Last year while working on my new introduction to A*, I decided to try something different for me. Usually I focus on the math, algorithms, and techniques, and let the reader figure out the code. However, for the A* page, I wrote a companion guide that shows how to implement A*. It's simple and unoptimized but I hope it's easy to understand, and shows all the tricky bits that I sometimes gloss over on the main page. While going back through my guide to hexagonal grids, I was improving the pseudocode examples on the page and realized I should probably help people who want to write code.
What usually happens is I have an explosion of questions and possibilities. What languages should I use? What grid variants should I support? What display styles should I implement? Dan Cook writes about alternating brainstorming and culling. I was deep in the brainstorming phase, and came up a crazy idea, that I should procedurally generate code, so that I could generate sample code on the page that matched your choice of grid and programming language, and then I decided I'd learn Haxe macros to do this, and run the code generator both on the server and in the browser, and then also procedurally generate unit tests, and …
… a few months later, I realized I had gone down an unnecessarily long route.
Implementing the code generator made me realize I could simplify the variants. That part was actually great. I learned a lot by thinking through all the different ways to structure the code, and found simpler ways of thinking about hex grids. As I simplified more and found a better class design, I realized I didn't need most of the code generator after all.
Once I realized this, it killed my motivation. I felt bad that I had spent so much time on something that didn't work out.
I had jumped right into the procedural code generator, because that part sounded like fun. And it was!! One mistake I often make with procedural generators is that I start with a cool process instead of starting with the end goal. I did that here. I should've started with the output I wanted to make, and then figured out how to get there.
The code generator project didn't really work out the way I wanted. I wasn't sure where to go from there. Should I add more languages? Should I add more grid variants? Should I add comments to the output? I realized that I was spiraling back into the brainstorming phase instead of culling. I switched to culling. No, I won't add Rust and Scheme and Haskell output. No, I won't add more grid variants. No, I won't add comments and modules and docstrings and instance methods. Instead, I'll write up what I have and share it.
Update: [2015-05-14] I added a bit about why I wanted to generate code (to show sample code on the hex grid guide)