6 commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
| 1e4a82df16 |
Log Emby's real status + response body on any proxied call failure
Every attempt to see Emby's actual error message by hand failed the same way: hls.js's own error reporting only ever exposes the generic HTTP reason phrase, never the real response body, and Emby's own live log view didn't reliably surface the exception's headline message either - across several different items and failure modes this session. Emby's own access log confirmed a real, several-KB text body really was being sent every time (Content-Type: text/plain, Content-Length: 5190) - it just never survived being reported back through either of those paths. Fixed at the root: EmbyClient.fetch - the one function every Emby call in this app goes through (fetchJson's JSON API calls and proxyResponse's streaming/image proxy alike) - now logs Emby's status and full response body itself straight to docker logs whenever a call fails. - server/src/emby/EmbyClient.ts: new private logErrorResponse, called from fetch() only when !res.ok. Reads the body via .clone() so the real caller still gets an untouched, fully readable response (a body can only be consumed once, and this had to not break either of the two very different ways this app's callers read one - fetchJson's .json(), proxyResponse's .text()/pipe). pathAndQuery is logged with api_key redacted - this can be called with Emby's own TranscodingUrl verbatim, which can carry one (same concern as CatalogService's logStreamRequest). - server/src/Logger.ts: new emby() method, same shape as party()/stream(). - server/test/embyClient.test.ts (new): body/status logged and readable by the real caller afterward, api_key redaction, and silence on a successful response (only failures are worth logging, same discipline as this app's existing logging). Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> |
|||
| 4eab0323d6 |
Fix CI: npm test's glob pattern doesn't expand on Node 20
"tsx --test test/**/*.test.ts" worked locally (Node v22) and failed on the CI runner (node:20-alpine, Node v20) with: Could not find '.../server/test/**/*.test.ts' node --test only gained native glob-pattern CLI arguments around Node 21/22; on Node 20 the literal glob string is treated as a single filename and, unsurprisingly, not found. Same script, two Node majors, different behavior - not a shell/quoting issue (both runs go through tsx's own CLI, no shell glob expansion involved either way). Fixed by listing the four test files explicitly instead of relying on a glob whose CLI support turned out to vary across the exact two Node installs this project runs on (this dev machine, and node:20-alpine - the CI container and server/Dockerfile's runtime image, deliberately matched). engines.node is >=18, so relying on a 21/22-only feature was already a latent bug even before it showed up. Verified locally (all 25 tests still pass with the explicit list) - the actual node:20-alpine behavior will confirm on the next CI run. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> |
|||
| fc678b313c |
Add a real server test suite and gate CI's Docker publish on it
- server/test/*.test.ts: formalizes this project's many ad-hoc scratchpad test scripts into a committed suite, one file per service (musicQueueService, partyService, socketGateway, catalogService), using Node's built-in node:test + node:assert/strict - zero new dependencies. Runs straight against TypeScript source via `tsx --test` (npm test), no compile step. - server/tsconfig.test.json: extends the base config, overrides rootDir to "." and adds test/**/*.ts to include, so `npm run typecheck` covers the test files too without pulling them into dist/ (npm run build still uses the unmodified base tsconfig, scoped to src/ only). - server/package.json: typecheck now runs against tsconfig.test.json; new test script. - .forgejo/workflows/publish-docker.yml: new `test` job (node:20-alpine container - the publish job's docker:27-cli has no Node) running server typecheck+test and client typecheck+build; publish now depends on it via needs: test. Removed the stale "no test suite yet" comment. - README.md, docs/architecture.md: documented the test suite, how to run it locally, and the new CI job/gating. Verified locally: server typecheck, server test (25/25 passing), server build, client typecheck, and client build all pass individually - the same commands the new CI job runs. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> |
|||
| 4a31ee741e |
Rewrite server + client in TypeScript, service-oriented server architecture
All checks were successful
Publier l'image Docker / publish (push) Successful in 18s
Full rewrite, requested directly. Two independent decisions since TypeScript can't compile itself away at runtime - "no build step" and "TypeScript" are mutually exclusive, not a small addition to the existing setup: - Server: plain tsc (no bundling need - Node resolves its own requires). server/dist mirrors server/src 1:1. - Client: esbuild (does need bundling) - picked over a heavier tool (Vite, webpack) since this app's whole client is a couple thousand lines; its built-in CSS-from-JS bundling and a `text` loader for importing an HTML template as a string cover everything needed via a ~30-line hand-written build script (client/build.mjs), not a config file for something built for much bigger apps. Both projects compile independently (no shared-types package - considered and rejected as more structural commitment than this app's size justifies; server/src/types.ts and client/src/types.ts have their own overlapping DTO copies instead). server/'s `npm run dev` uses tsx for a compile-free dev loop; `npm run build` (tsc) is what server/Dockerfile's build stage actually runs. client/'s `npm run