- foth
- Features
- Installation
- Embedded Usage
- Anti-Features
- Implementation Approach
- Implementation Overview
- Part 1 - Minimal initial-implementation.
- Part 2 - Hard-coded recursive word definitions.
- Part 3 - Allow defining minimal words via the REPL.
- Part 4 - Allow defining improved words via the REPL.
- Part 5 - Allow executing loops via
do
/loop
. - Part 6 - Allow conditional execution via
if
/then
. - Part 7 - Added minimal support for strings.
- Final Revision - Idiomatic Go, test-cases, and many new words
- BUGS
- loops - zero expected-iterations actually runs once
- See Also
- Github Setup
A simple implementation of a FORTH-like language, hence foth which is close to forth.
If you're new to FORTH then the wikipedia page is a good starting point, and there are more good reads online such as:
- Forth in 7 easy steps
- Just ignore any mention of the return-stack!
- Starting FORTH
- A complete book, but the navigation of this site is non-obvious.
In brief FORTH is a stack-based language, which uses Reverse Polish notation. The basic thing in Forth is the "word", which is a named data item, subroutine, or operator. Programming consists largely of defining new words, which are stored in a so-called "dictionary", in terms of existing ones. Iteratively building up a local DSL suited to your particular task.
This repository was created by following the brief tutorial posted within the following Hacker News thread, designed to demonstrate how you could implement something like FORTH, in a series of simple steps:
The comment-thread shows example-code and pseudo-code in C, of course this repository is written in Go.
The end-result of this work is a simple scripting-language which you could easily embed within your golang application, allowing users to write simple FORTH-like scripts. We implement the kind of features a FORTH-user would expect:
- Comments between
(
and)
are ignored, as expected.- Single-line comments
\
to the end of the line are also supported.
- Single-line comments
- Support for floating-point numbers (anything that will fit inside a
float64
). - Reverse-Polish mathematical operations.
- Including support for
abs
,min
,max
, etc.
- Including support for
- Support for printing the top-most stack element (
.
, orprint
). - Support for outputting ASCII characters (
emit
). - Support for outputting strings (
." Hello, World "
).- Some additional string-support for counting lengths, etc.
- Support for basic stack operations (
clearstack
,drop
,dup
,over
,swap
,.s
) - Support for loops, via
do
/loop
. - Support for conditional-execution, via
if
,else
, andthen
. - Support for declaring variables with
variable
, and getting/setting their values with@
and!
respectively. - Execute files specified on the command-line.
- If no arguments are supplied run a simple REPL instead.
- A standard library is loaded, from the present directory, if it is present.
- See what we load by default in foth/foth.4th.
- The use of recursive definitions, for example:
: factorial recursive dup 1 > if dup 1 - factorial * then ;
You can find binary releases of the final-version upon the project release page, but if you prefer you can install from source easily.
Either run this to download and install the binary:
$ go get github.com/skx/foth/foth@v0.5.0
Or clone this repository, and build the executable like so:
cd foth
go build .
./foth
The executable will try to load foth.4th from the current-directory, so you'll want to fetch that too. But otherwise it should work as you'd expect - the startup-file defines several useful words, so running without it is a little annoying but it isn't impossible.
Although this is a minimal interpreter it can be embedded within a Golang host-application, allowing users to write scripts to control it.
As an example of this I put together a simple demo:
This embeds the interpreter within an application, and defines some new words to allow the user to create graphics - in the style of turtle.
The obvious omission from this implementation is support for strings in the general case (string support is pretty limited to calling strlen, and printing strings which are constant and "inline").
We also lack the meta-programming facilities that FORTH users would expect, in a FORTH system it is possible to implement new control-flow systems, for example, by working with words and the control-flow directly. Instead in this system these things are unavailable, and the implementation of IF/DO/LOOP/ELSE/THEN are handled in the golang-code in a way users cannot modify.
Basically we ignore the common FORTH-approach of using a return-stack, and implementing a VM with "cells". Instead we just emulate the behaviour of the more advanced words:
- So we implement
if
ordo
/loop
in a hard-coded fashion.- That means we can't allow a user to define
while
, or similar. - But otherwise our language is flexible enough to allow real work to be done with it.
- That means we can't allow a user to define
The code evolves through a series of simple steps, contained in the comment-thread, ultimately ending with a final revision which is actually useful, usable, and pretty flexible.
