rdsh 0.12.0

// Native recursive grep: skips hidden/vendor dirs, caps output.
// Two phases: sequential walk (fixed order) then parallel grep over files.
// Stdout matches a sequential scan: per-file hits merge in walk order.

pub fn cmd_search(
    pattern: &str,
    dir: &str,
    max: usize,
    locale: crate::locale::Locale,
) -> anyhow::Result<()> {
    anyhow::ensure!(
        cfg!(unix),
        "{}",
        crate::locale::native_message(locale, crate::locale::NativeMessage::SearchUnavailable)
    );
    let root = std::fs::canonicalize(dir).map_err(|error| match locale {
        crate::locale::Locale::English => error.into(),
        crate::locale::Locale::Chinese | crate::locale::Locale::Japanese => anyhow::anyhow!(
            "{} {dir}: {error}",
            crate::locale::native_message(
                locale,
                crate::locale::NativeMessage::CannotAccessSearchDirectory
            )
        ),
    })?;
    let reader = crate::file_security::Root::new(&root);
    let files = collect_parallel(&root);
    let nfiles = files.len();
    let per_file: Vec<Vec<String>> = match reader {
        Some(reader) if nfiles >= 32 => grep_parallel(pattern, &files, &reader, max),
        Some(reader) => grep_chunk(pattern, &files, &reader, max),
        None => vec![],
    };
    // One locked, buffered stdout for the whole dump (same bytes out).
    use std::io::Write;
    let stdout = std::io::stdout();
    let mut out = std::io::BufWriter::new(stdout.lock());
    let mut shown = 0usize;
    let mut outer_done = false;
    for (path, fh) in files.iter().zip(&per_file) {
        if outer_done {
            break;
        }
        if fh.is_empty() {
            continue;
        }
        let prefix = format!("{}:", path.display());
        let display_path = std::path::Path::new(dir).join(path.strip_prefix(&root)?);
        for line in fh {
            if shown >= max {
                outer_done = true;
                break;
            }
            if let Some(hit) = line.strip_prefix(&prefix) {
                // The matched line comes from the searched tree; keep it from
                // rewriting the terminal around the result.
                let _ = writeln!(
                    out,
                    "{}:{}",
                    crate::terminal_safe::strip_control(&display_path.display().to_string()),
                    crate::terminal_safe::strip_control(hit)
                );
            }
            shown += 1;
        }
    }
    let _ = out.flush();
    match locale {
        crate::locale::Locale::English => eprintln!("[rdsh] {shown} hit(s) in {nfiles} file(s)"),
        crate::locale::Locale::Chinese => {
            eprintln!("[rdsh] 在 {nfiles} 个文件中找到 {shown} 个匹配")
        }
        crate::locale::Locale::Japanese => {
            eprintln!("[rdsh] {nfiles} 個のファイルで {shown} 件一致")
        }
    }
    Ok(())
}
const SKIP: &[&str] = &[
    ".git",
    "node_modules",
    "target",
    ".venv",
    "dist",
    "build",
    ".next",
];

fn collect_files(dir: &std::path::Path, out: &mut Vec<std::path::PathBuf>) {
    let entries = match std::fs::read_dir(dir) {
        Ok(e) => e,
        Err(_) => return,
    };
    for e in entries.filter_map(Result::ok) {
        // file_type is free from dirent, no stat. Files need no name check.
        let ft = match e.file_type() {
            Ok(t) => t,
            Err(_) => continue,
        };
        if ft.is_file() {
            out.push(e.path());
            continue;
        }
        if !ft.is_dir() {
            continue;
        }
        if is_skip_dir_name(&e.file_name()) {
            continue;
        }
        let p = e.path();
        collect_files(&p, out);
    }
}

#[derive(PartialEq)]
enum EntryKind {
    Dir,
    File,
    Skip,
}

/// True for hidden or vendor dirs, byte compare, no String alloc.
/// Same skip set as before, hidden check is first byte '.'.
fn is_skip_dir_name(name: &std::ffi::OsStr) -> bool {
    let b = name.as_encoded_bytes();
    if b.first() == Some(&b'.') {
        return true;
    }
    for s in SKIP {
        if b == s.as_bytes() {
            return true;
        }
    }
    false
}

