unicode.go raw

   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  
   5  // Package unicode provides Unicode encodings such as UTF-16.
   6  package unicode // import "golang.org/x/text/encoding/unicode"
   7  
   8  import (
   9  	"bytes"
  10  	"errors"
  11  	"unicode/utf16"
  12  	"unicode/utf8"
  13  
  14  	"golang.org/x/text/encoding"
  15  	"golang.org/x/text/encoding/internal"
  16  	"golang.org/x/text/encoding/internal/identifier"
  17  	"golang.org/x/text/internal/utf8internal"
  18  	"golang.org/x/text/runes"
  19  	"golang.org/x/text/transform"
  20  )
  21  
  22  // TODO: I think the Transformers really should return errors on unmatched
  23  // surrogate pairs and odd numbers of bytes. This is not required by RFC 2781,
  24  // which leaves it open, but is suggested by WhatWG. It will allow for all error
  25  // modes as defined by WhatWG: fatal, HTML and Replacement. This would require
  26  // the introduction of some kind of error type for conveying the erroneous code
  27  // point.
  28  
  29  // UTF8 is the UTF-8 encoding. It neither removes nor adds byte order marks.
  30  var UTF8 encoding.Encoding = utf8enc
  31  
  32  // UTF8BOM is an UTF-8 encoding where the decoder strips a leading byte order
  33  // mark while the encoder adds one.
  34  //
  35  // Some editors add a byte order mark as a signature to UTF-8 files. Although
  36  // the byte order mark is not useful for detecting byte order in UTF-8, it is
  37  // sometimes used as a convention to mark UTF-8-encoded files. This relies on
  38  // the observation that the UTF-8 byte order mark is either an illegal or at
  39  // least very unlikely sequence in any other character encoding.
  40  var UTF8BOM encoding.Encoding = utf8bomEncoding{}
  41  
  42  type utf8bomEncoding struct{}
  43  
  44  func (utf8bomEncoding) String() string {
  45  	return "UTF-8-BOM"
  46  }
  47  
  48  func (utf8bomEncoding) ID() (identifier.MIB, string) {
  49  	return identifier.Unofficial, "x-utf8bom"
  50  }
  51  
  52  func (utf8bomEncoding) NewEncoder() *encoding.Encoder {
  53  	return &encoding.Encoder{
  54  		Transformer: &utf8bomEncoder{t: runes.ReplaceIllFormed()},
  55  	}
  56  }
  57  
  58  func (utf8bomEncoding) NewDecoder() *encoding.Decoder {
  59  	return &encoding.Decoder{Transformer: &utf8bomDecoder{}}
  60  }
  61  
  62  var utf8enc = &internal.Encoding{
  63  	Encoding: &internal.SimpleEncoding{Decoder: utf8Decoder{}, Encoder: runes.ReplaceIllFormed()},
  64  	Name:     "UTF-8",
  65  	MIB:      identifier.UTF8,
  66  }
  67  
  68  type utf8bomDecoder struct {
  69  	checked bool
  70  }
  71  
  72  func (t *utf8bomDecoder) Reset() {
  73  	t.checked = false
  74  }
  75  
  76  func (t *utf8bomDecoder) Transform(dst, src []byte, atEOF bool) (nDst, nSrc int, err error) {
  77  	if !t.checked {
  78  		if !atEOF && len(src) < len(utf8BOM) {
  79  			if len(src) == 0 {
  80  				return 0, 0, nil
  81  			}
  82  			return 0, 0, transform.ErrShortSrc
  83  		}
  84  		if bytes.HasPrefix(src, []byte(utf8BOM)) {
  85  			nSrc += len(utf8BOM)
  86  			src = src[len(utf8BOM):]
  87  		}
  88  		t.checked = true
  89  	}
  90  	nDst, n, err := utf8Decoder.Transform(utf8Decoder{}, dst[nDst:], src, atEOF)
  91  	nSrc += n
  92  	return nDst, nSrc, err
  93  }
  94  
  95  type utf8bomEncoder struct {
  96  	written bool
  97  	t       transform.Transformer
  98  }
  99  
 100  func (t *utf8bomEncoder) Reset() {
 101  	t.written = false
 102  	t.t.Reset()
 103  }
 104  
 105  func (t *utf8bomEncoder) Transform(dst, src []byte, atEOF bool) (nDst, nSrc int, err error) {
 106  	if !t.written {
 107  		if len(dst) < len(utf8BOM) {
 108  			return nDst, 0, transform.ErrShortDst
 109  		}
 110  		nDst = copy(dst, utf8BOM)
 111  		t.written = true
 112  	}
 113  	n, nSrc, err := utf8Decoder.Transform(utf8Decoder{}, dst[nDst:], src, atEOF)
 114  	nDst += n
 115  	return nDst, nSrc, err
 116  }
 117  
 118  type utf8Decoder struct{ transform.NopResetter }
 119  
 120  func (utf8Decoder) Transform(dst, src []byte, atEOF bool) (nDst, nSrc int, err error) {
 121  	var pSrc int // point from which to start copy in src
 122  	var accept utf8internal.AcceptRange
 123  
 124  	// The decoder can only make the input larger, not smaller.
