[asguy]

Programming Forth caused the most massive change in my programming mindset (the LSD of programming languages). Even more so than Lisp. Thinking of the rapid rise of the base language to the problem domain (i.e. it's a DSL construction kit) opened my eyes to how to design useful and extensible APIs.

https://thinking-forth.sourceforge.net is worth reading, even if you have no desire to ever program any Forth.

[nequo]

What are your main lessons from Forth for API design? Is it something you could summarize?

[asguy]

There have been some great responses that cover the gist of it, but to add a couple more mundane specifics:

- Build up the thinnest abstraction that lets you express a solution in the problem domain.

- Allow the API user to express their intent, and not be focused on mechanics they don't need to understand to get the job done.

- Focus on getting the interfaces right before the implementation is perfect.

- A clean interface will allow you to make drastic changes to the implementation without the API consumer becoming aware.

[diffxx]

Not the OP, but for me one of the main lessons from Forth is that the boundary between high and low level code can be as thin as you make it. With Forth, you can define high level words using low level primitive words. You then weave these low level words together at a high level. After you have written a correct solution at a high level, bottlenecks can be optimized out by introducing new low level words. It's like you are solving a problem by designing an instruction set specifically for that problem. But unlike a hardware ISA, you can add new instructions tailor made for the specific problem you are presently solving.

Paul Graham has described lisp as a tool for writing fast programs fast (that's the gist at least). IIRC, he clarifies that each fast is a phase. Lisp allows you to write a fast, high level, prototype that might have poor runtime performance and then you refine the prototype so that the compiler can generate fast machine code.

Forth is similar but feels closer to the machine than Lisp because of the stack based threading model. In Forth, you often don't need manual memory management _or_ garbage collection and you can easily extend the system with new low level words. But it does require you to think differently about your program design so that it fits the stack based vm model.

Historically, Forth has been implemented in assembly but you can write a Forth that targets any host. The truly mind expanding thing is that you, an individual, can write a Forth compiler and you can also write it in such a way that it has multiple targets. For example, you can generate native, jvm and js targets from the same underlying source. Usually this can be done by translating the forth generated AST to source code for another high level language. This allows you to write a fast high level language quickly without getting bogged down in writing, say, fast x86 codegen or a garbage collector. To get started, you just target the host language with the best high level properties that you need (which will be problem and context specific). I don't enjoy writing c, but I have no problem translating to c if I need to for performance or for syscall access.

To illustrate the practical power of this. Suppose that you have an important subcomponent of your system that was written in, say, Node.js. It turns out that this is a critical bottleneck that cannot easily be optimized in javascript. You don't want to rewrite the whole system, but you do want to rewrite that component and have it seamlessly interoperate with the existing system. You could write a small Forth DSL for that subcomponent. This Forth will target javascript initially. You translate the subcomponent into Forth and reuse all of the existing tests (that were presumably also written in javascript). Then you rewrite the tests in Forth so that the entire subcomponent is now written in Forth. Now you write a new backend that translates to say, c or rust with node bindings. You can run the native implementation against a native implementation of the test suite since they're both written in high level Forth at this point. Then you can flip the switch between the native or javascript implementations and be confident that both implementations are identical because they have the same high level description.

Once you start seeing things this way though, you start realizing that you can write Forth style code in any language (and the reverse is also true). Forth is as much about a process for solving programming problems as it is a specific, concrete language. This is also why there is the old adage "once you've seen one Forth, you've seen one Forth."

[saulpw]

They just said it was like the LSD of programming languages. Your question is kind of like "What are your main takeaways from LSD? Can you summarize?" which realllly misses the point.