The picker no longer keeps a hidden, disabled copy of the opening camera.
That copy could be re-picked with ArrowDown on a closed list, could show
twice in the iOS picker, and left Firefox without an accessible value.
The list now selects the camera's own option, as before. A native list
sends no change for the option it already shows, so FOCUS (or NEAREST,
when that camera is the nearest) keeps the shown camera. Closing the list
without a choice keeps automatic selection.
While a user pick kept first claim on the slot of an empty automatic
day, a click or an arrow key onto a card already confirmed empty dropped
that pick but left the hand-over waiting for the next probe result. Until
then the map stayed blank and the Esri lease stayed held, where main
showed the next ranked day.
focus() and preview() now hand the slot over at once when no pick is
left. An arrow key onto an empty card no longer follows that day, as a
click already did, so focus stays on the card the user chose.
The previous change skipped the hand-over for good while a user pick was
pending. When that pick then probed empty, failed with a terminal error,
or lost its source, nothing previewed and the slot stayed blank, where
main would have draped the next ranked day. During the arrow-key
debounce the empty slot also handed the Esri lease back and took it
again 250 ms later.
The hand-over is now deferred: the pick keeps first claim on the slot
while its probe can still settle, and the next ranked day takes over as
soon as the pick turns out empty, errors or loses its source. Turning a
source off also drops a followed pick of that source. The layer keeps
the current basemap while a deferred pick waits, and Escape releases it.
Six tests cover both answer orders, the source toggle, a terminal error,
the debounce and the lease.
The centred Display/Radio row now applies only inside the phone shell
(720 px and below) on flight and bridge views; satellite and provider
cockpits, narrow short windows (under 390x450) and very short windows
(under 240 px) keep main's right-anchored place, so the row no longer
lands on the desktop rail, orbital zoom, LOOK AHEAD, Reset or the Stories
button. The row sizes to its launchers instead of spanning the width.
The first-visit Stories announcement stays hidden and unread while
Cockpit is open: a Stories tap that Cockpit refuses no longer marks it
seen, so it shows again after Exit.
The unit tests and the Cockpit QA cover the phone boundary, glyph and
side taps, the destination cue, short and narrow windows and the
announcement's return.
- The /frame handler reads the client's network key (IPv6 /64) before it
awaits the camera, so a client that drops the connection is still
charged to its own window instead of the shared fallback key.
- The budget is charged only after the Google key and finite catalog
coordinates are confirmed, and the key is read once.
- Places and the CCTV window share one default (GOOGLE_DEFAULT_PER_MIN).
- Tests: the catalog-only test checks heading/fov/pitch from the query at
the catalog location; the budget test covers unknown ids, forwarded
requests and /64 grouping; a real-socket test covers a dropped
connection. The duplicate pose test is removed.
- .env.example, SECURITY.md, CURRENT-STATE and the changelog fragment
describe the separate per-IP window, the catalog position and the
placeholder for seed cameras.
A point of an arrow, route or area outside the view frustum was not
sampled and stood in at 0 m, so a partly visible arrow could hide, point
at the wrong place or vanish behind the eye. Such points are sampled again
within the 32-sample budget and the one-second retry, as on main; lone
points outside the view still skip sampling. The 0 m stand-in exceptions
are removed. Also from review: shared viewPlanes for transit and the
renderer, a −430..9000 m segment test against the view planes, tighter
tests and shorter docs.
Sample frames with Director's normalized shots and runtime timing, count
frames with a tolerance bounded by float rounding, coerce legacy numeric
text, open the input without blocking, strip a BOM, remove a partial
output on failure, and escape control characters in errors.
Tests now check Cache-Control on the proxy's own 400, 405, 429 and
transport 502, a geocode 503 with ZERO_RESULTS on both geocode routes,
and an upstream 429 with an HTML body on both Places routes; the geocode
status cases move to the geocode suite and cover reverse-geocode too.
An unreadable Places 200 now throws into the existing catch instead of
repeating its 502 reply. Comments and the CURRENT-STATE rule are shorter
and match the code.
The separate layer test built a fake viewer with the old Cesium interface,
which fails on current main. Move it into src/data/infrastructure.test.mjs on
the fake ground renderer and input port, with one table and one parse of
dams.geojsonl for both checks. Compare source_url exactly and check
retrieved_at, and cover the dams attribution and both new records in the
find_infrastructure tests.
Also trim the dataset README section (keep both dams as polygons), drop the
outdated "does not refresh or relicense" sentence, keep provenance out of the
License cell, rename the changelog fragment after the change, and restore the
accent in the test name.
