Repository navigation
fix: chunk 3 — link and roster lifecycle, HID #132
Description
Part of the fix track (parent issue linked in the sidebar). These items change behaviour on purpose; each final answer was decided on 2026-10-16. The test checklist at the bottom is one hardware session: plugging and unplugging, a reflash, a reload, and a longer soak.
Owners: gateway/gateway-connection.ts, nodes/node-hub.ts (roster poll + GatewayWatch), nodes/node-model.ts, hid/hid-service.ts, hid/calibration/*, app/ipc.ts (rehydrate).
| Item | Today | After |
|---|---|---|
| M2 + G1 roster requests and the not-responding deadline | Each roster request (5 s poll, reconnect, Refresh) throws away replies still arriving, so those boards are marked offline. The poll interval equals the not-responding deadline, so one lost reply can flash "Gateway not responding" | A pending-request generation, so a new request never discards an arriving reply. The poll cadence and the not-responding deadline are designed together (e.g. re-ask once inside the window), so one lost reply never trips the alert |
| M4 stopped serial bridge | Callbacks from a stopped bridge still drive the link; after Disconnect/Retry an orphan can reopen the port and hold it | Callbacks from a bridge that isn't current are ignored; stopped is re-checked after each await |
| M7 retry saves last seen | The error → Retry path resets without saving last seen | Reset saves last seen first |
| M3 v2 bridge END | With a v2 bridge the END line still completes the roster, so every board reads offline | END is no longer reconciled after the v2 warning |
| X2 / R5 / S10 on the bus | Fresh PDU telemetry shows a board but doesn't count it as online; while unplugged, the Cockpit's "New on the bus" and About's "on the bus" treat the stale roster as live | Fresh telemetry counts as online; every "on the bus" surface goes quiet while the link is down |
| — | Dropped (user 2026-10-17): the hat/button list is the client's synced copy of HIDControls.h, not read from the device; axes already follow the device on every connect; a mismatched reflash isn't a realistic case | |
| N1 reload | A window reload drops the recording, replay file, serial-monitor flag, aircraft, telemetry, peak W and link-down time | The rehydrate bundle carries them |
| N6 calibrate badge | The per-axis badge opens calibration with the link down while the Calibrate button is disabled | The badge follows the button's link rule |
| NEW (user 2026-10-17) remove Refresh | Connection › Connected nodes has a Refresh button that asks for the roster now; the 5 s poll and every reconnect already do | Remove the button and everything only it uses: the nodes:refresh intent (contract, preload, IPC_ORDER, API guard), the store intent, the canRefresh fact on device:status. The poll and reconnect requests stay. Update the design board's Connection board too |
| X11 NACK names | Two NACK-name functions (5 reachable vs all) | One nackName() from the synced firmware table |
Tests that change on purpose: the GatewayWatch G1 mechanics test (gateway-health "counts from the oldest unanswered request").
Hardware test (one session)
- 30+ min soak with two panel groups + a PDU: no false offline/online rows, no "Gateway not responding" flash
- Disconnect → Connect and Retry repeatedly: the port is never left locked; last seen is saved each time
- Unplug the Gateway: "New on the bus" and Settings › About go quiet; plug back in: they return
- Reload the window (dev) while recording/replaying: the recording, replay, monitor flag, aircraft and gauges are still shown
- With the link down, an axis badge doesn't open calibration
- Connection › Connected nodes has no Refresh button; boards still update on their own
Activity
[-]fix: chunk 2 — cockpit membership, file v4 and power zones[/-][+]fix: chunk 3 — link and roster lifecycle, HID[/+]fix(nodes): steady roster polling — no lost replies, no false not-res…
fix(hid): one nackName from the synced firmware table (X11)
fix(hid): an axis badge opens calibration only while the link is up (N6)
fix(app): a window reload keeps the app's state main still holds (N1)
refactor(gateway): the serial monitor's flag is the monitor's own on/off
fix(hid): calibrate badge follows the link, one NACK name table, relo…
fix(nodes): a PDU's pages count only while newer than the roster's la…
fix(nodes): one "on the bus" rule — telemetry counts as online, quiet…