lauf

lauf is work-in-progress bytecode interpreter with a C API. It is currently highly unstable, not ready for production use, and does not support anything other than a recent clang and Linux.

Features

• Stack based bytecode API.

• Designed to quickly generate bytecode and execute it; suitable for e.g. evaluating constexpr functions in a C++ compiler.

• Extensible instruction set via lauf_builtin: things like add can be defined with your own semantics (e.g. overflow behavior).

• Support for fibers: cooperative user threads to implement stackful coroutines and generators.

• Fully sandboxed VM that supports low-level operations like pointer arithmetic.

FAQ

How fast is it?

On an M1 processor, recursive fib(35) takes about 325ms while the luajit interpreter (without JIT) takes about 535ms. Proper benchmarks are TODO.

Does it support just-in-time compilation?

Currently, no. I have started implementing it, but it’s a lot of work, and I’ll focus on adding features first.

Why is it called lauf?

"Lauf!" is German for "Run!" and lauf runs bytecode.

Documentation

Currently no documentation besides header comments. Everything should be considered unstable.