Merges master and resolves the collision with its shared playback helpers,
then closes two Codex findings.
Master extracted the Play/Stop button state and the inline position writer
that this branch had modified. Rather than forking private copies back out of
Xtream, both behaviours move into the shared helpers: `alsoOwns` lets a page
own an external session launched for a copy of the same film in another
playlist, and the resume latch — which stops a timeupdate emitted before the
engine reaches `startTime` from overwriting the point being resumed from —
now protects Stalker too, which had the same bug.
Auto-failover was offered on every engine, but only the built-in web players
raise the playback diagnostic that reaches `onPlaybackFailed()`: Embedded MPV
has its diagnostics suppressed and MPV/VLC play outside the app. The toggle is
now hidden there, in settings and in the sources menu, instead of promising a
switch that can never happen.
`setAutoFailover` also ignored `updateSettings()`, which patches memory first
and rejects if the write fails — the toggle looked saved, silently reverted on
restart, and the rejection surfaced only as an unhandled promise. It now
reports the failure like the settings form's own save paths.
The three VOD-details route specs each carried a near-identical 150-line
TestBed; they now share one harness, which is what makes room for the new
cases (1012 -> 584 lines).
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Master had moved ten commits ahead. Two conflicts, both resolved without
losing either side: the `XTREAM_PROBE_URL` handler this PR extracted into
`events/stream-probe.ts` stays extracted (master added performance capture
nearby but never touched that handler), and the mock-server scenario table
takes the union of master's `performance` row and this branch's `multisrc`
pair.
Two findings fixed alongside it.
Pinning the copy the route is already on makes discovery return that very
row, so prepending the current source listed one stream twice — a phantom
copy in the grouping and a chip that counted it.
And handing the "playing" badge back to the route row left an in-flight
switch valid, so a slow resolution could arrive afterwards and replace the
playback the user had just started with Play, Resume or Restart.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Carrying a position into a shorter cut of the same film — two hours into a
90-minute source — leaves the engine unable to ever report that time, so the
one-shot latch never released: every position save was suppressed for the
rest of the session, and the impossible start time kept being reported to
multi-source for the next switch.
The latch now also opens when a known duration puts the requested point out
of reach, while a reachable one still waits as before.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Two follow-ups to the previous round, both in code it introduced.
A width that matched exactly one sub-HD format ignored the height entirely,
so 640x480 came back as 360p — a measurement the numbers contradict. The
height now vetoes, but only in the direction that can be wrong: cropping
removes lines, so a SHORTER frame is a letterboxed master of that format and
the width still names it, while a taller one is a different shape and gets
no tag. That keeps the reason width is preferred in the first place.
And picking a source off the error screen cleared the failure state before
the switch resolved, so an alternative that could not be resolved left the
diagnostic on screen while the caption went back to claiming playback. The
flag now clears only once a switch actually starts something.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Two follow-ups to the previous round, both the same rule again.
The 800-wide band still answered from the width alone, so 800x600 and
800x450 were labelled 480p — published with `api` provenance, so read as a
measurement. Sub-HD formats are now matched against known shapes with the
same 5% tolerance the height path uses, and anything unrecognised carries no
tag at all.
And "Playing from ..." survived a playback failure: the inline host stays
mounted while the diagnostic is on screen, so the page named a source for a
stream it had just reported it could not play. The caption now clears on
failure and returns when the engine produces time again.
Splits the route playback spec along the "what it does" / "what it claims"
seam and lifts the repeated active-source stub into one helper.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Three findings, all of them the same rule: never state as fact something the
app has not established.
The "Playing from …" caption appeared as soon as discovery marked a source
active — before Play was pressed, and again after the player was closed. It
now requires a player that is actually running.
Probe answers were cached by URL alone, but the request now carries the
playlist's headers. Two playlists sharing a stream URL could therefore be
told the other's answer, marking a source dead without ever asking it.
And any width below 900 was labelled 480p, published with `api` provenance:
a 640x360 stream stated 480p as a fact, and a 720x576 PAL source likewise.
Widths below HD only resolve with the height — 720 is NTSC 480p or PAL 576p
— so an unrecognised shape now carries no quality tag at all.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Three findings from the latest review pass.
Repinning cleared the old rows and then wrote the new one, so a write that
failed after the clear left nothing persisted while the row still showed the
old pin. The order is reversed: the new key is stored first and the stale
ones retired only once it landed. Lookups are most-trusted-first, so a
leftover alias never outranks what was just written.
Starting a source from the picker, or letting a pin decide the primary Play,
never recorded the movie as recently viewed — unlike every other way of
playing it.
And closing an alternative's player and pressing Play started the route
stream while the controller still marked the alternative active, so the
picker and caption named a source that was not running.
Moves the discovery pass into the session module beside the switch and
failover mechanics, and splits the pin spec along the persistence/playback
seam, both to stay inside the file-size rule.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
A same-movie rediscovery read the pin, then held that snapshot across its
source lookup and applied it afterwards. A pin made while the lookup was out
was therefore overwritten by the older value: the row and the primary Play
action named a source the database no longer held.
The snapshot is now applied as soon as it is read, so a later write simply
wins on ordering rather than needing to be detected.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Five findings from the latest review pass.
Reading a pinned source's own position is a database round-trip, and the user
can navigate across it — the continuation then handed one film's source id to
whichever movie now owned the screen. Guarded, like every other await here.
Allowing a pinned copy to live in the current playlist made "is this the
route's own source?" a two-part question, and the ownership check still asked
only about the playlist: an external session for that copy was disowned, so
Stop vanished and its progress was dropped.
The yearless title alias is shared by every remake, so clearing every alias
before a write could delete a different film's pin. Writes and unpins now
touch only keys that name one film — plus the ambiguous row this session
actually read, which is the one the user is looking at and the one whose
absence would make an unpin come back.
