Introduction

I really like Advent of Code, as you probably can guess. I’ve solved every one of the 200 puzzles available at the time I write this (in late summer 2023). With the ninth edition coming in a few months, it’s a good time to review the eight existing years of Advent of Code and offer my impressions.

2015

The first installment of a long-running series is usually the strangest. Rules of the setting are looser or not established. The writing varies from the formula later installments will establish. So it is with Advent of Code’s first year. According to Eric Wastl, the sole author of Advent of Code, the original idea was just to entertain his friends with Christmas-themed puzzles. The first year accordingly has a more informal feel than what would come later. The difficulty spikes and troughs unexpectedly, as with the surprisingly easy Aunt Sue puzzle. Some puzzles lack the creative twists that became the series’ hallmark. Day 9, for instance, is just a traveling salesman problem small enough to brute-force. Overall, the year is probably the easiest of all; perhaps Wastl underestimated just how determined people would be to get those gold stars.

The narrative style is different, too. Later years would take the player on a convoluted adventure to save Christmas, but 2015 instead offers a series of disconnected Christmas scenarios. This might seem like mere flavor, but for me it gives Advent of Code much of its charm. Instead of a straightforward problem specification, you’re told to compute how long a lost Elf would take to wander through a snowstorm, or [emulate Tetris for some angry elephants](https://adventofcode.com/2022/day/17. Uncovering the programming problem wrapped in the silly plotline is half the fun, so I missed the absence of one to tie this year together.

But the charms of the series shine through nonetheless. Most of the classic puzzle topics appear: Conway’s Game of Life variations, made-up assembly languages, string manipulation, number theory, combinatorics, and deductive logic. The wry tone of the series establishes itself, as do the clever part 2 twists. (My favorite is day 13. In part 1, you have to seat people with strong opinions of each other at a dinner table in a way that maximizes their total happiness. Part 2 reveals that you forgot to reserve a place for yourself, but it’s not too hard, since you’re indifferent to who you sit next to). And the website had its classic blue-and-gold look from the start.

It’s easy to see in hindsight how Advent of Code became far more popular than anyone expected. A new edition for 2016 had to happen. But how would it improve on what came before?

2016

With a fancier narrative and tougher, more complex puzzles, naturally. As day 1 opens, you are airdropped a short distance from the Easter Bunny’s headquarters. He has stolen all 50 of Santa’s stars, and you need to retrieve them in time to save Christmas. This hilarious scenario (and my favorite from the seven years with proper plotlines) introduces the standard Advent of Code plot: some calamity has befallen Santa and scattered his stars, and you have to recover them before Christmas is ruined.

The puzzles start off harder than before. 2016’s day 1, which involves tracking the player’s position over a two-dimensional grid, is in my opinion the hardest day 1 yet. The first week or so of puzzles are tough but reasonable, but then comes day 11. This one requires you to find the most efficient way of moving a set of microchips and radioisotope thermoelectric generators through several floors of a building. The obvious method is to find a way to represent states and do breadth-first search, but this gets complicated and finicky. Part 2 is nigh impossible without doing a subtle optimization. Nothing about the puzzle is unfair, but for a second-week puzzle, its difficulty is extreme.

That aside, the puzzles provided a welcome challenge after the relatively straightforward ones of 2015. The only days I disliked required the player to compute MD5 hashes. They resulted in unavoidably slow solutions that just required the player to call their language’s standard library function for MD5.

2017

By 2017, Advent of Code had found its classic formula. MD5 hashes were gone, as were no-frills puzzle designs and sudden difficulty spikes. Everything this year was executed smoothly.

2017 turned down the difficulty from 2016. Many puzzles were simpler than the last year, and nothing anywhere near as hard as 2016’s day 11 appeared. Perhaps Wastl wanted to broaden the series’ appeal to inexperienced programmers. At any rate, the easier challenges provided a welcome change of pace.

Still, plenty of the puzzles offered a good challenge, such as day 23. This one asked you to implement a hideously inefficient program in a made-up assembly language. To solve part 2 in a reasonable amount of time, you had to reverse-engineer it. It was tough but fun to unravel the assembly code and discover a simple program for counting composite numbers within a numeric range.

