μLithp - a Lisp in 27 lines of Ruby
fogus.github.com
fogus.github.com
The implementation is as trivial as the concept is profound. The discussion should be about the lambda-calculus, not Lisp, which is a far more complex beast. And much less about Lisp dialects, Greenspun's, or the usual BS topics that invariably appear on any L-word thread.
Besides, we have better uses for you ;-)
You know what you might like, if you haven't seen it yet, is the Kernel programming language, a Scheme where everything, including macros, is first-class.
Once upon a time I wrote a lisp to lua compiler in 100 lines. https://github.com/meric/l2l/commit/a530f0133e002c3981937d21...
It's now over a thousand. https://github.com/meric/l2l/
However, whenever people post "Lisp in Ruby" stories, I always hope that it'll be a "Clojure in Ruby" implementation. I am surprised no-one has done it yet.
Optional static typing (a la Qi) is being implemented, too.
Clojure is immutable by default, which is a pretty big difference AIUI. It also has a pretty strong emphasis on concurrency, including a bunch of primitives for such. It uses STM.
A bunch of data structures are first class, like vectors, maps, and sets. There are literals for each of them, and you can use maps and sets as functions. You can also use keywords as functions against maps to e.g. retrieve the :name value from a list of maps:
user=> (map :name [{:name "foo"} {:name "bar"}])
("foo" "bar")
user=> (map #(* 2 (inc %)) [1 2 3])
(4 6 8)
Threading macros are somewhat closer to Haskell's syntactically straightforward function composition:
user=> (->> [1 2 3] ;; this'll be the last arg passed to each fn below
#_=> (map #(* 2 %))
#_=> (map inc))
(3 5 7)
Which data-structures are not first class "objects" in Scheme?
It may have been misleading to say that Clojure was immutable by default. It is immutable; it does not have set! or the moral equivalent. You can use concurrency primitives to update in place (sort of? AFAICT you're mutating the var itself, not the data it contains) but each primitive comes with its own semantics.
As far as first-class is concerned: in Clojure, maps, sets, and vectors are functions. They're functions of their data as well as being data, and there are literal representations for each.
Examples:
user=> ({:foo 1} :foo) ;; maps are functions
1
user=> (:foo {:foo 1}) ;; keywords are functions, too
1
user=> (filter :name [{:name "foo"} {:surname "baz"}]) ;; filter by key
({:name "foo"})
user=> ([\a \b \c] 2) ;; access vector by index
\c
user=> (#{1 2 3} 4) ;; test set membership
nil
user=> user=> (filter #{2 4 6} (range 1 10)) ;; filter by set membership
(2 4 6)
AFAICT there is not a well-understood definition of first-class outside of "first-class functions." If I had to nail it down, I'd probably include the notion of first-class as applied to functions, plus:
* literal representations (where applicable) * idiomatic, as demonstrated/enabled by the standard library * well-integrated/interop with other primitives
This is just off the top of my head. And sometimes this is relative to other constructs in the language, or fuzzier, like Haskell lists vs. maps vs. sets.
Why not just the same notion of "first-class" as functions? In other words, that the feature is actually represented by a run-time value that can be passed around freely and stored. This is the usual definition of "first-class" that I hear most people use in the programming languages world. Examples include first class control (continuations), first class references (boxes, mutable cons cells, etc.), first class environments, first class modules (see OCaml, units, etc.), first class labels, and so on.
In Scheme, for example, we could implement any data-structure as a closure, which accepts messages. There is no difficulty in writing such wrappers - it is just a closure which return another closure which accepts "messages", and following some protocol, returns, for example, another closures to be called for a certain action (generators, iterators, etc.)
We could teach the read function to recognize any kind of wired syntax we wish, and constructing appropriate data-structures with type-tags attached to them. But it will become a mess.
The real data-structures, however, its representation is very different thing. In Clojure, I suppose, it is based on built-in Java types and generic interfaces, such as Iterable or whatever it is. So, they are ordinary Java objects, without any magic in it.
