Source file src/regexp/syntax/parse.go

Documentation: regexp/syntax

     1  // Copyright 2011 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 syntax
     6  
     7  import (
     8  	"sort"
     9  	"strings"
    10  	"unicode"
    11  	"unicode/utf8"
    12  )
    13  
    14  // An Error describes a failure to parse a regular expression
    15  // and gives the offending expression.
    16  type Error struct {
    17  	Code ErrorCode
    18  	Expr string
    19  }
    20  
    21  func (e *Error) Error() string {
    22  	return "error parsing regexp: " + e.Code.String() + ": `" + e.Expr + "`"
    23  }
    24  
    25  // An ErrorCode describes a failure to parse a regular expression.
    26  type ErrorCode string
    27  
    28  const (
    29  	// Unexpected error
    30  	ErrInternalError ErrorCode = "regexp/syntax: internal error"
    31  
    32  	// Parse errors
    33  	ErrInvalidCharClass      ErrorCode = "invalid character class"
    34  	ErrInvalidCharRange      ErrorCode = "invalid character class range"
    35  	ErrInvalidEscape         ErrorCode = "invalid escape sequence"
    36  	ErrInvalidNamedCapture   ErrorCode = "invalid named capture"
    37  	ErrInvalidPerlOp         ErrorCode = "invalid or unsupported Perl syntax"
    38  	ErrInvalidRepeatOp       ErrorCode = "invalid nested repetition operator"
    39  	ErrInvalidRepeatSize     ErrorCode = "invalid repeat count"
    40  	ErrInvalidUTF8           ErrorCode = "invalid UTF-8"
    41  	ErrMissingBracket        ErrorCode = "missing closing ]"
    42  	ErrMissingParen          ErrorCode = "missing closing )"
    43  	ErrMissingRepeatArgument ErrorCode = "missing argument to repetition operator"
    44  	ErrTrailingBackslash     ErrorCode = "trailing backslash at end of expression"
    45  	ErrUnexpectedParen       ErrorCode = "unexpected )"
    46  	ErrNestingDepth          ErrorCode = "expression nests too deeply"
    47  	ErrLarge                 ErrorCode = "expression too large"
    48  )
    49  
    50  func (e ErrorCode) String() string {
    51  	return string(e)
    52  }
    53  
    54  // Flags control the behavior of the parser and record information about regexp context.
    55  type Flags uint16
    56  
    57  const (
    58  	FoldCase      Flags = 1 << iota // case-insensitive match
    59  	Literal                         // treat pattern as literal string
    60  	ClassNL                         // allow character classes like [^a-z] and [[:space:]] to match newline
    61  	DotNL                           // allow . to match newline
    62  	OneLine                         // treat ^ and $ as only matching at beginning and end of text
    63  	NonGreedy                       // make repetition operators default to non-greedy
    64  	PerlX                           // allow Perl extensions
    65  	UnicodeGroups                   // allow \p{Han}, \P{Han} for Unicode group and negation
    66  	WasDollar                       // regexp OpEndText was $, not \z
    67  	Simple                          // regexp contains no counted repetition
    68  
    69  	MatchNL = ClassNL | DotNL
    70  
    71  	Perl        = ClassNL | OneLine | PerlX | UnicodeGroups // as close to Perl as possible
    72  	POSIX Flags = 0                                         // POSIX syntax
    73  )
    74  
    75  // Pseudo-ops for parsing stack.
    76  const (
    77  	opLeftParen = opPseudo + iota
    78  	opVerticalBar
    79  )
    80  
    81  // maxHeight is the maximum height of a regexp parse tree.
    82  // It is somewhat arbitrarily chosen, but the idea is to be large enough
    83  // that no one will actually hit in real use but at the same time small enough
    84  // that recursion on the Regexp tree will not hit the 1GB Go stack limit.
    85  // The maximum amount of stack for a single recursive frame is probably
    86  // closer to 1kB, so this could potentially be raised, but it seems unlikely
    87  // that people have regexps nested even this deeply.
    88  // We ran a test on Google's C++ code base and turned up only
    89  // a single use case with depth > 100; it had depth 128.
    90  // Using depth 1000 should be plenty of margin.
    91  // As an optimization, we don't even bother calculating heights
    92  // until we've allocated at least maxHeight Regexp structures.
    93  const maxHeight = 1000
    94  
    95  // maxSize is the maximum size of a compiled regexp in Insts.
    96  // It too is somewhat arbitrarily chosen, but the idea is to be large enough
    97  // to allow significant regexps while at the same time small enough that
    98  // the compiled form will not take up too much memory.
    99  // 128 MB is enough for a 3.3 million Inst structures, which roughly
   100  // corresponds to a 3.3 MB regexp.
   101  const (
   102  	maxSize  = 128 << 20 / instSize
   103  	instSize = 5 * 8 // byte, 2 uint32, slice is 5 64-bit words
   104  )
   105  
   106  // maxRunes is the maximum number of runes allowed in a regexp tree
   107  // counting the runes in all the nodes.
   108  // Ignoring character classes p.numRunes is always less than the length of the regexp.
   109  // Character classes can make it much larger: each \pL adds 1292 runes.
   110  // 128 MB is enough for 32M runes, which is over 26k \pL instances.
   111  // Note that repetitions do not make copies of the rune slices,
   112  // so \pL{1000} is only one rune slice, not 1000.
   113  // We could keep a cache of character classes we've seen,
   114  // so that all the \pL we see use the same rune list,
   115  // but that doesn't remove the problem entirely:
   116  // consider something like [\pL01234][\pL01235][\pL01236]...[\pL^&*()].
   117  // And because the Rune slice is exposed directly in the Regexp,
   118  // there is not an opportunity to change the representation to allow
   119  // partial sharing between different character classes.
   120  // So the limit is the best we can do.
   121  const (
   122  	maxRunes = 128 << 20 / runeSize
   123  	runeSize = 4 // rune is int32
   124  )
   125  
   126  type parser struct {
   127  	flags       Flags     // parse mode flags
   128  	stack       []*Regexp // stack of parsed expressions
   129  	free        *Regexp
   130  	numCap      int // number of capturing groups seen
   131  	wholeRegexp string
   132  	tmpClass    []rune            // temporary char class work space
   133  	numRegexp   int               // number of regexps allocated
   134  	numRunes    int               // number of runes in char classes
   135  	repeats     int64             // product of all repetitions seen
   136  	height      map[*Regexp]int   // regexp height, for height limit check
   137  	size        map[*Regexp]int64 // regexp compiled size, for size limit check
   138  }
   139  
   140  func (p *parser) newRegexp(op Op) *Regexp {
   141  	re := p.free
   142  	if re != nil {
   143  		p.free = re.Sub0[0]
   144  		*re = Regexp{}
   145  	} else {
   146  		re = new(Regexp)
   147  		p.numRegexp++
   148  	}
   149  	re.Op = op
   150  	return re
   151  }
   152  
   153  func (p *parser) reuse(re *Regexp) {
   154  	if p.height != nil {
   155  		delete(p.height, re)
   156  	}
   157  	re.Sub0[0] = p.free
   158  	p.free = re
   159  }
   160  
   161  func (p *parser) checkLimits(re *Regexp) {
   162  	if p.numRunes > maxRunes {
   163  		panic(ErrLarge)
   164  	}
   165  	p.checkSize(re)
   166  	p.checkHeight(re)
   167  }
   168  
   169  func (p *parser) checkSize(re *Regexp) {
   170  	if p.size == nil {
   171  		// We haven't started tracking size yet.
   172  		// Do a relatively cheap check to see if we need to start.
   173  		// Maintain the product of all the repeats we've seen
   174  		// and don't track if the total number of regexp nodes
   175  		// we've seen times the repeat product is in budget.
   176  		if p.repeats == 0 {
   177  			p.repeats = 1
   178  		}
   179  		if re.Op == OpRepeat {
   180  			n := re.Max
   181  			if n == -1 {
   182  				n = re.Min
   183  			}
   184  			if n <= 0 {
   185  				n = 1
   186  			}
   187  			if int64(n) > maxSize/p.repeats {
   188  				p.repeats = maxSize
   189  			} else {
   190  				p.repeats *= int64(n)
   191  			}
   192  		}
   193  		if int64(p.numRegexp) < maxSize/p.repeats {
   194  			return
   195  		}
   196  
   197  		// We need to start tracking size.
   198  		// Make the map and belatedly populate it
   199  		// with info about everything we've constructed so far.
   200  		p.size = make(map[*Regexp]int64)
   201  		for _, re := range p.stack {
   202  			p.checkSize(re)
   203  		}
   204  	}
   205  
   206  	if p.calcSize(re, true) > maxSize {
   207  		panic(ErrLarge)
   208  	}
   209  }
   210  
   211  func (p *parser) calcSize(re *Regexp, force bool) int64 {
   212  	if !force {
   213  		if size, ok := p.size[re]; ok {
   214  			return size
   215  		}
   216  	}
   217  
   218  	var size int64
   219  	switch re.Op {
   220  	case OpLiteral:
   221  		size = int64(len(re.Rune))
   222  	case OpCapture, OpStar:
   223  		// star can be 1+ or 2+; assume 2 pessimistically
   224  		size = 2 + p.calcSize(re.Sub[0], false)
   225  	case OpPlus, OpQuest:
   226  		size = 1 + p.calcSize(re.Sub[0], false)
   227  	case OpConcat:
   228  		for _, sub := range re.Sub {
   229  			size += p.calcSize(sub, false)
   230  		}
   231  	case OpAlternate:
   232  		for _, sub := range re.Sub {
   233  			size += p.calcSize(sub, false)
   234  		}
   235  		if len(re.Sub) > 1 {
   236  			size += int64(len(re.Sub)) - 1
   237  		}
   238  	case OpRepeat:
   239  		sub := p.calcSize(re.Sub[0], false)
   240  		if re.Max == -1 {
   241  			if re.Min == 0 {
   242  				size = 2 + sub // x*
   243  			} else {
   244  				size = 1 + int64(re.Min)*sub // xxx+
   245  			}
   246  			break
   247  		}
   248  		// x{2,5} = xx(x(x(x)?)?)?
