decision/crates/ld-engine/src/edit.rs

//! Edit classification — a typo fix is not a decision.
//!
//! Once a proposal is under vote, its author may still correct it. The engine
//! tells two kinds of corrections apart:
//!
//! - **typographic**: accents, case, punctuation, spacing, emphasis, or at most
//!   two misspelt words fixed (edit distance ≤ 2, no negation touched) — and
//!   never a figure, a sign, a separator inside a number or a relation
//!   (`≥`, `+`, `%`…). The votes stay; the correction is shown with its diff.
//! - **substantive**: anything else. It opens a new revision and every vote
//!   cast on the previous one waits for its author to reconfirm it.
//!
//! The rule is deliberately conservative: when in doubt, it is substantive.

use serde::{Deserialize, Serialize};
use unicode_normalization::UnicodeNormalization;

#[derive(Clone, Copy, Debug, PartialEq, Eq, Serialize, Deserialize)]
#[serde(rename_all = "lowercase")]
pub enum EditKind {
    Unchanged,
    Typographic,
    Substantive,
}

/// Words whose presence flips a meaning; touching one is never a typo.
const NEGATIONS: &[&str] = &[
    // fr
    "ne", "n", "pas", "jamais", "non", "sans", "aucun", "aucune", "ni", "plus", "rien", "nul", "nulle", "personne",
    // en
    "not", "no", "never", "without", "none", "nor", // es
    "nunca", "sin", "ningun", "ninguna", "ninguno", "tampoco", "nada",
];

const MAX_TYPO_WORDS: usize = 2;
const MAX_TYPO_DISTANCE: usize = 2;

/// Lowercase, strip diacritics.
fn fold(s: &str) -> String {
    s.nfd().filter(|c| !is_combining(*c)).flat_map(char::to_lowercase).collect()
}

fn is_combining(c: char) -> bool {
    matches!(c as u32, 0x0300..=0x036F | 0x1AB0..=0x1AFF | 0x1DC0..=0x1DFF | 0x20D0..=0x20FF | 0xFE20..=0xFE2F)
}

/// Letters and digits only, folded: what survives every typographic change.
fn skeleton(s: &str) -> String {
    fold(s).chars().filter(|c| c.is_alphanumeric()).collect()
}

/// Symbols that carry a value or a relation, wherever they stand.
const RELATIONS: &[char] =
    &['+', '±', '×', '÷', '^', '<', '>', '=', '≤', '≥', '≠', '≈', '%', '‰', '€', '$', '£', '∞', '√'];
/// Signs and fraction bars: meaningful next to a figure ("-5", "1/5", "5-10").
const SIGNS: &[char] = &['-', '−', '–', '/'];
/// Separators: meaningful between two figures ("10 000", "10,000", "182.625").
const SEPARATORS: &[char] = &['.', ',', ' ', '\u{a0}', '\u{202f}', '\'', '’'];

/// What the text says in figures and symbols, as written. Two texts with the
/// same letters but a different figure, sign, separator inside a number or
/// relation ("≥" → "≤", "+" → "-", "50" → "5,0", "182.625" → "182 625") do not
/// say the same thing.
fn figures(s: &str) -> String {
    let cs: Vec<char> = s.chars().collect();
    let is_figure = |i: Option<usize>| i.and_then(|i| cs.get(i)).is_some_and(|c| c.is_numeric());
    cs.iter()
        .enumerate()
        .filter(|&(i, c)| {
            let (before, after) = (is_figure(i.checked_sub(1)), is_figure(Some(i + 1)));
            c.is_numeric()
                || RELATIONS.contains(c)
                || (SIGNS.contains(c) && (before || after))
                || (SEPARATORS.contains(c) && before && after)
        })
        .map(|(_, c)| *c)
        .collect()
}