watch` is esbuild's own watch mode. Server: one singleton service class per feature, generalizing the PartyManager/MusicQueueService pattern that already existed, rather than route handlers with logic inline: - AuthService (from embyAuth.js), CatalogService (NEW - previously ~15 route handlers' worth of Emby queries/DTO shaping inlined directly in index.js), PartyService (renamed from PartyManager), MusicQueueService (ported as-is, already service-shaped), SocketGateway (from socketHandler.js), EmbyClient (from embyProxy.js - low-level plumbing, not called a "service" since it has no state/feature of its own). - index.ts is now thin wiring only: every route reads the session, calls one service method, returns the result. - Real error-handling pass, not just a port: Express 4 doesn't catch a rejected Promise from an async handler on its own - a new asyncRoute() wrapper + a single error-handling middleware replace what used to be zero handling for a request that outright throws. CatalogService methods throw a CatalogFetchError (own HTTP status - getRawItem now preserves Emby's real 404 instead of a blanket 502) rather than touching res directly. Client: one folder per view (src/views/<name>/<Name>.ts + .html + .css) instead of one .js file with the markup inline as a template literal. A view's static shell moved to its .html (esbuild's text loader); per-item dynamic content (tile grids, episode/track lists) stays as template -literal-returning functions in the .ts, same as before. Genuinely shared CSS (tiles, the presence avatar, the series/album detail-header pattern, track-list rows) lives under src/styles/, imported by whichever views need it; src/styles/base.css holds the global primitives, imported once by app.ts. Deliberately not pushed to zero duplication/renaming shared classes across views for tidiness alone - not worth the risk in a change already this large. Player.ts keeps its pre-existing internal shape (the /party/:id dispatcher + the video renderer in one file, importing renderMusicPlayer from the sibling folder) - that pattern predates this rewrite and only needed types, not restructuring. Two real things the port itself caught, not just carried over: - A type bug: porting SocketGateway's "members" broadcast, reusing PartyService's PublicMember interface would have type-checked fine but was the wrong (larger) shape - the socket broadcast never included isPaused/positionTicks, only the REST response did. Caught by comparing against the original JS line by line, fixed with a dedicated SocketMemberDto type. - A real Express+TypeScript friction point: noUncheckedIndexedAccess (enabled in both tsconfigs - genuinely useful elsewhere, e.g. PartyService.advanceQueue's array indexing) makes every req.params.xxx read a type error, since Express's index-signature typing doesn't know a matched route's required :name segment is always present. Resolved once with a param(req, name) helper rather than 15 routes' worth of scattered assertions. Verified: both projects type-check cleanly in strict mode (tsc --noEmit - not just "esbuild didn't complain", which it wouldn't even with type errors present, since it transpiles without checking). Server: every unit/integration test from the JS version (party lifecycle logging, the music queue's timer/vote/pause behavior incl. real setTimeout auto-advance, the socket integration through SocketGateway+PartyService+ MusicQueueService together, the library-type filter, the playback -override forcing) re-run against the compiled server/dist output - identical results; the compiled server was actually started and confirmed listening cleanly. Client: since esbuild bundles everything into one app.js (no more individually-importable view modules), testing moved to driving the real compiled bundle through its actual boot -> login -> router -> view chain (mocked fetch/Socket.IO/hls.js, real clicks/hash navigation) - arguably more realistic than the old per-module approach. Confirmed this way: login, the sidebar, home's carousels, a video party (presence, host controls, leave), the full music hierarchy, a music party (presence, vote-skip, pause toggle, the inline browse-and-add panel), and the series page - all through client/dist's actual output, no thrown errors anywhere. Not verified: an actual `docker build` of the new multi-stage Dockerfile - this dev environment has no Docker at all. The build stage runs the exact npm run build commands already confirmed working locally, just inside node:20-alpine - the main remaining risk is something environment -specific, not the build commands themselves, but a real build is still the first thing to try after this lands. docs/architecture.md has the full writeup (new "TypeScript and a service-oriented architecture" section, plus a corrected top-level "Components" overview that had gone stale even before this change). Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> |
|||
| 5d535d18d8 |
Switch to Socket.IO; fix two real memory leaks
All checks were successful
Publier l'image Docker / publish (push) Successful in 10s
Requested: Socket.IO instead of raw WebSockets, and explicit attention
to memory leaks. Auditing for the second turned up two real, concrete
leaks, independent of the transport library:
1. Server-side: a raw `ws` server has no idea a client is gone unless
it gets a proper close frame - a sleeping laptop, a network drop, a
killed tab never fire 'close', so the PartyMember's socket
reference (and everything it closes over) sat in memory forever,
and the party was never cleaned up by anything else either
(removeIdleParties only looks at lastActivityUtc). Socket.IO's
built-in heartbeat (pingInterval/pingTimeout) detects and closes
these out on its own - a real, not just cosmetic, reason to prefer
it here.