   249  		size = int64(re.Max)*sub + int64(re.Max-re.Min)
   250  	}
   251  
   252  	if size < 1 {
   253  		size = 1
   254  	}
   255  	p.size[re] = size
   256  	return size
   257  }
   258  
   259  func (p *parser) checkHeight(re *Regexp) {
   260  	if p.numRegexp < maxHeight {
   261  		return
   262  	}
   263  	if p.height == nil {
   264  		p.height = make(map[*Regexp]int)
   265  		for _, re := range p.stack {
   266  			p.checkHeight(re)
   267  		}
   268  	}
   269  	if p.calcHeight(re, true) > maxHeight {
   270  		panic(ErrNestingDepth)
   271  	}
   272  }
   273  
   274  func (p *parser) calcHeight(re *Regexp, force bool) int {
   275  	if !force {
   276  		if h, ok := p.height[re]; ok {
   277  			return h
   278  		}
   279  	}
   280  	h := 1
   281  	for _, sub := range re.Sub {
   282  		hsub := p.calcHeight(sub, false)
   283  		if h < 1+hsub {
   284  			h = 1 + hsub
   285  		}
   286  	}
   287  	p.height[re] = h
   288  	return h
   289  }
   290  
   291  // Parse stack manipulation.
   292  
   293  // push pushes the regexp re onto the parse stack and returns the regexp.
   294  func (p *parser) push(re *Regexp) *Regexp {
   295  	p.numRunes += len(re.Rune)
   296  	if re.Op == OpCharClass && len(re.Rune) == 2 && re.Rune[0] == re.Rune[1] {
   297  		// Single rune.
   298  		if p.maybeConcat(re.Rune[0], p.flags&^FoldCase) {
   299  			return nil
   300  		}
   301  		re.Op = OpLiteral
   302  		re.Rune = re.Rune[:1]
   303  		re.Flags = p.flags &^ FoldCase
   304  	} else if re.Op == OpCharClass && len(re.Rune) == 4 &&
   305  		re.Rune[0] == re.Rune[1] && re.Rune[2] == re.Rune[3] &&
   306  		unicode.SimpleFold(re.Rune[0]) == re.Rune[2] &&
   307  		unicode.SimpleFold(re.Rune[2]) == re.Rune[0] ||
   308  		re.Op == OpCharClass && len(re.Rune) == 2 &&
   309  			re.Rune[0]+1 == re.Rune[1] &&
   310  			unicode.SimpleFold(re.Rune[0]) == re.Rune[1] &&
   311  			unicode.SimpleFold(re.Rune[1]) == re.Rune[0] {
   312  		// Case-insensitive rune like [Aa] or [Δδ].
   313  		if p.maybeConcat(re.Rune[0], p.flags|FoldCase) {
   314  			return nil
   315  		}
   316  
   317  		// Rewrite as (case-insensitive) literal.
   318  		re.Op = OpLiteral
   319  		re.Rune = re.Rune[:1]
   320  		re.Flags = p.flags | FoldCase
   321  	} else {
   322  		// Incremental concatenation.
   323  		p.maybeConcat(-1, 0)
   324  	}
   325  
   326  	p.stack = append(p.stack, re)
   327  	p.checkLimits(re)
   328  	return re
   329  }
   330  
   331  // maybeConcat implements incremental concatenation
   332  // of literal runes into string nodes. The parser calls this
   333  // before each push, so only the top fragment of the stack
   334  // might need processing. Since this is called before a push,
   335  // the topmost literal is no longer subject to operators like *
   336  // (Otherwise ab* would turn into (ab)*.)
   337  // If r >= 0 and there's a node left over, maybeConcat uses it
   338  // to push r with the given flags.
   339  // maybeConcat reports whether r was pushed.
   340  func (p *parser) maybeConcat(r rune, flags Flags) bool {
   341  	n := len(p.stack)
   342  	if n < 2 {
   343  		return false
   344  	}
   345  
   346  	re1 := p.stack[n-1]
   347  	re2 := p.stack[n-2]
   348  	if re1.Op != OpLiteral || re2.Op != OpLiteral || re1.Flags&FoldCase != re2.Flags&FoldCase {
   349  		return false
   350  	}
   351  
   352  	// Push re1 into re2.
   353  	re2.Rune = append(re2.Rune, re1.Rune...)
   354  
   355  	// Reuse re1 if possible.
   356  	if r >= 0 {
   357  		re1.Rune = re1.Rune0[:1]
   358  		re1.Rune[0] = r
   359  		re1.Flags = flags
   360  		return true
   361  	}
   362  
   363  	p.stack = p.stack[:n-1]
   364  	p.reuse(re1)
   365  	return false // did not push r
   366  }
   367  
   368  // literal pushes a literal regexp for the rune r on the stack.
   369  func (p *parser) literal(r rune) {
   370  	re := p.newRegexp(OpLiteral)
   371  	re.Flags = p.flags
   372  	if p.flags&FoldCase != 0 {
   373  		r = minFoldRune(r)
   374  	}
   375  	re.Rune0[0] = r
   376  	re.Rune = re.Rune0[:1]
   377  	p.push(re)
   378  }
   379  
   380  // minFoldRune returns the minimum rune fold-equivalent to r.
   381  func minFoldRune(r rune) rune {
   382  	if r < minFold || r > maxFold {
   383  		return r
   384  	}
   385  	m := r
   386  	r0 := r
   387  	for r = unicode.SimpleFold(r); r != r0; r = unicode.SimpleFold(r) {
   388  		m = min(m, r)
   389  	}
   390  	return m
   391  }
   392  
   393  // op pushes a regexp with the given op onto the stack
   394  // and returns that regexp.
   395  func (p *parser) op(op Op) *Regexp {
   396  	re := p.newRegexp(op)
   397  	re.Flags = p.flags
   398  	return p.push(re)
   399  }
   400  
   401  // repeat replaces the top stack element with itself repeated according to op, min, max.
   402  // before is the regexp suffix starting at the repetition operator.
   403  // after is the regexp suffix following after the repetition operator.
   404  // repeat returns an updated 'after' and an error, if any.
   405  func (p *parser) repeat(op Op, min, max int, before, after, lastRepeat string) (string, error) {
   406  	flags := p.flags
   407  	if p.flags&PerlX != 0 {
   408  		if len(after) > 0 && after[0] == '?' {
   409  			after = after[1:]
   410  			flags ^= NonGreedy
   411  		}
   412  		if lastRepeat != "" {
   413  			// In Perl it is not allowed to stack repetition operators:
   414  			// a** is a syntax error, not a doubled star, and a++ means
   415  			// something else entirely, which we don't support!
   416  			return "", &Error{ErrInvalidRepeatOp, lastRepeat[:len(lastRepeat)-len(after)]}
   417  		}
   418  	}
   419  	n := len(p.stack)
   420  	if n == 0 {
   421  		return "", &Error{ErrMissingRepeatArgument, before[:len(before)-len(after)]}
   422  	}
   423  	sub := p.stack[n-1]
   424  	if sub.Op >= opPseudo {
   425  		return "", &Error{ErrMissingRepeatArgument, before[:len(before)-len(after)]}
   426  	}
   427  
   428  	re := p.newRegexp(op)
   429  	re.Min = min
   430  	re.Max = max
   431  	re.Flags = flags
   432  	re.Sub = re.Sub0[:1]
   433  	re.Sub[0] = sub
   434  	p.stack[n-1] = re
   435  	p.checkLimits(re)
   436  
   437  	if op == OpRepeat && (min >= 2 || max >= 2) && !repeatIsValid(re, 1000) {
   438  		return "", &Error{ErrInvalidRepeatSize, before[:len(before)-len(after)]}
   439  	}
   440  
   441  	return after, nil
   442  }
   443  
   444  // repeatIsValid reports whether the repetition re is valid.
   445  // Valid means that the combination of the top-level repetition
   446  // and any inner repetitions does not exceed n copies of the
   447  // innermost thing.
   448  // This function rewalks the regexp tree and is called for every repetition,
   449  // so we have to worry about inducing quadratic behavior in the parser.
   450  // We avoid this by only calling repeatIsValid when min or max >= 2.
   451  // In that case the depth of any >= 2 nesting can only get to 9 without
   452  // triggering a parse error, so each subtree can only be rewalked 9 times.
   453  func repeatIsValid(re *Regexp, n int) bool {
   454  	if re.Op == OpRepeat {
   455  		m := re.Max
   456  		if m == 0 {
   457  			return true
   458  		}
   459  		if m < 0 {
   460  			m = re.Min
   461  		}
   462  		if m > n {
   463  			return false
   464  		}
   465  		if m > 0 {
   466  			n /= m
   467  		}
   468  	}
   469  	for _, sub := range re.Sub {
   470  		if !repeatIsValid(sub, n) {
   471  			return false
   472  		}
   473  	}
   474  	return true
   475  }
   476  
   477  // concat replaces the top of the stack (above the topmost '|' or '(') with its concatenation.
   478  func (p *parser) concat() *Regexp {
   479  	p.maybeConcat(-1, 0)
   480  
   481  	// Scan down to find pseudo-operator | or (.
   482  	i := len(p.stack)
   483  	for i > 0 && p.stack[i-1].Op < opPseudo {
   484  		i--
   485  	}
   486  	subs := p.stack[i:]
   487  	p.stack = p.stack[:i]
   488  
   489  	// Empty concatenation is special case.
   490  	if len(subs) == 0 {
   491  		return p.push(p.newRegexp(OpEmptyMatch))
   492  	}
   493  
   494  	return p.push(p.collapse(subs, OpConcat))
   495  }
   496  
   497  // alternate replaces the top of the stack (above the topmost '(') with its alternation.
   498  func (p *parser) alternate() *Regexp {
   499  	// Scan down to find pseudo-operator (.
   500  	// There are no | above (.
   501  	i := len(p.stack)
   502  	for i > 0 && p.stack[i-1].Op < opPseudo {
   503  		i--
   504  	}
   505  	subs := p.stack[i:]
   506  	p.stack = p.stack[:i]
   507  
   508  	// Make sure top class is clean.
   509  	// All the others already are (see swapVerticalBar).
   510  	if len(subs) > 0 {
   511  		cleanAlt(subs[len(subs)-1])
   512  	}
   513  
   514  	// Empty alternate is special case
   515  	// (shouldn't happen but easy to handle).
