The latter doesn't even mention FORTH, and describes some very archaic CPU architectures, but I found it fascinating because it builds things from the ground up.
Loeliger's 'Threaded Interpretive Languages' jumpstarted my career in the late 1970s: I built a networked water management system based on their code, which was my first big project, earning me £1,500. Note that there's a bug in their code (I no longer remember exactly where) so getting it off the ground was tricky.
In particular learning about threaded interpreters, sub-routine interpreters, etc is very eye opening. That and really internalizing that everything, even code, is really just numbers.
technically, "header only libraries" can be exceptions to C code not being in header files. See STB as an example https://github.com/nothings/stb. The advantage theoretically is that you can #include this library and use its code and types from just one file, its a decent model IMHO, but it can be jarring to someone unfamiliar with header only libraries.
They have plenty of downsides and only one very minor advantage: You need to copy only a single file instead of two into your source. I am still puzzled why anybody could think this is a good idea...
Your experiment is certainly a good tool to grok stack mechanics, but that is only one element of what makes a Forth what it is. You're missing out on other crucial ingredients: colon definition and immediateness.
You've implemented a stack calculator, which misses the big picture: the stack is an implementation detail that enables a lot of the nifty stuff, but Forth is a "braided language" - the whole is more than the sum of the parts. Each part of it is just implementation, and the implementation is kept simple, but all of it needs to be there to make it completely usable. The distinction is important and non-trivial to the design of the interpreter.
To get there, please implement some of the metaprogramming words found in one of the standardized Forths(and if you aren't sure which one, use an earlier spec like Forth83 since the core wordset is smaller, so you run into the "hard stuff" faster).
Forth used in anger towards an application actually tends to resemble Fortran: procedural idioms flinging around lots of named temporary variables. The stack, being just implementation, doesn't give any assistance for everyday programming, unless you extend the system to do so. This is a point on which modern concatenative languages have diverged and tried to add some rigor into it.
Interesting. I am currently in the process of writing an interpreter for a similar stack based language. I already wrote a compiler for this language to x86 assembly that can be compiled to an ELF. The language is used as an intermediate language for a C compiler that I am writing. It is maybe less Forth-like than your language. For more information see: https://github.com/FransFaase/MES-replacement For the interpreter have a look at: stack_c_interpreter.c
It's already pretty efficient but I'm working on it to make it even more efficient so I can use it as some sort of primitive fragment shader for an art project. This Forth variant is intended to execute Forth Haikus, as defined by the Forth Salon website.
I made my own lisp. It used to have a recursive interpreter but I recently finished converting it into a register/stack machine that's essentially a modified version of the Explicit Control Evaluator from Structure and Interpretation of Computer Programs.
Now I'm thinking about converting it into a CESK machine which I believe is a proper stack machine. Modifying the SICP machine has proven difficult due to the loose stack discipline. I'm told register machines are faster but keeping track of all those ad hoc pushes and pops is turning into a nightmare.
> What features would you add to make it more useful?
The neat part about all these machines is how the stack itself turns into some sort of code. You push some state indicating what you want the machine to do next. Then you set it up so that it evaluates a value. When it's done, that value just flows into whatever the next computation step is.
Making a copy of the stack and wrapping it into a callable that just plugs that value in is how first class continuations are implemented.
If you're interested in learning more about how FORTH works I, I can recommend two very old books.
Starting FORTH https://archive.org/details/LeoBrodieStartingFORTHIntroducti...
Threaded Interpretive Languages https://archive.org/details/R.G.LoeligerThreadedInterpretive...
The latter doesn't even mention FORTH, and describes some very archaic CPU architectures, but I found it fascinating because it builds things from the ground up.
Loeliger's 'Threaded Interpretive Languages' jumpstarted my career in the late 1970s: I built a networked water management system based on their code, which was my first big project, earning me £1,500. Note that there's a bug in their code (I no longer remember exactly where) so getting it off the ground was tricky.