Restart left the seeded position in the controller, so a failure before the
first timeupdate resolved the next source back at it.
And the alternative rows on the playback-error screen had a Check button
wired to nothing at all.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Three findings from the latest review pass.
`normalizeTitleKeys` strips bracketed segments as tag noise, so "Dune (1984)"
normalizes to exactly "dune" — an EXACT match for the 2021 film, ranked above
every fuzzy one, with the year never consulted because that tier skipped the
gate. Auto-failover could switch the user to the other film entirely. The
year is now read out of brackets too, and a stated disagreement rejects the
row on either tier.
Playback positions are keyed by (playlist, stream), so watching through a
pinned alternative stores progress under ITS ids while the page loads the
route copy's row. Starting the pin therefore resumed from a position
belonging to a different copy — usually zero. It now loads its own.
And a pin can point at another copy of the film inside the playlist being
viewed, which discovery excludes wholesale: the pinned row was absent from
the list, so nothing showed as pinned and Play ignored the preference. The
pin is now read before discovery, which keeps that one row.
Moves the pin-shaped decisions into the pin module, where the persistence
helpers already live.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Three findings, two of them regressions from the previous round.
Awaiting a pending discovery before failover let the user navigate during
that wait: the continuation then ran against whatever controller was current
and could answer one film's playback failure by starting another film's
alternative. Both waits — failover and pinned Play — now re-check that the
same movie still owns the screen.
Pinned Play also needed the wait it did not have. Pressing Play while the
pin lookup was still out concluded "nothing is pinned" and started the
route's own source, making a persisted preference depend on worker latency.
And the position bridge still accepted only the route's ids, so an external
player running an alternative had every progress update discarded: the
resume point stayed where playback began and a switch an hour in rewound the
lot. The session matcher and the bridge now share one ownership predicate,
since a page that shows Stop for a session whose progress it throws away is
the bug in two halves.
The test for the external case previously set the position signal directly,
which bypassed the very filter that was broken; it now drives the bridge.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Five findings from the latest review pass.
Writing a pin to every alias — last round's fix for stale aliases — was
wrong in the other direction: `title:{base}:` is shared by every remake, so
a known-year decision stored there answers for a different film. Pin Dune
(2021), open Dune (1984) before its year arrives, and it would start the
2021 source. A write now clears every alias and stores only the canonical
key, which retires the stale ones without making any of them ambiguous.
The probe checked a bare URL while playback sends the playlist's User-Agent,
Referer and Origin. A panel that requires them answers 401/403, so a stream
that plays perfectly was reported dead and penalised in failover ranking.
The switch toast interpolated the raw playlist name. Users routinely name a
playlist after the URL they pasted, so that line could put credentials over
the video; the notice now carries the same safe label the rows use.
External players have no timeupdate, so their polled position IS the live
one. Feeding it through the seed — which stops at the first value — froze
the resume point where playback started, and a switch an hour in rewound to
the beginning.
And auto-failover concluded "nowhere to go" when a stream failed before
discovery answered, stranding the user on the error screen.
Moves `switchTo` into the session module, which is where the rest of the
switch mechanics already live, and splits the route spec along the same
rendering/behaviour seam the other suites use.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Four findings from the latest review pass.
A pin lookup accepts several aliases of the same movie, but a write only
touched the most-trusted one — so after enrichment the title alias still
pointed at whatever was pinned before, and a reopen that read it (because
TMDB had not landed yet, or its request failed) started the source the user
had just replaced. Writes now go to every alias.
That alias set was also missing one. Enrichment supplies the year as well as
the id, so a pin set before either existed is stored yearless; the candidate
list skipped that form entirely and orphaned the row.
Discovery could lose whole playlists: one playlist listing a film in dozens
of categories produces identically ranked rows that fill the window before
another playlist is read. The collapse now happens in SQL, before the limit,
rather than in TypeScript afterwards where the missing rows are already gone.
And an abandoned source pick finishing late cleared the spinner from the row
the user was actually waiting on.
Removes `isExhausted()` from the host service — no caller outside its own
tests, where the assertion above it already proved the same thing.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Five findings from the round-5 review.
An external player launched for an alternative carries that playlist's ids,
so the page disowned its own session: the primary button never became Stop,
stopping found nothing to stop, and another click opened a second player.
Multi-source now tells playback which source is actually active, and the
matcher accepts either that or the route's own stream.
Stop also has to beat the pin. The primary action consults the pin first —
that is what makes a pin decide where playback starts — but while a session
is running the same button reads Stop, and consulting the pin there made the
control do the opposite of its label.
A pinned source started from the Resume button resolved at zero, because
nothing reports a live position until the first timeupdate. The controller is
now seeded from the persisted position, one-way: a live value always wins,
since the stored one lags it and applying it would rewind.
A pin whose write failed was still shown as pinned, promising a preference
that reopening the movie would not have.
Portal failures in this path logged raw errors. An Xtream error message
carries the stream URL, and that URL is built out of the username and
password, so they now go through the redacting logger.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Two findings from the review of the previous round.
A pin write is an IPC round-trip, and the user can navigate during it. The
continuation then applied one film's answer to another film's controller —
and because unpinning returns "nothing pinned", it would clear the pin the
new movie had just loaded and its Play action would quietly stop starting
from the preferred source. It now commits only while the same film is still
on screen, like every other async path here.
The short-title scan drops its row limit. FTS keeps its window because it
ranks by relevance, so what it keeps is what matters; a scan cannot rank, so
a window there silently decides which valid sources the user is allowed to
see. It also bought nothing: the GLOB cannot use an index, so SQLite reads
every row either way and the limit only truncated the answer. What bounds
the scan is its predicate — reaching it means the whole title is one or two
characters.
