PointClouds 1.0#152
Open
SimonDanisch wants to merge 16 commits into
Open
Conversation
lazarusA
reviewed
Sep 26, 2025
lazarusA
reviewed
Sep 26, 2025
| *.osm | ||
| london_drive.json | ||
| .qodo | ||
| !docs/Manifest.toml |
Collaborator
There was a problem hiding this comment.
I will test locally what's wrong with docs 😉
Member
|
Ok, I figured out the issue here. It turns out that you can't run |
Member
|
Looks like the image on CI is completely whited out, somehow? https://makieorg.github.io/Tyler.jl/previews/PR152/examples/map-3d#PointClouds |
Member
Author
|
Yeah looking into a fix right now! |
Member
|
Looks like the image now has a bit of color :D |
Member
Author
|
haha, what? :D |
The docs build currently fails because the registry GLMakie crashes the GPU with a hard context loss when rendering point clouds via `markerspace = :data` (FastPixel branch). The fix lives on the `sd/fastpixel-shader-fix` branch of MakieOrg/Makie.jl. Pin Makie and the two backends used in the docs to that branch so CI picks it up; this can be reverted once the fix lands on master and is tagged.
Pkg validates that every [sources] entry has a corresponding [deps] entry; without it `Pkg.instantiate` errors with "Sources for `Makie` not listed in `deps` or `extras` section". Makie was previously pulled in only transitively via CairoMakie/GLMakie.
Commit 20a0c69 ("trying to make loading more stable") removed the `wait(m.tiles; timeout=timeout)` call from `Base.wait(::AbstractMap)` together with `wait(map)` inside `Base.display`. The display change is intentional (inserting plots while the screen is still initializing crashes some AMD drivers); the wait change was incidental, and broke the existing "Tiles counts" tests which assert that the tile queue is empty and that the full set of prefetched tiles has been fetched once `wait(m)` returns. Restore the cache drain, but place it *after* the foreground-plot loop so the screen is already fully initialized by the time any extra plots are inserted — keeping the original AMD mitigation in place. The expected `fetched_tiles` counts were also stale (the prefetch behaviour has evolved): updated to the current deterministic values (48 → 71 for default OSM London, 71 → 75 for Google).
…olish The headline change is SSETiling3D: a Cesium-style screen-space-error quadtree replacing the single-zoom Tiling3D, used as the new default for Map3D. Tiles near the camera refine deeper while distant tiles stay coarse, giving uniform on-screen pixel density across the frustum and making the 3D map viable at low altitudes without overfetching at the horizon. Supporting pieces: * tile-fetching.jl: SSETiling3D itself; AABB-vs-frustum culling using per-tile (xmin..xmax, ymin..ymax, zmin..zmax); recursive descent from min_zoom tiles found by ground-plane frustum intersection rather than Tile(0,0,0) (the latter fanned out to billions of calls because low-z AABBs straddle every frustum plane). Bug fix in Tiling3D's distance measure (`norm(scalar .- vec)` broadcast nonsense → `norm(p - eye)`) and a near-plane lower bound that stops AMD's depth pass from crashing on absurd far/near ratios. Tile-queue push order reversed so foreground tiles drain before halo/background under the LIFO worker. Halo2DTiling defaults retuned (depth 8→2, halo 0.2→0.4). * provider/shared.jl: shared per-tile (zmin,zmax) cache + binary sidecar format used by ElevationProvider and GeoTilePointCloudProvider so SSE/culling can stop falling back to the global Mariana-to-Everest z range once tight bounds are known. ancestor_or_default_bounds() walks up the tile pyramid for nearest-known bounds. * provider/elevation: tightens bounds_cache from each fetched LERC tile; periodic mv-on-write checkpoint to disk. * provider/pointclouds: same hookup; AHN1 ships with a tight Netherlands default (-15..350 m) which prunes intermediate-LOD recursion ~150×. * map.jl: progressive-refinement hook — when a freshly arrived tile tightens provider bounds, re-fire update_tiles!() at a throttled rate so the view keeps refining while sitting still. Pan/zoom throttle 200ms → 100ms. * tile-plotting.jl: 3D move_z uses depth_shift to layer overlapping LODs (eliminates the speckled boundary between zoom levels) and the 2D path stops hiding higher-z tiles on zoom-out (eliminates the visible blink while finer tiles cull). * cam3d.jl: cursor-anchored pan and zoom (Cesium/Mapbox style), with per-frame translation capping so horizon-grabs can't teleport the camera. * map3d.jl: default fetching_scheme Tiling3D → SSETiling3D.
Member
Author
yay :)
|
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Weirdly enough, GLMakie crashes the AMD driver on linux/windows.
I have a ryzen 395+, which is pretty new, so could be AMDs drivers fault.
WGLMakie does work though.