shuire 0.2.0

#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub enum FileStatus {
    Added,
    Modified,
    Deleted,
    Renamed,
}

impl FileStatus {
    pub fn badge(&self) -> &'static str {
        match self {
            FileStatus::Added => "A",
            FileStatus::Modified => "M",
            FileStatus::Deleted => "D",
            FileStatus::Renamed => "R",
        }
    }
}

#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub enum LineKind {
    Context,
    Added,
    Removed,
    HunkHeader,
    FoldDown,
    FoldUp,
}

impl LineKind {
    pub fn is_code(&self) -> bool {
        matches!(
            self,
            LineKind::Context | LineKind::Added | LineKind::Removed
        )
    }

    pub fn is_navigable(&self) -> bool {
        matches!(
            self,
            LineKind::Context
                | LineKind::Added
                | LineKind::Removed
                | LineKind::FoldDown
                | LineKind::FoldUp
        )
    }

    pub fn is_fold(&self) -> bool {
        matches!(self, LineKind::FoldDown | LineKind::FoldUp)
    }

    pub fn is_hunk_boundary(&self) -> bool {
        matches!(
            self,
            LineKind::HunkHeader | LineKind::FoldDown | LineKind::FoldUp
        )
    }

    /// Rendered as a single row without body wrapping — fold markers and
    /// hunk headers are clipped/padded rather than wrapped.
    pub fn is_fixed_height(&self) -> bool {
        self.is_fold() || matches!(self, LineKind::HunkHeader)
    }
}

#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub enum TokenKind {
    Default,
    Keyword,
    String,
    Comment,
    Number,
    Constant,
    Type,
    Function,
    Variable,
    Operator,
    Punctuation,
}

#[derive(Debug, Clone)]
pub struct DiffLine {
    pub kind: LineKind,
    pub text: String,
    pub segments: Vec<(TokenKind, String)>,
    pub gutter: String,
    pub old_lineno: Option<u32>,
    pub new_lineno: Option<u32>,
    pub fold_count: u32,
    pub hunk_header: String,
    pub word_highlights: Vec<(usize, usize)>, // (start, end) byte offsets of changed ranges
    /// True when this Added/Removed line was matched to a counterpart on the
    /// opposite side within the same block (i.e. word-level diff applied).
    /// Unpaired lines represent pure deletes/inserts with no corresponding
    /// edit, and are rendered with a stronger background to distinguish them.
    pub paired: bool,
    /// Set from git's trailing `\ No newline at end of file` sentinel.
    pub no_newline: bool,
}

impl DiffLine {
    fn hunk(text: String) -> Self {
        Self {
            kind: LineKind::HunkHeader,
            text,
            segments: Vec::new(),
            gutter: format_gutter(None, None),
            old_lineno: None,
            new_lineno: None,
            fold_count: 0,
            hunk_header: String::new(),
            word_highlights: Vec::new(),
            paired: false,
            no_newline: false,
        }
    }

    pub fn body(
        kind: LineKind,
        text: String,
        old_lineno: Option<u32>,
        new_lineno: Option<u32>,
    ) -> Self {
        Self {
            kind,
            text,
            segments: Vec::new(),
            gutter: format_gutter(old_lineno, new_lineno),
            old_lineno,
            new_lineno,
            fold_count: 0,
            hunk_header: String::new(),
            word_highlights: Vec::new(),
            paired: false,
            no_newline: false,
        }
    }

    pub fn fold_down(old_start: u32, new_start: u32, count: u32, hunk_header: String) -> Self {
        Self {
            kind: LineKind::FoldDown,
            text: format!("▼ {} lines", count),
            segments: Vec::new(),
            gutter: format_gutter(None, None),
            old_lineno: Some(old_start),
            new_lineno: Some(new_start),
            fold_count: count,
            hunk_header: hunk_header.clone(),
            word_highlights: Vec::new(),
            paired: false,
            no_newline: false,
        }
    }

