Four stale doc/comment references caught by Copilot's pass:
- core/download_plugins/base.py: TYPE_CHECKING comment said the
shared dataclasses lived in core.soulseek_client. They were moved
to core.download_plugins.types in this PR. Comment updated.
- core/qobuz_client.py: reload_credentials docstring still referenced
soulseek_client.client('qobuz') after the global rename to
download_orchestrator. Updated to download_orchestrator.client(...).
- webui/static/helper.js: the older WHATS_NEW entries for the plugin
contract + engine refactor still claimed backward-compat
self.<source> attributes were preserved. Followup commits in the
same PR removed them. Each entry now flags the followup explicitly
and points at the "Drop Backward-Compat Per-Source Attrs" entry
above it so the changelog is internally consistent.
- docs/download-engine-refactor-plan.md: Compatibility commitments
section listed orchestrator.<source> attribute preservation as a
guarantee. Cin's review pass removed those attrs (and renamed the
global handle from soulseek_client to download_orchestrator) — both
are breaking changes for in-tree callers (which were migrated) and
in-flight branches (which will need to update). Section rewritten
to document the actual outcome.
@ -191,8 +191,9 @@ After Phase C: ~490 LOC of duplicated thread management deleted. Each affected c
## Compatibility commitments
- Backward-compat `orchestrator.soulseek` / `orchestrator.youtube` / etc. attributes preserved through every phase.
- Backward-compat `clear_all_searches`, `_make_request`, `signal_download_completion`, etc. (Soulseek-specific reaches) preserved through every phase.
- **Originally** the plan was to preserve `orchestrator.soulseek` / `orchestrator.youtube` / etc. attribute aliases through every phase. Cin's review pass removed them in favor of the generic `orchestrator.client('<name>')` accessor — this is a breaking change for any external code that reached into per-source attributes directly. Anything in-tree was migrated as part of the same PR; in-flight branches will need to update.
- The legacy `soulseek_client` global handle was renamed to `download_orchestrator` in the same review pass. Any module that imported / referenced `soulseek_client` was migrated; in-flight branches will need to update.
- Soulseek-specific helper methods (`clear_all_searches`, `_make_request`, `signal_download_completion`, etc.) still live on the orchestrator and continue to work — but reach the underlying SoulseekClient via `orchestrator.client('soulseek')` instead of `orchestrator.soulseek`.
- Frontend status dashboard keys (`deezer_dl` alias) preserved.
{title:'Internal: Migrate Album-Info Builders to Typed Path',desc:'internal — steps 2+3 of the typed metadata migration in one pr. two album-info builders now route through `Album.from_<source>_dict()` when the caller passes a known source: `_build_album_info` (used by every album-tracks lookup) and the embedded album section of `_build_single_import_context_payload` (used by single-track import context resolution). legacy duck-typed extraction stays as the fallback when source is empty/unknown, raw input isn\'t a dict, or the typed converter raises — so a converter bug can\'t break album resolution or import context. caller-provided album_id / album_name / artist_name fallbacks apply on the typed path the same way they did on legacy. zero behavior change for existing callers since they don\'t pass a source yet — opt-in only. 22 new tests pin the typed path, the legacy fallback, and parametrized coverage across registered providers.'},
{title:'Internal: Migrate Discography + Quality Scanner to Typed Path',desc:'internal — next round of the typed metadata migration. three more album-shape consumers now route through `Album.from_<source>_dict()` when the caller passes a known source: `_build_discography_release_dict` (artist discography release cards), `_build_artist_detail_release_card` (artist detail page release cards), and `_normalize_track_album` (quality scanner result normalization). legacy duck-typed extraction stays as the fallback when source is empty/unknown, raw input isn\'t a dict, or the typed converter raises — same safety contract as the prior migration steps. 20 new tests pin the typed path + legacy fallback + parametrized coverage across registered providers.'},
{title:'Fix: Maintenance Findings Badge Showed Inflated Count With Empty Findings Tab',desc:'discord report (husoyo): duplicate detector and cover art filler badges showed "364 findings" / "31 findings" after a scan, but clicking into the findings tab showed nothing. cause: `_create_finding` silently dedup-skipped re-discovered issues (when an equivalent row already existed with status pending/resolved/dismissed) but the caller incremented `result.findings_created` regardless of whether a row was actually inserted. so on a re-scan that found the same problems as a prior scan, the badge snapshot recorded 364 even though zero NEW pending rows hit the db. fix: `_create_finding` now returns a bool (True on insert, False on dedup-skip / db error). all 16 repair jobs updated to only increment `findings_created` on True. new `findings_skipped_dedup` counter added to job results and surfaced in the scan log: "Done: 2791 scanned, 0 fixed, 0 findings (363 already existed), 0 errors" — so re-scans show a real count, and you can see at a glance how many findings were carried over from prior scans. also fixed a missing `job_id` kwarg in the album tag consistency job that was silently breaking finding creation for that scan. companion ux improvement: findings tab now auto-switches its status filter from "pending" to "all status" when 0 pending rows exist but resolved/dismissed/auto-fixed rows do — with a small notice so you can see what carried over instead of staring at an "all clear" empty state.',page:'library'},
