an opinionated guide to scheme implementations -- wingolog

Hark, the beloved programing language! But hark, also: for Scheme is a language of many implementations. If Scheme were a countryside, it would have its cosmopolitan cities, its hipster dives, its blue-collar factory towns, quaint villages, western movie sets full of façades and no buildings, shacks in the woods, and single-occupancy rent-by-the-hour slums. It's a confusing and delightful place, but you need a guide.

And so it is that there is a steady rivulet of folks coming into the #scheme IRC channel asking what implementation to use. Mostly the regulars are tired of the question, but when they do answer it's in very guarded terms -- because the topic is a bit touchy, and a bit tribal. There's a subtle tension between partisans of the different implementations, but people are polite enough to avoid controversial, categorical statements. Unfortunately the end result of this is usually silence in the channel.

Well I'm tired of it! All that silence is boring, and so I thought I'd give you my opinionated guide to Scheme implementations. This is necessarily subjective and colored, as I co-maintain the Guile implementation and do all of my hacking in Guile. But hey, to each folk their stroke, right? What I say might still be valuable to you.

So without further introduction, here are seven recommendations for seven use cases.

The embedded Scheme

So you have a big project written in C or C++ and you want to embed a Scheme interpreter, to allow you to extend this project from the inside in a higher-level, dynamic language. This is the sort of thing where Lua really shines. So what Scheme to use?

Use Chibi Scheme! It's small and from all accounts works great for static linking. Include a copy of it in your project, and save yourself the headache of adding an external dependency to something that changes.

Embedding Scheme is a great option if you need to ship an application on Windows, iOS, or Android. The other possibility is shipping a native binary created by a Scheme that compiles natively, so...

The Scheme that can make a binary executable

Sometimes what you want to do is just ship an executable, and not rely on the user to install a runtime for a particular Scheme system. Here you should use Chicken or Gambit. Both of these Schemes have a compiler that produces C, which it then passes on to your system's C compiler to generate an executable.

Another possibility is to use Racket, and use its raco exe and raco dist commands to package together a program along with a dynamically-linked runtime into a distributable directory.

The dynamically linked Scheme

What if you want to extend a project written in C, C++, or the like, but you want to dynamically link to a Scheme provided with the user's system instead of statically linking? Here I recommend Guile. Its C API and ABI are stable, parallel-installable, well-documented, and binary packages are available on all Free Software systems. It's LGPL, so it doesn't impose any licensing requirements on your C program, or on the Scheme that you write.

The Scheme with an integrated development environment

What if you really like IDEs? For great justice, download you a Racket! It has a great debugger, an excellent Scheme editor, on-line help, and it's easy to run on all popular platforms. Incidentally, I would say that Racket is probably the most newbie-friendly Scheme there is, something that stems from its long association with education in US high schools and undergraduate programs. But don't let the pedagogical side of things fool you into thinking it is underpowered: besides its usefulness for language research, Racket includes a package manager with lots of real-world modules, and lots of people swear by it for getting real work done.

The Scheme for SICP

Many people come by #scheme asking which Scheme they should use for following along with Structure and Interpretation of Computer Programs. SICP was originally written for MIT Scheme, but these days it's easier to use Neil Van Dyke's SICP mode for Racket. It has nice support for SICP's "picture langauge". So do that!

The server-side Scheme

So you write the intertubes, do you? Well here you have a plethora of options. While I write all of my server-side software in Guile, and I think it's a fine implementation for this purpose, let me recommend Chicken for your consideration. It has loads of modules ("eggs", they call them) for all kinds of tasks -- AWS integration, great HTTP support, zeromq, excellent support for systems-level programming, and they have a good community of server-side hackers.

The Scheme for interactive development with Emacs

Install Guile! All right, this point is a bit of an advertisement, but it's my blog so that's OK. So the thing you need to do is install Paredit and Geiser. That page I linked to gives you the procedure. From there you can build up your program incrementally, starting with debugging the null program. It's a nice experience.

You and your Scheme

These recommendations are starting points. Most of these systems are multi-purpose: you can use Gambit as an interpreter for source code, Racket for server-side programing, or Chicken in Emacs via SLIME. And there are a panoply of other fine Schemes: Chez, Gauche, Kawa, and the list goes on and on (and even farther beyond that). The important thing in the beginning is to make a good starting choice, invest some time with your choice, and then (possibly) change implementations if needed.

To choose an implementation is to choose a tribe. Since Scheme is so minimal, you begin to rely on extensions that are only present in your implementation, and so through code you bind yourself to a world of code, people, and practice, loosely bound to the rest of the Scheme world through a fictional first-person-plural. This is OK! Going deep into a relationship with an implementation is the only way to do great work. The looser ties to the rest of the Scheme world in the form of the standards, the literature, the IRC channel, and the mailing lists provide refreshing conversation among fellow travellers, not marching orders for a phalanx.

So don't fear implementation lock-in -- treat your implementation's facilities as strengths, until you have more experience and know more about how your needs are served by your Scheme.

Finally...

Just do it! You wouldn't think that this point needs elaboration, but I see so many people floundering and blathering without coding, so it bears repeating. Some people get this strange nostalgic "Scheme is an ideal that I cannot reach" mental block that is completely unproductive, not to mention lame. If you're thinking about using Scheme, do it! Install your Scheme, run your REPL, and start typing stuff at it. And try to write as much of your program in Scheme as possible. Prefer extension over embedding. Use an existing implementation rather than writing your own. Get good at using your REPL (or the IDE, in the case of Racket). Read your implementation's manual. Work with other people's code so you get an idea of what kind of code experienced Schemers write. Read the source code of your implementation.

See you in #scheme, and happy hacking!