| 1 | package builtin |
| 2 | |
| 3 | import ( |
| 4 | "fmt" |
| 5 | "os" |
| 6 | "slices" |
| 7 | "strings" |
| 8 | |
| 9 | "reasonix/internal/fileutil" |
| 10 | fileenc "reasonix/internal/fileutil/encoding" |
| 11 | "reasonix/internal/tool" |
| 12 | ) |
| 13 | |
| 14 | // readFileEncoded reads a file and decodes its encoding to UTF-8. |
| 15 | // Returns the decoded content and the detected encoding kind so callers |
| 16 | // can re-encode on write to preserve the original charset. |
| 17 | func readFileEncoded(path string) (content string, enc fileenc.Kind, err error) { |
| 18 | b, err := os.ReadFile(path) |
| 19 | if err != nil { |
| 20 | return "", 0, err |
| 21 | } |
| 22 | enc, text := fileenc.DetectAndDecode(b) |
| 23 | return string(text), enc, nil |
| 24 | } |
| 25 | |
| 26 | // writeFileEncoded encodes content back to the given encoding and writes it. |
| 27 | // The write is atomic: a truncating write that fails midway (a Windows filter |
| 28 | // driver holding a transient lock, a full disk) would leave the user's source |
| 29 | // file empty or half-written. |
| 30 | func writeFileEncoded(path string, content string, enc fileenc.Kind) error { |
| 31 | data, err := fileenc.Encode(content, enc) |
| 32 | if err != nil { |
| 33 | return err |
| 34 | } |
| 35 | return fileutil.AtomicOverwriteFileStrict(path, data, 0o644) |
| 36 | } |
| 37 | |
| 38 | // matchLineEndings adapts an edit's old/new text to a CRLF file when the literal |
| 39 | // old_string isn't present but its CRLF form is. read_file strips '\r' (bufio |
| 40 | // ScanLines), so a model's multi-line old_string arrives LF-only while a |
| 41 | // Windows/CJK source stores '\r\n'; rewriting search and replacement to the |
| 42 | // file's ending fixes the match without rewriting the file's other line endings. |
| 43 | func matchLineEndings(content, old, new string) (string, string) { |
| 44 | if strings.Contains(content, old) || !strings.Contains(content, "\r\n") { |
| 45 | return old, new |
| 46 | } |
| 47 | if strings.Contains(content, toCRLF(old)) { |
| 48 | return toCRLF(old), toCRLF(new) |
| 49 | } |
| 50 | return old, new |
| 51 | } |
| 52 | |
| 53 | func toCRLF(s string) string { |
| 54 | return strings.ReplaceAll(strings.ReplaceAll(s, "\r\n", "\n"), "\n", "\r\n") |
| 55 | } |
| 56 | |
| 57 | func matchReplacementLineEndings(content, replacement string) string { |
| 58 | if strings.Contains(content, "\r\n") { |
| 59 | return toCRLF(replacement) |
| 60 | } |
| 61 | return replacement |
| 62 | } |
| 63 | |
| 64 | type editApplyResult struct { |
| 65 | updated string |
| 66 | applied int |
| 67 | matches int |
| 68 | fuzzy bool |
| 69 | receipt editReplacementReceipt |
| 70 | } |
| 71 | |
| 72 | type editRange struct { |
| 73 | start int |
| 74 | end int |
| 75 | } |
| 76 | |
| 77 | // editReplacementReceipt records only the span the tool actually matched and |
| 78 | // the span it wrote in its place. It deliberately excludes surrounding file |
| 79 | // content so a successful edit can ground the next model turn without widening |
| 80 | // provider-visible workspace data. |
| 81 | type editReplacementReceipt struct { |
| 82 | matched string |
| 83 | replacement string |
| 84 | occurrences int |
| 85 | fuzzy bool |
| 86 | } |
| 87 | |
| 88 | // applyOldStringEdit is the shared edit_file/multi_edit/Preview contract. It |
| 89 | // preserves the exact-match rule first, then falls back to a narrow fuzzy match |