{title:'Internal: Download Source Plugin Contract',desc:'internal — first step of a multi-step refactor on the multi-source download dispatcher. the orchestrator historically had 8 download sources (soulseek/youtube/tidal/qobuz/hifi/deezer/lidarr/soundcloud) hardcoded into 6+ different dispatch sites — `if username == "youtube" elif username == "tidal" elif ...` chains in `__init__`, search, download, get_all_downloads, cancel_download, etc. adding usenet (planned) would have meant 700+ lines of copy-paste across the same files. new `core/download_plugins/` package defines `DownloadSourcePlugin` (Protocol) — the canonical contract every source must satisfy: `is_configured`, `check_connection`, `search`, `download`, `get_all_downloads`, `get_download_status`, `cancel_download`, `clear_all_completed_downloads`. plus `DownloadPluginRegistry` — single source of truth for which sources exist, with name/alias resolution (legacy `deezer_dl` alias preserved). orchestrator now dispatches through the registry instead of hardcoded `[self.soulseek, self.youtube, ...]` lists; backward-compat `self.<source>` attributes still work so external callers reaching for source-specific internals (e.g. `orchestrator.soulseek._make_request`) keep working unchanged. zero behavior change for users — pure additive foundation that lets future PRs extract shared logic (background thread workers, search query normalization, post-processing context) into the contract instead of copy-pasted across all 8 sources. 19 new tests pin every plugin class\'s structural conformance to the contract — drift in any source will fail at the test boundary instead of at runtime against a live download.' },
{title:'Internal: Download Engine — Background Worker, State, Fallback',desc:'internal — followup to the download source plugin contract. lifts the duplicated thread-spawn boilerplate, per-source active_downloads dicts, and hybrid-fallback dispatch into a central `core/download_engine/` package. each streaming source (youtube, tidal, qobuz, hifi, deezer, soundcloud) used to hand-roll the same ~70 LOC of background thread management — semaphore-gated serialization, rate-limit sleep between downloads, state-dict updates for InProgress/Completed/Errored transitions, exception capture. ~490 LOC of copy-paste across 7 files. all of it gone now — `engine.worker.dispatch(source, target_id, impl_callable, ...)` owns thread spawning + semaphore + delay + state lifecycle. plugins provide only `_download_sync(download_id, target_id, display_name) → file_path`, the source-specific atomic download. per-source rate-limit policy declared via `RateLimitPolicy` (concurrency, delay) — engine reads at register time. cross-source state queries (`get_all_downloads`, `get_download_status`, `cancel_download`, `clear_all_completed_downloads`) read engine state directly instead of iterating per-source dicts. hybrid-mode search now goes through `engine.search_with_fallback(chain)` — same ordering / skip-unconfigured / swallow-per-source-exceptions semantics as before. every per-source migration commit gated by phase A pinning tests (54 tests across all 8 sources) so contract drift fails fast at the test boundary instead of at runtime against a live download. net: ~700 LOC removed across 6 client files, ~85 new engine + worker + rate-limit tests, suite green at every commit. zero behavior change for users — same downloads, same lifecycle states, same hybrid mode. backward-compat preserved for everything that reaches into `orchestrator.soulseek._make_request` etc. adding usenet now = one new client class + one registry entry, no orchestrator changes. follow-up: cin\'s metadata engine work may shape further refactors (e.g. extracting search retry / quality filter — left per-source for now since search code is genuinely 90% source-specific).' },
{title:'Internal: Download Source Plugin Contract',desc:'internal — first step of a multi-step refactor on the multi-source download dispatcher. the orchestrator historically had 8 download sources (soulseek/youtube/tidal/qobuz/hifi/deezer/lidarr/soundcloud) hardcoded into 6+ different dispatch sites — `if username == "youtube" elif username == "tidal" elif ...` chains in `__init__`, search, download, get_all_downloads, cancel_download, etc. adding usenet (planned) would have meant 700+ lines of copy-paste across the same files. new `core/download_plugins/` package defines `DownloadSourcePlugin` (Protocol) — the canonical contract every source must satisfy: `is_configured`, `check_connection`, `search`, `download`, `get_all_downloads`, `get_download_status`, `cancel_download`, `clear_all_completed_downloads`. plus `DownloadPluginRegistry` — single source of truth for which sources exist, with name/alias resolution (legacy `deezer_dl` alias preserved). orchestrator now dispatches through the registry instead of hardcoded `[self.soulseek, self.youtube, ...]` lists. (note: this PR initially preserved `self.<source>` attribute aliases for backward compat; followup commits in the same PR removed them — see the "Drop Backward-Compat Per-Source Attrs" entry above. external callers now reach individual clients via `orchestrator.client('<name>')`.) zero behavior change for end users — pure additive foundation that lets future PRs extract shared logic (background thread workers, search query normalization, post-processing context) into the contract instead of copy-pasted across all 8 sources. 19 new tests pin every plugin class\'s structural conformance to the contract — drift in any source will fail at the test boundary instead of at runtime against a live download.' },
