ratatui-wgpu

A wgpu based rendering backend for ratatui.

demo.mp4

This started out as a custom rendering backend for a game I'm developing, and I thought I'd make it available to the broader community as an alternative rendering target for TUI applications. One of its primary goals is to support serving TUI applications on the web.

Alternatives

egui_ratatui uses an egui widget as its backend, allowing it to run anywhere egui can run (including the web).
- Advantages: Egui is significantly more mature than this library and brings with it a host of widgets and builtin accessibility support.
- Disadvantages: You can't run custom shader code against its output.

Goals

The crate has the following goals in order of descending priority.

Allow custom shader code to be run against rendered text.
- See PostProcessor for details. You can also see the implementation of the shaders::DefaultPostProcessor or the hello_pipeline example for a demonstration of how this works.
Target WASM.
- The hello_web example demonstrates its usage for web. hello_webworker shows how to use this backend to render from a worker thread.
- You will likely want to enable the web feature if you intend to support Firefox.
Correct text rendering (including shaping, mixed bidi, and combining sequences).
- For color fonts (e.g. emojis) only colr v0 & v1 outlines and raster images are supported.
Reasonable performance.

Non-goals

Builtin-accessibility support.
- I'm willing to make concessions to enable accessibility (the native version of my game uses accesskit), but integrating directly with accessibility libraries is outside the scope of this library.

Known Limitations

No cursor rendering.
- The location of the cursor is tracked, and operations using it should behave as expected, but the cursor is not rendered to the screen.
Attempting to render more unique (utf8 character * BOLD|ITALIC) characters than can fit in the cache in a single draw call will cause incorrect rendering. This is ~3750 characters at the default font size with most fonts. If you need more than this, file a bug and I'll do the work to make rendering handle an unbounded number of unique characters.

To put that in perspective, rendering every printable ascii character in every combination of styles would take (95 * 4) 380 cache entries or ~10% of the cache.

Changelog

1.3 -> 1.4

Support colr v1.
- Drop skrifa.
Deprecate with_dimensions and replace with with_width_and_height.
- The order of arguments (height, width) to with_dimensions was confusing. The new function takes in a struct which explicitly specifies width & height.

1.2 -> 1.3

Support colr v0 and png/bitmap images in fonts, allowing emoji rendering.
- Switched tiny_skia -> raqote.
- Added skrifa.
- Added unicode-properties.
- Added png.
Fix an issue with incorrect handling of srgb conversion in the default shader.

1.1 -> 1.2

Added support for rtl and mixed bidi.
- Switched swash -> rustybuzz.
Added support for combining characters.
Added webworker example.
Added colors example.
Added dependency on ahash.
Dropped dependency on coolors.
Dropped dependency on palette.
Dropped dependency on guillotiere.

Dependencies

This crate attempts to be reasonable with its usage of external dependencies, although it is definitely not minimal.

ratatui & wgpu: This crate's core purpose is to create a backend for the former using the latter, so these are both necessary.
ahash (optional, default): Inserting into map & set tracking structures during the core rendering loop takes up a significant portion of the render time. Using ahash improves overall performance by ~3% for the colors example in my profiling (12% of execution time vs 15%).
bitvec: During rendering, I need to efficiently track dirty cells which need their background/contents repainted. Bitvec is an efficient structure for tracking which cells are dirty.
bytemuck: Working directly with byte slices is needed to interface with the shader code, and this nicely encapsulates work that would otherwise be unsafe.
indexmap: Used internally to implement an lru heap which has O(1) lookup for entries and to order glyphs in target cells for rendering. This could be replaced with a HashMap<Key, usize> + Vec<Value>, but doing so would require a lot of tedious & error prone book keeping when bubbling entries down the heap.
log: Integrating with standard logging infrastructure is very useful. This might be replaced with tracing, but I'm not going to go without some sort of logging.
png (optional, default): Some fonts embed png images as raster graphics for characters. The png crate is used to decode these images if they are present.
raqote: I don't want to implement path stroking & filling by hand and this library supports all the gradient modes required to render from a font's COLR table.
rustybuzz: Text shaping is hard and way out of scope for this library. There will always be an external dependency on some other library to do this for me. Rustybuzz happens to be (imo) the current best choice.
thiserror: I don't want to write the Error trait by hand. I might consider removing this if doing so doesn't turn out to be so bad.
unicode-bidi: I don't want to implement the unicode bidi algorithm by hand, and even if I did, most of the code would be based on a implementation like this anyways. This performs well enough even though cells have to be concatenated into a single string for processing. There are smarter ways to to this processing I'm sure, but I'll optimize when I need to.
unicode-properties: I need to check if a character is an emoji in order to know how to handle foreground colors and bold/italic styles.
unicode-width: I need to access the width of characters to figure out row layout and implementing this myself seems silly. This is already pulled in by ratatui, so it doesn't really increase the size of the dependency tree.
web-time: Used for crossplatform (web & native) time support in order to handle text blinking.

Jesterhearts/ratatui-wgpu