While it would certainly be possible to further improve the implementation I'm going to declare this project as "almost complete" for my own tastes:
- I'll make minor changes, as they occur to me.
- Comments, test-cases, and similar are fair game.
- Outright crashes will be resolved, if I spot any.
- But no major new features will be added.
If you wanted to extend things further then there are some obvious things to work upon:
- Adding more of the "standard" FORTH-words.
- For example we're missing
pow
, etc.
- For example we're missing
- Enhanced the string-support, to allow an input/read from the user, and other primitives.
- strcat, strstr, and similar C-like operations would be useful.
- Simplify the conditional/loop handling.
- Both of these probably involve using a proper return-stack.
- This would have the side-effect of allowing new control-flow primitives to be added.
- As well as more meta-programming.
Pull-requests adding additional functionality will be accepted with thanks.
Each subdirectory within this repository gets a bit further down the comment-chain.
In terms of implementation two files are largely unchanged in each example:
stack.go
, which contains a simple stack offloat64
numbers.main.go
, contains a simple REPL/driver.- The final few examples will also allow loading a startup-file, if present.
Each example builds upon the previous ones, with a pair of implementation files that change:
builtins.go
contains the forth-words implemented in golang.eval.go
is the workhorse which implements to FORTH-like interpreter.- This allows executing existing words, and defining new ones.
Part one of the implementation only deals with hard-coded execution of "words". It only supports the basic mathematical operations, along with the ability to print the top-most entry of the stack:
cd part1
go build .
./part1
> 2 3 + 4 5 + * print
45.000000
^D
See part1/ for details.
Part two allows the definition of new words in terms of existing ones, which can even happen recursively.
We've added dup
to pop an item off the stack, and push it back twice, which
has the ultimate effect of duplicating it.
To demonstrate the self-definition there is the new function square
which
squares the number at the top of the stack.
cd part2
go build .
./part2
> 3 square .
9.000000
> 3 dup + .
6.000000
^D
See part2/ for details.
Part three allows the user to define their own words, right from within the REPL!
This means we've removed the square
implementation, because you can add your own:
cd part3
go build .
./part3
> : square dup * ;
> : cube dup square * ;
> 3 cube .
27.000000
> 25 square .
625.000000
^D
See part3/ for details.
NOTE: We don't support using numbers in definitions, yet. That will come in part4!
Part four allows the user to define their own words, including the use of numbers, from within the REPL. Here the magic is handling the input of numbers when in "compiling mode".
To support this we switched our Words
array from int
to float64
, specifically to ensure that we could continue to support floating-point numbers.
cd part4
go build .
./part4
> : add1 1 + ;
> -100 add1 .
-99.000000
> 4 add1 .
5.000000
^D
See part4/ for details.
This part adds do
and loop
, allowing simple loops, and emit
which outputs the ASCII character stored in the topmost stack-entry.
Sample usage would look like this:
> : cr 10 emit ;
> : star 42 emit ;
> : stars 0 do star loop cr ;
> 4 stars
****
> 5 stars
*****
> 1 stars
*
> 10 stars
**********
^D
Here we've defined two new words cr
to print a return, and star
to output a single star.
We then defined the stars
word to use a loop to print the given number of stars.
(Note that the character *
has the ASCII code 42).
do
and loop
are pretty basic, allowing only loops to be handled which increment by one each iteration. You cannot use the standard i
token to get the current index, instead you can see them on the stack:
- Top-most entry is the current index.
- Second entry is the limit.
So to write out numbers you could try something like this, using dup
to duplicate the current offset within the loop:
> : l 10 0 do dup . loop ;
> l
0.000000
1.000000
2.000000
..
8.000000
9.000000
> : nums 10 0 do dup 48 + emit loop ;
> nums
0123456789>
See part5/ for details.
This update adds a lot of new primitives to our dictionary of predefined words:
drop
- Removes an item from the stack.swap
- Swaps the top-most two stack-items.words
- Outputs a list of all defined words.<
,<=
,=
(==
as a synonym),>
,>=
- Remove two items from the stack, and compare them appropriately.
- If the condition is true push
1
onto the stack, otherwise0
.
- The biggest feature here is the support for using
if
&then
, which allow conditional actions to be carried out.- (These are why we added the comparison operations.)
In addition to these new primitives the driver, main.go
, was updated to load and evaluate foth.4th on-startup if it is present.
Sample usage:
cd part6
go build .