    pub fn fold_up(old_start: u32, new_start: u32, count: u32, hunk_header: String) -> Self {
        Self {
            kind: LineKind::FoldUp,
            text: format!("▲ {} lines", count),
            segments: Vec::new(),
            gutter: format_gutter(None, None),
            old_lineno: Some(old_start),
            new_lineno: Some(new_start),
            fold_count: count,
            hunk_header,
            word_highlights: Vec::new(),
            paired: false,
            no_newline: false,
        }
    }
}

fn format_gutter(old: Option<u32>, new: Option<u32>) -> String {
    let fmt = |n: Option<u32>| match n {
        Some(v) => format!("{:>4}", v),
        None => "    ".to_string(),
    };
    format!("{} {} ", fmt(old), fmt(new))
}

#[derive(Debug, Clone)]
pub struct FileDiff {
    pub path: String,
    pub old_path: Option<String>,
    pub status: FileStatus,
    pub added: usize,
    pub removed: usize,
    pub lines: Vec<DiffLine>,
    pub auto_collapsed: bool,
    pub viewed: bool,
}

pub fn parse_unified(input: &str) -> Vec<FileDiff> {
    let mut files: Vec<FileDiff> = Vec::new();
    let mut cur: Option<FileDiff> = None;
    let mut pending_old: Option<String> = None;
    let mut old_ln: u32 = 0;
    let mut new_ln: u32 = 0;

    for line in input.lines() {
        if let Some(rest) = line.strip_prefix("diff --git ") {
            if let Some(f) = cur.take() {
                files.push(f);
            }
            let (a, b) = parse_diff_git_header(rest);
            let path = b.clone().or_else(|| a.clone()).unwrap_or_default();
            cur = Some(FileDiff {
                path: strip_prefix_ab(&path),
                old_path: a.map(|p| strip_prefix_ab(&p)),
                status: FileStatus::Modified,
                added: 0,
                removed: 0,
                lines: Vec::new(),
                auto_collapsed: false,
                viewed: false,
            });
            pending_old = None;
            old_ln = 0;
            new_ln = 0;
            continue;
        }

        let Some(file) = cur.as_mut() else { continue };

        if line.starts_with("new file") {
            file.status = FileStatus::Added;
            continue;
        }
        if line.starts_with("deleted file") {
            file.status = FileStatus::Deleted;
            continue;
        }
        if line.starts_with("rename ") || line.starts_with("similarity index") {
            if let Some(old_path) = line.strip_prefix("rename from ") {
                pending_old = Some(old_path.to_string());
            }
            if let Some(new_path) = line.strip_prefix("rename to ") {
                file.status = FileStatus::Renamed;
                if let Some(old) = pending_old.take() {
                    file.old_path = Some(old);
                }
                file.path = new_path.to_string();
            }
            continue;
        }

        // These are authoritative (the `diff --git` header's ` b/` split can
        // misfire on pathological paths) and also drive /dev/null → None for
        // added/deleted files.
        if let Some(rest) = line.strip_prefix("--- ") {
            file.old_path = parse_diff_path_header(rest);
            continue;
        }
        if let Some(rest) = line.strip_prefix("+++ ") {
            if let Some(path) = parse_diff_path_header(rest) {
                file.path = path;
            }
            continue;
        }

        if line.starts_with("index ")
            || line.starts_with("Binary files")
            || line.starts_with("old mode")
            || line.starts_with("new mode")
        {
            continue;
        }

