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.
The still preview gate looked only at the CCTV panel, so the Context
camera card, which reads the same frame, stopped updating while the
panel was collapsed or another phone sheet was open. A mounted card now
counts itself in the CCTV store, and the still is requested while the
panel or a card is shown; live video still follows the panel only.
Review follow-ups in the same area:
- visiblePoints is registered once, in createCesiumPorts, and drops
checks for input that cannot happen; the contract documents
horizonMarginM and when it answers false or null.
- The layer passes the mount and the four corners once each and asks
findCamera for the port.
- Phone sheet names come from a shared module instead of literals.
- Layer tests cover the fetch gate; the geometry cases moved to the
camera port tests. Comments and CURRENT-STATE match the code.
The picker's options are data: CctvControls computes the cameras in view
and the closing note for those left out, keeps a focused picker's list and
prepends a camera activated under it, and publishes the options, disabled
state and selection. CctvPicker renders them keyed by camera id and
applies the selection once the options are in place; it updates only when
the controls publish, as the old picker did on layer notifications. A
picker about to be disabled no longer counts as focused.
This completes the CCTV panel: everything it shows is rendered from the
gev-cctv store.
The frame lifecycle keeps its request state in an object instead of the
image's dataset and publishes it; the panel binds the image's src, class
and data attributes and the wrapper's loading and has-frame classes from
it, so preload tokens, kept pixels on a failed refresh and clearing on a
camera change are unchanged. The source badge is the pure
presentCctvSourceBadge(), the live switch and the typed summary are
published too. The video surface still draws into the panel's canvas.
presentCctvCalibration() joins the panel view: ADJUST, SAVE CAL, RESET CAL
and the seven pose chips render from the gev-cctv store. The inline editor
is store state: starting an edit publishes its field, starting value and
finish(), and the CctvCalValue chip renders the number input. The draft is
set once on mount so a refresh while typing keeps it; each edit gets its
own input. Commit and cancel rules, the captured base pose and the camera
guards stay with CctvControls.
formField.js had no other user and is removed.
CctvControls presents the switch, navigation and toggle buttons, the CAL
chip and the meta line with a pure presentCctvPanel() and publishes them
to the gev-cctv store; CctvPanel renders them. The controls keep their
listeners on the same elements, and still write the picker, frame, source
badge, calibration readout and summary, which carry no bindings.
A page that imports ui/cctv with its own markup no longer has these
elements painted for it, as with Street Level.