lean_ctx/tools/

ctx_edit.rs

use std::path::{Path, PathBuf};
use std::time::{SystemTime, UNIX_EPOCH};

use crate::core::cache::SessionCache;
use crate::core::tokens::count_tokens;

/// Parameters for a file edit operation: path, old/new strings, and flags.
pub struct EditParams {
    pub path: String,
    pub old_string: String,
    pub new_string: String,
    pub replace_all: bool,
    pub create: bool,
    /// Optional preimage guards. If provided, ctx_edit fails if the current file preimage differs.
    pub expected_md5: Option<String>,
    pub expected_size: Option<u64>,
    pub expected_mtime_ms: Option<u64>,
    /// Optional backup before writing.
    pub backup: bool,
    pub backup_path: Option<String>,
    /// Emit bounded diff evidence (redacted) by default.
    pub evidence: bool,
    pub diff_max_lines: usize,
    /// Reject invalid UTF-8 by default; allow lossy reads only when explicitly enabled.
    pub allow_lossy_utf8: bool,
}

struct ReplaceArgs<'a> {
    content: &'a str,
    old_str: &'a str,
    new_str: &'a str,
    occurrences: usize,
    replace_all: bool,
    old_tokens: usize,
    new_tokens: usize,
}

#[derive(Clone, Debug, PartialEq, Eq)]
struct FileFingerprint {
    size: u64,
    mtime_ms: u64,
    md5: String,
}

#[derive(Clone, Debug)]
struct FilePreimage {
    fp: FileFingerprint,
    permissions: std::fs::Permissions,
    bytes: Vec<u8>,
    text: String,
    uses_crlf: bool,
}

fn system_time_to_millis(t: SystemTime) -> u64 {
    t.duration_since(UNIX_EPOCH)
        .map_or(0, |d| d.as_millis() as u64)
}

fn read_file_bytes_limited(
    path: &Path,
    cap: usize,
) -> Result<(Vec<u8>, std::fs::Metadata), String> {
    if let Ok(meta) = std::fs::metadata(path) {
        if meta.len() > cap as u64 {
            return Err(format!(
                "ERROR: file too large ({} bytes, cap {} via LCTX_MAX_READ_BYTES): {}",
                meta.len(),
                cap,
                path.display()
            ));
        }
    }

    let mut file = std::fs::OpenOptions::new()
        .read(true)
        .open(path)
        .map_err(|e| format!("ERROR: cannot open {}: {e}", path.display()))?;

    use std::io::Read;
    let mut raw: Vec<u8> = Vec::new();
    let mut limited = (&mut file).take((cap as u64).saturating_add(1));
    limited
        .read_to_end(&mut raw)
        .map_err(|e| format!("ERROR: cannot read {}: {e}", path.display()))?;
    if raw.len() > cap {
        return Err(format!(
            "ERROR: file too large (cap {} via LCTX_MAX_READ_BYTES): {}",
            cap,
            path.display()
        ));
    }

    let meta = file
        .metadata()
        .map_err(|e| format!("ERROR: cannot stat {}: {e}", path.display()))?;
    Ok((raw, meta))
}

fn fingerprint_from_bytes(bytes: &[u8], meta: &std::fs::Metadata) -> FileFingerprint {
    FileFingerprint {
        size: bytes.len() as u64,
        mtime_ms: meta.modified().map_or(0, system_time_to_millis),
        md5: crate::core::hasher::hash_hex(bytes),
    }
}

fn read_preimage(path: &Path, cap: usize, allow_lossy_utf8: bool) -> Result<FilePreimage, String> {
    let (bytes, meta) = read_file_bytes_limited(path, cap)?;
    let permissions = meta.permissions();
    let fp = fingerprint_from_bytes(&bytes, &meta);

    let text = if allow_lossy_utf8 {
        String::from_utf8_lossy(&bytes).into_owned()
    } else {
        String::from_utf8(bytes.clone()).map_err(|_| {
            format!(
                "ERROR: file is not valid UTF-8 (binary/encoding). Refusing to edit: {}",
                path.display()
            )
        })?
    };
    let uses_crlf = text.contains("\r\n");

