1// Copyright 2013 The Go Authors. All rights reserved. 2// Use of this source code is governed by a BSD-style 3// license that can be found in the LICENSE file. 4 5package language 6 7import ( 8 "bytes" 9 "errors" 10 "fmt" 11 "sort" 12 13 "golang.org/x/text/internal/tag" 14) 15 16// isAlpha returns true if the byte is not a digit. 17// b must be an ASCII letter or digit. 18func isAlpha(b byte) bool { 19 return b > '9' 20} 21 22// isAlphaNum returns true if the string contains only ASCII letters or digits. 23func isAlphaNum(s []byte) bool { 24 for _, c := range s { 25 if !('a' <= c && c <= 'z' || 'A' <= c && c <= 'Z' || '0' <= c && c <= '9') { 26 return false 27 } 28 } 29 return true 30} 31 32// ErrSyntax is returned by any of the parsing functions when the 33// input is not well-formed, according to BCP 47. 34// TODO: return the position at which the syntax error occurred? 35var ErrSyntax = errors.New("language: tag is not well-formed") 36 37// ErrDuplicateKey is returned when a tag contains the same key twice with 38// different values in the -u section. 39var ErrDuplicateKey = errors.New("language: different values for same key in -u extension") 40 41// ValueError is returned by any of the parsing functions when the 42// input is well-formed but the respective subtag is not recognized 43// as a valid value. 44type ValueError struct { 45 v [8]byte 46} 47 48// NewValueError creates a new ValueError. 49func NewValueError(tag []byte) ValueError { 50 var e ValueError 51 copy(e.v[:], tag) 52 return e 53} 54 55func (e ValueError) tag() []byte { 56 n := bytes.IndexByte(e.v[:], 0) 57 if n == -1 { 58 n = 8 59 } 60 return e.v[:n] 61} 62 63// Error implements the error interface. 64func (e ValueError) Error() string { 65 return fmt.Sprintf("language: subtag %q is well-formed but unknown", e.tag()) 66} 67 68// Subtag returns the subtag for which the error occurred. 69func (e ValueError) Subtag() string { 70 return string(e.tag()) 71} 72 73// scanner is used to scan BCP 47 tokens, which are separated by _ or -. 74type scanner struct { 75 b []byte 76 bytes [max99thPercentileSize]byte 77 token []byte 78 start int // start position of the current token 79 end int // end position of the current token 80 next int // next point for scan 81 err error 82 done bool 83} 84 85func makeScannerString(s string) scanner { 86 scan := scanner{} 87 if len(s) <= len(scan.bytes) { 88 scan.b = scan.bytes[:copy(scan.bytes[:], s)] 89 } else { 90 scan.b = []byte(s) 91 } 92 scan.init() 93 return scan 94} 95 96// makeScanner returns a scanner using b as the input buffer. 97// b is not copied and may be modified by the scanner routines. 98func makeScanner(b []byte) scanner { 99 scan := scanner{b: b} 100 scan.init() 101 return scan 102} 103 104func (s *scanner) init() { 105 for i, c := range s.b { 106 if c == '_' { 107 s.b[i] = '-' 108 } 109 } 110 s.scan() 111} 112 113// restToLower converts the string between start and end to lower case. 114func (s *scanner) toLower(start, end int) { 115 for i := start; i < end; i++ { 116 c := s.b[i] 117 if 'A' <= c && c <= 'Z' { 118 s.b[i] += 'a' - 'A' 119 } 120 } 121} 122 123func (s *scanner) setError(e error) { 124 if s.err == nil || (e == ErrSyntax && s.err != ErrSyntax) { 125 s.err = e 126 } 127} 128 129// resizeRange shrinks or grows the array at position oldStart such that 130// a new string of size newSize can fit between oldStart and oldEnd. 131// Sets the scan point to after the resized range. 132func (s *scanner) resizeRange(oldStart, oldEnd, newSize int) { 133 s.start = oldStart 134 if end := oldStart + newSize; end != oldEnd { 135 diff := end - oldEnd 136 var b []byte 137 if n := len(s.b) + diff; n > cap(s.b) { 138 b = make([]byte, n) 139 copy(b, s.b[:oldStart]) 140 } else { 141 b = s.b[:n] 142 } 143 copy(b[end:], s.b[oldEnd:]) 144 s.b = b 145 s.next = end + (s.next - s.end) 146 s.end = end 147 } 148} 149 150// replace replaces the current token with repl. 151func (s *scanner) replace(repl string) { 152 s.resizeRange(s.start, s.end, len(repl)) 153 copy(s.b[s.start:], repl) 154} 155 156// gobble removes the current token from the input. 