12 Commits

Author	SHA1	Message	Date
cucadmuh	15cecb5d7d	feat(server): custom HTTP headers for reverse-proxy gates (#1095) (#1156)	2026-06-22 16:25:28 +03:00
Psychotoxical	a5313d5cb1	fix(cover): embed fanart.tv project key so from-source builds work (#1139) ... The project key was injected only at our build time (CI secret / option_env!), so AUR, Nix and any from-source build had no key baked in and the External Artwork toggle was inert there. Commit it as a source literal (like the Last.fm key, and as fanart.tv terms expect: the app ships a project key, users add their own on top via BYOK). Drops the now-redundant CI wiring.	2026-06-20 23:03:12 +02:00
Psychotoxical	b950d4704b	feat(cover): artist artwork from fanart.tv (off by default) (#1137) ... * feat(cover): add artist_artwork_lookup table + accessors Image-scraper P0 (design-review §12): additive library-SQLite migration 013 plus get/upsert/clear-per-server accessors for the external artist-artwork lookup (fanart.tv). Render never reads it; the on-demand cover ensure path and the mbid_ambiguous 24h negative cache use it. server_id = serverIndexKey. * feat(cover): add fanart.tv + getArtistInfo2 provider layer Image-scraper P0 (§7/§19/§23): new cover_cache/external.rs — Rust-side getArtistInfo2 tag-MBID resolution plus fanart.tv v3/music URL + first artistbackground fetch (BYOK client_key sent in addition to the project api_key per fanart.tv ToS, §22). Extract a shared build_subsonic_url helper in fetch.rs (cover URL behaviour unchanged). Add a dedicated low-concurrency fanart_http_sem so external HTTP never starves Navidrome (§26). URL builders unit-tested; wired into ensure_inner next. * feat(cover): wire fanart.tv external branch into ensure_inner Image-scraper P0 (§16): on-demand artist `fanart` ensures try fanart.tv before the Navidrome fallback. MBID resolved Rust-side via getArtistInfo2 (§23, tag MBID); on a miss it falls through WITHOUT a .fetch-failed marker so Navidrome stays the display fallback (§28). External tiers are written as {tier}-fanart.webp in the same entity dir (same cacheKind, §16) — 2000 + 512 (matryoshka §17); peek prefers them for the fanart surface. Dedicated low-concurrency fanart lane (§26); .miss-fanart ~30min negative marker. Additive IPC args (externalArtworkEnabled, surfaceKind), off by default and gated by PSYSONIC_FANART_KEY — inert until a render surface opts in (P1). Quality gate (§11), name->MusicBrainz (§19), and lookup-table writes are P1. * feat(cover): fanart-first peek for the fanart surface Image-scraper P0: for an artist `fanart` ensure, the early peek serves only the external {tier}-fanart.webp tiers; if none exist yet it returns None so ensure runs the external branch (fetch fanart) instead of short-circuiting on a cached Navidrome tier. Realises "fanart prioritised" (§18) for the opt-in surface; Navidrome stays the fallback inside the branch's miss path. * chore(cover): dev-only artist-fanart spike helper DEV-only window.psyFanartSpike(name) — resolves an artist by name and fires the real cover_cache_ensure with externalArtworkEnabled+surfaceKind=fanart to verify the P0 pipeline against a live server (with PSYSONIC_FANART_KEY set). Not wired in production. * feat(cover): §11 quality gate for the fanart surface Before an external fanart fetch, check whether a Navidrome tier already on disk is an HQ ~16:9 image (width >= 1280, aspect 1.6-2.0) and skip the fetch if so — square artist portraits never satisfy it, so the common case still fetches. Reads tier dimensions only (no full decode). Rust consts per the design review. Unit-tested predicate. * feat(cover): persist fanart resolution in artist_artwork_lookup (§12) Wire the lookup table as both the MBID resolution cache and the negative cache: a cached MBID skips the getArtistInfo2 round-trip; no_mbid/mbid_ambiguous back off 24h and miss 30min from updated_at before re-querying. Writes hit/miss/no_mbid with mbid/mbid_source/provider; transient network errors are not cached. All store reads/writes run off the async executor via spawn_blocking and no-op before login. Store reached via app.try_state::<LibraryRuntime>(). * feat(cover): compile-time fanart key fallback + album/name IPC args A runtime PSYSONIC_FANART_KEY still wins (dev), else the key baked in at build time via option_env! (release). Add additive artistName/albumTitle ensure args as context for the §19 name->MusicBrainz fallback (inert until the render passes them). Library backfill passes None. * feat(settings): add External Artwork Scraper toggle under Integrations Master toggle (themeStore, off by default per §20) in a new Integrations subsection, alongside the other opt-in third-party categories. Contacts fanart.tv only when enabled. i18n across all 9 locales. * feat(cover): wire fanart background into the fullscreen player (§28) New useArtistFanart hook resolves a fanart.tv 16:9 background via a dedicated cover_cache_ensure (surfaceKind=fanart) — it bypasses the shared peek/disk-src cache (the {tier}-fanart.webp surface is keyed differently) and reuses coverDiskUrl for the asset URL. Fullscreen background priority is now fanart -> Navidrome artist image (cover pipeline) -> album cover; the live useFsArtistPortrait probe is deleted (§28). Additive ensure opts (surfaceKind/artistName/albumTitle); externalArtworkEnabled is derived in ensureArgsFromRef from the master toggle and restricted to the artist fanart surface, so plain cover ensures are unaffected. * feat(cover): generalize external surface to fanart + banner (§13) surfaceKind='banner' fetches the fanart.tv musicbanner array -> {tier}-banner.webp in the same entity dir; fanart stays the 16:9 artistbackground. The ensure branch, peek, lookup rows (per-surface surface_kind), miss marker (.miss-<surface>) and tier suffix are all surface-parameterised. The §11 quality gate stays fanart-only (the banner strip has its own aspect). Unit test for the surface->fanart JSON key map. * feat(artist): fanart banner on the artist-detail header (§13, Option B) The artist-detail header gets an album-detail-style background layer: fanart.tv banner (musicbanner) -> the 16:9 fanart background cropped to the strip -> empty. Both via a shared useArtistExternalImage hook (useArtistBanner / useArtistFanart); each fetches on demand and shares the Rust cache, so the header and the fullscreen player warm each other's images. The header is its own stacking context (isolation) so a z-index:-1 banner clips behind the avatar + meta with no content wrapper; the album-style framing (padding/radius/clip) is applied only when a banner is shown, so the off-by-default case stays pixel-identical. * refactor(artist): reuse album-detail header structure for the fanart banner The artist-detail header now uses the same album-detail-* container classes as AlbumHeader (header/bg/overlay/content/hero) with the fanart banner as the background; the Back button moves inside the header. A surgical `artist-detail-bleed` cancels the artist page's .content-body padding so the banner is full-bleed to the container edges, matching the album header exactly instead of the earlier inset card. Reverts the experimental artist-specific bg CSS. * feat(cover): drop artist_artwork_lookup rows on clear-cover-cache (§12/B.4) cover_cache_clear_server already removed the server's whole cover dir (so the {tier}-fanart.webp / -banner.webp tiers + .miss-* markers go with it); also clear the artist_artwork_lookup rows for that server (off-thread) so no stale resolution state lingers. Automatic toggle-off purge deferred — turning the toggle off already hides external artwork (render is gated), and explicit cache-clear now cleans external state too. * feat(cover): name->MusicBrainz album-confirmed MBID resolution (§19) When getArtistInfo2 has no tag MBID and the ensure carries the artist name + an album in context (fullscreen), one MusicBrainz release-search query resolves the artist MBID: the primary artist across score>=90 releases wins, conflicting ids -> mbid_ambiguous (24h backoff), none -> no_mbid. Sends the required User-Agent; a single-permit musicbrainz_sem + >=1s spacing holds us under MB's rate limit. mbid_source=musicbrainz persisted. Banner surface (no album context) correctly skips this. Pure classify/escape helpers unit-tested. * fix(cover): enable banner surface in ensureArgsFromRef externalEnsureFields only set externalArtworkEnabled for surfaceKind 'fanart', so the 'banner' surface never fired — the artist-detail header always fell back to the fanart image instead of the fanart.tv musicbanner. Both external artist surfaces (fanart/banner) now enable the external branch. * feat(settings): optional BYOK personal fanart key field Add an optional personal fanart.tv API key field to the External Artwork Scraper block (shown when the toggle is on): a masked input, a saved/in-use status line, and the simple note that it is sent in addition to the app key. Persisted in themeStore and plumbed through cover_cache_ensure (externalArtworkByok); Rust prefers the settings key, falling back to the PSYSONIC_FANART_CLIENT_KEY dev env. i18n x9. * fix(cover): resolve artist-page fanart image collision on navigation The artist-detail header keyed its fanart/banner hooks on the route `id`, which flips immediately on navigation while `artist`/`albums` refetch a beat later. The mismatched ensure wrote the previous artist's image under the new artist's id (e.g. Sepultura's image under Lordi's id). - key on the loaded `artist.id`, not the route `id`, so id/name/album always describe the same artist - pick the §19 album context from an album that actually belongs to this artist (`albums.find(a => a.artistId === artist.id)`), so a stale album can't run a mismatched name→MusicBrainz query or cache a wrong `no_mbid` - reset `src` on every input change in `useArtistExternalImage` so a previous artist's image never lingers while the new one resolves * fix(cover): strip trailing album qualifier before MusicBrainz lookup Library titles like "Show No Mercy (2004 Remastered)" or "Album [Deluxe Edition]" failed the §19 MusicBrainz release query, blocking name-confirmed MBID resolution. `normalize_album_for_mb` strips a single trailing parenthetical/bracketed qualifier; leading qualifiers (e.g. "(What's the Story) Morning Glory?") are left intact. Unit-tested. * fix(cover): don't cache no_mbid when album context is unavailable The banner ensure could fire before the artist's albums loaded, with no album in context. The old code cached `no_mbid` there and the 24h backoff then blocked the later ensure that arrived *with* album context. Could-not-attempt is not tried-and-failed: the no-album branch now returns without persisting. * fix(cover): don't emit tier-ready for external fanart/banner surfaces `try_external_fanart` emitted `cover:tier-ready` with the `{tier}-{surface}.webp` path. That event is keyed by the canonical cover key (cacheKind/cacheEntityId/ tier, no surface), so the frontend `useCoverArtBridge` listener seeded the Navidrome artist cover's disk-src cache with the external image — leaking fanart/banner into the plain artist cover (avatar, fullscreen "navidrome-artist" fallback) even with the scraper toggled off. Remove the emit: the fanart/banner hooks read the path from the `cover_cache_ensure` return value, so no event is