- The flight after a count click checks the frame on both axes and at the
height of the highest mount, so a group on a slope or in a very wide
window stays in view; when it cannot, it flies to the nearest cell with
two or more cameras, never to a lone camera that joined the count.
- A count takes a click over a background fill no layer owns (earthquake
discs, fire perimeters), and still yields to a scene item another layer
registered.
- A count crossing the HUD edge refreshes the view only when the active
icon or a lone camera depends on it, and never mid-motion, so panning no
longer runs extra full view passes.
- A count restored at the HUD edge keeps the lone cameras its outline
covers, and leaves the others as icons.
The picker preselected the active camera, and a native select fires no
change for the option it already shows. So while CCTV was still following
the view, choosing the camera the picker showed did nothing, and the next
pan switched to another camera.
The layer now reports whether the active camera is the opening one
(activeCameraProvisional). While it is, the picker shows that camera from
a hidden, disabled first option, so choosing it from the list is a change
and ends automatic selection; closing the list keeps it. A same-camera
pick that ends the opening pick notifies listeners so the picker drops the
hidden option, and reopenNearestCamera passes { opening: true } so its own
notification already carries the opening state.
docs/CURRENT-STATE.md now says when automatic selection ends. Tests cover
the hidden option, confirming and dismissing the picker before a pan, the
notification, and focusNearestToPoint, which is now on main.
A card click on an unprobed day follows that day while the automatic
preview stays on the map. If the automatic day then probed empty, the
hand-over to the next ranked day overwrote that follow, so neither the
picked day nor the successor draped when their probes came back
present. The hand-over now leaves the slot to a day the user picked that
is still being probed or waiting out the arrow-key debounce.
Also covers the off-screen successor with empty days shown: once the
visible days probe empty, the successor's probe is queued (the re-request
that came with the merge from main) and the day drapes when present.
Main moved the screen annotation renderer onto the renderer ports, so the
height culling now goes through them too: the frustum test uses
frustumPlanes/sphereOutside from src/geo/frustum.js on the camera pose,
the horizon tests use the query port's isOccluded, and the file imports no
Cesium. The sampling tests run on main's scene doubles with their
assertions unchanged; the outside-frustum step turns the test camera
instead of stubbing the culling volume.
An endpoint outside the frustum is not sampled and projects at its 0 m
stand-in height. The horizon test then judged that stand-in, so an arrow
whose far end lies beyond the sea-level horizon but above it at its real
height disappeared with its label. Such an endpoint now skips the point
horizon test; the height-range test still drops endpoints that are
provably hidden, and settled or in-frustum anchors keep the point test.
Adds a regression with a real frustum and occluder (camera 100 m up, far
end 56 km away on 100 m terrain). CURRENT-STATE and the changelog now say
the 32-sample budget applies per projection pass, and the redundant helper
comment in the test is gone.
The fallback keeps the catalog position but takes heading, field of view
and pitch from the request again, as before, clamped as before. A catalog
camera with no pose (every world-cameras row on main) otherwise asked
Google for heading 0, pitch 0, FOV 80 while the browser draws it at the
id-hash heading, pitch -17 and FOV 74, and a calibrated camera lost its
calibration. Cost and the per-IP budget are unchanged.
- The Street View budget is built per proxy, so tests no longer depend on
GEV_RATELIMIT_GOOGLE_PER_MIN in the shell; the fixture pins it.
- New tests: a pose-less catalog camera, and the per-IP budget (no call
without a key, the cap, a separate budget per address).
- The client no longer sends lat/lon on /frame.
- CURRENT-STATE, the changelog fragment and SECURITY.md say the fallback
has its own per-client limit, and that unlisted (seed) cameras get the
placeholder.
Each Tarktee DATEX feed now reads its body as soon as its headers arrive, so a
failed or oversized body aborts the other request at once instead of waiting
for its headers up to the shared 30 s deadline. Status and redirect errors
name the feed again.
Ontario failures log the error class and system code, never the key-bearing
URL or upstream text. DATA_SOURCES.md and the camera licence field use the
Open Government Licence – Ontario statement the credit shows, and a test pins
that credit.
The focus-release rendering test now uses main's icon renderer and its
isVideoFeedCamera contract. The Tarktee cancellation tests disable the 30 s
deadline and time out after 2 s, so a regression fails fast instead of passing
late or hanging; a duplicated older test and an unused import are gone.
This PR's notes move into changelog.d/1266-cctv-shared-fix-batch.md: the 1133
fragment described main's tracking probe and terrain check, not this branch.
