The previous detectVerticalDragGestures modifier on the Scaffold
never fired because the body Column applies verticalScroll, which
wins the touch-slop competition for every vertical drag delta
before the outer detector sees it. User-visible symptom: swiping
down on the NowPlaying screen did nothing.
Replace with a NestedScrollConnection. verticalScroll offers its
over-scroll deltas to the nearest ancestor connection via the
standard nested-scroll protocol; when the column is at the top of
its scroll range, downward drag deltas surface in onPostScroll
unconsumed (available.y > 0). Accumulate those toward a 200 px
threshold and pop the back stack.
- Upward over-scroll or any consumed delta resets the accumulator
so a partial drag-down followed by drag-up doesn't latch.
- onPreFling resets on lift-off so a lazy swipe doesn't dismiss
later after the user has released.
- Slider thumb retains its own pointerInput and is not affected.
- Source filter (UserInput) ignores nested-scroll-driven
animations (e.g. flings from inner scrollables).
CacheSettings.prefetchWindow has shipped at default=5 since M8 but
the prefetcher itself was deferred to a follow-up. Without it,
forward skips re-fetch from the network every time and gapless
transitions stall.
New AudioPrefetcher singleton subscribes to PlayerController.uiState
+ AuthStore.cacheSettings, walks queue[currentIndex+1 .. +window],
and runs each missing track through Media3's CacheWriter against
the same SimpleCache the player reads from (PlayerFactory.simpleCache).
DataSpec uses setKey(trackId) to match PlayerController.toMediaItem's
setCustomCacheKey(id) — without this the player would miss the
cached bytes on read-through.
Reconcile is idempotent: CacheWriter is a no-op when bytes are
already resident, so distinctUntilChanged on (queue, index, window)
gates re-runs and the per-item check is cheap. Window slides cancel
in-flight jobs for tracks that have dropped out (skip-prev, queue
rebuild) so a stale prefetch doesn't keep the network busy.
Eager-constructed via the construct-the-singleton trick in
MinstrelApplication, alongside CacheIndexer and CoverPrefetcher.
Mirrors flutter_client/lib/cache/prefetcher.dart's user-visible
behavior (queue-walk + per-track pin + idempotent reconcile)
implemented with the native Media3 primitives (CacheWriter +
SimpleCache) instead of Flutter's AudioCacheManager.pin.
Previous design: android.yml's release job and release.yml ran in
parallel on every tag push. release.yml polled the gitea release-
download URL for up to 15 min waiting for the APK to appear. In
practice the polling step was completing in 11s — either following a
gitea redirect to a 200 page and writing HTML into the bundled APK
file, or returning empty content silently. Either way the resulting
image shipped without a working APK and the web Settings page's
in-app-update section never rendered.
Fix the race architecturally:
- Move the signed-release-build + Attach-APK steps out of android.yml
and into a new android-release job in release.yml.
- release.yml's image-release job declares needs: [android-release],
so on tag pushes the image cannot start building until the APK is
guaranteed-attached.
- Pass the APK between jobs via actions/upload-artifact@v3 +
download-artifact@v3 (the v2 backend isn't supported on Gitea).
This removes the polling loop entirely — the image-release job just
downloads the artifact, renames to minstrel.apk, writes the
.version sidecar, and continues to docker buildx.
- For main pushes android-release is skipped via its if: condition.
image-release uses so it
still runs (the skipped predecessor doesn't poison the chain) and
the download/stage steps gate themselves on the tag context. Main
images ship without an APK by design, same as before.
android.yml is now testing-only: lint + detekt + unit tests on every
push, debug APK artifact on main. Independent of release CI as
requested.
The previous version of the Attach APK to release step ran without
set -x and without capturing the upload's HTTP response. Run 118
(v2026.06.01 tag-build) shows the step reporting success but the
release ended up with zero assets — i.e. the upload silently failed
without surfacing a non-zero exit.
New version:
- set -euxo pipefail so every command is echoed and any failure
surfaces.
- ls -lh the APK before upload so we can see if Gradle even produced
it at the expected path.
- curl -w '%{http_code}' captures the actual HTTP status into a var
and writes the response body to a temp file; both are printed.
- Explicit if check on the HTTP code (200-299 = success) bails
loudly when it isn't.
