Discovered undocumented Qobuz API endpoints from bundle analysis:
- POST /playlist/subscribe { playlist_id }
- POST /playlist/unsubscribe { playlist_id }
This is the native "follow playlist" feature that syncs across all
Qobuz clients, separate from QBZ's existing "Copy to Library" (local
copy) and "Favorite" (local flag) features.
Added:
- subscribe_playlist/unsubscribe_playlist in both qbz-qobuz and legacy client
- v2_qobuz_subscribe_playlist / v2_qobuz_unsubscribe_playlist commands
- Bookmark toggle button in PlaylistDetailView (next to Copy button)
- i18n keys in all 5 locales
cc
USB hubs/docks (e.g. Razer USB4 Dock) may need longer than 300ms for
sample rate switching. When PipeWire hasn't finished reconfiguring,
96K/192K tracks play at wrong speed (half/quarter) and sound distorted.
Changes:
- Increase initial clock.force-rate delay from 300ms to 500ms
- After stream creation, query pw-metadata clock.rate to verify the
actual graph rate matches what was requested
- If mismatch detected, retry with additional 500ms delay and re-force
- Log warnings when rate can't be verified so users can report
cc
mupdf 0.5 had struct size mismatches on Arch Linux (mupdf-sys bindings
incompatible with system MuPDF headers). Upgrading to mupdf 0.6 fixes
the bindings.
- Removed the booklet feature gate — mupdf is now a direct dependency
again, matching the v1.1.19 configuration
- Removed stub implementation (always compile real MuPDF backend)
- sys-lib-libjpeg remains excluded (cross-distro libjpeg incompatibility)
cc
xdpyinfo returns combined multi-monitor resolution (e.g. 6000x1600 for
dual monitors), causing false positives like clamping 1926x1590 to
5700x1520. The 95% multiplier made things worse — the fallback size was
larger than any single monitor.
Replace with simple corruption guard: only clamp values above 8K
(obviously corrupt DB entries), fallback to 1920x1080. The window
manager already handles fitting windows to the actual display.
cc
The KWin window rule approach (kwinrulesrc + qdbus reconfigure) was
never actually active because GDK_BACKEND is unset on most KDE Wayland
sessions. In v1.1.18 this meant decorations(use_system_titlebar) was
used directly — GTK CSD appeared (ugly but functional, no CPU issues).
The hack itself, when triggered, caused double titlebar and a CPU busy
loop that blocked networking. Remove all KDE-specific detection and
KWin rule setup. Window now simply uses decorations(use_system_titlebar)
on all platforms, matching the effective behavior of v1.1.18.
The one-time cleanup of stale kwinrulesrc entries is preserved so users
who had the rule from v1.1.14-v1.1.19 get it cleaned up automatically.
cc
When BundleExtractionFailed (no network), the error box only showed
"Retry" with no way to enter offline mode. The timeout case and login
form both had the offline button, but the error state was missing it.
Added "Start Offline" button using existing offline-btn CSS and
i18n keys. Also replaced hardcoded strings with $t() calls.
cc
With ALSA Direct hw: there is no software mixer — volume must be
controlled at the DAC/hardware level. Software volume attenuation
would reduce bit depth and defeat the purpose of bit-perfect playback.
Backend: v2_set_volume forces volume to 1.0 when backend=Alsa+plugin=Hw
Frontend: volume slider disabled (grayed out at 100%), mute/step buttons
hidden, handleVolumeChange and handleToggleMute return early.
Settings callbacks notify +page.svelte when backend/plugin changes to
update the lock state in real-time.
i18n: volumeLockedHw tooltip added to all 5 locales.
cc
Two additional gaps found during audio settings sync audit:
1. convert_to_qbz_audio_settings() hardcoded alsa_hardware_volume to
false instead of reading settings.alsa_hardware_volume. This meant
the player never received the user's ALSA hardware volume preference.
2. v2_reset_audio_settings saved defaults to DB but never synced them
to the CoreBridge player, leaving stale settings in memory.
cc
Two bugs caused the exclusive mode to not work when QConnect was active:
1. Critical audio setters (backend_type, exclusive_mode, dac_passthrough,
output_device, alsa_plugin, pw_force_bitperfect, sample_rate,
alsa_hardware_volume) saved to the per-user store but never called
reload_settings() on the CoreBridge player. The Player kept using
stale settings (often defaults with backend_type=None), falling back
to CPAL/PipeWire instead of ALSA Direct.
2. After login, per-user audio settings were loaded into AudioSettingsState
but never pushed into the CoreBridge player, which was created at
startup with flat-path defaults.
Fix: Add sync_audio_settings_to_player() helper and call it from all
critical audio setters. Also sync settings in activate_session() and
activate_offline_session() right after audio_settings.init_at().
cc
Developer Mode: toggle to show/hide QConnect dev panel button in
the player bar (hidden by default, persisted in localStorage).