2. Client-side: router.js only ever did currentRoot.innerHTML = ''
between routes, destroying DOM but never telling the outgoing
view's JS to stop. Leaving the player any way other than its own
"Quitter" button (a sidebar link, browser back) left the 3s
heartbeat setInterval, the realtime connection, and the hls.js
instance all running against a detached <video> indefinitely, for
as long as the tab stayed open.
Fixes:
- server/src/socketHandler.js replaces wsHandler.js: same protocol
(see its own top comment), over Socket.IO. Rooms (socket.join/
io.to) replace manually iterating party.members to broadcast.
Error event renamed to "errorMessage" - Socket.IO reserves "error".
- server/src/index.js: socket.io server instead of a raw
WebSocketServer, same-origin (no CORS config needed).
- client/js/party.js: socket.io-client (window.io, loaded via
<script> in index.html) instead of hand-rolled WebSocket + JSON
framing - also gets automatic reconnection for free.
- router.js: route() handlers may now return a cleanup function;
render() calls whatever the previous view returned before switching
to the next route, regardless of how navigation happened.
- player.js returns a cleanup that clears the heartbeat, destroys
hls, disconnects the socket, and removes its document-level click
listener (now named, not inline-anonymous, specifically so it CAN
be removed - an anonymous one added fresh per visit would itself
have been a slow leak even with everything else fixed). The
"Quitter" button is now just navigate('/home') - cleanup runs the
same way regardless of how you leave.
Verified for real (see session history): a real Socket.IO client
connected to a really-running local server end to end. Then, with
document.addEventListener/removeEventListener and setInterval/
clearInterval instrumented, drove the router through repeated "join a
party, then navigate away without clicking Quitter" cycles - adds and
removes matched exactly every time (3/3 listeners over three cycles,
2/2 intervals over two), nothing left pending.
docs/architecture.md has the full writeup under "Realtime transport:
Socket.IO, and two real memory leaks".
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
|
|||
| 6ff9aa29e2 |
Pivot to a standalone app - Emby plugin hit a hard wall
No plugin API exists for adding a button to Emby's player or item
detail page (confirmed by an Emby developer on their own forum, and
by research since). The only way to reach that would be patching
Emby's shipped web client files directly - outside the plugin,
breaks on every image update, needs manual reapplication. Not worth
the ongoing cost, per discussion.
New direction: a standalone app with its own player. Reads Emby's
catalog/streams directly from the browser (no plugin, no menu
integration fights); a small Node backend only coordinates party
membership and relays playback state over WebSocket - it never
touches a media byte. Auth is Emby-only, as chosen: the browser signs
in against Emby directly and this backend validates whatever token it
gets handed by asking Emby itself (GET /Users/{userId}), never seeing
a password.
Control model changed too: now that we own the player, host controls
playback and everyone else follows (event-driven play/pause/seek +
a 3s heartbeat, drift-corrected on the follower side) - the old
plugin's "everyone keeps their own controls, shared reference
position" was specifically a concession to nudging pre-existing Emby
app sessions we didn't control, which no longer applies.
- server/: Express + ws backend (party lifecycle, WebSocket protocol,
Emby token validation)
- client/: plain-JS/ES-modules frontend (login, search, item detail,
party player) - no build step
- Smoke-tested locally end to end against a stubbed Emby URL: server
starts, routing/auth-gating works, error handling degrades
gracefully (network failure, invalid token) - see the session
transcript for the actual browser test run
- docs/architecture.md rewritten around the new design, including why
the plugin was abandoned and a fresh "assumptions to verify" list
(CORS between this app's origin and Emby's is the biggest unknown)
- src/WatchParty/ (the Emby plugin) kept as-is for reference, no
longer the active direction
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
|