        // `\ No newline at end of file` — attach to the preceding body line.
        if line.starts_with('\\') {
            if let Some(last) = file.lines.last_mut() {
                if last.kind.is_code() {
                    last.no_newline = true;
                }
            }
            continue;
        }

        if line.starts_with("@@") {
            let (old_start, new_start) = match parse_hunk_header(line) {
                Some((o, n)) => (o, n),
                None => {
                    old_ln = 0;
                    new_ln = 0;
                    file.lines.push(DiffLine::hunk(line.to_string()));
                    continue;
                }
            };

            let fold_new_start = if new_ln == 0 { 1 } else { new_ln };
            let fold_old_start = if old_ln == 0 { 1 } else { old_ln };
            let fold_count = new_start.saturating_sub(fold_new_start);
            old_ln = old_start;
            new_ln = new_start;

            if fold_count > 0 {
                let header = line.to_string();
                if fold_count > 20 {
                    file.lines.push(DiffLine::fold_down(
                        fold_old_start,
                        fold_new_start,
                        fold_count,
                        String::new(),
                    ));
                    file.lines.push(DiffLine::fold_up(
                        fold_old_start,
                        fold_new_start,
                        fold_count,
                        header,
                    ));
                } else {
                    file.lines.push(DiffLine::fold_down(
                        fold_old_start,
                        fold_new_start,
                        fold_count,
                        header,
                    ));
                }
            } else {
                file.lines.push(DiffLine::hunk(line.to_string()));
            }
            continue;
        }

        if let Some(rest) = line.strip_prefix('+') {
            file.added += 1;
            file.lines.push(DiffLine::body(
                LineKind::Added,
                rest.to_string(),
                None,
                Some(new_ln),
            ));
            new_ln += 1;
        } else if let Some(rest) = line.strip_prefix('-') {
            file.removed += 1;
            file.lines.push(DiffLine::body(
                LineKind::Removed,
                rest.to_string(),
                Some(old_ln),
                None,
            ));
            old_ln += 1;
        } else if let Some(rest) = line.strip_prefix(' ') {
            file.lines.push(DiffLine::body(
                LineKind::Context,
                rest.to_string(),
                Some(old_ln),
                Some(new_ln),
            ));
            old_ln += 1;
            new_ln += 1;
        }
    }

    if let Some(f) = cur.take() {
        files.push(f);
    }

    for f in &mut files {
        apply_auto_collapse(f);
        compute_word_diff_highlights(f);
    }

    files
}

/// Pair Removed/Added lines within each changed block by best similarity,
/// then compute word-level highlights for the chosen pairs. Lines that do
/// not clear the similarity threshold against any counterpart stay unpaired
/// and render with the stronger "no-pair" background.
///
/// Mirrors delta's approach: rather than assuming positional correspondence,
/// rank all candidate pairings by how similar the two lines are and greedily
/// assign starting from the best match. See delta `src/edits.rs::infer_edits`.
fn compute_word_diff_highlights(file: &mut FileDiff) {
    use crate::word_diff::score_pair;

    let mut i = 0;
    while i < file.lines.len() {
        if file.lines[i].kind != LineKind::Removed {
            i += 1;
            continue;
        }
        let removed_start = i;
        while i < file.lines.len() && file.lines[i].kind == LineKind::Removed {
            i += 1;
        }
        let removed_end = i;
        let added_start = i;
        while i < file.lines.len() && file.lines[i].kind == LineKind::Added {
            i += 1;
        }
        let added_end = i;

        let removed_count = removed_end - removed_start;
        let added_count = added_end - added_start;

        // Cap the all-pairs search so huge blocks don't balloon into
        // O(R*A*n*m) LCS work. Above the cap, skip word-diff entirely and
        // leave the lines unpaired — the stronger "no-pair" background still
        // conveys they are pure deletes/inserts.
        const MAX_BLOCK_PAIRS: usize = 400;
        if removed_count * added_count > MAX_BLOCK_PAIRS {
            continue;
        }

        struct Candidate {
            score: f32,
            r: usize,
            a: usize,
            rhl: Vec<(usize, usize)>,
            ahl: Vec<(usize, usize)>,
        }
        let mut candidates: Vec<Candidate> = Vec::new();
        for r in removed_start..removed_end {
            for a in added_start..added_end {
                let Some((score, rh, ah)) = score_pair(&file.lines[r].text, &file.lines[a].text)
                else {
                    continue;
                };
                candidates.push(Candidate {
                    score,
                    r,
                    a,
                    rhl: rh.into_iter().map(|h| (h.start, h.end)).collect(),
                    ahl: ah.into_iter().map(|h| (h.start, h.end)).collect(),
                });
            }
        }
        candidates.sort_by(|x, y| {
            y.score
                .partial_cmp(&x.score)
                .unwrap_or(std::cmp::Ordering::Equal)
        });

