WebUI: D&D drop on preset slot races, "alt vs. neu" diff missing until reload

[tostmann]

What

After dragging a radio station from the sidebar library and dropping it on a preset slot, the dual-row "old (HW) vs. new (Store)" diff view sometimes does not appear. The slot card still shows the previous state. The user has to either fully reload the page or click "Refresh Status" to see the diff.

Reported by @fred_feuerstein in the FHEM forum thread: "Teilweise kommt es vor, dass man nach Drag/Drop bspw. eines Radiosenders das Ergebnis nicht direkt auf der Speichertaste sieht (also diese Ansicht: alt / neu, wo man dann pushen kann). Auch hier muss man manchmal einfach mal einen Reload der Seite oder Refresh-Status machen."

Why this happens

The drop handler onPresetDrop in web-src/index.html:1853-2019 for a radio drop from the sidebar library flows through :1900-1945:

1. Sends PUT /api/speaker/{id}/preset/{slot} (writes to NVS immediately)
2. Calls markDirty(dst.dev) (:1933)
3. Calls loadPresetsFor(dst.dev) (:1937)
4. Calls kickPoll(dst.dev) (:1938)

loadPresetsFor (:1346-1375) fetches /api/speaker/{id}/presets AND /hardware-presets in parallel, then calls renderSpeakerSlots() + updatePushVisibility() — so the diff render should update.

Root cause: loadPresetsFor is async and silently catches all errors at :1374 (catch(e) { /* ignore */ }). When the parallel Promise.all resolves with {error: ...} for storeR (speaker briefly unreachable / ESP busy with migrate worker → 502), line :1352 reads storeR.presets || [] → empty array → renders all 6 slots as EMPTY. The user sees a momentarily wiped card, then the next 10s polling tick recovers. The catch eats the failure silently, so no toast either.

Proposed fixes

1. loadPresetsFor (:1346-1375): detect {error: ...} shape in storeR/hwR responses and bail without overwriting STATE.presetsByDev instead of normalizing to empty arrays. Retry with backoff.
2. onPresetDrop (:1933-1938): optimistically patch STATE.presetsByDev[dst.dev].store[dst.slot-1] from the local putBody before the GET completes, so the diff renders immediately regardless of network race. loadPresetsFor then reconciles asynchronously.
3. :1374: replace silent catch(e) { /* ignore */ } with at minimum console.warn and a retry-once.

Acceptance

• 10 consecutive D&D drops on the same slot show the diff immediately every time
• An intentionally-induced backend 502 during the drop does not blank the slot card

[tostmann]

Fixed in commit e0aa968.

Two underlying bugs:

1. loadPresetsFor() treated a failed /presets GET as an empty array (storeR.presets || []), which then convinced the auto-import-from-device path that the store was empty — silently overwriting the just-dropped slot with hardware state.
2. The PUT-then-GET sequence on drop meant the diff and push-button visibility only updated after the round-trip, sometimes blocked by (1) entirely.

Fixes:

• loadPresetsFor() now detects {error:...} shape and bails before touching state, with a single 600ms retry.
• Auto-import-from-device gated to true cold start only (no prior state for this device).
• New patchStoreSlot_() / swapStoreSlots_() helpers applied right after a successful PUT in all 4 drop paths (external preset, library drop, cross-speaker copy, within-speaker swap) — diff and push button appear instantly now.

Will ship in the next release tag.