/// Classify without stat in the common case (dirent type is free).
/// Skip symlinks and unknown types to keep repository traversal local.
fn entry_kind(e: &std::fs::DirEntry) -> EntryKind {
    let ft = match e.file_type() {
        Ok(t) => t,
        Err(_) => return EntryKind::Skip,
    };
    if ft.is_file() {
        return EntryKind::File;
    }
    if !ft.is_dir() {
        return EntryKind::Skip;
    }
    if is_skip_dir_name(&e.file_name()) {
        return EntryKind::Skip;
    }
    EntryKind::Dir
}

/// Walk subdirectories in parallel while preserving exact sequential order:
/// root entries keep their listing order and each subtree is joined in place.
/// Falls back to the plain sequential walk for narrow trees.
fn collect_parallel(dir: &std::path::Path) -> Vec<std::path::PathBuf> {
    let entries: Vec<std::fs::DirEntry> = match std::fs::read_dir(dir) {
        Ok(e) => e.filter_map(Result::ok).collect(),
        Err(_) => return vec![],
    };
    let mut subdirs = vec![];
    for e in &entries {
        if entry_kind(e) == EntryKind::Dir {
            subdirs.push(e.path());
        }
    }
    if subdirs.len() < 2 {
        let mut out = vec![];
        collect_files(dir, &mut out);
        return out;
    }
    enum Seg {
        Files(Vec<std::path::PathBuf>),
        Sub(std::path::PathBuf),
    }
    let mut segs: Vec<Seg> = Vec::new();
    let mut pending: Vec<std::path::PathBuf> = Vec::new();
    for e in &entries {
        match entry_kind(e) {
            EntryKind::Dir => {
                if !pending.is_empty() {
                    segs.push(Seg::Files(std::mem::take(&mut pending)));
                }
                segs.push(Seg::Sub(e.path()));
            }
            EntryKind::File => pending.push(e.path()),
            EntryKind::Skip => {}
        }
    }
    if !pending.is_empty() {
        segs.push(Seg::Files(pending));
    }
    let sub_idx: Vec<usize> = segs
        .iter()
        .enumerate()
        .filter(|(_, s)| matches!(s, Seg::Sub(_)))
        .map(|(i, _)| i)
        .collect();
    let mut resolved: Vec<Vec<std::path::PathBuf>> = Vec::new();
    resolved.resize_with(sub_idx.len(), Vec::new);
    let width = crate::inspect::parallelism().max(1);
    for (batch_no, batch) in sub_idx.chunks(width).enumerate() {
        std::thread::scope(|s| {
            let mut handles = vec![];
            for (k, seg_i) in batch.iter().enumerate() {
                if let Seg::Sub(p) = &segs[*seg_i] {
                    // slot in `resolved` = global index into sub_idx.
                    let pos = batch_no * width + k;
                    handles.push((
                        pos,
                        s.spawn(move || {
                            let mut v = vec![];
                            collect_files(p, &mut v);
                            v
                        }),
                    ));
                }
            }
            for (pos, h) in handles {
                resolved[pos] = h.join().unwrap_or_default();
            }
        });
    }
    let mut files = vec![];
    let mut ri = 0;
    for seg in segs {
        match seg {
            Seg::Files(v) => files.extend(v),
            Seg::Sub(_) => {
                files.extend(std::mem::take(&mut resolved[ri]));
                ri += 1;
            }
        }
    }
    files
}

/// Read at most 2MB+1 bytes as UTF-8. Returns None for missing files,
/// files over 2MB, and non-UTF-8 content — the same skip set as the old
/// metadata-check plus read_to_string combination (verified by diff).
fn read_capped(path: &std::path::Path, root: &crate::file_security::Root) -> Option<String> {
    use std::io::Read;
    let f = root.open(path)?;
    // One fstat on the open fd (no path lookup): skip oversized files
    // without reading, and pre-size the buffer to avoid regrowth.
    // Same skip set as the old metadata-check combination (verified by diff).
    let len = f.metadata().ok()?.len();
    if len > 2_000_000 {
        return None;
    }
    let mut buf = Vec::with_capacity(len as usize);
    f.take(2_000_001).read_to_end(&mut buf).ok()?;
    if buf.len() > 2_000_000 {
        return None;
    }
    String::from_utf8(buf).ok()
}

