One housekeeping note I should have put in the cover letter, sorry. Patch 59867, "avformat/dss: fix DSS SP demux of voice-activated (paused) recordings", which I sent back in June and which is still sitting in "new", is superseded by 2/2 of this series. Please mark it as such.
It was the same bug seen through a keyhole. In June I had found one way the running walk goes out of step - a block with frame_count == 0, emitted when a voice-activated recorder pauses - and patched that single case by skipping the empty block and re-anchoring on the next one. That was right as far as it went, and it is why the description in 59867 still reads correctly today. What I had not realised is that the empty block is not a special case at all. *Every* block re-states the framing, in three fields, and the walk can drift apart from them for reasons that have nothing to do with pauses - which is what 2/2 handles. The empty-block path falls out of it for free: a block declaring no frames is simply skipped, and the next one re-anchors, exactly as 59867 did by hand. So 2/2 is a strict superset, and applying both would be redundant. Apologies for the trail of patches on this format. This is the last one. :-) _______________________________________________ ffmpeg-devel mailing list -- [email protected] To unsubscribe send an email to [email protected]
