Add episodeCacheMode/count/days preferences (default: date, 60 days).
Apply the bound when reading instead of writing, so a preference change
takes effect without a refetch; the full parse cache stays intact so
fetch-more can page beyond the bound. Thread the window through
load/saveFeedsToFile and update tests and task docs.
loadMoreEpisodesForFeed's hot-cache path ran fully synchronously (no
await between getFeed and setFeeds), so the isLoadingMore spinner never
painted and keyboard input froze through every feed in a loadMoreAllFeeds
batch. Add await yieldToUI() before the setFeeds so the renderer gets a
macrotask turn to paint and process input between feeds.
Drop the "Loading…/Refreshing…/Downloading N" label text from the global
activity indicator — render the braille spinner only, per request.
Persisted feeds keep only episodes from the last 30 days (plus completed
downloads); older episodes live in volatile memory and survive refreshes
via union merge, with per-feed in-memory caches capped at 500. Refresh
batches run at FETCH_CONCURRENCY=4 with per-feed incremental apply (no
Promise.all barrier), config.json writes are trailing-edge debounced
(250ms, immediate flushPendingSave for unsubscribes), and cold
fetch-more refetches abort at FETCH_TIMEOUT_MS. A shared activity store
powers a global top-right indicator covering refresh, fetch-more,
subscribe, search, and downloads.
Also includes the in-flight incremental RSS parsing (chunked with
event-loop yields) and refresh spinner work this tree already carried.
ensureDecodeAround let a running decode pass close ANY forward gap in
place — at 4x pacing, skipping 30 min ahead meant ~7.5 min for the
frontier to arrive: bars held their last frame indefinitely. Now a gap
beyond 15s restarts the pass at the seek target (network coverage there
in ~1.6s, verified); smaller gaps close in place (cheaper than a
reconnect + range request). Seek-key holds are debounced (400ms) so
rapid re-seeks don't reconnect-spam the stream's server, and pending
seek-decode cancels on pause/stop so no ffmpeg restarts while paused.
Also fixes a pre-existing config-write race that intermittently failed
visualizer-toggle.test.ts: updateConfig re-resolved the config path and
re-read the patch state when the deferred write-chain drained, so a
queued save could land in a directory XDG_CONFIG_HOME had since been
pointed at (or carry state mutated after queueing). Path and patch
snapshot are now captured eagerly at call time.
Two fragility points, rebuilt at the root:
Playback: one resident mpv daemon (--idle --keep-open) with a persistent
IPC connection and observe_property state instead of spawn-per-episode and
connect-per-poll. Play/pause/seek are sub-ms commands; time-pos pushes at
~20Hz; external pauses arrive as events. Boot session restore preloads the
episode paused (loadfile + paused time-pos seek, since mpv defers --start
stream work until playback) so first Play is a ~400ms unpause instead of a
cold 4.3s open+seek. Load ops are mutex-serialized so a raced preload
cannot clobber an in-flight play.
Data throttling: mpv demuxer cache capped (cache-secs=90, max-bytes=40MiB)
so a paused preload no longer races to its 150MiB default (measured
45.7MB/12s); decoder paced at 4x realtime instead of 84x so playback start
isn't starved by the visualizer ripping the whole episode.
Visualization: replaced the paced-ring reader (AudioStreamReader) with a
position-indexed PCM cache (audio-pcm-cache). ffmpeg fills a cache indexed
by absolute playback time; reads at the player position are always exact.
Pause freezes the render loop, resume re-arms it — no coverage guessing,
no clamped-buffer freeze (the pause->broken-waveform->freeze bug). Seeks
and speed changes need no pipeline restarts; uncovered reads return empty
and the last frame holds.
Cover art: persistent per-URL disk cache under XDG cache dir; play() no
longer awaits a curl subprocess (up to 8s). Cache hit = one stat; misses
apply late via mpv video-add.
Test suite: 161 pass. New tests pin the position-index contract (sample-
exact window reads, hold-on-uncovered, pause-keeps-cache, seek segments),
the daemon contract (play/pause/resume/seek/stop, preload fast path,
EOF->replay), and cover cache/single-flight/404.
The waveform's ffmpeg decode + cavacore FFT pipeline lived inside
RealtimeWaveform, so switching away from the Player tab unmounted it and
killed the pipeline instantly — respawning ffmpeg on every return.
Move the pipeline into a module-level store (stores/visualizer.ts) that
outlives the page:
- losing Player-tab focus keeps the pipeline warm for
VISUALIZER_UNLOAD_DELAY_MS (30s), then tears it down (kills ffmpeg,
destroys the cava plan); regaining focus within the delay resumes the
warm pipeline with no restart churn; after an unload it restarts from
the current playback position.
