Hacker News new | comments | show | ask | jobs | submit login
Lisp in about 200 lines of Ruby (github.com)
94 points by r11t 2691 days ago | hide | past | web | 34 comments | favorite

Lisp in about 100 lines of Python:


Please instruct your httpd to serve *.py files as "text/plain". Right now it's forcing a download. Win32 FF.

It's being served as text/x-python, which I think is appropriate. So it's not the server that's "forcing" a download, it's your client. I suspect you're using FireFox, which seems AFAICT to be b0rken in this regard.

A micro-Lisp compiler (to x86 executable) in about 100 lines of OMeta: http://www.vpri.org/pdf/m2009011_chns_mng.pdf

examples given which run about 70% of the speed of equivalent C code.

That's a nice one, albeit several simplifications are made in there (e.g. from a quick glance I could not find anything dealing with scopes [the prefixing with "_V_" for variables might probably hint at its non-existence]).

Nevertheless, impressive work from one of the most interesting projects around.

This makes me vocalize something Ive wondered about...

Whats the smallest subset of lisp which you need to implement in order to bootstrap the language? So, given this small subset, one can implement the rest of say scheme or common lisp or arc on top of.

I vaguely recall Sussman or Ebelson implementing car and cadr in terms of lambdas, for example.

It seems this should be quite modular, given so many implementations of lisp [ in javascript, PHP, etc now Ruby ]. So that new lisps might be easily brought up over this kernel.

The lambda calculus is Turing-complete, so the following language is sufficient to encode any computable function:

exp ::= var | (lambda (var) exp) | (exp exp)

Check out


To see how to build numbers, booleans, lists, conditionals and recursion out of pure lambdas.

Even cooler is that these tricks work in languages like JavaScript too:


Well, you can define pairs like this (as long as you've got lexical scope):

  (setq cons (lambda (x xs)
               (lambda (msg)
                 (if (eq msg 'car) x xs)))

  (setq car (lambda (xs) (xs 'car)))
  (setq cdr (lambda (xs) (xs 'cdr)))
That only uses 'setq, 'lambda, 'if, 'eq, and quoted symbols. That pretty much gives you the untyped lambda calculus, plus some side effects. Throw in the appropriate macros and you've got the core of Scheme and Arc, without any I/O or efficient data representations.

You can take this pretty far. Essentially, you just need 'lambda and application. You could use church numerals for numbers, monads for side effects, the y-combinator for recursion, etc.

Under which lisp implementation do those forms work?

The definitions for CAR and CDR are identical and they don't make much sense (set the symbol 'car to the value of a function that takes angument xs, which is a function, and applies that to the symbol 'car)

And the definition for CONS is equally opaque.

How would you use those functions, CONS, CAR and CDR?

Since you pretty much nailed the uses of church numerals, monads and the y-combinator, I know you know what you're talking about, it just that Lisp's semantics is confusing me (it has common lisp keyword SETQ but none of its behavior)

Oops; I meant to write those in CL, but I forgot about functions having their own namespace (been playing too long in a lisp-1). It should work in CL if you use FUNCALL.

If mquander's explanation isn't enough, you might look at it using terms from the OO world: CAR takes a pair object and calls its 'car method.

So you mean a meta-circular OOP implementation, where the hosting language already has car, cdr and cons. Alright.

I thought what you wrote was supposed to be a primitive implementation of the functions.

No, not really, unless you're running the code in a metacircular interpreter.

Imagine encoding the pair ('hello . 'world) into a function. You might say:

  (lambda (msg)
    (if (eq msg 'car) 'hello 'world))
You could get 'hello by applying that function to the 'car symbol, or 'world by applying it to anything else. Wrapping that in another function lets us generalise to any CAR/CDR.

  (lambda (car-value cdr-value)
    (lambda (msg)
      (if (eq msg 'car) car-value cdr-value)))

Those look correct to me. The function returned by cons is just a switch that holds two values in a closure, and interprets what you pass it as a signal to determine which to retrieve. You could just as easily rewrite it to pass it 0 or 1 instead of car and cdr.

If you evaluate (car (cons 1 2)) and (cdr (cons 1 2)) on paper using substitution, you'll see how it works.

