Extend LegacyUiRedirects from an exact-match dictionary to a two-tier matcher:
Tier 1 keeps the exact Map (now 52 entries incl. the ?kind= browse roots),
Tier 2 adds 36 ordered segment-template PatternRules for id-carrying routes.
{id} is a strict positive integer (non-int/0/neg/overflow falls through), which
also makes the rule set collision-free by construction. New AppendQueryString
helper merges the incoming query into ?kind= targets with '&' (kills the
double-'?' bug); one-line Startup change keeps the redirect GET/HEAD-only 302
before UseRouting.
Completes phase-(a) Step 1 for every PARITY-OK route (#91 phase b); the
catch-all fallback replacing MapFallbackToPage stays with the removal PR.
/media/sources/* (#202) and /system/health remain deliberately un-redirected.
fixes#204
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Independent Codex review of the fix diff surfaced two real findings the fork pass missed:
- Medium: the #251 affected-playout QUERIES in ReplaceDecoTemplateItemsHandler and
UpdateDecoHandler still ran on the request `cancellationToken`, so a cancellation
landing after SaveChanges committed but before those queries executed would throw
before the CancellationToken.None enqueue — the edit committed but no playout Reset,
re-opening the stale-content bug in that window. Run the entire post-commit
invalidation (queries + enqueue) on CancellationToken.None so the side effect can't
be half-aborted once the data has changed.
- Low: UpdateDefaultDecoHandler enqueued a Reset for request.PlayoutId even when
ExecuteUpdateAsync matched 0 rows (nonexistent playout), creating a background build
request for an id that isn't there. Guard the enqueue on rows-updated > 0 so the
enqueued set equals the affected set. Added a regression test.
Also corrected the ReplaceProgramScheduleItemsHandler comments: the schedule-item
hierarchy is TPT (table-per-type), not TPH — the SetValues reconcile is safe either way
(same-runtime-type guard; no discriminator to corrupt), Codex confirmed.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
#252: ReplaceProgramScheduleItems deleted and re-inserted every item on every
save (even a no-op PUT-back), and PlayoutScheduleItemFillGroupIndex.ProgramScheduleItemId
is OnDelete(Cascade) — so every schedule save silently wiped persisted
fill-group/shuffle enumerator progression for all playouts using the schedule.
Switch to a positional in-place reconcile: for a same-typed slot, copy scalars via
CurrentValues.SetValues (BuildItem stays the single source of item construction, so
no field is dropped) and rebuild the watermark/graphics join rows, keeping the item
id — and with it the fill-group index. Subtype change / surplus falls back to
delete+insert for that slot only. The request DTO carries no stable item id, so
position is the only key here; true content-aware stable identity is deferred to the
shared concurrency/round-trip contract in #253.
#251: deco / deco-template CONTENT edits (and default-deco assignment) only take
effect on a playout Reset build — deco/break/default-filler content is applied during
Reset, a Continue keeps the frozen filler items, and BlockKey change-detection has no
deco dimension to self-heal. The editors enqueued nothing (a commented-out TODO in
ReplaceDecoTemplateItemsHandler), so filler/break content stayed stale indefinitely
until a manual Reset. Enqueue BuildPlayout(Reset) for exactly the affected playouts:
- ReplaceDecoTemplateItemsHandler: playouts via PlayoutTemplate.DecoTemplateId
- UpdateDecoHandler: playouts via Playout.DecoId and via deco-template items
- UpdateDefaultDecoHandler: the reassigned playout (adjacent same-class fix)
Post-commit enqueues use CancellationToken.None (audit #22 policy).
Tests: DecoInvalidationTests + ReplaceProgramScheduleItemsReconcileTests, each proven
non-vacuous against a negative control (inverted the primitive, verified 0 CS errors so
the --no-build run used a fresh dll). Full ErsatzTV.Tests suite green (1067).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Independent Codex review of PR #250 caught a cross-release the fork missed: the
outer catch in PrepareTroubleshootingPlaybackHandler released the troubleshooting
lock unconditionally, so an exception BEFORE this caller acquired it (e.g. request
cancellation during validation, or a DB error) would release a lock held by another
session.
