The three-line collapsed bio let its More control sit under the star and overflow buttons, and the collapsed state had no bottom gap before the Albums header. Move the actions up, reserve their width on all three collapsed lines, and add the same bottom gap the expanded state uses.
Move change detection into a desired CTE with UPDATE ... FROM so mass updates do not turn the trailing WHERE into a quadratic correlated scan, and so no-op passes write zero rows. Fold the trigger optimization into the 35->36 migration and re-check/rewrite it from onOpen, so a future migration cannot silently restore the full UPDATE triggers. Re-resolve artist detail art when the active source becomes ready, and cover artist search renames in a Room-level test.
The materialized art update rewrote every artist row on every library-changing sync. That fired the FTS5 content-sync triggers for each artist, which was the real cost: the same SQL is ~10ms with the triggers removed and ~4.7s for 100 changed artists with them in place. Only update rows whose cover lists actually changed, and replace the generated artist search triggers with UPDATE OF name, sortName so art-only writes never touch the FTS index. Existing databases get this through migration 36->37; fresh databases get it from the onCreate callback.
Cap the resolver at four covers, re-rank after deduplication so duplicate covers do not consume the grid budget, carry a thumbnail per fallback step, and advance to the grid when both the full image and its thumbnail fail. Negative-cache failed provider-image fetches per artist URL, clear materialized downloaded covers when all downloads are removed, and drop redundant plumbing.
Store the OpenSubsonic artistImageUrl and two materialized cover-id lists on each artist row, backfill them in the 35->36 migration, and refresh downloaded covers only for affected artists when downloads change. Icons and the artist hero now resolve provider art, server artist art, then the same materialized grid synchronously, with a definitive-failure fallback so a dead provider URL does not strand the letter placeholder.
Download artwork also caches the provider image for the track and album artists, falls back to the artist cover when that fetch fails, and keeps both keys across sweeps. Query timing for the materialization pass is down from roughly 3.4s to 1.9s on the 20k-song benchmark.
Load up to four thumbnail covers per visible artist row on demand and render the same 1-4 layout inside the circular artist icon. Lists stay on the expensive per-row query for now.
When an artist has no portrait, build a 1-4 cover hero from album credits on demand. Uses full-res for one cover and thumbnails for 2-4; three covers use a Y split.
Exercise context and block drags against the queue authority: all context positions across the up-next block, a far move that crosses the window, a context move while an up-next track plays, and moving the playing up-next track.
A context drag that crosses the up-next block moves the block's anchor in the repository but the old player.move used the combined drop position, leaving the two in different orders. Rebuild the window for context moves that happen while a block exists, in both the in-window and window-offset paths.
Also drop the redundant rebuildWindow snapshot parameter, the unused media-item context read-back, and add coverage for shuffled album context, ranged album moves and ranged up-next moves.
Use album gain for items whose queue entry references an album and track gain for playlists, singles and other mixed contexts, including shuffled album queues. Carry the resolved context on each media item so the audio thread does not need to inspect the queue.
Keep a track moved within an album as an album range so it stays in album context; moves out of playlists, artists, genres and other mixed contexts remain pinned song entries.
Use symmetric page margins for the artist bio, let lines below the action buttons use the full width, and stop the expanding bio from shifting the hero gradient.
Keep source line breaks, treat the triple-space paragraph separator some servers return as a blank line, justify wrapped bio lines, and append a bold Less control to the expanded bio.
Retry a failed download bio fetch instead of blacklisting the artist for
the rest of the process, and give a failed or offline refresh its own
message so it is not mistaken for a server with no biography.
Wait for the stored row before choosing the bio text, so a cached bio no
longer flashes the loading state on entry, and bound the page refresh so
a stalled response cannot leave the loading state up indefinitely.
Render an oversized first token instead of a blank collapsed line, and
check whether the whole remainder fits in one measurement.