    Ok(FilePreimage {
        fp,
        permissions,
        bytes,
        text,
        uses_crlf,
    })
}

fn verify_expected_preimage(pre: &FilePreimage, params: &EditParams) -> Result<(), String> {
    if let Some(expected) = params.expected_size {
        if expected != pre.fp.size {
            return Err(format!(
                "ERROR: preimage mismatch for {}: expected_size={}, actual_size={}",
                params.path, expected, pre.fp.size
            ));
        }
    }
    if let Some(expected) = params.expected_mtime_ms {
        if expected != pre.fp.mtime_ms {
            return Err(format!(
                "ERROR: preimage mismatch for {}: expected_mtime_ms={}, actual_mtime_ms={}",
                params.path, expected, pre.fp.mtime_ms
            ));
        }
    }
    if let Some(expected) = params.expected_md5.as_deref() {
        if expected != pre.fp.md5 {
            return Err(format!(
                "ERROR: preimage mismatch for {}: expected_md5={}, actual_md5={}",
                params.path, expected, pre.fp.md5
            ));
        }
    }
    Ok(())
}

fn ensure_preimage_still_matches(
    path: &Path,
    expected: &FileFingerprint,
    cap: usize,
) -> Result<(), String> {
    let (bytes, meta) = read_file_bytes_limited(path, cap)?;
    let now = fingerprint_from_bytes(&bytes, &meta);
    if &now != expected {
        return Err(format!(
            "ERROR: file changed since read (TOCTOU guard). Re-read and retry: {}\nexpected: size={}, mtime_ms={}, md5={}\nactual:   size={}, mtime_ms={}, md5={}",
            path.display(),
            expected.size,
            expected.mtime_ms,
            expected.md5,
            now.size,
            now.mtime_ms,
            now.md5
        ));
    }
    Ok(())
}

fn default_backup_path(path: &Path) -> Option<PathBuf> {
    let parent = path.parent()?;
    let filename = path.file_name()?.to_string_lossy();
    let pid = std::process::id();
    let nanos = SystemTime::now()
        .duration_since(UNIX_EPOCH)
        .map_or(0, |d| d.as_nanos());
    Some(parent.join(format!("{filename}.lean-ctx.bak.{pid}.{nanos}")))
}

fn write_atomic_bytes_with_permissions(
    path: &Path,
    bytes: &[u8],
    permissions: Option<&std::fs::Permissions>,
) -> Result<(), String> {
    if let Some(parent) = path.parent() {
        std::fs::create_dir_all(parent).map_err(|e| e.to_string())?;
    }

    let parent = path
        .parent()
        .ok_or_else(|| "invalid path (no parent directory)".to_string())?;
    let filename = path
        .file_name()
        .ok_or_else(|| "invalid path (no filename)".to_string())?
        .to_string_lossy();

    let pid = std::process::id();
    let nanos = SystemTime::now()
        .duration_since(UNIX_EPOCH)
        .map_or(0, |d| d.as_nanos());
    let tmp = parent.join(format!(".{filename}.lean-ctx.tmp.{pid}.{nanos}"));

    {
        use std::io::Write;
        let mut f = std::fs::OpenOptions::new()
            .write(true)
            .create_new(true)
            .open(&tmp)
            .map_err(|e| format!("ERROR: cannot write {}: {e}", tmp.display()))?;
        f.write_all(bytes)
            .map_err(|e| format!("ERROR: cannot write {}: {e}", tmp.display()))?;
        let _ = f.flush();
        let _ = f.sync_all();
    }

    if let Some(perms) = permissions {
        let _ = std::fs::set_permissions(&tmp, perms.clone());
    }

    #[cfg(windows)]
    {
        if path.exists() {
            let _ = std::fs::remove_file(path);
        }
    }

    std::fs::rename(&tmp, path).map_err(|e| {
        format!(
            "ERROR: atomic write failed: {} (tmp: {})",
            e,
            tmp.to_string_lossy()
        )
    })?;