Overall, I found 2017 fun and engaging. I should also note the ASCII art for the year has stunning animation, in my view the best in the series.

2018

After something of a breather in 2017, 2018 brought perhaps the toughest set of puzzles yet. This year’s plot involved traveling through time to recover stars. Naturally, that adventure ran through some fiendishly complex puzzles.

Several of the later puzzles pushed my skills to the limit. Day 17 is a notoriously hard physics simulation. Day 23 is meant to be solved by octary search, basically binary search in three dimensions. It was as hard (but fun) as it sounds. Day 15 requires the player to implement a combat video game; it’s probably the most complicated puzzle ever. The high difficulty was exhausting at times, but always rewarding.

Overall, the year demonstrated that Advent of Code would emphasize unconventional puzzle designs that rewarded a creative approach. The only hiccup was day 6. For the only time in series history, a bug with some people’s inputs made the puzzle unsolvable for them before it was fixed. All scores for this puzzle were canceled as a result. It’s mostly notable for being the one time there actually was a bug with the website rather than users’ code. Aside from that incident, this year featured some of the best puzzles yet.

2019

2019 was the most ambitious year yet, and also the most controversial. For the first and only time, Wastl changed the formula. Almost half the puzzles in 2019 would require players to work with Intcode: an invented programming language that uses integer opcodes to manipulate data and memory. To solve them, players would have to write an Intcode interpreter or compiler and use it to execute the Intcode program provided as each puzzle’s input.

The Intcode puzzles started simple, but got more and more complex. By late December, players found themselves using Intcode to implement simplified versions of Frogger and Breakout.Then came day 25. Most years have an easy problem on the last day, one meant to give you a relaxing diversion on Christmas morning. This year, it was an Intcode program that implemented an interactive adventure game. It has the genre’s classic elements: unfair deaths, a confusing layout, and an item-swapping puzzle at the end. You could either solve it interactively or write a script to do so automatically - a much richer challenge than a typical day 25 puzzle.

Intcode divided the community. If you search the official subreddit, you can find several threads debating its merits. Those in favor argued that the Intcode puzzles rewarded players for writing reusable code and offered a progressive challenge the classic puzzle format did not. Those against responded that Intcode was convoluted and hard to understand, locking out players who couldn’t keep up with the puzzles.

I think both claims are right. As original and interesting as Intcode was, it was also finicky and unforgiving. I remember wasting a few hours because I misunderstood a minor point in the specification. Overall, I think it succeeded in giving 2019 some unique puzzles, but I don’t blame people for disliking the concept. It probably would have been a lot easier if I’d taken a compilers class, as many participants probably have. All the same, I think it’s good to shake up a well-established formula. I’d like to see an experiment like Intcode again.

2020

2020 is special. It was the first year I took part in while it was ongoing. I stumbled across a link to a puzzle near the end of my fall semester of graduate school, and I decided to try solving it. It was day 12, a fairly easy puzzle that asks you to simulate moving a navigation beacon according to instructions. I had fun solving it, though I wasted a lot of time trying to implement a general rotation matrix to model changes of orientation. (I had yet to learn that Advent of Code only requires integral computations, so I only had to deal with multiples of 90 degrees).

I spent the rest of that winter break solving as many puzzles as I could, learning Python as I went. I had found Advent of Code at exactly the right time in my life. I had just switched from pursuing a master’s in public policy to a master’s in data science, and I knew I would have to learn to write effective code fast. Doing the puzzles increased my confidence that I could do that and confirmed to me that I would enjoy it.

Sentimentality aside, this was a good year, one that returned to the classic format after the Intcode experiment of 2019. There were yet more Game of Life variations, this time in four dimensions or with hexagonal coordinates. It was a fitting tribute to John Conway, who died that year. While the overall difficulty was reduced from the past two years, one puzzle broke the trend. (If you took part in this year, you already know which it is). Day 20, “Jurassic Jigsaw”, requires you to assemble dozens of tiles into an image by matching the patterns on their edges. Part 2 asks you to scan the image for the Loch Ness Monster, which can be recognized by a telltale pattern. It ranks among the very hardest puzzles, but I had fun designing a very complex depth-first search to solve it.

