[devdri]

The way it uses pin state (number passed as argument / result) for communicating with the outside world is genius. This way it very clearly matches what happens in hardware each cycle. Typical alternative is that the "cpu" calls read/write on the memory/peripherals - which gets tricky when the exact read/write cycle of the multi-cycle instruction matters for accuracy (e.g. racing the CRT beam).

I wonder if similar approach could realistically be applied to slightly newer generation hardware like GBA with ARM7TDMI.