed virtual machine. Also of interest is the Forth CPU I wrote, in VHDL, for an FPGA.

This interpreter is available at here.

Libforth build status


Command line switches must be given before any files, unless that switch takes a file as an argument.

  • -s file

This saves the working memory of the Forth interpreter to a file, which can later be loaded with the "-l" option. If a core file has been invalidated this will not be saved, invalidation occurs when an unrecoverable error has been detected that would prevent any recovery or meaningful execution with the current image.

  • -e string

Evaluate a Forth string passed in as an argument.

  • -t

After all the files have been read from and any core files have been loaded this will make the Forth interpreter read from stdin, the core file will be saved after stdin has been read from and there is no more work to do, if the "-d" or "-s" flags have been specified.

  • -h

Print out a short help message and exit unsuccessfully.

  • -v

Turn verbose mode on, more information will be printed out, to stderr, about what is happening in the interpreter. Usually the interpreter is as silent as possible.

  • -m size

Specify the virtual machines memory size in kilobytes, overriding the default memory size. This is mutually exclusive with "-l".

  • -l file

This option loads a forth core file generated from the "-d" option of a previous run. This core file is not portable and must be generated on the same platform as it was generated. It can only be specified once per run of the interpreter.

  • -L

The same as "-l", however the default core file name is used, "forth.core", so an argument does not have to be provided.

  • -S

The same as "-s", however the default core file name is used, "forth.core", so an argument does not have to be provided.

  • '-'

Stop processing any more command line options and treat all arguments after this as files to be executed, if there are any.

  • -V

Print version and interpreter information and exit successfully.

  • -f file

Process a file immediately. This allows options and file arguments to be intermingled.

  • -n

If the line editing library is compiled into the executable, which is a compile time option, then when reading from stdin this will use a line editor to read in a line at a time. This option implies -t.

  • -x

Enable signal handling, signal handling is experimental and not quite worked out yet. It is off by default as I find it annoying when programs catch signals when you really want to program to die. This cannot be enabled from within the Forth interpreter. This option should disappear once signal handling has been sorted out.

  • file...

If a file, or list of files, is given, read from them one after another and execute them. The dictionary and any stored Forth blocks will persist, as will values on the stack.

If no files are given to execute stdin will be read from.



Execute any commands given from stdin

./forth -t file1.4th file2.4th

Execute file "file1.4th", then "file2.4th", then read from stdin

./forth file1.4th

Execute file "file1.4th".

./forth -s file1.4th

Execute file "file1.4th", the produce a "forth.core" save file.

./forth -s -l forth.core

Load a "forth.core" file, read from stdin and execute any commands given, then dump the new core file to "forth.core".

The interpreter returns zero on success and non zero on failure.


The Forth interpreter and the library that implements it are released under the MIT license. Copyright (c) Richard Howe, 2016.


This program will return a non-zero value on failure, and zero on success.




If you find a bug, or would like to request a new feature, please Email me at:

    howe.r.j.89 [ at ] gmail . com

The interpreter has not been battle hardened yet so there is likely behavior that is non-standard (for no reason) or just outright incorrect.


This small Forth interpreter is based on a de-obfuscated entrant into the IOCCC by buzzard. The entry described a Forth like language which this derives from. You can use this library to evaluate Forth strings or as an embeddable interpreter. Work would need to be done to get useful information after doing those evaluations, but the library works quite well.

main.c is simply a wrapper around one the functions that implements a simple REPL.

This project implements a Forth interpreter library which can be embedded in other projects, it is incredibly minimalistic, but usable. To build the project a C compiler is needed, and a copy of Make, type:

make help

For a list of build options. By running:

make run

Will build the interpreter and run it, it will then read from stdin.

To build the documentation other programs may be needed, such as pandoc and the markdown script, but these steps are optional.

Forth is an odd language that has a loyal following groups, but it is admittedly not the most practical of language as it lacks nearly everything the modern programmer wants in a language; safety, garbage collection, modularity and clarity. It is however possible to implement a fully working interpreter in a one to two kilobytes of assembly, those kilobytes can make a functional and interactive programming environment, giving a high ratio of utility memory used.

From the Wikipedia article we can neatly summarize the language:

    "Forth is an imperative stack-based computer programming language 
    and programming environment. 
    Language features include structured programming, reflection (the
    ability to modify the program structure during program execution), 
    concatenative programming (functions are composed with juxtaposition) 
    and extensibility (the programmer can create new commands).


    A procedural programming language without type checking, Forth features
    both interactive execution of commands (making it suitable as a shell
    for systems that lack a more formal operating system) and the ability 
    to compile sequences of commands for later execution." 

Given the nature of the Forth language it does not make for a terribly good embeddable scripting language, but it is simple to implement and can be fun to use. This interpreter is based off a previous IOCCC in a file called buzzard.2.c, it is a descendant of that file.

Before using and understanding this library/interpreter it is useful to checkout more literature on Forth such as Thinking Forth by Leo Brodie for a philosophy of the language, Starting Forth (same Author), Jonesforth which is a specific implementation of the language in x86 assembly and Gforth, a more modern and portable implementation of the language.

It is important to realize that Forth is really more a philosophy and collection of ideas than a specific reference implementation or standard. It has been said that an intermediate Forth user is one who has implemented a Forth interpreter, something which cannot be said about other languages nor is possible given their complexity.