needed. (No disk-level overwrite — the suffixed files are never matched by `tier_exists`; this was frontend disk-src-cache cross-contamination.) * fix(cover): wait for the final external background before showing it, with fade-in The fullscreen player and artist-detail header flashed several backgrounds in sequence while the fanart resolved (upscaled album cover → Navidrome artist image → fanart), and the artist header could show the fanart first and then swap to the banner. - the album cover is no longer a background source — it only feeds the foreground thumbnail - the external-artwork hooks return `{ src, pending }` so callers can tell "still resolving" (hold back) from "resolved, no image" (fall back now) - fullscreen background: scraper on → fanart, empty while it resolves, Navidrome artist image only on a confirmed miss; scraper off → Navidrome artist image - artist header: the banner is preferred — nothing shows while it resolves (no fanart flash), fanart is the fallback only once the banner misses - both backgrounds preload the chosen image and fade it in (`onLoad` plus a `ref` `complete` check so an already-cached image, whose load event can fire before React attaches the handler, still appears). The header fade is a scoped inline opacity so the shared `album-detail-bg` class is untouched. * ci(release): pass PSYSONIC_FANART_KEY into the macOS + Linux builds * refactor(cover): extract external-artwork ensure into its own module Pure code move: the on-demand fanart/banner fetch, the quality gate, the surface-aware peek and the lookup-table cache move from cover_cache/mod.rs into cover_cache/external_ensure.rs. Behaviour unchanged; mod.rs 1877 -> 1488. * chore(cover): remove dev-only fanart spike helper The real render wiring now exercises the external ensure branch, so the dev-only window.psyFanartSpike helper is redundant. * feat(cover): purge external artwork on opt-out (B3) New cover_cache_purge_external command: when the External Artwork toggle is turned off, drop every fetched {tier}-{provider}.webp, .miss-{provider} marker and artist_artwork_lookup row across all configured servers, leaving the canonical Navidrome covers intact. Opting out now removes the third-party-sourced data instead of just hiding it (design-review §9/§12/B.4). * docs: changelog, credits and what's new for artist fanart (PR #1137)	2026-06-20 21:04:21 +02:00
cucadmuh	47832632fd	fix(cover): follow connect-URL flips in library cover backfill (#952) ... * fix(cover): follow connect-URL flips in library cover backfill The native cover backfill was configured once with a snapshot of the runtime connect URL. When a laptop moved off the LAN, the smart endpoint switch flipped the sticky connect URL to the public address (playback/UI covers rebuild it per request and followed it), but backfill kept fetching from the now-unreachable local address — flooding the log with "error sending request" failures. Make the connect cache observable (notify on effective flips) and have the backfill hook reconfigure when the resolved URL changes, forcing a pass so the .fetch-failed backoff from the stale address is cleared and those covers retry on the reachable endpoint. * docs(changelog): note cover backfill endpoint-switch fix (PR #952) * fix(cover): abort stale backfill pass + rerun on connect-URL flip Frontend reconfigure alone didn't fully cover the boot case: at startup the first backfill pass starts on the primary (LAN) URL before the reachability probe resolves, so when the probe flips to public the forced rerun was dropped by the pass_running guard and the slow LAN pass (every cover timing out) ran to completion with nothing re-running on the reachable address. set_session now bumps a session generation; the running pass checks it on every focus gate and abandons promptly when the URL flips (same server_index_key, new rest_base_url). try_schedule_full_pass records a rerun when a pass is in flight and drains it once the abandoned pass returns, so a fresh forced pass runs on the new address. * fix(cover-backfill): resolve connect URL per fetch instead of baking it into the queue The backfill worklist no longer carries a URL. Each cover fetch reads the current reachable address live from a single worker cell, so a LAN↔public flip is honoured even by the pass already in flight — its remaining covers download against the new endpoint without aborting/rebuilding the worklist. - Drop rest_base_url from CoverBackfillSession; add live base_url cell read in ensure_one. Remove the session-generation abort machinery (no longer needed). - New lightweight library_cover_backfill_set_base_url command pushes the URL on every connect-cache flip; a real change clears the stale .fetch-failed backoff and runs a forced pass so covers that timed out on the old address retry. - Split useLibraryCoverBackfill into a configure effect (server/creds/strategy) and a flip effect that only pushes the URL. - Keep a single rerun_pending flag for the boot case (flip mid-pass), since the finished pass re-arms the idle gate.	2026-06-02 16:01:01 +03:00
cucadmuh	2224ddbe78	feat(perf): add on-demand (ui) throughput to cover pipeline cpm (#947) ... * feat(perf): add on-demand (ui) throughput to cover pipeline cpm Cover cpm previously measured only the native backfill (lib) via the cover:library-progress done delta. Add a parallel UI series: every completed on-demand Rust cover ensure (grid / now-playing) records a timestamp, surfaced as a covers-per-minute rate. Both are exposed in the live cover diag, shown as separate Backfill (lib) / On-demand (ui) cards in the Monitor tab (each pinnable to the overlay) and as lib/ui rows in the Cover pipeline overlay block. * docs(changelog): note cover on-demand (ui) throughput (PR #947) Add CHANGELOG entry and credits line for the UI cover cpm metric. * fix(perf): source on-demand (ui) cpm from backend produced-cover count The JS ensure-queue counter never tracked: produced covers return hit:true (only misses/errors are hit:false), and ensure-queue dedup/HMR made client counting unreliable. Count on-demand covers natively in ensure_inner on the produce path (non-bulk, past the cache-hit gate), expose a cumulative uiEnsuredTotal in the pipeline stats, and derive the per-minute rate on the frontend from polled deltas — mirroring the lib backfill series. * perf(cover): measure cpm over trailing 5s instead of full minute A 60s rolling average added too much inertia, flattening real bursts and stalls in both the lib backfill and on-demand (ui) cover throughput. Compute the rate from the trailing 5s of samples (still extrapolated to per-minute), so the figure reacts promptly and decays to 0 within the window when idle.	2026-06-02 12:11:22 +03:00
cucadmuh	42aec6720c	fix(cover): stop per-song over-fetch + log failed cover downloads (#944) ... * fix(cover): stop per-song cover over-fetch (album/mf-* explosion) album_has_distinct_disc_covers returned true as soon as two tracks on the same disc had different cover ids. On Navidrome every song has its own mf-<id> coverArt, so almost every album was flagged "distinct disc covers" and backfill warmed one cover per track (~520k for ~170k tracks), filling album/ with mf-* dirs instead of ~one cover per album. Treat a release as having distinct disc covers only when each disc has a single consistent cover that differs across discs (genuine box set); per-song ids now collapse to one cover per album. Mirror the same fix in the TS albumHasDistinctDiscCovers used by on-demand warming. Adds regression tests on both sides. * docs(changelog): record per-song cover over-fetch fix (PR #944) Give the album/mf-* over-fetch fix its own [1.47.0] Fixed entry and a settingsCredits line under PR #944. * feat(cover): log failed cover downloads with album/artist name A non-200 (or network-failed) getCoverArt download was swallowed silently. Now the failure is logged with the resolved album/artist name and the server error, so a server refusing covers under backfill load (5xx/429/timeouts) is visible. - cover_resolve: describe_cover_entity() resolves a human label from the local index (album "Name" — Artist / artist "Name"), best-effort with id fallback. - cover_cache: log_cover_fetch_failure() in ensure_inner logs on the download error path; threads optional library_server_id through CoverCacheEnsureArgs so the name lookup happens only on failure. Backfill logs at normal level, on-demand misses at debug level.	2026-06-02 10:58:39 +03:00
cucadmuh	a63ba3c9cb	fix(cover-backfill): kill idle CPU spin and offline-cache menu re-walks (#943) ... * fix(cover-backfill): snapshot-diff worklist and live-tunable parallelism Aggressive cover backfill pegged one tokio worker at ~100% on large, fully-synced libraries while the download queues stayed empty. - Take two snapshots once per pass — the DB catalog (single GROUP BY) and the on-disk cover bucket (one directory walk) — and download the set-difference. No per-row `stat` syscalls and no re-scan loop; the empty cache case (heavy backfill) costs zero per-item disk hits. - Replace the front-loaded enumeration with a producer/consumer pipeline: the producer streams the catalog in chunks and feeds misses into a bounded channel; a fixed consumer pool keeps the download/encode pools saturated. - Make cover backfill parallelism runtime-tunable from the Performance Probe (threads slider + "Run full pass now"); HTTP download and CPU encode semaphores resize live. Not surfaced in app settings. - Add a "nothing changed" idle gate (catalog signature) so a settled pass is not re-run on every library:sync-idle, mirroring the analysis worker. - Cancel promptly on switch to lazy: consumers bail on enabled/focus change and the producer feeds via try_send so a full channel cannot deadlock. - Drop the per-item recursive disk walk from the ensure hot path. * fix(cover-backfill): cheap idle gate, settle on 404s, transient retries Follow-up to the snapshot-diff backfill: stop the periodic CPU spikes and the 89%-plateau wake storm on libraries whose covers can never reach 100%. - Idle gate is now disk-free: compare only the catalog COUNT(DISTINCT) instead of walking ~all cover dirs on every sync-idle. "Did the server change?" never touches the filesystem. Clear-cache commands re-arm the gate (rearm_idle_gate) since a clear leaves the catalog total unchanged, and the settings UI wakes the active server after a clear. - Settle the gate on any completed pass regardless of pending: remaining items are unfetchable-for-now (404), so the wake/sync-idle storm stops once the fetchable set is exhausted. - Stop auto-clearing .fetch-failed markers every pass (it defeated the 30-min backoff and re-attempted 404s forever). The manual "Run full pass now" sends force=true to clear them and retry; wake/sync-idle/configure stay opportunistic. - Rate-limit sync-idle passes (60s cooldown) as defence against chatty syncs. - Retry cover downloads up to 3x with backoff on transient failures (5xx / 429 / network), but never on a real 4xx so missing covers don't hammer the server. * fix(cover-cache): stop re-walking cover dirs from offline & cache menu The settings cover-cache section polled disk usage + progress every 15s for every server, each call doing a full recursive walk of the per-server cover directory. On a fully populated cache this caused periodic CPU spikes whenever that menu was open. - mod.rs: add a 10s TTL memo around the per-server cover dir walk (cached_dir_usage_for_server), shared by cover_cache_stats_server and library_cover_progress; invalidate on clear (per-server and clear-all). - CoverCacheStrategySection: recompute on entry only; rely on the cover:library-progress and cover:cache-cleared events for live updates; drop the per-cover cover:tier-ready refresh storm; turn the 15s loop into a 5-minute safety net. * fix(cover-backfill): keep emitting progress during the whole pass The producer finishes enumerating the worklist long before the consumer pool finishes downloading it, so progress was only emitted while feeding the channel — the "offline & cache" menu and overlay then froze through the entire drain phase. Replace the per-chunk emit with a 3s progress ticker that runs for the lifetime of the pass and is aborted once the consumers drain (final accurate emit still happens at settle). * docs(changelog): record cover-backfill idle-CPU fix (PR #943) Add [1.47.0] Fixed + Changed entries and a settingsCredits line for the cover-backfill idle CPU / offline & cache menu work.	2026-06-02 04:56:34 +03:00
