Adds columns to cached_tracks for the v2 format that stores Qobuz's raw
CMAF segments bit-identical on disk, with the per-track AES content key
wrapped via qbz-secrets (OS keyring or KDF fallback).
New columns (added via idempotent ALTER TABLE, safe to re-run):
- cache_format INTEGER DEFAULT 1 — 1 = legacy plain FLAC (existing),
2 = raw CMAF bundle (new)
- init_path TEXT — path to init.mp4 (FLAC header +
segment table)
- content_key_wrapped BLOB — wrapped 16-byte AES content key
- infos_wrapped BLOB — wrapped session infos salt
- format_id INTEGER — Qobuz format id (5/6/7/27)
- n_segments INTEGER — number of audio segments
Existing rows keep cache_format=1 and continue to play from the plain
FLAC at file_path — zero regression for users with populated caches
when the upgrade ships. New downloads will land as cache_format=2. The
two formats coexist until v1 rows age out naturally (user re-download
or subscription-lapse cache wipe).
Adds set_cmaf_bundle() / get_cmaf_bundle() helpers + a CmafBundleRow
struct so the download path and the playback path have one authoritative
DB surface to talk to.
Also adds qbz-secrets as a direct src-tauri dep so the next commit
(rewiring spawn_track_cache_download) can wrap keys right at the
download site.
cc
Foundation for the offline-cache hardening story. The offline cache
needs to persist per-track AES content keys to disk, but storing them
plaintext would make copying the SQLite DB enough to decrypt the
cached audio on another machine.
qbz-secrets provides a uniform wrap/unwrap API backed by:
1. **OS keyring** (preferred): libsecret/gnome-keyring on Linux,
Keychain on macOS, DPAPI on Windows. A 256-bit master AES key
lives inside the OS secure store; all wraps go through it. This
is the gold-standard device binding used by commercial music
apps (Spotify, Apple Music, Tidal) for the same class of data.
2. **HKDF fallback**: when the OS keyring is not reachable (headless
setups — Raspberry Pi, Docker, the future qbzd daemon on a
server), the master key is derived on demand via HKDF-SHA256
over service_name + machine_id + a persistent per-install UUID.
Weaker than keyring (reversible by anyone with filesystem access
to both sources) but still device-bound against cross-machine
copy attacks, and lets daemon/headless variants work at all.
Every wrap uses AES-256-GCM with a fresh random 12-byte nonce. The
on-disk envelope carries a version byte + backend marker + nonce +
ciphertext+tag, so (a) we detect tampering via GCM, (b) we refuse to
silently unwrap a blob produced by a backend we can't reach, and (c)
we have a forward-compat path for future key rotation.
4 unit tests cover: small-secret round-trip, 16-byte content-key
round-trip (the exact CMAF use case), GCM tamper detection, and
nonce-randomness (two wraps of the same input differ).
This crate intentionally has zero QBZ dependencies — it's a generic
primitive. Next commits wire it into the offline cache schema + the
DB migration.
cc
Prep work for offline-cache CMAF hardening. The full CMAF pipeline
(init-segment setup, full-download, decrypt) used to live in
src-tauri/src/commands_v2/helpers.rs — fine for playback but unreachable
from the offline-cache code without dragging Tauri deps around.
Moves:
- try_cmaf_full_download → qbz_qobuz::cmaf::download_full
- try_cmaf_streaming_setup → qbz_qobuz::cmaf::setup_streaming
- CmafStreamingInfo struct → qbz_qobuz::cmaf
Adds:
- qbz_qobuz::cmaf::download_raw + CmafRawBundle — fetches the same track
but returns the ENCRYPTED segments + init bytes + the unwrapped
content_key + the session infos, ready to be persisted to disk as-is.
The offline cache will use this so that cached files are bit-identical
to what Qobuz delivered (no decrypt-to-disk), with the content key
stored separately under an OS-keyring-wrapped blob. Coming in next
commits.
- qbz_qobuz::cmaf::decrypt_segments_into — shared decrypt helper used by
download_full and (in the next commits) by the offline-playback path
that reads encrypted segments from disk instead of CDN.