Fix: track ownership with a Handle-scoped `lockAcquired` flag and gate the catch on
it. Acquisition for the media-item path moves out of GetProcess up into Handle (after
validation succeeds), so one place owns the full lifecycle: acquire -> on Left release
-> on success hand off to StartTroubleshootingPlayback -> on any exception release only
if we own it. GetProcess is now lock-free.
Test: Handle_Should_Not_Release_Lock_It_Never_Acquired (exception before acquisition ->
no Unlock, no cross-release), verified non-vacuous against an inverted-condition
negative control. Strengthened the empty-path test to also assert acquisition happened.
Deferred (noted for close comment): worker-dispatch-failure lock leak and
BuildPlayout silent-skip observability are pre-existing / sanctioned -> #235.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Consume the EntityLocker ownership contract (#231/#241: Lock* returns true iff
this caller won the slot) at three lock-leak sites surfaced by adversarial-reviewer#20.
#233 (F3) — troubleshooting playback:
- PrepareTroubleshootingPlaybackHandler: both lock sites now acquire via
`if (!LockTroubleshootingPlayback())` (kills the check-then-set TOCTOU) and the
empty-media-path Left return releases the lock it acquired — previously it leaked,
wedging the status endpoint at "running" forever for a file gone from disk.
- TroubleshootController.TroubleshootPlayback: lock conflict is now 409 ProblemDetails
(was a bare 404, indistinguishable from a bad id); the Prepare-success -> enqueue
window releases the lock if we never hand off to StartTroubleshootingPlayback.
#234 (F4 + F5.2) — playout builds:
- ExtractEmbeddedSubtitlesHandler: try/finally releases exactly the playouts it
locked, on every terminal path (cancellation early-return, swallowed cancellation,
any exception) — no more permanent leaks after cancelled mid-extraction, and no
cross-release of playouts held by someone else.
- BuildPlayoutHandler: skips (logs, returns Right) when LockPlayout returns false
instead of building unlocked and cross-releasing the other owner's lock in finally.
Tests: handler-level release-discipline tests (Prepare empty-path, Extract
cancellation + no-cross-release, BuildPlayout skip + finally-release) via the
InMemoryTvContext harness, a TroubleshootController 409 test, and OpenApi contract
cases for the m3u8 endpoint's 409. OpenAPI regenerated. All non-vacuous (F3 verified
against a negative control).
Fixes#233, #234
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Move the Channels domain verbatim out of web/src/App.tsx into
web/src/screens/ChannelsScreen.tsx (zero-prop, self-sufficient, mirroring the
SchedulesScreen extraction). Pure structural move: no API, route, CSS, or
visual change. App.tsx retains only the import + the <ChannelsScreen /> dispatch.
- 14 symbols moved (ChannelViewFilter → ChannelTableRow); the Dashboard-owned
progressFromNowPlaying is inlined into the moved progressFromChannelState so
the screen has no import back into App.tsx (behavior-identical).
- 12 Channels behavior tests moved to a colocated ChannelsScreen.test.tsx with
its own scoped fetch mock (renders <ChannelsScreen /> directly, no
mockDashboardApi); App.test.tsx keeps one nav-smoke test for the route.
- Pruned 12 now-dead App.tsx imports; shared symbols (ChannelState,
messageFromError, ApiError, useChannelsQuery) verified still used and kept.
- Docs: spa-conventions §6 (extracted-screen own-fetch-mock convention),
decisions.md (single-file rationale; no web/src/channels/ sibling dir, unlike
Schedules; inlined helper; #238 deferral).