In Scheme or CL it depends on the implementation, the choices made by developers. So, for example, Gambit-C and MIT Scheme are quite different in how they implement hash-tables or vectors. My guess is that, say Alegro CL and CMUCL are also very different, yet they all conform to some standards (CLtL2, ANSI).
Having very different implementations with different set of compromises is a strength.
So, in my opinion, there is absolutely nothing special in this syntactic constructions, moreover, it is not that difficult to implement them using closures and macroses. In arc.arc you could see how strings and tables were implemented.
Another issue is, should we add all this wired syntax to what we call Lisp? In my opinion doing this ruins Lisp and the result is some very different in a look-and-feel language. Calling it Lisp is, well, confusing, at least to me. Arc is a Lisp, no doubt. CL and Scheme are Lisps, for sure. Clojure is Java with lisp-like syntax, if you wish.)
The old rules says that there must be very heavy reasons to add any new symbol or a keyword into a language, and that the same things shall look the same, and different - differently. For me, personally, Arc is a proper approach, while Clojure is, well, a mess.)
You're right, there is nothing special about what Clojure does; everything could be implemented in Scheme just fine. The real advantage comes from the fact that these things are there "out of the box" so that libraries are written to use them.
There is a lot of inertia when it comes to libraries in a language. In some sense, you only get one chance to get it "right" before everything becomes interdependent and you can no longer change anything. Clojure's advantage is simply that it presented a chance to start over and get things right from the beginning.
A dumb example would be a Option/Maybe type in Java. You could write a good approximation. But none of the standard libraries would use it, let alone any user supplied libs. Lisps are different in some very important ways, but since this is in part a meatspace phenomenon, it's still a challenge. Somebody went so far as to call it The Lisp Curse[0].
Anyway, syntax is a matter of taste, though I think people focus too much on it one way or the other— people don't like s-expressions b/c parens, people don't like Clojure b/c brackets/braces/whatever.
So while I agree that Clojure syntax isn't a huge game-changer, it can and does improve the quality of life for some people, myself included. Built-in literals are a big deal. And there is a lot to be said for being able to use these as primitives on day 1 of learning Clojure, esp. as your first Lisp.
[0]: http://www.winestockwebdesign.com/Essays/Lisp_Curse.html.
Personally I dislike Clojure's syntax - Scheme's feels more pure, cleaner. It's a bit like PERL vs. Python, a matter of taste largely.
I would call it more like Ruby vs. Python — c'mon, perl? that's just mean — but yes, we're quibbling. :)
The meta-circular evaluator is a prime example of mutual recursion. It consists of two functions, eval and apply, which call each other in a circular fashion until a result is reached.
Taking 100 MB to startup may be a problem. But not exactly for a webserver.
Or, I run a few hundred servers at work, also on VMs. I want to create a process that needs to run on all of them in addition to what the server is actually used for. Maybe for monitoring, maybe for server management, etc. and I want it to have a REST API. Suddenly using 100MB matters to me.
They have: http://rouge.io/ https://github.com/rouge-lang/rouge
Lisp begins with something like this:
(defmacro when (test &body forms)
`(cond (,test nil ,@forms)))
Lisp without lists at its core is nothing but another clumsy language. Look at Clojure - what a mess.
(defmacro when
"Evaluates test. If logical true, evaluates body in an implicit do."
{:added "1.0"}
[test & body]
(list 'if test (cons 'do body)))
l.eval [:label, :second, [:quote, [:lambda, [:x], [:car, [:cdr, :x]]]]]
l.eval [:second, [:quote, [1, 2, 3]]]
There is a classic function:
(define (filter fn l)
(cond ((null? l) '())
((fn (car l)) (cons (car l) (filter fn (cdr l))))
(else (filter fn (cdr l)))))
For the second piece of code - are't semicolons and comas somehow redundant?
Moreover, what is the point of writing something this way?
(let [x (f (first s))]
(if (nil? x)
(keep f (rest s))
(cons x (keep f (rest s)))))
Could you please support this assertion?
A minimal Lisp (the bare minimal) does not require sexprs though.
