| 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::{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 | pub old_count: usize, |
| 125 | pub new_start: usize, |
| 126 | #[cfg_attr(not(test), expect(dead_code))] |
| 127 | pub new_count: usize, |
| 128 | pub lines: Vec<HunkLine>, |
| 129 | } |
| 130 | |
| 131 | /// A line in a hunk |
| 132 | #[derive(Debug, Clone)] |
| 133 | pub enum HunkLine { |
| 134 | Context(String), |
| 135 | Add(String), |
| 136 | Remove(String), |
| 137 | } |
| 138 | |
| 139 | /// Tool for applying unified diff patches to files |
| 140 | pub struct ApplyPatchTool; |
| 141 | |
| 142 | #[derive(Debug, Clone)] |
| 143 | struct FilePatch { |
| 144 | path: String, |
| 145 | hunks: Vec<Hunk>, |
| 146 | delete_after: bool, |
| 147 | create_if_missing: bool, |
| 148 | /// The section's old header is `/dev/null`: it creates the file and has |
| 149 | /// nothing to anchor against, so it must not land on an existing one. |
| 150 | creates_new_file: bool, |
| 151 | } |
| 152 | |
| 153 | #[derive(Debug, Clone)] |
| 154 | struct PendingWrite { |
| 155 | path: PathBuf, |
| 156 | content: Option<String>, |
| 157 | original: Option<String>, |
| 158 | } |
| 159 | |
| 160 | #[derive(Debug, Default, Clone, Copy)] |
| 161 | struct PatchStats { |
| 162 | files_applied: usize, |
| 163 | files_total: usize, |
| 164 | hunks_applied: usize, |
| 165 | hunks_total: usize, |
| 166 | fuzz_used: usize, |
| 167 | hunks_with_fuzz: usize, |
| 168 | hunks_relocated: usize, |
| 169 | } |
| 170 | |
| 171 | fn is_zero(value: &usize) -> bool { |
| 172 | *value == 0 |
| 173 | } |
| 174 | |
| 175 | #[derive(Debug, Default, Clone)] |
| 176 | struct PatchStatsExt { |
| 177 | stats: PatchStats, |
| 178 | touched_files: Vec<String>, |
| 179 | file_summaries: Vec<FileSummary>, |
| 180 | header_path_mismatch: Option<String>, |
| 181 | } |
| 182 | |
| 183 | #[derive(Debug, Default, Clone)] |
| 184 | struct PatchShape { |
| 185 | has_hunks: bool, |
| 186 | header_files: Vec<String>, |
| 187 | /// Some section's old header is `/dev/null` (a file creation). |
| 188 | creates_new_file: bool, |
| 189 | } |
| 190 | |
| 191 | impl PatchShape { |
| 192 | fn file_count(&self) -> usize { |
| 193 | self.header_files.len() |
| 194 | } |
| 195 | } |
| 196 | |
| 197 | #[derive(Debug, Default, Clone, Copy)] |
| 198 | struct HunkApplyStats { |
| 199 | hunks_applied: usize, |
| 200 | fuzz_used: usize, |
| 201 | hunks_with_fuzz: usize, |
| 202 | hunks_relocated: usize, |
| 203 | } |
| 204 | |
| 205 | /// Result of applying a single hunk: how much positional fuzz was used, and |
| 206 | /// whether the hunk had to be relocated to a unique whole-file context match |
| 207 | /// (stale line numbers after earlier edits, #5003). |
| 208 | #[derive(Debug, Default, Clone, Copy)] |
| 209 | struct HunkApplyOutcome { |
| 210 | fuzz_used: usize, |
| 211 | relocated: bool, |
| 212 | } |
| 213 | |
| 214 | #[derive(Debug, Clone)] |
| 215 | enum ApplyPatchPreflightKind { |
| 216 | Replace, |
| 217 | PathOverride { |
| 218 | path: String, |
| 219 | hunks: Vec<Hunk>, |
| 220 | creates_new_file: bool, |
| 221 | }, |
| 222 | FilePatches(Vec<FilePatch>), |
| 223 | } |
| 224 | |
| 225 | /// Canonicalized `apply_patch` payload mode. |
| 226 | /// |
| 227 | /// `replace` is the preferred spelling for full-file replacements. `changes` |
| 228 | /// remains a compatibility alias for callers that learned the original tool |
| 229 | /// schema before the clearer name was introduced. |
| 230 | #[derive(Debug, Clone, Copy)] |
| 231 | pub(crate) enum NormalizedApplyPatchInput<'a> { |
| 232 | Patch(&'a str), |
| 233 | Replacement { |
| 234 | entries: &'a [Value], |
| 235 | source_field: &'static str, |
| 236 | }, |
| 237 | } |
| 238 | |
| 239 | /// Validate mutual exclusivity and normalize the legacy `changes` alias. |
| 240 | /// |
| 241 | /// This is the single parser used by execution, preflight, policy, approval, |
| 242 | /// and UI consumers so every surface agrees on the accepted input contract. |
| 243 | pub(crate) fn normalize_apply_patch_input( |
| 244 | input: &Value, |
| 245 | ) -> Result<NormalizedApplyPatchInput<'_>, ToolError> { |
| 246 | let provided: Vec<&'static str> = ["patch", "replace", "changes"] |
| 247 | .into_iter() |
| 248 | .filter(|field| input.get(*field).is_some()) |
| 249 | .collect(); |
| 250 | |
| 251 | if provided.len() > 1 { |
| 252 | let fields = provided |
| 253 | .iter() |
| 254 | .map(|field| format!("`{field}`")) |
| 255 | .collect::<Vec<_>>() |
| 256 | .join(", "); |
| 257 | return Err(ToolError::invalid_input(format!( |
| 258 | "Cannot use {fields} simultaneously. Choose exactly one of `patch`, `replace`, or the deprecated `changes` alias." |
| 259 | ))); |
| 260 | } |
| 261 | |
| 262 | let Some(field) = provided.first().copied() else { |
| 263 | return Err(ToolError::missing_field( |
| 264 | "patch, replace, or deprecated changes", |
| 265 | )); |
| 266 | }; |
| 267 | |
| 268 | if field == "patch" { |
| 269 | let patch = input |
| 270 | .get(field) |
| 271 | .and_then(Value::as_str) |
| 272 | .ok_or_else(|| ToolError::invalid_input("`patch` must be a string"))?; |
| 273 | return Ok(NormalizedApplyPatchInput::Patch(patch)); |
| 274 | } |
| 275 | |
| 276 | let entries = input.get(field).and_then(Value::as_array).ok_or_else(|| { |
| 277 | ToolError::invalid_input(format!( |
| 278 | "`{field}` must be an array of objects like {{path, content}}" |
| 279 | )) |
| 280 | })?; |
| 281 | if entries.is_empty() { |
| 282 | return Err(ToolError::invalid_input(format!( |
| 283 | "`{field}` cannot be empty" |
| 284 | ))); |
| 285 | } |
| 286 | |
| 287 | Ok(NormalizedApplyPatchInput::Replacement { |
| 288 | entries, |
| 289 | source_field: field, |
| 290 | }) |
| 291 | } |
| 292 | |
| 293 | #[derive(Debug, Clone)] |
| 294 | struct ApplyPatchPreflightPlan { |
| 295 | summary: ApplyPatchPreflight, |
| 296 | kind: ApplyPatchPreflightKind, |
| 297 | } |
| 298 | |
| 299 | // === Errors === |
| 300 | |
| 301 | #[derive(Debug, Error)] |
| 302 | enum ApplyHunkError { |
| 303 | #[error( |
| 304 | "Failed to find matching location for hunk (expected at line {expected_line}, adjusted to {adjusted_line} with offset {offset:+})" |
| 305 | )] |
| 306 | NoMatch { |
| 307 | expected_line: usize, |
| 308 | adjusted_line: usize, |
| 309 | offset: isize, |
| 310 | }, |
| 311 | #[error( |
| 312 | "Hunk context is ambiguous: matches at multiple locations {candidate_lines:?}, expected around line {expected_line}" |
| 313 | )] |
| 314 | ContextAmbiguous { |
| 315 | expected_line: usize, |
| 316 | candidate_lines: Vec<usize>, |
| 317 | }, |
| 318 | } |
| 319 | |
| 320 | #[async_trait] |
| 321 | impl ToolSpec for ApplyPatchTool { |
| 322 | fn name(&self) -> &'static str { |
| 323 | "apply_patch" |
| 324 | } |
| 325 | |
| 326 | fn model_visible(&self) -> bool { |
| 327 | true |
| 328 | } |
| 329 | |
| 330 | fn description(&self) -> &'static str { |
| 331 | "Apply a transactional unified-diff patch across one or more files, with fuzzy context matching and a rendered diff." |
| 332 | } |
| 333 | |
| 334 | fn input_schema(&self) -> Value { |
| 335 | json!({ |
| 336 | "type": "object", |
| 337 | "properties": { |
| 338 | "path": { |
| 339 | "type": "string", |
| 340 | "description": "Path to the file to patch (relative to workspace)" |
| 341 | }, |
| 342 | "patch": { |
| 343 | "type": "string", |
| 344 | "description": "Unified diff patch content" |
| 345 | }, |
| 346 | "replace": { |
| 347 | "type": "array", |
| 348 | "description": "Optional full file replacements (path + content).", |
| 349 | "items": { |
| 350 | "type": "object", |
| 351 | "properties": { |
| 352 | "path": { "type": "string" }, |
| 353 | "content": { "type": "string" } |
| 354 | }, |
| 355 | "required": ["path", "content"] |
| 356 | } |
| 357 | }, |
| 358 | "changes": { |
| 359 | "type": "array", |
| 360 | "description": "Deprecated compatibility alias for `replace` (full file replacements by path + content).", |
| 361 | "items": { |
| 362 | "type": "object", |
| 363 | "properties": { |
| 364 | "path": { "type": "string" }, |
| 365 | "content": { "type": "string" } |
| 366 | }, |
| 367 | "required": ["path", "content"] |
| 368 | } |
| 369 | }, |
| 370 | "fuzz": { |
| 371 | "type": "integer", |
| 372 | "description": "Maximum fuzz factor for fuzzy matching (default: 3)" |
| 373 | }, |
| 374 | "create_if_missing": { |
| 375 | "type": "boolean", |
| 376 | "description": "Create the file if it doesn't exist (for new file patches)" |
| 377 | }, |
| 378 | "expected_hash": { |
| 379 | "type": "string", |
| 380 | "description": format!( |
| 381 | "{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." |
| 382 | ) |
| 383 | } |
| 384 | }, |
| 385 | "oneOf": [ |
| 386 | { "required": ["patch"] }, |
| 387 | { "required": ["replace"] }, |
| 388 | { "required": ["changes"] } |
| 389 | ] |
| 390 | }) |
| 391 | } |
| 392 | |
| 393 | fn capabilities(&self) -> Vec<ToolCapability> { |
| 394 | vec![ |
| 395 | ToolCapability::WritesFiles, |
| 396 | ToolCapability::Sandboxable, |
| 397 | ToolCapability::RequiresApproval, |
| 398 | ] |
| 399 | } |
| 400 | |
| 401 | fn approval_requirement(&self) -> ApprovalRequirement { |
| 402 | ApprovalRequirement::Suggest |
| 403 | } |
| 404 | |
| 405 | async fn execute(&self, input: Value, context: &ToolContext) -> Result<ToolResult, ToolError> { |
| 406 | let mut input = input; |
| 407 | apply_param_aliases(&mut input, PATH_ALIASES, "File patch")?; |
| 408 | PATCH_PARAMS.reject_unknown(&input)?; |
| 409 | let input = input; |
| 410 | |
| 411 | let fuzz = optional_u64(&input, "fuzz", DEFAULT_FUZZ as u64)?.min(MAX_FUZZ as u64); |
| 412 | let fuzz = usize::try_from(fuzz).unwrap_or(DEFAULT_FUZZ); |
| 413 | let normalized = normalize_apply_patch_input(&input)?; |
| 414 | let create_if_missing = optional_bool(&input, "create_if_missing", false)?; |
| 415 | let preflight = preflight_apply_patch_plan(&input, normalized)?; |
| 416 | verify_patch_expected_hash(&input, &preflight.summary, context)?; |
| 417 | |
| 418 | if let NormalizedApplyPatchInput::Replacement { |
| 419 | entries, |
| 420 | source_field, |
| 421 | } = normalized |
| 422 | { |
| 423 | let entries = entries.to_vec(); |
| 424 | let build_context = context.clone(); |
| 425 | let (mut pending, stats) = patch_blocking(move || { |
| 426 | build_pending_writes_from_replace(&entries, source_field, &build_context) |
| 427 | }) |
| 428 | .await?; |
| 429 | normalize_pending_rust(&mut pending).await; |
| 430 | let pending = apply_pending_writes_blocking(pending).await?; |
| 431 | // Resolve absolute paths for LSP diagnostics query. |
| 432 | let abs_paths: Vec<PathBuf> = pending.iter().map(|p| p.path.clone()).collect(); |
| 433 | let diag_block = lsp_diagnostics_for_paths(context, &abs_paths).await; |
| 434 | let result = PatchResult { |
| 435 | success: true, |
| 436 | files_applied: stats.stats.files_applied, |
| 437 | files_total: stats.stats.files_total, |
| 438 | hunks_applied: stats.stats.hunks_applied, |
| 439 | hunks_total: stats.stats.hunks_total, |
| 440 | fuzz_used: stats.stats.fuzz_used, |
| 441 | hunks_with_fuzz: stats.stats.hunks_with_fuzz, |
| 442 | hunks_relocated: stats.stats.hunks_relocated, |
| 443 | touched_files: stats.touched_files.clone(), |
| 444 | file_summaries: stats.file_summaries.clone(), |
| 445 | message: build_summary_message(&stats), |
| 446 | }; |
| 447 | let mut tool_result = ToolResult::json(&result) |
| 448 | .map_err(|e| ToolError::execution_failed(e.to_string()))?; |
| 449 | tool_result = tool_result.with_metadata(apply_patch_result_metadata( |
| 450 | &preflight.summary, |
| 451 | &pending, |
| 452 | &stats, |
| 453 | )); |
| 454 | if !diag_block.is_empty() { |
| 455 | tool_result.content.push('\n'); |
| 456 | tool_result.content.push_str(&diag_block); |
| 457 | } |
| 458 | return Ok(tool_result); |
| 459 | } |
| 460 | |
| 461 | let file_patches = match preflight.kind { |
| 462 | ApplyPatchPreflightKind::Replace => { |
| 463 | unreachable!("replace input returned before patch execution") |
| 464 | } |
| 465 | ApplyPatchPreflightKind::PathOverride { |
| 466 | path, |
| 467 | hunks, |
| 468 | creates_new_file, |
| 469 | } => vec![FilePatch { |
| 470 | path, |
| 471 | hunks, |
| 472 | delete_after: false, |
| 473 | create_if_missing, |
| 474 | creates_new_file, |
| 475 | }], |
| 476 | ApplyPatchPreflightKind::FilePatches(file_patches) => file_patches, |
| 477 | }; |
| 478 | |
| 479 | let build_context = context.clone(); |
| 480 | let (mut pending, mut stats) = patch_blocking(move || { |
| 481 | build_pending_writes_from_patches(file_patches, &build_context, fuzz) |
| 482 | }) |
| 483 | .await?; |
| 484 | stats.header_path_mismatch = preflight.summary.header_path_mismatch.clone(); |
| 485 | normalize_pending_rust(&mut pending).await; |
| 486 | let pending = apply_pending_writes_blocking(pending).await?; |
| 487 | // Resolve absolute paths for LSP diagnostics query. |
| 488 | let abs_paths: Vec<PathBuf> = pending |
| 489 | .iter() |
| 490 | .filter(|p| p.content.is_some()) // skip deleted files |
| 491 | .map(|p| p.path.clone()) |
| 492 | .collect(); |
| 493 | let diag_block = lsp_diagnostics_for_paths(context, &abs_paths).await; |
| 494 | let result = PatchResult { |
| 495 | success: true, |
| 496 | files_applied: stats.stats.files_applied, |
| 497 | files_total: stats.stats.files_total, |
| 498 | hunks_applied: stats.stats.hunks_applied, |
| 499 | hunks_total: stats.stats.hunks_total, |
| 500 | fuzz_used: stats.stats.fuzz_used, |
| 501 | hunks_with_fuzz: stats.stats.hunks_with_fuzz, |
| 502 | hunks_relocated: stats.stats.hunks_relocated, |
| 503 | touched_files: stats.touched_files.clone(), |
| 504 | file_summaries: stats.file_summaries.clone(), |
| 505 | message: build_summary_message(&stats), |
| 506 | }; |
| 507 | let mut tool_result = |
| 508 | ToolResult::json(&result).map_err(|e| ToolError::execution_failed(e.to_string()))?; |
| 509 | tool_result = tool_result.with_metadata(apply_patch_result_metadata( |
| 510 | &preflight.summary, |
| 511 | &pending, |
| 512 | &stats, |
| 513 | )); |
| 514 | if !diag_block.is_empty() { |
| 515 | tool_result.content.push('\n'); |
| 516 | tool_result.content.push_str(&diag_block); |
| 517 | } |
| 518 | Ok(tool_result) |
| 519 | } |
| 520 | } |
| 521 | |
| 522 | /// Enforce the optional `expected_hash` precondition for a patch (#3979). |
| 523 | /// |
| 524 | /// **This is a whole-patch precondition on a single target file, not a per-file |
| 525 | /// guard.** A unified diff carries no place to attach a hash per file section, |
| 526 | /// and `File action="patch"` flattens its arguments into one object, so there |
| 527 | /// is no clean parameter shape for a hash-per-path map without inventing a |
| 528 | /// syntax the model has never seen. The guarded file is therefore the patch's |
| 529 | /// target: the explicit `path` argument when one is given, otherwise the first |
| 530 | /// file the patch touches. A multi-file patch still verifies only that one |
| 531 | /// file; for the rest, the existing hunk-context matching remains the check |
| 532 | /// that a stale patch fails on. |
| 533 | /// |
| 534 | /// Runs before any pending write is built or applied, so a mismatch leaves |
| 535 | /// every file in the patch untouched. |
| 536 | fn verify_patch_expected_hash( |
| 537 | input: &Value, |
| 538 | summary: &ApplyPatchPreflight, |
| 539 | context: &ToolContext, |
| 540 | ) -> Result<(), ToolError> { |
| 541 | let Some(expected) = optional_str(input, "expected_hash")? else { |
| 542 | return Ok(()); |
| 543 | }; |
| 544 | |
| 545 | let Some(target) = summary |
| 546 | .path_override |
| 547 | .as_deref() |
| 548 | .or_else(|| summary.touched_files.first().map(String::as_str)) |
| 549 | else { |
| 550 | return Err(ToolError::execution_failed( |
