AoC 2020 - Day 17: Conway Cubes

This type of puzzle (or at least a variant of it) is a staple in AoC. My initial assumption was to optimize for performance with large sets; but then I decided “oh, screw it and let’s see if really part two will have thousands of iterations”. From there, implementation was simple; and actually that was a good choice, because extending for part two was quite simple, and performance was not a problem (around two seconds). After a couple of optimizations, I was able to get in the below-second ballpark. Interestingly, the bulk optimization was to remove memoization, and instead use a generator (instead of small list) with an early exit in the loop using it. Another optimization, made the code uglier – but it was able to make it much faster (around a quarter of second) so I had to leave it. The previous solution was iterating over node neighbors twice: the first time to get all the candidates for being active, and then to get all their neighbors. However, this can be merged into one single (nested!) loop, recreating the graph. Not nice, but works better – as often with performance optimizations.

John Horton Conway (1937-2020)

I remember first implementing game of life in Turbo Pascal. I should still have that code, somewhere; I am pretty sure it would make me frown for its ugliness.
But I remember watching those patterns, moving around the screen, and hitting performance limits (I did not know about complexity at the time) of my computer when I wanted to make a version that could fully exploit the incredibly high resolution of my graphic card.
All of that, was out of an article I read on an old “Scientific American” found in my father’s collection.
Since then, I reimplemented it countless times; for my pleasure, in interviews, in AoC (at least in some variant). And it was always fun.
This time, however, it was a bit sad.
RIP, Conway