Spoilers and explanation of solution:

Each vertex here is one intersection in our hike. We don't actually care about the parts in-between, because there's only one way to go. The above is a visualisation of the final path, the red edges are the edges taken. Our graph looks "like that" because it's a hiking trail, not a maze, so there's no dead ends. This took about 2 seconds to generate, due to all the cloning needed to keep track of paths. The two veeery long edges on the ends are pretty obvious choices, but one might notice that pretty much every vertex takes the two maximum paths it has, given the restrictions of the path. There's still some mildly surprising paths, such as (99, 29) -> (89, 37) with a weight of 38. I'm wondering if there's a way to dismiss more paths... This graph is actually pretty free in terms of movement.

My actual solution takes ~150 ms to run (and 8 microseconds for part one with barely any optimization, damnn)

Anybody got some ideas to optimize today? I've got it down to 65ms (total) on my desktop, using A* with a visitation map. Each cell in the visitation map contains (in part 2) 16 entries; 4 per direction of movement, 1 for each level of straightaway. In part 2, I use a map with 11 entries per direction.

Optimizations I've implemented:

• use a 2D array instead of a hashset/map. No idea how much this saves, I did it in the first place.
• the minimum distance for a specific cell's direction + combo applies for higher combo levels as well for part 1. For part 2, if the current combo is greater than 4, we do the same*. Gains about 70(!!) ms
• A* heuristic weighting optimization, a weight of about 1% with a manhattan distance heuristic seems to gain about 15 ms (might be my input only tho)

*Correctness-wise: the reason we're splitting by direction is because there's a difference between being at a cell going up with a 3 combo but a really short path, and going right with a 0 combo but a long path. However, this is fine because a 3 combo in the same direction as a 0 combo is identical, just more restrictive.

Optimizations that could be done but I need to ensure correctness:

the same optimization for the combo, but for directions. If I'm on a specific combo+direction, does that imply something about the distance for another direction? Simply doing the same for every non-opposite direction isn't correct

Code: https://codeberg.org/Sekoia/adventofcode/src/branch/main/src/y2023/day17.rs

Warning: quite ugly, there's like 8 copy-pastes for adding to the queue

Is there a way to measure performance without depending on the hardware, i.e. two entirely different computers get the same score for the same code?

I could probably run the program on a server or something, but something local feels more reliable.

My Intel NUC server just died (whenever it's plugged in, it makes a buzzing noise, and the external power LED is off (the internal one is on tho)), so I need a new server box. Any recommendations?

I can salvage the RAM (16 GB DDR4) and hard drive (1TB HDD) off of this one, I believe.

I have a few selfhosted services, but I'm slowly adding more. Currently, they're all in subdomains like linkding.sekoia.example etc. However, that adds DNS records to fetch and means more setup. Is there some reason I shouldn't put all my services under a single subdomain with paths (using a reverse proxy), like selfhosted.sekoia.example/linkding?

According to https://lemmy.blahaj.zone/post/72658 I shouldn't be able to post but if you can see this...

I just want to say that the admins here are great and deserve appreciation, especially during this whole kerfuffle with Reddit :)

Have a good one, mods and admins!