map: point clouds, and Zarr dimensions set from the style panel

[amanbagrecha]

What

Point clouds: a new pointcloud layer kind for LAS / LAZ / COPC / EPT.

• Rendering: maplibre-gl-lidar (deck.gl PointCloudLayer), one hidden control per layer so each cloud styles on its own.
• COPC and EPT stream by viewport. LAS/LAZ ≤ 1.3 are read whole in the browser.
• LAS 1.4 and CRS problems: LAS 1.4, a CRS stored only in GeoTIFF keys, or a CRS the user names go through prepare.py's new pointcloud action. laspy rewrites the file as LAS 1.2 in lon/lat, thinned to 5M points and cached. The user names a CRS with options.crs, entered in the style panel when the file has none.
• Style panel: colour by elevation, intensity, classification or RGB; colormap and range; ASPRS class toggles; point size; elevation filter; height offset; tilt to 3D.
• Bundle: moves to deck.gl 9.4, which the lidar library needs. Rasters are unchanged and covered by the existing e2e tests. laz-perf.wasm now ships beside the bundle instead of loading from unpkg.
• MCP and file browser: both know the new kind and the new extensions.

Zarr dimensions: every non-map dimension (time, band, depth…) gets a value list, previous/play/next buttons and a slider. The style panel now opens when a layer is added or its card clicked. Testing against real stores turned up bugs, fixed here:

• Pyramids: multiscale stores (carbonplan/ndpyramid) failed with "Not found: v3 array or group".
• Unconsolidated v3: stores without consolidated metadata failed. The page now lists the bucket (ListObjectsV2) and falls back to Python inspect.
• String coordinates: band names showed as bare indexes. Picking a named dimension now recomputes the automatic colour range from that slice only.
• Placement: read from the coordinate arrays and passed to the layer. This covers descending latitude, 0..360 longitude, CF-attribute or WRF-style dimension names, and projected grids (a new crs action returns proj4 and lon/lat bounds).
• Extension-less URLs: a remote URL without an extension is checked for Zarr metadata before Python is asked.

Tested against

• Point clouds: giswqs autzen-classified.copc.laz (10.6M points) and madison.copc.laz, the USGS IA_FullState EPT, PDAL autzen_trim.las and autzen.laz, plus local LAS 1.2 and 1.4 files.
• Zarr: giswqs noaa-oisst-v2-monthly.zarr and stac-zarr/example.zarr (v3, unconsolidated), the carbonplan 4-D tavg-prec-month pyramid, the NEMO sos_abs pyramid, and local v3, UTM and WRF-style stores.

Tests

• test_map_e2e.py: three new cases cover point clouds in the browser and through Python, the CRS box, and Zarr placement plus dimension controls. 11 pass.
• test_prepare.py: five new cases cover the converter, thinning, the CRS refusal and override, inspect/plan, and the crs action. 32 pass.
• tests/test_map_doc.py and tests/test_map_runtime_contract.py: pass, with a new assertion that the vendored LAZ decoder is used.
• Known flake: test_an_open_map_file_follows_agent_edits_and_writes_its_own is flaky on main too: it failed 2 of 6 runs without this change.

Known limits

• Point clouds draw on deck.gl's canvas, above all map layers, whatever their order in the layer list.
• Classification codes above 31 read as unclassified in files Python converts, because LAS 1.2 can't hold them.

---

Note

Medium Risk
Large viewer and prepare.py changes (new conversion pipeline, Zarr placement) affect layer loading for multiple formats; mitigated by extensive new e2e and unit tests.

Overview
Adds point cloud layers (LAS/LAZ/COPC/EPT) end-to-end: new pointcloud map kind and style/options (color_scheme, class filters, options.crs), browser rendering via maplibre-gl-lidar, and a prepare.py pointcloud action that rewrites unsupported LAS (e.g. 1.4, missing CRS) to thinned LAS 1.2 in lon/lat with caching. MCP, discovery, and the style panel gain point-cloud controls (including CRS entry on failure).

Zarr loading and the style panel are expanded: richer lat/lon axis detection, projected-grid placement via a new crs action, multiscale pyramids, remote listing for unconsolidated stores, per-dimension slice UI (slider, pick list, play), and auto-clim that follows named dimensions. Extension-less remote URLs can be probed as Zarr before calling Python.

Azure blob SAS handling is normalized: unsigned() strips SAS query params for cache keys, signing replaces stale tokens, and download resume matches sources across token rotation.

^{Reviewed by Cursor Bugbot for commit 6f57182. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cursor]

Stale Bugbot comment from a previous run.

[cursor]

Cursor Bugbot has reviewed your changes using default effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit bf2e999. Configure here.}

[github-actions]

Coverage report

Click to see where and how coverage changed

File	Statements	Missing	Coverage	Coverage (new stmts)	Lines missing
fused_render/templates/map
discover.py					86, 88, 90
map_doc.py					146
map_tools.py					214
prepare.py					168-175, 180-185, 256-269, 282-350, 354-355, 360-366, 561-568, 613, 615, 617, 619-620, 639-652, 672, 674, 676, 678
Project Total

This report was generated by python-coverage-comment-action