utf8.mx raw

   1  // Copyright 2009 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  
   5  // Package utf8 implements functions and constants to support text encoded in
   6  // UTF-8. It includes functions to translate between runes and UTF-8 byte sequences.
   7  // See https://en.wikipedia.org/wiki/UTF-8
   8  package utf8
   9  
  10  // The conditions RuneError==unicode.ReplacementChar and
  11  // MaxRune==unicode.MaxRune are verified in the tests.
  12  // Defining them locally avoids this package depending on package unicode.
  13  
  14  // Numbers fundamental to the encoding.
  15  const (
  16  	RuneError = '\uFFFD'     // the "error" Rune or "Unicode replacement character"
  17  	RuneSelf  = 0x80         // characters below RuneSelf are represented as themselves in a single byte.
  18  	MaxRune   = '\U0010FFFF' // Maximum valid Unicode code point.
  19  	UTFMax    = 4            // maximum number of bytes of a UTF-8 encoded Unicode character.
  20  )
  21  
  22  // Code points in the surrogate range are not valid for UTF-8.
  23  const (
  24  	surrogateMin = 0xD800
  25  	surrogateMax = 0xDFFF
  26  )
  27  
  28  const (
  29  	t1 = 0b00000000
  30  	tx = 0b10000000
  31  	t2 = 0b11000000
  32  	t3 = 0b11100000
  33  	t4 = 0b11110000
  34  	t5 = 0b11111000
  35  
  36  	maskx = 0b00111111
  37  	mask2 = 0b00011111
  38  	mask3 = 0b00001111
  39  	mask4 = 0b00000111
  40  
  41  	rune1Max = 1<<7 - 1
  42  	rune2Max = 1<<11 - 1
  43  	rune3Max = 1<<16 - 1
  44  
  45  	// The default lowest and highest continuation byte.
  46  	locb = 0b10000000
  47  	hicb = 0b10111111
  48  
  49  	// These names of these constants are chosen to give nice alignment in the
  50  	// table below. The first nibble is an index into acceptRanges or F for
  51  	// special one-byte cases. The second nibble is the Rune length or the
  52  	// Status for the special one-byte case.
  53  	xx = 0xF1 // invalid: size 1
  54  	as = 0xF0 // ASCII: size 1
  55  	s1 = 0x02 // accept 0, size 2
  56  	s2 = 0x13 // accept 1, size 3
  57  	s3 = 0x03 // accept 0, size 3
  58  	s4 = 0x23 // accept 2, size 3
  59  	s5 = 0x34 // accept 3, size 4
  60  	s6 = 0x04 // accept 0, size 4
  61  	s7 = 0x44 // accept 4, size 4
  62  )
  63  
  64  const (
  65  	runeErrorByte0 = t3 | (RuneError >> 12)
  66  	runeErrorByte1 = tx | (RuneError>>6)&maskx
  67  	runeErrorByte2 = tx | RuneError&maskx
  68  )
  69  
  70  // first is information about the first byte in a UTF-8 sequence.