| 90 | // for the mismatches read_file commonly introduces or hides: trailing |
| 91 | // whitespace, tab-vs-spaces indentation, and copied read_file line prefixes. |
| 92 | // Non-replace_all edits still require exactly one match, including fuzzy |
| 93 | // matches. |
| 94 | func applyOldStringEdit(content, oldString, newString string, replaceAll bool) editApplyResult { |
| 95 | old, newStr := matchLineEndings(content, oldString, newString) |
| 96 | if replaceAll { |
| 97 | if count := strings.Count(content, old); count > 0 { |
| 98 | return editApplyResult{ |
| 99 | updated: strings.ReplaceAll(content, old, newStr), |
| 100 | applied: count, |
| 101 | matches: count, |
| 102 | receipt: editReplacementReceipt{ |
| 103 | matched: old, |
| 104 | replacement: newStr, |
| 105 | occurrences: count, |
| 106 | }, |
| 107 | } |
| 108 | } |
| 109 | ranges := fuzzyEditRanges(content, old) |
| 110 | if len(ranges) == 0 { |
| 111 | return editApplyResult{updated: content} |
| 112 | } |
| 113 | replacement := matchReplacementLineEndings(content, newStr) |
| 114 | return editApplyResult{ |
| 115 | updated: replaceEditRanges(content, ranges, replacement), |
| 116 | applied: len(ranges), |
| 117 | matches: len(ranges), |
| 118 | fuzzy: true, |
| 119 | receipt: editReplacementReceipt{ |
| 120 | matched: matchedRangeSample(content, old, ranges), |
| 121 | replacement: replacement, |
| 122 | occurrences: len(ranges), |
| 123 | fuzzy: true, |
| 124 | }, |
| 125 | } |
| 126 | } |
| 127 | |
| 128 | switch count := strings.Count(content, old); count { |
| 129 | case 0: |
| 130 | ranges := fuzzyEditRanges(content, old) |
| 131 | if len(ranges) != 1 { |
| 132 | return editApplyResult{updated: content, matches: len(ranges)} |
| 133 | } |
| 134 | return editApplyResult{ |
| 135 | updated: replaceEditRanges(content, ranges, matchReplacementLineEndings(content, newStr)), |
| 136 | applied: 1, |
| 137 | matches: 1, |
| 138 | fuzzy: true, |
| 139 | receipt: editReplacementReceipt{ |
| 140 | matched: matchedRangeSample(content, old, ranges), |
| 141 | replacement: matchReplacementLineEndings(content, newStr), |
| 142 | occurrences: 1, |
| 143 | fuzzy: true, |
| 144 | }, |
| 145 | } |
| 146 | case 1: |
| 147 | return editApplyResult{ |
| 148 | updated: strings.Replace(content, old, newStr, 1), |
| 149 | applied: 1, |
| 150 | matches: 1, |
| 151 | receipt: editReplacementReceipt{ |
| 152 | matched: old, |
| 153 | replacement: newStr, |
| 154 | occurrences: 1, |
| 155 | }, |
| 156 | } |
| 157 | default: |
| 158 | return editApplyResult{updated: content, matches: count} |
| 159 | } |
| 160 | } |
| 161 | |
| 162 | func matchedRangeSample(content, fallback string, ranges []editRange) string { |
| 163 | if len(ranges) == 0 { |
| 164 | return fallback |
| 165 | } |
| 166 | r := ranges[0] |
| 167 | if r.start < 0 || r.end < r.start || r.end > len(content) { |
| 168 | return fallback |
| 169 | } |
| 170 | actual := content[r.start:r.end] |
| 171 | sample := clipPostWriteSpan(actual, maxCapturedReceiptSpanBytes) |
| 172 | if len(sample) == len(actual) { |
| 173 | // Do not let a short substring keep an otherwise-dead large intermediate |
| 174 | // multi_edit buffer alive until all later steps finish. |
| 175 | return strings.Clone(sample) |
| 176 | } |
| 177 | return sample |
| 178 | } |
| 179 | |
| 180 | func oldStringNotFoundError(path, oldString, content string) (err error) { |
| 181 | defer func() { |
