The incremental/reconcile job docstrings still described last-write-wins and a
30-min cadence; update them to match the S2 behavior (Plex wins on divergence,
15-min incremental) so the "why" doesn't drift from the code.
Make Plex the source of truth for watch state:
- Write-back during playback: item_progress now pushes the resume position to
Plex on the periodic checkpoint too (not only the final flush), throttled to
one push per item every ~10s (in-memory, single-process rate limiter). The FE
now also fires an immediate (debounced) push on seek.
- Faster incremental pull: 30 → 15 min (plex_watch_sync_interval_min).
- Plex-authoritative pull: _pull_apply resolves a genuine divergence to Plex's
state instead of last-write-wins; the _same_state + synced_to_plex short-circuit
still prevents a freshly-pushed row from bouncing. Removed the now-unused
_LWW_TOLERANCE_S / _sift_ts.
Tests: 5 new in test_plex_watch_sync.py (Plex-wins resume/watched, no-bounce,
throttle final vs periodic).
The Plex player stalled ~2s a few seconds into every auto-advanced next
episode. Root cause: each auto-advance spins up a cold ffmpeg HLS remux
session, and for a fresh start (start_s==0) start_session returned
immediately with zero buffer — hls.js drained the first segment, then
rebuffered waiting for the still-warming ffmpeg to write the next one.
Once ffmpeg gets ahead (stream-copy is faster than realtime) it never
stalls again.
- stream.py: before reporting a session ready, wait for the first
_WARMUP_SEGMENTS (3) video segments to exist — the seg0 wait is now
unconditional (was start_s>0 only) and warm-up runs for all starts.
Best-effort and bounded: stop at the first segment not ready in time.
- PlexPlayer.tsx: a loading/buffering spinner. The existing overlay only
covered the INITIAL load (!ready); add a `buffering` state driven by the
video's waiting/stalled/seeking (set) and playing/canplay/seeked/pause
(clear) events so a mid-playback stall or in-flight seek reads as
"loading" instead of a frozen frame, and surface a fatal loadError even
once playback has started. Spinner uses the shared Loader2.
Merges roadmap E6·item1 (spinner while loading/seeking) into this fix,
since the server-side warm-up deliberately holds ready a beat longer.
Tests: 3 new in test_plex_stream.py (warm-up waits for the first few
segments; stops at the first missing one; no warm-up when seg0 never
arrives). Gate: backend pytest 460 (Docker lane), FE tsc/lint/knip/
prettier + vitest 483.
Fixes from several multi-agent deep-review passes over the E8 feed-filtering &
channel-manager epic (S1-S4):
- feed_years buckets by UTC year to match the UTC-pinned year-filter bounds
- tag suggestions: examples ∩ subscribed (drop tagged-then-unsubscribed channels
from the profile denominator); exclude hidden channels as suggestion targets
- channel manager: restore ghost chips + surface an error toast on a failed tag accept
- year filter: debounce the search term so the per-year facet doesn't refetch per keystroke
- channel tags: aria-label the "+N" overflow pills
- extract useTagSuggestions hook; add backend + hook tests (profile boundaries, cap,
language exclusion, unsubscribed/hidden handling, UTC bucketing, bulk-accept)
Bump VERSION to 0.73.0 + release notes.
With 2+ suggestions the narrow tags cell clipped the second ghost chip
mid-word right next to the edit "+", reading as the icon covering it. Route
the ghost chips through the same fit machinery the attached chips use: a
dedicated SuggestedChipsLane fits what the cell holds and collapses the rest
into a "+N" pill whose hover popover still one-tap accepts / dismisses each.
The lane is attached-XOR-ghost (suggestions are untagged-only, and accepting
one now drops the whole channel from the suggestion cache), so each lane owns
the full cell width instead of clipping the other.
Learn, per user tag, which auto TOPIC tags are characteristic of the channels
you've tagged it, then suggest that tag for still-untagged subscribed channels
whose topics cover enough of the profile. Language auto-tags are excluded from
the signal — near-universal within a language, they flooded every channel.
New GET /api/tags/suggestions (read-only, deterministic) backed by
app/sync/tagsuggest.py. Surfaced in the Channel Manager as dashed "ghost" chips
in the tags cell — one tap accepts (reuses the attach-tag endpoint), × dismisses
for the session — plus a header badge and a bulk "Accept N confident" (the
full-profile matches) behind a confirm. No silent writes (D4 = suggest-and-confirm).
Also reflow a pre-existing prettier drift in dateFilter.ts from E8·S2.
Add a Year filter to the feed rail: quick chips for recent years plus a
'More…' popover listing all years with per-year video counts, sortable by
year or count. Selecting a year narrows the feed to that whole calendar
year (a published_after/before range applied in the api layer) and is
mutually exclusive with the Upload-date control.
Backend: new /api/feed/years endpoint + feed_years() — one grouped query
that honours every filter EXCEPT the publish-date narrowing (dropped), so
the year list stays a live drill-down and switching years never zeroes the
others. FE keys the list on the date-stripped filters, so picking a year
re-queries the feed but not the list.
Wires year through FeedFilters, the filterParams/URL (?year) round-trip,
the active-filter chip row, and the panel layout. i18n HU+EN.
Tests: feed_years count query (date narrowing ignored, other filters
honoured); dateFilter year-clearing; urlState year round-trip.
The channel manager's tag-filter row wrapped the trailing "…" catalog-actions
menu onto a second line once the chips + Clear outgrew the width. Rework it as
one non-wrapping row: the TAGS editor trigger (left), a single-line chip lane
that collapses overflow into a "+N" pill whose popover holds the rest (still
toggleable), then Clear and the "…" menu pinned right — so "…" always stays on
the first line.
New TagFilterChips reuses the shared measure core (useChipOverflowFit +
fitChipsWithPill), the same compression the feed's applied-filter row uses.
Filter to channels you haven't personally tagged in the channel manager
(no user-owned ChannelTag; auto language/topic tags don't count):
- Feed: a "No tag" sentinel in the "Your tags" sidebar combobox (exclusive
with real tag picks) + a removable "Untagged" chip; new `untagged` bool
feed param / `notag` URL key, persisted + coerced on every filter path.
- Channel manager: a virtual "No tag" chip in the tag-chip row filters the
table to untagged channels (client-side, shared NO_TAG_SENTINEL).
Backend `untagged` is a plain channel-level narrowing, so feed_summary keeps
it applied in the base pass — the count/facet lockstep holds with no mirror.
Note: item E (per-tag counts) and item D (Any/All toggle) were already
shipped; S1 delivers only item B.
UX-8 layerStack: add an integration test wired through the real useLayer
hook (popover document-listener + stamp) and a PlayerModal-shaped window
listener, driven by a real dispatched keydown. The popover close is wrapped
in flushSync so its layer is popped synchronously between the two listeners
-- the production mid-dispatch race -- making the escapeHandled stamp the
only thing that keeps the player from also closing. Mutation-checked:
removing the guard fails exactly this case.
SavedViews: cover feedCanSeedDefault's "yield to a bookmarked filter URL"
branch (a filter-bearing search string -> no seed) and the per-source
seededRef guard against the 60s refetch (a changed-list refetch re-runs the
seed effect; the guard blocks re-seed). The refetch test waits for the
changed list to reach the component so the effect has its chance to wrongly
re-fire. Both mutation-checked.
Frontend gate green: 464 vitest (+4). Backend untouched.
The S4 Play-all wrapper had no test — a regression inverting the empty-queue
guard or dropping the notice would re-introduce the silent no-op it was added
to fix, and stay green. Adds a PlexCollectionView DOM test (spied api +
notify) asserting: a leaf queues as itself, a show expands to its episodes in
season->episode order, and an empty expansion surfaces plex.collection.playEmpty
instead of no-op'ing.
Extract the show->ordered-episode-keys flatten into one pure, unit-tested
helper (lib/plexQueue.ts) and use it at all three sites — the collection
Play-all (S4) and the show/season play-through pages in plexViews — so the
season-then-episode queue-order invariant lives in one place instead of three
hand-rolled copies. Behaviour-identical (the helper's ?? [] is a safe superset
of the collection site's previously non-null read).
Also: a show-collection whose shows expand to zero episodes had an enabled
Play-all that silently no-op'd — the same 'does nothing' state S4 removed for
the empty-queue case. Surface an honest info notice (plex.collection.playEmpty,
HU+EN) instead.
The S2 seed effect ran on the saved-views query's first resolve, but that
query is ungated and — since SavedViewsProvider mounts above <App>, which pins
the tab-account only after /api/me — resolves before the account is known on a
fresh tab. It then evaluated canSeedDefault() against a null account (which
reports 'stored, do not seed') and burned its one-shot guard, so the starred
default was never applied until a reload: exactly the fresh-tab miss S2 set out
to fix. Gate the effect on the reactive account id so it defers instead of
consuming the guard, and re-fires when the account pins. Covers feed + plex.