The switch-notice type moves to the module that builds it, which also
removes a circular type import between the two.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The packaging verifier bounded `iptvnator_mpv_helper --runtime-probe` with
RUNTIME_PROBE_TIMEOUT_MS (3s) — a constant it shares with the application's own
startup capability gate. Three seconds is a tight budget for a helper that
dlopens libmpv plus EGL/GL/GBM, and the Flatpak profile is closest to that edge
because the helper runs inside the sandbox against its bundled closure: on
#1277 the job failed three consecutive reruns and passed on the fourth with no
code change, while the concurrent master job passed.
Give the verifier its own budget rather than raising the shared one. The app's
probe is a blocking spawnSync on the Electron main process, so a hung helper
must not stall window creation, and a timeout there degrades gracefully to the
native-view fallback. Nothing waits on the packaging probe but the CI job,
which already has its own 120-minute bound, while a premature kill reports a
healthy package as broken.
- PACKAGE_VERIFICATION_PROBE_TIMEOUT_MS (15s) and
PACKAGE_VERIFICATION_PROBE_MAX_ATTEMPTS (2) join the frozen probe contract;
RUNTIME_PROBE_TIMEOUT_MS stays at 3s for the application gate.
- runBoundedRuntimeProbe() retries only on ETIMEDOUT, repeating the identical
bounded launch (same command, args, env, maxBuffer, killSignal) and
announcing the retry on stderr so a degrading trend stays visible.
Fail-closed behaviour is unchanged. A hard timeout is the one probe outcome
that says nothing about the payload; spawn errors (a missing helper, a wrapper
launched instead of the real ELF), termination by signal, nonzero exits and
malformed or wrong-protocol lines all still fail on the first attempt, and a
helper that keeps hanging still fails once both attempts are spent.
The four new/extended verifier tests cover retry-then-success (asserting the
second launch is identical to the first), exhausted timeouts still rejecting,
four non-timeout verdicts each probing exactly once, and the attempt bound
itself. Setting MAX_ATTEMPTS to 1 fails four of them.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Follow-on from keeping the session across a rediscovery. The rerun can
legitimately drop the row that is playing: enrichment supplies the release
year, and the year gate then rejects a copy the yearless search had admitted
— "Dune" 1984 while the user is watching the 2021 film.
Off the list is right; it is not the same film. Off the screen is not. It is
what is streaming, so it stays as a row and keeps the playing badge, rather
than letting the caption name a playlist that is not sending any bytes.
Also covers the new session key directly in the identity spec.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Six review findings, all in how multi-source decides what to show and what
it is playing.
Discovery: the current playlist is now excluded in SQL rather than after the
fact, so its own duplicate rows can no longer spend the whole row budget
before a single alternative is read. The short-title scan matches the token
as a word instead of a substring and orders by title length in a wider
window, so "Titanic" and "The Italian Job" cannot push the real "It" out of
it.
Session: metadata enrichment re-runs discovery for the film already on
screen. That is a refresh, not a new session — a second identity key
(playlistId:contentId) now separates the two, so the source the user
switched to keeps playing and stays named, the tried set stays burned, the
position survives and a switch in flight still commits.
Resume: the multi-source controller no longer records the engine's pre-seek
timeupdate at ~0. The playback service's one-shot latch now reports whether
the position can be believed, and until it can, the requested start time
stands in — so a switch during the initial seek does not restart the film.
UI: the in-player sources picker gets the same auto-failover setting and
match kind as the detail page's, instead of always rendering the default and
dropping the toggle. The caption counts distinct playlists, not stream
variants, since the popover groups a portal's copies under that portal.
Session mechanics and the pin toggle move into their own modules to keep the
host service inside the line budget.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The architecture doc and CLAUDE.md described the feature as first written, not
as it now behaves: pins were documented as a stored preference without saying
they decide playback, failover was described as stopping at the first
unresolvable candidate, the probe as HEAD-only, and discovery as pure FTS with
no mention that short titles cannot be tokenized at all.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
A movie that exists in several imported Xtream playlists now shows a
"Sources N" chip on its detail page and in the player. Switching playlist
mid-film keeps the timecode, a preferred source can be pinned per movie, and
a failed stream offers the alternatives instead of a dead end.
The governing rule is that a guess is never presented as a fact. Every
metadata value carries where it came from — `api` (the provider said so),
`parsed` (inferred from the title) or `probe` (we contacted the stream).
Facts render as plain tags, guesses are prefixed `~` in a warning colour, and
an unknown value renders no tag at all plus a "check" affordance. Ranking and
failover read through `factualOnly()`, so a filename claiming 4K is
structurally unable to outrank a source that was actually reached. A probe
that could not complete reports "unknown", never "unavailable".
Scope is deliberately narrow: Xtream to Xtream, movies only, Electron only.
Stalker never reaches the `content` table and M3U is a JSON blob whose search
forces live content; both are additive later, since the candidate type
already carries all three portal kinds. In the PWA every entry point is gated
off and the chip renders nothing.
Auto-failover is opt-in and off by default. Each source is tried at most once
per session, so it terminates structurally, and the switch is never silent —
the toast names the new playlist, offers an undo, and warns that the dub may
differ only when both sides state an audio track as fact.
Notable details:
- Playlist names are routinely the pasted URL, credentials included. They are
never rendered raw; a short host-only label is derived instead.
- Quality is derived from pixel width, not height: a 2.39:1 1080p master is
1920x800, and bucketing that by height would publish "720p" as a fact.
- Switching is a single `inlinePlayback.set()` so the player and engine
survive and re-seek; the carried position is read before the 15s
persistence throttle so it does not rewind.
- Sources from one playlist collapse into a group, since the same film often
appears there several times under different stream ids.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Moves the fictional movie catalog into `libs/shared/marketing-fixtures` so the
Xtream and Stalker mocks describe the same titles, and adds 20 rendered posters
plus the shared fixture types behind them.