| 182 | err = &tool.OperationError{Diagnostic: tool.OperationDiagnostic{Code: tool.WriteEvidenceStale, Path: path, Recovery: "re-read the target range, then retry with its current text"}, Cause: err} |
| 183 | }() |
| 184 | hint := oldStringNotFoundHint(oldString, content) |
| 185 | if line, text, ok := nearestContentLine(oldString, content); ok { |
| 186 | return fmt.Errorf("old_string not found in %s (nearest line %d: %q).%s", path, line, text, hint) |
| 187 | } |
| 188 | return fmt.Errorf("old_string not found in %s.%s", path, hint) |
| 189 | } |
| 190 | |
| 191 | func oldStringNotFoundHint(oldString, content string) string { |
| 192 | base := " Re-read the current file before retrying; if several related edits target the same area, combine the final replacements in one multi_edit call." |
| 193 | if !strings.Contains(content, "\r\n") { |
| 194 | return base |
| 195 | } |
| 196 | normalizedContent := strings.ReplaceAll(content, "\r\n", "\n") |
| 197 | normalizedOld := strings.ReplaceAll(oldString, "\r\n", "\n") |
| 198 | if strings.Contains(normalizedContent, normalizedOld) { |
| 199 | return " The target file uses CRLF line endings; edit_file/multi_edit normally normalize LF-only old_string for CRLF files, so this is likely stale context. Re-read the current file before retrying." |
| 200 | } |
| 201 | return " The target file uses CRLF line endings, but edit_file/multi_edit already tolerate LF-only old_string for CRLF files; check for stale, incomplete, or non-unique context before retrying." |
| 202 | } |
| 203 | |
| 204 | func oldStringNotUniqueError(path, oldString, content string, matches int, replaceAllHint bool) (err error) { |
| 205 | defer func() { |
| 206 | err = &tool.OperationError{Diagnostic: tool.OperationDiagnostic{Code: tool.WriteTargetAmbiguous, Path: path, Recovery: "read surrounding lines and use a unique anchor"}, Cause: err} |
| 207 | }() |
| 208 | lineHint := oldStringMatchLineSummary(oldString, content, 5) |
| 209 | if replaceAllHint { |
| 210 | return fmt.Errorf("old_string is not unique in %s (%d matches)%s; add nearby unique code, not just repeated separator lines, or set replace_all if every match should change", path, matches, lineHint) |
| 211 | } |
| 212 | return fmt.Errorf("old_string is not unique in %s (%d matches)%s; add nearby unique code, not just repeated separator lines", path, matches, lineHint) |
| 213 | } |
| 214 | |
| 215 | type lineSegment struct { |
| 216 | raw string |
| 217 | start int |
| 218 | end int |
| 219 | } |
| 220 | |
| 221 | type fuzzyMode struct { |
| 222 | stripOldReadPrefixes bool |
| 223 | trimTrailing bool |
| 224 | expandTabs bool |
| 225 | trimLeading bool |
| 226 | } |
| 227 | |
| 228 | func fuzzyEditRanges(content, old string) []editRange { |
| 229 | if old == "" || content == "" { |
| 230 | return nil |
| 231 | } |
| 232 | contentLines := splitLineSegments(content) |
| 233 | oldLines := splitLineSegments(old) |
| 234 | if len(oldLines) == 0 || len(oldLines) > len(contentLines) { |
| 235 | return nil |
| 236 | } |
| 237 | |
| 238 | oldHasReadPrefixes := allLinesHaveReadFilePrefix(oldLines) |
| 239 | modes := []fuzzyMode{ |
| 240 | {trimTrailing: true}, |
| 241 | {trimTrailing: true, expandTabs: true}, |
| 242 | } |
| 243 | if oldHasReadPrefixes { |
| 244 | modes = append(modes, |
| 245 | fuzzyMode{stripOldReadPrefixes: true, trimTrailing: true}, |
| 246 | fuzzyMode{stripOldReadPrefixes: true, trimTrailing: true, expandTabs: true}, |
