render/navara.js: one always-on-top engine.movers chevron bank (atlas of
every chevron icon at 32 px), MMSI pick ids, trail on engine.lines,
camera reads through the shared Navara layer view (extracted from the
aircraft renderer to src/layers/shared/render/navaraView.js, + camera
pickEllipsoid and canvas clientWidth). Registered in navaraMode.
QA: vessel qaDrawn; probe selected() falls back to getSelectedInfo;
qa-w1-layers vessel section; qa-vessels-engines side by side; --engine
for qa-vessel-cards/datum; qa-flights-engines military near pose 100 km.
Unit tests on the fake engine pass; browser runs NOT done (see progress).
Reconcile #880 with current main. Delegate the view-state builder while keeping enabled-only layers, disabledLayerCount, provenance, context semantics, and existing voice behavior. Add parity and roster-read regressions; retain package export, boundaries and changelog fragment.
Preserve original GEV voice, layer and packet contributor credit.
Carry reviewed #816 onto current main without conflicts; preserve i18n updates, replace obsolete changelog edit with a fragment, and add a constructor-wiring regression alongside the owner and rail tests.
Credit: raysung-2026 (report), asturksever (review), quanticsoul4772 (Windows diagnosis).
T11 review follow-up to ed9ea7e6. CPU hits rank as the engine draws them:
sets without a depth test over depth-tested ones, then by set order
(the tracked bank over the fleet, flights over local ADS-B), then depth —
so the CPU's top hit that checks a GPU answer is the instance drawn on
top. A model hit is no longer taken as answered: the model mover keeps
each instance's drawn diameter (after minimumPixelSize/maximumScale) and
the CPU checks the hit against its projected position within that radius
(+3 px); the CPU pick sees models at that radius too.
Review follow-up on #975. The shared same-site gate refuses any forwarding header, which is right for routes that spend a key, but terrain is keyless and needed for rendering: behind the reverse proxy the README documents, every /api/terrain/heights chunk got 403 and fell back to the geoid. Terrain now uses admitSameSiteKeyless, which keeps the Sec-Fetch-Site and exact-Origin checks (comparing Origin against a single X-Forwarded-Proto scheme) but does not refuse forwarding headers. The shared admitSameSite gate is unchanged for every other route. The changelog entry moves to changelog.d/975-terrain-cross-site-gate.md.
Co-authored-by: daikaginza <[email protected]>
src/layers/vessels/render/cesium.js holds the Cesium objects the layer
built itself, one for one: the translucent BillboardCollection of
chevrons (centred, screen-aligned, never depth-tested), the horizon
occluder, window projection, canvas size, viewer.trackedEntityChanged,
and the trail (data/trailRenderer's entity polyline, kept on Cesium: the
engine.lines Primitive path did not draw below the photoreal mesh in
round 6 — the plan's T13/T15 blocker; flights does the same).
The logic imports no Cesium: geo math, Rgba WHITE, the chevron artwork in
icons.js, clicks through the engine's selection dispatcher (own-rule
owner, same rule). Registered as CESIUM_LAYER_RENDERERS['ais-live-vessels'];
src/layers/vessels joins the boundary rule.
W1 replay vs 7c7422e2: vessels/flights/military/localAdsb counts, known
objects, screen points, requests and shot sha256 identical.
qa-vessel-cards --data synthetic 2/2 and qa-vessel-datum (?qaFeed=
vessels:fixture) 4/4, the same as 6eec8f6c.
Tests moved with the code (same meaning): cameraHandoff (the click body
is handleClick), inputOwnership (the guard precedes the click rule's
read of the pick), renderers.test (vessels in the extracted list),
importDirections (an unmoved example layer is satellites), the
ownership services fixture (renderer service).