/// Folded words (runs of letters/digits).
fn words(s: &str) -> Vec<String> {
    fold(s).split(|c: char| !c.is_alphanumeric()).filter(|w| !w.is_empty()).map(str::to_owned).collect()
}

pub fn classify(old: &str, new: &str) -> EditKind {
    if old == new {
        return EditKind::Unchanged;
    }
    // A figure, a sign or a relation is never a typo: it changes what the text says.
    if figures(old) != figures(new) {
        return EditKind::Substantive;
    }
    if skeleton(old) == skeleton(new) {
        return EditKind::Typographic;
    }
    let (a, b) = (words(old), words(new));
    if a.len() != b.len() {
        return EditKind::Substantive;
    }
    let mut changed = 0;
    for (x, y) in a.iter().zip(&b) {
        if x == y {
            continue;
        }
        changed += 1;
        let touches_meaning = changed > MAX_TYPO_WORDS
            || x.chars().any(|c| c.is_ascii_digit())
            || y.chars().any(|c| c.is_ascii_digit())
            || NEGATIONS.contains(&x.as_str())
            || NEGATIONS.contains(&y.as_str())
            || x.chars().count().min(y.chars().count()) < 3
            || levenshtein(x, y) > MAX_TYPO_DISTANCE;
        if touches_meaning {
            return EditKind::Substantive;
        }
    }
    EditKind::Typographic
}

pub fn levenshtein(a: &str, b: &str) -> usize {
    let b: Vec<char> = b.chars().collect();
    let mut prev: Vec<usize> = (0..=b.len()).collect();
    let mut cur = vec![0; b.len() + 1];
    for (i, ca) in a.chars().enumerate() {
        cur[0] = i + 1;
        for (j, cb) in b.iter().enumerate() {
            let cost = usize::from(ca != *cb);
            cur[j + 1] = (prev[j] + cost).min(prev[j + 1] + 1).min(cur[j] + 1);
        }
        std::mem::swap(&mut prev, &mut cur);
    }
    prev[b.len()]
}

#[derive(Clone, Copy, Debug, PartialEq, Eq, Serialize, Deserialize)]
#[serde(rename_all = "lowercase")]
pub enum DiffOp {
    Same,
    Added,
    Removed,
}

#[derive(Clone, Debug, PartialEq, Eq, Serialize, Deserialize)]
pub struct DiffPart {
    pub op: DiffOp,
    pub text: String,
}

/// Tokens that keep every character: words, and the runs between them.
fn tokens(s: &str) -> Vec<&str> {
    let mut out = Vec::new();
    let mut start = 0;
    let mut in_word: Option<bool> = None;
    for (i, c) in s.char_indices() {
        let w = c.is_alphanumeric();
        if in_word.is_some_and(|prev| prev != w) {
            out.push(&s[start..i]);
            start = i;
        }
        in_word = Some(w);
    }
    if start < s.len() {
        out.push(&s[start..]);
    }
    out
}

const MAX_DIFF_TOKENS: usize = 4000;

/// Word-level diff for display. Concatenating `Same` + `Added` gives `new`,
/// `Same` + `Removed` gives `old`.
pub fn diff(old: &str, new: &str) -> Vec<DiffPart> {
    let (a, b) = (tokens(old), tokens(new));
    if a.len() * b.len() > MAX_DIFF_TOKENS * MAX_DIFF_TOKENS / 4 {
        return vec![
            DiffPart { op: DiffOp::Removed, text: old.to_owned() },
            DiffPart { op: DiffOp::Added, text: new.to_owned() },
        ];
    }
    // LCS table, suffix form.
    let (n, m) = (a.len(), b.len());
    let mut lcs = vec![vec![0u32; m + 1]; n + 1];
    for i in (0..n).rev() {
        for j in (0..m).rev() {
            lcs[i][j] = if a[i] == b[j] { lcs[i + 1][j + 1] + 1 } else { lcs[i + 1][j].max(lcs[i][j + 1]) };
        }
    }
    let mut parts: Vec<DiffPart> = Vec::new();
    let mut push = |op: DiffOp, text: &str| match parts.last_mut() {
        Some(last) if last.op == op => last.text.push_str(text),
        _ => parts.push(DiffPart { op, text: text.to_owned() }),
    };
    let (mut i, mut j) = (0, 0);
    while i < n && j < m {
        if a[i] == b[j] {
            push(DiffOp::Same, a[i]);
            i += 1;
            j += 1;
        } else if lcs[i + 1][j] >= lcs[i][j + 1] {
            push(DiffOp::Removed, a[i]);
            i += 1;
        } else {
            push(DiffOp::Added, b[j]);
            j += 1;
        }
    }
    for t in &a[i..] {
        push(DiffOp::Removed, t);
    }
    for t in &b[j..] {
        push(DiffOp::Added, t);
    }
    parts
}

