Try one InsertKeys call per 500 content keys, falling back to per-key InsertKey only when the server answers 404; kv copy/sync chunk 500 for json mode.
An exact height match in one range denied the 16:9 canvas fallback to the others, so a cropped HDR10 or SDR track was dropped whenever an uncropped DV track matched the requested height. Also sweep the hybrid L5 temp files so a killed run cannot leave a stale RPU behind.
The server no longer spends its configured proxy providers on a remote client unless the API key sets server_proxy. It places the client by geolocating the connection address and asks for a proxy when that region does not match its own.
serve.users.<key>.server_cdm accepts a list of service tags; retry re-attaches the job service and a bare string no longer fails open. An unset client server_cdm now follows the server's per-service /tracks flag.
Add serve.services_refresh_interval; refresh pulls the repos and re-imports only the changed services, staging any with active jobs until they finish. Report skipped/broken services in the log and on /api/services.
Services set the original language, track names, repack flags, editions and track.data inside get_tracks; the client dropped them so --remote muxed with no default audio and could not select "orig". Also round-trip Song titles, subtitle DRM, PlayReady PSSH, exact bitrate and path-only attachments.
Opt-in decrypt_segments runs mp4decrypt --fragments-info per segment on a thread pool as segments land (DASH, single-init HLS), then decrypts the init alone; hides the post-download decrypt step with byte-identical output.
Tracks stayed in LICENSING while a key found in a local vault was pushed to remote vaults under a global lock. Queue those writes on one worker, snapshot content_keys at submit, log writer failures, move --export out of the DRM lock, and join the queue before the run exits.
Strip cue classes, <v>/<lang>/ruby and karaoke timestamps before conversion, decode entities for SRT/SSA/ASS, load pysubs2 as utf-8-sig, and route sdh_method: filter-subs through convert_before_strip instead of raising.
Request 4 MiB sub-ranges once a host cuts a body, skip the session's own urllib3 Retry inside download(), fail the ranged path at once on a 4xx, and cap resumes at 50 of 1 MiB each. Adds a deny fault profile.
Reset the attempt budget when a resumed cut delivers bytes, close the stream in a finally, skip the ranged path for a finished file, and add a throttle-cut profile to fault_server.
Uppercase entries in ALIASES never resolved, because get_tag lowercased the input but not the alias. Match real tags first so a service's own tag cannot be taken by another service's alias.
RFC 9110 clamps a range end past the resource to its last byte; the guard
rejected that valid response. The descriptive-audio filter also had no
emptiness check, so a title with only described audio failed with a
misleading missing-language error.
Adds Track.drm_preference so a service can pin a track to Widevine or
PlayReady when a manifest advertises both but only one will license it.
Honoured in DASH/HLS selection, the dl licence path and the remote/API
track round-trip. Also fixes quality-keyed CDM maps sending audio to the
default entry, and a licence callable binding the CDM before the swap.
Gate a title on audio or video languages that must exist, independent of what -l/-al/-vl select. --require-subs joins the same gate: its --s-lang conflict and download-all side effect are gone, so -sl now decides what to keep. Requiring a language that is also excluded fails at parse time.
A song skips the muxer, so the decrypted fragmented MP4 was the delivered file: its header states a length of zero and FLAC in an MP4 is not a FLAC stream, so playback stopped early and tagging failed.
Extract the DV stream and RPU once per title, skip the Level 6 edit when the RPU already holds the values, sample cropdetect in parallel with fewer frames, and let inject-rpu write the per-resolution output directly. Fill Level 6 MaxCLL/MaxFALL from the base-layer SEI when the RPU says unknown: the ffprobe probe read stream side data that raw HEVC never carries, so it always returned the defaults. Read the L6 block from dovi_tool's JSON instead of the summary text, which rounded the mastering display minimum.
Submit standalone mux tasks largest-first as soon as they are enqueued and build hybrid tracks while they run, each muxed from its own pool slot when ready; run mkvpropedit inside the mux thread instead of a serial pass at the end; load per-track DRM from a thread pool; decode m3u8 playlists as UTF-8 instead of letting requests run chardet over them.
Grow the service session pool to downloads * workers before track downloads, run mkvmerge jobs concurrently (muxing.concurrency, default 4), lock DRM preparation per key set instead of globally, and query local vaults first then remote vaults together.
get_services() normalised table names through Services.get_tag() and then handed the result back as an identifier, so kv copy and sync skipped any table whose name differed in case or matched an alias. Yield stored names, and resolve tables case-insensitively.
Delete the dead grouped-download block, the planner, and the music service hooks; Song now flows through the one download loop, gets tagged via tag_file, and answers -w with {disc}x{track} keys.
BREAKING CHANGE: GROUP_AUDIO_DOWNLOADS, get_music_track_options and get_music_collection_label are gone. A music service returns one Audio track per quality and lets the framework choose.
New post_scripts config and repeatable --postscript flag run a command per output, season/album folder, or run, substituting filename-template variables into pre-tokenized argv. The REST API rejects hook commands.
Closes#150