The saying "if you have seen one Forth implementation, you have seen one Forth implementation" applies, nearly every single Forth implementation has its own idea of how to go about things despite standardization efforts - in keeping with this, this library has its own idiosyncrasies.

This implementation, written in C, can be thought of as a hybrid between a fairly dumb stack based virtual machine with instructions such as "pop two values off the stack, add them, and push the result" and a small interpreter/compiler for the virtual machine. This simple kernel is then used to build a more compliant and usable Forth implementation by defining words that build upon those provided by the base system.

Other documentation

Apart from this file there are other sources of information about the project:

As can the code, which is small enough to be comprehensible:

And the forth startup code:

The startup code is well commented and shows how the core interpreter is extended to a more function Forth environment.

The source file libforth.c can be converted to a more readable webpage by first converting the source to markdown with convert script, the converting that to HTML in the usual fashion

Using the interpreter

main.c simple calls the function main_forth() in libforth.c, this function initializes a Forth environment and puts the user in a REPL where you can issue commands and define words. See the manual pages for list of command line options and library calls. All commands are given using Reverse Polish Notation (or RPN),




    4 2 * 2 +

And brackets are no longer needed. Numbers of pushed on to the variable stack automatically and commands (such as '*' and '+') take their operands off the stack and push the result. Juggling variables on the stack becomes easier over time. To pop a value from the stack and print it there is the '.' word.


    2 2 + . 



The simplicity of the language allows for a small interpreter, the loop looks something like this:

    1) Read in a space delimited Forth WORD.
    2) Is this WORD in the dictionary?
       FOUND)      Are we in IMMEDIATE mode?
                   IMMEDIATE-MODE) Execute WORD.
                                   goto 1;
                   COMPILE-MODE)   Compile WORD into the dictionary.
                                   goto 1;
       NOT-FOUND)  Is this actually a number?
                   YES) Are we in IMMEDIATE mode?
                        IMMEDIATE-MODE) Push Number onto the stack.
                                        goto 1;
                        COMPILE-MODE)   Compile a literal number.
                                        goto 1;
                   NO)  Error! Handle error
                        goto 1;

Given that we are reading in space delimited words if follows that the above expression:

    2 2 + .

Would not work if we did:

    2 2+ .


    2 2 +.

As "2+" and "+." would be parsed as words, which may or may not be defined and if they are do not have the behavior that we want. This is more apparent when we do any kind of string handling.

A Forth Word

The Forth execution model uses Threaded Code, the layout of a word header follows from this.

A Forth word is defined in the dictionary and has a particular format that varies between implementations. A dictionary is simply a linked list of Forth words, the dictionary is usually contiguous and can only grow. The format for our Forth words is as follows:


  • Word Header:
  • field <0 = Word Name (the name is stored before the main header)
  • field 0 = Previous Word
  • field 1 = Code Word (bits 0 - 7) | Hidden Flag (bit 8) | Word Name Offset (bit 9 - 14) | Compiling bit (bit 15)
  • field 2+ = Data field (if it exists).

And in more detail:

    |       Word Header          | Word Body |
    | NAME ...      | PWD | MISC | DATA ...  |

    NAME        = The name, or the textual representation, of a Forth
                  word, it is a variable length field that is ASCII NUL
                  terminated, the MISC field has an offset that points
                  to the begining of this field if taken off the PWD
                  position (not value). The offset is in machine words,
                  not characters.
    PWD         = A pointer to the previously declared word.
    MISC        = A complex field that can contains a CODE WORD, a
                  "hide" bit and the offset from the PWD field to the
                  beginning of NAME, as well as the compiling bit.
    DATA        = This could be anything, but it is most likely to be
                  a list of pointers to CODE WORDs of previously defined
                  words if this optional DATA field is present.

All fields are aligned on the Forth virtual machines word boundaries.

The MISC field is laid out as so:

    | <- Least Significant Bit                              Most Significant Bit -> |
    |  0 |  1 |  2 |  3 |  4 |  5 |  6 |  7 |  8 |  9 | 10 | 11 | 12 | 13 | 14 | 15 |
    |            CODE WORD             | HD |          NAME OFFSET             | CB |
    CODE WORD    = Bits 0-6 are a code word, this code word is always run
                   reguardless of whether we are in compiling or command
    HD           = Bit 7 is the Hide Bit, if this is true then when
                   compiling or executing words the word will be hidden from the 
    NAME OFFSET  = Bits 8 to 15 are the offset to the words name. To find the 
                   beginning of the words name we take this value away from
                   position of this words PWD header. This value is in
                   machine words, and so the beginning of the NAME must be aligned 
                   to the virtual machine words boundaries and not character, or byte, 
                   aligned. The length of this field, and the size of the input buffer, 
                   limit the maximum size of a word.
CB           = Compiling bit, if set this is a compiling word, if
               cleared it is an immediate word. 

Depending on the virtual machine word size, or cell size, there may be more bits above bit '15', the most significant bit, in the MISC field. These bits are not used and should be set to zero.

And the dictionary looks like this:

   [ Special 'fake' word ]
   | NAME  | PWD | Rest of the word ... |
     The rest of the dictionary
   | NAME  | PWD | Rest of the word ... |
   | NAME  | PWD | Rest of the word ... |
   [ Previous Word Register ]

Searching of the dictionary starts from the Previous Word Register and ends at a special 'fake' word.