   516  	if len(subs) == 0 {
   517  		return p.push(p.newRegexp(OpNoMatch))
   518  	}
   519  
   520  	return p.push(p.collapse(subs, OpAlternate))
   521  }
   522  
   523  // cleanAlt cleans re for eventual inclusion in an alternation.
   524  func cleanAlt(re *Regexp) {
   525  	switch re.Op {
   526  	case OpCharClass:
   527  		re.Rune = cleanClass(&re.Rune)
   528  		if len(re.Rune) == 2 && re.Rune[0] == 0 && re.Rune[1] == unicode.MaxRune {
   529  			re.Rune = nil
   530  			re.Op = OpAnyChar
   531  			return
   532  		}
   533  		if len(re.Rune) == 4 && re.Rune[0] == 0 && re.Rune[1] == '\n'-1 && re.Rune[2] == '\n'+1 && re.Rune[3] == unicode.MaxRune {
   534  			re.Rune = nil
   535  			re.Op = OpAnyCharNotNL
   536  			return
   537  		}
   538  		if cap(re.Rune)-len(re.Rune) > 100 {
   539  			// re.Rune will not grow any more.
   540  			// Make a copy or inline to reclaim storage.
   541  			re.Rune = append(re.Rune0[:0], re.Rune...)
   542  		}
   543  	}
   544  }
   545  
   546  // collapse returns the result of applying op to sub.
   547  // If sub contains op nodes, they all get hoisted up
   548  // so that there is never a concat of a concat or an
   549  // alternate of an alternate.
   550  func (p *parser) collapse(subs []*Regexp, op Op) *Regexp {
   551  	if len(subs) == 1 {
   552  		return subs[0]
   553  	}
   554  	re := p.newRegexp(op)
   555  	re.Sub = re.Sub0[:0]
   556  	for _, sub := range subs {
   557  		if sub.Op == op {
   558  			re.Sub = append(re.Sub, sub.Sub...)
   559  			p.reuse(sub)
   560  		} else {
   561  			re.Sub = append(re.Sub, sub)
   562  		}
   563  	}
   564  	if op == OpAlternate {
   565  		re.Sub = p.factor(re.Sub)
   566  		if len(re.Sub) == 1 {
   567  			old := re
   568  			re = re.Sub[0]
   569  			p.reuse(old)
   570  		}
   571  	}
   572  	return re
   573  }
   574  
   575  // factor factors common prefixes from the alternation list sub.
   576  // It returns a replacement list that reuses the same storage and
   577  // frees (passes to p.reuse) any removed *Regexps.
   578  //
   579  // For example,
   580  //
   581  //	ABC|ABD|AEF|BCX|BCY
   582  //
   583  // simplifies by literal prefix extraction to
   584  //
   585  //	A(B(C|D)|EF)|BC(X|Y)
   586  //
   587  // which simplifies by character class introduction to
   588  //
   589  //	A(B[CD]|EF)|BC[XY]
   590  func (p *parser) factor(sub []*Regexp) []*Regexp {
   591  	if len(sub) < 2 {
   592  		return sub
   593  	}
   594  
   595  	// Round 1: Factor out common literal prefixes.
   596  	var str []rune
   597  	var strflags Flags
   598  	start := 0
   599  	out := sub[:0]
   600  	for i := 0; i <= len(sub); i++ {
   601  		// Invariant: the Regexps that were in sub[0:start] have been
   602  		// used or marked for reuse, and the slice space has been reused
   603  		// for out (len(out) <= start).
   604  		//
   605  		// Invariant: sub[start:i] consists of regexps that all begin
   606  		// with str as modified by strflags.
   607  		var istr []rune
   608  		var iflags Flags
   609  		if i < len(sub) {
   610  			istr, iflags = p.leadingString(sub[i])
   611  			if iflags == strflags {
   612  				same := 0
   613  				for same < len(str) && same < len(istr) && str[same] == istr[same] {
   614  					same++
   615  				}
   616  				if same > 0 {
   617  					// Matches at least one rune in current range.
   618  					// Keep going around.
   619  					str = str[:same]
   620  					continue
   621  				}
   622  			}
   623  		}
   624  
   625  		// Found end of a run with common leading literal string:
   626  		// sub[start:i] all begin with str[0:len(str)], but sub[i]
   627  		// does not even begin with str[0].
   628  		//
   629  		// Factor out common string and append factored expression to out.
   630  		if i == start {
   631  			// Nothing to do - run of length 0.
   632  		} else if i == start+1 {
   633  			// Just one: don't bother factoring.
   634  			out = append(out, sub[start])
   635  		} else {
   636  			// Construct factored form: prefix(suffix1|suffix2|...)
   637  			prefix := p.newRegexp(OpLiteral)
   638  			prefix.Flags = strflags
   639  			prefix.Rune = append(prefix.Rune[:0], str...)
   640  
   641  			for j := start; j < i; j++ {
   642  				sub[j] = p.removeLeadingString(sub[j], len(str))
   643  				p.checkLimits(sub[j])
   644  			}
   645  			suffix := p.collapse(sub[start:i], OpAlternate) // recurse
   646  
   647  			re := p.newRegexp(OpConcat)
   648  			re.Sub = append(re.Sub[:0], prefix, suffix)
   649  			out = append(out, re)
   650  		}
   651  
   652  		// Prepare for next iteration.
   653  		start = i
   654  		str = istr
   655  		strflags = iflags
   656  	}
   657  	sub = out
   658  
   659  	// Round 2: Factor out common simple prefixes,
   660  	// just the first piece of each concatenation.
   661  	// This will be good enough a lot of the time.
   662  	//
   663  	// Complex subexpressions (e.g. involving quantifiers)
   664  	// are not safe to factor because that collapses their
   665  	// distinct paths through the automaton, which affects
   666  	// correctness in some cases.
   667  	start = 0
   668  	out = sub[:0]
   669  	var first *Regexp
   670  	for i := 0; i <= len(sub); i++ {
   671  		// Invariant: the Regexps that were in sub[0:start] have been
   672  		// used or marked for reuse, and the slice space has been reused
   673  		// for out (len(out) <= start).
   674  		//
   675  		// Invariant: sub[start:i] consists of regexps that all begin with ifirst.
   676  		var ifirst *Regexp
   677  		if i < len(sub) {
   678  			ifirst = p.leadingRegexp(sub[i])
   679  			if first != nil && first.Equal(ifirst) &&
   680  				// first must be a character class OR a fixed repeat of a character class.
   681  				(isCharClass(first) || (first.Op == OpRepeat && first.Min == first.Max && isCharClass(first.Sub[0]))) {
   682  				continue
   683  			}
   684  		}
   685  
   686  		// Found end of a run with common leading regexp:
   687  		// sub[start:i] all begin with first but sub[i] does not.
   688  		//
   689  		// Factor out common regexp and append factored expression to out.
   690  		if i == start {
   691  			// Nothing to do - run of length 0.
   692  		} else if i == start+1 {
   693  			// Just one: don't bother factoring.
   694  			out = append(out, sub[start])
   695  		} else {
   696  			// Construct factored form: prefix(suffix1|suffix2|...)
   697  			prefix := first
   698  			for j := start; j < i; j++ {
   699  				reuse := j != start // prefix came from sub[start]
   700  				sub[j] = p.removeLeadingRegexp(sub[j], reuse)
   701  				p.checkLimits(sub[j])
   702  			}
   703  			suffix := p.collapse(sub[start:i], OpAlternate) // recurse
   704  
   705  			re := p.newRegexp(OpConcat)
   706  			re.Sub = append(re.Sub[:0], prefix, suffix)
   707  			out = append(out, re)
   708  		}
   709  
   710  		// Prepare for next iteration.
   711  		start = i
   712  		first = ifirst
   713  	}
   714  	sub = out
   715  
   716  	// Round 3: Collapse runs of single literals into character classes.
   717  	start = 0
   718  	out = sub[:0]
   719  	for i := 0; i <= len(sub); i++ {
   720  		// Invariant: the Regexps that were in sub[0:start] have been
   721  		// used or marked for reuse, and the slice space has been reused
   722  		// for out (len(out) <= start).
   723  		//
   724  		// Invariant: sub[start:i] consists of regexps that are either
   725  		// literal runes or character classes.
   726  		if i < len(sub) && isCharClass(sub[i]) {
   727  			continue
   728  		}
   729  
   730  		// sub[i] is not a char or char class;
   731  		// emit char class for sub[start:i]...
   732  		if i == start {
   733  			// Nothing to do - run of length 0.
   734  		} else if i == start+1 {
   735  			out = append(out, sub[start])
   736  		} else {
   737  			// Make new char class.
   738  			// Start with most complex regexp in sub[start].
   739  			max := start
   740  			for j := start + 1; j < i; j++ {
   741  				if sub[max].Op < sub[j].Op || sub[max].Op == sub[j].Op && len(sub[max].Rune) < len(sub[j].Rune) {
   742  					max = j
   743  				}
   744  			}
   745  			sub[start], sub[max] = sub[max], sub[start]
   746  
   747  			for j := start + 1; j < i; j++ {
   748  				mergeCharClass(sub[start], sub[j])
   749  				p.reuse(sub[j])
   750  			}
   751  			cleanAlt(sub[start])
   752  			out = append(out, sub[start])
   753  		}
   754  
   755  		// ... and then emit sub[i].
   756  		if i < len(sub) {
   757  			out = append(out, sub[i])
   758  		}
   759  		start = i + 1
   760  	}
   761  	sub = out
   762  
   763  	// Round 4: Collapse runs of empty matches into a single empty match.
   764  	start = 0
   765  	out = sub[:0]
   766  	for i := range sub {
   767  		if i+1 < len(sub) && sub[i].Op == OpEmptyMatch && sub[i+1].Op == OpEmptyMatch {
   768  			continue
   769  		}
   770  		out = append(out, sub[i])
   771  	}
   772  	sub = out
   773  
   774  	return sub
   775  }
   776  
   777  // leadingString returns the leading literal string that re begins with.
   778  // The string refers to storage in re or its children.
