Is there a way to use index_axis on a reference to ArrayBase?

[LukeMathWalker]

I have been reviewing #606 and to better understand how the code works I was trying to rewrite it (it usually helps me).
The skeleton of what I was trying to do looks like this:

fn format_array<A, S, D, F>(
    view: &ArrayBase<S, D>,
    f: &mut fmt::Formatter,
    mut format: F,
    limit: Ix) -> fmt::Result
where
    F: FnMut(&A, &mut fmt::Formatter) -> fmt::Result,
    D: Dimension,
    S: Data<Elem=A>,
{
    match view.shape() {
        [] => format(view.iter().next().unwrap(), f)?,
        [_] => format_1d_array(&view, f, format, limit)?,
        shape => {
            let first_axis_length = shape[0];
            let indexes_to_be_printed: Vec<Option<usize>> = [...]
            write!(f, "[")?;
            for index in indexes_to_be_printed {
                match index {
                    Some(i) => format_array(
                        &view.index_axis(Axis(0), i), f, format, limit
                    )?,
                    None => {
                        writeln!(f, "...,")?
                    }
                }
            }
            write!(f, "]")?;
        }
    }
    Ok(())
}

If I try to compile it, I obviously get this error:

error[E0277]: the trait bound `D: dimension::remove_axis::RemoveAxis` is not satisfied
   --> src/arrayformat.rs:111:31
    |
111 |                         &view.index_axis(Axis(0), i), f, format, limit
    |                               ^^^^^^^^^^ the trait `dimension::remove_axis::RemoveAxis` is not implemented for `D`
    |
    = help: consider adding a `where D: dimension::remove_axis::RemoveAxis` bound

At this point I'd have liked to work with a dynamical dimension: get a "dynamically dimensioned" view of my ArrayBase reference and take the risk of a panic when using index_array (if I try to pass invalid parameters).
Is this even possible?
Looking at the signature of into_dyn, it seems I need to consume the array in the conversion, which is something I can't do.
Is there an alternative I can use that could work with this function skeleton?

[LukeMathWalker]

It's when I write stuff down that I get the "aha" moment.
This

let dyn_view = view.view().into_dyn()

should do the job.
I am still curious to know if there is a better way though.

[jturner314]

Yes, I think .view().into_dyn() is the best way right now. We could add a .dyn_view() method or something to make sure the shape/strides don't get copied twice (assuming the compiler doesn't optimize out the extra copy anyway), but unless it's called in a long, tight loop, the cost of an extra copy should be negligible anyway.

[LukeMathWalker]

Yeah, I think we should be fine. Thanks for confirming!