velyan/resvg

An SVG rendering library.

resvg

resvg is an SVG rendering library.

Purpose

resvg can be used as:

• a Rust library
• a C library (see capi)
• a CLI application (see tools/rendersvg)

to render SVG files based on a static SVG Full 1.1 subset to raster images or to a backend's canvas (e.g. to a QWidget via QPainter).

The core idea is to make a fast, small, portable, multiple-backend SVG library designed for edge-cases.

Another major difference from other SVG rendering libraries is that resvg does a lot of preprocessing before rendering. It converts an input SVG into a simplified one called Micro SVG and only then it begins rendering. So it's very easy to implement a new rendering backend. And you can also access Micro SVG as XML directly via usvg tool.

SVG support

resvg is aiming to support only the static SVG subset; e.g. no a, script, view or cursor elements, no events and no animations.

A list of unsupported features can be found here.

SVG Tiny 1.2 and SVG 2.0 are not supported and not planned.

Results of the resvg test suite:

You can find a complete table of supported features here. It also includes alternative libraries.

Results of the static subset of the SVG test suite:

Performance

Comparing performance between different SVG rendering libraries is like comparing apples and oranges. Everyone has a very different set of supported features, implementation languages, build flags, etc. You should do the benchmarks by yourself, on your images.

Because of that, we are only comparing the relative performance between resvg backends, so you can choose the one by yourself. Also, it helps with regression testing. But you should note, that results still may vary depending on OS, hardware, and backend's underlying library version.

• elementary icons are very simple, and we are basically testing primitives rendering.
• All icons were rendered with 4x zoom.
• Qt has a slow path stroking/filling algorithm, but a fast layer compositing. So on complex files it's pretty fast, as you can see below.

• Oxygen icons are absurdly complex. Many files are more than 200KiB and total size is 1.2GiB! Most of the files include complex features like clipping, masking, patterns, blur and multiple layers.
• All images were converted from .svgz to .svg beforehand.

The tests above we run on a single thread on Gentoo Linux with AMD 3700X and inside tmpfs. You can find the script here.

Used libraries: cairo 1.16.0, Qt 5.13.2, raqote 0.6.1, Skia m76

For more specific details checkout benches/README.md

Project structure

• resvg – rendering backends implementation
  • usvg – an SVG simplification tool
    • roxmltree – a DOM-like XML tree
      • xmlparser – an XML parser
    • svgtypes – SVG types parser and writer
    • simplecss – a simple CSS2 parser
    • ttf-parser – a TrueType/OpenType parser
    • xmlwriter – a simple XML writer
    • rctree – a DOM-like tree
  • resvg-qt – minimal bindings to Qt
  • resvg-skia – minimal bindings to Skia
  • svgfilters – a collection of SVG filters

All other dependencies aren't written by me for this project.

Directory structure

• benches - basic benchmarks for rendering operations
• bindings – minimal bindings to Qt and Skia used by resvg
• capi – C interface for resvg
• docs – basic documentation
• examples – usage examples for resvg as a library
• src – source code
• svg-tests - a collection of SVG files for testing
• svgfilters - SVG filters implementation
• testing-tools – scripts used for testing
• tools – useful tools
• usvg – an SVG simplification library used by resvg

Safety

• The library must not panic. Any panic should be considered a critical bug and should be reported. There are only a few methods that can produce a panic.
• The core library structure (see above) does not use any unsafe, but since all backends are implemented via FFI, we are stuck with unsafe anyway. Also, usvg uses unsafe for fonts memory mapping.

Testing

We are using regression testing to test resvg.

Basically, we will download a previous resvg version and check that the new one produces the same results (excluding the expected changes).

The downside of this method is that you need a network connection. On the other hand, we have 4 backends and each of them will produce slightly different results since there is no single correct 2D rendering technique. Bézier curves flattening, gradients rendering, bitmaps scaling, anti-aliasing - they are all backend-specific.
Not to mention the text rendering. We don't use the system fonts rendering, but a list of available, default fonts will still affect the results.

So a regression testing looks like the best choice between manual testing and overcomplicated automatic one. And before each release I'm testing all files manually anyway.

See testing-tools/regression/README.md for more details.

Also, the test files itself are located at the svg-tests directory.

License

resvg is licensed under the MPLv2.0.