  71  var first = [256]uint8{
  72  	//   1   2   3   4   5   6   7   8   9   A   B   C   D   E   F
  73  	as, as, as, as, as, as, as, as, as, as, as, as, as, as, as, as, // 0x00-0x0F
  74  	as, as, as, as, as, as, as, as, as, as, as, as, as, as, as, as, // 0x10-0x1F
  75  	as, as, as, as, as, as, as, as, as, as, as, as, as, as, as, as, // 0x20-0x2F
  76  	as, as, as, as, as, as, as, as, as, as, as, as, as, as, as, as, // 0x30-0x3F
  77  	as, as, as, as, as, as, as, as, as, as, as, as, as, as, as, as, // 0x40-0x4F
  78  	as, as, as, as, as, as, as, as, as, as, as, as, as, as, as, as, // 0x50-0x5F
  79  	as, as, as, as, as, as, as, as, as, as, as, as, as, as, as, as, // 0x60-0x6F
  80  	as, as, as, as, as, as, as, as, as, as, as, as, as, as, as, as, // 0x70-0x7F
  81  	//   1   2   3   4   5   6   7   8   9   A   B   C   D   E   F
  82  	xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, // 0x80-0x8F
  83  	xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, // 0x90-0x9F
  84  	xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, // 0xA0-0xAF
  85  	xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, // 0xB0-0xBF
  86  	xx, xx, s1, s1, s1, s1, s1, s1, s1, s1, s1, s1, s1, s1, s1, s1, // 0xC0-0xCF
  87  	s1, s1, s1, s1, s1, s1, s1, s1, s1, s1, s1, s1, s1, s1, s1, s1, // 0xD0-0xDF
  88  	s2, s3, s3, s3, s3, s3, s3, s3, s3, s3, s3, s3, s3, s4, s3, s3, // 0xE0-0xEF
  89  	s5, s6, s6, s6, s7, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, xx, // 0xF0-0xFF
  90  }
  91  
  92  // acceptRange gives the range of valid values for the second byte in a UTF-8
  93  // sequence.
  94  type acceptRange struct {
  95  	lo uint8 // lowest value for second byte.
  96  	hi uint8 // highest value for second byte.
  97  }
  98  
  99  // acceptRanges has size 16 to avoid bounds checks in the code that uses it.
 100  var acceptRanges = [16]acceptRange{
 101  	0: {locb, hicb},
 102  	1: {0xA0, hicb},
 103  	2: {locb, 0x9F},
 104  	3: {0x90, hicb},
 105  	4: {locb, 0x8F},
 106  }
 107  
 108  // FullRune reports whether the bytes in p begin with a full UTF-8 encoding of a rune.
 109  // An invalid encoding is considered a full Rune since it will convert as a width-1 error rune.
 110  func FullRune(p []byte) bool {
 111  	n := len(p)
 112  	if n == 0 {
 113  		return false
 114  	}
 115  	x := first[p[0]]
 116  	if n >= int(x&7) {
 117  		return true // ASCII, invalid or valid.
 118  	}
 119  	// Must be short or invalid.
 120  	accept := acceptRanges[x>>4]
 121  	if n > 1 && (p[1] < accept.lo || accept.hi < p[1]) {
 122  		return true
 123  	} else if n > 2 && (p[2] < locb || hicb < p[2]) {
 124  		return true
 125  	}
 126  	return false
 127  }
 128  
 129  // FullRuneInString is like FullRune but its input is a string.
 130  func FullRuneInString(s []byte) bool {
 131  	n := len(s)
 132  	if n == 0 {
 133  		return false
 134  	}
 135  	x := first[s[0]]
 136  	if n >= int(x&7) {
 137  		return true // ASCII, invalid, or valid.
 138  	}
 139  	// Must be short or invalid.
 140  	accept := acceptRanges[x>>4]
 141  	if n > 1 && (s[1] < accept.lo || accept.hi < s[1]) {
 142  		return true
 143  	} else if n > 2 && (s[2] < locb || hicb < s[2]) {
 144  		return true
 145  	}
 146  	return false
 147  }
 148  
 149  // DecodeRune unpacks the first UTF-8 encoding in p and returns the rune and
 150  // its width in bytes. If p is empty it returns ([RuneError], 0). Otherwise, if
 151  // the encoding is invalid, it returns (RuneError, 1). Both are impossible
 152  // results for correct, non-empty UTF-8.
 153  //
 154  // An encoding is invalid if it is incorrect UTF-8, encodes a rune that is
 155  // out of range, or is not the shortest possible UTF-8 encoding for the
 156  // value. No other validation is performed.
 157  func DecodeRune(p []byte) (r rune, size int) {
 158  	n := len(p)
 159  	if n < 1 {
 160  		return RuneError, 0
 161  	}
 162  	p0 := p[0]
 163  	x := first[p0]
 164  	if x >= as {
 165  		// The following code simulates an additional check for x == xx and
 166  		// handling the ASCII and invalid cases accordingly. This mask-and-or
 167  		// approach prevents an additional branch.
