| 1 | //! Patch tools: `apply_patch` for unified diff patching |
| 2 | //! |
| 3 | //! This tool provides precise file modifications using unified diff format, |
| 4 | //! supporting multi-hunk patches and fuzzy matching. |
| 5 | |
| 6 | use std::collections::HashSet; |
| 7 | use std::fs; |
| 8 | use std::path::PathBuf; |
| 9 | |
| 10 | use async_trait::async_trait; |
| 11 | use serde::{Deserialize, Serialize}; |
| 12 | use serde_json::{Value, json}; |
| 13 | use thiserror::Error; |
| 14 | |
| 15 | use super::diff_format::make_unified_diff; |
| 16 | use super::file::{ |
| 17 | EXPECTED_HASH_DESCRIPTION, PATCH_PARAMS, PATH_ALIASES, apply_param_aliases, content_hash, |
| 18 | }; |
| 19 | use super::rust_format::normalize_edit; |
| 20 | use super::spec::{ |
| 21 | ApprovalRequirement, ToolCapability, ToolContext, ToolError, ToolResult, ToolSpec, |
| 22 | lsp_diagnostics_for_paths, optional_bool, optional_str, optional_u64, |
| 23 | }; |
| 24 | use super::syntax_check::guard_edit; |
| 25 | |
| 26 | /// Maximum lines of context for fuzzy matching (increased for better tolerance) |
| 27 | const MAX_FUZZ: usize = 50; |
| 28 | /// Default fuzz when the caller does not specify one. Matches the tool schema's |
| 29 | /// documented default. Previously the default was `MAX_FUZZ` (50), so a hunk |
| 30 | /// with no `fuzz` argument could silently apply up to 50 lines from its stated |
| 31 | /// position — landing in the wrong region of a file with repeated blocks. |
| 32 | const DEFAULT_FUZZ: usize = 3; |
| 33 | |
| 34 | /// Minimum number of expected lines (context + removed) required before a hunk |
| 35 | /// may be relocated to a unique whole-file context match when its stated line |
| 36 | /// numbers are stale (#5003). Short anchors — a lone `}` or a 1-2 line snippet |
| 37 | /// — appear in too many places to relocate safely. |
| 38 | const MIN_ANCHOR_LINES: usize = 4; |
| 39 | |
| 40 | /// Reassemble hunk-processed logical lines back into file content, preserving |
| 41 | /// the base file's line-ending style (CRLF vs LF) and its trailing-newline |
| 42 | /// state. Processing round-trips through `str::lines()`, which strips both the |
| 43 | /// trailing `\n` and any `\r`; naively `join("\n")`-ing would silently delete |
| 44 | /// the file's final newline and flip a CRLF file to LF on every patch. |
| 45 | fn reassemble_preserving_newlines(lines: &[String], base_content: &str) -> String { |
| 46 | if lines.is_empty() { |
| 47 | return String::new(); |
| 48 | } |
| 49 | let terminator = if base_content.contains("\r\n") { |
| 50 | "\r\n" |
| 51 | } else { |
| 52 | "\n" |
| 53 | }; |
| 54 | // A newly created file (empty base) gets a conventional trailing newline; |
| 55 | // an existing file preserves whether it had one. |
| 56 | let trailing = base_content.is_empty() || base_content.ends_with('\n'); |
| 57 | let mut out = lines.join(terminator); |
| 58 | if trailing { |
| 59 | out.push_str(terminator); |
| 60 | } |
| 61 | out |
| 62 | } |
| 63 | /// Limit how much context we print in error messages. |
| 64 | const HUNK_PREVIEW_LINES: usize = 4; |
| 65 | const SNIPPET_RADIUS: usize = 2; |
| 66 | const FILE_LIST_LIMIT: usize = 6; |
| 67 | |
| 68 | // === Types === |
| 69 | |
| 70 | /// Result of applying a patch |
| 71 | #[derive(Debug, Clone, Serialize, Deserialize)] |
| 72 | pub struct PatchResult { |
| 73 | pub success: bool, |
| 74 | pub files_applied: usize, |
| 75 | pub files_total: usize, |
| 76 | pub hunks_applied: usize, |
| 77 | pub hunks_total: usize, |
| 78 | pub fuzz_used: usize, |
| 79 | #[serde(default)] |
| 80 | pub hunks_with_fuzz: usize, |
| 81 | #[serde(default, skip_serializing_if = "is_zero")] |
| 82 | pub hunks_relocated: usize, |
| 83 | #[serde(default, skip_serializing_if = "Vec::is_empty")] |
| 84 | pub touched_files: Vec<String>, |
| 85 | #[serde(default, skip_serializing_if = "Vec::is_empty")] |
| 86 | pub file_summaries: Vec<FileSummary>, |
| 87 | pub message: String, |
| 88 | } |
| 89 | |
| 90 | /// Per-file summary for patch application output. |
| 91 | #[derive(Debug, Clone, Serialize, Deserialize)] |
| 92 | pub struct FileSummary { |
| 93 | pub path: String, |
| 94 | pub hunks: usize, |
| 95 | pub hunks_applied: usize, |
| 96 | pub fuzz_used: usize, |
| 97 | pub hunks_with_fuzz: usize, |
| 98 | #[serde(default, skip_serializing_if = "is_zero")] |
| 99 | pub hunks_relocated: usize, |
| 100 | pub created: bool, |
| 101 | pub deleted: bool, |
| 102 | } |
| 103 | |
| 104 | /// No-mutation summary of what an `apply_patch` input intends to touch. |
| 105 | #[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)] |
| 106 | pub struct ApplyPatchPreflight { |
| 107 | pub touched_files: Vec<String>, |
| 108 | pub files_total: usize, |
| 109 | pub hunks_total: usize, |
| 110 | #[serde(default, skip_serializing_if = "Vec::is_empty")] |
| 111 | pub creates: Vec<String>, |
| 112 | #[serde(default, skip_serializing_if = "Vec::is_empty")] |
| 113 | pub deletes: Vec<String>, |
| 114 | #[serde(default, skip_serializing_if = "Option::is_none")] |
| 115 | pub path_override: Option<String>, |
| 116 | #[serde(default, skip_serializing_if = "Option::is_none")] |
| 117 | pub header_path_mismatch: Option<String>, |
| 118 | } |
| 119 | |
| 120 | /// A single hunk in a unified diff |
| 121 | #[derive(Debug, Clone)] |
| 122 | pub struct Hunk { |
| 123 | pub old_start: usize, |
| 124 | #[cfg_attr(not(test), expect(dead_code))] |
| 125 | pub old_count: usize, |
| 126 | #[cfg_attr(not(test), expect(dead_code))] |
| 127 | pub new_start: usize, |
| 128 | #[cfg_attr(not(test), expect(dead_code))] |
| 129 | pub new_count: usize, |
| 130 | pub lines: Vec<HunkLine>, |
| 131 | } |
| 132 | |
| 133 | /// A line in a hunk |
| 134 | #[derive(Debug, Clone)] |
| 135 | pub enum HunkLine { |
| 136 | Context(String), |
| 137 | Add(String), |
| 138 | Remove(String), |
| 139 | } |
| 140 | |
| 141 | /// Tool for applying unified diff patches to files |
| 142 | pub struct ApplyPatchTool; |
| 143 | |
| 144 | #[derive(Debug, Clone)] |
| 145 | struct FilePatch { |
| 146 | path: String, |
| 147 | hunks: Vec<Hunk>, |
| 148 | delete_after: bool, |
| 149 | create_if_missing: bool, |
| 150 | } |
| 151 | |
| 152 | #[derive(Debug, Clone)] |
| 153 | struct PendingWrite { |
| 154 | path: PathBuf, |
| 155 | content: Option<String>, |
| 156 | original: Option<String>, |
| 157 | } |
| 158 | |
| 159 | #[derive(Debug, Default, Clone, Copy)] |
| 160 | struct PatchStats { |
| 161 | files_applied: usize, |
| 162 | files_total: usize, |
| 163 | hunks_applied: usize, |
| 164 | hunks_total: usize, |
| 165 | fuzz_used: usize, |
| 166 | hunks_with_fuzz: usize, |
| 167 | hunks_relocated: usize, |
| 168 | } |
| 169 | |
| 170 | fn is_zero(value: &usize) -> bool { |
| 171 | *value == 0 |
| 172 | } |
| 173 | |
| 174 | #[derive(Debug, Default, Clone)] |
| 175 | struct PatchStatsExt { |
| 176 | stats: PatchStats, |
| 177 | touched_files: Vec<String>, |
| 178 | file_summaries: Vec<FileSummary>, |
| 179 | header_path_mismatch: Option<String>, |
| 180 | } |
| 181 | |
| 182 | #[derive(Debug, Default, Clone)] |
| 183 | struct PatchShape { |
| 184 | has_hunks: bool, |
| 185 | header_files: Vec<String>, |
| 186 | } |
| 187 | |
| 188 | impl PatchShape { |
| 189 | fn file_count(&self) -> usize { |
| 190 | self.header_files.len() |
| 191 | } |
| 192 | } |
| 193 | |
| 194 | #[derive(Debug, Default, Clone, Copy)] |
| 195 | struct HunkApplyStats { |
| 196 | hunks_applied: usize, |
| 197 | fuzz_used: usize, |
| 198 | hunks_with_fuzz: usize, |
| 199 | hunks_relocated: usize, |
| 200 | } |
| 201 | |
| 202 | /// Result of applying a single hunk: how much positional fuzz was used, and |
| 203 | /// whether the hunk had to be relocated to a unique whole-file context match |
| 204 | /// (stale line numbers after earlier edits, #5003). |
| 205 | #[derive(Debug, Default, Clone, Copy)] |
| 206 | struct HunkApplyOutcome { |
| 207 | fuzz_used: usize, |
| 208 | relocated: bool, |
| 209 | } |
| 210 | |
| 211 | #[derive(Debug, Clone)] |
| 212 | enum ApplyPatchPreflightKind { |
| 213 | Replace, |
| 214 | PathOverride { path: String, hunks: Vec<Hunk> }, |
| 215 | FilePatches(Vec<FilePatch>), |
| 216 | } |
| 217 | |
| 218 | /// Canonicalized `apply_patch` payload mode. |
| 219 | /// |
| 220 | /// `replace` is the preferred spelling for full-file replacements. `changes` |
| 221 | /// remains a compatibility alias for callers that learned the original tool |
| 222 | /// schema before the clearer name was introduced. |
| 223 | #[derive(Debug, Clone, Copy)] |
| 224 | pub(crate) enum NormalizedApplyPatchInput<'a> { |
| 225 | Patch(&'a str), |
| 226 | Replacement { |
| 227 | entries: &'a [Value], |
| 228 | source_field: &'static str, |
| 229 | }, |
| 230 | } |
| 231 | |
| 232 | /// Validate mutual exclusivity and normalize the legacy `changes` alias. |
| 233 | /// |
| 234 | /// This is the single parser used by execution, preflight, policy, approval, |
| 235 | /// and UI consumers so every surface agrees on the accepted input contract. |
| 236 | pub(crate) fn normalize_apply_patch_input( |
| 237 | input: &Value, |
| 238 | ) -> Result<NormalizedApplyPatchInput<'_>, ToolError> { |
| 239 | let provided: Vec<&'static str> = ["patch", "replace", "changes"] |
| 240 | .into_iter() |
| 241 | .filter(|field| input.get(*field).is_some()) |
| 242 | .collect(); |
| 243 | |
| 244 | if provided.len() > 1 { |
| 245 | let fields = provided |
| 246 | .iter() |
| 247 | .map(|field| format!("`{field}`")) |
| 248 | .collect::<Vec<_>>() |
| 249 | .join(", "); |
| 250 | return Err(ToolError::invalid_input(format!( |
| 251 | "Cannot use {fields} simultaneously. Choose exactly one of `patch`, `replace`, or the deprecated `changes` alias." |
| 252 | ))); |
| 253 | } |
| 254 | |
| 255 | let Some(field) = provided.first().copied() else { |
| 256 | return Err(ToolError::missing_field( |
| 257 | "patch, replace, or deprecated changes", |
| 258 | )); |
| 259 | }; |
| 260 | |
| 261 | if field == "patch" { |
| 262 | let patch = input |
| 263 | .get(field) |
| 264 | .and_then(Value::as_str) |
| 265 | .ok_or_else(|| ToolError::invalid_input("`patch` must be a string"))?; |
| 266 | return Ok(NormalizedApplyPatchInput::Patch(patch)); |
| 267 | } |
| 268 | |
| 269 | let entries = input.get(field).and_then(Value::as_array).ok_or_else(|| { |
| 270 | ToolError::invalid_input(format!( |
| 271 | "`{field}` must be an array of objects like {{path, content}}" |
| 272 | )) |
| 273 | })?; |
| 274 | if entries.is_empty() { |
| 275 | return Err(ToolError::invalid_input(format!( |
| 276 | "`{field}` cannot be empty" |
| 277 | ))); |
| 278 | } |
| 279 | |
| 280 | Ok(NormalizedApplyPatchInput::Replacement { |
| 281 | entries, |
| 282 | source_field: field, |
| 283 | }) |
| 284 | } |
| 285 | |
| 286 | #[derive(Debug, Clone)] |
| 287 | struct ApplyPatchPreflightPlan { |
| 288 | summary: ApplyPatchPreflight, |
| 289 | kind: ApplyPatchPreflightKind, |
| 290 | } |
| 291 | |
| 292 | // === Errors === |
| 293 | |
| 294 | #[derive(Debug, Error)] |
| 295 | enum ApplyHunkError { |
| 296 | #[error( |
| 297 | "Failed to find matching location for hunk (expected at line {expected_line}, adjusted to {adjusted_line} with offset {offset:+})" |
| 298 | )] |
| 299 | NoMatch { |
| 300 | expected_line: usize, |
| 301 | adjusted_line: usize, |
| 302 | offset: isize, |
| 303 | }, |
| 304 | #[error( |
| 305 | "Hunk context is ambiguous: matches at multiple locations {candidate_lines:?}, expected around line {expected_line}" |
| 306 | )] |
| 307 | ContextAmbiguous { |
| 308 | expected_line: usize, |
| 309 | candidate_lines: Vec<usize>, |
| 310 | }, |
| 311 | } |
| 312 | |
| 313 | #[async_trait] |
| 314 | impl ToolSpec for ApplyPatchTool { |
| 315 | fn name(&self) -> &'static str { |
| 316 | "apply_patch" |
| 317 | } |
| 318 | |
| 319 | fn model_visible(&self) -> bool { |
| 320 | true |
| 321 | } |
| 322 | |
| 323 | fn description(&self) -> &'static str { |
| 324 | "Apply a transactional unified-diff patch across one or more files, with fuzzy context matching and a rendered diff." |
| 325 | } |
| 326 | |
| 327 | fn input_schema(&self) -> Value { |
| 328 | json!({ |
| 329 | "type": "object", |
| 330 | "properties": { |
| 331 | "path": { |
| 332 | "type": "string", |
| 333 | "description": "Path to the file to patch (relative to workspace)" |
| 334 | }, |
| 335 | "patch": { |
| 336 | "type": "string", |
| 337 | "description": "Unified diff patch content" |
| 338 | }, |
| 339 | "replace": { |
| 340 | "type": "array", |
| 341 | "description": "Optional full file replacements (path + content).", |
| 342 | "items": { |
| 343 | "type": "object", |
| 344 | "properties": { |
| 345 | "path": { "type": "string" }, |
| 346 | "content": { "type": "string" } |
| 347 | }, |
| 348 | "required": ["path", "content"] |
| 349 | } |
| 350 | }, |
| 351 | "changes": { |
| 352 | "type": "array", |
| 353 | "description": "Deprecated compatibility alias for `replace` (full file replacements by path + content).", |
| 354 | "items": { |
| 355 | "type": "object", |
| 356 | "properties": { |
| 357 | "path": { "type": "string" }, |
| 358 | "content": { "type": "string" } |
| 359 | }, |
| 360 | "required": ["path", "content"] |
| 361 | } |
| 362 | }, |
| 363 | "fuzz": { |
| 364 | "type": "integer", |
| 365 | "description": "Maximum fuzz factor for fuzzy matching (default: 3)" |
| 366 | }, |
| 367 | "create_if_missing": { |
| 368 | "type": "boolean", |
| 369 | "description": "Create the file if it doesn't exist (for new file patches)" |
| 370 | }, |
| 371 | "expected_hash": { |
| 372 | "type": "string", |
| 373 | "description": format!( |
| 374 | "{EXPECTED_HASH_DESCRIPTION} Verifies the patch target — the `path` argument when given, otherwise the first file the patch touches; other files in a multi-file patch are not hash-checked." |
| 375 | ) |
| 376 | } |
| 377 | }, |
| 378 | "oneOf": [ |
| 379 | { "required": ["patch"] }, |
| 380 | { "required": ["replace"] }, |
| 381 | { "required": ["changes"] } |
| 382 | ] |
| 383 | }) |
| 384 | } |
| 385 | |
| 386 | fn capabilities(&self) -> Vec<ToolCapability> { |
| 387 | vec![ |
| 388 | ToolCapability::WritesFiles, |
| 389 | ToolCapability::Sandboxable, |
| 390 | ToolCapability::RequiresApproval, |
| 391 | ] |
