# Advent of Code 2023 - Day 17

By Eric Burden | December 17, 2023

It’s that time of year again! Just like last year, I’ll be posting my solutions to the Advent of Code puzzles. This year, I’ll be solving the puzzles in Kotlin. I’ll post my solutions and code to GitHub as well. If you haven’t given AoC a try, I encourage you to do so along with me!

# Day 17 - Clumsy Crucible

Find the problem description HERE.

## The Input - These City Streets

Today’s input parsing is… pretty much all done by the pre-processor. I did write a new function to handle grids of digits, though, so that’s cool. Check this out!

fun resourceAsGridOfDigits(fileName: String): List<List<Int>> =
resourceAsLines(fileName).map { it.map { char -> char.digitToInt() } }


Now I can use this function to read the file (much like the resourceAsGridOfChar from the last few days). With that integer grid in hand, I just pop it into a data class so I can attach new methods to it.

/**
* This class represents the grid of city blocks (and their heat loss amounts)
*
* @property grid The grid of heat loss amounts for each city block.
*/
data class CityBlocks(val grid: List<List<Int>>) {
// Convenience!
val rows: Int get() = grid.size
val cols: Int get() = grid.first().size

// Index a [LaserGrid] with an [Index2D].
operator fun get(idx: Index2D): Int = grid[idx.row][idx.col]

/**
* Check if a crucible is still on the grid
*
* @param crucible The crucible to check.
* @return Is the laser still on the grid?
*/
private fun isInBounds(crucible: Crucible): Boolean {
val (row, col) = crucible.position
return row in 0..<rows && col in 0..<cols
}
}

class Day17(input: List<List<Int>>) {

// Put the grid in the data class and get ready to rock.
private val parsed = CityBlocks(input)

}


I feel like I need a Grid superclass that I can attach all these convenience functions to so I don’t need to re-type them every day. The muscle memory is nice and all, but programming means making my own life easier!

## Part One - Roll That Beautiful Lava Bucket

Now that we have lava, there’s no choice but to roll it through (potentially) crowded city streets. But, on the plus side, we have lava! Also, the crucibles are hard to steer, apparently. Nothing can go wrong!

// For more on these re-used classes and functions, check out earlier
// day blog posts or just look at the code on GitHub.
import dev.ericburden.aoc2023.Utils.Index2D
import dev.ericburden.aoc2023.Utils.Offset2D
import dev.ericburden.aoc2023.Utils.plus
import java.util.PriorityQueue

/**
* This enum represents the four cardinal directions
*
* The four cardinal directions are north, south, east, and west. This
* enum also provides methods for turning from one cardinal direction
* to another.
*/
enum class CardinalDirection {
NORTH, SOUTH, EAST, WEST;

// Turn this direction to the left
fun turnLeft(): CardinalDirection = when (this) {
NORTH -> WEST
EAST -> NORTH
SOUTH -> EAST
WEST -> SOUTH
}

// Turn this direction to the right
fun turnRight(): CardinalDirection = when (this) {
NORTH -> EAST
EAST -> SOUTH
SOUTH -> WEST
WEST -> NORTH
}
}

/**
* This class represents a big ole crucible, rolling through the streets
*
* @property position The location of the crucible on the 2D grid.
* @property direction The direction the crucible is currently heading.
* @property run The number of blocks this crucible has moved in a straight line.
* @property kind The kind of crucible we have (REGULAR or ULTRA).
*/
data class Crucible(
val position: Index2D,
val direction: CardinalDirection,
val run: Int,
) {
/**
* Move the crucible forward.
*
* @return The crucible that would result from moving this crucible forward
* one step in the direction it is currently going.
*/
private fun advance(): Crucible {
// Update the position with offsets.
val position = position + when (direction) {
CardinalDirection.NORTH -> Offset2D(-1, 0)
CardinalDirection.EAST -> Offset2D(0, 1)
CardinalDirection.SOUTH -> Offset2D(1, 0)
CardinalDirection.WEST -> Offset2D(0, -1)
}

// Increase the run distance
val run = run + 1

// Everything else stays the same
return Crucible(position, direction, run)
}

/**
* Turn the crucible to the left without moving forward
*
* @return The crucible that results from turning this crucible to the left.
*/
private fun turnLeft(): Crucible {
val direction = direction.turnLeft()

// Reset the run to zero when we turn
return Crucible(position, direction, 0)
}

/**
* Turn the crucible to the right without moving forward
*
* @return The crucible that results from turning this crucible to the right.
*/
private fun turnRight(): Crucible {
val direction = direction.turnRight()

// Reset the run to zero when we turn
return Crucible(position, direction, 0)
}

/**
* Identify which states are possible for this crucible
*
* A crucible can always move left or right, but it can only move forward
* when the run length is less than three.
*
* @return A list of all the possible next states for this crucible based
* on its current state.
*/
fun nextStates() = if (run < 3) {
listOf(
)
} else {
listOf(
)
}
}

data class CityBlocks(val grid: List<List<Int>>) {
// val rows: Int get() = ...
// val cols: Int get() = ...