 168  		mask := rune(x) << 31 >> 31 // Create 0x0000 or 0xFFFF.
 169  		return rune(p[0])&^mask | RuneError&mask, 1
 170  	}
 171  	sz := int(x & 7)
 172  	accept := acceptRanges[x>>4]
 173  	if n < sz {
 174  		return RuneError, 1
 175  	}
 176  	b1 := p[1]
 177  	if b1 < accept.lo || accept.hi < b1 {
 178  		return RuneError, 1
 179  	}
 180  	if sz <= 2 { // <= instead of == to help the compiler eliminate some bounds checks
 181  		return rune(p0&mask2)<<6 | rune(b1&maskx), 2
 182  	}
 183  	b2 := p[2]
 184  	if b2 < locb || hicb < b2 {
 185  		return RuneError, 1
 186  	}
 187  	if sz <= 3 {
 188  		return rune(p0&mask3)<<12 | rune(b1&maskx)<<6 | rune(b2&maskx), 3
 189  	}
 190  	b3 := p[3]
 191  	if b3 < locb || hicb < b3 {
 192  		return RuneError, 1
 193  	}
 194  	return rune(p0&mask4)<<18 | rune(b1&maskx)<<12 | rune(b2&maskx)<<6 | rune(b3&maskx), 4
 195  }
 196  
 197  // DecodeRuneInString is like [DecodeRune] but its input is a string. If s is
 198  // empty it returns ([RuneError], 0). Otherwise, if the encoding is invalid, it
 199  // returns (RuneError, 1). Both are impossible results for correct, non-empty
 200  // UTF-8.
 201  //
 202  // An encoding is invalid if it is incorrect UTF-8, encodes a rune that is
 203  // out of range, or is not the shortest possible UTF-8 encoding for the
 204  // value. No other validation is performed.
 205  func DecodeRuneInString(s []byte) (r rune, size int) {
 206  	n := len(s)
 207  	if n < 1 {
 208  		return RuneError, 0
 209  	}
 210  	s0 := s[0]
 211  	x := first[s0]
 212  	if x >= as {
 213  		// The following code simulates an additional check for x == xx and
 214  		// handling the ASCII and invalid cases accordingly. This mask-and-or
 215  		// approach prevents an additional branch.
 216  		mask := rune(x) << 31 >> 31 // Create 0x0000 or 0xFFFF.
 217  		return rune(s[0])&^mask | RuneError&mask, 1
 218  	}
 219  	sz := int(x & 7)
 220  	accept := acceptRanges[x>>4]
 221  	if n < sz {
 222  		return RuneError, 1
 223  	}
 224  	s1 := s[1]
 225  	if s1 < accept.lo || accept.hi < s1 {
 226  		return RuneError, 1
 227  	}
 228  	if sz <= 2 { // <= instead of == to help the compiler eliminate some bounds checks
 229  		return rune(s0&mask2)<<6 | rune(s1&maskx), 2
 230  	}
 231  	s2 := s[2]
 232  	if s2 < locb || hicb < s2 {
 233  		return RuneError, 1
 234  	}
 235  	if sz <= 3 {
 236  		return rune(s0&mask3)<<12 | rune(s1&maskx)<<6 | rune(s2&maskx), 3
 237  	}
 238  	s3 := s[3]
 239  	if s3 < locb || hicb < s3 {
 240  		return RuneError, 1
 241  	}
 242  	return rune(s0&mask4)<<18 | rune(s1&maskx)<<12 | rune(s2&maskx)<<6 | rune(s3&maskx), 4
 243  }
 244  
 245  // DecodeLastRune unpacks the last UTF-8 encoding in p and returns the rune and
 246  // its width in bytes. If p is empty it returns ([RuneError], 0). Otherwise, if
 247  // the encoding is invalid, it returns (RuneError, 1). Both are impossible
 248  // results for correct, non-empty UTF-8.
