Using FFmpeg within a python application: ffmpeg tool or libav* libraries?

Question

I am working on a python project that uses ffmpeg as part of its core functionality. Essentially the functionality from ffmpeg that I use boils down to these two commands:

ffmpeg -i udp://<address:port> -qscale:v 2 -vf "fps=30" sttest%04d.jpg
ffmpeg -i udp://<address:port> -map data-re -codec copy -f data out.bin

Pretty simple stuff.

I am trying to create a self-contained program (which uses the above ffmpeg functionality) that can easily be installed on any particular system without relying on that system having the necessary dependencies, as hopefully I would package those dependencies with the program itself.

With that in mind, would it be best to use the libav* libraries to perform this functionality from within the program? Or would a wrapper (ffmpy) for the ffmpeg command line tool be a better option? My current thinking on the drawbacks of each is that using the libraries may be the best practice, but it seems overly complex to have to learn how to use them (and potentially learn C, which I've never learned, in the process) just to do those two basic things I mentioned above. The libraries overall are a bit of a bit of a black box to me and don't have very much documentation. But the problem with using a wrapper for ffmpeg would be that it essentially relies on calling a subprocess, which seems somewhat sloppy. Although I'm not sure why I feel so viscerally opposed to subprocesses.

Answer 1

It's somewhat of a matter of opinion, but I would suggest using the ffmpeg CLI in a subprocess as long as you're doing something that it supports well, only using the libav* libraries if you have some requirement that the CLI can't really satisfy.

Although you can get more flexibility with the libraries, the API is very intricate, and you would probably spend most of your time duplicating what the CLI already does (ffmpeg.c, just the main program gluing the libraries together, is around 4800 lines, not including its 3700 line option parser). And it's likely you would add a few bugs along the way — especially if you're lacking C knowledge. So if you can get the CLI to do what you need, that's undoubtedly the path of least resistance. There's no shame in subprocesses: that's the Unix way!