kajisho5/ffmpeg-skill v3.9.1 on GitHub

Automated release: version and notes generated from pull requests merged since 3.9.0.

Caches read in a dry run and stored after the inputs changed (#540).
- batch.py --dry-run serves an item a real run would serve from the cache, as before
  (cached: true), but its summary line no longer counts that row's cut.py calls as planned:
  they are what the run that stored them measured, and the row has no
  cut_possible_reencode_reason. The line now says the planned calls and the cached ones apart.
  cut_stream_copy counts both, as a real run's does. A cached row's verified under
  --dry-run is the stored run's, which the contract now says.
- render.py --cache makes a stage's key again when the stage ends and stores the artifact only
  when it is unchanged. A --brand logo or font (or any input) changed while the stage ran was
  drawn, stored under the old file's key, and served once the old file was put back. A stage
  whose key, made again when it ends, is not the key it started with now stores nothing and
  records a cache_inputs_changed skip; batch.py has done the same since #537. A change that
  is undone before the stage ends, the file's fingerprint restored, is not seen.
fix: a batch dry run says a cached row's cuts apart from its plans, and render stores a stage only under the key its inputs still have (#546)

What's Changed

fix: a batch dry run says a cached row's cuts apart from its plans, and render stores a stage only under the key its inputs still have by @kajisho5 in #546

Full Changelog: v3.9.0...v3.9.1