Load the biography when the artist page is entered. A stored biography is
rendered immediately while it refreshes in the background; otherwise the
page starts with the loading message and falls back to the empty state.
Collapse a biography to two lines and append a bold More only when the
text does not fit. Tapping that line expands the biography, after which
the first lines keep clear of the star and overflow buttons and the rest
use the full width.
Artist biographies come from getArtistInfo2, which the library sync does
not call. Store them in their own table so a page open can use the last
value immediately and downloads can keep it for offline use, without
pulling a large text column into every artist list query.
A missing row means the artist has never been fetched; a row with a null
biography means the server had none. That lets the page show the loading
state only until the first fetch completes.
The fallback art reads albums and credits, but a successful sync invalidates through the sources row, so observing those tables only adds refreshes for play, star, and sync writes that do not change artist art.
Avoid re-requesting search focus when the screen is restored, keep the cheaper album-credit-first art lookup, and test restoration plus association-table invalidation.
Keep the search field from stealing focus again when a result is dismissed, observe album and credit tables in artist paging, and simplify the cover fallback to one ranked union. Cover the association-table and invalidation paths in tests.
Fall back to the first album credit cover (then the first track credit) for artist art, keep list searches when opening a result, resume search at the end of the query, and navigate immediately from the artist chooser.
Use 'album' and 'track' instead of a 1/0 flag so the credit kind is readable
without a comment. This column is a constant in the SQL, not a stored value,
and only the per-artist credit query compares it, so the text costs nothing
meaningful.
Default the second row to a greyed "No artist bio available" placeholder.
Keep the collapsed two-line layout with a bold More and the expanded flow
that wraps around the action buttons for the next PR, which will supply the
server biography.
Keep the type and count on the first row with space reserved for the hanging
shuffle, and put the star and more buttons on the blank second row so they sit
at their old height.
Keep the type and count on the same row as the star and more buttons, and move
that row down to where it sat before the split. Give the Albums and Appears on
headers the album and microphone icons alongside the disc-title styling.
Move the star and more buttons from the meta line onto the first section
header so they sit below the hanging shuffle button again, and remove the
per-section counts now that the meta line carries the total.
Distinguish the album-credit rule (any downloaded song on the album) from the
track-credit rule (the credited track itself must be downloaded) in the
repository test, so a swap of those clauses fails.
Assert the combined meta line while loading, cover the downloaded partition
and the plural album/song credit branches, share the album-credit id query
with its plan test, and put the meta line through a translatable string.
Put the artist type and total album count on the single meta row beside the
star and more buttons, then list albums where the artist is an album artist
separately from albums where they only have a track credit. The hero shuffle
button and page-level actions stay where they were.
The split comes from an albumCredit flag on the shared credit rows, so the
online and downloaded album lists are partitioned by the same rule and the
two sections still sum to the stored album count.
Put the download-visibility comment on the constant it describes, keep the
no-scan guards beside the seek assertions, and give the identical-sync test a
positive control for both the artist aggregate and the album play recompute.
Share the artist page SQL with its plan test and assert the artist seeks, so
the test fails if the predicate push-down regresses. Add a query-log test that
an identical sync skips the aggregate, stop the downloaded album badge from
counting a song whose album row is missing, and correct comments that
overclaimed the remaining hand-written credit paths.
Replace the per-credit invalidation with one signal from the sync's write
transactions: if artists, albums or songs changed, the source is marked stale
and a single aggregate pass refreshes counts, plays and the credited flag in
one query. Identical syncs no longer run any of those scans, and a new credit
path cannot forget to invalidate the cache.
A changed sync still reads the credit unions once instead of separately for
counts and plays, and migration 33->34 remains a frozen snapshot rather than
sharing the runtime query text.
The artist page, shuffle, download attribution and artist badge each spelled
out the four credit paths. Reuse the shared credited song and album rows, and
keep only the download-aware album queries hand-written, where the per-credit
download visibility genuinely differs.
