On Day 18, we have lava rain.
The lava falls into a pool of water, and we estimate how long it’ll take to cool by calculating the surface area of the lava droplets. The droplets are composed of voxel cubes, and we only count surface area that water can reach- the laval droplets have internal air bubbles.
My solution desginates a square outside of the boundary of the droplet as “outside”, and then incrementally designates reachable empty squares as outside, within a box around the droplet.
Once this is done, we identify surfaces that touch the outside just by checking whether the neighbor is “outside” or “nothing”.
Visualization
The visualization shows the two phases of the problem- propagating squares defined as outside, and then scanning the droplet to identify outside-touching cubes.
This was a pretty easy solve for me, which is nice, because it gave me a bunch of extra mental energy to do some cool visualization work!
New Features
Wow! So much.
Bug Fixes
First of all, I had a bug in the line-drawing code in github.com/asymmetricia/pencil, which was not properly computing alpha as part of the process of aliasing. Fixed!
isovox
Next is isovox. My first pass at rendering this used fauxgl. fauxgl is
lovely, but when it comes to me and AoC visualizations, it feels a little
cheaty.
So, I wrote my own isometric voxel rendering, which can render three dimension isometric grids of colored cubes at arbitrary resolution, with alpha.
GIF Alpha Blending Palettes
The next challenge was taking these now rather complicated images and producing GIFs from them. I spent a lot of time playing with palettes and Floyd-Steinberg dithering. Ultimately I settled on augmenting the Tol vibrant palette with the colors from the WebSafe palette to use when blending/dithering. This produced pretty good results!
But it also produced 150MiB GIFs.
Direct MP4 Rendering
In the past, when my visualizations produced large GIFs, I’d just suck it up, keep them, and transcode them to MP4 for sharing. This produces somewhat sub-optimal results. Especially in this case, where the pipeline becomes:
- Go renders bitmap in memory
- Go palletizes bitmap to a GIF
- Go difference-optimizes the GIF
- Go renders the GIF to disk
- ffmpeg / handbrake renders the GIF to mp4 (manually)
Wow!
But it turns out you can just feed ffmpeg a stream of PNG files, and with the right options it will happily render them directly to a video file.
So, with the help of some goroutines, I now have a procedure to fire up ffmpeg, and render images then hand them off to ffmpeg to encode into video. And that’s what you see here!
See you next time, folks.