querySourceFeatures can return same cluster twice

[muster-mark]

maplibre-gl-js version: 5.2.0

browser: Google Chrome 136.0.7103.114 (Official Build) (64-bit)

Steps to Trigger Behavior

1. Create a source using { cluster: true }
2. Query features using map.querySourceFeatures("uk-mountains", { sourceLayer: "uk-mountains", filter: ["has", "point_count"], })

Expected Behavior

Clusters returned by querySourceFeatures should be unique, as identified by clusterId

Actual Behavior

In rare instances it can return the same cluster twice. Seems to be most likely to happen when zoomed out fully.

[HarelM]

Thanks for taking the time to report this issue.
Can you please provide a jsbin with the ability to reproduce this bug?
I would also check the sipercluster repo as this is probably the root cause of this issueas it is the underlayer which clustering is built upon in maplibre.

[muster-mark]

@HarelM I would have to give jsbin write access to my gists, which I'm not sure about. I created a codepen instead: https://codepen.io/markfisher/pen/azzrpWV

[HarelM]

Thanks!
Can you reduce the code to something smaller?
This amount of points is a bit hard to understand what's wrong...
Also you can check out some configuration of the super cluster:
https://github.com/mapbox/supercluster

I'm not sure query source feature is the right way to get non duplicated features BTW...

[muster-mark]

I have reduced it to just 2 points now (does not seem a coincidence that they are very close together). It only happens above a certain clusterRadius (the value in the codepen, 27 is the minimum value at which it occurs).
Oh, is there another method to get non-duplicated features? There is no suggestion that querySourceFeatures might return duplicated features... why would that be useful?

[neodescis]

In general, querySourceFeatures (and queryRenderedFeatures) can both return "duplicate" shapes (or different chunks of what was originally one shape) if they come from different tiles. This is true even for GeoJSON sources, as those shapes are tiled before being rendered by MapLibre.

That said, I'm not sure why you would see a duplicate cluster. Maybe its lat/lng is on the "margin" where tiles overlap?

[HarelM]

I believe this is an issue with supercluster as I believe I was able to reproduce the problem without all the maplibre wrapping code, and the cluster is returned twice from two different tiles, which I'm not sure it should be:

• Same cluster is returned from two tiles mapbox/supercluster#254

Feel free to try and tackle it there, I'll be happy to merge the updated version assuming this will be fixed there.

[muster-mark]

@HarelM Thanks for opening the issue on mapbox/supercluster. @neodescis I don't think it is anything to do with being on or near the edges of tiles, as I can shift the lat and lon of both points by the same amount and the issue still occurs.

[HarelM]

When trying to reproduce the issue I noticed it's not near a tile boundary, although this was my first suspect as well.