Merge main (87ab9d2b1, #1758 and the #1757 revert) into the renderer branch

Grounded aircraft turn 3D on the first settled frame: streaming tiles are not a miss. groundSnap's tests keep main's cases against the branch's query port; trackedModelForm.test.mjs (the branch's form of trackedModelRegime.test.mjs) takes main's edits.
This commit is contained in:
Sameh Khamis committed 2026-11-15 18:42:06 -08:00
commit 4b0958b9e8
5 files changed
+111 -56

No files matched your search

+15 -7
View File
@@ -11,11 +11,16 @@
* Discipline (CCTV-B9b style): scene.sampleHeight forces a synchronous offscreen
* pick render, so it is NEVER called per frame. Each grounded modeled plane gets
* ONE successful sample, cached per icao; the cache stops answering directly when
* the plane moves >MOVE_INVALIDATE_M from the sampled spot (taxiing). Misses (tiles
* still streaming, OSM fallback with no tileset skin, sample outside loaded
* geometry) are a retry-later signal with per-icao exponential backoff. A
* per-window budget stops a freshly modeled airport cluster from firing dozens of
* pick renders in one fleet tick.
* the plane moves >MOVE_INVALIDATE_M from the sampled spot (taxiing). A sample
* that misses (OSM fallback with no tileset skin, sample outside loaded
* geometry) is a retry-later signal with per-icao exponential backoff. Tiles
* still streaming is not a miss: no sample was taken, so it earns no backoff
* and the plane asks again on the next tick, sampling as soon as the tiles
* settle. Counting it as a miss had every grounded plane in view back off to
* RETRY_MAX_MS while the camera moved, then wait out that backoff after the
* tiles had loaded, so models replaced icons late and in batches. A per-window
* budget stops a freshly modeled airport cluster from firing dozens of pick
* renders in one fleet tick.
*
* What a caller gets while a sample is outstanding depends on whether this
* contact has ever HAD one, and the difference is the difference between two
@@ -124,7 +129,7 @@ const HELD_SNAP_MAX_DRIFT_M = 250;
* alternative because it errs UPWARD and stays visible, where a buried model
* errs into the terrain and cannot be seen at all. */
const HELD_SNAP_CONTRADICTION_M = 5;
/** First retry delay after a miss / tiles-not-ready (ms); doubles per consecutive miss. */
/** First retry delay after a sample finds no ground (ms); doubles per consecutive miss. */
const RETRY_BASE_MS = 2000;
/** Backoff cap (ms) — on the OSM fallback (no tile skin) sampleHeight misses forever. */
const RETRY_MAX_MS = 30000;
@@ -327,7 +332,10 @@ export function createGroundSnap({ groundFloor }) {
entries.set(icao, next);
return heldSnapM(next, surfacePos);
};
if (!_tilesReady(viewer)) return miss();
// Tiles still streaming: no sample is taken, so no backoff is earned (as
// with the budget above). The next tick asks again, and the first one after
// the tiles settle samples.
if (!_tilesReady(viewer)) return heldSnapM(entry, surfacePos);
windowCount += 1;
const at = toLonLatH(pos);
const exclude = Array.isArray(exclusions?.sets)