cucadmuh	9925771a86	feat(browse,cover,perf): lazy catalogs, cover pipeline, and Performance Probe (#890) ... * fix(cover): per-server cache stats and cover pipeline perf probe Stop count_cached_cover_ids from borrowing sibling bucket counts so Settings progress no longer attributes one server's disk cache to another. Add cover pipeline queue stats (ui ensure queue, ui vs lib HTTP/WebP semaphores) to Performance Probe overlay, with clearer ui/lib labels. * fix(browse): stabilize in-page infinite scroll and cap cover memory caches Extract useInpageScrollSentinel for album grids and song lists so sentinel reconnects do not spam loadMore during scroll. Harden useAlbumBrowseData with sync loading refs, tighter root margin, and hasMore termination when dedupe adds nothing. Pause middle-priority cover work during SQL pagination and bound diskSrc/resolve/ensure tail maps on long cold-cache sessions. * refactor(browse): unify in-page infinite scroll hooks and sentinel UI Extract shared transport (viewport ref, async pagination guards, client slice) and InpageScrollSentinel so Albums, New Releases, Artists, and song lists use one pagination pattern instead of duplicated IntersectionObserver wiring. * fix(browse): prioritize album SQL pagination over cover ensures Pause the entire webview ensure pump during grid page fetches, resume after SQL settles, add cover-queue backpressure before load-more, and re-probe the sentinel when pagination finishes so cold-cache scroll does not stall. * fix(browse): unblock covers, SQL spawn_blocking, and pagination retry Pair grid-pagination hold begin/end on stale fetches, resume the ensure pump after SQL, retry load-more when the cover backlog drains while the sentinel stays visible, and run album browse SQL on spawn_blocking so Tokio stays responsive during library_advanced_search. * feat(browse): All Albums client-slice scroll on local index (Artists-style) Load the filtered catalog once from SQLite when the library index is ready, then grow the visible grid with useClientSliceInfiniteScroll instead of offset SQL pagination per scroll. Network-only servers keep page mode. * fix(browse): lazy local catalog chunks instead of full 50k SQL fetch All Albums slice mode now loads 200 albums first, shows the grid immediately, then appends catalog chunks in the background as the user scrolls. Avoids the blocking library_advanced_search that hung the app on large libraries. * fix(browse): keep album covers loading during active grid scroll Pass high ensure priority and the in-page scroll root to AlbumCard on All Albums, stop pausing cover traffic for background catalog chunks, and never trim high-priority ensure jobs from the queue during scroll bursts. * fix(cover): viewport priority tiers and unstick ensure invoke pump All Albums uses IO-driven high/middle instead of blanket high; release only on unmount so scroll-ahead jobs are not dropped on reprioritize. Ensure queue shares one Rust flight per cover id, attaches duplicate waiters without consuming invoke slots, and times out wedged calls. Warm the first viewport slice on large grids; acquire CPU permits before spawn_blocking in cover_cache to avoid blocking-thread deadlocks. * fix(cover): wire in-page scroll root on New Releases and Lossless grids AlbumCard IO uses the same viewport id as VirtualCardGrid so cover ensure priority tracks visible in-page rows like All Albums. * fix(browse): lazy local artist catalog in 200-row chunks Replace runLocalBrowseAllArtists bulk fetch with paginated local-index chunks so large libraries do not hang on open; preserve text search, starred, letter filter, and client-slice scroll behavior. * feat(perf): add RSS and thread CPU groups to Performance Probe Extend performance_cpu_snapshot with process RSS (psysonic + WebKit children) and in-process thread CPU breakdown. Classify tokio-rt-worker and tokio-* workers separately from glib, audio/pipewire, reqwest, and other misc threads (Linux /proc only). * feat(perf): redesign Performance Probe with tabs, pins, and overlay layout Split the probe into Monitor (live metric cards, per-metric overlay pins, corner and opacity controls) and Toggles (diagnostic tree). Share live polling via perfLiveStore; label analysis/cover pipeline blocks in the HUD. * feat(perf): overlay sparklines, macOS CPU/memory, and sync fixes Add 1-minute pinned-metric sparklines with right-aligned growth and a shared poll clock. Enable macOS performance snapshots via sysinfo. Fix overlay infinite loop from unstable history snapshots, bar/sparkline tick jitter, and probe bar rescale flicker. * docs: CHANGELOG and credits for PR #890 * perf(probe): scoped CPU poll, adjustable interval, lazy thread groups Read only psysonic + WebKit children instead of the full process table; macOS uses sysctl host CPU and refreshes cached child PIDs. Add 0.5–10s poll slider (default 2s). Collect /proc thread groups only when the Monitor section is open or a thread metric is pinned. * feat(perf): three-way overlay mode switch (off / FPS / pinned) Add Monitor control for overlay visibility: hidden, FPS-only, or pinned metrics from Monitor. Live CPU poll runs only in pinned mode with live pins.	2026-05-29 04:40:31 +03:00
Frank Stellmacher	839c438a6d	fix(cover): sanitize server_index_key so Windows :port URLs work (#889) ... * fix(cover): sanitize server_index_key on disk so Windows accepts ":port" URLs `serverIndexKeyFromUrl` (frontend) strips the URL scheme and leaves the rest of the host as the index key — for a Navidrome instance running on the default `:4533` port that is `host:4533/...`. On Linux/macOS the `:` is fine as a path segment; on Windows `CreateDirectory` rejects the whole path with `ERROR_INVALID_NAME` (os error 123). Result: every `cover_cache_ensure` and `cover_cache_peek_batch` rejected its promise and the album / now-playing / mainstage / lightbox surfaces stayed blank. Empirically verified — switching the active server to a colon-free reverse-proxy URL made the covers load again without any other change. Centralize the fix in a new `cover_server_dir(root, key)` helper next to the existing `cover_entity_relative_dir`: it runs `sanitize_path_segment` on the server key the same way kind/entity ids are already cleaned, so `host:4533` becomes `host_4533` and embedded URL paths collapse into one flat bucket instead of nested directories. Every call site that wants the server bucket — `cover_dir`, `count_cached_cover_ids`, `dir_usage_for_server`, the clear-server command, and `clear_cover_fetch_failures` in the backfill worker — now goes through it. The on-disk layout changes (no more colons, no more nested URL paths), so bump `LAYOUT_STAMP` to `canonical-segment-v4`. The existing stamp-mismatch sweep at startup wipes the legacy buckets — users with a previously-working (colon-free) layout rebuild the cache lazily as they browse. Library, offline, and hot-cache data are not touched. Adds two unit tests covering the sanitization on `cover_server_dir` and the `cover_dir` passthrough. Follow-up to #878 (which introduced `cover_cache_layout.rs` with `sanitize_path_segment` applied only to kind/entity_id). * chore(windows): silence dead_code warnings in debug taskbar_win build `lib.rs` gates `taskbar_win::init` on `cfg(not(debug_assertions))` (PR #866 — debug runs alongside an installed release instance and must not fight it for the taskbar subclass). The `update_taskbar_icon` command still ships in debug and early-returns until `init` populates the COM/HWND atomics, so the init-only helpers (icon HICONs, button IDs, subclass plumbing, `make_buttons`, `subclass_proc`, `init` itself) all look unused — 14 dead_code warnings on every Windows debug `cargo build`. File-level `#![cfg_attr(debug_assertions, allow(dead_code))]` suppresses those warnings only in the debug profile. Release builds keep the strict dead-code check, so a real removal would still surface there. * docs(release): CHANGELOG for windows cover-cache server-key fix (PR #889)	2026-05-28 23:13:27 +02:00
Frank Stellmacher	9fa086c428	feat(server): dual server address (LAN + public) per profile (#880) ... * feat(server): ServerProfile.alternateUrl + shareUsesLocalUrl Additive optional fields on ServerProfile to back the dual-address work. 'alternateUrl' is the optional second endpoint (e.g. LAN counterpart of a public 'url' or vice versa); 'shareUsesLocalUrl' chooses which of the two goes into Orbit / entity / magic-string shares when both are set. Pure schema change — no runtime consumers yet. Single-address profiles keep the same persisted shape; the optional fields are simply absent. * feat(server): serverEndpoint module with LAN classification + IPv6 Introduces a single home for connect-/share-layer URL utilities that the dual-address work will build on: * normalizeServerBaseUrl(raw) — aligned with serverProfileBaseUrl * isLanUrl(url) — IPv4 (unchanged) + IPv6: ::1 loopback, fe80::/10 link-local, fc00::/7 ULA, and IPv4-mapped (both dot-decimal and the URL-API-normalized hex form, since new URL() rewrites ::ffff:1.2.3.4 to ::ffff:HHHH:HHHH) * allNormalizedAddresses(profile) — deduped [url, alternateUrl?] * serverAddressEndpoints(profile) — LAN-first ServerEndpoint[] isLanUrl moves out of useConnectionStatus.ts (the hook now imports + re-exports it for backward compatibility); OrbitStartModal points at the new path directly. 38 unit tests cover the IPv4/IPv6 matrix, dedupe, ordering, and edge cases. * feat(server): pickReachableBaseUrl with in-memory connect cache Adds the runtime connect layer on top of serverEndpoint: * pickReachableBaseUrl(profile) — sequential LAN-first ping via the existing pingWithCredentials; first OK wins. * ensureConnectUrlResolved(profile) — boot / switch / online-event entry point (same mechanism, intent-named). * In-memory cache keyed by profileId; sticky behaviour tries the cached endpoint first, falls back to the natural order on miss. * invalidateReachableEndpointCache(profileId?) — single-profile and whole-cache flushes for profile-edit / credentials-change / online. * getCachedConnectBaseUrl(profileId) — sync getter for getBaseUrl fallback path (next commit). The cache is **session-only**, never persisted. Single-address profiles keep the same shape (one endpoint, one ping). 