157// Caller must call scan after calling gobble. 158func (s *scanner) gobble(e error) { 159 s.setError(e) 160 if s.start == 0 { 161 s.b = s.b[:+copy(s.b, s.b[s.next:])] 162 s.end = 0 163 } else { 164 s.b = s.b[:s.start-1+copy(s.b[s.start-1:], s.b[s.end:])] 165 s.end = s.start - 1 166 } 167 s.next = s.start 168} 169 170// deleteRange removes the given range from s.b before the current token. 171func (s *scanner) deleteRange(start, end int) { 172 s.b = s.b[:start+copy(s.b[start:], s.b[end:])] 173 diff := end - start 174 s.next -= diff 175 s.start -= diff 176 s.end -= diff 177} 178 179// scan parses the next token of a BCP 47 string. Tokens that are larger 180// than 8 characters or include non-alphanumeric characters result in an error 181// and are gobbled and removed from the output. 182// It returns the end position of the last token consumed. 183func (s *scanner) scan() (end int) { 184 end = s.end 185 s.token = nil 186 for s.start = s.next; s.next < len(s.b); { 187 i := bytes.IndexByte(s.b[s.next:], '-') 188 if i == -1 { 189 s.end = len(s.b) 190 s.next = len(s.b) 191 i = s.end - s.start 192 } else { 193 s.end = s.next + i 194 s.next = s.end + 1 195 } 196 token := s.b[s.start:s.end] 197 if i < 1 || i > 8 || !isAlphaNum(token) { 198 s.gobble(ErrSyntax) 199 continue 200 } 201 s.token = token 202 return end 203 } 204 if n := len(s.b); n > 0 && s.b[n-1] == '-' { 205 s.setError(ErrSyntax) 206 s.b = s.b[:len(s.b)-1] 207 } 208 s.done = true 209 return end 210} 211 212// acceptMinSize parses multiple tokens of the given size or greater. 213// It returns the end position of the last token consumed. 214func (s *scanner) acceptMinSize(min int) (end int) { 215 end = s.end 216 s.scan() 217 for ; len(s.token) >= min; s.scan() { 218 end = s.end 219 } 220 return end 221} 222 223// Parse parses the given BCP 47 string and returns a valid Tag. If parsing 224// failed it returns an error and any part of the tag that could be parsed. 225// If parsing succeeded but an unknown value was found, it returns 226// ValueError. The Tag returned in this case is just stripped of the unknown 227// value. All other values are preserved. It accepts tags in the BCP 47 format 228// and extensions to this standard defined in 229// https://www.unicode.org/reports/tr35/#Unicode_Language_and_Locale_Identifiers. 230func Parse(s string) (t Tag, err error) { 231 // TODO: consider supporting old-style locale key-value pairs. 232 if s == "" { 233 return Und, ErrSyntax 234 } 235 if len(s) <= maxAltTaglen { 236 b := [maxAltTaglen]byte{} 237 for i, c := range s { 238 // Generating invalid UTF-8 is okay as it won't match. 239 if 'A' <= c && c <= 'Z' { 240 c += 'a' - 'A' 241 } else if c == '_' { 242 c = '-' 243 } 244 b[i] = byte(c) 245 } 246 if t, ok := grandfathered(b); ok { 247 return t, nil 248 } 249 } 250 scan := makeScannerString(s) 251 return parse(&scan, s) 252} 253 254func parse(scan *scanner, s string) (t Tag, err error) { 255 t = Und 256 var end int 257 if n := len(scan.token); n <= 1 { 258 scan.toLower(0, len(scan.b)) 259 if n == 0 || scan.token[0] != 'x' { 260 return t, ErrSyntax 261 } 262 end = parseExtensions(scan) 263 } else if n >= 4 { 264 return Und, ErrSyntax 265 } else { // the usual case 266 t, end = parseTag(scan) 267 if n := len(scan.token); n == 1 { 268 t.pExt = uint16(end) 269 end = parseExtensions(scan) 270 } else if end < len(scan.b) { 271 scan.setError(ErrSyntax) 272 scan.b = scan.b[:end] 273 } 274 } 275 if int(t.pVariant) < len(scan.b) { 276 if end < len(s) { 277 s = s[:end] 278 } 279 if len(s) > 0 && tag.Compare(s, scan.b) == 0 { 280 t.str = s 281 } else { 282 t.str = string(scan.b) 283 } 284 } else { 285 t.pVariant, t.pExt = 0, 0 286 } 287 return t, scan.err 288} 289 290// parseTag parses language, script, region and variants. 291// It returns a Tag and the end position in the input that was parsed. 292func parseTag(scan *scanner) (t Tag, end int) { 293 var e error 294 // TODO: set an error if an unknown lang, script or region is encountered. 