Would be neat to have an interactive example on gh-pages to demonstrate how the algorithm works:thinking:

The algorithm already minimizes whitespace, but I'm currently searching for a heuristic that prefers layouts that place whitespace closer to the bottom-right corner than leaving them strewn arbitrarily throughout the configuration.

An example of a "bad" configuration (generated automatically):

a "good" configuration (tweaked manually):

The second one is easily the more compact revision. Unfortunately, calculating a score based on whitespace "position" is a lot more difficult than it appears. I've tried sorting by a lot of different positional shenanigans with the boxes (note that this algorithm isn't strictly bound to a grid) but no satisfactory results have come up yet.

There are actually plenty of good box packing algorithms out there, but I made my own version without much regard for optimization since I had planned on using it for a spritesheet generator anyway. Surprisingly, I've found that arranging a mere six boxes creates upwards of 1000 permutations on average and already takes a few seconds to load. Seven boxes made Chromium lock up entirely in my own experience and required a system restart.

I haven't done benchmarks yet (which is another issue entirely), but I'm assuming there's a large performance bottleneck stemming from recursion and the subsequent mass object creation that's going on.

Basically, I'd love to create an imperative version of this algorithm or at least find some loopholes to keep memory usage to a bare minimum.

Vectors are currently modeled as arrays in the current implementation, adding an extra layer of versatility. The current algorithm should be expanded to cover any dimension.