        let mut r_used = vec![false; removed_count];
        let mut a_used = vec![false; added_count];
        for c in candidates {
            let ri = c.r - removed_start;
            let ai = c.a - added_start;
            if r_used[ri] || a_used[ai] {
                continue;
            }
            r_used[ri] = true;
            a_used[ai] = true;
            file.lines[c.r].paired = true;
            file.lines[c.r].word_highlights = c.rhl;
            file.lines[c.a].paired = true;
            file.lines[c.a].word_highlights = c.ahl;
        }
    }
}

fn parse_hunk_header(line: &str) -> Option<(u32, u32)> {
    let rest = line.strip_prefix("@@")?.trim_start();
    let mut parts = rest.split_whitespace();
    let old = parts.next()?.strip_prefix('-')?;
    let new = parts.next()?.strip_prefix('+')?;
    let old_start: u32 = old.split(',').next()?.parse().ok()?;
    let new_start: u32 = new.split(',').next()?.parse().ok()?;
    Some((old_start, new_start))
}

pub fn apply_auto_collapse(f: &mut FileDiff) {
    // Seed `viewed` as the initial state so lock/generated files start dim;
    // keeping `auto_collapsed` as a separate field lets the user un-view them
    // with `v` and preserves the origin (auto vs user) for future features.
    f.auto_collapsed = should_auto_collapse(&f.path) || has_generated_marker(&f.lines);
    if f.auto_collapsed {
        f.viewed = true;
    }
}

fn parse_diff_git_header(rest: &str) -> (Option<String>, Option<String>) {
    // Quoted form (git's default core.quotepath=true):
    //   `diff --git "a/<c-escaped>" "b/<c-escaped>"` — split on the ` "b/`
    //   delimiter between the two halves, then unquote each side.
    if let Some(a_body) = rest.strip_prefix("\"a/")
        && let Some(idx) = a_body.find("\" \"b/")
        && let Some(b_body) = a_body[idx + "\" \"b/".len()..].strip_suffix('"')
    {
        return (
            Some(unquote_c_path(&a_body[..idx])),
            Some(unquote_c_path(b_body)),
        );
    }

    // PATH may contain spaces (git uses `core.quotepath=false`), so split on
    // the ` b/` marker between the two halves rather than whitespace. The
    // result is approximate for pathological inputs — `--- a/...` / `+++ b/...`
    // overwrite it with the authoritative path.
    if let Some(a) = rest.strip_prefix("a/") {
        if let Some(idx) = a.find(" b/") {
            return (
                Some(a[..idx].to_string()),
                Some(a[idx + " b/".len()..].to_string()),
            );
        }
    }
    let mut parts = rest.split_whitespace();
    match parts.next() {
        Some(first) => {
            let last = parts.last().unwrap_or(first);
            (Some(strip_prefix_ab(first)), Some(strip_prefix_ab(last)))
        }
        None => (None, None),
    }
}

fn parse_diff_path_header(rest: &str) -> Option<String> {
    let raw = rest.split('\t').next().unwrap_or(rest).trim_end();
    if raw.is_empty() || raw == "/dev/null" {
        return None;
    }
    // Quoted form (core.quotepath=true): `"a/<c-escaped>"`.
    let path = if let Some(body) = raw.strip_prefix('"').and_then(|s| s.strip_suffix('"')) {
        unquote_c_path(body)
    } else {
        raw.to_string()
    };
    Some(strip_prefix_ab(&path))
}