 249  //
 250  // An encoding is invalid if it is incorrect UTF-8, encodes a rune that is
 251  // out of range, or is not the shortest possible UTF-8 encoding for the
 252  // value. No other validation is performed.
 253  func DecodeLastRune(p []byte) (r rune, size int) {
 254  	end := len(p)
 255  	if end == 0 {
 256  		return RuneError, 0
 257  	}
 258  	start := end - 1
 259  	r = rune(p[start])
 260  	if r < RuneSelf {
 261  		return r, 1
 262  	}
 263  	// guard against O(n^2) behavior when traversing
 264  	// backwards through strings with long sequences of
 265  	// invalid UTF-8.
 266  	lim := max(end-UTFMax, 0)
 267  	for start--; start >= lim; start-- {
 268  		if RuneStart(p[start]) {
 269  			break
 270  		}
 271  	}
 272  	if start < 0 {
 273  		start = 0
 274  	}
 275  	r, size = DecodeRune(p[start:end])
 276  	if start+size != end {
 277  		return RuneError, 1
 278  	}
 279  	return r, size
 280  }
 281  
 282  // DecodeLastRuneInString is like [DecodeLastRune] but its input is a string. If
 283  // s is empty it returns ([RuneError], 0). Otherwise, if the encoding is invalid,
 284  // it returns (RuneError, 1). Both are impossible results for correct,
 285  // non-empty UTF-8.
 286  //
 287  // An encoding is invalid if it is incorrect UTF-8, encodes a rune that is
 288  // out of range, or is not the shortest possible UTF-8 encoding for the
 289  // value. No other validation is performed.
 290  func DecodeLastRuneInString(s []byte) (r rune, size int) {
 291  	end := len(s)
 292  	if end == 0 {
 293  		return RuneError, 0
 294  	}
 295  	start := end - 1
 296  	r = rune(s[start])
 297  	if r < RuneSelf {
 298  		return r, 1
 299  	}
 300  	// guard against O(n^2) behavior when traversing
 301  	// backwards through strings with long sequences of
 302  	// invalid UTF-8.
 303  	lim := max(end-UTFMax, 0)
 304  	for start--; start >= lim; start-- {
 305  		if RuneStart(s[start]) {
 306  			break
 307  		}
 308  	}
 309  	if start < 0 {
 310  		start = 0
 311  	}
 312  	r, size = DecodeRuneInString(s[start:end])
 313  	if start+size != end {
 314  		return RuneError, 1
 315  	}
 316  	return r, size
 317  }
 318  
 319  // RuneLen returns the number of bytes in the UTF-8 encoding of the rune.
 320  // It returns -1 if the rune is not a valid value to encode in UTF-8.
 321  func RuneLen(r rune) int {
 322  	switch {
 323  	case r < 0:
 324  		return -1
 325  	case r <= rune1Max:
 326  		return 1
 327  	case r <= rune2Max:
 328  		return 2
 329  	case surrogateMin <= r && r <= surrogateMax:
 330  		return -1
 331  	case r <= rune3Max:
 332  		return 3
 333  	case r <= MaxRune:
 334  		return 4
 335  	}
 336  	return -1
 337  }
 338  
 339  // EncodeRune writes into p (which must be large enough) the UTF-8 encoding of the rune.
 340  // If the rune is out of range, it writes the encoding of [RuneError].
 341  // It returns the number of bytes written.
 342  func EncodeRune(p []byte, r rune) int {
 343  	// This function is inlineable for fast handling of ASCII.
 344  	if uint32(r) <= rune1Max {
 345  		p[0] = byte(r)
 346  		return 1
 347  	}
 348  	return encodeRuneNonASCII(p, r)
 349  }
 350  
 351  func encodeRuneNonASCII(p []byte, r rune) int {
 352  	// Negative values are erroneous. Making it unsigned addresses the problem.
