mirror of
https://github.com/FFmpeg/FFmpeg.git
synced 2026-11-06 05:17:42 +00:00
Piotr NiełacnyandClaude Opus 5.5
738961c650
With output-side -ss, audio/video filtering and subtitle encoding subtract the output start time before packets reach the scheduler. Streamcopy performs its subtraction later, in the muxer. Comparing these different timelines can starve copied audio long enough for early HLS segments to be written without audio packets. Register the upstream timestamp offset for each encoded mux stream and add it back in a separate last_dts_paced field. Use that field for both the trailing timestamp and choke comparisons, while retaining the raw last_dts for progress reporting. Register subtitle offsets as well, matching the packet timestamp rebasing introduced by7f2b4adeac. Add a generated-input HLS FATE test using native encoders. Declare the source, transcode and probe dependencies and register the test with FATE_FFMPEG_FFPROBE. Use a single stream listing in the reference, as required after88f7fd9ec5made program-stream output explicit. The scheduler regression was introduced byd119ae2fd8. Fixes #22765. Assisted-by: Fairy Signed-off-by: Piotr Niełacny <piotr.nielacny@gmail.com> Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>