| 551 | "File `patch` refused: expected_hash was supplied but the patch names no target file to verify it against, so nothing was written.".to_string(), |
| 552 | )); |
| 553 | }; |
| 554 | |
| 555 | let resolved = context.resolve_path(target)?; |
| 556 | if !resolved.exists() { |
| 557 | // Fail closed, matching `write`: a hash describes a file that was |
| 558 | // read, so a missing target means the guard cannot be honored. |
| 559 | return Err(ToolError::execution_failed(format!( |
| 560 | "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." |
| 561 | ))); |
| 562 | } |
| 563 | |
| 564 | let current = fs::read(&resolved).map_err(|e| { |
| 565 | ToolError::execution_failed(format!( |
| 566 | "File `patch` refused: could not read {target} to verify expected_hash ({e}); nothing was written." |
| 567 | )) |
| 568 | })?; |
| 569 | let actual = content_hash(¤t); |
| 570 | if actual == expected { |
| 571 | return Ok(()); |
| 572 | } |
| 573 | Err(ToolError::execution_failed(format!( |
| 574 | "File `patch` refused: {target} changed since it was read. \ |
| 575 | expected_hash was {expected} but the file is now {actual}, so nothing was written. \ |
| 576 | Recovery: call File with action=\"read\" path=\"{target}\" to get the current contents \ |
| 577 | and its content_hash, then rebuild the patch against them." |
| 578 | ))) |
| 579 | } |
| 580 | |
| 581 | /// Parse `apply_patch` input into a reusable, no-mutation preflight summary. |
| 582 | /// |
| 583 | /// This deliberately stops before workspace resolution or file reads. It is |
| 584 | /// suitable for policy checks, audit logs, diagnostics hooks, and future undo |
| 585 | /// planning that must know the target files before mutation. |
| 586 | pub fn preflight_apply_patch(input: &Value) -> Result<ApplyPatchPreflight, ToolError> { |
| 587 | let normalized = normalize_apply_patch_input(input)?; |
| 588 | Ok(preflight_apply_patch_plan(input, normalized)?.summary) |
| 589 | } |
| 590 | |
| 591 | fn preflight_apply_patch_plan( |
| 592 | input: &Value, |
| 593 | normalized: NormalizedApplyPatchInput<'_>, |
| 594 | ) -> Result<ApplyPatchPreflightPlan, ToolError> { |
| 595 | let create_if_missing = optional_bool(input, "create_if_missing", false)?; |
| 596 | |
| 597 | if let NormalizedApplyPatchInput::Replacement { |
| 598 | entries, |
| 599 | source_field, |
| 600 | } = normalized |
| 601 | { |
| 602 | return Ok(ApplyPatchPreflightPlan { |
| 603 | summary: preflight_replace(entries, source_field)?, |
| 604 | kind: ApplyPatchPreflightKind::Replace, |
| 605 | }); |
| 606 | } |
| 607 | |
| 608 | let NormalizedApplyPatchInput::Patch(patch_text) = normalized else { |
| 609 | unreachable!("replacement input returned before patch parsing") |
| 610 | }; |
| 611 | let path_override = optional_str(input, "path")?; |
| 612 | let patch_shape = inspect_patch_shape(patch_text); |
| 613 | validate_patch_shape(&patch_shape, path_override)?; |
| 614 | let header_path_mismatch = |
| 615 | path_override.and_then(|path| diff_header_mismatch(path, &patch_shape)); |
| 616 | |
| 617 | if let Some(path) = path_override { |
| 618 | let hunks = parse_unified_diff(patch_text)?; |
| 619 | if hunks.is_empty() { |
| 620 | return Err(ToolError::invalid_input( |
| 621 | "Patch did not contain any hunks (`@@ ... @@`). Provide a unified diff hunk.", |
| 622 | )); |
| 623 | } |
| 624 | return Ok(ApplyPatchPreflightPlan { |
| 625 | summary: ApplyPatchPreflight { |
| 626 | touched_files: vec![path.to_string()], |
| 627 | files_total: 1, |
| 628 | hunks_total: hunks.len(), |
| 629 | creates: if create_if_missing { |
| 630 | vec![path.to_string()] |
| 631 | } else { |
| 632 | Vec::new() |
| 633 | }, |
| 634 | deletes: Vec::new(), |
| 635 | path_override: Some(path.to_string()), |
| 636 | header_path_mismatch, |
| 637 | }, |
| 638 | kind: ApplyPatchPreflightKind::PathOverride { |
| 639 | path: path.to_string(), |
| 640 | hunks, |
| 641 | // `path` retargets the section; a `--- /dev/null` header |
| 642 | // still means "create", and must not land on an existing file. |
| 643 | creates_new_file: patch_shape.creates_new_file, |
| 644 | }, |
| 645 | }); |
| 646 | } |
| 647 | |
| 648 | let file_patches = parse_unified_diff_files(patch_text, create_if_missing)?; |
| 649 | if file_patches.is_empty() { |
| 650 | return Err(ToolError::invalid_input( |
| 651 | "No valid file patches found. Ensure the patch includes `---`/`+++` headers or provide `path`.", |
| 652 | )); |
| 653 | } |
| 654 | |
| 655 | let mut touched_files = Vec::new(); |
| 656 | let mut creates = Vec::new(); |
| 657 | let mut deletes = Vec::new(); |
| 658 | let mut hunks_total = 0; |
| 659 | for file_patch in &file_patches { |
| 660 | if file_patch.hunks.is_empty() { |
| 661 | return Err(ToolError::invalid_input(format!( |
| 662 | "Patch section for `{}` has no hunks (`@@ ... @@`).", |
| 663 | file_patch.path |
| 664 | ))); |
| 665 | } |
| 666 | push_unique(&mut touched_files, file_patch.path.clone()); |
| 667 | hunks_total += file_patch.hunks.len(); |
| 668 | if file_patch.create_if_missing && !file_patch.delete_after { |
| 669 | push_unique(&mut creates, file_patch.path.clone()); |
| 670 | } |
| 671 | if file_patch.delete_after { |
| 672 | push_unique(&mut deletes, file_patch.path.clone()); |
| 673 | } |
| 674 | } |
| 675 | |
| 676 | Ok(ApplyPatchPreflightPlan { |
| 677 | summary: ApplyPatchPreflight { |
| 678 | files_total: file_patches.len(), |
| 679 | touched_files, |
| 680 | hunks_total, |
| 681 | creates, |
| 682 | deletes, |
| 683 | path_override: None, |
| 684 | header_path_mismatch, |
| 685 | }, |
| 686 | kind: ApplyPatchPreflightKind::FilePatches(file_patches), |
| 687 | }) |
| 688 | } |
| 689 | |
| 690 | fn preflight_replace( |
| 691 | changes: &[Value], |
| 692 | source_field: &str, |
| 693 | ) -> Result<ApplyPatchPreflight, ToolError> { |
| 694 | let mut touched_files = Vec::new(); |
| 695 | for change in changes { |
| 696 | let path = change |
| 697 | .get("path") |
| 698 | .and_then(Value::as_str) |
| 699 | .ok_or_else(|| ToolError::missing_field(format!("{source_field}[].path")))?; |
| 700 | let _content = change |
| 701 | .get("content") |
| 702 | .and_then(Value::as_str) |
| 703 | .ok_or_else(|| ToolError::missing_field(format!("{source_field}[].content")))?; |
| 704 | push_unique(&mut touched_files, path.to_string()); |
| 705 | } |
| 706 | |
| 707 | Ok(ApplyPatchPreflight { |
| 708 | files_total: changes.len(), |
| 709 | touched_files, |
| 710 | hunks_total: 0, |
| 711 | creates: Vec::new(), |
| 712 | deletes: Vec::new(), |
| 713 | path_override: None, |
| 714 | header_path_mismatch: None, |
| 715 | }) |
| 716 | } |
| 717 | |
| 718 | fn apply_patch_result_metadata( |
| 719 | preflight: &ApplyPatchPreflight, |
| 720 | pending: &[PendingWrite], |
| 721 | stats: &PatchStatsExt, |
| 722 | ) -> Value { |
| 723 | let mut metadata = |
| 724 | serde_json::to_value(preflight).expect("ApplyPatchPreflight should serialize"); |
| 725 | if let Some(object) = metadata.as_object_mut() { |
| 726 | object.insert("event".to_string(), json!("apply_patch.preflight")); |
| 727 | object.insert( |
| 728 | "mutation".to_string(), |
| 729 | build_mutation_metadata(pending, &stats.file_summaries), |
| 730 | ); |
| 731 | } |
| 732 | metadata |
| 733 | } |
| 734 | |
| 735 | /// Preserve the exact applied before/after diff independently from approval |
| 736 | /// presentation. The TUI consumes this success-only metadata for its calm |
| 737 | /// File receipt; the normal model-facing result remains compact JSON. |
| 738 | fn build_mutation_metadata(pending: &[PendingWrite], summaries: &[FileSummary]) -> Value { |
| 739 | let mut matched = HashSet::new(); |
| 740 | let mut renames = Vec::new(); |
| 741 | |
| 742 | for (delete_index, (deleted, delete_summary)) in pending.iter().zip(summaries).enumerate() { |
| 743 | if !delete_summary.deleted || matched.contains(&delete_index) { |
| 744 | continue; |
| 745 | } |
| 746 | let Some(old_content) = deleted.original.as_deref() else { |
| 747 | continue; |
| 748 | }; |
| 749 | let Some((create_index, (created, create_summary))) = pending |
| 750 | .iter() |
| 751 | .zip(summaries) |
| 752 | .enumerate() |
| 753 | .find(|(index, (created, summary))| { |
| 754 | !matched.contains(index) |
| 755 | && summary.created |
| 756 | && created.content.as_deref() == Some(old_content) |
| 757 | }) |
| 758 | else { |
| 759 | continue; |
| 760 | }; |
| 761 | matched.insert(delete_index); |
| 762 | matched.insert(create_index); |
| 763 | // The destination's bytes are the ones this call wrote, so a rename |
| 764 | // carries the same size/sha256 facts as a created or updated file. |
| 765 | let mut rename = json!({ |
| 766 | "from": delete_summary.path, |
| 767 | "to": create_summary.path, |
| 768 | }); |
| 769 | if let Some(content) = created.content.as_deref() { |
| 770 | rename["size"] = json!(content.len()); |
| 771 | rename["sha256"] = json!(crate::hashing::sha256_hex(content.as_bytes())); |
| 772 | } |
| 773 | renames.push(rename); |
| 774 | } |
| 775 | |
| 776 | let mut files = Vec::new(); |
| 777 | for (index, summary) in summaries.iter().enumerate() { |
| 778 | if matched.contains(&index) { |
| 779 | continue; |
| 780 | } |
| 781 | let outcome = if summary.created { |
| 782 | "created" |
| 783 | } else if summary.deleted { |
| 784 | "deleted" |
| 785 | } else { |
| 786 | "updated" |
| 787 | }; |
| 788 | // Deleted entries carry neither size nor sha256: no bytes were written. |
| 789 | let written = pending |
| 790 | .get(index) |
| 791 | .and_then(|write| write.content.as_deref()) |
| 792 | .filter(|_| !summary.deleted) |
| 793 | .map(str::as_bytes); |
| 794 | files.push(crate::tools::file::mutation_file_entry( |
| 795 | &summary.path, |
| 796 | outcome, |
| 797 | written, |
| 798 | )); |
| 799 | } |
| 800 | |
| 801 | let mut diff_parts = Vec::new(); |
| 802 | for rename in &renames { |
| 803 | let from = rename["from"].as_str().unwrap_or("<file>"); |
| 804 | let to = rename["to"].as_str().unwrap_or("<file>"); |
| 805 | diff_parts.push(format!( |
| 806 | "diff --git a/{from} b/{to}\nsimilarity index 100%\nrename from {from}\nrename to {to}\n" |
| 807 | )); |
| 808 | } |
| 809 | for (index, (write, summary)) in pending.iter().zip(summaries).enumerate() { |
| 810 | if matched.contains(&index) { |
| 811 | continue; |
| 812 | } |
| 813 | let old = write.original.as_deref().unwrap_or(""); |
| 814 | let new = write.content.as_deref().unwrap_or(""); |
| 815 | let diff = make_unified_diff(&summary.path, old, new); |
| 816 | if !diff.is_empty() { |
| 817 | diff_parts.push(format!( |
| 818 | "diff --git a/{path} b/{path}\n{diff}", |
| 819 | path = summary.path |
| 820 | )); |
| 821 | } |
| 822 | } |
| 823 | |
| 824 | json!({ |
| 825 | "diff": diff_parts.join("\n"), |
| 826 | "files": files, |
| 827 | "renames": renames, |
| 828 | }) |
| 829 | } |
| 830 | |
| 831 | /// Parse a unified diff into hunks |
| 832 | fn parse_unified_diff(patch: &str) -> Result<Vec<Hunk>, ToolError> { |
| 833 | let mut hunks = Vec::new(); |
| 834 | let mut lines = patch.lines().peekable(); |
| 835 | |
| 836 | // Skip header lines (---, +++ etc) |
| 837 | while let Some(line) = lines.peek() { |
| 838 | if line.starts_with("@@") { |
| 839 | break; |
| 840 | } |
| 841 | lines.next(); |
| 842 | } |
| 843 | |
| 844 | // Parse hunks |
| 845 | while let Some(line) = lines.next() { |
| 846 | if line.starts_with("@@") { |
| 847 | let hunk = parse_hunk_header(line, &mut lines)?; |
| 848 | hunks.push(hunk); |
| 849 | } |
| 850 | } |
| 851 | |
| 852 | Ok(hunks) |
| 853 | } |
| 854 | |
| 855 | fn parse_unified_diff_files( |
| 856 | patch: &str, |
| 857 | create_if_missing: bool, |
| 858 | ) -> Result<Vec<FilePatch>, ToolError> { |
| 859 | let mut files = Vec::new(); |
| 860 | let mut lines = patch.lines().peekable(); |
| 861 | let mut current: Option<FilePatch> = None; |
| 862 | let mut old_path: Option<String> = None; |
| 863 | |
| 864 | while let Some(line) = lines.next() { |
| 865 | if line.starts_with("diff --git ") { |
| 866 | if let Some(file) = current.take() { |
| 867 | files.push(file); |
| 868 | } |
| 869 | old_path = None; |
| 870 | continue; |
| 871 | } |
| 872 | |
| 873 | if let Some(stripped) = line.strip_prefix("--- ") { |
| 874 | old_path = Some(stripped.trim().to_string()); |
| 875 | continue; |
| 876 | } |
| 877 | |
| 878 | if let Some(stripped) = line.strip_prefix("+++ ") { |
| 879 | let new_path = Some(stripped.trim().to_string()); |
| 880 | let (path, delete_after, create_flag) = |
| 881 | resolve_diff_paths(old_path.as_deref(), new_path.as_deref(), create_if_missing)?; |
| 882 | let creates_new_file = old_path.as_deref().is_some_and(is_dev_null_header); |
| 883 | old_path = None; |
| 884 | if let Some(file) = current.take() { |
| 885 | files.push(file); |
| 886 | } |
| 887 | current = Some(FilePatch { |
| 888 | path, |
| 889 | hunks: Vec::new(), |
| 890 | delete_after, |
| 891 | create_if_missing: create_flag, |
| 892 | creates_new_file, |
| 893 | }); |
| 894 | continue; |
| 895 | } |
| 896 | |
| 897 | if line.starts_with("@@") { |
| 898 | let Some(file) = current.as_mut() else { |
| 899 | if let Some(path) = old_path.as_deref() { |
| 900 | return Err(ToolError::invalid_input(format!( |
| 901 | "Patch hunk encountered after `--- {path}` but before a matching `+++` header. Each file section must include both headers." |
| 902 | ))); |
| 903 | } |
| 904 | return Err(ToolError::invalid_input( |
| 905 | "Patch hunk encountered before any file header. Add `---`/`+++` headers or provide `path`.", |
| 906 | )); |
| 907 | }; |
| 908 | let hunk = parse_hunk_header(line, &mut lines)?; |
| 909 | file.hunks.push(hunk); |
| 910 | } |
| 911 | } |
| 912 | |
| 913 | if let Some(file) = current { |
| 914 | files.push(file); |
| 915 | } |
| 916 | |
| 917 | Ok(files) |
| 918 | } |
| 919 | |
| 920 | fn resolve_diff_paths( |
| 921 | old_path: Option<&str>, |
| 922 | new_path: Option<&str>, |
| 923 | create_if_missing: bool, |
| 924 | ) -> Result<(String, bool, bool), ToolError> { |
| 925 | let old_norm = old_path.and_then(normalize_diff_path); |
| 926 | let new_norm = new_path.and_then(normalize_diff_path); |
| 927 | let delete_after = new_norm.is_none(); |
| 928 | let create_flag = create_if_missing || old_norm.is_none(); |
| 929 | let path = new_norm |
| 930 | .or(old_norm) |
| 931 | .ok_or_else(|| ToolError::invalid_input("Patch is missing both old and new file paths"))?; |
| 932 | Ok((path, delete_after, create_flag)) |
| 933 | } |
| 934 | |
| 935 | fn is_dev_null_header(raw: &str) -> bool { |
| 936 | let raw = raw.split_once('\t').map_or(raw, |(path, _timestamp)| path); |
| 937 | matches!(raw.trim(), "/dev/null" | "dev/null") |
| 938 | } |
| 939 | |
| 940 | fn normalize_diff_path(raw: &str) -> Option<String> { |
| 941 | let raw = raw.split_once('\t').map_or(raw, |(path, _timestamp)| path); |
| 942 | let raw = raw.trim(); |
| 943 | if raw.is_empty() { |
| 944 | return None; |
| 945 | } |
| 946 | if raw == "/dev/null" || raw == "dev/null" { |
| 947 | return None; |
| 948 | } |
| 949 | let raw = raw |
| 950 | .strip_prefix("a/") |
| 951 | .or_else(|| raw.strip_prefix("b/")) |
| 952 | .unwrap_or(raw); |
| 953 | Some(raw.to_string()) |
| 954 | } |
| 955 | |
| 956 | /// Parse a hunk header and its content |
| 957 | fn parse_hunk_header<'a, I>( |
| 958 | header: &str, |
| 959 | lines: &mut std::iter::Peekable<I>, |
| 960 | ) -> Result<Hunk, ToolError> |
| 961 | where |
| 962 | I: Iterator<Item = &'a str> + Clone, |
| 963 | { |
| 964 | // Parse @@ -old_start,old_count +new_start,new_count @@ |
| 965 | let parts: Vec<&str> = header.split_whitespace().collect(); |
| 966 | if parts.len() < 3 { |
| 967 | return Err(ToolError::invalid_input(format!( |
| 968 | "Invalid hunk header: {header}. Expected numeric unified-diff form `@@ -old_start,old_count +new_start,new_count @@` (example: `@@ -12,3 +12,5 @@`)." |