The shared song and album rows use UNION ALL so the artist list filter can push
its artist predicate into every branch instead of materialising a deduplicated
union; callers that need a set use DISTINCT or COUNT(DISTINCT).
The credit recompute was keyed on song-to-album moves and relation-table
writes, so a server that changed songs.artistId or albums.artistId while the
plural refs stayed the same left the counts attributed to the old lead. Fold
the lead ids into the change check for both songs and albums.
The artist badge now counts albums from album lead ids and album_artists even
when an album has no songs, and only counts song-credited albums that still
exist, matching what albumsForArtist lists. Preserve the derived albumCount
when the server re-upserts an artist row so an identical sync does not rewrite
it twice, relax the migration guard to either relation table, and share one
album-count order between the plain and downloaded artist lists.
Artist badges and play rankings only counted album credits, while the artist
page, shuffle queue and download attribution already included track credits.
Derive albumCount and the artist plays from the same four-way credit union,
count lead ids that the server omits from the plural refs, and bump the song's
credited artists on a local scrobble even when the track has no album.
Sort the downloaded artist list by the downloaded album count it displays
instead of the synced total, and backfill the artist totals on upgrade so the
fix applies before the next sync.
The preamp prompt promised extra level unconditionally, but a track whose gain has no peak is capped at unity. Say so in the supporting text so the setting does not overpromise for Opus/R128 libraries.
A peak is the only headroom check the gain code has. When the server sends a gain without one (Navidrome does this for Opus R128 gains, which carry no peak tags by design), a positive gain was applied unchecked and the 16-bit sink saturated.
Cap a peakless gain at unity instead: attenuation still applies, but a boost needs a peak to prove it fits. The cap is unchanged when a peak is present.
Update the tests that expected a peakless boost, and cover the positive preamp that can no longer lift a peakless track and the attenuation that still applies.
The adjacent artwork effect keys on the item and queue layout, but the predictions now also depend on the repeat mode. Toggling repeat left the cached neighbours stale, so the edge swipe still resisted until the track changed. Key the effect on the repeat mode too, with a route test that toggles repeat and expects the wrapped previous art.
Lift the wrap arithmetic out of next(), previous() and previousTrack() into nextPosition()/previousPosition() so upcomingItem()/previousItem() cannot drift from the actions again. A one-track queue no longer predicts itself as its own neighbour, online or offline: the wrap would land on the current position, and the actions are a no-op there, so the strip offered a dead swipe.
The now playing strip took its neighbours from previousItem() and upcomingItem(), which returned null at the queue edges online even when repeat would wrap. Swiping there resisted and only a hard fling advanced the track, leaving no artwork to slide in. Mirror next() and previousTrack() so the online path wraps with repeat All, and with repeat One for the next slot.
Use right/down chevrons in the download manager and order downloaded
tracks by disc and track number, matching the album screen. Move sync
concurrency to the end of the network settings, let the artist hero name
wrap to six lines, and leave the album hero title unbounded.
Restore the both-tables-empty fallback in the credit recompute, skip
redundant marker writes, append albumless tracks after album tracks, and make
the paged artist query observe sources. Add tests for an interrupted sync and
the partial-relation migration guard.
Left-join albums in the artist queue queries so albumless credited songs can
be played, downloaded and deleted. Add sortName to the starred index, persist
a per-source stale-credit marker so an interrupted sync cannot strand artist
counts or hide artists, and recompute artist play data in one grouped pass.
Add artists.credited, updated at the end of a sync that changed artists or
credits and backfilled on upgrade when relation data exists. Filter the
artist lists and the downloaded artist list on it, which keeps partially
synced libraries visible and hides display-only rows once credits are known.
Filter the paged, search and home artist lists to artists referenced by the
plural credits, falling back to the full list while the relation tables are
still empty. Display names keep using the singular tag and navigation keeps
using the plural refs.
