rumpy

A numpy-compatible Python module implemented in Rust on top of ndarray, exposed to rustpython-vm as the module numpy.

What it is

Drop-in import numpy as np for embedded RustPython interpreters — no CPython, no PyO3, no C extension. Pure Rust array core with a Python surface that matches numpy's API where it makes sense.

Dtypes

bool, int8/16/32/64, uint8/16/32/64, float16/32/64, complex64/128.

Promotion follows numpy.result_type (see src/promote.rs).

Usage

use rustpython_vm::Interpreter;

let interp = Interpreter::with_init(Default::default(), |vm| {
    vm.add_native_module("numpy".to_owned(), Box::new(rumpy::module_def));
});

interp.enter(|vm| {
    let scope = vm.new_scope_with_builtins();
    vm.run_code_string(scope, "import numpy as np; print(np.arange(6).reshape(2,3))", "<embed>".into())
        .unwrap();
});

Layout

• src/ — Rust array core, ops, linalg, fft, npy/npz I/O, indexing, einsum.
• py-src/ — Python shims loaded into the numpy module namespace (dtypes, polynomial, testing, etc.).
• tests/ — Rust-side correctness tests, including parity checks against CPython numpy via pyo3.

Features

• safe-locks — wraps array storage with a real lock (via rustpython-vm/threading). Off by default; embedders that don't run the VM on multiple OS threads can skip it.

Status

Pre-1.0. The numeric core and most numpy surface area work; gaps and rough edges exist. File issues with a failing snippet.