| 969 | ))); |
| 970 | } |
| 971 | |
| 972 | let old_range = parts[1].trim_start_matches('-'); |
| 973 | let new_range = parts[2].trim_start_matches('+'); |
| 974 | |
| 975 | let (old_start, old_count) = parse_range(old_range)?; |
| 976 | let (new_start, new_count) = parse_range(new_range)?; |
| 977 | |
| 978 | // Parse hunk lines |
| 979 | let mut hunk_lines = Vec::new(); |
| 980 | let expected_lines = old_count.max(new_count) + old_count.min(new_count); |
| 981 | let mut old_remaining = old_count; |
| 982 | let mut new_remaining = new_count; |
| 983 | |
| 984 | for _ in 0..expected_lines * 2 { |
| 985 | // `--- ` opens the next file section once this hunk's counts are |
| 986 | // spent, or when `+++ ` follows (a header after a miscounted hunk). |
| 987 | // Otherwise it is a removed line whose text starts `-- `. Reading it |
| 988 | // as a removal unconditionally glued a second section for the same |
| 989 | // file onto the first hunk (B7). |
| 990 | let next_file_header = { |
| 991 | let mut ahead = lines.clone(); |
| 992 | ahead.next().is_some_and(|line| line.starts_with("--- ")) |
| 993 | && ((old_remaining == 0 && new_remaining == 0) |
| 994 | || ahead.next().is_some_and(|line| line.starts_with("+++ "))) |
| 995 | }; |
| 996 | if next_file_header { |
| 997 | break; |
| 998 | } |
| 999 | // Allow for more lines than expected |
| 1000 | match lines.peek() { |
| 1001 | Some(line) if line.starts_with("@@") => break, |
| 1002 | Some(line) if line.starts_with('-') => { |
| 1003 | hunk_lines.push(HunkLine::Remove(line[1..].to_string())); |
| 1004 | old_remaining = old_remaining.saturating_sub(1); |
| 1005 | lines.next(); |
| 1006 | } |
| 1007 | Some(line) if line.starts_with('+') => { |
| 1008 | hunk_lines.push(HunkLine::Add(line[1..].to_string())); |
| 1009 | new_remaining = new_remaining.saturating_sub(1); |
| 1010 | lines.next(); |
| 1011 | } |
| 1012 | Some(line) if line.starts_with(' ') || line.is_empty() => { |
| 1013 | let content = if line.is_empty() { "" } else { &line[1..] }; |
| 1014 | hunk_lines.push(HunkLine::Context(content.to_string())); |
| 1015 | old_remaining = old_remaining.saturating_sub(1); |
| 1016 | new_remaining = new_remaining.saturating_sub(1); |
| 1017 | lines.next(); |
| 1018 | } |
| 1019 | Some(line) |
| 1020 | if line.starts_with("diff ") |
| 1021 | || line.starts_with("--- ") |
| 1022 | || line.starts_with("+++ ") => |
| 1023 | { |
| 1024 | // Start of a new file patch - don't consume, let outer loop handle it |
| 1025 | break; |
| 1026 | } |
| 1027 | Some(line) if !line.starts_with('\\') => { |
| 1028 | // Treat as context line without leading space |
| 1029 | hunk_lines.push(HunkLine::Context((*line).to_string())); |
| 1030 | old_remaining = old_remaining.saturating_sub(1); |
| 1031 | new_remaining = new_remaining.saturating_sub(1); |
| 1032 | lines.next(); |
| 1033 | } |
| 1034 | Some(_) => { |
| 1035 | lines.next(); // Skip "\ No newline at end of file" etc |
| 1036 | } |
| 1037 | None => break, |
| 1038 | } |
| 1039 | } |
| 1040 | |
| 1041 | Ok(Hunk { |
| 1042 | old_start, |
| 1043 | old_count, |
| 1044 | new_start, |
| 1045 | new_count, |
| 1046 | lines: hunk_lines, |
| 1047 | }) |
| 1048 | } |
| 1049 | |
| 1050 | /// Parse a range like "10,5" or "10" into (start, count) |
| 1051 | fn parse_range(range: &str) -> Result<(usize, usize), ToolError> { |
| 1052 | let parts: Vec<&str> = range.split(',').collect(); |
| 1053 | let start = parts[0].parse::<usize>().map_err(|_| { |
| 1054 | ToolError::invalid_input(format!( |
| 1055 | "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`.", |
| 1056 | parts[0] |
| 1057 | )) |
| 1058 | })?; |
| 1059 | let count = if parts.len() > 1 { |
| 1060 | parts[1].parse::<usize>().map_err(|_| { |
| 1061 | ToolError::invalid_input(format!( |
| 1062 | "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`.", |
| 1063 | parts[1] |
| 1064 | )) |
| 1065 | })? |
| 1066 | } else { |
| 1067 | 1 |
| 1068 | }; |
| 1069 | Ok((start, count)) |
| 1070 | } |
| 1071 | |
| 1072 | fn inspect_patch_shape(patch: &str) -> PatchShape { |
| 1073 | let mut shape = PatchShape::default(); |
| 1074 | let mut seen = HashSet::new(); |
| 1075 | let mut old_path: Option<String> = None; |
| 1076 | let mut old_is_dev_null = false; |
| 1077 | let mut hunk_old_remaining = 0usize; |
| 1078 | let mut hunk_new_remaining = 0usize; |
| 1079 | |
| 1080 | for line in patch.lines() { |
| 1081 | if line.starts_with("@@") { |
| 1082 | shape.has_hunks = true; |
| 1083 | if let Some((old_count, new_count)) = hunk_line_counts_for_shape(line) { |
| 1084 | hunk_old_remaining = old_count; |
| 1085 | hunk_new_remaining = new_count; |
| 1086 | } |
| 1087 | continue; |
| 1088 | } |
| 1089 | |
| 1090 | if hunk_old_remaining > 0 || hunk_new_remaining > 0 { |
| 1091 | advance_hunk_shape_counts(line, &mut hunk_old_remaining, &mut hunk_new_remaining); |
| 1092 | continue; |
| 1093 | } |
| 1094 | |
| 1095 | if let Some(stripped) = line.strip_prefix("--- ") { |
| 1096 | old_is_dev_null = is_dev_null_header(stripped); |
| 1097 | old_path = normalize_diff_path(stripped); |
| 1098 | continue; |
| 1099 | } |
| 1100 | |
| 1101 | if let Some(stripped) = line.strip_prefix("+++ ") { |
| 1102 | shape.creates_new_file |= old_is_dev_null && !is_dev_null_header(stripped); |
| 1103 | old_is_dev_null = false; |
| 1104 | let new_path = normalize_diff_path(stripped); |
| 1105 | let resolved = new_path.or(old_path.clone()); |
| 1106 | if let Some(path) = resolved |
| 1107 | && seen.insert(path.clone()) |
| 1108 | { |
| 1109 | shape.header_files.push(path); |
| 1110 | } |
| 1111 | old_path = None; |
| 1112 | } |
| 1113 | } |
| 1114 | |
| 1115 | shape |
| 1116 | } |
| 1117 | |
| 1118 | fn hunk_line_counts_for_shape(header: &str) -> Option<(usize, usize)> { |
| 1119 | let parts: Vec<&str> = header.split_whitespace().collect(); |
| 1120 | if parts.len() < 3 { |
| 1121 | return None; |
| 1122 | } |
| 1123 | let (_, old_count) = parse_range(parts[1].trim_start_matches('-')).ok()?; |
| 1124 | let (_, new_count) = parse_range(parts[2].trim_start_matches('+')).ok()?; |
| 1125 | Some((old_count, new_count)) |
| 1126 | } |
| 1127 | |
| 1128 | fn advance_hunk_shape_counts(line: &str, old_remaining: &mut usize, new_remaining: &mut usize) { |
| 1129 | if line.starts_with('\\') { |
| 1130 | return; |
| 1131 | } |
| 1132 | if line.starts_with('+') { |
| 1133 | *new_remaining = new_remaining.saturating_sub(1); |
| 1134 | } else if line.starts_with('-') { |
| 1135 | *old_remaining = old_remaining.saturating_sub(1); |
| 1136 | } else { |
| 1137 | *old_remaining = old_remaining.saturating_sub(1); |
| 1138 | *new_remaining = new_remaining.saturating_sub(1); |
| 1139 | } |
| 1140 | } |
| 1141 | |
| 1142 | fn validate_patch_shape(shape: &PatchShape, path_override: Option<&str>) -> Result<(), ToolError> { |
| 1143 | if !shape.has_hunks { |
| 1144 | return Err(ToolError::invalid_input( |
| 1145 | "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 @@`).", |
| 1146 | )); |
| 1147 | } |
| 1148 | |
| 1149 | match path_override { |
| 1150 | Some(_) if shape.file_count() > 1 => Err(ToolError::invalid_input(format!( |
| 1151 | "Patch references multiple files ({}) but `path` was provided. Remove `path` to apply a multi-file patch, or provide a single-file patch.", |
| 1152 | format_file_list(&shape.header_files), |
| 1153 | ))), |
| 1154 | None if shape.file_count() == 0 => Err(ToolError::invalid_input( |
| 1155 | "Patch contains hunks but no file headers (`---`/`+++`). Provide `path` or add headers.", |
| 1156 | )), |
| 1157 | _ => Ok(()), |
| 1158 | } |
| 1159 | } |
| 1160 | |
| 1161 | fn diff_header_mismatch(path_override: &str, shape: &PatchShape) -> Option<String> { |
| 1162 | if shape.file_count() != 1 { |
| 1163 | return None; |
| 1164 | } |
| 1165 | let header_path = &shape.header_files[0]; |
| 1166 | let override_norm = normalize_diff_path(path_override).unwrap_or_else(|| path_override.into()); |
| 1167 | if &override_norm == header_path { |
| 1168 | None |
| 1169 | } else { |
| 1170 | Some(format!( |
| 1171 | "Note: patch headers reference `{header_path}` but `path` overrides to `{override_norm}`." |
| 1172 | )) |
| 1173 | } |
| 1174 | } |
| 1175 | |
| 1176 | fn build_summary_message(stats: &PatchStatsExt) -> String { |
| 1177 | let mut parts = Vec::new(); |
| 1178 | if stats.stats.hunks_total > 0 { |
| 1179 | parts.push(format!( |
| 1180 | "Applied {}/{} hunks across {} file(s).", |
| 1181 | stats.stats.hunks_applied, stats.stats.hunks_total, stats.stats.files_applied |
| 1182 | )); |
| 1183 | } else { |
| 1184 | parts.push(format!( |
| 1185 | "Applied {} file change(s).", |
| 1186 | stats.stats.files_applied |
| 1187 | )); |
| 1188 | } |
| 1189 | |
| 1190 | if !stats.touched_files.is_empty() { |
| 1191 | parts.push(format!( |
| 1192 | "Files: {}.", |
| 1193 | format_file_list(&stats.touched_files) |
| 1194 | )); |
| 1195 | } |
| 1196 | |
| 1197 | if stats.stats.fuzz_used > 0 { |
| 1198 | parts.push(format!( |
| 1199 | "Fuzz used on {} hunk(s) (total fuzz: {}).", |
| 1200 | stats.stats.hunks_with_fuzz, stats.stats.fuzz_used |
| 1201 | )); |
| 1202 | } |
| 1203 | |
| 1204 | if stats.stats.hunks_relocated > 0 { |
| 1205 | parts.push(format!( |
| 1206 | "{} hunk(s) applied with stale line numbers (auto-relocated to unique context).", |
| 1207 | stats.stats.hunks_relocated |
| 1208 | )); |
| 1209 | } |
| 1210 | |
| 1211 | if let Some(note) = stats.header_path_mismatch.as_deref() { |
| 1212 | parts.push(note.to_string()); |
| 1213 | } |
| 1214 | |
| 1215 | parts.join(" ") |
| 1216 | } |
| 1217 | |
| 1218 | fn format_file_list(files: &[String]) -> String { |
| 1219 | if files.is_empty() { |
| 1220 | return "<none>".to_string(); |
| 1221 | } |
| 1222 | let mut shown: Vec<String> = files.iter().take(FILE_LIST_LIMIT).cloned().collect(); |
| 1223 | let remaining = files.len().saturating_sub(shown.len()); |
| 1224 | if remaining > 0 { |
| 1225 | shown.push(format!("... (+{remaining} more)")); |
| 1226 | } |
| 1227 | shown.join(", ") |
| 1228 | } |
| 1229 | |
| 1230 | fn push_unique(target: &mut Vec<String>, value: String) { |
| 1231 | if !target.iter().any(|existing| existing == &value) { |
| 1232 | target.push(value); |
| 1233 | } |
| 1234 | } |
| 1235 | |
| 1236 | fn build_pending_writes_from_replace( |
| 1237 | changes: &[Value], |
| 1238 | source_field: &str, |
| 1239 | context: &ToolContext, |
| 1240 | ) -> Result<(Vec<PendingWrite>, PatchStatsExt), ToolError> { |
| 1241 | let mut pending = Vec::new(); |
| 1242 | let mut stats = PatchStatsExt::default(); |
| 1243 | for change in changes { |
| 1244 | let path = change |
| 1245 | .get("path") |
| 1246 | .and_then(Value::as_str) |
| 1247 | .ok_or_else(|| ToolError::missing_field(format!("{source_field}[].path")))?; |
| 1248 | let content = change |
| 1249 | .get("content") |
| 1250 | .and_then(Value::as_str) |
| 1251 | .ok_or_else(|| ToolError::missing_field(format!("{source_field}[].content")))?; |
| 1252 | |
| 1253 | let resolved = context.resolve_path(path)?; |
| 1254 | let original = if resolved.exists() { |
| 1255 | Some(read_file_content(&resolved)?) |
| 1256 | } else { |
| 1257 | None |
| 1258 | }; |
| 1259 | let created = original.is_none(); |
| 1260 | |
| 1261 | pending.push(PendingWrite { |
| 1262 | path: resolved, |
| 1263 | content: Some(content.to_string()), |
| 1264 | original, |
| 1265 | }); |
| 1266 | |
| 1267 | stats.stats.files_total += 1; |
| 1268 | stats.stats.files_applied += 1; |
| 1269 | push_unique(&mut stats.touched_files, path.to_string()); |
| 1270 | stats.file_summaries.push(FileSummary { |
| 1271 | path: path.to_string(), |
| 1272 | hunks: 0, |
| 1273 | hunks_applied: 0, |
| 1274 | fuzz_used: 0, |
| 1275 | hunks_with_fuzz: 0, |
| 1276 | hunks_relocated: 0, |
| 1277 | created, |
| 1278 | deleted: false, |
| 1279 | }); |
| 1280 | } |
| 1281 | |
| 1282 | Ok((pending, stats)) |
| 1283 | } |
| 1284 | |
| 1285 | fn build_pending_writes_from_patches( |
| 1286 | file_patches: Vec<FilePatch>, |
| 1287 | context: &ToolContext, |
| 1288 | fuzz: usize, |
| 1289 | ) -> Result<(Vec<PendingWrite>, PatchStatsExt), ToolError> { |
| 1290 | let mut pending: Vec<PendingWrite> = Vec::new(); |
| 1291 | // Where each file's write already sits in `pending`: a second section for |
| 1292 | // the same file applies on top of the first instead of re-reading the |
| 1293 | // disk and silently dropping the earlier hunks (B7). |
| 1294 | let mut pending_index: std::collections::HashMap<PathBuf, usize> = |
| 1295 | std::collections::HashMap::new(); |
| 1296 | let mut stats = PatchStatsExt::default(); |
| 1297 | stats.stats.files_total = file_patches.len(); |
| 1298 | |
| 1299 | for file_patch in file_patches { |
| 1300 | if file_patch.hunks.is_empty() { |
| 1301 | return Err(ToolError::invalid_input(format!( |
| 1302 | "Patch section for `{}` has no hunks (`@@ ... @@`).", |
| 1303 | file_patch.path |
| 1304 | ))); |
| 1305 | } |
| 1306 | |
| 1307 | let resolved = context.resolve_path(&file_patch.path)?; |
| 1308 | if let Some(&index) = pending_index.get(&resolved) { |
| 1309 | let Some(current) = pending[index].content.clone() else { |
| 1310 | return Err(ToolError::invalid_input(format!( |
| 1311 | "Patch has a section for `{}` after a section that deletes it; merge them into one section.", |
| 1312 | file_patch.path |
| 1313 | ))); |
| 1314 | }; |
| 1315 | if file_patch.creates_new_file && !current.is_empty() { |
| 1316 | return Err(file_already_exists_error(&file_patch.path, &resolved)); |
| 1317 | } |
| 1318 | let mut lines: Vec<String> = current.lines().map(String::from).collect(); |
| 1319 | let apply_stats = |
| 1320 | apply_hunks_to_lines(&mut lines, &file_patch.hunks, fuzz, &file_patch.path)?; |
| 1321 | if file_patch.delete_after { |
| 1322 | ensure_delete_consumes_file(&lines, &file_patch.path)?; |
| 1323 | } |
| 1324 | stats.stats.hunks_applied += apply_stats.hunks_applied; |
| 1325 | stats.stats.hunks_total += file_patch.hunks.len(); |
| 1326 | stats.stats.fuzz_used += apply_stats.fuzz_used; |
| 1327 | stats.stats.hunks_with_fuzz += apply_stats.hunks_with_fuzz; |
| 1328 | stats.stats.hunks_relocated += apply_stats.hunks_relocated; |
| 1329 | if let Some(summary) = stats |
| 1330 | .file_summaries |
| 1331 | .iter_mut() |
| 1332 | .rev() |
| 1333 | .find(|summary| summary.path == file_patch.path) |
| 1334 | { |
| 1335 | summary.hunks += file_patch.hunks.len(); |
| 1336 | summary.hunks_applied += apply_stats.hunks_applied; |
| 1337 | summary.fuzz_used += apply_stats.fuzz_used; |
| 1338 | summary.hunks_with_fuzz += apply_stats.hunks_with_fuzz; |
| 1339 | summary.hunks_relocated += apply_stats.hunks_relocated; |
| 1340 | summary.deleted |= file_patch.delete_after; |
| 1341 | summary.created &= !file_patch.delete_after; |
| 1342 | } |
| 1343 | pending[index].content = (!file_patch.delete_after) |
| 1344 | .then(|| reassemble_preserving_newlines(&lines, ¤t)); |
| 1345 | // One file, however many sections: count it once. |
| 1346 | stats.stats.files_total = stats.stats.files_total.saturating_sub(1); |
| 1347 | continue; |
| 1348 | } |
| 1349 | let original = if resolved.exists() { |
| 1350 | Some(read_file_content(&resolved)?) |
| 1351 | } else { |
| 1352 | None |
| 1353 | }; |
| 1354 | |
| 1355 | if original.is_none() && !file_patch.create_if_missing { |
| 1356 | return Err(ToolError::execution_failed(format!( |
| 1357 | "File `{}` does not exist at `{}`. Set create_if_missing=true for new files or include headers for file creation.", |
| 1358 | file_patch.path, |
| 1359 | resolved.display(), |
| 1360 | ))); |
| 1361 | } |
| 1362 | |
| 1363 | if file_patch.delete_after && original.is_none() { |
| 1364 | return Err(ToolError::execution_failed(format!( |
| 1365 | "File `{}` does not exist at `{}` to delete.", |
| 1366 | file_patch.path, |
| 1367 | resolved.display(), |
| 1368 | ))); |
