recursive/tools/

apply_patch.rs

//! `apply_patch`: apply a structured multi-file patch atomically.
//!
//! The patch format is the **V4A** flavour popularised by Anthropic and used
//! by OpenAI's Codex CLI: no line numbers, context lines locate the change
//! in the file. This is materially more robust than unified diff for
//! LLM-authored edits — models reliably mis-count line numbers, but they
//! reliably preserve a few lines of context.
//!
//! ```text
//! *** Begin Patch
//! *** Update File: src/foo.rs
//! @@ optional anchor
//!  context line (unchanged, must match the file exactly)
//! -line to remove (must match the file exactly)
//! +line to add
//!  context line
//! *** Add File: new/file.rs
//! +line one
//! +line two
//! *** Delete File: doomed/file.rs
//! *** End Patch
//! ```
//!
//! Reliability strategy:
//!   - **Strict matching.** Context+remove lines must match the live file
//!     exactly. No fuzzy matching.
//!   - **Single-match.** If a hunk's pattern appears 0 or >1 times in the
//!     file, the patch is rejected (the model can add an `@@ anchor` for
//!     disambiguation or include more context).
//!   - **Atomic.** All hunks are applied in memory first; only after every
//!     file's final state is computed do we touch the disk. A failure in
//!     hunk N rolls the whole patch back; nothing on disk changes.
//!   - **Sandboxed.** Every path goes through `resolve_within`.
//!   - **Useful errors.** The tool returns structured diagnostics (which
//!     file, which hunk, what was being matched) so the model can self-correct.

use async_trait::async_trait;
use serde_json::{json, Value};
use std::collections::HashSet;
use std::path::PathBuf;

use super::{resolve_within, Tool};
use crate::error::{Error, Result};
use crate::llm::ToolSpec;

const BEGIN: &str = "*** Begin Patch";
const END: &str = "*** End Patch";
const UPDATE: &str = "*** Update File: ";
const ADD: &str = "*** Add File: ";
const DELETE: &str = "*** Delete File: ";
const HUNK_SEP: &str = "@@";

/// Normalize a unified-diff-style hunk header to a V4A anchor.
///
/// Unified-diff headers look like `@@ -14,6 +14,28 @@ anchor text` or
/// `@@ -14 +14 @@ anchor text`. The line-number range is discarded; only
/// the trailing text after the second `@@` is returned as the anchor.
///
/// Returns `None` when the trailing text is empty or whitespace-only.
fn normalize_hunk_header(rest: &str) -> Option<String> {
    let trimmed = rest.trim();
    if trimmed.is_empty() {
        return None;
    }

    // Check for unified-diff pattern: `-N[,M] +N[,M] @@ [trailing]`
    // We scan manually (no regex crate).
    let bytes = trimmed.as_bytes();
    let len = bytes.len();
    let mut i = 0;

    // Skip leading whitespace.
    while i < len && bytes[i].is_ascii_whitespace() {
        i += 1;
    }

    // Must start with '-'.
    if i < len && bytes[i] == b'-' {
        i += 1;
        // Consume digits and optional comma+digits (the "old" range).
        let mut has_digits = false;
        while i < len && bytes[i].is_ascii_digit() {
            has_digits = true;
            i += 1;
        }
        if i < len && bytes[i] == b',' {
            i += 1;
            while i < len && bytes[i].is_ascii_digit() {
                has_digits = true;
                i += 1;
            }
        }
        if !has_digits {
            // Not a unified-diff header after all; return whole trimmed.
            return Some(trimmed.to_string());
        }

        // Skip whitespace.
        while i < len && bytes[i].is_ascii_whitespace() {
            i += 1;
        }

        // Must have '+'.
        if i < len && bytes[i] == b'+' {
            i += 1;
            // Consume digits and optional comma+digits (the "new" range).
            let mut has_digits2 = false;
            while i < len && bytes[i].is_ascii_digit() {
                has_digits2 = true;
                i += 1;
            }
            if i < len && bytes[i] == b',' {
                i += 1;
                while i < len && bytes[i].is_ascii_digit() {
                    has_digits2 = true;
                    i += 1;
                }
            }
            if !has_digits2 {
                return Some(trimmed.to_string());
            }

            // Skip whitespace.
            while i < len && bytes[i].is_ascii_whitespace() {
                i += 1;
            }

            // Must have '@@'.
            if i + 1 < len && bytes[i] == b'@' && bytes[i + 1] == b'@' {
                i += 2;
                // Skip whitespace after the second @@.
                while i < len && bytes[i].is_ascii_whitespace() {
                    i += 1;
                }
                let trailing = trimmed[i..].trim().to_string();
                if trailing.is_empty() {
                    return None;
                }
                return Some(trailing);
            }
        }
    }

    // Not a unified-diff header; return the whole trimmed string.
    Some(trimmed.to_string())
}