9 new unit tests cover single + dual address, LAN-first preference, fallthrough, unreachable + stale-cache clear, sticky hit, and the two invalidate flavours. * feat(server): getBaseUrl reads connect cache, falls back to primary url Dual-address profiles need 'getBaseUrl' to return the runtime-probed connect URL (LAN at home, public elsewhere), not the literal primary 'url'. To keep the sync getter shape that ~60 call sites depend on, the store now reads 'getCachedConnectBaseUrl(server.id)' from the serverEndpoint cache. If no probe has run yet (very early boot, before switchActiveServer populates the cache), it falls back to the normalized primary URL — identical to today's behaviour. Single-address profiles see no difference; their cache entry equals serverProfileBaseUrl(url). * feat(server): switch + connection-status probe via ensureConnectUrlResolved Both surfaces that previously called pingWithCredentials(server.url, ...) directly now route through the dual-address connect layer: * switchActiveServer awaits ensureConnectUrlResolved(server), reads the identity (type/serverVersion/openSubsonic) off the returned ping, and hands probe.baseUrl (not server.url) to scheduleInstantMixProbeForServer so the AudioMuse probe also hits the reachable endpoint. * useConnectionStatus.check() does the same on every 120 s tick — sticky cache fast-paths the steady state, and a network change naturally flips the active endpoint without a manual retry. * useConnectionStatus.retry() and the 'online' event flush the cached entry for the active profile first, so the next probe starts from the natural LAN-first order instead of revalidating a stale URL. Single-address profiles behave identically: one endpoint in the list, one ping per check. No behaviour change for them. * feat(server): PickReachableResult.ping + connectBaseUrlForServer helper Two additive extensions to the serverEndpoint surface that the upcoming CONNECT migrations all need: * 'ping: PingWithCredentialsResult' on the OK branch of PickReachableResult so callers (switchActiveServer, useConnectionStatus) can read type/serverVersion/openSubsonic from the same probe instead of issuing a second pingWithCredentials. * connectBaseUrlForServer(server) — sync getter for the connect URL of *any* saved profile (active or not). Reads the cache, falls back to normalized primary url on miss. Becomes the canonical helper for non-active-server HTTP traffic (apiForServer, stream URLs, cover fetches, library bind session). Tests: pingOk() fixture now returns identity fields; the single toEqual-asserting test in pickReachableBaseUrl became a guarded read to keep equality concise while still asserting on the new ping field. * feat(connect): apiForServer + stream URLs route via connectBaseUrlForServer Non-active-server HTTP traffic (the apiForServer entry point, used by QueuePanel cross-server cover fetches and share-paste resolution) and the stream-URL builders previously read 'server.url' directly, bypassing the dual-address connect cache. Both now go through connectBaseUrlForServer, which serves the cached LAN/public endpoint when one exists and falls back to the normalized primary url otherwise. buildStreamUrl(id) now uses the baseUrl it already pulled from getBaseUrl() (which is connect-cache aware as of 2a6d8283) instead of re-normalizing server.url — same value for single-address profiles, correctly dual-address for the rest. Single-address callers see no behaviour change. * feat(connect): cover fetch + cover-cache ensure use connect URL Both cover-fetch surfaces previously read 'scope.url' / 'server.url' straight into HTTP requests, which would freeze the cover pipeline on the primary URL even when a LAN endpoint was reachable: * buildCoverArtFetchUrl(ref, tier) — for the 'server' scope (queue rows cached against a non-active profile) and the 'playback' scope (cross-server playback) now resolves the connect URL via connectBaseUrlForServer before handing off to buildCoverArtUrlForServer. * ensureArgsFromRef in coverCache.ts — restBaseUrl for the Tauri 'cover_cache_ensure' invoke now uses the cached connect URL for both 'server' and 'active'/'playback' scopes. Scope.url itself remains the index-stable primary URL — that's what storageKeys (INDEX) consume. Only the HTTP base shifts to the connect endpoint, exactly the split the spec calls out. * feat(connect): library bind + server-test probe via ensureConnectUrlResolved Two more 'rebind / re-test an existing saved server' paths still pinged the literal primary URL — both go through the dual-address connect layer now: * bindIndexedServer (librarySession.ts): used to ping server.url then pass serverProfileBaseUrl(server) as the bind 'baseUrl' to Rust. Now: one ensureConnectUrlResolved call covers both — probe.baseUrl feeds librarySyncBindSession directly, so Rust's per-server library sync uses whichever endpoint actually answered. Drops the pingWithCredentials and serverProfileBaseUrl imports. * testConnection (ServersTab.tsx): same shape — probe via the connect layer, read identity from probe.ping, hand probe.baseUrl to scheduleInstantMixProbeForServer so the AudioMuse probe also hits the connect endpoint. Single-address profiles still ping once. Add/edit save handlers (handleAddServer / handleEditServer) keep the direct pingWithCredentials against the user-entered data.url — that flow is the dual-address verify hook, which PR 2 extends. * feat(server): serverShareBaseUrl — public-by-default share URL picker Adds the share-layer companion to connectBaseUrlForServer. Different intent: connect picks the reachable endpoint for HTTP, share picks the URL that goes into Orbit invites / entity share payloads / queue share links / magic strings — where a guest opening the link is not on the host's LAN. Logic per spec §5.1: * Single-address profile → that one address (normalized). * Both set, default flag → public. * Both set, shareUsesLocalUrl flag → local. * Edge cases (both LAN, both public, missing one of the two): fall back to the first endpoint in the list so the function is total. Empty profile returns the normalized url (possibly empty) — defensive, never throws. 7 unit tests pin all six branches plus the empty case. Call-site migration (Orbit, copyEntityShareLink, QueuePanel, magic string srv field, findServerIdForShareUrl, Orbit LAN warning) lands in the next commits. * feat(share): Orbit + entity + queue shares use serverShareBaseUrl Four share-encoding surfaces previously read 'getBaseUrl()' (connect) or the raw 'server.url' (primary) when embedding the host into outgoing share links — both wrong for a dual-address profile: * copyEntityShareLink (track / album / artist / composer) — was getBaseUrl(); now reads serverShareBaseUrl(active). Guests opening the link are off-LAN, so public is the right default. * OrbitStartModal — was raw server?.url for both 'buildOrbitShareLink' and the LAN warning. Now goes through serverShareBaseUrl; the LAN warning correspondingly reflects the address the guest will see, not the host's primary. * OrbitSharePopover — same fix on the host-only share popover. * QueuePanel.handleCopyQueueShare — was getBaseUrl(); same migration. Single-address profiles return exactly the same string from serverShareBaseUrl as serverProfileBaseUrl(server.url) did before, so no behaviour change for the common case. shareUsesLocalUrl flips the default to LAN for the rare 'share into a LAN-only group' use; that checkbox lands with the form UI in the next sub-phase. findServerIdForShareUrl + paste-side share matchers are migrated in the next commit (read-side of the same contract). * feat(share): paste-side resolves dual-address profiles + connect URL Read-side counterpart to the encode-side migration: * findServerIdForShareUrl(servers, shareSrv) now matches a profile when shareSrv normalizes to either profile.url OR profile.alternateUrl. Without this, a paste of a link generated against the host's LAN address (shareUsesLocalUrl=true) would fail to find the local saved profile even though it's the same server. * resolveSharedSong / resolveShareSearchAlbum / resolveShareSearchArtist in enqueueShareSearchPayload now hand connectBaseUrlForServer(lookup.server) to the *WithCredentials HTTP calls instead of the raw lookup.server.url. The looked-up profile may be dual-address; this routes the song / album / artist fetch through whichever endpoint is currently reachable. Indirect consumers (shareServerOriginLabel, shareQueueServerContext) read the same findServerIdForShareUrl, so no code change needed there — they now match both addresses transparently. * feat(verify): serverFingerprint + same-server verification Pure-logic core of dual-address verify. Three exports: * fetchServerFingerprint(baseUrl, user, pass) — one ping (envelope 'version' extracted alongside type/serverVersion/openSubsonic) plus four soft-fail optional calls in parallel (getMusicFolders, getUser, getLicense, getIndexes). Optional failures collapse to null fields, not whole-fingerprint failure. Subsonic-generic — never branches on type === 'navidrome'. indexesDigest is a hash of letter-count plus sorted first 20 artist ids, so two probes against the same library see the same digest without comparing full payloads. * compareFingerprints(a, b) — strict on the ping triple (type case-insensitive, serverVersion exact, openSubsonic boolean); envelope apiVersion informational only. Body signals counted only when both sides have a non-null value; empty musicFolders [] on both sides still counts as a matching signal. Result: 'match' (>=1 common signal all agreeing) / 'mismatch' (any common differs) / 'insufficient' (0 common). No 'save anyway' for insufficient in v1. * verifySameServerEndpoints(profile, user, pass) — single-address short-circuits to ok:true (nothing to verify). Otherwise parallel fingerprint probes, then pairwise compare. Ping-fail on any endpoint reports the offending host for the UI. 20 unit tests cover the compare matrix (every body signal in every direction), Navidrome- vs minimal-Subsonic-shape probes, ping-fail, and all four verify outcomes (ok / unreachable / mismatch / insufficient). HTTP mocked via vi.stubGlobal('fetch') + plain-object Response shape (avoiding any Response-polyfill dependency). * feat(boundary): resolve_host_addresses Tauri command + TS wrapper Adds one additive invoke for dual-address form hints. Spec §9.1 + contracts.md §6. Rust side (src-tauri/src/lib_commands/app_api/network.rs): * #[tauri::command] resolve_host_addresses(hostname: String) -> Result<Vec<String>, String> * tokio::net::lookup_host with a port suffix (':0'); discards the port from each SocketAddr, dedupes addresses via HashSet. * strip_port helper handles 'host:port', 'ipv4:port', '[ipv6]:port', '[ipv6]' (no port), and bare 'ipv6' (left as-is for lookup_host to wrap). 