Unable to handle shader compilation errors

[Herschel]

wgpu currently has no API to inspect errors when compiling a shader module. On desktop, Device::create_shader_module will panic on error. On web, the different browsers will print out error info to the console, but execution continues, and errors will usually occur later during pipeline creation with the errored ShaderModule objects.

The WebGPU spec requires a GPUShaderModule.compilationInfo() method that should return any errors generated during compilation.

Ideally, either:

• change Device::create_shader_module to idiomatically return a Result,
• or match the spec and add a ShaderModule::compilation_info method. The user can then inspect this info and return an error as appropriate.

[kvark]

create_shader_module follows the same logic as all the other methods that could fail - it triggers the error on the device. The default handler panics, but you can set your own handler -

wgpu/wgpu/src/lib.rs

Line 314 in 5f1b3e5

fn device_on_uncaptured_error(

compilation_info isn't implemented yet, and we definitely need this.

As for Result, we need to land @pythonesque work first and then see.

[Herschel]

Thanks for the tip about onUncapturedError! Unfortunately looks like it's a TODO on wgpu web backend.

[michael-brennan2005]

Sorry if this isn't the right etiquette, but is there any current work being done to implement compilation_info? This is presenting an obstacle to the renderer I'm working on currently.

[Wumpf]

not that I know of no. But there's also some movement on the Naga side on how errors should be handled #4429

[stefnotch]

I would love to take on this issue.

[stefnotch]

The PR for this is up at #5410, and am happily awaiting reviews

[nnmm]

What I'm using, based on the comments in this thread, is

    // Set up error handling. By using `on_uncaptured_error()`, create_shader_module()
    // will return instead of exiting.
    let shader_err: Arc<Mutex<Option<wgpu::Error>>> = Arc::new(Mutex::new(None));
    {
        let shader_err = Arc::clone(&shader_err);
        device.on_uncaptured_error(Box::new(move |err| {
            *shader_err.lock().unwrap() = Some(err);
        }));
    }
    let shader = device.create_shader_module(wgpu::ShaderModuleDescriptor {
        label: Some(&shader_path.display().to_string()),
        source: wgpu::ShaderSource::Wgsl(Cow::Owned(shader_src)),
    });
    if let Some(err) = shader_err.lock().unwrap().take() {
        let comp_info = pollster::block_on(shader.get_compilation_info());
        Err(comp_info)
    } else {
        Ok(shader)
    }

This returns a Result<ShaderModule, CompilationInfo>. However, it's obviously not pretty, and I'm not sure how to reset the on_uncaptured_error() callback after I'm done handling the error that I care about.

[teoxoy]

However, it's obviously not pretty, and I'm not sure how to reset the on_uncaptured_error() callback after I'm done handling the error that I care about.

You should use push_error_scope & pop_error_scope.

[nnmm]

You should use push_error_scope & pop_error_scope.

I tried that, but it didn't work for me. It does work on the main thread, but not in the thread that calls a notify-rs callback. Edit: Seems related to #5375.

Code

    device.push_error_scope(wgpu::ErrorFilter::Validation);

    let shader = device.create_shader_module(wgpu::ShaderModuleDescriptor {
        label: Some(&shader_path.display().to_string()),
        source: wgpu::ShaderSource::Wgsl(shader_src.into()),
    });

    if let Some(_) = pollster::block_on(device.pop_error_scope()) {
        let comp_info = pollster::block_on(shader.get_compilation_info());
        return Err(comp_info);
    }

Error message

[2024-09-10T20:16:10Z ERROR wgpu_core::device::global] Device::create_render_pipeline error: Error matching ShaderStages(FRAGMENT) shader requirements against the pipeline
[2024-09-10T20:16:10Z ERROR wgpu::backend::wgpu_core] Handling wgpu errors as fatal by default
thread 'notify-rs inotify loop' panicked at /home/user/.cargo/registry/src/index.crates.io-6f17d22bba15001f/wgpu-22.1.0/src/backend/wgpu_core.rs:3411:5:
wgpu error: Validation Error

Caused by:
  In Device::create_render_pipeline
    Error matching ShaderStages(FRAGMENT) shader requirements against the pipeline
      Unable to find entry point 'fs_main'


note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace