The MIDI protocol is very specific to music, it works in the context of "note on, key pressed this hard. other note on. note off. note off." etc.

The "normal" connectors are fairly bulky by modern standards (large DIN-style, like on really old keyboards and mice).

The electric level would have been a good basis, so one could imagine a world where MIDI was extended to also transport keyboard keypresses etc, but it would have meant extending the protocol quite a bit, or weirdly mapping concepts onto the music-specific basics. (EDIT: actually, it might be kind of overkill, and thus more expensive than simpler methods. What's useful in a large studio or stage environment isn't really needed on my desk)

In the days of the dinosaurs, some audio cards had a connector that could be used as either a joystick port or a MIDI port (https://en.wikipedia.org/wiki/Game_port ).

I was looking at one yesterday whilst have a clear out ...

MIDI produces great value with encoding a domain specific mapping. It's one of the differences with OSC which, by being too open-ended, provides no standard interpretation..

Vaguely in this vein - there was a game on the Atari ST called "MIDI Maze" which actually used MIDI as a network protocol!

With the thread about Quake Path Tracing engine, I read about FPS history, and Midi Maze is one of the early famous ones

https://en.wikipedia.org/wiki/First-person_shooter

I'm not an expert so I might be wrong but probably because one solution that fits for all can be a bad idea. Driver complexity is one of the reasons most devices are not compatible with all existing OSs by default.

MIDI is - and should be with future standards - simple, slow and reliable: Musical Instruments don't need much/many volume/types of information, so writing the drivers should be fairly simple.

Having said that, some midi to usb converters are still not up to full compliance with the 80s MIDI 1 standard

Does that mean that the converters implement the protocol themselves? I expected them to be roughly FTDI USB UART chips, tied to the MIDI baudrate.

I'm not entirely sure, but the fact that there are compatibility/compliance issues leads me to believe it's a mixture of both.

Either way, the drivers may be the issue: Parsing the serial correctly etc.

I sometimes wonder whether the ongoing USB-C trainwreck is partly due do trying to be everything for everybody. Leading to complex and flaky HW and SW.

I'm not sure I understand your comment. USB-C is still just USB, which was already a universal messaging transport solution. Driver complexity hasn't gone up just because a new physical plug got invented, so presumably you're thinking of the increased number of devices that now use USB, with a C connector, but the fact that it can do "even more device types!" now makes literally zero difference when the spec was already designed to allow all devices: the only difference is that HW/SW is now finally capable of transmitting reliably enough to handle the massive loads those now-supported devices require (hdmi, ethernet, etc.)

But now your USB-C cable can brick your device.

Sysex calls already allowed that since day 1.