// ---- Tool ------------------------------------------------------------------

#[derive(Debug, Clone)]
pub struct ApplyPatch {
    pub root: PathBuf,
}

impl ApplyPatch {
    pub fn new(root: impl Into<PathBuf>) -> Self {
        Self { root: root.into() }
    }
}

#[async_trait]
impl Tool for ApplyPatch {
    fn spec(&self) -> ToolSpec {
        ToolSpec {
            name: "apply_patch".into(),
            description: concat!(
                "Apply a V4A-format patch atomically. Format:\n",
                "*** Begin Patch\n",
                "*** Update File: <path>\n",
                "[@@ optional_anchor_line]\n",
                " context_line\n",
                "-line_to_remove\n",
                "+line_to_add\n",
                " context_line\n",
                "*** Add File: <path>\n",
                "+line_one\n",
                "+line_two\n",
                "*** Delete File: <path>\n",
                "*** End Patch\n",
                "Context lines must match the file exactly. If a hunk's pattern ",
                "appears multiple times in the file, add an @@ anchor (some unique ",
                "line that appears earlier in the file) before the hunk. ",
                "Both `@@ <unique_line>` (V4A) and ",
                "`@@ -N,M +N,M @@ <unique_line>` (unified-diff style) headers are ",
                "accepted; the line-number range is ignored, only the anchor text ",
                "after the final `@@` is used to locate the hunk."
            )
            .into(),
            parameters: json!({
                "type": "object",
                "properties": {
                    "patch": {"type": "string", "description": "The V4A patch text"},
                    "dry_run": {"type": "boolean", "description": "If true, validate and resolve the patch but do not write any files. Returns the same success/error message it would produce in a normal run.", "default": false}
                },
                "required": ["patch"]
            }),
        }
    }

    fn side_effect_class(&self) -> crate::tools::ToolSideEffect {
        crate::tools::ToolSideEffect::Mutating
    }

    async fn execute(&self, args: Value) -> Result<String> {
        let input = args["patch"].as_str().ok_or_else(|| Error::BadToolArgs {
            name: "apply_patch".into(),
            message: "missing `patch`".into(),
        })?;
        let dry_run = args
            .get("dry_run")
            .and_then(|v| v.as_bool())
            .unwrap_or(false);
        let patch = parse_patch(input).map_err(|e| Error::Tool {
            name: "apply_patch".into(),
            message: e,
        })?;
        let writes = stage(&patch, &self.root).await.map_err(|e| Error::Tool {
            name: "apply_patch".into(),
            message: e,
        })?;
        if dry_run {
            let file_count = writes.len();
            let hunk_count: usize = patch
                .ops
                .iter()
                .map(|op| match op {
                    FileOp::Update { hunks, .. } => hunks.len(),
                    _ => 1,
                })
                .sum();
            let paths: Vec<&str> = writes.iter().map(|w| w.rel_path.as_str()).collect();
            return Ok(format!(
                "dry-run: would apply {} hunk(s) across {} file(s): {}",
                hunk_count,
                file_count,
                paths.join(", ")
            ));
        }
        commit(writes).await.map_err(|e| Error::Tool {
            name: "apply_patch".into(),
            message: e,
        })
    }
}

// ---- AST -------------------------------------------------------------------

#[derive(Debug, Clone, PartialEq, Eq)]
pub(crate) enum HunkLine {
    Context(String),
    Add(String),
    Remove(String),
}

#[derive(Debug, Clone, PartialEq, Eq)]
pub(crate) struct Hunk {
    pub anchor: Option<String>,
    pub lines: Vec<HunkLine>,
}

#[derive(Debug, Clone, PartialEq, Eq)]
pub(crate) enum FileOp {
    Update { path: String, hunks: Vec<Hunk> },
    Add { path: String, content: String },
    Delete { path: String },
}

#[derive(Debug, Clone, Default, PartialEq, Eq)]
pub(crate) struct Patch {
    pub ops: Vec<FileOp>,
}

// ---- Parser ----------------------------------------------------------------

pub(crate) fn parse_patch(input: &str) -> std::result::Result<Patch, String> {
    let mut lines = input.lines().peekable();

    // Skip leading blank lines.
    while matches!(lines.peek(), Some(l) if l.trim().is_empty()) {
        lines.next();
    }

    let begin = lines.next().ok_or("empty patch")?;
    if begin.trim() != BEGIN {
        return Err(format!("first line must be `{BEGIN}`, got `{begin}`"));
    }

    let mut patch = Patch::default();
    let mut current_file: Option<FileOp> = None;

    loop {
        let Some(line) = lines.next() else {
            return Err(format!("patch terminated without `{END}`"));
        };

        if line.trim() == END {
            if let Some(op) = current_file.take() {
                patch.ops.push(op);
            }
            break;
        }

