Hacker News Reader

Comment by jmull | Hacker News Reader

To be fair, those memory mapped locations of the C-64 were APIs… for accessing the graphics, sound, and i/o capabilities of the system. It was a pretty nice design. One abstraction made the entire system programmable in a flexible, intuitive way and played well with the CPU’s native machine language. You just needed a good memory map. (A knack for memorizing useful memory locations didn’t hurt either)

View on HN

Casey Muratori proposes going back to the memory-mapped-io-as-universal-hardware-api in "The Thirty Million Line Problem" https://www.youtube.com/watch?v=kZRE7HIO3vk

It may have been possible to use them as an API, but it was different in the sense that you were interacting with the hardware directly. There were some nasty consequences to that (it was difficult to maintain compatibility between hardware models, nevermind hardware vendors). There were also some cool consequences to that (you could bend the hardware to do a lot of undocumented stuff).

That said, even firmware calls (the closest thing to a modern software API) were a different beast back then. You could buy documented assembly dumps of the ROM for some computers. Not only was it useful for figuring out how things worked, but you could figure out alternate entry points to use that firmware in undocumented ways.

A memory mapped API isn't inherently more (or less) subject to compatibility issues and versions than any other kind of API. With the C-64 it was hard-wired but that's not any more necessary than burning the OS on to ROM. (Which is to say: totally necessary at the time, for a machine of that capability and cost, but not at all necessary for systems that have the APIs we're comparing it to.

Reply on news.ycombinator.com