Two findings from PoBruno's Docker Engine on Linux run (#208), and one from
re-running the suite in the container after rebasing:
- The launcher pinned its own allowedHosts, including '.local'. Vite reads a
leading dot as a suffix match, so `Host: anything.local` reached the page and
the data proxies. Main has since moved the policy into build/allowedHosts.js,
which drops suffix entries on purpose and applies before the app's routes, so
the launcher's override is deleted rather than patched. As a side effect
GEV_ALLOWED_HOSTS now also works from `.env`, as it does under `npm run dev`.
- A checkout the container cannot write - typically a clone made as root on a
NAS or Proxmox host - made Save Keys answer 500 and never created .gev-cache,
with nothing logged. The launcher now warns at start, naming the owner, the
ids it runs as, and the two fixes.
- On Docker Desktop the bind-mounted repo always shows an owner other than the
container user, so git refuses it as "dubious ownership" and the tests that
shell out to git fail (src/data/bundledJson.test.mjs, new on main). The image
now marks /app, and only /app, as a safe directory.
Verified in a container on this base: `anything.local` 403 on the page and on
/api routes, localhost 200, GEV_ALLOWED_HOSTS honoured from the environment and
from .env, the warning on a root-owned /app run as 1000:1000, and
bundledJson.test.mjs 6/6.
Reported-by: PoBruno
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Docker delivers the host's own browser to the container from the container
network's gateway, never from 127.0.0.1, so the loopback-only admission gate
answered every POWER UP request with 403 and no key could be saved. The README
and CHANGELOG claimed otherwise; both are corrected here.
- keySetupCore: GEV_KEY_SETUP_TRUSTED_PEERS, an exact-address list (no
wildcards, prefixes or hostnames) admitted alongside loopback. Unset trusts
nothing, and the sharing, Host, Origin, Content-Type and proxy-header checks
all still apply to a listed address. Twelve assertions pin each refusal.
- scripts/docker-start.mjs: the container entry point, shaped after
scripts/pinokio-start.mjs. It reads the default gateway from /proc/net/route,
trusts exactly that address, and starts Vite. No gateway (--network none)
trusts nothing. The pure helpers are unit-tested and a contract test pins the
Dockerfile CMD and the loopback-only publish.
- Dockerfile: install git. `npm test` builds fixtures with `git init`, so the
suite failed 71 tests inside the image and the README's claim that it runs in
the container was wrong.
- SECURITY.md and .env.example document the setting and why it is safe only
behind a 127.0.0.1 publish; README gains the Linux/NAS id note and .dockerignore
drops macOS AppleDouble sidecars, both reported on the PR.
Verified in Docker on a clean build: saving a key from the panel writes the
bind-mounted .env (0600 in the container), it survives `compose down`/`up`, and
REMOVE comments it back out. A foreign Origin, a foreign Host, a proxy header
and a sibling container on the same network are each refused. `npm run doctor`,
`npm test` (5067 passed, 1 skipped, 0 failed) and `npm run build` are green
inside the container.
Reported-by: @jklaze
Co-Authored-By: Bogdan Petrut <[email protected]>
Co-Authored-By: Claude Opus 5 <[email protected]>
Doctor and dev-fresh told every user without an OpenAI key that voice could
use ChatGPT OAuth, whether or not Codex was installed. Say voice is off
until a key is added or a Codex sign-in is selected, and mark that path
experimental.
The login child inherited the whole server environment, including
OPENAI_API_KEY and every other provider credential. Pass only what Codex
needs to find its home, open a browser and reach OpenAI through a proxy.
Voice-card captions from main priced transcription usage into the API-key
dollar meter. In a ChatGPT OAuth session that could trip the $2 warning or
$5 cap while the readout said COST UNKNOWN. Count them with the session's
other OAuth usage instead.
