For this particular problem, I'm using ffmpeg-0.5 on a Linux x86_64 platform. The input file is an MPEG-2 containing a video stream, and a KLV data stream. When I convert the file to any other video format, the video stream is parsed correctly, but the data stream is discarded.
Given my inputs, that's what should happen. However, the ffmpeg diagnostics don't even list the data stream. ffmpeg is apparently detecting the data, as the KLV packets are removed from the new file (when viewing with od, or when demuxing with klvdemux) but there is no indication of that detection relayed to the user.
What confuses me is that I've seen instances of data streams being listed in diagnostic output in the archives. How does that work?
Questions that might need answering:
Am I using the right release for examining data streams?
If so: are there any limitations on data stream formats? (I was under the impression that KLV data was straight-forward, but the question is worth asking.) And, is there some undocumented flag that I'm missing in my execution (or configuration)?
(I'm eventually going to want to copy the data stream into new MPEG files, but that's a question for later.)
An example run follows. (It's excessively simple, but the directly relevant bit is the input program stream.)