Overzoomed line label has characters of reversed string

[wipfli]

maplibre-gl-js version: 4.7.1

browser: chrome ubuntu

Steps to Trigger Behavior

1. open https://jsbin.com/cacihux/edit?html,output
2. see how the 50 line label sometimes appears as 05

Link to Demonstration

https://jsbin.com/cacihux/edit?html,output

Expected Behavior

characters of string appear in the normal order

Actual Behavior

characters of string appear in reversed order

[wipfli]

The rotation seems also off by 180 degrees when the string is reversed.

[HarelM]

Can you check what is the first version experiencing this issue? Maybe we'll be able to find the offending PR.

[wipfli]

3.0.0

[wipfli]

And more precisely, 3.0.0-pre.5

[wipfli]

Possibly related pull request is #2260

[HarelM]

@ChrisLoer any chance you can take a look at this issue?

[ChrisLoer]

I think this is a line that doubles directly back on itself? At least I see there are overlapping features making up this line:

Removing the text-offset: [0, 1] shows the doubling back in this video:

double_back_fort.mov

I'm not sure the exact mechanics, but it's kind of not surprising to me that the orientation would get flipped if it just can't calculate a meaningful normal to the line.

[ChrisLoer]

Well maybe I'm wrong about doubling back, but I'm definitely confused by the intended geometry here -- it looks like all of these features have just two vertices but they overlap each other sometimes going in the same direction (but with different lengths) and sometimes going in opposite directions?

{ "type": "Feature", "properties": { "mvt_id": null, "edge_id": 45185, "length": 12.907566720770523, "visits": 49, "after_list": [ 45185, 49 ] }, "geometry": { "type": "LineString", "coordinates": [ [ 2839.0, 3066.0 ], [ 2945.0, 3138.0 ] ] } },
{ "type": "Feature", "properties": { "mvt_id": null, "edge_id": 45186, "length": 6.9214614238149998, "visits": 49, "after_list": [ 45186, 49 ] }, "geometry": { "type": "LineString", "coordinates": [ [ 2839.0, 3066.0 ], [ 2783.0, 3028.0 ] ] } },
{ "type": "Feature", "properties": { "mvt_id": null, "edge_id": 45159, "length": 11.855569527575458, "visits": 50, "after_list": [ 45159, 50 ] }, "geometry": { "type": "LineString", "coordinates": [ [ 2716.0, 2483.0 ], [ 2648.0, 2579.0 ] ] } },
{ "type": "Feature", "properties": { "mvt_id": null, "edge_id": 45160, "length": 22.855715638137209, "visits": 50, "after_list": [ 45160, 50 ] }, "geometry": { "type": "LineString", "coordinates": [ [ 2716.0, 2483.0 ], [ 2866.0, 2313.0 ] ] } },
{ "type": "Feature", "properties": { "mvt_id": null, "edge_id": 45157, "length": 22.855715638137209, "visits": 50, "after_list": [ 45157, 50 ] }, "geometry": { "type": "LineString", "coordinates": [ [ 2866.0, 2313.0 ], [ 2716.0, 2483.0 ] ] } },
{ "type": "Feature", "properties": { "mvt_id": null, "edge_id": 45158, "length": 3.4125469570288272, "visits": 50, "after_list": [ 45158, 50 ] }, "geometry": { "type": "LineString", "coordinates": [ [ 2866.0, 2313.0 ], [ 2889.0, 2288.0 ] ] } },
{ "type": "Feature", "properties": { "mvt_id": null, "edge_id": 45155, "length": 3.4125469570288272, "visits": 50, "after_list": [ 45155, 50 ] }, "geometry": { "type": "LineString", "coordinates": [ [ 2889.0, 2288.0 ], [ 2866.0, 2313.0 ] ] } },
{ "type": "Feature", "properties": { "mvt_id": null, "edge_id": 45156, "length": 5.3431562053574986, "visits": 50, "after_list": [ 45156, 50 ] }, "geometry": { "type": "LineString", "coordinates": [ [ 2889.0, 2288.0 ], [ 2916.0, 2243.0 ] ] } },

[HarelM]

Thanks for looking into this @ChrisLoer, @wipfli can you check if the data is valid?

[wipfli]

What is the best way to validate a vector tile?

[HarelM]

Look at the points and at the features in it and make sure the data makes sense in terms of geometry.
Probably using this library:
https://github.com/mapbox/pbf

[wipfli]

@ChrisLoer what tool did you use to generate the above geojson from the pbf?
Your description of the data by the way is accurate. The tile contains linestrings each having only 2 vertices (one linear segment) and all linestrings should have a reversed counterpart. Since there are no polygons there is not much potential for invalid geometries.

[ChrisLoer]

I downloaded the PBF, opened it in QGIS, and then exported it as GeoJSON. I think you could also use ogr2ogr to do it more directly, but I sniffed around in QGIS first to get a sense for what was in it. Another tool I think is super useful for looking at tiles is (although not so much in this case) @stevage 's Vector Inspector (https://stevage.github.io/vector-inspector/#?url=%20https://wipfli.github.io/debug-assets/reversed-label/16/34323/22952.pbf&loc=16.7/47.366937/8.545262)

I didn't see anything that looked invalid about the geometry, it's more that it's kind of edge-casey and "weird" in a way that seems like it might be triggering the behavior. If you look at addToLineVertexArray, getAnchors, updateLineLabels or placeGlyphAlongLine, you'll see a lot of logic that's trying to iterate over segments of a line -- in this case all of those loops should reduce to a single operation that either fits something on the segment or doesn't, but it kind of seems like maybe there's a logic error in there that's letting it... traverse the same segment backwards? or accidentally traverse on to an adjacent entry in the line vertex array?

I do think turning off the offset kind of highlights the root issue because you can see the doubling back happen incrementally, and I think it rules out the findOffsetIntersectionPoint as the source b/c it skips that whole pathway.