./part6
> : hot 72 emit 111 emit 116 emit 10 emit ;
> : cold 67 emit 111 emit 108 emit 100 emit 10 emit ;
> : test_hot 0 > if hot then ;
> : test_cold 0 <= if cold then ;
> : test dup test_hot test_cold ;
> 10 test
Hot
> 0 test
Cold
> -1 test
Cold
> 10 test_hot
Hot
> 10 test_cold
> -1 test_cold
Cold
^D
See part6/ for the code.
NOTE: The if
handler allows:
: foo $COND IF word1 [word2 .. wordN] then [more_word1 more_word2 ..] ;
This means if the condition is true then we run word1
, word2
.. and otherwise we skip them, and continue running after the then
statement. Specifically note there is no support for else
. That is why we call the test_host
and test_cold
words in our test
definition. Each word tests separately.
As an example:
> : foo 0 > if star star then star star cr ;
If the test-passes, because you give a positive number, you'll see FOUR stars. if it fails you just get TWO:
> 2 foo
****
> 1 foo
****
> 0 foo
**
> -1 foo
**
This is because the code is synonymous with the following C-code:
if ( x > 0 ) {
printf("*");
printf("*");
}
printf("*");
printf("*");
printf("\n");
I found this page useful, it also documents invert
which I added for completeness:
This update adds a basic level of support for strings.
- When we see a string we store it in an array of strings.
- We then push the offset of the new string entry onto the stack.
- This allows it to be referenced and used.
- Three new words are added:
strings
Return the number of strings we've seen/stored.strlen
show the length of the string at the given address.strprn
print the string at the given address.
Sample usage:
cd part7
go build .
./part7
> : steve "steve" ;
> steve strlen .
5
> steve strprn .
steve
^D
See part7/ for the code.
The final version, stored beneath foth/, is pretty similar to the previous part from an end-user point of view, however there have been a lot of changes behind the scenes:
- We've added near 100% test-coverage.
- We've added a simple lexer to tokenize our input.
- This was required to allow us to ignore comments, and handle string literals.
- Merely splitting input-strings at whitespace characters would have made either of those impossible to handle correctly.
- The
if
handling has been updated to support anelse
-branch, the general form is now:$COND IF word1 [ .. wordN ] else alt_word1 [.. altN] then [more_word1 more_word2 ..]
- It is now possible to use
if
,else
,then
,do
, andloop
outside word-definitions.- i.e. Immediately in the REPL.
do
/loop
loops can be nested.- And the new words
i
andm
used to return the current index and maximum index, respectively.
- And the new words
- There were many new words defined in the go-core:
.s
to show the stack-contents.clearstack
to clear the stack.debug
to change the debug-flag.debug?
to reveal the status.dump
dumps the compiled form of the given word.- You can view the definitions of all available words this:
#words 0 do i dump loop
#words
to return the number of defined words.- Variables can be declared, by name, with
variable
, and the value of the variable can be set/retrieved with@
and!
respectively.- See this demonstrated in the standard library
- There were some new words defined in the standard library
- e.g.
abs
,even?
,negate
,odd?
,
- e.g.
- Removed all calls to
os.Exit()
- We now return
error
objects where appropriate, allowing the caller to detect problems.
- We now return
- It is now possible to redefine existing words.
- Execute any files specified on the command line.
- If no files are specified run the REPL.
- We've added support for recursive definitions, in #16 for example allowing:
: factorial recursive dup 1 > if dup 1 - factorial * then ;
See foth/ for the implementation.
A brief list of known-issues:
The handling of loops isn't correct when there should be zero-iterations:
> : star 42 emit ;
> : stars 0 do star loop 10 emit ;
> 3 stars
***
> 1 stars
*
> 0 stars
*
^D
NOTE: In gforth
the result of 0 0 do ... loop
is actually an infinite loop, which is perhaps worse!
In our stars
definition we handle this case by explicitly testing the loop
value before we proceed, only running the loop if the value is non-zero.
This repository was put together after experimenting with a scripting language, an evaluation engine, putting together a TCL-like scripting language, writing a BASIC interpreter and creating yet another lisp.
I've also played around with a couple of compilers which might be interesting to refer to:
- Brainfuck compiler:
- A math-compiler:
This repository is configured to run tests upon every commit, and when pull-requests are created/updated. The testing is carried out via .github/run-tests.sh which is used by the github-action-tester action.