   779  func (p *parser) leadingString(re *Regexp) ([]rune, Flags) {
   780  	if re.Op == OpConcat && len(re.Sub) > 0 {
   781  		re = re.Sub[0]
   782  	}
   783  	if re.Op != OpLiteral {
   784  		return nil, 0
   785  	}
   786  	return re.Rune, re.Flags & FoldCase
   787  }
   788  
   789  // removeLeadingString removes the first n leading runes
   790  // from the beginning of re. It returns the replacement for re.
   791  func (p *parser) removeLeadingString(re *Regexp, n int) *Regexp {
   792  	if re.Op == OpConcat && len(re.Sub) > 0 {
   793  		// Removing a leading string in a concatenation
   794  		// might simplify the concatenation.
   795  		sub := re.Sub[0]
   796  		sub = p.removeLeadingString(sub, n)
   797  		re.Sub[0] = sub
   798  		if sub.Op == OpEmptyMatch {
   799  			p.reuse(sub)
   800  			switch len(re.Sub) {
   801  			case 0, 1:
   802  				// Impossible but handle.
   803  				re.Op = OpEmptyMatch
   804  				re.Sub = nil
   805  			case 2:
   806  				old := re
   807  				re = re.Sub[1]
   808  				p.reuse(old)
   809  			default:
   810  				copy(re.Sub, re.Sub[1:])
   811  				re.Sub = re.Sub[:len(re.Sub)-1]
   812  			}
   813  		}
   814  		return re
   815  	}
   816  
   817  	if re.Op == OpLiteral {
   818  		re.Rune = re.Rune[:copy(re.Rune, re.Rune[n:])]
   819  		if len(re.Rune) == 0 {
   820  			re.Op = OpEmptyMatch
   821  		}
   822  	}
   823  	return re
   824  }
   825  
   826  // leadingRegexp returns the leading regexp that re begins with.
   827  // The regexp refers to storage in re or its children.
   828  func (p *parser) leadingRegexp(re *Regexp) *Regexp {
   829  	if re.Op == OpEmptyMatch {
   830  		return nil
   831  	}
   832  	if re.Op == OpConcat && len(re.Sub) > 0 {
   833  		sub := re.Sub[0]
   834  		if sub.Op == OpEmptyMatch {
   835  			return nil
   836  		}
   837  		return sub
   838  	}
   839  	return re
   840  }
   841  
   842  // removeLeadingRegexp removes the leading regexp in re.
   843  // It returns the replacement for re.
   844  // If reuse is true, it passes the removed regexp (if no longer needed) to p.reuse.
   845  func (p *parser) removeLeadingRegexp(re *Regexp, reuse bool) *Regexp {
   846  	if re.Op == OpConcat && len(re.Sub) > 0 {
   847  		if reuse {
   848  			p.reuse(re.Sub[0])
   849  		}
   850  		re.Sub = re.Sub[:copy(re.Sub, re.Sub[1:])]
   851  		switch len(re.Sub) {
   852  		case 0:
   853  			re.Op = OpEmptyMatch
   854  			re.Sub = nil
   855  		case 1:
   856  			old := re
   857  			re = re.Sub[0]
   858  			p.reuse(old)
   859  		}
   860  		return re
   861  	}
   862  	if reuse {
   863  		p.reuse(re)
   864  	}
   865  	return p.newRegexp(OpEmptyMatch)
   866  }
   867  
   868  func literalRegexp(s string, flags Flags) *Regexp {
   869  	re := &Regexp{Op: OpLiteral}
   870  	re.Flags = flags
   871  	re.Rune = re.Rune0[:0] // use local storage for small strings
   872  	for _, c := range s {
   873  		if len(re.Rune) >= cap(re.Rune) {
   874  			// string is too long to fit in Rune0.  let Go handle it
   875  			re.Rune = []rune(s)
   876  			break
   877  		}
   878  		re.Rune = append(re.Rune, c)
   879  	}
   880  	return re
   881  }
   882  
   883  // Parsing.
   884  
   885  // Parse parses a regular expression string s, controlled by the specified
   886  // Flags, and returns a regular expression parse tree. The syntax is
   887  // described in the top-level comment.
   888  func Parse(s string, flags Flags) (*Regexp, error) {
   889  	return parse(s, flags)
   890  }
   891  
   892  func parse(s string, flags Flags) (_ *Regexp, err error) {
   893  	defer func() {
   894  		switch r := recover(); r {
   895  		default:
   896  			panic(r)
   897  		case nil:
   898  			// ok
   899  		case ErrLarge: // too big
   900  			err = &Error{Code: ErrLarge, Expr: s}
   901  		case ErrNestingDepth:
   902  			err = &Error{Code: ErrNestingDepth, Expr: s}
   903  		}
   904  	}()
   905  
   906  	if flags&Literal != 0 {
   907  		// Trivial parser for literal string.
   908  		if err := checkUTF8(s); err != nil {
   909  			return nil, err
   910  		}
   911  		return literalRegexp(s, flags), nil
   912  	}
   913  
   914  	// Otherwise, must do real work.
   915  	var (
   916  		p          parser
   917  		c          rune
   918  		op         Op
   919  		lastRepeat string
   920  	)
   921  	p.flags = flags
   922  	p.wholeRegexp = s
   923  	t := s
   924  	for t != "" {
   925  		repeat := ""
   926  	BigSwitch:
   927  		switch t[0] {
   928  		default:
   929  			if c, t, err = nextRune(t); err != nil {
   930  				return nil, err
   931  			}
   932  			p.literal(c)
   933  
   934  		case '(':
   935  			if p.flags&PerlX != 0 && len(t) >= 2 && t[1] == '?' {
   936  				// Flag changes and non-capturing groups.
   937  				if t, err = p.parsePerlFlags(t); err != nil {
   938  					return nil, err
   939  				}
   940  				break
   941  			}
   942  			p.numCap++
   943  			p.op(opLeftParen).Cap = p.numCap
   944  			t = t[1:]
   945  		case '|':
   946  			if err = p.parseVerticalBar(); err != nil {
   947  				return nil, err
   948  			}
   949  			t = t[1:]
   950  		case ')':
   951  			if err = p.parseRightParen(); err != nil {
   952  				return nil, err
   953  			}
   954  			t = t[1:]
   955  		case '^':
   956  			if p.flags&OneLine != 0 {
   957  				p.op(OpBeginText)
   958  			} else {
   959  				p.op(OpBeginLine)
   960  			}
   961  			t = t[1:]
   962  		case '$':
   963  			if p.flags&OneLine != 0 {
   964  				p.op(OpEndText).Flags |= WasDollar
   965  			} else {
   966  				p.op(OpEndLine)
   967  			}
   968  			t = t[1:]
   969  		case '.':
   970  			if p.flags&DotNL != 0 {
   971  				p.op(OpAnyChar)
   972  			} else {
   973  				p.op(OpAnyCharNotNL)
   974  			}
   975  			t = t[1:]
   976  		case '[':
   977  			if t, err = p.parseClass(t); err != nil {
   978  				return nil, err
   979  			}
   980  		case '*', '+', '?':
   981  			before := t
   982  			switch t[0] {
   983  			case '*':
   984  				op = OpStar
   985  			case '+':
   986  				op = OpPlus
   987  			case '?':
   988  				op = OpQuest
   989  			}
   990  			after := t[1:]
   991  			if after, err = p.repeat(op, 0, 0, before, after, lastRepeat); err != nil {
   992  				return nil, err
   993  			}
   994  			repeat = before
   995  			t = after
   996  		case '{':
   997  			op = OpRepeat
   998  			before := t
   999  			min, max, after, ok := p.parseRepeat(t)
  1000  			if !ok {
  1001  				// If the repeat cannot be parsed, { is a literal.
  1002  				p.literal('{')
  1003  				t = t[1:]
  1004  				break
  1005  			}
  1006  			if min < 0 || min > 1000 || max > 1000 || max >= 0 && min > max {
  1007  				// Numbers were too big, or max is present and min > max.
  1008  				return nil, &Error{ErrInvalidRepeatSize, before[:len(before)-len(after)]}
  1009  			}
  1010  			if after, err = p.repeat(op, min, max, before, after, lastRepeat); err != nil {
  1011  				return nil, err
  1012  			}
  1013  			repeat = before
  1014  			t = after
  1015  		case '\\':
  1016  			if p.flags&PerlX != 0 && len(t) >= 2 {
  1017  				switch t[1] {
  1018  				case 'A':
  1019  					p.op(OpBeginText)
  1020  					t = t[2:]
  1021  					break BigSwitch
  1022  				case 'b':
  1023  					p.op(OpWordBoundary)
  1024  					t = t[2:]
  1025  					break BigSwitch
  1026  				case 'B':
  1027  					p.op(OpNoWordBoundary)
  1028  					t = t[2:]
  1029  					break BigSwitch
  1030  				case 'C':
  1031  					// any byte; not supported
  1032  					return nil, &Error{ErrInvalidEscape, t[:2]}
  1033  				case 'Q':
  1034  					// \Q ... \E: the ... is always literals
  1035  					var lit string
  1036  					lit, t, _ = strings.Cut(t[2:], `\E`)
  1037  					for lit != "" {
  1038  						c, rest, err := nextRune(lit)
  1039  						if err != nil {
  1040  							return nil, err
  1041  						}
  1042  						p.literal(c)
  1043  						lit = rest
  1044  					}
  1045  					break BigSwitch
  1046  				case 'z':
  1047  					p.op(OpEndText)
  1048  					t = t[2:]
  1049  					break BigSwitch
  1050  				}
  1051  			}
  1052  
  1053  			re := p.newRegexp(OpCharClass)
  1054  			re.Flags = p.flags
  1055  
  1056  			// Look for Unicode character group like \p{Han}
  1057  			if len(t) >= 2 && (t[1] == 'p' || t[1] == 'P') {
  1058  				r, rest, err := p.parseUnicodeClass(t, re.Rune0[:0])
  1059  				if err != nil {
  1060  					return nil, err
  1061  				}
  1062  				if r != nil {
  1063  					re.Rune = r
  1064  					t = rest
  1065  					p.push(re)
  1066  					break BigSwitch
  1067  				}
  1068  			}
  1069  
  1070  			// Perl character class escape.
  1071  			if r, rest := p.parsePerlClassEscape(t, re.Rune0[:0]); r != nil {
  1072  				re.Rune = r
  1073  				t = rest
  1074  				p.push(re)
  1075  				break BigSwitch
  1076  			}
  1077  			p.reuse(re)
  1078  
  1079  			// Ordinary single-character escape.