Defining words adds them to the dictionary, we can defined words with the ':' words like this:

    : two-times 2 * ;

Which defined the word "two-times", a word that takes a value from the stack, multiplies it by two and pushes the results back onto the stack.

The word ':' performs multiple actions; it is an immediate word that reads in the next space delimited word from the input stream and creates a header for that word. It also switches the interpreter into compile mode, compiling words will be compiled into that word definition instead of being executed, immediate words are executed as normal. ';' is also an immediate word, it compiles a special word exit into the dictionary which returns from a word call and switches the interpreter back into command mode. This type of behavior is typical of Forth implementations.

Memory Map and Special Registers

The way this interpreter works is that is emulates an idealized machine, one built for executing Forth directly. As such it has to make compromises and treats certain sectors of memory as being special, as shown below (numbers are given in hexadecimal and are multiples of the virtual machines word-size which is either 16, 32 or 64 bit depending on compile time options.

Where the dictionary ends and the variable and return stacks begin depends on how much memory was allocated to the interpreter (with a minimum of 2048 words), the default is 32768 words, and the following diagram assumes this:

    | 0-3F      | 40-7BFF       |7C00-7DFF|7E00-7FFF|
    | Registers | Dictionary... | V stack | R stack |

    V stack = The Variable Stack
    R stack = The Return Stack

Each may be further divided into special sections:


At the beginning of the Forth virtual machine there is a section used for registers, modifying them arbitrary can cause undefined behavior to occur which will most likely cause the virtual machine to be terminated.

              DECIMAL  HEX
               0-1      0-1    Unused
               2-5      2-5    Push integer word
DIC            6        6      Dictionary pointer
RSTK           7        7      Return stack pointer
STATE          8        8      Interpreter state; compile/command mode
BASE           9        9      Base conversion variable
PWD            10       A      Pointer to last defined word 
SOURCE_ID      11       B      Input source selector (-1 = string input, 
                               0 = file input)
SIN            12       C      String input pointer
SIDX           13       D      String input index  (index into SIN)
SLEN           14       E      String input length (length of SIN)
START_ADDR     15       F      Pointer to start of VM
FIN            16       10     File input pointer
FOUT           17       11     File output pointer 
STDIN          18       12     File pointer to stdin, if available
STDOUT         19       13     File pointer to stdout, if available
STDERR         20       14     File pointer to stderr, if available
ARGC           21       15     Count of arguments passed to program,
                               if available
ARGV           22       16     An array of pointers to NUL terminated
                               ASCII strings, if available, of ARGC
DEBUG          23       17     Turn debugging on/off if enabled
INVALID        24       18     If non zero, this interpreter is invalid 
TOP            25       19     Stored version of top of stack
INSTRUCTION    26       1A     Stored version of instruction pointer
STACK_SIZE     27       1B     Size of the variable stack
ERROR_HANDLER  28       1C     Action to take on error
THROW          29       1D     Used for throw/catch
SIGNAL_HANDLER 30       1E     Used for signal handling
SCRATCH_X      31       1F     Scratch variable for the user

Some registers will need more explaining.


This register is used when a signal is caught, it is up to the C environment to call forth_signal from a signal handler in the C environment to let the Forth interpreter know a signal has been caught.


Scratch X is a variable that can be used by the user, be warned that other words defined in the base library might use this register as well.


Apart from the constraints that the dictionary begins after where the registers are and before where V stack is there are no set demarcations for each region, although currently the defined word region ends before 0x200 leaving room between that and 0x7BFF for user defined words.

    | 40-???            | ???-???          | ???-7BFF                |
    | Special read word | Interpreter word | Defined word ...        |

    Special read word = A word called on entrance to the interpreter, 
                        it calls itself recursively (as a tail call). This
                        word cannot be 'found', it does not have a name.
    Interpreter word  = Any named (not 'invisible' ones) interpreter word 
                        gets put here.
    Defined word      = A list of words that have been defined with ':'

Glossary of Forth words

Each word is also given with its effect on the variable stack, any other effects are documented (including the effects on other stacks). Each entry looks like this:

  • word ( y -- z )

Where 'word' is the word being described, the contents between the parenthesis describe the stack effects, this word expects one number to be one the stack, 'y', and returns a number to the stack 'z'.

Some of the descriptions of the variables in the stack effect comment have a meaning:

|  Name   |                   Meaning                       |
| addr    | An address in cells                             |
| c-addr  | Character address                               |
| r-addr  | A raw address                                   |
| file-id | A file id, used as a handle for file operations |
| ior     | A error status returned by file operations      |
| char    | Single byte / character                         |
| u       | An unsigned value                               |
| x       | A signed value                                  |
| c" xxx" | The word parses a word (not a stack effect)     |

The addressing modes complicate certain operations of Forth words. A normal Forth address, as passed to words like '!', '@', 'c@' and 'c!' are relative to the beginning of the start of the Forth cores memory, or at register zero. '!' and '@' take their addresses in cells (addr), whereas 'c@' and 'c!' take their addresses in character address (c-addr).

Raw addresses are however normal addresses, as understood by C programs and the computer. Operations on raw addresses can possibly result in undefined behavior, and all objects they may operate on that exists outside of the Forth core cannot be guaranteed to exists after a core file has been reloaded (that is pointers to objects outside of Forth core should not be saved for further use). Raw-addresses are also in character, not cell, units.

