[FFmpeg-user] ffmpeg architecture question

[Paul B Mahol]

On 4/17/20, Mark Filipak <markfilipak.windows+ffmpeg at gmail.com> wrote:
> Another cogent point:
>
> Suppose I put 'datascope' before a filter that would pass the original frame
> (say, based on a
> color), but that the filter won't pass the 'scope' image (because it doesn't
> contain that color). I
> haven't tried it, but I'll bet that the 'scope' image doesn't appear at all
> and that the frame is
> dropped.
>
> (Note that if I moved the 'datascope' to after the filter, it would work as
> expected.)
>
> Supposing that I'm correct, and considering the prior experiments that I did
> conduct, the
> non-recursive nature of ffmpeg filter complexes is an important
> architectural feature that's not
> documented.
>
> Understand that I'm not criticizing ffmpeg. ffmpeg works how it works and
> that's fine. It just
> doesn't work like an oscilloscope. It could have been designed to work like
> an oscilloscope, but it
> wasn't. That's okay, but it really should be documented.
>
> Do you agree, Paul, or am I mistaken?

Tried this simple command?

ffmpeg -f lavfi -i testsrc2 -f lavfi -i testsrc -lavfi
"[0:v]select='mod(n,2)'[a];[1:v]select='1-mod(n,2)'[b];[a][b]interleave"
-f null -

Interleave filter use frame pts/timestamps for picking frames.