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

View as plain text