| 1369 | } |
| 1370 | |
| 1371 | // A `--- /dev/null` section has no old lines to anchor on, so on an |
| 1372 | // existing file it would splice the "new" content onto the top of |
| 1373 | // the old one and report success. `git apply` refuses this too. |
| 1374 | if file_patch.creates_new_file && original.as_deref().is_some_and(|c| !c.is_empty()) { |
| 1375 | return Err(file_already_exists_error(&file_patch.path, &resolved)); |
| 1376 | } |
| 1377 | |
| 1378 | let base_content = original.clone().unwrap_or_default(); |
| 1379 | let mut lines: Vec<String> = if base_content.is_empty() { |
| 1380 | Vec::new() |
| 1381 | } else { |
| 1382 | base_content.lines().map(String::from).collect() |
| 1383 | }; |
| 1384 | |
| 1385 | let apply_stats = |
| 1386 | apply_hunks_to_lines(&mut lines, &file_patch.hunks, fuzz, &file_patch.path)?; |
| 1387 | if file_patch.delete_after { |
| 1388 | ensure_delete_consumes_file(&lines, &file_patch.path)?; |
| 1389 | } |
| 1390 | stats.stats.hunks_applied += apply_stats.hunks_applied; |
| 1391 | stats.stats.hunks_total += file_patch.hunks.len(); |
| 1392 | stats.stats.fuzz_used += apply_stats.fuzz_used; |
| 1393 | stats.stats.hunks_with_fuzz += apply_stats.hunks_with_fuzz; |
| 1394 | stats.stats.hunks_relocated += apply_stats.hunks_relocated; |
| 1395 | stats.stats.files_applied += 1; |
| 1396 | push_unique(&mut stats.touched_files, file_patch.path.clone()); |
| 1397 | stats.file_summaries.push(FileSummary { |
| 1398 | path: file_patch.path.clone(), |
| 1399 | hunks: file_patch.hunks.len(), |
| 1400 | hunks_applied: apply_stats.hunks_applied, |
| 1401 | fuzz_used: apply_stats.fuzz_used, |
| 1402 | hunks_with_fuzz: apply_stats.hunks_with_fuzz, |
| 1403 | hunks_relocated: apply_stats.hunks_relocated, |
| 1404 | created: original.is_none() && !file_patch.delete_after, |
| 1405 | deleted: file_patch.delete_after, |
| 1406 | }); |
| 1407 | |
| 1408 | pending_index.insert(resolved.clone(), pending.len()); |
| 1409 | if file_patch.delete_after { |
| 1410 | pending.push(PendingWrite { |
| 1411 | path: resolved, |
| 1412 | content: None, |
| 1413 | original, |
| 1414 | }); |
| 1415 | } else { |
| 1416 | let new_content = reassemble_preserving_newlines(&lines, &base_content); |
| 1417 | pending.push(PendingWrite { |
| 1418 | path: resolved, |
| 1419 | content: Some(new_content), |
| 1420 | original, |
| 1421 | }); |
| 1422 | } |
| 1423 | } |
| 1424 | |
| 1425 | Ok((pending, stats)) |
| 1426 | } |
| 1427 | |
| 1428 | fn file_already_exists_error(label: &str, resolved: &Path) -> ToolError { |
| 1429 | ToolError::execution_failed(format!( |
| 1430 | "Patch creates `{label}` (`--- /dev/null`), but it already exists at `{}`. Nothing was changed. To change the existing file, send a patch with context lines from its current contents; to replace it wholesale, use a write/replace action.", |
| 1431 | resolved.display(), |
| 1432 | )) |
| 1433 | } |
| 1434 | |
| 1435 | /// A `+++ /dev/null` section deletes the file, so its hunks must remove every |
| 1436 | /// line. Deleting after a partial match would silently drop the rest. |
| 1437 | fn ensure_delete_consumes_file(remaining: &[String], label: &str) -> Result<(), ToolError> { |
| 1438 | if remaining.is_empty() { |
| 1439 | return Ok(()); |
| 1440 | } |
| 1441 | Err(ToolError::execution_failed(format!( |
| 1442 | "Patch deletes `{label}` (`+++ /dev/null`), but its hunks leave {} line(s) of the file unaccounted for. Nothing was changed. Include every line of the file as `-` lines to delete it, or send a patch with a real `+++` path to edit it.", |
| 1443 | remaining.len(), |
| 1444 | ))) |
| 1445 | } |
| 1446 | |
| 1447 | /// Normalize the Rust files a patch rewrites (#6205), before the write and |
| 1448 | /// before the result's diff is built, so the rendered diff and the bytes on |
| 1449 | /// disk are the same text and the model's next anchor matches reality. |
| 1450 | async fn normalize_pending_rust(pending: &mut [PendingWrite]) { |
| 1451 | for entry in pending.iter_mut() { |
| 1452 | let (Some(content), Some(original)) = (entry.content.as_ref(), entry.original.as_ref()) |
| 1453 | else { |
| 1454 | continue; |
| 1455 | }; |
| 1456 | if let Some(normalized) = normalize_edit(&entry.path, original, content).await { |
| 1457 | entry.content = Some(normalized); |
| 1458 | } |
| 1459 | } |
| 1460 | } |
| 1461 | |
| 1462 | /// Run the patch's synchronous file work (reads, the syntax gate, writes and |
| 1463 | /// rollback) on the blocking pool rather than an async worker (#6559 D01-06). |
| 1464 | async fn patch_blocking<T: Send + 'static>( |
| 1465 | work: impl FnOnce() -> Result<T, ToolError> + Send + 'static, |
| 1466 | ) -> Result<T, ToolError> { |
| 1467 | tokio::task::spawn_blocking(work) |
| 1468 | .await |
| 1469 | .map_err(|error| ToolError::execution_failed(format!("patch task failed: {error}")))? |
| 1470 | } |
| 1471 | |
| 1472 | /// [`apply_pending_writes`] on the blocking pool, handing `pending` back for |
| 1473 | /// the result's diagnostics and metadata. |
| 1474 | async fn apply_pending_writes_blocking( |
| 1475 | pending: Vec<PendingWrite>, |
| 1476 | ) -> Result<Vec<PendingWrite>, ToolError> { |
| 1477 | patch_blocking(move || apply_pending_writes(&pending).map(|()| pending)).await |
| 1478 | } |
| 1479 | |
| 1480 | fn apply_pending_writes(pending: &[PendingWrite]) -> Result<(), ToolError> { |
| 1481 | // Syntax gate (#6204) ahead of the first write, not per file: a patch is |
| 1482 | // transactional, so one unparseable result must leave every file in the |
| 1483 | // patch untouched rather than half-applied and rolled back. |
| 1484 | for entry in pending { |
| 1485 | if let Some(content) = entry.content.as_ref() { |
| 1486 | let display = entry.path.display().to_string(); |
| 1487 | guard_edit(&entry.path, &display, entry.original.as_deref(), content)?; |
| 1488 | } |
| 1489 | } |
| 1490 | |
| 1491 | let mut applied = Vec::new(); |
| 1492 | |
| 1493 | for entry in pending { |
| 1494 | let result = if let Some(content) = entry.content.as_ref() { |
| 1495 | let parent_result = if let Some(parent) = entry.path.parent() { |
| 1496 | fs::create_dir_all(parent).map_err(|e| { |
| 1497 | ToolError::execution_failed(format!( |
| 1498 | "Failed to create directory {}: {}", |
| 1499 | parent.display(), |
| 1500 | e |
| 1501 | )) |
| 1502 | }) |
| 1503 | } else { |
| 1504 | Ok(()) |
| 1505 | }; |
| 1506 | |
| 1507 | parent_result.and_then(|()| { |
| 1508 | crate::utils::write_atomic_workspace(&entry.path, content.as_bytes()).map_err(|e| { |
| 1509 | ToolError::execution_failed(format!( |
| 1510 | "Failed to write {}: {}", |
| 1511 | entry.path.display(), |
| 1512 | e |
| 1513 | )) |
| 1514 | }) |
| 1515 | }) |
| 1516 | } else if entry.path.exists() { |
| 1517 | fs::remove_file(&entry.path).map_err(|e| { |
| 1518 | ToolError::execution_failed(format!( |
| 1519 | "Failed to delete {}: {}", |
| 1520 | entry.path.display(), |
| 1521 | e |
| 1522 | )) |
| 1523 | }) |
| 1524 | } else { |
| 1525 | Ok(()) |
| 1526 | }; |
| 1527 | |
| 1528 | if let Err(err) = result { |
| 1529 | let failures = rollback_pending_writes(&applied); |
| 1530 | if failures.is_empty() { |
| 1531 | return Err(err); |
| 1532 | } |
| 1533 | // A failed restore leaves the patch partly applied; say which |
| 1534 | // files still hold it rather than implying nothing changed. |
| 1535 | return Err(ToolError::execution_failed(format!( |
| 1536 | "{err}. Rolling back the files already written also failed, so the patch is \ |
| 1537 | partly applied: {}", |
| 1538 | failures.join("; ") |
| 1539 | ))); |
| 1540 | } |
| 1541 | |
| 1542 | applied.push(entry.clone()); |
| 1543 | } |
| 1544 | |
| 1545 | Ok(()) |
| 1546 | } |
| 1547 | |
| 1548 | /// Restore every already-applied entry, newest first. Returns one line per |
| 1549 | /// entry that could not be restored (#6559 D01-06: these were discarded). |
| 1550 | fn rollback_pending_writes(applied: &[PendingWrite]) -> Vec<String> { |
| 1551 | let mut failures = Vec::new(); |
| 1552 | for entry in applied.iter().rev() { |
| 1553 | let restored = match entry.original.as_ref() { |
| 1554 | Some(content) => crate::utils::write_atomic_workspace(&entry.path, content.as_bytes()), |
| 1555 | None => match fs::remove_file(&entry.path) { |
| 1556 | Err(error) if error.kind() == std::io::ErrorKind::NotFound => Ok(()), |
| 1557 | other => other, |
| 1558 | }, |
| 1559 | }; |
| 1560 | if let Err(error) = restored { |
| 1561 | failures.push(format!("{} ({error})", entry.path.display())); |
| 1562 | } |
| 1563 | } |
| 1564 | failures |
| 1565 | } |
| 1566 | |
| 1567 | fn read_file_content(path: &PathBuf) -> Result<String, ToolError> { |
| 1568 | fs::read_to_string(path).map_err(|e| { |
| 1569 | ToolError::execution_failed(format!("Failed to read {}: {}", path.display(), e)) |
| 1570 | }) |
| 1571 | } |
| 1572 | |
| 1573 | fn preview_expected_lines(hunk: &Hunk, limit: usize) -> Vec<String> { |
| 1574 | let mut preview = Vec::new(); |
| 1575 | for line in hunk.lines.iter().filter_map(|line| match line { |
| 1576 | HunkLine::Context(s) => Some((" ", s)), |
| 1577 | HunkLine::Remove(s) => Some(("-", s)), |
| 1578 | HunkLine::Add(_) => None, |
| 1579 | }) { |
| 1580 | if preview.len() >= limit { |
| 1581 | break; |
| 1582 | } |
| 1583 | preview.push(format!(" {}{}", line.0, line.1)); |
| 1584 | } |
| 1585 | if preview.is_empty() { |
| 1586 | preview.push(" <no context lines in hunk>".to_string()); |
| 1587 | } |
| 1588 | preview |
| 1589 | } |
| 1590 | |
| 1591 | fn snippet_around(lines: &[String], line_1_based: usize, radius: usize) -> Vec<String> { |
| 1592 | if lines.is_empty() { |
| 1593 | return vec![" <empty file>".to_string()]; |
| 1594 | } |
| 1595 | |
| 1596 | let center = line_1_based |
| 1597 | .saturating_sub(1) |
| 1598 | .min(lines.len().saturating_sub(1)); |
| 1599 | let start = center.saturating_sub(radius); |
| 1600 | let end = (center + radius).min(lines.len().saturating_sub(1)); |
| 1601 | |
| 1602 | lines[start..=end] |
| 1603 | .iter() |
| 1604 | .enumerate() |
| 1605 | .map(|(idx, line)| { |
| 1606 | let line_no = start + idx + 1; |
| 1607 | format!(" {line_no:>4}: {line}") |
| 1608 | }) |
| 1609 | .collect() |
| 1610 | } |
| 1611 | |
| 1612 | fn format_hunk_no_match_error( |
| 1613 | lines: &[String], |
| 1614 | hunk: &Hunk, |
| 1615 | err: &ApplyHunkError, |
| 1616 | max_fuzz: usize, |
| 1617 | ) -> String { |
| 1618 | match err { |
| 1619 | ApplyHunkError::NoMatch { |
| 1620 | expected_line, |
| 1621 | adjusted_line, |
| 1622 | offset, |
| 1623 | } => { |
| 1624 | let expected_preview = preview_expected_lines(hunk, HUNK_PREVIEW_LINES).join("\n"); |
| 1625 | let file_preview = snippet_around(lines, *adjusted_line, SNIPPET_RADIUS).join("\n"); |
| 1626 | format!( |
| 1627 | "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." |
| 1628 | ) |
| 1629 | } |
| 1630 | ApplyHunkError::ContextAmbiguous { |
| 1631 | expected_line, |
| 1632 | candidate_lines, |
| 1633 | } => { |
| 1634 | let candidates = candidate_lines |
| 1635 | .iter() |
| 1636 | .map(|line| line.to_string()) |
| 1637 | .collect::<Vec<_>>() |
| 1638 | .join(", "); |
| 1639 | format!( |
| 1640 | "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." |
| 1641 | ) |
| 1642 | } |
| 1643 | } |
| 1644 | } |
| 1645 | |
| 1646 | fn apply_hunks_to_lines( |
| 1647 | lines: &mut Vec<String>, |
| 1648 | hunks: &[Hunk], |
| 1649 | fuzz: usize, |
| 1650 | file_label: &str, |
| 1651 | ) -> Result<HunkApplyStats, ToolError> { |
| 1652 | let mut stats = HunkApplyStats::default(); |
| 1653 | let mut cumulative_offset: isize = 0; |
| 1654 | |
| 1655 | for (idx, hunk) in hunks.iter().enumerate() { |
| 1656 | match apply_hunk(lines, hunk, fuzz, &mut cumulative_offset) { |
| 1657 | Ok(outcome) => { |
| 1658 | stats.hunks_applied += 1; |
| 1659 | if outcome.fuzz_used > 0 { |
| 1660 | stats.fuzz_used += outcome.fuzz_used; |
| 1661 | stats.hunks_with_fuzz += 1; |
| 1662 | } |
| 1663 | if outcome.relocated { |
| 1664 | stats.hunks_relocated += 1; |
| 1665 | } |
| 1666 | } |
| 1667 | Err(e) => { |
| 1668 | let detail = format_hunk_no_match_error(lines, hunk, &e, fuzz); |
| 1669 | return Err(ToolError::execution_failed(format!( |
| 1670 | "Failed to apply hunk {}/{} for `{}`: {}", |
| 1671 | idx + 1, |
| 1672 | hunks.len(), |
| 1673 | file_label, |
| 1674 | detail |
| 1675 | ))); |
| 1676 | } |
| 1677 | } |
| 1678 | } |
| 1679 | |
| 1680 | Ok(stats) |
| 1681 | } |
| 1682 | |
| 1683 | /// Apply a hunk to the file content with fuzzy matching |
| 1684 | fn apply_hunk( |
| 1685 | lines: &mut Vec<String>, |
| 1686 | hunk: &Hunk, |
| 1687 | max_fuzz: usize, |
| 1688 | cumulative_offset: &mut isize, |
| 1689 | ) -> Result<HunkApplyOutcome, ApplyHunkError> { |
| 1690 | // Build expected old lines from hunk |
| 1691 | let old_lines: Vec<&str> = hunk |
| 1692 | .lines |
| 1693 | .iter() |
| 1694 | .filter_map(|line| match line { |
| 1695 | HunkLine::Context(s) | HunkLine::Remove(s) => Some(s.as_str()), |
| 1696 | HunkLine::Add(_) => None, |
| 1697 | }) |
| 1698 | .collect(); |
| 1699 | |
| 1700 | // Build new lines from hunk |
| 1701 | let new_lines: Vec<String> = hunk |
| 1702 | .lines |
| 1703 | .iter() |
| 1704 | .filter_map(|line| match line { |
| 1705 | HunkLine::Context(s) | HunkLine::Add(s) => Some(s.clone()), |
| 1706 | HunkLine::Remove(_) => None, |
| 1707 | }) |
| 1708 | .collect(); |
| 1709 | |
| 1710 | // Try to find the location with fuzzy matching |
| 1711 | // Apply cumulative offset from previous hunks, clamping to valid range. |
| 1712 | // A pure insertion (`@@ -N,0 +M,K @@`, as `git diff -U0` emits) names |
| 1713 | // the line it follows, so it goes *after* line N, not before it. Only a |
| 1714 | // header that says `,0` qualifies: an omitted or nonzero old count keeps |
| 1715 | // the "before line N" anchor. `new_start` settles the common hand-written |
| 1716 | // `@@ -N,0 +N,K @@` ("new text starts at line N"): when it points at |
| 1717 | // line N itself rather than after it, the text goes before line N. |
| 1718 | let pure_insertion = hunk.old_count == 0 && old_lines.is_empty(); |
| 1719 | let starts_at_named_line = hunk.new_start > 0 |
| 1720 | && cumulative_offset |
| 1721 | .checked_neg() |
| 1722 | .and_then(|back| (hunk.new_start - 1).checked_add_signed(back)) |
| 1723 | == Some(hunk.old_start.saturating_sub(1)); |
| 1724 | let base_idx = if pure_insertion && !starts_at_named_line { |
| 1725 | hunk.old_start |
| 1726 | } else { |
| 1727 | hunk.old_start.saturating_sub(1) |
| 1728 | }; |
| 1729 | // Use checked_add_signed to safely handle negative offsets without |
| 1730 | // risking isize overflow on adversarial input. |
| 1731 | let start_idx = base_idx |
| 1732 | .checked_add_signed(*cumulative_offset) |
| 1733 | .unwrap_or(0) |
| 1734 | .min(lines.len()); |
| 1735 | |
| 1736 | for fuzz in 0..=max_fuzz { |
| 1737 | // Try at exact position first, then nearby |
| 1738 | let search_range = if fuzz == 0 { |
| 1739 | vec![start_idx] |
| 1740 | } else { |
| 1741 | let min = start_idx.saturating_sub(fuzz); |
| 1742 | let max = (start_idx + fuzz).min(lines.len()); |
| 1743 | (min..=max).collect() |
| 1744 | }; |
| 1745 | |
| 1746 | for pos in search_range { |
| 1747 | if matches_at_position(lines, &old_lines, pos) { |
| 1748 | // Apply the hunk |
| 1749 | let end_pos = pos + old_lines.len(); |
| 1750 | lines.splice(pos..end_pos, new_lines.clone()); |
| 1751 | |
| 1752 | // Update cumulative offset: new lines added minus old lines removed |
| 1753 | let delta = new_lines.len() as isize - old_lines.len() as isize; |
| 1754 | *cumulative_offset += delta; |
| 1755 | |
| 1756 | return Ok(HunkApplyOutcome { |
| 1757 | fuzz_used: fuzz, |
| 1758 | relocated: false, |
| 1759 | }); |
| 1760 | } |
| 1761 | } |
| 1762 | } |
| 1763 | |
| 1764 | // Special case: adding to empty file or new hunk at end |