Vessels' click rule (sibling ownership, cyclone background, card hit
test, unkeyed records) does not map onto the plain or look-under rules:
an owner with onClick(ctx) gets every click with the top pick, no gesture
rule (its old LEFT_CLICK handler), and is another layer to the rest.
getNearby returns the contact's label as `id` (the callsign when it has
one) and its key as `icao24`; the drawn rows are keyed by icao24, so the
model lookup missed and every contact was measured at its sprite. Under a
3D model that sprite is hidden and keeps the pre-floor height by design
(T7): the "buried at -27 m" contacts on Cesium (the geoid at AUS) were
those hidden sprites, at HEAD and on 7c7422e2, whose script measured every
sprite. The Cesium report also gives each contact's model state and the
floor cell the layer reads.
AUS, 100 s after enabling flights, both engines: 3 low contacts, 0 buried;
N3752 / DAL848 drawn by their models at 130.2 / 135.4 m over a 123.5 /
128.6 m mesh, the DEM-prior cell warm (123.5 / 128.6 m).
The instance colour (sRGB, as Cesium takes colours: the military amber
#FFB800) was mixed into the linear material colour as is, so a tinted
model drew paler and brighter than Cesium's: the military footprint gate
read 124 % (more pixels cleared the contrast threshold over water) and
the median luminance +9 %. White (flights) is the same either way, which
is why the flights gate passed. The tint is now sRGBTransferEOTF-decoded,
as the sprite sets already decode theirs.
qa-flights-engines' model gate adds the highlights: the p90 luminance of
the model pixels within ±10 % of Cesium, with the near-white pixel count
reported (10b review low 2).
Models on, near pose (computed scale ratio ~1.0000 on both):
- military (from 100 km): footprint 9,902 vs 8,610 px (115 %), median
luminance 158.9 vs 155.3, p90 190.9 vs 184.5 — 18/18;
- flights: footprint 104 %, median 189.8 vs 196.7, p90 223.1 vs 247.8
(near-white px 83 vs 510: Cesium's white models saturate where Navara's
tone mapping rolls them off) — 20/20.
Models off: flights 16/16, military 15/15. Side-by-sides regenerated.
- The DEM prior column read a fresh import of src/data/groundFloor.js, a
module the app does not use for its layers (they read the scene's
surface services), so it was null for every contact. It now reads
window.__godsEyeView.surfaceServices.groundFloor. The terrain proxy
answers 200 and the AUS cells resolve within one or two polls (123-128 m);
the "proxy 503" reading in the task-11 report was this misread.
- A contact a 3D model draws is measured at the model's translation (the
probe's drawn() reports it now): its sprite is hidden and keeps the
pre-floor height by design (T7), so measuring the sprite reported a
buried aircraft that is not drawn.
Navara 0.2.1 can answer a click with the previous gesture's pick. The
check let that answer through when its instance was drawn within twice its
pick radius of the click — the followed aircraft at the centre beside the
clicked one (the 20/24 flights checklist run: a click selected the
previously followed aircraft). A GPU answer for a sprite or point set now
has to be drawn under the click (no slack) and be the instance the CPU
pick finds on top there (lines left out: they are not in the GPU pick);
otherwise the CPU pick answers. Models, which the CPU pick does not see,
are still taken as answered. The CPU drill takes `includeLines`.
Test: 20 gestures alternating a click on A and a click on its neighbour B
answered with A's stale pick — B every time. In the app, 40 scripted
click/follow/release gestures on Navara selected the clicked aircraft
every time with and without the fix (the stale answer is rare).
Task 12b. src/layers/localAdsb/render/navara.js implements the Local
ADS-B renderer interface of 12a on Navara:
- markers: the shared aircraft renderer's glyph sets in magenta #ff4fd8
(class silhouette and scale, ×3 near / ×0.5 far, always on top), drawn
under every sprite layer as Cesium draws the entities; each frame writes
what Cesium's CallbackProperties gave (position, screen rotation,
visibility) and the 2,000 km display distance;
- the 978 MHz UAT ring: its own sprite set, a white ring atlas tinted
#ffb3ef, Cesium's 24 px point plus its 1.5 px outline (27 px), the same
distance scale and distance, pickable as its aircraft;
- models (the Flights GLBs, magenta MIX tint), trail and head: the shared
aircraft renderer's; the camera never follows a local aircraft.