    Ok(())
}

macro_rules! static_regex {
    ($pattern:expr) => {{
        static RE: std::sync::OnceLock<regex::Regex> = std::sync::OnceLock::new();
        RE.get_or_init(|| {
            regex::Regex::new($pattern).expect(concat!("BUG: invalid static regex: ", $pattern))
        })
    }};
}

fn redact_sensitive_diff(input: &str) -> String {
    let patterns: Vec<(&str, &regex::Regex)> = vec![
        (
            "Bearer token",
            static_regex!(r"(?i)(bearer\s+)[a-zA-Z0-9\-_\.]{8,}"),
        ),
        (
            "Authorization header",
            static_regex!(r"(?i)(authorization:\s*(?:basic|bearer|token)\s+)[^\s\r\n]+"),
        ),
        (
            "API key param",
            static_regex!(
                r#"(?i)((?:api[_-]?key|apikey|access[_-]?key|secret[_-]?key|token|password|passwd|pwd|secret)\s*[=:]\s*)[^\s\r\n,;&"']+"#
            ),
        ),
        ("AWS key", static_regex!(r"(AKIA[0-9A-Z]{12,})")),
        (
            "Private key block",
            static_regex!(
                r"(?s)(-----BEGIN\s+(?:RSA\s+)?PRIVATE\s+KEY-----).+?(-----END\s+(?:RSA\s+)?PRIVATE\s+KEY-----)"
            ),
        ),
        (
            "GitHub token",
            static_regex!(r"(gh[pousr]_)[a-zA-Z0-9]{20,}"),
        ),
        (
            "Generic long secret",
            static_regex!(
                r#"(?i)(?:key|token|secret|password|credential|auth)\s*[=:]\s*['"]?([a-zA-Z0-9+/=\-_]{32,})['"]?"#
            ),
        ),
    ];

    let mut out = input.to_string();
    for (label, re) in &patterns {
        out = re
            .replace_all(&out, |caps: &regex::Captures| {
                if let Some(prefix) = caps.get(1) {
                    format!("{}[REDACTED:{}]", prefix.as_str(), label)
                } else {
                    format!("[REDACTED:{label}]")
                }
            })
            .to_string();
    }
    out
}

fn build_diff_evidence(old: &str, new: &str, label: &str, max_lines: usize) -> String {
    let diff = similar::TextDiff::from_lines(old, new)
        .unified_diff()
        .context_radius(3)
        .header(label, label)
        .to_string();
    let diff = redact_sensitive_diff(&diff);

    let mut out = String::new();
    for (i, line) in diff.lines().enumerate() {
        if i >= max_lines {
            out.push_str(&format!("\n... diff truncated (max_lines={max_lines})"));
            break;
        }
        out.push_str(line);
        out.push('\n');
    }
    out.trim_end_matches('\n').to_string()
}

/// Performs a string replacement edit on a file with CRLF/LF and whitespace tolerance.
pub fn handle(cache: &mut SessionCache, params: &EditParams) -> String {
    let file_path = &params.path;

    if params.create {
        return handle_create(cache, file_path, &params.new_string, params);
    }

    let cap = crate::core::limits::max_read_bytes();
    let path = Path::new(file_path);
    let pre = match read_preimage(path, cap, params.allow_lossy_utf8) {
        Ok(p) => p,
        Err(e) => return e,
    };
    if let Err(e) = verify_expected_preimage(&pre, params) {
        return e;
    }
    let content = &pre.text;

    if params.old_string.is_empty() {
        return "ERROR: old_string must not be empty (use create=true to create a new file)".into();
    }

    let uses_crlf = pre.uses_crlf;
    let old_str = &params.old_string;
    let new_str = &params.new_string;

    let occurrences = content.matches(old_str).count();

    if occurrences > 0 {
        let args = ReplaceArgs {
            content,
            old_str,
            new_str,
            occurrences,
            replace_all: params.replace_all,
            old_tokens: count_tokens(&params.old_string),
            new_tokens: count_tokens(&params.new_string),
        };
        return do_replace(cache, path, &pre, params, cap, &args);
    }