| 392 | } |
| 393 | |
| 394 | fn approval_requirement(&self) -> ApprovalRequirement { |
| 395 | ApprovalRequirement::Suggest |
| 396 | } |
| 397 | |
| 398 | async fn execute(&self, input: Value, context: &ToolContext) -> Result<ToolResult, ToolError> { |
| 399 | let mut input = input; |
| 400 | apply_param_aliases(&mut input, PATH_ALIASES, "File patch")?; |
| 401 | PATCH_PARAMS.reject_unknown(&input)?; |
| 402 | let input = input; |
| 403 | |
| 404 | let fuzz = optional_u64(&input, "fuzz", DEFAULT_FUZZ as u64)?.min(MAX_FUZZ as u64); |
| 405 | let fuzz = usize::try_from(fuzz).unwrap_or(DEFAULT_FUZZ); |
| 406 | let normalized = normalize_apply_patch_input(&input)?; |
| 407 | let create_if_missing = optional_bool(&input, "create_if_missing", false)?; |
| 408 | let preflight = preflight_apply_patch_plan(&input, normalized)?; |
| 409 | verify_patch_expected_hash(&input, &preflight.summary, context)?; |
| 410 | |
| 411 | if let NormalizedApplyPatchInput::Replacement { |
| 412 | entries, |
| 413 | source_field, |
| 414 | } = normalized |
| 415 | { |
| 416 | let (mut pending, stats) = |
| 417 | build_pending_writes_from_replace(entries, source_field, context)?; |
| 418 | normalize_pending_rust(&mut pending).await; |
| 419 | apply_pending_writes(&pending)?; |
| 420 | // Resolve absolute paths for LSP diagnostics query. |
| 421 | let abs_paths: Vec<PathBuf> = pending.iter().map(|p| p.path.clone()).collect(); |
| 422 | let diag_block = lsp_diagnostics_for_paths(context, &abs_paths).await; |
| 423 | let result = PatchResult { |
| 424 | success: true, |
| 425 | files_applied: stats.stats.files_applied, |
| 426 | files_total: stats.stats.files_total, |
| 427 | hunks_applied: stats.stats.hunks_applied, |
| 428 | hunks_total: stats.stats.hunks_total, |
| 429 | fuzz_used: stats.stats.fuzz_used, |
| 430 | hunks_with_fuzz: stats.stats.hunks_with_fuzz, |
| 431 | hunks_relocated: stats.stats.hunks_relocated, |
| 432 | touched_files: stats.touched_files.clone(), |
| 433 | file_summaries: stats.file_summaries.clone(), |
| 434 | message: build_summary_message(&stats), |
| 435 | }; |
| 436 | let mut tool_result = ToolResult::json(&result) |
| 437 | .map_err(|e| ToolError::execution_failed(e.to_string()))?; |
| 438 | tool_result = tool_result.with_metadata(apply_patch_result_metadata( |
| 439 | &preflight.summary, |
| 440 | &pending, |
| 441 | &stats, |
| 442 | )); |
| 443 | if !diag_block.is_empty() { |
| 444 | tool_result.content.push('\n'); |
| 445 | tool_result.content.push_str(&diag_block); |
| 446 | } |
| 447 | return Ok(tool_result); |
| 448 | } |
| 449 | |
| 450 | let file_patches = match preflight.kind { |
| 451 | ApplyPatchPreflightKind::Replace => { |
| 452 | unreachable!("replace input returned before patch execution") |
| 453 | } |
| 454 | ApplyPatchPreflightKind::PathOverride { path, hunks } => vec![FilePatch { |
| 455 | path, |
| 456 | hunks, |
| 457 | delete_after: false, |
| 458 | create_if_missing, |
| 459 | }], |
| 460 | ApplyPatchPreflightKind::FilePatches(file_patches) => file_patches, |
| 461 | }; |
| 462 | |
| 463 | let (mut pending, mut stats) = |
| 464 | build_pending_writes_from_patches(file_patches, context, fuzz)?; |
| 465 | stats.header_path_mismatch = preflight.summary.header_path_mismatch.clone(); |
| 466 | normalize_pending_rust(&mut pending).await; |
| 467 | apply_pending_writes(&pending)?; |
| 468 | // Resolve absolute paths for LSP diagnostics query. |
| 469 | let abs_paths: Vec<PathBuf> = pending |
| 470 | .iter() |
| 471 | .filter(|p| p.content.is_some()) // skip deleted files |
| 472 | .map(|p| p.path.clone()) |
| 473 | .collect(); |
| 474 | let diag_block = lsp_diagnostics_for_paths(context, &abs_paths).await; |
| 475 | let result = PatchResult { |
| 476 | success: true, |
| 477 | files_applied: stats.stats.files_applied, |
| 478 | files_total: stats.stats.files_total, |
| 479 | hunks_applied: stats.stats.hunks_applied, |
| 480 | hunks_total: stats.stats.hunks_total, |
| 481 | fuzz_used: stats.stats.fuzz_used, |
| 482 | hunks_with_fuzz: stats.stats.hunks_with_fuzz, |
| 483 | hunks_relocated: stats.stats.hunks_relocated, |
| 484 | touched_files: stats.touched_files.clone(), |
| 485 | file_summaries: stats.file_summaries.clone(), |
| 486 | message: build_summary_message(&stats), |
| 487 | }; |
| 488 | let mut tool_result = |
| 489 | ToolResult::json(&result).map_err(|e| ToolError::execution_failed(e.to_string()))?; |
| 490 | tool_result = tool_result.with_metadata(apply_patch_result_metadata( |
| 491 | &preflight.summary, |
| 492 | &pending, |
| 493 | &stats, |
| 494 | )); |
| 495 | if !diag_block.is_empty() { |
| 496 | tool_result.content.push('\n'); |
| 497 | tool_result.content.push_str(&diag_block); |
| 498 | } |
| 499 | Ok(tool_result) |
| 500 | } |
| 501 | } |
| 502 | |
| 503 | /// Enforce the optional `expected_hash` precondition for a patch (#3979). |
| 504 | /// |
| 505 | /// **This is a whole-patch precondition on a single target file, not a per-file |
| 506 | /// guard.** A unified diff carries no place to attach a hash per file section, |
| 507 | /// and `File action="patch"` flattens its arguments into one object, so there |
| 508 | /// is no clean parameter shape for a hash-per-path map without inventing a |
| 509 | /// syntax the model has never seen. The guarded file is therefore the patch's |
| 510 | /// target: the explicit `path` argument when one is given, otherwise the first |
| 511 | /// file the patch touches. A multi-file patch still verifies only that one |
| 512 | /// file; for the rest, the existing hunk-context matching remains the check |
| 513 | /// that a stale patch fails on. |
| 514 | /// |
| 515 | /// Runs before any pending write is built or applied, so a mismatch leaves |
| 516 | /// every file in the patch untouched. |
| 517 | fn verify_patch_expected_hash( |
| 518 | input: &Value, |
| 519 | summary: &ApplyPatchPreflight, |
| 520 | context: &ToolContext, |
| 521 | ) -> Result<(), ToolError> { |
| 522 | let Some(expected) = optional_str(input, "expected_hash")? else { |
| 523 | return Ok(()); |
| 524 | }; |
| 525 | |
| 526 | let Some(target) = summary |
| 527 | .path_override |
| 528 | .as_deref() |
| 529 | .or_else(|| summary.touched_files.first().map(String::as_str)) |
| 530 | else { |
| 531 | return Err(ToolError::execution_failed( |
| 532 | "File `patch` refused: expected_hash was supplied but the patch names no target file to verify it against, so nothing was written.".to_string(), |
| 533 | )); |
| 534 | }; |
| 535 | |
| 536 | let resolved = context.resolve_path(target)?; |
| 537 | if !resolved.exists() { |
| 538 | // Fail closed, matching `write`: a hash describes a file that was |
| 539 | // read, so a missing target means the guard cannot be honored. |
| 540 | return Err(ToolError::execution_failed(format!( |
| 541 | "File `patch` refused: expected_hash was supplied but {target} does not exist, so there is no snapshot to verify and nothing was written. Recovery: drop `expected_hash` when creating files." |
| 542 | ))); |
| 543 | } |
| 544 | |
| 545 | let current = fs::read(&resolved).map_err(|e| { |
| 546 | ToolError::execution_failed(format!( |
| 547 | "File `patch` refused: could not read {target} to verify expected_hash ({e}); nothing was written." |
| 548 | )) |
| 549 | })?; |
| 550 | let actual = content_hash(¤t); |
| 551 | if actual == expected { |
| 552 | return Ok(()); |
| 553 | } |
| 554 | Err(ToolError::execution_failed(format!( |
| 555 | "File `patch` refused: {target} changed since it was read. \ |
| 556 | expected_hash was {expected} but the file is now {actual}, so nothing was written. \ |
| 557 | Recovery: call File with action=\"read\" path=\"{target}\" to get the current contents \ |
| 558 | and its content_hash, then rebuild the patch against them." |
| 559 | ))) |
| 560 | } |
| 561 | |
| 562 | /// Parse `apply_patch` input into a reusable, no-mutation preflight summary. |
| 563 | /// |
| 564 | /// This deliberately stops before workspace resolution or file reads. It is |
| 565 | /// suitable for policy checks, audit logs, diagnostics hooks, and future undo |
| 566 | /// planning that must know the target files before mutation. |
| 567 | pub fn preflight_apply_patch(input: &Value) -> Result<ApplyPatchPreflight, ToolError> { |
| 568 | let normalized = normalize_apply_patch_input(input)?; |
| 569 | Ok(preflight_apply_patch_plan(input, normalized)?.summary) |
| 570 | } |
| 571 | |
| 572 | fn preflight_apply_patch_plan( |
| 573 | input: &Value, |
| 574 | normalized: NormalizedApplyPatchInput<'_>, |
| 575 | ) -> Result<ApplyPatchPreflightPlan, ToolError> { |
| 576 | let create_if_missing = optional_bool(input, "create_if_missing", false)?; |
| 577 | |
| 578 | if let NormalizedApplyPatchInput::Replacement { |
| 579 | entries, |
| 580 | source_field, |
| 581 | } = normalized |
| 582 | { |
| 583 | return Ok(ApplyPatchPreflightPlan { |
| 584 | summary: preflight_replace(entries, source_field)?, |
| 585 | kind: ApplyPatchPreflightKind::Replace, |
| 586 | }); |
| 587 | } |
| 588 | |
| 589 | let NormalizedApplyPatchInput::Patch(patch_text) = normalized else { |
| 590 | unreachable!("replacement input returned before patch parsing") |
| 591 | }; |
| 592 | let path_override = optional_str(input, "path")?; |
| 593 | let patch_shape = inspect_patch_shape(patch_text); |
| 594 | validate_patch_shape(&patch_shape, path_override)?; |
| 595 | let header_path_mismatch = |
| 596 | path_override.and_then(|path| diff_header_mismatch(path, &patch_shape)); |
| 597 | |
| 598 | if let Some(path) = path_override { |
| 599 | let hunks = parse_unified_diff(patch_text)?; |
| 600 | if hunks.is_empty() { |
| 601 | return Err(ToolError::invalid_input( |
| 602 | "Patch did not contain any hunks (`@@ ... @@`). Provide a unified diff hunk.", |
| 603 | )); |
| 604 | } |
| 605 | return Ok(ApplyPatchPreflightPlan { |
| 606 | summary: ApplyPatchPreflight { |
| 607 | touched_files: vec![path.to_string()], |
| 608 | files_total: 1, |
| 609 | hunks_total: hunks.len(), |
| 610 | creates: if create_if_missing { |
| 611 | vec![path.to_string()] |
| 612 | } else { |
| 613 | Vec::new() |
| 614 | }, |
| 615 | deletes: Vec::new(), |
| 616 | path_override: Some(path.to_string()), |
| 617 | header_path_mismatch, |
| 618 | }, |
| 619 | kind: ApplyPatchPreflightKind::PathOverride { |
| 620 | path: path.to_string(), |
| 621 | hunks, |
| 622 | }, |
| 623 | }); |
| 624 | } |
| 625 | |
| 626 | let file_patches = parse_unified_diff_files(patch_text, create_if_missing)?; |
| 627 | if file_patches.is_empty() { |
| 628 | return Err(ToolError::invalid_input( |
| 629 | "No valid file patches found. Ensure the patch includes `---`/`+++` headers or provide `path`.", |
| 630 | )); |
| 631 | } |
| 632 | |
| 633 | let mut touched_files = Vec::new(); |
| 634 | let mut creates = Vec::new(); |
| 635 | let mut deletes = Vec::new(); |
| 636 | let mut hunks_total = 0; |
| 637 | for file_patch in &file_patches { |
| 638 | if file_patch.hunks.is_empty() { |
| 639 | return Err(ToolError::invalid_input(format!( |
| 640 | "Patch section for `{}` has no hunks (`@@ ... @@`).", |
| 641 | file_patch.path |
| 642 | ))); |
| 643 | } |
| 644 | push_unique(&mut touched_files, file_patch.path.clone()); |
| 645 | hunks_total += file_patch.hunks.len(); |
| 646 | if file_patch.create_if_missing && !file_patch.delete_after { |
| 647 | push_unique(&mut creates, file_patch.path.clone()); |
| 648 | } |
| 649 | if file_patch.delete_after { |
| 650 | push_unique(&mut deletes, file_patch.path.clone()); |
| 651 | } |
| 652 | } |
| 653 | |
| 654 | Ok(ApplyPatchPreflightPlan { |
| 655 | summary: ApplyPatchPreflight { |
| 656 | files_total: file_patches.len(), |
| 657 | touched_files, |
| 658 | hunks_total, |
| 659 | creates, |
| 660 | deletes, |
| 661 | path_override: None, |
| 662 | header_path_mismatch, |
| 663 | }, |
| 664 | kind: ApplyPatchPreflightKind::FilePatches(file_patches), |
| 665 | }) |
| 666 | } |
| 667 | |
| 668 | fn preflight_replace( |
| 669 | changes: &[Value], |
| 670 | source_field: &str, |
| 671 | ) -> Result<ApplyPatchPreflight, ToolError> { |
| 672 | let mut touched_files = Vec::new(); |
| 673 | for change in changes { |
| 674 | let path = change |
| 675 | .get("path") |
| 676 | .and_then(Value::as_str) |
| 677 | .ok_or_else(|| ToolError::missing_field(format!("{source_field}[].path")))?; |
| 678 | let _content = change |
| 679 | .get("content") |
| 680 | .and_then(Value::as_str) |
| 681 | .ok_or_else(|| ToolError::missing_field(format!("{source_field}[].content")))?; |
| 682 | push_unique(&mut touched_files, path.to_string()); |
| 683 | } |
| 684 | |
| 685 | Ok(ApplyPatchPreflight { |
| 686 | files_total: changes.len(), |
| 687 | touched_files, |
| 688 | hunks_total: 0, |
| 689 | creates: Vec::new(), |
| 690 | deletes: Vec::new(), |
| 691 | path_override: None, |
| 692 | header_path_mismatch: None, |
| 693 | }) |
| 694 | } |
| 695 | |
| 696 | fn apply_patch_result_metadata( |
| 697 | preflight: &ApplyPatchPreflight, |
| 698 | pending: &[PendingWrite], |
| 699 | stats: &PatchStatsExt, |
| 700 | ) -> Value { |
| 701 | let mut metadata = |
| 702 | serde_json::to_value(preflight).expect("ApplyPatchPreflight should serialize"); |
| 703 | if let Some(object) = metadata.as_object_mut() { |
| 704 | object.insert("event".to_string(), json!("apply_patch.preflight")); |
| 705 | object.insert( |
| 706 | "mutation".to_string(), |
| 707 | build_mutation_metadata(pending, &stats.file_summaries), |
| 708 | ); |
| 709 | } |
| 710 | metadata |
| 711 | } |
| 712 | |
| 713 | /// Preserve the exact applied before/after diff independently from approval |
| 714 | /// presentation. The TUI consumes this success-only metadata for its calm |
| 715 | /// File receipt; the normal model-facing result remains compact JSON. |
| 716 | fn build_mutation_metadata(pending: &[PendingWrite], summaries: &[FileSummary]) -> Value { |
| 717 | let mut matched = HashSet::new(); |
| 718 | let mut renames = Vec::new(); |
| 719 | |
| 720 | for (delete_index, (deleted, delete_summary)) in pending.iter().zip(summaries).enumerate() { |
| 721 | if !delete_summary.deleted || matched.contains(&delete_index) { |
| 722 | continue; |
| 723 | } |
| 724 | let Some(old_content) = deleted.original.as_deref() else { |
| 725 | continue; |