Adds a dom test for the defer-then-seed-on-pin path the old test masked by
pre-pinning the account.
A collection's Play-all/Shuffle queue held only playable leaves
(movies/episodes), filtering out shows — so on a show-collection the queue
was empty and Play-all was hidden/dead. Build the queue by expanding each
show member into its episodes (season then episode order), concatenated in
member order; leaves still queue as themselves. Shows are fetched lazily on
click (parallel plexShow fetches), not on page load. Gate Play-all/Shuffle
and the poster on "has any playable member" (leaves OR shows), and disable
them while the queue builds. Scope: option A — no member manual-order
migration. Adds plex.collection.playFailed (HU/EN).
ActiveFilterChips (feed applied-filter row) and ChannelTagsControl both
hand-rolled the same single-line chip-fit measurer: a hidden mirror + a
layout-effect + ResizeObserver reading chip right edges, the real rem gap,
the pill width and the lane width. The policy half was already extracted
(decideChipLayout); the DOM-measurement half was still duplicated.
Promote that half to useChipOverflowFit(rowRef, chipsRef, pillRef, deps,
decide, initial): it owns the effect + observer + initial measure and reads
the common geometry, then hands it to each caller's own policy. The feed
row layers its leading label + trailing "Clear all" reservation on top and
keeps decideChipLayout; the tag cell's chips-only policy is extracted as
fitChipsWithPill (pure, unit-tested) alongside it. Add a jsdom test for the
hook (stubbed offsets, re-measure on deps). Behaviour unchanged — verified
in the browser: feed all/count modes + popover, tag "+N" fit + hover reveal.
The feed applied its starred default via a localStorage mirror
(LS.defaultViewFilters) written by SavedViewsProvider and read by
FeedFiltersProvider on load. That raced the login effect: on a genuinely
fresh browser (no LS.filters and no mirror yet) the feed could seed
DEFAULT_FILTERS and never re-apply the default until a reload. Plex had no
default-application at all.
Replace the mirror with a single path: SavedViewsProvider applies the
starred default via cfg.setFilters on the views query's FIRST resolve, but
only into a genuinely fresh module (no stored filters for the account, and
— feed only — no URL filter override). Applying persists it, so it happens
once and the last-view-wins path takes over, matching the documented
storage semantics. Because it runs off cfg.setFilters it delivers the Plex
port for free. Extract pickDefaultToSeed (pure, unit-tested) and cover the
first-resolve/stored/no-default cases for both modules in a provider test.
Removes the now-dead LS.defaultViewFilters key and its mirror read.
UX-8: one Escape must close only the topmost layer, never both an open
popover AND the player beneath it. The popover's Escape listener is on
`document` and fires before the player's on `window`; closing the popover
flushes its layer off the shared stack synchronously, so the player's
`anyLayerAbove()` lookup read "nothing above" and closed too. Add a
consumed-event flag to layerStack (`markEscapeHandled`/`escapeHandled`):
the acting topmost layer stamps the shared KeyboardEvent, and both players
check the stamp before the (now-mutated) stack. Immune to listener order
and mid-dispatch stack mutation. Cover with layerStack.dom.test.ts.
UX-7: the E4 `.reveal` sweep already covered the card action rows; two
residual hover-only traps remained — invisible-but-tappable/Tab-focusable
on touch. Add a group-agnostic `.reveal-hover` recipe (display:none under
`(hover: none)`) paired with each site's own hover/focus-within reveal:
- VideoCard play/Continue/Restart overlay: also reveal on
group-focus-within (keyboard), and drop it on touch — the card tap
already plays/resumes, so no permanent scrim over every mobile thumbnail.
- Carousel prev/next arrows: gone on touch, where the strip swipes.
The notifyCollectionEditError helper localizes the backend's duplicate-name
409 to plex.collection.nameTaken; it was applied to the New-collection panel
but two sibling collection-write sites still routed 409 through
notifyActionError, surfacing the server's raw English detail under the HU UI:
- PlexCollectionView.doRename (rename collision)
- PlexCollectionEditor.createNew (create-in-editor duplicate name)
Both now use notifyCollectionEditError, matching the create path. Also adds a
direct dom test for the extracted PlexLibraryPicker covering the enabled gate,
emptyHint, results, honest error, and no-results branches.
- CLAUDE.md: new "Collaboration" section (plan-first, options-with-marked-
optimal-and-why, mockups for design decisions, challenge unsound requests,
visual UAT before done, terse otherwise).
- CLAUDE.md: clarify release is user-owned in timing, agent-run in execution.
- CLAUDE.md + docs/development.md: document the ruff format --check gate step
(already enforced in .gitea/workflows/check.yml, was missing from the docs).
- CLAUDE.md: refresh the stale footer to point at the now-existing
docs/development.md and backend/app/README.md.
Backend: create-collection happy path (write-through + editable) and the
create resync-failure 422 (honest reconcile detail), mirroring the rename/remove
tests. Frontend: extract notifyCollectionEditError (the 409->nameTaken vs
generic-fallback mapping that was inline in the New-collection create()) and
unit-test it; wire create() through it.
Renders PageScroller and drives restoreScroll's rAF synchronously with fake
metrics: asserts a forward nav (new entry id) opens at top and Back restores
the offset, and that a new content key within one entry resets to top while
returning restores — the per-(entry,key) contract only the id primitives were
covering before.
AddItemPanel (Collection page) and NewCollectionPanel (browse) had near-
duplicate debounced library-search + result-list rendering. Extract it into
one PlexLibraryPicker that owns the query, the loading/error/empty states and
the row markup; callers keep their own chrome and pass term + renderAction.
No behaviour change.
Adds ruff's formatter to the gate. line-length 100 matches the frontend
prettier printWidth. The reformat itself lands as the next commit (kept
separate so it can go in .git-blame-ignore-revs).
knip flagged src/lib/storage.ts readMerged as an unused export — its only
caller is readAccountMerged in the same module, so it doesn't need to be
exported. This aborted the 0.72.0 publish gate after the merge; no behaviour
change.
- collections: the resync-failed-after-successful-write path now returns 422
(honest "reconcile on next sync" detail), NOT 502. The frontend treats
502/503/504 as gateway-unreachable and raises a sticky offline banner
(api/core.ts RETRIABLE_GATEWAY); the app answered, so a 422 action-error
toast is correct. Removes the false-offline regression the 502 introduced,
and — since 422 isn't retried — the create double-write vector with it.
- tests: make the mutate-then-resync rollback assertion non-vacuous — the
resync stub now flushes a partial local change before raising, so the test
proves db.rollback() actually undoes it; split into a write-failed (422)
case and a resync-failed (422 + reconcile detail + rollback) case.
- New-collection dup hint is now muted-tone with its own nameMaybeTaken copy
(per-library uniqueness means the name may be free here) instead of the red
authoritative nameTaken, which the still-enabled create flow contradicts.
- history: cover the navCounter post-reload collision guard (ensureNavId
advances the counter past a higher persisted sfScroll id).
Round of USER-triggered deep-review fixes (11 CONFIRMED + 1 PLAUSIBLE, all
verified) on the 22-commit branch:
- collections: split the mutate-then-resync error surface. A resync READ that
fails after a successful Plex mutation no longer reports a false 422 "Plex
write failed" + rollback; it returns 502 "sync pending" (the next full sync
reconciles the mirror). New _write_or_pending_error helper; delete is exempt
(no Plex read). Covered by new happy-path / 422-rollback / 502 tests.
- New-collection dup-name guard is now advisory only: existingNames is the
search-scoped, all-library union, but the backend enforces uniqueness PER
library — so it no longer blocks a valid same-name collection in another
library, and the backend 409 is localized to nameTaken.
- member-remove now confirms before the write-through (matches delete-collection).
- add-title / new-collection library search render an error state on request
failure instead of a misleading "No matches".
- history: seed navCounter past the current entry's persisted sfScroll on load
so post-reload ids can't collide with back-stack entries and mis-restore scroll.
- PageScroller: LRU-cap the navId-keyed savedScroll map (was unbounded).
- share isIntrinsicReadOnly between the browse tile and the info page so the
lock/edit affordance can't drift.
Two follow-ups after testing the collection editing against real Plex, which revealed that
Plex CONFLATES same-named collections (two rating_keys resolve to one merged member list) —
a genuine footgun.
- Duplicate-name guard: create and rename now reject (409) a title that already exists in the
target section (case-insensitive), before any Plex write. create() can also derive the
library from the seed item, so the caller needn't know the section key. The create dialogs
surface the 409, and the New-collection panel warns inline before submit.