In future revisions of Forth this memory addressing might be simplified, if a simple way of doing this can be achieved.

Internal words

There are three types of words.

'Invisible' words

These invisible words have no name but are used to implement the Forth. They are all immediate words.

  • push ( -- x )

Push the next value in the instruction stream onto the variable stack, advancing the instruction stream.

  • const ( -- x )

Push the next location after this instruction, not the instruction stream, onto the variable stack.

  • run ( -- )

Save the current instruction stream pointer onto the return stack and set the pointer instruction stream pointer to point to value after run.

Immediate words

These words are named and are immediate words.

  • ':' ( c" xxx" -- )

Read in a new word from the input stream and compile it into the dictionary.

  • 'immediate' ( -- )

Make the previously declared word immediate. Unlike in most Forth implementations this is used after the words name is given not after the final ';' has been reached.


    : word immediate ... ;

Instead of:

    : word ... ; immediate
  • '\' ( c" \n" -- )

A comment, ignore everything until the end of the line.

Compiling words

  • 'read' ( c" xxx" -- )

read is a complex word that implements most of the input interpreter, it reads in a Forth word (up to 31 characters), if this word is in the dictionary it will either execute the word if we are in command mode or compile a pointer to the executable section of the word if in compile mode. If this word is not in the dictionary it is checked if it is a number, if it is then in command mode we push this value onto the variable stack, if in compile mode then we compile a literal into the dictionary. If it is none of these we print an error message and attempt to read in a new word.

  • '@' ( address -- x )

Pop an address and push the value at that address onto the stack.

  • '!' ( x address -- )

Given an address and a value, store that value at that address.

  • 'c@' ( char-address -- char )

Pop a character address and push the character value at that address onto the stack. Note that this access is not checked for being within range of the virtual machines memory, but it is still relative to the start address of virtual machine memory.

  • 'c!' ( char char-address -- )

Given a character address, store a character value at that address, like 'c@' the address is relative to the virtual machines starting address.

  • '-' ( x y -- z )

Pop two values, subtract 'y' from 'x' and push the result onto the stack.

  • '+' ( x y -- z )

Pop two values, add 'y' to 'x' and push the result onto the stack.

  • 'and' ( x y -- z )

Pop two values, compute the bitwise 'AND' of them and push the result on to the stack.

  • 'or' ( x y -- z )

Pop two values, compute the bitwise 'OR' of them and push the result on to the stack.

  • 'xor' ( x y -- z )

Pop two values, compute the bitwise 'XOR' of them and push the result on to the stack.

  • 'invert' ( x y -- z )

Perform a bitwise negation on the top of the stack.

  • 'lshift' ( x y -- z )

Pop two values, compute 'y' shifted by 'x' places to the left and push the result on to the stack.

  • 'rshift' ( x y -- z )

Pop two values, compute 'y' shifted by 'x' places to the right and push the result on to the stack.

  • '*' ( x y -- z )

Pop two values, multiply them and push the result onto the stack.

  • '/' ( x y -- z )

Pop two values, divide 'x' by 'y' and push the result onto the stack. If 'y' is zero and error message is printed and 'x' and 'y' will remain on the stack, but execution will continue on as normal.

  • 'u<' ( x y -- z )

Pop two unsigned values, compare them (y < x) and push the result onto the stack, the comparison will be unsigned.

  • 'u>' ( x y -- z )

Pop two values, compare them (y > x) and push the result onto the stack. The comparison will be unsigned.

  • '_exit' ( -- )

Pop the return stack and set the instruction stream pointer to that value.

  • 'exit' ( -- )

This does the same as _exit, the reason there are two exits instead of one is so that the word 'see', defined in forth.fth can differentiate between an exit that occurs in the definition of a word, and one that occurs at the end of a words definition. ('see' is a decompiler for Forth).

For example:

: test 0 begin dup 10 > if exit then dup . cr 1+  again ;

Gets Compiled to:

Address        Contents 
         ._____._____._____._____. <- Start of Word
  X      | 't' | 'e' | 's' | 't' | 
  X+1    |  0  |  0  |  0  |  0  | NUL terminates 'test' string
  X+2    | Previous Word Pointer | AKA 'PWD' field
  X+3    |       MISC Field      | <- Execution Starts here
  X+4    | Literal               | Literals a compiled as a pointer to
         ._______________________. a 'literal' word and the literal in
  X+5    | 0                     | the next field.
  X+6    | Pointer to 'dup'      |
  X+7    | literal               |
  X+8    | 10                    |
  X+9    | Pointer to '>'        |
  X+10   | Pointer to 'branch?'  | 'if' gets compiled to 'branch?'
         ._______________________. and '2' so it jumps over 'exit'
  X+11   | 2                     | if the previous test fails. This
         ._______________________. is encoded as the jump destination
  X+12   | Pointer to 'exit'     | less one as an increment happens
         ._______________________. after the word is executed.
  X+13   | Pointer to 'dup'      |
  X+14   | Pointer to '.'        |
  X+15   | Pointer to 'cr'       |
  X+16   | Pointer to '1+'       |
  X+17   | Pointer to 'branch'   |
  X+18   | -12                   |
  X+19   | Pointer to '_exit'    |
         ._______________________. <- End of Word

The decompiler knows that the end of a word is demarcated by a pointer to _exit, and that pointers to exit can occur within the body of the definition.

