> Run your code on the example input. > Seriously -- make extra super 100% sure your code works on the example input. Write some boilerplate code to make it ea

> Run your code on the example input. > Seriously -- make extra super 100% sure your code works on the example input. Write some boilerplate code to make it easy to switch between your input and the example input. > Test your code as you go. Printing the output of intermediate steps to the console is a great way of catching bugs. Honestly, just set up whatever you need to be able to write unit tests in your lang of choice. These problems are _so_ amenable to a piecewise approach driven by tests. I'm not like a big TDD advocate or anything, but these problems are great practice for that style of coding - it's just so damn useful to know each of your small pieces of code work. Parameterized tests are amazing for AoC, because you can get a handful of test cases basically for free from the puzzle description. If your code doesn't work once you've got all the samples working, you either have some weird edge case that you didn't consider, or you've got one of the brute-force killer puzzles. Even for today's, I wound up with 43 different test cases. The vast majority of those are from the puzzle text, and adding them didn't really make the puzzle take that much longer. (Obviously, if you're optimizing for solve speed, you probably wouldn't bother with this approach, but I'm not). https://github.com/epiccoleman/advent_of_code_ex/blob/master/test/aoc_2023/day01/day01_test.exs https://github.com/epiccoleman/advent_of_code_ex/blob/master... Another thing of note is that every puzzle basically operates on a list of strings, so it's pretty easy to genericize certain parts of the work of solving puzzles. I have a script which generates a module for the solution in my repo, with separate functions for each part that receive the input, and a test file that has tests for part 1 and part 2. The tests read the input file and pass it as a list of strings (lines) to the part_1 and part_2 functions, so that all the boilerplate is already done, and I get to just focus on writing the guts of the part_1 and part_2 functions (which usually get broken down into several other functions, which can also be tested individually).