    // Direct match failed -- try CRLF/LF normalization
    if uses_crlf && !old_str.contains('\r') {
        let old_crlf = old_str.replace('\n', "\r\n");
        let occ = content.matches(&old_crlf).count();
        if occ > 0 {
            let new_crlf = new_str.replace('\n', "\r\n");
            let args = ReplaceArgs {
                content,
                old_str: &old_crlf,
                new_str: &new_crlf,
                occurrences: occ,
                replace_all: params.replace_all,
                old_tokens: count_tokens(&params.old_string),
                new_tokens: count_tokens(&params.new_string),
            };
            return do_replace(cache, path, &pre, params, cap, &args);
        }
    } else if !uses_crlf && old_str.contains("\r\n") {
        let old_lf = old_str.replace("\r\n", "\n");
        let occ = content.matches(&old_lf).count();
        if occ > 0 {
            let new_lf = new_str.replace("\r\n", "\n");
            let args = ReplaceArgs {
                content,
                old_str: &old_lf,
                new_str: &new_lf,
                occurrences: occ,
                replace_all: params.replace_all,
                old_tokens: count_tokens(&params.old_string),
                new_tokens: count_tokens(&params.new_string),
            };
            return do_replace(cache, path, &pre, params, cap, &args);
        }
    }

    // Still not found -- try trimmed trailing whitespace per line
    let normalized_content = trim_trailing_per_line(content);
    let normalized_old = trim_trailing_per_line(old_str);
    if !normalized_old.is_empty() && normalized_content.contains(&normalized_old) {
        let line_sep = if uses_crlf { "\r\n" } else { "\n" };
        let adapted_new = adapt_new_string_to_line_sep(new_str, line_sep);
        let adapted_old = find_original_span(content, &normalized_old);
        if let Some(original_match) = adapted_old {
            let occ = content.matches(&original_match).count();
            let args = ReplaceArgs {
                content,
                old_str: &original_match,
                new_str: &adapted_new,
                occurrences: occ,
                replace_all: params.replace_all,
                old_tokens: count_tokens(&params.old_string),
                new_tokens: count_tokens(&params.new_string),
            };
            return do_replace(cache, path, &pre, params, cap, &args);
        }
    }

    let preview = if old_str.len() > 80 {
        format!("{}...", &old_str[..77])
    } else {
        old_str.clone()
    };
    let hint = if uses_crlf {
        " (file uses CRLF line endings)"
    } else {
        ""
    };

    let escalation = auto_escalate_reread(cache, file_path);

    format!(
        "ERROR: old_string not found in {file_path}{hint}. \
         Make sure it matches exactly (including whitespace/indentation).\n\
         Searched for: {preview}{escalation}"
    )
}

/// Auto-escalation: when old_string is not found and the file was previously read
/// in a compressed mode, re-read in full and return the content so the agent
/// can immediately retry with the correct old_string.
fn auto_escalate_reread(cache: &mut SessionCache, path: &str) -> String {
    let entry = cache.get(path);
    let last_mode = entry.map(|e| e.last_mode.clone()).unwrap_or_default();

    if last_mode.is_empty() || last_mode == "full" {
        return String::new();
    }

    let Ok(fresh_content) = std::fs::read_to_string(path) else {
        return String::new();
    };
    cache.store(path, &fresh_content);
    cache.mark_full_delivered(path);

    let line_count = fresh_content.lines().count();
    const MAX_LINES: usize = 300;

    let content_preview = if line_count <= MAX_LINES {
        fresh_content
    } else {
        let lines: Vec<&str> = fresh_content.lines().collect();
        let head = &lines[..MAX_LINES / 2];
        let tail = &lines[line_count - MAX_LINES / 2..];
        let omitted = line_count - MAX_LINES;
        format!(
            "{}\n[... {omitted} lines omitted ...]\n{}",
            head.join("\n"),
            tail.join("\n")
        )
    };