  • 'key' ( -- char )

Get a value from the input and put it onto the stack.

  • '_emit' ( char -- status )

Put a character to the output stream returning a success value.

  • 'r>' ( -- x )

Pop a value from the return stack and push it to the variable stack.

  • '>r' ( x -- )

Pop a value from the variable stack and push it to the return stack.

  • 'branch' ( -- )

Jump unconditionally to the destination next in the instruction stream.

  • '?branch' ( bool -- )

Pop a value from the variable stack, if it is zero the jump to the destination next in the instruction stream, otherwise skip over it.

  • '(.)' ( x -- status )

Pop a value from the variable stack and print it to the output either as a ASCII decimal or hexadecimal value depending on the BASE register. A return status is pushed onto the stack, greater or equal to zero is a success, negative is a failure. Failure can occur because of an invalid base in the BASE register, or because the output could not be written to.

  • ''' ( -- )

Push the next value in the instruction stream onto the variable stack and advance the instruction stream pointer over it.

  • ',' ( x -- )

Write a value into the dictionary, advancing the dictionary pointer.

  • '=' ( x y -- z )

Pop two values, perform a test for equality and push the result.

  • 'swap' ( x y -- y z )

Swap two values on the stack.

  • 'dup' ( x -- x x )

Duplicate a value on the stack.

  • 'drop' ( x -- )

Drop a value.

  • 'over' ( x y -- x y x )

Duplicate the value that is next on the stack.

  • 'find' ( -- execution-token )

Find a word in the dictionary pushing a pointer to that word onto the variable stack.

  • 'depth' ( -- depth )

Push the current stack depth onto the stack, the value is the depth of the stack before the depth value was pushed onto the variable stack.

  • 'sp@' ( -- addr )

Push the address of the stack pointer onto the stack, before sp@ was executed:

1 2 sp@ . . .


2 2 1
  • 'sp!' ( addr -- )

Set the address of the stack pointer.

  • 'clock' ( -- x )

Push the difference between the startup time and now, in milliseconds. This can be used for timing and implementing sleep functionality, the counter will not increase the interpreter is blocking and waiting for input, although this is implementation dependent.

  • 'evaluator' ( c-addr u 0 | file-id 0 1 -- x )

This word is a primitive used to implement 'evaluate' and 'include-file', it takes a boolean to decide whether it will read from a file (1) or a string (0), and then takes either a forth string, or a file-id.

  • 'system' ( c-addr u -- status )

Execute a command with the systems command interpreter.

  • 'raise' ( u -- ??? )

Raises a signal, what happens after the signal is raised is undefined.

  • 'date' ( -- date )

Push the date onto the stack, the order of the arguments are:

Is day light savings time? 
Days since start of year
Week day
Day of month
Seconds (note, this can include lead seconds!)

The time is in UTC time.

  • 'memory-copy' ( r-addr1 r-addr2 u -- )

Operates on two raw addresses. Copy 'u' characters from r-addr2 to r-addr1.

  • 'memory-locate' ( r-addr char u -- r-addr | 0 )

Locate a character in a block of memory 'u' characters wide, returning a pointer to that character or zero if that address cannot be found.

  • 'memory-set' ( r-addr char u -- )

Set 'u' character of memory starting at 'r-addr' to 'char'.

  • 'memory-compare' ( r-addr1 r-addr2 u -- x )

Compare two blocks of memory 'u' units wide.

  • 'allocate' ( u -- r-addr status )

Allocate a block of memory.

  • 'free' ( r-addr -- status )

Free a block of memory.

  • 'getenv' ( c-addr u -- r-addr u )

Get an environment variable given a string, it returns '0 0' if the variable was not found.

File Access Words

The following compiling words are part of the File Access Word set, a few of the fields need explaining in the stack comments. "file-id" refers to a previously opened file as returned by "open-file", "ior" refers to a return status provided by the file operations. "fam" is a file access method,

  • 'close-file' ( file-id -- ior )

Close an already opened file.

  • 'open-file' ( c-addr u fam -- file-id ior )

Open a file, given a Forth string (the 'c-addr' and the 'u' arguments), and a file access method, which is defined within "forth.fth". Possible file access methods are "w/o", "r/w" and "r/o" for read only, read-write and write only respectively.

  • 'delete-file' ( c-addr u -- ior )

Delete a file on the file system given a Forth string.

  • 'read-file' ( c-addr u file-id -- ior )

Read in 'u' characters into 'c-addr' given a file identifier.

  • 'write-file' ( c-addr u file-id -- ior )

Write 'u' characters from 'c-addr' to a given file identifier.

  • 'file-position' ( file-id -- ud ior )

Get the file position offset from the beginning of the file given a file identifier.

  • 'reposition-file' ( ud file-id -- ior )

Reposition a files offset relative to the beginning of the file given a file identifier.

  • 'flush-file' ( file-id -- ior )

Attempt to flush any buffered information written to a file.

  • 'rename-file' ( c-addr1 u1 c-addr2 u2 -- ior )

Rename a file on the file system named by the first string ('c-addr1' and 'u1') to the second string ('c-addr2' and 'u2').

  • 'temporary-file' ( -- file-id ior )

Open up a new temporary file for writing and reading.

Defined words

Defined words are ones which have been created with the ':' word, some words get defined before the user has a chance to define their own to make their life easier.

