Question about possibility of HTTP streaming support

[CharlesWiltgen]

Hey @sbooth, in 2022 you said, "The HTTP input source pathway is the least-tested so it's definitely possible SFBAudioEngine is doing something incorrectly with NSURLSessionDataTask, especially when it comes to seeking."

Fair enough! Is that still true? I've been using happily using SFBAudioEngine to build a Yet Another Music Player, but I'm starting to think that I incorrectly assumed that HTTP streaming would be achievable. That's totally on me, of course, but it would be great to get a sanity-check. Thank you!

[sbooth]

In 2022 🤣 has it been that long!?

It's been a while since I've thought about this but at the time I remember thinking that with the current design it would be difficult to get HTTP streaming working because a decoder is expected to be ready to provide samples as soon as -openReturningError: returns; there is not a provision for a buffering period or anything asynchronous is nature. Things have changed a bit with SFBAudioEngine since then so perhaps it is achievable without too much rearchitecting, but I would have to take another look.

[CharlesWiltgen]

@sbooth, I consider this answered and closable. Here's a quick update:

HTTP streaming with SFBAudioEngine works great in Poppy now (local files, plus remote streams from all media servers and internet radio). The constraint you described (decoder must be sample-ready when ‑openReturningError: returns) was the hard part. Once I worked around that, the rest was straightforward. At a high level:

• Poppy has a custom SFBInputSource subclass backed by a custom async fetch + byte-cache layer, rather than the built-in HTTP path
• All async work happens before SFBAudioEngine's -openReturningError: is called, so by the time SFB asks for length, seekability, or bytes, real answers are ready
• Seeking is handled outside SFBAudioEngine, via fresh HTTP range requests, so we never ask a single URLSessionDataTask to seek mid-stream

That's it! The implementation is interesting (I think), but conceptually it's just "prepare asynchronously, then present a synchronous-looking input source to SFB". Feel free to close this, and let me know if you ever want to compare notes.

Thanks again for SFBAudioEngine! If you're looking for a great music experience on iPhone (and soon other Apple devices), Poppy's shaping up to be nice showcase for it.