171e117f7ceba3667f479b68538739f511f577d7
11
Commits
-
Recover blocked searches, fall through to the next peer, rebuild the UI
Measured against a live instance: 47 failed jobs against 33 imported. Three independent causes, all of which ended the request. The Soulseek server filters some artist names. "Michael Jackson Billie Jean" and bare "Michael Jackson" both return zero responses; "Billie Jean" returns 1083 files. The Weeknd, Kanye West and Frank Ocean behave the same way. Since pick_and_enqueue sent exactly one query of `artist title`, every request for a filtered artist failed reporting "searched 0 files" -- 28 of 29 occurrences. search_queries() now walks a ladder, loosest last, ending at the bare title, and escalates only when a search underdelivers so the common case still costs one search. Dropping the artist from the query does not mean ignoring it: match_ratio() looks for it in the folder path, where an organised share puts it. A rejected transfer ended the request. pick_and_enqueue kept a running maximum and discarded every runner-up, so "Transfer rejected: File not shared." -- a per-file condition where candidate two almost always works -- was terminal, as were timeouts and slskd's 500 on an unreachable peer. The ranked list is now persisted on the job row, ordered to put a different peer next, and try_next_candidate() advances through it. A job fails only once the list is exhausted, and then reports every peer it tried and what each one said. norm() destroyed every non-ASCII character, so "Beyonce" never matched Beyonce, nor Bjork Bjork, nor Dont Don't. It now folds accents and drops apostrophes while leaving & as a separator. Also: availability outranks format in score(), because a FLAC behind 5000 queued users used to beat an immediately available MP3; transfers that stall or that slskd forgets are abandoned instead of sitting in `downloading` forever; searches are deleted only once finished, since deleting one mid-flight makes slskd throw an optimistic-concurrency error and lose the record; a failed search POST is retried rather than failing the job on one blip. The UI moves to a sidebar shell -- rail at tablet, bottom bar on phones -- with neutral chrome and the sampled artwork colour confined to hero areas rather than repainting the whole interface. Adds URL routing, a primitive set under components/ui, and a request sheet showing the attempt ladder. Fixes hover:bg-glass-hi, which was never a valid class, so track rows and the search field had no hover or focus state at all; a reduced-motion path that could strand elements at opacity 0; the login page's hardcoded palette; and the missing 404 and error boundaries. tests/test_acquisition.py covers this path against a faked slskd. It had none. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Aryan SinghandClaude Opus 5
2026-09-02 11:29:59 -05:00 -
Publish arm64 alongside amd64
muse is now the front door of a public project, and it was the only image in that stack without an arm64 build. Raspberry Pis, arm64 NASes and Ampere VPSes are a large share of self-hosters, so amd64-only made the whole stack un-runnable for them — gluetun, slskd, soularr, lidarr, navidrome and the socket proxy all already publish both. Three parts, because changing `platforms:` alone would have failed: - setup-qemu-action registers the binfmt handlers buildx needs to produce arm64 on an amd64 runner. Without it the build errors with "exec format error"; it does not quietly fall back to amd64. - The web stage is pinned to --platform=$BUILDPLATFORM so the Next.js build runs natively once instead of being emulated per platform. It is the expensive step, and its output is static HTML/JS/CSS with no native code, so the bytes are identical either way. This is what makes the old comment's "3-5x slower for nothing" no longer true. - platforms: linux/amd64,linux/arm64. Verified locally before pushing: both platforms build in ~52s, and the loaded arm64 image reports aarch64 with uvloop, watchfiles, httptools and pydantic_core all importing, `import app` succeeding, and the static UI present. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Aryan SinghandClaude Opus 5
2026-08-06 00:51:31 -05:00 -
Bind invites to a provider, and email them