- A "New collection" button on the Collections browse surface (admin only) opens a panel:
a name field (with the live duplicate check) + a search to pick the FIRST title — Plex can't
create an empty collection, it needs a seed. Creating opens the new collection's detail page
straight away, so you can keep adding titles there. This is the B-route the user wanted,
vs. the only prior path (a movie's "add to collection").
Verified live: the 409 guard (creating an existing name is blocked, no Plex write), and the
full New-collection flow (name + first title → create → lands on the detail page); test
collections created and deleted, real account left clean.
Gate: tsc/lint/knip/prettier clean, 424 vitest; backend ruff clean, 7 collections tests
(+ a duplicate-name-rejection assertion).
Clicking a collection tile now opens a dedicated Collection info page (like the Plex-web
one) instead of just applying it as a filter. New GET /collections/{rk} returns the hero
aggregates (year range, most-restrictive member cert, total runtime) + the members ordered
by release year then title; a new PlexCollectionView renders the hero (art/title/meta/
summary+More) and the member grid, with a Play-all and Shuffle over the playable leaves
(reusing the player's existing queue + auto-advance).
For a non-external, admin-enabled collection the page offers rename / delete / add-title
(a search picker) / remove-item, all through the pre-existing admin endpoints that write
THROUGH to Plex (each edit awaits and re-fetches — no optimism, since a Plex write can 422).
An eligible collection shows "Enable editing" first (the DB editable flag). Read-only ones
(smart / imdb / tmdb / … auto-lists) are marked with a lock on both the info page AND the
browse tile, up front.
New `{ kind: "collection" }` sub-view wired through plexSubUrl (bookmarkable ?v=collection:rk),
PlexBrowse (render branch, Backspace back-nav, tile handler), with a round-trip test.
Verified live: tile→page, hero + chronological members (James Bond 1962→2021, PG-13, 55h),
Play/Shuffle, Enable editing → edit controls + remove-X, the add search (marks already-in
titles "Added"), and the read-only lock on external collections. Real Plex mutations wired
but not executed against the live account.
Gate: tsc/lint/knip/prettier clean, 424 vitest; backend ruff clean, 6 collections tests
(+ a detail-hero/ordering assertion).
A new Discover rail listing what the user last FULLY watched, newest first: a movie as its
own card, a show as the specific EPISODE watched (carrying its show title via leaf_card). Sits
above Recently added. Movies honour the sidebar filters like the other rails; episodes carry no
filterable metadata so they don't. Clicking a card plays it (the existing onCard path already
routes episodes to the player). Bilingual heading added.
Verified live: the rail renders watched movies + episodes with Watched badges.
Gate: tsc/lint/knip/prettier clean, 423 vitest; backend ruff clean, 13 discover tests (+ a
recently-watched movie+episode assertion).
The Surprise rail drew a random 20+20 sample including already-watched titles. Its point is
to surface something to WATCH, so exclude seen titles (a watched movie, a fully-watched show
— a partially-watched or never-started show still counts as unseen) and cap at 3 movies + 3
shows (6 total), interleaved.
Verified live: /surprise returns 6 items, 3+3, none watched.
Gate: backend ruff clean, 12 discover tests (+ a watched-exclusion/cap assertion).
The collection browse-tile montage queried only movie members, so show-library
collections (IMDb Popular, TMDb Trending, … in the TV library) rendered as bare colour
tiles. Union the art from PlexShow members as well (a collection lives in one library, so
exactly one side has art); fall back to the collection's own poster, then the colour hash.
Runtime stays movie-only (shows have no single duration).
Verified live: every collection now shows a montage (0 art-less tiles, was ~all the show
collections).
Gate: backend ruff clean, 5 collections tests (+ a show-collection art assertion).
The full-bleed 4-poster collage drowned the on-tile text. Split PosterTile in two: the top
half shows up to 2 montage posters, the bottom half is a solid colour-hash band that carries
the label, so the text is always readable. Genre tiles drop the count (just the label now);
collection tiles show the name on one truncated line plus a sub-line of item count · total
runtime. A group with no art falls back to an all-colour tile.
Backend: `/collections` cards gain `total_runtime_s` (a new `runtime_by_key` grouped sum) and
their montage art is capped at 2. Frontend: PosterTile reworked; a `collectionMeta` helper
builds the count·runtime sub-line via the existing formatRuntime.
Verified live: collection and genre tiles render 2 posters over a readable colour band with
the right sub-lines; colour fallback intact for art-less groups.
Gate: tsc/lint/knip/prettier clean, 423 vitest; backend ruff clean, collections tests green
(+ a new art/runtime assertion).
The Browse-by-genre tiles and the Collections browse tiles were flat color-hash
gradients. Give each a 2×2 collage of its top films' poster art, with the color hash
kept as the fallback so a group with no matching art (a show-only or empty collection)
never renders blank.
Backend: a shared `top_art_keys` windowed query returns, per key (a genre value or a
collection rating_key), the top few items' rating_keys — best-rated then newest, in one
query. Wired into the Discover genre tiles and the collections union endpoint; the client
renders the posters from the existing /api/plex/image/{key} endpoint, so no new image
plumbing. Frontend: a shared PosterTile component (collage + scrim + label, color-hash
fallback) replaces the two ad-hoc colored-tile implementations.
Verified live on localdev: genre tiles and collection tiles render real poster collages,
with the color fallback correctly kicking in for groups without art.
Gate: tsc/lint/knip/prettier clean, 423 vitest; backend ruff clean, 11 discover tests
(incl. a new genre-art assertion).
The app scrolls one persistent <main> (PageScroller); its saved offsets were keyed by a
content `scrollKey` that is the same for a Plex page's Discover and grid views, so opening
a Browse-by-genre tile inherited the scrolled-down Discover offset (landing mid-page) and
Back couldn't tell the two views apart.
Tie scroll memory to the history entry instead: every pushState that opens a main-content
view stamps a fresh `sfScroll` id into history.state (setPage, Plex sub-view open); the
initial/legacy entries get one lazily via replaceState. PageScroller now saves/restores per
`${entryId}:${scrollKey}` — the entry id makes Back restore exactly where you left an entry
and a forward navigation (new id) open at the top, while the scrollKey still resets a new
feed filter to the top within the same entry. A small nav bus lets a Plex sub-view open
(which doesn't re-render the scroller) trigger the restore; `history.scrollRestoration` is
set to manual since the browser can't see our inner scroller.
Verified live: genre tile opens the grid at the top and Back returns to the Discover scroll;
feed→plex→Back restores the feed scroll; a feed filter change still resets to the top.
Gate: tsc/lint/knip/prettier clean, 5 new nav-id unit tests, full vitest green.
The floating header wrapper is pointer-events-none (so clicks fall through its empty
area to the content below); every interactive child must re-enable pointer events.
PlexViewToggle rendered its SegmentedToggle without the pointer-events-auto wrapper the
sibling PlexScopeToggle has, so the whole Discover/Browse switch was click-dead — the
buttons rendered and were correctly wired, but no real mouse click ever reached them.
Add `className="pointer-events-auto shrink-0"` to the toggle and a regression test that
asserts the root re-enables pointer events (plus the render/onChange/hide-on-deeper-view
contract). Verified live on localdev: real clicks now switch Discover⟷Browse (?v=browse)
in both directions; a header-wide hit-test audit shows no other dead controls.
Gate: tsc/lint/knip/prettier clean, 418 vitest.
Third review round, scoped to the round-2 fix commit. The production ternary was
confirmed correct; two sibling-surface findings:
- The sidebar CollectionPicker's clear-all still wrote collectionTitles:{} while the
chip-row remove (fixed in round 2) wrote undefined — a divergence that leaves the
{}-vs-absent cache split on the picker path. Collapse an empty map to undefined there
too, so both no-collection routes key-match the pristine state.
- The queryKeys lang test was near-tautological (a variadic-spread factory can't drop a
passed arg), and nothing asserted the CALL SITE keys carry lang. Add a real guard in
PlexDiscover.dom.test.tsx that inspects the registered query keys for scope+lang, and
reword the factory test's comment to say it only shape-pins the segment pass-through.
Gate: tsc/lint/knip/prettier clean, 413 vitest.
Second review round, scoped to the round-1 fix commit (where round-1's own fixes
introduced new code):
- The collection chip-remove pruned collectionTitles to an empty {} on last removal,
which still hashes differently from the pristine no-collection state (key absent) in
the React Query keys → a residual cache split. Collapse an empty map to undefined so
the key-shape matches the pristine one.
- Replace the tautological "order-independent signature" test (both objects spread from
EMPTY_PLEX_FILTERS inherited the same key order, so it proved nothing) with full
literals in genuinely different insertion order that would fail an order-sensitive
JSON.stringify signature.