| 1765 | if old_lines.is_empty() && (lines.is_empty() || start_idx >= lines.len()) { |
| 1766 | let delta = new_lines.len() as isize; |
| 1767 | lines.extend(new_lines); |
| 1768 | *cumulative_offset += delta; |
| 1769 | return Ok(HunkApplyOutcome { |
| 1770 | fuzz_used: 0, |
| 1771 | relocated: false, |
| 1772 | }); |
| 1773 | } |
| 1774 | |
| 1775 | // #5003 — positional search failed. The line numbers are probably stale |
| 1776 | // because an earlier edit (this patch or a previous one) shifted the file |
| 1777 | // and the model regenerated the patch from outdated read_file output. |
| 1778 | // If the hunk carries enough anchor lines, look for a unique whole-file |
| 1779 | // content match and relocate there; anything that matched within `fuzz` |
| 1780 | // of `start_idx` was already tried above, so any unique match found here |
| 1781 | // is genuinely relocated. Ambiguous matches are refused (applying to the |
| 1782 | // wrong copy of a repeated block would corrupt the file). |
| 1783 | let anchor_matches: Vec<usize> = if old_lines.len() >= MIN_ANCHOR_LINES { |
| 1784 | (0..=lines.len().saturating_sub(old_lines.len())) |
| 1785 | .filter(|&pos| matches_at_position(lines, &old_lines, pos)) |
| 1786 | .collect() |
| 1787 | } else { |
| 1788 | Vec::new() |
| 1789 | }; |
| 1790 | match anchor_matches.as_slice() { |
| 1791 | [pos] => { |
| 1792 | let end_pos = pos + old_lines.len(); |
| 1793 | lines.splice(*pos..end_pos, new_lines.clone()); |
| 1794 | let delta = new_lines.len() as isize - old_lines.len() as isize; |
| 1795 | *cumulative_offset += delta; |
| 1796 | Ok(HunkApplyOutcome { |
| 1797 | fuzz_used: 0, |
| 1798 | relocated: true, |
| 1799 | }) |
| 1800 | } |
| 1801 | [] => Err(ApplyHunkError::NoMatch { |
| 1802 | expected_line: hunk.old_start, |
| 1803 | adjusted_line: start_idx + 1, // Convert back to 1-indexed |
| 1804 | offset: *cumulative_offset, |
| 1805 | }), |
| 1806 | multiple => Err(ApplyHunkError::ContextAmbiguous { |
| 1807 | expected_line: hunk.old_start, |
| 1808 | candidate_lines: multiple.iter().map(|&p| p + 1).collect(), |
| 1809 | }), |
| 1810 | } |
| 1811 | } |
| 1812 | |
| 1813 | /// Check if `old_lines` match at the given position |
| 1814 | fn matches_at_position(lines: &[String], old_lines: &[&str], pos: usize) -> bool { |
| 1815 | if pos + old_lines.len() > lines.len() { |
| 1816 | return false; |
| 1817 | } |
| 1818 | |
| 1819 | for (i, old_line) in old_lines.iter().enumerate() { |
| 1820 | // Normalize whitespace for comparison |
| 1821 | let file_line = lines[pos + i].trim_end(); |
| 1822 | let expected = old_line.trim_end(); |
| 1823 | if file_line != expected { |
| 1824 | return false; |
| 1825 | } |
| 1826 | } |
| 1827 | |
| 1828 | true |
| 1829 | } |
| 1830 | |
| 1831 | // === Unit Tests === |
| 1832 | |
| 1833 | #[cfg(test)] |
| 1834 | mod tests { |
| 1835 | use super::*; |
| 1836 | use tempfile::tempdir; |
| 1837 | |
| 1838 | /// #6559 D01-06: a rollback that cannot restore a file used to be |
| 1839 | /// discarded, so the error read as if nothing had changed. Deleting `x` |
| 1840 | /// and creating `x/y.txt` leaves a directory at `x`; when the third write |
| 1841 | /// fails, restoring the file `x` over that directory fails too. |
| 1842 | #[test] |
| 1843 | fn failed_rollback_is_reported_as_a_partly_applied_patch() { |
| 1844 | let tmp = tempdir().expect("tempdir"); |
| 1845 | let root = tmp.path(); |
| 1846 | fs::write(root.join("x"), "original\n").expect("seed x"); |
| 1847 | fs::write(root.join("blocker"), "a file, not a directory\n").expect("seed blocker"); |
| 1848 | let pending = vec![ |
| 1849 | PendingWrite { |
| 1850 | path: root.join("x"), |
| 1851 | content: None, |
| 1852 | original: Some("original\n".to_string()), |
| 1853 | }, |
| 1854 | PendingWrite { |
| 1855 | path: root.join("x").join("y.txt"), |
| 1856 | content: Some("new\n".to_string()), |
| 1857 | original: None, |
| 1858 | }, |
| 1859 | PendingWrite { |
| 1860 | path: root.join("blocker").join("z.txt"), |
| 1861 | content: Some("never written\n".to_string()), |
| 1862 | original: None, |
| 1863 | }, |
| 1864 | ]; |
| 1865 | |
| 1866 | let message = apply_pending_writes(&pending) |
| 1867 | .expect_err("the third write cannot create its directory") |
| 1868 | .to_string(); |
| 1869 | |
| 1870 | assert!(message.contains("partly applied"), "{message}"); |
| 1871 | assert!( |
| 1872 | message.contains(&root.join("x").display().to_string()), |
| 1873 | "{message}" |
| 1874 | ); |
| 1875 | } |
| 1876 | |
| 1877 | /// The receipt entry a created/updated file must carry: its size and |
| 1878 | /// SHA-256 are those of the bytes now on disk. |
| 1879 | fn on_disk_receipt(root: &std::path::Path, path: &str, outcome: &str) -> Value { |
| 1880 | let bytes = fs::read(root.join(path)).expect("receipt target exists"); |
| 1881 | json!({ |
| 1882 | "path": path, |
| 1883 | "outcome": outcome, |
| 1884 | "size": bytes.len(), |
| 1885 | "sha256": crate::hashing::sha256_hex(&bytes), |
| 1886 | }) |
| 1887 | } |
| 1888 | |
| 1889 | fn parse_patch_result(result: ToolResult) -> PatchResult { |
| 1890 | serde_json::from_str(&result.content).expect("patch result json") |
| 1891 | } |
| 1892 | |
| 1893 | #[test] |
| 1894 | fn test_parse_range() { |
| 1895 | assert_eq!(parse_range("10,5").unwrap(), (10, 5)); |
| 1896 | assert_eq!(parse_range("10").unwrap(), (10, 1)); |
| 1897 | assert_eq!(parse_range("1,0").unwrap(), (1, 0)); |
| 1898 | } |
| 1899 | |
| 1900 | #[test] |
| 1901 | fn test_parse_unified_diff() { |
| 1902 | let patch = r"--- a/test.txt |
| 1903 | +++ b/test.txt |
| 1904 | @@ -1,3 +1,3 @@ |
| 1905 | line1 |
| 1906 | -line2 |
| 1907 | +modified line2 |
| 1908 | line3 |
| 1909 | "; |
| 1910 | |
| 1911 | let hunks = parse_unified_diff(patch).unwrap(); |
| 1912 | assert_eq!(hunks.len(), 1); |
| 1913 | assert_eq!(hunks[0].old_start, 1); |
| 1914 | assert_eq!(hunks[0].old_count, 3); |
| 1915 | assert_eq!(hunks[0].new_start, 1); |
| 1916 | assert_eq!(hunks[0].new_count, 3); |
| 1917 | } |
| 1918 | |
| 1919 | /// #6205 — a patch that lands unformatted Rust in an already-clean file is |
| 1920 | /// normalized before the write, and the rendered diff shows the normalized |
| 1921 | /// text, so the model's next patch context matches the bytes on disk. |
| 1922 | #[tokio::test] |
| 1923 | async fn patch_normalizes_rust_in_an_already_clean_file() { |
| 1924 | let tmp = tempdir().expect("tempdir"); |
| 1925 | let ctx = ToolContext::new(tmp.path().to_path_buf()); |
| 1926 | let file = tmp.path().join("clean.rs"); |
| 1927 | fs::write(&file, "fn main() {\n let x = 1;\n}\n").expect("write"); |
| 1928 | |
| 1929 | 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"; |
| 1930 | let result = ApplyPatchTool |
| 1931 | .execute(json!({"path": "clean.rs", "patch": patch}), &ctx) |
| 1932 | .await |
| 1933 | .expect("execute"); |
| 1934 | |
| 1935 | assert_eq!( |
| 1936 | fs::read_to_string(&file).expect("read"), |
| 1937 | "fn main() {\n let x = 1;\n let y = 2;\n}\n" |
| 1938 | ); |
| 1939 | let diff = result.metadata.as_ref().expect("metadata")["mutation"]["diff"] |
| 1940 | .as_str() |
| 1941 | .expect("diff") |
| 1942 | .to_string(); |
| 1943 | assert!(diff.contains("+ let y = 2;"), "{diff}"); |
| 1944 | assert!(!diff.contains("let y=2;"), "{diff}"); |
| 1945 | } |
| 1946 | |
| 1947 | /// #6204 — a patch whose result does not parse is refused before any file |
| 1948 | /// is written, so a multi-file patch cannot land half-applied. |
| 1949 | #[tokio::test] |
| 1950 | async fn patch_refuses_a_hunk_that_breaks_rust_syntax() { |
| 1951 | let tmp = tempdir().expect("tempdir"); |
| 1952 | let ctx = ToolContext::new(tmp.path().to_path_buf()); |
| 1953 | let file = tmp.path().join("main.rs"); |
| 1954 | let original = "fn main() {\n println!(\"hi\");\n}\n"; |
| 1955 | fs::write(&file, original).expect("write"); |
| 1956 | |
| 1957 | let patch = "--- a/main.rs\n+++ b/main.rs\n@@ -1,3 +1,2 @@\n fn main() {\n println!(\"hi\");\n-}\n"; |
| 1958 | let error = ApplyPatchTool |
| 1959 | .execute(json!({"path": "main.rs", "patch": patch}), &ctx) |
| 1960 | .await |
| 1961 | .expect_err("a patch that breaks Rust syntax must be refused"); |
| 1962 | |
| 1963 | let message = error.to_string(); |
| 1964 | assert!(message.contains("Rust syntax error at line"), "{message}"); |
| 1965 | assert_eq!( |
| 1966 | fs::read_to_string(&file).expect("read"), |
| 1967 | original, |
| 1968 | "a refused patch must leave the file untouched" |
| 1969 | ); |
| 1970 | } |
| 1971 | |
| 1972 | #[test] |
| 1973 | fn input_schema_exposes_replace_and_deprecated_changes_alias() { |
| 1974 | let schema = ApplyPatchTool.input_schema(); |
| 1975 | |
| 1976 | assert_eq!(schema["properties"]["replace"]["type"], "array"); |
| 1977 | assert_eq!(schema["properties"]["changes"]["type"], "array"); |
| 1978 | assert!( |
| 1979 | schema["properties"]["changes"]["description"] |
| 1980 | .as_str() |
| 1981 | .is_some_and(|description| description.contains("Deprecated")) |
| 1982 | ); |
| 1983 | assert_eq!( |
| 1984 | schema["oneOf"], |
| 1985 | json!([ |
| 1986 | { "required": ["patch"] }, |
| 1987 | { "required": ["replace"] }, |
| 1988 | { "required": ["changes"] } |
| 1989 | ]) |
| 1990 | ); |
| 1991 | } |
| 1992 | |
| 1993 | #[test] |
| 1994 | fn test_preflight_apply_patch_with_path_override() { |
| 1995 | let patch = r"@@ -1,2 +1,2 @@ |
| 1996 | old |
| 1997 | -value |
| 1998 | +new-value |
| 1999 | "; |
| 2000 | |
| 2001 | let preflight = preflight_apply_patch(&json!({ |
| 2002 | "path": "src/lib.rs", |
| 2003 | "patch": patch |
| 2004 | })) |
| 2005 | .expect("preflight"); |
| 2006 | |
| 2007 | assert_eq!(preflight.touched_files, vec!["src/lib.rs"]); |
| 2008 | assert_eq!(preflight.files_total, 1); |
| 2009 | assert_eq!(preflight.hunks_total, 1); |
| 2010 | assert_eq!(preflight.path_override.as_deref(), Some("src/lib.rs")); |
| 2011 | } |
| 2012 | |
| 2013 | #[test] |
| 2014 | fn test_preflight_apply_patch_multi_file_create_and_delete() { |
| 2015 | let patch = r"diff --git a/new.rs b/new.rs |
| 2016 | --- /dev/null |
| 2017 | +++ b/new.rs |
| 2018 | @@ -0,0 +1 @@ |
| 2019 | +fn added() {} |
| 2020 | diff --git a/old.rs b/old.rs |
| 2021 | --- a/old.rs |
| 2022 | +++ /dev/null |
| 2023 | @@ -1 +0,0 @@ |
| 2024 | -fn old() {} |
| 2025 | "; |
| 2026 | |
| 2027 | let preflight = preflight_apply_patch(&json!({ "patch": patch })).expect("preflight"); |
| 2028 | |
| 2029 | assert_eq!(preflight.touched_files, vec!["new.rs", "old.rs"]); |
| 2030 | assert_eq!(preflight.files_total, 2); |
| 2031 | assert_eq!(preflight.hunks_total, 2); |
| 2032 | assert_eq!(preflight.creates, vec!["new.rs"]); |
| 2033 | assert_eq!(preflight.deletes, vec!["old.rs"]); |
| 2034 | } |
| 2035 | |
| 2036 | #[test] |
| 2037 | fn test_preflight_apply_patch_timestamp_headers_strip_metadata() { |
| 2038 | let patch = "diff --git a/src/lib.rs b/src/lib.rs\n\ |
| 2039 | --- a/src/lib.rs\t2026-06-26 10:00:00 +0000\n\ |
| 2040 | +++ b/src/lib.rs\t2026-06-26 10:01:00 +0000\n\ |
| 2041 | @@ -1,1 +1,1 @@\n\ |
| 2042 | -old\n\ |
| 2043 | +new\n"; |
| 2044 | |
| 2045 | let preflight = preflight_apply_patch(&json!({ "patch": patch })).expect("preflight"); |
| 2046 | |
| 2047 | assert_eq!(preflight.touched_files, vec!["src/lib.rs"]); |
| 2048 | assert_eq!(preflight.files_total, 1); |
| 2049 | assert_eq!(preflight.hunks_total, 1); |
| 2050 | } |
| 2051 | |
| 2052 | #[test] |
| 2053 | fn test_preflight_apply_patch_ignores_forged_headers_inside_hunk_shape() { |
| 2054 | let patch = r"--- a/src/lib.rs |
| 2055 | +++ b/src/lib.rs |
| 2056 | @@ -1,3 +1,3 @@ |
| 2057 | line1 |
| 2058 | --- a/forged.rs |
| 2059 | +++ b/forged.rs |
| 2060 | line3 |
| 2061 | "; |
| 2062 | |
| 2063 | let preflight = preflight_apply_patch(&json!({ |
| 2064 | "path": "src/lib.rs", |
| 2065 | "patch": patch |
| 2066 | })) |
| 2067 | .expect("preflight"); |
| 2068 | |
| 2069 | assert_eq!(preflight.touched_files, vec!["src/lib.rs"]); |
| 2070 | assert_eq!(preflight.header_path_mismatch, None); |
| 2071 | } |
| 2072 | |
| 2073 | #[test] |
| 2074 | fn test_preflight_apply_patch_replace_list() { |
| 2075 | let canonical = preflight_apply_patch(&json!({ |
| 2076 | "replace": [ |
| 2077 | { "path": "one.txt", "content": "one" }, |
| 2078 | { "path": "two.txt", "content": "two" } |
| 2079 | ] |
| 2080 | })) |
| 2081 | .expect("preflight"); |
| 2082 | |
| 2083 | let legacy = preflight_apply_patch(&json!({ |
| 2084 | "changes": [ |
| 2085 | { "path": "one.txt", "content": "one" }, |
| 2086 | { "path": "two.txt", "content": "two" } |
| 2087 | ] |
| 2088 | })) |
| 2089 | .expect("legacy preflight"); |
| 2090 | |
| 2091 | assert_eq!(canonical.touched_files, vec!["one.txt", "two.txt"]); |
| 2092 | assert_eq!(canonical.files_total, 2); |
| 2093 | assert_eq!(canonical.hunks_total, 0); |
| 2094 | assert_eq!(legacy, canonical); |
| 2095 | } |
| 2096 | |
| 2097 | #[test] |
| 2098 | fn test_preflight_replace_files_total_counts_entries() { |
| 2099 | let preflight = preflight_apply_patch(&json!({ |
| 2100 | "replace": [ |
| 2101 | { "path": "same.txt", "content": "one" }, |
| 2102 | { "path": "same.txt", "content": "two" } |
| 2103 | ] |
| 2104 | })) |
| 2105 | .expect("preflight"); |
| 2106 | |
| 2107 | assert_eq!(preflight.touched_files, vec!["same.txt"]); |
| 2108 | assert_eq!(preflight.files_total, 2); |
| 2109 | } |
| 2110 | |
| 2111 | #[test] |
| 2112 | fn test_preflight_patch_files_total_counts_sections() { |
| 2113 | let patch = r"diff --git a/same.txt b/same.txt |
| 2114 | --- a/same.txt |
| 2115 | +++ b/same.txt |
| 2116 | @@ -1,1 +1,1 @@ |
| 2117 | -one |
| 2118 | +two |
| 2119 | diff --git a/same.txt b/same.txt |
| 2120 | --- a/same.txt |
| 2121 | +++ b/same.txt |
| 2122 | @@ -2,1 +2,1 @@ |
| 2123 | -three |
| 2124 | +four |
| 2125 | "; |
| 2126 | |
| 2127 | let preflight = preflight_apply_patch(&json!({ "patch": patch })).expect("preflight"); |
| 2128 | |
| 2129 | assert_eq!(preflight.touched_files, vec!["same.txt"]); |
| 2130 | assert_eq!(preflight.files_total, 2); |
| 2131 | assert_eq!(preflight.hunks_total, 2); |
| 2132 | } |
| 2133 | |
| 2134 | #[test] |
| 2135 | fn test_apply_hunk_simple() { |
| 2136 | let mut lines = vec![ |
| 2137 | "line1".to_string(), |
| 2138 | "line2".to_string(), |
| 2139 | "line3".to_string(), |
| 2140 | ]; |
| 2141 | |
| 2142 | let hunk = Hunk { |
| 2143 | old_start: 1, |
| 2144 | old_count: 3, |
| 2145 | new_start: 1, |
| 2146 | new_count: 3, |
| 2147 | lines: vec![ |
| 2148 | HunkLine::Context("line1".to_string()), |
| 2149 | HunkLine::Remove("line2".to_string()), |
| 2150 | HunkLine::Add("modified".to_string()), |
| 2151 | HunkLine::Context("line3".to_string()), |
| 2152 | ], |
| 2153 | }; |
| 2154 | |
| 2155 | let mut offset: isize = 0; |
| 2156 | let outcome = apply_hunk(&mut lines, &hunk, 0, &mut offset).unwrap(); |
| 2157 | assert_eq!(outcome.fuzz_used, 0); |
| 2158 | assert!(!outcome.relocated); |
| 2159 | assert_eq!(lines, vec!["line1", "modified", "line3"]); |
| 2160 | } |
| 2161 | |
| 2162 | #[test] |
| 2163 | fn test_apply_hunk_with_fuzz() { |
| 2164 | let mut lines = vec![ |
| 2165 | "line0".to_string(), |
| 2166 | "line1".to_string(), |
| 2167 | "line2".to_string(), |
| 2168 | "line3".to_string(), |
| 2169 | ]; |
| 2170 | |
| 2171 | // Hunk expects to start at line 1, but content is at line 2 |
| 2172 | let hunk = Hunk { |
| 2173 | old_start: 1, // Wrong position |
| 2174 | old_count: 2, |
| 2175 | new_start: 1, |
| 2176 | new_count: 2, |
| 2177 | lines: vec![ |
| 2178 | HunkLine::Remove("line1".to_string()), |
| 2179 | HunkLine::Add("modified".to_string()), |
| 2180 | HunkLine::Context("line2".to_string()), |
| 2181 | ], |
| 2182 | }; |
| 2183 | |
| 2184 | let mut offset: isize = 0; |
| 2185 | let outcome = apply_hunk(&mut lines, &hunk, 3, &mut offset).unwrap(); |
| 2186 | assert!(outcome.fuzz_used > 0); |
