getting one of the closest point for Closest::Indeterminate

[kraktus]

Hello, thanks for the great lib!

When using geo::algorithm::closest_point::ClosestPoint, is there a one of the closest point when the result is Closest:: Indeterminate? If defined

[michaelkirk]

I'm sorry but I don't understand your question. Can you try to rephrase it or give a more specific example?

The docs for Indeterminate if you haven't seen them: https://docs.rs/geo/latest/geo/enum.Closest.html#variant.Indeterminate

[kraktus]

Sorry, some words got missing in my issue. When there is an "infinite number of closest points", is there a way a method that retrieves one of those?

My real use case was:

• I want to compute the haversine distance between a point and a polygon, but since it is not defined, I approximate it by
• Computing the closest point between the polygon and the point
• computing the haversine distance between one polygon and the closest point of the polygon

extract from my code

// licence: AGPLv3
            let arp_pt = airport.to_point();
            let Closest::SinglePoint(closest_point) = polygon.closest_point(&arp_pt) else {
                panic!("no closest point")
            };
            let distance_m = arp_pt.haversine_distance(&closest_point) as usize;

In case of of Indeterminate, I would be fine with any one of closest points. For now I bailed out, fortunately it didn't happen in practice

[michaelkirk]

I'm wondering if we should get rid of the Indeterminate case.

Here's where it's currently used in the euclidean implementation:

         if line_length == F::zero() {
            // if we've got a zero length line, technically the entire line
            // is the closest point...
            return Closest::Indeterminate;
        }

Arguably we could do something like:

         if line_length == F::zero() {
            // if we've got a zero length line, technically the entire line
            // is the closest point...
-            return Closest::Indeterminate;
+            return Closest::SinglePoint(line.start_point());
        }

A more interesting third case would be to accumulate multiple points if there are multiple distinct points that are closest.

[michaelkirk]

There remains a need to handle empty geometries - so I guess Indeterminate is still useful there.

We could express that case as Option<Closest>::None

Taking that, and the ability for multiple points to be closest, we could end up with something like:

enum Closest {
  Single(Point),
  Intersection(Point),
  Many(Vec<Point>)
}

pub trait ClosestPoint<F: GeoFloat, Rhs = Point<F>> {
    /// Find the closest point between `self` and `p`.
    fn closest_point(&self, p: &Rhs) -> Option<Closest<F>>;
}

Also, note that technically the ClosestPoint trait accepts a Rhs parameter, implying

If it's there to support higher dimension arguments like: line_string.closest_point(line_string)

That would entail further augmenting our return, since the closest thing might not be a point, but a contiguous segment (or a contiguous area if we're talking polygons).

To be frank, I think we should get rid of the Rhs argument. Or if we go that route, rename the method to closest rather than closest_point.