- Cover the collectionTitles chip-remove prune (usePlexActiveFilters.dom.test.tsx) and
the plexSurprise/plexDiscover lang-key split (queryKeys.test.ts) with regression tests.
Gate: tsc/lint/knip/prettier clean, 412 vitest.
Self-review of the branch diff (feature/plex-filter-bar-overhaul) against main:
- Surprise rail query key omitted `lang` while the backend /surprise now filters
against the language-specific genre column → a HU/EN switch under an active genre
filter left the rail stale. Key it by scope+lang+filters, matching plexDiscover.
- Chip-removing a collection left its entry in `collectionTitles` (hashed into the
plex query keys), splitting the cache from the pristine no-collection state and
growing the titles map unboundedly. Prune the title on remove, like CollectionPicker.
- The saved-views default-mirror effect gained `cfg` in its deps (re-memoized on every
live filter edit) → a synchronous localStorage write per keystroke. Depend on the
module-stable mirror fn instead.
- Collections-browse search input had no accessible name (placeholder only). Add
aria-label + type="search".
- Extract the tile hue-hash + gradient duplicated between PlexDiscover and
plexBrowseLists into a shared lib/hueTile so the two surfaces can't drift.
- Document migration 0073 downgrade() as lossy and de-dup non-feed rows before
recreating the narrow (user_id, name) constraint (verified by the drift test).
- Add unit tests for coercePlexFilters (legacy single-collection migration) and
plexFiltersSignature (order-independence, title-ignoring).
Gate: tsc/lint/knip/prettier clean, 410 vitest; backend ruff + migration drift +
saved-views tests green.
Plex now has the same personal-views system as the feed — the rail header icon
opens the quick-pick popover, the rail title opens the manager modal (list /
create / rename / delete / reorder / set-default / apply), and the active view
leads the chip row with a "modified" dot.
One SavedViewsProvider serves both modules: it switches `source` (feed ⟷ plex)
by the active page (both filter providers are global, so it reads whichever the
page owns) and runs the same logic on per-source wiring — the live filters,
apply, a stable signature (feed filtersToParams / plex plexFiltersSignature),
coercion (asFilters / coercePlexFilters), and what to strip on save. Moved below
PlexProvider in the tree so it can read plex state; resets sticky/modal on a
source switch. The shared popover/modal/chip are unchanged; the modal describes
a plex view's captured filters via a pure buildPlexActiveFilters (extracted from
usePlexActiveFilters — no duplication). api.savedViews/createView take a source;
`SavedView.filters` is now an opaque blob. New coercePlexFilters/plexFilters-
Signature (also dedups PlexProvider's collection migration). Empty-state copy
genericised (no longer says "feed"). A Plex view captures the rail filters (not
watch-state/sort/scope).
Verified end-to-end on localdev: create/apply/preview/default/delete on plex,
feed views stay separate + unregressed (402 vitest), no console errors.
Saved "smart views" were feed-only; the Plex module is getting them too. Add a
`source` discriminator to saved_views ("feed" | "plex"), default "feed" so every
existing row and any source-less client stays feed. Widen the per-user unique
name constraint to (user_id, source, name) so a feed view and a Plex view can
share a name, and scope create's position numbering + the is_default "clear the
others" to (user, source).
Migration 0073 (ADD source NOT NULL server_default 'feed' + swap the unique
constraint; reversible). Endpoints: GET/POST take `source` (default feed).
Tests: source isolation, per-source default, per-source name uniqueness.
The header library-scope pill gains two entry points — Playlist and Collection —
next to All/Movies/Shows, a second way into each area (the rail filter is the
other). Picking Playlist shows a grid of the user's playlists (a tile opens that
playlist's view); Collection shows a searchable grid of collections (a tile
filters the title grid by that collection). Pure front-end — reuses the existing
plexPlaylists / plexCollections(union) list endpoints.
Two new top-level sub-views (`playlists` / `collections`, bookmarkable as
?v=playlists / ?v=collections) rendered by PlexBrowse via the new plexBrowseLists
surfaces (hue-tiled like the Discover genre tiles). The PlexViewBridge now carries
the view `kind`; PlexScopeToggle is a 5-way that derives its active segment from
it (browse-mode when in one, else scope) and drives the two views through the
bridge — picking a scope from a browse mode returns to the title grid. The
Discover/Browse toggle hides in these modes; Backspace + the search→grid flip
cover them.
Tests: PlexScopeToggle (5-way + browse-mode drive/active), plexSubUrl round-trip
for the two new views. Gate: tsc/lint/knip/prettier + 402 vitest.
The Discover ⟷ Browse switch now sits in the header next to the library-scope
pill (All / Movies / Shows), instead of on the chip row / atop the Discover
surface.
It's driven through a small one-way PlexViewBridge: PlexBrowse still OWNS the
sub-view (useHistorySubview) and publishes its top-level view + opener to the
bridge; the always-mounted Header reads it to render PlexViewToggle. A global
sub-view provider was deliberately avoided — PlexBrowse's hook remounts on every
plex entry and re-seeds from history (a page switch clears ?v + drops _sub), so
each entry starts at the Discover root; lifting it out would let the view go
stale across navigation. The toggle hides on deeper views (show/info/person/…)
and on non-plex pages (bridge cleared on unmount). sortView is just SortControl
now; the grid toolbar no longer carries the toggle.
Verified: toggle in the header, switches discover⟷grid, hides on info + on Feed,
and Feed→Plex resets to Discover (no stale view); no console errors.
Discover now behaves like the Browse grid — the active metadata filters (genre /
rating / year / collection / studio / …) narrow every rail: Continue watching,
Recently added, Surprise me, the genre tiles, and Because you watched. Same
LibraryFilters / meta_filter_conds the grid uses, applied to each rail's movie +
show query (movie duration included); a rail may come back empty, which the client
already hides. Genre-language matches the grid.
Because Discover honours filters itself, adding a filter no longer force-flips the
Discover landing to the grid (only a header SEARCH does, since Discover has no
search rail) — so you stay on Discover with narrowed rails.
Frontend threads `filters` into api.plexDiscover/plexSurprise + their query keys
(plexSurprise is variadic now) and the watched-toggle setQueryData; PlexDiscover
reads them from usePlex. Tests: two discover filter-narrowing cases; PlexDiscover
dom test updated for the provider default + 3-arg call.
A title can belong to several Plex collections, so the collection filter becomes
multi-select combining with Any (OR, default) / All (AND) — the same shape as the
genre/actor filters.
Backend (query.py): `collection` (single) → `collections` CSV + `collection_mode`
on LibraryFilters; meta_filter_conds ORs/ANDs `collection_keys.contains` per id.
Frontend: PlexFilters `collection`/`collectionTitle` → `collections` + optional
`collectionMode` + `collectionTitles` map; CollectionPicker is now a multi-select
Combobox with the any/all match footer; one chip per collection (usePlexActive-
Filters); the info-page "Browse collection" strip unions into the list; applyFilter
now shallow-merges label maps so adding one keeps the others' titles. PlexProvider
migrates a legacy persisted single `collection` into the new shape on read.
Test: test_plex_collections_filter (single / any-union / all-intersection).
The playlists section used to render an always-open list plus a dedicated
create input, growing the rail taller with every playlist. Replace it with a
single Combobox field (reusing the shared component, like the feed's "Your
tags"): the list of playlists only appears when the dropdown is opened, and
typing a name that doesn't exist offers a "Create …" row. Picking a playlist
opens it (the existing playlistOpen hand-off); the create row creates then opens
the new one. `selected` stays empty — this picks an action, not a persisted value.
The field's input carried min-w-16 (64px). Flex line-packing sizes a flex item
at its min-width, so in a narrow rail field (~202px) one chip (~119px) + the
64px-reserved input + the chevron overflowed the line, wrapping the chevron
down while the (grown, empty-looking) input filled the first line — the "breaks
to two rows even though it'd fit" glitch seen on the Plex Collection widget and
the feed's channel picker (both are Combobox). Lowering the floor to min-w-8
(32px) keeps chip + input + chevron on one line; flex-1 still grows the input to
fill the remaining width. Verified in-browser: field height 54px→36px (one row).
The "Show" (watch-state) filter leaves the left rail and becomes a FilterSelect
pill next to the item count, mirroring the feed's Watch-state control — a
top-level "what am I looking at" choice, always in reach, not buried in the rail.
It still emits a removable chip when non-default (same as the feed).
Bring the Plex module's filter surface in line with the feed's final shape:
- Flatten the rail: PlexSidebar passes `collapsible={false}` to FilterRail so
the widgets become plain always-open label + control rows (no collapse
chevrons; reorder/hide in edit mode stay). Drop the rail's own "Clear all" —
the chip row above the grid already owns it.
