const fn constructors

[whatisaphone]

const fn has landed in Rust stable! It would be really helpful in one of my projects if nalgebra adopted it for constructors. The shortlist of types I need is Point2, Point3, Vector2, Vector3, UnitComplex, UnitQuaternion. But it probably makes sense for every type in the library.

Things like UnitComplex::new might not work due to compile-time restrictions on FP math, but I'm happy to work around that with UnitComplex::from_cos_sin_unchecked or whatever else makes sense.

Adding the keyword is a simple enough change (at least I hope!), but it will break on older compilers and I don't know your policy on Rust version compatibility.

[jswrenn]

I don't believe this is possible quite yet! The current level of const_fn support doesn't include functions with trait bounds. Not only do we need this feature to mark constructors as const in nalgebra (because of the pervasive N: Scalar, R: Dim, C: Dim, S: Storage bounds), but we need this feature to make the constructors in generic-array const, too.

[whatisaphone]

Aw, you're right :(

I guess this becomes the tracking issue then.

In the meantime I just learned about the union transmute hack, so that's the alternative for now.

related rust-lang/rust#24111

[sebcrozet]

Revisiting this issue to see if the situation improved since last year. It appears that const fn on functions with trait bounds is still a problem:

error[E0723]: trait bounds other than `Sized` on const fn parameters are unstable
    --> src/base/construction.rs:842:14
     |
842  |           impl<N> MatrixMN<N, $R, $C>
     |                ^
...

[clarfonthey]

Side note, it would be nice if we could have a feature to enable these on nightly even though it'd be a mega pain to get working with macros.