Recovered from 63ecc324 (authored on the native-port branch, never merged to
pre-release). Backend-only, no Slint. Scales the prefetch cap down on slow
links and during audio underruns, and exposes the per-segment download signal
the offline detector reuses as a positive liveness override (#467).
cc
Extracts two self-contained, framework-agnostic modules out of
src-tauri so they can be reused by other clients without duplicating
logic:
- ImageCacheService / ImageCacheStats move into the qbz-cache crate
(LRU disk cache, MD5-keyed, SQLite index).
- The credential store (keyring + AES-256-GCM encrypted file for Qobuz
credentials and the OAuth token) moves into a new qbz-credentials
crate.
src-tauri/image_cache.rs and src-tauri/credentials/mod.rs become thin
re-exports, keeping their Tauri managed-state wrappers. The Tauri app
is behavior-identical; both modules were already self-contained (no
Tauri and no crate:: references).
cc
Task 2 of the Slint POC readiness audit. AppRuntime now performs
framework-agnostic per-user session activation: activate, activate_at,
activate_offline, deactivate, plus is_session_active, active_user_id,
and a with_session_store accessor.
This is the minimal version per the 2026-05-18 audit correction. It
opens only the session store and does the portable scaffolding (user
paths, directories, last_user_id marker, runtime state machine). It
does not touch Tauri's session_lifecycle, does not initialize the
src-tauri-side per-user stores, and does not run the flat-to-user
migration. The shipping app keeps identical behavior; this is additive.
activate_at takes explicit directories and is the testable core;
activate resolves real per-user paths through UserDataPaths.
cc
Adds a framework-agnostic composition root for non-Tauri UI shells
(Slint, TUI, headless) as Task 1 of the Slint POC readiness audit
(session-20-slint-poc-readiness-audit).
AppRuntime<A> owns an Arc<QbzCore<A>> plus the runtime state machine,
built with the same recipe as the Tauri CoreBridge but with no Tauri
dependency. Scope is composition and accessors only; per-user store
ownership and session activation are deferred to Task 2.
qbz-app now depends on qbz-core, qbz-player, qbz-audio, and qbz-models.
The Tauri app does not consume this module; it is purely additive.
cc
Google Cast closes a control connection that goes roughly 10 seconds without a heartbeat PING. QBZ pinged every 25 seconds, so any cast connection left idle between connect and load — for example while the track downloads, or across a delayed second play attempt — was dropped by the receiver. The next LOAD then failed with EPIPE (broken pipe): the device chimed on connect but never played audio, and the heartbeat thread logged repeating broken-pipe warnings.
Pinging every 5s is the cadence the Cast protocol expects and keeps the connection alive through those idle gaps.
Fixes#439.
cc
NVIDIA-only systems no longer get a hardware-acceleration recommendation that visibly stutters under WebKitGTK. The three NVIDIA-only branches (Wayland+GNOME, Wayland, X11) collapse into one CPU-rendering profile, since XWayland does not rescue the stutter either.
NVIDIA + iGPU hybrid now allows DMA-BUF: WebKit composites on the stable iGPU stack, so the instability that affects NVIDIA-only systems does not apply. The recommendation also tells the user to pick the iGPU explicitly in the rendering GPU selector instead of Auto.
cc
`/favorite/getNewReleases` includes albums whose stream availability is
still in the future, so they appear in the Release Watch grid even
though `get_album` 404s as soon as the user clicks them. Same shape for
regional restrictions and label takedowns surfaced via `streamable=false`.
Add `release_date_stream` and `streamable` to the Album struct (both
optional with serde defaults so existing call sites stay happy) and
filter the response inside `v2_get_release_watch` after the artist
blacklist pass: skip any album marked `streamable=false` or whose
`release_date_stream` is strictly greater than today (UTC). The total
count gets adjusted alongside so pagination math stays consistent.
cc
Bundle token extraction requires the network. When the host is fully
offline at startup, the failure used to propagate out of CoreBridge::new
and leave the bridge as None, which made every queue/playback V2 command
panic at `bridge.get().expect("CoreBridge not initialized")`.