Really? Why not? I realize that McCarthy's original spec used recursion equations, but that notation is more complicated than s-expressions.
Above was HN submission from early last year, Lisp in 32 lines of Ruby, which was the original blog post of this code by Fogus.
module.exports = function() {
var atom = function(val) { return val instanceof Array ? false : true; };
var env = {
label: function(sexpr, senv) { senv[sexpr[1]] = evaluate( sexpr[2] ); },
quote: function(sexpr, senv) { return sexpr[1]; },
"==": function(sexpr, senv) { return evaluate(sexpr[1], senv) == evaluate(sexpr[2], senv); },
head: function(sexpr, senv) { return evaluate(sexpr[1], senv)[0]; },
tail: function(sexpr, senv) { return evaluate(sexpr[1], senv).slice(1); },
conc: function(sexpr, senv) { return [evaluate(sexpr[1])].concat(evaluate(sexpr[2])); },
"if": function(sexpr, senv) { return evaluate(sexpr[1], senv) ? evaluate(sexpr[2], senv) : evaluate(sexpr[3], senv); },
atom: function(sexpr, senv) { return atom(sexpr[1]); },
lambda: function(sexpr, senv) {
return function(lexpr, lenv) {
for(var i=0; i<sexpr[1].length; ++i) lenv[sexpr[1][i]] = evaluate(lexpr[i+1], lenv);
return evaluate(sexpr[2], lenv);
};
}
};
var evaluate = function(sexpr, senv) {
senv = senv || env;
if( atom(sexpr) ) return senv[sexpr] !== undefined ? senv[sexpr] : sexpr;
else return senv[sexpr[0]](sexpr, senv);
};
this.evaluate = evaluate;
};
var notlisp = require("./notlisp.js");
var l = new notlisp();
l.evaluate( ["label", "second", ["lambda", ["x"], ["head", ["tail", "x"]]]] );
l.evaluate( ["second", ["quote", [1, 2, 3]] ] );Any obvious choice? (If not, I imagine I'd like to learn the most widely-used)
Learn yourself some Clojure for great good.
Then I can get busy with that ahead of time (I've been sweating a Python release)!
Thank you.
Yep!
Happy hacking: learning some Lisp variant is a great day/month/year/lifetime in any programmer's life.
(book & code) - http://billhails.net/Book/ | Previous HN post - http://news.ycombinator.com/item?id=1747132
20 years later: for Ruby users, situation improves! :)
:)
Lisp is very simple and very powerful. Those who don't learn it are doomed to reinvent it, badly (cough XSLT cough).
It's a bit mind-bending, but from this simple foundation comes the power to easily create and assimilate ever higher-level abstractions (which is what programming is really all about). Parts of a program can be used as a template for creating other parts as needed. You can assemble these parts and shape the language into something very specific for the problem you are solving.
Other languages come part-way, but don't fully embrace this concept (code=data) because it has a cost: it seems difficult to learn because it requires a change in thinking; it seems difficult to write (and read) because you're directly encoding a data structure and it all looks a bit 'samey'. Where other languages have 'shapes' in their code that give the trained eye an indication of what the code is expressing (such as for...do loops and indentation), lisp code has one shape: a tree [1].
I don't know if I've reached beyond your understanding. I haven't meant to, but the takeaway should be this: lisp programs are self-aware and can operate on themselves because a lisp program is just data like any other. This makes it easy for the interpreter/compiler (hence 'lisp in x lines of code'), but more difficult for the programmer (at first). As with all things, practice and familiarity overcome these initial difficulties. It is worth it.
I know a bit about programming but it only comes down to PHP or jQuery variables, functions, and loops. Lisp looks very interesting and I had already heard about it (through pg obviously). It might just be the language that will fascinate me if I ever took programming more seriously.
This is a really good explanation, and I just want to add a small bit to it, having had to use Lisp as a primary language quite recently due to some AI projects in cognitive modeling. One thing that I've found personally is that by understanding Lisp, you can essentially gain a deep understanding of every __other__ programming language.