fn grep_one(
    pattern: &str,
    path: &std::path::Path,
    root: &crate::file_security::Root,
    max: usize,
) -> Vec<String> {
    let mut hits = vec![];
    if max == 0 {
        return hits;
    }
    if let Some(text) = read_capped(path, root) {
        for (i, line) in text.lines().enumerate() {
            if line.contains(pattern) {
                hits.push(format!(
                    "{}:{}: {}",
                    path.display(),
                    i + 1,
                    truncate(line, 240)
                ));
                if hits.len() == max {
                    break;
                }
            }
        }
    }
    hits
}

/// An ordered chunk never contributes more than the global output limit.
/// Retain at most workers × max hits, even when every line in every file matches.
fn grep_chunk(
    pattern: &str,
    files: &[std::path::PathBuf],
    root: &crate::file_security::Root,
    mut remaining: usize,
) -> Vec<Vec<String>> {
    files
        .iter()
        .map(|path| {
            let hits = grep_one(pattern, path, root, remaining);
            remaining -= hits.len();
            hits
        })
        .collect()
}

fn grep_parallel(
    pattern: &str,
    files: &[std::path::PathBuf],
    root: &crate::file_security::Root,
    max: usize,
) -> Vec<Vec<String>> {
    let threads = crate::inspect::parallelism();
    if threads <= 1 {
        return grep_chunk(pattern, files, root, max);
    }
    let chunk = files.len().div_ceil(threads);
    let mut out: Vec<Vec<Vec<String>>> = Vec::new();
    std::thread::scope(|s| {
        let mut handles = vec![];
        for c in files.chunks(chunk) {
            handles.push(s.spawn(move || grep_chunk(pattern, c, root, max)));
        }
        for h in handles {
            out.push(h.join().unwrap_or_default());
        }
    });
    out.into_iter().flatten().collect()
}

fn truncate(s: &str, n: usize) -> String {
    let mut rest = s.chars();
    let mut t: String = rest.by_ref().take(n).collect();
    if rest.next().is_some() {
        t.push('>');
    }
    t
}

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

    /// 切り詰めは文字数基準。バイト長で判定すると、240文字以下の非ASCII行に
    /// 偽のマーカーが付く（日本語81文字＝243バイト）。
    #[test]
    fn truncate_marks_only_when_chars_are_actually_cut() {
        let cjk = "あ".repeat(81);
        assert_eq!(truncate(&cjk, 240), cjk);
        assert_eq!(
            truncate(&"a".repeat(241), 240),
            format!("{}>", "a".repeat(240))
        );
        assert_eq!(
            truncate(&"あ".repeat(241), 240),
            format!("{}>", "あ".repeat(240))
        );
    }

    #[test]
    fn worker_cap() {
        // Issue #85-4: walker/grep threads follow cores, clamped 1..=8.
        assert!((1..=8).contains(&crate::inspect::parallelism()));
    }

    #[cfg(unix)]
    #[test]
    fn bounded_workers_preserve_global_prefix_and_skip_reads_at_zero() {
        let temporary = std::env::temp_dir().join(format!(
            "rdsh-search-limit-{}",
            crate::local_http::random_token().unwrap()
        ));
        std::fs::create_dir(&temporary).unwrap();
        let root = std::fs::canonicalize(&temporary).unwrap();
        let reader = crate::file_security::Root::new(&root).unwrap();
        let mut files = Vec::new();
        let mut expected = Vec::new();
        for i in 0..65 {
            let path = root.join(format!("file-{i}.txt"));
            // Some chunks start with files containing no matches.
            let text = if i % 4 == 0 {
                "ordinary\n"
            } else {
                "match A\nordinary\nmatch B\n"
            };
            std::fs::write(&path, text).unwrap();
            if i % 4 != 0 {
                expected.push(format!("{}:1: match A", path.display()));
                expected.push(format!("{}:3: match B", path.display()));
            }
            files.push(path);
        }
        for max in [0, 1, 3, 19, 300] {
            let hits = grep_parallel("match", &files, &reader, max)
                .into_iter()
                .flatten()
                .collect::<Vec<_>>();
            assert!(hits.len() <= crate::inspect::parallelism() * max);
            assert_eq!(
                hits.into_iter().take(max).collect::<Vec<_>>(),
                expected.iter().take(max).cloned().collect::<Vec<_>>()
            );
        }
        assert!(grep_one("match", &root.join("missing"), &reader, 0).is_empty());
        std::fs::remove_dir_all(temporary).unwrap();
    }