Verified: web vitest 587 passed, eslint clean, tsc/vite build clean,
check:api no drift. #212 empty-lineup bare-create success+failure coverage
preserved. #238 TopBar dead-button left as-is (its owned bug; shell redesign
is epic phase 4 / #247).
refs #244#243
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Codex re-review of the prior fix commit found the Edit gate closed the exact
repro but two paths remained. One is reachable: Delete was the only
schedule-switch path that guardedSwitch's `saving` guard didn't cover — deleting
mid-save runs applySwitch to the next schedule while the in-flight items PUT is
still outstanding, and that PUT's completion handler then overwrites the next
schedule's draft with the deleted schedule's response. Delete is now
`disabled={saving}`, consistent with the Select, Edit, and guardedSwitch.
Regression: the deferred-PUT test now also asserts Delete is disabled in-flight
and re-enables after the save settles.
Also widens the test mock's onRequest return type to `Response | Promise<Response>
| null` (removes the `as unknown as Response` cast — a test-only type hole the
re-review flagged).
Deferred to #248: the other residual path (properties dialog not focus-trapped, so
keyboard focus can escape to underlying Add/Save mid-save) is a pre-existing,
cross-cutting overlay.tsx a11y gap affecting all dialogs — out of scope for this
targeted blocker fix.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Two findings from the PR #242 adversarial review (Codex + Claude fork):
1. ChannelsEmptyState swallowed mutationError: on the fresh-install path #212
targets, a bare-create can 4xx (e.g. no default ffmpeg profile), but the empty
branch never rendered the error alert the non-empty screen shows — the user saw
only a spinner re-enable. The empty state now renders the same ctv-channels-error
alert. Regression: App.test.tsx asserts the error surfaces + no navigation on an
empty lineup.
2. SchedulesScreen Edit button was not gated on `saving` (Codex): during an items
save PUT the draft is still dirty, so a discard-to-open → shuffle-flip could let
the in-flight PUT resolve AFTER the shuffle reload and clobber the normalized
draft with the pre-shuffle body. Edit is now disabled while saving (consistent
with the schedule Select). Regression: a deferred PUT proves Edit is disabled
in-flight and re-enables once the save settles.
Deferred as nits (both reviews rate low): vetoed navigateToPath leaves two stray
history entries (cosmetic, rare — fixing means refactoring central nav); rapid
double-Back is best-effort (inherent popstate non-cancellability).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Adversarial review (fork + Codex, both flagged) of PR #241:
- SPA (both reviewers): removing PENDING_GRACE_TICKS wholesale reintroduced a
stuck scan button. A 202'd scan that finishes between 10s polls (short/empty
library) is never observed active, so its optimistic pending flag wedged the
button disabled until reload. Restore a BOUNDED grace net (pruneGraceExpiredPending)
— re-scoped honestly: it absorbs the inherent queue->observed-active lag and the
fast-completion race, NOT the removed lying-200 compensation (the POST now returns
409/404/422 honestly). Bounds pending to PENDING_GRACE_TICKS * pollMs (~30s).
- SPA 409 (Codex): on "already scanning" the button was cleared+reconciled, but a
scan-status still lagging the in-progress scan re-enabled the button and let the
user fire repeated 409s. Keep the pending flag on 409 (no toast) so the button
stays disabled; polling promotes or expires it.
- Scheduler (Codex): the Plex-Shows tail-token batch and the local/Jellyfin/Emby
scan enqueues had no compensating unlock — a WriteAsync failure after LockLibrary
(cancellation on shutdown) stranded the library lock. Wrap each acquired-lock
enqueue in try/catch → UnlockLibrary → rethrow (the Plex catch covers both writes,
since the library message carries Unlock: false and the un-enqueued networks
message was the sole releaser).
Tests: two new App.test.tsx cases — grace-window expiry re-enables the button, and
409 keeps it disabled through the queue->active lag.
