esbuild compiles an optional method call reached through an optional chain
(`a?.b?.()`) in a parameter's default, for browsers it lowers optional
chaining for (Chrome and Edge before 91), into code that reads a variable out
of its scope. The weather shell's texture limit was read that way, so every
satellite frame threw a ReferenceError in the built app and the layer showed
"Weather imagery unavailable"; the leader and card reveal took their default
time the same way. A test now refuses the form anywhere under src/.
On a phone the field kept 60px on its right for the language picker even
where the application shows none, so it sat left of centre. It now keeps
16px either side, and leaves the picker its room only while it is there.
A canvas the panorama moved off (a second view drew its own) is no longer
ignored for good: its draws are counted, so an open waiting on one goes on,
and once a single view is left its canvas is taken back. Before, after the
second view left, the next 360° open waited forever and the panorama stayed
on the departed view's canvas.
A mount whose view unmounts it inside its first draw now fails as closed,
as every other unmounted mount does, rather than with a TypeError.
Both found in review; each new test fails without its fix.
Co-authored-by: Manjunath Paparaju <[email protected]>
Second review follow-ups:
- A mount woken without being drawn unmounts only its own presentation:
an earlier mount overtaken by a second mount of the same viewer fails
as closed and leaves the second one mounted and presented.
- A mount whose last view left fails only if no view has subscribed
since; one subscribed in the same tick draws it.
- Mount keys come from photoViewers.js's nextViewKey(), and a keyless
presentation is stamped from the same counter when any picker first
sees it, kept until it presents null: its rank no longer depends on the
picker's lifetime or the order it subscribes in.
- A canvas handed back while a panorama's open waits on its draw takes
the panorama: the open builds its renderer there instead of falling
back to the front view. A canvas the panorama moved off is not moved
back to, so two views drawing their own cannot take it in turns.
- The guard test also refuses compound writes (+=, ||=, ??=, &&=…),
destructuring writes to properties, Object.assign (other than onto a
new Error), Object.defineProperty and Reflect.set, and a catch binding
that shadows loadImage's image.
Nothing changes in the running app.
Co-authored-by: Manjunath Paparaju <[email protected]>
Review follow-ups to the previous commit:
- photoViewers.js ranks a presentation without a mountKey as mounted when
it began presenting (after the latest key seen, before any later mount),
so a photo viewer mounting while a keyless one shows is drawn and the
keyless one superseded; supersede is queued whenever more than one
presents after any change, not only when the pick changes.
- A panorama's open still waiting on a draw when the last view leaves
fails as closed and leaves no photo shown, instead of uploading to a
canvas never drawn. A mount waiting on its first draw then also fails as
closed and is unmounted, so a view subscribing later sees nothing.
- A view handing back another canvas element than the renderer's drops
the renderer and draws the panorama again on the new one, as after a
lost context.
- The guard test allows a .src write in loadImage only on the one `image`
binding, created by new Image() and never rebound or shadowed, and
refuses `(expr).src =` and bracket property writes anywhere.
- The test fakes' shared view is reset before each test
(resetPhotoView), so a viewer left by an earlier test is neither drawn
with nor superseded by a later test's.
Nothing changes in the running app.
Co-authored-by: Manjunath Paparaju <[email protected]>
Follow-ups to the photo viewer's move to a Vue view (#1868):
- A panorama's open waits for a draw of the canvas counted from before it
is shown, so a view that draws at once inside its listener no longer
leaves the open pending; and when the last view unsubscribes, an open
waiting on a draw goes on instead of hanging.
- photoViewers.js supersedes the mounts not drawn on a microtask, after
the winning presentation has reached every subscriber: a second
subscriber sees the earlier mount, then the later one, never null
between.
- A presentation without a mountKey counts as the latest mount (among
several without, the one that began presenting last), and one without
actions or supersede is passed over without throwing. Both fields are
documented on ViewerAdapter.subscribeView in registry.js and in the
store's comment.
- The guard test also refuses .width, .height, .ariaLabel, .id and .value
writes in photoViewer.js, and .src writes outside loadImage, where only
the off-page Image it creates may take one.