| 247 | ) |
| 248 | } |
| 249 | |
| 250 | for _, mode := range modes { |
| 251 | normOld := make([]string, len(oldLines)) |
| 252 | for i, line := range oldLines { |
| 253 | normOld[i] = normalizeFuzzyLine(line.raw, lineHasNewline(line.raw), mode, mode.stripOldReadPrefixes) |
| 254 | } |
| 255 | var ranges []editRange |
| 256 | for i := 0; i <= len(contentLines)-len(oldLines); { |
| 257 | if fuzzyWindowMatches(contentLines[i:i+len(oldLines)], oldLines, normOld, mode) { |
| 258 | ranges = append(ranges, editRange{ |
| 259 | start: contentLines[i].start, |
| 260 | end: fuzzyWindowEnd(contentLines[i+len(oldLines)-1], oldLines[len(oldLines)-1]), |
| 261 | }) |
| 262 | i += len(oldLines) |
| 263 | continue |
| 264 | } |
| 265 | i++ |
| 266 | } |
| 267 | if len(ranges) > 0 { |
| 268 | return ranges |
| 269 | } |
| 270 | } |
| 271 | return nil |
| 272 | } |
| 273 | |
| 274 | func fuzzyWindowMatches(contentWindow, oldLines []lineSegment, normOld []string, mode fuzzyMode) bool { |
| 275 | for i, contentLine := range contentWindow { |
| 276 | oldHasNewline := lineHasNewline(oldLines[i].raw) |
| 277 | if oldHasNewline && !lineHasNewline(contentLine.raw) { |
| 278 | return false |
| 279 | } |
| 280 | got := normalizeFuzzyLine(contentLine.raw, oldHasNewline, mode, false) |
| 281 | if got != normOld[i] { |
| 282 | return false |
| 283 | } |
| 284 | } |
| 285 | return true |
| 286 | } |
| 287 | |
| 288 | func splitLineSegments(s string) []lineSegment { |
| 289 | if s == "" { |
| 290 | return nil |
| 291 | } |
| 292 | var lines []lineSegment |
| 293 | start := 0 |
| 294 | for i, r := range s { |
| 295 | if r == '\n' { |
| 296 | end := i + 1 |
| 297 | lines = append(lines, lineSegment{raw: s[start:end], start: start, end: end}) |
| 298 | start = end |
| 299 | } |
| 300 | } |
| 301 | if start < len(s) { |
| 302 | lines = append(lines, lineSegment{raw: s[start:], start: start, end: len(s)}) |
| 303 | } |
| 304 | return lines |
| 305 | } |
| 306 | |
| 307 | func lineHasNewline(line string) bool { |
| 308 | return strings.HasSuffix(line, "\n") |
| 309 | } |
| 310 | |
| 311 | func fuzzyWindowEnd(contentLast, oldLast lineSegment) int { |
| 312 | if lineHasNewline(oldLast.raw) || !lineHasNewline(contentLast.raw) { |
| 313 | return contentLast.end |
| 314 | } |
| 315 | end := contentLast.end - 1 |
| 316 | if end > contentLast.start && contentLast.raw[len(contentLast.raw)-2] == '\r' { |
| 317 | end-- |
| 318 | } |
| 319 | return end |
| 320 | } |
| 321 | |
| 322 | func normalizeFuzzyLine(line string, includeNewline bool, mode fuzzyMode, stripReadPrefix bool) string { |
| 323 | body := strings.TrimSuffix(line, "\n") |
| 324 | if stripReadPrefix { |
| 325 | body, _ = stripReadFileLinePrefix(body) |
| 326 | } |
| 327 | if mode.trimTrailing { |
| 328 | body = strings.TrimRight(body, " \t\r") |
| 329 | } |
| 330 | if mode.expandTabs { |
| 331 | body = strings.ReplaceAll(body, "\t", " ") |
| 332 | } |
| 333 | if mode.trimLeading { |
| 334 | body = strings.TrimLeft(body, " \t") |
| 335 | } |
| 336 | if includeNewline { |
| 337 | return body + "\n" |
| 338 | } |
| 339 | return body |
| 340 | } |
| 341 | |
| 342 | func allLinesHaveReadFilePrefix(lines []lineSegment) bool { |
| 343 | if len(lines) == 0 { |
| 344 | return false |
| 345 | } |
| 346 | for _, line := range lines { |
| 347 | body := strings.TrimSuffix(line.raw, "\n") |
| 348 | if _, ok := stripReadFileLinePrefix(body); !ok { |