Supporting changes made while getting it green:
- `shared-marketing-fixtures` is classified Tier B in the coverage policy. Not
Tier A: it is fictional fixture data, so a statement percentage over it means
nothing, and a Tier A entry would pull it into the merged coverage map and the
ratchet. Tier B still runs its spec in CI. `stalker-mock-server` needs no entry
of its own — it is already Tier C and the Tier B/C runner falls back to
`pnpm nx test <project>`, so its new `marketing-poster-url.spec.ts` runs.
- Two release-capture defects the catalog reorder introduced, both fixed in
`tools/release/capture-app-driver.ts`:
- VOD stream ids are `MARKETING_VOD_STREAM_ID_BASE + index` and the generator
now lists the showcase movies first, so 62000-62002 became Black Harbor, The
Paper Astronaut and Summer Static while the dashboard seeding still mapped
those ids to the previous titles' backdrops.
- the raw `tsx` spawn of the Xtream mock lacked `--tsconfig
tsconfig.base.json`, so the mock could not resolve
`@iptvnator/shared/marketing-fixtures` and the capture never started. Both
mock projects' own serve targets already passed the flag.
Split epg.events.ts (514), the embedded MPV frame-copy adapter (428) and two of its specs (547, 539) below the 400-line hard limit, and teach the baseline generator to skip files that already carry a justified file-wide eslint-disable max-lines.
The generated baseline list is unchanged: 128 entries before and after. No behavior change.
settings.component.ts had grown to 819 lines — past the CLAUDE.md target (<300)
and hard maximum, passing lint only because it sat in the max-lines baseline.
The behaviour moves into facades the template binds to directly, following the
precedent already in this folder: new app-update (218), form (197), epg (123),
embedded-mpv (74) and remote-control (37) facades, with playlist-reset extended
to 143 and settings-options to 200. The component is now a 259-line coordinator
holding capability flags, section nav, players() and the cross-facade flows.
settings.component.ts is removed from the max-lines baseline.
No behaviour change. One ordering detail: applyChangedSettings now applies
language/theme before kicking off the EPG re-fetch; changeTheme only touches DOM
theme sync and translate.use does not touch the form, so the two are
independent.
Adds contract-focused regression coverage for the Electron HTTP server,
remote-control events, settings events, and managed download paths, and makes
Tier A coverage fail closed when instrumentation fails or a runtime-owning
production file disappears from a project or from the merged Istanbul report.
The old `coverage:ci` exited 0 despite a `Failed to collect coverage`
diagnostic: libs/m3u-state/src/lib/effects.ts was simply absent from the merged
map. All 30 Tier A reports are now required, the merged map covers 710 files,
and effects.ts is reported as 0/159 instead of silently disappearing.
Also fixes remote static-file path containment for encoded, malformed, NUL,
POSIX and Win32-style traversal inputs, with behavior-preserving testability
seams.
Statements 69.27% -> 69.54%; http-server.ts 0% -> 90.21%,
remote-control.events.ts 0% -> 96.55%, settings.events.ts 59.25% -> 96.29%.
Supersedes #1250, whose 22 -> 26 bump failed the image build: `corepack enable`
exits 127 because Node 25 unbundled Corepack.
Both stages move to node:24-alpine, the current LTS line — Node 26 stays
Current until October 2026, which is the wrong target for a self-hosted runtime
image. pnpm is installed globally at the exact `packageManager` version, with
the `+sha512...` suffix stripped, so the next base-image major is a one-line
change instead of a broken build.
Rebuilt from #1251 so the group could merge, on top of the transitive-CVE
overrides from #1258. Supersedes #1230 and #1251.
Carries axios 1.16.0 -> 1.18.1, closing seven runtime-scope advisories
including the proxy-credential leak on redirects, and sharp 0.34.5 -> 0.35.3
for the libvips CVEs.
`esModuleInterop` moves to tsconfig.base.json. artplayer 5.4.0 switched from a
Parcel build exposing `module.exports.default` to UMD assigning
`module.exports` directly; the flag was only set in apps/web, so every lib
compiled `import Artplayer from 'artplayer'` to `.default` and got undefined.
Production was never affected — esbuild resolves the ESM entry.
Two packages are deliberately held back, each for its own PR:
- epg-parser ^0.5.0 — grouped as a minor, but 0.x minors are breaking and this
one reshapes the parse output (`channel.name` -> `displayName`, icons/urls
become objects, `credits` becomes role-keyed, dates switch to ISO). Its only
consumer is the uncovered web-backend `/parse-xml` endpoint.
- electron-builder ^26.15.3 — rewrote the snap target, and the resulting snap
cannot start (`command.sh` execs a `desktop-init.sh` that never lands at the
snap root under our core22 strict config). Its two required fixes go with it:
the `engines` node floor from @electron/rebuild 4, and resolving upstream
node-gyp instead of the dropped `@electron/node-gyp` fork.
The persisted subtitle preference only ever had an owner behind the default-off
shared web-controls flag, so with the shipping controls it did nothing: Video.js
never read it, ArtPlayer declared the input but never used it, and the HTML5
player only ran a one-shot pass after play() resolved — before hls.js had added
its text tracks. No portal host bound the input at all, so it never reached
Xtream or Stalker pages either.
Extract the source-local track controllers into the adapter-free
WebVideoSourceTracks and have WebVideoSourceControlsBridge wrap it, so both
controls modes apply the preference through the same code. The preference-off
players bind it directly (VjsLegacyTracks for Video.js), and
WebPlayerViewComponent reads the preference from SettingsStore instead of an
input so every host inherits it.