- The test fakes draw every viewer through one shared view behind
photoViewers.js's picker, as the application does, so provider and
layer tests see supersede.
Nothing changes in the running app.
Co-authored-by: Manjunath Paparaju <[email protected]>
While a selected vessel, aircraft or military aircraft is off screen, behind
the globe or out of its layer's feed, its Context card shows Find it. One
still in its feed is flown to (an aircraft followed). One out of its feed is
looked up on its own track at the layer's time (the world clock's while the
layer follows it), the card moves to that reading and says how old it is, and
the camera flies there; once the feed has it again its layer selects it again.
With no reading at or before the time the card says so and the camera stays.
A vessel its layer announces is named by its own id, as a click names it,
not by the context store's.
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.
An open bigger view now counts as a viewer of the camera's still, so it
keeps refreshing after the layout turns phone and the Cameras sheet is
not the one open. The card releases it in the watcher's cleanup, which
also runs when the card unmounts in the same tick the popup closes.
Following review, the scan now also:
- reads every SFC's script blocks, not only modules;
- matches a createElement call Prettier wraps over lines, and any case of
the tag name;
- fails on a button, select, input or textarea written as markup in a
string, comments aside, apart from the globe panel's standalone page
(plain HTML a host serves, with no Vue or kit), which the test names and
requires to still be the reason;
- catches a Material class built in a template string;
- reads an input's type when bound to a quoted literal or named in a
v-bind object, in any case, and unquoted before `/>`;
- catches a Material class named in a v-bind object.
Each new path was confirmed to fail when its violation is put back.
docs/VUE-MIGRATION.md now names the globe panel page as the exception,
lists the two text areas among the native fields left, and notes that the
scene dialog's field takes text-like types from its callers only.
Co-authored-by: Manjunath Paparaju <[email protected]>
viewControls.test.mjs now reads every SFC outside the kit for a raw
<button>, a tag whose class (static or bound) names a Material Symbols
style, a native checkbox, and a native range other than the radio tuner's
and the world timeline's. It also reads every module under src outside the
kit, tests and test support for a button, select, input or textarea built
with createElement, or a Material Symbols class written in script. Every
check passes on main as it stands; each was confirmed to fail when its
violation is put back (a button, a glyph span, a checkbox, a range, a
script-built button and a script-written glyph class).
docs/VUE-MIGRATION.md says that no view hand-writes a control the kit has,
names the native fields left and points at the scan.
Co-authored-by: Manjunath Paparaju <[email protected]>
createApplicationControls takes startLayers, the layer ids that turn on when
the address carries no link (none by default). They restore through the
passive start-up path, so they are never written as the user's own choice, and
a link's own layers always win.
- 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.
Fixes from review of the move to StreetLevelPhotoViewer:
- Each mount and each panorama canvas now takes a key no viewer reuses,
and the view keys its root by the mount. Before, switching KartaView and
Panoramax in one tick let Vue patch the old viewer's canvas, whose
destroyed renderer had lost its context, into the new one, so 360°
photos fell back to their front view.
- Of the viewers mounting in the host, the view draws the one mounted
last, by mount key (photoViewers.js, which subscribePhotoViewer() now
uses), whatever order they present in. A viewer still mounting when a
later one is drawn is superseded: it unmounts and its mount fails as
closed, rather than staying pending and being drawn once the later one
goes.
- A panorama's first image is uploaded once the view has drawn its canvas
shown (the view hands the canvas back on every draw), so it is never
sized into a hidden canvas.
- The tests' fake view now draws as Vue does: once per tick with the final
value, keeping elements while their keys are unchanged and handing every
element back on each draw. StreetLevelPhotoViewer.test.mjs covers the
switch in one tick and the overtaken mount with the real view.
- The guard test's check that the photo viewer writes no DOM covers
inserting, moving, removing and replacing elements, attributes, classes,
styles and the hidden, disabled and text properties.