{title:'Internal: Download Engine — Background Worker, State, Fallback',desc:'internal — followup to the download source plugin contract. lifts the duplicated thread-spawn boilerplate, per-source active_downloads dicts, and hybrid-fallback dispatch into a central `core/download_engine/` package. each streaming source (youtube, tidal, qobuz, hifi, deezer, soundcloud) used to hand-roll the same ~70 LOC of background thread management — semaphore-gated serialization, rate-limit sleep between downloads, state-dict updates for InProgress/Completed/Errored transitions, exception capture. ~490 LOC of copy-paste across 7 files. all of it gone now — `engine.worker.dispatch(source, target_id, impl_callable, ...)` owns thread spawning + semaphore + delay + state lifecycle. plugins provide only `_download_sync(download_id, target_id, display_name) → file_path`, the source-specific atomic download. per-source rate-limit policy declared via `RateLimitPolicy` (concurrency, delay) — engine reads at register time. cross-source state queries (`get_all_downloads`, `get_download_status`, `cancel_download`, `clear_all_completed_downloads`) read engine state directly instead of iterating per-source dicts. hybrid-mode search now goes through `engine.search_with_fallback(chain)` — same ordering / skip-unconfigured / swallow-per-source-exceptions semantics as before. every per-source migration commit gated by phase A pinning tests (54 tests across all 8 sources) so contract drift fails fast at the test boundary instead of at runtime against a live download. net: ~700 LOC removed across 6 client files, ~85 new engine + worker + rate-limit tests, suite green at every commit. zero behavior change for end users — same downloads, same lifecycle states, same hybrid mode. (per-source attribute aliases like `orchestrator.soulseek` were initially preserved here for backward compat; followup commits in the same PR cycle removed them — see the "Drop Backward-Compat Per-Source Attrs" entry above. soulseek-specific internals are now reached via `orchestrator.client('soulseek')._make_request(...)`.) adding usenet now = one new client class + one registry entry, no orchestrator changes. follow-up: cin\'s metadata engine work may shape further refactors (e.g. extracting search retry / quality filter — left per-source for now since search code is genuinely 90% source-specific).' },
{title:'Discogs Collection in "Your Albums"',desc:'discord request: pull your discogs collection into the your albums section on discover, similar to spotify liked albums. set your discogs personal access token on settings → connections (already there from prior work) and add discogs as one of the configured sources via the gear button on your albums. background fetcher pulls your full collection (all folders, all pages — capped at 5000 releases), normalizes artist names (strips discogs `(N)` disambiguation suffix), dedupes against any spotify/tidal/deezer-saved versions of the same album. clicking a discogs-only album opens with discogs context — full release detail (year, format, label, country, tracklist) from the /releases endpoint. clicking an album that exists in both your spotify saved AND discogs collection prefers spotify (download flow is more direct). discogs is physical-media-first so many releases won\'t have streaming equivalents — those still show in the grid but the modal flow may need to fall back to a name search to find a downloadable digital version.',page:'discover'},
{title:'Drop Redundant "Your Spotify Library" Section on Discover',desc:'discover page used to show two near-identical sections: "Your Albums" (cross-source aggregator across spotify/deezer/etc) AND "Your Spotify Library" (spotify-only). same UI, same grid, same filter / sort / download-missing controls — the spotify-only one was a strict subset of what your albums already covers. removed it. spotify saved albums still surface via the your albums section with spotify as one of its configured sources (gear button → configure sources). backend collection / storage is unchanged — the watchlist scanner still populates the spotify_library_albums cache for your albums to read.',page:'discover'},
{title:'Library Disk Usage on Stats Page',desc:'discord request (samuel [KC]): show how much disk space the library takes. new card on stats → system statistics shows total bytes + per-format breakdown (FLAC vs MP3 vs M4A bars). data comes from `tracks.file_size` populated during deep scan from whatever the media server already returns (plex MediaPart.size, jellyfin MediaSources[].Size, navidrome song.size, soulsync standalone os.path.getsize) — zero filesystem walk overhead. existing libraries see "Run a Deep Scan to populate" until the next deep scan fills in sizes; partial coverage shown as "X tracks measured (+Y pending)". migration is additive (NULL on legacy rows) so upgrading users have nothing to do.',page:'stats'},
Broque Thomas
3 weeks ago