 125  	n := len(src)
 126  	if len(dst) < n {
 127  		err = transform.ErrShortDst
 128  		n = len(dst)
 129  		atEOF = false
 130  	}
 131  	for nSrc < n {
 132  		c := src[nSrc]
 133  		if c < utf8.RuneSelf {
 134  			nSrc++
 135  			continue
 136  		}
 137  		first := utf8internal.First[c]
 138  		size := int(first & utf8internal.SizeMask)
 139  		if first == utf8internal.FirstInvalid {
 140  			goto handleInvalid // invalid starter byte
 141  		}
 142  		accept = utf8internal.AcceptRanges[first>>utf8internal.AcceptShift]
 143  		if nSrc+size > n {
 144  			if !atEOF {
 145  				// We may stop earlier than necessary here if the short sequence
 146  				// has invalid bytes. Not checking for this simplifies the code
 147  				// and may avoid duplicate computations in certain conditions.
 148  				if err == nil {
 149  					err = transform.ErrShortSrc
 150  				}
 151  				break
 152  			}
 153  			// Determine the maximal subpart of an ill-formed subsequence.
 154  			switch {
 155  			case nSrc+1 >= n || src[nSrc+1] < accept.Lo || accept.Hi < src[nSrc+1]:
 156  				size = 1
 157  			case nSrc+2 >= n || src[nSrc+2] < utf8internal.LoCB || utf8internal.HiCB < src[nSrc+2]:
 158  				size = 2
 159  			default:
 160  				size = 3 // As we are short, the maximum is 3.
 161  			}
 162  			goto handleInvalid
 163  		}
 164  		if c = src[nSrc+1]; c < accept.Lo || accept.Hi < c {
 165  			size = 1
 166  			goto handleInvalid // invalid continuation byte
 167  		} else if size == 2 {
 168  		} else if c = src[nSrc+2]; c < utf8internal.LoCB || utf8internal.HiCB < c {
 169  			size = 2
 170  			goto handleInvalid // invalid continuation byte
 171  		} else if size == 3 {
 172  		} else if c = src[nSrc+3]; c < utf8internal.LoCB || utf8internal.HiCB < c {
 173  			size = 3
 174  			goto handleInvalid // invalid continuation byte
 175  		}
 176  		nSrc += size
 177  		continue
 178  
 179  	handleInvalid:
 180  		// Copy the scanned input so far.
 181  		nDst += copy(dst[nDst:], src[pSrc:nSrc])
 182  
 183  		// Append RuneError to the destination.
 184  		const runeError = "\ufffd"
 185  		if nDst+len(runeError) > len(dst) {
 186  			return nDst, nSrc, transform.ErrShortDst
 187  		}
 188  		nDst += copy(dst[nDst:], runeError)
 189  
 190  		// Skip the maximal subpart of an ill-formed subsequence according to
 191  		// the W3C standard way instead of the Go way. This Transform is
 192  		// probably the only place in the text repo where it is warranted.
 193  		nSrc += size
 194  		pSrc = nSrc
 195  
 196  		// Recompute the maximum source length.
 197  		if sz := len(dst) - nDst; sz < len(src)-nSrc {
 198  			err = transform.ErrShortDst
 199  			n = nSrc + sz
 200  			atEOF = false
 201  		}
 202  	}
 203  	return nDst + copy(dst[nDst:], src[pSrc:nSrc]), nSrc, err
 204  }
 205  
 206  // UTF16 returns a UTF-16 Encoding for the given default endianness and byte
 207  // order mark (BOM) policy.
