Source file src/regexp/syntax/parse.go

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

View as plain text