- Move Sort + the Discover/Browse toggle into the applied-filter chip row,
pinned right directly above the grid (a `sortView` fragment, like the feed's),
and give ActiveFilterChips `flex-1` so it gets the shared responsive overflow
(wide: fit + "+N" pill; narrow: single "N filters" count pill) — the chips can
no longer crowd the controls off-screen.
ItemCount leads its own toolbar row, matching the feed. No behaviour change to
Discover, which keeps its standalone view toggle.
Flattened, more compact YT feed filter rail and a responsive applied-filter row:
- Rail: removed the in-rail Clear all; Upload date is now a single-select dropdown
(Any time + more presets + reveal-on-select Custom range); widgets are plain
label + control rows (non-collapsible), with the "Your tags" title opening the
tag manager.
- Sort + View moved into the applied-filter row (pinned right, above the grid)
with responsive overflow: "+N" pill on wide lanes, a single count pill on narrow.
- Earlier branch work also merges here: typeable combobox filters, hidden-channel
empty->Library hint, Mine/Library scope toggle in the header, per-category
any/all tag match, and the saved-views redesign (popover + modal + chip).
Three USER-triggered deep-review rounds (trend 14 -> 4 -> 1); round 2 caught a real
coerce-on-seed migration bug. Gate-green: tsc, 397 vitest, lint, prettier, backend
pytest. Not yet released -- prod stays 0.71.0.
The re-review of the fix round surfaced 4 findings; fixing the coverage one
exposed that the C1 migration was ineffective:
- FeedFiltersProvider.loadAccountFilters: run coerceTagModes on the RAW stored
blob, not the object merged over DEFAULT_FILTERS. The merge injects
DEFAULT_TAG_MODES (all "or"), which coerceTagModes prefers over the legacy
scalar `tagMode` — so the migration never fired. Now a persisted legacy
tagMode:'and' actually migrates into tagModes.topic (proven by a new test).
- test_feed_views: the tag_mode_other=and case was vacuous (only chC carried
either "other" tag, so AND == OR). Give chA an "other" tag too so OR-other
{chA,chC} differs from AND-other {chC} — the conjunctive path now narrows.
- DateFilterSelect: reset the custom reveal in useLayoutEffect (not useEffect)
so the correction commits before paint — no one-frame "Custom range" flicker
on the trigger after an external clear.
- SavedViewsProvider: extract the sticky-view tie-break to a pure, tested
lib/savedViews.ts (preferSticky).
New tests: savedViews (preferSticky), FeedFiltersProvider.dom (legacy tagMode
migration on the localStorage/default-view seed).
- test_feed_views: extend _seed_tagged so a channel carries 2+ tags in a
non-topic category, and add language-AND / other-AND equivalence cases — the
per-category any/all generalization was only exercised for topic.
- tagModes.test: coerceTagModes migration (legacy global tagMode -> topic),
per-category preservation, invalid-value coercion, defaults.
- urlState.dom.test: per-category tag-mode URL round-trip + legacy ?tagmode=and
migration into the topic slot.
- Combobox.dom.test: the any/all match footer (hidden below 2 selections, reports
a mode change at 2+).
- chipOverflow.test: the all/some/count fit boundaries and the +N reservation.
Correctness / behaviour:
- Move the Personal Views modal from inside the collapsible filter rail to the
feed page (moduleRegistry FeedPage), so opening it from the always-visible
toolbar view chip works while the rail is unpinned (its default) instead of
no-op'ing and latching modalOpen into a later stale pop-open.
- Migrate a persisted legacy global `tagMode` into per-category `tagModes` on the
localStorage seed paths too (loadAccountFilters), matching the saved-view and
URL paths — an upgrading user's topic any/all no longer silently reverts to 'or'.
- DateFilterSelect: reset the custom reveal on an external clear while the popover
is closed (guard on !open, not maxAgeDays), so the trigger reverts to "Any time"
instead of lingering on "Custom range" with empty fields.
- PersonalViewsModal delete: only advance the detail selection when the delete
actually happened (remove() now resolves a boolean) — cancelling the confirm no
longer switches the selected view.
- SavedViewsProvider: prefer the explicitly-applied (sticky) view among several
that capture identical filters, so apply(B) isn't re-attributed to an earlier
duplicate A (wrong "modified" chip name).
A11y:
- Give the clickable panel title a descriptive accessible name (titleActionLabel,
wired to "Personal views") instead of announcing only "Filters".
- Give the "+N" overflow pill an aria-label ("N more filters").
- Use t("views.reorder") for the saved-view drag handle instead of a hardcoded
English "reorder" literal.
Also extract the chip-fit decision into a pure, testable lib/chipOverflow.ts and
measure the inter-chip gap live (rem-safe) instead of a frozen 6px constant.
Sort and the view switcher are tightly bound to the grid, so pin them to the
right of the applied-filter row (directly above the cards) instead of the filter
toolbar; they render there on both the feed and channel-scoped views.
ActiveFilterChips now shares that row with the controls and trims itself so it
never crowds them, measuring fit off a hidden mirror of the full chip set:
- wide -> show as many chips as fit, then a "+N" pill whose popover holds the rest;
- narrow -> collapse every chip into one "N filters" pill whose popover holds them all.
Both popovers list the (overflowed / all) chips, each removable, plus Clear all.
Rework the feed filter rail into simpler label + control rows:
- Remove the rail's own "Clear all" (the chip row above the cards already
surfaces one whenever any filter is active).
- Replace the Upload date chip group with a single-select dropdown
(DateFilterSelect): an "Any time" clear, more presets (2 weeks / 3 months /
2 years / 5 years) and a "Custom range" that reveals from/to on select; the
trigger then reads the chosen range. Selection logic is the pure lib/dateFilter
(applyDateChip -> selectDateChip, no toggle-off; tests updated).
- Widgets are no longer collapsible: PanelGroup gains `collapsible` (feed passes
false) so groups render as always-open label + control rows; reorder + hide in
edit mode stay. Other rails (Plex/Playlists) keep the collapsible islands.
- A non-collapsible group's title can act as its own trigger (PanelItem
onTitleClick): "Your tags" opens the tag manager, replacing the inline Manage
button.
Retire the standalone 'Saved views' sidebar panel. The saved-views feature now
lives in three lighter surfaces, coordinated by a new SavedViewsProvider (query +
CRUD + the active/sticky view + modal state):
- the filter panel's header ICON opens a quick-pick popover (apply a view, active
one checked, plus 'Manage views…');
- the header TITLE ('Filters') opens a new Personal Views modal — a master/detail
manager: left = the reorderable list + a save-current-filters input, right = the
selected view's name, its captured filters (shown as the same chips the toolbar
row uses), default toggle, apply, and delete;
- a SelectedViewChip leads the applied-filter row, naming the active view with a
warn-coloured 'modified' dot once the live filters diverge from it; clicking it
reopens the picker.
SidePanel/FilterRail gain optional iconExpanded + onTitleClick slots (feed-only,
hidden for the demo account); the collapsed tab keeps its plain pin behaviour.
SavedViewsWidget is deleted and dropped from the panel layout.
Verified live: icon popover lists HH + Manage; applying HH surfaces the chip and
switches scope; removing a filter lights the dirty dot; the title opens the
master/detail modal with the captured-filters summary. 374 vitest, tsc, eslint
(0 err).
The tag 'Match: Any/All' toggle was a single GLOBAL setting that only ever
affected topics, shown as its own row under the Topic widget. Make it
per-category — language, topic, and 'your tags' each get their own independent
any/all — and move it into each filter combobox's popover footer (like
SortControl's direction segment), reclaiming the standalone Match row.
Backend: the feed query already groups selected tags by category; replace the
single tag_mode with per-category tag_mode_language/topic/other params and apply
the conjunctive (HAVING all) branch per category. Mirrored across filtered_query,
feed_summary.passes(), and the /facets drill-down exclusion (kept in lockstep;
equivalence test updated to tag_mode_topic). No schema change.
Frontend: FeedFilters.tagMode -> tagModes {language,topic,other} (pure leaf
lib/api/tagModes with a coerceTagModes migration for the legacy global value);
serialized to the API and the shareable URL per category (legacy ?tagmode=and
still resolves into the topic slot); saved-view blobs migrate on read. Combobox
gains an optional matchMode/onMatchModeChange footer segment, shown once 2+
values are picked; the three tag comboboxes wire it to their category.
Verified: backend feed tests green (12, incl. the count/facets equivalence net);
live, the footer appears at 2+ selections, toggles aria-pressed, serializes
tagmode_topic=and to the URL, and the request carries tag_mode_topic=and (200).
