ruzu

What this project actually is

I wanted to know one thing: could frontier LLMs carry a genuinely huge piece of software from C++ to Rust? Not generate a plausible-looking file, not translate a self-contained algorithm, carry a whole codebase across, one that is:

large: hundreds of thousands of lines across a dozen subsystems; deeply stateful: an HLE kernel with schedulers, fibers, IPC and services; unforgiving: a GPU command processor, a shader recompiler, and a JIT where a single wrong bit produces a black screen rather than a stack trace; built on C++ idioms with no direct Rust equivalent: inheritance hierarchies, raw pointer graphs, shared_ptr cycles, destructor ordering, std::variant.

I had serious doubts it would go anywhere. It did: the experiment ended up producing an emulator that boots and runs commercial titles, and that reaches the performance of the C++ original.

"Faithful" was the whole point, and it is a much harder target than "works". The port is held to structural parity with the C++ source: the Rust file tree mirrors the upstream directory tree, methods live in the file their C++ counterpart lives in, constants stay next to the code that owns them, and lifecycle order is preserved literally. A maintainer looking at any Rust file should be able to answer which C++ file is this, and what is still missing? The contract the work is held to is written down in CLAUDE.md — it is the interesting artefact of this project as much as the code is.

Now ruzu can fly on its own.