In-tree behaviour is unchanged: try_cmaf_full_download /
try_cmaf_streaming_setup in helpers.rs are now thin wrappers that hand
the CoreBridge's QobuzClient to the crate functions. Same API surface
for playback.rs callers, same log tags, same on-wire behaviour.
v2_cmaf_stream stays in src-tauri because it binds to qbz-player's
BufferWriter + qbz-cache's AudioCache — those are playback-only deps
that don't belong in qbz-qobuz.
Net: -100 LOC out of src-tauri, +390 LOC in a properly-layered crate
module with docs explaining the two variants and why the raw one exists.
cc
Sweep of remaining <QualityBadge compact> usages in row/list contexts.
Same fix as the label popular-tracks one: switch to the formatQuality
text representation so every row lines up visually.
Replaced in:
- SearchView.svelte — track rows in track tab & album tab
- PurchasesView.svelte — album list rows
- PurchaseAlbumDetailView.svelte — album metadata header (matches the
text-only treatment already used by AlbumDetailView)
- VirtualizedFavoritesAlbumGrid.svelte — favorites album list rows
Kept as-is (the badge is a first-class visual element, not a text
indicator in a list):
- Player widgets — NowPlayingBar, ExpandedPlayer, ImmersiveControls,
ImmersivePlayer
- Immersive visualizer panels (all 13)
- SearchView "Most Popular" promoted card
- AlbumReactivePanel
cc
LabelView's popular tracks section was rendering the <QualityBadge>
component with compact mode — same one we dropped elsewhere. Switch
to the formatQuality() text representation used by ArtistDetailView's
popular tracks so both views look identical.
Also align .track-quality CSS (font-size / color / min-width) with
ArtistDetailView's rules.
cc
/label/get is gone from the Qobuz v9.7.0.3 Android app (source:
c30/b.java doesn't declare it anymore). With the Qobuz beta rollout
the server-side deprecation timer is probably running, so we yank our
dependency now.
Removed:
- LABEL_GET path constant (qbz-qobuz + src-tauri/api)
- get_label() client method (qbz-qobuz + src-tauri/api)
- get_label() wrapper in QbzCore + CoreBridge
- v2_get_label tauri command + its lib.rs registration
- LabelDetail backend type (qbz-models + src-tauri/api/models)
- commands::search::get_label (dead, was never registered)
- Unused LabelDetail import in src/routes/+page.svelte
Preserved:
- Frontend LabelDetail interface (it's a UI view-model shaped by
LabelReleasesView, not a backend contract)
- /api/labels/{id} HTTP route on qbzd — now composes /label/page +
/label/getAlbums internally so external HTTP clients still see the
same { id, name, description, image, albums: { items, total, ... } }
shape
cc
LabelReleasesView now fetches /label/page (metadata) + /label/getAlbums
(paginated catalog) in parallel instead of hitting the legacy
/label/get one-shot. loadMore() paginates through /label/getAlbums.
LabelView's loadLabelAlbumsAndDescription() now only fetches the album
list via /label/getAlbums; the description comes exclusively from the
already-in-flight /label/page call (loadLabelPage).
Both views still present the same LabelDetail-shaped view-model;
nothing UI-visible should change. Prepares for deletion of the legacy
/label/get stack in the next commit.
cc
Closes the gap between our client and the Qobuz Android app's current
label API (source: c30/b.java). Qobuz is rolling out a web beta so the
legacy /label/get is likely on borrowed time — adding the v9.7.0.3
endpoints now so the migration can happen in follow-up commits.
New endpoints (paths + client + core + bridge + v2 command + registered):
- /label/getAlbums → v2_get_label_albums
paginated catalog with sort/order/genre_ids/from_date/to_date
- /label/getNextReleases → v2_get_label_next_releases
- /label/getAwardedReleases → v2_get_label_awarded_releases
- /label/getPlaylists → v2_get_label_playlists
- /label/getTopArtists → v2_get_label_top_artists
- /label/story → v2_get_label_story
- /label/getList → v2_get_label_list (POST, mirrors
track/getList signing)
New response types in qbz-models:
- LabelListPage<T> — generic { has_more, items: [T], total?, offset?,
limit? } envelope, tolerant to either V2 list or SearchResultsPage
shapes until samples land
- LabelStoryResponse, LabelGetListResponse — best-effort with optional
fields so a shape mismatch is non-fatal
The legacy /label/get + get_label() + v2_get_label + LabelDetail are
retained untouched for now; frontend migration and deletion come next.
cc
Fix#326 — multi-track selection in Local Library. Works on the tracks
tab and inside the album detail view (the more useful context per user
feedback). Bulk play next / play later / add to playlist.