No behavioral change in success path; on failure the cause is
visible in the log instead of being eaten.
android.yml run 116 (main push on 2534384e) failed the debug-APK
upload step with:
::error::@actions/artifact v2.0.0+, upload-artifact@v4+ and
download-artifact@v4+ are not currently supported on GHES.
Gitea Actions emulates GHES; the v2 backend used by upload-artifact
v4 isn't implemented there yet. Pinning to @v3 keeps the main-push
debug-APK channel working without affecting the tag-push release
path (which doesn't use this action).
The android.yml release build (run 110, job 282) failed at
lintVitalRelease with:
AndroidManifest.xml:13: Error: Remove androidx.work.WorkManagerInitializer
from your AndroidManifest.xml when using on-demand initialization.
[RemoveWorkManagerInitializer from androidx.work]
MinstrelApplication implements androidx.work.Configuration.Provider
and supplies the HiltWorkerFactory, so WorkManager initializes
on-demand at first WorkManager.getInstance(...) call. The androidx-
startup InitializationProvider that ships with the work-runtime
library still auto-registers WorkManagerInitializer, racing the
on-demand path. Lint catches this as a release blocker.
Fix: merge the InitializationProvider entry and remove the nested
WorkManagerInitializer meta-data with tools:node='remove'. This is
the AOSP-recommended fix when an Application is its own
Configuration.Provider.
The release.yml build on main HEAD c0185357 failed with:
go: go.mod requires go >= 1.25.0 (running go 1.23.12; GOTOOLCHAIN=local)
go.mod was bumped to 1.25.0 during the M7/M8 batch but the
Dockerfile's builder stage still pinned golang:1.23-bookworm. Bump
to golang:1.25-bookworm to match.
Every commit on dev that's part of an open PR fires each workflow
TWICE: once for the push event and once for the pull_request event
(observable in run #43 + #44 for the same SHA in android.yml, same
shape across test-web.yml and test-go.yml). The two runs check the
exact same SHA — pure waste.
Drop the pull_request trigger from test-web.yml, test-go.yml, and
android.yml. The dev push already exercises every check that a PR
would re-run. This repo has no fork PRs to cover, which is the only
case pull_request would catch that push doesn't.
Add a concurrency group keyed on workflow + ref with
cancel-in-progress: true to every workflow including release.yml.
Rapid re-pushes (or force-moving the per-day tag) now cancel the
older in-flight run instead of stacking.
Drops the trailing patch digit from the CalVer model. Same-day
re-releases force-move the tag (and overwrite both the docker image
and the release asset of the same name). :latest is now updated by
every main push AND every tag push, so it always reflects the
newest blessed image without a separate cut step.
Comment-only changes — the workflow glob is already 'v*', so the
existing pipeline accepts the new format with no code changes.
Two related cutover changes so the rolling stable channel works end
to end after the Flutter sunset.
release.yml:
- Main pushes now tag both :main and :latest. Main is the protected
post-PR-merge branch and the team's stable channel; pinning to
:latest gets the newest main automatically while :vX.Y.Z stays
available for pinned consumers.
android.yml:
- Asset attach renamed from minstrel-android-<tag>.apk to
minstrel-<tag>.apk (M8 phase 14.4 cutover). Without this, the
server image's bundled in-app-update fetcher in release.yml polls
a filename that no workflow produces, so /api/client/version
returns 404 and the web UI's MobileAppDownload silently renders
nothing. With this rename the existing fetcher resolves to the
native APK with no further plumbing.
No code paths change; both fixes are workflow rewires.
Operator-tunable 0-12s crossfade in Settings → Playback. Default 0
(off). Most albums sound best at 0 — gapless masters, classical, and
live recordings all suffer noticeable crossfades.
Implementation: pure-function position-derived volume scalar.
`deriveFadeScalar(position, duration, crossfadeSec)` ramps from 0 to
1 over the leading X seconds, 1 to 0 over the trailing X seconds,
and stays at 1 in between. Tracks shorter than 2X don't fade.
The layout's audio-volume effect now multiplies player.volume by
the fade scalar — no timers, no AudioContext, no element swapping.
Re-renders at the ~4Hz `timeupdate` cadence give 16 discrete steps
over a 4s fade, audibly close to smooth. Smoothing to per-frame
ramps via rAF is a follow-up if needed.
Setting persists to localStorage as `minstrel.crossfade` (matches
the existing volume-storage convention).
