WIP: Adds mediainfo sniffing and auto-thumbnailing with ffmpeg.

[tomatohater]

Don't merge this. Tests are not written yet and error handling needs to be improved, but I'd like to get some eyes on this before going much further. Thoughts?

---

Per issue #5 , this merge adds support for metadata extraction (duration, height, width) and thumbnailing if ffmpeg is installed and a path specified in settings.

A few notes:

• Currently, only works with local file storage (but that was already the case since os.path is used). We can improve the whole thing to be storage independent in the future.
• I had some ideas about making the sniffer/thumbnailer backends swappable. Like maybe use lighter alternative to ffmpeg for audio file.
• I might actually decouple metadata sniffing from thumbnailing... making them completely independent from eachother.
• The sniffed mediainfo is stored in the model, but not exposed in the UI. It's json data stored in a TextField for max compatibility. It would be nicer if is was a JSONField.
• I might make the new Django settings more robust.
• I'd like to do some testing with LARGE media files to see performance implications. It may make sense to do this work asynchronously, but that may complicate things.

[nimasmi]

Thanks, @tomatohater, this initially looks good.

I haven't gone so far as testing it, though I think that some exception handling should be introduced (in media_sniff() perhaps?) in case the return from ffmpeg isn't as expected.

[nimasmi]

NB this leaves no way to leave a thumbnail blank. I'm happy with this, but just pointing it out for discussion if anyone has an opinion.

[tomatohater]

I agree with you here. I think I'll add a "No thumbnail" checkbox to the Media admin UI.

[nimasmi]

I agree with your comment that just dumping this data dictionary into a text field feels wasteful.

[tomatohater]

I felt compelled to store the results of the media info extraction alongside the Media object. Maybe it should be called "raw_mediainfo" or something. I'm not sure how wasteful it is. Probably less wasteful than re-extracting via ffmpeg each time.

I don't love that it's a Textfield.... I'd prefer a JSONField but wanted to maximize compatibility.

[tomatohater]

Thanks for your comments, @nimasmi. I wanted to get feedback on the direction before going too far. I'll add the tests and exception handling to this PR (as well as the other tweaks already mentioned).

[thibaudcolas]

Marking this PR as draft since it’s missing a lot to be considered for merging.

I see @thenewguy closed #5 on the basis that this was outside of the scope of this project (#5 (comment)) – won’t close this PR for now but we should discuss this before anyone spends more time on this.