        // File header? Flush whatever we were building and start a new op.
        if let Some(path) = line.strip_prefix(UPDATE) {
            if let Some(op) = current_file.take() {
                patch.ops.push(op);
            }
            let path = path.trim().to_string();
            if path.is_empty() {
                return Err(format!("`{UPDATE}` with empty path"));
            }
            current_file = Some(FileOp::Update {
                path,
                hunks: vec![Hunk {
                    anchor: None,
                    lines: vec![],
                }],
            });
            continue;
        }
        if let Some(path) = line.strip_prefix(ADD) {
            if let Some(op) = current_file.take() {
                patch.ops.push(op);
            }
            let path = path.trim().to_string();
            if path.is_empty() {
                return Err(format!("`{ADD}` with empty path"));
            }
            current_file = Some(FileOp::Add {
                path,
                content: String::new(),
            });
            continue;
        }
        if let Some(path) = line.strip_prefix(DELETE) {
            if let Some(op) = current_file.take() {
                patch.ops.push(op);
            }
            let path = path.trim().to_string();
            if path.is_empty() {
                return Err(format!("`{DELETE}` with empty path"));
            }
            current_file = Some(FileOp::Delete { path });
            continue;
        }

        // Body of the current file.
        let op = current_file.as_mut().ok_or_else(|| {
            format!("body line `{line}` before any `*** Update/Add/Delete File:` header")
        })?;

        match op {
            FileOp::Update { hunks, .. } => {
                if let Some(rest) = line.strip_prefix(HUNK_SEP) {
                    // New hunk; first hunk is pre-allocated above, so we only push if the current is non-empty.
                    let anchor = normalize_hunk_header(rest);
                    let last = hunks.last_mut().unwrap();
                    if last.lines.is_empty() {
                        last.anchor = anchor;
                    } else {
                        hunks.push(Hunk {
                            anchor,
                            lines: vec![],
                        });
                    }
                    continue;
                }

                let parsed = if let Some(rest) = line.strip_prefix('-') {
                    HunkLine::Remove(rest.to_string())
                } else if let Some(rest) = line.strip_prefix('+') {
                    HunkLine::Add(rest.to_string())
                } else if let Some(rest) = line.strip_prefix(' ') {
                    HunkLine::Context(rest.to_string())
                } else if line.is_empty() {
                    // Unprefixed blank lines are treated as a blank context line.
                    // (Models sometimes emit `\n` instead of ` \n`.)
                    HunkLine::Context(String::new())
                } else {
                    return Err(format!("unexpected line in Update hunk: `{line}`"));
                };
                hunks.last_mut().unwrap().lines.push(parsed);
            }
            FileOp::Add { content, .. } => {
                let body_line = if let Some(rest) = line.strip_prefix('+') {
                    rest
                } else if line.is_empty() {
                    ""
                } else {
                    return Err(format!(
                        "Add File body lines must start with `+`, got `{line}`"
                    ));
                };
                if !content.is_empty() {
                    content.push('\n');
                }
                content.push_str(body_line);
            }
            FileOp::Delete { .. } => {
                // No body allowed.
                if !line.trim().is_empty() {
                    return Err(format!("Delete File body must be empty, got `{line}`"));
                }
            }
        }
    }

    // Validation: paths unique, hunks non-empty for Update.
    let mut seen: HashSet<&String> = HashSet::new();
    for op in &patch.ops {
        let path = match op {
            FileOp::Update { path, .. } | FileOp::Add { path, .. } | FileOp::Delete { path } => {
                path
            }
        };
        if !seen.insert(path) {
            return Err(format!("path `{path}` appears in patch more than once"));
        }
        if let FileOp::Update { hunks, .. } = op {
            if hunks.iter().all(|h| h.lines.is_empty()) {
                return Err(format!("Update File `{path}` has no hunks"));
            }
            for (i, h) in hunks.iter().enumerate() {
                if h.lines.is_empty() {
                    return Err(format!("Update File `{path}` hunk {} is empty", i + 1));
                }
                if !h
                    .lines
                    .iter()
                    .any(|l| matches!(l, HunkLine::Add(_) | HunkLine::Remove(_)))
                {
                    return Err(format!(
                        "Update File `{path}` hunk {} has no +/- lines",
                        i + 1
                    ));
                }
            }
        }
    }

    Ok(patch)
}

// ---- Applier ---------------------------------------------------------------

struct StagedWrite {
    abs_path: PathBuf,
    rel_path: String,
    kind: StagedKind,
}

enum StagedKind {
    WriteText(String),
    Delete,
}

async fn stage(
    patch: &Patch,
    root: &std::path::Path,
) -> std::result::Result<Vec<StagedWrite>, String> {
    let mut staged = Vec::new();
    for op in &patch.ops {
        let path = match op {
            FileOp::Update { path, .. } | FileOp::Add { path, .. } | FileOp::Delete { path } => {
                path
            }
        };
        let abs = resolve_within(root, path).map_err(|e| e.to_string())?;

