fix(mosaic): Make MosaicLayer repeat over world copies (antimeridian)

[willjnz]

Hey @kylebarron, great work - thank you very much.

I tried the beta v0.8 because I needed what I thought was fixed by #518. It turns out that fix doesn't make MosaicLayer repeat across world copies. This PR applies the same fix to MosaicLayer.

There are two independent layers of "which things are visible" in this codebase:

1. Inside an already-open COG: RasterTileLayer decides which pixel tiles (x, y, z) of that one file are visible. Fixed by fix: traverse tiles across world copies (#517) #518.
2. One level up: MosaicLayer's MosaicTileset2D decides which sources (whole COGs - e.g. STAC items) are visible at all, via a spatial query over their bounding boxes, before any RasterTileLayer for that source exists.

#518 only fixed layer 1. If a source is never selected at layer 2, it never gets a RasterTileLayer built for it (renderSource never runs), so #518's fix never gets a chance to run.

The root cause: MosaicTileset2D.getTileIndices queried its Flatbush spatial index once, using the viewport's raw (possibly unwrapped) longitude bounds. Once a WebMercatorViewport with repeat: true pans past ±180°, viewportBounds extends past that range while sources are still indexed at their true, unwrapped bbox - so items on the far side of the antimeridian were never selected.

The fix repeats the query at world-copy offsets - same subViewports.length > 1 gate and MAX_MAPS walk-until-empty pattern as raster-tile-traversal.ts - shifting the search bounds by worldOffset * 360° of longitude instead of translating a bounding volume, since Flatbush operates on WGS84 degrees, not deck.gl common space. Documented as a second, independent layer in dev-docs/world-copies.md.

Test example

Before (v0.8.0-beta.2)	After (this PR)

Note: there is still a gap right at the antimeridian, because COGs with bounds of -180 to +180 (antimeridian-spanning) are excluded - a known separate issue, unrelated to this fix. The geomad layer has some COGs that render badly before and after this PR for that reason; LULC does not.

Minimal app to test

import { COGLayer, MosaicLayer } from "@developmentseed/deck.gl-geotiff";
import type { StyleSpecification } from "maplibre-gl";
import "maplibre-gl/dist/maplibre-gl.css";
import { Map as MaplibreMap } from "react-map-gl/maplibre";
import { DeckGLOverlay } from "./deckgl-overlay.js";

type LulcSource = { id: string; bbox: [number, number, number, number]; href: string };

// One COG on each side of the antimeridian (dep_ls_lulc STAC collection).
const SOURCES: LulcSource[] = [
  {
    id: "dep_ls_lulc_063_019_2025",
    bbox: [177.380649859963, -19.29852907893266, 178.24303253271776, -18.4776694259545],
    href: "https://s3.us-west-2.amazonaws.com/dep-public-staging/dep_ls_lulc/0-0-9/063/019/2025/dep_ls_lulc_063_019_2025_classification.tif",
  },
  {
    id: "dep_ls_lulc_067_019_2025",
    bbox: [-179.169819449018, -19.29852907893266, -178.30743677626327, -18.4776694259545],
    href: "https://s3.us-west-2.amazonaws.com/dep-public-staging/dep_ls_lulc/0-0-9/067/019/2025/dep_ls_lulc_067_019_2025_classification.tif",
  },
];

const BASEMAP_STYLE: StyleSpecification = {
  version: 8,
  sources: {
    osm: {
      type: "raster",
      tiles: ["https://tile.openstreetmap.org/{z}/{x}/{y}.png"],
      tileSize: 256,
      attribution: "© OpenStreetMap contributors",
    },
  },
  layers: [{ id: "osm", type: "raster", source: "osm" }],
};

const layer = new MosaicLayer<LulcSource>({
  id: "lulc-mosaic",
  sources: SOURCES,
  renderSource: (source) => new COGLayer({ id: `lulc-${source.id}`, geotiff: source.href }),
});

export default function App() {
  return (
    <div style={{ position: "relative", width: "100%", height: "100%" }}>
      <MaplibreMap
        initialViewState={{ longitude: 180, latitude: -18.9, zoom: 7 }}
        mapStyle={BASEMAP_STYLE}
        renderWorldCopies={true}
      >
        <DeckGLOverlay layers={[layer]} />
      </MaplibreMap>
    </div>
  );
}

Before (v0.8.0-beta.2)	After (this PR)

[kylebarron]

Thanks @willjnz; this looks reasonable but I'll need to take a closer look before merging. I'm focused on creating presentations for FOSS4G at the moment