    format!(
        "\n\n[auto-escalation] Last read used mode=\"{last_mode}\". \
         Full content ({line_count}L) below — retry edit with exact text from here:\n\n{content_preview}"
    )
}

fn do_replace(
    cache: &mut SessionCache,
    path: &Path,
    pre: &FilePreimage,
    params: &EditParams,
    cap: usize,
    args: &ReplaceArgs<'_>,
) -> String {
    if args.occurrences > 1 && !args.replace_all {
        return format!(
            "ERROR: old_string found {} times in {}. \
             Use replace_all=true to replace all, or provide more context to make old_string unique."
            ,
            args.occurrences,
            path.display()
        );
    }

    let new_content = if args.replace_all {
        args.content.replace(args.old_str, args.new_str)
    } else {
        args.content.replacen(args.old_str, args.new_str, 1)
    };

    if let Err(e) = ensure_preimage_still_matches(path, &pre.fp, cap) {
        return e;
    }

    let backup_path = if params.backup {
        let bp = params
            .backup_path
            .as_deref()
            .map(PathBuf::from)
            .or_else(|| default_backup_path(path));
        let Some(bp) = bp else {
            return format!("ERROR: cannot compute backup path for {}", path.display());
        };
        if let Err(e) = write_atomic_bytes_with_permissions(&bp, &pre.bytes, Some(&pre.permissions))
        {
            return format!("ERROR: cannot create backup {}: {e}", bp.display());
        }
        Some(bp.to_string_lossy().to_string())
    } else {
        None
    };

    if let Err(e) =
        write_atomic_bytes_with_permissions(path, new_content.as_bytes(), Some(&pre.permissions))
    {
        return e;
    }

    cache.invalidate(&params.path);

    if let Ok(mut bt) = crate::core::bounce_tracker::global().lock() {
        bt.record_edit(&params.path);
    }

    let old_lines = args.content.lines().count();
    let new_lines = new_content.lines().count();
    let line_delta = new_lines as i64 - old_lines as i64;
    let delta_str = if line_delta > 0 {
        format!("+{line_delta}")
    } else {
        format!("{line_delta}")
    };

    let old_tokens = args.old_tokens;
    let new_tokens = args.new_tokens;

    let replaced_str = if args.replace_all && args.occurrences > 1 {
        format!("{} replacements", args.occurrences)
    } else {
        "1 replacement".into()
    };

    let short = path.file_name().map_or_else(
        || path.to_string_lossy().to_string(),
        |f| f.to_string_lossy().to_string(),
    );

    let post_mtime_ms = std::fs::metadata(path)
        .ok()
        .and_then(|m| m.modified().ok())
        .map_or(0, system_time_to_millis);
    let post_fp = FileFingerprint {
        size: new_content.len() as u64,
        mtime_ms: post_mtime_ms,
        md5: crate::core::hasher::hash_hex(new_content.as_bytes()),
    };

    let mut out = format!(
        "✓ {short}: {replaced_str}, {delta_str} lines ({old_tokens}→{new_tokens} tok)\n\
preimage: bytes={}, mtime_ms={}, md5={}\n\
postimage: bytes={}, mtime_ms={}, md5={}",
        pre.fp.size, pre.fp.mtime_ms, pre.fp.md5, post_fp.size, post_fp.mtime_ms, post_fp.md5
    );
    if let Some(bp) = backup_path {
        out.push_str(&format!("\nbackup: {bp}"));
    }
    if params.evidence {
        let diff = build_diff_evidence(args.content, &new_content, &short, params.diff_max_lines);
        out.push_str("\n\nevidence (diff, redacted, bounded):\n```diff\n");
        out.push_str(&diff);
        out.push_str("\n```");
    }
    out
}

fn handle_create(
    cache: &mut SessionCache,
    file_path: &str,
    content: &str,
    params: &EditParams,
) -> String {
    let path = Path::new(file_path);
    let cap = crate::core::limits::max_read_bytes();