/// Decode a C-style quoted git path: `\XXX` octal triples are collected into
/// a byte buffer and re-parsed as UTF-8; `\t`, `\n`, `\r`, `\"`, `\\` expand
/// to their literal counterparts; everything else is passed through.
///
/// Git emits this form under the default `core.quotepath=true` for any path
/// containing non-ASCII bytes; shuire's own git invocation disables that, but
/// diffs arriving via `--from-file` / stdin can still carry it.
fn unquote_c_path(s: &str) -> String {
    let bytes = s.as_bytes();
    let mut out: Vec<u8> = Vec::with_capacity(bytes.len());
    let mut i = 0;
    while i < bytes.len() {
        if bytes[i] == b'\\' && i + 1 < bytes.len() {
            let rest = &bytes[i + 1..];
            if rest.len() >= 3 && rest[..3].iter().all(|b| (b'0'..=b'7').contains(b)) {
                let v = ((rest[0] - b'0') << 6) | ((rest[1] - b'0') << 3) | (rest[2] - b'0');
                out.push(v);
                i += 4;
                continue;
            }
            let mapped = match rest[0] {
                b'n' => Some(b'\n'),
                b't' => Some(b'\t'),
                b'r' => Some(b'\r'),
                b'"' => Some(b'"'),
                b'\\' => Some(b'\\'),
                b'a' => Some(0x07),
                b'b' => Some(0x08),
                b'f' => Some(0x0c),
                b'v' => Some(0x0b),
                _ => None,
            };
            if let Some(b) = mapped {
                out.push(b);
                i += 2;
                continue;
            }
        }
        out.push(bytes[i]);
        i += 1;
    }
    String::from_utf8(out).unwrap_or_else(|_| s.to_string())
}

fn strip_prefix_ab(p: &str) -> String {
    if let Some(s) = p.strip_prefix("a/") {
        return s.to_string();
    }
    if let Some(s) = p.strip_prefix("b/") {
        return s.to_string();
    }
    p.to_string()
}

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

    const SAMPLE: &str = "diff --git a/src/foo.rs b/src/foo.rs
index 1111111..2222222 100644
--- a/src/foo.rs
+++ b/src/foo.rs
@@ -1,3 +1,4 @@
 fn a() {}
-fn b() {}
+fn bb() {}
+fn c() {}
diff --git a/new.txt b/new.txt
new file mode 100644
index 0000000..3333333
--- /dev/null
+++ b/new.txt
@@ -0,0 +1,2 @@
+hello
+world
";

    #[test]
    fn parses_two_files() {
        let files = parse_unified(SAMPLE);
        assert_eq!(files.len(), 2);
        assert_eq!(files[0].path, "src/foo.rs");
        assert_eq!(files[0].status, FileStatus::Modified);
        assert_eq!(files[0].added, 2);
        assert_eq!(files[0].removed, 1);
        assert_eq!(files[1].path, "new.txt");
        assert_eq!(files[1].status, FileStatus::Added);
        assert_eq!(files[1].added, 2);
    }

    #[test]
    fn parses_paths_with_spaces() {
        // `core.quotepath=false` keeps spaces literal in diff headers.
        let input = "diff --git a/with space.txt b/with space.txt
index 111..222 100644
--- a/with space.txt\t
+++ b/with space.txt\t
@@ -1 +1 @@
-old
+new
";
        let files = parse_unified(input);
        assert_eq!(files.len(), 1);
        assert_eq!(files[0].path, "with space.txt");
        assert_eq!(files[0].old_path.as_deref(), Some("with space.txt"));
    }

    #[test]
    fn parses_paths_with_multiple_spaces_and_unicode() {
        let input = "diff --git a/emoji 🎉.md b/emoji 🎉.md
index 111..222 100644
--- a/emoji 🎉.md
+++ b/emoji 🎉.md
@@ -1 +1 @@
-old
+new
";
        let files = parse_unified(input);
        assert_eq!(files.len(), 1);
        assert_eq!(files[0].path, "emoji 🎉.md");
    }

