Looking through the docs it seems a very nice lisp. Clean, simple, yet covers all the basics without a lot of legacy warts. let with less parens and limitless if branches are very nice. Definitely one of the nicer looking JS lisps I've seen.
Yes, the performance is by far one of the best parts of Lumen.
Think of it this way. It generates exactly the code you write. It's nearly a one-to-one translation between the input forms and output code. But that's why it's fast.
If you want to see just how fast, try `time LUMEN_HOST=luajit make -B`
On my box, LuaJIT is able to recompile the entirety of Lumen and run all the tests in 1.3 seconds flat.
As for Urn, I'd be happy to answer specific comparison questions. I've been putting together a little intro guide to Lumen, but for now the best way to jump in is to study test.l and learn by example: https://github.com/sctb/lumen/blob/master/test.l
EDIT: By the way, Lumen runs on Torch too. If you want to do some hardcore machine learning, it's as easy as installing torch and running `LUMEN_HOST=th bin/lumen`.
Try evaluating `(require 'torch)` and `(require 'nn)`, then dive into the Torch guides. You can bring Lumen's full macro system to bear on any ML problem, which greatly simplifies the boilerplate generation involved with standardizing datasets and such.
It's been few days since this post, don't know if you're still checking the thread.
I'm really a beginner at lisp, so perhaps my questions won't make much sense? I'd like to have Lua host application (in Love2d) with lisp-like runtime. The lisp-like code is expected to be changed often during the runtime by modifications of structure and constants in AST, so compile-time macros are not useful to me.
On the other hand, the lisp-like code would not be written from scratch in text form. It would be built by executing code-transformation functions that would always construct a syntactically valid AST. For example, let's say there is a function for drawing an arc of single color. Now user wants to draw a rainbow. I would offer a function that could be executed to wrap existing operation in a loop. In rainbow example the loop iteration would also modify a vertical displacement and color hue shift.
My two questions - is lisp (and Lumen specifically) a good choice for self-modifying live coding runtime? And is generated Lua source in Lumen project suitable for embedding Lumen into Lua host application?
The idea came from wondering what a Lisp-like language might look like if you based it on 'everything is a table' instead of 'everything is a list'. A few old comments where this came up:
The core idea was to explore Lua's table-ness in a Lisp, with JavaScript along for the ride for browser support. I think it was an interesting experiment—interesting enough to possibly continue exploring in a more “serious” implementation, if someone were so inclined.
As a Common Lisp user, I wish more of these one-off lisps would try to be compatible with Common Lisp or Scheme.
Both CL and Scheme have pretty good library ecosystems, though realistically not as rich as more popular languages. Coming out with a brand new, incompatible, Lisp that has an even smaller community isn't very helpful, IMO. Why not at least take advantage of the existing libraries and community?
We'll implement this ourselves. Let's dive right in. Start off by cloning Lumen.
$ git clone https://github.com/sctb/lumen
$ cd lumen
$ git checkout aedf2fd2c209bfc7926e0d4eae4becd8a647fb80
$ vim main.l
(Commit aedf2fd is just the latest commit at the time of this writing. I make it explicit here so that anyone can follow along in the future.)
Now that main.l is open, we begin by defining a global function `read-file` that simply returns a file as a string:
(define-global read-file (path)
((get system 'read-file) path))
If you run `make && bin/lumen`, you'll see the function is now defined when Lumen starts:
> read-file
function
> (read-file "README.md")
"Lumen\n=\nLumen is a very small, self-hosted Lisp for Lua and JavaScript. It provides a flexible compilation environment with an extensible reader, macros, and extensible special forms, but otherwise attempts..."
Now we define `read-from-file`, which reads Lumen source code and returns it as a form that can be evaluated:
Pretty good! We're already doing some basic compiler-type stuff. That's Lumen's power: It's a flexible compiler. (IMO one of the finest in the world due to its simplicity.)
Now, what can we do with these forms? Well, we can expand them:
> (print (compile (expand (read-from-file "reader.l"))))
local delimiters = {["("] = true, [")"] = true, [";"] = true, ["\n"] = true, ["\r"] = true}
local whitespace = {["\r"] = true, [" "] = true, ["\n"] = true, ["\t"] = true}
local function stream(str, more)
return {more = more, pos = 0, len = _35(str), string = str}
end
local function peek_char(s)
local ____id9 = s
local __pos6 = ____id9.pos
local __len3 = ____id9.len
local __string3 = ____id9.string
if __pos6 < __len3 then
return char(__string3, __pos6)
end
end
local function read_char(s)
local __c21 = peek_char(s)
if __c21 then
s.pos = s.pos + 1
return __c21
end
end
...