    #[cfg(unix)]
    #[test]
    fn recursive_search_does_not_follow_repository_symlinks() {
        use std::os::unix::fs::symlink;
        let root = std::env::temp_dir().join(format!(
            "rdsh-search-{}",
            crate::local_http::random_token().unwrap()
        ));
        let repo = root.join("repo");
        let outside = root.join("outside");
        std::fs::create_dir_all(&repo).unwrap();
        std::fs::create_dir_all(&outside).unwrap();
        std::fs::write(outside.join("credentials.json"), "DUMMY_SECRET").unwrap();
        std::fs::write(repo.join("README.md"), "safe").unwrap();
        symlink(&outside, repo.join("linked-dir")).unwrap();
        symlink(outside.join("credentials.json"), repo.join("linked.json")).unwrap();
        let files = collect_parallel(&repo);
        assert_eq!(files, vec![repo.join("README.md")]);
        std::fs::remove_dir_all(root).unwrap();
    }

    #[cfg(unix)]
    #[test]
    fn search_rejects_files_replaced_after_collection() {
        let temporary = std::env::temp_dir().join(format!(
            "rdsh-search-swap-{}",
            crate::local_http::random_token().unwrap()
        ));
        let project = temporary.join("project");
        std::fs::create_dir_all(&project).unwrap();
        let root = std::fs::canonicalize(&project).unwrap();
        let slot = root.join("slot.txt");
        let outside = temporary.join("secret.txt");
        std::fs::write(&slot, "safe").unwrap();
        std::fs::write(&outside, "DUMMY_SEARCH_SECRET").unwrap();
        let collected = collect_parallel(&root);
        let reader = crate::file_security::Root::new(&root).unwrap();
        assert_eq!(collected, vec![slot.clone()]);
        std::fs::remove_file(&slot).unwrap();
        std::os::unix::fs::symlink(&outside, &slot).unwrap();
        assert!(grep_one("DUMMY_SEARCH_SECRET", &collected[0], &reader, 10).is_empty());
        std::fs::remove_file(&slot).unwrap();
        std::fs::hard_link(&outside, &slot).unwrap();
        assert!(
            grep_parallel("DUMMY_SEARCH_SECRET", &collected, &reader, 10)
                .into_iter()
                .flatten()
                .next()
                .is_none()
        );
        std::fs::remove_dir_all(temporary).unwrap();
    }

    #[cfg(unix)]
    #[test]
    fn capped_boundary() {
        let dir = std::env::temp_dir().join(format!(
            "rdsh-search-cap-{}",
            crate::local_http::random_token().unwrap()
        ));
        std::fs::create_dir(&dir).unwrap();
        let dir = std::fs::canonicalize(dir).unwrap();
        let reader = crate::file_security::Root::new(&dir).unwrap();
        let exact = dir.join("rdsh-cap-exact.txt");
        let over = dir.join("rdsh-cap-over.txt");
        let bin = dir.join("rdsh-cap-bin.bin");
        std::fs::write(&exact, vec![120u8; 2_000_000]).unwrap();
        let mut big = vec![120u8; 2_000_001];
        big.push(121);
        std::fs::write(&over, big).unwrap();
        std::fs::write(&bin, [0u8, 159, 146, 150]).unwrap();
        assert!(read_capped(&exact, &reader).is_some());
        assert!(read_capped(&over, &reader).is_none());
        assert!(read_capped(&bin, &reader).is_none());
        assert!(read_capped(&dir.join("rdsh-cap-missing.txt"), &reader).is_none());
        std::fs::remove_dir_all(dir).unwrap();
    }
}