295 t.LangID, e = getLangID(scan.token) 296 scan.setError(e) 297 scan.replace(t.LangID.String()) 298 langStart := scan.start 299 end = scan.scan() 300 for len(scan.token) == 3 && isAlpha(scan.token[0]) { 301 // From http://tools.ietf.org/html/bcp47, <lang>-<extlang> tags are equivalent 302 // to a tag of the form <extlang>. 303 lang, e := getLangID(scan.token) 304 if lang != 0 { 305 t.LangID = lang 306 copy(scan.b[langStart:], lang.String()) 307 scan.b[langStart+3] = '-' 308 scan.start = langStart + 4 309 } 310 scan.gobble(e) 311 end = scan.scan() 312 } 313 if len(scan.token) == 4 && isAlpha(scan.token[0]) { 314 t.ScriptID, e = getScriptID(script, scan.token) 315 if t.ScriptID == 0 { 316 scan.gobble(e) 317 } 318 end = scan.scan() 319 } 320 if n := len(scan.token); n >= 2 && n <= 3 { 321 t.RegionID, e = getRegionID(scan.token) 322 if t.RegionID == 0 { 323 scan.gobble(e) 324 } else { 325 scan.replace(t.RegionID.String()) 326 } 327 end = scan.scan() 328 } 329 scan.toLower(scan.start, len(scan.b)) 330 t.pVariant = byte(end) 331 end = parseVariants(scan, end, t) 332 t.pExt = uint16(end) 333 return t, end 334} 335 336var separator = []byte{'-'} 337 338// parseVariants scans tokens as long as each token is a valid variant string. 339// Duplicate variants are removed. 340func parseVariants(scan *scanner, end int, t Tag) int { 341 start := scan.start 342 varIDBuf := [4]uint8{} 343 variantBuf := [4][]byte{} 344 varID := varIDBuf[:0] 345 variant := variantBuf[:0] 346 last := -1 347 needSort := false 348 for ; len(scan.token) >= 4; scan.scan() { 349 // TODO: measure the impact of needing this conversion and redesign 350 // the data structure if there is an issue. 351 v, ok := variantIndex[string(scan.token)] 352 if !ok { 353 // unknown variant 354 // TODO: allow user-defined variants? 355 scan.gobble(NewValueError(scan.token)) 356 continue 357 } 358 varID = append(varID, v) 359 variant = append(variant, scan.token) 360 if !needSort { 361 if last < int(v) { 362 last = int(v) 363 } else { 364 needSort = true 365 // There is no legal combinations of more than 7 variants 366 // (and this is by no means a useful sequence). 367 const maxVariants = 8 368 if len(varID) > maxVariants { 369 break 370 } 371 } 372 } 373 end = scan.end 374 } 375 if needSort { 376 sort.Sort(variantsSort{varID, variant}) 377 k, l := 0, -1 378 for i, v := range varID { 379 w := int(v) 380 if l == w { 381 // Remove duplicates. 382 continue 383 } 384 varID[k] = varID[i] 385 variant[k] = variant[i] 386 k++ 387 l = w 388 } 389 if str := bytes.Join(variant[:k], separator); len(str) == 0 { 390 end = start - 1 391 } else { 392 scan.resizeRange(start, end, len(str)) 393 copy(scan.b[scan.start:], str) 394 end = scan.end 395 } 396 } 397 return end 398} 399 400type variantsSort struct { 401 i []uint8 402 v [][]byte 403} 404 405func (s variantsSort) Len() int { 406 return len(s.i) 407} 408 409func (s variantsSort) Swap(i, j int) { 410 s.i[i], s.i[j] = s.i[j], s.i[i] 411 s.v[i], s.v[j] = s.v[j], s.v[i] 412} 413 414func (s variantsSort) Less(i, j int) bool { 415 return s.i[i] < s.i[j] 416} 417 418type bytesSort struct { 419 b [][]byte 420 n int // first n bytes to compare 421} 422 423func (b bytesSort) Len() int { 424 return len(b.b) 425} 426 427func (b bytesSort) Swap(i, j int) { 428 b.b[i], b.b[j] = b.b[j], b.b[i] 429} 430 431func (b bytesSort) Less(i, j int) bool { 432 for k := 0; k < b.n; k++ { 433 if b.b[i][k] == b.b[j][k] { 434 continue 435 } 436 return b.b[i][k] < b.b[j][k] 437 } 438 return false 439} 440 441// parseExtensions parses and normalizes the extensions in the buffer. 442// It returns the last position of scan.b that is part of any extension. 443// It also trims scan.b to remove excess parts accordingly. 