Integrations: editable device name input for Qobuz Connect with
default "Qbz - {hostname}". Backend commands v2_qconnect_get/set
_device_name and v2_get_hostname. Custom name is used during the
WS bootstrap handshake.
cc
Place QConnect badge as direct child of badges-group with 3px gap
matching DAC/EXC spacing. Two-row vertical layout: icon on top,
Qobuz/Connect text below. Wire props from +page.svelte to both
NowPlayingBar instances. Add qconnect i18n keys to de/fr/pt locales.
cc
resolve_core_shuffle_order was reordering the server's shuffle_order
to put the current track first. This destroyed the server's concept
of "already played" vs "upcoming" — CoreBridge showed all tracks as
upcoming while the web player correctly split them.
Now the server's raw shuffle_order is passed through unchanged.
CoreBridge's set_queue_with_order finds shuffle_position from
current_index, preserving the played/upcoming split.
cc
mupdf-sys 0.5.0 bindings are incompatible with the current system
MuPDF library (struct size assertions overflow). Gate mupdf behind
a booklet feature flag (disabled by default). When disabled, booklet
commands return an error instead of crashing the build.
cc
- setAutoplayMode now sends autoplay preference to QConnect server
- queue_load_tracks syncs local autoplay mode after loading
- QConnect panel shows autoplay_mode status
Backend _if_remote methods for QConnect autoplay protocol:
- set_autoplay_mode_if_remote: toggles autoplay on the QConnect server
- autoplay_load_tracks_if_remote: loads track IDs into the autoplay pool
Both follow the established _if_remote pattern with diagnostics.
Prevents double-bootstrap on initial connect. The Subscribed
handler now only calls bootstrap_remote_presence when the
connection was previously lost (has_disconnected=true), not
on the first connect where connect() already bootstraps.
Also sets has_disconnected=true in the Disconnected handler.
New _if_remote backend commands for set_volume and mute_volume.
handleVolumeChange and handleToggleMute in both main view and miniplayer
now check for peer renderer before setting volume locally. Media
control set_volume also routed through the guard. When no peer is
active, volume is set locally and reported to QConnect server.
When a peer renderer is active, the MPRIS/media-key stop command now
sends SetPlayerState(STOPPED) to the QConnect server instead of stopping
local playback. Falls back to local stop on error or when no peer is
active.
Replace immediate seek-on-drag with the NowPlayingBar pattern:
preview the position visually during drag, commit the seek only on
mouseup. Both PlayerControlsCompact and MiniPlayerFooter now use
dragPreviewTime/effectiveTime, matching NowPlayingBar behavior.
Replace local isMuted/previousVolume state in PlayerControlsCompact and
MiniPlayerFooter with the global toggleMute from playerStore. Muted
state is now derived from volume === 0, matching NowPlayingBar behavior.
Pre-mute volume is persisted in localStorage by the store.
When a peer renderer is active, shuffle/repeat/queue-play/history-play
commands are now routed to the QConnect server instead of being applied
locally. Two new backend commands (toggle_shuffle_if_remote,
cycle_repeat_if_remote) follow the existing _if_remote pattern.
Guards added to both +page.svelte (main + immersive) and miniplayer.
When QBZ is the local renderer in a QConnect session and the web player
presses next, the server sends SET_STATE with the new track. The backend
correctly starts playback, but the frontend's gapless transition handler
saw the track_id change, treated it as a local gapless event, and called
nextTrack() — which advanced the CoreBridge queue every 1s polling cycle,
flooding the queue and destroying the shuffle order.
Root cause: shouldQconnectSuppressLocalPlaybackAutomation only suppressed
when a peer renderer was active (activeRendererId !== localRendererId).
When QBZ IS the renderer, remoteControlMode stayed false, so all local
automation (gapless, auto-advance, resume-from-stop) remained active and
fought the server.
Fix: suppress local playback automation whenever QConnect is connected,
regardless of which renderer is active. The server is the authority for
track transitions in all QConnect scenarios.
cc
SrvrCtrlQueueTracksReordered and SrvrCtrlQueueTracksRemoved were not
triggering AskForQueueState, so the local queue never reflected
reorder/remove actions performed via QConnect. Same pattern as the
existing shuffle resync.
cc
navigateTo() was passing the tab id to the navigation store but never
setting the local homeTab state that HomeView reads. Editor's Picks
and For You always landed on Tab-Home.
Backend now sends 512 spectral bands (from 4096-point FFT).
SpectralRibbon was still expecting 190, causing the length check
to fail and silently discard all spectral data.
Add linebed to the per-panel FPS settings dropdown in Settings.
