You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Some tiles loses wood and water polygons on z14.
At the same time, adjacent tiles usually do not lose these polygons. As the result there's a hole in the middle of the forest the size of one or two tiles.
This issue was reproduced with full planet pbf and with a small region. This happens to water polygons too. Disappearing parts of Woods and Waters are always same between builds.
You may see this here: z14 lat: 55.85876 lng: 37.72924
I tested both reports with the config and Lua from the first post (landcover, high_resolution: true), on current master (4e0f03d) and on 0ebc1d7 (master as of the original report).
Original report (Moscow bbox): with today's Geofabrik central-fed-district extract I could not find any tile-sized holes at z14 in the bbox, on either build, in memory or with --store. Tile 14/9909/5113 (the location given) is fully covered. The OSM data may have changed since August 2025. If you still see it with current master and current data, the OSM id of a missing polygon would help.
Second report (way 846976779, unchanged since 2020): reproduces. The part of the way in z14 tile 8148/5240, just across the tile edge (about 0.0125% of the tile), is dropped on master and on 0ebc1d7:
So this looks like the ring-collapse-on-scaling problem that #933 and #941 both address. On the Moscow bbox, #933 and #941 also give identical output: compared to master they restore wood in 6 z14 tiles, up to 3.8% of a tile (14/9917/5115).
Activity
SomeoneElseOSM commentedon Aug 28, 2025
Just to throw in something that may be an example of something similar, part of one of the polygons is missing from a tile at https://map.atownsend.org.uk/vector/index_svwd01.html#19/54.2651968/-0.9640996 . The underlying green forest https://www.openstreetmap.org/way/763248810#map=20/54.2651970/-0.9641000&layers=V is shown but the mixed tree cover of the wood https://www.openstreetmap.org/way/846976779#map=20/54.2651959/-0.9640986&layers=V is not.
The top of the problem area corresponds to a tile boundary, but unfortunately it's not large enough to really see at vector zooms < 14. For style and spec see https://github.com/SomeoneElseOSM/SomeoneElse-vector-web-display/tree/main/resources, but note that names and destinations etc. are parameterised.
oberhamsi commentedon Sep 19, 2025
this can also be observed with the shortbread versatiles style near athens, greece, at zoom=8
one tile only has partial ocean. the original shapefiles for the ocean are correct.
https://shortbread-tiles.org/styles/#versatiles
dktue commentedon Oct 28, 2026
I tested both reports with the config and Lua from the first post (
landcover,high_resolution: true), on current master (4e0f03d) and on 0ebc1d7 (master as of the original report).Original report (Moscow bbox): with today's Geofabrik
central-fed-districtextract I could not find any tile-sized holes at z14 in the bbox, on either build, in memory or with--store. Tile 14/9909/5113 (the location given) is fully covered. The OSM data may have changed since August 2025. If you still see it with current master and current data, the OSM id of a missing polygon would help.Second report (way 846976779, unchanged since 2020): reproduces. The part of the way in z14 tile 8148/5240, just across the tile edge (about 0.0125% of the tile), is dropped on master and on 0ebc1d7:
coordinates_geom.cpppart)So this looks like the ring-collapse-on-scaling problem that #933 and #941 both address. On the Moscow bbox, #933 and #941 also give identical output: compared to master they restore wood in 6 z14 tiles, up to 3.8% of a tile (14/9917/5115).