GeoHashUtils.geoHashesForCircle is not generating correctly

[Lucio Piccoli]

the GeoHashUtils.geoHashesForCircle is missing some big chunks of hashes because it seems to believe that is outside of the area. But my test show that it should be there.

   @Test
public void shouldCalculateHashesForCircleBrisbane2000() {
    double lat = -27.4669214;
    double lon = 153.02381279999997;
    double radius =  2000.0;
    int suitableHashLength = GeoHashUtils.suitableHashLength(radius, lat, lon);

    System.out.println(suitableHashLength);
    long start = System.currentTimeMillis();
    Set<String> hashesForCircle = GeoHashUtils.geoHashesForCircle(suitableHashLength, lat, lon, radius);
    long end = System.currentTimeMillis();
    System.out.println("time ms:" + (end-start));


    System.out.println("size: " + hashesForCircle.size());
    System.out.println(hashesForCircle);
    for (String hash : hashesForCircle) {
        double[] point = GeoHashUtils.decode(hash);
        double distance = GeoGeometry.distance(point, new double[]{lon,lat});
        assertThat(hash, distance, lessThan(radius));
        System.out.println( distance );            
    }

    String r7hg9y = "r7hg9y"; 
    double distance = GeoGeometry.distance( GeoHashUtils.decode(r7hg9y), new double[]{lon,lat});
    System.out.println( r7hg9y + ": "  +distance ); 
    assertThat(r7hg9y, distance, lessThan(radius));
    assertTrue( "this should be in circle", hashesForCircle.contains( r7hg9y ) );
 }

[Jilles van Gurp]

Hi, thanks for reporting. I usually use a simple google maps hack to debug this type of issues. I've added the tool to the repository just now under tools. If I render your circle, I get cluster of geohashes in Brisbane that sort of looks alright. If you bump the geohashlength by one, you can see better what is happening. You'll have to manually paste in the hashes arrays below.

The chunk that appears to be missing would be partially outside the circle and that's why it is omitted. The fix is to use longer geohashes. This will fill the hole with smaller geohashes and it is more clear this way what it is doing.

For reference, here are the two arrays that I generated from your test coordinate:

The first commmented one is the original one and the second one has one character more in the geohash. Just modify the html and open it in a browser to render.