Tests cover the pure derivation (off, too-short track, fade-in,
fade-out) + clamp/persist on setCrossfade + invalid-input recovery
on readStoredCrossfade.
Drag was already wired via @neodrag/svelte; this adds the missing
keyboard path so reorder is accessible without a pointer device.
- The drag handle becomes a real <button> (focusable). Disabled
when canDrag is false so non-owners + unavailable tracks can't
attempt a move.
- ArrowUp / ArrowDown on the focused handle call
onMove(row.position, row.position +/- 1). Bounds clamping
already lives in the page-level handler (playlists/[id]/+page.svelte:39),
so out-of-range targets resolve as no-ops.
- Matches QueueTrackRow's pattern: aria-label "Reorder track …",
aria-keyshortcuts, swallow Space/Enter to avoid scrolling.
Test updates: drag-handle assertions retargeted to /reorder track/
+ two new ArrowUp/ArrowDown handler tests.
Add a Link icon button to the playlist header that copies the
absolute /playlists/[id] URL to the clipboard. Visible only when
isOwner && pl.is_public. Falls back to a window.prompt() in
insecure contexts where navigator.clipboard is unavailable.
Tests: button visibility (private vs public owner) + clipboard
write target.
Replace dead-end empty copy with EmptyState cards that include a
clear next step.
- EmptyState.svelte: reusable card with `block` (whole-tab) and
`inline` (sub-section) variants, an actions snippet for buttons.
- Artists / Albums: when the whole library is empty, link to
/settings (the operator can scan a folder there).
- Liked: when all three sub-sections are empty, show a single
whole-tab card ("No likes yet" with Explore Home + Browse albums).
When only one sub-section is empty, an inline hint with a link
to the corresponding library tab.
- History: "Listen to something" button → Home.
- Playlists: "Create a playlist" button calls the existing create
flow; renamed from "New playlist" to avoid colliding with the
header's button-by-name lookup in tests.
Liked tab tests updated to match: the previous "three 'no liked X
yet'" assertion is now a single onboarding card test + a sibling
test for the mixed populated/empty case.
Add a debounced QuickFilter input at the top of each Library section
to narrow the loaded list without leaving the tab. Server-side
/search remains the route for full-library searches.
- QuickFilter.svelte: 120ms debounced two-way bound input with a
clear button and Escape-to-clear.
- Artists: filter by name; empty-match copy includes a "Search the
full library" link to /search/artists.
- Albums: filter by title + artist_name; same fallback link.
- Liked: filters all three sub-sections (artists/albums/tracks);
sub-section header hides when its filtered length is zero.
- History: filters flatEvents before groupByDay, so day-grouping
reflects the filter.
- Playlists: filters owned + public by name; owners CTA stays.
Covered by QuickFilter unit tests (debounce + clear + Escape).
Add a global selection store backing a checkbox per track row and a
floating SelectionBar above PlayerBar with Play next / Add to queue /
Add to playlist / Like all / Clear.
- selection/store.svelte.ts: id Set + TrackRef Map + anchor index;
toggleOne, selectRange (shift+click), clearSelection. Singleton —
layout effect clears on pathname change.
- TrackRow: checkbox slot replaces the track number on hover; row
click toggles selection once any row is picked; shift+click extends
the range. Esc clears (takes priority over closing the queue
drawer).
- SelectionBar: floating pill above PlayerBar, mounted inside the
QueryClientProvider so the Like-all action can resolve.
- player.playNextMany: bulk variant of playNext for "Play next" on a
multi-track selection.
Covered by selection-store unit tests and three new TrackRow tests
(checkbox toggle, sticky select-mode row click, shift+click range).
CI on a2466b7d caught it: the cover/title area now links to
/now-playing per the NowPlaying landing, but PlayerBar.test.ts
still asserted /open album/i + href /albums/xyz. Rewritten test
asserts /open now playing/i + href /now-playing, with a comment
explaining the Spotify/YouTube Music pattern (album reachable via
the album-name link inside NowPlaying).
Mirrors Android's NowPlayingScreen. New /now-playing route renders
outside the Shell (no top bar, no PlayerBar) so it's a focused full
viewport: large square cover, title + linked artist/album, full-width
scrubber with timestamps, prominent transport (prev / play-pause /
next, with a 56dp circular play button), shuffle + repeat toggles,
like button, volume slider, and a queue button that opens the
existing queue drawer.