- playback signals move to utils/audio-signals.ts (module-level, no
useAudio() owner needed) so the store reacts to play/pause/seek/speed
while no Player page is mounted; useAudio re-imports them.
- render a braille spinner as the loading state for the visualizer
(first play / after unload); stale bars stay on screen during warmup
restarts so the waveform never blanks out for the network-bound cold
start.
- seed the smooth position clock at pipeline start: with the position
still frozen at 0 while mpv opens the stream, the reader sampled a
1-sample window that could never fill, starving the bars until mpv's
first poll.
Pins the store contract in tests/visualizer-store.test.ts: loading→bars,
warm resume without restart, 30s unload, and bars while position is
frozen at 0.
mpv can pause or resume OUTSIDE PodTUI — system sleep/lock, AirPod
removal, device swap, OS media keys, the Now Playing center. The poll
previously only reflected commands PodTUI sent, so the UI stayed stuck
on "playing" (or "paused") with a frozen position clock.
The poll now reads mpv's live pause state each tick: an external pause
reconciles the UI to paused (persisting progress, syncing media
controls) while keeping the poll armed; an external resume brings the
UI back to playing. A paused player is polled at a throttled rate
(PAUSE_WATCH_TICKS) so the watch costs ~1 IPC read per second instead
of hammering mpv; a dead process (track end / crash) finalizes the
track.
Store the playback volume in app settings (config.json) whenever it
changes and re-apply the previous session's level at boot, instead of
always starting at the old 70% fallback.
- AppSettings gains volume (default 1 = 100%); both default-settings
copies and the volume signal default are raised from 0.7 to 1.
- doSetVolume persists via the app store (mirrors playbackSpeed).
- The boot sync awaits the app store's async config load (new
whenReady()) so a persisted level is applied even when settings load
finishes after useAudio mounts.
- tests/volume-persistence.test.ts: default, clamp, and cross-session
reuse (fresh module instance simulates the next launch).
Reload the episode that was loaded in the player when the previous run
ended (persisted on play/load and synchronously at exit) into the Player
tab paused at its saved position — never autostarted. Episodes at or
above 98% completion are skipped, as are empty-player and unsubscribed
episodes.
- useAudio gains load(episode) (sets currentEpisode/position/Now Playing
without starting the backend) and restoreLastSession(), triggered once
at boot and serialized through a chain so a late-finishing boot restore
can't clobber later state.
- togglePlayback branches on a startedPlayback flag: a restored episode
starts the backend from the saved position; a paused one resumes.
- stop() clears the marker; the exit teardown writes it synchronously
(process.exit bypasses async writes).
- feed/progress stores expose whenReady() so restore waits for the async
boot loads; feed store gains findEpisode().
- app-persistence serializes last-player marker writes and exposes
waitForLastPlayerWrite() for deterministic tests.
- tests/restore-session.test.ts: real modules + local RSS feed server;
the real useAudio is imported via a ?restore-test query suffix to
bypass the suite's mock.module leak across shared bun workers.
Episode search results gain d (download), D (delete), x (unsubscribe)
and enter (play for subscribed shows); downloads of unsubscribed shows
are recorded with the show's metadata under a deterministic synthetic
feed id and listed under an "Unsubscribed Show Downloads" section in
My Shows and the settings Download Manager. Classified at render time
by feed id or feed URL, so subscribing re-classifies the downloads and
unsubscribing purges them.
Self-rescheduling refresh timer (default 30 min, configurable via a
Preferences item, re-read on every tick, skips in-flight refreshes).
fetchEpisodes returns null on network failure/timeout so a failed
refresh can never wipe a feed's episodes (addFeed/refreshFeed/
refreshAllFeeds all treat null as unchanged); feeds still refresh on
launch.
Regression coverage for the search input's focus flag: it must track
the renderable's REAL focus (useInputFocusNav FOCUSED/BLURRED), not a
flag that outlives it. Clicking off the input drops inputFocused and
keyboard control resumes; s re-enters typing; Esc defocuses; clicking
the input refocuses.
Documents the observed full-suite CPU-contention flake in the header:
both tests occasionally time out in openSearch under suite load while
passing reliably in isolation.
Podcast Index (api.podcastindex.org) ships as a disabled, key-less source
and is only consulted as a fallback when primary search results are fewer
than 3 — never on the hot path, never when disabled or credential-less.
