Minutely updated vector tiles demo

That looks like a huge improvement as far as tile size is concerned. It looks like it is now only 50% larger than Versatiles unpacked. My to-Hamburg-zoom-measure is now

Vector tile provider Bytes transferred Bytes unpacked
tilekiln demo 29 MB 65 MB
Versatiles 22 MB 46 MB
JawgMaps 2.5 MB 11 MB

I am also seeing weird artifacts, like the whole of North America and Spain missing on zoom level 1, west Africa having an unexpected shape:

1 Like

On my desktop zooming-in and out, I succeeded to see east of Canada. But problems persist at low zoom level.

  • zoom 0.93, the planet is gone except for a small island.
  • zoom 1.27 all of north-america is gone
  • zoom 2.27 all planet looks ok
  • zoom 3.27 artefacts for east of Canada are back
  • zoom 4.27 east of Canada map is back

I find it curious that so many of those thresholds end in decimal .27. Hm. Maybe some kind of fractal edge…built into the math of the renderer or coincidence? Not critical. Simply my inner monologue running publicly (again).

1 Like

This is a bug in Maplibre. Some other changes appear to have suppressed the problem, but I have a copy of the tiles saved for debugging later. The tiles themselves are fine, it’s just sometimes Maplibre has seemingly random problems with geometries that report as valid to everything else.

1 Like

Many ways are missing (highway=track?), like this one:

https://pnorman.github.io/tilekiln-shortbread-demo/#17/49.38723/8.739258

Is that intentional?

Worth noting too that the Maplibre implementations are different - you’ll find stuff that renders fine in JS but not in Native.

1 Like

I can’t view the map with Firefox. Chrome is working fine.
Can anybody else confirm this behaviour?

FF is fine here.

1 Like

Ok, then the problem is on my side.

highway=steps are also missing which leads to “broken” footpaths (example) but I assume that is an upstream style issue rather than a problem with this demo.

It is not a style issue, highway=steps and highway=track are not included in the vector tiles.

According to the documentation, they should be Vector Tile Schema 1.0 :: Shortbread . Is this a bug, or a known missing feature? (Has it been reported?)

Before the set of changes (tile regeneration), highway=track was present in the vector tiles, as can also be seen here on the left of the screenshot:

Screenshot location: https://www.openstreetmap.org/#map=17/37.86153/-1.57520
So I don’t think it’s an issue with Vector Tile Schema 1.0 :: Shortbread.

highway=service + service=driveway not rendered either. I opened a Github issue to clarify if this is the intended behavior or not, but not sure if it was the correct place.

Edit: Yep, not the correct place. Don’t know where to report this. Perhaps here GitHub - versatiles-org/versatiles-style: create map styles?

I’ve noticed a change that isn’t showing in the demo, at this location. (The demo is currently showing the old junction arrangement, with a traffic island, but it’s been changed to a T junction.) The change was made after you regenerated the tiles. The change is showing in OSM Carto.

It has been fixed and will start appearing. I’m not immediately doing a reload.

2 Likes

Updates had been off for a few days. They’re running now.

Versatiles isn’t what’s used to create the tiles, so no. I’m sure that when/if @pnorman feels ready to invite issues then he’ll point to any issue tracker that’s created.

1 Like

A public issue tracker for the tile generation already exists, it is here:

As apparently this bug has not been noticed yet, It would be helpful to create a bug report at maplibre, otherwise I don’t think it will be fixed anytime soon.

Could it be that geometries need to be simplified more to be under the hard 65535 vertex limit in MapLibre? On JS there’s only a warning logged:

Max vertices per segment is 65535: bucket requested 67242
4 Likes