A tile whose terrain lookup cast some lines kept castRequested set, so a
later refresh skipped it and the lines that failed stayed draped over
trees and roofs until the tile was rebuilt. A refresh now retries any tile
with lines still draped, and redraws it only when more of its lines can be
cast, so a tile that cannot be cast fully costs a cache check, not a
rebuild, per refresh. castRequested still stops the redraw from casting
again at once.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Fixesbilawalsidhu/gods-eye-view#927. Each stdio MCP server made its own
random panel key, and panel_request accepts only the key in the panel
page. A client that starts two server processes for one connection and
reads the page from one while sending the page's requests to the other
(seen in Claude Desktop on Windows) therefore always got "Only the God's
Eye View panel may make this request".
The stdio entry now passes a key shared by the install, created once in
.gev-cache/mcp-panel-key. The file is staged 0600 and published with
link(), so processes starting together agree on one key and never read a
half-written one. If it cannot be stored, the process keeps a key of its
own, as before. createLocalMcpServer still makes a fresh key unless given
one, so in-process servers (the /mcp HTTP plugin, tests) are unchanged.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
- A photo takes a deferred navigation ticket when it starts opening
(begin/reassert, the application's _beginDeferredNavigation and
_reassertNavigationHandoff) and frames only if that ticket still owns
the camera once loaded, so a slow photo no longer flies the camera back
over newer navigation or clears newer aircraft tracking.
- A failed mesh refresh keeps the last good sample and the distance it was
taken from, and is retried after the miss cooldown, instead of recording
the failed distance and never refreshing again.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
- A floating portable panel saved its measured box, so dragging a
collapsed window stored its 50 px header strip and a reload clamped it
to the minimum. It now saves the inline size the user chose. A unit
test and a collapse -> drag -> reload -> expand step in the panel gate
pin it.
- Coverage awaits the tile fetch and the one-time terrain table load
together, so a tile that fails before the table loads is reported at
once instead of surfacing as an unhandled rejection.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
- Photo framing and FOLLOW claim the camera through the application's
navigation, injected with attachNavigation({ run, subscribeHandoff }):
aircraft and satellite tracking is released first, the cockpit refuses,
and FOLLOW stops when another feature takes the camera. No UI imports
in the layer.
- The mesh sampler follows the mesh floor's rules: it waits for the
visible tileset to finish streaming (visibleTilesetLoaded, now exposed
by the mesh floor service), needs a real bare-earth prior and keeps to
the shared meshFloorSampleWithinPrior window, which cameraFollow now
uses too. A cell is probed again once the camera is half as far from
it (down to 40 m), so a coarse early sample is replaced and redrawn.
- The MCP globe panel's CSP allows the Mapillary Graph API and its image
CDN (https://*.xx.fbcdn.net).
- The tile proxy reads bodies with the shared readResponseBytesCapped.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Trim the doc text this PR adds (README, CHANGELOG, CURRENT-STATE,
KNOWN-ISSUES, SECURITY, DATA_SOURCES, .env.example, the CI job comment),
drop code comments that restate the code, cut design-note JSDoc to a line
or two, and drop review tags from test names. Comments only; no code
changes.
Also: README key count (seven keys, two metered), CURRENT-STATE's gate
paragraph (CI runs a production build with a dummy token), the stacked
doc above meshBoxInRange and the orphaned inMeshRange line in
coverage.js, and a duplicate ALPR row in DATA_SOURCES that a merge had
brought back.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The fleet draws a grounded contact at its delayed position lifted onto the floor under it, but a new billboard was created at the delayed position without that lift. A taxiing aircraft whose reported fix sat lower than the ground under its delayed position started on the fix's floor and then rose on the next fleet tick. Creation now applies the same floor the fleet pass does.
The military fleet also renders behind real time and projects a fresh contact's first fix back to that delayed time, but its billboard was created at the raw fix. A new contact therefore appeared ahead of where the next fleet tick put it and jumped back. The snapshot renderer now receives the layer's motion and creates the billboard at the dead-reckoned delayed position.
The fleet renders behind real time, so a newly seen aircraft is displayed at its fix projected back to that delayed time. Its billboard was created at the raw fix instead, so it appeared ahead of where the next fleet tick put it and then jumped back about one render delay of travel before moving forward.
