kajisho5/ffmpeg-skill v3.1.8

on GitHub

Automated release: version and notes generated from pull requests merged since 3.1.7.

Fixed

• render.py with both an export and a chapters stage failed with output_exists on a fresh
  output path (its dry run said completed), leaving the delivery without chapters: the chapters
  stage replaced a file export.py had written as a child process, which the guard took for someone
  else's. A file a stage of this run wrote now counts as the run's own (#415).
• metadata.py (and so render's chapters stage) wrote an MP4 with its index at the end, undoing
  the export's faststart; MP4/MOV/M4A/M4V output is now +faststart (#415).
• The re-encoding tools named a .webm source's default output .webm, which their x264/AAC
  encodes cannot be written into: the dry run said completed and the run failed in the muxer. The
  default is now <stem>_<suffix>.mp4; metadata.py, a stream copy, keeps .webm (#415).
• waveform.py --title/--srt/--text --dry-run planned the text at font size 0 without saying the
  numbers were placeholders; a note now says so (#415).
• batch.py never removed its .work intermediates, so two sources with one stem (clip.mp4,
  clip.mov) refused each other's, and a later run was refused on the previous run's. Each item now
  works in a directory of its own, removed when it ends (#415).
• batch.py marked an item ok with an output that was never written when the last step names no
  {out} (a recipe ending in check.py), and its cache entry then missed on every run. Such a step
  passes its input on and the last written file is delivered. A recipe in which no step writes (a QC
  pass with check.py or probe.py) reports each row with written: false and a note, and its
  cache entries no longer miss as output_missing on every run (#415).
• batch.py's final-name collision check missed names that differ only in case, which are one file
  on macOS and Windows (#415).
• After a full disk, a missing encoder or an unreadable input, export.py's subtitle/data fallback,
  color.py --retag, cut.py's copy paths and sync.py's copy paths encoded again (up to three
  times) and blamed the subtitles or the copy; they now report the first failure, as
  run_keeping_subtitles() does since #389 (#415).
• A timeout or interrupt after the deliverable was written (export.py --normalize, render.py's
  chapters and check stages) said "nothing was written" while the file stayed; the message now names
  the kept file and the document carries it as output. export.py --normalize exited 1 when its
  loudness pass timed out or was interrupted; it now exits 124 / 130 / 143 like render.py (#415).
• ffprobe calls ignored a --timeout below 120 s, and a probe that hit its own 120 s ceiling advised
  raising --timeout, which could not help. Probes now take min(120 s, --timeout) (#415).
• Follow-ups from the #428 review: export.py --preset copy keeps a .webm source's extension;
  render.py exits 124/130/143 when an executed plan's tool or a pack destination is stopped, and
  carries the delivery an export stage kept as output; export.py says the export is kept when a
  read after the write is stopped; batch.py names a delivered intermediate by its own container
  (.mp4), delivers under the output lock through a sibling temp across filesystems, refuses an
  existing output in --dry-run too, handles stems near the 255-byte name limit, and refuses an
  output_dir that is the input folder (which let suffix: "" with --overwrite replace the sources);
  its cache key covers the files a step names (check.py --compare REF, a LUT), so editing or
  removing one reruns the items; check.py --dry-run on a not-yet-written file no longer crashes; a child killed by a signal exits 128 + the
  signal (#415).
• Dry-run honesty, process handling, faststart and batch/render orchestration (#415) (#428)

What's Changed

• Dry-run honesty, process handling, faststart and batch/render orchestration (#415) by @kajisho5 in #428

Full Changelog: v3.1.7...v3.1.8