#![feature(test, iter_array_chunks)]
extern crate test;
use std::ops::Range;

use aoc2023::{boilerplate, common::*};
use itertools::Itertools;

const DAY: usize = 5;
type I = i64;
type Mapping = Vec<(Range<I>, Range<I>, I)>;
type Parsed = (Vec<I>, Vec<Mapping>);

fn parse_input(raw: &str) -> Parsed {
    let mut groups = raw.split("\n\n");
    let seeds = groups.next().unwrap().trim_start_matches("seeds: ").split(' ').map(parse_num).collect();
    let ranges = groups
        .map(|g| {
            g.lines()
                .skip(1)
                .map(|l| l.split(' ').map(parse_num).collect_tuple().unwrap())
                // We swap the order of dst and src here because it just makes much more sense to me this way.
                .map(|(dst, src, len)| (src..src + len, dst..dst + len, dst - src))
                .collect()
        })
        .collect();
    (seeds, ranges)
}

fn resolve(start: I, mappings: &Vec<Mapping>) -> I {
    mappings.iter().fold(start, |i, map| map.iter().find_map(|(range, _, offset)| range.contains(&i).then_some(i + offset)).unwrap_or(i))
}

fn part1((seeds, mappings): &Parsed) -> I {
    seeds.iter().map(|&s| resolve(s, mappings)).min().unwrap()
}

fn resolve_backwards(start: I, mappings: &Vec<Mapping>) -> I {
    mappings.iter().fold(start, |i, map| map.iter().find_map(|(_, range, offset)| range.contains(&i).then_some(i - offset)).unwrap_or(i))
}

fn part2((seeds, mappings): &Parsed) -> I {
    let seed_ranges = seeds.iter().array_chunks().map(|[&a, &b]| a..a + b).collect_vec();
    let mut mappings = mappings.clone();
    mappings.reverse();

    let mut destinations = mappings.remove(0);
    destinations.sort_by_key(|(_, range, _)| range.start);

    let mut ranges = Vec::<(Range<I>, I)>::new();
    for (_, range, offset) in destinations {
        ranges.push((ranges.last().map(|(r, _)| r.end).unwrap_or(0)..range.start, 0));
        ranges.push((range, offset));
    }

    // By calculating the smallest range in the input data, we can perform something similar to binary search.
    // We start at location 0 and move in steps of this size,
    // and when we first encounter a location that corresponds to a starting seed we start binary searching,
    // i.e. we half the step size and move back by that until we find the lowest location that has a starting seed.
    let mut step = mappings.iter().flatten().map(|(r, _, _)| r.try_len().unwrap()).min().unwrap() as I;
    ranges
        .into_iter()
        .find_map(|(range, offset)| {
            let mut best = I::MAX;
            let mut location = range.start + step;
            while location < range.end && step != 0 {
                if has_starting_seed(location, offset, &seed_ranges, &mappings) {
                    best = best.min(location);
                    step = (step >> 1) + 1;
                    location -= step;
                } else {
                    step >>= (best != I::MAX) as usize;
                    location += step;
                }
            }
            (best != I::MAX).then_some(best)
        })
        .unwrap()
}

fn has_starting_seed(start: I, offset: i64, seed_ranges: &Vec<Range<I>>, mappings: &Vec<Mapping>) -> bool {
    let seed = resolve_backwards(start - offset, &mappings);
    // If seed == s, the entire resolution didn’t hit a single mapping, so we don’t need to check seeds.
    seed != start && seed_ranges.iter().any(|r| r.contains(&seed))
}

boilerplate! {
    TEST_INPUT == "seeds: 79 14 55 13

seed-to-soil map:
50 98 2
52 50 48

soil-to-fertilizer map:
0 15 37
37 52 2
39 0 15

fertilizer-to-water map:
49 53 8
0 11 42
42 0 7
57 7 4

water-to-light map:
88 18 7
18 25 70

light-to-temperature map:
45 77 23
81 45 19
68 64 13

temperature-to-humidity map:
0 69 1
1 0 69

humidity-to-location map:
60 56 37
56 93 4",
    tests: {
        part1: { TEST_INPUT => 35 },
        part2: { TEST_INPUT => 46 },
    },
    bench1 == 462648396,
    bench2 == 2520479,
    bench_parse: |(v1, v2): &Parsed| (v1.len(), v2.len()) => (20, 7),
}