The preference means different things depending on who renders the caption UI:
shared controls stay authoritative for the session, while vendor chrome is
source-default — the preference seeds each new source and is released once the
media reports playing, so the engine own caption menu keeps working. Mode
selection is an optional playbackStarted probe passed to the HLS, native and
Shaka helpers; in that mode the HLS helper deselects the track rather than
hiding it, since subtitleDisplay would silently override the vendor menu.
Closes#1155
Closes 13 runtime-scope Dependabot advisories that Dependabot cannot fix itself:
every vulnerable package here is transitive, so the bot has no lever until each
parent publishes a release widening its own pin.
Overrides added (pinned-source form, matching existing convention):
- @xmldom/xmldom 0.8.11 -> 0.8.13 (5 high) via video.js -> mpd-parser
- fast-uri 3.1.0 -> 3.1.4 (4 high) via electron-conf -> ajv
- js-yaml 4.1.1 -> 4.3.0 (2) via electron-updater
- form-data 4.0.5 -> 4.0.6 (1 high) via axios
- ajv 8.17.1 -> 8.18.0 (1) via electron-conf
Every target stays inside its parent's declared semver range. For xmldom,
fast-uri and js-yaml the newest published version is outside that range
(0.9.x / 4.x / 5.x), so "latest" would have broken them; the new doc
records that constraint.
Deliberately excluded: axios and uuid are direct deps already covered by open
Dependabot PRs (#1251, #1252). undici is labelled runtime scope but every path
to it is build tooling (electron -> @electron/get, @angular/build,
@module-federation/dts-plugin) and it is not in the packaged app.
Reachability: xmldom arrives via video.js -> VHS -> mpd-parser, but the app
routes every .mpd to Shaka, which uses its own DASH parser, so that one is
defence in depth. The genuinely reachable one is js-yaml, which
electron-updater uses to parse latest.yml from releases.
Adds docs/architecture/dependency-security-overrides.md and a .changes note.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
* fix(tmdb): series cast was the latest season only, not the show
TMDB documents a TV id's `credits` as the credits of the LATEST SEASON.
We requested exactly that and rendered it as "the cast", so every
long-running show lost every regular who had left: The Boys showed
whoever appears in the newest season, not the ensemble.
The TV details request now also appends `aggregate_credits`, which spans
the whole run — but per TMDB omits the newest season, so neither payload
alone is the cast. `unifiedTvCast` unions them: whole-run billing order
first, then people who appear only in the newest season, deduplicated by
person id. Characters come from the aggregate `roles[]` shape.
Deliberately NO cache-key bump. Rows cached before this simply lack
`aggregate_credits` and keep the previous behaviour until they expire,
which avoids invalidating every user's details cache twice — the roadmap
schedules one consolidated bump once the remaining append_to_response
additions (images, certifications, alternative_titles) land together.
Movies are untouched: /movie/{id} has no aggregate_credits and its
`credits` is already the full cast.
Tests: departed regulars retained, newest-season arrivals appended after
show billing order, characters read from roles[], no duplicates across
the two payloads, graceful fallback for pre-aggregate cache rows.
Refs docs/architecture/tmdb-roadmap.md A2.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* fix(tmdb): reserve cast slots so newest-season arrivals survive the cap
The union was appended aggregate-first and then truncated to ten, so on
exactly the shows it was built for — long-running ones, where the
whole-run cast alone exceeds the limit — every newest-season arrival was
sliced back off. The original fixture had two aggregate members and
could not catch it.
unifiedTvCast now holds back up to three slots for the top-billed
arrivals instead of appending them where the cap discards them, and
gives the slots back when nobody is new.
Tests: a 12-member aggregate plus two arrivals keeps both arrivals and
top billing; an aggregate with no arrivals still gets all ten slots.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* test(tmdb): split the series-cast suite out of the merge spec
The merge conflict resolution put both new describes back into
tmdb-merge.spec.ts, pushing it to 499 lines — past the 400-line
max-lines cap. The aggregate-credits suite moves to its own file.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* fix(tmdb): stop the cast union from shrinking, and bound what it caches
Three follow-ups from a review pass over the aggregate-credits union:
- The reserved arrival slots were subtracted from the aggregate even when
the aggregate was shorter than the cap, so a show with four regulars and
five newcomers returned seven names instead of nine. The reservation is
a floor for arrivals now, not a quota.
- An aggregate member's character came from the first role with any text,
so a one-episode cameo could outrank the part the actor is known for.
Pick the role with the most episodes.
- aggregate_credits carries a show's whole-run cast AND crew, and details
payloads are cached verbatim — orders of magnitude of JSON for a list
the merge truncates to ten people. Cache the billing-order prefix and
drop the crew nothing reads.
Extracting the people-related helpers into tmdb-credits.ts keeps
tmdb-merge.ts under the line cap.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* fix(tmdb): keep the aggregate ids the arrival check depends on
Trimming the cached cast to its top 40 broke the property it was supposed
to preserve: `known` is built from the aggregate ids, so a returning actor
billed below the cut read as a new arrival on the cached path and took a
reserved slot. The same show then showed a different top ten on its second
open than on its first.
Keep the whole cast, and cut the two things nothing reads instead: the
aggregate crew, and every `roles[]` entry except the one the merge picks
(most episodes). A merge over the trimmed payload now provably returns
what a merge over the full one does — covered by a test that runs both.
Also points CLAUDE.md and the doc's module table at tmdb-credits.ts, where
the credit helpers now live.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* docs(tmdb): add the release note for the series-cast fix
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* fix(tmdb): let the cache trim reuse the merge's own role choice
The trim picked the role with the most episodes; the merge picks the
NAMED role with the most episodes. TMDB uses unnamed roles for uncredited
appearances, so a member whose blank role outranked their real one lost
their character on every render after the first.
Both now call pickAggregateRole, which is the point — two copies of the
same choice are what let them drift.
Also adds a test pinning the property the earlier truncation defect broke:
the displayed cast is the cap or everyone available, whichever is smaller.