A failed fallback leaves primary results intact.
Credentials are user-supplied: enabling the source pops a dialog that
asks for the free key+secret, prefilled masked (first 3 chars + "...")
when already stored; toggling off never clears them. Secrets prefer the
macOS keychain (security CLI, encrypted at rest) with a plaintext
config.json fallback when the keychain is unavailable; sources carry only
a hasCredentials/credentialStorage marker, and legacy plaintext keys in
existing configs are migrated on load.
Auth follows the documented scheme: X-Auth-Key, X-Auth-Date (epoch) and
Authorization = sha1(key + secret + date). Dead feeds are filtered, feed
URLs are used directly, and episode-scope search is a no-op (no endpoint).
Shows that left Apple Podcasts (e.g. Daily Wire's in 2021) come back from
the iTunes Search API as metadata-only stub records with feedUrl null.
mapItunesResult dropped them, so The Ben Shapiro Show — the #1 hit for
'ben shapiro' — never appeared in search while sibling shows did.
- Keep feedUrl-less results (feedUrl "" + directoryUrl pointing at the
Apple page) so delisted shows stay findable.
- Resolve the real feed from the Apple page at subscribe time
(itunes-feed-resolver: anchor on the collection's adamId, forward-scan
for the embedded feedUrl; Apple serves page variants where the
showOffer block sits thousands of chars after the adamId).
- addFeed refuses feedless stubs whose feed can't be resolved instead of
adding a broken feed; SearchPage surfaces the failure via toast.
- Tests: stub mapping, extractor variants, and an end-to-end subscribe
over a local HTTP server.
Mirrors the Feed tab's '[Fetch More]' row inside a drilled show's episode
list (My Shows depth 1): shows only while the show's cache holds episodes
beyond its loaded window, and advances just that show's window by 50 on
Enter (or automatically at the bottom in auto mode). Same fetchMoreMode
preference drives both behaviors. Adds a store-contract test for the
per-feed pagination path.
Feed and My Shows rows could grow to 4+ lines when a long title was
shrunk by the current pane: flexible text wrapped instead of
truncating, shifting every row below while scrolling. Add
wrapMode=none + truncate to flexible text and flexShrink=0 to
fixed-width cells so rows stay one line tall. Feed rows also move
the podcast name onto its own line. Adds a rendered-layout
regression test at 70 columns (35-col current pane).
Two fixes to refresh order stability (My Shows / Feed sort by
lastUpdated):
- A refresh that fetches identical episodes no longer bumps
lastUpdated (id-set comparison via sameEpisodes), so unchanged
feeds keep their position instead of reordering every cycle.
- refreshAllFeeds now fetches in parallel and applies ONE atomic
update instead of a per-feed setFeeds, which re-sorted the list
once per completion and made order flap until the batch finished.
Adds feed-refresh regression tests with mocked clock.
Change the parent|current|preview split from 1:2:2 (20/40/40) to
2:5:3 (20/50/30) so the focused list gets more room. 2-pane tabs
now give current the combined 80%. Updates ratio comments and the
PaneRow test expectations.
The decode head lagged the player by a constant ε (ffmpeg startup
latency) because ffmpeg paced at -readrate <speed> started behind mpv
and, advancing at the same rate, never caught up — bars were a few
seconds behind for the entire playback.
Add -readrate_initial_burst LEAD_SECONDS so ffmpeg emits 3s of audio
immediately on start, then paces at realtime after. The decode head
leads the player by a stable ~3s from the very first frame; read()
samples at the exact player position and always finds fresh samples.
Add a sustained render-loop test that simulates ~5s of real playback,
asserting ffmpeg stays alive, the decode head maintains a positive
lead, and read() returns full windows. Uses real wall-clock time
(documented exception) since ffmpeg's decode pacing can't be tested
deterministically.
MouseEvent coordinates are terminal-absolute, so the bar's seek handler
must subtract its own absolute left edge (renderable ref) instead of
treating the mouse x as bar-local. Clicking was off by the width of the
parent/Up pane (~20%) plus chrome. Also drop the empty played-text node
at position 0, which rendered a phantom space column and shifted the
drawn bar one char when playback started.
- ProgressBar: full-width played/remaining bar in the player pane, click-to-seek;
waveform no longer handles seeking or the played/future color split (pure visual)
- bars: 2 terminal rows per bar (16 levels) via barChars, partial block in the
top row so the column renders continuously
- fix bars maxing at audio start: disable cava autosens (silence gain-ramp),
pre-warm the malloc'd FFT window with zeros, skip partial FFT windows
- createBarScaler peak follower + power curve replaces cava autosens for
level-to-height mapping (src/utils/bar-mapping.ts, unit-tested)
- Current pane gets muted left/right borders only (no full box, no accent
ring); border colors are passed only when a border is requested, since
opentui flips borderless boxes to bordered when borderColor is supplied.