Collect every credited artist id when resolving album and song sort keys so
the singular id and plural refs can both be looked up, and add regression
tests for the compilation and stale-sort-name cases.
Drop the FTS tables before the backfill so the old triggers do not index rows
that are about to be discarded, and rely on createSearchIndex's rebuild
instead of rebuilding a second time. Treat a whitespace-only artist sortName
as absent, and use the sort keys in the downloads tree too.
Parse the OpenSubsonic sortName attributes, store sort keys on artists,
albums and songs, and sort on them instead of the display names. Resolve
album and track artist sort names from the artists table, and include the
keys in the FTS content so romaji sort names are searchable as well.
Schema 31 rebuilds the sort indexes and search tables, and backfills
existing rows from their display names.
Use songCount as the total for both artist status queries and stop excluding
albumless tracks from download attribution, so the Artists badge and the
artist page agree. Backfill song counts in the 29->30 migration for offline
upgrades, and recompute only when a song moves albums or its credits change
rather than on every play-metadata sync.
Add artists.songCount, recomputed after syncs that change albums or songs,
and derive the Artists tab status map from the download rows plus that count
instead of walking every song. The badge query is now proportional to
downloads rather than library size.
Rewrite artist queue membership as a UNION of indexed song-id branches
instead of an OR that scanned every song, and recompute artist play data from
an explicit credited-album set instead of scanning all albums per artist.
Run PRAGMA optimize after syncing songs and before play aggregation, and
derive downloaded artist lists and filters from downloads rather than all
songs. Count downloaded albums per artist with artist-first branches.
Extend the query timing harness with the artist playback, download and status
queries plus a partial-download fixture.
Keep the server's combined legacy artist/albumArtist string when it is
present, prefer the OpenSubsonic displayArtist when it is not, and otherwise
join the plural artist refs with Navidrome's separator. The normalized value
is stored in the display columns, so song rows, album cards, now playing,
menu subtitles, downloads, search and artist ordering all get it.
Rewrite the artist album lookup as indexed UNION branches instead of ORs
that scanned the song table, and let artistName take the first indexed row.
Keep the type label and album count blank until the album flow has emitted,
so the page no longer flashes "Artist" and "0 albums" before showing the
real values.
A multi-artist album or song now adds one row per artist, so the sheet can
exceed the screen and clip the remaining actions. Cap it at 560dp and scroll
the content inside.
Credit all album artists in play aggregation and recorded plays, include the
relation rows in the downloaded artist list, and fall back to the album artist
display name in download groups.
Also keep the chooser scrollable, close it when the track changes, keep the
singular id in the album-artist label, stop a nested artists container from
swallowing later artist elements, and make the refs encoding length-prefixed
so tag values cannot corrupt it.
OpenSubsonic servers send plural artists/albumArtists refs and can omit the
singular artistId when an album or song has more than one artist. Parse and
store the ordered refs on the item and in indexed album_artists/song_artists
tables, so artist pages, shuffles, downloads and the artist filters include
every credited artist.
Album and now playing artist taps open a chooser when several artists are
credited; context menus get one artist row per artist.
The toggle's behaviour was untested: deleting the Modifier.keepScreenOn(),
hard-coding the row's request or emptying the Switch handler all left the
suite green, so none of the three new code paths were actually pinned.
- NowPlayingKeepScreenOnTest renders Now Playing and asserts a view in the
activity tree holds the screen on while the flag is set and releases it
once the flag clears.
- SettingsScreenTest now covers off->on and on->off on the row, and the
Switch itself, instead of only the off->on row click.
- Remember the keep-screen-on flow so NowPlayingRoute stops tearing down
and rebuilding its DataStore collector on every position tick.
Review follow-ups.
- The snap clamped nothing, so an out-of-range stored value would show a
decibel label the slider cannot reach: clamps now live in the snap
rather than duplicated in the step.
