Stephen Tetley wrote:
Events in FRP / Yampa are typically key presses / mouse movement, so a MIDI controller generating Note-on / Note-off events would be a direct analogue to key presses.
More problematic is that FRP models hybrid (continuous and discrete) systems. For me at least, MIDI seems essentially discrete - a stream of control events. In MIDI files control events are twinned with a time stamp so they can be played. Presumably events are instantaneous in real-time interactive MIDI - not something I've looked at.
Working with an FRP system like Yampa might add a lot of complexity, which admittedly you should be able to ignore - but initially it might be difficult to identify what parts are needed for a "mostly" discrete system like MIDI. (If you are time-stamping MIDI events yourself you will presumably need to sample a running clock which seems like a continuous behaviour...)
Unfortunately I can't think of any systems in Haskell that are "more discrete than continuous" so you might have to choose a FRP system anyway.
Concerning FRP, I would like to advertise my reactive-banana library here, which tries to follow Conal Elliott's semantics with behaviors and events. http://www.haskell.org/haskellwiki/Reactive-banana I intend to do audio / MIDI programming in the future, so it's going to be well-supported for your particular purpose, even. That said, I haven't started to use it for MIDI myself yet, so I appreciate any kind of feedback! If you want to learn reactive-banana, I recommend that you have a look at the source code of the model implementation in Reactive.Banana.Model. It's intended to be really simple to understand and it's the authoritative reference for the semantics of the actual implementation (which is far from simple to understand). As you can see, the model uses infinite lists. The advantage of the actual implementation, especially for MIDI, is that it is *real-time*, something which is tricky to do with infinite lists. Still, you could probably use the model as a guide for cooking up your own FRP library.
Incidentally, I've been working on a MIDI "animation language" for the last couple of days based on the animation language in Paul Hudak's book. I've wanted continuous behaviours to model modulating volumes (crescendos, decrescendos) and panning, but I've found the work tough going for modelling the note lists where I want the system discrete in both input (specification) and output.
Consider me interested. How does your approach compare to Conal-style FRP with behaviors and events? Best regards, Heinrich Apfelmus -- http://apfelmus.nfshost.com