Above the compass block the centred row covered the DEST route-direction
label and chevron at 320-390 px, and its fixed +182px offset pushed it off
the top of short landscape screens. Place it between the roll scale and the
reticle from a viewport-aware visor inset instead, keep it on screen below
the vision controls on short landscape phones, and centre Cyber's expanded
panel on the Reset/Exit axis.
The unit test now asserts the pointer-events contract and Cyber centring;
the browser QA adds the DEST cue to the clearance check, hit-tests each
launcher glyph and covers 740x360 and 667x240.
Acaray was stored as a LineString, which the Dams layer skips: it got no
marker, label, card or analyst record. Store the closed way as a polygon
(counter-clockwise ring from the same nine nodes) and add a layer test for
both Paraguay dams.
Also tidy the dataset README (drop the landing-order notes and the analyst
record count, note why the Yacyretá marker sits over the reservoir), keep one
credited changelog bullet, credit the OpenStreetMap API next to Open
Infrastructure Map, and drop "other" from the Three Gorges mission.
Every Google route now sets no-store right after the same-site gate and
switches to private, max-age=300 only on a confirmed answer, so the
proxy's own 400, 405, 429 and 502 replies are no longer left without a
header. Both Places routes answer an unreadable 200 body with 502 instead
of caching an empty result, as the geocode routes already did. The test
pins the allowlist with OVER_DAILY_LIMIT and UNKNOWN_ERROR and covers the
unreadable and empty Places bodies; the changelog fragment is shorter.
- Join a space-separated negative latitude to --origin before parseArgs,
so `--origin -33.86,151.21,0` works as the usage and docs show.
- Read the input through one descriptor, regular files only, at most one
byte past the 5 MiB scene limit.
- Use geodetic ENU axes at the origin (ellipsoid normal at its lat/lon,
raised to its height) instead of Cesium's frame at the raised point.
- Reuse the shots already resolved, serialize the sampled shot table
directly, and drop the typeof-only export test.
- Docs: timing matches Director seeking, with the playback differences;
add a manual OpenUSD/Blender check; shorten the overview and test list.
- Name the G3 record "AutoVu SharpV G3" and the SharpX record "XGA", so a
bare SharpV or SharpX tag shows no specs without a separate wrapper.
- Reuse the CCTV layer's Bosch MIC-7522-Z30 record instead of a copy.
- Drop the brand from the SPEC line when the equipment line already names
it, so long lines fit the card; pass the summary to entity properties.
- Fall back to the OSM `brand` tag when `manufacturer` is missing.
- Reject non-string or overlong model/manufacturer values before matching.
- Remove the unused `brandAliases` parameter; shorten JSDoc and comments.
- Cover the tile `model` path and the megapixel summary in tests; fold
duplicate tests into existing ones.
- Docs and changelog: the default hourly extract has no `model`; any
source that supplies it shows the SPEC line.
Main released #1228 and #1233 and deleted their changelog fragments, so they
stay deleted here. Main also rewrote the tracking arrival probe in
scripts/track-regression.mjs, which supersedes this branch's version; main's
file is kept unchanged.
- A count click stands aside when an overlay label or another layer's card
already took the click, and CCTV now claims a count click in the shared
pick registry, so the flights layer no longer reads it as an empty-map
click and keeps tracking its aircraft.
- The flight to a group halves the distance only while the group stays in
frame; otherwise it flies to the nearest ground cell of the group, so a
tilted click no longer ends with no camera on screen.
- Counts hidden under the HUD no longer push the active icon aside, and a
lone camera that would move a merged count under the HUD leaves the
original count and its icon as they were.
- Cameras on one mount are spread before counts merge, so a spread icon no
longer slides under a neighbouring count.
- The ground module comment says how a camera without a batch height takes
the drawn terrain.
- When F or a panel's collapse button folds a panel that holds the focus,
the focus moves to that panel's visible rail entry. The same happens when
opening Data Layers folds a peer panel that held it, and in Cockpit the
focus moves to Contact's expand button when that card folds.
- The `shownBy: 'active'` rail gate is gone: nothing removes `.active` from
Data Layers any more, so a rail entry now follows `panel.hidden`.
- CURRENT-STATE says what F does, and the panel catalog comment no longer
says F toggles `.active`.
Bugs:
- Cameras on one mount stay side by side after their geometry is built,
so a click or a settled view no longer folds them back into one icon.
- A count click yields to a card drawn over the marker and to another
layer's object under it, waits while the cockpit or tracking holds the
view, and flies at most halfway, so a long chain of cameras still splits.