Wanted shape: the owner signs in with GitHub, everyone invited later signs in with Google. That was previously a convention, not a boundary -- may_login checked only that a live row existed, so any enabled provider could satisfy any invite. Now an invite is bound to the provider it was issued for. friend@gmail.com invited for Google cannot be redeemed over GitHub, and there is never a second, weaker door onto an existing account. The per-request re-check still passes provider=None, because there the question is only whether the account is still live. Two related fixes fell out: - upsert_user no longer refreshes `provider` from the login. It is part of the identity an admin issued and may_login now enforces it, so overwriting it on sign-in would let a successful login silently rewrite the very rule that permitted it. - a Google invite must be a full email address. A bare username would authorise something Google can never return, and the invitee would see only "not authorised" with no way to work out why. INVITE EMAILS. Adding a user now mails them, when SMTP is configured and there is an address -- for Google there always is, since the identity IS one. The response reports whether the mail actually went and why not, because "Invited" alone reads as success while the person waits for an email nobody sent. The link carries NO token, deliberately. The row already names the exact identity permitted and Google will not hand us an unverified address, so possession of the mail is not what authorises anyone: a forwarded invite grants nothing, nothing expires, and resend is always safe. A signed expiring token would only buy the ability to withdraw an unused invitation, which deleting the row already does. Resend is offered in the panel on anyone who has an address and has never signed in -- the likeliest reason being that they missed it. tests/test_invites.py: 29 assertions driving a real SMTP conversation against a sink server, plus the full provider-binding matrix. 119 tests pass across four suites. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Aryan SinghandClaude Opus 5
2026-08-05 12:05:44 -05:00 -
Add review, discover, email; rebuild the design system
Three features and a typographic pass over everything. REVIEW. Picking a file and trusting it are different questions, so they now use different numbers. The existing rank decides which candidate to fetch and is right to weigh peer upload slots and queue length -- but none of that says anything about whether the file is the correct recording, and a fast peer must not be able to raise the confidence of a wrong match. So confidence() scores the filename alone: title 0.75, artist 0.25, album +0.05, and a variant tag you did not ask for multiplies by 0.8 AND caps below the auto threshold. Confidence gates the IMPORT, not the download. The transfer completes either way and the file waits in downloads, so approving is instant and rejecting is a delete. Holding the transfer instead would mean the peer is usually gone by the time anyone looks. The two cases it exists for: "02 Creep.flac" in an unsorted folder scores 0.75 -- title exactly right, artist simply absent, the commonest honest case there is -- and must be reviewed rather than binned. And "Creep (Live at Glastonbury)" scores 0.80 with BOTH artist and title perfect; the cap is what stops a live recording auto-filing over the studio version. Ask for "Creep (Acoustic)" and it auto-imports at 0.97, because then the variant is the request. DISCOVER. Related artists seeded from what you have actually imported, plus a recap of top artists, genres and decades. Built on muse's own request history, never on listening data: Navidrome owns play counts and history, and a second source of truth for those would disagree with the one you listen through. EMAIL. Optional, off unless MUSE_SMTP_HOST is set. Sends on import and on review-needed. smtplib is blocking so it runs off the event loop, and every failure is logged and swallowed -- a notification is never worth failing an import over. Google accounts need no configuration, since the identity is an address. Each user owns their own setting under Settings, deliberately not an admin route: an address you cannot change yourself is someone else's chore. DESIGN. A real type scale (display/title/headline/body/callout/caption) whose tracking tightens as size grows, which is the single biggest tell of typography nobody set. Specular top edges on glass surfaces -- blur alone reads as a grey rectangle. A two-part elevation scale, contact shadow plus soft light. Semantic colours are now per-theme: the amber tuned for the dark background failed contrast outright on white, which was exactly the text a reviewer most needs to read. The tab underline became a sliding pill so the eye tracks one object, and Review only appears when something is actually waiting. Add tests/test_matching.py: 13 assertions on the confidence bands rather than the exact scores. 90 tests pass across the three suites. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Aryan SinghandClaude Opus 5
2026-08-05 02:10:33 -05:00 -
Fix a path traversal out of the music library
