Track the active source id and probe when it changes, instead of a remembered probed id.
The behaviour is the same for switching sources, but the intent is now explicit: switch
source, probe it.
An existing token-auth source with an already-synced library would fail playback before
any API call ran: stream and cover-art URLs are built by uri() and fetched by
ExoPlayer/Coil, outside stream(). Probe the active source once when it uses token auth so
the 41 flips and persists before any URL is built.
Also base the retry on the flag the request actually used (concurrent requests no longer
lose their retry), report the fallback from ping() so Add Source flips the toggle instead
of reporting success and saving token auth on, and make the callback non-null.
Navidrome refuses the Subsonic token scheme for LDAP users with error code 41. Catch it
once in SubsonicClient: flip that client to the password scheme, tell SourceRepository to
persist the change for the source and show a toast, then retry the request. Only 41 is
retried, the flip is idempotent, and a client never retries more than once.
Spell out the sync contract: browsing, filtering and sorting never touch the network,
only media bytes and images are on-demand, and a server-derived sort key has to land in a
local column at sync time rather than being requested lazily.
- resolve() no longer discards a stored seed when the bitmap cannot be decoded (it used to
return null, dropping the theme for pre-v12 rows); it falls back to the stored pair
- fold the busy flag into the seed cache value so eviction is atomic (a busy entry could be
evicted from the separate flag LRU while its seed stayed, silently disabling the scrim)
- the artist top bar samples heroBarColor at scroll 0 to match the static HeroGradient, and
drops the wrong scroll estimate and per-pixel recomposition
- rememberOverlaidNameBusy keys its state so the cached flag is read on key change, not only
the first composition; bottomBandBusy bails out on an empty bitmap
- ArtworkSeedStore now carries the flag; artwork_seeds gains a nullable nameBusy column
(DB v12, migration 11->12) so it survives cold starts with no decode
- resolve() computes the flag from the same bitmap it extracts the colour seed from, so
there is no second decode, and persists it alongside the seed
- overlaidNameBusy reuses the seed job and reads the cached flag synchronously, so the
scrim is known before the art shows; drop the fade
- bottomBandBusy now uses perceptual (gamma-space) luma, edge density and a bright
fraction, which catches busy art on dark backgrounds (issue #40) that mean/variance missed
The busy check can only resolve after the art's first decode, so the scrim appeared in a
single frame. Animate it over 180ms so it eases in behind the name.
ArtworkTheme swapped between a bare content() and a MaterialTheme wrapper depending on
whether a scheme existed. When the first track's artwork (or a screen's) resolved, that
branch change remounted the whole subtree, reloading images, resetting scroll and
restarting the mini player. Always provide a MaterialTheme, falling back to the ambient
scheme when there is no artwork, and cover it with a test.
- the artist top bar now uses the default app-bar title style and the art-derived
heroBarColor instead of a hardcoded headline/black/white, so it follows the artist theme
- add a bottom scrim behind the overlaid artist name only when the art under it is bright
or noisy, measured from the seed bitmap (so no extra decode)
- SyncManager now uses a conflated request channel instead of tryLock, so a
request that arrives during a sync (e.g. add-source switching the active
server) is queued and runs afterwards against the then-active source
instead of being dropped.
- Paged loops count raw server rows, not accepted entities, so an empty-id
row on an otherwise full page no longer ends the scan early and prunes
the unseen tail.
- The cap now fails when the page just past it is non-empty, not only when
it is full, matching the bound the error message reports.
- fetchRank rethrows the cap instead of downgrading an over-cap album scan
to empty ranks.
- Docs: the mutex lives in SyncManager, and the prune still holds the
existing-id set.
A full page at the cap is not proof of truncation, so a library whose
size is an exact multiple of the page size failed for no reason. Each
paged loop now keeps going one page past the cap and only throws when
that page comes back non-empty; a library up to the cap syncs normally.
Add a boundary test for the exact-cap case.
Also note the accepted per-batch observer invalidation churn in the sync
code and the architecture doc.
- Paged loops now throw when they hit maxPages with a full last page, so a
truncated enumeration aborts the entity's collect and skips its prune
instead of deleting rows the sync never saw. The cap is injectable for
tests.