 208  //
 209  // When decoding from UTF-16 to UTF-8, if the BOMPolicy is IgnoreBOM then
 210  // neither BOMs U+FEFF nor noncharacters U+FFFE in the input stream will affect
 211  // the endianness used for decoding, and will instead be output as their
 212  // standard UTF-8 encodings: "\xef\xbb\xbf" and "\xef\xbf\xbe". If the BOMPolicy
 213  // is UseBOM or ExpectBOM a staring BOM is not written to the UTF-8 output.
 214  // Instead, it overrides the default endianness e for the remainder of the
 215  // transformation. Any subsequent BOMs U+FEFF or noncharacters U+FFFE will not
 216  // affect the endianness used, and will instead be output as their standard
 217  // UTF-8 encodings. For UseBOM, if there is no starting BOM, it will proceed
 218  // with the default Endianness. For ExpectBOM, in that case, the transformation
 219  // will return early with an ErrMissingBOM error.
 220  //
 221  // When encoding from UTF-8 to UTF-16, a BOM will be inserted at the start of
 222  // the output if the BOMPolicy is UseBOM or ExpectBOM. Otherwise, a BOM will not
 223  // be inserted. The UTF-8 input does not need to contain a BOM.
 224  //
 225  // There is no concept of a 'native' endianness. If the UTF-16 data is produced
 226  // and consumed in a greater context that implies a certain endianness, use
 227  // IgnoreBOM. Otherwise, use ExpectBOM and always produce and consume a BOM.
 228  //
 229  // In the language of https://www.unicode.org/faq/utf_bom.html#bom10, IgnoreBOM
 230  // corresponds to "Where the precise type of the data stream is known... the
 231  // BOM should not be used" and ExpectBOM corresponds to "A particular
 232  // protocol... may require use of the BOM".
 233  func UTF16(e Endianness, b BOMPolicy) encoding.Encoding {
 234  	return utf16Encoding{config{e, b}, mibValue[e][b&bomMask]}
 235  }
 236  
 237  // mibValue maps Endianness and BOMPolicy settings to MIB constants. Note that
 238  // some configurations map to the same MIB identifier. RFC 2781 has requirements
 239  // and recommendations. Some of the "configurations" are merely recommendations,
 240  // so multiple configurations could match.
 241  var mibValue = map[Endianness][numBOMValues]identifier.MIB{
 242  	BigEndian: [numBOMValues]identifier.MIB{
 243  		IgnoreBOM: identifier.UTF16BE,
 244  		UseBOM:    identifier.UTF16, // BigEnding default is preferred by RFC 2781.
 245  		// TODO: acceptBOM | strictBOM would map to UTF16BE as well.
 246  	},
 247  	LittleEndian: [numBOMValues]identifier.MIB{
 248  		IgnoreBOM: identifier.UTF16LE,
 249  		UseBOM:    identifier.UTF16, // LittleEndian default is allowed and preferred on Windows.
 250  		// TODO: acceptBOM | strictBOM would map to UTF16LE as well.
 251  	},
 252  	// ExpectBOM is not widely used and has no valid MIB identifier.
 253  }
 254  
 255  // All lists a configuration for each IANA-defined UTF-16 variant.
 256  var All = []encoding.Encoding{
 257  	UTF8,
 258  	UTF16(BigEndian, UseBOM),
 259  	UTF16(BigEndian, IgnoreBOM),
 260  	UTF16(LittleEndian, IgnoreBOM),
 261  }
 262  
 263  // BOMPolicy is a UTF-16 encoding's byte order mark policy.
 264  type BOMPolicy uint8
 265  
 266  const (
 267  	writeBOM   BOMPolicy = 0x01
 268  	acceptBOM  BOMPolicy = 0x02
 269  	requireBOM BOMPolicy = 0x04
 270  	bomMask    BOMPolicy = 0x07
 271  
 272  	// HACK: numBOMValues == 8 triggers a bug in the 1.4 compiler (cannot have a
 273  	// map of an array of length 8 of a type that is also used as a key or value
 274  	// in another map). See golang.org/issue/11354.
 275  	// TODO: consider changing this value back to 8 if the use of 1.4.* has
 276  	// been minimized.
 277  	numBOMValues = 8 + 1
 278  
 279  	// IgnoreBOM means to ignore any byte order marks.
 280  	IgnoreBOM BOMPolicy = 0
 281  	// Common and RFC 2781-compliant interpretation for UTF-16BE/LE.
 282  
 283  	// UseBOM means that the UTF-16 form may start with a byte order mark, which
 284  	// will be used to override the default encoding.
 285  	UseBOM BOMPolicy = writeBOM | acceptBOM
 286  	// Common and RFC 2781-compliant interpretation for UTF-16.