In particular learning about threaded interpreters, sub-routine interpreters, etc is very eye opening. That and really internalizing that everything, even code, is really just numbers.
1. You should add a URL when you you create a post on HN. You can indent code two spaces on HN, eg:
2. Your readme mentions a repl but I don't see it in the source code.3. I'm not an expert in C but I thought header files shouldn't have code in them. The code should be in a .c file
4. Maybe move the code from USAGE into its own .c file.
technically, "header only libraries" can be exceptions to C code not being in header files. See STB as an example https://github.com/nothings/stb. The advantage theoretically is that you can #include this library and use its code and types from just one file, its a decent model IMHO, but it can be jarring to someone unfamiliar with header only libraries.
They have plenty of downsides and only one very minor advantage: You need to copy only a single file instead of two into your source. I am still puzzled why anybody could think this is a good idea...
And you also can name your single file to be included .c instead of "hiding the truth".
You can include a .c file just fine.
Your experiment is certainly a good tool to grok stack mechanics, but that is only one element of what makes a Forth what it is. You're missing out on other crucial ingredients: colon definition and immediateness.
I wrote a series of articles that can help in that kind of discovery: http://tumbleforth.hardcoded.net/
You've implemented a stack calculator, which misses the big picture: the stack is an implementation detail that enables a lot of the nifty stuff, but Forth is a "braided language" - the whole is more than the sum of the parts. Each part of it is just implementation, and the implementation is kept simple, but all of it needs to be there to make it completely usable. The distinction is important and non-trivial to the design of the interpreter.
To get there, please implement some of the metaprogramming words found in one of the standardized Forths(and if you aren't sure which one, use an earlier spec like Forth83 since the core wordset is smaller, so you run into the "hard stuff" faster).
Forth used in anger towards an application actually tends to resemble Fortran: procedural idioms flinging around lots of named temporary variables. The stack, being just implementation, doesn't give any assistance for everyday programming, unless you extend the system to do so. This is a point on which modern concatenative languages have diverged and tried to add some rigor into it.
Interesting. I am currently in the process of writing an interpreter for a similar stack based language. I already wrote a compiler for this language to x86 assembly that can be compiled to an ELF. The language is used as an intermediate language for a C compiler that I am writing. It is maybe less Forth-like than your language. For more information see: https://github.com/FransFaase/MES-replacement For the interpreter have a look at: stack_c_interpreter.c
At WHY2025, I gave a talk about the reasons why am working on this. See: https://www.youtube.com/watch?v=akzyyO5wvm0
Another thread on small forth interpreters from just 15 days ago:
https://news.ycombinator.com/item?id=45039301
Forth can be beautifully and efficiently implemented in portable c++ using the using continuation passing style via the clang musttail attribute.
Have a look at Tails (not my project):
[1] https://github.com/snej/tails
I recently wrote one, in C, using tail calls to implement dispatch with CPS: https://tia.mat.br/posts/2025/08/30/forth-haiku.html
It's already pretty efficient but I'm working on it to make it even more efficient so I can use it as some sort of primitive fragment shader for an art project. This Forth variant is intended to execute Forth Haikus, as defined by the Forth Salon website.
> Has anyone else built something similar?
I made my own lisp. It used to have a recursive interpreter but I recently finished converting it into a register/stack machine that's essentially a modified version of the Explicit Control Evaluator from Structure and Interpretation of Computer Programs.
https://github.com/lone-lang/lone/blob/master/source/lone/li...
Now I'm thinking about converting it into a CESK machine which I believe is a proper stack machine. Modifying the SICP machine has proven difficult due to the loose stack discipline. I'm told register machines are faster but keeping track of all those ad hoc pushes and pops is turning into a nightmare.
> What features would you add to make it more useful?
The neat part about all these machines is how the stack itself turns into some sort of code. You push some state indicating what you want the machine to do next. Then you set it up so that it evaluates a value. When it's done, that value just flows into whatever the next computation step is.
Making a copy of the stack and wrapping it into a callable that just plugs that value in is how first class continuations are implemented.