Serve desktop and Android builds from the latest Gitea release on /download, and fix the desktop compile and CI flakes that blocked v1.3.0.
Co-authored-by: Cursor <cursoragent@cursor.com>
Labels that wrapped to two lines pushed their inputs down, the storage
select had its own label style, and the checkbox sat in the middle of its
cell. The section now lays label, input and hint rows out with subgrid, and
the select no longer stretches past its column.
Also add a web-fake launch config that runs the web app on the fake API.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The web check only accepted FLAC and WAV, so AIFF files were flagged
"needs attention" although the server already counts them as lossless.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
E2e failed to launch Chromium on the runner (missing libnspr4.so). Install
with --with-deps and cache ~/.cache/ms-playwright, since the primary download
host is unreachable from the runner and the fallback takes minutes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Introduced a new warning for users uploading non-FLAC or WAV formats, indicating that these formats should only be used as a last resort.
- Added functionality to highlight files needing format attention in the upload queue.
- Updated the UI to provide clearer guidance on acceptable audio formats during uploads.
- Improved documentation to reflect changes in upload requirements and user interface updates.
Reviewing the upload queue is now a per-user flag (reviews_uploads), not part
of the admin role: several people can hold it, a moderator can have it, and an
admin may not. Only admins grant or revoke it; existing admins keep it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Quality upgrades, batch covers, staging preview, and resume after reload are in the queue. Loudness and crossfade work on web and Android, and the documented checks now have suites.
Co-authored-by: Cursor <cursoragent@cursor.com>
- POST /admin/users/{id}/reset-recommendations clears the taste profile
(onboarding, CF vector, bandit), play stats and wave impressions; likes,
dislikes and history stay, wave sessions stay so a playing wave keeps going
- moderators may reset plain users, admins anyone including themselves;
the reset is written to the audit log
- users panel gets the menu action with a confirmation; docs updated
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- appearance settings get "collapse search" (off keeps the field always open)
and separate switches for the activity and mood pickers on the home page;
stored in ui_prefs (search_collapse, home_activities, home_moods)
- Collection and Admin headers in the sidebar fold their items; the state is
remembered in the browser, a folded Admin shows the review queue badge;
icon-only and phone layouts always show every item
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
On an Android 16 emulator the release APK crashed at launch with
"Cannot read property 'NativeModule' of undefined": libexpo-modules-core.so
was missing for the ABI the device picked. Two build defects:
- librnskia.so was linked with 4 KB segment alignment (Skia 2.0.0-next.4
lacks ANDROID_SUPPORT_FLEXIBLE_PAGE_SIZES), so 16 KB page devices cannot
load it; a pnpm patch adds the flag
- the APK carried incomplete x86, x86_64 and armeabi-v7a folders from the
React Native libraries; an Expo config plugin keeps only arm64-v8a, the ABI
the native modules are built for (the APK also halves in size)
scripts/check-apk.mjs now fails the release on another ABI, a missing core
library or a library not aligned to 16 KB.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The server kept the git.example.lan placeholder from .env.example, so the
/android page and in-app updates had no release. install.sh and update.sh now
fill GITEA_URL and GITEA_REPO from the clone's origin (https, ssh and scp
forms, credentials stripped) when they are empty or still the placeholder,
and the API treats the placeholder as not configured.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Leaving the field scheduled a close 120 ms later; coming back and typing
within that window let the old timer hide the fresh suggestions. Focus now
cancels the pending close. This also made the search box test flaky.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- settings get a search position (left, center by default, right); the top bar
is three zones and the suggestions open towards the free space
- buttons, chips, tabs, menu items and the home screen controls are not
text-selectable, so clicks and double clicks no longer highlight labels
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- /android is a public page with the latest APK, the server address to copy
and three install steps; linked from the sign-in screen and the profile menu
- failed audio analysis can be retried from the library page, which also shows
the most common error; the analyzer no longer burns attempts on files that
are missing because the disk is not mounted
- the top bar search folds into a round button and slides open on click, "/"
or Ctrl+K
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Every shell child got the same z-index, so the page painted over the search
suggestions and the profile menu and swallowed their clicks: suggestions could
not be picked, and the language switch and settings item did nothing.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
On the test server the host music folder was not mounted, so every stream and
fingerprint failed with no such file, and a scan of the empty folder would have
marked the whole library missing and purged it after 30 days.