7 unit tests cover each shape. * Lookup failure → Ok(vec![]) so a DNS hiccup doesn't surface as an error toast or block the save flow — form-hint only. Frontend wrapper (src/api/network.ts): * resolveHostAddresses(hostname) — trims, invokes, swallows errors to an empty array so consumers get a clean total function. §04 boundary note: additive command, no breaking change. Added to the invoke_handler! generate_handler list in lib.rs at the natural alphabetical-ish slot near migration_run. * feat(form): AddServerForm — second optional address + share flag + DNS hint UI side of dual-address. AddServerForm now carries the four new moving parts spec §6 calls out: * Second address field ('Second address (optional)') under the primary URL. Placeholder flips to suggest the opposite kind of address based on the primary's LAN classification. * Contextual hint under the second field when it's empty — 'add a public address for outside-home use' (primary is LAN) or 'add a local address for faster home access' (primary is public). Hint disappears once the field has content. * Two-LAN client-side check on submit: when both addresses classify as LAN, save is blocked with a toast. Reverse case (both public) intentionally allowed per spec §6.3. * shareUsesLocalUrl checkbox — visibility rule per spec §5.3: hidden until the second address has content; shows with the persisted value on edit; cleared when the user empties the second address before save. DNS hint: on primary-URL blur we call the Tauri resolve_host_addresses command and classify the response by isLanUrl. Literal IPs skip DNS (already classified locally). DNS miss → no hint surfaces (never blocks save). Used only for the hint text — connect still goes through pingWithCredentials. i18n: 9 new keys in en + ru (serverAlternateUrl*, shareUsesLocalUrl*, serverBothLanError). Other locales fall back to English. onSave signature widened to 'void | Promise<void>' — ServersTab / Login both accept that already. The save flow itself still calls the legacy pingWithCredentials path; verify wiring lands in 2e. * feat(verify): wire verifySameServerEndpoints into add + dual-edit flows Save flow now blocks persisting a dual-address profile that fails the same-server check. Spec §6.4 + §7.4. handleAddServer: * When data.alternateUrl is non-empty, runs verifySameServerEndpoints BEFORE the existing ping. mismatch / insufficient / unreachable each surface a localized toast and abort the save (no addServer call). * Single-address adds keep the legacy single-ping path — no extra network round-trip when there's nothing to verify. handleEditServer: * Unconditional save remains the default — but if the edit either introduces / changes alternateUrl OR changes url / username / password while alternateUrl is set, verify runs first and may block. * After persist, invalidates the reachable-endpoint cache for this profile id so any sticky cached connect URL from before the edit is re-probed on the next access (credentials may have changed, alternate may have appeared). i18n: 4 new toast keys in en + ru (dualAddressVerifying placeholder for future progress UI, plus mismatch / insufficient / unreachable). Other locales fall back to English. announceVerifyResult helper keeps the toast routing in one place so handleAddServer and handleEditServer share the same surface. * i18n(settings): dual-address keys across all 9 locales Adds the 13 new dual-address keys to the remaining 7 locales: de · fr · es · nl · nb · ro · zh. * serverAlternateUrl + placeholderPublic / placeholderLocal * serverAlternateUrlHintAddPublic / HintAddLocal (contextual hints under the second-address field) * serverBothLanError (client-side two-LAN validation) * dualAddressVerifying / Mismatch / Insufficient / Unreachable (toast strings; Unreachable carries the {{host}} interpolation) * shareUsesLocalUrl + Desc (checkbox label + short description) Placeholders (example.com URL / 192.168.1.100:4533) kept identical across all locales — same shape the existing serverUrlPlaceholder already uses. Single coordinated sweep so every supported language ships dual- address fully translated rather than falling back to English. * feat(cover): cover_cache_rename_server_bucket Tauri command Adds the disk-side companion to the upcoming URL-change remigration flow (dual-server-address spec §8.3). When a user edits the primary url so the derived index key changes, the SQLite migration already re-tags rows via the existing migration_run command — this command moves the cover-cache bucket on disk so cached WebP tiles stay reachable under the new key. Behaviour: * old_key == new_key → no-op. * Old bucket missing → no-op success (nothing to migrate). * New bucket missing → simple fs::rename (fastest path). * Both exist → recursive merge with 'prefer existing' on file collision (the newer bucket wins; data is never lost). * Emits 'cover:bucket-renamed' with {oldKey, newKey} on success so the frontend disk-src cache can invalidate stale URLs. Sanitization: rejects empty, backslash, and '..' path segments at the FS boundary. Forward slashes are legitimate (a Navidrome mounted at a subpath like 'music.example.com/navidrome' produces an index key with one); they're handled by Path::join. 4 unit tests cover key sanitization (accepts real keys, rejects traversal/backslash) and the merge semantics (unique-file move + prefer-existing on collision). Registered in lib.rs invoke handler. * feat(remap): rewriteFrontendStoreKeysForRemap — explicit oldKey→newKey path Adds the URL-change remigration entry point next to the existing UUID→indexKey migration. Same plumbing (offline store, hot cache, analysis-strategy maps) but driven by explicit { oldKey, newKey }[] mappings instead of being derived from the current servers list. Also folds in the player-side cleanup the existing path didn't have: * queueServerId — if currently bound to a remapped index key, repoint to the new one so playback continues through the rename instead of looking unbound. * analysisStrategyStore — runs the same map-key swap inline (the 'migrateServerOverrides' helper handles UUID→indexKey, not indexKey→indexKey). Same prefer-existing-on-collision semantics as the disk-side cover_cache_rename_server_bucket — if a destination key already carries data, we keep it. 8 unit tests cover no-ops, the four store rewrites, the queue repoint, the 'leave other servers untouched' guarantee, and the collision case. * feat(remap): serverUrlRemigration orchestrator — 4-stage pipeline Single-file orchestrator that runs the full URL-change remigration when a profile edit shifts the primary url's derived index key. Spec §8 + contracts.md §4. Two exports: * indexKeyRemapForUrlChange(prev, next): IndexKeyRemap | null — short-circuits scheme-only edits ('http://x' ↔ 'https://x'), trailing-slash differences, and alternateUrl-only changes by normalizing both sides through serverIndexKeyFromUrl and returning null when they match. Empty urls also return null. * runIndexKeyRemigration(remap): Promise<IndexKeyRemigrationResult> — runs inspect → run → frontend-rewrite → cover-rename in order, aborts on the first failure and reports the offending stage. Failure ordering rationale: * inspect / run failures stop the destructive step — DB is untouched, the caller can retry safely. * frontend rewrite swallows errors (best-effort; zustand persist catches up on rehydrate next session). * cover-rename failure is reported but does NOT roll back the DB rows (that would be even more destructive); covers under the old key recover via the existing cover backfill on next access. 10 tests cover the detect logic (5 cases including the path-suffix case) plus all four pipeline stages with mocked Tauri invoke — including verifying the legacyId/indexKey mapping shape lands on both inspect and run calls. * feat(remap): wire URL-change remigration into ServersTab edit flow handleEditServer now detects a primary-url index-key change BEFORE any other save logic (verify, persist, etc.) and orchestrates the full remigration when one is needed. Flow: * indexKeyRemapForUrlChange(prev, next) — null for scheme-only edits / alternateUrl-only edits / no-op edits, so the common case falls straight through without prompting the user. * On a real remap → confirm modal via useConfirmModalStore.request (danger style, plain-language oldKey → newKey copy, explicit 'cannot be undone'). User cancel → abort save entirely. * On confirm → runIndexKeyRemigration pipeline. Failure surfaces a stage-specific toast (inspect / run / cover-rename — wording per spec §8.4 + the recoverability matrix in serverUrlRemigration.ts). * On success → fall through to the existing dual-address verify, then auth.updateServer + invalidate cache + ping (unchanged). i18n: 7 new keys (urlRemigrationTitle / Message with oldKey + newKey interpolation / Confirm / Progress / FailureInspect / FailureRun / FailureCoverRename) — full sweep across all 9 locales (en, de, ru, fr, es, nl, nb, ro, zh). Test fixup: the offline-store collision test in rewriteFrontendStoreKeys needed an 'as unknown as' cast — the existing OfflineTrackMeta type doesn't carry the test-only marker property, and a one-step cast TS rejected as not overlapping enough. * feat(magic): magic-string v2 encode/decode with dual-address fields Adds the v2 wire format for server invites. Spec §10 + contracts.md §5.2. Encode is opportunistic: payloads with alternateUrl set OR shareUsesLocalUrl=true emit v2; everything else stays v1 byte-identical. This means single-address profiles keep producing exactly the same invite they did before — older receivers can't tell the difference. v2 shape: { v: 2, url, alt?, shareLocal?, u, w, n? } * url — the host's share URL (public by default; LAN if shareLocal), NOT necessarily the host's primary URL. Receiver treats it as the primary of the new profile (their own index key). * alt — the host's alternate address (the other half of the pair). * shareLocal — mirrors the host's shareUsesLocalUrl preference so onward shares from the receiver behave the same way. Decode accepts both v1 and v2. v2-only fields are left undefined when decoding a v1 payload (no '|| undefined' shim — kept absent so zustand persist diffs stay clean). Defensive: an empty alt on a v2 invite decodes as if the field were absent, never as ''. Test updates: 'rejects a payload with the wrong version' now targets v: 3 (v: 2 became valid). 