    #[test]
    fn parses_added_file_with_space_in_name() {
        let input = "diff --git a/new file.txt b/new file.txt
new file mode 100644
index 0000000..3333333
--- /dev/null
+++ b/new file.txt
@@ -0,0 +1 @@
+hello
";
        let files = parse_unified(input);
        assert_eq!(files.len(), 1);
        assert_eq!(files[0].path, "new file.txt");
        assert_eq!(files[0].status, FileStatus::Added);
        assert_eq!(files[0].old_path, None);
    }

    #[test]
    fn parses_hunk_header_variants() {
        assert_eq!(parse_hunk_header("@@ -1,3 +1,4 @@"), Some((1, 1)));
        assert_eq!(parse_hunk_header("@@ -1 +1 @@"), Some((1, 1)));
        assert_eq!(parse_hunk_header("@@ -0,0 +1,2 @@"), Some((0, 1)));
        assert_eq!(
            parse_hunk_header("@@ -10,5 +20,7 @@ fn foo()"),
            Some((10, 20))
        );
        assert_eq!(parse_hunk_header("@@ broken"), None);
    }

    #[test]
    fn tracks_line_numbers_in_gutter() {
        let files = parse_unified(SAMPLE);
        let f = &files[0];
        let body: Vec<_> = f.lines.iter().filter(|l| l.kind.is_code()).collect();
        assert_eq!(body[0].gutter, "   1    1 ");
        assert_eq!(body[1].gutter, "   2      ");
        assert_eq!(body[2].gutter, "        2 ");
    }

    #[test]
    fn no_fold_for_first_hunk_at_line_1() {
        let files = parse_unified(SAMPLE);
        assert!(!files[0].lines.iter().any(|l| l.kind.is_fold()));
    }

    #[test]
    fn inserts_fold_between_hunks() {
        let input = "diff --git a/f.rs b/f.rs
index aaa..bbb 100644
--- a/f.rs
+++ b/f.rs
@@ -1,2 +1,3 @@
 line1
+added
 line2
@@ -20,2 +21,2 @@ fn bar()
 line20
-old
+new
";
        let files = parse_unified(input);
        let folds: Vec<_> = files[0].lines.iter().filter(|l| l.kind.is_fold()).collect();
        assert_eq!(folds.len(), 1);
        assert_eq!(folds[0].kind, LineKind::FoldDown);
        assert_eq!(folds[0].new_lineno, Some(4));
        assert_eq!(folds[0].fold_count, 17);
    }

    #[test]
    fn pair_selection_matches_by_similarity_not_order() {
        // Three removed lines, three added lines in a different order.
        // Naive positional pairing would pair r0-a0 etc (all mismatches);
        // similarity-based pairing should correctly match the related lines.
        let input = "diff --git a/f.rs b/f.rs
index a..b 100644
--- a/f.rs
+++ b/f.rs
@@ -1,6 +1,6 @@
-fn alpha_original(x: i32) {}
-fn beta_original(y: i32) {}
-fn gamma_original(z: i32) {}
+fn gamma_original(z: i64) {}
+fn alpha_original(x: i64) {}
+fn beta_original(y: i64) {}
";
        let files = parse_unified(input);
        let f = &files[0];
        let removed: Vec<&DiffLine> = f
            .lines
            .iter()
            .filter(|l| l.kind == LineKind::Removed)
            .collect();
        let added: Vec<&DiffLine> = f
            .lines
            .iter()
            .filter(|l| l.kind == LineKind::Added)
            .collect();
        assert_eq!(removed.len(), 3);
        assert_eq!(added.len(), 3);
        // Every line has a same-name counterpart differing only in i32/i64,
        // so all six should be marked paired with non-empty word_highlights.
        for line in removed.iter().chain(added.iter()) {
            assert!(line.paired, "expected paired for text={:?}", line.text);
            assert!(
                !line.word_highlights.is_empty(),
                "expected word_highlights for text={:?}",
                line.text,
            );
        }
    }