The year’s plot, which involves the player frantically trying to salvage a vacation plan gone awry, is the most realistic so far. Aside from the part with the Loch Ness Monster, anyway.

2021

I began 2021 with far more experience than I had in 2020. And I needed that experience, because this year cranked the difficulty back up to near-2018 levels. I kept up for the first week or so, but I began to fall behind when the puzzles began requiring graph traversal. I knew nothing about graphs or pathfinding algorithms back then, and had to learn fast. I fumbled my way through Dijkstra’s algorithm well enough to solve some of the early puzzles, but I soon fell several days behind.

I persevered, making steady progress as 2022 began. I had to put Advent of Code aside when grad school resumed, but by late spring I had earned all 50 stars. It was the first year I finished completely, and by far the most satisfying to complete.

It took weeks to crawl through days 21 through 24, one of the toughest stretches of puzzles in the whole series. Most of these puzzles demand creativity to pare down huge problem spaces. Day 24, for instance, asks you to find the highest number that meets a condition from trillions of possibilities. I solved it only after painfully reverse-engineering the made-up assembly language provided as the input. Day 23, to me, is the hardest puzzle in the entire series. It asks you to solve a variation on the Towers of Hanoi puzzle. The secret is that it’s pretty easy to solve by hand. (I used cut-up Post-it notes as game pieces). But I couldn’t resist the challenge of writing a program to solve it, one that would work on a general input. It took a week of focused effort to do so. It was painful at time, but I grew so much as a programmer from these puzzles that it was all worth it.

2022

The most recent year, as of this writing, has the player accompany a jungle expedition searching for star fruit, which the reindeer apparently eat. The more relaxed narrative suits the puzzles, which are a bit easier than the previous year’s. The puzzle design seemed more thoughtful than ever; most of this year’s puzzles had simple rules that lacked accidental complexity. Instead, the difficulty emerged organically from the constraints of the problems.

Day 16, perhaps the hardest puzzle, is a good example. It asks you to open a series of pressure valves in rooms connected by tunnels in a way that releases as much pressure as possible. Representing the situation as a simple graph isn’t enough; the number of possible states quickly becomes unmanageable. After asking for advice, I solved the problem by creating a simpler graph that represented distances between pairs of valves, then writing a traverse algorithm that discarded paths that had no chance of releasing the most pressure. This complex design process emerged from a simple problem statement, and I found it more satisfying than trying to meet a complicated specification.

Conclusion

Advent of Code continues to grow more popular. In all likelihood, it will do so for many years to come. Looking back over the eight years and 200 puzzles so far, it’s easy to see why. The puzzles offer endless variations on classic computer science problems, often with an inventive “part 2 twist.” The story tying each year together, while inessential, keeps the mood light. The design of the website makes it easy to check answers, and the carefully worded puzzle descriptions provide enough guidance to find a solution. It success reminds me how many computer science discoveries came from just playing around with interesting problems. I’ll probably keep participating as long as I can, because it never stops being fun.