7 new tests cover the v1/v2 wire-format choice, both v2 round-trip shapes, the v1-decode-into-v2-fields- undefined backward-compat case, and the empty-alt edge. * feat(magic): wire magic-string v2 through invite-encode + paste consumers Five surfaces now produce / accept v2 invites with the dual-address fields end-to-end: Encode side (admin → user invite): * MagicStringModal — looks up the saved profile that matches the serverUrl prop (primary OR alternateUrl normalize-match) and encodes through serverShareBaseUrl + alternateUrl + the share flag. Falls back to plain v1 for single-address profiles. * UserForm — same lookup + encode plumbing on the in-form 'Save & get magic string' flow. Decode side (paste invite into form): * AddServerForm useEffect (initialInvite from outer state) + handleMagicStringChange (live paste) both pull alternateUrl and shareUsesLocalUrl off the decoded payload into form state, so the second-address field + checkbox surface immediately on a v2 paste. * Login form state gains hidden alternateUrl + shareUsesLocalUrl fields (not user-editable — Login stays single-address by design); v2 paste / initial-invite both populate them so they persist with the new profile via addServer. handleQuickConnect likewise forwards the existing dual-address shape for already-saved profiles. * attemptConnect signature widened with the two optional fields; addServer / updateServer call sites conditionally include them only when alternateUrl is non-empty so single-address profiles keep their lean persisted shape. Both encode helpers share the same magicPayloadAddressFields lookup (MagicStringModal + UserForm) — the duplication is acceptable for two co-located small files but a single shared helper would be the natural place to consolidate if a third encoder appears. * fix(form): magic-string submit forwards decoded alternateUrl + share flag AddServerForm.submit's magic-string branch was forwarding only name/url/username/password from the decoded payload, silently dropping alternateUrl + shareUsesLocalUrl. A v2 invite pasted into the magic-string field and saved would land in the store as a single- address profile even though handleMagicStringChange had already populated the dual-address fields into form state for display. Now the magic branch picks the dual-address fields off the decoded payload (same conditional spread as the non-magic branch a few lines down) so v2 invites round-trip end-to-end through the form. * fix(verify): extractUserId drops username fallback to avoid false mismatches Spec §7.2 footnote: 'fallback (normalized username) only if no id on either side'. The old extract unconditionally fell back to the authenticated username whenever user.id was empty — but the comparator can't tell username-fallback apart from a server-supplied id. So a server pair where endpoint A returns user.id='42' and endpoint B only returns user.username='frank' would land in compare with userId values '42' vs 'frank' and report mismatch on what is actually the same user on the same server. Now extractUserId returns null when user.id is absent. Both sides returning null means compareFingerprints skips userId as a common signal (correct behaviour per the §7.3 'both sides have a value' rule). One new test pins it; the call site drops the redundant fallbackUsername argument. * fix(remap): rewriteFrontendStoreKeysForRemap migrates coverStrategyStore Spec §8.2 lists 'analysis/cover strategy maps' as targets of the URL-change remigration. The For-Remap path covered analysis but missed cover — a user-set cover strategy override (e.g. 'aggressive' for one server) would silently drop when the user edited that profile's primary URL, because the key tagged under the old index key never made it to the new one. Same shape as the analysis remap inline: prefer-existing on collision (newer key wins), delete the legacy entry afterwards. One new test pins the move; setup adds a fresh useCoverStrategyStore reset between cases. * fix(cover): is_safe_index_key rejects absolute paths + drive letters Defense-in-depth gap in cover_cache_rename_server_bucket. The old sanitizer only rejected backslashes and '..' segments — '/etc/passwd' on Unix and 'C:\Windows' on Windows both passed. Path::join with an absolute argument REPLACES the base path, so root.join('/etc/passwd') on Unix would walk out of cover-cache entirely. Real index keys never reach this state — they come out of serverIndexKeyFromUrl which strips schemes and trailing slashes, so no leading separator and no drive-letter prefix is ever produced. But the comment promises 'defense in depth' and that defense is worth completing. Now rejects: * empty keys (would join to the cover-cache root itself) * leading '/' or '\' * Windows drive-letter prefix 'X:' (case-insensitive ASCII letter) * backslashes anywhere (separators are forward-slash only) * '..' segments after split('/') One new Rust test covers all four cases. * fix(connection): isLan badge reflects active endpoint, not primary url useConnectionStatus.isLan was reading isLanUrl(server.url) — the *primary* address — so a dual-address profile that had fallen over to its public alternate kept advertising 'LAN' in the badge until the user looked at the server URL itself. Spec call-site-checklist line 25: 'active endpoint kind for badge (LAN vs extern)'. Now: ensureConnectUrlResolved returns probe.endpoint.kind on success; the hook tracks the latest kind in local state and surfaces that. Pre-first-probe fall-through keeps the old primary-URL classification so the badge has something to render at mount time. Three tests pin the three states (LAN-active after probe, public- active after fallback, primary-fallback before first probe completes); plus one for the online-event handler that invalidates the cache and re-probes. * fix(server): dedupe concurrent pickReachableBaseUrl calls for same profile Multiple call sites probe on roughly the same beat — useConnectionStatus 120-s tick + online handler + initial mount, switchActiveServer, bindIndexedServer, and ServersTab.testConnection. Two near-simultaneous probes for the same profile both observed an empty cache, both pinged every endpoint, and raced to write the sticky URL with last-write-wins. On a dual-address profile the slower probe could stomp the correct LAN sticky a millisecond after it was set. Now: an in-flight Map<profileId, Promise<PickReachableResult>>; a second call for the same id during an active probe returns the same promise instead of starting a fresh one. The finally-clear keeps the dedup window narrow (one settled probe → next call probes fresh). invalidateReachableEndpointCache deliberately doesn't touch the in-flight map — the racing probe's own finally will clean up. Three tests pin it: two concurrent calls share one ping; a fresh call after the previous settled does ping again; plus the existing shareBaseUrl two-LAN-with-flag-on test + the two-public-default test for completeness. * refactor(magic): extract magicPayloadAddressFields to single shared helper The lookup that turns a serverUrl into the v2-encode-ready { url, alternateUrl?, shareUsesLocalUrl? } shape was duplicated byte-identically between MagicStringModal.tsx and UserForm.tsx — both Navidrome admin user-mgmt encode sites. Workdocs §03 / §1a 'one source of truth for behavior'. Now lives in src/utils/server/serverMagicString.ts as a named export. Both call sites import + pass the current servers list (via useAuthStore.getState().servers) — keeps the helper pure so it's straight to unit-test if a third encoder ever appears. * fix(cover): wire cover:bucket-renamed listener for URL-change remigration cover_cache_rename_server_bucket emits cover:bucket-renamed on successful rename / merge, but no frontend listener was wiring it in. After a URL-change remigration the in-memory disk-src cache still held entries pointing at `{root}/{oldKey}/…/.webp` paths the disk no longer carries — the next read would serve a stale URL until the entry naturally aged out. New: forgetDiskSrcForServer(serverIndexKey) in diskSrcCache drops every cover entry under one server key in one pass (the existing forgetDiskSrcPrefix needs a non-empty coverArtId and can't blanket- clear a server). useCoverArtBridge subscribes to cover:bucket- renamed and calls it with oldKey. Tests: forgetDiskSrcForServer happy path, empty-key defensive no-op, no-match no-op; plus a regression-pin on the existing forgetDiskSrcPrefix so the two helpers don't drift in semantics. * test(dual-server): fill the high-impact coverage gaps from the branch review Adds the tests the review identified as missing on the hot-path UI flows + the partial-fail / alt-only edges: * AddServerForm.test.tsx (new) — five component tests: single-address save / dual-address save / two-LAN block-with- toast / v2 magic-string paste forwarding alt + share flag / edit-flow stripping alt+flag when the field is emptied. * Login.test.tsx (new) — v2 invite paste persists alternateUrl + shareUsesLocalUrl onto the saved profile; v1 invite leaves those fields undefined. * shareLink.test.ts — alternateUrl-only match for findServerIdForShareUrl (the dual-address paste case where the host shared the LAN URL via shareUsesLocalUrl=true); 'first hit wins' across two profiles where both match. * serverFingerprint.test.ts — partial-fail mix (folders+user ok, license+indexes rejected) so a Promise.allSettled regression wouldn't go silent. * cover_cache/mod.rs — rename_bucket_inner extracted as a testable FS-only helper (the Tauri command wrapper now just locks state + calls it + emits the event), plus six tests covering empty/unsafe keys, no-op-when-missing, no-op-when- equal, simple-rename, merge-with-prefer-existing. Also: every test fixture that I authored on this branch using 'frank' / 'frank@example.com' as a stand-in username/email swapped to 'tester' / 'tester@example.com' — keeping personal names out of test data is the codebase convention (see factories.ts) and is now a memory rule. * fix(test): typed mock access for pingWithCredentials in useConnectionStatus vi.mocked(...) returns the mock typed properly — bare .mock fails tsc because the imported function signature isn't a vi.Mock at import time. * docs(changelog): dual server address (PR #880) * fix(test): align diskSrcCache tests with main cover storage key API After rebase onto main, forgetDiskSrcPrefix takes a CoverArtRef-shaped argument and storage keys include cacheKind; merge coverDiskUrl tests from main with dual-address forgetDiskSrcForServer coverage.	2026-05-28 14:36:25 +03:00
cucadmuh	df3533bb5a	fix(cover): Windows thumbnails, tier fallback, PNG decode, coverArt id (#878) ... * fix(cover): tier fallback for sparse surfaces and Windows asset URLs Sparse UI (player bar, queue) now reads disk covers via the same tier ladder as dense grids, so a warm 800.webp satisfies a 128px request. Reject non-asset convertFileSrc results on Windows, widen Tauri asset scope, and seed ladder keys on cover:tier-ready. applyDiskPath uses seedGridDiskSrcCache only to avoid notify/subscriber infinite loops. * fix(artist): top-track thumb uses album coverArt already warm in grid Song coverArt ids often differ from album cover ids (e.g. Octastorium in the grid vs empty track thumb). Prefer the album row's coverArt on artist pages and ensure high priority for 32px dense cells. * fix(cover): albumId for playback/queue; no broken img until disk URL ready Prefer albumId over track-id coverArt (Navidrome). Wire queue to CoverArtImage with playback scope. CoverArtImage renders a placeholder div until asset src exists to avoid the browser broken-image icon. * fix(test): add song id to resolveArtistPageSongCoverArtId fixture Pick<SubsonicSong, …> requires id; fixes tsc in CI/build. * fix(cover): resolve albumId for Now Playing and artist top tracks Prefer albumId when album.coverArt echoes track id; use sparse surface on artist suggestion thumbs; apply resolveSubsonicSongCoverArtId across playback surfaces (Now Playing, fullscreen, mobile, mini). * fix(cover): decode PNG from Subsonic before WebP tier encode Enable `png` in the image crate — some servers return PNG cover art; failed decode left `.fetch-failed` and empty thumbs for those albums. * refactor(cover): consolidate cover id resolution and align tests Move resolveSubsonicSongCoverArtId helpers to src/cover/resolveCoverArtId.ts with resolvePlaybackTrackCoverArtId for player surfaces; co-locate tests; fix FullscreenPlayer expectations for albumId-first resolution. * docs: CHANGELOG and credits for PR #878 * fix(cover): keep per-track coverArt when distinct from song id Address PR #878 review (b): albumId only when coverArt is missing or echoes track id; pin case with unit test; comment isRawFsPath symmetry. * chore(cover): address PR #878 review nits (scope, tests, rename) Narrow asset scope to cover-cache dirs only; add diskSrcCache Windows-path tests; rename ArtistTopTrackCover; CHANGELOG symptom-first wording. * fix(cover): restore asset scope to app data dirs (Windows regression) $APPDATA/cover-cache/** did not match Tauri scope resolution — covers were blocked after load. Use $APPDATA/** and $APPLOCALDATA/** (no $DATA). * fix(cover): Windows asset URLs — restore DATA scope, path normalize Regression after review nits: dropped $DATA/** and strict isAssetProtocolUrl blocked valid http://asset.localhost URLs on Windows. Normalize C:/ paths before convertFileSrc; CoverArtImage/Hero hide broken img on load error. * fix(cover): disk peek fallbacks when cache folder id differs Small surfaces resolve albumId while cover-cache often stores WebP under track id or album.coverArt from the grid. Peek batch now tries legacy ids; playback scope resolves server index key by URL key, not UUID-only lookup. * fix(cover): Navidrome al-* vs mf-* disk id mismatch UI used mf-* coverArtId while library backfill only cached al-* folders. Prefer album id for display/peek when coverArt is mf-*; backfill now queues both distinct album_id and cover_art_id values. * fix(cover): mf→al disk peek when mf folder missing in cache Navidrome Subsonic often returns mf-* coverArtId while backfill only creates al-* folders. Peek mf first, then al-* from hints; load albumId from library when Subsonic omits it; ensure fallback uses al-* id. * feat(cover): CoverArtRef, segment disk layout, library-index backfill Normalize cover caching around stable entity ids from the local library and Navidrome fetch ids. Disk paths live in psysonic_core::cover_cache_layout (album/<entityId>/); UI uses CoverArtRef with cacheEntityId + fetchCoverArtId. - Remove SQLite/mf peek helpers (diskPeekIds, peekCoverOnDisk, mergeDiskIdHints) - Backfill reads album/artist rows from library SQLite (bare Navidrome ids ok) - Use stored cover_art_id for HTTP; per-disc dirs only when discs differ - Migrate call sites to albumCoverRef / albumCoverRefForPlayback * feat(cover): central CoverEntry resolver (artist, album, track) Add resolveEntry.ts and Rust CoverEntry helpers as the single source of truth for cache_entity_id vs fetch_cover_art_id. ref.ts delegates to them; resolveCoverArtId becomes a thin compatibility shim. * feat(cover): resolve cover entries from local library index Add library_resolve_cover_entry IPC and cover_resolve.rs so album, artist, and track covers use SQLite cover_art_id + disc detection. TypeScript helpers in resolveEntryLibrary.ts prefer the index over live API fields when rows exist. * feat(cover): library-first hooks for grids and playback UI Add useAlbumCoverRef, useArtistCoverRef, useTrackCoverRef, and usePlaybackTrackCoverRef — sync fallback then SQLite index upgrade. Wire album/artist cards, album header, song card, and all player surfaces to resolve covers from the local library when indexed. * feat(cover): complete library-first migration across all UI surfaces Add Album/Artist/TrackCoverArtImage, useLibraryCoverPrefetch, and batch resolve helpers. Migrate grids, search, home, playback sidecars, warm peek, playlists, and share flows to hooks that upgrade from SQLite. Backfill normalizes album rows through cover_resolve; document paths in COVER_PATHS.md. Radio remains a deliberate non-library exception. * fix(cover): stop render loop from unstable serverScope in library hooks Default param `{ kind: 'active' }` created a new object every render, so every grid cell re-ran library_resolve IPC and setState in a loop. Use COVER_SCOPE_ACTIVE singleton, coverScopeKey deps, and guarded sync updates. * chore(cover): remove COVER_PATHS.md from app tree (lives in workdocs) Audit doc is team spec — see workdocs 2026-05-cover-art-pipeline/cover-paths-audit.md. * fix(cover): unstick library backfill after route changes (PR #870 regression) useCoverNavigationPriority cleanup called beginNavigation instead of end, leaking navigationHoldDepth so ui_priority_hold never released and backfill never downloaded. Also skip disk check after cover_resolve normalization. * fix(cover): segment progress, cap backfill CPU, include artists in catalog Progress and disk size now scan album/ and artist/ segments (canonical 800.webp). Prune legacy flat server/al-* dirs on startup and backfill pass. Backfill: max 2 concurrent ensures; JPEG decode and WebP encode run on the blocking pool behind a shared 2-permit semaphore so Tokio workers stay cool. Artists were missing because the catalog only read the empty artist table; add distinct artist_id from track and album rows. Paginate with a composite (kind, id) cursor so album and artist rows are not skipped. * fix(cover): drop legacy prune; backfill per-disc and artist catalog Remove prune_legacy_* and cover_cache_catalog_entry — layout is only cover_dir (album|artist segments); stale flat dirs clear on LAYOUT_STAMP change. Backfill: artists from track/album artist_id; expand albums to per-CD mf-* slots when discs differ; fix resolve_album_cover_entry when album row is missing. * fix(cover): reduce library IPC storms and fix multi-disc player art Skip per-row library_resolve on live search and artist album grids; warm grids from API coverArt after mount instead of blocking layout. Dedupe and cap concurrent library_resolve calls. Restore per-disc cache keys in the player and queue when track mf-* art differs from the album bucket. * fix(cover): skip library resolve on advanced and full search rows Use API coverArt for album/artist rails and lazy viewport artwork so result pages do not fire hundreds of library_resolve IPC calls at once. * fix(cover): default libraryResolve off for browse grids and rails Skip per-card library_resolve on album/artist/song browse UI by default; keep it on album/artist headers, playback queue rows, and orbit approval. * fix(cover): split UI/backfill CPU pools and restore mainstage hero carousel Library backfill no longer shares the 2-permit JPEG/WebP semaphore with visible cover ensures. Hero initializes albums from props, re-binds scroll visibility after mount, updates backdrop on slide change, and uses library resolve for correct cover art on the banner. * fix(analysis): resume full-library scan after candidates phase Reset the SQL cursor when entering full-library mode so tracks with partial analysis are not skipped. Tighten TS backfill completion and CPU queue watermarking; align cover-cache key tests with album-scoped storage keys. * fix(library): remove useless map_err in cover_resolve (clippy) CI treats clippy::useless-conversion as error on rusqlite optional() chains. * fix(cover): satisfy clippy on cover_cache_ensure IPC args Pass CoverCacheEnsureArgs as a single Tauri parameter instead of nine positional fields; align frontend invoke payload with { args }.	2026-05-28 03:15:08 +03:00
cucadmuh	418b25914a	feat(cover): unify cover pipeline and stabilize mainstage/now-playing (#870) ... * chore(cover): scaffold cover module and rust cover_cache stub Wave 0: src/cover/ skeleton per contracts.md §12, stub IPC commands in cover_cache/mod.rs (no-op returns until phase B). * feat(cover): add unified cover module and tier resolver (phase A) Wave 1A: tiers, storage keys, resolveJs with cold/sibling races, useCoverArt, CoverArtImage, layoutSizes, playback scope helpers, coverSiblings tier ladder, deprecated shims on subsonicStreamUrl. * feat(cover): rust disk cache and tier-ready events (phase B) Wave 1B: cover_cache module with WebP tier encode, HTTP canonical 800 fetch, cover_cache_* commands, cover:tier-ready / cover:evicted events, disk layout tests. * feat(cover): prefetch hook, tier-ready handoff, library backfill IPC (phase B/C) Wave 2: useCoverArtPrefetch, cover:tier-ready/evicted bridge, one-time IDB cover key clear, prefetch registry drain, MainApp wiring. * feat(cover): migrate dense grids to CoverArtImage and prefetch (phase D) Wave 3A: dense surfaces use layout-native displayCssPx, surface=dense, coverPrefetchRegister on Home/Albums/search; AlbumCard cell width from grid. * feat(cover): migrate sparse surfaces and integrations (phase E sparse) Wave 3B: sparse CoverArtImage/useCoverArt, lightbox tier 2000, ArtistHeroCover, MPRIS/Discord/export integrations, playback chrome and detail heroes. * feat(cover): revalidation scheduler and disk pressure gate (phase E+) Wave 4: coverCacheMaxMb settings (en/ru), StorageTab disk usage, cover_cache_configure, useCoverRevalidateScheduler, playbackServer uses cover fetchUrl; pressure watermarks. * docs: CHANGELOG and credits for cover art pipeline PR #869 * fix(cover): stop webview getCoverArt storm on dense grids (429) Dense surfaces no longer put rotating getCoverArt URLs in img src; load disk via Rust ensure + convertFileSrc. Tier-ready notifies listeners instead of invalidating IDB. Throttle background prefetch and cap Home registry. * fix(cover): omit empty img src until cover URL is ready React 19 warns on src=""; CoverArtImage uses undefined until disk/IDB resolves; queue current track shows placeholder when src is still empty. * fix(cover): disk cache by host index key, parallel ensure, asset protocol Bind cover storage to serverIndexKey (library host), rename cover IPC/events, fix REST base URL and Tauri flat args, enable protocol-asset for disk paths, add prioritized ensure queue, and wipe legacy profile-UUID cache once. Limit Vite dep scan to index.html so research/target HTML is ignored. * fix(cover): WebP tiers, disk peek, home cache, asset URLs for mainstage Encode lossy WebP (~82), write only missing tiers, library cover backfill, and cover_cache_peek_batch for fast paint from disk. diskSrcCache + CSP asset protocol; no IDB fallback when server is up. Session Home feed cache with warm peek on return; BecauseYouLike deduped cover hook and high prefetch. * feat(cover): per-server cache strategy and native library backfill Move cover disk cache settings to Offline & cache with Lazy/Aggressive per server, per-server clear, and no size cap. Run full-catalog backfill on the Rust runtime (sync-idle wake, bounded HTTP, bulk 800px writes without flooding the webview). Drop global prefetch limits from auth store and waveform clear from the offline storage block. * fix(build): CSP connect-src for Subsonic API; quieter prod nix build Prod webview blocked axios ping after cover CSP (missing connect-src). Drop cargo tauri -v in flake build, raise Vite chunk limit, ignore tsbuildinfo. * fix(cover): complete WebP ladder in library bulk backfill Aggressive backfill now writes all derived tiers (128–800), skips IDs only when the full ladder exists (not 800 alone), avoids fetch-failed markers on bulk HTTP errors, and stops the pass when the active server changes. * fix(cover,home): navigation-priority backfill and Because You Like UX Pause library cover backfill while navigating; split peek/ensure traffic so grids and rails win over bulk work. Disk src lookup, grid warm hooks, and non-blocking mainstage prime for faster visible covers. Because You Like: session snapshot, staggered horizontal skeleton row, text hidden until cover is ready, and layout