I'd use a convenient platform, often costing well under $200US, as the development platform, in the same way as we old-timers once used a Z80 board as a development platform for a much simpler target. Back then, the OS was CP/M, which allowed one simply to solder a wire-wrap socket into the target board and plug it into the host system where the code could be tried and debugged module-by-module, and subsequently programmed into the target independently of the host platform. The same physical CPU ran the code during the development phase.

I know this can't be done in quite the same simple way with so many different MCU's, but a large CPU/MCU can still run the code with timing/addressing differences handled with a different header file or similar feature, and the code never becomes encumbered by limitations of a simulator as it would using a PC platform.

Real-world physical parameters such as interrupt interactions, etc, can be explored in a way it could not easily be accomplished with a PC-based simulator and cross-assembler/cross-compiler. Moreover, it would, ultimately, be a lot less costly, which would appeal to me and my clients.

RE