--- title: "Advent of Code So Far" author: "Ryan Heslin" date: "2023-08-25" categories: ["Advent of Code" ] urlcolor: "blue" --- # Introduction I really like [Advent of Code](https://adventofcode.com/), as you probably can guess. I've solved every one of the 200 puzzles available at the time I write this (in late summer 2023). With the ninth edition coming in a few months, it's a good time to review the eight existing years of Advent of Code and offer my impressions. # 2015 The first installment of a long-running series is usually the strangest. Rules of the setting are looser or not established. The writing varies from the formula later installments will establish. So it is with Advent of Code's first year. According to [Eric Wastl](http://was.tl/), the sole author of Advent of Code, the original idea was just to entertain his friends with Christmas-themed puzzles. The first year accordingly has a more informal feel than what would come later. The difficulty spikes and troughs unexpectedly, as with the surprisingly easy [Aunt Sue](https://adventofcode.com/2015/day/16) puzzle. Some puzzles lack the creative twists that became the series' hallmark. [Day 9](https://adventofcode.com/2015/day/9), for instance, is just a [traveling salesman problem](https://en.wikipedia.org/wiki/Travelling_salesman_problem) small enough to brute-force. Overall, the year is probably the easiest of all; perhaps Wastl underestimated just how determined people would be to get those gold stars. The narrative style is different, too. Later years would take the player on a convoluted adventure to save Christmas, but 2015 instead offers a series of disconnected Christmas scenarios. This might seem like mere flavor, but for me it gives Advent of Code much of its charm. Instead of a straightforward problem specification, you're told to compute how long [a lost Elf would take to wander through a snowstorm](https://adventofcode.com/2022/day/24), or [emulate Tetris for some angry elephants](https://adventofcode.com/2022/day/17. Uncovering the programming problem wrapped in the silly plotline is half the fun, so I missed the absence of one to tie this year together. But the charms of the series shine through nonetheless. Most of the classic puzzle topics appear: [Conway's Game of Life](https://en.wikipedia.org/wiki/Conway%27s_Game_of_Life) variations, made-up assembly languages, string manipulation, number theory, combinatorics, and deductive logic. The wry tone of the series establishes itself, as do the clever part 2 twists. (My favorite is [day 13](https://adventofcode.com/2015/day/13). In part 1, you have to seat people with strong opinions of each other at a dinner table in a way that maximizes their total happiness. Part 2 reveals that you forgot to reserve a place for yourself, but it's not too hard, since you're indifferent to who you sit next to). And the website had its classic blue-and-gold look from the start. It's easy to see in hindsight how Advent of Code became far more popular than anyone expected. A new edition for 2016 had to happen. But how would it improve on what came before? # 2016 With a fancier narrative and tougher, more complex puzzles, naturally. As day 1 opens, you are airdropped a short distance from the Easter Bunny's headquarters. He has stolen all 50 of Santa's stars, and you need to retrieve them in time to save Christmas. This hilarious scenario (and my favorite from the seven years with proper plotlines) introduces the standard Advent of Code plot: some calamity has befallen Santa and scattered his stars, and you have to recover them before Christmas is ruined. The puzzles start off harder than before. 2016's day 1, which involves tracking the player's position over a two-dimensional grid, is in my opinion the hardest day 1 yet. The first week or so of puzzles are tough but reasonable, but then comes [day 11](https://adventofcode.com/2016/day/11). This one requires you to find the most efficient way of moving a set of microchips and [radioisotope thermoelectric generators](https://en.wikipedia.org/wiki/Radioisotope_thermoelectric_generator) through several floors of a building. The obvious method is to find a way to represent states and do breadth-first search, but this gets complicated and finicky. Part 2 is nigh impossible without doing a subtle optimization. Nothing about the puzzle is unfair, but for a second-week puzzle, its difficulty is extreme. That aside, the puzzles provided a welcome challenge after the relatively straightforward ones of 2015. The only days I disliked required the player to compute [MD5 hashes](https://en.wikipedia.org/wiki/MD5). They resulted in unavoidably slow solutions that just required the player to call their language's standard library function for MD5. # 2017 By 2017, Advent of Code had found its classic formula. MD5 hashes were gone, as were no-frills puzzle designs and sudden difficulty spikes. Everything this year was executed smoothly. 