Today's puzzle required us to find low points and basins in a heightmap like this:


0 is the lowest possible value, and 9 is the highest.

Today's grid is dense, with fixed bounds, so we could use an array, but I prefer being able to index the map directly, rather than converting back and forth between indexes and Point tuples.

Point = tuple[x: int, y: int]
Heightmap = Table[Point, int]

Parsing the input wasn't too tricky.

func parseHeightMap(input: string): Heightmap =
let lines = splitLines(input)
for y, line in lines:
for x, ch in line:
result[(x, y)] = parseInt($ch)

We need to find the sum of the risk levels of all the low points on our heightmap.

A low point is defined as a location that has a lower height than all of it's adjacent locations.

We can start by creating an iterator that yields the adjacent points for a given point.

iterator adjacentPoints(p: Point): Point =
let (x, y) = p
yield (x - 1, y)
yield (x + 1, y)
yield (x, y - 1)
yield (x, y + 1)

This iterator doesn't know anything about the boundaries of the heightmap though, so it will happily return points outside of the boundaries.

We can account for this by overloading the [] operator for Heightmap, to ensure that any accessing missing keys will return 9, rather than throwing a runtime error. This gives the heightmap a "walled" effect.

func `[]`(map: Heightmap, p: Point): int =
getOrDefault(map, p, 9)

It's cool that core bits of the syntax are implemented as procs that you can overload, but the tradeoff is that you have to learn type specific sematics for [] (and other operators).

In practice, if you're unsure which overload will be used, you can get more info by hovering over a specific usage of [], and even jump to the definition.

We can use these new functions to determine whether a given point is a low point.

func isLowPoint(heights: Heightmap, point: Point): bool =
result = true
for next in adjacentPoints(point):
if heights[point] >= heights[next]:
return false

Finally we can calculate the risk level directly to solve the first part of the puzzle.

func part1(input: string): int =
let heights = parseHeightMap(input)
for point, height in heights:
if isLowPoint(heights, point):
result += 1 + height

The next part requires us to multiply the sizes of the three largest basins together.

A basin is defined as a set of points that all flow down to the same low point. Locations of height 9 do not belong to any basins.

We have a guarantee that other locations will be part of exactly one basin, so we know that basins must be bounded by 9s. If a boundary location was lower than 9 then it would be part of multiple basins!

I suspect that the intention here was for us to flow "uphill" from the low points we found in part one, but looking at the examples it seemed simpler to flood fill across everything that wasn't a 9.

func findBasins(heights: HeightMap): seq[seq[Point]] =
var visited: HashSet[Point]

for point, height in heights:
if height >= 9 or point in visited: continue
var basin: seq[Point]
var stack = @[point]

while stack.len > 0:
let point = stack.pop

for next in point.adjacentPoints:
if heights[next] >= 9 or next in visited: continue

basin = @[]

There's quite a bit to break down here. The rough idea is to iterate over every point in the map, checking whether it can be a part of basin (<= 9) and whether it's already part of a basin we've identified (in visited).

If the point is valid and we haven't already visited it, then we start a stack based flood fill (depth-first search). When the stack is empty, we add the basin to the result, and continue.

Once we've discovered all basins, we can get the answer by multiplying the sizes of the three largest.

func part2(input: string): int =
let heights = parseHeightMap(input)
let basins = findBasins(heights)
var sizes = basins.mapIt(it.len)
sort(sizes, SortOrder.Descending)
sizes[0] * sizes[1] * sizes[2]

I tried a couple of things differently with Nim today.

I find the proc keyword an odd decision on Nim's part. Variants of function (func, fun, fn, defn, def) are common across many languages and it feels like there needs to be an important distinction to justify using a different term. Whilst reading about Nim yesterday, I discovered that there is in fact a func keyword, too.

It's a shorthand for creating a proc with the {.noSideEffect.} pragma. I don't typically write procs that perform side effects, but I'm not keen on typing out that pragma alongside every proc, so I decided to try func out for today's puzzles.

As I understand, it's a relatively recent addition to the language, so you don't see it around much whilst reading the standard library, or other people's code.

I have mixed feelings.

I prefer reading and writing func, and I like that I get the no side effect guarantee without pragmas, however, I found myself having to constantly switch back and forth between func and proc whenever I wanted to drop an echo to debug.

I can imagine for codebases with longer lifespans than this one, those guarantees are important enough to be worth it, but for Advent of Code, it just created extra work.

I've noticed that the documentation doesn't often use the method call syntax, and instead seems to prefer simple function calls (e.g. len(xs) instead of xs.len). I tried to use less of that syntax today, to see whether it felt any different.

It wasn't a huge change, but I think it's easier to write code that reads naturally when you are using method calls.

Here's all that code together:

GitHub Day 9