`year` was appended to the album directory name AFTER safe() had run, so it was the one path component reaching the filesystem unchecked: POST /api/request {"artist":"E","title":"p","year":"../../../../tmp/pwned"} -> library/E/Singles (../../../../tmp/pwned)/p.flac -> normalises to ../tmp/pwned)/p.flac Any authenticated user could do it, guests included, and it wrote a real audio file to an arbitrary path the process could reach. Existing files were not overwritten -- tag_and_file returns early when the destination exists -- so this created new files and directories rather than clobbering. Fixed in two layers, because the first is a property of five call sites that must each stay correct and the second cannot be forgotten: 1. safe_year() whitelists \d{4}, applied in enqueue_job (the choke point for all three request paths) and again in tag_and_file. 2. tag_and_file resolves the destination and refuses anything that is not under the library root, before mkdir rather than after. Also from the same pass: - Cap import at 500 pasted lines / 1000 expanded tracks. URL expansion runs BEFORE the quota check -- it has to, since a playlist URL is one line and 200 tracks -- so without a cap a paste of 10,000 album URLs was 10,000 requests to Deezer from this IP before any quota was consulted. - Clamp search limit (1-50), jobs limit (1-1000) and activity days (1-365). search limit fanned out one /album/{id} enrichment per distinct album. - safe_cover() requires https and <=300 chars. img-src permits data:, so the field was free storage for megabytes of base64 per job. - Truncate filename components to 150 chars; a long title used to fail the import with ENAMETOOLONG after the download had already completed. - Treat Deezer's 403 as no-results. Its WAF rejects injection-looking queries, which surfaced to the user as a 502 "service broken". Add tests/: 42 security assertions and 35 authorization assertions covering the role matrix, last-admin protection, mid-session revocation, job scoping and OAuth CSRF. All 77 pass. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>Aryan SinghandClaude Opus 5
2026-08-05 01:45:41 -05:00 -
Stop the admin panel hanging when slskd is down
/api/admin/system probed slskd with the shared client, whose 30s timeout is sized for searches and transfers. So the panel sat on skeletons for the full 30s exactly when slskd was unreachable -- the case you open it to diagnose -- and the 5s UI poll stacked requests behind it. Probes now get 4s: above a LAN round-trip, under the poll interval. httpx.ConnectTimeout also stringifies to "", so the banner rendered "slskd unreachable -- " with nothing after the dash. Fall back to the exception class name. Add dev.sh, because the documented dev commands could not work: MUSE_DB defaults to /config/muse.db which a Mac cannot write, and the auth preflight refuses to boot without a client ID. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Aryan SinghandClaude Opus 5
2026-08-05 01:36:10 -05:00 -
Add multi-user: Google OAuth, roles, quotas, admin panel
Phase 2. The schema landed last commit; this makes it real. AUTH is now multi-provider. MUSE_AUTH takes a comma list, so github,google runs both. The callback URL is unchanged — /auth/callback for every provider, with the provider carried in the state COOKIE rather than the query string — so adding Google needs no edit to an already-registered GitHub app, and the callback can't be steered to a different provider by the redirect. With more than one enabled, /auth/login shows a chooser built from style attributes only, which the CSP already permits, so it needs no stylesheet and no extra hash. Google identities are the VERIFIED email. An unverified one is refused: anyone can put someone else's address on an account they control, and the allowlist is keyed on exactly that string. WHO MAY LOG IN is now two rules — the MUSE_ALLOWED_USERS bootstrap list, then a non-disabled users row. Inviting is adding a row; no email, no token, no expiry to manage. MUSE_ALLOWED_USERS stays mandatory only while the table is empty, so a working multi-user install isn't held hostage by an env var it has outgrown, but an empty database with no allowlist still refuses to start because nobody could ever get in. The rule is one function used at login AND on every request: checking only at sign-in isn't revocation, it's a delay. QUOTAS are requests per calendar month and bytes of imported audio, 0 = unlimited. Calendar month, not a rolling window — "12 left until the 1st" is something a person can plan around. Batches are checked against the whole batch, so a 30-track album with 5 requests left is refused up front rather than accepted and abandoned halfway. Storage is measured from the file actually written, not the size the peer advertised. Admins are exempt. ADMIN PANEL: host CPU/memory/disk read straight from /proc — no psutil, no compiled dependency, and in a container /proc reports the host, which is what you want for a homelab box. Plus library totals, slskd connection state, a 30-day chart, per-user and per-artist breakdowns, and inline role/quota editing. The chart is hand-rolled SVG: a charting library would add ~150KB for one stacked bar chart and most of them inject <style> at runtime, which this app's CSP blocks. Guard rails: the last active admin cannot be demoted, disabled or deleted; disabling revokes sessions immediately; deleting a user keeps their jobs, because the history records what is in the library and why. Requests are private by default — users see their own, admins see all — and retrying someone else's job returns 404 rather than 403, since a distinct error would confirm the id exists. Also drops AnimatePresence mode="wait" from the tab container. It gates the incoming panel on the outgoing one's exit animation reporting finished, so anything that stalls the frame loop leaves the UI stuck on the old tab with no way forward — found while testing, where a backgrounded page froze navigation entirely. A cosmetic transition should never gate navigation. Verified: admin endpoints 200/403/403/401 for admin/user/guest/anon; last-admin demote, disable and delete all 409; disable revokes mid-session; quota refuses the 3rd of 2 and a 30-track batch, then succeeds once raised; users see only their own jobs; cross-user retry 404s; bad role/provider/duplicate/empty patch all rejected; multi-provider chooser and both redirects correct; three CSRF variants 400; preflight refuses unknown providers, missing secrets, and an empty database with no allowlist. CPU, memory and disk confirmed populated inside the linux/amd64 container. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Aryan SinghandClaude Opus 5
2026-08-04 20:54:12 -05:00 -
Rebuild the UI on Next.js, and give search real metadata
The old page was hand-written HTML/CSS/JS. That was fine for a search box and a list, but the roadmap here is multi-user, an admin dashboard with usage graphs, and an import-review queue — all of which get painful in vanilla DOM code. Moves to Next.js 16 / React 19 / Tailwind 4 / Framer Motion. Built as a STATIC EXPORT served by FastAPI. Running `next start` alongside uvicorn would add a Node runtime, a second process to supervise, and a second thing that can be down while the other is up. There is no use for SSR — every view is behind auth and renders from live JSON — so the deployment stays one container, one process, one healthcheck. Image is 60MB; Node only exists in the build stage. Search now carries what a music UI needs: - year and genre, grafted on from /album/{id} concurrently, deduplicated by album and cached forever since a release date is immutable - 30-second previews, so you can hear a track before committing a Soulseek search to it. Not a library player — that is Navidrome's job. - autocomplete over songs/albums/artists, debounced and aborted per keystroke so a slow response to "rad" cannot land after "radiohead" - album and artist pages, with `Get all` to queue a release in one request - the accent colour is sampled from the artwork, so the palette follows the music. Deezer's CDN sends ACAO:*, which is what makes canvas sampling work with no image proxy. /api/album/{id} deliberately calls Deezer twice: the track list embedded in the album response has no track_position, and using the list index instead misnumbers every multi-disc release — verified against the White Album, where disc 2 track 1 would have been filed as 18. Schema is multi-user ready ahead of the feature: users table with roles and quotas, requested_by on jobs, and the first login becomes admin. Auth is still the single-user GitHub allowlist. CSP survives all of this intact. Next emits inline hydration scripts, which script-src 'self' blocks; rather than per-request nonces, their SHA-256 is computed at startup from the built page — a static export never varies between requests, so hashes need no maintenance and no HTML rewriting. The one concession is style-src-attr 'unsafe-inline', because Framer Motion writes its pre-hydration state into style="" and without it every element flashes in. That permits style attributes only. img-src/media-src now name Deezer's CDN explicitly instead of allowing https: wholesale, closing an exfiltration channel that survived connect-src 'self'. /_next/* sits behind the session too — there is no local login page to style, so the bundle need not be public. Verified: unauthenticated /→302 and every /api/*→401 including the four new endpoints, forged state→400, forged cookie→401, misconfigured auth refuses to start, and the production image hydrates with zero CSP violations. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>Aryan SinghandClaude Opus 5
2026-08-04 14:26:31 -05:00 -
Add single-user GitHub OAuth, hardened for internet exposure