374 vitest, tsc, ruff, eslint (0 err).
The scope toggle used to sit at the top of the filter rail, which is buried when
the rail is collapsed/unpinned. Relocate it to the header, just right of the feed
search box — mirroring Plex's scope toggle placement — so the primary 'what am I
browsing' control is always visible.
New FeedScopeToggle adapter (SegmentedToggle over filters.scope, same as the Plex
one over its scope); rendered in Header on the feed page between the search box and
the sync chip. Removed the toggle from the Sidebar rail's now-unused `top` slot and
dropped the dead `top` prop from FilterRail. Same filters.scope field, so behaviour
is unchanged.
Verified live: the toggle sits beside the feed search and switches scope; the rail
no longer carries it.
Filtering by a channel under the Mine scope returns nothing when that channel's
subscription is hidden (or you don't subscribe to it) — its videos only live in
the shared Library, since the Mine feed excludes hidden subscriptions by design.
Previously this left a bare 'No videos match these filters', which reads as a
bug when you know you're subscribed.
Add an empty-state hint + a 'View in Library' button (switches scope to all) when
there's a channel filter under Mine and zero results. Bilingual EN/HU.
Verified live: ?channel=<ColdFusion>&show=all under Mine shows the hint; the
button switches to Library and surfaces the 36 videos.
The <input> carried only floating-ui's reference ref, so the component's own
inputRef stayed null. The field container's onMouseDown then preventDefault()-ed
every press (its `e.target !== inputRef.current` guard was always true) and the
compensating inputRef.current?.focus() was a no-op — so clicking the box opened
the list but never focused the input, and mouse users could not type to filter.
Tab focus still worked, which is why it read as 'click, can't type'.
Feed a merged callback ref that assigns the node to both floating-ui's
setReference and inputRef. Fixes the Channel, Your tags, Topic, and Language
sidebar filters at once (all use Combobox); the channel list's 50-row cap is now
reachable past 50 via search. Single-select commit no longer re-focuses (that now
actually fires and would re-open the just-closed list). Regression test presses
the field chrome and asserts the input receives focus.
Verified live: clicking the Channel field focuses it and typing 'cold' narrows
the list to ColdFusion / wehatethecold.
Collects the toolbar/header/layout redesign and assorted bugfixes on the
bug/channel-back-header-overlap branch: new FilterSelect / ItemCount /
pill-SortControl primitives, removed ModuleName, the --sbw / --content-mw /
--rail-pinned-w chrome-alignment machinery, the centred search box, and the
two deep-review rounds (menu-role a11y, sort aria-pressed, rem rail threshold,
accessible trigger names, e2e helper fix, added coverage).
Not a release: VERSION stays 0.71.0 and this is not deployed to prod.
Second gate round over the full branch surfaced 4 verified findings (1 was a
regression the round-1 role change introduced); a 5th (--sbw load jump) was
dropped as refuted (the header only mounts after auth+data, so --sbw is set
before it ever paints).
- e2e (medium): the round-1 switch of FilterSelect rows from role=option to
menuitemcheckbox moved their state from aria-selected to aria-checked, but
the Playwright helper expectContentType still asserted aria-selected, so the
channel specs would fail whenever the e2e lane runs (not part of the fast
gate, so it stayed green). Assert aria-checked.
- a11y (low): the FilterSelect and SortControl pill triggers set
aria-label={label}, which overrides the visible "Label: value" text, so a
screen reader announced only the group name, never the current selection.
Carry the value (and, for sort, the direction) in the accessible name,
matching ViewSwitcher. Updated the unit-test trigger name queries to match.
- tests (low): cover the PageShell font-scale MutationObserver recompute path
(a post-mount root font-size change → --rail-pinned-w re-computes once, no
loop) — previously only the pre-render and resize paths were exercised.
Verified live: the triggers now announce e.g. "Sort: Date, Descending" /
"Watch state: All". Gate green: tsc (app+node+e2e), 373 vitest, prettier,
eslint (0 err).
Epic-end deep-review of the branch surfaced six verified findings; fix them.
a11y (medium): FilterSelect and SortControl opened their AnchoredPopover in
mode:"listbox" but never passed a role, so the surface fell through to
role="menu" while the rows rendered role="option"+aria-selected — options
orphaned outside a listbox, so AT could not announce the selected value.
Switch both to the house menu pattern (as ViewSwitcher already uses): a menu
surface with menuitemradio (single) / menuitemcheckbox (multi) rows carrying
aria-checked, scoped by a role=group. Trigger stays a button+haspopup=menu.
a11y (low): the SortControl asc/desc direction segment signalled the active
direction by colour alone. Add aria-pressed so the current direction is
programmatically exposed, matching SegmentedToggle.
correctness (low): PageShell converted a rem maxWidth to px with a hardcoded
16 to decide the pinned-rail overlay threshold, but index.css scales the root
font by --font-scale (default 1.06 ~ 17px, slider 0.9-1.3), so the threshold
was wrong at every text size on the rem pages (playlists is the only one with
a pinned rail). Use the live root font-size, scale the 36rem slack the same
way, and recompute when --font-scale changes (a guarded MutationObserver — the
slider fires no resize).
tests: cover the two gaps the review flagged — ItemCount (plural, 1M+ overflow,
undefined icon-only, tooltip) and the PageShell rail-pinned-w/content-mw
threshold math incl. the font-scale scaling. Updated the FilterSelect/
SortControl role queries to the new menuitemradio/checkbox roles.
Verified in the running app: menu surfaces carry role=group + menuitemradio/
checkbox with 0 orphaned options, and the sort direction segment exposes
aria-pressed. Dropped two review findings (max-width duplication, pill-class
reuse) as refuted on adversarial verification. Gate green: tsc, 372 vitest,
prettier, eslint (0 err).
The floating header band already centres on --content-mw (the width the
page's cards use), but its search box + scope-toggle/sync-chip group was
packed to the left edge (default justify-start), leaving it offset from
the centred card grid on max-width pages (Plex/Channels/Playlists).
Add justify-center to the band so the group centres within it: once the
column is wider than the box's max-w, the box sits centred over the
cards instead of hugging the left edge. Shared header row, so it applies
to every module with a SearchBar. Verified at 1920px on Plex: group
centre == content-band centre (431px margin each side).
On a page that centres its content at a max width, if the window is wide enough for the rail to fit
in the content's LEFT margin without overlapping it, pinning the rail now leaves it OVERLAY — the
content doesn't shift. It reverts to the old in-flow "push" only when there isn't room: a full-width
page (feed) or a window too narrow for the rail to clear the centred content.
PageShell computes --rail-pinned-w from the page's maxWidth vs the viewport (innerWidth >= maxWidth +
~576px → overlay 2.25rem, else in-flow 16.75rem) on a resize listener; the SidePanel slot reads it.
Verified at 1920px: pinning the playlists rail leaves the table at left 487 (moved 0); the rail sits
256px wide ending at 312 — 175px clear of the content. Feed (full width) still pins in-flow.
Every page's floating header (search bar + sync chip) and the toolbar band now centre on the SAME
max-width the content column uses, instead of spanning the whole column while a max-w page centres
its grid. So the search bar lines up with the left edge of the cards/table on every page, and sits
centred with the content — not against the full page width.
- PageShell takes a `maxWidth`, published as --content-mw; the header adds max-w-[--content-mw] mx-auto
inside its left-4 / right-(1rem+scrollbar) span, so it centres on that width (`none` = full width).
- moduleRegistry PAGE_CONTENT carries each page's content max width (plex 1600px, channels 96rem,
playlists 64rem, downloads 56rem, notifications/users 48rem; feed stays full width) — one source.
- Playlists: its content (header, filter row, table) was left-aligned max-w-5xl → now mx-auto, so the
table is centred and the header lines up with it.
Verified at 1920px: header left edge == content left edge — feed 340, channels 213, playlists 487.
Plex browse now uses the same fixed-width ItemCount pill as the feed (its title count → "N items"),
placed as the leading element, and the toolbar gap tightened 2 → 1.5 to match. One consistent
toolbar shape across the source modules.
- New ItemCount: a generic, FIXED-WIDTH count pill (glass, non-interactive) — "192,050 items"
(neutral noun, dynamic plural), min-w-[9rem] so a changing number never shifts the controls
beside it; a million or more collapses to "1M+" with the exact figure in the title tooltip.
- Feed toolbar: ItemCount is the fixed leading element now; dropped the mx-1 separators so every
gap is a uniform 1.5 (Content-types↔Source used to be wider than Watch-state↔Content-types).