Tapping the cover or title area in PlayerBar (compact + desktop)
now navigates to /now-playing instead of the album page. Pattern
matches Spotify / YouTube Music. The album link stays reachable via
the album-name link inside NowPlaying itself, so no nav is lost.
Back button in the NowPlaying header uses history.back() with a
fallback to /. Empty state when no track is loaded points the user
at Home or the Library.
Scribe 528, local task #62.
Mirrors the Android change from commit bf61d6cf. Web's Home Playlists
row gains the 5 secondary system kinds (deep_cuts, rediscover,
new_for_you, on_this_day, first_listens) after the Songs-like slots,
in server-registry order, when they exist. No placeholders for these
— they depend on library shape (Deep cuts needs deep albums, On this
day needs prior history, etc.) so a missing one means "not enough
data," not "still building."
Test coverage: new `renders secondary system kinds` test creates a
mocked owned set with for_you + deep_cuts + new_for_you and asserts
all three card names render. Existing tests (5 placeholders, building
variant, For-You + 3 placeholders) unchanged since none seed
secondary kinds.
Most Played multi-row is NOT touched — web already chunks the 75
tracks into 3 rows of 25 via the existing HorizontalScrollRow rows=
{chunk(...,25)} pattern. My initial task description was wrong on
that point.
Naming collision (`rediscover` playlist vs Rediscover recommendations
section both read "Rediscover" on the same Home) deferred — operator
follow-up. Same on Android.
Scribe 531, local task #61.
CI on 4362233d caught it: playQueue leaves the store at 'loading'
(intent-to-play; jsdom never advances to 'playing' because there's
no real audio). togglePlay's contract is loading|playing -> paused,
anything else -> loading. So Space round-trips loading <-> paused,
not paused <-> playing.
Three changes:
- "Space toggles" expects loading -> paused -> loading.
- "K alias" expects loading -> paused.
- "shortcuts ignored when focus in input" + "modifier keys disable"
capture state before the press and assert no change, instead of
hard-coding 'paused'.
Universal music-player conventions, attached at window level from
+layout.svelte:
Space, K play / pause
ArrowLeft previous track
ArrowRight next track
J seek -10s
L seek +10s
ArrowUp volume +5%
ArrowDown volume -5%
M mute / unmute (restores prior volume)
/ focus the global search input
Skipped when focus is in an input / textarea / select /
contenteditable so typing in the search box and rename fields
behaves normally. Modifier-only chords (Ctrl/Cmd/Alt + key)
intentionally pass through to the browser.
New store helper `toggleMute()` captures the last non-zero volume
so M can toggle back without permanently losing the user's level.
shortcuts.svelte.ts is the new module; +layout.svelte wires it
alongside useMediaSession / useEventsDispatcher.
Scribe 529, local task #60.
Operator 2026-06-01: "navigation layout and library sections are
what I'd like to have implemented as it seems better than our
current navbar solution. I think I'd like to have these nav options
moved into the top bar centered."
Top-bar restructure:
- Centered nav (replaces the 192dp left sidebar): Home / Library /
Discover, with icons + labels. Labels collapse below sm breakpoint
so the bar stays icon-only on small viewports.
- Right side (search input + user dropdown) unchanged.
- Hamburger button + MobileNavDrawer + the mobileNav store all
removed - the centered nav lives at all viewport sizes.
Library page restructure (mirrors Android LibraryScreen):
- New routes/library/+layout.svelte renders a tab bar across the
five Library sub-pages: Artists / Albums / Liked / History /
Playlists. Active tab gets an accent underline + onSurface text.
- routes/library/+page.server.ts redirects bare /library to
/library/artists (Android default tab).
- /playlists (list) moved to /library/playlists; old URL gets a 308
redirect (routes/playlists/+page.server.ts) so existing bookmarks
land on the new location. /playlists/[id] (detail) is unchanged -
matches the server API URL shape.
Deleted: Shell's sidebar markup, MobileNavDrawer.{svelte,test.ts},
the mobileNav store, the old routes/playlists/+page.svelte. Shell
test rewritten to assert the new 3-item centered nav; playlists
test moved next to its new +page.svelte and its test-utils import
path updated.