And we can switch languages:
> (set target 'js)
"js"
> (print (compile (expand (read-from-file "reader.l"))))
var delimiters = {"(": true, ")": true, ";": true, "\n": true, "\r": true};
var whitespace = {"\r": true, " ": true, "\n": true, "\t": true};
var stream = function (str, more) {
return {more: more, pos: 0, len: _35(str), string: str};
};
var peek_char = function (s) {
var ____id12 = s;
var __pos8 = ____id12.pos;
var __len4 = ____id12.len;
var __string4 = ____id12.string;
if (__pos8 < __len4) {
return char(__string4, __pos8);
}
};
var read_char = function (s) {
var __c28 = peek_char(s);
if (__c28) {
s.pos = s.pos + 1;
return __c28;
}
};
...
English translation: "When compiling, read the forms from runtime.l, macros.l, and main.l, join them together, then compile the result."
Let's compile this file and see what happens:
$ bin/lumen -c lumen.l
environment = {{}}
target = "lua"
function nil63(x)
return x == nil
end
function is63(x)
return not nil63(x)
end
function no(x)
return nil63(x) or x == false
end
function yes(x)
return not no(x)
end
function either(x, y)
if is63(x) then
return x
else
return y
end
end
...
Why does it generate Lua? Because on my system, Lumen happens to default to running on LuaJIT, and the host language is the default. If Lua wasn't installed on your system, you'd be seeing JS instead.
To get a specific language, pass the `-t` parameter:
$ bin/lumen -c lumen.l -t js
environment = [{}];
target = "js";
nil63 = function (x) {
return x === undefined || x === null;
};
is63 = function (x) {
return ! nil63(x);
};
no = function (x) {
return nil63(x) || x === false;
};
yes = function (x) {
return ! no(x);
};
either = function (x, y) {
if (is63(x)) {
return x;
} else {
return y;
}
};
...
This has been a short tour of why Lumen has fascinated me for the last three years, and hope you find its mysteries delightful. It is only through abstraction that we can bend computers to our will, and Lumen is a decisive step forward.
Very nice project - thanks for sharing!
How would you say this compares to maru, which does a similar thing?
Maru uses gcc rather than Node.js or lua, and compiles it's own (x86) binaries. It focusses on as minimal syntax as possible, to expose new methods of composition:
http://piumarta.com/software/maru/
If I've got this correct, lumen still relies on having the lua/Node/luaJIT runtime underneath?
Regarding Maru the Lisp, the goals are similar-ish to Lumen’s. Lumen started life as an experiment: “what would a table-based Lisp look like?” Maru on the other hand is a (very fine!) traditional Lisp. It has cons cells, for example, and a GC. Lumen elides these features by virtue of running on a host that already implements them.
You’ve got me excited to try writing a Lumen to C compiler now...
Actually, that highlights one interesting difference. It’s possible to implement a C backend for Lumen relatively easily. It would just spit out C, which is fed straight to gcc. The problem is self-hosting. It’s easy to emit some C functions. It’s hard to emit code which is capable of compiling itself, i.e. implementing Lumen’s runtime and compiler systems.
I would say that Lumen’s power is largely thanks to its brevity. In fact it’s so small that I almost overlooked it, that first day many years ago. It didn’t seem like something so small could be production-quality.
I’ll keep studying Maru and perhaps report back with more thoughts.
The vid made me smile :)
Yes, maru was conceived as part of the STEPS program, to provide a minimal runtime base for an OS. As such it's focus was on being able to produce DSLs for a higher level of abstraction, and hence needed to provide the ability to extend composition (http://piumarta.com/freeco11/freeco11-piumarta-oecm.pdf).
Personally I've been interested in replacing the GC part of maru with a structural method of managing memory (http://concurrency.ch/Content/publications/Blaeser_ETH_Diss_...), this would require extending the language with the Composita primitives, but not really sure how this'll go yet.
The Lumen-C compiler sounds good! Yes the tricky part is the self compilation, I'm wondering if you'll just end up with a maru-like system if you follow this down?
We did a search when we picked the name, and found https://github.com/xopxe/Lumen. I emailed Jorge to ask if he was ok with us using the name for a Lisp-to-Lua project and he sent a delightful reply that began "It's impossible to avoid conflicts with such a cool name", and pointed out that there were other projects called lumen that he discovered when he published on Github. Who knows how far back the lumens go.
By now finding great names that aren't in use yet (let alone not in use in a similar context) is almost impossible. Fast fwd 100 years from now & we'll have way more artifacts with names. I think we'll be fine.