- Remove the Up / <current tab> / Detail titles above the panes; the current
pane's title now renders once, top-left in the parent column's header slot.
- Drop the parentLabel/previewLabel props from PaneRow and all callers.
- Remove the tab/depth indicator from the bottom-left of the status bar.
- Tests measure column widths from the border glyphs and assert the
left/right edges render muted regardless of focus.
Make parent|current|preview fully borderless: no scrollbox borders and no
accent border highlight on the current column. focused still gates
scroll-following but never surfaces a separator. Update tests to measure
column widths from the header-label row and assert no border glyphs render.
tsconfig used jsx:"preserve"+jsxImportSource, which trips a TS 5.9
internal crash ("Expected sourceFile.imports[0] to be the synthesized JSX
runtime import") so tsc could never run clean. Switch to jsx:"react-jsx"
with the package's real jsx-runtime types, and fix the real errors that
surfaced:
- delete dead src/components/Navigation.tsx (imported ./Tab that doesn't
exist; the component has no importers)
- PlaybackControls: fix relative path to @/utils/audio-player
- SourceBadge: drop dead module-level typeColor (bare 'theme')
- command palette: bind to the real 'command' keybind (:) instead of the
never-defined 'command_list' action (palette was unreachable)
- yazi-pane-row test: destroy() must return Promise<void> as typed
Also fold in the package.json lint fix (bun tsc --noEmit) and doc polish.
Move the keybind router out of Shell into a reusable createDispatcher
in @/utils/dispatch, covering tabs (digits 1-6, [/]), h/l depth
drill/pop + fixed-pane swipe, modes, audio transport, quit/help, and
command, delegating pane/list actions over the nav.action event bus.
Shell now renders the active page full-width with a bottom status bar
carrying the tab strip; the old sidebar pane and its leftover sidebar
action handling are removed. Adds tests covering the dispatcher's
routing behavior.
Convert Discover, Feed, MyShows, and Settings pages from bespoke 3-column
flexbox JSX to render entirely via the shared <YaziPaneRow> parent|current|
preview primitive. Each page now supplies parent/current/preview content
accessors plus labels; the pane row handles layout, borders, focus styling,
and the muted placeholder for the blank parent slot at depth 0.
Replace Solid's children() helper in YaziPaneRow with a local normalizeContent
that hands the raw accessor to a reactive
expression. children() flattens to a stable resolved-nodes array and does not
re-resolve on truthy<->truthy root swaps (e.g. depth switching a pane's root
between a list fragment and an editor), which would freeze the previous
subtree. The insert effect from the reactive expression disposes and
remounts on element-identity change instead.
Add tests/yazi-pages-depth.test.ts covering the nav-store depth-stack
contract the pages depend on (push/pop stack growth, parent-slot data model
across root→child→grandchild→pop transitions).
Add the shared parent | current | preview row primitive that implements
yazi's mgr.ratio = [1, 3, 3] contract via Yoga flexGrow, so every list
tab renders an identical, layout-stable shell regardless of terminal
size.
- New YaziPaneRow component: three bordered columns grow at 1/7 : 3/7
: 3/7 with flexBasis=0 (content can never stretch its slot). The
current column carries the accent focus ring when `focused` is
truthy; parent and preview stay muted. The blank parent keeps its
1/7 slot with a muted placeholder rather than collapsing.
- PANE_RATIO.current 4 -> 3 to match yazi's [1, 3, 3].
- Add render-based tests covering the 1:3:3 ratios (incl. null
parent/preview) and the focused/unfocused accent ring behavior.
- Configure the bun test preload for the solid JSX transform.
- Extract navigation state and logic into navigation-store.ts for
testability and separation of concerns
- NavigationContext now only wraps the store as a Solid provider
- Remove SIDEBAR_PANE from the model; depth-tabs have a single
focusable content pane (center/current column)
- Split LayerGraph and page imports into layer-graph.ts so the
navigation utils stay free of JSX (unit-testable)
- Update Shell keybind dispatch: remove sidebar pane handling,
simplify swipe to fixed-pane tabs only, clarify depth-tab h/l
- Add nav-model.test.ts covering depth stack, tab/pane focus, and
selection semantics
