[ygra]

Oh, that's something I also did in QBasic ages ago. I since lost the source code, but it basically worked by querying the screen from bottom to top, finding snow pixels (white) and moving them down, unless there was an obstacle. The initial snow on the ground (and the snowman) were using a different color that was almost white so it wouldn't detect as snow. It worked fairly well in 320×200 at the time.

I've tried the same approach in Turbo Pascal with BGI in the hopes of having a faster language and higher resolution available. It turned out to be quite a bit slower, likely because drawing and querying pixels was a bit more involved when using an adapter layer like BGI.

A few weeks ago I tried rewriting all that in TypeScript for fun and also trying to integrate it as an easter-egg with our graph drawing library (which renders with SVG) and first had to figure out how to efficiently support arbitrary obstacles that are initially SVG as well as a potentially changing viewport of the whole scene. I got sidetracked and didn't finish it, but proper collision handling was so easy back then (just look at the pixel color), but now with vector graphics and reading pixels being a very slow operation in many cases, it was surprisingly complicated.