// var hashes=["r7hg9z","r7hgcb","r7hgdj","r7hgdn","r7hgdp","r7hgdq","r7hgdr","r7hgf0","r7hgf1","r7hgf2"];
var hashes=["r7hg9ve","r7hg9vf","r7hg9vg","r7hg9vj","r7hg9vk","r7hg9vm","r7hg9vn","r7hg9vp","r7hg9vq","r7hg9vr","r7hg9vs","r7hg9vt","r7hg9vu","r7hg9vv","r7hg9vw","r7hg9vx","r7hg9vy","r7hg9vz","r7hg9wx","r7hg9wz","r7hg9xp","r7hg9xq","r7hg9xr","r7hg9xw","r7hg9xx","r7hg9xy","r7hg9xz","r7hg9y1","r7hg9y2","r7hg9y3","r7hg9y4","r7hg9y5","r7hg9y6","r7hg9y7","r7hg9y8","r7hg9y9","r7hg9yb","r7hg9yc","r7hg9yd","r7hg9ye","r7hg9yf","r7hg9yg","r7hg9yh","r7hg9yj","r7hg9yk","r7hg9ym","r7hg9yn","r7hg9yp","r7hg9yq","r7hg9yr","r7hg9ys","r7hg9yt","r7hg9yu","r7hg9yv","r7hg9yw","r7hg9yx","r7hg9yy","r7hg9yz","r7hg9z","r7hgc8n","r7hgc8p","r7hgc8r","r7hgc8x","r7hgc8z","r7hgcb","r7hgcc0","r7hgcc1","r7hgcc2","r7hgcc3","r7hgcc4","r7hgcc5","r7hgcc6","r7hgcc7","r7hgcc9","r7hgccd","r7hgcce","r7hgccf","r7hgccg","r7hgcch","r7hgccj","r7hgcck","r7hgccm","r7hgccn","r7hgccp","r7hgccq","r7hgccr","r7hgccs","r7hgcct","r7hgccu","r7hgccv","r7hgccw","r7hgccx","r7hgccy","r7hgccz","r7hgcf5","r7hgcfh","r7hgcfj","r7hgcfm","r7hgcfn","r7hgcfp","r7hgcfq","r7hgcfr","r7hgcfx","r7hgdhb","r7hgdhc","r7hgdhf","r7hgdhg","r7hgdhu","r7hgdhv","r7hgdhy","r7hgdhz","r7hgdj","r7hgdkb","r7hgdm0","r7hgdm1","r7hgdm2","r7hgdm3","r7hgdm4","r7hgdm5","r7hgdm6","r7hgdm7","r7hgdm8","r7hgdm9","r7hgdmb","r7hgdmc","r7hgdmd","r7hgdme","r7hgdmf","r7hgdmg","r7hgdmk","r7hgdms","r7hgdmt","r7hgdmu","r7hgdmv","r7hgdmw","r7hgdmy","r7hgdmz","r7hgdn","r7hgdp","r7hgdq","r7hgdr","r7hgdw2","r7hgdw8","r7hgdw9","r7hgdwb","r7hgdwc","r7hgdx0","r7hgdx1","r7hgdx2","r7hgdx3","r7hgdx6","r7hgdx8","r7hgdx9","r7hgdxb","r7hgdxc","r7hgdxd","r7hgdxf","r7hgf0","r7hgf1","r7hgf2","r7hgf30","r7hgf31","r7hgf32","r7hgf33","r7hgf34","r7hgf35","r7hgf36","r7hgf37","r7hgf38","r7hgf39","r7hgf3b","r7hgf3c","r7hgf3d","r7hgf3e","r7hgf3f","r7hgf3g","r7hgf3h","r7hgf3j","r7hgf3k","r7hgf3m","r7hgf3n","r7hgf3p","r7hgf3q","r7hgf3r","r7hgf3s","r7hgf3t","r7hgf3u","r7hgf3v","r7hgf3w","r7hgf3x","r7hgf3y","r7hgf40","r7hgf41","r7hgf42","r7hgf43","r7hgf44","r7hgf45","r7hgf46","r7hgf47","r7hgf48","r7hgf49","r7hgf4d","r7hgf4e","r7hgf4f","r7hgf4g","r7hgf4h","r7hgf4j","r7hgf4k","r7hgf4m","r7hgf4n","r7hgf4p","r7hgf4q","r7hgf4r","r7hgf4s","r7hgf4t","r7hgf4u","r7hgf4v","r7hgf4w","r7hgf4x","r7hgf4y","r7hgf60","r7hgf61","r7hgf62","r7hgf63","r7hgf64","r7hgf65","r7hgf66","r7hgf67","r7hgf68","r7hgf69","r7hgf6h","r7hgf6j","r7hgf80","r7hgf81","r7hgf82","r7hgf83","r7hgf84","r7hgf88","r7hgf89","r7hgf8b","r7hgf8c","r7hgf90","r7hgf92"];

[Lucio Piccoli]

hi jilles
thanks for your response.
after reading the code i think i all i need is to get the smallest hashcode size to provide partial contained.
so in the splitAndFilter function i want only this check
else if (nw || ne || sw || se)
instead of the fully contained
if (nw && ne && sw && se)
i am more interested in getting total coverage outside the geometry which can be filter out later if needed.
i hope that makes some sense.
thanks again.

[Jilles van Gurp]

Yes that makes sense. Feel free to contribute a pull request that makes this optional behavior. I think that could be useful to some people.

I opted to fill the polygon from the inside and use smaller ones to fill the gaps. This makes sense if you don't want any false positives. If you care less about that, you could consider using a bounding box instead. There's a point where simply using a bounding box is probably more efficient.