Ref #232.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Editing a schedule's properties to change shuffleScheduleItems for the active
schedule left the open items draft stale: hidden Fixed/Flood start values could
be saved back and the inspector kept offering start-type controls the schedule
no longer supports. onScheduleSaved now detects a shuffle flip on the active
schedule and reloads the items via GET so the server's EnforceProperties
re-normalizes the draft. The schedule-level flags already refresh from the save
response (setBoot maps `saved` into the list).
Decision: chose "block opening the edit dialog while the item draft is dirty"
(confirm-to-discard, guardedSwitch semantics) over confirm-at-reload — the
smaller fully-consistent change. It guarantees the properties editor only ever
opens over a clean baseline draft, so the post-save reload is lossless and
avoids the awkward state where a cancelled discard leaves a now-shuffled
schedule holding Fixed values.
Regression: SchedulesScreen.test.tsx — Fixed item on a non-shuffled schedule →
edit properties to shuffle=true → a fresh items GET fires, the Fixed option is
gone, and a subsequent Save's PUT carries no Fixed startType.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
canLeaveCurrentScreen() was only consulted in App's navigate() (sidebar/nav
clicks); browser Back/Forward switched screens unguarded. A popstate can't be
cancelled, so App's popstate handler now, on a vetoed guard, re-pushes the
pre-pop path (tracked in currentPathRef, updated on every approved navigation)
and leaves activeRoute untouched — undoing the browser's URL change. The same
handler covers the synthetic pop navigateToPath() dispatches. Re-pushing is
safe: only one screen is mounted at a time and the guard-registering screen
(schedules) owns no internal popstate listener, so no sub-path screen's
pathname state can desync. Effect cleanup keeps StrictMode double-mount from
double-registering.
Docs: spa-conventions §8 rewritten from navigate-only to describe popstate
coverage. Regression: App.test.tsx dirties the schedules draft, simulates
popstate → confirm called; cancel keeps route + re-pushes path; accept
switches route and unmounts the draft.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
ChannelsScreen returned ChannelsEmptyState before the action bar that owns
"New blank channel", so a fresh install could never create its first channel.
The empty state now offers both create paths (bare-create + ChannelBuilder),
reusing the exact createBlankChannel handler (number = max+1 → 1 on empty,
group "ErsatzTV", default ffmpeg profile).
Regression: App.test.tsx bare-creates from a [] lineup, asserts the POST
payload (number "1") + navigation to /app/edit-channel/{id}.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Scan queue handler now returns a QueueLibraryScanResult enum
(Queued|NotFound|SyncDisabled|AlreadyScanning) instead of a lying bool;
LibrariesController.ScanLibrary maps them to 202/404/422/409 with ProblemDetails.
Guard the lock->enqueue with the EnqueueWithTraktLock compensating-unlock pattern.
ScannerService now releases every library/collection lock in a finally so a handler
exception can't leak the lock. Plex "Shows" scheduler batch (one lock, two messages)
now has only the trailing SynchronizePlexNetworks carry the single release
(Unlock flag), mirroring the scheduler Trakt tail-token precedent.
Guard the other lock->enqueue producers (Create/UpdateLocalLibrary, UpdateTraktList)
with compensating unlock. SPA drops the PENDING_GRACE_TICKS heuristic now that the
POST reports 202/409/404/422 directly: 202 -> pending+poll, 409 -> reconcile (no
error toast), 404/422 -> surface error.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The six plain-bool lock flags (Plex, Trakt, Emby/Jellyfin/Plex collections,
troubleshooting playback) used a non-atomic check-then-set, so two concurrent
Lock* callers could both win. Convert them to int flags mutated only via
Interlocked.CompareExchange, so the caller that wins the 0->1 transition is the
sole owner and the only one that fires the change event. The three
ConcurrentDictionary-backed kinds (Library/Playout/RemoteMediaSource) were
already atomic; drop their redundant ContainsKey pre-checks.