Registered by registerNavaraLayerRenderers (import()); the Navara catalog
now offers flights, military and local ADS-B.
- ?qaFeed=local-adsb:fixture: the local receivers' answer in the server's
format (src/bench/w1Feeds syntheticLocalReceivers: 40 aircraft near JFK,
10 on 978 MHz UAT) through the layer's own decoder-feed session
(sources.localAdsbFeeds, null by default: the layer makes its own).
- qa-w1-layers --layer=local-adsb: a click selects a marker and shows its
card with the receiver line, the camera does not follow, Escape leaves
the selection (it never answered Escape), empty map releases it.
qa-w1-layers local-adsb: Cesium 12/12, Navara 13/13, --compare parity
(global pose: the 2.27 px camera offset only). qa-flights-engines
--layer=local-adsb: 14/14 (fills exact; UAT rings side by side in
docs/migration/phase3/localAdsb).
A sprite/point set uploads its instances by update range. When the camera
went far enough to move the set's origin, update() rewrote every instance
(relative to the new origin) — and a commit later in the same frame
cleared the ranges and kept only its own, so the GPU never got the
rewrite: the instances not committed again drew offset by the origin move
(a military glyph at the orbit pose drew 45 px from its point after the
global → city → orbit sequence; the CPU picks, the probe and Cesium had it
in place). Ranges now add up until three uploads them (it clears them
after the upload).
Found by the military side-by-side (11b); flights' sprites had the same
exposure since 10b.
Task 12a. The Local ADS-B logic (index.js, models.js) no longer imports
Cesium. src/layers/localAdsb/render/cesium.js holds what it built, one
for one: the CustomDataSource of marker entities (billboard with
per-frame position/rotation/visibility CallbackProperties, the 2,000 km
DistanceDisplayCondition, the 978 MHz UAT ring as a PointGraphics), the
model PrimitiveCollection (models load, pose and leave through the shared
aircraft renderer's model part, adopting that collection), the selected
aircraft's trail (data/trailRenderer, as before) and live head segment,
and the camera reads. Registered for 'local-adsb' in the renderer
registry, created at init(viewer, {engine}).
- Clicks go through the engine's selection dispatcher (drill 8) instead of
the layer's ScreenSpaceEventHandler, with the layer's own rule kept as is
— a new `lookUnder` owner kind: its marker on top, or under picks no
layer owns and under trail lines, is selected (and re-selected);
another layer's contact on top, or empty map, releases only its own
selection; a trail line alone does nothing; no gesture rule. To the
other owners it is another layer, as its pick-owner registration made
it before. (The dispatcher's `drillThrough` would have taken the marker
from under another layer's contact, which the old handler never did.)
- models.js reads the camera through the renderer's view (frustum and
horizon tests, ground snap owner) and draws through renderer.models
(loadFleet/add/remove/destroy/setPose: the same MIX-tinted glTF options,
the same HPR frame with the Flights nose offset). The shared aircraft
renderer gains a loadModel seam (default Cesium.Model.fromGltfAsync).
- Cesium math → src/geo; the model tint is an Rgba equal to the old
Cesium.Color; check-import-directions: src/layers/localAdsb/ joins the
rendered layers.
Tests moved with the code, same meaning:
- index.test: the services fixture creates the layer's Cesium renderer and
the selection dispatcher; the click harness scripts the picks on a fake
engine registered for the fake viewer (was a fake ScreenSpaceEventHandler)
and the app's pick owners (trails, a Flights contact) are registered in
the pick registry (was an injected picking service).
- cockpitModels.test: the GLB loads are the aircraft renderer's loadModel
calls (Cesium.Model.fromGltfAsync by default), each asynchronous:false.
