Comment by jart | Hacker News Reader

Comment by jart | Hacker News Reader

Parent Full thread

Author here. You know it when you see it. LISP is a tool that people want to use. People write production software using it. For example, this website (Hacker News) was written in LISP. Languages like BrainFuck are usually only called programming languages if the word "esoteric" is attached. There's a real consensus of opinion around this this being a generally held view.

There should ideally be a more compelling explanation of why Brainfuck is an esoteric toy and LISP isn't, than me simply saying "but that's what people believe". I considered going into more depth on this subject. I found a C to Brainfuck compiler written in OCaml and used it to compile a Brainfuck metacircular evaluator into Brainfuck. It ended up being 88,439 bytes of code, and it sadly didn't work, because the compiler isn't very polished.

Let's compare ratios. Brainfuck is 99 bytes, whereas Brainfuck written in Brainfuck is 88439 bytes. That's 893x larger. On the other hand, LISP is 436 bytes of systems code and LISP written in LISP is 1370 bytes of LISP. That's 3.14x larger. It's cool that the ratio between layers of simulation with LISP is pretty close to pi. It helps explain why LISP is viewed by human beings as being such a powerful tool, since it lets you accomplish more impact with less effort, without having the castles you build turn into a leaning tower of pisa. Since after all, the purpose of tools since prometheus has been to give people more power.

Would you consider binary lambda calculus [1] [2] a real language? The 650 byte interpreter written in obfuscated C also sports garbage collection and might fit in a bootsector when translated to x86 code. Its metacircular interpreter is only 29 bytes (232 bits), and looks somewhat readable with some syntactic sugar [4] (analogous to how x86 looks more readable as assembler source).

[1] https://www.ioccc.org/2012/tromp/hint.html

[2] https://tromp.github.io/cl/Binary_lambda_calculus.html

[3] https://github.com/tromp/AIT/blob/master/int.lam

But does either include a JIT? Can they JIT themselves?

Counter argument: ever seen the proof for simple addition in English? The verbosity and/or opacity of the definition of a system isn't a great proxy for value or ease of use. Principia Mathematica uses over 1000 pages to reach the proof that 1+1=2, formally defining and explaining everything needed along the way, establishing fundamentals and primitive types and so on.

Brainfuck isn't a great interpreter language. It is a great perspective language. If you can break your tasks down sufficiently that you can express your program in brainfuck, you can take that understanding back to a more performant and expressive language. Doing Rosetta code exercises in esoteric languages forces your brain to model programs with more nuance and detail. If raw machine level performance was the only thing that mattered, we'd be writing everything in assembly. Different languages can have different utility, though, and while brainfuck is less useful as software, it is an excellent tool, not just a toy.

Isn't the distinction as simple as "does the language let you define anything and use recursion?", where I'm going to be a bit handwavy with "define anything" but I think we all get what I mean.

You can't name functions in Brainfuck or recursively call them. You can do so in LISP and Forth. That is also the entire reason Brainfuck explodes in size: it lacks the "semiotic" capabilities required to compress source code (and therefore, structure and refactor code, which is what humans need).

To me it's missing the point to describe that as Brainfuck versus Lisp. It just turns out that lambda functions and linked lists are higher level abstractions than arrays and pointers. I'd like to see some more variations on that concept and what the size is if using another esolang that has lambda functions.

Reply on news.ycombinator.com