  1080  			if c, t, err = p.parseEscape(t); err != nil {
  1081  				return nil, err
  1082  			}
  1083  			p.literal(c)
  1084  		}
  1085  		lastRepeat = repeat
  1086  	}
  1087  
  1088  	p.concat()
  1089  	if p.swapVerticalBar() {
  1090  		// pop vertical bar
  1091  		p.stack = p.stack[:len(p.stack)-1]
  1092  	}
  1093  	p.alternate()
  1094  
  1095  	n := len(p.stack)
  1096  	if n != 1 {
  1097  		return nil, &Error{ErrMissingParen, s}
  1098  	}
  1099  	return p.stack[0], nil
  1100  }
  1101  
  1102  // parseRepeat parses {min} (max=min) or {min,} (max=-1) or {min,max}.
  1103  // If s is not of that form, it returns ok == false.
  1104  // If s has the right form but the values are too big, it returns min == -1, ok == true.
  1105  func (p *parser) parseRepeat(s string) (min, max int, rest string, ok bool) {
  1106  	if s == "" || s[0] != '{' {
  1107  		return
  1108  	}
  1109  	s = s[1:]
  1110  	var ok1 bool
  1111  	if min, s, ok1 = p.parseInt(s); !ok1 {
  1112  		return
  1113  	}
  1114  	if s == "" {
  1115  		return
  1116  	}
  1117  	if s[0] != ',' {
  1118  		max = min
  1119  	} else {
  1120  		s = s[1:]
  1121  		if s == "" {
  1122  			return
  1123  		}
  1124  		if s[0] == '}' {
  1125  			max = -1
  1126  		} else if max, s, ok1 = p.parseInt(s); !ok1 {
  1127  			return
  1128  		} else if max < 0 {
  1129  			// parseInt found too big a number
  1130  			min = -1
  1131  		}
  1132  	}
  1133  	if s == "" || s[0] != '}' {
  1134  		return
  1135  	}
  1136  	rest = s[1:]
  1137  	ok = true
  1138  	return
  1139  }
  1140  
  1141  // parsePerlFlags parses a Perl flag setting or non-capturing group or both,
  1142  // like (?i) or (?: or (?i:.  It removes the prefix from s and updates the parse state.
  1143  // The caller must have ensured that s begins with "(?".
  1144  func (p *parser) parsePerlFlags(s string) (rest string, err error) {
  1145  	t := s
  1146  
  1147  	// Check for named captures, first introduced in Python's regexp library.
  1148  	// As usual, there are three slightly different syntaxes:
  1149  	//
  1150  	//   (?P<name>expr)   the original, introduced by Python
  1151  	//   (?<name>expr)    the .NET alteration, adopted by Perl 5.10
  1152  	//   (?'name'expr)    another .NET alteration, adopted by Perl 5.10
  1153  	//
  1154  	// Perl 5.10 gave in and implemented the Python version too,
  1155  	// but they claim that the last two are the preferred forms.
  1156  	// PCRE and languages based on it (specifically, PHP and Ruby)
  1157  	// support all three as well. EcmaScript 4 uses only the Python form.
  1158  	//
  1159  	// In both the open source world (via Code Search) and the
  1160  	// Google source tree, (?P<expr>name) and (?<expr>name) are the
  1161  	// dominant forms of named captures and both are supported.
  1162  	startsWithP := len(t) > 4 && t[2] == 'P' && t[3] == '<'
  1163  	startsWithName := len(t) > 3 && t[2] == '<'
  1164  
  1165  	if startsWithP || startsWithName {
  1166  		// position of expr start
  1167  		exprStartPos := 4
  1168  		if startsWithName {
  1169  			exprStartPos = 3
  1170  		}
  1171  
  1172  		// Pull out name.
  1173  		end := strings.IndexRune(t, '>')
  1174  		if end < 0 {
  1175  			if err = checkUTF8(t); err != nil {
  1176  				return "", err
  1177  			}
  1178  			return "", &Error{ErrInvalidNamedCapture, s}
  1179  		}
  1180  
  1181  		capture := t[:end+1]        // "(?P<name>" or "(?<name>"
  1182  		name := t[exprStartPos:end] // "name"
  1183  		if err = checkUTF8(name); err != nil {
  1184  			return "", err
  1185  		}
  1186  		if !isValidCaptureName(name) {
  1187  			return "", &Error{ErrInvalidNamedCapture, capture}
  1188  		}
  1189  
  1190  		// Like ordinary capture, but named.
  1191  		p.numCap++
  1192  		re := p.op(opLeftParen)
  1193  		re.Cap = p.numCap
  1194  		re.Name = name
  1195  		return t[end+1:], nil
  1196  	}
  1197  
  1198  	// Non-capturing group. Might also twiddle Perl flags.
  1199  	var c rune
  1200  	t = t[2:] // skip (?
  1201  	flags := p.flags
  1202  	sign := +1
  1203  	sawFlag := false
  1204  Loop:
  1205  	for t != "" {
  1206  		if c, t, err = nextRune(t); err != nil {
  1207  			return "", err
  1208  		}
  1209  		switch c {
  1210  		default:
  1211  			break Loop
  1212  
  1213  		// Flags.
  1214  		case 'i':
  1215  			flags |= FoldCase
  1216  			sawFlag = true
  1217  		case 'm':
  1218  			flags &^= OneLine
  1219  			sawFlag = true
  1220  		case 's':
  1221  			flags |= DotNL
  1222  			sawFlag = true
  1223  		case 'U':
  1224  			flags |= NonGreedy
  1225  			sawFlag = true
  1226  
  1227  		// Switch to negation.
  1228  		case '-':
  1229  			if sign < 0 {
  1230  				break Loop
  1231  			}
  1232  			sign = -1
  1233  			// Invert flags so that | above turn into &^ and vice versa.
  1234  			// We'll invert flags again before using it below.
  1235  			flags = ^flags
  1236  			sawFlag = false
  1237  
  1238  		// End of flags, starting group or not.
  1239  		case ':', ')':
  1240  			if sign < 0 {
  1241  				if !sawFlag {
  1242  					break Loop
  1243  				}
  1244  				flags = ^flags
  1245  			}
  1246  			if c == ':' {
  1247  				// Open new group
  1248  				p.op(opLeftParen)
  1249  			}
  1250  			p.flags = flags
  1251  			return t, nil
  1252  		}
  1253  	}
  1254  
  1255  	return "", &Error{ErrInvalidPerlOp, s[:len(s)-len(t)]}
  1256  }
  1257  
  1258  // isValidCaptureName reports whether name
  1259  // is a valid capture name: [A-Za-z0-9_]+.
  1260  // PCRE limits names to 32 bytes.
  1261  // Python rejects names starting with digits.
  1262  // We don't enforce either of those.
  1263  func isValidCaptureName(name string) bool {
  1264  	if name == "" {
  1265  		return false
  1266  	}
  1267  	for _, c := range name {
  1268  		if c != '_' && !isalnum(c) {
  1269  			return false
  1270  		}
  1271  	}
  1272  	return true
  1273  }
  1274  
  1275  // parseInt parses a decimal integer.
  1276  func (p *parser) parseInt(s string) (n int, rest string, ok bool) {
  1277  	if s == "" || s[0] < '0' || '9' < s[0] {
  1278  		return
  1279  	}
  1280  	// Disallow leading zeros.
  1281  	if len(s) >= 2 && s[0] == '0' && '0' <= s[1] && s[1] <= '9' {
  1282  		return
  1283  	}
  1284  	t := s
  1285  	for s != "" && '0' <= s[0] && s[0] <= '9' {
  1286  		s = s[1:]
  1287  	}
  1288  	rest = s
  1289  	ok = true
  1290  	// Have digits, compute value.
  1291  	t = t[:len(t)-len(s)]
  1292  	for i := 0; i < len(t); i++ {
  1293  		// Avoid overflow.
  1294  		if n >= 1e8 {
  1295  			n = -1
  1296  			break
  1297  		}
  1298  		n = n*10 + int(t[i]) - '0'
  1299  	}
  1300  	return
  1301  }
  1302  
  1303  // can this be represented as a character class?
  1304  // single-rune literal string, char class, ., and .|\n.
  1305  func isCharClass(re *Regexp) bool {
  1306  	return re.Op == OpLiteral && len(re.Rune) == 1 ||
  1307  		re.Op == OpCharClass ||
  1308  		re.Op == OpAnyCharNotNL ||
  1309  		re.Op == OpAnyChar
  1310  }
  1311  
  1312  // does re match r?
  1313  func matchRune(re *Regexp, r rune) bool {
  1314  	switch re.Op {
  1315  	case OpLiteral:
  1316  		return len(re.Rune) == 1 && re.Rune[0] == r
  1317  	case OpCharClass:
  1318  		for i := 0; i < len(re.Rune); i += 2 {
  1319  			if re.Rune[i] <= r && r <= re.Rune[i+1] {
  1320  				return true
  1321  			}
  1322  		}
  1323  		return false
  1324  	case OpAnyCharNotNL:
  1325  		return r != '\n'
  1326  	case OpAnyChar:
  1327  		return true
  1328  	}
  1329  	return false
  1330  }
  1331  
  1332  // parseVerticalBar handles a | in the input.
  1333  func (p *parser) parseVerticalBar() error {
  1334  	p.concat()
  1335  
  1336  	// The concatenation we just parsed is on top of the stack.
  1337  	// If it sits above an opVerticalBar, swap it below
  1338  	// (things below an opVerticalBar become an alternation).
  1339  	// Otherwise, push a new vertical bar.
  1340  	if !p.swapVerticalBar() {
  1341  		p.op(opVerticalBar)
  1342  	}
  1343  
  1344  	return nil
  1345  }
  1346  
  1347  // mergeCharClass makes dst = dst|src.
  1348  // The caller must ensure that dst.Op >= src.Op,
  1349  // to reduce the amount of copying.
  1350  func mergeCharClass(dst, src *Regexp) {
  1351  	switch dst.Op {
  1352  	case OpAnyChar:
  1353  		// src doesn't add anything.