Fix#327 — clicking a track in the queue panel now plays that exact
track even under shuffle. Adds play_upcoming_at with shuffle-aware
index resolution, plus dead From-impl cleanup.
The reset `$effect` was tracking its own writes. When the select-mode
button set `trackSelectMode = true`, the effect re-ran (because
`resetMultiSelect()` reads both `trackSelectMode` and `selectedTrackIds`
inside the effect body) and immediately cleared the state again — so
the button visibly did nothing.
Wrap the reset call in `untrack()` so the effect only rebinds on tab /
album changes, not on its own output.
cc
Extends the tracks-tab multi-select to the album detail view. User
requested this because it's actually more useful there (picking a
subset of an album's tracks for a playlist is the common case).
- Select toggle button added to the album-actions row (next to play /
shuffle), using the same `action-btn-circle` style so it sits with
the existing circular actions.
- TrackRow inside `albumSections` now receives `selectable`,
`selected`, `onToggleSelect` — TrackRow already supports these, they
were just not wired here.
- Same BulkActionBar rendered below the track list; the existing bulk
handlers (`handleBulkPlayNext` / `handleBulkPlayLater` /
`handleBulkAddToPlaylist`) were updated to look up selected ids from
both `tracks` and `albumTracks`, so it works in either context.
- $effect watches `activeTab` and `selectedAlbum?.id`, clearing
selection whenever the user switches tab or enters/leaves an album.
Prevents stale selection bleeding across views.
cc
Mirrors the pattern already used in FavoritesView so users can select
multiple local tracks and bulk-add them to the queue (play next / play
later) or to a playlist — the same way it works for Qobuz favorites
and playlists.
- New `trackSelectMode` + `selectedTrackIds` state + handlers (toggle
mode, toggle row, bulk play next/later, bulk add to playlist).
- New `SquareCheckBig` toolbar button on the tracks tab (matches the
one in FavoritesView).
- VirtualizedTrackList wired with `selectable`, `selectedIds`,
`onToggleSelect` props (already supported, just unused here).
- `<BulkActionBar>` rendered below the list when selection is active.
- New `onBulkAddToPlaylist` prop, wired in +page.svelte to
`openAddToPlaylist(trackIds, true)` so the existing playlist picker
modal handles the rest.
- Bulk play next/later go through `cmdAddTracksToQueue` /
`cmdAddTracksToQueueNext` (route to qbzd when remote, otherwise local
bulk command). Plex tracks are filtered out only for "add to
playlist" (per existing per-row guard); they're fine in the queue.
cc
The frontend was resolving queue-panel clicks via
`actualIndex = current_index + trackIndex`, mixing the unshuffled
canonical position with the shuffle-aware display position. On shuffle
mode, the resulting index pointed at a different track than the one
the user clicked, which is exactly what issue #327 reports.
Solution:
- New `QueueManager::play_upcoming_at(upcoming_index)` resolves the
click position through `shuffle_order` + `shuffle_position` when
shuffle is on, and falls back to `current_index + upcoming_index + 1`
otherwise.
- Exposed through `QbzCore::play_upcoming_at` (emits QueueUpdated) and
`CoreBridge::play_upcoming_at`.
- New `v2_play_queue_upcoming_at` tauri command.
- `QueuePanel.onPlayTrack` now carries the `originalIndex` so the
caller knows the exact upcoming-list position the user clicked.
- `handleQueueTrackPlay` calls the new command directly — no more
client-side index arithmetic.
Also deletes the two dead `From<crate::queue::QueueTrack>` impls in
commands_v2/queue.rs (canonical queue lives in qbz-player now), per
the zero-legacy V2 policy.
Tests: two new unit tests cover the linear and shuffle-aware paths.
Session persistence is untouched — the new method only consults
existing state.
cc
AirPlay was scaffolded in both crates/qbz-cast/src/airplay and
src-tauri/src/cast/airplay but never worked at runtime: the RAOP
sender required an Apple MFi-licensed stack with no free/open
alternative, and the backend state was already disabled via the
commented `.manage(airplay_state)` line in src-tauri/src/lib.rs.
The CastPicker UI already hid the AirPlay tab pending streaming
support that was never going to ship.