Co-authored-by: Manjunath Paparaju <[email protected]>
The photo viewer KartaView and Panoramax share (photoViewer.js) no longer
builds its root, picture, panorama canvas and ‹ › buttons in script. It
presents them through subscribeView(), which the street-level layer
gathers in subscribePhotoViewer(); the application shell publishes that
to the gev-street-level-photo store, and StreetLevelPhotoViewer, beside
the Street Level panel in the layer panels, teleports the viewer into the
viewer host with the same classes, data attributes, labels and order. Its
‹ and › are ghost UiButtons. The view forwards keys, pointers, wheel, the
step clicks and a lost WebGL context to the viewer and hands each element
back: the canvas the panorama is drawn on (a fresh, keyed one after a lost
context, which an open now waits for), and the frame, picture, root and
buttons the viewer measures and keeps focus with. The viewer keeps every
decision: loading, upgrades, size limits, keys, drag, pinch, zoom, Esc on
the viewer's box and focus when a step button disables. The street-level
layer still needs no Vue; with no view subscribed these photos cannot
mount, which docs/VUE-MIGRATION.md records.
The provider and viewer tests draw the viewer with a fake view in
kartaViewFakes.mjs; StreetLevelPhotoViewer.test.mjs runs the real view on
a page, and streetLevelPhotoControls.test.mjs pins the kit controls. The
panel page's lost-graphics notice in globePanelRuntime.js stays built in
script, as the document now says.
Co-authored-by: Manjunath Paparaju <[email protected]>
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.
The event layer's tests now check the button's pressed state: true once a replay claims the camera, false once the operator releases it. Before, only the panel view's test covered it, with hand-built records.
Co-authored-by: Manjunath Paparaju <[email protected]>
The reconstruction panel's buttons (previous and next beat, Play, Play
scene, Cinematic, Full story, Open original, Lower gorge, the field
reports and Open public map) are ghost UiButtons that keep their
data-action hooks. Cinematic is a pressed toggle: the event layer now
presents its state as the record's `pressed` instead of writing an
aria-pressed attribute, so it is no longer an exception in
viewControls.test.mjs; bhote-koshi.css still reads aria-pressed, so it
takes no pressed class.
The embedded media cards (frame, leader line, title, status, original
link) and the hidden YouTube preload are drawn by BhoteKoshiMediaCards,
which bhoteKoshiEmbeddedMedia.js renders into its overlay root with h()
and render() on every change and on a locale change, in place of the
markup it built with createElement and innerHTML. Placement after each
frame, provider loading, preloads, timers and teardown are unchanged. The
script still builds the SDK script and preconnect tags, the overlay root
and the Facebook player node, which is warmed hidden before its card
exists and carried loaded into the card, where the SDK replaces it.
bhoteKoshiControls.test.mjs pins it; the media tests run on a DOM.
Co-authored-by: Manjunath Paparaju <[email protected]>
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.
The Space Missions panel's buttons (a mission's roster row, FOCUS, the
replay's pause, PREV, NEXT and SHOW ALL / DESELECT, beside REPLAY ASCENT
and the two closes that already were), the Contacts panel's PREV, FOCUS,
NEXT and contact rows, the Radio panel's PREV, PLAY, NEXT and STOP, the
RTL-SDR card's Locate, Change device, Tune and the two seeks, and the key
setup's POWER UP chip, REMOVE, voice auth toggle and SAVE KEYS are ghost
UiButtons keeping their ids, classes, data-*, aria-*, titles and
listeners. PLAY and Locate keep their own active class (their words
change with their state), so they gain no aria-pressed. The chip's bolt
is a UiIcon with :size="null", so provider-settings.css still sizes it.
The card's FM frequency stays a native number field (the kit has none)
and the radio tuner's range stays native, as documented. None of these
owners renders synchronously, so no slot needs redrawWithView.
src/ui/kit/missionsRadioControls.test.mjs pins it; docs/VUE-MIGRATION.md
describes the surface and no longer calls the key setup chip a plain
button. radioPanel.test.mjs reads PLAY's Core classes without the kit's.
Co-authored-by: Manjunath Paparaju <[email protected]>