Adds season/episode picker and sources list to the shared Detail
composable, switches FluxaAppState's selectedDetail to carry a
DetailRequestUiModel directly (instead of a full CatalogItemUiModel)
so the standalone Screen.Detail route can seed it without going
through a catalog item first, and threads a detailRequest +
onDetailNavigationEvent pair through FluxaAppHost.
AppRoutesHost now builds a DetailRequestUiModel from Screen.Detail and
hands the shared Detail screen's navigation decisions (play a
specific stream vs. go pick sources, based on cs3: id prefixes) back
to Screen.Player / Screen.Sources, reconstructing the full Stream
object via AndroidDetailDataSource.resolveStream() so headers/
subtitles aren't lost. TV still uses the legacy DetailRoute for now.
Diagnosed against NuvioMobile's open-source discover implementation
(same addons, more catalogs visible there). Two real gaps:
1. buildDiscoverCatalogOptions() excluded any catalog requiring an
extra parameter beyond genre. Nuvio's equivalent (supportsDiscover())
explicitly whitelists "skip" (pagination) as never blocking a
catalog even when required — Fluxa's filter didn't, silently
dropping every catalog that declares skip as required, which is a
common Stremio addon SDK convention. Added "skip" to the allowed set.
2. The Type filter (movie/series) was hardcoded in the shared Discover
UI instead of derived from what addons actually expose. Nuvio builds
its type list from distinct catalog types across installed addons,
which is why "anime" shows up there. Added a new
buildDiscoverContentTypes() helper (same extras filter, minus the
"all"/type restriction), threaded contentTypes through the
readDiscoverCatalogFilters effect, Rust's DiscoverState (new field,
needed since the effect completion flows back through the engine's
own state, not directly to Kotlin), HomeViewModel, and the shared
DiscoverUiState/DiscoverScreen, with a movie/series fallback so the
dropdown isn't empty before the first fetch completes.
AppRoutesHost was mounting a fresh FluxaAppHost per Screen navigation,
so every tab switch among Home/Search/Discover/Calendar recreated
CatalogHomeStore and re-triggered CatalogAction.Refresh, which force-
reloads the whole Home catalog over the network regardless of which
tab was actually opened. Hoist the shared shell out of AnimatedContent
so it stays mounted across those four screens and only remounts when
leaving them entirely (Detail/Sources/Player/Settings/etc., same as
before).
Discover regressions from the last pass, now fixed to match native:
- Filter chips replaced with dropdown menus (DropdownMenu/MenuItem)
matching native's Type/Catalog/Genre picker pattern.
- Grid restored to a fixed 3-column layout instead of adaptive sizing.
- Added the inline search box native's Explore screen has, reusing
SearchResultRows/SearchResults (now internal, not private) so typing
in Discover shows grouped search results instead of the filtered
browse grid, same as native.
Wires Screen.Search/Explore/Calendar to FluxaAppHost on mobile, same
pattern as Home. TV keeps using the existing native screens untouched.
To preserve native feature parity in the shared UI:
- Search: per-addon grouped result rows (was a flat grid), a clear
history action, and a query-specific "no results for X" message with
a recent-history fallback, matching native's MobileSearchScreen.
- Discover: filter chips are now a scrollable LazyRow instead of a
plain Row, avoiding the same overflow-wrap bug fixed on Home's row
headers when there are many genre options.
- Calendar: added month navigation (prev/next + localized month
title), day-grouped list headers via the existing cross-platform
localizedShortMonthDay/localizedMonthTitle helpers, replacing the
flat undated list.
Deliberately not replicated (documented simplification, not a bug):
native's Search and Explore screens each embed a blended search+browse
UX (explore filters shown pre-typing in Search; inline search box in
Explore). Shared keeps Search and Discover as clean separate screens
since both capabilities are still fully reachable via their own tabs.
Also skipped: Explore's list/grid view toggle, swipe-to-change-month
and the day-grid calendar view (list view covers the same data).
java.text.DateFormatSymbols/SimpleDateFormat are JVM-only, so the
calendar screens' weekday names, month titles, and short dates
couldn't move to KMP as-is. Adds an expect/actual trio in :core
(localizedShortWeekdayNames, localizedMonthTitle, localizedShortMonthDay)
— Android actual keeps the existing DateFormatSymbols/SimpleDateFormat
behavior, iOS actual uses NSDateFormatter/NSCalendar/NSLocale.
MobileCalendarDateUtils.kt's shortWeekdays/monthTitle/
formatCalendarListDate now delegate to these instead of doing the
java.text work directly — same signatures, so all 6 call sites
(TvCalendarScreen + 5 mobile calendar files) are untouched. This is
groundwork only: buildMonthCells/shiftMonth still use java.util.Calendar
and the calendar screens themselves aren't ported — that needs the
screens' JavaCalendar-typed state converted to kotlinx-datetime, a
separate, larger change.
Verified compiling for android, iosArm64, iosSimulatorArm64, and both
app product flavors still build.