        match op {
            FileOp::Update { hunks, .. } => {
                let current = tokio::fs::read_to_string(&abs)
                    .await
                    .map_err(|e| format!("update `{path}`: {e}"))?;
                let updated =
                    apply_hunks(&current, hunks).map_err(|e| format!("update `{path}`: {e}"))?;
                staged.push(StagedWrite {
                    abs_path: abs,
                    rel_path: path.clone(),
                    kind: StagedKind::WriteText(updated),
                });
            }
            FileOp::Add { content, .. } => {
                if abs.exists() {
                    return Err(format!("add `{path}`: file already exists"));
                }
                staged.push(StagedWrite {
                    abs_path: abs,
                    rel_path: path.clone(),
                    kind: StagedKind::WriteText(content.clone()),
                });
            }
            FileOp::Delete { .. } => {
                if !abs.exists() {
                    return Err(format!("delete `{path}`: file does not exist"));
                }
                staged.push(StagedWrite {
                    abs_path: abs,
                    rel_path: path.clone(),
                    kind: StagedKind::Delete,
                });
            }
        }
    }
    Ok(staged)
}

async fn commit(staged: Vec<StagedWrite>) -> std::result::Result<String, String> {
    let mut summary = Vec::new();
    for w in &staged {
        match &w.kind {
            StagedKind::WriteText(content) => {
                if let Some(parent) = w.abs_path.parent() {
                    tokio::fs::create_dir_all(parent)
                        .await
                        .map_err(|e| format!("mkdir for `{}`: {e}", w.rel_path))?;
                }
                tokio::fs::write(&w.abs_path, content)
                    .await
                    .map_err(|e| format!("write `{}`: {e}", w.rel_path))?;
                summary.push(format!("wrote {} ({} bytes)", w.rel_path, content.len()));
            }
            StagedKind::Delete => {
                tokio::fs::remove_file(&w.abs_path)
                    .await
                    .map_err(|e| format!("delete `{}`: {e}", w.rel_path))?;
                summary.push(format!("deleted {}", w.rel_path));
            }
        }
    }
    Ok(format!(
        "applied {} change(s):\n{}",
        staged.len(),
        summary.join("\n")
    ))
}

/// Find up to 3 unique lines near the match location that could serve as anchors.
/// We look 5 lines before and 5 lines after the first matched location,
/// and keep only lines that appear exactly once in the whole file.
fn find_unique_anchor_suggestions(
    lines: &[String],
    match_start: usize,
    pattern: &[&str],
) -> Vec<String> {
    let mut suggestions = Vec::new();
    let window = 5;

    // Collect candidate lines: 5 before through 5 after the match location
    let start = match_start.saturating_sub(window);
    let end = (match_start + pattern.len() + window).min(lines.len());

    for i in start..end {
        let line = &lines[i];
        if line.is_empty() {
            continue;
        }

        // Count occurrences of this exact line in the whole file
        let count = lines.iter().filter(|l| l.as_str() == line.as_str()).count();
        if count == 1 {
            suggestions.push(line.clone());
        }

        if suggestions.len() >= 3 {
            break;
        }
    }

    suggestions
}

fn apply_hunks(content: &str, hunks: &[Hunk]) -> std::result::Result<String, String> {
    let trailing_newline = content.ends_with('\n');
    let mut lines: Vec<String> = content.lines().map(|s| s.to_string()).collect();

    for (idx, hunk) in hunks.iter().enumerate() {
        // Pattern: context + remove lines in order. Replacement: context + add.
        let pattern: Vec<&str> = hunk
            .lines
            .iter()
            .filter_map(|l| match l {
                HunkLine::Context(s) | HunkLine::Remove(s) => Some(s.as_str()),
                HunkLine::Add(_) => None,
            })
            .collect();

        let replacement: Vec<String> = hunk
            .lines
            .iter()
            .filter_map(|l| match l {
                HunkLine::Context(s) | HunkLine::Add(s) => Some(s.clone()),
                HunkLine::Remove(_) => None,
            })
            .collect();

        if pattern.is_empty() {
            // Pure-add hunk with no anchor: prepend to file. Discouraged but allowed.
            // We only allow this if the file is currently empty AND there's no anchor.
            if !lines.is_empty() || hunk.anchor.is_some() {
                return Err(format!(
                    "hunk {} has no context/remove lines; need at least one for matching",
                    idx + 1
                ));
            }
            lines.extend(replacement);
            continue;
        }

        // Find all positions where pattern matches.
        let mut matches: Vec<usize> = Vec::new();
        if pattern.len() <= lines.len() {
            for start in 0..=(lines.len() - pattern.len()) {
                if lines[start..start + pattern.len()]
                    .iter()
                    .zip(pattern.iter())
                    .all(|(actual, expected)| actual == expected)
                {
                    matches.push(start);
                }
            }
        }

