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

View as plain text