QbzCore::init() now treats bundle extraction as best-effort: on failure
the core still finishes initialization, the queue manager and player stay
usable for local/offline playback, and API methods return
CoreError::NotInitialized cleanly instead of crashing the runtime. Adds
is_api_initialized and try_init_api so callers can rebuild the API
client lazily once the network returns.
cc
Three bugs were stacking up to make session resume drop the saved offset
for streamed Qobuz tracks:
1. Position save was debounced, not throttled. Each playback tick reset
the 5s timer, so the save never fired while a track played
continuously — the persisted position froze at whatever value got
written during the first quiet window after track start (typically
~2s). Switch to throttle: fire the first call immediately, then once
per 5s window, with a tail call to flush the latest value when the
window ends.
2. togglePlay tried to apply the restore offset via a separate cmdSeek
right after cmdPlayTrack. For Qobuz streaming the audio thread's
Seek handler bails when current_audio_data is unset — streaming
uses current_streaming_source instead — so the seek was silently
dropped and the track started at 0:00. Pass start_position_secs
directly to v2_play_track. The streaming backend (legacy + V2
player) waits for enough buffer to cover the offset plus 8s
headroom, then eagerly decode-and-discards decoder output up to
the offset before engaging the audio engine — no race, no
first-pull underrun. Plex and local tracks keep the old play+seek
path; they load full audio_data so Seek works as-is.
3. The seekbar showed 0:00 from app launch until the user hit play,
even when a saved position was queued. setPendingSessionRestore
now pre-positions currentTime when positionSecs > 0 so the UI
reflects "paused at X" the moment the app opens.
Other plumbing:
- AudioCommand::PlayStreaming carries start_position_secs and
content_length (used to estimate bytes-per-second for the resume
buffer target).
- Player::play_streaming{,_dynamic} grow a start_position_secs
parameter; QConnect and api_server callers pass 0 (remote-driven
playback always starts fresh).
- cmdPlayTrack forwards startPositionSecs as a 5th optional arg.
cc
The Plex aggregation CTE used MAX(codec)/MAX(bit_depth) directly,
which produced two bugs that surfaced in the Albums grid for many
Plex-only libraries:
- format = "Unknown". Plex frequently omits the stream-level codec
when its analyzer hasn't seen the file yet. MAX(NULL) collapsed
the row's format to NULL, which the frontend parser mapped to
AudioFormat::Unknown — the Library grid then showed UNKNOWN even
though the file's container was clearly "flac" / "mp3".
- bit_depth = NULL. Plex commonly omits bitDepth on older releases
entirely; the album row inherited the gap and the UI rendered
"--/44.1" next to an otherwise valid quality string.
Two fallbacks applied to the plex_aggregated CTE only — the local
CTE is untouched because local indexing always writes a non-null
format string and a real bit depth via Symphonia:
- format: COALESCE(MAX(codec), MAX(container)). The `container`
column is populated for the same media and typically carries the
same value ("flac", "mp3"), so it's a safe fallback when codec
is fully NULL across the album.
- bit_depth: COALESCE(MAX(bit_depth), CASE WHEN <lossless> AND
sample_rate <= 48000 THEN 16 ELSE NULL END). Lossless formats
at CD-range rates are virtually always 16-bit redbook, so we
default to 16 in that narrow window. Higher rates stay NULL so
we don't lie about 24 vs 32. The per-track view shows the real
value when the user clicks in (lazy hydration via
plex_get_track_metadata).
cc
The /track/getList endpoint caps at 50 IDs per call. Sending more in a
single POST either errors out or silently truncates, both of which take
the entire call site down.
For QConnect this surfaced as: pushing a >50-track playlist from the
Qobuz Android app to QBZ caused the materialization step (corebridge
get_tracks_batch hydration) to return an error, abort, and leave only
the 1-track placeholder queue from the earlier align-cursor pass. The
visible symptom was "only the first track loads on Connect playlists
above 50 tracks".