Which people make the cut at the cap is the reservation's job and is
deliberate; the count is not negotiable.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* docs(tmdb): state the aggregate-credits contract as TMDB actually words it
TMDB describes the endpoint in one sentence that contradicts itself: "it
does not return the newest season. Instead, it is a view of all the entire
cast & crew for all episodes belonging to a TV show." The doc and the code
comment asserted the first half as settled fact.
The union never depended on that reading — arrivals are a set difference,
so under "whole run" they are simply empty — but the comment implied an
assumption the code does not make. Say what TMDB says, note the ambiguity,
and note why either reading is safe.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
The custom minimize/maximize/close controls stayed hidden forever after
leaving HTML-element (video player) fullscreen on Windows: window state was
polled at event time, and isFullScreen() can still report the pre-transition
value while 'leave-full-screen' fires, leaving a stale push with no later
event to correct it. The same polling on the companion flag cleared
isMaximized during fullscreen transitions and stuck the maximize/restore
glyph on the wrong icon.
attachWindowStateEvents now seeds the state once at window creation and each
event patches only the flag it names, sending a copy per push. The
enter/leave-html-full-screen variants are wired too.
Regression coverage: app-window-state.spec.ts (9 cases, 6 of which fail
against the old implementation) and an Electron E2E case that toggles HTML
element fullscreen and asserts the controls come back.
#1261 replaced this one-off capture with a manifest-driven script, so the
v0.20 version is dead weight: hard-coded slugs, paths and output directory,
none of the fail-closed guards, and two `RegExp`-from-string constructions of
the kind CodeQL flags (one of which it flagged on the replacement before that
was rewritten to use predicates).
Removing it also drops its `tools/eslint/max-lines-baseline.mjs` entry, so the
baseline no longer carries a file that does not exist.
Not a pure dead-code deletion, and worth stating: two capabilities go with it,
neither reachable from the new pipeline — `createDesignedCopy` (title/kicker
overlays on captured frames) and `createHeroImage` (a 1600x900 canvas collage
built from three screenshots). The v0.20 assets they produced are already
committed under apps/website/public/blog/v0-20/, so nothing published breaks;
a future release wanting the same collage needs it ported deliberately rather
than resurrected here.
Docs: docs/architecture/xtream-mock-server.md now points at
capture-release-screenshots.ts and notes its mock-identity check.
Verified: no references to the removed file remain anywhere in the repo, and
`pnpm run lint` passes for all 42 projects with the shortened baseline.
Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
* feat(tmdb): metadata cache panel with a clear button in settings
Adds "Metadata cache — N entries · X MB" with a Clear button to
Settings > Metadata (TMDB), next to the API key it belongs to.
Three things it is good for: dropping stale or wrong metadata so the next
open refetches it, seeing what the cache actually costs on disk, and
reclaiming rows that a lookup-key version bump has orphaned — a bump makes
rows unreachable, not deleted, so nothing else would ever collect them.
Sizing the cache is a full table scan (LENGTH() on TEXT counts characters,
so the SUM casts to BLOB to get bytes), which is why stats load lazily and
only once the TMDB section is the active one rather than on every settings
open. Clearing is always safe: enrichment refetches on demand, so the only
cost is the next few requests.
Works in both environments — the PWA has no bridge, so the service reports
and clears its session-scoped in-memory map instead.
i18n: 4 keys across all 19 locales via the tools/i18n workflow;
placeholder integrity verified. Contract fixtures updated for both the
preload bridge and the DB-worker payload shapes.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* fix(tmdb): make cache clearing durable and stop reporting failures as empty
Four review findings, all real:
- A metadata write already in flight when the user cleared would land
afterwards and silently restore what they removed. Writes now carry the
generation they started in; a write that outlives a clear is dropped
(PWA) or undone (Electron).
- The PWA byte count used String.length, i.e. UTF-16 code units, so
localized payloads under-reported and disagreed with the SQLite BLOB
byte count. TextEncoder now measures actual bytes.
- A failed stats read returned a valid zero-entry result, so the panel
claimed an empty cache and disabled Clear while rows were still there.
getStats/clear now return null on failure and the panel says so instead
of inventing state.
- No behavioural coverage existed for either side.
Tests: SQL ops (entry/byte reporting, empty table, missing row, delete
count) and the service (encoded bytes, clear count, and a write racing a
clear).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* fix(tmdb): make the cache clear precise and version skew visible
Review follow-ups on the cache panel:
- A write that was in flight when the user cleared used to trigger a
second full-table clear once it landed, which also deleted anything
written in between. clear() now waits for the writes issued before it
and lets the single clear take them; later writes survive.
- An Electron shell without the maintenance ops fell through to the
renderer map, which is always empty there — it reported an empty cache
and disabled the Clear button while SQLite was full. Both operations
now report unsupported instead.
- Component coverage for the panel (deferred scan, clear + re-read,
failed clear, failed read) and Electron-path service coverage.
