Misleading error message when f64 is used in shader

[TrueDoctor]

When using a double-precision float in a shader without enabling the wgpu SHADER_F64 feature, the validation fails with this error message:

thread 'main' panicked at 'wgpu error: Validation Error

Caused by:
    In Device::create_shader_module
      note: label = `gpu::eval`

Shader validation error:
  ┌─ gpu::eval:1:1
  │
1 │
  │   naga::Constant [6]


    Constant [6] '' is invalid
    The type doesn't match the constant

', /home/dennis/.cargo/registry/src/index.crates.io-6f17d22bba15001f/wgpu-0.16.1/src/backend/direct.rs:3019:5

https://github.com/gfx-rs/naga/blob/f31093fb3b0c8183069037fcac99a90d3adef2bd/src/valid/mod.rs#L306-L310

This throws an invalid type error even though spirv-val and naga-cli pass.
Does it make sense to throw a more descriptive error here?

https://github.com/gfx-rs/naga/blob/f31093fb3b0c8183069037fcac99a90d3adef2bd/src/valid/type.rs#L217-L224

[eddyb]

My two cents: the message for InvalidType is simply wrong (not a mismatch but rather the type itself is not supported) and it should be wrapping a TypeError which itself would have all the necessary information to explain what went wrong with the constant's type.

[teoxoy]

I think the changes in gfx-rs/naga#2266 (which is part of the 0.13 release) resolved this issue. You should now get the proper error (capability error). Could you try it again?

[TrueDoctor]

Yes, I should have time to try that out in the next few days