libav.js coordination

[Yahweasel]

Hello,

This is not a specific report or request. More like a communiqué.

I'm the author of libav.js, which is essentially the browser/WebAssembly version of the same project. It's not infrequent that I get people using libav.js only because there is or was nothing better at the time for Node, so I appreciate a similar project for Node existing.

Unfortunately, while our goals our the same, our APIs could not be more different. Mine is lower level, directly exposing pointers (well, WebAssembly addresses, so quasipointers), and is mostly a reflection of the C API. It has no classes, but is also asynchronous (part of why it has no classes), so that the actual work can be done on WebWorkers and/or with threads.

What I'm wondering is if there's any enthusiasm from you, me, or users to bridge the gap. Perhaps a cross-compatible API could be developed, perhaps a compatibility layer for one or the other, or perhaps TransAVormer, which already bridges the gap between libav.js and WebCodecs, could bridge more gaps.

As I said, I have no specific report or request. I just wanted to create this issue to open the discussion, if there's a discussion to be had.

[longnguyen2004]

Hello, you might recognize me :) I just found out about this project too, and are considering switching to it.

I'd say the low level API is pretty similar to what libav.js have, only with a small layer of abstraction to make it somewhat object oriented (although the abstraction leaks out in some ways, as can be seen by (packet | frame).alloc() and (packet | frame).unref() being exposed, instead of being tied to the garbage collector).

Probably the biggest difference is how non-file inputs are handled. This project exposes AVIOContext, while libav.js doesn't, and relies on global callbacks instead. I've always seen libav.js method as a bit of a hack (which it probably is). Is it possible to expose AVIOContext in libav.js in some ways?

[Yahweasel]

Given that I/O was one of the very first things I had to implement, I suspect that the short answer is that I did what was more expedient at the time. I do recall looking into making an AVIOContext driver and rejecting that idea for some reason, but that could've been before Asyncify even worked particularly well, so it's entirely possible that that would be a better technique in modern Emscripten. PR's welcome ;)

[seydx]

Hi @Yahweasel,

First of all, sorry for the late response - this got buried under other projects 🫣

Thanks for reaching out! I appreciate libav.js and the work you've put into it.

While our APIs are quite different by design (node-av focuses on a class-based approach with both low-level and high-level APIs), I'm not opposed to the idea of some form of compatibility layer or shared abstractions.

I don't have the bandwidth to actively drive such an effort myself, but if someone from the community wants to explore this (e.g., a shared high-level interface or TransAVormer integration), I'd be happy to provide guidance and review PRs.

Feel free to keep this issue open for discussion :)