GoKawiilContext
I've been working on Pizza Legacy, an open-source reimplementation of the 1994 DOS game Pizza Tycoon . The game has a close-zoom street view of the cities, and when you scroll around it you can see a steady stream of cars driving through the streets. Maybe 20 or 30 tiny sprites at a time, but they navigate the road network, queue behind each other at intersections, and generally look like a living city. Yes, it was a bit buggy because sometimes they would drive through each other, but it was good enough to just give some sense of life to the map. All that on a 25 MHz 386 CPU.
The first thing I implemented in 2010 when I started this project was that close zoom level, but it took 14 years before I finally had the cars driving around on it, in a way that I was happy about; I had multiple attempts over the years but every time I ran into problems I got stuck building an overly complicated system that was hard to reason about and no fun to work on.
One attempt in 2017 involved each tile keeping track of which positions were occupied, and every car had to ask the grid for permission before moving, reserving and freeing slots as it went. It basically turned into a shared locking system just to move a few pixels, with cars and tiles constantly trying to stay in sync.
All the while I had this nagging thought in the back of my mind: the original Pizza Tycoon ran this on a 25 MHz CPU, so why were my versions always so complicated?
Finally I went to the assembly (which I had spent many years slowly understanding better and documenting) to figure out what the original was doing, with the help of LLMs which were (a couple of years ago) this new and exciting technology that could better understand assembly than I could.
Now that I finally have it working I can see where I went wrong: I went into it with a brain full of modern concepts: scene graphs, path finding, collision detection, and of course plenty of CPU to run it all!