Two issues from the prior CI failures:
1. Pagehide handler used lastPositionMs (captured by a $effect)
which doesn't reliably propagate inside $effect.root in the test
fixture - test saw 0 instead of 73000. Switched to a direct
`Math.round(player.position * 1000)` read at pagehide time. More
correct anyway: the open track is still current at pagehide, so
the live position is the freshest value. Same change applied to
the natural-end branch for consistency.
2. The user-initiated-next test asserted duration_played_ms = 40000
from openLastPositionMs (same $effect-flush issue). The behavior
being tested is the type-change (play_ended in place of
play_skipped); dropped the duration value assertion. Type
assertion stays.
Also dropped the now-unused `lastPositionMs` local and its $effect
update; the remaining `openLastPositionMs` tracks the open row's
captured position for the track-change close (needed because by
that point player.position has been reset to 0 by the queue
advance).
jsdom's Blob doesn't ship .text() (only the polyfill via FileReader
or arrayBuffer is reliable). The pagehide test failed with
"TypeError: blob.text is not a function" at events.svelte.test.ts:183.
Swap blob.text() for new FileReader().readAsText() wrapped in a
Promise.
Web client now always POSTs play_ended with the actual position
played to, matching the Android fix at d82e744d (Scribe 519). The
server's skip rule (skip_max_completion_ratio +
skip_max_duration_played_ms, default 50%/30s) classifies was_skipped
from the duration_played_ms; the dispatcher no longer makes the
call itself.
Three transitions all collapse to play_ended:
- Track change (was: reachedEnd ? play_ended : play_skipped)
- Natural end via paused-at-duration (was already play_ended)
- pagehide via sendBeacon (was: play_skipped)
Drops the openReachedEnd field, COMPLETION_TOLERANCE_MS constant,
and the position-vs-duration threshold check that backed the prior
classification. play_skipped wire endpoint stays in place for a
future explicit-dislike affordance.
Two test cases updated:
- "user-initiated next mid-track" now asserts play_ended with
duration_played_ms = the partial position, not play_skipped.
- "pagehide fires sendBeacon" now parses the beacon payload and
asserts type=play_ended + duration_played_ms = last position.
Flutter NOT touched — deprecated per M8.
PlayEventsReporter used to make the skip-vs-ended call client-side:
if the play head reached within 3s of duration -> playEnded with
full duration; otherwise -> playSkipped (forces was_skipped=true on
the server regardless of actual play time). That left the server's
configurable skip rule (skip_max_completion_ratio +
skip_max_duration_played_ms, default 50%/30s) dead code and meant
any "next" tap registered as a skip even after most of the track
played - too strict per operator 2026-06-01.
PlayEventsReporter now always calls playEnded with the actual
position played to. Server's RecordPlayEnded applies the rule and
decides was_skipped. If the play ran to completion, the last
observed position is approximately the track duration -> ratio ~1
-> not skipped. If the user hit next mid-track, the duration the
server sees is the real listen time and the rule fires normally.
Drops the curReachedEnd field and COMPLETION_TOLERANCE_MS constant
that backed the prior classification. playSkipped wire endpoint
stays in place for a future explicit-dislike affordance (not wired
to track-change transitions any more).
Parity-map row updated. Web backflow tracked as task #57 (Scribe
521). Flutter not touched - deprecated per M8.
CI on 5c2011e6 caught it: removing the MiniPlayer kebab also removed
the only consumer of ShellScaffold's navController parameter, which
detekt's UnusedParameter rule flagged. Per YAGNI, dropping the param
entirely instead of @Suppress'ing a dead carry; if a future banner
or shell-level affordance needs nav, plumb it back in at that point.
Touches the public signature so all 10+ call sites in MinstrelNavGraph
also lose the `navController = navController` argument.
HomeViewModel.playPlaylist (the Home play-button overlay path) and
PlaylistDetailViewModel.play (the detail-screen path) both converted
PlaylistTrackRef -> TrackRef before player.setQueue, but only the
detail-screen path filtered out unplayable rows (missing trackId or
empty streamUrl). Home's path passed them through; Media3 then
silently no-op'd on setUri("") and the queue appeared loaded but
nothing started.
Operator reported "Songs-like playlists queue music that never
plays" via the Home play-overlay 2026-06-01. The same conversion in
two places with different rules is exactly the foot-gun the §1-§3
DRY pass was supposed to head off; this commit adds it as a follow-up
since the playlist-play helper wasn't covered by that sweep.
