12 unstable releases (3 breaking)

0.4.0 Jan 21, 2024
0.4.0-rc.0 Jan 20, 2024
0.3.0 Dec 28, 2023
0.2.0 Oct 22, 2023
0.1.0-alpha.2 Sep 30, 2023

#154 in Visualization

Download history 26/week @ 2023-11-05 15/week @ 2023-11-12 35/week @ 2023-11-19 41/week @ 2023-11-26 27/week @ 2023-12-03 16/week @ 2023-12-10 13/week @ 2023-12-17 174/week @ 2023-12-24 39/week @ 2023-12-31 27/week @ 2024-01-07 51/week @ 2024-01-14 80/week @ 2024-01-21 36/week @ 2024-01-28 18/week @ 2024-02-04 74/week @ 2024-02-11 235/week @ 2024-02-18

365 downloads per month
Used in 4 crates

MIT license

4.5K SLoC

vsvg: core library for pen-plotter graphics

github Latest version Documentation GitHub

vsvg is the core crate of the project. It implements data structures, I/O, and algorithms to import, create, manipulate, and export vector graphics for pen-plotter generative art. It is similar in purpose to the core library that is part of vpype, but with many improvements (see bellow).

This crate, to ether with its companion vsvg-viewer, is part of the vsvg project and powers the whiskers interactive sketch environment.


use vsvg::{DocumentTrait, LayerTrait, PathTrait};

fn main() {
    /* == Document == */
    let mut doc = vsvg::Document::default();

    // push a path to layer 1
    doc.push_path(1, vec![(0., 0.), (100., 100.), (200., 0.), (0., 0.)]);

    /* == Layers == */
    let mut layer = vsvg::Layer::default();
    layer.metadata_mut().name = "Layer 2".to_string();

    // vsvg uses kurbo internally, and its API is compatible with it
    layer.push_path(kurbo::Circle::new((50., 50.), 30.));
    doc.layers_mut().insert(2, layer);

    /* == Path == */
    // Amongst various ways to create a path, the SVG <path> syntax is supported.
    let mut path = vsvg::Path::from_svg("M 200 200 L 200 400 Q 500 300 200 200 Z").unwrap();
    path.metadata_mut().color = vsvg::Color::DARK_GREEN;
    path.metadata_mut().stroke_width = 3.0;
    doc.push_path(3, path);

    // save to SVG

This generates the following SVG:

demo SVG

Compared to vpype

This crate offers similar functionalities as the core vpype package and its Document/LineCollection structures. It brings however a host of improvements listed below. In the future, a Python wrapper over vsvg will replace vpype's core package, leading to vpype 2.0.


cargo add vsvg


By "fast", I mean "much faster than vpype". Consider the following command:

vpype read 300_beziers.svg show

On my computer (a 2021 MacBook Pro M1 Max), vpype takes about 570ms just to start. Then it needs another 140ms to load this benchmark file containing 300 Bézier curves. Then it kicks off the viewer (this part is actually rather quick).

As of today vsvg takes ~3.6ms to start and load the same SVG file. It feels instantaneous.

In another test with a rather pathological 90MB SVG, vsvg takes 1.4s to load the file, where it takes upward of 30s for vpype.

Support for Béziers

The core path data structure used by vsvg is kurbo::BezPath. It is similar to SVG's <path> element in that it consists of a series of draw commands such as "move to", "line to", "quad bézier to", "cubic bézier to", and "close". This is a major departure from vpype's approach where everything was linearised to polylines. The major advantages of this path representation is a much greater accuracy while remaining more compact that polyline for all curved elements.

The bezpath example from whiskers provides insights on how kurbo::BezPath are used in vsvg.

Note that kurbo::BezPath does not include arc/circle/ellipses commands. When drawing such primitive (or loading SVGs containing them), they are converted into cubic Béziers. Although this approximation is not 100% accurate, it's vastly superior to polylines. For example, a circle is typically approximated by 4 cubic Béziers (16 coordinates), where hundreds of line segment would be needed for a lesser accuracy.

Compound paths

A subtle (but decisive) advantage of the kurbo::BezPath data structure is that it can contain multiple sub-paths, a.k.a. Inkscape's "Combine" path operation. This happens by using multiple "move to" commands. This means that shapes such as polygons with hole can be modelled in a semantically correct way. Previously, such a shape was represented by distinct, unrelated paths.

Amongst other thing, this will enable hatching algorithm that correctly handle polygons with hole, e.g. where the holes aren't actually hatched. This is currently impossible to do with vpype. vsketch's Shape currently enables this, but at the cost of a lot of internal/external complexity.

Improved linearisation

vsvg retains the ability to convert curved elements into polylines (a.k.a. linearisation), as this step is still required by some operations. For example, exporting to G-code (a feature not yet implemented) typically requires fully linearised path data. Likewise, the vsvg-viewer crate requires linearised paths for rendering.

The linearisation process is better than vpype, with an improved tolerance handling. The segment size adapts based on the curvature instead of respecting an absolute maximum segment length, which minimises the number of points needed when curves are nearly straight.

To conveniently support linearisation, vsvg provides a hierarchy of data structures (FlattenedDocument/FlattenedLayer/FlattenedPath) which generally behaves identically to the kurbo::BezPath-based main hierarchy.

Path-level metadata

vsvg maintains per-path metadata such a color and stroke width. This is an improvement over vpype, which only offers per-layer metadata.


Although I have yet to sort out the details, the entire stack of the vsvg project is compatible with WebAssembly targets. This means that most of what vsvg is capable natively could be made available on the web, including display and interactivity. This opens up lots of very nice opportunities to build web-based tooling, generators, etc.

Is that the future of vpype?

Maybe. At least I'd very much like to.

vpype is currently made of two packages: vpype and vpype-cli. The former implements the "engine" and the API, that's then used by vpype-cli and plug-ins to offer a CLI interface. What this project explores is basically to opportunity to entirely re-implement the vpype package in Rust, which would aptly be renamed vpype-core in the process. This would dramatically improve the performance of vpype, thanks the Rust being compiled to native code and having much better concurrency.


~408K SLoC