This is an expansion of the discussion from the "Iterative looping" thread. What is ultimately necessary for that is a way to route the output of Mobius back into itself after passing through other effects plugins, or even out to hardware effects and back. I think this is a very interesting and necessary thing to do. But it's going to require some changes I'll need to explain, and to understand those, there needs to be a good understanding of how latency compensation works.

It might help to think of audio blocks or buffers as little pieces of recording tape. The audio
interface hardware is listening to you the performer and recording what you play on these pieces
of tape. When it fills one up, it takes it out of the recorder and sends it to the application,
then it loads another piece of tape and starts recording that one, this process continues
forever.

What this means is that the application, in this case Mobius, is receiving pieces of tape
containing audio are not "live". They were recorded in the past. What the performer is playing
at that moment is still being recorded on another piece of tape in the audio interface. Mobius won't receive that until the hardware fills up the tape and passes it along. Mobius is always one piece of tape behind where the performer actually is in their live performance.

Now let's assume these tapes are each 1 second long. The performer is slowly playing the C major scale and holding each note for 1 second. C D E F G A B C over and over. It isn't the most thrilling performance, but he's a minimalist, you have to trust his artistry. So the audio interface is busy recording these pieces of tape and handing them off to Mobius. When the performer is playing a D, mobius is still hearing a C because it is one tape behind.

This goes on for an hour, and the performer decides to spice things up with a little looping. He presses Record on the footswitch. At that moment the performer is playing a C. Mobius wakes up and sees that it is supposed to start recording a loop now, so it looks at it's input tape and sees an B, because it is one tape behind. This isn't what the performer wants to record though. Mobius has to be told that it is always one tape behind, so it gets ready to record but doesn't actually do anything yet, it ignores that tape containing the B and waits for the next one. When the C tape comes in it starts recording.

This goes on for 8 notes until the performer reaches another C. After holding that final C for exactly one second he presses Record on the footswitch again to end the loop. What he wants to start hearing is that C-C scale loop start playing right away when he ended the recording. And he immediately wants to start overdubbing notes that are a third above the notes in the loop: E F G A B C D E. Natural minor is for babies, we're going phrygian on this bus.

What he wants to hear is C/E, D/F, E/G, etc. when he's playing an 8 note overdub, then end the overdub at the end of the final E, and hear that pattern repeat. Now we have a problem, and it isn't boredom.

Remember those pieces of tape the audio interface is recording and passing off to Mobius? It's also doing the same thing on the way out. When mobius wants to play something, it instantly makes a little piece of tape and passes it off to the audio interface. It doesn't have to wait to make the tape, all the notes are recorded and ready, it just makes a 1 second tape containing a note and gives it to the audio interface. The audio interface can't do anything instantly however, it has to feed that tape into another player that will slowly play the tape over one second, and this is what you hear coming out of the speakers.

So while the performer is hearing CDEFGABC over and over, what Mobius is actually playing
is DEFGABCD over and over. It has to be one tape ahead of the performer because the audio interface is busy playing the last tape Mobius sent it, even though Mobius is ready to send the next one.

This in a nutshell is Input Latency and Output Latency. Mobius always has to be delaying the recording of what the performer is actually playing, and it has to be playing ahead of what the performer is actually hearing.

I'm going to pause and let that sink in before talking about feedback. Isn't this fun?