  1354  	case OpAnyCharNotNL:
  1355  		// src might add \n
  1356  		if matchRune(src, '\n') {
  1357  			dst.Op = OpAnyChar
  1358  		}
  1359  	case OpCharClass:
  1360  		// src is simpler, so either literal or char class
  1361  		if src.Op == OpLiteral {
  1362  			dst.Rune = appendLiteral(dst.Rune, src.Rune[0], src.Flags)
  1363  		} else {
  1364  			dst.Rune = appendClass(dst.Rune, src.Rune)
  1365  		}
  1366  	case OpLiteral:
  1367  		// both literal
  1368  		if src.Rune[0] == dst.Rune[0] && src.Flags == dst.Flags {
  1369  			break
  1370  		}
  1371  		dst.Op = OpCharClass
  1372  		dst.Rune = appendLiteral(dst.Rune[:0], dst.Rune[0], dst.Flags)
  1373  		dst.Rune = appendLiteral(dst.Rune, src.Rune[0], src.Flags)
  1374  	}
  1375  }
  1376  
  1377  // If the top of the stack is an element followed by an opVerticalBar
  1378  // swapVerticalBar swaps the two and returns true.
  1379  // Otherwise it returns false.
  1380  func (p *parser) swapVerticalBar() bool {
  1381  	// If above and below vertical bar are literal or char class,
  1382  	// can merge into a single char class.
  1383  	n := len(p.stack)
  1384  	if n >= 3 && p.stack[n-2].Op == opVerticalBar && isCharClass(p.stack[n-1]) && isCharClass(p.stack[n-3]) {
  1385  		re1 := p.stack[n-1]
  1386  		re3 := p.stack[n-3]
  1387  		// Make re3 the more complex of the two.
  1388  		if re1.Op > re3.Op {
  1389  			re1, re3 = re3, re1
  1390  			p.stack[n-3] = re3
  1391  		}
  1392  		mergeCharClass(re3, re1)
  1393  		p.reuse(re1)
  1394  		p.stack = p.stack[:n-1]
  1395  		return true
  1396  	}
  1397  
  1398  	if n >= 2 {
  1399  		re1 := p.stack[n-1]
  1400  		re2 := p.stack[n-2]
  1401  		if re2.Op == opVerticalBar {
  1402  			if n >= 3 {
  1403  				// Now out of reach.
  1404  				// Clean opportunistically.
  1405  				cleanAlt(p.stack[n-3])
  1406  			}
  1407  			p.stack[n-2] = re1
  1408  			p.stack[n-1] = re2
  1409  			return true
  1410  		}
  1411  	}
  1412  	return false
  1413  }
  1414  
  1415  // parseRightParen handles a ) in the input.
  1416  func (p *parser) parseRightParen() error {
  1417  	p.concat()
  1418  	if p.swapVerticalBar() {
  1419  		// pop vertical bar
  1420  		p.stack = p.stack[:len(p.stack)-1]
  1421  	}
  1422  	p.alternate()
  1423  
  1424  	n := len(p.stack)
  1425  	if n < 2 {
  1426  		return &Error{ErrUnexpectedParen, p.wholeRegexp}
  1427  	}
  1428  	re1 := p.stack[n-1]
  1429  	re2 := p.stack[n-2]
  1430  	p.stack = p.stack[:n-2]
  1431  	if re2.Op != opLeftParen {
  1432  		return &Error{ErrUnexpectedParen, p.wholeRegexp}
  1433  	}
  1434  	// Restore flags at time of paren.
  1435  	p.flags = re2.Flags
  1436  	if re2.Cap == 0 {
  1437  		// Just for grouping.
  1438  		p.push(re1)
  1439  	} else {
  1440  		re2.Op = OpCapture
  1441  		re2.Sub = re2.Sub0[:1]
  1442  		re2.Sub[0] = re1
  1443  		p.push(re2)
  1444  	}
  1445  	return nil
  1446  }
  1447  
  1448  // parseEscape parses an escape sequence at the beginning of s
  1449  // and returns the rune.
  1450  func (p *parser) parseEscape(s string) (r rune, rest string, err error) {
  1451  	t := s[1:]
  1452  	if t == "" {
  1453  		return 0, "", &Error{ErrTrailingBackslash, ""}
  1454  	}
  1455  	c, t, err := nextRune(t)
  1456  	if err != nil {
  1457  		return 0, "", err
  1458  	}
  1459  
  1460  Switch:
  1461  	switch c {
  1462  	default:
  1463  		if c < utf8.RuneSelf && !isalnum(c) {
  1464  			// Escaped non-word characters are always themselves.
  1465  			// PCRE is not quite so rigorous: it accepts things like
  1466  			// \q, but we don't. We once rejected \_, but too many
  1467  			// programs and people insist on using it, so allow \_.
  1468  			return c, t, nil
  1469  		}
  1470  
  1471  	// Octal escapes.
  1472  	case '1', '2', '3', '4', '5', '6', '7':
  1473  		// Single non-zero digit is a backreference; not supported
  1474  		if t == "" || t[0] < '0' || t[0] > '7' {
  1475  			break
  1476  		}
  1477  		fallthrough
  1478  	case '0':
  1479  		// Consume up to three octal digits; already have one.
  1480  		r = c - '0'
  1481  		for i := 1; i < 3; i++ {
  1482  			if t == "" || t[0] < '0' || t[0] > '7' {
  1483  				break
  1484  			}
  1485  			r = r*8 + rune(t[0]) - '0'
  1486  			t = t[1:]
  1487  		}
  1488  		return r, t, nil
  1489  
  1490  	// Hexadecimal escapes.
  1491  	case 'x':
  1492  		if t == "" {
  1493  			break
  1494  		}
  1495  		if c, t, err = nextRune(t); err != nil {
  1496  			return 0, "", err
  1497  		}
  1498  		if c == '{' {
  1499  			// Any number of digits in braces.
  1500  			// Perl accepts any text at all; it ignores all text
  1501  			// after the first non-hex digit. We require only hex digits,
  1502  			// and at least one.
  1503  			nhex := 0
  1504  			r = 0
  1505  			for {
  1506  				if t == "" {
  1507  					break Switch
  1508  				}
  1509  				if c, t, err = nextRune(t); err != nil {
  1510  					return 0, "", err
  1511  				}
  1512  				if c == '}' {
  1513  					break
  1514  				}
  1515  				v := unhex(c)
  1516  				if v < 0 {
  1517  					break Switch
  1518  				}
  1519  				r = r*16 + v
  1520  				if r > unicode.MaxRune {
  1521  					break Switch
  1522  				}
  1523  				nhex++
  1524  			}
  1525  			if nhex == 0 {
  1526  				break Switch
  1527  			}
  1528  			return r, t, nil
  1529  		}
  1530  
  1531  		// Easy case: two hex digits.
  1532  		x := unhex(c)
  1533  		if c, t, err = nextRune(t); err != nil {
  1534  			return 0, "", err
  1535  		}
  1536  		y := unhex(c)
  1537  		if x < 0 || y < 0 {
  1538  			break
  1539  		}
  1540  		return x*16 + y, t, nil
  1541  
  1542  	// C escapes. There is no case 'b', to avoid misparsing
  1543  	// the Perl word-boundary \b as the C backspace \b
  1544  	// when in POSIX mode. In Perl, /\b/ means word-boundary
  1545  	// but /[\b]/ means backspace. We don't support that.
  1546  	// If you want a backspace, embed a literal backspace
  1547  	// character or use \x08.
  1548  	case 'a':
  1549  		return '\a', t, err
  1550  	case 'f':
  1551  		return '\f', t, err
  1552  	case 'n':
  1553  		return '\n', t, err
  1554  	case 'r':
  1555  		return '\r', t, err
  1556  	case 't':
  1557  		return '\t', t, err
  1558  	case 'v':
  1559  		return '\v', t, err
  1560  	}
  1561  	return 0, "", &Error{ErrInvalidEscape, s[:len(s)-len(t)]}
  1562  }
  1563  
  1564  // parseClassChar parses a character class character at the beginning of s
  1565  // and returns it.
  1566  func (p *parser) parseClassChar(s, wholeClass string) (r rune, rest string, err error) {
  1567  	if s == "" {
  1568  		return 0, "", &Error{Code: ErrMissingBracket, Expr: wholeClass}
  1569  	}
  1570  
  1571  	// Allow regular escape sequences even though
  1572  	// many need not be escaped in this context.
  1573  	if s[0] == '\\' {
  1574  		return p.parseEscape(s)
  1575  	}
  1576  
  1577  	return nextRune(s)
  1578  }
  1579  
  1580  type charGroup struct {
  1581  	sign  int
  1582  	class []rune
  1583  }
  1584  
  1585  // parsePerlClassEscape parses a leading Perl character class escape like \d
  1586  // from the beginning of s. If one is present, it appends the characters to r
  1587  // and returns the new slice r and the remainder of the string.
  1588  func (p *parser) parsePerlClassEscape(s string, r []rune) (out []rune, rest string) {
  1589  	if p.flags&PerlX == 0 || len(s) < 2 || s[0] != '\\' {
  1590  		return
  1591  	}
  1592  	g := perlGroup[s[0:2]]
  1593  	if g.sign == 0 {
  1594  		return
  1595  	}
  1596  	return p.appendGroup(r, g), s[2:]
  1597  }
  1598  
  1599  // parseNamedClass parses a leading POSIX named character class like [:alnum:]
  1600  // from the beginning of s. If one is present, it appends the characters to r
  1601  // and returns the new slice r and the remainder of the string.