- Chrome-to-card alignment (app-wide, via the shared Header + PageShell): the floating header and
the toolbar band now span exactly the card grid's width — left-4, and a right edge of
1rem + the scrollbar gutter (measured once into --sbw), so the fixed chrome lines up with the
scrolled cards instead of running ~15px past them. Wraps cleanly to multiple rows on mobile.
Renames the count testid feed-result-count → item-count (E2Es updated).
The user no longer wants the header's "FEED" pill (module label + accent dot + ◀/▶ module-step
arrows). Drop it everywhere — the header now starts with the module's SearchBar; module switching
stays on the nav rail.
Deletes ModuleName.tsx, the now-unused stepModule() in lib/modules, and the Header's
module-order/label/setPage plumbing that only fed it.
The earlier bump measured the view-switcher's BUTTON (16px, inherited) instead of its visible
label span (text-sm = 14px), and 14px is the size the user actually finds comfortable. Walk the
over-sized triggers back to text-sm: FilterSelect, SortControl, ViewSwitcher, and the Combobox
input (which was 14px before the bump anyway). The secondary text already at text-sm/text-xs
(count, chips, widget headers, section labels) was correct and stays.
The user prefers FilterSelect's rounded-full glass pill over the ViewSwitcher's squarer
rounded-lg trigger. Give the ViewSwitcher trigger the same look — glass-card/glass-hover,
rounded-full, px-3 py-1, open→border-accent, and a 16px label (was text-sm) — so the two read
as one control family. Applies to every ViewSwitcher consumer (feed, channel manager, Plex
playlists).
Native <select>s, the Combobox input box, and playlist/list cards are a different control class
(form fields / variable-height inputs / cards) and keep their own shape.
The view-switcher's 16px reads comfortably; the toolbar and filter controls sat at 12px. Bring
the marked controls up to match, and lift the general scale on the shared filter surfaces.
- Toolbar (feed + channel + Plex, via the shared components): FilterSelect and SortControl trigger
pills text-xs → text-base (16px, the view-switcher size); the result count → text-sm; the
active-filter chips → text-sm with the "FILTERED BY" label text-2xs → text-xs.
- Filterbar: the Combobox picker input → text-base (matching the toolbar pills) + its chips → text-sm;
the panel/widget headers (SidePanel collapsed-rail + panel title, PanelGroup widget titles) up one step.
Channel info and the manager tables were already at 14–16px (readable), so they were left as-is.
The floating-header → toolbar → filter-chips → content stack carried ~2× the whitespace it
needed. Halve it, app-wide where the chrome is shared:
- PageShell: the --hdr-h gap below the floating header 8px → 4px (every page's header).
- Feed: toolbar wrapper pt-4 → pt-2, the toolbar's own pb-3 → pb-1.5, and the grid's top pt-2
→ pt-1. Measured on the feed: header→toolbar 24→12px, toolbar→chips 12→6px, chips→cards 12→8px.
- Plex / Playlists / ChannelDiscovery / DownloadCenter / Notifications / AdminUsers: toolbar
wrapper pt-3 → pt-2, so every page's top band matches the tighter feed.
PageShell.dom.test updated for the 4px gap.
Rebuilds SortControl's internals (API unchanged, so all five consumers — feed, Plex,
playlists, and the two table/card ColumnSortControl sites — get it for free) as a labeled
"Label: key ↓ ▾" pill on the AnchoredPopover platform, matching FilterSelect. The listbox
lists the sort keys; an asc/desc segment sits in the panel footer (with the trailing slot —
feed reshuffle, playlists dirty-first — beside it).
Two-dimensional key+direction, so it stays its own component rather than reusing FilterSelect:
the trigger is a single plain button (no nested button), picking a key keeps the panel open so
key AND direction can be set in one visit, and picking "no sort" (tables' defaultLabel) closes
it. Directionless keys (shuffle) hide the segment and drop the trigger arrow.
Drops the now-redundant external "Sort" label spans in Feed and PlexBrowse (the trigger carries
the label now). Rewrites SortControl.dom.test for the popover UI.
Replaces the native <select> "Source" control with a labeled FilterSelect pill, matching
the Watch state / Content types pills — one consistent toolbar idiom. Single-select over
the three librarySource values (organic / all / search-only).
Sort and View are intentionally left as-is: Sort is a two-dimensional key+direction control
(plus reshuffle and a tables-only "no sort") shared across five surfaces, and View is already
on the same AnchoredPopover platform with an icon-only trigger tuned for this packed toolbar —
FilterSelect's text trigger would regress it. Neither fits FilterSelect's one-dimensional model.
Adds a reusable FilterSelect: a "Label: value ▾" pill that opens a listbox (a checkmark
row for single-select, a checkbox row for multi), on the AnchoredPopover platform. It
replaces the two feed-toolbar chip walls:
- Watch state (single-select, default All) — picking a row replaces + closes.
- Content types (multi-select, default Normal; requireOne keeps at least one on) — rows
toggle and the panel stays open.
The trigger summarises multi picks as "First +N". The component is source-agnostic, meant
for reuse anywhere a small known option set currently renders as a chip group.
FilterSelect takes a `testId` → the trigger gets it, each option gets `${testId}-${value}`,
so the feed/channel E2Es keep stable selectors. They now open the dropdown and assert the
option's aria-selected instead of the old inline aria-pressed chips (new expectContentType/
toggleContentType helpers).
Replaces the hand-rolled channel search list and the flat tag chip walls with the
searchable, ARIA-correct Combobox built for the Plex filter bar (E5), so the YT feed
filters get the same input+listbox+chips UX with facet counts.
- ChannelPicker now wraps Combobox (selectedLabels keeps picks named); drops its bespoke
input + scroll-fade list.
- Sidebar language/topic/tags: each Combobox owns only its slice of the shared
filters.tags list (setTagSelection replaces that list's ids, leaves the rest). Topic
keeps its any/all match toggle, Your tags keeps Manage. TagChip removed.
PageShell is reused (never remounted) across feed↔channel navigation — same
component position in App.tsx — so a one-shot useLayoutEffect kept its
ResizeObserver bound to the FIRST header node. On channel-open that node detached
and fired the observer with a 0 box, rewriting --hdr-h to 0+0+8=8px on the
still-mounted content div. Back restored the pt-(--hdr-h) pad class with that
stale value, sliding the content toolbar up under the header — stuck until reload.
Hold both the header and content nodes in state (callback refs) and key the
measurement effect to the live nodes, so it re-runs every time the header
mounts/unmounts; drop the stale pad when the header is absent (channel page).
Adds a regression test: header removed then restored must re-measure, not 8px.
Full-epic /deep-review (11 finders + 12 adversarial verifiers) surfaced no HIGH; every
mechanism finding calibrated to LOW on verification. Fix all 13:
Correctness / reliability:
- Combobox: Enter now commits the VISUALLY highlighted row (highlightIndex ?? activeIndex),
not the bare keyboard activeIndex — so arrow-then-hover-then-Enter no longer toggles a
different option than the one painted, and hover-with-no-arrow is no longer a silent no-op.
- parse_person: dedup the known_for pool by (id, media_type), not the bare id — TMDB movie
and tv id spaces are independent, so a same-id movie+show pair is two distinct credits
(matches the collision-safe key the rest of the module already uses).
- person-enrich: a per-person in-flight + 60s cooldown guard (people._should_schedule_enrich)
dedups concurrent opens and rate-limits re-triggering while TMDB is degraded (a transient
TmdbError leaves the watermark NULL), so bursts can't pile 8s tasks onto the shared threadpool.
- PlexInfo compact()/money(): format with i18n.language (keeping currency USD) instead of a
hardcoded en/en-US, matching the app's Intl convention.
- delete the dead `plex.filter.add` i18n key (both locales) — the +Add button that used it is gone.
Reuse:
- _continue_watching reuses rollup.agg_subquery(include_resumed=True) + query.apply_show_state_filter
instead of a third hand-inlined copy of the per-show rollup + in_progress predicate, so the
Discover rail can't drift from the grid's definition.
Test coverage (the gaps a regression would ship through):
- resolve_library_guids scope-exclusion (a credit in a non-scoped/disabled library is dropped)
+ the []-vs-None distinction; parse_person cross-media-id retention; enrich_person transient
(NULL watermark) and 404 (stamp) branches; _enrich_person_bg body (fresh session close/rollback,
None-guard) + the cooldown dedup; continue_watching fully-watched-show exclusion; Combobox
Enter-commits-highlighted + async/controlled-query mode; PopoverItem aria-checked contract;
RatingsRow missing-IMDb + imdb==audience branches.
Backend lane 413→420, FE vitest 351→357; all gates green. No migration.
