Rust crate

The computational core is the rebasis crate, which has no Python dependency (the Python bindings live in the separate rebasis-py crate). The Python API is a thin layer over it, and the names correspond:

Python	Rust
rb.Space.hats(x), rb.Space.voigts(...)	Space::hats(&x), Space::profiles(&centers, &[Profile::voigt(σ, γ)?], cutoff)
rb.measure(k, b)	measure(&k, &b, Normalization::Integral, Weight::Uniform) → Csr
rb.measure_derivative(k, b, "center")	measure_derivative(&k, &b, norm, weight, Parameter::Center)
rb.Projection(src, dst)	Projection::new(&src, &dst, weight)
space.to_wavenumber(jacobian=True)	space.mapped(CoordinateMap::NM_CMINV, true)

Results are compressed sparse row matrices (Csr). apply_axis applies one along any axis of an ndarray array, e.g. the wavelength axis 0 of a sasktran2 radiance (wavelength, los, stokes), and apply_transpose_axis applies its transpose, for reverse-mode derivatives. Space::concatenate combines kernels of different shapes into one space. The crate’s API reference is on docs.rs.