 287  
 288  	// ExpectBOM means that the UTF-16 form must start with a byte order mark,
 289  	// which will be used to override the default encoding.
 290  	ExpectBOM BOMPolicy = writeBOM | acceptBOM | requireBOM
 291  	// Used in Java as Unicode (not to be confused with Java's UTF-16) and
 292  	// ICU's UTF-16,version=1. Not compliant with RFC 2781.
 293  
 294  	// TODO (maybe): strictBOM: BOM must match Endianness. This would allow:
 295  	// - UTF-16(B|L)E,version=1: writeBOM | acceptBOM | requireBOM | strictBOM
 296  	//    (UnicodeBig and UnicodeLittle in Java)
 297  	// - RFC 2781-compliant, but less common interpretation for UTF-16(B|L)E:
 298  	//    acceptBOM | strictBOM (e.g. assigned to CheckBOM).
 299  	// This addition would be consistent with supporting ExpectBOM.
 300  )
 301  
 302  // Endianness is a UTF-16 encoding's default endianness.
 303  type Endianness bool
 304  
 305  const (
 306  	// BigEndian is UTF-16BE.
 307  	BigEndian Endianness = false
 308  	// LittleEndian is UTF-16LE.
 309  	LittleEndian Endianness = true
 310  )
 311  
 312  // ErrMissingBOM means that decoding UTF-16 input with ExpectBOM did not find a
 313  // starting byte order mark.
 314  var ErrMissingBOM = errors.New("encoding: missing byte order mark")
 315  
 316  type utf16Encoding struct {
 317  	config
 318  	mib identifier.MIB
 319  }
 320  
 321  type config struct {
 322  	endianness Endianness
 323  	bomPolicy  BOMPolicy
 324  }
 325  
 326  func (u utf16Encoding) NewDecoder() *encoding.Decoder {
 327  	return &encoding.Decoder{Transformer: &utf16Decoder{
 328  		initial: u.config,
 329  		current: u.config,
 330  	}}
 331  }
 332  
 333  func (u utf16Encoding) NewEncoder() *encoding.Encoder {
 334  	return &encoding.Encoder{Transformer: &utf16Encoder{
 335  		endianness:       u.endianness,
 336  		initialBOMPolicy: u.bomPolicy,
 337  		currentBOMPolicy: u.bomPolicy,
 338  	}}
 339  }
 340  
 341  func (u utf16Encoding) ID() (mib identifier.MIB, other string) {
 342  	return u.mib, ""
 343  }
 344  
 345  func (u utf16Encoding) String() string {
 346  	e, b := "B", ""
 347  	if u.endianness == LittleEndian {
 348  		e = "L"
 349  	}
 350  	switch u.bomPolicy {
 351  	case ExpectBOM:
 352  		b = "Expect"
 353  	case UseBOM:
 354  		b = "Use"
 355  	case IgnoreBOM:
 356  		b = "Ignore"
 357  	}
 358  	return "UTF-16" + e + "E (" + b + " BOM)"
 359  }
 360  
 361  type utf16Decoder struct {
 362  	initial config
 363  	current config
 364  }
 365  
 366  func (u *utf16Decoder) Reset() {
 367  	u.current = u.initial
 368  }
 369  
 370  func (u *utf16Decoder) Transform(dst, src []byte, atEOF bool) (nDst, nSrc int, err error) {
 371  	if len(src) < 2 && atEOF && u.current.bomPolicy&requireBOM != 0 {
 372  		return 0, 0, ErrMissingBOM
 373  	}
 374  	if len(src) == 0 {
 375  		return 0, 0, nil
 376  	}
 377  	if len(src) >= 2 && u.current.bomPolicy&acceptBOM != 0 {
 378  		switch {
 379  		case src[0] == 0xfe && src[1] == 0xff:
 380  			u.current.endianness = BigEndian
 381  			nSrc = 2
 382  		case src[0] == 0xff && src[1] == 0xfe:
 383  			u.current.endianness = LittleEndian
 384  			nSrc = 2
 385  		default:
 386  			if u.current.bomPolicy&requireBOM != 0 {
 387  				return 0, 0, ErrMissingBOM
 388  			}
 389  		}
 390  		u.current.bomPolicy = IgnoreBOM
 391  	}
 392  
 393  	var r rune
 394  	var dSize, sSize int
 395  	for nSrc < len(src) {
 396  		if nSrc+1 < len(src) {
 397  			x := uint16(src[nSrc+0])<<8 | uint16(src[nSrc+1])
 398  			if u.current.endianness == LittleEndian {
 399  				x = x>>8 | x<<8
 400  			}
 401  			r, sSize = rune(x), 2
 402  			if utf16.IsSurrogate(r) {
 403  				if nSrc+3 < len(src) {
 404  					x = uint16(src[nSrc+2])<<8 | uint16(src[nSrc+3])
 405  					if u.current.endianness == LittleEndian {
 406  						x = x>>8 | x<<8
 407  					}
 408  					// Save for next iteration if it is not a high surrogate.