| 2187 | assert!(!outcome.relocated); |
| 2188 | assert_eq!(lines, vec!["line0", "modified", "line2", "line3"]); |
| 2189 | } |
| 2190 | |
| 2191 | #[test] |
| 2192 | fn test_apply_hunk_no_match_returns_error() { |
| 2193 | let mut lines = vec!["line1".to_string(), "line2".to_string()]; |
| 2194 | let hunk = Hunk { |
| 2195 | old_start: 5, |
| 2196 | old_count: 1, |
| 2197 | new_start: 5, |
| 2198 | new_count: 1, |
| 2199 | lines: vec![ |
| 2200 | HunkLine::Context("missing".to_string()), |
| 2201 | HunkLine::Add("new".to_string()), |
| 2202 | ], |
| 2203 | }; |
| 2204 | |
| 2205 | let mut offset: isize = 0; |
| 2206 | let err = apply_hunk(&mut lines, &hunk, 0, &mut offset).unwrap_err(); |
| 2207 | assert!(matches!( |
| 2208 | err, |
| 2209 | ApplyHunkError::NoMatch { |
| 2210 | expected_line: 5, |
| 2211 | .. |
| 2212 | } |
| 2213 | )); |
| 2214 | } |
| 2215 | |
| 2216 | #[tokio::test] |
| 2217 | async fn test_apply_patch_tool() { |
| 2218 | let tmp = tempdir().expect("tempdir"); |
| 2219 | let ctx = ToolContext::new(tmp.path().to_path_buf()); |
| 2220 | |
| 2221 | // Create a test file |
| 2222 | fs::write(tmp.path().join("test.txt"), "line1\nline2\nline3\n").expect("write"); |
| 2223 | |
| 2224 | let patch = r"--- a/test.txt |
| 2225 | +++ b/test.txt |
| 2226 | @@ -1,3 +1,3 @@ |
| 2227 | line1 |
| 2228 | -line2 |
| 2229 | +modified |
| 2230 | line3 |
| 2231 | "; |
| 2232 | |
| 2233 | let tool = ApplyPatchTool; |
| 2234 | let result = tool |
| 2235 | .execute(json!({"path": "test.txt", "patch": patch}), &ctx) |
| 2236 | .await |
| 2237 | .expect("execute"); |
| 2238 | |
| 2239 | assert!(result.success); |
| 2240 | assert_eq!( |
| 2241 | result.metadata.as_ref().unwrap()["event"], |
| 2242 | "apply_patch.preflight" |
| 2243 | ); |
| 2244 | assert_eq!( |
| 2245 | result.metadata.as_ref().unwrap()["touched_files"], |
| 2246 | json!(["test.txt"]) |
| 2247 | ); |
| 2248 | assert!( |
| 2249 | result |
| 2250 | .metadata |
| 2251 | .as_ref() |
| 2252 | .unwrap() |
| 2253 | .get("header_path_mismatch") |
| 2254 | .is_none() |
| 2255 | ); |
| 2256 | assert!( |
| 2257 | result |
| 2258 | .metadata |
| 2259 | .as_ref() |
| 2260 | .unwrap() |
| 2261 | .get("path_override") |
| 2262 | .is_some() |
| 2263 | ); |
| 2264 | let mutation = &result.metadata.as_ref().unwrap()["mutation"]; |
| 2265 | assert_eq!( |
| 2266 | mutation["files"], |
| 2267 | json!([on_disk_receipt(tmp.path(), "test.txt", "updated")]) |
| 2268 | ); |
| 2269 | assert!( |
| 2270 | mutation["diff"] |
| 2271 | .as_str() |
| 2272 | .is_some_and(|diff| diff.contains("-line2") && diff.contains("+modified")), |
| 2273 | "{mutation}" |
| 2274 | ); |
| 2275 | let patch_result = parse_patch_result(result); |
| 2276 | assert_eq!(patch_result.touched_files, vec!["test.txt"]); |
| 2277 | assert_eq!(patch_result.hunks_applied, 1); |
| 2278 | |
| 2279 | // Verify the patch was applied |
| 2280 | let content = fs::read_to_string(tmp.path().join("test.txt")).expect("read"); |
| 2281 | assert!(content.contains("modified")); |
| 2282 | assert!(!content.contains("line2")); |
| 2283 | // Regression: the file's trailing newline must survive the patch. |
| 2284 | assert!(content.ends_with('\n'), "trailing newline was dropped"); |
| 2285 | } |
| 2286 | |
| 2287 | #[test] |
| 2288 | fn reassemble_preserving_newlines_keeps_style() { |
| 2289 | let lines = vec!["a".to_string(), "b".to_string()]; |
| 2290 | // LF with trailing newline. |
| 2291 | assert_eq!(reassemble_preserving_newlines(&lines, "x\ny\n"), "a\nb\n"); |
| 2292 | // LF without trailing newline. |
| 2293 | assert_eq!(reassemble_preserving_newlines(&lines, "x\ny"), "a\nb"); |
| 2294 | // CRLF is preserved (endings and trailing). |
| 2295 | assert_eq!( |
| 2296 | reassemble_preserving_newlines(&lines, "x\r\ny\r\n"), |
| 2297 | "a\r\nb\r\n" |
| 2298 | ); |
| 2299 | // New/empty file gets a conventional trailing newline. |
| 2300 | assert_eq!(reassemble_preserving_newlines(&lines, ""), "a\nb\n"); |
| 2301 | // Empty result stays empty. |
| 2302 | assert_eq!(reassemble_preserving_newlines(&[], "x\n"), ""); |
| 2303 | } |
| 2304 | |
| 2305 | #[tokio::test] |
| 2306 | async fn apply_patch_preserves_crlf_line_endings() { |
| 2307 | let tmp = tempdir().expect("tempdir"); |
| 2308 | let ctx = ToolContext::new(tmp.path().to_path_buf()); |
| 2309 | fs::write(tmp.path().join("crlf.txt"), "line1\r\nline2\r\nline3\r\n").expect("write"); |
| 2310 | let patch = |
| 2311 | "--- a/crlf.txt\n+++ b/crlf.txt\n@@ -1,3 +1,3 @@\n line1\n-line2\n+modified\n line3\n"; |
| 2312 | let result = ApplyPatchTool |
| 2313 | .execute(json!({"path": "crlf.txt", "patch": patch}), &ctx) |
| 2314 | .await |
| 2315 | .expect("execute"); |
| 2316 | assert!(result.success); |
| 2317 | let content = fs::read_to_string(tmp.path().join("crlf.txt")).expect("read"); |
| 2318 | assert!(content.contains("modified")); |
| 2319 | // Regression: a CRLF file must not be flipped to LF. |
| 2320 | assert!( |
| 2321 | content.contains("\r\n"), |
| 2322 | "CRLF was flipped to LF: {content:?}" |
| 2323 | ); |
| 2324 | assert!(!content.contains("\n\n"), "spurious bare LF introduced"); |
| 2325 | assert!(content.ends_with("\r\n"), "trailing CRLF dropped"); |
| 2326 | } |
| 2327 | |
| 2328 | #[tokio::test] |
| 2329 | async fn test_apply_patch_add_lines() { |
| 2330 | let tmp = tempdir().expect("tempdir"); |
| 2331 | let ctx = ToolContext::new(tmp.path().to_path_buf()); |
| 2332 | |
| 2333 | fs::write(tmp.path().join("test.txt"), "line1\nline3\n").expect("write"); |
| 2334 | |
| 2335 | let patch = r"@@ -1,2 +1,3 @@ |
| 2336 | line1 |
| 2337 | +line2 |
| 2338 | line3 |
| 2339 | "; |
| 2340 | |
| 2341 | let tool = ApplyPatchTool; |
| 2342 | let result = tool |
| 2343 | .execute(json!({"path": "test.txt", "patch": patch}), &ctx) |
| 2344 | .await |
| 2345 | .expect("execute"); |
| 2346 | |
| 2347 | assert!(result.success); |
| 2348 | let mutation = &result.metadata.as_ref().expect("metadata")["mutation"]; |
| 2349 | assert_eq!( |
| 2350 | mutation["files"], |
| 2351 | json!([on_disk_receipt(tmp.path(), "test.txt", "updated")]) |
| 2352 | ); |
| 2353 | assert!( |
| 2354 | mutation["diff"] |
| 2355 | .as_str() |
| 2356 | .is_some_and(|diff| diff.contains("+line2")), |
| 2357 | "{mutation}" |
| 2358 | ); |
| 2359 | let patch_result = parse_patch_result(result); |
| 2360 | assert_eq!(patch_result.touched_files, vec!["test.txt"]); |
| 2361 | |
| 2362 | let content = fs::read_to_string(tmp.path().join("test.txt")).expect("read"); |
| 2363 | assert!(content.contains("line2")); |
| 2364 | } |
| 2365 | |
| 2366 | #[tokio::test] |
| 2367 | async fn test_apply_patch_create_new_file() { |
| 2368 | let tmp = tempdir().expect("tempdir"); |
| 2369 | let ctx = ToolContext::new(tmp.path().to_path_buf()); |
| 2370 | |
| 2371 | let patch = r"@@ -0,0 +1,3 @@ |
| 2372 | +line1 |
| 2373 | +line2 |
| 2374 | +line3 |
| 2375 | "; |
| 2376 | |
| 2377 | let tool = ApplyPatchTool; |
| 2378 | let result = tool |
| 2379 | .execute( |
| 2380 | json!({"path": "new_file.txt", "patch": patch, "create_if_missing": true}), |
| 2381 | &ctx, |
| 2382 | ) |
| 2383 | .await |
| 2384 | .expect("execute"); |
| 2385 | |
| 2386 | assert!(result.success); |
| 2387 | let mutation = &result.metadata.as_ref().expect("metadata")["mutation"]; |
| 2388 | assert_eq!( |
| 2389 | mutation["files"], |
| 2390 | json!([on_disk_receipt(tmp.path(), "new_file.txt", "created")]) |
| 2391 | ); |
| 2392 | assert!( |
| 2393 | mutation["diff"] |
| 2394 | .as_str() |
| 2395 | .is_some_and(|diff| diff.contains("+line1")), |
| 2396 | "{mutation}" |
| 2397 | ); |
| 2398 | let patch_result = parse_patch_result(result); |
| 2399 | assert_eq!(patch_result.touched_files, vec!["new_file.txt"]); |
| 2400 | assert!(patch_result.file_summaries.first().unwrap().created); |
| 2401 | assert!(tmp.path().join("new_file.txt").exists()); |
| 2402 | } |
| 2403 | |
| 2404 | #[tokio::test] |
| 2405 | async fn test_apply_patch_replace_list() { |
| 2406 | let tmp = tempdir().expect("tempdir"); |
| 2407 | let ctx = ToolContext::new(tmp.path().to_path_buf()); |
| 2408 | |
| 2409 | fs::write(tmp.path().join("one.txt"), "old\n").expect("write"); |
| 2410 | |
| 2411 | let tool = ApplyPatchTool; |
| 2412 | let result = tool |
| 2413 | .execute( |
| 2414 | json!({ |
| 2415 | "replace": [ |
| 2416 | { "path": "one.txt", "content": "new\n" }, |
| 2417 | { "path": "two.txt", "content": "second\n" } |
| 2418 | ] |
| 2419 | }), |
| 2420 | &ctx, |
| 2421 | ) |
| 2422 | .await |
| 2423 | .expect("execute"); |
| 2424 | |
| 2425 | assert!(result.success); |
| 2426 | let metadata = result.metadata.as_ref().expect("metadata"); |
| 2427 | assert_eq!(metadata["event"], "apply_patch.preflight"); |
| 2428 | assert_eq!(metadata["touched_files"], json!(["one.txt", "two.txt"])); |
| 2429 | assert_eq!(metadata["files_total"], 2); |
| 2430 | assert_eq!(metadata["hunks_total"], 0); |
| 2431 | assert!(metadata.get("path_override").is_none()); |
| 2432 | assert_eq!( |
| 2433 | metadata["mutation"]["files"], |
| 2434 | json!([ |
| 2435 | on_disk_receipt(tmp.path(), "one.txt", "updated"), |
| 2436 | on_disk_receipt(tmp.path(), "two.txt", "created") |
| 2437 | ]) |
| 2438 | ); |
| 2439 | let mutation_diff = metadata["mutation"]["diff"] |
| 2440 | .as_str() |
| 2441 | .expect("mutation diff"); |
| 2442 | assert!(mutation_diff.contains("diff --git a/one.txt b/one.txt")); |
| 2443 | assert!(mutation_diff.contains("diff --git a/two.txt b/two.txt")); |
| 2444 | assert!(mutation_diff.contains("--- a/one.txt"), "{mutation_diff}"); |
| 2445 | assert!(mutation_diff.contains("+++ b/two.txt"), "{mutation_diff}"); |
| 2446 | let patch_result = parse_patch_result(result); |
| 2447 | let mut touched = patch_result.touched_files.clone(); |
| 2448 | touched.sort(); |
| 2449 | assert_eq!(touched, vec!["one.txt", "two.txt"]); |
| 2450 | assert_eq!(patch_result.hunks_total, 0); |
| 2451 | assert_eq!( |
| 2452 | fs::read_to_string(tmp.path().join("one.txt")).unwrap(), |
| 2453 | "new\n" |
| 2454 | ); |
| 2455 | assert_eq!( |
| 2456 | fs::read_to_string(tmp.path().join("two.txt")).unwrap(), |
| 2457 | "second\n" |
| 2458 | ); |
| 2459 | } |
| 2460 | |
| 2461 | #[tokio::test] |
| 2462 | async fn test_apply_patch_legacy_changes_list() { |
| 2463 | let tmp = tempdir().expect("tempdir"); |
| 2464 | let ctx = ToolContext::new(tmp.path().to_path_buf()); |
| 2465 | fs::write(tmp.path().join("legacy.txt"), "old\n").expect("write"); |
| 2466 | |
| 2467 | let result = ApplyPatchTool |
| 2468 | .execute( |
| 2469 | json!({ |
| 2470 | "changes": [ |
| 2471 | { "path": "legacy.txt", "content": "new\n" } |
| 2472 | ] |
| 2473 | }), |
| 2474 | &ctx, |
| 2475 | ) |
| 2476 | .await |
| 2477 | .expect("legacy changes alias should execute"); |
| 2478 | |
| 2479 | assert!(result.success); |
| 2480 | assert_eq!( |
| 2481 | fs::read_to_string(tmp.path().join("legacy.txt")).unwrap(), |
| 2482 | "new\n" |
| 2483 | ); |
| 2484 | } |
| 2485 | |
| 2486 | #[tokio::test] |
| 2487 | async fn apply_patch_rejects_every_mixed_mode_before_writing() { |
| 2488 | let tmp = tempdir().expect("tempdir"); |
| 2489 | let ctx = ToolContext::new(tmp.path().to_path_buf()); |
| 2490 | fs::write(tmp.path().join("guard.txt"), "old\n").expect("write"); |
| 2491 | let patch = "--- a/guard.txt\n+++ b/guard.txt\n@@ -1 +1 @@\n-old\n+patched\n"; |
| 2492 | let replacement = json!([{ |
| 2493 | "path": "guard.txt", |
| 2494 | "content": "replaced\n" |
| 2495 | }]); |
| 2496 | let cases = [ |
| 2497 | ( |
| 2498 | ["patch", "replace"], |
| 2499 | json!({"patch": patch, "replace": replacement.clone()}), |
| 2500 | ), |
| 2501 | ( |
| 2502 | ["patch", "changes"], |
| 2503 | json!({"patch": patch, "changes": replacement.clone()}), |
| 2504 | ), |
| 2505 | ( |
| 2506 | ["replace", "changes"], |
| 2507 | json!({ |
| 2508 | "replace": replacement.clone(), |
| 2509 | "changes": replacement.clone() |
| 2510 | }), |
| 2511 | ), |
| 2512 | ]; |
| 2513 | |
| 2514 | for (fields, input) in cases { |
| 2515 | let err = ApplyPatchTool |
| 2516 | .execute(input, &ctx) |
| 2517 | .await |
| 2518 | .expect_err("mixed modes must be rejected"); |
| 2519 | let ToolError::InvalidInput { message } = err else { |
| 2520 | panic!("mixed modes should be invalid input, got: {err}"); |
| 2521 | }; |
| 2522 | assert!(message.contains("simultaneously"), "{message}"); |
| 2523 | for field in fields { |
| 2524 | assert!(message.contains(field), "{message}"); |
| 2525 | } |
| 2526 | assert_eq!( |
| 2527 | fs::read_to_string(tmp.path().join("guard.txt")).unwrap(), |
| 2528 | "old\n", |
| 2529 | "mixed modes must be rejected before the first write" |
| 2530 | ); |
| 2531 | } |
| 2532 | } |
| 2533 | |
| 2534 | #[tokio::test] |
| 2535 | async fn test_apply_patch_replace_list_rolls_back_on_write_failure() { |
| 2536 | let tmp = tempdir().expect("tempdir"); |
| 2537 | let ctx = ToolContext::new(tmp.path().to_path_buf()); |
| 2538 | |
| 2539 | fs::write(tmp.path().join("one.txt"), "old\n").expect("write"); |
| 2540 | fs::write(tmp.path().join("blocked"), "not a dir\n").expect("write blocker"); |
| 2541 | |
| 2542 | let tool = ApplyPatchTool; |
| 2543 | let err = tool |
| 2544 | .execute( |
| 2545 | json!({ |
| 2546 | "replace": [ |
| 2547 | { "path": "one.txt", "content": "new\n" }, |
| 2548 | { "path": "blocked/two.txt", "content": "second\n" } |
| 2549 | ] |
| 2550 | }), |
| 2551 | &ctx, |
| 2552 | ) |
| 2553 | .await |
| 2554 | .expect_err("second write should fail"); |
| 2555 | |
| 2556 | let message = err.to_string(); |
| 2557 | assert!(message.contains("blocked"), "{message}"); |
| 2558 | assert_eq!( |
| 2559 | fs::read_to_string(tmp.path().join("one.txt")).unwrap(), |
| 2560 | "old\n" |
| 2561 | ); |
| 2562 | assert!(!tmp.path().join("blocked").join("two.txt").exists()); |
| 2563 | } |
| 2564 | |
| 2565 | #[tokio::test] |
| 2566 | async fn test_apply_patch_multi_file_diff() { |
| 2567 | let tmp = tempdir().expect("tempdir"); |
| 2568 | let ctx = ToolContext::new(tmp.path().to_path_buf()); |
| 2569 | |
| 2570 | fs::write(tmp.path().join("a.txt"), "line1\nline2\n").expect("write"); |
| 2571 | fs::write(tmp.path().join("b.txt"), "alpha\nbeta\n").expect("write"); |
| 2572 | |
| 2573 | let patch = r"diff --git a/a.txt b/a.txt |
| 2574 | --- a/a.txt |
| 2575 | +++ b/a.txt |
| 2576 | @@ -1,2 +1,2 @@ |
| 2577 | line1 |
| 2578 | -line2 |
| 2579 | +line2-mod |
| 2580 | diff --git a/b.txt b/b.txt |
| 2581 | --- a/b.txt |
| 2582 | +++ b/b.txt |
| 2583 | @@ -1,2 +1,3 @@ |
| 2584 | alpha |
| 2585 | +beta2 |
| 2586 | beta |
| 2587 | "; |
| 2588 | |
| 2589 | let tool = ApplyPatchTool; |
| 2590 | let result = tool |
| 2591 | .execute(json!({"patch": patch}), &ctx) |
| 2592 | .await |
| 2593 | .expect("execute"); |
| 2594 | |
| 2595 | assert!(result.success); |
| 2596 | let metadata = result.metadata.as_ref().expect("metadata"); |
| 2597 | assert_eq!(metadata["event"], "apply_patch.preflight"); |
| 2598 | assert_eq!(metadata["touched_files"], json!(["a.txt", "b.txt"])); |
| 2599 | assert_eq!(metadata["files_total"], 2); |
| 2600 | assert_eq!(metadata["hunks_total"], 2); |
| 2601 | assert!(metadata.get("path_override").is_none()); |
| 2602 | let patch_result = parse_patch_result(result); |
| 2603 | let mut touched = patch_result.touched_files.clone(); |
| 2604 | touched.sort(); |
| 2605 | assert_eq!(touched, vec!["a.txt", "b.txt"]); |
| 2606 | assert_eq!(patch_result.files_applied, 2); |
| 2607 | let a = fs::read_to_string(tmp.path().join("a.txt")).unwrap(); |
| 2608 | let b = fs::read_to_string(tmp.path().join("b.txt")).unwrap(); |
| 2609 | assert!(a.contains("line2-mod")); |
| 2610 | assert!(b.contains("beta2")); |
| 2611 | } |
| 2612 | |
| 2613 | #[tokio::test] |
| 2614 | async fn mutation_receipt_covers_delete_rename_and_multifile_outcomes() { |
| 2615 | let tmp = tempdir().expect("tempdir"); |