From gekh's review of db79027, each with a test that fails before:
- P2: image cones and the photo marker no longer disappear where the
ground lies below the WGS84 ellipsoid (NYC -22 m, Colombo -95 m). The
horizon cull's occluder is WGS84 shrunk by 1 km (HORIZON_CULL_DEPTH_M),
below any real surface, so a camera near or under the ellipsoid no
longer treats every lower point as behind the horizon; the far side of
the Earth is still hidden from orbit (tested).
- Mesh range across the date line: the camera's longitude is taken to the
box's side of +-180 before clamping (meshBoxInRange), so a tile just
across the line is not measured from its far edge.
- MapillaryJS no longer tracks window resizes (trackResize: false): the
panel's ResizeObserver resizes the viewer and skips it while hidden, so
a window resize after closing a photo cannot ask for a tile at z=NaN.
- A key-less install sends no tile requests before the key status is
known: coverage waits for the first status answer and the provider
refreshes it once the status lands.
- constructCatalog takes the source method list from MAPILLARY_SOURCE_
METHODS; sequences.js imports view.js once; the provider's unused
re-exports are gone.
Verified: unit 5970 pass; strict fixture gate on a production build 28 ok
+ 2 "no Google 3D" skips and the panel gate 6/6 (QA_FAIL_ON_RETRY=1);
live gate with a real token 25 ok + 3 fixture-only skips.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
CI on fb024d5 failed step 24 with QA_FAIL_ON_RETRY=1: the press after a
double-click dock landed on the globe because the panel, showing a photo
that was still settling, moved 116 px up and lost 46 px of height between
the pre-press hit check and the press (box 1058,382 330x471 -> 1058,266
330x425, from the new diagnostics and the uploaded screenshot). That is
the panel's own content changing its layout, not the rail race fixed in
9dde7b4.
- Fail-on-retry now fails a miss only when the panel's box was the same
at the hit check and at the press (the pointer missed a header that
stood still); a miss the panel moving explains is retried with a
::warning:: naming the move. New harness test with a moving panel.
- Step 24 waits for the open photo and its sequence to finish loading
before pressing the header again.
Verified: strict fixture gate on a production build 28 ok + 2 "no Google
3D" skips and the panel gate 6/6, both with QA_FAIL_ON_RETRY=1.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
- The CI gate builds the app and serves it with `vite preview`, with a
dummy non-secret client token (a fork PR never gets secrets), and
answers the Graph API, image tiles and thumbnails from fixtures: 41
photos on a line under the parked camera and one generated 640x320 JPEG
(scripts/fixtures/street-level). All 7 photo steps now run in CI, plus a
real click on a coverage line and on a cone, the framing flight stopping
on close, the inert modal, the viewer returning into the panel, and a
reload keeping a collapsed panel collapsed (share ui token t).
- Fixture interception is limited to Mapillary hosts via CDP
Fetch.enable: page-wide request interception stalled Cesium's workers,
so the old fixture mode never drew anything on the globe.
- Failure evidence (screenshots, UI state, console, failed requests,
render errors) is written to qa-artifacts/ with keys and tokens redacted
and uploaded on failure; qa-artifacts/ is gitignored.
- Cesium render-loop errors fail the gate; Google 3D branches that cannot
run print as skips, and --strict rejects any other skip.
- Panel presses log their diagnostics, warn on a retry, and CI now fails
on any retry (QA_FAIL_ON_RETRY=1) since the rail settles; a press with
no pointerdown is a failure, not a retry.
- Assertion-gating sleeps 27 -> 1; the rejected-key "no requests" check
has a positive control.