Delete:
- crates/qbz-cast/src/airplay/ (entire directory)
- src-tauri/src/cast/airplay/ (entire directory)
- 10x v2_airplay_* commands from commands_v2/library.rs + their
tauri::generate_handler entries
- AirPlayError from qbz-cast errors
- 'airplay' from CastProtocol + every case arm in castStore.ts,
CastPicker.svelte, AudioOutputBadges.svelte
- Corresponding i18n keys if present
Zero behavior change for Chromecast / DLNA / QConnect / MPRIS. If
AirPlay ever becomes feasible we'll revisit from scratch with a
real streaming backend.
cc
Split the monolithic v2_search_all command (44 cognitive complexity,
critical on Sonar) into four small single-purpose helpers:
- parse_results_page<T>(response, key, limit) — generic DeserializeOwned
accessor that replaces four copy-pasted .get().and_then().unwrap_or
blocks returning an empty SearchResultsPage.
- filter_page<T, F>(page, keep) — closure-based retain-with-total-
adjustment that replaces three near-identical post-filter blocks.
- most_popular_from_item(item, blacklist) — per-item match over the
three supported types; each arm is a flat parse + single blacklist
check, no nested continues.
- pick_most_popular(response, blacklist) — traverses the items array
via find_map, replacing the deeply-nested and_then/for/match/return
stack that tripped the cognitive-complexity threshold.
Semantics are preserved: same filter criteria (including the track
"no performer" edge case), same skip-on-blacklist behaviour, same
return shape. v2_search_all itself is now ~30 lines, each helper is
<15 complexity.
cc
Bugs (3, all same pattern): Promise-typed variables were used as
boolean conditions, which is always truthy regardless of the resolved
value. Make the nullness check explicit:
- src/lib/services/imageCacheService.ts:36 (Map.get result)
- src/lib/stores/awardCatalogStore.ts:108 (inflight Promise|null)
- src/lib/stores/windowControlsStore.ts:228 (detectInFlight)
Hotspots:
- playerStore.test.ts, queueStore.test.ts: http -> https in test
fixtures (harmless in-memory fixtures, but keeps the linter quiet)
- playbackTargetStore.ts:30: replaced the trailing-slash trim regex
with a bounded while-endsWith loop to drop the ReDoS warning
- playlistSuggestionsService.ts:181 & trackActions.ts:551: kept
Math.random for shuffle/seed (non-cryptographic by design); added
explanatory comments and marked the Sonar hotspots as SAFE.
Closes Reliability delta from 3.0 (C) to 1.0 (A) on next scan.
cc
Per the original instruction: single line for 1.2.7 + the full 1.2.6
body (which already carries the 1.2.5 content below it). The previous
version over-expanded the 1.2.7 section into three paragraphs.
cc
Resolved via two `nix build` iterations against the v1.2.7 tag.
Users tracking main get a zero-friction build; users pinning to the
v1.2.7 git tag still need to substitute the hashes themselves since
the hashes were unknown at tag time (same as every prior release).
cc
The v1.2.6 regenerated tray glyphs were monochrome outlines of the Q:
the selection logic (prefer_dark_tray() -> dark-variant) picks
tray-dark.png on dark panels, which rendered as a black Q on dark
bg (issue visible in Plasma on emerge qbz-bin install).
Switch to a full-colour render of the final icon.svg (matte-vinyl
disc + blue rim + inner highlight). Full-colour rim gives definition
on both light and dark panels, so one PNG works for every variant:
the three tray files (tray.png, tray-light.png, tray-dark.png) now
point to the same render and the selection logic becomes a no-op.
cc
CodeQL rust/cleartext-logging (alert #69) flagged the info-level log
that prints the radio session id on creation. The value is only
useful for local troubleshooting; demote to debug! so it stays out
of default-level logs (same treatment as the qbzd user_id logs).
cc
The 0.10.x advisory was fixed in v1.2.7 by bumping 0.10.0 -> 0.10.1.
The same advisory also applies to 0.9.x with a separate patched
version (0.9.3). Applying the bump to both lockfiles so the alerts
for the transitive 0.9.x deps close too.
cc
Settings menu > Report Issue now opens a modal instead of jumping
straight to the GitHub new-issue page. The modal explains that
sharing the (manually masked) logs speeds up triage and offers two
actions:
- Go to logs → navigates to Settings > Developer Mode and
auto-opens the Logs modal so the user can review and upload.
