Getting it

Three ways in, and only the third needs anything installed.

A release. https://github.com/GregoireHENRY/kalast/releases — one archive per platform. Unpack it and run it from inside the folder:

./kalast                                   # the editor
./kalast examples/two_spheres/main.py      # a Python example
./kalast examples/crater_self_shadow/step.rs   # a Rust one
./kalast some/shape.obj                    # a mesh

There is nothing to install and nothing is written outside the folder. The archive carries its own Python, with kalast and its dependencies already in it, and the .rs examples come compiled. Editing one, or opening a .rs of your own, means compiling it: if the machine has no cargo the editor fetches a minimal toolchain into toolchain/ beside the executable and reuses it afterwards. On macOS that also wants Apple’s command line tools for the linker (xcode-select --install).

The package, to use kalast from your own environment:

pip install kalast          # Python
cargo add kalast            # Rust

A clone, to work on kalast itself — that is what Compilation below is about. If you are reading this inside an unpacked release, it is not for you.

Structure

• kalast/: Python wrapper. Provides Pythonic usage of Kalast (e.g. object references) for users less familiar with Rust. Built with maturin.

• src/: Rust core of the simulation. Written to be usable standalone by Rust users, independent of Python — the Python wrapper must not compromise its speed.

• shaders/: wgpu shaders (.wgsl) used by the rendering pipeline.

• examples/: Examples of usage of Kalast. Scripts under examples/old/ are earlier/superseded versions kept for reference, not maintained as user-facing examples.

• res/: resources folder (if missing get it from cloud-as.oma.be).

• out/: default output directory for simulation results.

Compilation

Create a virtual environment to install dependencies and compile code, I recommend astral uv for Python. Then, from within you venv, run the following.

Build the kalast rust dynamic library kalast/_rs.cpython-314-darwin.so (example for Mac).

maturin develop

Use --release for anything measured or run in earnest – debug is 2-15x slower, worst on the per-pixel frame-export loops (measured 22.6 -> 53.1 it/s at 3.1M facets with export on). Keep plain maturin develop while implementing a feature, and rebuild with --release once it works.

maturin develop --release

Beyond the default opt-level = 3 there is nothing worth adding: lto = "fat" + codegen-units = 1 was measured on this project and gave no improvement (render loop 52.7 vs 55.1 it/s, mesh load+flatten 1.07 vs 1.10 s, both inside run-to-run noise) while pushing the build from ~59 s to ~87 s. Recorded in Cargo.toml so it is not retried blindly.

Run a Python example.

python -i examples/two_spheres/main.py

Import kalast from Python and start writing your own scripts.

import kalast;