muse drives a download client and writes to the media library, so exposing it publicly needs more than a login form. It fails closed. Auth is on unless MUSE_AUTH=off is set explicitly, and when it is on with incomplete config the app refuses to start rather than serving open. The mode worth engineering against is not a broken login page, it is a misconfigured deploy quietly publishing an unauthenticated download client. Sessions are server-side: the cookie carries a 256-bit random token and nothing else, and only its SHA-256 is stored. There is no signed payload to get subtly wrong, a stolen database yields no usable cookies, and logout revokes for real. The allowlist is re-checked on every request so removing a user takes effect immediately rather than at session expiry. OAuth state is compared with secrets.compare_digest, so a callback the user did not initiate is rejected. MUSE_BASE_URL comes from config rather than X-Forwarded-* because those headers are attacker-controllable and this one is a redirect target. CSP is default-src 'self' with no unsafe-inline, which is why the stylesheet and script moved out of index.html into static files -- an injected <script> now cannot execute. Also frame-ancestors 'none', base-uri 'none', nosniff, DENY, no-referrer and HSTS. FastAPI's /docs, /redoc and /openapi.json are disabled so the login page does not hand out a map of every endpoint. /healthz is the only unauthenticated route and returns a bare ok, so the container healthcheck works without leaking anything. Verified: misconfigured auth refuses to boot; unauthenticated / redirects and /api/* returns 401; a forged state is rejected 400; a junk cookie is 401; a valid session for a non-allowlisted user is 401; all headers present; and the UI renders with no CSP violations in the console. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
aryansinghandClaude Opus 5
2026-08-04 12:09:51 -05:00 -
Add bulk import, and a bounded queue to make it safe
Requesting songs one at a time does not scale past a handful, so this adds a paste-a-list import. It accepts "Artist - Title", en dashes, "Title by Artist", numbered lists, bare titles, and Deezer album/playlist URLs, which get expanded into their tracks. The queue is the load-bearing part. Every request previously spawned its own asyncio task, so a 50-track paste would have launched 50 concurrent Soulseek searches -- each 10-40s of polling and real load on slskd -- and drowned the client. Work now goes through an asyncio.Queue drained by MUSE_WORKERS consumers, default 3. Requests return immediately and the batch drains at a civil rate. The queue is in memory, so jobs left 'queued' or 'searching' are re-queued at startup rather than stranded by a restart. Ranking needed two fixes found by testing against the real API: Import was calling deezer_search(limit=1), which fetches two rows, so the ranker was choosing between a demo and a rehearsal take while the studio version -- six times the popularity -- sat unseen further down. It now pulls a real candidate pool. More seriously, the Deezer ranker had no title-relevance term at all: only artist match, popularity and penalties. A different song by the right artist therefore competed on equal footing, and "Oasis - Live Forever" resolved to "Cloudburst" as soon as a variant penalty tipped the balance. Title overlap now dominates, and the variant penalty is small enough only to break ties between versions of the same song. Derivative terms match on word boundaries rather than substrings, so a bare "live" catches "Come As You Are (Live)" without penalising "Deliver Me" or "Olive", and a term is ignored when the user typed it -- "oasis live forever" still works. Verified against live slskd: six tracks pasted, six resolved to the correct song, five imported with correct album/year/track and link count 2, in about two minutes. The sixth returned an empty Soulseek search and is retryable from the UI. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-04 03:23:17 -05:00 -
muse: request a single song, get it tagged in your library
Lidarr's smallest unit is a MusicBrainz release, so asking for one track means monitoring a whole discography. Soulseek trades individual files but has no concept of a request. muse bridges the two: search, click, and the song lands in the library correctly tagged. The design point is that the user picks the exact track before anything downloads, so muse holds artist, album, year and track number as ground truth and tags deterministically rather than inferring them afterwards from whatever a stranger named the file. Search is Deezer rather than MusicBrainz, decided on measurement: MusicBrainz put eight covers above the real recording for "radiohead creep" and never surfaced Queen for "bohemian rhapsody" across 90 results, because its index matches a word in a cover's title as readily as in the artist field. Re-ranking cannot rescue a result set missing the right row. Imports are os.link(), so downloads keep seeding from the Soulseek share while the library gets a second name for the same bytes.
2026-08-04 03:05:45 -05:00