  • 'state' ( -- addr )

Push the address of the register that controls the interpreter state onto the stack, this value can be written to put the interpreter into compile or command modes.

  • ';' ( -- )

Write 'exit' into the dictionary and switch back into command mode.

  • 'base' ( -- addr )

This pushes the address of a variable used for input and output conversion of numbers, this address can be written to and read, valid numbers to write are 0 and 2 to 36 (not 1).

  • 'pwd' ( -- pointer )

Pushes a pointer to the previously define word onto the stack.

  • 'h' ( -- pointer )

Push a pointer to the dictionary pointer register.

  • 'r' ( -- pointer )

Push a pointer to the register pointer register.

  • 'here' ( -- dictionary-pointer )

Push the current dictionary pointer (equivalent to "h @").

  • '[' ( -- )

Immediately switch into command mode.

  • ']' ( -- )

Switch into compile mode

  • '>mark' ( -- location )

Write zero into the head of the dictionary and advance the dictionary pointer, push a address to the zero written into the dictionary. This is usually used after in a word definition that changes the control flow, after a branch for example.

  • ':noname' ( -- execution-token )

This creates a word header for a word without a name and switches to compile mode, the usual ';' finishes the definition. It pushes a execution token onto the stack that can be written into the dictionary and run, or executed directly.

  • 'if' ( bool -- )

Begin an if-else-then statement. If the top of stack is true then we execute all between the if and a corresponding 'else' or 'then', otherwise we skip over it.

Abstract Examples:

    : word ... bool if do-stuff ... else do-other-stuff ... then ... ;

    : word ... bool if do-stuff ... then ... ;

and a concrete examples:

    : test-word if 2 2 + . cr else 3 3 * . cr ;
    0 test-word
    4             # prints 4
    1 test-word
    9             # prints 9

Is a simple and contrived example.

  • 'else' ( -- )

See 'if'.

  • 'then' ( -- )

See 'if'.

  • 'begin' ( -- )

This marks the beginning of a loop.

  • 'until' ( bool -- )

Loop back to the corresponding 'begin' if the top of the stack is zero, continue on otherwise.

  • "')'" ( -- char )

Push the number representing the ')' character onto the stack.

  • 'tab' ( -- )

Print a tab.

  • 'cr' ( -- )

Prints a newline.

  • '(' ( -- )

This will read the input stream until encountering a ')' character, it is used for comments.

  • 'allot' ( amount -- )

Allocate a number of cells in the dictionary.

  • 'tuck' ( x y -- y x y )

The stack comment documents this word entirely.

  • 'nip' ( x y -- y )

The stack comment documents this word entirely.

  • 'rot' ( x y z -- z x y )

The stack comment documents this word entirely. This word rotates three items on the variable stack.

  • '-rot' ( x y z -- y z x )

The stack comment documents this word entirely. This word rotates three items on the variable stack, in the opposite direction of "rot".

  • 'emit' ( x -- )

Write a single character out to the output stream.

  • '?' ( addr -- )

Print the contents of addr to the screen.

Library of Forth words

The file forth.fth contains many defined words, however those words are documented within that file and so as to avoid duplication will not be mentioned here. This file is not loaded automatically, and so should be run like this:


    ./forth -t forth.fth 


forth.exe -t forth.fth

Glossary of Forth terminology

  • Word vs Machine-Word

Usually in computing a 'word' refers to the natural length of integer in a machine, the term 'machine word' is used to invoke this specific meaning, a word in Forth is more analogous to a function, but there are different types of Forth words; immediate and compiling words, internal and defined words and finally visible and invisible words.

The distinction between a machine word and a Forth word can lead to some confusion.

  • The dictionary

There is only one dictionary in a normal Forth implementation, it is a data structure that can only grow in size (or at least it can in this implementation) and holds all of the defined words.

  • The stack

When we referring to a stack, or the stack, we refer to the variable stack unless otherwise stated (such as the return stack). The variable, or the stack, holds the result of recent operations such as addition or subtraction.

  • The return stack

Forth implementations are two (or more) stack machines. The second stack is the return stack which holds the usual function call return values as well as temporary variables.

  • Defined Words

A defined word is one that is not implement directly by the interpreter but has been create with the ':' word. It can be an immediate word, but does not have to be.

  • Compile mode

In this mode we compile words unless those words are immediate words, if the are then we immediately execute them.

  • Command mode

In this mode, regardless of whether we are in command or compile mode we execute words or push them on to the stack.

  • A block.

A Forth block is primitive way of managing persistent storage and this version of block interface is more primitive than most. A block is a contiguous range of bytes, usually 1024 of them as in this instance, and they can be written or read from disk. Flushing of dirty blocks is not performed in this implementation and must be done 'manually'.

Porting this interpreter

The interpreter code is written in C99, and is written to be portable, however porting it to embedded platforms that lack a C standard library (which is most of them) would mean replacing the most of the C standard library functions used, and implementing a new I/O mechanism for reading, printing and block storage.

The interpreter has been tested on the following platforms:

  • Linux x86-64 with,
    • Debian Jessie (8.x)
    • GCC version 4.9.2
  • Windows 7 x86-64 (not recently)
  • Linux ARM 32-bit Little Endian (not recently)
  • OSX Sierra 10.12.1 (tested by Rikard Lang).