| 726 | }; |
| 727 | let Some((create_index, (_, create_summary))) = pending |
| 728 | .iter() |
| 729 | .zip(summaries) |
| 730 | .enumerate() |
| 731 | .find(|(index, (created, summary))| { |
| 732 | !matched.contains(index) |
| 733 | && summary.created |
| 734 | && created.content.as_deref() == Some(old_content) |
| 735 | }) |
| 736 | else { |
| 737 | continue; |
| 738 | }; |
| 739 | matched.insert(delete_index); |
| 740 | matched.insert(create_index); |
| 741 | renames.push(json!({ |
| 742 | "from": delete_summary.path, |
| 743 | "to": create_summary.path, |
| 744 | })); |
| 745 | } |
| 746 | |
| 747 | let mut files = Vec::new(); |
| 748 | for (index, summary) in summaries.iter().enumerate() { |
| 749 | if matched.contains(&index) { |
| 750 | continue; |
| 751 | } |
| 752 | let outcome = if summary.created { |
| 753 | "created" |
| 754 | } else if summary.deleted { |
| 755 | "deleted" |
| 756 | } else { |
| 757 | "updated" |
| 758 | }; |
| 759 | files.push(json!({ "path": summary.path, "outcome": outcome })); |
| 760 | } |
| 761 | |
| 762 | let mut diff_parts = Vec::new(); |
| 763 | for rename in &renames { |
| 764 | let from = rename["from"].as_str().unwrap_or("<file>"); |
| 765 | let to = rename["to"].as_str().unwrap_or("<file>"); |
| 766 | diff_parts.push(format!( |
| 767 | "diff --git a/{from} b/{to}\nsimilarity index 100%\nrename from {from}\nrename to {to}\n" |
| 768 | )); |
| 769 | } |
| 770 | for (index, (write, summary)) in pending.iter().zip(summaries).enumerate() { |
| 771 | if matched.contains(&index) { |
| 772 | continue; |
| 773 | } |
| 774 | let old = write.original.as_deref().unwrap_or(""); |
| 775 | let new = write.content.as_deref().unwrap_or(""); |
| 776 | let diff = make_unified_diff(&summary.path, old, new); |
| 777 | if !diff.is_empty() { |
| 778 | diff_parts.push(format!( |
| 779 | "diff --git a/{path} b/{path}\n{diff}", |
| 780 | path = summary.path |
| 781 | )); |
| 782 | } |
| 783 | } |
| 784 | |
| 785 | json!({ |
| 786 | "diff": diff_parts.join("\n"), |
| 787 | "files": files, |
| 788 | "renames": renames, |
| 789 | }) |
| 790 | } |
| 791 | |
| 792 | /// Parse a unified diff into hunks |
| 793 | fn parse_unified_diff(patch: &str) -> Result<Vec<Hunk>, ToolError> { |
| 794 | let mut hunks = Vec::new(); |
| 795 | let mut lines = patch.lines().peekable(); |
| 796 | |
| 797 | // Skip header lines (---, +++ etc) |
| 798 | while let Some(line) = lines.peek() { |
| 799 | if line.starts_with("@@") { |
| 800 | break; |
| 801 | } |
| 802 | lines.next(); |
| 803 | } |
| 804 | |
| 805 | // Parse hunks |
| 806 | while let Some(line) = lines.next() { |
| 807 | if line.starts_with("@@") { |
| 808 | let hunk = parse_hunk_header(line, &mut lines)?; |
| 809 | hunks.push(hunk); |
| 810 | } |
| 811 | } |
| 812 | |
| 813 | Ok(hunks) |
| 814 | } |
| 815 | |
| 816 | fn parse_unified_diff_files( |
| 817 | patch: &str, |
| 818 | create_if_missing: bool, |
| 819 | ) -> Result<Vec<FilePatch>, ToolError> { |
| 820 | let mut files = Vec::new(); |
| 821 | let mut lines = patch.lines().peekable(); |
| 822 | let mut current: Option<FilePatch> = None; |
| 823 | let mut old_path: Option<String> = None; |
| 824 | |
| 825 | while let Some(line) = lines.next() { |
| 826 | if line.starts_with("diff --git ") { |
| 827 | if let Some(file) = current.take() { |
| 828 | files.push(file); |
| 829 | } |
| 830 | old_path = None; |
| 831 | continue; |
| 832 | } |
| 833 | |
| 834 | if let Some(stripped) = line.strip_prefix("--- ") { |
| 835 | old_path = Some(stripped.trim().to_string()); |
| 836 | continue; |
| 837 | } |
| 838 | |
| 839 | if let Some(stripped) = line.strip_prefix("+++ ") { |
| 840 | let new_path = Some(stripped.trim().to_string()); |
| 841 | let (path, delete_after, create_flag) = |
| 842 | resolve_diff_paths(old_path.as_deref(), new_path.as_deref(), create_if_missing)?; |
| 843 | old_path = None; |
| 844 | if let Some(file) = current.take() { |
| 845 | files.push(file); |
| 846 | } |
| 847 | current = Some(FilePatch { |
| 848 | path, |
| 849 | hunks: Vec::new(), |
| 850 | delete_after, |
| 851 | create_if_missing: create_flag, |
| 852 | }); |
| 853 | continue; |
| 854 | } |
| 855 | |
| 856 | if line.starts_with("@@") { |
| 857 | let Some(file) = current.as_mut() else { |
| 858 | if let Some(path) = old_path.as_deref() { |
| 859 | return Err(ToolError::invalid_input(format!( |
| 860 | "Patch hunk encountered after `--- {path}` but before a matching `+++` header. Each file section must include both headers." |
| 861 | ))); |
| 862 | } |
| 863 | return Err(ToolError::invalid_input( |
| 864 | "Patch hunk encountered before any file header. Add `---`/`+++` headers or provide `path`.", |
| 865 | )); |
| 866 | }; |
| 867 | let hunk = parse_hunk_header(line, &mut lines)?; |
| 868 | file.hunks.push(hunk); |
| 869 | } |
| 870 | } |
| 871 | |
| 872 | if let Some(file) = current { |
| 873 | files.push(file); |
| 874 | } |
| 875 | |
| 876 | Ok(files) |
| 877 | } |
| 878 | |
| 879 | fn resolve_diff_paths( |
| 880 | old_path: Option<&str>, |
| 881 | new_path: Option<&str>, |
| 882 | create_if_missing: bool, |
| 883 | ) -> Result<(String, bool, bool), ToolError> { |
| 884 | let old_norm = old_path.and_then(normalize_diff_path); |
| 885 | let new_norm = new_path.and_then(normalize_diff_path); |
| 886 | let delete_after = new_norm.is_none(); |
| 887 | let create_flag = create_if_missing || old_norm.is_none(); |
| 888 | let path = new_norm |
| 889 | .or(old_norm) |
| 890 | .ok_or_else(|| ToolError::invalid_input("Patch is missing both old and new file paths"))?; |
| 891 | Ok((path, delete_after, create_flag)) |
| 892 | } |
| 893 | |
| 894 | fn normalize_diff_path(raw: &str) -> Option<String> { |
| 895 | let raw = raw.split_once('\t').map_or(raw, |(path, _timestamp)| path); |
| 896 | let raw = raw.trim(); |
| 897 | if raw.is_empty() { |
| 898 | return None; |
| 899 | } |
| 900 | if raw == "/dev/null" || raw == "dev/null" { |
| 901 | return None; |
| 902 | } |
| 903 | let raw = raw |
| 904 | .strip_prefix("a/") |
| 905 | .or_else(|| raw.strip_prefix("b/")) |
| 906 | .unwrap_or(raw); |
| 907 | Some(raw.to_string()) |
| 908 | } |
| 909 | |
| 910 | /// Parse a hunk header and its content |
| 911 | fn parse_hunk_header<'a, I>( |
| 912 | header: &str, |
| 913 | lines: &mut std::iter::Peekable<I>, |
| 914 | ) -> Result<Hunk, ToolError> |
| 915 | where |
| 916 | I: Iterator<Item = &'a str>, |
| 917 | { |
| 918 | // Parse @@ -old_start,old_count +new_start,new_count @@ |
| 919 | let parts: Vec<&str> = header.split_whitespace().collect(); |
| 920 | if parts.len() < 3 { |
| 921 | return Err(ToolError::invalid_input(format!( |
| 922 | "Invalid hunk header: {header}. Expected numeric unified-diff form `@@ -old_start,old_count +new_start,new_count @@` (example: `@@ -12,3 +12,5 @@`)." |
| 923 | ))); |
| 924 | } |
| 925 | |
| 926 | let old_range = parts[1].trim_start_matches('-'); |
| 927 | let new_range = parts[2].trim_start_matches('+'); |
| 928 | |
| 929 | let (old_start, old_count) = parse_range(old_range)?; |
| 930 | let (new_start, new_count) = parse_range(new_range)?; |
| 931 | |
| 932 | // Parse hunk lines |
| 933 | let mut hunk_lines = Vec::new(); |
| 934 | let expected_lines = old_count.max(new_count) + old_count.min(new_count); |
| 935 | |
| 936 | for _ in 0..expected_lines * 2 { |
| 937 | // Allow for more lines than expected |
| 938 | match lines.peek() { |
| 939 | Some(line) if line.starts_with("@@") => break, |
| 940 | Some(line) if line.starts_with('-') => { |
| 941 | hunk_lines.push(HunkLine::Remove(line[1..].to_string())); |
| 942 | lines.next(); |
| 943 | } |
| 944 | Some(line) if line.starts_with('+') => { |
| 945 | hunk_lines.push(HunkLine::Add(line[1..].to_string())); |
| 946 | lines.next(); |
| 947 | } |
| 948 | Some(line) if line.starts_with(' ') || line.is_empty() => { |
| 949 | let content = if line.is_empty() { "" } else { &line[1..] }; |
| 950 | hunk_lines.push(HunkLine::Context(content.to_string())); |
| 951 | lines.next(); |
| 952 | } |
| 953 | Some(line) |
| 954 | if line.starts_with("diff ") |
| 955 | || line.starts_with("--- ") |
| 956 | || line.starts_with("+++ ") => |
| 957 | { |
| 958 | // Start of a new file patch - don't consume, let outer loop handle it |
| 959 | break; |
| 960 | } |
| 961 | Some(line) if !line.starts_with('\\') => { |
| 962 | // Treat as context line without leading space |
| 963 | hunk_lines.push(HunkLine::Context((*line).to_string())); |
| 964 | lines.next(); |
| 965 | } |
| 966 | Some(_) => { |
| 967 | lines.next(); // Skip "\ No newline at end of file" etc |
| 968 | } |
| 969 | None => break, |
| 970 | } |
| 971 | } |
| 972 | |
| 973 | Ok(Hunk { |
| 974 | old_start, |
| 975 | old_count, |
| 976 | new_start, |
| 977 | new_count, |
| 978 | lines: hunk_lines, |
| 979 | }) |
| 980 | } |
| 981 | |
| 982 | /// Parse a range like "10,5" or "10" into (start, count) |
| 983 | fn parse_range(range: &str) -> Result<(usize, usize), ToolError> { |
| 984 | let parts: Vec<&str> = range.split(',').collect(); |
| 985 | let start = parts[0].parse::<usize>().map_err(|_| { |
| 986 | ToolError::invalid_input(format!( |
| 987 | "Invalid line number `{}` in hunk header. Expected numeric unified-diff form `@@ -old_start,old_count +new_start,new_count @@` (example: `@@ -12,3 +12,5 @@`); use positive integers like `12` or `12,3`.", |
| 988 | parts[0] |
| 989 | )) |
| 990 | })?; |
| 991 | let count = if parts.len() > 1 { |
| 992 | parts[1].parse::<usize>().map_err(|_| { |
| 993 | ToolError::invalid_input(format!( |
| 994 | "Invalid line count `{}` in hunk header. Expected numeric unified-diff form `@@ -old_start,old_count +new_start,new_count @@` (example: `@@ -12,3 +12,5 @@`); use positive integers like `3`.", |
| 995 | parts[1] |
| 996 | )) |
| 997 | })? |
| 998 | } else { |
| 999 | 1 |
| 1000 | }; |
| 1001 | Ok((start, count)) |
| 1002 | } |
| 1003 | |
| 1004 | fn inspect_patch_shape(patch: &str) -> PatchShape { |
| 1005 | let mut shape = PatchShape::default(); |
| 1006 | let mut seen = HashSet::new(); |
| 1007 | let mut old_path: Option<String> = None; |
| 1008 | let mut hunk_old_remaining = 0usize; |
| 1009 | let mut hunk_new_remaining = 0usize; |
| 1010 | |
| 1011 | for line in patch.lines() { |
| 1012 | if line.starts_with("@@") { |
| 1013 | shape.has_hunks = true; |
| 1014 | if let Some((old_count, new_count)) = hunk_line_counts_for_shape(line) { |
| 1015 | hunk_old_remaining = old_count; |
| 1016 | hunk_new_remaining = new_count; |
| 1017 | } |
| 1018 | continue; |
| 1019 | } |
| 1020 | |
| 1021 | if hunk_old_remaining > 0 || hunk_new_remaining > 0 { |
| 1022 | advance_hunk_shape_counts(line, &mut hunk_old_remaining, &mut hunk_new_remaining); |
| 1023 | continue; |
| 1024 | } |
| 1025 | |
| 1026 | if let Some(stripped) = line.strip_prefix("--- ") { |
| 1027 | old_path = normalize_diff_path(stripped); |
| 1028 | continue; |
| 1029 | } |
| 1030 | |
| 1031 | if let Some(stripped) = line.strip_prefix("+++ ") { |
| 1032 | let new_path = normalize_diff_path(stripped); |
| 1033 | let resolved = new_path.or(old_path.clone()); |
| 1034 | if let Some(path) = resolved |
| 1035 | && seen.insert(path.clone()) |
| 1036 | { |
| 1037 | shape.header_files.push(path); |
| 1038 | } |
| 1039 | old_path = None; |
| 1040 | } |
| 1041 | } |
| 1042 | |
| 1043 | shape |
| 1044 | } |
| 1045 | |
| 1046 | fn hunk_line_counts_for_shape(header: &str) -> Option<(usize, usize)> { |
| 1047 | let parts: Vec<&str> = header.split_whitespace().collect(); |
| 1048 | if parts.len() < 3 { |
| 1049 | return None; |
| 1050 | } |
| 1051 | let (_, old_count) = parse_range(parts[1].trim_start_matches('-')).ok()?; |
| 1052 | let (_, new_count) = parse_range(parts[2].trim_start_matches('+')).ok()?; |
| 1053 | Some((old_count, new_count)) |
| 1054 | } |
| 1055 | |
| 1056 | fn advance_hunk_shape_counts(line: &str, old_remaining: &mut usize, new_remaining: &mut usize) { |
| 1057 | if line.starts_with('\\') { |
| 1058 | return; |
| 1059 | } |
| 1060 | if line.starts_with('+') { |
| 1061 | *new_remaining = new_remaining.saturating_sub(1); |
| 1062 | } else if line.starts_with('-') { |
| 1063 | *old_remaining = old_remaining.saturating_sub(1); |
| 1064 | } else { |
| 1065 | *old_remaining = old_remaining.saturating_sub(1); |
| 1066 | *new_remaining = new_remaining.saturating_sub(1); |
| 1067 | } |
| 1068 | } |
| 1069 | |
| 1070 | fn validate_patch_shape(shape: &PatchShape, path_override: Option<&str>) -> Result<(), ToolError> { |
| 1071 | if !shape.has_hunks { |
| 1072 | return Err(ToolError::invalid_input( |
| 1073 | "Patch must include at least one hunk header in numeric unified-diff form (`@@ -old_start,old_count +new_start,new_count @@`, example: `@@ -12,3 +12,5 @@`).", |
| 1074 | )); |
| 1075 | } |
| 1076 | |
| 1077 | match path_override { |
| 1078 | Some(_) if shape.file_count() > 1 => Err(ToolError::invalid_input(format!( |
| 1079 | "Patch references multiple files ({}) but `path` was provided. Remove `path` to apply a multi-file patch, or provide a single-file patch.", |
| 1080 | format_file_list(&shape.header_files), |
| 1081 | ))), |
| 1082 | None if shape.file_count() == 0 => Err(ToolError::invalid_input( |
| 1083 | "Patch contains hunks but no file headers (`---`/`+++`). Provide `path` or add headers.", |
| 1084 | )), |
| 1085 | _ => Ok(()), |
| 1086 | } |
| 1087 | } |
| 1088 | |
| 1089 | fn diff_header_mismatch(path_override: &str, shape: &PatchShape) -> Option<String> { |
| 1090 | if shape.file_count() != 1 { |
| 1091 | return None; |
| 1092 | } |
| 1093 | let header_path = &shape.header_files[0]; |
| 1094 | let override_norm = normalize_diff_path(path_override).unwrap_or_else(|| path_override.into()); |
| 1095 | if &override_norm == header_path { |
| 1096 | None |
| 1097 | } else { |
| 1098 | Some(format!( |
| 1099 | "Note: patch headers reference `{header_path}` but `path` overrides to `{override_norm}`." |
| 1100 | )) |
| 1101 | } |
| 1102 | } |
| 1103 | |
| 1104 | fn build_summary_message(stats: &PatchStatsExt) -> String { |
| 1105 | let mut parts = Vec::new(); |
| 1106 | if stats.stats.hunks_total > 0 { |
| 1107 | parts.push(format!( |
| 1108 | "Applied {}/{} hunks across {} file(s).", |
| 1109 | stats.stats.hunks_applied, stats.stats.hunks_total, stats.stats.files_applied |