The last of the deep-review deferred reuse items:
- extract the scroll-affordance machinery (watch the scroller AND its swappable/
lazily-growing children, diff two edge booleans) that useScrollFade (vertical) and
Carousel's useHScroll (horizontal) each hand-duplicated into one shared
lib/useScrollEdges(axis); both hooks now just map start/end to their own shape.
- migrate PlexInfo's two remaining hand-rolled `flex overflow-x-auto` strips — the
cast row and the collection strips — to the S1 <Carousel> primitive, so they get the
edge fade + hover/focus scroll buttons (only on the clipped side) for free, matching
the Discover rails / CastStrip / RelatedStrip.
Pure refactor, no behaviour change. Gate green (tsc, vitest 351, eslint, prettier,
knip). Browser-verified on a movie info page: cast row + collection strip render as
Carousels with the edge fade and the right-scroll button on the clipped cast row.
Address the deep-review's deferred backend items so E5 is review-complete:
- person-enrich OFF the request path (ABUSE finding): person_detail no longer makes
the TMDB call inline (even bounded to 8s it sat on the serve thread, un-throttled,
demo-reachable — it violated the module's "no live TMDB call at serve time"
invariant). It now schedules a FastAPI BackgroundTask (people._enrich_person_bg,
fresh session, best-effort) after the response; the first open serves the row as-is
and the bio lands by the next open. The inline try/rollback recovery is gone with it.
- reuse: a shared tmdb.resolve_library_guids owns the collision-safe guid→library
bucketing (movie vs tv ids are independent sequences) that resolve_recs and
people._resolve_known_for each hand-duplicated; a new rollup.item_state_map dedups
the "PlexState-by-item_id" fetch across the person + discover paths.
- test gaps: enrich_batch transient TmdbError leaves the watermark NULL to retry;
guid_to_tmdb_id resolves a tvdb source + tv media_type via /find's tv_results;
_continue_watching includes a NULL-status started movie and excludes a hidden one;
/api/search/youtube returns 422 (not 5xx) when ingest fails (route-level, B-core-1).
Backend lane 403 → 413, ruff clean. No migration; serve contracts unchanged.
The E5 filter pickers were a hand-rolled floating-ui combobox (MultiSelectChips)
with no combobox ARIA, no arrow-nav, and options in a body-portal — keyboard/SR
users could not operate it, and the bare role=listbox/option wiring actively
misled screen readers (deep-review A11Y-1, CONFIRMED 3/3 medium). Rather than
one-off-hack the a11y, add combobox as a first-class mode of the AnchoredPopover
platform and rebuild the pickers on a reusable component.
Platform (AnchoredPopover):
- new mode "combobox": reuses the whole positioning core (offset/flip/shift/size/
hide, isPositioned + transform-clobber flash guards, FloatingPortal + fullscreen
re-parent, topmost-only Escape layer). Differs only in the focus model — no
useClick (open is field-driven), useRole role="combobox" so Floating UI wires the
full ARIA tree (input role=combobox + aria-expanded/controls/autocomplete/
activedescendant, listbox surface, option ids), and useListNavigation virtual:true
so ↑/↓ move an aria-activedescendant pointer while DOM focus STAYS on the input.
- PopoverSurface: combobox surfaces host the item registry but are NOT focus-trapped.
- PopoverItem: virtual mode carries a stable id (the activedescendant target), never
takes roving DOM focus, keeps the input focused on click (mousedown preventDefault),
and its checked/selected truth wins over getItemProps' single-select guess.
Component (Combobox.tsx, new): the universal react-select/Overseerr combi-input —
chips-in-field + typed search + checklist. Supports multi/single select, client OR
server (async) filtering, custom rows, the big-vocabulary render cap + refine footer,
and a selected-label cache so a server-narrowed single-select still names its chip.
Rollout: MultiSelectChips (genre/age-rating/studio) → Combobox; the CollectionPicker
"+Add" popover → a single-select async Combobox (also fixes its arrow-keys-eaten
story), unifying the two filter idioms. PlexSearchBox is a documented future consumer.
Gate green (tsc, vitest 351, eslint, prettier, knip). Browser E2E on localdev:
role=combobox confirmed on both pickers; ↑/↓ moves aria-activedescendant with focus
staying on the input; Enter + click select; multi-select chips accumulate; Backspace
removes the last chip; async collection server-search picks + closes. No backend change.
Three verified deep-review findings (the 4th, the filter combobox a11y gap, is
deferred to a dedicated session on the combobox architecture):
- SEC (medium): TmdbClient sent the api_key as a URL query param, so an HTTP
error's httpx exception string embedded the full URL incl. the key, which
enrich_batch logged. _get now raises status+path only (never str(e)/the URL)
and severs the secret-bearing cause with `from None`. Regression test asserts
neither the status nor the network error path contains the key.
- STATE (low): person_detail's lazy-enrich except now db.rollback()s, so a
failed enrich_person commit can't poison the session and 500 the subsequent
SELECTs (SQLAlchemy 2.x deactivates the tx on a commit-time flush failure and
does not auto-reset it). Test locks that the page renders despite a poisoned
enrich.
- ABUSE (low): the request-path lazy person enrich now caps its TMDB client at
an 8s timeout (was the 30s default) so a slow TMDB can't tie up a worker;
TmdbClient gained a `timeout` param (batch keeps 30s). The fuller fix (move
person enrichment off the request path) is logged for later.
The combi-input popover flashed in the top-left corner for a frame, then
jumped into place. Two causes, both confirmed with a per-frame capture in
the browser:
1. The panel painted before floating-ui computed a position — gated its
visibility on `isPositioned` (hidden until positioned), same as
PopoverSurface.
2. The bigger one: the `popIn` entrance animation animates `transform`
(scale/translateY), and it sat on the SAME element floating-ui positions
via `transform` — so during the 0.16s animation the keyframes OVERRODE
the translate(x,y) and the panel rendered at the origin. Moved the glass
styling + popIn onto an inner child, leaving the positioned element's
transform untouched.
Frame capture after the fix: frame 0 hidden, first visible frame already at
translate(82,474) — never visible at the origin. Also gave the size
middleware a min width for a narrow rail.
The AnchoredPopover-based version had three UAT-caught bugs: the field
showed an ugly "Add" placeholder; the first click couldn't type (the
platform's FloatingFocusManager + focusItemOnOpen pulled focus onto the
list, blurring the field until a second click); and an outside click didn't
close the list. A visual pass also caught the dropdown rendering BEHIND the
sidebar panel (z-menu 20 < z-panel 35).
Rebuilt on floating-ui primitives directly — no focus manager, so focus
stays in the field; open is driven by field focus; useDismiss closes on
outside-press/Escape; the list is a z-popover (60) FloatingPortal sized to
the field width. Placeholder is now a plain "Search" (empty once chips
exist). Verified in the browser: type-on-first-click, filter, chip add,
outside-close, and the list painting above the sidebar.
User-reported issues after the E5 build:
- useCrossfadeBackdrop: the static branch (one image / crossfade off /
reduced motion) painted #root but returned NO cleanup, so the backdrop
persisted after leaving the page (e.g. switching modules). Every painted
branch now returns clear() so unmount always resets the default backdrop.
- scheduler: give the tmdb_enrich job a proper label + description (was the
raw "tmdb_enrich" id) in EN/HU.
- CastStrip: cap crew names to ONE line (truncate) so the row of circular
photos sits on one baseline instead of going wavy on 2-line names.
- PlexInfo: lay Details / Keywords / Crew out as side-by-side boxes
(items-start grid) instead of full-width stacked panels with dead space;
narrower Crew grid for the boxed width; each panel independently hideable
from the customize menu (new Details/Keywords/Crew toggles). Tighter
vertical spacing.
- MultiSelectChips: reshape the trigger from a "+Add" button into an
Overseerr-style combi input — removable chips inside a typeable field that
filters the checklist, opening on focus (backspace removes the last chip).
- Discover / person rails: reduce the vertical gap between rails.
Follow-ups from the epic-end /code-review of the E5 branch:
- tmdb.resolve_recs + people._resolve_known_for: match recommendations/
credits to the library bucketed by media_type. TMDB movie ids and tv ids
are independent sequences, so a bare `tmdb:<id>` match could deep-link a
movie rec to an unrelated show (or vice-versa) on a numeric collision.
Regression test added (test_resolve_recs_matches_by_media_type_not_bare_id).
- routes/search: wrap the _ingest_candidates call so a DB/enrich failure
degrades to a 422 with a message instead of a bare 500 hidden behind the
FE's generic "search failed" (extends the B-core-1 rationale); rollback on
failure. Stamp enriched_at with its own timestamp taken after the enrich,
not the stub-insert placeholder.
- routes/plex/discover: key serialized cards back by rating_key (globally
unique in Plex) instead of id(obj); fix the stale rows 3-tuple annotation.
peter