Show HN: AttaLambda: a language where types and data are made of untyped lambdas
44 points - last Monday at 4:21 PM
I made a programming language!
I call it AttaLambda.
The idea is this: a usable Lisp-shaped language where all the meaningful computation is done in untyped lambda calculus. Logic, arithmetic, data structures, control flow, even the types β all untyped lambdas. A small, explicit Racket layer sits at the boundary to handle the outside world, plus some macros for syntactic sugar.
This is its story:
A couple years ago, I wanted to play with untyped lambda calculus and go beyond where tutorials usually stop. They show booleans, numbers, arithmetic, maybe the Y-combinator β and then stop. I wanted them to keep going.
So I started a project called All The Lambdas. Using Racket set to lazy, I used only one Racket construct for actual computation β lambda β and built integers, rationals, lists, binary digit-list number encodings, search algorithms, and more.
Then I found Functional Programming Through Lambda Calculus by Greg Michaelson. In it, Michaelson sketches the bones of a language built in untyped lambda calculus, including a type system where typed objects are themselves pair functions containing a type tag and value.
I found that intriguing and implemented and extended the idea, still entirely with untyped lambdas. I don't have a background in programming language theory, so I was figuring it out as I went.
Then I stopped tinkering with it for a while.
Recently I came back and thought: why not turn this into a real usable language with the help of coding agents? I reused most of All The Lambdas as the foundation.
Thus AttaLambda was born.
Some additional details:
* Rat, its number type, uses binary digit-list encodings instead of Church numerals, so numbers scale with their number of binary digits rather than their value
* errors are lambda-encoded values, not Racket exceptions, and propagate through the language like ordinary data
* the Racket host only performs irreducibly external operations; even things like HTTP parsing, routing, and response construction stay in the pure lambda world
* recursion uses lambda-calculus recursion: no loops or true self-reference, just the Y-combinator underneath
* automated purity checks catch accidental cheating, like native computation leaking into the pure parts
* syntax like multi-argument lambdas, let, cond, and list is just macro sugar that reduces to unary lambdas and application
A couple code examples: (short of print, every single thing here reduces to unary untyped lambdas)
Factorial:
#lang attalambda
(rec factorial n =
(cond
((eq n 0) 1)
(else (mult n (factorial (sub n 1))))))
(print (factorial 10))
Which prints: 3628800
Or an exact harmonic sum: #lang attalambda
(print
(reduce add 0
(map (lambda (n)
(unwrap-ok (div 1 n)))
(range 1 8))))
Which prints exactly: 363/140
As far as I know, no programming language combines all these features: Michaelson-style type tags built from untyped lambdas, exact rationals backed by binary digit lists, errors as lambda values, and real-world programs where almost all computation stays inside the lambda core. None of those pieces are individually new, but I don't know of another language combining them this way.Download: https://github.com/kserrec/attalambda/releases/tag/v0.7.0
Code: https://github.com/kserrec/attalambda
Original All The Lambdas: https://github.com/kserrec/all_the_lambdas
Comments
I wrote a wee lambda calculus interpreter to test out the book forms. I did cheat a wee bit and added:
def <name> = <exp> ==> substitution of <exp> for <name> before evaluation
if <exp1> then <exp2> else <exp3> fi ==> ((<exp1> <exp2>) <exp3>)
to make things more readable.
I think a few folk have got as far as implementing lists and recursion, but you're the first dynamic type implementer I've come across.
I think lambda calculus is fun. My book is about functional programming, and so isn't much use for theory, for which I recommend Barendregdt's "The Lambda Calculus. Its Syntax and Semantics".
Best wishes
Greg Michaelson