lbjScheme
An embeddable Scheme interpreter in pure Java
This will become my most complete Scheme implementation so far, striving for 100% R5RS compliance when done.
There will be (at least) three evaluators:
-
A very simple interpreter that even expands macros on the fly as they are encountered (done)
-
An analyzing interpreter that performs an analyzer step before interpreting; macros are expanded during analysis (done)
-
A compiler that uses the same analyzer, but instead of interpreting the AST will emit bytecode for a VM that is then used to run the program (todo)
The first two will remain quite simple and incomplete, more as a test harness for the builtins, while the compiler will support first class continuations. Let's see how it works out. In the future there might even be a compiler that generates Java or .NET bytecode, or C sources or whatever. Focus is on keeping this thing embeddable though, as there are lots of great Scheme-to-something compilers out there already.
Everything in this repo is published under the BSD license unless stated otherwise in the respective file.
How to build
Make sure you have the following installed on your system:
- A working JDK 1.7 or higher
- Apache Maven
Download the .zip file, extract and run "mvn package" in the extracted folder. Once Maven has finished building and testing lbjScheme, you will find the generated .jar file in the target folder.
FAQ
Q: Which of the "advanced features" of Scheme are supported?
A: The full numeric tower, tail calls and lexical scoping work. Exactness of numbers is preserved as long as feasible. The 1.0 release will have first-class continuations and the ability to freely redefine builtins without others breaking. Macros will be CL style in the 1.0 release, i.e. (defmacro name (params) ...) instead of Scheme (hygienic) macros. Maybe in a later version I'll support Scheme macros too. There are a few builtins that are not part of the standard, and a few parts of SRFI-1, but none of the SRFIs are officially part of lbjScheme at the moment.
Q: How fast is this thing?
A: Not my primary concern. At the moment, this is a very simple interpreter with no performance optimizations so I can keep the code clean and simple. The next evaluator will be a compiler though, and maybe I'll add a JVM bytecode backend, too. Let's see. First thing is to get lbjScheme R5RS compliant.
Q: Is there a console REPL?
A: The .jar file, when started via "java -jar target/lbjscheme.jar", will open a GUI window or start a console REPL depending on where it was started from. There are a few command line options for the console REPL that I'll refactor a bit in the near future so that the GUI will support them too, and add a force-console mode too. Nothing fancy yet.
Q: Why did you write this?
A: Why not?
Q: Seriously, why?
A: This is my Scheme interpreter. There are many like it, but this one is mine. I'm doing this because I want to learn more about the Scheme language, compiler and interpreter implementation techniques, and so on. Reading SICP challenged my mind in so many ways that I wanted to build my own interpreter to toy around with. This is the culmination of lots of little side projects evolving in parallel, and now I'm closer than ever to completing my very own Scheme implementation, one that I fully understand, without any magic going on inside some source code taken from other Open Source projects. Maybe I'll even add a Scheme-to-C compiler that uses its own garbage collector. Lots of cool stuff to experiment with. Apart from that, if you truly want to learn a programming language, write a compiler for it. And Scheme is one of the few languages out there that manage to be both powerful and simple, so the concepts you learn in the process of implementing Scheme change how you think about programming in general.