Phase 0 of the iOS migration: core now targets android, iosArm64, and
iosSimulatorArm64 via a new fluxa.kmp.library convention plugin.
AppStrings and the TTL cache lose their JVM-only dependencies
(org.json, java.time, java.util.concurrent) in favor of
kotlinx-serialization, kotlinx-datetime, and small expect/actual
platform shims, with Locale-returning APIs kept as Android-only
extension functions since only Android UI consumes them today.
- Use Chapters for Skipping: gates the existing chapter-derived skip
segment fallback (already computed by deriveSkipSegmentsFromChapters)
behind a new toggle instead of applying it unconditionally. Also
brought TV's Skip Segments section up to parity with mobile (IntroDB +
API key, AniSkip, chapter toggle, auto-skip), since TV was missing all
but auto-skip.
- Japanese audio for anime: overrides the resolved preferred audio
language to "ja" for anime-genre content when enabled, at the single
entry point both audio and subtitle track resolution already share.
- Autoplay countdown: the next-episode card now counts down and fires
automatically, instead of only advancing on click.
- Auto-retry next source: extends the existing Cloudstream-only
auto-fallback-on-error path to also apply generally when enabled.
Both onConnectStremio and MobileNuvioLoginView already existed but had no
UI trigger anywhere in Settings — onConnectStremio was a dead parameter
never invoked, and the Nuvio login view was only reachable from the
initial onboarding screen. Added a Stremio connection tile (TV + mobile)
and a Nuvio import row (mobile only, since TV's login screen has no
Nuvio path) that reuse the existing Screen.Login(startOnNuvio) navigation
already used by onboarding.
Mirrors the existing Trakt/MyAnimeList/Simkl connection pattern: a
fluxa://oauth/anilist deep link, authorize-URL launch, token exchange
through a new ExternalOAuthClient.exchangeAnilistCode call, and profile
fields (anilistAccessToken/RefreshToken/TokenExpiresAt) merged via the
existing authExchangeRequested headless effect. Adds a matching
AuthProvider::Anilist arm in fluxa_core's token_merge_plan_json so the
shared Rust merge path recognizes the new provider too.
Connect/disconnect tiles added to both TV and mobile Settings, folded
into the existing multi-provider "Disconnect" action.
Note: app/build.gradle.kts also carries a small pre-existing uncommitted
fix (rustCrateDir path) unrelated to this change, included here only
because it was already sitting in the working tree on the same file.
Adds the missing torrentSpeedPreset settings row (the field already
controlled preload size) and a new Wi-Fi-only toggle. The Android-side
gating for Wi-Fi-only lives in TorrentStreamManager.startStream (see the
next commit); this piece wires the active profile's torrentWifiOnly flag
through the startTorrentStream headless effect.
torrentMaxConnections and torrentCachePreset are intentionally left
unexposed: the underlying rqbit torrent engine has no per-torrent
connection-count cap or disk-cache-size knob to back them, so a toggle
for either would be a no-op.
These profile fields (notificationsEnabled, automaticUpdates, subtitleShadow,
trailerOnHomeHeroEnabled/DelaySeconds) already drove real behavior but had no
visible Settings UI on TV or mobile. Also gates the background update-check
loop in MainActivity behind automaticUpdates, and wires a TV "Check for
updates" action mirroring the existing mobile flow.
Give back, search, cast, picture-in-picture, external-player, and
calendar month-navigation buttons localized accessibility labels so
TalkBack announces their action. Adds the supporting common.* i18n keys.
- Return UpdateReport(updatedPlugins, failedPlugins) from auto-updater so failures
are tracked separately from successes
- Show a distinct notification when one or more plugins fail to update, using
separate notification IDs so the success and failure toasts don't stomp each other
- Add installKey() to Plugin for stable deduplication by (name, repositoryUrl) pair;
use it in isPluginInstalled to avoid duplicate installs on re-run
- Add i18n keys notification.plugins_update_failed_title/body in en-US and tr-TR
mpv:
- Set ao=audiotrack explicitly so the AudioTrack output is guaranteed
(avoids openSLES which lacks encoded passthrough)
- Add audio-channels=7.1,5.1,stereo so mpv negotiates multichannel
instead of defaulting to auto-safe stereo downmix
- Build audio-spdif at init time by querying AudioManager output device
encodings; only advertises AC3/E-AC3/DTS/DTS-HD/TrueHD formats the
device actually reports
ExoPlayer:
- Guard setVideoEffects(emptyList()) behind a videoEffectsActive flag:
only call it when IptPqc2ToneMapEffect was actually installed, never
on a clean path — the empty-list call triggers the GL VideoFrameProcessor
pipeline in media3 1.10.1 which tone-maps HDR even for an empty effect list
- detectPreferredVideoMimeType now accepts Context and checks display +
decoder DV capability first; on fully DV-capable devices it returns
VIDEO_DOLBY_VISION so DV variant tracks are preferred rather than
deprioritized behind H.265/H.264