Define the ownership contract (tokenless single-owner discipline, no interface
change) on IEntityLocker and in docs/decisions.md: a true from Lock* confers
ownership of exactly one release; Unlock* on an unlocked slot returns false,
fires no event, and logs a warning (the double-release / non-owner tripwire).
Adds EntityLockerTests (real locker, parallel-caller races) proving exactly one
winner per kind, one-releaser-per-slot, and event-fires-once-per-transition.
Ref #231. Scan-lifecycle call-site fixes that consume this contract land in the
same PR (#232); the BuildPlayout/subtitle finally-gating is #234.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Closes the reviewer-verification gap from adversarial-reviewer#18: the
locked-path 409 guard tests already existed for Delete, EraseItems,
EraseItemsAndHistory, Update, and UpdateDefaultDeco, but were missing for
PlayoutController.ReplaceAlternateSchedules and .ReplaceTemplates even
though their production guards (entityLocker.IsPlayoutLocked) were in
place. Added two tests mirroring the existing pattern exactly (409
ConflictObjectResult + ProblemDetails + DidNotReceive() on the mediator
command). ApiErrorResponseMetadataTests and OpenApiErrorResponseContractTests
already carried 409 rows for both PUT endpoints, so no changes were needed
there.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
- build & docs-reminder jobs -> runs-on: small (dedicated small-jobs runner,
server-management#574). Gitea dispatches a job as a runner task even when
its 'if' skips it; the PR-run skip of 'Build & push image' waited up to
31 min for an ubuntu-latest slot, stalling every PR run.
- concurrency scoped per event+ref with cancel-in-progress for PRs: runs
parallelize across PRs, superseded syncs auto-cancel. Previously one global
group serialized ALL runs (single-runner relic). Main/tag builds still
serialize within their ref; don't push main + v* tag simultaneously
(shared :buildcache / smoke container) — tag after main is green.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Bug 2 (client side): array position becomes the persisted index on the next
PUT-replace, so the schedules editor must ingest strictly by the server-provided
`index` rather than trusting response row order — otherwise a reload + re-save could
silently reshuffle the lineup. Applied at both ingest points (GET load and the
replace response). Pinned by a shuffled-response-order test.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Bug 1 (500 on watermark/graphics save): Replace/Add handlers projected the
freshly-built entity graph, whose ProgramScheduleItemWatermark / -GraphicsElement
join rows carry only foreign-key ids — the Watermark/GraphicsElement navs are null,
and Mapper.ProjectToViewModel dereferences them unguarded, throwing an NRE that the
controller surfaced as a 500 on PUT/POST. Both handlers now reload the persisted
item(s) through the read-side include chain before projecting. Extracted that chain
into ProgramScheduleItemQueryExtensions.IncludeScheduleItemDetails() so GET, Replace
and Add share one source of truth.
Masking: PersistItems returned a lazy LanguageExt Map, and the existing round-trip
test only checked .IsRight — never enumerating it, so the deferred NRE never fired.
The new ScheduleItemWriteProjectionTests force enumeration (as the controller's
.ToList()/serialization does) and seed watermark/graphics via a separate context so
the handler's fresh factory context has nothing pre-tracked.
Bug 2 (server side): GetProgramScheduleItemsHandler now .OrderBy(i => i.Index) —
it previously returned id order, which is not index order.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Handoff file reduced to static kickoff prompt + append-only lessons lore;
queue/arc/session log live in pinned tracker ersatztv#237 with in-progress
claim labels and end-of-session triage. Decision recorded in decisions.md;
docs index updated.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Seed a non-null FixedStartTimeBehavior (Flexible) on the Fixed-start item and
assert it survives envelope A; assert the Marathon-order Duration item's
DiscardToFillAttempts is zeroed by the deliberate FixDiscardToFillAttempts
server normalization (Random/Shuffle keep the value, all else -> 0).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
timothyandClaude Opus 4.8