    let mut preimage: Option<FilePreimage> = None;
    if path.exists() {
        let pre = match read_preimage(path, cap, params.allow_lossy_utf8) {
            Ok(p) => p,
            Err(e) => return e,
        };
        if let Err(e) = verify_expected_preimage(&pre, params) {
            return e;
        }
        if let Err(e) = ensure_preimage_still_matches(path, &pre.fp, cap) {
            return e;
        }
        preimage = Some(pre);
    }

    if let Some(parent) = path.parent() {
        if !parent.exists() {
            if let Err(e) = std::fs::create_dir_all(parent) {
                return format!("ERROR: cannot create directory {}: {e}", parent.display());
            }
        }
    }

    let backup_path = if params.backup {
        if let Some(pre) = &preimage {
            let bp = params
                .backup_path
                .as_deref()
                .map(PathBuf::from)
                .or_else(|| default_backup_path(path));
            let Some(bp) = bp else {
                return format!("ERROR: cannot compute backup path for {}", path.display());
            };
            if let Err(e) =
                write_atomic_bytes_with_permissions(&bp, &pre.bytes, Some(&pre.permissions))
            {
                return format!("ERROR: cannot create backup {}: {e}", bp.display());
            }
            Some(bp.to_string_lossy().to_string())
        } else {
            None
        }
    } else {
        None
    };

    let perms = preimage.as_ref().map(|p| &p.permissions);
    if let Err(e) = write_atomic_bytes_with_permissions(path, content.as_bytes(), perms) {
        return e;
    }

    cache.invalidate(file_path);

    let lines = content.lines().count();
    let tokens = count_tokens(content);
    let short = path.file_name().map_or_else(
        || path.to_string_lossy().to_string(),
        |f| f.to_string_lossy().to_string(),
    );

    let mut out = format!("✓ created {short}: {lines} lines, {tokens} tok");
    if let Some(bp) = backup_path {
        out.push_str(&format!("\nbackup: {bp}"));
    }
    out
}

fn trim_trailing_per_line(s: &str) -> String {
    s.lines().map(str::trim_end).collect::<Vec<_>>().join("\n")
}

fn adapt_new_string_to_line_sep(s: &str, sep: &str) -> String {
    let normalized = s.replace("\r\n", "\n");
    if sep == "\r\n" {
        normalized.replace('\n', "\r\n")
    } else {
        normalized
    }
}

/// Find the original (un-trimmed) span in `content` that matches `normalized_needle`
/// after trailing-whitespace trimming per line.
fn find_original_span(content: &str, normalized_needle: &str) -> Option<String> {
    let needle_lines: Vec<&str> = normalized_needle.lines().collect();
    if needle_lines.is_empty() {
        return None;
    }

    let content_lines: Vec<&str> = content.lines().collect();

    'outer: for start in 0..content_lines.len() {
        if start + needle_lines.len() > content_lines.len() {
            break;
        }
        for (i, nl) in needle_lines.iter().enumerate() {
            if content_lines[start + i].trim_end() != *nl {
                continue 'outer;
            }
        }
        let sep = if content.contains("\r\n") {
            "\r\n"
        } else {
            "\n"
        };
        return Some(content_lines[start..start + needle_lines.len()].join(sep));
    }
    None
}

#[cfg(test)]
mod tests {
    use super::*;
    use std::io::Write;
    use tempfile::NamedTempFile;

    fn make_temp(content: &str) -> NamedTempFile {
        let mut f = NamedTempFile::new().unwrap();
        f.write_all(content.as_bytes()).unwrap();
        f
    }

    fn mk_params(path: &Path, old: &str, new: &str, replace_all: bool, create: bool) -> EditParams {
        EditParams {
            path: path.to_string_lossy().to_string(),
            old_string: old.to_string(),
            new_string: new.to_string(),
            replace_all,
            create,
            expected_md5: None,
            expected_size: None,
            expected_mtime_ms: None,
            backup: false,
            backup_path: None,
            evidence: false,
            diff_max_lines: 200,
            allow_lossy_utf8: false,
        }
    }

