Context

Hello 👋
While using a Chart Library called Nivo (that uses d3-delaunay), we faced an issue with one of our Charts that had empty data.
Basically under the hood the code uses Delaunay.from with an empty array.

Using delaunay.find with an empty array returns NaN each time which kind of breaks the Chart Library workflow because it needed a number used as an index.

Repro

To reproduce the behavior you need to do the following steps :

import { Delaunay } from 'd3-delaunay';

const delaunay = Delaunay.from([]);
console.log(delaunay.find(0, 0)); // ⬅️ This returns NaN

Possible fix

Looking at the code I think I spotted where this can come from :

Basically in the case of points.length that is equal to 0, points.length >>> 1 is still equal to 0 so the operation returns NaN.
Also one other case that could happen is having the points.length equal to 1 but inedges[i] is equal to -1. With this case point.length >>> 1 will return 0 as well, so again trying to divide by 0 will return NaN.

So the fix would be to check those 2 cases separately to return -1 to match the other edge case you covered in the find function.

I can open a PR with a fix and the tests, I just wanted to make sure we want this fix or if the NaN case is actually something you wanted to keep 😄