- Create issue report → opens the GitHub issues page.
A persistent "Do not show this again" checkbox stores the preference
in localStorage (qbz-hide-report-issue-modal). When the flag is set,
the menu item skips the modal and goes straight to GitHub, matching
the previous behaviour.
Wiring: new settingsIntentStore lets UserCard cue SettingsView to
open at the `developer` section with showLogsModal=true, consumed and
cleared on mount so a plain Settings navigation afterwards uses the
normal default section.
i18n: new reportIssue block added to all 5 locales (en/es/de/fr/pt).
cc
New resting state (= previous hover):
fill --accent-hover
glyph --text-primary
ring --text-primary (2px)
New hover state (= previous resting):
fill --accent-primary
glyph --accent-hover
ring --text-primary (2px)
Ring colour/thickness stays constant across both states; only fill
and glyph flip.
cc
Glyph colour in normal/inactive state is now var(--accent-hover) —
i.e. the same colour the button fills to on hover. Explicit color:
var(--text-primary) on the hover rule to keep the hover path intact
after inverting the resting glyph colour.
cc
Per design direction: normal/inactive state of .action-btn-circle.primary
now uses the exact same stroke as the hover state (2px inset
--text-primary). The sole visual difference between resting and hover
is now the background fill (--accent-primary vs --accent-hover).
cc
Revert the four commits that touched .action-btn-circle since 1.2.6:
- d673dc20 thicken stroke 1->2px (3px primary)
- a9eaf646 thicken only resting stroke of primary
- eb9f82ea bump primary resting stroke to 5px
- 5a52ef55 switch primary to real border
All rules back to their pre-1.2.6-release state. Awaiting a new
direction from the user.
cc
Inset box-shadow renders the circular stroke as a stair-stepped /
hairline edge on WebKitGTK — no matter how thick we set it, sub-pixel
AA eats most of it and on HiDPI it disappears entirely. Switch the
primary .action-btn-circle to an actual 5px border with
box-sizing: border-box so the outer dimensions stay 44x44 but the
stroke is rasterised through border AA (which respects border-radius
cleanly).
Also swap --border-subtle for --alpha-40 so the ring has guaranteed
contrast against the accent colour in every theme.
Secondary buttons and is-active stay on inset box-shadow — they're
thinner and the background contrast is higher there, so the shadow
path still reads fine.
cc
Settings > Developer Mode > View Logs now has a toggle (default ON,
persisted to localStorage as qbz-paste-bundle-diagnostics) that
prepends the full System Diagnostics snapshot plus a 10-second cast
scan to the terminal paste upload, so bug reporters share one link
instead of two. Console-tab uploads are unaffected.
- New service module diagnosticsSnapshot.ts centralises the snapshot
payload (runtime + system + playback + qconnect + optional cast
scan) so the panel and the paste flow don't drift apart.
- LogsModal shows "Collecting diagnostics..." while the scan is
running, then "Uploading..." once it hands off to the paste server.
Also revert the auto-run cast scan on the diagnostics panel: bundling
into the paste covers the "one-shot bug report" use case, and the
panel opens instantly again.
cc
New v2_get_system_info command and matching UI section in the
System Diagnostics panel. What it reports:
- os (linux/macos/windows)
- arch (x86_64, aarch64)
- kernel_version from /proc/sys/kernel/osrelease
- Linux distro: id, version_id, pretty_name from /etc/os-release
(falls back to /run/host/os-release so it works inside Flatpak)
- install_method: flatpak / snap / appimage / dev / native, detected
via FLATPAK_ID + /.flatpak-info, SNAP, APPIMAGE env vars, and
debug_assertions as the last fallback
- flatpak_runtime + version: parsed from /.flatpak-info
- loaded library versions (from /proc/self/maps, symlink-resolved):
webkit2gtk-4.1, gtk-3 (or 4), glibc, alsa, pipewire, pulseaudio
All info is derived from local files/env; no sensitive data, no IDs,
no hashes. Included in the clipboard export.
cc
- Volume: playerStore.getVolume() already returns 0-100, the extra *100
was reporting "10000%". Just Math.round() the stored value.