- trackingClickGesture.test: the dispatcher's empty click goes to the
plain owners (filter(plain)).
- renderers.test: local-adsb has an extracted Cesium renderer.
New: the dispatcher's look-under rule beside a plain owner (7 click cases).
W1 replay (flights, military, localAdsb) vs a fresh 7c7422e2 run: counts,
known objects, screen points, tracking poses, requests and every shot's
sha256 identical.
On Navara each contact row now carries the surface probe's source and tile
size and the coarse DEM floor (cachedGroundFloor). The buried contacts
(SWA4871 60.5 m and AAL2592 -27 m against a 121-126 m mesh) have no DEM
prior (null: the terrain proxy answers 503 here) and the mesh floor sampler
latches nothing without one, so their sprite heights stay the barometric
fallback; the same rows bury on Cesium when the same contacts are low.
The near-pose finding (Navara models faint, Cesium ~23 bright aircraft)
came from the side-by-side's frozen page clock: the fleet tick ran once per
pose, so Cesium never handed its aircraft to their models and drew 3×
glyphs, against real models on Navara. With the hand-off driven (the move
ends every 10 frames while a pose loads), both engines draw models, and
the comparison showed three real differences, all fixed:
- Orientation: Navara's glTF loader is three's GLTFLoader, which keeps
the scene Y-up; the instance turn is now Cesium's whole axis correction
(Y_UP_TO_Z_UP × Z_UP_TO_X_UP), not Z_UP_TO_X_UP alone. Every fleet model
drew 90° off its heading.
- Size: minimumPixelSize is taken at Cesium's distance (Camera
distanceToBoundingSphere: view depth less the radius at the model's own
scale), in the mover and in ModelHandle.computedScale; the straight-line
distance drew off-centre models up to 16 % larger.
- Colour: the instance colour tints by the set's blend amount (Cesium's
colorBlendAmount) and its alpha is the model's opacity (Cesium: a
model colour's alpha makes it translucent), not a blend weight; a model
is translucent only while faded. An ambient share of the tinted surface
(MODEL_AMBIENT 0.5, Cesium's image-based lighting) keeps a tinted
aircraft near-white at a low sun (median luminance 139 → 190; Cesium 197).
qa-flights-engines (now --layer=flights|military) gates the near pose
with --models=on: computed scale within ±5 % per model (ratio 1.000007),
footprint against the same view with the layer off within ±20 % (1553 vs
1496 px, 104 %), median luminance within ±10 % (189.8 vs 196.7). The probe's
drawn() reports a model's computedScale. Flights models on: 19/19.
Task 11b. src/layers/military/render/navara.js: the shared Navara aircraft
renderer with the military's sprite slot (SPRITE_LAYER_ORDER 'military',
under flights) and jet.glb's nose offset; registered by
registerNavaraLayerRenderers (now a table of lazily imported factories,
flights and military), so the Navara catalog offers military and the main
chunk stays without three/Navara.
- The follow passes between aircraft layers on Navara as on Cesium: the
follow-handle → tracked-handle map and the "follow being started" handle
are shared by every aircraft renderer of the page, so a layer whose follow
another layer took reads the new owner's tracked handle (and its
gevSelectionOrigin) — before, it read a bare follow handle and reported
'programmatic'.
- trackById on a selected-but-not-followed contact follows it again (as
flights since 10b; on Navara any camera command ends a follow).
- qaDrawn (the probe's drawn()) for military, shared with flights
(src/layers/aircraft/qaDrawn.js).
- ?qaFeed=military:synthetic[:count]: seeded readsb rows (1 000 by
default, 150 near JFK, 12 on the ground, 8 types), the first 25 hexes
shared with the synthetic civil flights.
- qa-w1-layers: sections 4–6 read the --layer's module (were 'flights');
section 5 keeps the queries beside the voice answers (the voice block
overwrote them, so --compare never compared queries); for military: the
follow passes military → flights → military, and flights leaves the 25
shared aircraft to military while it is on (9 975 / 10 000 / 9 975).
qa-w1-layers --layer=military: Cesium 25/25, Navara 26/26, --compare parity
(queries included; the known 2.3 px camera offset at the global pose only).