    #[test]
    fn unpaired_lines_stay_unpaired() {
        // Single removed line with no remotely similar added line.
        let input = "diff --git a/f.rs b/f.rs
index a..b 100644
--- a/f.rs
+++ b/f.rs
@@ -1,2 +1,2 @@
-fn delete_me_completely() {}
+struct WhollyNewThing;
";
        let files = parse_unified(input);
        let f = &files[0];
        let removed = f
            .lines
            .iter()
            .find(|l| l.kind == LineKind::Removed)
            .unwrap();
        let added = f.lines.iter().find(|l| l.kind == LineKind::Added).unwrap();
        assert!(!removed.paired);
        assert!(!added.paired);
        assert!(removed.word_highlights.is_empty());
        assert!(added.word_highlights.is_empty());
    }

    #[test]
    fn auto_collapses_lock_files() {
        assert!(should_auto_collapse("Cargo.lock"));
        assert!(should_auto_collapse("sub/package-lock.json"));
        assert!(should_auto_collapse("vendor/foo.go"));
        assert!(!should_auto_collapse("src/main.rs"));
    }

    #[test]
    fn parses_no_newline_at_eof_marker() {
        let input = "diff --git a/f.txt b/f.txt
index 111..222 100644
--- a/f.txt
+++ b/f.txt
@@ -1 +1 @@
-old line
\\ No newline at end of file
+new line
\\ No newline at end of file
";
        let files = parse_unified(input);
        let f = &files[0];
        let removed = f
            .lines
            .iter()
            .find(|l| l.kind == LineKind::Removed)
            .unwrap();
        let added = f.lines.iter().find(|l| l.kind == LineKind::Added).unwrap();
        assert!(removed.no_newline);
        assert!(added.no_newline);
    }

    #[test]
    fn auto_collapsed_files_seed_viewed() {
        let input = "diff --git a/package-lock.json b/package-lock.json
new file mode 100644
index 0000000..3333333
--- /dev/null
+++ b/package-lock.json
@@ -0,0 +1,2 @@
+{
+}
diff --git a/src/main.rs b/src/main.rs
index 111..222 100644
--- a/src/main.rs
+++ b/src/main.rs
@@ -1 +1 @@
-old
+new
";
        let files = parse_unified(input);
        let lock = files
            .iter()
            .find(|f| f.path == "package-lock.json")
            .unwrap();
        let main = files.iter().find(|f| f.path == "src/main.rs").unwrap();
        assert!(lock.auto_collapsed);
        assert!(lock.viewed);
        assert!(!main.auto_collapsed);
        assert!(!main.viewed);
    }

    #[test]
    fn unquotes_core_quotepath_headers() {
        // Output of `git diff` with default `core.quotepath=true` wraps
        // non-ASCII paths in `"..."` and octal-escapes each UTF-8 byte.
        let input = "diff --git \"a/\\346\\227\\245\\346\\234\\254\\350\\252\\236.md\" \"b/\\346\\227\\245\\346\\234\\254\\350\\252\\236.md\"
index 111..222 100644
--- \"a/\\346\\227\\245\\346\\234\\254\\350\\252\\236.md\"
+++ \"b/\\346\\227\\245\\346\\234\\254\\350\\252\\236.md\"
@@ -1 +1 @@
-old
+new
";
        let files = parse_unified(input);
        assert_eq!(files.len(), 1);
        assert_eq!(files[0].path, "日本語.md");
        assert_eq!(files[0].old_path.as_deref(), Some("日本語.md"));
    }

    #[test]
    fn unquote_c_path_handles_simple_escapes() {
        assert_eq!(unquote_c_path("foo"), "foo");
        assert_eq!(unquote_c_path("a\\tb"), "a\tb");
        assert_eq!(unquote_c_path("a\\\"b"), "a\"b");
        assert_eq!(unquote_c_path("a\\\\b"), "a\\b");
        // `\040` = space
        assert_eq!(unquote_c_path("a\\040b"), "a b");
    }