The fix is at the helper level rather than in QConnect: split inputs
larger than 50 into 50-ID windows fetched serially, preserve input
order in the returned Vec. Single-window path is unchanged for the
common case. Same return shape, same error type, so the four call
sites (corebridge.materialize_remote_queue, qbzd qconnect, V2
v2_get_tracks_batch, qbzd catalog) all benefit transparently without
edits to QConnect orchestration code (auto-advance, deletion timing,
queue ordering are untouched).
Serial loop instead of the concurrent fan-out used by get_playlist —
lower risk for the first fix; if QConnect playlists of 200+ tracks
start feeling slow we can revisit with try_join_all.
cc
Clearing the error flag also drops any pending stream_error_message.
A fast record_stream_error -> set_stream_error(false) -> polling-tick
sequence will silently swallow the toast — that is the intent (a
recovered error is not user-actionable), but it surprises future
readers, so document it inline.
Pairs the existing stream_error AtomicBool with an Option<String> that
the audio thread can populate when a stream init failure has a
meaningful cause to surface to the user (currently: the macOS
CoreAudio rate-mismatch path).
set_stream_error(false) also clears the message; record_stream_error
sets both atomically; take_stream_error_message drains so the UI
fires its notification exactly once per error.
evaluate_stream_recreate runs on every track change on the audio
thread, and each call issued two HAL syscalls
(resolve_output_device_id + get_nominal_sample_rate). Cache the result
per-thread for 750ms — long enough to absorb the bulk of repeats
within an album, short enough to notice a system rate change.
Cache invalidates on device-name change, so switching the output
device still triggers a fresh query.
User-reported: Local Library Albums view showed the folder name as
the album title even when the album tag in the track metadata had
a different value — e.g. folder `Alle Songs` shadowed metadata
`ALBUM.`. Cause: the three metadata-grouped album queries (full-load
`get_albums_metadata_grouped`, chunked `get_albums_metadata_page`,
and `count_albums_metadata_for_page`) projected the title via
`COALESCE(album_group_title, album, 'Unknown Album')`.
`album_group_title` is the snapshot the scanner wrote — it
contains the tag's album when the tag was present at scan time,
otherwise the folder name. So whenever a tag was missing or
changed after scan, the folder name would win over the live
`album` column.
Fixed the priority by querying `album` first and falling back to
`album_group_title` only when the live metadata is empty / NULL /
'Unknown Album':
COALESCE(
NULLIF(NULLIF(TRIM(album), ''), 'Unknown Album'),
album_group_title,
'Unknown Album'
)
Folder-grouped queries (which use `album_group_key` first and
intentionally prefer folder-derived titles) are unchanged.
Closes#411
cc
Two bugs from the same session:
- The plex_aggregated CTE prefixed `group_key` with `'plex:'`, but
`plex/mod.rs::plex_album_key()` already returns `"plex:<hash>"`.
Result: album.id was `"plex:plex:<hash>"` from the chunked path,
which `handleAlbumClick` couldn't resolve, so opening an album
showed NO TRACKS. Fixed: use `COALESCE(album_key, 'plex:' ||
rating_key)` so populated rows pass through untouched and only
the rating_key fallback gets the prefix added. GROUP BY mirrors
the same expression for consistency.
- On AlbumCardLibraryLite the scrim (`.cover-wrap::after`) and the
overlay container (`.overlay`) were both at `z-index: 2`. Pseudo-
elements paint after all DOM children at the same z-index, so
the scrim ended up ON TOP of the buttons, hiding them on hover.
Re-laid the layering so it's monotonic top-to-bottom:
1 cover img
2 scrim
3 overlay (action buttons)
4 select-checkbox + source-badge-slot
Plus: restored the dark `rgba(0,0,0,0.85)` hover scrim itself,
removed the `::after` accent-border ring + title-colour cue. Per the
2026-05-12 user call ("Hasta quisiera el overlay de regreso") — the
Plex thumbnail downscale (9f417999) made the scrim's opacity
transition cheap again since the underlying paint is now 220x220
not 1000x1000. Buttons go back to opacity 0 → 1 on hover with a
single transition on `.action-buttons`. White-outline circles
(transparent fill) since the scrim provides contrast.
cc
Plex cache lives in a separate SQLite database
(`<data_dir>/qbz/plex_cache.db`), which is why `v2_library_get_albums_page`
previously only saw local_tracks and the LocalLibraryView chunked
path had to fall back to the legacy virtualizer when Plex was on.
