About 20 sites called scene.sampleHeight, globe.getHeight,
scene.clampToHeight or the MostDetailed variants directly, each with its own
idea of a miss (undefined, NaN, a throw). They now ask heightsFor(scene) in
src/queries/heights.js:
- now(point, { against, exclude }) answers from what is loaded, or null;
against is 'surface' (what is drawn) or 'terrain' (loaded terrain only).
- clampNow(position) moves a position onto the drawn surface, or null.
- batch() and clampBatch() wait for the most detailed tiles: null per
missed point, a rejection when the whole query fails.
- supported() reports { surface, clamp } without allocating.
Cesium still answers now() synchronously, so nothing changes on screen and
nothing is slower: in the live app 1,000 calls cost the same through the
service as direct (terrain 0.5-0.7 ms, surface 2-6 ms a call, run-to-run
noise larger than the difference). Every site already treated a miss as
"hold the last value or fall back"; a renderer that can only answer from a
cache or asynchronously can now plug in at one place.
- A Bhote Koshi evidence point whose terrain sample misses stays at its
fallback elevation instead of reading the miss as sea level.
- The mesh floor sampler skips a scene without depth-texture support
instead of spending its budget on probes that throw.
- The world annotation probe's trace reads hit/missed.
- A source scan forbids direct height calls outside the service.
The service is exported as ./queries/heights and documented in
docs/APPLICATION.md.
Every layer and tool bound its own ScreenSpaceEventHandler and picked the
same click again: with the usual layers on, one click ran about 26
scene.pick and 2 drillPick passes (120-215 ms in NYC with 18 layers). Pointer
input now comes in through src/input/dispatcher.js, one handler per scene.
A click is picked lazily and at most once per kind of query, and every
subscriber reads the same ClickEvent (pick, hit, id, ownerId, gesture,
drill, ground). The same click now costs 3 picks and 2 drills, 17-70 ms.
- Ambient onClick subscribers run only while no tool holds the pointer,
checked once per event; tools subscribe with their lease. Moves and
releases always reach ambient subscribers, so hover and drags end cleanly.
- Camera input and the viewer's stock select/track clicks are ref-counted
holds (suspendCamera, suppressStockClicks). The dispatcher owns the
controller's enableInputs, so a Cesium flight that ends no longer turns
input back on under a drag or the cockpit.
- The draw tool, imagery box, CCTV gizmo, cockpit, directions and every
selecting layer use it; bindTrackingClickGesture is gone (the gesture is
on the event). A source scan forbids new handlers, enableInputs writes and
stock-click removal outside the dispatcher.
- The sensor-style pick correction transit had now applies to every layer.
- Radio's 8-pick near-miss ring is one 17x17 windowed drill.
- A missed pick passed to pickContact is no longer picked again, and a
touch tap's contact search shares the click's drill.
- A gizmo drag whose release Cesium never reported ends on the next press
instead of leaving the camera held.
The dispatcher is exported as ./input with the pointer-lease functions, and
documented in docs/APPLICATION.md.
Aircraft draw from GPU icon sets, so the billboard mirror, the billboard
treatment and the aircraft visual-state module had no callers left. 3D models
are now records the layer holds under the contact's id (modelRecord.js):
show and placement are written through showModel/placeModel and read back
from the record; renderer-measured values stay on record.model.
The tracked-model click check only matches when a tracked model exists, so
an id-only pick is never mistaken for a click on the tracked plane.
The flights icon planning and presentation move into
`createAircraftIcons` (src/data/aircraftIconLayer.js), and the motion
regime both layers dead-reckon from into src/data/aircraftMotionRegime.js.
Flights keeps only its grounded-floor pass on top.
Military now draws every contact as a row of that icon set, like
flights: the billboards, the cull positions and the rotation-pass
signature go. Tracking hides the row and untracking re-plans it from
the tracked course. Queries read the drawn row position and judge the
horizon as the shader does. Snapshots that arrive before init or after
destroy are ignored, and per-contact lookups find no contact then
instead of throwing.
With about 120 live military contacts, preRender p95 drops from
0.7-0.8 ms to 0.2 ms on the same views. Render time and frame gaps are
unchanged or slightly lower.
Every flight is now a row in the layer's icon set. Its motion segment
follows the same regime `_deadReckon` picks (interpolate between fixes,
warm up backward from the oldest, coast forward from the newest), and
one shared `_motionRegime` makes that choice for both. The shader moves
every untracked contact between polls, so the 12 Hz fleet tick no
longer touches each billboard. It now only re-plans expired rows,
floors grounded contacts, advances focus while needed, and runs the 3D
model regime.
Grounded contacts the floor lifts are pinned to a hold at the floored
position. Their nose glides on the GPU. Contacts within 50 km re-floor
every tick, and farther ones take turns about once a second. Low
contacts are judged for the horizon at no lower than 12 m above the
ellipsoid, as before. Below hover speed a contact's nose holds. A row
whose fleet model is drawn keeps its focus updated, because the model
reads it. Untracking re-plans and re-floors the icon from the tracked
course.