And the different virtual machine word size options (32 and 64 bit machine words) have been tested. There is no reason it should not also work on 16-bit platforms.

libforth is also available as a Linux Kernel Module, on a branch of libforth, see This is module is very experimental, and it is quite possible that it will make your system unstable.

Standards compliance

This Forth interpreter is in no way compliant with any of the standards relating to Forth, such as ANS Forth, or previous Forth standardization efforts. However attempts to support words and behavior typical of these standards are made (although there is still a lot to do).

Some important deviations are:

  • ok

'ok' is not printed after a successful command execution , this is for two reasons, firstly because of limitations in the implementation, and secondly there is no reason for cluttering up the output window with this. The implementation should be silent by default.

  • Control structures

Various control structures in Forth behave slightly differently than in normal Forths. This should be fixed, given time.

forth.core magic

The Forth interpreters core files can be detected by the file utility by creating a magic file.

Save the following file on Unix systems to /etc/magic:

# libforth core files start with 0xFF '4' 'T' 'H'
0  string \xFF\x34\x54\x48 libforth core file
>4 byte   2                16-bit
>4 byte   4                32-bit
>4 byte   8                64-bit
## File version, version 4 is current as of 2017/03/08
>5 byte   x                version=[%d]
>5 byte   <4               ancient 
>5 byte   4                current
>5 byte   >4               futuristic
## Endianess test
>6 byte   0                big-endian
>6 byte   1                little-endian
>6 byte   >1               INVALID-ENDIANESS
## Size is stored as the base-2 logarithm of the size
>7 byte   x                size=[2^%d]
## Extra tests could be added, such as whether the core file is still valid

Coding Standards

The coding standards used for both the C and Forth code deviate from what is normal, the Forth code mostly so. Nonetheless they should be abided by regardless of what one may think of them. The C code is also written in a literate style (although the documentation is generated from it, rather than the code being generated from the documentation), this imposes extra requirements on comments.

C Coding Standards

The libforth Forth interpreter is primarily a library first and should be thought of as such, it is possible that multiple instances of the Forth interpreter are active at the same time, as such absolutely no global state can be used within the Forth library. Be very careful in what functions you export, by default all functions should be declared as static.

Global state can be used in the [main.c][] file which contains a wrapper around libforth. This wrapper is used to make the forth executable, and there is only ever one instance of the interpreter in use at a time.

The library is also written in C99 and made to be portable. Any non-portable code should be placed in [main.c][] and should be optionally compiled only if the correct macro for that functionality is defined. The macros _WIN32 and __unix__ are currently used to test for the operating system in use for example.

The program doxygen is used to extract and built the API documentation for libforth. At minimum all functions and macros that are exported by the library should be documented using doxygen style markup.

All functions exported by libforth.c should be tested, and should be prefixed with "forth_". Unit tests against libforth.c belong in unit.c. The same portability guidelines that apply to libforth.c also apply to unit.c.

Try to use assertions and checks on code liberally, even if the check is expensive to do. Expensive checks should be disabled with by defining the NDEBUG macro, like assert is.

The Linux kernel coding standards are to be roughly followed. Line lengths should be limited to 80 characters in length (this is much more important given documentation is generated from the source).

The main exception to the rule on code indentation is the forth_run function, which mostly consists of a large switch statement that implements the Forth virtual machine.

Bellow is a cut down version of the switch statement:

switch(w) { /* switches on instruction */
/* ...More instructions... */
case DEFINE:
	m[STATE] = 1; /* compile mode */
	if(forth_get_word(o, o->s) < 0)
		goto end;
	compile(o, RUN, (char*)o->s, true);
/* ...More instructions... */
case ADD:     f = *S-- + f;                   break;
case AND:     f = *S-- & f;                   break;
case OR:      f = *S-- | f;                   break;
case XOR:     f = *S-- ^ f;                   break;
case INV:     f = ~f;                         break;
/* ...More instructions... */

Of note is that simple instructions like ADD and OR take up a single line, having worked on the virtual machine for a while this aids in readability for me. The DEFINE instruction is a lot longer and so is split up into multiple lines.

A AWK script, specifically GAWK, is used to turn the C code into a single PDF document, by first generating markdown from it. The script, called convert, is simple. The script by default indents any C code which in markdown means it is treated as a code block, to turn the indentation on or off the commands "/**" turns if off and the command "**/" turns it back on, these obviously double as both C comment blocks, but also make doxygen look for tags within the comments. The script only looks for comment commands at the beginning of a line.

The following C program:

## A title using markdown syntax
This comment will not be indented in the resulting mark down
file and can be used to describe any of the following code.
int main(void) 
	printf("Hello, World\n"); /* a normal C comment */
	return 0;

Will be turned into:

## A title using markdown syntax
This comment will not be indented in the resulting mark down
file and can be used to describe any of the following code.
	int main(void) 
		printf("Hello, World\n"); /* a normal C comment */
		return 0;

The comment blocks are used to explain the code in detail.

Forth Coding Standards

The way code is formatted is idiosyncratic for Forth, the main difference is that tabs are used to indent code. Spaces and new lines are used to breakup longer word definitions, for short definitions only the entire definition should take up only two lines, as so:

: foo ( n -- n : add bar to 'n')
	bar + ;

All functions exposed to the user should be unit tested (which currently is not the case, but that is not an excuse not to test them). Unit tests go in the file unit.fth.