- The screenshot test only rendered the dialog without asserting it
opened, so deleting the row's click would have left it passing while
the capture showed the plain list. It now requires the dialog's Reset.
- "Adjust level on top of the ReplayGain" was not grammatical, and the
mode prompt claimed the server normalises when it only supplies the
tags.
- Drop a test assertion that restated the production formula.
The in-repo comment rule allows a comment only when it names an external
system, version or issue, so both of these now name Compose Material 3,
which is what they are actually about.
Fifty-nine steps across a phone-width track is a tick every few dps,
which reads as a smudge on the bar. The steps are what make the slider
snap in 0.5 dB increments, so they stay; only the tick drawing is
suppressed, through a SliderDefaults.Track that skips drawTick.
- Call it ReplayGain, which is the name of the standard, in the row, the
prompt and the preamp prompt.
- Give the slider room under the prompt instead of sitting against it.
- Put Reset in the button row, where it lines up with Cancel and Done,
rather than above them, and centre the decibel value that is left in
the content.
- Add a screenshot test that scrolls to the section and opens the dialog,
which also composes the new rows; the existing screenshot only captures
the top of the list, so it never rendered them.
The slider alone is awkward for the two things people actually do with a
preamp: zero it, and nudge it by a known amount.
- Minus and plus buttons flank the slider, stepping one grid notch (0.5
dB) and disabling at the ends.
- A Reset button sits under it, beside the current value, disabled when
the value is already zero.
- Both go through the same snapping the slider does, so a button cannot
land off the grid.
Review follow-ups.
- Remove the second "Playback" section header the rows were added
under; the section already had one, so the screen rendered it twice.
- Snap the preamp to the 0.5 dB grid in the dialog. Material's slider
snaps in pixel space and converts back, so the value it reports is near
the configured grid but not on it: -1.0 dB arrives as -0.99999905, was
displayed verbatim and was stored that way until the next drag. The
step count is now derived from the range rather than hardcoded.
- Format the decibel value with the resource locale so translated
layouts get their own decimal separator.
- The controls move to `ReplayGainControls.kt`, since the file now holds
the grid as well as the labels, and the snap is covered by a test that
reproduces the off-grid value without it.
Final slice: the screen that drives the mode and preamp preferences the
playback side already reads. Until now `setReplayGainMode` and
`setReplayGainPreampDb` had no production caller, so Off and the preamp
were unreachable.
- `SettingsViewModel` exposes both preferences and writes them back.
- The Playback section gains a "Replay gain" row opening an Off / Track /
Album choice, and a "Preamp" row opening a slider across the +-15 dB
range in 0.5 dB steps.
- `ReplayGainMode` is ordered Off, Track, Album so the choice list reads
that way; the stored value is the name, so the order is not persisted
and the default is still Track.
- The labels live in their own file so the signed decibel formatting can
be tested, as `formatTrackTime` does.
Float output is the one way the audio processor chain is skipped on this
build, and it is a flag the renderers factory is handed. Nothing sets it,
so nothing happens today; if it is ever turned on, this says so once at
sink construction rather than leaving replay gain silently inapplicable
to high-resolution tracks.
Offload has no equivalent check: it is not a flag this code sets, and
AudioSink.setOffloadMode is only reachable by building the audio renderer
by hand. The factory comment names it instead.
Review follow-ups for the apply slice.
- A negative-infinite gain produced 0 dB, not unity: 10^(inf/20) is 0.0,
which is finite, so the result guard never fired and the track played
silent — the opposite of what the guard is for. A large-enough finite
negative gain underflows the same way, and an infinite gain paired with
a peak survived as the cap. The gain is now rejected before the pow and
an underflowed result falls back to unity.
- A peak carrying no data shadowed a usable peak in the other slot, so
trackPeak=0 with albumPeak=0.5 skipped the cap entirely — the gonic
shape on the default mode. Both peaks are filtered before one is chosen.
