osm basemap: degrade sub-pixel shapes to dots, and stop compositing toward black

[AdaWorldAPI]

Follow-up to #127. Two independent causes of the black core in the z12 overview, both reported from a live Berlin view where the outskirts rendered correctly — which is the clue that made it two problems rather than one.

1. Edges that cannot resolve

At z12 a Berlin building is a fraction of a pixel. Its outline was still stroked, and stroke colours are dark by design (other = #39424f), so in the dense core thousands landed on the same pixels and saturated them — while the fill, the colour that carries the meaning, never survived. Density made the map darker, which is backwards; the outskirts stayed legible only by being sparse.

dot_if_subpixel measures each simplified shape in the same 2^(z−24) world-pixel units encode_tile_bin projects with, and below 1.5 px collapses it to one centroid point carrying its class.

• No wire change — a dot is npoints == 1, which both client parsers previously skipped at n<2.
• A resolution rule, not a zoom rule — a no-op once a shape exceeds a pixel, so no zoom branch, and it disables itself as you zoom in. The under-budget city view is byte-identical.
• Client, both paths: GL gains uDot[10] + uMode==4, drawn between the fill and line passes so road structure stays legible over the dot texture; Canvas2D gains a dots Path2D map in the same slot.
• The dot takes the class fill colour, falling back to stroke for the fill-less line classes. Without that fallback road/rail/other would be transparent dots — recreating the exact disappearance being fixed.

2. Alpha over a dark ground

Every fill was dark and alpha'd (…80, …4d, …66), and #map's background — which is the land, since GL clears transparent and Canvas2D only clears — was #11151c. Each fill therefore composited toward black rather than replacing it: one polygon read as a faint tint, a stack of them converged on the ground colour.

Opaque fills also restore FILL_PASS's back-to-front ordering as the thing that decides what wins — what that ordering was written for, and what alpha was quietly defeating.

Palette is now opaque and OSM-Carto-shaped over #f2efe9 land: blue is water, green is vegetation, beige-grey is built.

Verification

• cargo check -p cockpit-server — exit 0, no new warnings.
• The embedded JS extracts and passes node --check (936 lines). It lives inside a Rust string literal, so nothing in the Rust build can see it — this gap is worth knowing about.
• Both shaders compile and link in real headless WebGL2 (swiftshader), with uFill/uStroke/uDot/uMode/uView/uTrans/uPointPx and aPos/aCls all present.
  • The first run of that check reported a false PASS: it grepped the dumped DOM, which also contains the unexecuted 'PASS:' literal from the script source. The result is now read from the output div only. Recording it because a detector that can't distinguish success from failure is the failure mode worth naming.
• Not verified: the live visual result. Dot density and the palette want a real browser look, and the 1.5 px threshold is a first estimate, not a measured one.

Draft pending that visual check.

---

Generated by Claude Code

[cursor]

Bugbot couldn't run - usage limit reached

Bugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit.

A user or team admin can review and increase usage limits in the Cursor dashboard.

(requestId: serverGenReqId_7047aeb8-60e0-4b31-a96a-c1d554e4066f)