* feat: key shared hls sessions by origin id
* feat: support hls content encoding
* fix: coderabbitai findings
* Fix: Custom headers are lost on provider://
In the manual redirect path, the abstract provider:// URL is compared against the resolved HTTP URL.
This causes the request to be marked as "scrubbed" before the first real redirect even occurs, stripping out headers like X-API-Key.
* fix(strm): name every flat/tmdb version after the folder it lands in
Under `flat: true` the movie folder is deduplicated by TMDB id, but
`final_filename` was derived from each item's own provider title before the
folder-reuse branch. Providers routinely list the same film twice with the tag
written two ways, so the second listing landed in the first listing's folder
under a name that does not start with the folder name.
That is precisely the condition Jellyfin's Emby.Naming VideoListResolver checks
in IsEligibleForMultiVersion: the file name must start with the folder name, and
the remainder must be empty or start with '-', '_', '.', or a [bracketed] token.
Any ineligible file makes GetVideosGroupedByVersion return the folder's videos
ungrouped, so *every* file in it becomes a separate movie.
The first item to claim a flat folder now also fixes the file name for every
item that follows it into that folder. The existing add_quality_to_filename
suffix, or the existing [Version id#N] collision suffix when the qualities tie,
is what keeps the versions apart. Kodi and Emby shared the same defect and are
fixed with them.
* fix(quality): classify resolution from the frame, not its height alone
VideoResolution was bucketed from ffprobe's height only. Widescreen releases keep
the full raster width but encode fewer active lines, so a 2.40:1 film mastered at
1080p (1920x796) was tagged '720p HD', and a 2.40:1 UHD master (3840x1600) was
tagged '1440p QHD'.
Classify both axes and keep the higher tier. Height still decides on its own when
the width is missing, which also keeps anamorphic frames (1440x1080) correct.
* fix(strm): drop the category from flat movie file names
The category suffix was added to flat movie file names 'to avoid collisions',
but with folder reuse now naming files after their folder, the only files that
can collide are versions of the same movie -- same TMDB folder, same quality
string -- and the existing [Version id#N] pass already separates those.
It is not free: Jellyfin and Emby render whatever follows the folder name as the
version label, so the category was showing up in the version picker next to the
quality. Removing it makes the label read as the quality alone.
Items without a TMDB id keep the category, because it is what makes their folder
unique -- but they now carry it in the file name as well. Previously the folder
got '[Category]' while the file did not, so the file name did not start with the
folder name and version detection was broken for exactly those items.
* fix(strm): keep the version label when the writer truncates a long name
The [Version id#N] suffix is the only thing separating two versions that share a
base and a quality string, but it was appended last and the writer truncates the
file stem to 250 chars -- so for a long enough title the suffix was cut off, both
versions resolved to the same path, and the second silently overwrote the first.
The collision detector could not see it either, since it compares untruncated
names.
Trim the shared base to fit instead, so the label always survives. The 250-char
limit is now a named constant used by both the naming and the writing side.
Also hoist the per-item flag lookups out of the naming loop and drop a redundant
HashSet lookup (insert already reports whether the entry was present).
* docs: scope the no-tmdb category note to jellyfin/emby
Shared HLS cache session feature and aligns the runtime, cache, provisioning, recovery, custom-response, and documentation paths with the new shared-session model.
* perf: binary-search EPG programme lookup
Replaces the linear position() scan with a binary search via partition_point for EPG programmes. Reduces complexity to O(log p).
* perf: cache interned PlaylistItemType label
Avoids repetitive hash-map lookups during sorting and filtering. The five fixed labels are now interned once via OnceLock and returned as a cheap Arc clone.
* perf: avoid per-event String allocation in web-UI XMLTV parser
Uses borrowed Cow types instead of allocating a new String for every element name. Tracks the active text state using a small TextTag enum.
* perf: throttle per-chunk download control poll
Throttles the control_signal lock polling to a 200ms interval on multi-GB downloads to prevent CPU thrashing. Immediate actions (pause/cancel) remain guaranteed via control_notify.
* perf: pre-size filtered channel buffers in apply_filter_to_playlist
Pre-allocates the filtered channel vector based on the source group length. Eliminates step-by-step reallocations on large playlists.
* fix: surface playlist cache load failures instead of silent empty fallback
Replaces silent error swallowing with real error logging (distinguishing missing files from corruption) while preserving the empty fallback behavior.
* perf: remove redundant first drain loop in get_remote_content_as_file
Removes a dead, redundant loop. The stream body is now consumed only once, allowing the idle timeout to protect the transfer correctly.
* perf: back off instead of busy-spinning on empty client-stream chunks
Introduces a short Sleep backoff (Poll::Pending) when a provider sends continuous empty keep-alive chunks. Prevents the CPU from spinning hot.
* perf: make library orphan cleanup linear instead of O(n^2)
Reduces the complexity of cleaning up orphaned entries from O(n^2) to linear O(n). Loads the metadata library only once and pre-computes partitions.
* perf: indexed batched merge for series expansion instead of O(n^2) scans
Replaces expensive O(n^2) scans during series expansion with index-based batch processing. Builds temporary lookup indexes and merges groups linearly.
* perf: pre-size group flatten buffers to reduce reallocation churn
Pre-allocates target merge vectors and category indexes to eliminate repeated memory reallocations when flattening large target playlists.
* perf: stream custom provider playlist endpoint instead of materializing it
Switches from full in-memory materialization to lazy conversion and streaming of JSON/binary responses. Massively reduces the memory footprint.
* refactor: centralize InputType batch/family checks behind enum helpers
Moves scattered matches! checks for M3u/Xtream types into centralized enum helpers (is_batch(), etc.) to prevent logic drift.
* refactor: centralize XMLTV tag/attribute literals into named constants
Replaces hardcoded inline strings in the XMLTV parser with named constants (EPG_TAG_TITLE, etc.) as a single source of truth.
* refactor: derive GeoIpUpdateError Display/Error via thiserror
Replaces manual Display and Error implementations with the thiserror macro, allowing new error variants to self-document their formatting.
* refactor: attach per-cluster facts to XtreamCluster instead of scattered matches
Centralizes redundant API actions and collection name mappings directly onto the XtreamCluster struct instead of using scattered match blocks.
* refactor: centralize PlaylistItemType video/series classification
Introduces is_video() / is_series() on the enum to unify duplicate matches! checks across filters, probes, and tracking modules.
* feat: add per-stage summary logs to VOD/series resolve queueing
Adds a concise debug summary per processing stage (total counters and elapsed time) to make hot-path diagnostics easier without raising the log level.
* refactor: generate ItemField get/set from a single direct-field list
Uses a callback macro to synchronously generate both getter and setter blocks for directly-bound fields (Name, Url, etc.) from a single master list.
* refactor: derive input-type routing from a single capability descriptor
Bundles InputType behaviors (backends, connection rules) into a single capabilities() descriptor to prevent bugs when adding new variants.
* refactor: derive output-format routing from a single capability descriptor
Centralizes output format features (filter, EPG, and cache support) into a single TargetCapabilities descriptor instead of spreading logic across no-op match arms.
* Fixed sanitizing url scheme batch and provder
* local date in Stream history
* Fixed EPG programme time display to render in local timezone for improved readability.
* Improved stream history date range filtering to accurately use local date boundaries.
* Enhanced credential sanitization for stream URLs to properly mask sensitive information.
* Fix EPG path selection for mixed Xtream/M3U targets
* Fix EPG selection and complete Xtream series_info for input/custom requests
* Add configurable interner GC tuning