| 1110 | )); |
| 1111 | } else { |
| 1112 | parts.push(format!( |
| 1113 | "Applied {} file change(s).", |
| 1114 | stats.stats.files_applied |
| 1115 | )); |
| 1116 | } |
| 1117 | |
| 1118 | if !stats.touched_files.is_empty() { |
| 1119 | parts.push(format!( |
| 1120 | "Files: {}.", |
| 1121 | format_file_list(&stats.touched_files) |
| 1122 | )); |
| 1123 | } |
| 1124 | |
| 1125 | if stats.stats.fuzz_used > 0 { |
| 1126 | parts.push(format!( |
| 1127 | "Fuzz used on {} hunk(s) (total fuzz: {}).", |
| 1128 | stats.stats.hunks_with_fuzz, stats.stats.fuzz_used |
| 1129 | )); |
| 1130 | } |
| 1131 | |
| 1132 | if stats.stats.hunks_relocated > 0 { |
| 1133 | parts.push(format!( |
| 1134 | "{} hunk(s) applied with stale line numbers (auto-relocated to unique context).", |
| 1135 | stats.stats.hunks_relocated |
| 1136 | )); |
| 1137 | } |
| 1138 | |
| 1139 | if let Some(note) = stats.header_path_mismatch.as_deref() { |
| 1140 | parts.push(note.to_string()); |
| 1141 | } |
| 1142 | |
| 1143 | parts.join(" ") |
| 1144 | } |
| 1145 | |
| 1146 | fn format_file_list(files: &[String]) -> String { |
| 1147 | if files.is_empty() { |
| 1148 | return "<none>".to_string(); |
| 1149 | } |
| 1150 | let mut shown: Vec<String> = files.iter().take(FILE_LIST_LIMIT).cloned().collect(); |
| 1151 | let remaining = files.len().saturating_sub(shown.len()); |
| 1152 | if remaining > 0 { |
| 1153 | shown.push(format!("... (+{remaining} more)")); |
| 1154 | } |
| 1155 | shown.join(", ") |
| 1156 | } |
| 1157 | |
| 1158 | fn push_unique(target: &mut Vec<String>, value: String) { |
| 1159 | if !target.iter().any(|existing| existing == &value) { |
| 1160 | target.push(value); |
| 1161 | } |
| 1162 | } |
| 1163 | |
| 1164 | fn build_pending_writes_from_replace( |
| 1165 | changes: &[Value], |
| 1166 | source_field: &str, |
| 1167 | context: &ToolContext, |
| 1168 | ) -> Result<(Vec<PendingWrite>, PatchStatsExt), ToolError> { |
| 1169 | let mut pending = Vec::new(); |
| 1170 | let mut stats = PatchStatsExt::default(); |
| 1171 | for change in changes { |
| 1172 | let path = change |
| 1173 | .get("path") |
| 1174 | .and_then(Value::as_str) |
| 1175 | .ok_or_else(|| ToolError::missing_field(format!("{source_field}[].path")))?; |
| 1176 | let content = change |
| 1177 | .get("content") |
| 1178 | .and_then(Value::as_str) |
| 1179 | .ok_or_else(|| ToolError::missing_field(format!("{source_field}[].content")))?; |
| 1180 | |
| 1181 | let resolved = context.resolve_path(path)?; |
| 1182 | let original = if resolved.exists() { |
| 1183 | Some(read_file_content(&resolved)?) |
| 1184 | } else { |
| 1185 | None |
| 1186 | }; |
| 1187 | let created = original.is_none(); |
| 1188 | |
| 1189 | pending.push(PendingWrite { |
| 1190 | path: resolved, |
| 1191 | content: Some(content.to_string()), |
| 1192 | original, |
| 1193 | }); |
| 1194 | |
| 1195 | stats.stats.files_total += 1; |
| 1196 | stats.stats.files_applied += 1; |
| 1197 | push_unique(&mut stats.touched_files, path.to_string()); |
| 1198 | stats.file_summaries.push(FileSummary { |
| 1199 | path: path.to_string(), |
| 1200 | hunks: 0, |
| 1201 | hunks_applied: 0, |
| 1202 | fuzz_used: 0, |
| 1203 | hunks_with_fuzz: 0, |
| 1204 | hunks_relocated: 0, |
| 1205 | created, |
| 1206 | deleted: false, |
| 1207 | }); |
| 1208 | } |
| 1209 | |
| 1210 | Ok((pending, stats)) |
| 1211 | } |
| 1212 | |
| 1213 | fn build_pending_writes_from_patches( |
| 1214 | file_patches: Vec<FilePatch>, |
| 1215 | context: &ToolContext, |
| 1216 | fuzz: usize, |
| 1217 | ) -> Result<(Vec<PendingWrite>, PatchStatsExt), ToolError> { |
| 1218 | let mut pending = Vec::new(); |
| 1219 | let mut stats = PatchStatsExt::default(); |
| 1220 | stats.stats.files_total = file_patches.len(); |
| 1221 | |
| 1222 | for file_patch in file_patches { |
| 1223 | if file_patch.hunks.is_empty() { |
| 1224 | return Err(ToolError::invalid_input(format!( |
| 1225 | "Patch section for `{}` has no hunks (`@@ ... @@`).", |
| 1226 | file_patch.path |
| 1227 | ))); |
| 1228 | } |
| 1229 | |
| 1230 | let resolved = context.resolve_path(&file_patch.path)?; |
| 1231 | let original = if resolved.exists() { |
| 1232 | Some(read_file_content(&resolved)?) |
| 1233 | } else { |
| 1234 | None |
| 1235 | }; |
| 1236 | |
| 1237 | if original.is_none() && !file_patch.create_if_missing { |
| 1238 | return Err(ToolError::execution_failed(format!( |
| 1239 | "File `{}` does not exist at `{}`. Set create_if_missing=true for new files or include headers for file creation.", |
| 1240 | file_patch.path, |
| 1241 | resolved.display(), |
| 1242 | ))); |
| 1243 | } |
| 1244 | |
| 1245 | if file_patch.delete_after && original.is_none() { |
| 1246 | return Err(ToolError::execution_failed(format!( |
| 1247 | "File `{}` does not exist at `{}` to delete.", |
| 1248 | file_patch.path, |
| 1249 | resolved.display(), |
| 1250 | ))); |
| 1251 | } |
| 1252 | |
| 1253 | let base_content = original.clone().unwrap_or_default(); |
| 1254 | let mut lines: Vec<String> = if base_content.is_empty() { |
| 1255 | Vec::new() |
| 1256 | } else { |
| 1257 | base_content.lines().map(String::from).collect() |
| 1258 | }; |
| 1259 | |
| 1260 | let apply_stats = |
| 1261 | apply_hunks_to_lines(&mut lines, &file_patch.hunks, fuzz, &file_patch.path)?; |
| 1262 | stats.stats.hunks_applied += apply_stats.hunks_applied; |
| 1263 | stats.stats.hunks_total += file_patch.hunks.len(); |
| 1264 | stats.stats.fuzz_used += apply_stats.fuzz_used; |
| 1265 | stats.stats.hunks_with_fuzz += apply_stats.hunks_with_fuzz; |
| 1266 | stats.stats.hunks_relocated += apply_stats.hunks_relocated; |
| 1267 | stats.stats.files_applied += 1; |
| 1268 | push_unique(&mut stats.touched_files, file_patch.path.clone()); |
| 1269 | stats.file_summaries.push(FileSummary { |
| 1270 | path: file_patch.path.clone(), |
| 1271 | hunks: file_patch.hunks.len(), |
| 1272 | hunks_applied: apply_stats.hunks_applied, |
| 1273 | fuzz_used: apply_stats.fuzz_used, |
| 1274 | hunks_with_fuzz: apply_stats.hunks_with_fuzz, |
| 1275 | hunks_relocated: apply_stats.hunks_relocated, |
| 1276 | created: original.is_none() && !file_patch.delete_after, |
| 1277 | deleted: file_patch.delete_after, |
| 1278 | }); |
| 1279 | |
| 1280 | if file_patch.delete_after { |
| 1281 | pending.push(PendingWrite { |
| 1282 | path: resolved, |
| 1283 | content: None, |
| 1284 | original, |
| 1285 | }); |
| 1286 | } else { |
| 1287 | let new_content = reassemble_preserving_newlines(&lines, &base_content); |
| 1288 | pending.push(PendingWrite { |
| 1289 | path: resolved, |
| 1290 | content: Some(new_content), |
| 1291 | original, |
| 1292 | }); |
| 1293 | } |
| 1294 | } |
| 1295 | |
| 1296 | Ok((pending, stats)) |
| 1297 | } |
| 1298 | |
| 1299 | /// Normalize the Rust files a patch rewrites (#6205), before the write and |
| 1300 | /// before the result's diff is built, so the rendered diff and the bytes on |
| 1301 | /// disk are the same text and the model's next anchor matches reality. |
| 1302 | async fn normalize_pending_rust(pending: &mut [PendingWrite]) { |
| 1303 | for entry in pending.iter_mut() { |
| 1304 | let (Some(content), Some(original)) = (entry.content.as_ref(), entry.original.as_ref()) |
| 1305 | else { |
| 1306 | continue; |
| 1307 | }; |
| 1308 | if let Some(normalized) = normalize_edit(&entry.path, original, content).await { |
| 1309 | entry.content = Some(normalized); |
| 1310 | } |
| 1311 | } |
| 1312 | } |
| 1313 | |
| 1314 | fn apply_pending_writes(pending: &[PendingWrite]) -> Result<(), ToolError> { |
| 1315 | // Syntax gate (#6204) ahead of the first write, not per file: a patch is |
| 1316 | // transactional, so one unparseable result must leave every file in the |
| 1317 | // patch untouched rather than half-applied and rolled back. |
| 1318 | for entry in pending { |
| 1319 | if let Some(content) = entry.content.as_ref() { |
| 1320 | let display = entry.path.display().to_string(); |
| 1321 | guard_edit(&entry.path, &display, entry.original.as_deref(), content)?; |
| 1322 | } |
| 1323 | } |
| 1324 | |
| 1325 | let mut applied = Vec::new(); |
| 1326 | |
| 1327 | for entry in pending { |
| 1328 | let result = if let Some(content) = entry.content.as_ref() { |
| 1329 | let parent_result = if let Some(parent) = entry.path.parent() { |
| 1330 | fs::create_dir_all(parent).map_err(|e| { |
| 1331 | ToolError::execution_failed(format!( |
| 1332 | "Failed to create directory {}: {}", |
| 1333 | parent.display(), |
| 1334 | e |
| 1335 | )) |
| 1336 | }) |
| 1337 | } else { |
| 1338 | Ok(()) |
| 1339 | }; |
| 1340 | |
| 1341 | parent_result.and_then(|()| { |
| 1342 | crate::utils::write_atomic_workspace(&entry.path, content.as_bytes()).map_err(|e| { |
| 1343 | ToolError::execution_failed(format!( |
| 1344 | "Failed to write {}: {}", |
| 1345 | entry.path.display(), |
| 1346 | e |
| 1347 | )) |
| 1348 | }) |
| 1349 | }) |
| 1350 | } else if entry.path.exists() { |
| 1351 | fs::remove_file(&entry.path).map_err(|e| { |
| 1352 | ToolError::execution_failed(format!( |
| 1353 | "Failed to delete {}: {}", |
| 1354 | entry.path.display(), |
| 1355 | e |
| 1356 | )) |
| 1357 | }) |
| 1358 | } else { |
| 1359 | Ok(()) |
| 1360 | }; |
| 1361 | |
| 1362 | if let Err(err) = result { |
| 1363 | rollback_pending_writes(&applied); |
| 1364 | return Err(err); |
| 1365 | } |
| 1366 | |
| 1367 | applied.push(entry.clone()); |
| 1368 | } |
| 1369 | |
| 1370 | Ok(()) |
| 1371 | } |
| 1372 | |
| 1373 | fn rollback_pending_writes(applied: &[PendingWrite]) { |
| 1374 | for entry in applied.iter().rev() { |
| 1375 | match entry.original.as_ref() { |
| 1376 | Some(content) => { |
| 1377 | let _ = crate::utils::write_atomic_workspace(&entry.path, content.as_bytes()); |
| 1378 | } |
| 1379 | None => { |
| 1380 | let _ = fs::remove_file(&entry.path); |
| 1381 | } |
| 1382 | } |
| 1383 | } |
| 1384 | } |
| 1385 | |
| 1386 | fn read_file_content(path: &PathBuf) -> Result<String, ToolError> { |
| 1387 | fs::read_to_string(path).map_err(|e| { |
| 1388 | ToolError::execution_failed(format!("Failed to read {}: {}", path.display(), e)) |
| 1389 | }) |
| 1390 | } |
| 1391 | |
| 1392 | fn preview_expected_lines(hunk: &Hunk, limit: usize) -> Vec<String> { |
| 1393 | let mut preview = Vec::new(); |
| 1394 | for line in hunk.lines.iter().filter_map(|line| match line { |
| 1395 | HunkLine::Context(s) => Some((" ", s)), |
| 1396 | HunkLine::Remove(s) => Some(("-", s)), |
| 1397 | HunkLine::Add(_) => None, |
| 1398 | }) { |
| 1399 | if preview.len() >= limit { |
| 1400 | break; |
| 1401 | } |
| 1402 | preview.push(format!(" {}{}", line.0, line.1)); |
| 1403 | } |
| 1404 | if preview.is_empty() { |
| 1405 | preview.push(" <no context lines in hunk>".to_string()); |
| 1406 | } |
| 1407 | preview |
| 1408 | } |
| 1409 | |
| 1410 | fn snippet_around(lines: &[String], line_1_based: usize, radius: usize) -> Vec<String> { |
| 1411 | if lines.is_empty() { |
| 1412 | return vec![" <empty file>".to_string()]; |
| 1413 | } |
| 1414 | |
| 1415 | let center = line_1_based |
| 1416 | .saturating_sub(1) |
| 1417 | .min(lines.len().saturating_sub(1)); |
| 1418 | let start = center.saturating_sub(radius); |
| 1419 | let end = (center + radius).min(lines.len().saturating_sub(1)); |
| 1420 | |
| 1421 | lines[start..=end] |
| 1422 | .iter() |
| 1423 | .enumerate() |
| 1424 | .map(|(idx, line)| { |
| 1425 | let line_no = start + idx + 1; |
| 1426 | format!(" {line_no:>4}: {line}") |
| 1427 | }) |
| 1428 | .collect() |
| 1429 | } |
| 1430 | |
| 1431 | fn format_hunk_no_match_error( |
| 1432 | lines: &[String], |
| 1433 | hunk: &Hunk, |
| 1434 | err: &ApplyHunkError, |
| 1435 | max_fuzz: usize, |
| 1436 | ) -> String { |
| 1437 | match err { |
| 1438 | ApplyHunkError::NoMatch { |
| 1439 | expected_line, |
| 1440 | adjusted_line, |
| 1441 | offset, |
| 1442 | } => { |
| 1443 | let expected_preview = preview_expected_lines(hunk, HUNK_PREVIEW_LINES).join("\n"); |
| 1444 | let file_preview = snippet_around(lines, *adjusted_line, SNIPPET_RADIUS).join("\n"); |
| 1445 | format!( |
| 1446 | "could not find matching context near line {expected_line} (searched around line {adjusted_line} with offset {offset:+} and fuzz up to {max_fuzz}). Expected context preview:\n{expected_preview}\nFile snippet near line {adjusted_line}:\n{file_preview}\nHints: the line numbers may be stale after earlier edits — call File with action=\"read\" to re-check the current contents, ensure the patch matches the file, increase `fuzz`, or regenerate the patch." |
| 1447 | ) |
| 1448 | } |
| 1449 | ApplyHunkError::ContextAmbiguous { |
| 1450 | expected_line, |
| 1451 | candidate_lines, |
| 1452 | } => { |
| 1453 | let candidates = candidate_lines |
| 1454 | .iter() |
| 1455 | .map(|line| line.to_string()) |
| 1456 | .collect::<Vec<_>>() |
| 1457 | .join(", "); |
| 1458 | format!( |
| 1459 | "could not find matching context near line {expected_line}: the hunk's context appears at multiple locations (lines {candidates}), and the line numbers may be stale after earlier edits, so it is not safe to relocate automatically. Hints: call File with action=\"read\" to inspect the candidate locations above, then regenerate the patch with more surrounding context lines that uniquely identify the target block." |
| 1460 | ) |
| 1461 | } |
| 1462 | } |
| 1463 | } |
| 1464 | |
| 1465 | fn apply_hunks_to_lines( |
| 1466 | lines: &mut Vec<String>, |
| 1467 | hunks: &[Hunk], |
| 1468 | fuzz: usize, |
| 1469 | file_label: &str, |
| 1470 | ) -> Result<HunkApplyStats, ToolError> { |
| 1471 | let mut stats = HunkApplyStats::default(); |
| 1472 | let mut cumulative_offset: isize = 0; |
| 1473 | |
| 1474 | for (idx, hunk) in hunks.iter().enumerate() { |
| 1475 | match apply_hunk(lines, hunk, fuzz, &mut cumulative_offset) { |
| 1476 | Ok(outcome) => { |
| 1477 | stats.hunks_applied += 1; |
| 1478 | if outcome.fuzz_used > 0 { |
| 1479 | stats.fuzz_used += outcome.fuzz_used; |
| 1480 | stats.hunks_with_fuzz += 1; |
| 1481 | } |
| 1482 | if outcome.relocated { |
| 1483 | stats.hunks_relocated += 1; |
| 1484 | } |
| 1485 | } |
| 1486 | Err(e) => { |
| 1487 | let detail = format_hunk_no_match_error(lines, hunk, &e, fuzz); |
| 1488 | return Err(ToolError::execution_failed(format!( |
| 1489 | "Failed to apply hunk {}/{} for `{}`: {}", |
| 1490 | idx + 1, |