If I had the chance to do my entire Computer Science career over again, I would have learned Lisp as my first language, and then compared every other subsequent language against it. Because of the deep symmetry between code and data, it is one of the easiest languages for implementing and re-implementing programming language features on top of.
You've asked a profound and important question, not so much about Lisp as about systems in general. A system is simple when it doesn't have many pieces. But if the pieces it starts with are very general, and the system provides good ways to put them together, then you can build complex things out of a tiny initial set.
Think of a classic Lego set. There are only a few "primitives" (different kinds of Lego piece). And there is an easy and standard way to put them together. Starting from these simple initial conditions you can construct very complex structures. Lego is simple and powerful at the same time.
How do you build a complex Lego structure? You start by putting a few pieces together to form a cluster. And then you make a second cluster and you join the two clusters together to make a third, and so on. The critical property here, the reason why Lego stays simple all the way, is that the "operation" you use to join two clusters is exactly the same operation you use to join two individual pieces – you interlock their knobby bits. In this sense your clusters are still "Lego pieces", just custom ones that didn't ship with the initial set: they still have knobby bits suitable for joining up with any other piece. This quality is sometimes called "regularity", meaning that the initial properties of the system are preserved as you put things together into ever more complex forms.
Suppose that weren't the case and that each time you reached a certain level of complexity you had to learn a new technique for building further. That would make Lego much less simple-and-powerful. You could still build complex things, but it would be harder and more complicated. And there would probably be a bunch of things that, while you technically could build them, it would be so hard and complicated to do that nobody would bother.
Programming languages also consist of a set of initial pieces and ways of putting them together. But most languages don't have the quality of regularity that Lego has. Their initial pieces aren't universal enough to get away with having only a few, so they need many more of them, and that means they're not simple, and that means they're not simple-and-powerful. There are, however, a few programming languages that do have this regularity. Lisp is one of them. Smalltalk, Forth, and APL are others.
The really interesting thing is that this kind of simple-and-powerful system enables you to do qualitatively different things as complexity grows — not because other languages can't do them, but because they make it too much trouble to bother. For example, because Lisp programs are Lisp pieces in the same sense that Lego structures are Lego pieces, it's easy to write Lisp programs that take other Lisp programs as their inputs and do meaningful things with them. That is a powerful technique, and because it's so easy, Lisp programmers exploit it heavily, more than is practical in most other languages, and that means they can do more with less.
This quality of simple-and-powerful is in my view very important and underutilized for managing software complexity. We don't yet understand it all that well, because the programming languages that achieved dominance to date don't have it.
http://ocw.mit.edu/courses/electrical-engineering-and-comput...
Ignore the hairdos, it's from the 80s, the information itself ages better than the fashion.
I have been playing with a Clojure, a Lisp dialect, for six or so months (it seems so much longer), but this was still really interesting to me.
He could probably set up his publish templates to accomplish the same formatting.
(huge org mode nerd, checking in)
--- EDIT ---
And what do you know, Eric Schulte already implemented this for org mode! It's checked in under contrib.
Instructions: http://eschulte.me/org-docco/org-docco.html
Org source: http://orgmode.org/w/?p=org-mode.git;a=blob_plain;f=contrib/...
Fallacy. You haven't eliminated the possibility that line count is important, but also that too few lines is also bad. It's widely understood that line count, even though it's a hazy and exploitable metric, is indeed important. It's also widely understood that being too terse is as bad as being too verbose. Time to brush up on some old-school CS curriculum books. (Mythical Man-Month? I'm not sure if the above is discussed there, but it's a good place to start.)
Now, wasn't it more enjoyable the way I put it first?
On the otherhand, I've been interested in doing a lisp in low memory environments (uC's, etc). I've done a dialect in C using a semi-space garbage collector. But I'm curious if anyone's done any work on lisps in resource constrained environments.
Some times it's cool to do things just because you can, maybe he was bored or he always wanted to implement a lisp dialect.
It might not be that useful, but i'm sure that the autor had a blast doing it, otherwise he wouldn't have done it.