use icu::{
    collator::{Collator, CollatorOptions, Strength},
    segmenter::GraphemeClusterSegmenter,
};
use itertools::Itertools;
use std::collections::HashSet;

pub fn anagrams_for<'a>(word: &str, possible_anagrams: &'a [&str]) -> HashSet<&'a str> {
    let mut options = CollatorOptions::new();
    // Ignore case sensitivity, accents, etc.
    options.strength = Some(Strength::Primary);
    // Strictly speaking an anagram only makes sense within a given locale,
    // but we don't know it, so leave it as undefined.
    let collator = Collator::try_new(Default::default(), options).unwrap();
    let word_anagram = first_anagram(&word, &collator);
    possible_anagrams
        .iter()
        .copied()
        .filter(|possibility| {
            let possibility_anagram = first_anagram(&possibility, &collator);
            let is_anagram = collator.compare(&word_anagram, &possibility_anagram).is_eq();
            let is_same = collator.compare(&possibility, &word).is_eq();
            is_anagram && !is_same
        })
        .collect()
}

// We define the first anagram of a string as the concatenation of
// sorting all of its grapheme clusters. To do this fully correctly,
// we would want to know what language we should use when considering
// whether two grapheme clusters are "equal".
fn first_anagram(word: &str, collator: &Collator) -> String {
    let segmenter = GraphemeClusterSegmenter::new();
    let mut graphemes: Vec<&str> = segmenter
        .segment_str(&word)
        .tuple_windows()
        .map(|(i, j)| &word[i..j])
        .collect();
    graphemes.sort_by(|a, b| collator.compare(a, b));
    graphemes.into_iter().collect()
}