Template meta-programming Lisp (Scheme)

Lisp S-expressions are encoded into template applications, then interpreted and evaluated into new template types, or compiled into C++ code with a minimal (and maximally stupid) run-time system.

The compiler does not do any optimizations yet.

The interpreter is probably the most interesting as that interprets a program and produces an output string at compile time - the compiled output is only a call to puts() with a constant string. Ironically the compiler ends up a bit more boring - it's your typical application of templates to generate C++ code, and doesn't even evaluate anything until runtime.

Samples

See example_compile.cpp and example_interpret.cpp for a simple stand-alone demonstration without using the helper or test scripts.

Dealing with error messages

Template errors are notoriously difficult to read. To translate the template types back into S-expressions, run your compiler output through filter.sh.

Note: the filter code is fairly old now, it needs some updating with recent compilers.

Evaluating expressions inline

runlisp.py converts S-expressions from the command line into template typedefs and inserts them into the interpreter or compiler. See ./runlisp.py --help for usage and more flags.

For smaller expressions (up to maybe 1k characters) the parsing can be done with templates too, minimizing the cheating done by through Python - use the --compile-time-parsing flag to activate it.

The metacircular evaluator is too big to be parsed with ~16GB of RAM though, and if you're experimenting with compile-time parsing larger expressions you probably want to use something like ulimit -Sv $[4*1024*1024] to limit virtual memory allowed by the compiler to 4GB (adjust limit to taste).

Compiling and running the metacircular evaluator from SICP

mceval.sh wraps runlisp.py --compile to compile and launch the metacircular evaluator. It'll take a while to compile on first run, and will then print ;;; M-Eval when it is ready for input.

Running tests