// operator fun get(idx: Index2D): Int = ...

// private fun isInBounds(crucible: Crucible): Boolean { ... }

/**
* Find the shortest path from a starting [Crucible] state to the end index
*
* This is an implementation of Dijkstra's algorithm for finding the
* shortest path. I _considered_ going full A* here, but this runs in
* a sufficiently reasonable amount of time, so I'll leave it as-is.
*
* @param start The starting state of our crucible.
* @param end The index we're pathfinding to.
*/
fun shortestPath(start: Crucible, end: Index2D): Int {
// Minimum priority queue of <heat loss>, <crucible> pairs, ordered by
// minimum heat loss.
val queue =
PriorityQueue<Pair<Int, Crucible>> { o1, o2 -> o1.first - o2.first }

// Mapping of minimum heat attained per crucible state. This can't just
// be position, because the path from a particular location will vary
// depending on the direction and run length as well.
val minHeatLosses = HashMap<Crucible, Int>()
minHeatLosses[start] = 0

// Find that path!
while (queue.isNotEmpty()) {
// Get the state with the minimum heat loss so far from the queue
val (heatLost, crucible) = queue.remove()

// If we've reached the end, we win!
if (crucible.position == end) return heatLost

// Otherwise, identify all the possible next states for the
// current crucible.
val nextStates = crucible.nextStates().filter { isInBounds(it) }

// For each possible future state...
for (state in nextStates) {
// Check to see if the heat loss for reaching this new state
// can be less than any previous amount of heat lost reaching
// this state.
val newHeatLoss = heatLost + this[state.position]
val previousMinHeatLoss = minHeatLosses[state] ?: Int.MAX_VALUE

// If so, we've found a new shortest path to this state. Add
// it to the queue!
if (newHeatLoss < previousMinHeatLoss) {
minHeatLosses[state] = newHeatLoss
}

}
}

// When disaster strikes! If we search through all possible paths and
// don't reach the end _ever_ we get this exception.
throw Exception("Could not find a path to $end from$start!")
}
}

class Day17(input: List<List<Int>>) {

// private val parsed = ...

// In part one, we use regular, everyday giant rolling crucibles to move
// lava around city streets. Totally safe.
fun solvePart1(): Int {
val topLeftCrucible = Crucible(Index2D(0, 0), CardinalDirection.EAST, 0)
val bottomRightIdx = Index2D(parsed.rows - 1, parsed.cols - 1)
return parsed.shortestPath(topLeftCrucible, bottomRightIdx)
}

}


It just wouldn’t be Advent of Code without Dijkstra’s algorithm. I was wondering when we’d see the old guy, and here he is! There’s a bit of a twist with needing to turn the crucible sometimes, but nothing our nextStates logic can’t handle.

## Part Two - Go Ultra, or Go Home

Turns out, what we really need to get all the lava where it needs to go are larger, more difficult to steer crucibles. It’s time to go ultra! The main differences here are that the rules for determining possible next states has changed a bit. Also, we can’t always stop. Yeah… Well, the one thing that doesn’t change, actually, is our pathfinding algorithm. So, that’s nice.

// For more on these re-used classes and functions, check out earlier
// day blog posts or just look at the code on GitHub.
import dev.ericburden.aoc2023.Utils.Index2D
import dev.ericburden.aoc2023.Utils.Offset2D
import dev.ericburden.aoc2023.Utils.plus
import java.util.PriorityQueue

// enum class CardinalDirection { ... }

// In part two, we have ULTRA crucibles, which behave a little differently
enum class CrucibleKind { REGULAR, ULTRA; }

/**
* This class represents a big ole crucible, rolling through the streets
*
* @property position The location of the crucible on the 2D grid.
* @property direction The direction the crucible is currently heading.
* @property run The number of blocks this crucible has moved in a straight line.
* @property kind The kind of crucible we have (REGULAR or ULTRA).
*/
data class Crucible(
val position: Index2D,
val direction: CardinalDirection,
val run: Int,
val kind: CrucibleKind = CrucibleKind.REGULAR
) {
/**
* Move the crucible forward.
*
* @return The crucible that would result from moving this crucible forward
* one step in the direction it is currently going.
*/
private fun advance(): Crucible {
// Update the position with offsets.
val position = position + when (direction) {
CardinalDirection.NORTH -> Offset2D(-1, 0)
CardinalDirection.EAST -> Offset2D(0, 1)
CardinalDirection.SOUTH -> Offset2D(1, 0)
CardinalDirection.WEST -> Offset2D(0, -1)
}

// Increase the run distance
val run = run + 1

// Everything else stays the same
return Crucible(position, direction, run, kind)
}