- The canonical IPC and settings sections of the enrichment doc, plus
the matching CLAUDE.md lines, now list the maintenance ops.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* fix(tmdb): drop Promise.allSettled from the cache clear
The web target compiles against lib es2018, so allSettled broke the
Windows frontend build (TS2550). The pending writes swallow their own
errors, so a plain Promise.all over neutralized promises does the job.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* fix(tmdb): keep a synchronous bridge throw inside the cache write
Moving the write into a tracked promise dropped the try/catch that used
to cover the call itself, so a bridge that threw synchronously would
escape set(). Wrap it in an async IIFE, which turns that back into a
rejection the same handler swallows.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* fix(tmdb): retry the cache size read when the section is reopened
The effect skipped the read once cacheError was set, so one transient
IPC failure left the panel showing "could not read the cache" for the
life of the settings page — and the only enabled control that could
shift it was the destructive Clear button. Gate on the stats signal
alone: reopening the section retries.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* fix(tmdb): queue writes that start while the cache is being cleared
Awaiting the in-flight writes closed one side of the race and left the
other open: a set() that started during that wait dispatched its IPC
immediately, was absent from the snapshot, and could reach SQLite just
before the delete — so a row written after the user clicked Clear was
removed anyway.
clear() now holds its own promise for the whole operation and set() waits
on it, which puts such a write on the far side of the delete. Rows are
stamped when they are dispatched rather than when set() was called, since
a write may have waited. Covered by a test that fails without the guard.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* docs(tmdb): add the release note for the cache panel
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* test(tmdb): cover the cache panel with an Electron E2E
The panel drives IPC and SQLite, and nothing exercised that path end to
end. The new test seeds a row through the preload bridge — enrichment
itself needs a TMDB key that CI does not have — then opens the section,
asserts the reported size, clears, and reads the database back to confirm
the row is gone rather than merely hidden.
Verified both ways: dropping the DELETE from clearTmdbMetadata fails it.
Settings nav buttons gained a data-test-id so the section can be opened
without matching translated labels.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
* fix(tmdb): stop a broken provider tmdb_id from suppressing enrichment
Providers ship dead and stale tmdb_id values, and enrich() trusted them
unconditionally:
parseProviderTmdbId(query.tmdbId) ?? await resolveIdBySearch(...)
A garbage-but-integer id short-circuited the title search entirely. The
details fetch then 404'd, the outer catch swallowed it, and the item was
left permanently unenriched — no plot, no cast, no artwork — for a title
the search would have matched. Failed detail fetches cache nothing, so
the wasted request repeated on every re-open. The stale-but-valid case
was worse: it never threw, nothing sanity-checked the resolved title, and
we confidently rendered another film's metadata.
enrich() now treats the provider id as a hint. If it fails to resolve, or
resolves to something whose title matches none of the search variants we
would have queried, the confidence-gated title search gets its turn — and
proven-bad ids are negative-cached (7d, language-independent row) so the
404 is not repeated forever.
Deliberately NOT a hard rejection on title mismatch: TMDB returns titles
in the REQUEST language, so a Russian provider title legitimately fails
the name check against an en-US payload. A mismatch only lets the search
compete; when the search finds nothing confident, the provider payload is
kept. The change can therefore only add enrichment, never remove it.
Extracts the search resolution and the bad-id cache into
TmdbIdResolverService — tmdb-enrichment.service.ts was at 290 lines
against the 300-line target, and the resolver is independently testable.
Tests: new tmdb-enrichment.service.spec.ts covers the happy path issuing
exactly one details call and no search, 404 fallback, stale-id override,
the keep-the-payload safety property, bad-id skip, and the no-match case;
matcher spec covers detailsMatchProviderTitle and the namespaced cache key.
Refs docs/architecture/tmdb-roadmap.md A1.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* fix(tmdb): only blame a provider id when TMDB confirms it does not exist
Review found the bad-id negative cache too eager in two ways, both of
which could deny enrichment to items whose id was fine.
1. Any failure recorded the verdict. A 401, 429, 5xx or an offline blip
would mark a perfectly valid id as dead for seven days, so after the
service recovered — or the user fixed their API key — titles that the
search cannot resolve confidently stayed unenriched until the marker
expired. TmdbApiService now throws a typed TmdbApiError carrying the
status, and only a confirmed 404 is recorded.
2. Title mismatches were recorded too. That id EXISTS; it is merely wrong
for this item. The row is keyed by id alone and shared across
playlists, so a stale mapping on one item disabled the direct lookup
for every other item that legitimately used the same id. Mismatches
are no longer cached at all — the search verdict is cached anyway, so
the repeat cost is a single details fetch.
Documents the row kind in the cache contract, which listed only two of
the (now six) lookup_key shapes.
Tests: 404 records, 429 does not, network error does not, mismatch does
not.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* fix(tmdb): keep provider details when the competing search fails
detailsForProviderId only runs the search to see whether it can beat a
title-mismatched provider payload. A throw from that best-effort search
(offline, rate limit, 5xx) propagated to enrich()'s outer catch and threw
away details we already had — the searched-details fetch right below it
was already tolerant. Fail to the details in hand instead.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* fix(tmdb): decide a suspect provider id on evidence, not on the title
The title check alone was both too weak and too dangerous.
Too weak: normalizeTitle strips trailing years, so "Blade Runner 2049"
carrying the 1982 film's id matched and the wrong film was rendered —
exactly the stale-id case this was meant to catch.
Too dangerous: an ALL-CAPS leading token reads as a language tag, so
"IT - Chapter Two" normalizes to "chapter two". The correct payload
failed the name check, and a year-less search for "chapter two" would
confidently return the 1979 film and overwrite it. Master trusted the
provider id here and got it right.
assessProviderId weighs both signals: title or year agrees means use the
details; both years known and incompatible means the search may take
over; a title-only mismatch is inconclusive and keeps the details. The
search branch now always has a year, so its own gate corroborates
whatever it returns instead of matching on name alone.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* fix(tmdb): do not search after a transient provider-id failure
enrich() reads a null from detailsForProviderId as "the id is unusable,
try the search". A 401/429/5xx/offline failure gave it that null, so an
outage turned into a second request that would fail too — and if it did
come back, a title match replaced a provider id that was probably fine.
Only a 404 falls through to the search now; everything else rethrows and
leaves the id retryable.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* docs(tmdb): add the release note for the provider-id fix
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
Favorites and recently-viewed rows store Stalker items as full JSON
snapshots, so a vclub-style embedded series[] episode list froze at the
moment the row was written: a series favorited when only episode 1 was out
kept showing one episode forever when opened from favorites, recents,
Continue Watching, or any dashboard rail.