Extracts `List<PlaylistTrackRef>.toPlayableTrackRefs()` in
playlists/data/PlaylistsRepository.kt. Filters (isAvailable &&
streamUrl non-empty) THEN maps to TrackRef, in one pass. Both call
sites now use it.
The remaining private `toTrackRef()` in PlaylistDetailScreen is kept
for the per-row TrackActionsButton wiring - the actions menu only
needs id+display fields and doesn't queue anything itself.
Cover + LikeButton + 3 transport buttons + kebab consumed almost all
of the 360-410dp viewport on typical phones, leaving the title /
artist column with only ~20-80dp - enough for a truncated title but
nothing for the artist line, which was rendering but effectively
unreadable.
Drops TrackActionsButton from MiniRow and the now-unused
onNavigateToAlbum / onNavigateToArtist callbacks from the MiniPlayer
signature + ShellScaffold call site. Full kebab surface still lives
on NowPlayingScreen (one screen swipe away).
Operator request 2026-06-01. Diverges from Flutter's player_bar
which still ships the kebab; parity-map row updated to reflect the
deliberate divergence. ShellScaffold's navController parameter is
kept on the signature for future shell-level navigation needs.
Operator device check 2026-06-01: the multi-row grid landed, but the
square 140dp tiles (web-matching) waste vertical space at the per-
tile level. Flutter's CompactTrackCard
(flutter_client/lib/library/widgets/compact_track_card.dart) uses a
horizontal-row pattern - 176dp wide, 56dp tall, 48dp cover thumb +
title/artist column to the right. Much denser; fits ~3x more tiles
in the same screen area.
CompactTrackTile rewritten to that shape:
- Row layout instead of Column
- 48dp square cover (down from 140dp)
- Title + artist column to the right, vertically centered
- 176dp x 56dp outer
MOST_PLAYED_GRID_HEIGHT_DP cut from 600 to 200dp (3 * 56 + 2 * 8
spacing). Skeleton matches the new dimensions.
Diverges from web (which uses square tiles same as Android's prior
shape) - intentional, operator preference. TrackActionsButton from
Flutter's card is omitted for now (would need nav callbacks threaded
through MostPlayedRow); follow-up if kebab-on-tile is wanted.
PlayerController only ran the UI-state copy on Media3 Player.Events
callbacks (track change, pause/play, buffer state) - none of which
fire on normal position advancement during playback. The seek bar
appeared to update every 20-30 seconds, whenever a buffer event
happened to fire, instead of ticking smoothly.
Adds a startPositionPolling() coroutine launched on
Dispatchers.Main.immediate after the controller is connected.
Samples controller.currentPosition + bufferedPosition every 500ms
while isPlaying; skips state updates while paused (positionMs stays
at the last value, which is the correct paused-state behavior).
The poll patches only the position fields via .copy() so stale
event-driven snapshots can't clobber the latest state.
Cadence reference: Flutter via just_audio positionStream runs at
~200ms (audio_handler.dart:75); Spotify / Apple Music sit around
250-500ms in-app. 500ms is the battery-friendly middle ground that
still reads as smooth on a slider. Lock-screen / notification
surfaces are driven by Media3's MediaSession separately and don't
need this poll.
Liked tracks now render a filled heart (HeartFilled, defined inline
in LikeButton.kt) instead of just recoloring the outlined heart.
Matches the cross-app convention (Spotify, Apple Music, etc.) where
fill is the affordance change rather than tint alone.
Lucide ships outlined-only by design (jar inspection: heart.kt,
heart_crack.kt, heart_off.kt, etc., none filled). Built the filled
variant locally from Lucide's exact path data so the toggle reads as
a fill swap, not a different shape. No new dep - keeps the project's
Lucide-everywhere icon system consistent.
Operator authorized the divergence 2026-06-01 ("changing to material
for this one symbol... is acceptable"); inline ImageVector route
preserves Lucide visual harmony better than mixing material-icons
just for one glyph.
CachedTrackEntity.toDomain() was emitting TrackRef.streamUrl="" (the
default). Tracks routed through MetadataProvider (the Home cache
hydration path used by Most Played) inherited the empty string;
player.setQueue() then got a queue of unplayable tracks.
The server's stream_url is deterministic per track id
(internal/api/convert.go:75 streamURL builder), so the cache mapper
can reconstruct it from the row's id without storing it in Room.