Add per-panel default override system to fpsConfig so linebed
defaults to 60fps while other panels stay at 15fps.
Translations already existed in all 5 locale files.
Terrain extends vertically from 12% to 88% of canvas height (76%
spread). Width contained: back 48%, front 72%. 200 lines fill
the tall rectangle with visible spacing between them.
Three issues from user feedback:
1. Right side empty: 512 log bands span 20Hz-20kHz but musicvid only
goes to ~1752Hz. Now display first 350 bands (~3.5kHz), cutting
the dead treble zone that wasted 60% of visual width.
2. Viewpoint too low: terrain vertical spread reduced to 33% of height
(was 54%), centered on canvas. This simulates a much higher camera
looking down at the table rather than crouching at table level.
3. Mountains too tall: maxAmplitude 0.35 → 0.18. Peaks should be
modest bumps on the terrain surface, not towering spikes.
Also reduced to 150 lines for more visible spacing between them.
User feedback with annotated screenshot identified three issues:
1. Perspective (red): Was infinite conveyor belt (back 20%, front 105%).
Now table shape: back 50%, front 82%, contained within canvas.
2. Peak distribution (green): Exponential 6/3 with per-frame normalization
crushed everything except loudest bin. Reduced to 3.5/1.8 so moderate
peaks stay visible across the full spectrum width.
3. Back edge (blue): Opacity faded to 0.06 at back, hiding where lines
begin. Now 0.45 minimum — all lines clearly visible including back edge.
Also: thinner uniform lines, moderate amplitude (0.35 vs 0.55).
The 3D mountain effect in musicvid comes from many temporally-close
lines stacking in perspective. At 15fps with 120 lines, adjacent lines
differed too much — no "body" to the mountains.
- NUM_LINES 120 → 200 (match musicvid)
- Remove frame throttle, render at full rAF rate (~60fps)
so adjacent lines are ~17ms apart, creating solid ridges
- Perspective: backY 0.25, frontY 0.98, depth power 1.8
packs back lines tightly (2-3px apart) for solid terrain surface
Implement transformToVisualBins() with power 2.5 exponent from
musicvid.org to spread bass energy across full visual width instead
of clustering in the first ~30 bins. Add edge tapering (10 bands)
to eliminate staircase/square artifact at spectrum boundaries.
Based on frame analysis of musicvid.org's actual LineBed rendering:
- Revert isometric angle to centered symmetric trapezoid perspective
(vanishing point at center-top, NOT upper-right offset)
- Increase to 120 lines for denser terrain surface
- Lower temporal smoothing from 0.35 to 0.05 (musicvid uses 0.03)
for much more responsive, detailed peaks
- Push snapshots every frame instead of every 2nd for fluid scroll
- Wider front lines (110% canvas width) for edge-to-edge coverage
- Higher amplitude (0.45 * height) for dramatic mountain peaks
Replace centered symmetric perspective with angled isometric view:
- Vanishing point at upper-right (75%, 8%) matching camera(-80,150,-10)
- Lines converge from lower-left toward upper-right vanishing point
- Front line extends beyond left edge for full-width coverage
- Non-linear depth curve (pow 1.8) for realistic compression
- Eliminates the "square" artifact from symmetric convergence
New spectrum data now appears at the BACK (top/horizon) and scrolls
toward the FRONT (bottom/viewer), creating the "audio flowing from
far away toward you" effect. Previously data flowed bottom-to-top
which looked like the terrain was moving away from the viewer.
Also slightly increased amplitude scaling for more dramatic peaks.
3D terrain-style frequency visualization using stacked spectrum lines
with perspective. Uses viz:spectral (190 bands) for high-resolution
terrain, with quadratic curve smoothing and black occlusion fill for
depth. Track info displayed bottom-right matching other panels.
TrackRow, AlbumDetailView header, and ImageLightbox were loading artwork
directly via WebKitGTK HTTPS, which fails on systems with TLS database
issues. Switch all three to use:cachedSrc action to proxy through the
Rust backend image cache, matching the pattern used by other components.
is_disc_folder() was matching any folder name containing "Disc N" or
"CDN" anywhere in the name. This caused albums like "100 Popular
Classics, Disc 1" or "Now 75 - CD1" under Various Artists to be treated
as disc subfolders, collapsing all tracks into a single album group.
Now is_disc_folder() only returns true when the folder name consists
exclusively of disc-related tokens (disc/cd/disk, digits, and modifiers
like "bonus"). Album names with extra words are correctly left as-is.
Users must re-scan their library for the fix to take effect.
- dependabot.yml: target pre-release branch for npm and cargo updates
- redirect-pr-to-prerelease.yml: auto-redirect PRs to main → pre-release
with a friendly comment explaining the workflow
vicrodh