Solutions to Advent of Code 2024 puzzles in SWI Prolog. My aim this year is to try and make use of Prolog's unique features more and avoid writing functional code which is tempting when working with immutable data structures.
-
I tried in earnest to use Prolog's distinguishing features as much as possible, some times with great success (e.g. Q17).
-
The code is mostly short, but also dense and laden with 2 letter variables.
-
Here are the questions in which I used CLP(FD/B) to significant advantage: 7.1, 8, 13, 17, 24.1.
-
Here are the questions in which I used SLG resolution (aka tabling) to significant advantage: 10, 11, 19.
-
The code I've committed for 21.1 is wrong: I made the mistake comitting a refactor.
-
My plan for 24 was to use CLP(B) to model both the given circuit and a working binary adder, and then for each incorrect output, test the dependent endpoints exhaustively until a fix is found. However, implementing this with CLP(B) turned out to be difficult and I eventually gave up on it in favour of a Python Z3 version.
I did AoC 2023 in Prolog too, and I feel this code is a significant improvement over previous attempts. I think that Prolog makes these puzzles harder to solve overall. Prolog is a kind of thought scaffolding which requires a significant amount of sympathy for the interpreter. For example, in question 24, working out what CLP(B) was doing during backtracking and what constraints were active at different parts of the code got quite exhausting. Other than what felt like unrewarded cognitive load, speed of the code was a constant issue, and the unavoidable verbosity that comes with having to chain predicates together using intermediary variables, when processing lists.
I think perhaps the biggest benefit of using Prolog for these puzzles is as an aide to tacitly understanding a different way of expressing computing: Prolog clauses are logical but always have a procedural interpretation which is "executed" through SLD/SLG resolution.