New withStalkerSnapshotRefresh() store feature renders the stored snapshot
immediately and re-fetches the item from the portal in the background via a
title search (get_ordered_list&type=vod&search=..., matched by id, paginated
up to 5 pages, wildcard-category retry), patching fresh episodes and cmd into
the active selection. The patch is guarded on both the item id and the active
playlist id, since Stalker ids are only unique per portal.
Only the in-memory selection is patched — the stored snapshot row is
deliberately left alone, because every entry path into the detail view runs
this refresh and writing it back would add an uncontrolled background writer
to the whole-playlist read-modify-write that every favorite/recent mutation
performs.
Also fixes the stalker-mock-server embedded-series scenario, which generated
series[] as objects the app's vclub adapters filter out instead of the
episode-number arrays real portals send.
Regular type=series and Ministra is_series items are unaffected; Xtream is
unaffected (get_series_info is never cached).
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
`setSelectedItem` mirrored every selection's id into `selectedSerialId`, and
`serialSeasonsResource` fires a `get_ordered_list&type=series&movie_id=<id>`
portal request on every change of that id. Opening any Stalker detail page —
plain VOD, vclub items with embedded `series[]` (whose result
`mapRegularSeriesSeasons` discards), Ministra `is_series` items, and ITV
channel clicks — therefore issued a pointless request, on every entry path
(browse, favorites, recent, dashboard, search).
Set `selectedSerialId` only when `selectedContentType === 'series'`, clearing
it otherwise. The gate is deliberately on content type alone, not item shape:
under the `series` content type `serialSeasonsResource` is the only episode
source (the detail templates render `<app-stalker-series-view />` with no
`vodWithSeries` input, and `isVodSeries()` requires content type `vod`), so
gating on `is_series`/`series[]` would leave a series-section item carrying
either field with a silently empty episode list.
Adds selection-state and request-level regression coverage, and corrects a
stale invariant in the Stalker architecture docs.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* docs(tmdb): record the TMDB capability roadmap
Backlog for the TMDB subsystem produced by a multi-agent audit that
cross-checked our code against the official API reference and against how
Plex/Jellyfin/Emby/Stremio/Kodi present metadata: unused API surface,
zero-extra-call wins already sitting in cached payloads, effort-ranked
themes, a top-8 shortlist, implementation sketches, and an explicit
"deliberately not building" section with reasons.
Three entries are defects in shipped code rather than features and are
sequenced first: a broken provider tmdb_id suppresses enrichment
entirely, series cast is latest-season-only (needs aggregate_credits),
and cache retention exceeds the six-month TMDB ToS limit.
Cross-linked from tmdb-metadata-enrichment.md.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* docs(tmdb): correct the A1 sketch to match what shipped
The sketch proposed discarding details on a title mismatch and recording
a bad-id verdict on any failure. Neither survived review: TMDB returns
titles in the request language, so a localized provider title fails the
check legitimately, and the badProviderId row is keyed by id alone and
shared across playlists.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
Two fields TMDB already sends us and the merge threw away — no new API
calls, no cache-key bump, they light up on existing cached payloads.
Series detail views (Xtream and Stalker) gain a production-status chip:
"Ended" tells you a show is finished before you commit to it, "Returning"
that it is not. TMDB returns `status` as an ENGLISH string even under
language=ru-RU, so it is normalized to a stable token
(normalizeSeriesStatus) and rendered through translated labels
(seriesStatusLabelKey). Unknown values are dropped rather than shown, so
a status TMDB adds later can never leak raw English into 19 locales.
Person pages render `deathday`, which mapPersonProfile has always parsed
into ActorProfile and no template ever read.
i18n: 7 keys across all 19 locales via the tools/i18n workflow.
Tests: status normalization (token mapping, case-insensitivity, the
British "cancelled" spelling, unknown/missing dropped).
Docs: tmdb-metadata-enrichment.md, CLAUDE.md.
Refs docs/architecture/tmdb-roadmap.md C1 and the zero-extra-call tier.
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
Pipeline audit follow-up: reduce wasted runner time on PRs and tighten CI
security, without reducing what actually gets validated.
Runner-time waste:
- Concurrency with PR-only cancel-in-progress on CI, E2E, and docker-build,
so a new push cancels the previous commit's still-running checks. Non-PR
runs use the unique run_id as the group, because GitHub keeps at most one
pending run per group even with cancel-in-progress: false — a shared ref
group could silently drop a queued master run.
- paths-ignore for docs-only changes (Markdown, docs/, .plans/, .codex/,
.claude/) on the Electron build matrix and the E2E suites; E2E also skips
apps/website/**. The build workflow keeps apps/website/** because its Linux
job builds the website to verify AppStream assets. Tag pushes are
unaffected: GitHub does not evaluate paths filters for tags.
- PRs lint affected projects only; master pushes keep the full run-many.
Lint-global inputs (eslint.config.mjs, tools/eslint/**) now mark all 41
lint projects affected, including the run-commands targets database and
packaging, so the max-lines baseline cannot be widened without lint.
Hardening:
- Explicit least-privilege permissions on CI, E2E, and build-and-make; the
create-release job keeps its job-level contents: write. The repository
default workflow token was switched to read-only.
- New actionlint job (image pinned by digest, shellcheck at warning+), with
the shared-anchor false positive suppressed in .github/actionlint.yaml.
Fixed one real finding: unquoted $GITHUB_OUTPUT.
- .github/dependabot.yml: weekly cadence, minor+patch grouped per ecosystem
(npm, GitHub Actions, Docker), majors stay individual PRs.
Docs updated: CLAUDE.md, docs/architecture/nx-workspace-boundaries.md, and
docs/architecture/validation-map.md now describe affected-lint on PRs and the
E2E path-filter exceptions.