That fallback meant Plex-dominant users (which is the common case)
got none of the chunked-store / recycling-pool perf work.
Now consolidated server-side via SQL `ATTACH DATABASE` + `UNION ALL`:
Backend (qbz-library):
- `get_albums_metadata_page` accepts `plex_cache_path: Option<&Path>`.
When the file exists it `ATTACH`es the plex DB to the connection,
runs a CTE that aggregates `plex_cache.plex_cache_tracks` by
`album_key` into the same column shape as local_aggregated, and
UNIONs both before the filter/sort/pagination step. DETACHes at
the end so subsequent queries don't see stale attachment.
Plex group_key prefixed with `'plex:'` so it doesn't collide with
local group keys; `'plex'` literal in the `source` column drives
the frontend's SourceBadge mapping.
- Schema differences smoothed via `CAST` in the plex arm:
`duration_ms / 1000 → total_duration` (matches local seconds),
`sampling_rate_hz → sample_rate` (INTEGER → REAL).
- `count_albums_metadata_for_page` mirrors the same UNION so total
is consistent across empty-page edge cases.
Backend (Tauri command):
- `v2_library_get_albums_page` accepts an `include_plex` bool.
When true, resolves the plex cache path the same way
`plex/mod.rs::open_plex_cache_db` does and passes it through.
Frontend:
- Chunked store's `LocalAlbumsQueryParams` carries an `includePlex`
flag, included in the `paramsEqual` check so toggling Plex
visibility invalidates the chunk cache.
- LocalLibraryView drops the `!isPlexLibraryEnabled()` guard from
both the `$effect` that calls `setParams` and the render
condition that picks chunked vs legacy. The new path serves
Plex-only, local-only, and mixed libraries the same way.
cc
Phase A of the Plex-style port: the Albums grid no longer keeps the
full collection (1200+ rows on the user's library) in reactive Svelte
state. Instead, it pulls a single chunk at a time from a new Tauri
command, with the recycling-grid pool firing `requestIndex` for
indices whose data hasn't arrived yet.
Why this matters: Phase B (recycling pool against an in-memory array)
removed mount/unmount churn during scroll but left two costs in
place — every reactive tick still ran the `$derived` filter/sort
pipeline over all 1200 records, and the records themselves still
sat in Svelte's reactivity graph. Under software compositing the
combined cost was still the dominant scroll-time spike.
Server-side pagination eliminates both: sort + filter run in SQL,
only the visible chunk's records are in memory, and the per-tick
$derived work disappears.
Backend:
- `qbz-library/src/models.rs` — new `AlbumsMetadataPage` struct
(`{ albums, total }`) returned by the paginated query.
- `qbz-library/src/database.rs` — `get_albums_metadata_page`:
reuses the existing track-aggregation CTE from
`get_albums_metadata_grouped`, adds a HAVING-style search filter
over the aggregated title/artist (mirroring the legacy
`matchesAlbumSearchFast` semantics), validated allowlist-based
`ORDER BY` for artist/title/year × asc/desc, and
`LIMIT ?/OFFSET ?` on the filtered set. Total comes from
`COUNT(*) OVER ()` in the same query for one round-trip; a
fallback `count_albums_metadata_for_page` covers the empty-page
case where the window function has no row to attach to.
- `src-tauri/src/commands_v2/library.rs` — new
`v2_library_get_albums_page` Tauri command, registered in
`lib.rs`. `include_qobuz_downloads` is read from
`DownloadSettingsStore` to match the legacy
`v2_library_get_albums_metadata` behavior.