| 1491 | hunks.len(), |
| 1492 | file_label, |
| 1493 | detail |
| 1494 | ))); |
| 1495 | } |
| 1496 | } |
| 1497 | } |
| 1498 | |
| 1499 | Ok(stats) |
| 1500 | } |
| 1501 | |
| 1502 | /// Apply a hunk to the file content with fuzzy matching |
| 1503 | fn apply_hunk( |
| 1504 | lines: &mut Vec<String>, |
| 1505 | hunk: &Hunk, |
| 1506 | max_fuzz: usize, |
| 1507 | cumulative_offset: &mut isize, |
| 1508 | ) -> Result<HunkApplyOutcome, ApplyHunkError> { |
| 1509 | // Build expected old lines from hunk |
| 1510 | let old_lines: Vec<&str> = hunk |
| 1511 | .lines |
| 1512 | .iter() |
| 1513 | .filter_map(|line| match line { |
| 1514 | HunkLine::Context(s) | HunkLine::Remove(s) => Some(s.as_str()), |
| 1515 | HunkLine::Add(_) => None, |
| 1516 | }) |
| 1517 | .collect(); |
| 1518 | |
| 1519 | // Build new lines from hunk |
| 1520 | let new_lines: Vec<String> = hunk |
| 1521 | .lines |
| 1522 | .iter() |
| 1523 | .filter_map(|line| match line { |
| 1524 | HunkLine::Context(s) | HunkLine::Add(s) => Some(s.clone()), |
| 1525 | HunkLine::Remove(_) => None, |
| 1526 | }) |
| 1527 | .collect(); |
| 1528 | |
| 1529 | // Try to find the location with fuzzy matching |
| 1530 | // Apply cumulative offset from previous hunks, clamping to valid range. |
| 1531 | let base_idx = if hunk.old_start > 0 { |
| 1532 | hunk.old_start - 1 |
| 1533 | } else { |
| 1534 | 0 |
| 1535 | }; |
| 1536 | // Use checked_add_signed to safely handle negative offsets without |
| 1537 | // risking isize overflow on adversarial input. |
| 1538 | let start_idx = base_idx |
| 1539 | .checked_add_signed(*cumulative_offset) |
| 1540 | .unwrap_or(0) |
| 1541 | .min(lines.len()); |
| 1542 | |
| 1543 | for fuzz in 0..=max_fuzz { |
| 1544 | // Try at exact position first, then nearby |
| 1545 | let search_range = if fuzz == 0 { |
| 1546 | vec![start_idx] |
| 1547 | } else { |
| 1548 | let min = start_idx.saturating_sub(fuzz); |
| 1549 | let max = (start_idx + fuzz).min(lines.len()); |
| 1550 | (min..=max).collect() |
| 1551 | }; |
| 1552 | |
| 1553 | for pos in search_range { |
| 1554 | if matches_at_position(lines, &old_lines, pos) { |
| 1555 | // Apply the hunk |
| 1556 | let end_pos = pos + old_lines.len(); |
| 1557 | lines.splice(pos..end_pos, new_lines.clone()); |
| 1558 | |
| 1559 | // Update cumulative offset: new lines added minus old lines removed |
| 1560 | let delta = new_lines.len() as isize - old_lines.len() as isize; |
| 1561 | *cumulative_offset += delta; |
| 1562 | |
| 1563 | return Ok(HunkApplyOutcome { |
| 1564 | fuzz_used: fuzz, |
| 1565 | relocated: false, |
| 1566 | }); |
| 1567 | } |
| 1568 | } |
| 1569 | } |
| 1570 | |
| 1571 | // Special case: adding to empty file or new hunk at end |
| 1572 | if old_lines.is_empty() && (lines.is_empty() || start_idx >= lines.len()) { |
| 1573 | let delta = new_lines.len() as isize; |
| 1574 | lines.extend(new_lines); |
| 1575 | *cumulative_offset += delta; |
| 1576 | return Ok(HunkApplyOutcome { |
| 1577 | fuzz_used: 0, |
| 1578 | relocated: false, |
| 1579 | }); |
| 1580 | } |
| 1581 | |
| 1582 | // #5003 — positional search failed. The line numbers are probably stale |
| 1583 | // because an earlier edit (this patch or a previous one) shifted the file |
| 1584 | // and the model regenerated the patch from outdated read_file output. |
| 1585 | // If the hunk carries enough anchor lines, look for a unique whole-file |
| 1586 | // content match and relocate there; anything that matched within `fuzz` |
| 1587 | // of `start_idx` was already tried above, so any unique match found here |
| 1588 | // is genuinely relocated. Ambiguous matches are refused (applying to the |
| 1589 | // wrong copy of a repeated block would corrupt the file). |
| 1590 | let anchor_matches: Vec<usize> = if old_lines.len() >= MIN_ANCHOR_LINES { |
| 1591 | (0..=lines.len().saturating_sub(old_lines.len())) |
| 1592 | .filter(|&pos| matches_at_position(lines, &old_lines, pos)) |
| 1593 | .collect() |
| 1594 | } else { |
| 1595 | Vec::new() |
| 1596 | }; |
| 1597 | match anchor_matches.as_slice() { |
| 1598 | [pos] => { |
| 1599 | let end_pos = pos + old_lines.len(); |
| 1600 | lines.splice(*pos..end_pos, new_lines.clone()); |
| 1601 | let delta = new_lines.len() as isize - old_lines.len() as isize; |
| 1602 | *cumulative_offset += delta; |
| 1603 | Ok(HunkApplyOutcome { |
| 1604 | fuzz_used: 0, |
| 1605 | relocated: true, |
| 1606 | }) |
| 1607 | } |
| 1608 | [] => Err(ApplyHunkError::NoMatch { |
| 1609 | expected_line: hunk.old_start, |
| 1610 | adjusted_line: start_idx + 1, // Convert back to 1-indexed |
| 1611 | offset: *cumulative_offset, |
| 1612 | }), |
| 1613 | multiple => Err(ApplyHunkError::ContextAmbiguous { |
| 1614 | expected_line: hunk.old_start, |
| 1615 | candidate_lines: multiple.iter().map(|&p| p + 1).collect(), |
| 1616 | }), |
| 1617 | } |
| 1618 | } |
| 1619 | |
| 1620 | /// Check if `old_lines` match at the given position |
| 1621 | fn matches_at_position(lines: &[String], old_lines: &[&str], pos: usize) -> bool { |
| 1622 | if pos + old_lines.len() > lines.len() { |
| 1623 | return false; |
| 1624 | } |
| 1625 | |
| 1626 | for (i, old_line) in old_lines.iter().enumerate() { |
| 1627 | // Normalize whitespace for comparison |
| 1628 | let file_line = lines[pos + i].trim_end(); |
| 1629 | let expected = old_line.trim_end(); |
| 1630 | if file_line != expected { |
| 1631 | return false; |
| 1632 | } |
| 1633 | } |
| 1634 | |
| 1635 | true |
| 1636 | } |
| 1637 | |
| 1638 | // === Unit Tests === |
| 1639 | |
| 1640 | #[cfg(test)] |
| 1641 | mod tests { |
| 1642 | use super::*; |
| 1643 | use tempfile::tempdir; |
| 1644 | |
| 1645 | fn parse_patch_result(result: ToolResult) -> PatchResult { |
| 1646 | serde_json::from_str(&result.content).expect("patch result json") |
| 1647 | } |
| 1648 | |
| 1649 | #[test] |
| 1650 | fn test_parse_range() { |
| 1651 | assert_eq!(parse_range("10,5").unwrap(), (10, 5)); |
| 1652 | assert_eq!(parse_range("10").unwrap(), (10, 1)); |
| 1653 | assert_eq!(parse_range("1,0").unwrap(), (1, 0)); |
| 1654 | } |
| 1655 | |
| 1656 | #[test] |
| 1657 | fn test_parse_unified_diff() { |
| 1658 | let patch = r"--- a/test.txt |
| 1659 | +++ b/test.txt |
| 1660 | @@ -1,3 +1,3 @@ |
| 1661 | line1 |
| 1662 | -line2 |
| 1663 | +modified line2 |
| 1664 | line3 |
| 1665 | "; |
| 1666 | |
| 1667 | let hunks = parse_unified_diff(patch).unwrap(); |
| 1668 | assert_eq!(hunks.len(), 1); |
| 1669 | assert_eq!(hunks[0].old_start, 1); |
| 1670 | assert_eq!(hunks[0].old_count, 3); |
| 1671 | assert_eq!(hunks[0].new_start, 1); |
| 1672 | assert_eq!(hunks[0].new_count, 3); |
| 1673 | } |
| 1674 | |
| 1675 | /// #6205 — a patch that lands unformatted Rust in an already-clean file is |
| 1676 | /// normalized before the write, and the rendered diff shows the normalized |
| 1677 | /// text, so the model's next patch context matches the bytes on disk. |
| 1678 | #[tokio::test] |
| 1679 | async fn patch_normalizes_rust_in_an_already_clean_file() { |
| 1680 | let tmp = tempdir().expect("tempdir"); |
| 1681 | let ctx = ToolContext::new(tmp.path().to_path_buf()); |
| 1682 | let file = tmp.path().join("clean.rs"); |
| 1683 | fs::write(&file, "fn main() {\n let x = 1;\n}\n").expect("write"); |
| 1684 | |
| 1685 | let patch = "--- a/clean.rs\n+++ b/clean.rs\n@@ -1,3 +1,4 @@\n fn main() {\n let x = 1;\n+ let y=2;\n }\n"; |
| 1686 | let result = ApplyPatchTool |
| 1687 | .execute(json!({"path": "clean.rs", "patch": patch}), &ctx) |
| 1688 | .await |
| 1689 | .expect("execute"); |
| 1690 | |
| 1691 | assert_eq!( |
| 1692 | fs::read_to_string(&file).expect("read"), |
| 1693 | "fn main() {\n let x = 1;\n let y = 2;\n}\n" |
| 1694 | ); |
| 1695 | let diff = result.metadata.as_ref().expect("metadata")["mutation"]["diff"] |
| 1696 | .as_str() |
| 1697 | .expect("diff") |
| 1698 | .to_string(); |
| 1699 | assert!(diff.contains("+ let y = 2;"), "{diff}"); |
| 1700 | assert!(!diff.contains("let y=2;"), "{diff}"); |
| 1701 | } |
| 1702 | |
| 1703 | /// #6204 — a patch whose result does not parse is refused before any file |
| 1704 | /// is written, so a multi-file patch cannot land half-applied. |
| 1705 | #[tokio::test] |
| 1706 | async fn patch_refuses_a_hunk_that_breaks_rust_syntax() { |
| 1707 | let tmp = tempdir().expect("tempdir"); |
| 1708 | let ctx = ToolContext::new(tmp.path().to_path_buf()); |
| 1709 | let file = tmp.path().join("main.rs"); |
| 1710 | let original = "fn main() {\n println!(\"hi\");\n}\n"; |
| 1711 | fs::write(&file, original).expect("write"); |
| 1712 | |
| 1713 | let patch = "--- a/main.rs\n+++ b/main.rs\n@@ -1,3 +1,2 @@\n fn main() {\n println!(\"hi\");\n-}\n"; |
| 1714 | let error = ApplyPatchTool |
| 1715 | .execute(json!({"path": "main.rs", "patch": patch}), &ctx) |
| 1716 | .await |
| 1717 | .expect_err("a patch that breaks Rust syntax must be refused"); |
| 1718 | |
| 1719 | let message = error.to_string(); |
| 1720 | assert!(message.contains("Rust syntax error at line"), "{message}"); |
| 1721 | assert_eq!( |
| 1722 | fs::read_to_string(&file).expect("read"), |
| 1723 | original, |
| 1724 | "a refused patch must leave the file untouched" |
| 1725 | ); |
| 1726 | } |
| 1727 | |
| 1728 | #[test] |
| 1729 | fn input_schema_exposes_replace_and_deprecated_changes_alias() { |
| 1730 | let schema = ApplyPatchTool.input_schema(); |
| 1731 | |
| 1732 | assert_eq!(schema["properties"]["replace"]["type"], "array"); |
| 1733 | assert_eq!(schema["properties"]["changes"]["type"], "array"); |
| 1734 | assert!( |
| 1735 | schema["properties"]["changes"]["description"] |
| 1736 | .as_str() |
| 1737 | .is_some_and(|description| description.contains("Deprecated")) |
| 1738 | ); |
| 1739 | assert_eq!( |
| 1740 | schema["oneOf"], |
| 1741 | json!([ |
| 1742 | { "required": ["patch"] }, |
| 1743 | { "required": ["replace"] }, |
| 1744 | { "required": ["changes"] } |
| 1745 | ]) |
| 1746 | ); |
| 1747 | } |
| 1748 | |
| 1749 | #[test] |
| 1750 | fn test_preflight_apply_patch_with_path_override() { |
| 1751 | let patch = r"@@ -1,2 +1,2 @@ |
| 1752 | old |
| 1753 | -value |
| 1754 | +new-value |
| 1755 | "; |
| 1756 | |
| 1757 | let preflight = preflight_apply_patch(&json!({ |
| 1758 | "path": "src/lib.rs", |
| 1759 | "patch": patch |
| 1760 | })) |
| 1761 | .expect("preflight"); |
| 1762 | |
| 1763 | assert_eq!(preflight.touched_files, vec!["src/lib.rs"]); |
| 1764 | assert_eq!(preflight.files_total, 1); |
| 1765 | assert_eq!(preflight.hunks_total, 1); |
| 1766 | assert_eq!(preflight.path_override.as_deref(), Some("src/lib.rs")); |
| 1767 | } |
| 1768 | |
| 1769 | #[test] |
| 1770 | fn test_preflight_apply_patch_multi_file_create_and_delete() { |
| 1771 | let patch = r"diff --git a/new.rs b/new.rs |
| 1772 | --- /dev/null |
| 1773 | +++ b/new.rs |
| 1774 | @@ -0,0 +1 @@ |
| 1775 | +fn added() {} |
| 1776 | diff --git a/old.rs b/old.rs |
| 1777 | --- a/old.rs |
| 1778 | +++ /dev/null |
| 1779 | @@ -1 +0,0 @@ |
| 1780 | -fn old() {} |
| 1781 | "; |
| 1782 | |
| 1783 | let preflight = preflight_apply_patch(&json!({ "patch": patch })).expect("preflight"); |
| 1784 | |
| 1785 | assert_eq!(preflight.touched_files, vec!["new.rs", "old.rs"]); |
| 1786 | assert_eq!(preflight.files_total, 2); |
| 1787 | assert_eq!(preflight.hunks_total, 2); |
| 1788 | assert_eq!(preflight.creates, vec!["new.rs"]); |
| 1789 | assert_eq!(preflight.deletes, vec!["old.rs"]); |
| 1790 | } |
| 1791 | |
| 1792 | #[test] |
| 1793 | fn test_preflight_apply_patch_timestamp_headers_strip_metadata() { |
| 1794 | let patch = "diff --git a/src/lib.rs b/src/lib.rs\n\ |
| 1795 | --- a/src/lib.rs\t2026-06-26 10:00:00 +0000\n\ |
| 1796 | +++ b/src/lib.rs\t2026-06-26 10:01:00 +0000\n\ |
| 1797 | @@ -1,1 +1,1 @@\n\ |
| 1798 | -old\n\ |
| 1799 | +new\n"; |
| 1800 | |
| 1801 | let preflight = preflight_apply_patch(&json!({ "patch": patch })).expect("preflight"); |
| 1802 | |
| 1803 | assert_eq!(preflight.touched_files, vec!["src/lib.rs"]); |
| 1804 | assert_eq!(preflight.files_total, 1); |
| 1805 | assert_eq!(preflight.hunks_total, 1); |
| 1806 | } |
| 1807 | |
| 1808 | #[test] |
| 1809 | fn test_preflight_apply_patch_ignores_forged_headers_inside_hunk_shape() { |
| 1810 | let patch = r"--- a/src/lib.rs |
| 1811 | +++ b/src/lib.rs |
| 1812 | @@ -1,3 +1,3 @@ |
| 1813 | line1 |
| 1814 | --- a/forged.rs |
| 1815 | +++ b/forged.rs |
| 1816 | line3 |
| 1817 | "; |
| 1818 | |
| 1819 | let preflight = preflight_apply_patch(&json!({ |
| 1820 | "path": "src/lib.rs", |
| 1821 | "patch": patch |
| 1822 | })) |
| 1823 | .expect("preflight"); |
| 1824 | |
| 1825 | assert_eq!(preflight.touched_files, vec!["src/lib.rs"]); |
| 1826 | assert_eq!(preflight.header_path_mismatch, None); |
| 1827 | } |
| 1828 | |
| 1829 | #[test] |
| 1830 | fn test_preflight_apply_patch_replace_list() { |
| 1831 | let canonical = preflight_apply_patch(&json!({ |
| 1832 | "replace": [ |
| 1833 | { "path": "one.txt", "content": "one" }, |
| 1834 | { "path": "two.txt", "content": "two" } |
| 1835 | ] |
| 1836 | })) |
| 1837 | .expect("preflight"); |
| 1838 | |
| 1839 | let legacy = preflight_apply_patch(&json!({ |
| 1840 | "changes": [ |
| 1841 | { "path": "one.txt", "content": "one" }, |
| 1842 | { "path": "two.txt", "content": "two" } |
| 1843 | ] |
| 1844 | })) |
| 1845 | .expect("legacy preflight"); |
| 1846 | |
| 1847 | assert_eq!(canonical.touched_files, vec!["one.txt", "two.txt"]); |
| 1848 | assert_eq!(canonical.files_total, 2); |
| 1849 | assert_eq!(canonical.hunks_total, 0); |
| 1850 | assert_eq!(legacy, canonical); |
| 1851 | } |
| 1852 | |
| 1853 | #[test] |
| 1854 | fn test_preflight_replace_files_total_counts_entries() { |
| 1855 | let preflight = preflight_apply_patch(&json!({ |
| 1856 | "replace": [ |
| 1857 | { "path": "same.txt", "content": "one" }, |
| 1858 | { "path": "same.txt", "content": "two" } |
| 1859 | ] |
| 1860 | })) |
| 1861 | .expect("preflight"); |
| 1862 | |
| 1863 | assert_eq!(preflight.touched_files, vec!["same.txt"]); |
| 1864 | assert_eq!(preflight.files_total, 2); |
| 1865 | } |
| 1866 | |
| 1867 | #[test] |
| 1868 | fn test_preflight_patch_files_total_counts_sections() { |
| 1869 | let patch = r"diff --git a/same.txt b/same.txt |
| 1870 | --- a/same.txt |
| 1871 | +++ b/same.txt |
| 1872 | @@ -1,1 +1,1 @@ |
| 1873 | -one |