/**
* Turn the crucible to the left without moving forward
*
* @return The crucible that results from turning this crucible to the left.
*/
private fun turnLeft(): Crucible {
val direction = direction.turnLeft()

// Reset the run to zero when we turn
return Crucible(position, direction, 0, kind)
}

/**
* Turn the crucible to the right without moving forward
*
* @return The crucible that results from turning this crucible to the right.
*/
private fun turnRight(): Crucible {
val direction = direction.turnRight()

// Reset the run to zero when we turn
return Crucible(position, direction, 0, kind)
}

/**
* Identify which states are possible for this crucible
*
* @return A list of all the possible next states for this crucible based
* on its current state.
*/
fun nextStates() = when (kind) {
// A regular crucible (part one) can always move left or right, but
// it can only move forward when the run length is less than three.
CrucibleKind.REGULAR -> if (run < 3) {
listOf(
)
} else {
listOf(
)
}

// An _ultra_ crucible (part two) can _only_ move forward if its run
// length is less than four (or it's on the first space). With run
// lengths of 4-9 (inclusive) it can move forward, left, or right just
// like a regular crucible. At a run length of 10, it must turn.
CrucibleKind.ULTRA -> if (run < 4 && (position.row != 0 || position.col != 0)) {
} else if (run < 10) {
listOf(
)
} else {
listOf(
)
}
}

/**
* Indicates whether this crucible can stop
*
* A regular crucible can stop any time, but an _ultra_ crucible can only
* stop after moving at least four blocks in a straight line.
*
* @return A flag indicating whether this crucible can stop moving.
*/
fun canStop() = when (kind) {
CrucibleKind.REGULAR -> true
CrucibleKind.ULTRA -> run >= 4
}
}

data class CityBlocks(val grid: List<List<Int>>) {
// val rows: Int get() = ...
// val cols: Int get() = ...

// operator fun get(idx: Index2D): Int = ...

// private fun isInBounds(crucible: Crucible): Boolean { ... }

fun shortestPath(start: Crucible, end: Index2D): Int {
// Minimum priority queue of <heat loss>, <crucible> pairs, ordered by
// minimum heat loss.
val queue =
PriorityQueue<Pair<Int, Crucible>> { o1, o2 -> o1.first - o2.first }

// Mapping of minimum heat attained per crucible state. This can't just
// be position, because the path from a particular location will vary
// depending on the direction and run length as well.
val minHeatLosses = HashMap<Crucible, Int>()
minHeatLosses[start] = 0

// Find that path!
while (queue.isNotEmpty()) {
// Get the state with the minimum heat loss so far from the queue
val (heatLost, crucible) = queue.remove()

// ****************************************************************

// Turns out, I did need to change the pathfinding algorithm a
// _smidge_. Need to check to see if we can stop before declaring
// we've found our way through.
if (crucible.position == end && crucible.canStop()) return heatLost

// ****************************************************************

// Otherwise, identify all the possible next states for the
// current crucible.
val nextStates = crucible.nextStates().filter { isInBounds(it) }

// For each possible future state...
for (state in nextStates) {
// Check to see if the heat loss for reaching this new state
// can be less than any previous amount of heat lost reaching
// this state.
val newHeatLoss = heatLost + this[state.position]
val previousMinHeatLoss = minHeatLosses[state] ?: Int.MAX_VALUE

// If so, we've found a new shortest path to this state. Add
// it to the queue!
if (newHeatLoss < previousMinHeatLoss) {
minHeatLosses[state] = newHeatLoss
}

}
}

// When disaster strikes! If we search through all possible paths and
// don't reach the end _ever_ we get this exception.
throw Exception("Could not find a path to $end from$start!")
}
}

class Day17(input: List<List<Int>>) {

// private val parsed = ...

// fun solvePart1(): Int { ... }

// In part two, we use _ultra_ crucibles, which are way more ultra than
// regular crucibles in every conceivable way.
fun solvePart2(): Int {
val topLeftCrucible = Crucible(
Index2D(0, 0), CardinalDirection.EAST, 0, CrucibleKind.ULTRA
)
val bottomRightIdx = Index2D(parsed.rows - 1, parsed.cols - 1)
return parsed.shortestPath(topLeftCrucible, bottomRightIdx)
}

}


We’ve got a bit of a “strategy pattern” thing going with our Crucible now, where the behaviour changes based on what kind of Crucible it is. Again, I started to reach for an abstract class/subclass setup, but I can’t figure out a way to get those shared functions (in this case, advance, turnLeft, and turnRight) to live in the abstract class and still return instances of the subclass that used the method. I could have those methods mutate the object directly, but that seems like it shouldn’t be required. Probably a skill issue.

## Wrap Up

Hello Dijkstra, my old friend! Today’s puzzle was in classic Advent of Code style, pathfinding with rules for which next states are possible. If we see another of these (and we’re likely to), I think I’m going to try abstracting out some of the basic operations I use on grids all the time, like indexing in two dimensions and getting the number of rows and columns.