recognizers 2.1.0

A collection of recognizers


Matthias Trute Version 2.1.0 - 2017-09-14

This package is a collection of recognizer examples. To play with them a simple implementation is provided. It works unchanged on (old) gforth's, swiftforth and VFX. They are not (yet) able to compile code or are otherwise integrated into the underlying system.

Many examples contain test units that illustrate what the recognizer does.

More information about recognizers can be found at and

Running the examples with e.g. gforth should not give any errors beside a possible redefinition of {

$ gforth rec-double-paren.4th 
redefined {  Gforth 0.7.2, Copyright (C) 1995-2008 Free Software Foundation, Inc.
Gforth comes with ABSOLUTELY NO WARRANTY; for details type `license'
Type `bye' to exit

The code examples use rev 4 of the recognizer RFD. Package version 1.x use the rev 3 of it.


This file in combination with Stack.4th gives a playground for making experiments with recognizers. It has been tested with gforth versions that currently come with the linux distributions, MPE's vfxlin and Forth Inc's Swiftforth for Linux.


This file contains four recognizers. They deal with numbers in various formats and use >NUMBER for the actual number conversion.

rec-char ( addr len -- n rectype-num | rectype-null )

Forth 2012 defined the 'c' syntax for single characters. That means that char c or [char] c can be replaced with 'c'.

rec-snum ( addr len -- n rectype-num | rectype-null )

This recognizer handles single cell numbers. It checks whether a strings consists of digits only and that the value range of the number fits into a single cell. The sign and base prefix characters work as expected.

rec-dnum ( addr len -- d rectype-dnum | rectype-null )

This recognizer handles double cell numbers. It accepts the standard double number format: digits with a trailing dot in addition to the base prefix and sign character.

These three recognizers are combined into the rec-num recognizer that can be used to handle all number formats in one call.

As a special extension, the 0x prefix is used to switch to hex too. It requires the string wordset (COMPARE) to be present.


This file contains a dictionary lookup recognizer. It uses FIND for the actual work thus uses the search order if present.


Another dictionary lookup recognizer. This one does not depend on FIND and searches the standard forth wordlist only. It returns name tokens instead of execution tokens.

It requires TRAVERSE-WORDLIST and the NAME>x words from Forth 2012 to work. From the future, quotations are used too.


Uses " as the string delimiter. Everything between two " within SOURCE is a string. It can replace the forth command S" completly. Instead of S" foo" use "foo". The space after S" is no longer needed, it is now part of the string. S" foo" and " foo" differ with the leading space in the latter.

The string lives as long as SOURCE is unchanged! More sophisticated implementations may use a string stack. Compilation to the dictionary is done by SLITERAL. Postponing throws an exception simply because that is not specified in the standard.

A typical use is

"hello world" type
hello world
: test "hello world" type ; test
hello world

that prints the string resp. compiles it and prints it at runtime.


This is not really a recognizer but an API wrapper for the often found not-found hook. It discards any input and calls a deferred word not-found. This word is expected to never return properly by e.g. printing an error message and throwing an exception.

With this recognizer as the last one in the stack, programs that use the not-found hook can easily adapted.


Implements the multiline (( )) comment block.

The (( switches the system recognizer stack to one that searches only one otherwise hidden wordlist. This wordlist contains only words that are allowed to be executed in comments. For now only )) that switches back to normal operation.

Since the recognizer stack switch is unaffected from REFILL operations, multiline comments work too.

This recognizer needs a system that has recognizers native support. It does not depend on SOURCE however. It re-uses the REFILL-PARSE-EVALUATE loop found in the text interpreter. Some tests are provided.


Simple formatting of source code in the spirit von Don Knuth' literate programming. You can combine the real source code and its documentation in one file that the standard forth interpreter can handle directly.

Actually testing this recognizer requires a Forth that supports recognizers already.

Author: Julian Fondren, 2014 License: probably public domain.


check and convert for the hh:mm:ss notation returning a double cell number for the number of seconds it represents.