TODO

• clipper:
  • line
  • tri (article)
• srgb conversion (palette!)
• compile naga IR with cranelift (JIT demo)
• remove VariableType:
  • disregard, will try cranelift first.
  • Variable can store a &'module TypeInner directly and an optional ty_name: Option<&'module str>
  • we might need a trait for things that can fetch &'module TypeInner given the &'module ShaderModule.
• do we want to change some fields from Vec<Option<T>> to [Option<T>; N] where there's a limit? (e.g. color attachments).

Issues

Issues that can be raised with wgpu (marked items are patched in a local copy of wgpu):

• InstanceInterface::create_surface returns a wgpu::CreateSurfaceError, but these can't be constructed other than by a conversion from wgpu_core::CreateSurfaceError.
• custom RequestAdapterError
• RenderPassDescriptor generic over label, with map_label
• RenderPassInterface::end is never called. core implements its own Drop that calls it. But shouldn't this be called by the wrapper?
• TextureViewDescriptor::map_label
• TextureViewInterface::create_view has no knowledge of the texture's descriptor. It would need to store it, but the wrapping type does that already.
• InstanceInterface::wgsl_language_features is behind a feature flag. This means when I implement the interface without the flag, but another crate in the dependency graph enables it, it breaks compilation of my crate.
• InstanceInterface::new doesn't serve a purpose?
• naga really needs something that returns TypeLayout for a TypeInner. type resolution sometimes can't give you a handle, so Layouter won't always do.
• naga: can matrices and vectors really only be multiplied by floats, or by their scalar type? (http://localhost:8001/wgpu/naga/ir/enum.BinaryOperator.html#arithmetic-type-rules)
• naga: make naga::ir::Binding Copy. This might be useful, but we don't need it anymore.
• naga: bools have size and alignment 1. See naga_interpreter::tests::naga_bool_width_is_32bit.
• naga: way to convert Alignment to NonZeroU32 (or anything usable). cranelift wants to know this, just using round_up won't always do. Alternative: just offer a getter for the log2 of the alignment