444func parseExtensions(scan *scanner) int { 445 start := scan.start 446 exts := [][]byte{} 447 private := []byte{} 448 end := scan.end 449 for len(scan.token) == 1 { 450 extStart := scan.start 451 ext := scan.token[0] 452 end = parseExtension(scan) 453 extension := scan.b[extStart:end] 454 if len(extension) < 3 || (ext != 'x' && len(extension) < 4) { 455 scan.setError(ErrSyntax) 456 end = extStart 457 continue 458 } else if start == extStart && (ext == 'x' || scan.start == len(scan.b)) { 459 scan.b = scan.b[:end] 460 return end 461 } else if ext == 'x' { 462 private = extension 463 break 464 } 465 exts = append(exts, extension) 466 } 467 sort.Sort(bytesSort{exts, 1}) 468 if len(private) > 0 { 469 exts = append(exts, private) 470 } 471 scan.b = scan.b[:start] 472 if len(exts) > 0 { 473 scan.b = append(scan.b, bytes.Join(exts, separator)...) 474 } else if start > 0 { 475 // Strip trailing '-'. 476 scan.b = scan.b[:start-1] 477 } 478 return end 479} 480 481// parseExtension parses a single extension and returns the position of 482// the extension end. 483func parseExtension(scan *scanner) int { 484 start, end := scan.start, scan.end 485 switch scan.token[0] { 486 case 'u': // https://www.ietf.org/rfc/rfc6067.txt 487 attrStart := end 488 scan.scan() 489 for last := []byte{}; len(scan.token) > 2; scan.scan() { 490 if bytes.Compare(scan.token, last) != -1 { 491 // Attributes are unsorted. Start over from scratch. 492 p := attrStart + 1 493 scan.next = p 494 attrs := [][]byte{} 495 for scan.scan(); len(scan.token) > 2; scan.scan() { 496 attrs = append(attrs, scan.token) 497 end = scan.end 498 } 499 sort.Sort(bytesSort{attrs, 3}) 500 copy(scan.b[p:], bytes.Join(attrs, separator)) 501 break 502 } 503 last = scan.token 504 end = scan.end 505 } 506 // Scan key-type sequences. A key is of length 2 and may be followed 507 // by 0 or more "type" subtags from 3 to the maximum of 8 letters. 508 var last, key []byte 509 for attrEnd := end; len(scan.token) == 2; last = key { 510 key = scan.token 511 end = scan.end 512 for scan.scan(); end < scan.end && len(scan.token) > 2; scan.scan() { 513 end = scan.end 514 } 515 // TODO: check key value validity 516 if bytes.Compare(key, last) != 1 || scan.err != nil { 517 // We have an invalid key or the keys are not sorted. 518 // Start scanning keys from scratch and reorder. 519 p := attrEnd + 1 520 scan.next = p 521 keys := [][]byte{} 522 for scan.scan(); len(scan.token) == 2; { 523 keyStart := scan.start 524 end = scan.end 525 for scan.scan(); end < scan.end && len(scan.token) > 2; scan.scan() { 526 end = scan.end 527 } 528 keys = append(keys, scan.b[keyStart:end]) 529 } 530 sort.Stable(bytesSort{keys, 2}) 531 if n := len(keys); n > 0 { 532 k := 0 533 for i := 1; i < n; i++ { 534 if !bytes.Equal(keys[k][:2], keys[i][:2]) { 535 k++ 536 keys[k] = keys[i] 537 } else if !bytes.Equal(keys[k], keys[i]) { 538 scan.setError(ErrDuplicateKey) 539 } 540 } 541 keys = keys[:k+1] 542 } 543 reordered := bytes.Join(keys, separator) 544 if e := p + len(reordered); e < end { 545 scan.deleteRange(e, end) 546 end = e 547 } 548 copy(scan.b[p:], reordered) 549 break 550 } 551 } 552 case 't': // https://www.ietf.org/rfc/rfc6497.txt 553 scan.scan() 554 if n := len(scan.token); n >= 2 && n <= 3 && isAlpha(scan.token[1]) { 555 _, end = parseTag(scan) 556 scan.toLower(start, end) 557 } 558 for len(scan.token) == 2 && !isAlpha(scan.token[1]) { 559 end = scan.acceptMinSize(3) 560 } 561 case 'x': 562 end = scan.acceptMinSize(1) 563 default: 564 end = scan.acceptMinSize(2) 565 } 566 return end 567} 568 569// getExtension returns the name, body and end position of the extension. 570func getExtension(s string, p int) (end int, ext string) { 571 if s[p] == '-' { 572 p++ 573 } 574 if s[p] == 'x' { 575 return len(s), s[p:] 576 } 577 end = nextExtension(s, p) 578 return end, s[p:end] 579} 580 581// nextExtension finds the next extension within the string, searching 582// for the -<char>- pattern from position p. 583// In the fast majority of cases, language tags will have at most 584// one extension and extensions tend to be small. 585func nextExtension(s string, p int) int { 586 for n := len(s) - 3; p < n; { 587 if s[p] == '-' { 588 if s[p+2] == '-' { 589 return p 590 } 591 p += 3 592 } else { 593 p++ 594 } 595 } 596 return len(s) 597} 598