Gambit-C 4.0b22

Gambit-C 4.0b22 offers you a complete and useful version of the Gambit programming system in which the compiler generates portable C code. Gambit consists of two main programs: gsi, the Gambit Scheme interpreter, and gsc, the Gambit Scheme compiler. The interpreter contains the complete execution and debugging environment. The compiler is the interpreter extended with the capability of generating executable files. The compiler can produce standalone executable or compiled modules which can be loaded at run time. Interpreted code and compiled code can be freely mixed. 

Major Features:

1. Conformance. The Gambit-C system conforms to the R4RS, R5RS and IEEE Scheme standards and implements all optional features. Tail calls and first class continuations conform to the Scheme semantics. The full numeric tower is implemented, including: arbitrary precision integers (bignums), rationals, inexact reals (floating point numbers), and complex numbers. Several extensions to Scheme are provided, including: lightweight threads, a foreign-function interface (FFI), and extended I/O capabilities (Unicode, networking, subprocesses, ...).
2. Portability. Because the system is mostly written in Scheme and the compiler generates portable C code, it is easy to port the system as well as programs compiled with it to any platform with a decent C or C++ compiler. There is no reliance on a particular C compiler, although the system can take advantage of some gcc-specific constructs. The C code generated is oblivious to the endianness and word size of the target environment (32 and 64 bit architectures are currently supported).
3. Performance. The compiler includes several powerful program transformations such as user procedure inlining, partial-evaluation, and lambda-lifting. With appropriate declarations in the source code the executable programs generated by the compiler run roughly as fast as equivalent C programs. Programs containing no declarations are also optimized without compromising the Scheme semantics by speculatively inlining predefined procedures (see the benchmarks page for a comparison with other Scheme implementations and other languages). Thanks to an efficient implementation of continuations, the thread system is very efficient and can support millions of concurrent threads. The bignum library uses sophisticated algorithms that are in some cases asymptotically faster than those of GMP (GNU Multiple Precision Arithmetic Library). [Note: With the release of GMP 4.3.0 in April 2009, this claim is no longer true; see the benchmarks page for some crude comparisons on large integers.]
4. Reliability. Stable releases typically have very few bugs. Bug tracking is done with bugzilla. Gambit-C has been used to develop large-scale real-world software in academic and commercial settings.