- a scan of an empty source folder with known files aborts with a clear error
- fingerprinting skips such a source with one warning instead of one per file
- LIBRARY_HOST mounts use rslave, so a disk mounted after start shows up
- update.sh refuses to deploy with an empty LIBRARY_HOST
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Primary buttons, selected segments and context chips use accent-fill, as on
the web, so their labels meet WCAG AA contrast.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
An axe-core audit of every screen in both themes found low contrast on white
text over the accent and on caption text, track rows using row roles without
cells, and unlabelled progress bars. Accent fills now use a darker token,
caption text meets 4.5:1, track lists are lists and progress bars are labelled.
The queue drawer slides in from the right as designed, takes focus, closes on
Escape and notes that the wave never ends.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Day-to-day operation after install: users and roles, storages and scans, the
upload queue, wave metrics and training, updates, backups and restore,
monitoring and common situations.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Three minutes of wave sessions with stream starts, play events and next batches
alongside a steady suggest load; thresholds follow the documented p95 targets.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A backup container dumps the database every night, keeps 14 daily and 8 weekly
copies and once a week restores a copy into a scratch database to check tables
and row counts. update.sh takes a copy before updating, restore.sh brings one
back and keeps the old database aside. CI runs the dump, verify and restore
cycle on the database the end-to-end journey filled.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The library page compares listen-through, early-skip and like rates per
impression model version, so v1 and the trained ranker can be compared online.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Suggest fetches three times as many candidates and reranks them in the API:
what the user plays, finishes and likes moves up a few positions and what they
dislike moves down, with Meilisearch relevance still the base order.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Deleting a missing <uid>_next index is accepted with 202 and fails inside the
task with index_not_found. That error was not treated as not-found, so the first
reindex aborted and search stayed unavailable until indexes existed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The wave logs each served candidate's feature vector to rec_impression_features.
Nightly training joins them with impression outcomes, validates on the last week
and activates the model only when it beats the v1 order and the active model.
The service loads the active ranker and uses its order in place of the v1 score.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- GET /onboarding offers the library's top artists (with album covers)
and genres and says whether to show it (not done and < 10 plays);
PUT /me/onboarding stores the choice or a skip (migration 00009).
- The recommender seeds the taste from it while the user has fewer than
30 scored tracks: chosen artists' tracks +1, chosen genres +0.5.
- Web: onboarding replaces the wave on the home page until done or
skipped; artist grid with covers, genre chips, at least 3 artists.
- e2e journey and the recommender DB test cover it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Nightly Beta(α, β) per user and source from wave impressions (reward:
complete or like, ×0.99 daily decay) stored in user_profiles_rec.bandit;
the wave adds 0.2·(θ − 0.5) per candidate source, sources without data
stay neutral. The user's exploration share now follows how well `fresh`
does against the other sources.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
v0.7.3 crawled for 35 minutes at the 6 GB cap and was killed without a
log. The container now gets 7 GB, the Gradle heap is 2 GB, the Metro
node process is capped at 1.5 GB and the job times out after 60 minutes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Imported tracks are created directly under their final name with
O_EXCL (never overwriting; " (2)" on conflict), fsync'ed and checked
against SHA-256 — no hidden .pipu-tmp file next to them any more.
- On any failure the agent removes its own partial file and the empty
directories it created for it.
- The write check uses access(W_OK) and the read-only mount flag instead
of creating and deleting .pipu-write-test in the storage root.
- Tests assert the storage tree holds only audio files after successful
and failed writes and after the write check.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Changing the activity or mood used to keep the current track playing
and only replace the upcoming ones, so nothing seemed to happen. Now the
first track of the new batch starts immediately (web and Android). The
interrupted track is not reported to the wave as a skip and is recorded
in history only if it played for at least 30 s, so a context change does
not teach the recommender a false dislike.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
mrleo1nid
