ffmpeg-more-verbose-log

A patch that adds more helpful log messages(with loglevel VERBOSE and WARNING) to ffmpeg for debugging or understanding ffmpeg. It also modifies some defaults to be more sane. Can be directly applied to the master branch of ffmpeg at the time of writing(Oct 8, 2023).

I am very bad at C and this is only for ad-hoc ffmpeg hacking, sorry for the messy code...

Details

• Detailed logs about options parsing: what the parsed options are(help strings), applied to what OptionGroup/XXXContext or is a func_arg, and is consumed by whom at what time.
• Detailed logs about format conversions: how the AVFrame format changes along the filterchain, and what yuv2rgb matrix is used to do the conversions.
• when and what FFCodec is initialized, when and what encoder/decoder/muxer/demuxer threads are created.
• add NAL sizes(by bytes) to -loglevel trace output
• better vf_showinfo: also dumps the frame metadata, ASCII unregistered SEI messages, more verbose lines, also with comparison of deprecated pkt_pos and pkt_pos from ffmpeg FrameData(mentioned in the deprecation commit message).
• provide more information when stream codec tag finding fails.
• aligns fields of various trace logs to make them easier to read.
• various modifications to log messages to make them easier to understand without digging into source code.

Behavioral changes

• In auto_scale filters that do yuv<->rgb conversion, when the input AVFrame's colorspace is BGR or UNSPECIFIED, use bt709 instead of bt470bg. Otherwise yuv frames with bt709 colorspace would get wrong colors after going through yuv->rgb->yuv in filtergraphs.
• Made an explicit exception for the fourcc tag dvh1 and HEVC codec id. Otherwise ffmpeg would refuse to tag hvc1 stream as dvh1 in Dolby Vision muxing.