    #[test]
    fn replace_single_occurrence() {
        let f = make_temp("fn hello() {\n    println!(\"hello\");\n}\n");
        let mut cache = SessionCache::new();
        let result = handle(
            &mut cache,
            &mk_params(f.path(), "hello", "world", false, false),
        );
        assert!(result.contains("ERROR"), "should fail: 'hello' appears 2x");
    }

    #[test]
    fn replace_all() {
        let f = make_temp("aaa bbb aaa\n");
        let mut cache = SessionCache::new();
        let result = handle(&mut cache, &mk_params(f.path(), "aaa", "ccc", true, false));
        assert!(result.contains("2 replacements"));
        let content = std::fs::read_to_string(f.path()).unwrap();
        assert_eq!(content, "ccc bbb ccc\n");
    }

    #[test]
    fn not_found_error() {
        let f = make_temp("some content\n");
        let mut cache = SessionCache::new();
        let result = handle(
            &mut cache,
            &mk_params(f.path(), "nonexistent", "x", false, false),
        );
        assert!(result.contains("ERROR: old_string not found"));
    }

    #[test]
    fn create_new_file() {
        let dir = tempfile::tempdir().unwrap();
        let path = dir.path().join("sub/new_file.txt");
        let mut cache = SessionCache::new();
        let result = handle(
            &mut cache,
            &mk_params(&path, "", "line1\nline2\nline3\n", false, true),
        );
        assert!(result.contains("created new_file.txt"));
        assert!(result.contains("3 lines"));
        assert!(path.exists());
    }

    #[test]
    fn unique_match_succeeds() {
        let f = make_temp("fn main() {\n    let x = 42;\n}\n");
        let mut cache = SessionCache::new();
        let result = handle(
            &mut cache,
            &mk_params(f.path(), "let x = 42", "let x = 99", false, false),
        );
        assert!(result.contains("✓"));
        assert!(result.contains("1 replacement"));
        let content = std::fs::read_to_string(f.path()).unwrap();
        assert!(content.contains("let x = 99"));
    }

    #[test]
    fn crlf_file_with_lf_search() {
        let f = make_temp("line1\r\nline2\r\nline3\r\n");
        let mut cache = SessionCache::new();
        let result = handle(
            &mut cache,
            &mk_params(f.path(), "line1\nline2", "changed1\nchanged2", false, false),
        );
        assert!(result.contains("✓"), "CRLF fallback should work: {result}");
        let content = std::fs::read_to_string(f.path()).unwrap();
        assert!(
            content.contains("changed1\r\nchanged2"),
            "new_string should be adapted to CRLF: {content:?}"
        );
        assert!(
            content.contains("\r\nline3\r\n"),
            "rest of file should keep CRLF: {content:?}"
        );
    }

    #[test]
    fn lf_file_with_crlf_search() {
        let f = make_temp("line1\nline2\nline3\n");
        let mut cache = SessionCache::new();
        let result = handle(
            &mut cache,
            &mk_params(f.path(), "line1\r\nline2", "a\r\nb", false, false),
        );
        assert!(result.contains("✓"), "LF fallback should work: {result}");
        let content = std::fs::read_to_string(f.path()).unwrap();
        assert!(
            content.contains("a\nb"),
            "new_string should be adapted to LF: {content:?}"
        );
    }

    #[test]
    fn trailing_whitespace_tolerance() {
        let f = make_temp("  let x = 1;  \n  let y = 2;\n");
        let mut cache = SessionCache::new();
        let result = handle(
            &mut cache,
            &mk_params(
                f.path(),
                "  let x = 1;\n  let y = 2;",
                "  let x = 10;\n  let y = 20;",
                false,
                false,
            ),
        );
        assert!(
            result.contains("✓"),
            "trailing whitespace tolerance should work: {result}"
        );
        let content = std::fs::read_to_string(f.path()).unwrap();
        assert!(content.contains("let x = 10;"));
        assert!(content.contains("let y = 20;"));
    }