Task 11a. The military flights logic no longer imports Cesium: it draws
through the shared aircraft renderer (src/layers/aircraft/render/cesium.js)
with the military's parameters in src/layers/military/render/cesium.js
(sprite-order slot 'military', jet.glb's 180° nose offset), registered for
'cesium' in the renderer registry and created at init(viewer, {engine}).
The same pure move as flights in 10a:
- Fleet: the BillboardCollection and its sprite-order slot (enable keeps
restoreSpriteOrder, as before); sprites stay Billboard handles.
- Models: fleet/tracked/preload glTF loads (asynchronous:false, MIX tint at
MODEL_COLOR_BLEND_AMOUNT, IR boost as `unlit` → the UNLIT CustomShader),
the model collection, setPose. The per-class specs keep their own nose
offset: setPose takes an optional offsetDeg (default the layer's), on
both aircraft renderers (same arithmetic, test added).
- Tracked: the ENU billboard Entity with its position/colour/rotation
callbacks (#FFD166, transparent amber once the model owns the visual),
viewFrom, followed through viewer.trackedEntity + the one exact tracked
frame; refocusTrackedById through follow.refocus.
- Trail: the entity trail and the floored live head segment.
- View: camera/scene reads, frame number, frustum and horizon tests,
render requests, frame events.
- Clicks and Escape go through the engine's selection dispatcher (the
ScreenSpaceEventHandler and keydown listener are gone); the cross-layer
untrack watches the renderer's follow. Same rules: cockpit clicks are
inert, the tracked visual/model is a no-op, another layer's pick is not
empty space, an unowned pick is.
- Colours are engine-neutral Rgba (MIL_ICON_COLOR #FFB800,
TRACKED_ICON_COLOR #FFD166, AMBER_TRANSPARENT), NearFarScalar →
nearFarScale; Cesium math → src/geo.
- check-import-directions: src/layers/military/ joins the rendered layers
(cesium only in render/cesium*.js).
- The military registry and the flights' duplicate suppression are
untouched.
Source-text tests updated with the moved code, same meaning:
- militaryAwareness.test: military's hasContact guard reads fleet.visible().
- trackingClickGesture.test: the military gesture rules are the
dispatcher's (selection gesture before select, click classifier before
the empty-space clear), military answers the empty click with a
user-origin clear — the same assertion flights already had.
- ownership.test: the services fixture names selection and renderer; the
app-composed services build the layer's Cesium renderer.
- renderers.test: flights and military have extracted Cesium renderers.
New tests: the military logic over a plain-data renderer and a fake engine
(amber fleet, click-select, #FFD166 tracked glyph, no-op on the tracked
visual, empty-click and Escape release, the keyboard left on destroy); the
plain-data renderer moved to src/testSupport for both aircraft layers;
setPose with a per-model offset.
W1 replay vs 7c7422e2 (the 10a reference run): every layer's counts, known
objects, screen points, tracking pose after 2 s, requests and the
city/global/orbit shots (sha256) identical. qa-military-names (it checks
the installations layer): the same verdict as 7c7422e2 (the zoom-markers
assertion fails on both). Main chunk 2,890,763 B (−2.9 KB).
The fire and camera labels in the context store, getStrongestFire() and
getSelectedInfo() were translated at the source, so the voice model got
"Пожар · FRP 12 МВт" or "Cámara ALPR" in track_entity,
get_entity_context and pointer_context while every other entity name in
those payloads is English. Keep them English, as before the interface
was translated; the voice card translates them where it shows them.
The pointer chip, its live-region line and the get_entity_context card
title now translate the English entity labels the FIRMS and ALPR layers
register ("Fire · FRP {frp} MW", "ALPR camera"). Other names, such as
callsigns and place names, show as they came.
summarizeAwarenessCohort() translated `reason`, which is retained in the
awareness context snapshot. Mark the four reasons with msg() instead; the
awareness panel already shows them through translateKnown().