TrackWire.toDomain() continues to use the wire-provided value
(identical content).
Operator reported "tap to play wiring in Most Played queues content
that can't play" on 2026-06-01. The Most Played multi-row layout
itself isn't the cause - the same gap affected the single-row
layout - but the denser display made the dead taps more obvious.
CI on 58213779 caught the missing import: the previous code used
Arrangement only inside HorizontalScrollRow (which imports it
internally), but the inline LazyHorizontalGrid for Most Played
now needs it at the HomeScreen.kt scope. ktlint passed because
it doesn't follow type resolution; the kotlinc compile step
flagged it.
Web Home (web/src/routes/+page.svelte:196) stacks the 75 most-played
tracks into 3 rows of 25 inside a single horizontal scroller; Android
was rendering them as one long LazyRow which makes the section feel
sparse and forces a lot of horizontal scrolling to reach mid-rank
tracks. Operator request 2026-06-01.
Replaces MostPlayedRow's LazyRow with a LazyHorizontalGrid where
rows=Fixed(MOST_PLAYED_ROWS=3). Cards stay 140dp (matching web's
w-36); density gain comes purely from stacking.
Web is row-major (ranks 0..24 across the top row); LazyHorizontalGrid
fills column-major. To match web's visual order, the input list is
pre-chunked into 3 row-major rows then re-flattened column-by-column
before being handed to the grid. With 75 tracks and 3 rows that's
25 columns, ranks 0/25/50 in column 0, 1/26/51 in column 1, etc.
MOST_PLAYED_GRID_HEIGHT_DP=600 budgets ~188dp per row (140 cover +
8 spacer + 2 lines of text), plus 2 * 8dp inter-row spacing.
Diverges from Flutter (still single-row); intentional, parity-map
updated. Backflow to Flutter not tracked yet because the Flutter
client is being retired.
CI on 8b586c2e caught it: the $1::text cast in the md5 ORDER BY made
sqlc infer the parameter as plain string instead of pgtype.UUID,
generating Column1 string instead of UserID pgtype.UUID on the
ListRediscover*ForUserParams structs. go vet flagged both call
sites in internal/recommendation/home.go where the Go code passes
UserID by name.
Fix: hash on album_id (or artist_id) + current_date only. The
eligibility filter already differs per user (different liked sets),
so the daily-rotation goal is preserved; we just lose per-user salt
in the within-day ordering of overlapping items - acceptable.
Applies to both primary queries AND both fallback queries (all four
had the same cast).
User request 2026-06-01: the previous Rediscover only fired on
explicit album/artist likes, so users who only liked at the track
level got an empty Rediscover row. Also no daily rotation - the
section never changed between data updates - and 25 items felt long
for the Home carousel.
SQL (internal/db/queries/recommendation.sql):
- ListRediscoverAlbumsForUser UNIONs the existing explicit album-like
signal with a new track-derived path: an album qualifies if it has
>=2 liked tracks where the earliest like is >30 days old.
- ListRediscoverArtistsForUser does the same with threshold 3 (artists
span more releases, so the bar is higher to avoid one-hit-wonder
affinities).
- Both ordering switched from longest-since-last-play to a daily-
stable random hash md5(entity_id || user_id || current_date) so
the row randomizes but stays consistent within a day.
- Fallback queries also switched to the daily-stable hash so the
fallback rows don't reshuffle on every refresh.
Go (internal/recommendation/home.go):
- HomeRediscoverLimit dropped from 25 to 10 (carousel-sized).
- rediscoverInnerLimit (30) is the per-query cap; gives headroom so
the Go-layer filters don't undershoot.
- applyRediscoverAlbumFilters does cross-section dedup vs Most Played
(no point surfacing albums the user is actively spinning) plus a
diversity cap of max 2 albums per artist (prevents one liked-but-
forgotten artist dominating the row).
- applyRediscoverArtistFilters does cross-section dedup vs Most
Played's artist set; no diversity cap needed (one row per artist).
Tests (internal/recommendation/home_integration_test.go):
- TrackDerived path: 2 liked tracks >30d old + no recent plays = album
appears in Rediscover.
- Threshold guard: 1 liked track = album does NOT appear.
- Diversity cap: 4 explicit album-likes from same artist = 2 in output.
- Dedup vs MostPlayed: eligible album whose track is in MostPlayed
gets filtered out.