        // Anchor filtering: the anchor text must appear in some line at or
        // before the END of the matched region. Spec'd anchors sit *above*
        // the hunk, but models sometimes inline them into the first context
        // line — both should disambiguate equally well.
        if let Some(anchor) = &hunk.anchor {
            matches.retain(|&start| {
                let end = start + pattern.len();
                lines[..end].iter().any(|l| l.contains(anchor.as_str()))
            });
        }

        let pos = match matches.len() {
            0 => {
                return Err(format!(
                    "hunk {} pattern not found{}.\nFirst 3 pattern lines:\n{}",
                    idx + 1,
                    hunk.anchor
                        .as_deref()
                        .map(|a| format!(" (anchor `{a}`)"))
                        .unwrap_or_default(),
                    pattern
                        .iter()
                        .take(3)
                        .map(|l| format!("    {l}"))
                        .collect::<Vec<_>>()
                        .join("\n")
                ));
            }
            1 => matches[0],
            n => {
                // Find unique anchor suggestions around the first match location
                let suggestions = find_unique_anchor_suggestions(&lines, matches[0], &pattern);

                let suggestion_text = if suggestions.is_empty() {
                    String::new()
                } else {
                    format!(
                        "\nAdd an `@@ <anchor>` line above the hunk with one of these unique nearby lines:\n{}",
                        suggestions
                            .iter()
                            .map(|s| format!("  @@ {}", s))
                            .collect::<Vec<_>>()
                            .join("\n")
                    )
                };

                return Err(format!(
                    "hunk {} pattern matches {} locations; add an `@@ anchor` line above the hunk to disambiguate{}",
                    idx + 1, n, suggestion_text
                ));
            }
        };

        let end = pos + pattern.len();
        lines.splice(pos..end, replacement);
    }

    let mut result = lines.join("\n");
    if trailing_newline {
        result.push('\n');
    }
    Ok(result)
}

// ---- Tests -----------------------------------------------------------------

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

    fn p(input: &str) -> Patch {
        parse_patch(input).expect("parse")
    }

    // -- normalize_hunk_header ------------------------------------------

    #[test]
    fn parses_v4a_anchor() {
        assert_eq!(normalize_hunk_header("fn foo"), Some("fn foo".to_string()));
    }

    #[test]
    fn parses_unified_header_with_anchor() {
        assert_eq!(
            normalize_hunk_header("-10,5 +10,7 @@ pub use foo;"),
            Some("pub use foo;".to_string())
        );
    }

    #[test]
    fn parses_unified_header_without_anchor() {
        assert_eq!(normalize_hunk_header("-10,5 +10,7 @@"), None);
    }

    #[test]
    fn parses_unified_header_singular_counts() {
        assert_eq!(
            normalize_hunk_header("-10 +10 @@ fn bar"),
            Some("fn bar".to_string())
        );
    }

    #[test]
    fn normalize_empty_returns_none() {
        assert_eq!(normalize_hunk_header(""), None);
        assert_eq!(normalize_hunk_header("  "), None);
    }

    #[test]
    fn normalize_non_unified_returns_whole() {
        assert_eq!(
            normalize_hunk_header("some random text"),
            Some("some random text".to_string())
        );
    }

    #[test]
    fn normalize_unified_with_extra_whitespace() {
        assert_eq!(
            normalize_hunk_header("  -14,6 +14,28 @@  pub use mock::MockProvider;  "),
            Some("pub use mock::MockProvider;".to_string())
        );
    }

    // -- parser ----------------------------------------------------------

    #[test]
    fn parse_update_single_hunk() {
        let patch = p("\
*** Begin Patch
*** Update File: a.txt
 keep
-old
+new
*** End Patch
");
        assert_eq!(patch.ops.len(), 1);
        match &patch.ops[0] {
            FileOp::Update { path, hunks } => {
                assert_eq!(path, "a.txt");
                assert_eq!(hunks.len(), 1);
                assert_eq!(hunks[0].anchor, None);
                assert_eq!(
                    hunks[0].lines,
                    vec![
                        HunkLine::Context("keep".into()),
                        HunkLine::Remove("old".into()),
                        HunkLine::Add("new".into()),
                    ]
                );
            }
            _ => panic!("wrong op"),
        }
    }

    #[test]
    fn parse_multi_hunk_with_anchors() {
        let patch = p("\
*** Begin Patch
*** Update File: a.txt
@@ fn foo
 a
-b
+B
@@ fn bar
 c
-d
+D
*** End Patch
");
        match &patch.ops[0] {
            FileOp::Update { hunks, .. } => {
                assert_eq!(hunks.len(), 2);
                assert_eq!(hunks[0].anchor.as_deref(), Some("fn foo"));
                assert_eq!(hunks[1].anchor.as_deref(), Some("fn bar"));
            }
            _ => panic!(),
        }
    }