    #[test]
    fn crlf_with_trailing_whitespace() {
        let f = make_temp("  const a = 1;  \r\n  const b = 2;\r\n");
        let mut cache = SessionCache::new();
        let result = handle(
            &mut cache,
            &mk_params(
                f.path(),
                "  const a = 1;\n  const b = 2;",
                "  const a = 10;\n  const b = 20;",
                false,
                false,
            ),
        );
        assert!(
            result.contains("✓"),
            "CRLF + trailing whitespace should work: {result}"
        );
        let content = std::fs::read_to_string(f.path()).unwrap();
        assert!(content.contains("const a = 10;"));
        assert!(content.contains("const b = 20;"));
    }

    #[test]
    fn rejects_invalid_utf8_by_default() {
        let mut f = NamedTempFile::new().unwrap();
        f.write_all(&[0xff, 0xfe, 0xfd]).unwrap();
        let mut cache = SessionCache::new();
        let result = handle(&mut cache, &mk_params(f.path(), "a", "b", false, false));
        assert!(
            result.contains("not valid UTF-8"),
            "expected utf8 rejection, got: {result}"
        );
    }

    #[test]
    fn allows_lossy_utf8_only_when_enabled() {
        let mut f = NamedTempFile::new().unwrap();
        f.write_all(&[0xff, 0xfe, 0xfd]).unwrap();
        let mut cache = SessionCache::new();
        let mut p = mk_params(f.path(), "a", "b", false, false);
        p.allow_lossy_utf8 = true;
        let result = handle(&mut cache, &p);
        assert!(
            !result.contains("not valid UTF-8"),
            "lossy mode should avoid utf8 hard error, got: {result}"
        );
    }

    #[test]
    fn expected_md5_mismatch_fails_without_writing() {
        let f = make_temp("aaa\n");
        let mut cache = SessionCache::new();
        let mut p = mk_params(f.path(), "aaa", "bbb", false, false);
        p.expected_md5 = Some("deadbeef".to_string());
        let result = handle(&mut cache, &p);
        assert!(
            result.contains("preimage mismatch"),
            "expected preimage mismatch, got: {result}"
        );
        let content = std::fs::read_to_string(f.path()).unwrap();
        assert_eq!(content, "aaa\n");
    }

    #[test]
    fn backup_is_created_when_enabled() {
        let f = make_temp("aaa\n");
        let mut cache = SessionCache::new();
        let mut p = mk_params(f.path(), "aaa", "bbb", false, false);
        p.backup = true;
        let out = handle(&mut cache, &p);
        assert!(out.contains("backup:"), "expected backup path, got: {out}");
        let bp = out
            .lines()
            .find_map(|l| l.strip_prefix("backup: "))
            .expect("backup line");
        let backup_content = std::fs::read_to_string(bp).unwrap();
        assert_eq!(backup_content, "aaa\n");
        let content = std::fs::read_to_string(f.path()).unwrap();
        assert_eq!(content, "bbb\n");
    }

    #[test]
    fn evidence_diff_is_emitted_when_enabled() {
        let f = make_temp("line1\nline2\n");
        let mut cache = SessionCache::new();
        let mut p = mk_params(f.path(), "line2", "changed2", false, false);
        p.evidence = true;
        p.diff_max_lines = 50;
        let out = handle(&mut cache, &p);
        assert!(out.contains("```diff"), "expected diff fence, got: {out}");
        assert!(
            out.contains("preimage:"),
            "expected preimage metadata, got: {out}"
        );
        assert!(
            out.contains("postimage:"),
            "expected postimage metadata, got: {out}"
        );
    }

    #[test]
    fn detects_toctou_via_preimage_guard() {
        let f = make_temp("aaa\n");
        let cap = crate::core::limits::max_read_bytes();
        let pre = read_preimage(f.path(), cap, false).unwrap();
        std::fs::write(f.path(), "bbb\n").unwrap();
        let err = ensure_preimage_still_matches(f.path(), &pre.fp, cap).unwrap_err();
        assert!(err.contains("TOCTOU guard"), "unexpected error: {err}");
    }
}