[BUG] Broken yellow polygon covers the Adriatic sea.

[hyperknot]

Describe the bug
Broken yellow polygon covers the Adriatic sea on given zoom levels.

To Reproduce
Full log included. Nothing changed since the last few month, except the source data.

Expected behavior
Shouldn't be a broken polygon in the Adriatic sea.

Screenshots

Environment (please complete the following information):
Visible in the full log.

Additional context
I understand that this might not be a complete Issue, and also my build is old. But what broke here is related to the source data, not the binary of planetiler.

build git hash: cf49b86
build version: 0.8.3-SNAPSHOT
build timestamp: 2024-08-26T22:54:11.710Z

Of course, I'm happy to upgrade to latest commit if it helps.

planetiler.log

[msbarry]

Can you download and share the actual mvt? I need that to isolate if planetiler is emitting an invalid polygon or if maplibre is failing to render a valid polygon

Also what version of maplibre are you using? I fixed a bug in earcut a little while ago that would look like that mapbox/earcut#177

[hyperknot]

MapLibre is 5.1.1

The tile is one of:
https://tiles.openfreemap.org/planet/20250212_001001_pt/7/70/48.pbf
https://tiles.openfreemap.org/planet/20250212_001001_pt/7/71/48.pbf
https://tiles.openfreemap.org/planet/20250212_001001_pt/7/69/48.pbf

[hyperknot]

It's visible in Maputnik Inspector:

Feature ID: 93501283, starts at zoom level 7

[msbarry]

From that screenshot the feature ID is 93501283, which means it's https://www.openstreetmap.org/relation/9350128. It looks like the relation had natural=beach tag on it from Jan 29 to Feb 5 which explains what you're seeing here.

It also looks like that relation has since been removed:

[hyperknot]

Oh, great, then it's already solved.

Do you have any idea how to possibly detect these in the future? Or it's not really possible other than rendering the map client side with MapLibre and comparing the rendered results with pixelmatch?

[msbarry]

There's a lot of things that could go wrong, and a lot of legitimate changes that could trigger a red flag...

You could define a set of sanity checks and have them run against the output, like a bunch of points/polygons you expect to be ocean, etc. That's how planetiler unit tests verify it's working end to end in CI (just Monaco though)

Alternatively, you could use overture data for some layers like water, since they do some amount of QA before each release.

[hyperknot]

Thanks! But it's not as simple as looking at this part for example, right?

data errors:
 render_snap_fix_input 23,687,992
 merge_snap_fix_input  22,545
 omt_fix_water_before_ne_intersect 1,711
 feature_polygon_osm_invalid_multipolygon_empty_after_fix  309
 osm_multipolygon_duplicate_member 101
 omt_park_area_osm_invalid_multipolygon_empty_after_fix  12
 feature_centroid_if_convex_osm_invalid_multipolygon_empty_after_fix 12
 osm_boundary_duplicate_member 11
 omt_boundary_poly_osm_invalid_multipolygon_empty_after_fix  8
 feature_point_on_surface_osm_invalid_multipolygon_empty_after_fix 1
 omt_water_osm_invalid_multipolygon_empty_after_fix  1

[msbarry]

Most of those get auto-recovered from, but some of them result in the feature getting dropped. None of them would have caught this case since it was a valid polygon that rendered correctly, it just turned the Adriatic Sea into a beach.

[hyperknot]

Thank you for the explanation and the help!