Exactly. SICP shows some similar code, though I don't remember which part..


Pair-structured memory, a (very simple) recursive-descent parser, eval, apply, read, print, a few mathematical primitives, not too much else. Once you have the bare bones, _The Art of the Interpreter_ provides a good skeleton. Don't worry about a garbage collector yet - either use a language with one (Scheme, Python, Lua, Ruby, OCaml, etc.) or just let it leak memory for now. (Garbage collectors aren't deep magic either, but one thing at a time, you know?)

Implementing a toy Lisp is absolutely worthwhile. It shouldn't take too long* , and it will teach you several deep things. Don't worry about making it efficient until you have it together enough to be useful. (If you want a mini language that's mature, just use Lua.) This is about discovery. Many interpreters for mini-languages don't need to be efficient, anyway. Who writes 20,k-long files in a DSL?

* Though if you've never done an interpreter at all, a half-assed Forth is even simpler. Either way, implementing half-assed Lisps, Prologs, Js, etc. is a great way to feel out a language. Worrying about getting it efficient upfront can be a red herring.

According to pg, seven: (quote, atom, eq, car, cdr, cons, and cond)

Eighth paragraph: http://www.paulgraham.com/ilc03.html

Ya that should be the minimum required to bootstrap. I always thought the definitive minimum code for lisp was here:


Or at least from a historic perspective.

Quote as a predefined macro could be important though, in a Lisp, if you expect to be doing a lot of macro magic, and I don't see it here.

This is cool, though, and very readable; I know what all the code is doing with a quick glance. Cheers to the author.

Well, lambda calculus is Turing complete, so you could just stick with a call-by-value version of that. However, you'd want to add macros if you wanted to extend the syntax of the language. Since you mentioned car and cdr in lambdas, they'd be defined like this:

NIL = (lambda (x) (lambda (y) y))

CONS = (lambda (ar) (lambda (dr) (lambda (op) (op ar dr))))

CAR = (lambda (expr) (expr (lambda (ar dr) ar)))

CDR = (lambda (expr) (expr (lambda (ar dr) dr)))

Since nobody mentioned it, there is a very interesting paper among what are know the "lambda papers":


What you are looking for is answered in "The Art of the Interpreter, the Modularity Complex (Part Zero, One, and Two)", Steele and Sussman, MIT AI Lab Memo 453, May 1978.


More LISP details can be obtained by the book "LISP in Small Pieces" by Christian Queinnec (1996), and an excellent description of the eval/apply functions can be found in "The Architecture of Symbolic Computers" by Peter M. Kogge, 1990.

Btw: to satisfy my more or less latent pedantism: The second SICP author is called Abelson not Ebelson.

hehe, I had this open two tabs to the left of HN :)


Nice link.

Frustrating to have to login to ACM to read McCarthys - 'A micro-manual for LISP - not the whole truth'

Thankfully, found this hidden away at [pdf d/l link] : http://www.ee.ryerson.ca/~elf/pub/misc/micromanualLISP.pdf

thank you, I'd been looking for that one :)

I threw something like this together in python some time ago: http://bitbucket.org/statictype/code/src/tip/pylisp/

Not as elegant though.

I wonder if this says more about Ruby, or more about Lisp?

Does lisp not support negative numbers? :)

Fixed in 1.4.2... Thanks for pointing that out!

X in (some small number) of lines of Y

This is generally an indicator that language_power(X) > language_power(Y)

I think you have that the wrong way round.

Taking a specific example, "Lisp in 100 lines of Ruby" means that once you have Ruby, you have Lisp. That means that anything you can do in Lisp can be done in Ruby, and hence Ruby is at least as powerful. That means:

X in (some small number) of lines of Y


language_power(X) <= language_power(Y)

Exactly the opposite of what you said.

Greenspun's Law in action ;-)

Greenspun's Tenth Rule of Programming: "Any sufficiently complicated C or Fortran program contains an ad-hoc, informally-specified bug-ridden slow implementation of half of Common Lisp." http://philip.greenspun.com/research/

A upgrade to Ruby?

Thanks for that. It's been my motivational poster for a while now.

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | DMCA | Apply to YC | Contact