The Local ADS-B layer copied the SDR controller's `message`, which is
Radio card text in the UI language, into its `error` stat. The controller
now publishes an English `errorText` next to the translated `message` for
every receiver error, and the layer reports that. A receiver is any
object with getState(); one that does not publish errorText still shows
its error from `message`, as before. The Radio card text is unchanged,
and the Layers panel still shows the error translated.
The Directions row errors (pointer blocked, no route, routing timed out,
no response from the routing service) and the route request failures
("Routing failed: {error}", "{error} — try again in a moment",
"Routing unavailable (HTTP {status})", the rate-limit and fallback
texts) were translated inside getStats(), so the layer snapshot that
voice tools and the analyst read carried Spanish or Russian text.
Fixed texts are marked with msg() and the interpolated ones are built
in English. The Layers panel translates them where it shows them,
through a ROUTING_ERROR_PATTERNS block in statusText.js that the
proxy's refusal texts move into. English reads as before. In other
locales two things differ slightly: a proxy text inside "Routing
failed: X" that is not one of the known refusals is now looked up in
the catalog instead of shown as given, and the "— try again in a
moment" shape is recognised on every layer's error, not only on
Directions.
The CCTV frame proxy marks a Street View fallback with X-CCTV-Source, but the portable source dropped the header and get_cctv_snapshot always reported a current view credited to the camera's agency. The source now keeps the recognised provenance, and the tool labels Street View frames as historical Google imagery, refuses synthetic placeholders, and reports unknown provenance without claiming a current view.
On Navara the military-awareness module has no controls, so its
releaseCameraOwnership was skipped silently and a city pill, voice
fly_to_location or reset ended the follow but kept the aircraft selected.
The trackers it would release are released directly instead, as on
Cesium (selection and follow both cleared).
destroy() now expires every retry attempt in flight, clearing its deadline
timer and aborting its signal at once rather than when the attempt settles,
and leaves no retry marked as running.
A successful terrain retry overwrote the cached fallback without releasing
it. The replaced fallback is now disposed as soon as the real terrain takes
its place on screen, or, when it stayed installed behind a 3D tileset, once
the next globe activation replaces it.
Inside a Navara follow the orientation controls frame the follow's target
and retarget its viewFrom instead of moving the camera, and the release
cancels only a running flight (a cancel is a camera command, which ends
the follow). Cesium's trackedEntity path is unchanged. Updated test:
cameraOrientationControls.test.mjs (orientation navigation starts a
flight before expecting its cancel; no flight, no cancel).
FollowHandle.target() returns the followed point, checked by the contract
on both engines. The contract's model check now waits for ready() before
looking for the model's pixels: the magenta point and sprites could fill
the region before the glTF loaded (qa-globe-contract cesium 73/73,
navara 79/79).
An Esri or keyless terrain attempt that never settled kept the single-flight
retry open forever, so neither provider was retried again. Each attempt now
counts as failed after 60 seconds and the retry backs off as usual; its
signal is aborted, and a provider that lands after the deadline is released
rather than installed or cached.
CURRENT-STATE now says only the construction fallback to OSM is retried, not
the switch made after repeated Esri tile failures.
An activation that re-showed a fallback while a retry was in flight armed a
second timer, so two retries could run at once and land out of order, leaving
the cache and the screen on different providers. Retries now run one at a
time; a request made meanwhile re-arms once the current one ends, and a
result that is obsolete or no longer wanted is disposed.
Imagery and terrain retries are now handled as each settles, so a stalled
terrain attempt no longer keeps recovered imagery off screen, or the reverse.
Main now takes release notes as one fragment per pull request (#787). The browser asks Google through /api/google/geocode, so the no-geocoder-request check now watches that route.
Kevin Rajan