Complexes

[andy-thomason]

This PR implements the Complexes type and provides complex number support via num_complex.

Rcplx wraps num_complex::Complex<f64> the same way that Rfloat wraps f64.

fn mycplx(x: Rcplx) -> Rcplx {
  -x
}

fn mycplxvec(x: Complexes) -> Complexes {
  x
}

fn mycplxvec2() -> Complexes {
  Complexes::from_values([(0., 0.), (1., -1.)])
}

This PR provides only basic support.

[Ilia-Kosenkov]

I know it is against the naming convention, but shouldn't it be c64 inline with f64 and i64/u64?

[andy-thomason]

Rust likes types to start with a capital letter. But we can override this with a #[allow()]

[Ilia-Kosenkov]

I know, I am just curious and asking for opininons. I am ok with either options.

[Ilia-Kosenkov]

Maybe 'complex number'? 'Complex type' can be misleading.

[andy-thomason]

Ok.

[andy-thomason]

Also, next step is to ditch FromRobj

[andy-thomason]

Are you good with this @clauswilke @yutannihilation @Ilia-Kosenkov ?

[clauswilke]

I gave this a quick look and it generally seems fine to me. One issue that caught my eye is Rcomplex vs. Rcplx. It's not immediately clear to me why there are two types, which one an enduser would be supposed to use, and whether the difference is documented. I only found some documentation for Rcplx.

If there is a good reason for having these two types, then I think this should be documented somewhere. Also, users should be told which one to use when.

[andy-thomason]

One issue that caught my eye is Rcomplex vs. Rcplx.

Currently, we either wrap num_complex or a simple struct with the same layout in Rcplx.

num_complex gives us complex arithmetic, functions, nan tests and more but
is not the same as Rcomplex even though it has the same layout.

We could have implemented arithmetic for Rcomplex, but it would have been a lot
more work and would involve a lot of transmute calls.

[clauswilke]

Currently, we either wrap num_complex or a simple struct with the same layout in Rcplx.

That's fine, and I have no concerns with the technical aspects, but I think this should be stated clearly in the docs somewhere.

If I understand correctly, endusers would never normally use Rcplx, only Rcomplex. So I would suggest adding these thoughts to the documentation:

• In the documentation for Rcplx, mention that endusers would typically use Rcomplex.
• In the documentation for Rcomplex, mention that it wraps two different types, depending on use case. Also mention how this can be configured via features.

[yutannihilation]

I'm not very familiar with complex math, but this generally looks good to me.