| 1874 | +two |
| 1875 | diff --git a/same.txt b/same.txt |
| 1876 | --- a/same.txt |
| 1877 | +++ b/same.txt |
| 1878 | @@ -2,1 +2,1 @@ |
| 1879 | -three |
| 1880 | +four |
| 1881 | "; |
| 1882 | |
| 1883 | let preflight = preflight_apply_patch(&json!({ "patch": patch })).expect("preflight"); |
| 1884 | |
| 1885 | assert_eq!(preflight.touched_files, vec!["same.txt"]); |
| 1886 | assert_eq!(preflight.files_total, 2); |
| 1887 | assert_eq!(preflight.hunks_total, 2); |
| 1888 | } |
| 1889 | |
| 1890 | #[test] |
| 1891 | fn test_apply_hunk_simple() { |
| 1892 | let mut lines = vec![ |
| 1893 | "line1".to_string(), |
| 1894 | "line2".to_string(), |
| 1895 | "line3".to_string(), |
| 1896 | ]; |
| 1897 | |
| 1898 | let hunk = Hunk { |
| 1899 | old_start: 1, |
| 1900 | old_count: 3, |
| 1901 | new_start: 1, |
| 1902 | new_count: 3, |
| 1903 | lines: vec![ |
| 1904 | HunkLine::Context("line1".to_string()), |
| 1905 | HunkLine::Remove("line2".to_string()), |
| 1906 | HunkLine::Add("modified".to_string()), |
| 1907 | HunkLine::Context("line3".to_string()), |
| 1908 | ], |
| 1909 | }; |
| 1910 | |
| 1911 | let mut offset: isize = 0; |
| 1912 | let outcome = apply_hunk(&mut lines, &hunk, 0, &mut offset).unwrap(); |
| 1913 | assert_eq!(outcome.fuzz_used, 0); |
| 1914 | assert!(!outcome.relocated); |
| 1915 | assert_eq!(lines, vec!["line1", "modified", "line3"]); |
| 1916 | } |
| 1917 | |
| 1918 | #[test] |
| 1919 | fn test_apply_hunk_with_fuzz() { |
| 1920 | let mut lines = vec![ |
| 1921 | "line0".to_string(), |
| 1922 | "line1".to_string(), |
| 1923 | "line2".to_string(), |
| 1924 | "line3".to_string(), |
| 1925 | ]; |
| 1926 | |
| 1927 | // Hunk expects to start at line 1, but content is at line 2 |
| 1928 | let hunk = Hunk { |
| 1929 | old_start: 1, // Wrong position |
| 1930 | old_count: 2, |
| 1931 | new_start: 1, |
| 1932 | new_count: 2, |
| 1933 | lines: vec![ |
| 1934 | HunkLine::Remove("line1".to_string()), |
| 1935 | HunkLine::Add("modified".to_string()), |
| 1936 | HunkLine::Context("line2".to_string()), |
| 1937 | ], |
| 1938 | }; |
| 1939 | |
| 1940 | let mut offset: isize = 0; |
| 1941 | let outcome = apply_hunk(&mut lines, &hunk, 3, &mut offset).unwrap(); |
| 1942 | assert!(outcome.fuzz_used > 0); |
| 1943 | assert!(!outcome.relocated); |
| 1944 | assert_eq!(lines, vec!["line0", "modified", "line2", "line3"]); |
| 1945 | } |
| 1946 | |
| 1947 | #[test] |
| 1948 | fn test_apply_hunk_no_match_returns_error() { |
| 1949 | let mut lines = vec!["line1".to_string(), "line2".to_string()]; |
| 1950 | let hunk = Hunk { |
| 1951 | old_start: 5, |
| 1952 | old_count: 1, |
| 1953 | new_start: 5, |
| 1954 | new_count: 1, |
| 1955 | lines: vec![ |
| 1956 | HunkLine::Context("missing".to_string()), |
| 1957 | HunkLine::Add("new".to_string()), |
| 1958 | ], |
| 1959 | }; |
| 1960 | |
| 1961 | let mut offset: isize = 0; |
| 1962 | let err = apply_hunk(&mut lines, &hunk, 0, &mut offset).unwrap_err(); |
| 1963 | assert!(matches!( |
| 1964 | err, |
| 1965 | ApplyHunkError::NoMatch { |
| 1966 | expected_line: 5, |
| 1967 | .. |
| 1968 | } |
| 1969 | )); |
| 1970 | } |
| 1971 | |
| 1972 | #[tokio::test] |
| 1973 | async fn test_apply_patch_tool() { |
| 1974 | let tmp = tempdir().expect("tempdir"); |
| 1975 | let ctx = ToolContext::new(tmp.path().to_path_buf()); |
| 1976 | |
| 1977 | // Create a test file |
| 1978 | fs::write(tmp.path().join("test.txt"), "line1\nline2\nline3\n").expect("write"); |
| 1979 | |
| 1980 | let patch = r"--- a/test.txt |
| 1981 | +++ b/test.txt |
| 1982 | @@ -1,3 +1,3 @@ |
| 1983 | line1 |
| 1984 | -line2 |
| 1985 | +modified |
| 1986 | line3 |
| 1987 | "; |
| 1988 | |
| 1989 | let tool = ApplyPatchTool; |
| 1990 | let result = tool |
| 1991 | .execute(json!({"path": "test.txt", "patch": patch}), &ctx) |
| 1992 | .await |
| 1993 | .expect("execute"); |
| 1994 | |
| 1995 | assert!(result.success); |
| 1996 | assert_eq!( |
| 1997 | result.metadata.as_ref().unwrap()["event"], |
| 1998 | "apply_patch.preflight" |
| 1999 | ); |
| 2000 | assert_eq!( |
| 2001 | result.metadata.as_ref().unwrap()["touched_files"], |
| 2002 | json!(["test.txt"]) |
| 2003 | ); |
| 2004 | assert!( |
| 2005 | result |
| 2006 | .metadata |
| 2007 | .as_ref() |
| 2008 | .unwrap() |
| 2009 | .get("header_path_mismatch") |
| 2010 | .is_none() |
| 2011 | ); |
| 2012 | assert!( |
| 2013 | result |
| 2014 | .metadata |
| 2015 | .as_ref() |
| 2016 | .unwrap() |
| 2017 | .get("path_override") |
| 2018 | .is_some() |
| 2019 | ); |
| 2020 | let mutation = &result.metadata.as_ref().unwrap()["mutation"]; |
| 2021 | assert_eq!( |
| 2022 | mutation["files"], |
| 2023 | json!([{ "path": "test.txt", "outcome": "updated" }]) |
| 2024 | ); |
| 2025 | assert!( |
| 2026 | mutation["diff"] |
| 2027 | .as_str() |
| 2028 | .is_some_and(|diff| diff.contains("-line2") && diff.contains("+modified")), |
| 2029 | "{mutation}" |
| 2030 | ); |
| 2031 | let patch_result = parse_patch_result(result); |
| 2032 | assert_eq!(patch_result.touched_files, vec!["test.txt"]); |
| 2033 | assert_eq!(patch_result.hunks_applied, 1); |
| 2034 | |
| 2035 | // Verify the patch was applied |
| 2036 | let content = fs::read_to_string(tmp.path().join("test.txt")).expect("read"); |
| 2037 | assert!(content.contains("modified")); |
| 2038 | assert!(!content.contains("line2")); |
| 2039 | // Regression: the file's trailing newline must survive the patch. |
| 2040 | assert!(content.ends_with('\n'), "trailing newline was dropped"); |
| 2041 | } |
| 2042 | |
| 2043 | #[test] |
| 2044 | fn reassemble_preserving_newlines_keeps_style() { |
| 2045 | let lines = vec!["a".to_string(), "b".to_string()]; |
| 2046 | // LF with trailing newline. |
| 2047 | assert_eq!(reassemble_preserving_newlines(&lines, "x\ny\n"), "a\nb\n"); |
| 2048 | // LF without trailing newline. |
| 2049 | assert_eq!(reassemble_preserving_newlines(&lines, "x\ny"), "a\nb"); |
| 2050 | // CRLF is preserved (endings and trailing). |
| 2051 | assert_eq!( |
| 2052 | reassemble_preserving_newlines(&lines, "x\r\ny\r\n"), |
| 2053 | "a\r\nb\r\n" |
| 2054 | ); |
| 2055 | // New/empty file gets a conventional trailing newline. |
| 2056 | assert_eq!(reassemble_preserving_newlines(&lines, ""), "a\nb\n"); |
| 2057 | // Empty result stays empty. |
| 2058 | assert_eq!(reassemble_preserving_newlines(&[], "x\n"), ""); |
| 2059 | } |
| 2060 | |
| 2061 | #[tokio::test] |
| 2062 | async fn apply_patch_preserves_crlf_line_endings() { |
| 2063 | let tmp = tempdir().expect("tempdir"); |
| 2064 | let ctx = ToolContext::new(tmp.path().to_path_buf()); |
| 2065 | fs::write(tmp.path().join("crlf.txt"), "line1\r\nline2\r\nline3\r\n").expect("write"); |
| 2066 | let patch = |
| 2067 | "--- a/crlf.txt\n+++ b/crlf.txt\n@@ -1,3 +1,3 @@\n line1\n-line2\n+modified\n line3\n"; |
| 2068 | let result = ApplyPatchTool |
| 2069 | .execute(json!({"path": "crlf.txt", "patch": patch}), &ctx) |
| 2070 | .await |
| 2071 | .expect("execute"); |
| 2072 | assert!(result.success); |
| 2073 | let content = fs::read_to_string(tmp.path().join("crlf.txt")).expect("read"); |
| 2074 | assert!(content.contains("modified")); |
| 2075 | // Regression: a CRLF file must not be flipped to LF. |
| 2076 | assert!( |
| 2077 | content.contains("\r\n"), |
| 2078 | "CRLF was flipped to LF: {content:?}" |
| 2079 | ); |
| 2080 | assert!(!content.contains("\n\n"), "spurious bare LF introduced"); |
| 2081 | assert!(content.ends_with("\r\n"), "trailing CRLF dropped"); |
| 2082 | } |
| 2083 | |
| 2084 | #[tokio::test] |
| 2085 | async fn test_apply_patch_add_lines() { |
| 2086 | let tmp = tempdir().expect("tempdir"); |
| 2087 | let ctx = ToolContext::new(tmp.path().to_path_buf()); |
| 2088 | |
| 2089 | fs::write(tmp.path().join("test.txt"), "line1\nline3\n").expect("write"); |
| 2090 | |
| 2091 | let patch = r"@@ -1,2 +1,3 @@ |
| 2092 | line1 |
| 2093 | +line2 |
| 2094 | line3 |
| 2095 | "; |
| 2096 | |
| 2097 | let tool = ApplyPatchTool; |
| 2098 | let result = tool |
| 2099 | .execute(json!({"path": "test.txt", "patch": patch}), &ctx) |
| 2100 | .await |
| 2101 | .expect("execute"); |
| 2102 | |
| 2103 | assert!(result.success); |
| 2104 | let mutation = &result.metadata.as_ref().expect("metadata")["mutation"]; |
| 2105 | assert_eq!( |
| 2106 | mutation["files"], |
| 2107 | json!([{ "path": "test.txt", "outcome": "updated" }]) |
| 2108 | ); |
| 2109 | assert!( |
| 2110 | mutation["diff"] |
| 2111 | .as_str() |
| 2112 | .is_some_and(|diff| diff.contains("+line2")), |
| 2113 | "{mutation}" |
| 2114 | ); |
| 2115 | let patch_result = parse_patch_result(result); |
| 2116 | assert_eq!(patch_result.touched_files, vec!["test.txt"]); |
| 2117 | |
| 2118 | let content = fs::read_to_string(tmp.path().join("test.txt")).expect("read"); |
| 2119 | assert!(content.contains("line2")); |
| 2120 | } |
| 2121 | |
| 2122 | #[tokio::test] |
| 2123 | async fn test_apply_patch_create_new_file() { |
| 2124 | let tmp = tempdir().expect("tempdir"); |
| 2125 | let ctx = ToolContext::new(tmp.path().to_path_buf()); |
| 2126 | |
| 2127 | let patch = r"@@ -0,0 +1,3 @@ |
| 2128 | +line1 |
| 2129 | +line2 |
| 2130 | +line3 |
| 2131 | "; |
| 2132 | |
| 2133 | let tool = ApplyPatchTool; |
| 2134 | let result = tool |
| 2135 | .execute( |
| 2136 | json!({"path": "new_file.txt", "patch": patch, "create_if_missing": true}), |
| 2137 | &ctx, |
| 2138 | ) |
| 2139 | .await |
| 2140 | .expect("execute"); |
| 2141 | |
| 2142 | assert!(result.success); |
| 2143 | let mutation = &result.metadata.as_ref().expect("metadata")["mutation"]; |
| 2144 | assert_eq!( |
| 2145 | mutation["files"], |
| 2146 | json!([{ "path": "new_file.txt", "outcome": "created" }]) |
| 2147 | ); |
| 2148 | assert!( |
| 2149 | mutation["diff"] |
| 2150 | .as_str() |
| 2151 | .is_some_and(|diff| diff.contains("+line1")), |
| 2152 | "{mutation}" |
| 2153 | ); |
| 2154 | let patch_result = parse_patch_result(result); |
| 2155 | assert_eq!(patch_result.touched_files, vec!["new_file.txt"]); |
| 2156 | assert!(patch_result.file_summaries.first().unwrap().created); |
| 2157 | assert!(tmp.path().join("new_file.txt").exists()); |
| 2158 | } |
| 2159 | |
| 2160 | #[tokio::test] |
| 2161 | async fn test_apply_patch_replace_list() { |
| 2162 | let tmp = tempdir().expect("tempdir"); |
| 2163 | let ctx = ToolContext::new(tmp.path().to_path_buf()); |
| 2164 | |
| 2165 | fs::write(tmp.path().join("one.txt"), "old\n").expect("write"); |
| 2166 | |
| 2167 | let tool = ApplyPatchTool; |
| 2168 | let result = tool |
| 2169 | .execute( |
| 2170 | json!({ |
| 2171 | "replace": [ |
| 2172 | { "path": "one.txt", "content": "new\n" }, |
| 2173 | { "path": "two.txt", "content": "second\n" } |
| 2174 | ] |
| 2175 | }), |
| 2176 | &ctx, |
| 2177 | ) |
| 2178 | .await |
| 2179 | .expect("execute"); |
| 2180 | |
| 2181 | assert!(result.success); |
| 2182 | let metadata = result.metadata.as_ref().expect("metadata"); |
| 2183 | assert_eq!(metadata["event"], "apply_patch.preflight"); |
| 2184 | assert_eq!(metadata["touched_files"], json!(["one.txt", "two.txt"])); |
| 2185 | assert_eq!(metadata["files_total"], 2); |
| 2186 | assert_eq!(metadata["hunks_total"], 0); |
| 2187 | assert!(metadata.get("path_override").is_none()); |
| 2188 | assert_eq!( |
| 2189 | metadata["mutation"]["files"], |
| 2190 | json!([ |
| 2191 | { "path": "one.txt", "outcome": "updated" }, |
| 2192 | { "path": "two.txt", "outcome": "created" } |
| 2193 | ]) |
| 2194 | ); |
| 2195 | let mutation_diff = metadata["mutation"]["diff"] |
| 2196 | .as_str() |
| 2197 | .expect("mutation diff"); |
| 2198 | assert!(mutation_diff.contains("diff --git a/one.txt b/one.txt")); |
| 2199 | assert!(mutation_diff.contains("diff --git a/two.txt b/two.txt")); |
| 2200 | assert!(mutation_diff.contains("--- a/one.txt"), "{mutation_diff}"); |
| 2201 | assert!(mutation_diff.contains("+++ b/two.txt"), "{mutation_diff}"); |
| 2202 | let patch_result = parse_patch_result(result); |
| 2203 | let mut touched = patch_result.touched_files.clone(); |
| 2204 | touched.sort(); |
| 2205 | assert_eq!(touched, vec!["one.txt", "two.txt"]); |
| 2206 | assert_eq!(patch_result.hunks_total, 0); |
| 2207 | assert_eq!( |
| 2208 | fs::read_to_string(tmp.path().join("one.txt")).unwrap(), |
| 2209 | "new\n" |
| 2210 | ); |
| 2211 | assert_eq!( |
| 2212 | fs::read_to_string(tmp.path().join("two.txt")).unwrap(), |
| 2213 | "second\n" |
| 2214 | ); |
| 2215 | } |
| 2216 | |
| 2217 | #[tokio::test] |
| 2218 | async fn test_apply_patch_legacy_changes_list() { |
| 2219 | let tmp = tempdir().expect("tempdir"); |
| 2220 | let ctx = ToolContext::new(tmp.path().to_path_buf()); |
| 2221 | fs::write(tmp.path().join("legacy.txt"), "old\n").expect("write"); |
| 2222 | |
| 2223 | let result = ApplyPatchTool |
| 2224 | .execute( |
| 2225 | json!({ |
| 2226 | "changes": [ |
| 2227 | { "path": "legacy.txt", "content": "new\n" } |
| 2228 | ] |
| 2229 | }), |
| 2230 | &ctx, |
| 2231 | ) |
| 2232 | .await |
| 2233 | .expect("legacy changes alias should execute"); |
| 2234 | |
| 2235 | assert!(result.success); |
| 2236 | assert_eq!( |
| 2237 | fs::read_to_string(tmp.path().join("legacy.txt")).unwrap(), |
| 2238 | "new\n" |
| 2239 | ); |
| 2240 | } |
| 2241 | |
| 2242 | #[tokio::test] |
| 2243 | async fn apply_patch_rejects_every_mixed_mode_before_writing() { |
| 2244 | let tmp = tempdir().expect("tempdir"); |
| 2245 | let ctx = ToolContext::new(tmp.path().to_path_buf()); |
| 2246 | fs::write(tmp.path().join("guard.txt"), "old\n").expect("write"); |
| 2247 | let patch = "--- a/guard.txt\n+++ b/guard.txt\n@@ -1 +1 @@\n-old\n+patched\n"; |
| 2248 | let replacement = json!([{ |
| 2249 | "path": "guard.txt", |
| 2250 | "content": "replaced\n" |
| 2251 | }]); |
| 2252 | let cases = [ |
| 2253 | ( |
| 2254 | ["patch", "replace"], |
| 2255 | json!({"patch": patch, "replace": replacement.clone()}), |
| 2256 | ), |
| 2257 | ( |
| 2258 | ["patch", "changes"], |
| 2259 | json!({"patch": patch, "changes": replacement.clone()}), |
| 2260 | ), |
| 2261 | ( |
| 2262 | ["replace", "changes"], |
| 2263 | json!({ |
| 2264 | "replace": replacement.clone(), |
| 2265 | "changes": replacement.clone() |
| 2266 | }), |
| 2267 | ), |
| 2268 | ]; |
| 2269 | |
| 2270 | for (fields, input) in cases { |
| 2271 | let err = ApplyPatchTool |
| 2272 | .execute(input, &ctx) |