  1602  func (p *parser) parseNamedClass(s string, r []rune) (out []rune, rest string, err error) {
  1603  	if len(s) < 2 || s[0] != '[' || s[1] != ':' {
  1604  		return
  1605  	}
  1606  
  1607  	i := strings.Index(s[2:], ":]")
  1608  	if i < 0 {
  1609  		return
  1610  	}
  1611  	i += 2
  1612  	name, s := s[0:i+2], s[i+2:]
  1613  	g := posixGroup[name]
  1614  	if g.sign == 0 {
  1615  		return nil, "", &Error{ErrInvalidCharRange, name}
  1616  	}
  1617  	return p.appendGroup(r, g), s, nil
  1618  }
  1619  
  1620  func (p *parser) appendGroup(r []rune, g charGroup) []rune {
  1621  	if p.flags&FoldCase == 0 {
  1622  		if g.sign < 0 {
  1623  			r = appendNegatedClass(r, g.class)
  1624  		} else {
  1625  			r = appendClass(r, g.class)
  1626  		}
  1627  	} else {
  1628  		tmp := p.tmpClass[:0]
  1629  		tmp = appendFoldedClass(tmp, g.class)
  1630  		p.tmpClass = tmp
  1631  		tmp = cleanClass(&p.tmpClass)
  1632  		if g.sign < 0 {
  1633  			r = appendNegatedClass(r, tmp)
  1634  		} else {
  1635  			r = appendClass(r, tmp)
  1636  		}
  1637  	}
  1638  	return r
  1639  }
  1640  
  1641  var anyTable = &unicode.RangeTable{
  1642  	R16: []unicode.Range16{{Lo: 0, Hi: 1<<16 - 1, Stride: 1}},
  1643  	R32: []unicode.Range32{{Lo: 1 << 16, Hi: unicode.MaxRune, Stride: 1}},
  1644  }
  1645  
  1646  // unicodeTable returns the unicode.RangeTable identified by name
  1647  // and the table of additional fold-equivalent code points.
  1648  func unicodeTable(name string) (*unicode.RangeTable, *unicode.RangeTable) {
  1649  	// Special case: "Any" means any.
  1650  	if name == "Any" {
  1651  		return anyTable, anyTable
  1652  	}
  1653  	if t := unicode.Categories[name]; t != nil {
  1654  		return t, unicode.FoldCategory[name]
  1655  	}
  1656  	if t := unicode.Scripts[name]; t != nil {
  1657  		return t, unicode.FoldScript[name]
  1658  	}
  1659  	return nil, nil
  1660  }
  1661  
  1662  // parseUnicodeClass parses a leading Unicode character class like \p{Han}
  1663  // from the beginning of s. If one is present, it appends the characters to r
  1664  // and returns the new slice r and the remainder of the string.
  1665  func (p *parser) parseUnicodeClass(s string, r []rune) (out []rune, rest string, err error) {
  1666  	if p.flags&UnicodeGroups == 0 || len(s) < 2 || s[0] != '\\' || s[1] != 'p' && s[1] != 'P' {
  1667  		return
  1668  	}
  1669  
  1670  	// Committed to parse or return error.
  1671  	sign := +1
  1672  	if s[1] == 'P' {
  1673  		sign = -1
  1674  	}
  1675  	t := s[2:]
  1676  	c, t, err := nextRune(t)
  1677  	if err != nil {
  1678  		return
  1679  	}
  1680  	var seq, name string
  1681  	if c != '{' {
  1682  		// Single-letter name.
  1683  		seq = s[:len(s)-len(t)]
  1684  		name = seq[2:]
  1685  	} else {
  1686  		// Name is in braces.
  1687  		end := strings.IndexRune(s, '}')
  1688  		if end < 0 {
  1689  			if err = checkUTF8(s); err != nil {
  1690  				return
  1691  			}
  1692  			return nil, "", &Error{ErrInvalidCharRange, s}
  1693  		}
  1694  		seq, t = s[:end+1], s[end+1:]
  1695  		name = s[3:end]
  1696  		if err = checkUTF8(name); err != nil {
  1697  			return
  1698  		}
  1699  	}
  1700  
  1701  	// Group can have leading negation too.  \p{^Han} == \P{Han}, \P{^Han} == \p{Han}.
  1702  	if name != "" && name[0] == '^' {
  1703  		sign = -sign
  1704  		name = name[1:]
  1705  	}
  1706  
  1707  	tab, fold := unicodeTable(name)
  1708  	if tab == nil {
  1709  		return nil, "", &Error{ErrInvalidCharRange, seq}
  1710  	}
  1711  
  1712  	if p.flags&FoldCase == 0 || fold == nil {
  1713  		if sign > 0 {
  1714  			r = appendTable(r, tab)
  1715  		} else {
  1716  			r = appendNegatedTable(r, tab)
  1717  		}
  1718  	} else {
  1719  		// Merge and clean tab and fold in a temporary buffer.
  1720  		// This is necessary for the negative case and just tidy
  1721  		// for the positive case.
  1722  		tmp := p.tmpClass[:0]
  1723  		tmp = appendTable(tmp, tab)
  1724  		tmp = appendTable(tmp, fold)
  1725  		p.tmpClass = tmp
  1726  		tmp = cleanClass(&p.tmpClass)
  1727  		if sign > 0 {
  1728  			r = appendClass(r, tmp)
  1729  		} else {
  1730  			r = appendNegatedClass(r, tmp)
  1731  		}
  1732  	}
  1733  	return r, t, nil
  1734  }
  1735  
  1736  // parseClass parses a character class at the beginning of s
  1737  // and pushes it onto the parse stack.
  1738  func (p *parser) parseClass(s string) (rest string, err error) {
  1739  	t := s[1:] // chop [
  1740  	re := p.newRegexp(OpCharClass)
  1741  	re.Flags = p.flags
  1742  	re.Rune = re.Rune0[:0]
  1743  
  1744  	sign := +1
  1745  	if t != "" && t[0] == '^' {
  1746  		sign = -1
  1747  		t = t[1:]
  1748  
  1749  		// If character class does not match \n, add it here,
  1750  		// so that negation later will do the right thing.
  1751  		if p.flags&ClassNL == 0 {
  1752  			re.Rune = append(re.Rune, '\n', '\n')
  1753  		}
  1754  	}
  1755  
  1756  	class := re.Rune
  1757  	first := true // ] and - are okay as first char in class
  1758  	for t == "" || t[0] != ']' || first {
  1759  		// POSIX: - is only okay unescaped as first or last in class.
  1760  		// Perl: - is okay anywhere.
  1761  		if t != "" && t[0] == '-' && p.flags&PerlX == 0 && !first && (len(t) == 1 || t[1] != ']') {
  1762  			_, size := utf8.DecodeRuneInString(t[1:])
  1763  			return "", &Error{Code: ErrInvalidCharRange, Expr: t[:1+size]}
  1764  		}
  1765  		first = false
  1766  
  1767  		// Look for POSIX [:alnum:] etc.
  1768  		if len(t) > 2 && t[0] == '[' && t[1] == ':' {
  1769  			nclass, nt, err := p.parseNamedClass(t, class)
  1770  			if err != nil {
  1771  				return "", err
  1772  			}
  1773  			if nclass != nil {
  1774  				class, t = nclass, nt
  1775  				continue
  1776  			}
  1777  		}
  1778  
  1779  		// Look for Unicode character group like \p{Han}.
  1780  		nclass, nt, err := p.parseUnicodeClass(t, class)
  1781  		if err != nil {
  1782  			return "", err
  1783  		}
  1784  		if nclass != nil {
  1785  			class, t = nclass, nt
  1786  			continue
  1787  		}
  1788  
  1789  		// Look for Perl character class symbols (extension).
  1790  		if nclass, nt := p.parsePerlClassEscape(t, class); nclass != nil {
  1791  			class, t = nclass, nt
  1792  			continue
  1793  		}
  1794  
  1795  		// Single character or simple range.
  1796  		rng := t
  1797  		var lo, hi rune
  1798  		if lo, t, err = p.parseClassChar(t, s); err != nil {
  1799  			return "", err
  1800  		}
  1801  		hi = lo
  1802  		// [a-] means (a|-) so check for final ].
  1803  		if len(t) >= 2 && t[0] == '-' && t[1] != ']' {
  1804  			t = t[1:]
  1805  			if hi, t, err = p.parseClassChar(t, s); err != nil {
  1806  				return "", err
  1807  			}
  1808  			if hi < lo {
  1809  				rng = rng[:len(rng)-len(t)]
  1810  				return "", &Error{Code: ErrInvalidCharRange, Expr: rng}
  1811  			}
  1812  		}
  1813  		if p.flags&FoldCase == 0 {
  1814  			class = appendRange(class, lo, hi)
  1815  		} else {
  1816  			class = appendFoldedRange(class, lo, hi)
  1817  		}
  1818  	}
  1819  	t = t[1:] // chop ]
  1820  
  1821  	// Use &re.Rune instead of &class to avoid allocation.
  1822  	re.Rune = class
  1823  	class = cleanClass(&re.Rune)
  1824  	if sign < 0 {
  1825  		class = negateClass(class)
  1826  	}
  1827  	re.Rune = class
  1828  	p.push(re)
  1829  	return t, nil
  1830  }
  1831  
  1832  // cleanClass sorts the ranges (pairs of elements of r),
  1833  // merges them, and eliminates duplicates.
  1834  func cleanClass(rp *[]rune) []rune {
  1835  
  1836  	// Sort by lo increasing, hi decreasing to break ties.
  1837  	sort.Sort(ranges{rp})
  1838  
  1839  	r := *rp
  1840  	if len(r) < 2 {
  1841  		return r
  1842  	}
  1843  
  1844  	// Merge abutting, overlapping.
  1845  	w := 2 // write index
  1846  	for i := 2; i < len(r); i += 2 {
  1847  		lo, hi := r[i], r[i+1]
  1848  		if lo <= r[w-1]+1 {
  1849  			// merge with previous range
  1850  			if hi > r[w-1] {
  1851  				r[w-1] = hi
  1852  			}
  1853  			continue
  1854  		}
  1855  		// new disjoint range
  1856  		r[w] = lo
  1857  		r[w+1] = hi
  1858  		w += 2
  1859  	}
  1860  
  1861  	return r[:w]
  1862  }
  1863  
  1864  // inCharClass reports whether r is in the class.
  1865  // It assumes the class has been cleaned by cleanClass.
  1866  func inCharClass(r rune, class []rune) bool {
  1867  	_, ok := sort.Find(len(class)/2, func(i int) int {
  1868  		lo, hi := class[2*i], class[2*i+1]
  1869  		if r > hi {
  1870  			return +1
  1871  		}
  1872  		if r < lo {
  1873  			return -1
  1874  		}
  1875  		return 0
  1876  	})
  1877  	return ok
  1878  }
  1879  
  1880  // appendLiteral returns the result of appending the literal x to the class r.