2017 turned down the difficulty from 2016. Many puzzles were simpler than the last year, and nothing anywhere near as hard as 2016's day 11 appeared. Perhaps Wastl wanted to broaden the series' appeal to inexperienced programmers. At any rate, the easier challenges provided a welcome change of pace. Still, plenty of the puzzles offered a good challenge, such as [day 23](https://adventofcode.com/2017/day/23). This one asked you to implement a hideously inefficient program in a made-up assembly language. To solve part 2 in a reasonable amount of time, you had to reverse-engineer it. It was tough but fun to unravel the assembly code and discover a simple program for counting composite numbers within a numeric range. Overall, I found 2017 fun and engaging. I should also note the ASCII art for the year has stunning animation, in my view the best in the series. # 2018 After something of a breather in 2017, 2018 brought perhaps the toughest set of puzzles yet. This year's plot involved traveling through time to recover stars. Naturally, that adventure ran through some fiendishly complex puzzles. Several of the later puzzles pushed my skills to the limit. [Day 17](https://adventofcode.com/2018/day/17) is a notoriously hard physics simulation. [Day 23](https://adventofcode.com/2018/day/23) is meant to be solved by octary search, basically binary search in three dimensions. It was as hard (but fun) as it sounds. [Day 15](https://adventofcode.com/2018/day/15) requires the player to implement a combat video game; it's probably the most complicated puzzle ever. The high difficulty was exhausting at times, but always rewarding. Overall, the year demonstrated that Advent of Code would emphasize unconventional puzzle designs that rewarded a creative approach. The only hiccup was day 6. For the only time in series history, a bug with some people's inputs made the puzzle unsolvable for them before it was fixed. All scores for this puzzle were canceled as a result. It's mostly notable for being the _one_ time there actually was a bug with the website rather than users' code. Aside from that incident, this year featured some of the best puzzles yet. # 2019 2019 was the most ambitious year yet, and also the most controversial. For the first and only time, Wastl changed the formula. Almost half the puzzles in 2019 would require players to work with Intcode: an invented programming language that uses integer [opcodes](https://en.wikipedia.org/wiki/Opcode) to manipulate data and memory. To solve them, players would have to write an Intcode interpreter or compiler and use it to execute the Intcode program provided as each puzzle's input. The Intcode puzzles started simple, but got more and more complex. By late December, players found themselves using Intcode to implement simplified versions of [Frogger](https://en.wikipedia.org/wiki/Frogger) and [Breakout](https://en.wikipedia.org/wiki/Breakout_(video_game)).Then came day 25. Most years have an easy problem on the last day, one meant to give you a relaxing diversion on Christmas morning. This year, it was an Intcode program that implemented an interactive adventure game. It has the genre's classic elements: unfair deaths, a confusing layout, and an item-swapping puzzle at the end. You could either solve it interactively or write a script to do so automatically - a much richer challenge than a typical day 25 puzzle. Intcode divided the community. If you search the [official subreddit](https://www.reddit.com/r/adventofcode/), you can find several threads debating its merits. Those in favor argued that the Intcode puzzles rewarded players for writing reusable code and offered a progressive challenge the classic puzzle format did not. Those against responded that Intcode was convoluted and hard to understand, locking out players who couldn't keep up with the puzzles. I think both claims are right. As original and interesting as Intcode was, it was also finicky and unforgiving. I remember wasting a few hours because I misunderstood a minor point in the specification. Overall, I think it succeeded in giving 2019 some unique puzzles, but I don't blame people for disliking the concept. It probably would have been a lot easier if I'd taken a compilers class, as many participants probably have. All the same, I think it's good to shake up a well-established formula. I'd like to see an experiment like Intcode again. # 2020 2020 is special. It was the first year I took part in while it was ongoing. I stumbled across a link to a puzzle near the end of my fall semester of graduate school, and I decided to try solving it. It was [day 12](https://adventofcode.com/2020/day/12), a fairly easy puzzle that asks you to simulate moving a navigation beacon according to instructions. I had fun solving it, though I wasted a lot of time trying to implement a general rotation matrix to model changes of orientation. (I had yet to learn that Advent of Code only requires integral computations, so I only had to deal with multiples of 90 degrees). I spent the rest of that winter break solving as many puzzles as I could, learning Python as I went. I had found Advent of Code at exactly the right time in my life. I had just switched