aligned with loaded cards. * feat(random-albums,library): local-first album fetch + cover art pipeline Random Albums теперь запрашивает локальный SQLite-индекс (ORDER BY RANDOM() LIMIT N) вместо сетевого запроса к серверу. При готовом индексе спиннер исчезает практически мгновенно; сеть используется только как фолбэк. - advanced_search.rs: добавляет `("random", _) => RANDOM()` в allowlist сортировок - browseTextSearch.ts: runLocalRandomAlbums — SQLite-рандом для Albums - RandomAlbums.tsx: doFetchRandomAlbums local-first для обоих путей (без жанра и с жанром через runLocalAlbumsByGenres + JS-shuffle); speculative reserve прогревает следующий батч в фоне после каждого Refresh Также: обновление пайплайна обложек (coverTraffic, peekQueue, ensureQueue, diskSrcLookup, warmDiskPeek, prefetchRegistry, useCoverArt, useWarmGridCovers, useCoverNavigationPriority, resolveIntersectionScrollRoot и сопутствующие компоненты/хуки). * fix(random-albums): prevent double-load on Zustand rehydration useEffect([selectedGenres, load]) fired twice on every visit: first with default store values, then again ~50 ms later when Zustand rehydrated mixMinRatingFilterEnabled/minAlbum/minArtist from localStorage. Previously this was invisible because the first network fetch took ~1.5 s, so loadingRef.current was still true on the second fire. With the new local-first SQLite path the first load completes in ~50 ms, leaving the guard cleared before rehydration triggers a second random batch. Fix: ref-pattern — keep loadRef.current fresh on every render, effect depends only on selectedGenres. Manual Refresh and genre-filter changes still call the latest closure correctly. * fix(random-albums): stop warmCoverDiskSrcBatch in fillReserve from causing visual flash fillReserve вызывал warmCoverDiskSrcBatch для обложек резервного батча, что вызывало bumpDiskSrcCache() для каждой новой обложки (~30+ вызовов). Это будило всех подписчиков useCoverArt на текущей странице, провоцируя видимую перерисовку примерно через ~1.5 с после загрузки (когда filterAlbumsByMixRatings делает сетевые запросы к рейтингам артистов). - fillReserve: убран warmCoverDiskSrcBatch — обложки прогреваются лениво при consume резерва через primeAlbumCoversForDisplay - reserve-путь в load(): добавлен primeAlbumCoversForDisplay перед setAlbums (аналогично non-reserve пути; при уже прогретом кэше — мгновенно) * feat(because-you-like): reserve-first pattern — instant display on return visits Каждый визит на Mainstage после первого теперь отдаёт готовую заготовку мгновенно, вместо spinner → сетевые запросы → контент. Архитектура: - resolvePicks / fetchBecauseYouLike вынесены на уровень модуля (выход из замыкания useEffect); читают текущий localStorage, возвращают { anchor, recs, nextAnchorHistory, nextPicksHistory } - fillBecauseReserve — fire-and-forget фоновая функция: запускается сразу после отображения результата, кладёт следующий батч в _becauseReserve. Covers намеренно не прогреваются (bumpDiskSrcCache на текущей странице не нужен); они прогреваются через primeAlbumCoversForDisplay при consume. - useLayoutEffect: если reserve готов — не сбрасывает стейт в skeleton (контент появляется без мигания) - useEffect: reserve-first path — consume → primeCovers → setState → fill; full-fetch path сохранён как fallback при первом визите или промахе Поведение: - Визит 1: full fetch (как раньше) → показ → fillReserve R1 - Визит 2+: consume R1 → мгновенный показ → fillReserve R2 - При сетевом сбое: restore из session cache (как раньше) * fix(because-you-like): initialise state from reserve — no skeleton flash on remount При ремаунте компонент стартовал с refreshing=true/anchor=null/recs=[] и показывал skeleton на один тик до того как useEffect отработает. Теперь useState() использует lazy initializers, которые читают _becauseReserve прямо в первом рендере: если reserve валиден — state сразу refreshing=false, anchor=X, recs=[...] и skeleton не показывается вообще. Covers уже в diskSrcCache (из предыдущего показа) и появляются без дополнительных запросов. useLayoutEffect упрощён: вызывает hasValidReserve() и сбрасывает в skeleton только если reserve отсутствует (для случая navigation без ремаунта). * fix(because-you-like): apply reserve in useLayoutEffect to handle async pool arrival Lazy initializers не могли применить reserve при первом рендере, потому что mostPlayed/recentlyPlayed/starred приходят из Home.tsx асинхронно — pool=[] на первом рендере, poolKey не совпадает с reserve. useLayoutEffect теперь активно ставит стейт из reserve (а не просто не сбрасывает): когда pool обновляется до реальных данных, useLayoutEffect срабатывает синхронно до paint, проверяет reserve и сразу применяет anchor/recs/refreshing=false. При отсутствии reserve — сбрасывает в skeleton как прежде. * fix(because-you-like): reserve > cache > skeleton — eliminate skeleton flash on mount Корневая причина: Home.tsx загружает mostPlayed асинхронно через useEffect, поэтому на первом рендере pool=[], poolKey=''. Reserve хранится с реальным poolKey → mismatch → lazy initializers запускали skeleton. Теперь двухуровневый fallback без зависимости от poolKey: 1. reserve (serverId + poolKey совпадают) → мгновенный новый батч 2. becauseYouLikeCache (только serverId) → stale-while-revalidate, контент доступен сразу с mount, обновляется тихо в фоне 3. skeleton → только при полном отсутствии данных (первый визит) Применяется одинаково в lazy useState initializers, useLayoutEffect и full-fetch path useEffect (не сбрасывать в skeleton пока есть cached контент). * fix(because-you-like): key reserve by serverId only; guard useEffect on empty pool Проблема: reserve хранился с poolKey, но на первом рендере pool=[] → poolKey='' → mismatch → показывался кэш (предыдущий набор) ~500ms пока Home.tsx не загружал mostPlayed. Исправления: - BecauseReserve: убран poolKey — reserve валиден для любого pool-состояния на том же сервере. Pool (топ-артисты) меняется медленно; один раз показать reserve с чуть устаревшим anchor лучше чем показывать предыдущий набор 500ms - hasValidReserve: проверяет только serverId - fillBecauseReserve: убран poolKey из сигнатуры и хранилища - useEffect: guard pool.length === 0 → возврат без fetch/consume; effect перезапустится когда pool заполнится (реальные deps изменятся) → reserve применяется из useLayoutEffect ещё до pool, без стале-флэша Итоговый порядок: reserve (instant, serverId) > cache (stale-while-revalidate) > skeleton (только первый визит) * fix(home): remove mix-rating deps from feed useEffect — prevent Zustand rehydration double-fetch Корень: useAuthStore(mixMinRatingFilterEnabled/Album/Artist) были в deps useEffect. Zustand persist реhydrates асинхронно — сначала activeServerId, потом mix-rating значения. Это вызывало двойной запуск эффекта: - Первый запуск: homeFeedCache hit → показывает набор предыдущего просмотра - Второй запуск (после rehydration): cache miss или повторный fetch с реальными mix-настройками → ~500ms → новый набор Итог: Hero, AlbumRow, BecauseYouLikeRail показывали предыдущий набор первые ~500ms при каждом возврате на Mainstage. Fix: убраны mixMinRatingFilterEnabled/Album/Artist из deps. getMixMinRatingsConfigFromAuth() читается внутри эффекта через getState() — всегда актуальные значения без пересоздания замыкания. Mix-настройки по-прежнему применяются при fetch, но не вызывают двойной запуск при rehydration. * feat(home): local-first discover songs via SQLite ORDER BY RANDOM() Добавлена runLocalRandomSongs (аналог runLocalRandomAlbums для треков) в browseTextSearch.ts — использует libraryAdvancedSearch с sort random, field уже поддерживается Rust-кодом через wildcarded ("random", _) ветку. В Home.tsx: discoverSongs теперь сначала пробует локальный индекс, и только при недоступности (индекс не готов, ошибка) падает обратно на getRandomSongs.view. Ускоряет первую загрузку Mainstage — треки берутся из SSD вместо сети. * fix(home): pre-populate state from cache at mount — eliminate empty-state flash on return visits Причина: Home.tsx размонтируется при навигации. При возврате первый рендер всегда с пустыми массивами (heroAlbums=[], mostPlayed=[] и т.д.), потом useEffect читает homeFeedCache и заполняет state. Даже один кадр с пустым состоянием вызывает перерисовку Hero и BecauseYouLikeRail (pool=[]). Решение: getInitialHomeFeed() читает homeFeedCache синхронно через useAuthStore.getState() (не hook) в lazy useState initializers. К моменту повторного визита store уже rehydrated — все state получают кэшированные данные до первого рендера. Дополнительно: wasPrePopulated предотвращает повторный applyFeedSnapshot в useEffect когда state уже заполнен — иначе новые ссылки на массивы вызывали бы ненужные ре-рендеры дочерних компонентов с теми же данными. * fix(mainstage): keep refresh without return flicker Keep Home and Because You Like visually stable during a single visit while still refreshing data for the next re-enter. Improve mainstage cover warmup by ensuring and pre-decoding above-the-fold artwork so hero and top rails appear instantly after navigation. * fix(mainstage): stabilize because rail and hero background framing Measure Because You Like layout before first paint to avoid width snap flicker, and render hero background as centered cover-fit images so the frame no longer jumps from top to middle on mount. * fix(now-playing): prewarm track data and prevent stale carry-over Warm Now Playing fetch caches and playback cover art on track change so entering the page no longer waits on first-load requests. Gate key-based sections (top songs, tour, Last.fm) by the active track/artist keys to avoid briefly rendering values from the previous track. * fix(cover,test): refresh playback scope and default tauri cover mocks Recompute playback cover scope when queue/server context changes so now-playing art resolves against the correct server after handoffs. Add default cover-cache invoke handlers to the shared Tauri test harness to prevent unhandled rejections in suites that mount cover-aware UI. * fix(cover,now-playing,test): align prewarm scopes and tighten tauri mocks Make cover-cache invoke defaults opt-in for tests, align radio prewarm scope with active rendering scope, and add targeted hook tests for prewarm + playback-scope reactivity. Also harden Rust cover URL building to avoid panic on malformed base URLs. * test(cover): hoist mocked useCoverArt and clean EOF whitespace Fix the new playback-scope hook test to use a hoisted vi.mock-safe stub and keep branch-wide diff checks clean by removing an accidental trailing blank line. * fix(cover): align playback ensure auth and harden backfill retry flow Use playback-server credentials for playback-scoped cover ensures, persist fetch-failed markers for bulk library backfill failures, and avoid advancing backfill cursor when UI-priority hold interrupts a batch. * fix(ci): resolve clippy lint and update frontend node runtime Move fetch helper before the test module to satisfy clippy's items-after-test-module rule, and modernize frontend CI to setup-node v6 with lts/* instead of pinned Node 20. * chore(settings): simplify cover and analytics strategy copy Move strategy summaries below tables, simplify Lazy/Aggressive wording, keep analytics warning always visible, and localize Russian texts to plain language without technical jargon.	2026-05-26 19:35:08 +03:00