RFC 9110 permits a server to ignore Range and answer 200 with the whole
resource. When the item's slice starts at byte 0 and spans exactly the
response body, that 200 is byte-identical to the 206, so keep it instead of
burning all five attempts and failing the track. Every other 200 (nonzero
start, span mismatch, open-ended, encoded) still fails the attempt to
protect the merge.
The rnet branch zeroed content_length only for gzip/deflate/br, so an
unknown coding (zstd, future tokens) kept a bogus length and, worse, the
drop-the-partial-instead-of-Range-resuming guard could never fire on the
rnet path because content_encoded stayed False. Use the same fail-closed
_is_content_encoded check the requests branch uses.
Any batch failure sets the process-global DOWNLOAD_CANCELLED, which
TokenBucket.consume waits on, so a test elsewhere could void every timing
assertion here; add the autouse clear fixture the other downloader test
modules already use. Also capture the debt sleep via a stub instead of
timing a real Event.wait, whose occasional spurious early wake tripped the
1.0s lower bound.
Spawned children re-import the module and start with no TokenBucket, so
--download-processes silently dropped a configured --speed-limit /
global_speed_limit for any 24+ segment batch. Fall back to the in-process
path while a limit is set (same gate as the ranged-parallel paths), log the
fallback, and document the interaction.
Catch the branch up with dev (45 commits) so testers can run the perf work
against current dev.
Conflict resolutions:
- downloaders/requests.py: keep the branch's racing-gated read1, adaptive
controller speed reporting and byte-range slice guards; take dev's
content-encoding checks (raw reads are skipped on an encoded body) and its
rolling SpeedWindow for the no-controller speed fallback. The single-URL
ranged-parallel gate now requires no speed limit and no caller Range.
- service.py / session.py: keep the branch's TimeoutHTTPAdapter and retry
config, raise POOL_MAX_SIZE to 64 so dev's pool-starvation fix survives
(pool_block=True stalls threads on a pool smaller than the in-flight count).
- docs: dev split configuration.md into per-area pages, so the branch's rnet
network option rows moved to reference/configuration/network.md.
Reading .raw skips urllib3's decoding, so brotli-served subtitles landed compressed and converted to empty files. Also drops a partial rather than Range-resuming an encoded body, since a Range indexes the encoded bytes.
sort_subtitles resolved only the all token, while the video and audio sorts
resolved all and best. by_language now asks matching_languages rather than
repeating the RFC 4647 preference, with no behaviour change.
subtitle.language_priority and audio.language_priority prepend to the CLI
languages, so the named languages sort first and nothing is filtered.
--exact-lang now applies to the sorts too, through matching_languages, which
follows the same RFC 4647 rule as Tracks.by_language and ignores invalid tags.
Each dl run works in temp/task_<id>/, removed on every exit path. Orphans from hard-killed processes are swept via lock liveness, and env clear temp skips directories whose task is still running.
TVDB covers TV far better than the existing sources, so add it as a
provider and let users decide where it sits in the lookup chain.
- new TVDB v4 provider: search, get_by_id, get_external_ids and
find_by_imdb_id, with a process-level token cache and 401 re-auth
- metadata_providers config both orders and filters the providers used,
so TVDB can be primary rather than a TMDB fallback. Cross-validation
keeps its own fixed trust ranking, unchanged by search order
- --tvdb-order renumbers episodes into a TVDB season order. A service
does not always use aired order, so the source order is detected
rather than assumed, and the renumber is refused when an order that
omits episodes would give two of them the same filename
- --tvdb sets the series ID directly, skipping the lookup
- the Matroska TVDB2 tag now populates on every lookup path, including
--imdb, which previously produced no ID tags when IMDxAPI was off
Search matches on year as well as title so an identically named remake
cannot win on title alone.
sort_subtitles applied the language sort before the type sort. Python sorts
are stable, so the last pass wins and type became the major key: every forced
track, then every normal, then every SDH, each block ordered by language. In a
title with many languages this put a track far from its own SDH version, for
example Finnish at track 5 and Finnish SDH at track 42.
Add a group_by parameter that selects the major key. "type" is the default and
keeps the current order byte for byte. "language" swaps the two passes so each
language stays next to its own variants. Read it from the new subtitle.group_by
config key.
Also correct the docs: type_priority was described as sorting the variants
within each language, which was never true. It ranks the types across all
languages.
Only merge back tracks from another manifest, so a drifted id no longer duplicates a rung or wipes the main manifest's own subtitles. track_is_encrypted now recognises an ISM ProtectionHeader.
Error messages named the resolved language for a user who only ever defaulted to orig. Audio tracks now carry is_original, resolved through Tracks.by_language so it agrees with the downloader.
Detect the dot/space style from the full folder template instead of per segment, so `{artist}/{album} ({year})` no longer yields `Some.Artist/Some Album (2025)`. Also drops empty `[]` groups like the existing `()` handling and removes redundant cleanup passes.
UNITY_MATRIX emitted 11 int32s instead of 9, oversizing mvhd/tkhd so pymp4 desynced and fTTML subtitles yielded no cues. Also rewrites PIFF senc to CENC so shaka can decrypt Smooth, labels HEV1 as hev1, and raises when a fragment survives decryption still carrying a sample-encryption box.
Smooth fragments set tfhd sample_description_index=2 against a server init carrying a second, identical sample entry. A lone entry left that dangling: mp4decrypt skipped every such fragment silently and demuxers stopped after the first, yielding a file that looked fine but was almost entirely encrypted.
Only fall back to Title.language when the manifest labels nothing, and reject und as firmly as a missing tag. Exported direct-URL subtitles now replace the re-parsed set, and fetch failures stay distinct from parse failures.
Detect an encv/enca sample entry surviving decryption (moov-scoped 4CC scan) and force a repack so mkvmerge isn't first-FourCC-locked to a generic encrypted codec; warn if one persists after repacking.
Closes#142
Also throttle hedge-median recompute to 0.5s, cache replace_fields format
patterns and short-circuit when no $ token, and stream HLS merge with
copyfileobj. Adds test_dash_collapse and live A/B tooling.
Segments carrying their own Range (DASH SegmentBase, HLS EXT-X-BYTERANGE) now
fail the attempt when a server ignores the Range and returns 200, instead of
writing the whole parent resource as the segment and corrupting the merge. The
resume path no longer clobbers such a slice's Range, and retries rewrite the
slice whole. All segment GETs request Accept-Encoding: identity and the rnet
path mirrors the gzip guard, so transparent CDN (de)compression can't break the
Content-Length truncation check. Adds tests pinning these plus the adaptive
start-at-cap wiring and the small-segment tail-probe gate.
by_language now applies RFC 4647 Lookup in exact mode, preferring the string-equal tag so --exact-lang separates base/regional pairs (ar/ar-EG, en/en-US) instead of collapsing them, and dedupes by id to prevent the duplicate-track-id crash. Adds a language-selection regression test.
Merge cues at the string level instead of round-tripping through
pycaption, which dropped inline tags and cue settings. STYLE and
REGION blocks are kept, duplicate boundary cues splice correctly,
and a bad timing line now skips the cue instead of crashing.
Also rank pysubs2 above SubtitleEdit for ASS to TTML/VTT since
SubtitleEdit flattens inline styling there. Documented the measured
backend fidelity table and added regression tests for the merge.
imSp4rky