On the same views with about 6,700 OpenSky flights, frame gaps at p95
drop from 28-40 ms to 17 ms, and the fleet tick from 16-37 ms to about
2 ms at most.
Icon rows live in typed-array columns: one packed motion segment (hold,
lerp or arc, with a course glide), style, colour and pick ids per row.
The primitive draws every row in one instanced call, dead-reckons in the
vertex shader, matches the billboard treatment the fleet layers use, and
picks and drill-picks like a billboard. Cyber sonar covers icon sets.
Installation selection fills and outlines record the installation they
belong to in the layer's installationMarkers store instead of an
installationId expando, and the click handler resolves them from there.
Named-marker points keep their { installationId } pick payload, which is
the app's own object, so sibling layers treat a point click as before; the
marker remembers its record id for visit() instead of reading the point.
The local GeoJSON package tracks the entities each layer loaded in a
per-instance set instead of a __localLayerId tag, and passes the stem
base to registerEntityContext as targetPosition instead of hanging it on
the entity; FIRMS does the same, and voice targeting reads the context
record. QA scripts find a layer's features by their context id.
The changelog fragment named share-link token 6 (sea marks); the layer
reserves 8. CURRENT-STATE said voice could toggle and query the layer and
listed it among the analyst record sources, but it is in VOICE_OFF_LAYERS.
Say voice is off for now, and that the layer covers FAA prohibited and
restricted special-use airspace only, not the UAS Facility Map / LAANC grids
or TFRs. A test pins these claims to the token ledger and the voice manifest.
Earthquake analyst records come from the events the layer drew, gated on
its own enabled flag, instead of reading each entity's position and
property bag back; the discs no longer carry the bag. CCTV coverage rays
and projection planes record their camera id in a layer pick store
instead of a cctvCameraId entity property, with the same ownership checks.
The tracked flight, military and local ADS-B billboards record their size
and distance scaling as the entity record's sprite, which the close-range
camera guard and the screen annotations' tracked footprint now read in
place of the billboard's property getters.
Flights, military and local ADS-B now write their models' show and world
matrix through setModel in src/data/modelMirror.js, and read them back
through modelState: the fleet and tracked handoffs, the tracked visual and
trail anchors, detection anchors and modelAnchorWorld. A matrix is copied
into the app's record and into the model's own matrix, so the shared scratch
can be passed directly and models still never share one matrix object.
What Cesium measures (ready, computedScale) is still read from the model,
and colour stays live as on billboards, because cyber sonar writes it too.
Test seams mirror the models they install, as they do billboards.
From the whole-globe view the camera could end up facing empty space with
only stars on screen. Cesium's free look (Shift+drag) turns the camera in
place, and from 18,000 km the planet's edge is only about 15 degrees from
nadir, so a few drags turned it out of frame. The rolled pose was then
saved into the share hash, so a reload or shared link opened on stars too.
- Allow free look only below 1,000 km. Orbit, tilt and zoom are unchanged.
- Measure how much of the frame shows the planet's surface by casting a
32x18 grid of view rays at the WGS84 ellipsoid. While the scope mask is
on (the default) only the rays inside its keyhole count, since the mask
hides or dims the rest of the frame.
- When a camera move ends above 1,000 km with no ray on the surface, turn
straight down at the same position and heading. Limb glow and a sliver
up to about one grid cell thick can count as out of view. The end of a
move comes from Cesium's moveEnd and from a check of its own that
samples the pose every 250 ms after moveStart and acts once it has held
for half a second: at some of these poses the camera drifts by a
rounding error every frame, so Cesium never sees it stop and never
raises moveEnd. The zoom-out guard from #1522 keeps wheel and pinch
zoom-outs in view; this catches the other ways a move can end looking
away, such as a saved or shared view that did, or a
drag that Cesium turns into an in-place look below its 7,500 km
trackball height (one that starts on the stars, or a tilt drag with the
screen centre off the planet). Views with more surface, tracked or
locked cameras and the 2D and Columbus modes are left alone.
- Voice tilt and in-place turns of a free 3D globe camera above 1,000 km
stop when about a tenth of the frame still shows the planet (or about as
much as it showed when it started, two grid cells of sampling noise
allowed), never below about 1% of the visible frame; a view that already
shows less only keeps or gains what it has. A turn that cannot go half a
degree further without crossing that limit is refused with a reason
instead of reported as done, and a turn it accepts turns at least that
half degree. A turn that keeps or gains surface is not refused, even
after the window has been resized, except that near the limit, where the
planet's edge crosses the sample grid, sampling wobble can occasionally
refuse one (most often above about 15,000 km and on phone or ultrawide
frames); nothing then turns away from the planet. A turn that starts
where the last voice turn left the camera keeps that turn's limit, or a
higher one if the planet now fills more, so repeating the request does
not take any more; once anything else has moved or turned the camera, a
turn is measured from what is on screen. A request made while the camera
is tracked is judged by the pose voice's release of the follow camera
will leave, before that release, so a refused turn keeps the tracking.
The frame measure and the look-step limit live in src/globeInView.js,
shared by the viewer and the voice camera verbs.