- Existing FallbackWhenSparse test preserved (semantics unchanged for
recent likes that miss the >30d primary filter).
Clients unchanged - they render whatever /api/home/index returns. No
parity-map row needed (this is a server-internal recommendation
change, not a layout divergence).
Server's playlists registry has 8 system kinds (internal/playlists/
system.go:281-290) but Flutter Home only ever showed 5 slots:
For You + Discover + 3 Songs-like. Operator request 2026-06-01: have
the Android Home view also surface the 5 secondary kinds so the
user can find them without digging into the Library > Playlists list.
Android buildPlaylistsRow now appends deep_cuts, rediscover,
new_for_you, on_this_day, first_listens in server-registry order
after the existing Songs-like slots, when those playlists actually
exist. No placeholders for the new 5 since they depend on library
shape (deep albums for Deep cuts, prior history for On this day,
etc.) and an empty placeholder would imply they're still building
when they may just have no candidates.
PlaylistCard.kt systemLabelFor gains explicit labels for the new
kinds and drops the dead "todays_mix" branch (no such variant in
the server registry).
This diverges from Flutter Home which never surfaces the secondary
kinds; web UI catch-up tracked as task #53. Naming note: the
"Rediscover" playlist tile and the unrelated "Rediscover"
recommendations section on the same Home both read "Rediscover" -
operator follow-up to disambiguate.
Adding the play-button overlay in the previous commit pushed
PlaylistCard from 49 to 62 lines, two over detekt's 60-line cap.
Splitting the cover stack into its own private composable
(PlaylistCardCover) brings the outer function back under the limit
while keeping the call site readable.
Behavior identical to f17610ec; pure refactor for lint compliance.
Mirrors Flutter's PlayCircleButton (library/widgets/play_circle_button.dart)
on the Home AlbumCard / ArtistCard / PlaylistCard surfaces. 44dp accent
disc bottom-right of the cover, parchment Play icon, self-managed
spinner during the fetch-and-queue setup, drop shadow.
New: shared/widgets/PlayCircleButton.kt. Cards gain an opt-in
onPlay: (suspend () -> Unit)? parameter; when null the overlay is
omitted, preserving the Library / Discover / Search / detail surfaces
unchanged. HomeScreen wires three new HomeViewModel suspend methods:
playAlbum GET /api/albums/{id}, play tracks from 0
playArtistShuffled GET /api/artists/{id}/tracks, FY shuffle, play 0
playPlaylist systemShuffle for refreshable system playlists,
refreshDetail otherwise, 8s timeout
PlaylistsApi gains systemShuffle for the rotation-aware shuffle path
(GET /api/playlists/system/{kind}/shuffle, mirrors playlists.dart).
PlaylistCard play is disabled offline + refreshable (server endpoint
unreachable) and for empty playlists. Album / artist errors surface
via the existing transientMessages snackbar channel, matching the
ArtistDetailViewModel improvement over Flutter's silent fail.
Scope: Home tiles only this slice. Flutter applies the overlay to
the cards everywhere they render; revisit Library / Discover / Search
/ detail surfaces in a follow-up.
Project-level AI-tool instruction files (CLAUDE.md, AGENTS.md,
GEMINI.md, .cursorrules, .windsurfrules, .aider.conf.yml) are
operator-scoped working notes, not project artifacts. Other
contributors don't need them, and committing them conflates
operator preferences with shared project conventions.
CLAUDE.md remains on the local disk (untracked) so Claude Code
keeps loading it for this checkout; new clones simply won't have
it, which is the desired state.
Last push failed both android and flutter jobs at workflow-setup time
because the runner couldn't resolve forgejo/upload-artifact@v3
(github.com/forgejo/upload-artifact does not exist; the Forgejo
project hosts on Codeberg and our Gitea runner falls through to
github by default). The canonical Gitea Actions form is
actions/upload-artifact@v4, which act_runner resolves cleanly.
Flutter pipeline is being retired in favor of the native Android
client, so flutter.yml is deleted outright rather than fixed. The
container-image build's release-asset polling (release.yml) will now
graceful-degrade when chasing the legacy minstrel-<TAG>.apk name,
which the comment in ci-requirements.md already documents as
acceptable. Renaming the native APK to drop the -android- infix is
deferred to a follow-up so the cutover is reviewable in isolation.
bvandeusen