from pursuing a master's in public policy to a master's in data science, and I knew I would have to learn to write effective code fast. Doing the puzzles increased my confidence that I could do that and confirmed to me that I would enjoy it. Sentimentality aside, this was a good year, one that returned to the classic format after the Intcode experiment of 2019. There were yet more Game of Life variations, this time [in four dimensions](https://adventofcode.com/2020/day/17) or [with hexagonal coordinates](https://adventofcode.com/2020/day/24). It was a fitting tribute to John Conway, who died that year. While the overall difficulty was reduced from the past two years, one puzzle broke the trend. (If you took part in this year, you already know which it is). [Day 20](https://adventofcode.com/2020/day/20), "Jurassic Jigsaw", requires you to assemble dozens of tiles into an image by matching the patterns on their edges. Part 2 asks you to scan the image for the [Loch Ness Monster](https://en.wikipedia.org/wiki/Loch_Ness_Monster), which can be recognized by a telltale pattern. It ranks among the very hardest puzzles, but I had fun designing a very complex depth-first search to solve it. The year's plot, which involves the player frantically trying to salvage a vacation plan gone awry, is the most realistic so far. Aside from the part with the Loch Ness Monster, anyway. # 2021 I began 2021 with far more experience than I had in 2020. And I needed that experience, because this year cranked the difficulty back up to near-2018 levels. I kept up for the first week or so, but I began to fall behind when the puzzles began requiring graph traversal. I knew nothing about graphs or pathfinding algorithms back then, and had to learn fast. I fumbled my way through [Dijkstra's algorithm](https://en.wikipedia.org/wiki/Dijkstra%27s_algorithm) well enough to solve some of the early puzzles, but I soon fell several days behind. I persevered, making steady progress as 2022 began. I had to put Advent of Code aside when grad school resumed, but by late spring I had earned all 50 stars. It was the first year I finished completely, and by far the most satisfying to complete. It took weeks to crawl through days 21 through 24, one of the toughest stretches of puzzles in the whole series. Most of these puzzles demand creativity to pare down huge problem spaces. Day 24, for instance, asks you to find the highest number that meets a condition from trillions of possibilities. I solved it only after painfully reverse-engineering the made-up assembly language provided as the input. Day 23, to me, is the hardest puzzle in the entire series. It asks you to solve a variation on the [Towers of Hanoi](https://en.wikipedia.org/wiki/Tower_of_Hanoi) puzzle. The secret is that it's pretty easy to solve by hand. (I used cut-up Post-it notes as game pieces). But I couldn't resist the challenge of writing a program to solve it, one that would work on a general input. It took a week of focused effort to do so. It was painful at time, but I grew so much as a programmer from these puzzles that it was all worth it. # 2022 The most recent year, as of this writing, has the player accompany a jungle expedition searching for star fruit, which the reindeer apparently eat. The more relaxed narrative suits the puzzles, which are a bit easier than the previous year's. The puzzle design seemed more thoughtful than ever; most of this year's puzzles had simple rules that lacked [accidental complexity](https://en.wikipedia.org/wiki/No_Silver_Bullet). Instead, the difficulty emerged organically from the constraints of the problems. [Day 16](https://adventofcode.com/2022/day/16), perhaps the hardest puzzle, is a good example. It asks you to open a series of pressure valves in rooms connected by tunnels in a way that releases as much pressure as possible. Representing the situation as a simple graph isn't enough; the number of possible states quickly becomes unmanageable. After asking for advice, I solved the problem by creating a simpler graph that represented distances between pairs of valves, then writing a traverse algorithm that discarded paths that had no chance of releasing the most pressure. This complex design process emerged from a simple problem statement, and I found it more satisfying than trying to meet a complicated specification. # Conclusion Advent of Code continues to grow more popular. In all likelihood, it will do so for many years to come. Looking back over the eight years and 200 puzzles so far, it's easy to see why. The puzzles offer endless variations on classic computer science problems, often with an inventive "part 2 twist." The story tying each year together, while inessential, keeps the mood light. The design of the website makes it easy to check answers, and the carefully worded puzzle descriptions provide enough guidance to find a solution. It success reminds me how many computer science discoveries came from just playing around with interesting problems. I'll probably keep participating as long as I can, because it never stops being fun.