- Move sync serialization into SyncManager (tryLock coalesces duplicate
requests) and construct SyncService there; SourceRepository only exposes
the active MusicSource. Removes the repo <-> sync import cycle.
- Re-throw CancellationException instead of reporting a cancelled sync as
failed.
- Add buffer(1) to the outer albums/songs/playlistSongs flows so their
flowOn channel cannot hold 64 batches.
- Rank prefetch is best-effort (missing frequent/recent no longer aborts
the album sync).
- Skip playlist entries without an id, keeping positions contiguous.
- Chunk stale-disc deletes like the other prunes.
- Update the implementation plan line that still described the old
single-transaction sync.
- SourceRepository now holds a Mutex around sync(), so overlapping
requestSync() calls can no longer interleave per-batch writes and
stale-set prunes. The per-call SyncService lock could never do this.
- parse() wraps SAXParseException/IOException as SubsonicException, so a
truncated or malformed response is a clean sync failure instead of a
raw parser exception escaping the empty-search fallback.
- A response with no status attribute is accepted again, matching the
old DOM behaviour (only status="failed" fails).
- Simplify the SAX handler (drop sawRoot and StatusHandler), make the
entity mappers internal, and drop an unreachable guard in write().
Tests: pagination and empty-search caching for songs, album pagination,
empty-id filtering, cancellation, SAX error paths and structural edge
cases, cross-source prune isolation, emptied playlists, disc-set shrink,
multi-chunk deletes, change detection via total_changes, playlist id
filtering, and SyncManager cache invalidation on failure.
The DOM-based sync materialised the whole library in memory before
writing: SubsonicSource accumulated every paged response into one list
and SubsonicClient built a full DOM per response, so peak memory grew
with the library size (hundreds of MB for large libraries).
Replace it with a streaming pipeline:
- SubsonicXml is now a SAX reader that maps entities straight from the
stream. Album disc titles are accumulated from child elements and
attached when the album closes.
- SubsonicClient exposes stream()/check() instead of returning a DOM.
- MusicSource returns batches as Flow<List<T>>; SubsonicSource emits one
batch per response page with backpressure so only one batch is held.
- SyncService syncs each entity sequentially, commits one transaction
per batch, and prunes with an existing-minus-seen stale set. Playlist
song tails are pruned once the whole flow has been counted, and
orphaned playlist_songs rows are removed with a single delete.
Sync is no longer atomic: on failure the committed batches remain and
the queue's library cache is now invalidated on failure too.
- QueueRepository.sourceName returns null for a blank album/playlist name so the now
playing title fallback applies instead of rendering empty
- resolve the now playing source title with remember + LaunchedEffect instead of
produceState so it resets synchronously when the queue context changes
- extract the repeated status bar scrim into Modifier.statusBarScrim() and use it in
AppRoot, LibraryScreen and HeroDetailScaffold
On the library no navigation happens, so ours was registered before the NavHost's and
lost; keying the DisposableEffect on the overlay flags re-adds it last as soon as the
overlay opens.
The NavHost registers its own back callback after ours, so on a pushed route it won
the back and popped the route under the now playing overlay. Re-register our callback
on every navigation (DisposableEffect on the current route) so it is always the most
recent, and toggle its enabled state with the overlay/queue.
Keeps the NavHost composed (the dropped-NavHost version left the still-transparent
overlay over black) while stopping the controller from winning the back over the
overlay's BackHandler. Back now: queue -> now playing -> pop the route.
- restore the now playing overlay (the destination had too much moving during the
transition); while the overlay is open, omit the NavHost so it has no back callback
to steal the back from the overlay's BackHandler
- first disc header gets a little top padding (16dp), the rest keep 40dp
- keep the fab glyph container 28dp so the filter dot aligns as before
The overlay fought the NavHost over the back: OnBackPressedDispatcher hands the back
to the last-registered enabled callback, and the NavHost registers its own as part of
its composition, so it intermittently popped the route underneath. A destination lets
the back stack own it.
