🎄 Advent Of Code 2023

Hi! These are my Rust solutions for the Advent of Code 2023.

Day	Name	Source	Part 1	Part 2	Time 1	Time 2
1	Trebuchet?!	01.rs	⭐	⭐	33.1 µs	42.1 µs
2	Cube Conundrum	02.rs	⭐	⭐	26.5 µs	37.2 µs
3	Gear Ratios	03.rs	⭐	⭐	39.1 µs	30.7 µs
4	Scratchcards	04.rs	⭐	⭐	82.4 µs	84.0 µs
5	If You Give A Seed A Fertilizer	05.rs	⭐	⭐	18.7 µs	23.5 µs
6	Wait For It	06.rs	⭐	⭐	174 ns	180 ns
7	Camel Cards	07.rs	⭐	⭐	150 µs	157 µs
8	Haunted Wasteland	08.rs	⭐	⭐	193 µs	1.2 ms
9	Mirage Maintenance	09.rs	⭐	⭐	99.5 µs	96.5 µs
10	Pipe Maze	10.rs	⭐	⭐	87.6 µs¹	138 µs¹
11	Cosmic Expansion	11.rs	⭐	⭐	3.1 ms	3.1 ms
12	Hot Springs	12.rs	⭐	⭐	1.7 ms	21.3 ms
13	Point of Incidence	13.rs	⭐	⭐	46.0 µs	48.5 µs
14	Parabolic Reflector Dish	14.rs	⭐	⭐	48.8 µs	25.6 ms
15	Lens Library	15.rs	⭐	⭐	54.8 µs	266 µs
16	The Floor Will Be Lava	16.rs	⭐	⭐	664 µs	162 ms
17	Clumsy Crucible	17.rs	⭐	⭐	18.1 ms	58.3 ms
18	Lavaduct Lagoon	18.rs	⭐	⭐	19.2 µs	26.1 µs
19	Aplenty	19.rs	⭐	⭐	229 µs	195 µs
	...

Key: ⭐ Completed    🎁 In progress    😔 Gave up

Benchmarked on Intel i7-11800H @ 2.30 GHz (over many samples).

_{¹ I noticed after some refactoring that benchmark times got ~2x slower, with the simpler part taking significantly longer. A ~2x speed increase (relative to initial benchmark) was obtained by only testing one part at a time (with the other commented out for dead-code removal)! This may be an extreme sensitivity to the layout of the linked binary and how this is loaded into the instruction cache? Or bad branch prediction?}

Verdict

Doing this year in Rust was fun. As expected, the result is fast and efficient. Seeing solutions complete in microseconds, or even nanoseconds, was satisfying to say the least. I was surprised by how much I enjoyed using iterators, maybe a little too much. However, it's easy to get lost in premature optimization, static dispatch and avoiding heap allocation. Knowing your data structures, and choosing the right one for the job, makes a much more significant difference than the language itself. Unit tests were brilliant for testing solutions, before running them on the input data. The language is great for expressing the problem domain, although I do miss the conciseness of solutions in other languages (such as Python) that are often a fraction of the number of lines. The time saved writing such a solution is a valid trade off, despite taking orders of magnitude longer to run (i.e. a few seconds...).

Usage

Install Rust.

Run the following commands to run the project:

# Test all solutions on example data
cargo test

# Run all solutions with optimisations
cargo all --release

# Benchmark solutions for day 1
cargo solve 1 --release --time

The Rust compiler will automatically download the required dependencies and compile each solution into its own binary that can be found in the target/debug or target/release directory, depending on whether the --release flag was used.

To read the problem input from stdin, instead of a hardcoded file path, make a small modification to the solution macro in day.rs that steps up the main function of each binary:

- let input = advent_of_code::template::read_input(DAY);
+ let input = advent_of_code::template::read_stdin();

Acknowledgments

This repository uses a modified version of this template. Thanks Felix!

License

Distributed under the MIT Licence. See LICENCE for more information.