#[cfg(test)]
mod tests {
    use super::*;

    #[test]
    fn the_glued_word_of_the_seed_is_a_typo() {
        let old = "Toute Certification doit s'accompagner de latransmission de ce document";
        let new = "Toute Certification doit s'accompagner de la transmission de ce document";
        assert_eq!(classify(old, new), EditKind::Typographic);
    }

    #[test]
    fn accents_case_punctuation_emphasis() {
        assert_eq!(classify("Je m'engage a verifier.", "Je m’engage à vérifier !"), EditKind::Typographic);
        assert_eq!(classify("un **être humain**", "un être humain"), EditKind::Typographic);
        assert_eq!(classify("same", "same"), EditKind::Unchanged);
    }

    #[test]
    fn a_figure_a_sign_or_a_relation_is_never_a_typo() {
        // Same letters, different meaning (independent reviews, 2026-10-21).
        for (old, new) in [
            ("au moins 50 voix", "au moins 5,0 voix"),
            ("1 5 jours", "15 jours"),
            ("10 000 membres", "10,000 membres"),
            ("un écart de -5", "un écart de 5"),
            ("1/5 des voix", "1.5 des voix"),
            ("50 % des voix", "50 ‰ des voix"),
            ("avoir >= 5 certifications", "avoir <= 5 certifications"),
            ("DU(n) + c² × M(n)", "DU(n) - c² × M(n)"),
            ("182.625 jours", "182 625 jours"),
            ("c²", "c2"),
        ] {
            assert_eq!(classify(old, new), EditKind::Substantive, "{old} → {new}");
        }
        // Punctuation around figures, list markers, dashes between words stay typographic.
        assert_eq!(classify("50 voix pour", "50 voix pour !"), EditKind::Typographic);
        assert_eq!(classify("- une liste", "* une liste"), EditKind::Typographic);
        assert_eq!(classify("1. premier point", "1) premier point"), EditKind::Typographic);
        assert_eq!(classify("libre - le texte", "libre — le texte"), EditKind::Typographic);
    }

    #[test]
    fn misspelt_words() {
        assert_eq!(classify("un certifieurs existant", "un certifieur existant"), EditKind::Typographic);
        assert_eq!(classify("toute certifcation", "toute certification"), EditKind::Typographic);
        // « personne » is also a negation (ne … personne): conservative.
        assert_eq!(classify("la persone", "la personne"), EditKind::Substantive);
    }

    #[test]
    fn meaning_changes() {
        // A number.
        assert_eq!(classify("tous les 5 jours", "tous les 6 jours"), EditKind::Substantive);
        // A negation, even one letter apart.
        assert_eq!(classify("je peux certifier", "je ne peux certifier"), EditKind::Substantive);
        assert_eq!(classify("plus de deux", "pas de deux"), EditKind::Substantive);
        // A different word.
        assert_eq!(classify("au moins deux canaux", "au moins trois canaux"), EditKind::Substantive);
        // Three fixes are no longer a typo.
        assert_eq!(classify("aaaa bbbb cccc", "aaab bbbc cccd"), EditKind::Substantive);
        // A word added.
        assert_eq!(classify("en visio", "en visio directe"), EditKind::Substantive);
    }

    #[test]
    fn diff_reconstructs_both_sides() {
        let old = "J'ai rencontré la personne physiquement (préférable).";
        let new = "J'ai rencontré la personne en visio (préférable).";
        let parts = diff(old, new);
        let rebuild = |keep: DiffOp| -> String {
            parts.iter().filter(|p| p.op == DiffOp::Same || p.op == keep).map(|p| p.text.as_str()).collect()
        };
        assert_eq!(rebuild(DiffOp::Removed), old);
        assert_eq!(rebuild(DiffOp::Added), new);
        assert!(parts.iter().any(|p| p.op == DiffOp::Removed && p.text.contains("physiquement")));
    }

    #[test]
    fn levenshtein_basics() {
        assert_eq!(levenshtein("kitten", "sitting"), 3);
        assert_eq!(levenshtein("", "abc"), 3);
        assert_eq!(levenshtein("été", "ete"), 2);
    }
}