Rust WGPU hot-reload

demo.mp4

Demo Boids

https://github.com/Azkellas/rust_wgpu_hot_reload/assets/29731210/b3befb4d-6ef5-437e-a737-b99d32dbaa65

Demo Raymarching

https://github.com/Azkellas/rust_wgpu_hot_reload/assets/29731210/592edece-d19e-48a1-9bab-e0e711511992

Hot reload both shaders and rust code while developing a WGPU application with egui integration.

Package a single binary/wasm for native and web targets.

---

Features

• build to native and wasm
• hot reload shader (instant), embed shader files in release and wasm builds
• hot reload rust (~4sec to rebuild and relink library)
• hot reload ui with egui integration
• shader preprocessor
  • #import "file.wgsl"

---

Running the project

To have both shader and rust hot reload, run:

• cargo run --features reload in one terminal
• cargo watch -w lib -x "rustc --crate-type=dylib -p lib" in another terminal. (requires cargo-watch)

Alternatively use cargo runcc -c runcc.yml to run both commands at the same time. (requires cargo-runcc)

You can also run cargo run if you only care about shader hot reloading.

cargo run --release as usual to build a single executable for your native target. For wasm builds, see below.

---

Building for the web

This project targets webgpu first, and defaults to webgl when not available.

# In charge of building and hosting a webserver. Should expose to localhost:8000
cargo run-wasm [--release] [--features ...]

---

Wgsl preprocessing

This project contains a small homemade preprocessor for wgsl files. It currently allows to include other files by using #import "path/to/file.wgsl" in your shader files.

This syntax follows the bevy preprocessor syntax, which is roughly supported by wgsl-analyzer.

---

Using the template

The project comes with a PipelineFuncs trait. Hopefully it should be enough for your needs. You just need to replace the current implementation with yours in lib/lib.rs: demo_pipelines::demo::Pipeline as CurrentPipeline;.

---

Project architecture

The project is divided in two crates: lib and src.

src should only contain the minimal code necessary to start the application and the windowing system, allowing a maximum of code to be hot-reloaded in lib which is built as a dynamic library and reloaded at runtime whenever changes are saved.

Entry point functions should be written in lib/src/lib.rs, be public and have the #[no_mangle] attribute to be hot-reloadable. Note that they cannot be generic. Example:

#[no_mangle]
pub fn get_pipeline_name(pipeline: &CurrentPipeline) -> String {
    pipeline.get_name().to_owned()
}

Then each use of the lib in src should be done via the module hot_lib::library_bridge that makes the bridge between the binary and the dynamic library.

See hot-lib-reloader-rs for more information about hot reloading Rust code and its limitations.

---

Troubleshooting

• Since the wasm and native targets use different flags, switching from one target to the other takes time as many dependencies need to be rebuilt. Be careful to set rust-analyzer to the same target you're building to, otherwise they will compete against each other and create a lot of unnecessary recompiling.

• Rust requires the dylib to be present at compile time to link, so starting the hot-reload mode with runcc can crash if the bin finishes compiling when the no dll/so is present yet. In which case you just have to let the library finish building in dynamic mode and restart runcc.

• wgpu does not use the idiomatic rust way Error to handle errors. See here for more info, or have a look at shader_build.rs::ShaderBuilder::create_module for an example.

---

References:

• hot-lib-reloader-rs
• wgpu and its boids example
• learn-wgpu
• egui