- add Routes.NOW_PLAYING with a vertical slide, opened by tapping the mini player or
flicking it up; back or a downward fling pops it
- hide the mini player while on the now-playing route so it fills the screen
- the static scrim now follows the route instead of a drag progress
- drop the nowPlayingOpen/progress overlay state and its test
- register the now playing and queue back callback from a LaunchedEffect instead of BackHandler: the NavHost registers its own callback from a LaunchedEffect, so a BackHandler composed before it loses the back and pops the route underneath
- NowPlayingBackTest registers the overlay handler before the NavHost (the losing order) and asserts back closes the overlay without popping the route
- more top padding on disc headers, except the first which sits flush
- LIST_OPTIONS_WEIGHT 0.15 -> 0.11
- Now Playing: the type line reads "songs" for the library songs list, source line stays "Library"
- Album detail: align the disc icon over the track numbers and the disc label with the track titles (16dp start + 18dp icon + 16dp gap, matching the ListItem columns)
- Add NowPlayingBackTest covering back-from-overlay not popping the destination
The bundled Material icon set has no weight axis, so the FAB now draws the
three-bar glyph with a configurable stroke thickness instead of using Segment.
- FAB: use the Rounded Segment icon so the glyph reads thick
- Now Playing: two-line header, an uppercase "Now playing: <type>" label (disc-title style) above the source, both left aligned
- CoverArt: remember the aspect ratio every image loads at and seed the non-square placeholder from it, so the detail hero does not start square and snap
- derive the now playing fallback title from the current item for every kind except the songs list, so a track reordered into the album queue (now a Song entry) still shows its album
- reset the resolved title to the fallback when the queue context changes, so crossing albums does not briefly show the previous name
- start the now playing scrim fade later (0.85) so it darkens the sheet, not the library, while the sheet is still sliding up
- reuse the thumbnail's onSuccess painter ratio instead of loading a second thumbnail painter for the aspect ratio
- drop the now-unneeded ExperimentalLayoutApi opt-in on AddSourceScreen (imePadding is stable)
- CoverArt: use the thumbnail's intrinsic ratio until the full-res painter loads, so non-square art no longer snaps
- Now Playing: move the top scrim out of the sliding screen into MainNavigation, static under the status bar, fading in as the screen settles at the top
- Now Playing: titlebar shows the playback source (album/playlist name, or Library) instead of "Now playing"
- Add server: keep the test/save bar above the keyboard on first launch
- Titlebars: use the theme default title style instead of headlineMedium
- Library: clicking an item in a searched tab dismisses the search
- Library FAB: use the AutoMirrored Sort icon at 28dp so the glyph reads a little thicker
The header was the same size as the track title, so render it uppercase
labelLarge with letter spacing to separate it from the tracks. Use the
outlined FiberSmartRecord icon.
DiscFull reads as an error icon, so use FiberSmartRecord, render it on
the "Disc N" fallback too, sit it 8dp from the edge with 8dp before the
text, and size it 20dp.
Use Icons.Rounded.DiscFull, which is distinct from the album tab's
Icons.Rounded.Album, and only beside a real title rather than the
"Disc N" fallback.
When two discs of an album share a title, showing "Live" alone is
ambiguous, so render "Live: Disc 1"/"Live: Disc 2". Unique titles stay
as-is. Also drop the header from titleLarge to a semibold titleMedium so
longer titles fit on one line.
Navidrome and gonic expose OpenSubsonic discTitles on the album response.
Parse them into a transient Album.discTitles map and persist each album's
titles in a child discs table (sourceId, albumId, disc, title) cascading
from albums. Songs keep only their disc number, and the album screen joins
by disc to render a title when one exists, falling back to "Disc N".
Titles are read opportunistically: a server that doesn't send discTitles
just leaves the table empty, so no API version negotiation is needed.
Titles apply to single- and multi-disc albums, are pruned when removed,
and are stored per album.
Schema 10 -> 11.
Drives the pager across the Artists to Playlists transition and asserts the
Songs icon only moves one way. Fails on the min-width clamp version (738-727-745)
and passes on the natural-48dp version. LibraryTabs is internal so the test can
drive its PagerState.
A min-width clamp kept an incoming tab at 48dp until its label passed ~4dp while the outgoing label shrank immediately, so the tab row dipped left and then grew right. With the tab naturally 48dp wide (8dp/16dp padding around the 24dp icon) the width tracks the label continuously and the row moves one way.
austinried