 353  	switch i := uint32(r); {
 354  	case i <= rune2Max:
 355  		_ = p[1] // eliminate bounds checks
 356  		p[0] = t2 | byte(r>>6)
 357  		p[1] = tx | byte(r)&maskx
 358  		return 2
 359  	case i < surrogateMin, surrogateMax < i && i <= rune3Max:
 360  		_ = p[2] // eliminate bounds checks
 361  		p[0] = t3 | byte(r>>12)
 362  		p[1] = tx | byte(r>>6)&maskx
 363  		p[2] = tx | byte(r)&maskx
 364  		return 3
 365  	case i > rune3Max && i <= MaxRune:
 366  		_ = p[3] // eliminate bounds checks
 367  		p[0] = t4 | byte(r>>18)
 368  		p[1] = tx | byte(r>>12)&maskx
 369  		p[2] = tx | byte(r>>6)&maskx
 370  		p[3] = tx | byte(r)&maskx
 371  		return 4
 372  	default:
 373  		_ = p[2] // eliminate bounds checks
 374  		p[0] = runeErrorByte0
 375  		p[1] = runeErrorByte1
 376  		p[2] = runeErrorByte2
 377  		return 3
 378  	}
 379  }
 380  
 381  // AppendRune appends the UTF-8 encoding of r to the end of p and
 382  // returns the extended buffer. If the rune is out of range,
 383  // it appends the encoding of [RuneError].
 384  func AppendRune(p []byte, r rune) []byte {
 385  	// This function is inlineable for fast handling of ASCII.
 386  	if uint32(r) <= rune1Max {
 387  		return append(p, byte(r))
 388  	}
 389  	return appendRuneNonASCII(p, r)
 390  }
 391  
 392  func appendRuneNonASCII(p []byte, r rune) []byte {
 393  	// Negative values are erroneous. Making it unsigned addresses the problem.
 394  	switch i := uint32(r); {
 395  	case i <= rune2Max:
 396  		return append(p, t2|byte(r>>6), tx|byte(r)&maskx)
 397  	case i < surrogateMin, surrogateMax < i && i <= rune3Max:
 398  		return append(p, t3|byte(r>>12), tx|byte(r>>6)&maskx, tx|byte(r)&maskx)
 399  	case i > rune3Max && i <= MaxRune:
 400  		return append(p, t4|byte(r>>18), tx|byte(r>>12)&maskx, tx|byte(r>>6)&maskx, tx|byte(r)&maskx)
 401  	default:
 402  		return append(p, runeErrorByte0, runeErrorByte1, runeErrorByte2)
 403  	}
 404  }
 405  
 406  // RuneCount returns the number of runes in p. Erroneous and short
 407  // encodings are treated as single runes of width 1 byte.
 408  func RuneCount(p []byte) int {
 409  	np := len(p)
 410  	var n int
 411  	for ; n < np; n++ {
 412  		if c := p[n]; c >= RuneSelf {
 413  			// non-ASCII slow path
 414  			return n + RuneCountInString([]byte(p[n:]))
 415  		}
 416  	}
 417  	return n
 418  }
 419  
 420  // RuneCountInString is like [RuneCount] but its input is a string.
 421  func RuneCountInString(s []byte) (n int) {
 422  	for range s {
 423  		n++
 424  	}
 425  	return n
 426  }
 427  
 428  // RuneStart reports whether the byte could be the first byte of an encoded,
 429  // possibly invalid rune. Second and subsequent bytes always have the top two
 430  // bits set to 10.
 431  func RuneStart(b byte) bool { return b&0xC0 != 0x80 }
 432  
 433  // Valid reports whether p consists entirely of valid UTF-8-encoded runes.
 434  func Valid(p []byte) bool {
 435  	// This optimization avoids the need to recompute the capacity
 436  	// when generating code for p[8:], bringing it to parity with
 437  	// ValidString, which was 20% faster on long ASCII strings.
 438  	p = p[:len(p):len(p)]
 439  
 440  	// Fast path. Check for and skip 8 bytes of ASCII characters per iteration.