- Drop the claim that a period-less flush comes from a sink reset; it does
not, and the guard stands on its own.
- Trim two comments back to naming the Media3 types.
- Revert the observeStar rename, which was for a path that no longer
applies the gain.
The in-repo comment rule now allows a comment only when it names an
external system, version or issue, so the ones carrying rationale move
to this message.
- why the gain rides on the media item: PlaybackService can then apply it
the moment the item becomes current instead of after a library round
trip
- that ReplayGainController is written from the playback thread by
onFlush and from the app thread by the preference collector
- that an empty passthrough input is the pipeline's own shared buffer,
which cannot be put into itself
The Playback section already documents the star button's dependence on
unstable Media3 APIs and what to re-check when bumping the library. The
gain path needs the same treatment: it relies on flush carrying the
timeline and period, on the sink configuring per media item, and on the
processor chain being reached at all.
Review follow-ups for the apply slice.
- Drop the media item transition application. It never supplied a gain
the flush had not already supplied, and as a second writer it raced the
playback thread: it applied the incoming item's gain while the outgoing
item's buffers were still queued, scaling the previous track's tail by
the next track's gain (reproduced at +12 dB). The flush is the only
writer now.
This reverses the earlier decision to keep it as a fallback. That was
made on the premise that it was insurance against a flush that could
not be resolved. The flush covers every transition, and a period-less
flush only happens while the sink is being reset, so the premise does
not hold.
- Ignore a flush that carries no period rather than reading it as "this
item has no gain". Resetting to unity there would have been wrong if a
bare flush ever arrived mid-playback; the test that pinned that
behaviour now pins the opposite.
- Guard the pass-through branch against an empty input, which is the
pipeline's own shared buffer and cannot be put into itself.
- The controller is written from the playback thread now, so its state
updates and the recompute that follows them are synchronized rather
than two independent volatile writes.
- Add ReplayGainSinkTest, which drives the real DefaultAudioSink and
asserts it hands the processor the item being configured. That is the
media3 behaviour this whole feature rests on, and it was previously
asserted only by comments.
The app-thread transition callback was only ever a stopgap: media3 posts
it after the playback thread has already rendered, so the first buffers
of a new item could be scaled by the previous item's gain.
media3 forwards the playback timeline and the current period to
AudioProcessor.flush, and DefaultAudioSink flushes the processors on
every sink configuration, which happens at each media item transition
even when the format is unchanged. That is early enough to set the gain
before any of the new item's buffers are processed, on the playback
thread. See androidx/media#418.
The processor resolves the period back to its MediaItem and reports it,
so the gain maths and the preferences stay in ReplayGainController. That
controller is now written from two threads, so its state is volatile.
The transition callback stays as the fallback for a flush that carries no
period: without it a stale gain would persist for a whole track rather
than for a few milliseconds.
Review follow-ups.
- `ReplayGainController.applyItem` owns reading the gain off a
`MediaItem`, so the transition, the controller and the extras read are
exercised together rather than only in isolation.
- drop the leftover `setTrack(null)` on a missing active source: now that
the transition sets the gain from the item, that line only raced it and
could reset a playing item to unity.
- tests for a zero gain surviving the extras round trip (zero is a real
gain, not an absent one), for the latest item's gain being the one left
applied, and for a missing item applying unity.
The gain was resolved by querying the library from the media item
transition callback, so the buffers the player had already begun
queueing for the new track could be scaled by the previous track's gain.
A quiet track followed by a hot one could clip at the boundary for as
long as that query took.
Carry the gain on `QueueItem` so `queueMediaItem` puts it on the
`MediaItem` extras, and apply it from the item that just became current,
with no library round trip.
That narrows the window rather than closing it: media3 delivers
`onMediaItemTransition` on the app thread after the playback thread has
already rendered for that iteration, so the very first buffers after a
transition can still carry the previous item's gain. The gap is now one
handler hop instead of a database query.