| 2273 | .await |
| 2274 | .expect_err("mixed modes must be rejected"); |
| 2275 | let ToolError::InvalidInput { message } = err else { |
| 2276 | panic!("mixed modes should be invalid input, got: {err}"); |
| 2277 | }; |
| 2278 | assert!(message.contains("simultaneously"), "{message}"); |
| 2279 | for field in fields { |
| 2280 | assert!(message.contains(field), "{message}"); |
| 2281 | } |
| 2282 | assert_eq!( |
| 2283 | fs::read_to_string(tmp.path().join("guard.txt")).unwrap(), |
| 2284 | "old\n", |
| 2285 | "mixed modes must be rejected before the first write" |
| 2286 | ); |
| 2287 | } |
| 2288 | } |
| 2289 | |
| 2290 | #[tokio::test] |
| 2291 | async fn test_apply_patch_replace_list_rolls_back_on_write_failure() { |
| 2292 | let tmp = tempdir().expect("tempdir"); |
| 2293 | let ctx = ToolContext::new(tmp.path().to_path_buf()); |
| 2294 | |
| 2295 | fs::write(tmp.path().join("one.txt"), "old\n").expect("write"); |
| 2296 | fs::write(tmp.path().join("blocked"), "not a dir\n").expect("write blocker"); |
| 2297 | |
| 2298 | let tool = ApplyPatchTool; |
| 2299 | let err = tool |
| 2300 | .execute( |
| 2301 | json!({ |
| 2302 | "replace": [ |
| 2303 | { "path": "one.txt", "content": "new\n" }, |
| 2304 | { "path": "blocked/two.txt", "content": "second\n" } |
| 2305 | ] |
| 2306 | }), |
| 2307 | &ctx, |
| 2308 | ) |
| 2309 | .await |
| 2310 | .expect_err("second write should fail"); |
| 2311 | |
| 2312 | let message = err.to_string(); |
| 2313 | assert!(message.contains("blocked"), "{message}"); |
| 2314 | assert_eq!( |
| 2315 | fs::read_to_string(tmp.path().join("one.txt")).unwrap(), |
| 2316 | "old\n" |
| 2317 | ); |
| 2318 | assert!(!tmp.path().join("blocked").join("two.txt").exists()); |
| 2319 | } |
| 2320 | |
| 2321 | #[tokio::test] |
| 2322 | async fn test_apply_patch_multi_file_diff() { |
| 2323 | let tmp = tempdir().expect("tempdir"); |
| 2324 | let ctx = ToolContext::new(tmp.path().to_path_buf()); |
| 2325 | |
| 2326 | fs::write(tmp.path().join("a.txt"), "line1\nline2\n").expect("write"); |
| 2327 | fs::write(tmp.path().join("b.txt"), "alpha\nbeta\n").expect("write"); |
| 2328 | |
| 2329 | let patch = r"diff --git a/a.txt b/a.txt |
| 2330 | --- a/a.txt |
| 2331 | +++ b/a.txt |
| 2332 | @@ -1,2 +1,2 @@ |
| 2333 | line1 |
| 2334 | -line2 |
| 2335 | +line2-mod |
| 2336 | diff --git a/b.txt b/b.txt |
| 2337 | --- a/b.txt |
| 2338 | +++ b/b.txt |
| 2339 | @@ -1,2 +1,3 @@ |
| 2340 | alpha |
| 2341 | +beta2 |
| 2342 | beta |
| 2343 | "; |
| 2344 | |
| 2345 | let tool = ApplyPatchTool; |
| 2346 | let result = tool |
| 2347 | .execute(json!({"patch": patch}), &ctx) |
| 2348 | .await |
| 2349 | .expect("execute"); |
| 2350 | |
| 2351 | assert!(result.success); |
| 2352 | let metadata = result.metadata.as_ref().expect("metadata"); |
| 2353 | assert_eq!(metadata["event"], "apply_patch.preflight"); |
| 2354 | assert_eq!(metadata["touched_files"], json!(["a.txt", "b.txt"])); |
| 2355 | assert_eq!(metadata["files_total"], 2); |
| 2356 | assert_eq!(metadata["hunks_total"], 2); |
| 2357 | assert!(metadata.get("path_override").is_none()); |
| 2358 | let patch_result = parse_patch_result(result); |
| 2359 | let mut touched = patch_result.touched_files.clone(); |
| 2360 | touched.sort(); |
| 2361 | assert_eq!(touched, vec!["a.txt", "b.txt"]); |
| 2362 | assert_eq!(patch_result.files_applied, 2); |
| 2363 | let a = fs::read_to_string(tmp.path().join("a.txt")).unwrap(); |
| 2364 | let b = fs::read_to_string(tmp.path().join("b.txt")).unwrap(); |
| 2365 | assert!(a.contains("line2-mod")); |
| 2366 | assert!(b.contains("beta2")); |
| 2367 | } |
| 2368 | |
| 2369 | #[tokio::test] |
| 2370 | async fn mutation_receipt_covers_delete_rename_and_multifile_outcomes() { |
| 2371 | let tmp = tempdir().expect("tempdir"); |
| 2372 | let ctx = ToolContext::new(tmp.path().to_path_buf()); |
| 2373 | fs::write(tmp.path().join("old.txt"), "same\n").expect("old"); |
| 2374 | fs::write(tmp.path().join("update.txt"), "before\n").expect("update"); |
| 2375 | fs::write(tmp.path().join("delete.txt"), "gone\n").expect("delete"); |
| 2376 | |
| 2377 | let patch = r"diff --git a/old.txt b/old.txt |
| 2378 | --- a/old.txt |
| 2379 | +++ /dev/null |
| 2380 | @@ -1 +0,0 @@ |
| 2381 | -same |
| 2382 | diff --git a/new.txt b/new.txt |
| 2383 | --- /dev/null |
| 2384 | +++ b/new.txt |
| 2385 | @@ -0,0 +1 @@ |
| 2386 | +same |
| 2387 | diff --git a/update.txt b/update.txt |
| 2388 | --- a/update.txt |
| 2389 | +++ b/update.txt |
| 2390 | @@ -1 +1 @@ |
| 2391 | -before |
| 2392 | +after |
| 2393 | diff --git a/create.txt b/create.txt |
| 2394 | --- /dev/null |
| 2395 | +++ b/create.txt |
| 2396 | @@ -0,0 +1 @@ |
| 2397 | +fresh |
| 2398 | diff --git a/delete.txt b/delete.txt |
| 2399 | --- a/delete.txt |
| 2400 | +++ /dev/null |
| 2401 | @@ -1 +0,0 @@ |
| 2402 | -gone |
| 2403 | "; |
| 2404 | |
| 2405 | let result = ApplyPatchTool |
| 2406 | .execute(json!({"patch": patch}), &ctx) |
| 2407 | .await |
| 2408 | .expect("execute"); |
| 2409 | let mutation = &result.metadata.as_ref().expect("metadata")["mutation"]; |
| 2410 | assert_eq!( |
| 2411 | mutation["files"], |
| 2412 | json!([ |
| 2413 | { "path": "update.txt", "outcome": "updated" }, |
| 2414 | { "path": "create.txt", "outcome": "created" }, |
| 2415 | { "path": "delete.txt", "outcome": "deleted" } |
| 2416 | ]) |
| 2417 | ); |
| 2418 | assert_eq!( |
| 2419 | mutation["renames"], |
| 2420 | json!([{ "from": "old.txt", "to": "new.txt" }]) |
| 2421 | ); |
| 2422 | let exact = mutation["diff"].as_str().expect("exact mutation diff"); |
| 2423 | assert!(exact.contains("rename from old.txt"), "{exact}"); |
| 2424 | assert!(exact.contains("rename to new.txt"), "{exact}"); |
| 2425 | assert!(exact.contains("--- a/update.txt"), "{exact}"); |
| 2426 | assert!(exact.contains("+++ b/create.txt"), "{exact}"); |
| 2427 | assert!(exact.contains("--- a/delete.txt"), "{exact}"); |
| 2428 | |
| 2429 | assert!(!tmp.path().join("old.txt").exists()); |
| 2430 | assert_eq!( |
| 2431 | fs::read_to_string(tmp.path().join("new.txt")).expect("renamed target"), |
| 2432 | "same\n" |
| 2433 | ); |
| 2434 | assert_eq!( |
| 2435 | fs::read_to_string(tmp.path().join("update.txt")).expect("updated"), |
| 2436 | "after\n" |
| 2437 | ); |
| 2438 | assert!(tmp.path().join("create.txt").exists()); |
| 2439 | assert!(!tmp.path().join("delete.txt").exists()); |
| 2440 | } |
| 2441 | |
| 2442 | #[tokio::test] |
| 2443 | async fn test_apply_patch_requires_headers_without_path() { |
| 2444 | let tmp = tempdir().expect("tempdir"); |
| 2445 | let ctx = ToolContext::new(tmp.path().to_path_buf()); |
| 2446 | let tool = ApplyPatchTool; |
| 2447 | |
| 2448 | let patch = r"@@ -1,1 +1,1 @@ |
| 2449 | -old |
| 2450 | +new |
| 2451 | "; |
| 2452 | |
| 2453 | let err = tool |
| 2454 | .execute(json!({"patch": patch}), &ctx) |
| 2455 | .await |
| 2456 | .unwrap_err(); |
| 2457 | match err { |
| 2458 | ToolError::InvalidInput { message } => { |
| 2459 | assert!(message.contains("no file headers")); |
| 2460 | assert!(message.contains("Provide `path`")); |
| 2461 | } |
| 2462 | other => panic!("expected invalid input, got: {other}"), |
| 2463 | } |
| 2464 | } |
| 2465 | |
| 2466 | #[tokio::test] |
| 2467 | async fn test_path_override_rejects_multi_file_diff() { |
| 2468 | let tmp = tempdir().expect("tempdir"); |
| 2469 | let ctx = ToolContext::new(tmp.path().to_path_buf()); |
| 2470 | let tool = ApplyPatchTool; |
| 2471 | |
| 2472 | let patch = r"diff --git a/a.txt b/a.txt |
| 2473 | --- a/a.txt |
| 2474 | +++ b/a.txt |
| 2475 | @@ -1,1 +1,1 @@ |
| 2476 | -one |
| 2477 | +one-mod |
| 2478 | diff --git a/b.txt b/b.txt |
| 2479 | --- a/b.txt |
| 2480 | +++ b/b.txt |
| 2481 | @@ -1,1 +1,1 @@ |
| 2482 | -two |
| 2483 | +two-mod |
| 2484 | "; |
| 2485 | |
| 2486 | let err = tool |
| 2487 | .execute(json!({"path": "a.txt", "patch": patch}), &ctx) |
| 2488 | .await |
| 2489 | .unwrap_err(); |
| 2490 | match err { |
| 2491 | ToolError::InvalidInput { message } => { |
| 2492 | assert!(message.contains("multiple files")); |
| 2493 | assert!(message.contains("a.txt")); |
| 2494 | assert!(message.contains("b.txt")); |
| 2495 | } |
| 2496 | other => panic!("expected invalid input, got: {other}"), |
| 2497 | } |
| 2498 | } |
| 2499 | |
| 2500 | #[tokio::test] |
| 2501 | async fn test_apply_patch_summary_reports_fuzz() { |
| 2502 | let tmp = tempdir().expect("tempdir"); |
| 2503 | let ctx = ToolContext::new(tmp.path().to_path_buf()); |
| 2504 | let tool = ApplyPatchTool; |
| 2505 | |
| 2506 | fs::write(tmp.path().join("test.txt"), "line0\nline1\nline2\nline3\n").expect("write"); |
| 2507 | |
| 2508 | let patch = r"@@ -1,2 +1,2 @@ |
| 2509 | -line1 |
| 2510 | +modified |
| 2511 | line2 |
| 2512 | "; |
| 2513 | |
| 2514 | let result = tool |
| 2515 | .execute(json!({"path": "test.txt", "patch": patch, "fuzz": 3}), &ctx) |
| 2516 | .await |
| 2517 | .expect("execute"); |
| 2518 | assert!(result.success); |
| 2519 | let patch_result = parse_patch_result(result); |
| 2520 | assert_eq!(patch_result.hunks_with_fuzz, 1); |
| 2521 | assert!(patch_result.fuzz_used > 0); |
| 2522 | assert!(patch_result.message.contains("Fuzz used")); |
| 2523 | let summary = patch_result.file_summaries.first().unwrap(); |
| 2524 | assert_eq!(summary.hunks_with_fuzz, 1); |
| 2525 | } |
| 2526 | |
| 2527 | #[tokio::test] |
| 2528 | async fn test_path_override_header_mismatch_note() { |
| 2529 | let tmp = tempdir().expect("tempdir"); |
| 2530 | let ctx = ToolContext::new(tmp.path().to_path_buf()); |
| 2531 | let tool = ApplyPatchTool; |
| 2532 | |
| 2533 | fs::write(tmp.path().join("override.txt"), "old\n").expect("write"); |
| 2534 | |
| 2535 | let patch = r"--- a/other.txt |
| 2536 | +++ b/other.txt |
| 2537 | @@ -1,1 +1,1 @@ |
| 2538 | -old |
| 2539 | +new |
| 2540 | "; |
| 2541 | |
| 2542 | let result = tool |
| 2543 | .execute(json!({"path": "override.txt", "patch": patch}), &ctx) |
| 2544 | .await |
| 2545 | .expect("execute"); |
| 2546 | let metadata = result.metadata.as_ref().expect("metadata"); |
| 2547 | assert!( |
| 2548 | metadata["header_path_mismatch"] |
| 2549 | .as_str() |
| 2550 | .unwrap() |
| 2551 | .contains("headers reference `other.txt`") |
| 2552 | ); |
| 2553 | let patch_result = parse_patch_result(result); |
| 2554 | assert!( |
| 2555 | patch_result |
| 2556 | .message |
| 2557 | .contains("headers reference `other.txt`") |
| 2558 | ); |
| 2559 | assert!( |
| 2560 | patch_result |
| 2561 | .message |
| 2562 | .contains("path` overrides to `override.txt`") |
| 2563 | ); |
| 2564 | } |
| 2565 | |
| 2566 | #[test] |
| 2567 | fn test_apply_patch_tool_properties() { |
| 2568 | let tool = ApplyPatchTool; |
| 2569 | assert_eq!(tool.name(), "apply_patch"); |
| 2570 | assert!(!tool.is_read_only()); |
| 2571 | assert!(tool.is_sandboxable()); |
| 2572 | assert_eq!(tool.approval_requirement(), ApprovalRequirement::Suggest); |
| 2573 | } |
| 2574 | |
| 2575 | #[test] |
| 2576 | fn test_multi_hunk_offset_tracking() { |
| 2577 | // File with 6 lines |
| 2578 | let mut lines: Vec<String> = vec![ |
| 2579 | "line1".to_string(), |
| 2580 | "line2".to_string(), |
| 2581 | "line3".to_string(), |
| 2582 | "line4".to_string(), |
| 2583 | "line5".to_string(), |
| 2584 | "line6".to_string(), |
| 2585 | ]; |
| 2586 | |
| 2587 | // Hunk 1: Add 2 lines after line1 (offset becomes +2) |
| 2588 | let hunk1 = Hunk { |
| 2589 | old_start: 1, |
| 2590 | old_count: 2, |
| 2591 | new_start: 1, |
| 2592 | new_count: 4, |
| 2593 | lines: vec![ |
| 2594 | HunkLine::Context("line1".to_string()), |
| 2595 | HunkLine::Add("new_a".to_string()), |
| 2596 | HunkLine::Add("new_b".to_string()), |
| 2597 | HunkLine::Context("line2".to_string()), |
| 2598 | ], |
| 2599 | }; |
| 2600 | |
| 2601 | // Hunk 2: Modify line5 (originally at position 5, now at position 7 due to +2 offset) |
| 2602 | let hunk2 = Hunk { |
| 2603 | old_start: 5, // Original position in the diff |
| 2604 | old_count: 1, |
| 2605 | new_start: 7, |
| 2606 | new_count: 1, |
| 2607 | lines: vec![ |
| 2608 | HunkLine::Remove("line5".to_string()), |
| 2609 | HunkLine::Add("modified5".to_string()), |
| 2610 | ], |
| 2611 | }; |
| 2612 | |
| 2613 | let mut offset: isize = 0; |
| 2614 | |
| 2615 | // Apply first hunk |
| 2616 | let outcome1 = apply_hunk(&mut lines, &hunk1, 3, &mut offset).unwrap(); |
| 2617 | assert_eq!(outcome1.fuzz_used, 0); |
| 2618 | assert!(!outcome1.relocated); |
| 2619 | assert_eq!(offset, 2); // Added 2 lines (4 new - 2 old) |
| 2620 | assert_eq!( |
| 2621 | lines, |
| 2622 | vec![ |
| 2623 | "line1", "new_a", "new_b", "line2", "line3", "line4", "line5", "line6" |
| 2624 | ] |
| 2625 | ); |
| 2626 | |
| 2627 | // Apply second hunk - this would fail without offset tracking! |
| 2628 | let outcome2 = apply_hunk(&mut lines, &hunk2, 3, &mut offset).unwrap(); |
| 2629 | assert_eq!(outcome2.fuzz_used, 0); |
| 2630 | assert!(!outcome2.relocated); |
| 2631 | assert!(lines.contains(&"modified5".to_string())); |
| 2632 | assert!(!lines.contains(&"line5".to_string())); |
| 2633 | } |
| 2634 | |
| 2635 | #[test] |
| 2636 | fn test_apply_hunk_relocates_to_unique_context_match() { |
| 2637 | // #5003 - stale line numbers (hunk says line 1, content is at line 20). |
| 2638 | let mut lines: Vec<String> = (0..25).map(|i| format!("line{i}")).collect(); |
| 2639 | let hunk = Hunk { |
| 2640 | old_start: 1, // stale line number |
| 2641 | old_count: 5, |
| 2642 | new_start: 1, |
| 2643 | new_count: 5, |
| 2644 | lines: vec![ |
| 2645 | HunkLine::Context("line19".to_string()), |
| 2646 | HunkLine::Context("line20".to_string()), |
| 2647 | HunkLine::Remove("line21".to_string()), |
| 2648 | HunkLine::Add("line21-modified".to_string()), |
| 2649 | HunkLine::Context("line22".to_string()), |
| 2650 | HunkLine::Context("line23".to_string()), |
| 2651 | ], |
| 2652 | }; |
| 2653 | |
| 2654 | let mut offset: isize = 0; |
| 2655 | let outcome = apply_hunk(&mut lines, &hunk, 1, &mut offset).unwrap(); |
| 2656 | assert!( |
| 2657 | outcome.relocated, |
| 2658 | "expected relocation for stale line numbers" |
| 2659 | ); |
| 2660 | assert_eq!(outcome.fuzz_used, 0); |
| 2661 | assert_eq!(lines[21], "line21-modified"); |
| 2662 | assert!(!lines.contains(&"line21".to_string())); |
| 2663 | } |
| 2664 | |
| 2665 | #[test] |
