Advent of Code 2021 in Kotlin - Day 20


The Day 20 problem seemed to be quite straightforward at first sight and the base can be passed here really quickly. The magic is in boundaries conditions of the task and the given sample, that doesn’t include such situation, so we have to deal with it on ourselves.


To solve the problem of enhancing the image, we need some more information that the locations of lighten pixel (that we store in Image::enlighten field). We need to take care of the current background state, as it can also change during the enhancement. However, it can only change from fully empty to fully filled with enlighten pixels, so it’s enough if we remember only a Boolean flag for this state in fillInfty.

It’s not so obvious at first to include the infinity of image also in its computations, but it was definitely the hardest part of this task (and to realize what’s going on when base sample is working but the final answer is wrong).


object Day20 : AdventDay() {
  override fun solve() {
    val data = reads<String>() ?: return
    val algorithm = data.toAlgorithm()
    val image = data.toImage()

    image.enhance(algorithm, times = 2).enlighten.size.printIt()
    image.enhance(algorithm, times = 50).enlighten.size.printIt()

private val Char.isLight: Boolean get() = this == '#'

private fun List<String>.toAlgorithm() = take(1).single().let { Image.Algorithm(it) }

private fun List<String>.toImage() = drop(2).flatMapIndexed { y, line ->
  line.mapIndexedNotNull { x, c -> if (c.isLight) Pixel(x, y) else null }
}.toSet().let { Image(it, fillInfty = false) }

private data class Pixel(val x: Int, val y: Int) {
  infix fun on(s: Image.Surface) = x in s.x && y in s.y

private class Image(val enlighten: Set<Pixel>, val fillInfty: Boolean) {
  private val surface = with(enlighten) {
    Surface(minOf { it.x }..maxOf { it.x }, minOf { it.y }..maxOf { it.y })

  fun enhance(algorithm: Algorithm, times: Int) = (1..times)
    .fold(this) { img, _ -> img.enhanceStep(algorithm) }

  private fun enhanceStep(algorithm: Algorithm): Image = buildSet {
    for (x in surface.x + 1) for (y in surface.y + 1) Pixel(x, y).let {
      val encoding = encoding(it)
      val state = algorithm(encoding)
      if (state) add(it)
  }.let { Image(it, if (fillInfty) algorithm(0b111111111) else algorithm(0b000000000)) }

  private fun encoding(p: Pixel) = sequence {
    for (yi in -1..1) for (xi in -1..1) yield(Pixel(p.x + xi, p.y + yi))
    .map { if (it on surface) it in enlighten else fillInfty }
    .fold(0) { acc, b -> 2 * acc + if (b) 1 else 0 }

  private operator fun Int) = first - i..last + i

  class Surface(val x: IntRange, val y: IntRange)

  class Algorithm(data: String) {
    private val lightOn: Set<Int> = data
      .mapIndexedNotNull { idx, c -> if (c.isLight) idx else null }.toSet()

    operator fun invoke(x: Int) = x in lightOn

Extra notes

We used extension properties as well extension functions in our solution to make it more readable. For example, it’s more convenient to define the

private val Char.isLight: Boolean get() = this == '#'

if we check for this equality a few times in a file, and it can be precisely named. That’s just a single line that enables nice syntax like c.isLight instead of writing the symbol explicitly with ==.

Notice also the definition of infix fun on for Pixel class that was later used in encoding method. We can define such functions in Kotlin for every type and give them the names, which make reading code more pleasant. Remember about that, when writing your libraries in Kotlin, just to give the developers possibility to use infix notation for functions with single argument.

Once again, we should see and remember how the builders for collections in Kotlin can be used. Let’s see that the usage of buildSet { } contains just single nested instruction, while it is a nested for loop with the let { } usage on pixel - it’s really efficient approach of going through the image and building the new one at the same time.

We came up with also some tricky local definition of extension function for IntRange that made it expand in both directions by just writing

private operator fun Int) = first - i..last + i

Then, it was used to process the pixels from the border in standard loop by iterating like

for (x in surface.x + 1) for (y in surface.y + 1) 

This kind of approach is really cool and removes a lot of code repetitions, but we need to define them usually with private visibility, as they might have been understood differently in different contexts, e.g. we could have also

private operator fun Int) = first..last + i

or even

private operator fun Int) = first + i..last + i

so always remember to make sure, that the other developers will understand what you meant or just forbid using your definitions outside your world, to make code safe 😉.

Student of Computer Science

My interests include robotics (mainly with Arduino), mobile development for Android (love Kotlin) and Java SE/EE applications development.