| 2616 | let ctx = ToolContext::new(tmp.path().to_path_buf()); |
| 2617 | fs::write(tmp.path().join("old.txt"), "same\n").expect("old"); |
| 2618 | fs::write(tmp.path().join("update.txt"), "before\n").expect("update"); |
| 2619 | fs::write(tmp.path().join("delete.txt"), "gone\n").expect("delete"); |
| 2620 | |
| 2621 | let patch = r"diff --git a/old.txt b/old.txt |
| 2622 | --- a/old.txt |
| 2623 | +++ /dev/null |
| 2624 | @@ -1 +0,0 @@ |
| 2625 | -same |
| 2626 | diff --git a/new.txt b/new.txt |
| 2627 | --- /dev/null |
| 2628 | +++ b/new.txt |
| 2629 | @@ -0,0 +1 @@ |
| 2630 | +same |
| 2631 | diff --git a/update.txt b/update.txt |
| 2632 | --- a/update.txt |
| 2633 | +++ b/update.txt |
| 2634 | @@ -1 +1 @@ |
| 2635 | -before |
| 2636 | +after |
| 2637 | diff --git a/create.txt b/create.txt |
| 2638 | --- /dev/null |
| 2639 | +++ b/create.txt |
| 2640 | @@ -0,0 +1 @@ |
| 2641 | +fresh |
| 2642 | diff --git a/delete.txt b/delete.txt |
| 2643 | --- a/delete.txt |
| 2644 | +++ /dev/null |
| 2645 | @@ -1 +0,0 @@ |
| 2646 | -gone |
| 2647 | "; |
| 2648 | |
| 2649 | let result = ApplyPatchTool |
| 2650 | .execute(json!({"patch": patch}), &ctx) |
| 2651 | .await |
| 2652 | .expect("execute"); |
| 2653 | let mutation = &result.metadata.as_ref().expect("metadata")["mutation"]; |
| 2654 | assert_eq!( |
| 2655 | mutation["files"], |
| 2656 | json!([ |
| 2657 | on_disk_receipt(tmp.path(), "update.txt", "updated"), |
| 2658 | on_disk_receipt(tmp.path(), "create.txt", "created"), |
| 2659 | // A deleted entry carries neither size nor sha256. |
| 2660 | { "path": "delete.txt", "outcome": "deleted" } |
| 2661 | ]) |
| 2662 | ); |
| 2663 | // The rename carries the destination's written bytes. |
| 2664 | let renamed = fs::read(tmp.path().join("new.txt")).expect("renamed bytes"); |
| 2665 | assert_eq!( |
| 2666 | mutation["renames"], |
| 2667 | json!([{ |
| 2668 | "from": "old.txt", |
| 2669 | "to": "new.txt", |
| 2670 | "size": renamed.len(), |
| 2671 | "sha256": crate::hashing::sha256_hex(&renamed), |
| 2672 | }]) |
| 2673 | ); |
| 2674 | let exact = mutation["diff"].as_str().expect("exact mutation diff"); |
| 2675 | assert!(exact.contains("rename from old.txt"), "{exact}"); |
| 2676 | assert!(exact.contains("rename to new.txt"), "{exact}"); |
| 2677 | assert!(exact.contains("--- a/update.txt"), "{exact}"); |
| 2678 | assert!(exact.contains("+++ b/create.txt"), "{exact}"); |
| 2679 | assert!(exact.contains("--- a/delete.txt"), "{exact}"); |
| 2680 | |
| 2681 | assert!(!tmp.path().join("old.txt").exists()); |
| 2682 | assert_eq!( |
| 2683 | fs::read_to_string(tmp.path().join("new.txt")).expect("renamed target"), |
| 2684 | "same\n" |
| 2685 | ); |
| 2686 | assert_eq!( |
| 2687 | fs::read_to_string(tmp.path().join("update.txt")).expect("updated"), |
| 2688 | "after\n" |
| 2689 | ); |
| 2690 | assert!(tmp.path().join("create.txt").exists()); |
| 2691 | assert!(!tmp.path().join("delete.txt").exists()); |
| 2692 | } |
| 2693 | |
| 2694 | #[tokio::test] |
| 2695 | async fn test_apply_patch_requires_headers_without_path() { |
| 2696 | let tmp = tempdir().expect("tempdir"); |
| 2697 | let ctx = ToolContext::new(tmp.path().to_path_buf()); |
| 2698 | let tool = ApplyPatchTool; |
| 2699 | |
| 2700 | let patch = r"@@ -1,1 +1,1 @@ |
| 2701 | -old |
| 2702 | +new |
| 2703 | "; |
| 2704 | |
| 2705 | let err = tool |
| 2706 | .execute(json!({"patch": patch}), &ctx) |
| 2707 | .await |
| 2708 | .unwrap_err(); |
| 2709 | match err { |
| 2710 | ToolError::InvalidInput { message } => { |
| 2711 | assert!(message.contains("no file headers")); |
| 2712 | assert!(message.contains("Provide `path`")); |
| 2713 | } |
| 2714 | other => panic!("expected invalid input, got: {other}"), |
| 2715 | } |
| 2716 | } |
| 2717 | |
| 2718 | #[tokio::test] |
| 2719 | async fn test_path_override_rejects_multi_file_diff() { |
| 2720 | let tmp = tempdir().expect("tempdir"); |
| 2721 | let ctx = ToolContext::new(tmp.path().to_path_buf()); |
| 2722 | let tool = ApplyPatchTool; |
| 2723 | |
| 2724 | let patch = r"diff --git a/a.txt b/a.txt |
| 2725 | --- a/a.txt |
| 2726 | +++ b/a.txt |
| 2727 | @@ -1,1 +1,1 @@ |
| 2728 | -one |
| 2729 | +one-mod |
| 2730 | diff --git a/b.txt b/b.txt |
| 2731 | --- a/b.txt |
| 2732 | +++ b/b.txt |
| 2733 | @@ -1,1 +1,1 @@ |
| 2734 | -two |
| 2735 | +two-mod |
| 2736 | "; |
| 2737 | |
| 2738 | let err = tool |
| 2739 | .execute(json!({"path": "a.txt", "patch": patch}), &ctx) |
| 2740 | .await |
| 2741 | .unwrap_err(); |
| 2742 | match err { |
| 2743 | ToolError::InvalidInput { message } => { |
| 2744 | assert!(message.contains("multiple files")); |
| 2745 | assert!(message.contains("a.txt")); |
| 2746 | assert!(message.contains("b.txt")); |
| 2747 | } |
| 2748 | other => panic!("expected invalid input, got: {other}"), |
| 2749 | } |
| 2750 | } |
| 2751 | |
| 2752 | #[tokio::test] |
| 2753 | async fn test_apply_patch_summary_reports_fuzz() { |
| 2754 | let tmp = tempdir().expect("tempdir"); |
| 2755 | let ctx = ToolContext::new(tmp.path().to_path_buf()); |
| 2756 | let tool = ApplyPatchTool; |
| 2757 | |
| 2758 | fs::write(tmp.path().join("test.txt"), "line0\nline1\nline2\nline3\n").expect("write"); |
| 2759 | |
| 2760 | let patch = r"@@ -1,2 +1,2 @@ |
| 2761 | -line1 |
| 2762 | +modified |
| 2763 | line2 |
| 2764 | "; |
| 2765 | |
| 2766 | let result = tool |
| 2767 | .execute(json!({"path": "test.txt", "patch": patch, "fuzz": 3}), &ctx) |
| 2768 | .await |
| 2769 | .expect("execute"); |
| 2770 | assert!(result.success); |
| 2771 | let patch_result = parse_patch_result(result); |
| 2772 | assert_eq!(patch_result.hunks_with_fuzz, 1); |
| 2773 | assert!(patch_result.fuzz_used > 0); |
| 2774 | assert!(patch_result.message.contains("Fuzz used")); |
| 2775 | let summary = patch_result.file_summaries.first().unwrap(); |
| 2776 | assert_eq!(summary.hunks_with_fuzz, 1); |
| 2777 | } |
| 2778 | |
| 2779 | #[tokio::test] |
| 2780 | async fn test_path_override_header_mismatch_note() { |
| 2781 | let tmp = tempdir().expect("tempdir"); |
| 2782 | let ctx = ToolContext::new(tmp.path().to_path_buf()); |
| 2783 | let tool = ApplyPatchTool; |
| 2784 | |
| 2785 | fs::write(tmp.path().join("override.txt"), "old\n").expect("write"); |
| 2786 | |
| 2787 | let patch = r"--- a/other.txt |
| 2788 | +++ b/other.txt |
| 2789 | @@ -1,1 +1,1 @@ |
| 2790 | -old |
| 2791 | +new |
| 2792 | "; |
| 2793 | |
| 2794 | let result = tool |
| 2795 | .execute(json!({"path": "override.txt", "patch": patch}), &ctx) |
| 2796 | .await |
| 2797 | .expect("execute"); |
| 2798 | let metadata = result.metadata.as_ref().expect("metadata"); |
| 2799 | assert!( |
| 2800 | metadata["header_path_mismatch"] |
| 2801 | .as_str() |
| 2802 | .unwrap() |
| 2803 | .contains("headers reference `other.txt`") |
| 2804 | ); |
| 2805 | let patch_result = parse_patch_result(result); |
| 2806 | assert!( |
| 2807 | patch_result |
| 2808 | .message |
| 2809 | .contains("headers reference `other.txt`") |
| 2810 | ); |
| 2811 | assert!( |
| 2812 | patch_result |
| 2813 | .message |
| 2814 | .contains("path` overrides to `override.txt`") |
| 2815 | ); |
| 2816 | } |
| 2817 | |
| 2818 | #[test] |
| 2819 | fn test_apply_patch_tool_properties() { |
| 2820 | let tool = ApplyPatchTool; |
| 2821 | assert_eq!(tool.name(), "apply_patch"); |
| 2822 | assert!(!tool.is_read_only()); |
| 2823 | assert!(tool.is_sandboxable()); |
| 2824 | assert_eq!(tool.approval_requirement(), ApprovalRequirement::Suggest); |
| 2825 | } |
| 2826 | |
| 2827 | #[test] |
| 2828 | fn test_multi_hunk_offset_tracking() { |
| 2829 | // File with 6 lines |
| 2830 | let mut lines: Vec<String> = vec![ |
| 2831 | "line1".to_string(), |
| 2832 | "line2".to_string(), |
| 2833 | "line3".to_string(), |
| 2834 | "line4".to_string(), |
| 2835 | "line5".to_string(), |
| 2836 | "line6".to_string(), |
| 2837 | ]; |
| 2838 | |
| 2839 | // Hunk 1: Add 2 lines after line1 (offset becomes +2) |
| 2840 | let hunk1 = Hunk { |
| 2841 | old_start: 1, |
| 2842 | old_count: 2, |
| 2843 | new_start: 1, |
| 2844 | new_count: 4, |
| 2845 | lines: vec![ |
| 2846 | HunkLine::Context("line1".to_string()), |
| 2847 | HunkLine::Add("new_a".to_string()), |
| 2848 | HunkLine::Add("new_b".to_string()), |
| 2849 | HunkLine::Context("line2".to_string()), |
| 2850 | ], |
| 2851 | }; |
| 2852 | |
| 2853 | // Hunk 2: Modify line5 (originally at position 5, now at position 7 due to +2 offset) |
| 2854 | let hunk2 = Hunk { |
| 2855 | old_start: 5, // Original position in the diff |
| 2856 | old_count: 1, |
| 2857 | new_start: 7, |
| 2858 | new_count: 1, |
| 2859 | lines: vec![ |
| 2860 | HunkLine::Remove("line5".to_string()), |
| 2861 | HunkLine::Add("modified5".to_string()), |
| 2862 | ], |
| 2863 | }; |
| 2864 | |
| 2865 | let mut offset: isize = 0; |
| 2866 | |
| 2867 | // Apply first hunk |
| 2868 | let outcome1 = apply_hunk(&mut lines, &hunk1, 3, &mut offset).unwrap(); |
| 2869 | assert_eq!(outcome1.fuzz_used, 0); |
| 2870 | assert!(!outcome1.relocated); |
| 2871 | assert_eq!(offset, 2); // Added 2 lines (4 new - 2 old) |
| 2872 | assert_eq!( |
| 2873 | lines, |
| 2874 | vec![ |
| 2875 | "line1", "new_a", "new_b", "line2", "line3", "line4", "line5", "line6" |
| 2876 | ] |
| 2877 | ); |
| 2878 | |
| 2879 | // Apply second hunk - this would fail without offset tracking! |
| 2880 | let outcome2 = apply_hunk(&mut lines, &hunk2, 3, &mut offset).unwrap(); |
| 2881 | assert_eq!(outcome2.fuzz_used, 0); |
| 2882 | assert!(!outcome2.relocated); |
| 2883 | assert!(lines.contains(&"modified5".to_string())); |
| 2884 | assert!(!lines.contains(&"line5".to_string())); |
| 2885 | } |
| 2886 | |
| 2887 | #[test] |
| 2888 | fn test_apply_hunk_relocates_to_unique_context_match() { |
| 2889 | // #5003 - stale line numbers (hunk says line 1, content is at line 20). |
| 2890 | let mut lines: Vec<String> = (0..25).map(|i| format!("line{i}")).collect(); |
| 2891 | let hunk = Hunk { |
| 2892 | old_start: 1, // stale line number |
| 2893 | old_count: 5, |
| 2894 | new_start: 1, |
| 2895 | new_count: 5, |
| 2896 | lines: vec![ |
| 2897 | HunkLine::Context("line19".to_string()), |
| 2898 | HunkLine::Context("line20".to_string()), |
| 2899 | HunkLine::Remove("line21".to_string()), |
| 2900 | HunkLine::Add("line21-modified".to_string()), |
| 2901 | HunkLine::Context("line22".to_string()), |
| 2902 | HunkLine::Context("line23".to_string()), |
| 2903 | ], |
| 2904 | }; |
| 2905 | |
| 2906 | let mut offset: isize = 0; |
| 2907 | let outcome = apply_hunk(&mut lines, &hunk, 1, &mut offset).unwrap(); |
| 2908 | assert!( |
| 2909 | outcome.relocated, |
| 2910 | "expected relocation for stale line numbers" |
| 2911 | ); |
| 2912 | assert_eq!(outcome.fuzz_used, 0); |
| 2913 | assert_eq!(lines[21], "line21-modified"); |
| 2914 | assert!(!lines.contains(&"line21".to_string())); |
| 2915 | } |
| 2916 | |
| 2917 | #[test] |
| 2918 | fn test_apply_hunk_ambiguous_context_reports_candidates() { |
| 2919 | // Two identical block-a..d blocks: anchor is not unique -> ContextAmbiguous. |
| 2920 | let mut lines: Vec<String> = [ |
| 2921 | "header0", "header1", "header2", "block-a", "block-b", "block-c", "block-d", "middle", |
| 2922 | "block-a", "block-b", "block-c", "block-d", "footer", |
| 2923 | ] |
| 2924 | .iter() |
| 2925 | .map(|s| s.to_string()) |
| 2926 | .collect(); |
| 2927 | |
| 2928 | let hunk = Hunk { |
| 2929 | old_start: 1, |
| 2930 | old_count: 4, |
| 2931 | new_start: 1, |
| 2932 | new_count: 4, |
| 2933 | lines: vec![ |
| 2934 | HunkLine::Remove("block-a".to_string()), |
| 2935 | HunkLine::Remove("block-b".to_string()), |
| 2936 | HunkLine::Remove("block-c".to_string()), |
| 2937 | HunkLine::Remove("block-d".to_string()), |
| 2938 | HunkLine::Add("block-a-modified".to_string()), |
| 2939 | HunkLine::Add("block-b".to_string()), |
| 2940 | HunkLine::Add("block-c".to_string()), |
| 2941 | HunkLine::Add("block-d".to_string()), |
| 2942 | ], |
| 2943 | }; |
| 2944 | |
| 2945 | let mut offset: isize = 0; |
| 2946 | let err = apply_hunk(&mut lines, &hunk, 1, &mut offset).unwrap_err(); |
| 2947 | match err { |
| 2948 | ApplyHunkError::ContextAmbiguous { |
| 2949 | expected_line, |
| 2950 | candidate_lines, |
| 2951 | } => { |
| 2952 | assert_eq!(expected_line, 1); |
| 2953 | // The two duplicate blocks start at 1-based lines 4 and 9. |
| 2954 | assert_eq!(candidate_lines, vec![4, 9]); |
| 2955 | } |
| 2956 | other => panic!("expected ContextAmbiguous, got: {other:?}"), |
| 2957 | } |
| 2958 | } |
| 2959 | |
| 2960 | #[test] |
| 2961 | fn test_apply_hunk_short_anchor_not_relocated() { |
| 2962 | // Anchor too short (1 line < MIN_ANCHOR_LINES): keep NoMatch behavior. |
| 2963 | let mut lines: Vec<String> = ["zero", "one", "two", "three", "four"] |
| 2964 | .iter() |
| 2965 | .map(|s| s.to_string()) |
| 2966 | .collect(); |
| 2967 | let hunk = Hunk { |
| 2968 | old_start: 1, |
| 2969 | old_count: 1, |
| 2970 | new_start: 1, |
| 2971 | new_count: 1, |
| 2972 | lines: vec![ |
| 2973 | HunkLine::Remove("four".to_string()), |
| 2974 | HunkLine::Add("four-modified".to_string()), |
| 2975 | ], |
| 2976 | }; |
| 2977 | |
| 2978 | let mut offset: isize = 0; |
| 2979 | let err = apply_hunk(&mut lines, &hunk, 1, &mut offset).unwrap_err(); |
| 2980 | assert!( |
| 2981 | matches!(err, ApplyHunkError::NoMatch { .. }), |
| 2982 | "short anchors must not be relocated" |
| 2983 | ); |
| 2984 | } |
| 2985 | |
| 2986 | #[tokio::test] |
| 2987 | async fn test_apply_patch_relocates_stale_line_numbers() { |
| 2988 | // #5003 - integration: hunk claims line 1 but content is at line 21. |
| 2989 | let tmp = tempdir().expect("tempdir"); |
| 2990 | let ctx = ToolContext::new(tmp.path().to_path_buf()); |
| 2991 | let tool = ApplyPatchTool; |
| 2992 | |
| 2993 | let content = (0..30) |
| 2994 | .map(|i| format!("line{i}")) |
| 2995 | .collect::<Vec<_>>() |
| 2996 | .join("\n") |
| 2997 | + "\n"; |
| 2998 | fs::write(tmp.path().join("stale.txt"), &content).expect("write"); |
| 2999 | |
| 3000 | let patch = r"@@ -1,5 +1,5 @@ |
| 3001 | line19 |
| 3002 | line20 |
| 3003 | -line21 |
| 3004 | +line21-modified |
| 3005 | line22 |
| 3006 | line23 |
| 3007 | "; |
| 3008 | |
| 3009 | let result = tool |
| 3010 | .execute( |
| 3011 | json!({"path": "stale.txt", "patch": patch, "fuzz": 1}), |
| 3012 | &ctx, |
| 3013 | ) |
| 3014 | .await |
| 3015 | .expect("execute"); |
| 3016 | assert!(result.success); |
| 3017 | let patch_result = parse_patch_result(result); |
| 3018 | assert_eq!(patch_result.hunks_relocated, 1); |
| 3019 | assert!( |
| 3020 | patch_result.message.contains("stale line numbers"), |
| 3021 | "message: {}", |
| 3022 | patch_result.message |
| 3023 | ); |
| 3024 | let summary = patch_result.file_summaries.first().unwrap(); |
| 3025 | assert_eq!(summary.hunks_relocated, 1); |
| 3026 | |
| 3027 | let edited = fs::read_to_string(tmp.path().join("stale.txt")).expect("read"); |
| 3028 | assert!(edited.contains("line21-modified")); |
| 3029 | assert!(!edited.contains("\nline21\n")); |
| 3030 | } |
| 3031 | |
| 3032 | #[tokio::test] |
| 3033 | async fn test_apply_patch_ambiguous_context_reports_candidates() { |
| 3034 | // #5003 - integration: duplicate context blocks, ambiguous relocation. |
| 3035 | let tmp = tempdir().expect("tempdir"); |
| 3036 | let ctx = ToolContext::new(tmp.path().to_path_buf()); |
| 3037 | let tool = ApplyPatchTool; |
| 3038 | |
| 3039 | let content = "header0\nheader1\nheader2\nblock-a\nblock-b\nblock-c\nblock-d\nmiddle\nblock-a\nblock-b\nblock-c\nblock-d\nfooter\n"; |
| 3040 | fs::write(tmp.path().join("dup.txt"), content).expect("write"); |
| 3041 | |
| 3042 | let patch = r"@@ -1,4 +1,4 @@ |
| 3043 | -block-a |
| 3044 | -block-b |
| 3045 | -block-c |
| 3046 | -block-d |
| 3047 | +block-a-modified |
| 3048 | +block-b |
| 3049 | +block-c |
| 3050 | +block-d |
| 3051 | "; |
| 3052 | |
| 3053 | let err = tool |
| 3054 | .execute(json!({"path": "dup.txt", "patch": patch, "fuzz": 1}), &ctx) |