| 2666 | fn test_apply_hunk_ambiguous_context_reports_candidates() { |
| 2667 | // Two identical block-a..d blocks: anchor is not unique -> ContextAmbiguous. |
| 2668 | let mut lines: Vec<String> = [ |
| 2669 | "header0", "header1", "header2", "block-a", "block-b", "block-c", "block-d", "middle", |
| 2670 | "block-a", "block-b", "block-c", "block-d", "footer", |
| 2671 | ] |
| 2672 | .iter() |
| 2673 | .map(|s| s.to_string()) |
| 2674 | .collect(); |
| 2675 | |
| 2676 | let hunk = Hunk { |
| 2677 | old_start: 1, |
| 2678 | old_count: 4, |
| 2679 | new_start: 1, |
| 2680 | new_count: 4, |
| 2681 | lines: vec![ |
| 2682 | HunkLine::Remove("block-a".to_string()), |
| 2683 | HunkLine::Remove("block-b".to_string()), |
| 2684 | HunkLine::Remove("block-c".to_string()), |
| 2685 | HunkLine::Remove("block-d".to_string()), |
| 2686 | HunkLine::Add("block-a-modified".to_string()), |
| 2687 | HunkLine::Add("block-b".to_string()), |
| 2688 | HunkLine::Add("block-c".to_string()), |
| 2689 | HunkLine::Add("block-d".to_string()), |
| 2690 | ], |
| 2691 | }; |
| 2692 | |
| 2693 | let mut offset: isize = 0; |
| 2694 | let err = apply_hunk(&mut lines, &hunk, 1, &mut offset).unwrap_err(); |
| 2695 | match err { |
| 2696 | ApplyHunkError::ContextAmbiguous { |
| 2697 | expected_line, |
| 2698 | candidate_lines, |
| 2699 | } => { |
| 2700 | assert_eq!(expected_line, 1); |
| 2701 | // The two duplicate blocks start at 1-based lines 4 and 9. |
| 2702 | assert_eq!(candidate_lines, vec![4, 9]); |
| 2703 | } |
| 2704 | other => panic!("expected ContextAmbiguous, got: {other:?}"), |
| 2705 | } |
| 2706 | } |
| 2707 | |
| 2708 | #[test] |
| 2709 | fn test_apply_hunk_short_anchor_not_relocated() { |
| 2710 | // Anchor too short (1 line < MIN_ANCHOR_LINES): keep NoMatch behavior. |
| 2711 | let mut lines: Vec<String> = ["zero", "one", "two", "three", "four"] |
| 2712 | .iter() |
| 2713 | .map(|s| s.to_string()) |
| 2714 | .collect(); |
| 2715 | let hunk = Hunk { |
| 2716 | old_start: 1, |
| 2717 | old_count: 1, |
| 2718 | new_start: 1, |
| 2719 | new_count: 1, |
| 2720 | lines: vec![ |
| 2721 | HunkLine::Remove("four".to_string()), |
| 2722 | HunkLine::Add("four-modified".to_string()), |
| 2723 | ], |
| 2724 | }; |
| 2725 | |
| 2726 | let mut offset: isize = 0; |
| 2727 | let err = apply_hunk(&mut lines, &hunk, 1, &mut offset).unwrap_err(); |
| 2728 | assert!( |
| 2729 | matches!(err, ApplyHunkError::NoMatch { .. }), |
| 2730 | "short anchors must not be relocated" |
| 2731 | ); |
| 2732 | } |
| 2733 | |
| 2734 | #[tokio::test] |
| 2735 | async fn test_apply_patch_relocates_stale_line_numbers() { |
| 2736 | // #5003 - integration: hunk claims line 1 but content is at line 21. |
| 2737 | let tmp = tempdir().expect("tempdir"); |
| 2738 | let ctx = ToolContext::new(tmp.path().to_path_buf()); |
| 2739 | let tool = ApplyPatchTool; |
| 2740 | |
| 2741 | let content = (0..30) |
| 2742 | .map(|i| format!("line{i}")) |
| 2743 | .collect::<Vec<_>>() |
| 2744 | .join("\n") |
| 2745 | + "\n"; |
| 2746 | fs::write(tmp.path().join("stale.txt"), &content).expect("write"); |
| 2747 | |
| 2748 | let patch = r"@@ -1,5 +1,5 @@ |
| 2749 | line19 |
| 2750 | line20 |
| 2751 | -line21 |
| 2752 | +line21-modified |
| 2753 | line22 |
| 2754 | line23 |
| 2755 | "; |
| 2756 | |
| 2757 | let result = tool |
| 2758 | .execute( |
| 2759 | json!({"path": "stale.txt", "patch": patch, "fuzz": 1}), |
| 2760 | &ctx, |
| 2761 | ) |
| 2762 | .await |
| 2763 | .expect("execute"); |
| 2764 | assert!(result.success); |
| 2765 | let patch_result = parse_patch_result(result); |
| 2766 | assert_eq!(patch_result.hunks_relocated, 1); |
| 2767 | assert!( |
| 2768 | patch_result.message.contains("stale line numbers"), |
| 2769 | "message: {}", |
| 2770 | patch_result.message |
| 2771 | ); |
| 2772 | let summary = patch_result.file_summaries.first().unwrap(); |
| 2773 | assert_eq!(summary.hunks_relocated, 1); |
| 2774 | |
| 2775 | let edited = fs::read_to_string(tmp.path().join("stale.txt")).expect("read"); |
| 2776 | assert!(edited.contains("line21-modified")); |
| 2777 | assert!(!edited.contains("\nline21\n")); |
| 2778 | } |
| 2779 | |
| 2780 | #[tokio::test] |
| 2781 | async fn test_apply_patch_ambiguous_context_reports_candidates() { |
| 2782 | // #5003 - integration: duplicate context blocks, ambiguous relocation. |
| 2783 | let tmp = tempdir().expect("tempdir"); |
| 2784 | let ctx = ToolContext::new(tmp.path().to_path_buf()); |
| 2785 | let tool = ApplyPatchTool; |
| 2786 | |
| 2787 | let content = "header0\nheader1\nheader2\nblock-a\nblock-b\nblock-c\nblock-d\nmiddle\nblock-a\nblock-b\nblock-c\nblock-d\nfooter\n"; |
| 2788 | fs::write(tmp.path().join("dup.txt"), content).expect("write"); |
| 2789 | |
| 2790 | let patch = r"@@ -1,4 +1,4 @@ |
| 2791 | -block-a |
| 2792 | -block-b |
| 2793 | -block-c |
| 2794 | -block-d |
| 2795 | +block-a-modified |
| 2796 | +block-b |
| 2797 | +block-c |
| 2798 | +block-d |
| 2799 | "; |
| 2800 | |
| 2801 | let err = tool |
| 2802 | .execute(json!({"path": "dup.txt", "patch": patch, "fuzz": 1}), &ctx) |
| 2803 | .await |
| 2804 | .unwrap_err(); |
| 2805 | let message = err.to_string(); |
| 2806 | assert!( |
| 2807 | message.contains("multiple locations"), |
| 2808 | "expected ambiguity error, got: {message}" |
| 2809 | ); |
| 2810 | // The two duplicate blocks start at 1-based lines 4 and 9. |
| 2811 | assert!( |
| 2812 | message.contains("lines 4, 9"), |
| 2813 | "expected candidate lines, got: {message}" |
| 2814 | ); |
| 2815 | let unchanged = fs::read_to_string(tmp.path().join("dup.txt")).expect("read"); |
| 2816 | assert_eq!( |
| 2817 | unchanged, content, |
| 2818 | "ambiguous hunk must not modify the file" |
| 2819 | ); |
| 2820 | } |
| 2821 | #[tokio::test] |
| 2822 | async fn test_apply_patch_relocates_after_crlf_chinese_edit_shift() { |
| 2823 | // #5003 - issue scenario: a C-style file with CRLF line endings and |
| 2824 | // Chinese comments. A first edit shifts line numbers; a second patch |
| 2825 | // still uses stale line numbers but its context is unique in the |
| 2826 | // file, so apply_patch relocates instead of failing. |
| 2827 | let tmp = tempdir().expect("tempdir"); |
| 2828 | let ctx = ToolContext::new(tmp.path().to_path_buf()); |
| 2829 | let tool = ApplyPatchTool; |
| 2830 | |
| 2831 | let mut lines = vec!["/* 扭矩控制模块 */".to_string()]; |
| 2832 | for i in 0..60 { |
| 2833 | lines.push(format!("int cfg_{i} = {i}; // 配置项 {i}")); |
| 2834 | } |
| 2835 | let content = lines.join("\r\n") + "\r\n"; |
| 2836 | fs::write(tmp.path().join("app_foc.c"), &content).expect("write"); |
| 2837 | |
| 2838 | // First edit: insert 8 lines right after cfg_9 (old line 11). This |
| 2839 | // shifts every later line number by +8. |
| 2840 | let patch1 = r"@@ -11,1 +11,9 @@ |
| 2841 | int cfg_9 = 9; // 配置项 9 |
| 2842 | +int new_a = 100; // 新增配置 A |
| 2843 | +int new_b = 101; // 新增配置 B |
| 2844 | +int new_c = 102; // 新增配置 C |
| 2845 | +int new_d = 103; // 新增配置 D |
| 2846 | +int new_e = 104; // 新增配置 E |
| 2847 | +int new_f = 105; // 新增配置 F |
| 2848 | +int new_g = 106; // 新增配置 G |
| 2849 | +int new_h = 107; // 新增配置 H |
| 2850 | "; |
| 2851 | let r1 = tool |
| 2852 | .execute( |
| 2853 | json!({"path": "app_foc.c", "patch": patch1, "fuzz": 0}), |
| 2854 | &ctx, |
| 2855 | ) |
| 2856 | .await |
| 2857 | .expect("first patch"); |
| 2858 | assert!(r1.success, "first patch should apply: {}", r1.content); |
| 2859 | |
| 2860 | // Second edit: claims line 31 (stale - cfg_21 now), but the replaced |
| 2861 | // block cfg_30..cfg_34 actually lives at lines 40-44. Positional |
| 2862 | // search with fuzz=1 fails; the unique whole-file anchor relocates. |
| 2863 | let patch2 = r"@@ -31,5 +31,5 @@ |
| 2864 | int cfg_30 = 30; // 配置项 30 |
| 2865 | int cfg_31 = 31; // 配置项 31 |
| 2866 | -int cfg_32 = 32; // 配置项 32 |
| 2867 | +int cfg_32 = 3200; // 配置项 32 已修改 |
| 2868 | int cfg_33 = 33; // 配置项 33 |
| 2869 | int cfg_34 = 34; // 配置项 34 |
| 2870 | "; |
| 2871 | let r2 = tool |
| 2872 | .execute( |
| 2873 | json!({"path": "app_foc.c", "patch": patch2, "fuzz": 1}), |
| 2874 | &ctx, |
| 2875 | ) |
| 2876 | .await |
| 2877 | .expect("second patch"); |
| 2878 | assert!(r2.success, "second patch should relocate: {}", r2.content); |
| 2879 | let pr2 = parse_patch_result(r2); |
| 2880 | assert_eq!(pr2.hunks_relocated, 1, "second patch must be relocated"); |
| 2881 | |
| 2882 | let edited = fs::read_to_string(tmp.path().join("app_foc.c")).expect("read"); |
| 2883 | assert!( |
| 2884 | edited.contains("int cfg_32 = 3200;"), |
| 2885 | "replacement must land" |
| 2886 | ); |
| 2887 | assert!( |
| 2888 | edited.contains("int cfg_33 = 33;"), |
| 2889 | "context after the edit must stay intact" |
| 2890 | ); |
| 2891 | assert!( |
| 2892 | edited.contains("int new_h = 107;"), |
| 2893 | "first edit must survive" |
| 2894 | ); |
| 2895 | } |
| 2896 | |
| 2897 | // === Content-hash patch guard (#3979) === |
| 2898 | // |
| 2899 | // The guard is a whole-patch precondition on the patch target, not a |
| 2900 | // per-file check — see `verify_patch_expected_hash`. These pin that it |
| 2901 | // fires before anything is written. |
| 2902 | |
| 2903 | const GUARD_PATCH: &str = "--- a/test.txt |
| 2904 | +++ b/test.txt |
| 2905 | @@ -1,3 +1,3 @@ |
| 2906 | line1 |
| 2907 | -line2 |
| 2908 | +modified |
| 2909 | line3 |
| 2910 | "; |
| 2911 | |
| 2912 | #[tokio::test] |
| 2913 | async fn patch_with_matching_expected_hash_proceeds() { |
| 2914 | let tmp = tempdir().expect("tempdir"); |
| 2915 | let ctx = ToolContext::new(tmp.path().to_path_buf()); |
| 2916 | let body = "line1\nline2\nline3\n"; |
| 2917 | fs::write(tmp.path().join("test.txt"), body).expect("write"); |
| 2918 | |
| 2919 | ApplyPatchTool |
| 2920 | .execute( |
| 2921 | json!({ |
| 2922 | "path": "test.txt", |
| 2923 | "patch": GUARD_PATCH, |
| 2924 | "expected_hash": content_hash(body.as_bytes()), |
| 2925 | }), |
| 2926 | &ctx, |
| 2927 | ) |
| 2928 | .await |
| 2929 | .expect("matching hash must not block the patch"); |
| 2930 | |
| 2931 | let updated = fs::read_to_string(tmp.path().join("test.txt")).expect("read"); |
| 2932 | assert!(updated.contains("modified"), "{updated}"); |
| 2933 | } |
| 2934 | |
| 2935 | #[tokio::test] |
| 2936 | async fn patch_with_stale_expected_hash_rejects_without_writing() { |
| 2937 | let tmp = tempdir().expect("tempdir"); |
| 2938 | let ctx = ToolContext::new(tmp.path().to_path_buf()); |
| 2939 | let body = "line1\nline2\nline3\n"; |
| 2940 | fs::write(tmp.path().join("test.txt"), body).expect("write"); |
| 2941 | |
| 2942 | let err = ApplyPatchTool |
| 2943 | .execute( |
| 2944 | json!({ |
| 2945 | "path": "test.txt", |
| 2946 | "patch": GUARD_PATCH, |
| 2947 | "expected_hash": content_hash(b"a different file entirely\n"), |
| 2948 | }), |
| 2949 | &ctx, |
| 2950 | ) |
| 2951 | .await |
| 2952 | .expect_err("stale hash must reject"); |
| 2953 | |
| 2954 | let message = err.to_string(); |
| 2955 | assert!(message.contains("changed since it was read"), "{message}"); |
| 2956 | assert!(message.contains("nothing was written"), "{message}"); |
| 2957 | assert_eq!( |
| 2958 | fs::read_to_string(tmp.path().join("test.txt")).expect("read"), |
| 2959 | body, |
| 2960 | "a rejected patch must not modify the file" |
| 2961 | ); |
| 2962 | } |
| 2963 | |
| 2964 | #[tokio::test] |
| 2965 | async fn patch_without_expected_hash_is_unchanged() { |
| 2966 | let tmp = tempdir().expect("tempdir"); |
| 2967 | let ctx = ToolContext::new(tmp.path().to_path_buf()); |
| 2968 | fs::write(tmp.path().join("test.txt"), "line1\nline2\nline3\n").expect("write"); |
| 2969 | |
| 2970 | ApplyPatchTool |
| 2971 | .execute(json!({ "path": "test.txt", "patch": GUARD_PATCH }), &ctx) |
| 2972 | .await |
| 2973 | .expect("absent expected_hash keeps the pre-#3979 behavior"); |
| 2974 | |
| 2975 | let updated = fs::read_to_string(tmp.path().join("test.txt")).expect("read"); |
| 2976 | assert!(updated.contains("modified"), "{updated}"); |
| 2977 | } |
| 2978 | |
| 2979 | #[tokio::test] |
| 2980 | async fn multi_file_patch_guards_the_first_file_and_writes_nothing_on_mismatch() { |
| 2981 | let tmp = tempdir().expect("tempdir"); |
| 2982 | let ctx = ToolContext::new(tmp.path().to_path_buf()); |
| 2983 | fs::write(tmp.path().join("one.txt"), "one\n").expect("write"); |
| 2984 | fs::write(tmp.path().join("two.txt"), "two\n").expect("write"); |
| 2985 | |
| 2986 | let patch = "diff --git a/one.txt b/one.txt |
| 2987 | --- a/one.txt |
| 2988 | +++ b/one.txt |
| 2989 | @@ -1 +1 @@ |
| 2990 | -one |
| 2991 | +ONE |
| 2992 | diff --git a/two.txt b/two.txt |
| 2993 | --- a/two.txt |
| 2994 | +++ b/two.txt |
| 2995 | @@ -1 +1 @@ |
| 2996 | -two |
| 2997 | +TWO |
| 2998 | "; |
| 2999 | |
| 3000 | let err = ApplyPatchTool |
| 3001 | .execute( |
| 3002 | json!({ |
| 3003 | "patch": patch, |
| 3004 | "expected_hash": content_hash(b"stale\n"), |
| 3005 | }), |
| 3006 | &ctx, |
| 3007 | ) |
| 3008 | .await |
| 3009 | .expect_err("stale hash on the first file must reject the whole patch"); |
| 3010 | assert!(err.to_string().contains("one.txt"), "{err}"); |
| 3011 | |
| 3012 | // Transactional: the unguarded second file must not have been |
| 3013 | // written either. |
| 3014 | assert_eq!( |
| 3015 | fs::read_to_string(tmp.path().join("one.txt")).unwrap(), |
| 3016 | "one\n" |
| 3017 | ); |
| 3018 | assert_eq!( |
| 3019 | fs::read_to_string(tmp.path().join("two.txt")).unwrap(), |
| 3020 | "two\n" |
| 3021 | ); |
| 3022 | } |
| 3023 | |
| 3024 | #[tokio::test] |
| 3025 | async fn patch_expected_hash_on_a_missing_target_fails_closed() { |
| 3026 | let tmp = tempdir().expect("tempdir"); |
| 3027 | let ctx = ToolContext::new(tmp.path().to_path_buf()); |
| 3028 | |
| 3029 | let err = ApplyPatchTool |
| 3030 | .execute( |
| 3031 | json!({ |
| 3032 | "path": "absent.txt", |
| 3033 | "patch": GUARD_PATCH, |
| 3034 | "create_if_missing": true, |
| 3035 | "expected_hash": content_hash(b"anything"), |
| 3036 | }), |
| 3037 | &ctx, |
| 3038 | ) |
| 3039 | .await |
| 3040 | .expect_err("guarded patch of a missing file must fail closed"); |
| 3041 | |
| 3042 | assert!(err.to_string().contains("does not exist"), "{err}"); |
| 3043 | assert!(!tmp.path().join("absent.txt").exists()); |
| 3044 | } |
| 3045 | } |
| 3046 |