| 349 | return false |
| 350 | } |
| 351 | } |
| 352 | return true |
| 353 | } |
| 354 | |
| 355 | func stripReadFileLinePrefix(line string) (string, bool) { |
| 356 | i := 0 |
| 357 | for i < len(line) && (line[i] == ' ' || line[i] == '\t') { |
| 358 | i++ |
| 359 | } |
| 360 | j := i |
| 361 | for j < len(line) && line[j] >= '0' && line[j] <= '9' { |
| 362 | j++ |
| 363 | } |
| 364 | if j == i || !strings.HasPrefix(line[j:], "\u2192") { |
| 365 | return line, false |
| 366 | } |
| 367 | return line[j+len("\u2192"):], true |
| 368 | } |
| 369 | |
| 370 | func replaceEditRanges(content string, ranges []editRange, replacement string) string { |
| 371 | updated := content |
| 372 | for _, v := range slices.Backward(ranges) { |
| 373 | r := v |
| 374 | updated = updated[:r.start] + replacement + updated[r.end:] |
| 375 | } |
| 376 | return updated |
| 377 | } |
| 378 | |
| 379 | func nearestContentLine(oldString, content string) (int, string, bool) { |
| 380 | oldLines := splitLineSegments(oldString) |
| 381 | if len(oldLines) == 0 { |
| 382 | return 0, "", false |
| 383 | } |
| 384 | target := strings.TrimSpace(normalizeFuzzyLine(oldLines[0].raw, false, fuzzyMode{trimTrailing: true, expandTabs: true}, true)) |
| 385 | if target == "" { |
| 386 | return 0, "", false |
| 387 | } |
| 388 | bestLine := 0 |
| 389 | bestScore := 0 |
| 390 | bestText := "" |
| 391 | for i, line := range splitLineSegments(content) { |
| 392 | text := strings.TrimSuffix(line.raw, "\n") |
| 393 | score := commonPrefixLen(strings.TrimSpace(strings.ReplaceAll(text, "\t", " ")), target) |
| 394 | if score > bestScore { |
| 395 | bestLine = i + 1 |
| 396 | bestScore = score |
| 397 | bestText = text |
| 398 | } |
| 399 | } |
| 400 | if bestScore < 3 { |
| 401 | return 0, "", false |
| 402 | } |
| 403 | return bestLine, bestText, true |
| 404 | } |
| 405 | |
| 406 | func oldStringMatchLineSummary(oldString, content string, limit int) string { |
| 407 | if limit <= 0 { |
| 408 | return "" |
| 409 | } |
| 410 | target := firstNonEmptyLine(oldString) |
| 411 | if target == "" { |
| 412 | return "" |
| 413 | } |
| 414 | var matches []int |
| 415 | for i, line := range splitLineSegments(content) { |
| 416 | text := strings.TrimSuffix(line.raw, "\n") |
| 417 | text = strings.TrimSuffix(text, "\r") |
| 418 | if strings.Contains(text, target) { |
| 419 | matches = append(matches, i+1) |
| 420 | } |
| 421 | } |
| 422 | if len(matches) == 0 { |
| 423 | return "" |
| 424 | } |
| 425 | var b strings.Builder |
| 426 | b.WriteString("; matching lines include ") |
| 427 | for i, line := range matches { |
| 428 | if i >= limit { |
| 429 | b.WriteString(", ...") |
| 430 | break |
| 431 | } |
| 432 | if i > 0 { |
| 433 | b.WriteString(", ") |
| 434 | } |
| 435 | fmt.Fprint(&b, line) |
| 436 | } |
| 437 | return b.String() |
| 438 | } |
| 439 | |
| 440 | func firstNonEmptyLine(s string) string { |
| 441 | for _, line := range splitLineSegments(s) { |
| 442 | text := strings.TrimSpace(strings.TrimSuffix(line.raw, "\n")) |
| 443 | text = strings.TrimSuffix(text, "\r") |
| 444 | if text != "" { |
| 445 | return text |
| 446 | } |
| 447 | } |
| 448 | return "" |
| 449 | } |
| 450 | |
| 451 | func commonPrefixLen(a, b string) int { |
| 452 | n := min(len(b), len(a)) |
| 453 | for i := range n { |
| 454 | if a[i] != b[i] { |
| 455 | return i |
| 456 | } |
| 457 | } |
| 458 | return n |
| 459 | } |
| 460 |