- Cast scan: run automatically the first time the panel is opened so
the user doesn't have to press the button. Button now doubles as
"re-scan" while disabled during an in-flight scan. A spinner
placeholder shows while the initial scan is in progress.
- Track info: capture a single snapshot on open/refresh instead of
live-subscribing to the player store — live updates were unnecessary
for a diagnostic dump and slowed the settings page. Use "Refresh"
to take a fresh snapshot.
cc
Extend the System Diagnostics panel with three new sections aimed at
what debug tickets usually ask for, and include them in the clipboard
export:
- Playback: is-playing, volume, position/duration, current track
(title, artist, album, source, local flag, quality/format/bit-depth/
sample-rate). No track IDs or UUIDs — defensive regex redacts any
hex/UUID-shaped strings before export so paste sites and CodeQL
don't flag them as secrets.
- Qobuz Connect: running/transport_connected, role (controller /
local-renderer / observer / none), active renderer friendly
name/brand/model (no UUIDs or IPs), visible renderer count, last
error (id-redacted).
- Cast Discovery: on-demand 10-second multicast scan for Chromecast
and DLNA; reports per-protocol count and the friendly names that
were discovered. No hosts, no ports, no UUIDs.
Playback rows stay live via a player-store subscription while the
panel is open.
cc
3px was still not enough against a low-contrast --border-subtle ring.
Push it to 5px so the stroke is unmistakable at default resolution
and survives HiDPI scaling.
cc
Previous commit bumped every stroke state to the same thickness. The
hover and secondary-button states were fine; only the primary's
resting stroke (using --border-subtle, a low-contrast colour) read as
jagged from sub-pixel AA and disappeared on HiDPI.
Revert secondary buttons, is-active, and primary-hover to their
original 1-2px. Keep the primary resting state at 3px so it survives
AA at the default resolution and stays visible on HiDPI.
cc
The 1px inset stroke on .action-btn-circle read as jagged at standard
resolution (sub-pixel antialiasing artifacts) and effectively invisible
on HiDPI displays. Bump the resting stroke to 2px (secondary) and 3px
(primary play button) so it survives both; hover states scale with it
to keep the visual hierarchy. The extra pixels eat slightly into the
concentric fill, per the requested trade-off.
Applies to all reuse sites: Album/Artist/Playlist/Label/FavQ detail
headers, DynamicSuggest, ForYouTab, LocalLibrary, etc.
cc
The previous CSS only disabled user-select while .dragging was active,
but the drag class only gets applied after crossing the 5px threshold
in onPointerMove. During those first few pixels the browser's default
selection behaviour would highlight card titles and artist names,
which the user then has to click away to clear.
Disable user-select on the scroll-container permanently. Text is
still selectable in the album/artist/playlist detail views — only the
carousel rows themselves are locked.
cc
Includes:
- feat(window): optional track metadata in window title (#324)
- fix(window,login): grant set-title capability + shrink login logo 20%
- fix(cast): keep scanning spinner visible for the full discovery window
Users whose favorites tab_order was persisted before 'labels' became
a favorites tab see the sidebar render without the Labels entry,
while the TitleBar nav (which uses a fresh default) shows it fine.
When loading the persisted tab_order, diff against the canonical
default and splice back any tab the user is missing at its default
position. Preserves any reordering the user applied.
cc
Advisory: rand is unsound with a custom logger using rand::rng().
Affects 0.10.x only; 0.10.1 fixes the issue. The older rand versions
still present in the lockfiles (0.7.3, 0.8.5, 0.9.2) are transitive
dependencies that do not use the rand::rng() API surface introduced
in 0.10 and are not exploitable via this advisory.
The lru 0.12.5 advisory (GHSA-rhfx-m35p-ff5j) is a transitive dep of
ratatui 0.29.0 (used by qbzd). Can't be bumped without upstream
ratatui dropping the old lru; tracked separately.
cc
Multicast discovery (DLNA / Chromecast) typically needs 5-15s before
devices announce themselves. The previous code set loading=false
after the first pollDevices() call, so users saw an empty "No devices
found" state just a second or two after opening the picker — even
though discovery was still in progress.
Add a separate `scanning` flag that stays true for 15 seconds after
startDiscovery (cleared explicitly on stopDiscovery and on error).
The empty state now shows the spinner + "Loading devices" while
(loading || scanning), so users get visual feedback during the full
multicast window instead of a misleading "no devices found".
cc