Verified locally: strict fixture gate on the production build 28 ok + 2
"no Google 3D" skips (twice, fail-on-retry on); panel gate 5/5 plus no
render errors; live gate with a real token 25 ok + 3 fixture-only skips.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A mutation review left 18 earlier survivors; each now has a test that
fails with the mutant applied:
- the expanded viewer returns into the panel on shrink, close and destroy
(the fake DOM had no parentNode, so reinsertion was never exercised); the
fake DOM gains offsetParent and a MutationObserver that honours its
options, and a <body> child added while expanded goes inert;
- each pose moves the marker, drives FOLLOW and selects a new sequence;
close() retires an open still mounting or opening; a provider toggled
while the layer is off activates nothing; the adapter drops late and
superseded poses;
- share-link defaults: "any date" (0 days), the codec max pinned to the
filter's MAX_SINCE_DAYS (layerState cannot import it across package
boundaries, so a test ties the two), Mapillary on when a link omits it;
- Esc that something else handled, or typed in a field, keeps the
selection; a key problem outranks LOADING on the pill;
- the marker's terrain height (mesh, DEM, clamp then re-place);
- proxy: LRU memory cache and its half-budget rule, Retry-After capped,
a request after a cancelled fetch starts a fresh one;
- date-line tile ranking with and without focus; the horizon cull idles
while the camera stands still; the old zoom goes as soon as the new one
lands; a rate limit holds every refresh inside its wait.
One real bug found and fixed: Esc inside a contenteditable cleared the
selected sequence (selection.js now ignores isContentEditable targets).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The right rail took its top from the left stack's bounding rect while the
left stack's `top` was still transitioning (150 ms): the left layout sets
its target and calls onAligned() synchronously, the right pass runs a
frame later mid-transition, and nothing re-ran it when the transition
ended. The rail then sat ~32 px off and jumped on the next unrelated
relayout - under the pointer in CI's panel gates, which is why a header
press missed on every run.
settledLeftRailTop() reads the end value of a running `top` transition
(falling back to dataset.safeTopPct, then the rect). A settled stack is
read from its rect as before, which keeps Cyber's CSS top override (and
cockpit's transition: none) exactly as they were. The Cyber test's
pinned pattern follows the new helper.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
From the IC8 review of #768, each with tests that fail before:
- P2: the expanded viewer is a real modal: while it is on screen every
other <body> child is inert (only the ones it set are restored, on
shrink, close and destroy, and when Clean View/recording/cockpit hide
it). The voice pill and toasts are not clickable while it is expanded.
- P2: the tile proxy follows redirects by hand, at most 3 hops, each one
HTTPS on exactly https://tiles.mapillary.com without credentials;
anything else is a 502 and the token never leaves that origin.
- P2: the real dev and preview server smoke tests request
/api/mapillary/status and a tile, and the browser gate first checks the
server's real status route before any interception, so CI fails if the
Mapillary plugin is not registered.
- P2: both gates press panel headers through one shared helper
(scripts/qa-panelDrag.mjs): re-resolve the header until it is under the
pointer, record what the press hit, wait on the real floating/docked
state, and retry only a press that provably missed the header.
Verified: unit 5910 pass; fixture gate 17/17; panel gate 5/5; live gate
with a real Mapillary token 22/22.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
From the IC8 review of #768, each with tests that fail before:
- P1: a mesh batch no longer rebuilds the selected sequence's cones. The
drawn cones are indexed by mesh cell; a batch that touches none of them
is ignored and the affected cones are moved in place (no removeAll/add).
- P2: mesh sampling keeps to its 6 ms slices. Requests are taken in inside
the budget, only cells within range of the camera are queued, the queue
is a persistent min-heap re-ranked in one pass only after a 100 m camera
move, and the deadline starts before any work. Coverage keeps a bounding
box per line and densifies a line only once it comes within range. On a
synthetic dense tile (600 x 40 points): slice median ~10 ms -> 5 ms, max
18-30 ms -> 6.7 ms, refresh cost 11-24 ms -> 1.5-3 ms.
- P2: image cones and the photo marker are horizon-culled like the
overview dots (one createHorizonCull in view.js, shared by all three).
- P2: a selection change recolours the previous zoom's lines still on
screen and the old lines a cast swap keeps showing.
- P2: distances (metresBetween), line densification and ground-corner
requests go the short way round +-180 deg (wrapLon in tileMath.js), so
nearest ordering, cone thinning, mesh range and cast lines are right
across the date line.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