The three layers now add their billboards through addBillboard and write
the mirrored fields through setBillboard, so selection, cards, queries,
focus, view thinning and the FIRMS horizon cull read what was pushed from
the app instead of from Cesium. Vessels and FIRMS state their sprites' own
sizes (32 px ship artwork, the glow canvas at twice the core size), which
the mirror needs. A vessel helper, iconPosition, replaces the repeated
mirror-or-visual position lookups, and the vessel visibility pass writes
show only when it changes, as CCTV and FIRMS already did.
A billboard never given a position now has none in the mirror rather than
the origin.
Flights and military now add their billboards through addBillboard and
write show, position, scale, rotation, size and distance scaling through
setBillboard, which keeps the app's copy in src/data/billboardMirror.js.
Their reads go to billboardState instead of the Cesium billboard, so the
layers no longer read back what they pushed to the renderer.
A mirrored billboard needs an explicit width and height, because unset
Cesium reports the loaded image's size. Colour is not mirrored yet: the
cyber-sonar theme writes it too and tells the writers apart by the live
value, so it moves with the effects port.
Re-pin the Realtime tool payload hash for the merged tree (voice geometry plus pitchDeg). To fit main's 55,500 prompt ceiling with Overpass on, the pitchDeg schema drops its -90..90 range (the runner already clamps to -89..-5, now pinned by a test) and its description is shorter. The Overpass prompt measures 55,492.
Submarine cables wrote each pick's info both to a per-load WeakMap and to a
__gevTeleGeography expando, always together; one cablePicks store replaces
both and resolves the same objects. The voice view-target prewarm marks its
viewer in a WeakSet instead of a __gevViewTargetPrewarmInstalled property.
The aircraft layers' model spec key, IR boost and belly offset and the
billboards' large-icon flag move into aircraftVisuals, one store shared by
flights, military and local ADS-B. Installation markers' last drawn height
and selection move into installationMarkers. The CCTV viewshed QA tag is now
the viewshedCameraId record field, which qa-cctv-v2 reads through
entityRecords. Test fixtures drop a placement flag nothing reads.
Picks up #1527, which restores the 97 format-scope entries an earlier merge
dropped; this branch had kept the same 97. The resolved list is main's plus
this PR's 12 test and helper files, 1,222 paths, all present.
The guard's own tests pass even if createApplicationControls stops
installing it. This checks the ceiling and frame listeners are in place
after construction and gone after teardown.
Merge commit 34906c9a resolved the end-of-list conflict in scripts/format-scope.json with the branch's side, which dropped the 97 entries main had appended since Oct 26. Those files all still exist; 64 of them are tests that runtime discovery does not cover, so format:check had stopped checking them.
Each entry goes back at its position on main before the merge. All 1680 files pass npm run format:check.
Picks up #1308 (Overpass admission), #114 (Sea Marks and Ski Pistes),
#1206 (radio country coverage) and #1435. Resolutions:
- SECURITY.md: main's rewritten cross-site paragraph, with voice place
search listed beside /api/overpass among the keyless map routes.
- scripts/format-scope.json: both sides. A merge on main dropped 97 entries
(being restored there); keeping them, this PR's 12 and main's 2 new ones
keeps every file in format:check. All 1,222 paths exist.
- Translation catalogs: main's two new keys and this PR's 47, no overlap.
- Layer count 36 (33 + OSM Places + Sea Marks + Ski Pistes); both schema
pins re-derived, the first-run length is exactly 30,919 + 292 + 107.
Typed chat sessions no longer end on the voice inactivity timeout, and a
failed or stopped reply keeps the conversation: Stop cancels only the
reply in progress, and map actions that already ran stay in the model's
context. Each upstream request and map action gets its own deadline, old
tool results shrink instead of the history overflowing, and step limits,
a missing or rejected key, the OAuth lock and throttling each get their
own message.
- Server: OpenAI chat is off under GEV_PREFER_CODEX_OAUTH; chat throttles
default to the provider's throttle; replies carry API_KEY_MISSING,
API_KEY_REJECTED, CODEX_OAUTH_REQUIRED and RATE_LIMITED codes; upstream
Retry-After passes through, and a rejected key (Gemini's invalid-key 400
included) cools down for a minute
- Pinokio forwards the chat throttle settings
- UI: the reply reveal publishes only the message list; the border orbit
runs only while busy; layers attach with a button or Enter; the log is
aria-busy while a reply types out; auto-scroll respects a reader who
scrolled up; Escape closes the attach menu; the chat gear says Settings;
a note says chat bills the OpenAI key when voice uses ChatGPT sign-in
- Layout does less work while panel height animates; unused blurs removed
- Gemini voice start ends chat; the voice preview clears on every new
session
- Tests for the route gates, client limits, attachments and the fixes
The ALPR layer's private per-entity fields (record, clamp, canvas and display
anchors, wedge, native appearance, pick position, shown, surface check) move
from gevAlpr* expandos into alprMarkers, a store made by the new
createEntityStore(). The shared entity records are built on the same factory.