Seemingly misleading tilestats when a source has more than 1 geometry type

[zabop]

Recently I noticed that I find the tilestats section of the metadata.json slightly surprising.

Short summary: I created a GeoJSON file with 3 geometries: a point, a line, and a polygon. I processed it using tippecanoe (creating both an mbtiles file and a directory tree of pbf files). All 3 input geometries show up in QGIS when the mbtiles file is loaded. However, the tilestats section of metadata.json only mentions Point as geometry.

---

Steps to reproduce the issue

Create example GeoJSON file:

#!/usr/bin/python3

import geopandas as gpd
import shapely.geometry

point = shapely.geometry.Point(12, 34)
linestring = shapely.geometry.LineString([[10,20],[-30,40]])
polygon = shapely.geometry.Polygon([[0,0],[0,10],[10,10],[0,0]])

gpd.GeoSeries([point,linestring,polygon]).set_crs(4326).to_file("example.geojson")

Tile data to an mbtiles file:

tippecanoe -Z2 -z10 -o out.mbtiles example.geojson

Tile data to a directory tree:

tippecanoe -Z2 -z10 -e out example.geojson

Open mbtiles file in QGIS:

Features show up as expected. Inspect the tilestats section of metadata.json:

cat out/metadata.json | jq '.json' | jq -c '. | fromjson' | jq '.tilestats'

Get:

{
  "layerCount": 1,
  "layers": [
    {
      "layer": "example",
      "count": 3,
      "geometry": "Point",
      "attributeCount": 0,
      "attributes": []
    }
  ]
}

It's nice that we have "count": 3 (we started with 3 features). However, why do we have "geometry": "Point"?

I think it would be nice to have ["Point","LineString","Polygon"] instead, for example. Or some other value which enables users to know that not only Point geometries where used to create the tiles. Alternatively, I think an option forcing tippecanoe to fail if tilestats doesn't accurately represent input geometries (perhaps --be-strict-with-tilestats?) would be very useful.

Thanks for sharing some insights on the matter :) Tippecanoe is awesome, thanks a lot for developing it.