Speedtest failures print an exit-code note instead of an empty section,
the report file writes progressively again (utf8 rewrite at the end),
and pytest-asyncio's py3.14 deprecation warning is filtered.
Adds machine info to the report header, multiprocess and strided-split
runs, hedge/slow-tail cases, the rnet path, and the leak regression
tests; ps1 reports save as utf8 and expected failures no longer render
as NativeCommandError blocks.
A leaked racer could retry and reopen its .!dev once dl.py cleared the
global cancel flag for the next track (WinError 32 at cleanup). A
batch-local abort event set at teardown makes its exit unconditional.
Also join terminated mp children and detect one that dies silently.
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.
bench_downloader.py sweeps the unified downloader over synthetic
loopback segments (latency/rate shaping, fault injection, hedge and
sweep-safety checks) or real URLs, with --procs and --no-read1 A/B
flags. live_manifest_test.py drives the full DASH/HLS/ISM pipeline
against 11 public streams on 8 CDNs; bench-urls/ holds reproducible
real-CDN URL lists.
Unconditional read1 for every segmented download paid a per-TLS-record
Python iteration on all segments and capped single-process throughput
at ~160 MB/s (253 MB/s with blocking full-chunk reads, measured on a
fast CDN). A segment can only be claimed once a hedge racer exists for
it, so fine-grained claimed() checks are needed only during an actual
race: download() now takes a racing predicate and workers pass the
hedged-set membership check.
One CPython interpreter caps sustained segment throughput at ~160 MB/s;
spawned children each run the existing thread-pool path on a contiguous
chunk of the batch, reaching NIC line rate on fast CDNs. Opt-in via
--download-processes (default 1), engaged only for 24+ segment batches.
Sessions are rebuilt per child from a picklable spec; RnetSession falls
back to a single process when it has no impersonate preset.
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.
Upstream migrated the free IMDxAPI from imdbapi.dev to api.tiffara.com.
Endpoints and JSON schema are unchanged, so only BASE_URL and the
user-facing docs/docstrings needed updating.
- Define keep_* flags early in the track selection loop to accurately trace which track types (video, audio, subs, chapters) should be kept.
- Guard specific track filtering and validation blocks in dl.py using keep_* variables.
- Prevent fatal 'not found in video/audio/subtitle tracks' errors when using skip options like --no-video (-nv) or --no-subs (-ns).
- Simplify and clean up the trailing track cleanup logic to rely on the pre-determined keep_* flags.
Share one best-audio selector across the episode and music paths so best/all works for songs, descriptive audio is capped to one track, and codec_priority is honored. Guard --channels and 16:9 canvas math against tracks missing channel/width metadata. Extend --best-available to video languages and full resolution misses in both normal and hybrid mode, and count only deliverable ranges when reporting missing resolutions. Rank video by resolution before bitrate, and trim group audio selections to one track per song with a warning.
BREAKING CHANGE: a missing resolution with -r HYBRID now errors without --best-available, matching non-hybrid behavior.
superseded racers now exit on the next network arrival and the pool
waits for them on clean completion, so Windows can delete stray .!dev
files at merge time (WinError 32)
superseded racers now exit on the next network arrival and the pool
waits for them on clean completion, so Windows can delete stray .!dev
files at merge time (WinError 32)
Warm up after half a tick and double the worker target while probes pay off, dropping to AIMD +2 on first plateau/error burst. 33% faster in shaped benchmarks; backoff semantics unchanged.
imSp4rky