Fast Parse VS AdventOfCode

Perhaps use Li Haoyi's fastparse instead? https://github.com/com-lihaoyi/fastparse

You might want to adapt Li Haoyi’s XML parser for fastparse.

Mostly a mess of pattern matching. I really need to make some generic tree utilities. Haven't been able to find a decent parser combinator that works in Scala 3 (I usually use fastparse which depends heavily on Scala 2 macros, and scala-parser-combinators works in Scala 3, but I've had a lot of trouble getting it to not be too greedy), so I used the state monad from cats to parse at the bottom of the file, which I think turned out fairly nice.

Agreed! I would say that parser combinators are the sweet spot and the right choice in most cases.

Scala has them as well, e.g.: https://com-lihaoyi.github.io/fastparse/

And the good thing is, you don't have to learn a completely new language/syntax, you can use the host language's syntax and you have full IDE support as well.

Here's an example implementation in Rust, using rayon's parallel iterators to automatically distribute the computation of a range of hashes over all available cores. It solves both parts together in around 3 seconds on my M2 MacBook Pro. Single-threaded, it takes around 25s in Rust and 35s in Python.

Here are mine: https://github.com/benediktwerner/AdventOfCode

I love doing optimizations like this but I don't really have enough time anymore to really do it. In 2020 I spent a fair amount of time writing super optimized solutions for the first 10 or so days (repo solving all those days together in 150us i.e. <1ms) but even back then, I eventually stopped since I had other things to do and it took more and more time as the days went on.

Python 3 116/125

Wow, my final solution looks almost exactly the same. Though I calculated my initial answers by hand.

That's certainly not necessary though and really not what makes most of the speed, especially for problems like today's. I don't really look at other leaderboard competitor's solutions much but at least my solutions (ranked ~20 today) are almost always completely vanilla Python (the only exception is networkx for the occasional graph problem but even that isn't really that much of a speedup if you know the common graph algorithms) and always self-contained.

Interesting approach. I guess mine could be considered a bit nicer, doing it recursively and returning the number to add to the left or right.

My priority Q/heap Dijkstra in Python runs just fine. I don't remember and can't check rn if it was instant or took a second but it definitely wasn't longer than that.