 441  	for len(p) >= 8 {
 442  		// Combining two 32 bit loads allows the same code to be used
 443  		// for 32 and 64 bit platforms.
 444  		// The compiler can generate a 32bit load for first32 and second32
 445  		// on many platforms. See test/codegen/memcombine.go.
 446  		first32 := uint32(p[0]) | uint32(p[1])<<8 | uint32(p[2])<<16 | uint32(p[3])<<24
 447  		second32 := uint32(p[4]) | uint32(p[5])<<8 | uint32(p[6])<<16 | uint32(p[7])<<24
 448  		if (first32|second32)&0x80808080 != 0 {
 449  			// Found a non ASCII byte (>= RuneSelf).
 450  			break
 451  		}
 452  		p = p[8:]
 453  	}
 454  	n := len(p)
 455  	for i := 0; i < n; {
 456  		pi := p[i]
 457  		if pi < RuneSelf {
 458  			i++
 459  			continue
 460  		}
 461  		x := first[pi]
 462  		if x == xx {
 463  			return false // Illegal starter byte.
 464  		}
 465  		size := int(x & 7)
 466  		if i+size > n {
 467  			return false // Short or invalid.
 468  		}
 469  		accept := acceptRanges[x>>4]
 470  		if c := p[i+1]; c < accept.lo || accept.hi < c {
 471  			return false
 472  		} else if size == 2 {
 473  		} else if c := p[i+2]; c < locb || hicb < c {
 474  			return false
 475  		} else if size == 3 {
 476  		} else if c := p[i+3]; c < locb || hicb < c {
 477  			return false
 478  		}
 479  		i += size
 480  	}
 481  	return true
 482  }
 483  
 484  // ValidString reports whether s consists entirely of valid UTF-8-encoded runes.
 485  func ValidString(s []byte) bool {
 486  	// Fast path. Check for and skip 8 bytes of ASCII characters per iteration.
 487  	for len(s) >= 8 {
 488  		// Combining two 32 bit loads allows the same code to be used
 489  		// for 32 and 64 bit platforms.
 490  		// The compiler can generate a 32bit load for first32 and second32
 491  		// on many platforms. See test/codegen/memcombine.go.
 492  		first32 := uint32(s[0]) | uint32(s[1])<<8 | uint32(s[2])<<16 | uint32(s[3])<<24
 493  		second32 := uint32(s[4]) | uint32(s[5])<<8 | uint32(s[6])<<16 | uint32(s[7])<<24
 494  		if (first32|second32)&0x80808080 != 0 {
 495  			// Found a non ASCII byte (>= RuneSelf).
 496  			break
 497  		}
 498  		s = s[8:]
 499  	}
 500  	n := len(s)
 501  	for i := 0; i < n; {
 502  		si := s[i]
 503  		if si < RuneSelf {
 504  			i++
 505  			continue
 506  		}
 507  		x := first[si]
 508  		if x == xx {
 509  			return false // Illegal starter byte.
 510  		}
 511  		size := int(x & 7)
 512  		if i+size > n {
 513  			return false // Short or invalid.
 514  		}
 515  		accept := acceptRanges[x>>4]
 516  		if c := s[i+1]; c < accept.lo || accept.hi < c {
 517  			return false
 518  		} else if size == 2 {
 519  		} else if c := s[i+2]; c < locb || hicb < c {
 520  			return false
 521  		} else if size == 3 {
 522  		} else if c := s[i+3]; c < locb || hicb < c {
 523  			return false
 524  		}
 525  		i += size
 526  	}
 527  	return true
 528  }
 529  
 530  // ValidRune reports whether r can be legally encoded as UTF-8.
 531  // Code points that are out of range or a surrogate half are illegal.
 532  func ValidRune(r rune) bool {
 533  	switch {
 534  	case 0 <= r && r < surrogateMin:
 535  		return true
 536  	case surrogateMax < r && r <= MaxRune:
 537  		return true
 538  	}
 539  	return false
 540  }
 541