The trade-off is staleness: if a sync changes a track's gain while that
track is queued, the item keeps the old value until it is rebuilt.
Applying a new gain mid-track is of marginal value anyway.
Review follow-ups for the apply slice.
The parse guard in the base branch keeps non-finite gains and peaks out
of the library, but nothing here should depend on that: a value already
stored, or written by a future caller, must not be able to silence a
track or blast it at full scale.
- the gain maths ignores a non-finite peak and falls back to unity when
the result is not finite.
- the processor falls back to unity for a non-finite gain, so
`roundToInt` can never be handed a NaN.
- the preamp is sanitised on both read and write, since `coerceIn` passes
NaN straight through.
- the renderers factory is pulled out of `PlaybackService` into
`ReplayGainRenderersFactory` so a test can assert the sink really is
built with the processor in its chain; the wiring was otherwise
untested.
- a null active source clears the gain rather than leaving the previous
track's applied.
Second slice: read the gain synced in the first one and scale the decoded
PCM by it.
- `ReplayGainProcessor` is a `BaseAudioProcessor` that scales 16-bit and
float samples, saturating on overflow rather than wrapping. It is
installed through `DefaultAudioSink`'s audio processor chain, so it runs
ahead of the default silence-skip and speed processors.
- `replayGainLinearGain` picks the track or album gain according to the
mode, falls back to the other gain when one is missing, adds the preamp,
and caps the result at the peak's headroom so a boost cannot clip. A
non-positive peak means the server has no peak data and is ignored.
- `ReplayGainMode` and the preamp live in `UserPreferences`, defaulting to
track mode and 0 dB. The settings UI is still to come.
- `PlaybackService` resolves the current song's gain alongside its star
state and applies it on every media item transition.
The processor only runs while the audio processor chain does, which
`DefaultAudioSink` skips when float output is enabled. That stays off, as
it is by default.
A second <replayGain> element for the same song cleared whatever the
first one set, because each field was assigned unconditionally and an
absent attribute parses to null. No server is known to split the object
across elements, but the merge costs nothing and the alternative loses
fields silently.
Also covers peaks arriving without gains, which the existing cases did
not.
`toFloatOrNull` accepts `"Infinity"`, `"NaN"` and overflowing literals
like `"1e40"`, so a server (or a corrupt tag it read) could put a
non-finite gain or peak into the library. An infinite peak later caps the
applied gain at negative infinity, i.e. silence; an infinite gain clips
every sample to full scale.
Nothing consumes these columns yet, so this is only about not storing
values that cannot mean anything.
Review follow-ups for the replay gain slice.
`upsertChanged` never checked that its hand-maintained column list and
the row values line up, and SQLite binds a missing trailing value as NULL
without error: adding a column to `songColumns` but not to `Song.values()`
would have written NULLs silently. It now asserts the sizes match before
preparing, so that mistake fails the sync instead.
The replay gain sync test also now re-syncs identical values and asserts
no rows are rewritten, pinning the REAL change-detection clause; an
update there rewrites the `song_search` FTS row for no reason.
Store the per-track replay gain a server reports on OpenSubsonic's
`Child.replayGain` so playback can normalise loudness later.
- add trackGain/albumGain/trackPeak/albumPeak (REAL) to the songs table
and the 27 -> 28 migration
- parse the nested `<replayGain>` element, folding the child-element
dispatch into a single `songChild` shared by the search and album paths
- teach the sync upsert to bind Float, which it previously rejected
baseGain and fallbackGain are omitted: the servers that send the object
do not populate them, so they would only ever be null columns.
Nothing applies the values yet; this slice only mirrors them.
The `search3`-less fallback emitted one batch per album, and
`SyncService.writeSongs` opens a transaction per batch, so syncing a large
library that way ran one transaction per album — thousands of fsyncs, and an
invalidation round for every list observing `songs` with each.