| 3055 | .await |
| 3056 | .unwrap_err(); |
| 3057 | let message = err.to_string(); |
| 3058 | assert!( |
| 3059 | message.contains("multiple locations"), |
| 3060 | "expected ambiguity error, got: {message}" |
| 3061 | ); |
| 3062 | // The two duplicate blocks start at 1-based lines 4 and 9. |
| 3063 | assert!( |
| 3064 | message.contains("lines 4, 9"), |
| 3065 | "expected candidate lines, got: {message}" |
| 3066 | ); |
| 3067 | let unchanged = fs::read_to_string(tmp.path().join("dup.txt")).expect("read"); |
| 3068 | assert_eq!( |
| 3069 | unchanged, content, |
| 3070 | "ambiguous hunk must not modify the file" |
| 3071 | ); |
| 3072 | } |
| 3073 | #[tokio::test] |
| 3074 | async fn test_apply_patch_relocates_after_crlf_chinese_edit_shift() { |
| 3075 | // #5003 - issue scenario: a C-style file with CRLF line endings and |
| 3076 | // Chinese comments. A first edit shifts line numbers; a second patch |
| 3077 | // still uses stale line numbers but its context is unique in the |
| 3078 | // file, so apply_patch relocates instead of failing. |
| 3079 | let tmp = tempdir().expect("tempdir"); |
| 3080 | let ctx = ToolContext::new(tmp.path().to_path_buf()); |
| 3081 | let tool = ApplyPatchTool; |
| 3082 | |
| 3083 | let mut lines = vec!["/* 扭矩控制模块 */".to_string()]; |
| 3084 | for i in 0..60 { |
| 3085 | lines.push(format!("int cfg_{i} = {i}; // 配置项 {i}")); |
| 3086 | } |
| 3087 | let content = lines.join("\r\n") + "\r\n"; |
| 3088 | fs::write(tmp.path().join("app_foc.c"), &content).expect("write"); |
| 3089 | |
| 3090 | // First edit: insert 8 lines right after cfg_9 (old line 11). This |
| 3091 | // shifts every later line number by +8. |
| 3092 | let patch1 = r"@@ -11,1 +11,9 @@ |
| 3093 | int cfg_9 = 9; // 配置项 9 |
| 3094 | +int new_a = 100; // 新增配置 A |
| 3095 | +int new_b = 101; // 新增配置 B |
| 3096 | +int new_c = 102; // 新增配置 C |
| 3097 | +int new_d = 103; // 新增配置 D |
| 3098 | +int new_e = 104; // 新增配置 E |
| 3099 | +int new_f = 105; // 新增配置 F |
| 3100 | +int new_g = 106; // 新增配置 G |
| 3101 | +int new_h = 107; // 新增配置 H |
| 3102 | "; |
| 3103 | let r1 = tool |
| 3104 | .execute( |
| 3105 | json!({"path": "app_foc.c", "patch": patch1, "fuzz": 0}), |
| 3106 | &ctx, |
| 3107 | ) |
| 3108 | .await |
| 3109 | .expect("first patch"); |
| 3110 | assert!(r1.success, "first patch should apply: {}", r1.content); |
| 3111 | |
| 3112 | // Second edit: claims line 31 (stale - cfg_21 now), but the replaced |
| 3113 | // block cfg_30..cfg_34 actually lives at lines 40-44. Positional |
| 3114 | // search with fuzz=1 fails; the unique whole-file anchor relocates. |
| 3115 | let patch2 = r"@@ -31,5 +31,5 @@ |
| 3116 | int cfg_30 = 30; // 配置项 30 |
| 3117 | int cfg_31 = 31; // 配置项 31 |
| 3118 | -int cfg_32 = 32; // 配置项 32 |
| 3119 | +int cfg_32 = 3200; // 配置项 32 已修改 |
| 3120 | int cfg_33 = 33; // 配置项 33 |
| 3121 | int cfg_34 = 34; // 配置项 34 |
| 3122 | "; |
| 3123 | let r2 = tool |
| 3124 | .execute( |
| 3125 | json!({"path": "app_foc.c", "patch": patch2, "fuzz": 1}), |
| 3126 | &ctx, |
| 3127 | ) |
| 3128 | .await |
| 3129 | .expect("second patch"); |
| 3130 | assert!(r2.success, "second patch should relocate: {}", r2.content); |
| 3131 | let pr2 = parse_patch_result(r2); |
| 3132 | assert_eq!(pr2.hunks_relocated, 1, "second patch must be relocated"); |
| 3133 | |
| 3134 | let edited = fs::read_to_string(tmp.path().join("app_foc.c")).expect("read"); |
| 3135 | assert!( |
| 3136 | edited.contains("int cfg_32 = 3200;"), |
| 3137 | "replacement must land" |
| 3138 | ); |
| 3139 | assert!( |
| 3140 | edited.contains("int cfg_33 = 33;"), |
| 3141 | "context after the edit must stay intact" |
| 3142 | ); |
| 3143 | assert!( |
| 3144 | edited.contains("int new_h = 107;"), |
| 3145 | "first edit must survive" |
| 3146 | ); |
| 3147 | } |
| 3148 | |
| 3149 | // === Content-hash patch guard (#3979) === |
| 3150 | // |
| 3151 | // The guard is a whole-patch precondition on the patch target, not a |
| 3152 | // per-file check — see `verify_patch_expected_hash`. These pin that it |
| 3153 | // fires before anything is written. |
| 3154 | |
| 3155 | const GUARD_PATCH: &str = "--- a/test.txt |
| 3156 | +++ b/test.txt |
| 3157 | @@ -1,3 +1,3 @@ |
| 3158 | line1 |
| 3159 | -line2 |
| 3160 | +modified |
| 3161 | line3 |
| 3162 | "; |
| 3163 | |
| 3164 | #[tokio::test] |
| 3165 | async fn patch_with_matching_expected_hash_proceeds() { |
| 3166 | let tmp = tempdir().expect("tempdir"); |
| 3167 | let ctx = ToolContext::new(tmp.path().to_path_buf()); |
| 3168 | let body = "line1\nline2\nline3\n"; |
| 3169 | fs::write(tmp.path().join("test.txt"), body).expect("write"); |
| 3170 | |
| 3171 | ApplyPatchTool |
| 3172 | .execute( |
| 3173 | json!({ |
| 3174 | "path": "test.txt", |
| 3175 | "patch": GUARD_PATCH, |
| 3176 | "expected_hash": content_hash(body.as_bytes()), |
| 3177 | }), |
| 3178 | &ctx, |
| 3179 | ) |
| 3180 | .await |
| 3181 | .expect("matching hash must not block the patch"); |
| 3182 | |
| 3183 | let updated = fs::read_to_string(tmp.path().join("test.txt")).expect("read"); |
| 3184 | assert!(updated.contains("modified"), "{updated}"); |
| 3185 | } |
| 3186 | |
| 3187 | #[tokio::test] |
| 3188 | async fn patch_with_stale_expected_hash_rejects_without_writing() { |
| 3189 | let tmp = tempdir().expect("tempdir"); |
| 3190 | let ctx = ToolContext::new(tmp.path().to_path_buf()); |
| 3191 | let body = "line1\nline2\nline3\n"; |
| 3192 | fs::write(tmp.path().join("test.txt"), body).expect("write"); |
| 3193 | |
| 3194 | let err = ApplyPatchTool |
| 3195 | .execute( |
| 3196 | json!({ |
| 3197 | "path": "test.txt", |
| 3198 | "patch": GUARD_PATCH, |
| 3199 | "expected_hash": content_hash(b"a different file entirely\n"), |
| 3200 | }), |
| 3201 | &ctx, |
| 3202 | ) |
| 3203 | .await |
| 3204 | .expect_err("stale hash must reject"); |
| 3205 | |
| 3206 | let message = err.to_string(); |
| 3207 | assert!(message.contains("changed since it was read"), "{message}"); |
| 3208 | assert!(message.contains("nothing was written"), "{message}"); |
| 3209 | assert_eq!( |
| 3210 | fs::read_to_string(tmp.path().join("test.txt")).expect("read"), |
| 3211 | body, |
| 3212 | "a rejected patch must not modify the file" |
| 3213 | ); |
| 3214 | } |
| 3215 | |
| 3216 | #[tokio::test] |
| 3217 | async fn patch_without_expected_hash_is_unchanged() { |
| 3218 | let tmp = tempdir().expect("tempdir"); |
| 3219 | let ctx = ToolContext::new(tmp.path().to_path_buf()); |
| 3220 | fs::write(tmp.path().join("test.txt"), "line1\nline2\nline3\n").expect("write"); |
| 3221 | |
| 3222 | ApplyPatchTool |
| 3223 | .execute(json!({ "path": "test.txt", "patch": GUARD_PATCH }), &ctx) |
| 3224 | .await |
| 3225 | .expect("absent expected_hash keeps the pre-#3979 behavior"); |
| 3226 | |
| 3227 | let updated = fs::read_to_string(tmp.path().join("test.txt")).expect("read"); |
| 3228 | assert!(updated.contains("modified"), "{updated}"); |
| 3229 | } |
| 3230 | |
| 3231 | #[tokio::test] |
| 3232 | async fn multi_file_patch_guards_the_first_file_and_writes_nothing_on_mismatch() { |
| 3233 | let tmp = tempdir().expect("tempdir"); |
| 3234 | let ctx = ToolContext::new(tmp.path().to_path_buf()); |
| 3235 | fs::write(tmp.path().join("one.txt"), "one\n").expect("write"); |
| 3236 | fs::write(tmp.path().join("two.txt"), "two\n").expect("write"); |
| 3237 | |
| 3238 | let patch = "diff --git a/one.txt b/one.txt |
| 3239 | --- a/one.txt |
| 3240 | +++ b/one.txt |
| 3241 | @@ -1 +1 @@ |
| 3242 | -one |
| 3243 | +ONE |
| 3244 | diff --git a/two.txt b/two.txt |
| 3245 | --- a/two.txt |
| 3246 | +++ b/two.txt |
| 3247 | @@ -1 +1 @@ |
| 3248 | -two |
| 3249 | +TWO |
| 3250 | "; |
| 3251 | |
| 3252 | let err = ApplyPatchTool |
| 3253 | .execute( |
| 3254 | json!({ |
| 3255 | "patch": patch, |
| 3256 | "expected_hash": content_hash(b"stale\n"), |
| 3257 | }), |
| 3258 | &ctx, |
| 3259 | ) |
| 3260 | .await |
| 3261 | .expect_err("stale hash on the first file must reject the whole patch"); |
| 3262 | assert!(err.to_string().contains("one.txt"), "{err}"); |
| 3263 | |
| 3264 | // Transactional: the unguarded second file must not have been |
| 3265 | // written either. |
| 3266 | assert_eq!( |
| 3267 | fs::read_to_string(tmp.path().join("one.txt")).unwrap(), |
| 3268 | "one\n" |
| 3269 | ); |
| 3270 | assert_eq!( |
| 3271 | fs::read_to_string(tmp.path().join("two.txt")).unwrap(), |
| 3272 | "two\n" |
| 3273 | ); |
| 3274 | } |
| 3275 | |
| 3276 | #[tokio::test] |
| 3277 | async fn patch_expected_hash_on_a_missing_target_fails_closed() { |
| 3278 | let tmp = tempdir().expect("tempdir"); |
| 3279 | let ctx = ToolContext::new(tmp.path().to_path_buf()); |
| 3280 | |
| 3281 | let err = ApplyPatchTool |
| 3282 | .execute( |
| 3283 | json!({ |
| 3284 | "path": "absent.txt", |
| 3285 | "patch": GUARD_PATCH, |
| 3286 | "create_if_missing": true, |
| 3287 | "expected_hash": content_hash(b"anything"), |
| 3288 | }), |
| 3289 | &ctx, |
| 3290 | ) |
| 3291 | .await |
| 3292 | .expect_err("guarded patch of a missing file must fail closed"); |
| 3293 | |
| 3294 | assert!(err.to_string().contains("does not exist"), "{err}"); |
| 3295 | assert!(!tmp.path().join("absent.txt").exists()); |
| 3296 | } |
| 3297 | |
| 3298 | /// B7: two sections for the same file both land; the second applies on |
| 3299 | /// top of the first instead of re-reading the disk and dropping it. |
| 3300 | #[tokio::test] |
| 3301 | async fn two_sections_for_one_file_apply_both() { |
| 3302 | let tmp = tempdir().expect("tempdir"); |
| 3303 | let ctx = ToolContext::new(tmp.path().to_path_buf()); |
| 3304 | fs::write(tmp.path().join("f.txt"), "a1\na2\na3\nmid\nb1\nb2\nb3\n").expect("write"); |
| 3305 | let patch = r"--- a/f.txt |
| 3306 | +++ b/f.txt |
| 3307 | @@ -1,3 +1,3 @@ |
| 3308 | a1 |
| 3309 | -a2 |
| 3310 | +A2 |
| 3311 | a3 |
| 3312 | --- a/f.txt |
| 3313 | +++ b/f.txt |
| 3314 | @@ -5,3 +5,3 @@ |
| 3315 | b1 |
| 3316 | -b2 |
| 3317 | +B2 |
| 3318 | b3 |
| 3319 | "; |
| 3320 | let result = ApplyPatchTool |
| 3321 | .execute(json!({"patch": patch}), &ctx) |
| 3322 | .await |
| 3323 | .expect("execute"); |
| 3324 | assert!(result.success, "{}", result.content); |
| 3325 | assert_eq!( |
| 3326 | fs::read_to_string(tmp.path().join("f.txt")).expect("read"), |
| 3327 | "a1\nA2\na3\nmid\nb1\nB2\nb3\n" |
| 3328 | ); |
| 3329 | let summary = parse_patch_result(result); |
| 3330 | assert_eq!(summary.files_total, 1, "one file, two sections"); |
| 3331 | assert_eq!(summary.files_applied, 1); |
| 3332 | assert_eq!(summary.hunks_applied, 2); |
| 3333 | } |
| 3334 | |
| 3335 | #[tokio::test] |
| 3336 | async fn creation_patch_on_an_existing_file_is_refused() { |
| 3337 | let tmp = tempdir().expect("tempdir"); |
| 3338 | let ctx = ToolContext::new(tmp.path().to_path_buf()); |
| 3339 | fs::write(tmp.path().join("foo.rs"), "fn keep() {}\n").expect("write"); |
| 3340 | let patch = "--- /dev/null\n+++ b/foo.rs\n@@ -0,0 +1,2 @@\n+fn a() {}\n+fn b() {}\n"; |
| 3341 | |
| 3342 | let err = ApplyPatchTool |
| 3343 | .execute(json!({"patch": patch}), &ctx) |
| 3344 | .await |
| 3345 | .expect_err("a /dev/null creation must not land on an existing file"); |
| 3346 | |
| 3347 | assert!(err.to_string().contains("already exists"), "{err}"); |
| 3348 | assert_eq!( |
| 3349 | fs::read_to_string(tmp.path().join("foo.rs")).expect("read"), |
| 3350 | "fn keep() {}\n" |
| 3351 | ); |
| 3352 | } |
| 3353 | |
| 3354 | #[tokio::test] |
| 3355 | async fn pure_insertion_hunk_lands_after_the_named_line() { |
| 3356 | let tmp = tempdir().expect("tempdir"); |
| 3357 | let ctx = ToolContext::new(tmp.path().to_path_buf()); |
| 3358 | fs::write(tmp.path().join("f.txt"), "l1\nl2\nl3\nl4\n").expect("write"); |
| 3359 | // `git diff -U0` form: insert after line 3. |
| 3360 | let patch = "@@ -3,0 +4,1 @@\n+inserted\n"; |
| 3361 | |
| 3362 | ApplyPatchTool |
| 3363 | .execute(json!({"path": "f.txt", "patch": patch}), &ctx) |
| 3364 | .await |
| 3365 | .expect("execute"); |
| 3366 | |
| 3367 | assert_eq!( |
| 3368 | fs::read_to_string(tmp.path().join("f.txt")).expect("read"), |
| 3369 | "l1\nl2\nl3\ninserted\nl4\n" |
| 3370 | ); |
| 3371 | } |
| 3372 | |
| 3373 | #[tokio::test] |
| 3374 | async fn delete_section_covering_part_of_a_file_is_refused() { |
| 3375 | let tmp = tempdir().expect("tempdir"); |
| 3376 | let ctx = ToolContext::new(tmp.path().to_path_buf()); |
| 3377 | let original = "1\n2\n3\n4\n5\n"; |
| 3378 | fs::write(tmp.path().join("big.txt"), original).expect("write"); |
| 3379 | let patch = "--- a/big.txt\n+++ /dev/null\n@@ -1,3 +0,0 @@\n-1\n-2\n-3\n"; |
| 3380 | |
| 3381 | let err = ApplyPatchTool |
| 3382 | .execute(json!({"patch": patch}), &ctx) |
| 3383 | .await |
| 3384 | .expect_err("a partial delete must not remove the file"); |
| 3385 | |
| 3386 | assert!(err.to_string().contains("2 line(s)"), "{err}"); |
| 3387 | assert_eq!( |
| 3388 | fs::read_to_string(tmp.path().join("big.txt")).expect("still there"), |
| 3389 | original |
| 3390 | ); |
| 3391 | } |
| 3392 | |
| 3393 | #[tokio::test] |
| 3394 | async fn creation_patch_with_a_path_override_on_an_existing_file_is_refused() { |
| 3395 | let tmp = tempdir().expect("tempdir"); |
| 3396 | let ctx = ToolContext::new(tmp.path().to_path_buf()); |
| 3397 | fs::write(tmp.path().join("foo.rs"), "fn keep() {}\n").expect("write"); |
| 3398 | let patch = "--- /dev/null\n+++ b/foo.rs\n@@ -0,0 +1,2 @@\n+fn a() {}\n+fn b() {}\n"; |
| 3399 | |
| 3400 | let err = ApplyPatchTool |
| 3401 | .execute(json!({"path": "foo.rs", "patch": patch}), &ctx) |
| 3402 | .await |
| 3403 | .expect_err("`path` must not turn a /dev/null creation into a prepend"); |
| 3404 | |
| 3405 | assert!(err.to_string().contains("already exists"), "{err}"); |
| 3406 | assert_eq!( |
| 3407 | fs::read_to_string(tmp.path().join("foo.rs")).expect("read"), |
| 3408 | "fn keep() {}\n" |
| 3409 | ); |
| 3410 | } |
| 3411 | |
| 3412 | #[tokio::test] |
| 3413 | async fn creation_section_after_an_edit_of_the_same_file_is_refused() { |
| 3414 | let tmp = tempdir().expect("tempdir"); |
| 3415 | let ctx = ToolContext::new(tmp.path().to_path_buf()); |
| 3416 | fs::write(tmp.path().join("foo.txt"), "a\nb\n").expect("write"); |
| 3417 | let patch = "--- a/foo.txt\n+++ b/foo.txt\n@@ -1,2 +1,2 @@\n a\n-b\n+c\n\ |
| 3418 | --- /dev/null\n+++ b/foo.txt\n@@ -0,0 +1,1 @@\n+new\n"; |
| 3419 | |
| 3420 | let err = ApplyPatchTool |
| 3421 | .execute(json!({"patch": patch}), &ctx) |
| 3422 | .await |
| 3423 | .expect_err("a second, creating section must not prepend to the edited file"); |
| 3424 | |
| 3425 | assert!(err.to_string().contains("already exists"), "{err}"); |
| 3426 | assert_eq!( |
| 3427 | fs::read_to_string(tmp.path().join("foo.txt")).expect("read"), |
| 3428 | "a\nb\n" |
| 3429 | ); |
| 3430 | } |
| 3431 | |
| 3432 | #[tokio::test] |
| 3433 | async fn insertion_hunks_without_a_zero_old_count_keep_the_before_anchor() { |
| 3434 | let tmp = tempdir().expect("tempdir"); |
| 3435 | let ctx = ToolContext::new(tmp.path().to_path_buf()); |
| 3436 | for (patch, expected) in [ |
| 3437 | // Omitted old count (defaults to 1): before line 2, as before. |
| 3438 | ("@@ -2 +2,1 @@\n+x\n", "l1\nx\nl2\nl3\n"), |
| 3439 | // Hand-written "new text starts at line 1": at the top. |
| 3440 | ("@@ -1,0 +1,1 @@\n+x\n", "x\nl1\nl2\nl3\n"), |
| 3441 | // `git diff -U0` form: after line 1. |
| 3442 | ("@@ -1,0 +2,1 @@\n+x\n", "l1\nx\nl2\nl3\n"), |
| 3443 | ] { |
| 3444 | fs::write(tmp.path().join("f.txt"), "l1\nl2\nl3\n").expect("write"); |
| 3445 | ApplyPatchTool |
| 3446 | .execute(json!({"path": "f.txt", "patch": patch}), &ctx) |
| 3447 | .await |
| 3448 | .expect("execute"); |
| 3449 | assert_eq!( |
| 3450 | fs::read_to_string(tmp.path().join("f.txt")).expect("read"), |
| 3451 | expected, |
| 3452 | "{patch:?}" |
| 3453 | ); |
| 3454 | } |
| 3455 | } |
| 3456 | } |
| 3457 |