Frontend:
- `src/lib/discovery-v2/localAlbumsChunkedStore.svelte.ts` — new
class-backed store with `version` + `total` as `$state`, sparse
`Map<chunkIdx, LocalAlbumLite[]>`, in-flight `Set<chunkIdx>`
dedup, and a generation counter that drops stale responses if
the user switches filters mid-flight. `getAlbum(globalIdx)` is
sync (null for missing chunks); `requestIndex(globalIdx)` fires
the fetch.
- `src/lib/components/views/LocalLibraryView.svelte` — when the
Albums tab is in grid mode AND grouping is off AND no quality
filters are active (the common case), render through
`AlbumGridPool` bound to the chunked store. Other paths
(grouped, filtered, list mode) keep the legacy
`VirtualizedAlbumList` so we don't lose features we haven't
server-side yet. A `$effect` mirrors search/sort/source state
into `setParams` whenever they change; the store no-ops on
unchanged params.
Plex consolidation (other source listed in the audit) deferred to a
follow-up: `plex_cache.db` is a separate SQLite file, so unifying
needs `ATTACH DATABASE` + `UNION ALL` from the same aggregation,
which is a clean delta once the local-only path is verified.
cc
The error strings flow up through stream_error and end up in user-
visible logs. Replace the internal-shaped messages ("opened at XHz
while device nominal rate is YHz") with text that names the cause,
the remediation, and the reason the fallback path is refused.
CPAL caches device configs and can briefly return a stale sample rate
right after CoreAudio changes the device nominal rate. Before this
change, opening at the stale rate failed the post-open validation and
returned Err, leaving the user with no audio on the first track after
a rate change.
Drop the bad sink, re-resolve the device, re-query nominal rate and
try once more with a 50ms backoff. Two attempts in total. Turns the
racy window into transparent recovery.
Extract the boolean decision out of evaluate_stream_recreate into a
pure function so the truth table can be tested without faking a
MixerDeviceSink or AudioSettings mutex. Eight cases cover the matrix:
no stream, unchanged format, format change with and without each
bit-perfect backend, and the CoreAudio shared-mode rate-mismatch path.
The rate-mismatch detection block was duplicated between the Play and
Stream handlers, and each handler acquired thread_settings.lock() three
times in a row. Extract into one helper that reads settings once and
returns every flag the recreate decision needs.
Also drops a dead else-branch that logged "Creating initial audio
stream" — unreachable inside if stream_opt.is_some() — and the
redundant info-log that re-announced the rate-mismatch reason at
recreate time (the warn-level log inside the helper already says
"Recreating stream to avoid wrong-speed playback").
The field tracks the decoded source format, not the output stream's
hardware rate. The names are now explicit so future readers don't
conflate the two — on macOS shared mode these can legitimately differ
since Rodio resamples internally.
Two cooperating changes that flatten the long sort tail in the Tracks tab:
1. search_with_filter now ORDER BYs (album, COALESCE(album_artist, artist),
disc_number, track_number, title) COLLATE NOCASE — matching the default
album-grouped browsing. SQLite handles 100K rows in well under 100ms;
the JS sort took several seconds for the same volume.
2. groupTracks replaces the chained localeCompare with decorate-sort-
undecorate using precomputed composite keys (lowercase strings with a
sentinel separator and zero-padded numeric segments). The Intl collator
instantiation per comparison was the bulk of the per-pass cost.
Both changes preserve the existing ordering semantics for the album, artist
and name modes; locale-specific accent ordering becomes lowercase ASCII
order, which is acceptable for library browsing.
Tier 3 of three.
cc
Lightweight COUNT(*) helper for the tree-mode scan-root rows that
needed a recursive descendant count. Same network-folder filter as
the listing primitives.
cc
list_folder_children, list_folder_tracks, list_folder_tracks_recursive
now apply the same network-folder exclusion that flat-mode search
honors, so the tree rail and recursive playback see the same source
of truth. Tests cover the on/off cases.
cc
Verify the recursive listing returns an empty Vec (not an error) when
the requested folder path has no descendants — frontend treats empty
as "nothing to play/queue" and skips the toast.
cc
vicrodh