    #[test]
    fn parse_unified_header_hunk() {
        // A hunk header in unified-diff style should be accepted.
        let patch = p("\
*** Begin Patch
*** Update File: a.txt
@@ -10,5 +10,7 @@ pub use foo;
 a
-b
+B
*** End Patch
");
        match &patch.ops[0] {
            FileOp::Update { hunks, .. } => {
                assert_eq!(hunks.len(), 1);
                assert_eq!(hunks[0].anchor.as_deref(), Some("pub use foo;"));
            }
            _ => panic!(),
        }
    }

    #[test]
    fn parse_unified_header_no_anchor() {
        let patch = p("\
*** Begin Patch
*** Update File: a.txt
@@ -10,5 +10,7 @@
 a
-b
+B
*** End Patch
");
        match &patch.ops[0] {
            FileOp::Update { hunks, .. } => {
                assert_eq!(hunks.len(), 1);
                assert_eq!(hunks[0].anchor, None);
            }
            _ => panic!(),
        }
    }

    #[test]
    fn parse_add_file_collects_plus_lines() {
        let patch = p("\
*** Begin Patch
*** Add File: new.txt
+hello
+world
*** End Patch
");
        match &patch.ops[0] {
            FileOp::Add { path, content } => {
                assert_eq!(path, "new.txt");
                assert_eq!(content, "hello\nworld");
            }
            _ => panic!(),
        }
    }

    #[test]
    fn parse_delete_file() {
        let patch = p("\
*** Begin Patch
*** Delete File: doomed.txt
*** End Patch
");
        assert!(matches!(&patch.ops[0], FileOp::Delete { path, .. } if path == "doomed.txt"));
    }

    #[test]
    fn parse_rejects_missing_begin() {
        let e = parse_patch("hello\n*** End Patch\n").unwrap_err();
        assert!(e.contains("first line"));
    }

    #[test]
    fn parse_rejects_missing_end() {
        let e = parse_patch("*** Begin Patch\n*** Update File: a\n keep\n").unwrap_err();
        assert!(e.contains("terminated"));
    }

    #[test]
    fn parse_rejects_duplicate_path() {
        let e = parse_patch(
            "\
*** Begin Patch
*** Add File: a.txt
+x
*** Delete File: a.txt
*** End Patch
",
        )
        .unwrap_err();
        assert!(e.contains("more than once"));
    }

    #[test]
    fn parse_rejects_body_without_header() {
        let e = parse_patch("*** Begin Patch\n+stray\n*** End Patch\n").unwrap_err();
        assert!(e.contains("before any"));
    }

    #[test]
    fn parse_rejects_hunk_without_changes() {
        let e = parse_patch(
            "\
*** Begin Patch
*** Update File: a
 just
 context
*** End Patch
",
        )
        .unwrap_err();
        assert!(e.contains("no +/- lines"));
    }

    // -- applier (in-memory) --------------------------------------------

    #[test]
    fn applies_single_hunk_round_trip() {
        let original = "line1\nold\nline3\n";
        let patch = p("\
*** Begin Patch
*** Update File: f
 line1
-old
+new
 line3
*** End Patch
");
        let FileOp::Update { hunks, .. } = &patch.ops[0] else {
            panic!()
        };
        let updated = apply_hunks(original, hunks).unwrap();
        assert_eq!(updated, "line1\nnew\nline3\n");
    }

    #[test]
    fn applies_multi_hunk() {
        let original = "alpha\nbeta\ngamma\ndelta\n";
        let hunks = vec![
            Hunk {
                anchor: None,
                lines: vec![
                    HunkLine::Context("alpha".into()),
                    HunkLine::Remove("beta".into()),
                    HunkLine::Add("BETA".into()),
                ],
            },
            Hunk {
                anchor: None,
                lines: vec![
                    HunkLine::Context("gamma".into()),
                    HunkLine::Remove("delta".into()),
                    HunkLine::Add("DELTA".into()),
                ],
            },
        ];
        assert_eq!(
            apply_hunks(original, &hunks).unwrap(),
            "alpha\nBETA\ngamma\nDELTA\n"
        );
    }

    #[test]
    fn errors_when_pattern_not_found() {
        let e = apply_hunks(
            "hello\nworld\n",
            &[Hunk {
                anchor: None,
                lines: vec![
                    HunkLine::Context("foo".into()),
                    HunkLine::Remove("bar".into()),
                    HunkLine::Add("baz".into()),
                ],
            }],
        )
        .unwrap_err();
        assert!(e.contains("not found"));
    }

    #[test]
    fn errors_when_pattern_ambiguous() {
        let original = "x\ny\nx\ny\n";
        let e = apply_hunks(
            original,
            &[Hunk {
                anchor: None,
                lines: vec![
                    HunkLine::Context("x".into()),
                    HunkLine::Remove("y".into()),
                    HunkLine::Add("Y".into()),
                ],
            }],
        )
        .unwrap_err();
        assert!(e.contains("matches"));
    }