 409  					if isHighSurrogate(rune(x)) {
 410  						r, sSize = utf16.DecodeRune(r, rune(x)), 4
 411  					}
 412  				} else if !atEOF {
 413  					err = transform.ErrShortSrc
 414  					break
 415  				}
 416  			}
 417  			if dSize = utf8.RuneLen(r); dSize < 0 {
 418  				r, dSize = utf8.RuneError, 3
 419  			}
 420  		} else if atEOF {
 421  			// Single trailing byte.
 422  			r, dSize, sSize = utf8.RuneError, 3, 1
 423  		} else {
 424  			err = transform.ErrShortSrc
 425  			break
 426  		}
 427  		if nDst+dSize > len(dst) {
 428  			err = transform.ErrShortDst
 429  			break
 430  		}
 431  		nDst += utf8.EncodeRune(dst[nDst:], r)
 432  		nSrc += sSize
 433  	}
 434  	return nDst, nSrc, err
 435  }
 436  
 437  func isHighSurrogate(r rune) bool {
 438  	return 0xDC00 <= r && r <= 0xDFFF
 439  }
 440  
 441  type utf16Encoder struct {
 442  	endianness       Endianness
 443  	initialBOMPolicy BOMPolicy
 444  	currentBOMPolicy BOMPolicy
 445  }
 446  
 447  func (u *utf16Encoder) Reset() {
 448  	u.currentBOMPolicy = u.initialBOMPolicy
 449  }
 450  
 451  func (u *utf16Encoder) Transform(dst, src []byte, atEOF bool) (nDst, nSrc int, err error) {
 452  	if u.currentBOMPolicy&writeBOM != 0 {
 453  		if len(dst) < 2 {
 454  			return 0, 0, transform.ErrShortDst
 455  		}
 456  		dst[0], dst[1] = 0xfe, 0xff
 457  		u.currentBOMPolicy = IgnoreBOM
 458  		nDst = 2
 459  	}
 460  
 461  	r, size := rune(0), 0
 462  	for nSrc < len(src) {
 463  		r = rune(src[nSrc])
 464  
 465  		// Decode a 1-byte rune.
 466  		if r < utf8.RuneSelf {
 467  			size = 1
 468  
 469  		} else {
 470  			// Decode a multi-byte rune.
 471  			r, size = utf8.DecodeRune(src[nSrc:])
 472  			if size == 1 {
 473  				// All valid runes of size 1 (those below utf8.RuneSelf) were
 474  				// handled above. We have invalid UTF-8 or we haven't seen the
 475  				// full character yet.
 476  				if !atEOF && !utf8.FullRune(src[nSrc:]) {
 477  					err = transform.ErrShortSrc
 478  					break
 479  				}
 480  			}
 481  		}
 482  
 483  		if r <= 0xffff {
 484  			if nDst+2 > len(dst) {
 485  				err = transform.ErrShortDst
 486  				break
 487  			}
 488  			dst[nDst+0] = uint8(r >> 8)
 489  			dst[nDst+1] = uint8(r)
 490  			nDst += 2
 491  		} else {
 492  			if nDst+4 > len(dst) {
 493  				err = transform.ErrShortDst
 494  				break
 495  			}
 496  			r1, r2 := utf16.EncodeRune(r)
 497  			dst[nDst+0] = uint8(r1 >> 8)
 498  			dst[nDst+1] = uint8(r1)
 499  			dst[nDst+2] = uint8(r2 >> 8)
 500  			dst[nDst+3] = uint8(r2)
 501  			nDst += 4
 502  		}
 503  		nSrc += size
 504  	}
 505  
 506  	if u.endianness == LittleEndian {
 507  		for i := 0; i < nDst; i += 2 {
 508  			dst[i], dst[i+1] = dst[i+1], dst[i]
 509  		}
 510  	}
 511  	return nDst, nSrc, err
 512  }
 513