  1881  func appendLiteral(r []rune, x rune, flags Flags) []rune {
  1882  	if flags&FoldCase != 0 {
  1883  		return appendFoldedRange(r, x, x)
  1884  	}
  1885  	return appendRange(r, x, x)
  1886  }
  1887  
  1888  // appendRange returns the result of appending the range lo-hi to the class r.
  1889  func appendRange(r []rune, lo, hi rune) []rune {
  1890  	// Expand last range or next to last range if it overlaps or abuts.
  1891  	// Checking two ranges helps when appending case-folded
  1892  	// alphabets, so that one range can be expanding A-Z and the
  1893  	// other expanding a-z.
  1894  	n := len(r)
  1895  	for i := 2; i <= 4; i += 2 { // twice, using i=2, i=4
  1896  		if n >= i {
  1897  			rlo, rhi := r[n-i], r[n-i+1]
  1898  			if lo <= rhi+1 && rlo <= hi+1 {
  1899  				if lo < rlo {
  1900  					r[n-i] = lo
  1901  				}
  1902  				if hi > rhi {
  1903  					r[n-i+1] = hi
  1904  				}
  1905  				return r
  1906  			}
  1907  		}
  1908  	}
  1909  
  1910  	return append(r, lo, hi)
  1911  }
  1912  
  1913  const (
  1914  	// minimum and maximum runes involved in folding.
  1915  	// checked during test.
  1916  	minFold = 0x0041
  1917  	maxFold = 0x1e943
  1918  )
  1919  
  1920  // appendFoldedRange returns the result of appending the range lo-hi
  1921  // and its case folding-equivalent runes to the class r.
  1922  func appendFoldedRange(r []rune, lo, hi rune) []rune {
  1923  	// Optimizations.
  1924  	if lo <= minFold && hi >= maxFold {
  1925  		// Range is full: folding can't add more.
  1926  		return appendRange(r, lo, hi)
  1927  	}
  1928  	if hi < minFold || lo > maxFold {
  1929  		// Range is outside folding possibilities.
  1930  		return appendRange(r, lo, hi)
  1931  	}
  1932  	if lo < minFold {
  1933  		// [lo, minFold-1] needs no folding.
  1934  		r = appendRange(r, lo, minFold-1)
  1935  		lo = minFold
  1936  	}
  1937  	if hi > maxFold {
  1938  		// [maxFold+1, hi] needs no folding.
  1939  		r = appendRange(r, maxFold+1, hi)
  1940  		hi = maxFold
  1941  	}
  1942  
  1943  	// Brute force. Depend on appendRange to coalesce ranges on the fly.
  1944  	for c := lo; c <= hi; c++ {
  1945  		r = appendRange(r, c, c)
  1946  		f := unicode.SimpleFold(c)
  1947  		for f != c {
  1948  			r = appendRange(r, f, f)
  1949  			f = unicode.SimpleFold(f)
  1950  		}
  1951  	}
  1952  	return r
  1953  }
  1954  
  1955  // appendClass returns the result of appending the class x to the class r.
  1956  // It assume x is clean.
  1957  func appendClass(r []rune, x []rune) []rune {
  1958  	for i := 0; i < len(x); i += 2 {
  1959  		r = appendRange(r, x[i], x[i+1])
  1960  	}
  1961  	return r
  1962  }
  1963  
  1964  // appendFoldedClass returns the result of appending the case folding of the class x to the class r.
  1965  func appendFoldedClass(r []rune, x []rune) []rune {
  1966  	for i := 0; i < len(x); i += 2 {
  1967  		r = appendFoldedRange(r, x[i], x[i+1])
  1968  	}
  1969  	return r
  1970  }
  1971  
  1972  // appendNegatedClass returns the result of appending the negation of the class x to the class r.
  1973  // It assumes x is clean.
  1974  func appendNegatedClass(r []rune, x []rune) []rune {
  1975  	nextLo := '\u0000'
  1976  	for i := 0; i < len(x); i += 2 {
  1977  		lo, hi := x[i], x[i+1]
  1978  		if nextLo <= lo-1 {
  1979  			r = appendRange(r, nextLo, lo-1)
  1980  		}
  1981  		nextLo = hi + 1
  1982  	}
  1983  	if nextLo <= unicode.MaxRune {
  1984  		r = appendRange(r, nextLo, unicode.MaxRune)
  1985  	}
  1986  	return r
  1987  }
  1988  
  1989  // appendTable returns the result of appending x to the class r.
  1990  func appendTable(r []rune, x *unicode.RangeTable) []rune {
  1991  	for _, xr := range x.R16 {
  1992  		lo, hi, stride := rune(xr.Lo), rune(xr.Hi), rune(xr.Stride)
  1993  		if stride == 1 {
  1994  			r = appendRange(r, lo, hi)
  1995  			continue
  1996  		}
  1997  		for c := lo; c <= hi; c += stride {
  1998  			r = appendRange(r, c, c)
  1999  		}
  2000  	}
  2001  	for _, xr := range x.R32 {
  2002  		lo, hi, stride := rune(xr.Lo), rune(xr.Hi), rune(xr.Stride)
  2003  		if stride == 1 {
  2004  			r = appendRange(r, lo, hi)
  2005  			continue
  2006  		}
  2007  		for c := lo; c <= hi; c += stride {
  2008  			r = appendRange(r, c, c)
  2009  		}
  2010  	}
  2011  	return r
  2012  }
  2013  
  2014  // appendNegatedTable returns the result of appending the negation of x to the class r.
  2015  func appendNegatedTable(r []rune, x *unicode.RangeTable) []rune {
  2016  	nextLo := '\u0000' // lo end of next class to add
  2017  	for _, xr := range x.R16 {
  2018  		lo, hi, stride := rune(xr.Lo), rune(xr.Hi), rune(xr.Stride)
  2019  		if stride == 1 {
  2020  			if nextLo <= lo-1 {
  2021  				r = appendRange(r, nextLo, lo-1)
  2022  			}
  2023  			nextLo = hi + 1
  2024  			continue
  2025  		}
  2026  		for c := lo; c <= hi; c += stride {
  2027  			if nextLo <= c-1 {
  2028  				r = appendRange(r, nextLo, c-1)
  2029  			}
  2030  			nextLo = c + 1
  2031  		}
  2032  	}
  2033  	for _, xr := range x.R32 {
  2034  		lo, hi, stride := rune(xr.Lo), rune(xr.Hi), rune(xr.Stride)
  2035  		if stride == 1 {
  2036  			if nextLo <= lo-1 {
  2037  				r = appendRange(r, nextLo, lo-1)
  2038  			}
  2039  			nextLo = hi + 1
  2040  			continue
  2041  		}
  2042  		for c := lo; c <= hi; c += stride {
  2043  			if nextLo <= c-1 {
  2044  				r = appendRange(r, nextLo, c-1)
  2045  			}
  2046  			nextLo = c + 1
  2047  		}
  2048  	}
  2049  	if nextLo <= unicode.MaxRune {
  2050  		r = appendRange(r, nextLo, unicode.MaxRune)
  2051  	}
  2052  	return r
  2053  }
  2054  
  2055  // negateClass overwrites r and returns r's negation.
  2056  // It assumes the class r is already clean.
  2057  func negateClass(r []rune) []rune {
  2058  	nextLo := '\u0000' // lo end of next class to add
  2059  	w := 0             // write index
  2060  	for i := 0; i < len(r); i += 2 {
  2061  		lo, hi := r[i], r[i+1]
  2062  		if nextLo <= lo-1 {
  2063  			r[w] = nextLo
  2064  			r[w+1] = lo - 1
  2065  			w += 2
  2066  		}
  2067  		nextLo = hi + 1
  2068  	}
  2069  	r = r[:w]
  2070  	if nextLo <= unicode.MaxRune {
  2071  		// It's possible for the negation to have one more
  2072  		// range - this one - than the original class, so use append.
  2073  		r = append(r, nextLo, unicode.MaxRune)
  2074  	}
  2075  	return r
  2076  }
  2077  
  2078  // ranges implements sort.Interface on a []rune.
  2079  // The choice of receiver type definition is strange
  2080  // but avoids an allocation since we already have
  2081  // a *[]rune.
  2082  type ranges struct {
  2083  	p *[]rune
  2084  }
  2085  
  2086  func (ra ranges) Less(i, j int) bool {
  2087  	p := *ra.p
  2088  	i *= 2
  2089  	j *= 2
  2090  	return p[i] < p[j] || p[i] == p[j] && p[i+1] > p[j+1]
  2091  }
  2092  
  2093  func (ra ranges) Len() int {
  2094  	return len(*ra.p) / 2
  2095  }
  2096  
  2097  func (ra ranges) Swap(i, j int) {
  2098  	p := *ra.p
  2099  	i *= 2
  2100  	j *= 2
  2101  	p[i], p[i+1], p[j], p[j+1] = p[j], p[j+1], p[i], p[i+1]
  2102  }
  2103  
  2104  func checkUTF8(s string) error {
  2105  	for s != "" {
  2106  		rune, size := utf8.DecodeRuneInString(s)
  2107  		if rune == utf8.RuneError && size == 1 {
  2108  			return &Error{Code: ErrInvalidUTF8, Expr: s}
  2109  		}
  2110  		s = s[size:]
  2111  	}
  2112  	return nil
  2113  }
  2114  
  2115  func nextRune(s string) (c rune, t string, err error) {
  2116  	c, size := utf8.DecodeRuneInString(s)
  2117  	if c == utf8.RuneError && size == 1 {
  2118  		return 0, "", &Error{Code: ErrInvalidUTF8, Expr: s}
  2119  	}
  2120  	return c, s[size:], nil
  2121  }
  2122  
  2123  func isalnum(c rune) bool {
  2124  	return '0' <= c && c <= '9' || 'A' <= c && c <= 'Z' || 'a' <= c && c <= 'z'
  2125  }
  2126  
  2127  func unhex(c rune) rune {
  2128  	if '0' <= c && c <= '9' {
  2129  		return c - '0'
  2130  	}
  2131  	if 'a' <= c && c <= 'f' {
  2132  		return c - 'a' + 10
  2133  	}
  2134  	if 'A' <= c && c <= 'F' {
  2135  		return c - 'A' + 10
  2136  	}
  2137  	return -1
  2138  }
  2139  

View as plain text