    #[test]
    fn ambiguous_error_shows_unique_anchor_suggestions() {
        // File with duplicated context and unique function definitions nearby
        let original = "UNIQUE_FUNC_A()\n    x\n    y\n}\nUNIQUE_FUNC_B()\n    x\n    y\n}\nUNIQUE_FUNC_C() {}\n";
        let hunks = vec![Hunk {
            anchor: None,
            lines: vec![
                HunkLine::Context("    x".into()),
                HunkLine::Remove("    y".into()),
                HunkLine::Add("    Y".into()),
            ],
        }];
        let e = apply_hunks(original, &hunks).unwrap_err();

        // Should mention multiple matches
        assert!(e.contains("matches"));
        // Should contain anchor suggestion section
        assert!(e.contains("@@ <anchor>"));
        // Should contain at least one unique suggestion that exists in the original
        assert!(
            e.contains("UNIQUE_FUNC_A")
                || e.contains("UNIQUE_FUNC_B")
                || e.contains("UNIQUE_FUNC_C")
        );
    }

    #[test]
    fn anchor_disambiguates_repeated_context() {
        // Two regions, same shape, anchor picks the second.
        let original = "fn foo {\n  x\n  y\n}\nfn bar {\n  x\n  y\n}\n";
        let hunks = vec![Hunk {
            anchor: Some("fn bar".into()),
            lines: vec![
                HunkLine::Context("  x".into()),
                HunkLine::Remove("  y".into()),
                HunkLine::Add("  Y".into()),
            ],
        }];
        let out = apply_hunks(original, &hunks).unwrap();
        assert_eq!(out, "fn foo {\n  x\n  y\n}\nfn bar {\n  x\n  Y\n}\n");
    }

    #[test]
    fn preserves_no_trailing_newline() {
        let original = "a\nb";
        let hunks = vec![Hunk {
            anchor: None,
            lines: vec![
                HunkLine::Remove("a".into()),
                HunkLine::Add("A".into()),
                HunkLine::Context("b".into()),
            ],
        }];
        assert_eq!(apply_hunks(original, &hunks).unwrap(), "A\nb");
    }

    // -- end-to-end (tool) ---------------------------------------------

    #[tokio::test]
    async fn tool_updates_existing_file() {
        let tmp = TempDir::new().unwrap();
        std::fs::write(tmp.path().join("hello.txt"), "before\n").unwrap();
        let tool = ApplyPatch::new(tmp.path());
        let result = tool
            .execute(json!({"patch": "\
*** Begin Patch
*** Update File: hello.txt
-before
+after
*** End Patch
"}))
            .await
            .unwrap();
        assert!(result.contains("wrote hello.txt"));
        assert_eq!(
            std::fs::read_to_string(tmp.path().join("hello.txt")).unwrap(),
            "after\n"
        );
    }

    #[tokio::test]
    async fn tool_adds_new_file() {
        let tmp = TempDir::new().unwrap();
        let tool = ApplyPatch::new(tmp.path());
        tool.execute(json!({"patch": "\
*** Begin Patch
*** Add File: sub/new.txt
+hi
+there
*** End Patch
"}))
            .await
            .unwrap();
        assert_eq!(
            std::fs::read_to_string(tmp.path().join("sub/new.txt")).unwrap(),
            "hi\nthere"
        );
    }

    #[tokio::test]
    async fn tool_deletes_existing_file() {
        let tmp = TempDir::new().unwrap();
        std::fs::write(tmp.path().join("doomed.txt"), "x").unwrap();
        ApplyPatch::new(tmp.path())
            .execute(json!({"patch": "\
*** Begin Patch
*** Delete File: doomed.txt
*** End Patch
"}))
            .await
            .unwrap();
        assert!(!tmp.path().join("doomed.txt").exists());
    }

    #[tokio::test]
    async fn tool_is_atomic_on_failure() {
        // Two-file patch: first hunk succeeds (would update a.txt), second fails.
        // After apply, neither file should be modified.
        let tmp = TempDir::new().unwrap();
        std::fs::write(tmp.path().join("a.txt"), "original-a\n").unwrap();
        std::fs::write(tmp.path().join("b.txt"), "original-b\n").unwrap();
        let err = ApplyPatch::new(tmp.path())
            .execute(json!({"patch": "\
*** Begin Patch
*** Update File: a.txt
-original-a
+changed-a
*** Update File: b.txt
-nonexistent-line
+changed-b
*** End Patch
"}))
            .await
            .unwrap_err();
        assert!(matches!(err, Error::Tool { .. }));
        assert_eq!(
            std::fs::read_to_string(tmp.path().join("a.txt")).unwrap(),
            "original-a\n"
        );
        assert_eq!(
            std::fs::read_to_string(tmp.path().join("b.txt")).unwrap(),
            "original-b\n"
        );
    }

    #[tokio::test]
    async fn tool_rejects_add_when_file_exists() {
        let tmp = TempDir::new().unwrap();
        std::fs::write(tmp.path().join("exists.txt"), "x").unwrap();
        let err = ApplyPatch::new(tmp.path())
            .execute(json!({"patch": "\
*** Begin Patch
*** Add File: exists.txt
+content
*** End Patch
"}))
            .await
            .unwrap_err();
        assert!(matches!(err, Error::Tool { .. }));
    }