    #[test]
    fn handles_dev_null_clears_old_path_for_added_file() {
        let input = "diff --git a/new.txt b/new.txt
new file mode 100644
index 0000000..3333333
--- /dev/null
+++ b/new.txt
@@ -0,0 +1 @@
+hello
";
        let files = parse_unified(input);
        // `--- /dev/null` must clear the stale old_path left by the
        // `diff --git a/new.txt b/new.txt` header.
        assert_eq!(files[0].old_path, None);
    }
}

/// Layer 2: Check first 20 lines of diff content for generated-file markers.
fn has_generated_marker(lines: &[DiffLine]) -> bool {
    const MARKERS: &[&str] = &[
        "@GENERATED",
        "DO NOT EDIT",
        "AUTO-GENERATED",
        "AUTO GENERATED",
        "CODE GENERATED BY",
        "GENERATED BY ORVAL",
        "PRISMA CLIENT",
        "THIS FILE IS AUTO-GENERATED",
        "THIS FILE WAS AUTOMATICALLY GENERATED",
        "GENERATED CODE",
        "GENERATED SIGNEDSOURCE",
    ];
    for line in lines.iter().filter(|l| l.kind.is_code()).take(20) {
        let upper = line.text.to_uppercase();
        if MARKERS.iter().any(|m| upper.contains(m)) {
            return true;
        }
    }
    false
}

#[derive(Debug, Clone, Copy)]
pub struct SplitRow {
    pub left: Option<usize>,
    pub right: Option<usize>,
}

pub fn build_split_rows(file: &FileDiff) -> Vec<SplitRow> {
    let mut rows: Vec<SplitRow> = Vec::new();
    let mut removed_buf: Vec<usize> = Vec::new();
    let mut added_buf: Vec<usize> = Vec::new();

    let flush = |removed: &mut Vec<usize>, added: &mut Vec<usize>, out: &mut Vec<SplitRow>| {
        let max = removed.len().max(added.len());
        for i in 0..max {
            out.push(SplitRow {
                left: removed.get(i).copied(),
                right: added.get(i).copied(),
            });
        }
        removed.clear();
        added.clear();
    };

    for (idx, line) in file.lines.iter().enumerate() {
        match line.kind {
            LineKind::Removed => {
                if !added_buf.is_empty() {
                    flush(&mut removed_buf, &mut added_buf, &mut rows);
                }
                removed_buf.push(idx);
            }
            LineKind::Added => {
                added_buf.push(idx);
            }
            _ => {
                flush(&mut removed_buf, &mut added_buf, &mut rows);
                rows.push(SplitRow {
                    left: Some(idx),
                    right: Some(idx),
                });
            }
        }
    }
    flush(&mut removed_buf, &mut added_buf, &mut rows);
    rows
}

fn should_auto_collapse(path: &str) -> bool {
    let name = path.rsplit('/').next().unwrap_or(path);

    const LOCK_FILES: &[&str] = &[
        "package-lock.json",
        "yarn.lock",
        "pnpm-lock.yaml",
        "Cargo.lock",
        "Gemfile.lock",
        "Pipfile.lock",
        "poetry.lock",
        "composer.lock",
        "go.sum",
    ];
    if LOCK_FILES.contains(&name) {
        return true;
    }

    if name.ends_with(".min.js")
        || name.ends_with(".min.css")
        || name.ends_with(".map")
        || name.ends_with(".generated.ts")
        || name.ends_with(".gen.go")
    {
        return true;
    }

    const DIR_PATTERNS: &[&str] = &[
        "vendor/",
        "node_modules/",
        "dist/",
        "build/",
        "/generated/",
        "/__generated__/",
        "/gen/",
    ];
    for pat in DIR_PATTERNS {
        if path.contains(pat) || path.starts_with(pat.trim_start_matches('/')) {
            return true;
        }
    }

    false
}