- The active camera's icon is never covered by a count, transparent
corners of the marker no longer catch clicks, and a count hidden under
the HUD keeps its lone cameras.
- A lone camera joins only its nearest count and never bridges two; cameras
close enough to draw as icons take part too, at their drawn icon size.
- Groups across the antimeridian keep their longitude; the flight frames
the highest mount; markers on ground below the ellipsoid stay visible;
counts fade outside the Scope circle.
- Mounts less than a metre apart count as one mount.
- A drawn-terrain height survives a late fallback answer, is read for an
unresolved camera and for one projected off the window, is re-read when
the terrain under it finishes loading or the terrain source changes, and
the read budget no longer goes to the same misses each pass.
Cleanup: the glyph draws from one green SVG sprite instead of recolouring
pixels, marker-only constants move next to their use, layout tests check
behaviour rather than formulas, and docs and comments match the code.
docs/I18N.md names the one fixed notation of count badges as an exception.
- The first height that replaces the entry frame's now moves only the
anchor's height. The anchor keeps its place and its eased lift, so the
camera no longer jumps to a corrected fix or steps back.
- A detail line is measured with the title's glyph widths. They are a
little wide for its smaller type, so a long detail line is no longer
clipped.
- The first aircraft update is checked in the existing test, and the copy
is removed; test fixtures are simpler.
- Comments, the release note and CURRENT-STATE match the code.
- A clock correction starts a new epoch, and playback may first draw its
second report. The turn point now resets on a new epoch as well as on a
break, so a standing train no longer turns around after a correction.
- recallDirection and earlierFixes are gone: comparing each earlier report
with the newest turned a standing train on GPS jitter. A train bound when
its line arrives late shows no direction until it has moved 15 m.
- Tests share reportDirections, iconOf and the playback helper.
- Comments and CURRENT-STATE match the code.
When the terrain height batch loses a chunk to the rate limit, the cameras
in it keep a sea-level fallback height. In the mountains that put them far
under the surface, so the view pass judged them by a point that was not
where they are drawn: a tilted view dropped the count marker, a click on
the marker framed empty ground, and icons drifted on tilt and pan.
- A camera with no real ground height takes the height of the terrain the
globe draws once it is near the screen (up to 300 reads a pass, through
the height service), and a later answer from the batch replaces it.
- A count marker is anchored to the drawn terrain under its centre, and a
click frames that ground.
- Cameras on one mount draw side by side, 30 px apart, each with its own
callout, instead of as one icon with several callouts.
The photoreal map stack is not covered: terrain is not the surface there.
A ground cell with one camera always drew its icon, so at some zooms a
camera poked out from under a count marker with its own callout.
An icon closer than the merge gap to a count marker now joins that marker:
the count grows by one and the icon, its callout and its card are not drawn.
Two lone cameras with no count beside them stay two icons, and the active
camera always keeps its own icon.
- The first floor height that replaces the entry frame's height re-anchors
the camera and resets the smoothed rise, so the subject stands on the
floor within a second instead of hovering above it.
- A provider's thenable gets one rejection handler; frame fields and eye
are read once each.
- A progress label's width counts the smaller detail line and full-width
and astral glyphs.
- Tests cover the first camera update of an aircraft and a vessel, a
double-click on a selected vessel, a tone-only Live Signals change and a
getter that throws on its second read; repeated cases share helpers.
- Comments, the release note and CURRENT-STATE say which height the camera
holds while it waits for the surface and how a Live Signals row keeps
its time.
- GPS jitter under 15 m at a standing train no longer turns it, in reports
and in playback; a reversal needs at least 15 m of movement.
- A standing train keeps its operator's last bearing through reports that
leave it out.
- Playback resets its turn point at a break, so a jump to a corrected
position is not a reversal.
- A train bound while its line is still loading takes its direction from
the movement it already made.
- A vehicle in view that leaves a Transit hold goes on from its own clock
instead of rewinding.
- Tests cover the 15 m boundary, the late line, the hold release and the
icon after Transit is turned back on; duplicate jitter tests are merged.
- FOLLOW_MIN_MOVE_M is gone; comments and CURRENT-STATE match the code.
A count marker now says what it counts and how much:
- Its height grows with the count, from 40 px for two cameras to 58 px for
several thousand, with clear fill around the digits.
- A small single-camera map icon sits under the top of the ring and the
number just below the centre, so the two never touch. The icon is built
from the same sprite as a single camera on the map.
- Thousands print as 1.7K or 12K in every UI language.
- The camera layer is green: count markers and idle single cameras share
the colour, the active camera stays gold and live-video cameras violet.