    #[tokio::test]
    async fn tool_rejects_delete_when_file_missing() {
        let tmp = TempDir::new().unwrap();
        let err = ApplyPatch::new(tmp.path())
            .execute(json!({"patch": "\
*** Begin Patch
*** Delete File: ghost.txt
*** End Patch
"}))
            .await
            .unwrap_err();
        assert!(matches!(err, Error::Tool { .. }));
    }

    #[tokio::test]
    async fn tool_rejects_sandbox_escape() {
        let tmp = TempDir::new().unwrap();
        let err = ApplyPatch::new(tmp.path())
            .execute(json!({"patch": "\
*** Begin Patch
*** Add File: ../escape.txt
+evil
*** End Patch
"}))
            .await
            .unwrap_err();
        // Wrapped as a Tool error (resolve_within returned BadToolArgs internally,
        // but we re-wrapped it). Just check it failed.
        assert!(matches!(err, Error::Tool { .. }));
        assert!(!tmp.path().parent().unwrap().join("escape.txt").exists());
    }

    #[tokio::test]
    async fn tool_round_trips_a_real_change() {
        let tmp = TempDir::new().unwrap();
        std::fs::write(
            tmp.path().join("lib.rs"),
            "\
fn add(a: i32, b: i32) -> i32 {
    a + b
}

fn sub(a: i32, b: i32) -> i32 {
    a - b
}
",
        )
        .unwrap();
        let tool = ApplyPatch::new(tmp.path());
        tool.execute(json!({"patch": "\
*** Begin Patch
*** Update File: lib.rs
@@ fn sub
 fn sub(a: i32, b: i32) -> i32 {
-    a - b
+    a.saturating_sub(b)
 }
*** End Patch
"}))
            .await
            .unwrap();
        let got = std::fs::read_to_string(tmp.path().join("lib.rs")).unwrap();
        assert!(got.contains("a.saturating_sub(b)"));
        assert!(got.contains("a + b"), "add() unchanged");
    }

    #[tokio::test]
    async fn tool_applies_patch_with_unified_header() {
        // End-to-end test using a unified-diff-style header.
        let tmp = TempDir::new().unwrap();
        std::fs::write(tmp.path().join("greeting.txt"), "hello\nworld\n").unwrap();
        let tool = ApplyPatch::new(tmp.path());
        tool.execute(json!({"patch": "\
*** Begin Patch
*** Update File: greeting.txt
@@ -1,3 +1,4 @@ hello
 hello
-world
+earth
*** End Patch
"}))
            .await
            .unwrap();
        let got = std::fs::read_to_string(tmp.path().join("greeting.txt")).unwrap();
        assert_eq!(got, "hello\nearth\n");
    }

    #[tokio::test]
    async fn dry_run_returns_summary_and_does_not_write() {
        let tmp = TempDir::new().unwrap();
        std::fs::write(tmp.path().join("target.txt"), "before\n").unwrap();
        let tool = ApplyPatch::new(tmp.path());
        let result = tool
            .execute(json!({"patch": "\
*** Begin Patch
*** Update File: target.txt
-before
+after
*** End Patch
", "dry_run": true}))
            .await
            .unwrap();
        assert!(result.contains("dry-run: would apply"));
        assert!(result.contains("target.txt"));
        // File must be unchanged
        assert_eq!(
            std::fs::read_to_string(tmp.path().join("target.txt")).unwrap(),
            "before\n"
        );
    }

    #[tokio::test]
    async fn dry_run_errors_on_ambiguous_anchor() {
        let tmp = TempDir::new().unwrap();
        std::fs::write(tmp.path().join("dup.txt"), "x\ny\nx\ny\n").unwrap();
        let tool = ApplyPatch::new(tmp.path());
        let err = tool
            .execute(json!({"patch": "\
*** Begin Patch
*** Update File: dup.txt
 x
-y
+Y
*** End Patch
", "dry_run": true}))
            .await
            .unwrap_err();
        assert!(matches!(err, Error::Tool { .. }));
        let msg = format!("{}", err);
        assert!(
            msg.contains("matches"),
            "error should mention multiple matches: {msg}"
        );
    }

    #[tokio::test]
    async fn dry_run_false_is_same_as_omitting_field() {
        let tmp = TempDir::new().unwrap();
        std::fs::write(tmp.path().join("test.txt"), "old\n").unwrap();
        let tool = ApplyPatch::new(tmp.path());
        tool.execute(json!({"patch": "\
*** Begin Patch
*** Update File: test.txt
-old
+new
*** End Patch
", "dry_run": false}))
            .await
            .unwrap();
        assert_eq!(
            std::fs::read_to_string(tmp.path().join("test.txt")).unwrap(),
            "new\n"
        );
    }
}