Rounding Vectors / Points.

[Binero]

Is there a way yet to round vectors or points, similar to std's round function implemented for f32 and f64? I cannot seem to find it, but the crate might use a non-standard naming convention, or assume other means of rounding.

Furthermore, if it is not present, would a round function be something to consider? How would we go about adding it, given a rounding operation might not necessarily make sense on any matrix.

[sebcrozet]

This is currently not implemented.

Instead of defining a round method, it would be more useful to define a method apply that takes a closure Fn(N) -> N as parameter. Each component is then replaced by the result of the closure (the modification is in-place so this is not a map).

[Binero]

Hmm I was about to tinker a bit to resolve this issue but it seems that currently the master branch is broken, as the tests do not compile?

[sebcrozet]

Could you paste the error you get please? I've made a fresh clone of the nalgebra repo and test do seem to compile on the master branch.

[Binero]

On stable:

error: cannot find macro `quickcheck!` in this scope
  --> tests/similarity.rs:15:1
   |
15 | quickcheck!(
   | ^^^^^^^^^^

error: aborting due to previous error(s)

On nightly, I get 1096 errors. They all seem to be of the unused result variety.

As for the original issue, wouldn't apply be essentially the same as iter_mut?

[sebcrozet]

Tests and benchmarks can only be computed on nightly because we use #[feature(test)] which is unstable. The errors you have on nightly are strange though. Did you enable any features? Could you try with those features enabled:

cargo test --features "debug arbitrary serde-serialize"

Also, if this still doesn't work, could you try to compile tests on the linalg branch?

[milibopp]

It is a new issue on nightly. I already reported it yesterday in rust-lang/rust#43806. There is already a pull request in, so it should be fixed very soon. For the time being we could comment out the unused_results lint perhaps?

[sebcrozet]

@aepsil0n Yes, or maybe we could just make it a warn instead of a deny so that we don't forget to deny it again later.
@Binero I forgot to answer your second question. The apply I propose would no be iter_mut exactly, but the same as:

for e in matrix.iter_mut() {
    *e = f(*e)
}

Assuming f is a Fn(N) -> N where N is the scalar type.
I am not sure there exists a clean one-liner way of doing this.