Sections of related code should be enclosed with decoration that makes it obvious that the code belongs together, like so:

( ============ The Foo Library ============ )
( The Foo library implements the standard 
words 'foo', 'bar' and 'foobar' )
( ... )
: foo ( n -- n )
	bar + ;
( ... )
( ============ The Foo Library ============ )

The standard way of making a word immediate in Forth is to place the word immediate after a word definition. This is still possible to do in libforth, however the preferred method is to place it just after the word being defined:

: example immediate ; ( preferred way )
: example ; immediate ( traditional way )

I prefer the first way of making a word immediate as it implies that it is part of the word definition and not something that acts on the word.


As mentioned in the standards compliance section, this Forth does things in a non-standard way. Apart from that:

  • Passing invalid pointers to instructions like open-file or system can cause undefined behavior (your program will most likely crash). There is no simple way to handle this (apart from not doing it).
  • The core interpreter does not currently make use of the throw and catch mechanism when handling certain errors (like division by zero), in effect there are two error handlers. These mechanisms need unifying.


  • A few environment variables could be used to specify start up files for the interpreter and user specific startup files.
  • Signal handling should be improved.
  • Error handling could be improved - the latest word definition should be erased if an error occurs before the terminating ';'. And trap handling should be done in pure forth, instead of as a hybrid which is currently is.
  • A series simple Unix utilities, like in could be made in Forth, this would demonstrate that this Forth is usable and practical. Tools like "cat", "tr", "grep", and various other text formatting tools could be done in pure C, others like "kill", "mount" or "ls" would require operating system dependent code to be added, this would need to be contained somehow, so the core remains pure C.
  • Allow the main memory to be resized. A Forth primitive MORE-CORE could be added.
  • Move the stack space so it is before the dictionary, this would mean we could write to the end of memory without worrying about hitting the stack.
  • Make a debian package for the Forth interpreter.
  • For a Forth only related "To-Do" list see the end of the file forth.fth.
  • A compiler for the virtual machine itself should be made, as a separate program. This could be used to make a more advanced read-evaluate loop.
  • Core files are currently not portable across machines of different words sizes or endianess, which needs addressing, or ignoring conclusively.
  • Use either liballocs or libffi for adding foreign function interfaces to programs, see:
  • The Forth could be made to be more standards compliant (especially where it comes to I/O and addressing (but also things like CASE statements).
  • A document (part of this document) for the coding standards used in this document needs to be created and added here. Things that will need documenting are:
  • How comments should look like (literate and doxygen comments)
  • Tabs vs Spaces (use tabs)
  • Levels of indentation
  • Coding standards for Forth (stack comments, indentation, etcetera).
  • And much more.
  • Improve error handling by adding 'throw' and 'catch' to the virtual machine as both C functions callable within the confines of 'libforth.c' and as virtual machine instructions.
  • To fix the problem with a mismatching between C addresses and Forth addresses which currently exists all modes of addressing should be made to be relative, or fixed up by the forth_init. Currently access to memory outside of the Forth address space can only be done by copying memory to and from the C and the Forth addresses spaces with words that accept real addresses (like memory-copy).

Virtual Machine To-Do points

The virtual machine needs changing, in someways drastically, to accommodate the following behaviors:

  1. Error handling should be purely Forth based.

Currently error handling and recovery is a weird mishmash between C and Forth, more of the error handling should be moved to the Forth environment using the normal THROW/CATCH mechanisms (which should be turned into virtual machine instructions and C functions, which can then be called to make a throw occur.

  1. The virtual machine should use character based addressing. Currently it uses cell based addressing, which causes all kinds of confusion.

  2. Words currently not defined need to be hidden, until the terminating ';'.


The [main.c][] file depends, optionally, on various submodules. These submodules need improving.


  • libline needs porting to Windows
  • The API needs improving so there is more control on whether or not raw mode is turned on or off, whether a terminal is being read from or not, ...

Experimental single binary

The build system can be used to build a single binary which contains the contents of "forth.fth", with the command "make libforth". This is a multistage build process that is experimental, so should not be relied upon. It involves a bootstrapping process using the first forth executable to build a second.

The process works like this:

  1. The executable "forth" is made.

    make forth


gcc -std=c99 main.c unit.c libforth.c -o forth
  1. The executable is used to generate a code file containing the contents of "forth.fth" is made.

    ./forth -s forth.core

  2. A forth word, core2c is used to generate a C file called "core.gen.c":

    ./forth -l forth.core -e 'c" forth.core" c" core.gen.c" core2c'

  3. The forth program is recompiled with an extra define, which means that initialization of a minimal forth environment is replaced with the core file we just made:

    gcc -DUSE_BUILT_IN_CORE -std=c99 main.c unit.c libforth.c core.gen.c -o libforth

The new executable, libforth, behaves the same as forth but with a built in core file.


  • The compilation should result in a small executable, and when statically linked against musl under Linux (x86-84), the stripped executable is around 50kb in size.
  • It is quite possible to make Forth programs that corrupt memory that they should, this is not a design flaw in this interpreter but more part of the Forth philosophy. If you want memory safety (and most of the time you should) you should use a different language, or implementation.
<style type="text/css">body{margin:40px auto;max-width:850px;line-height:1.6;font-size:16px;color:#444;padding:0 10px}h1,h2,h3{line-height:1.2}</style>


libforth: A small Forth interpreter that can be used as a library written in c99





No packages published