Tilecover with the same min and max zoom creates unpredictable results

[tcql]

trying to cover the US with min_zoom:7, max_zoom: 7 - no coastal tiles, and lots of scattered random ones

if I generate tiles with min_zoom: 6, max_zoom: 7 then use tilebelt to get the child-tiles for z6 tiles, I get what I was expecting

Interestingly, both runs result in the same number of tiles, which suggests there's just some data flow issue where the coastal tiles get their coordinates scrambled up

[bkowshik]

I came across a similar issue today with the following code snippet:

// File: index.js
var feature = {"type":"Feature","geometry":{"type":"Polygon","coordinates":[[[77.3440933227539,13.239277104266526],[77.3440933227539,13.239611302005722],[77.34443664550781,13.239611302005722],[77.34443664550781,13.239277104266526],[77.3440933227539,13.239277104266526]]]},"properties":{}};
var tiles = tileCover.geojson(feature.geometry, {min_zoom: 25, max_zoom: 25});
console.log(JSON.stringify(tiles));

I ran the script and piped to tippecanoe and mbview.

node index.js | tippecanoe -pf -pk -fo tileset.mbtiles && mbview tileset.mbtiles -q

Staggered tiles viewed with mbview

Close up view of staggered tiles

[bkowshik]

Also, I have seen the following behaviour on a couple of occasions. Basically, things are great except a few extra projection tiles somewhere around the corners of the feature that is tile-covered.

A z7 tile covered by z15 tiles

An extra tile on the top-left corner

[bkowshik]

Added a fixture and a failing test case with: https://github.com/mapbox/tile-cover/commit/a8fdfe57ca73383579b3aa5af8d9e577637e2980. The idea behind the fixture is:

1. When you take a z20 tile as a feature
2. And tile-cover it with min_zoom and max_zoom limits of 20
3. You should get the same feature as the tile

I am not sure if this fixture is helpful in creating a fix, but I am documenting it for reference.

Input fixture on the left and output tiles on the right

---

cc: @tcql @mourner

[mourner]

@bkowshik I think this can be an expected output — it may return adjacent tiles because the tile coordinate in mercator is unprojected to latitude/longitude bbox, which is subject to floating point precision errors, and then tiles that had a border exactly matching a tile side can start overlapping it slightly.

However, the cases in the first two comments look like serious bugs.

[mourner]

@bkowshik It looks like https://github.com/mapbox/tile-cover/issues/73#issuecomment-260911228 is different from the original issue, and tile-cover generates the tiles properly but tippecanoe somehow rounds the coordinates because they're so tiny (cc @ericfischer could this be the case?). However, I don't think you should ever use a zoom as high as z25 — it translates to about 30-50cm width for a single tile.

[mourner]

@tcql I can't reproduce the original issue with the following script — everything looks good:

var tileCover = require('./');
var geojson = require('./test/fixtures/world/USA.geo.json');
var tiles = tileCover.geojson(geojson.features[0].geometry, {min_zoom: 7, max_zoom: 7});
console.log(JSON.stringify(tiles));

node debug.js | geojsonio

Can you set up a broken test case?

[e-n-f]

@mourner Yes, it looks like the misaligned grid is probably from Tippecanoe tiny polygon reduction and could be eliminated with --no-tiny-polygon-reduction, at the cost of probably making the low zooms not work at all. I don't think this could explain the case of the one extra stray tile at the top left, but I'll try some tests and find out.

[mourner]

@tcql I have to close this as not reproducible on the tile-cover side. Please feel free to reopen after we have a reproducible test case.