Order of "types" for generic benchmark is off

[dvdplm]

When setting up a generic benchmark using the types macro argument I noticed that the order is not respected when the benchmark runs and the types in question are generic.

With an invocation like so:

#[divan::bench(types = [U256, U1024, U2048, U4096])
fn gcd<T>(bencher: Bencher<'_, '_>) {
  // SNIP
}

…I would expect the benchmark to run over the types in the order listed (U256, U1024, U2048, U4096). Instead, what I see is:

divan              fastest       │ slowest       │ median        │ mean          │ samples │ iters
╰─ benches                       │               │               │               │         │
   ╰─ gcd                        │               │               │               │         │
      ├─ Uint<16>  199.8 ns      │ 3.569 µs      │ 223.7 ns      │ 228.2 ns      │ 273839  │ 2190712
      ├─ Uint<32>  274.9 ns      │ 4.593 µs      │ 322.7 ns      │ 327.9 ns      │ 320700  │ 1282800
      ├─ Uint<4>   140.3 ns      │ 1.21 µs       │ 155.9 ns      │ 157.5 ns      │ 273881  │ 4382096
      ╰─ Uint<64>  468.4 ns      │ 4.889 µs      │ 520.7 ns      │ 531 ns        │ 168361  │ 673444

What seems to be happening is that the proc macro resolves the types into Uint<const L: usize> and then orders them lexicographically (?) such that the "4" in "Uint<4>" ends up after "Uint<16>".

Is this something that is easy to fix? I think the least surprising behaviour would be to respect the order entered, e.g. types = [Vec<u8>, ZedType, AbcType] would run using first Vec<u8>, then ZedType and lastly AbcType.

[nvzqz]

This will be fixed in the next release soon. Thanks for pointing it out!