I want a new mixer with new functionality;
I am kind of lazy, and have ended up starting building the mixer on top of my preexisting MIDI synthesizer;
(note: just because it is MIDI doesn't necessarily mean this really has much to do with music, vs say, using it as a command-language for controlling a game mixer...).
all this then leads to a seeming "solution" of sorts:
hacking MIDI in some nasty ways, basically to allow more functionality (basically, by hacking said functionality directly into the MIDI command-stream).
types of functionality being added:
support for loading arbitrary samples / patches (basically, the ability to pass the synth either a patch name, or raw sample data);
support for a larger max number of channels (probably setting a limit somewhere between 1024 and 16384, vs the original 16);
support for 3D spatial positioning;
I debated the specifics for a little while, but settled mostly on using the 0xFD byte mostly as an escape-code for new wider command-space (still nicer IMO).
For example, older style events:
Vs newer wider events:
where an SV is a signed variable-length integer value (1 or more bytes).
as well as adding things like vector-based ControllerChange events:
Int Vector ControllerChange
Float Vector ControllerChange
and, other commands like:
yes, it is ugly, but probably still cleaner than basically hacking in extensions via a pile of ControllerChange events (and the occasional SysEx).
note that there is little particular need for interop in this case (since this data will unlikely leave the program), and I had partly considered throwing out MIDI up-front, but mostly didn't really want to have to deal with the hassle of providing a "this is not MIDI" flag for parts of the pipeline already set up to assume MIDI, seemed easier mostly to just hack it.
similarly, this hacked form assumes the ability to send full 8-bit data.
now, whether or not there is a whole lot of a point is debatable...