Racket v5.3.2

HN uses Arc which is based on MzScheme/Racket, IIRC.

Well, I haven't used it in production, but I used it for a research project last semester. It was actually pretty good, and there are definitely some very nice features. The standard library was pretty good except it missed some functional features I wanted like immutable vectors (or even just functional operations on mutable vectors).

The language is extremely flexible, and you can take advantage of this to write some really elegant and maintainable code. It's one of the more productive languages I've used--of the languages I've used extensively, it's better than JavaScript, Java or Python.

So I liked it quite a bit over all. However, it doesn't quite clear the threshold where I would use the language voluntarily. I thought it had lackluster support for functional programming, especially from what is supposed to be a functional language. This was evident in the library (e.g. the lack of immutable vectors and the like) and also in the way certain pretty fundamental structures (like ports) worked.

Also, for a dynamically typed language, I thought it had too many different types. It was too easy to get mixed up between bytestrings and strings, for example. It was also not obvious when you should use or expect nil vs #f. That said, the features for creating your own types (mostly structs) were very good--I found they were very lightweight but still flexible enough to be useful.

I am actually a big fan of the syntax. Not just the s-expressions (although, with paredit, they are pretty awesome) but also the reader macros. That said, I'm also rather partial to infix operators, which are obviously not present.

So over all, it's a great language. However, I would still choose Haskell or OCaml over it for virtually any project. The main exception would be projects very heavy on metaprogramming, where Racket really shines.

> Who's using this in production?

Here are two examples that have been presented at CUFP. The Starfire Optical Range[1, 2] uses it in production, I think as part of the control software for their telescopes. Naughty Dog, a major video game developer, uses it for scripting their games[3, 4].

[1]: http://en.wikipedia.org/wiki/Starfire_Optical_Range [2]: http://cufp.org/archive/2006/abstracts.html#RichardCleis [3]: http://cufp.org/videos/functional-mzscheme-dsls-game-develop... [4]: http://www.slideshare.net/naughty_dog/statebased-scripting-i...

I use it to run Arc, which I use as my go-to general-purpose programming language. I use it to do math, Project Euler problems (not exactly the same thing), sometimes to control the computer [I sometimes run programs that click the mouse for me, and used to make it type as well], to generate repetitive code[1] for usually experimental purposes, and occasionally to make complicated projects like a sudoku solver, or a thing that reads and writes WAV files.

Racket is pretty good, is pretty fast. It can add the integers from 1 to 100 million in 407 milliseconds on my machine. If you need incredibly speedy numerical code, it is still possible to load machine code, perhaps from a C dynamic library, and run it. My main objection is that it doesn't have real-time garbage collection. We'll see about that.

[1] Don't get excited, I haven't written a compiler or a working assembler (yet). I mean things like this:

  (pbcopy:tostring:on x '(rax rbx rcx rdx rdi rsi rbp rsp r8 r9 r10 r11 r12 r13 r14 r15)
    (prn "        " "mov [rel teh_end + " (* index 8) "], " x))
  --> clipboard now contains the following:
        mov [rel teh_end + 0], rax
        mov [rel teh_end + 8], rbx
        mov [rel teh_end + 16], rcx
        mov [rel teh_end + 24], rdx
        [...]

I'm using Racket right now to develop a machine learning algorithm. The new math and plot libraries are absolutely fantastic. For the brave or foolhardy it makes a viable alternative to R, NumPy, or Julia. (I haven't used Julia. I have found NumPy impossible to install. I want to punch something every time I use the steaming mess that is R.) Racket is far more a real language than the alternatives in this space but it lacks their wide range of numeric libraries.

Racket is also a really interesting language to learn. Nothing else does meta-programming like Racket, or mixes typed and untyped languages (see Typed Racket). However the most fun I've had with Racket is writing machine code directly into memory and executing it straight from Racket.

I used to use Racket in production and prototyping (few years ago), as a "glue language" and "business-logic" for a web-app, for the parts that are not IO, memory, compute intensive, etc. My metrics for language choice are:

1. Simple to learn and remember (they tend to correspond)

2. Module System; Support for modularization of complexity

3. Simplicity of coding and maintaining simple modules

4. Simplicity of coding and maintaining complex modules

5. Available libraries and their quality

In comparison to (I can only code efficiently in) Java, Python, Javascript/CoffeeScript/Node, Common Lisp, Scheme, and Haskell

- Racket is best for 2. (especially if you add a bit of meta-data to a module) and 4.

- Racket is much better than Haskell, Java, and Common Lisp for all 5 metrics Except for Haskell, which happens to be very nice for certain things that benefit from its type and class system

- For 3., Racket is overly complex, especially with the macro facility and huge number of types and structures. Node/Javascript/Coffee is better by far; Python is a little better than Racket because of less language features

- Racket is weak in 5. - Racket has many libraries of dubious quality, and I've hit undocumented issues a few times; Only Java was worse; However glue/business-logic doesn't require many libraries I now use NodeJS because I've been dealing with simpler stuff.

"I Write Like"[1] runs on Racket on a VM with 256 MB RAM.

[1] http://iwl.me, source code: http://www.codingrobots.org/p/iwl/

I don't know if this counts as in production, but we use racket extensively on my team (we do predictive market simulations). I wish I could give you a deep technical reason why racket works where nothing else would do - but the truth is that I'm the one who picks the technology, and I like writing code in racket.

biggest problems thus far are 1) the lack of external libraries (exempli gratia, nothing like python's boto for talking to ec2), but you probably saw that one coming. and 2) the language is almost too big, with too many datatypes: without even mentioning macros (which racket does quite well), the class/object system, the functional and mutable datatypes, contracts, pattern matching, typed racket, et cetera, can keep you pretty busy. Myself and the others might not just know racket well enough yet, but it's easy for one person to write unreadable code without ever leaving standard language constructs.

But this is also the best thing about racket - the language is flexible enough that you can find a natural way to express whatever you need to. We've also been playing with distributed places (racket processes on remote machines) with some success, too.

This is exactly how I feel and a similar background, though I used Gambit and Bigloo more than Chicken.

I wonder how many people are in this situation.

Well, 'years ago' is a long time. You'll probably find out that Chicken has evolved a lot since then. As has the list of available eggs.

I'm not affiliated with them, but I know Naughty Dog, the developers behind the Uncharted series, Jak and Daxter, Crash Bandicoot, and the upcoming The Last of Us, have long used Lisp dialects in all their games. The game engine driving Uncharted uses a port of MzScheme for the majority of their development pipeline and game logic.

They've talked about how it improves their workflow for years. You can find some of their articles and presentations if you look for references to GOAL (game-oriented assembly language) and GOOL (game-oriented object language), their in-house Lisp dialects.