Naga: Should not be a structural type system

[stefnotch]

Description
Currently naga will accept using a struct as another struct with the same layout.

Could it be that type checking only checks if the layout of a type is the same? And that aliases could be an issue, because they're represented as a copy of a given type in the naga::Module.

Repro steps

Add the following to naga/tests/wgsl_errors.rs. Naga creates a module and says it's valid, instead of throwing the expected error.

#[test]
fn different_types() {
    check_validation! {
        "@vertex
fn vtx_main() -> @builtin(position) vec4f {
  // I'm assigning a Foo to a Bar
  let x: Bar = Foo(1);
  return vec4f(f32(x.a));
}
 
struct Foo { a: u32 };
struct Bar { a: u32 }; // Same layout":
        Err(naga::valid::ValidationError::Type  {
            ..
        })
    }
}

Expected vs observed behavior
I expected it to match the Chrome behaviour of throwing an error.

Platform

• Windows 11
• WGSL trunk on commit 583cc6a (one of the latest commits)

[teoxoy]

Thanks for filing! I thought we had an issue tracking this but I can't seem to find it.