Accumulate what each `getAlbum` returns and flush at `PAGE_SIZE`, so the
fallback writes the same ~500-row batches every other fetch already does.
Albums are fetched concurrently and complete out of order, so the accumulator
sits behind a mutex and the send happens outside it.
Batches are in-flight memory, so the flush is per rows reached rather than
per album page. Measured per `Song` (strings, boxed longs and genre list
included): ~0.7 KiB, so a 500-row batch is ~334 KiB plus ~51 KiB of the
`names`/`genres` structures `writeSongs` derives, where a whole album page
(~5,500 rows) would be ~3.8 MiB. `albumSongs()` is also a bare `channelFlow`,
whose channel holds 64 elements whatever their size, so 500-row batches would
have made its worst case ~21 MiB; it now carries the same `.buffer(1)` that
`entityBatches()` already uses on itself. The rest of the in-flight rows
scale with `syncConcurrency`, so that leaves the default's batch in hand and
a few MiB at the highest setting.
`SubsonicSourceTest.albumSongsAreEmittedInBoundedBatches` asserts the emitted
sizes, [500, 100] for 600 songs spread over 200 three-song albums, which pins
the carried remainder as well as the flush: verified it fails on the
pre-change code (200 batches of 3) and against a mutant that drops the
remainder ([500, 99]). `docs/architecture.md`'s page cap was 20x out of date,
so it is corrected alongside. `ktlintCheck`, `testDebugUnitTest` and
`lintDebug` pass.
The GitHub job's store is ~7.3 GiB but gc-max-store-size-linux is 6G, and
nothing roots the devshells, so cache-nix-action's pre-save `nix store gc`
deletes whichever paths cross the cap before every save. The snapshots land
missing the JDK and the Android SDK components, so both `nix develop` steps
re-download or rebuild them on every run.
Raise the cap to 10G, which clears the store, and bump the key to nix-v3-. The
first run after that downloads the full closure once and saves a complete
snapshot; the gutted nix-v2- snapshots are not restored, so there is one cold
run per ref and v3 is warm from then on.
Compressed at ~0.47x its store the snapshot is ~4.7 GiB at worst, inside the
10 GB per-repo budget. What pushes the repo over that budget today is the four
stale gradle- (v1) entries, which no restore-key can reach and which age out by
LRU.
A scrobble writes albums.playCount/played, which invalidates the album
list's PagingSource. Room then refreshes from anchor - initialLoadSize/2,
and with placeholders off that page is the presented list: LazyPagingItems
index zero becomes the page's first album, so the grid re-groups its rows
and every cover moves by (oldStart - newStart) % 3 columns, while the last
row's width becomes the loaded count % 3 instead of the library size % 3.
Enable placeholders for the album pager so an album's index is its
absolute position, and draw a placeholder cell for an unloaded slot. The
other lists are single column, so a reload re-anchors them by key and they
never shifted.
This only keeps the grid still; the reloads themselves are still
unnecessary and are tracked separately.
Media3 marks a progressive stream live when the data source reports no
length and the extractor no duration, and a live item makes the media
session publish PLAYBACK_POSITION_UNKNOWN, so the notification progress
never advances even though the item carries the library duration.
The app only declares a stream length for a transcode of a track that has
one in a length-estimable format (`declaredStreamLength`), so a chunked
response stays live for an Opus/Ogg/Vorbis transcode (a length estimate
breaks the end-seek they rely on), for a transcode of a track whose
library duration is missing, and for a direct play the server does not
give a Content-Length. The now playing screen shows progress in all of
those because it falls back to the metadata duration; the notification
had nothing to advance.
Wrap the ExoPlayer in a ForwardingPlayer that reports an item with a
declared duration as finite. The session then publishes the real position
and the metadata duration, so the notification shows progress for every
unknown-length stream, not just Opus. The stream is still not seekable;
only the reported liveness changes.
austinried