• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // Copyright 2011 the V8 project authors. All rights reserved.
2 // Use of this source code is governed by a BSD-style license that can be
3 // found in the LICENSE file.
4 
5 // Features shared by parsing and pre-parsing scanners.
6 
7 #ifndef V8_SCANNER_H_
8 #define V8_SCANNER_H_
9 
10 #include "src/allocation.h"
11 #include "src/char-predicates.h"
12 #include "src/checks.h"
13 #include "src/globals.h"
14 #include "src/hashmap.h"
15 #include "src/list.h"
16 #include "src/token.h"
17 #include "src/unicode-inl.h"
18 #include "src/utils.h"
19 
20 namespace v8 {
21 namespace internal {
22 
23 
24 class ParserRecorder;
25 
26 
27 // Returns the value (0 .. 15) of a hexadecimal character c.
28 // If c is not a legal hexadecimal character, returns a value < 0.
HexValue(uc32 c)29 inline int HexValue(uc32 c) {
30   c -= '0';
31   if (static_cast<unsigned>(c) <= 9) return c;
32   c = (c | 0x20) - ('a' - '0');  // detect 0x11..0x16 and 0x31..0x36.
33   if (static_cast<unsigned>(c) <= 5) return c + 10;
34   return -1;
35 }
36 
37 
38 // ---------------------------------------------------------------------
39 // Buffered stream of UTF-16 code units, using an internal UTF-16 buffer.
40 // A code unit is a 16 bit value representing either a 16 bit code point
41 // or one part of a surrogate pair that make a single 21 bit code point.
42 
43 class Utf16CharacterStream {
44  public:
Utf16CharacterStream()45   Utf16CharacterStream() : pos_(0) { }
~Utf16CharacterStream()46   virtual ~Utf16CharacterStream() { }
47 
48   // Returns and advances past the next UTF-16 code unit in the input
49   // stream. If there are no more code units, it returns a negative
50   // value.
Advance()51   inline uc32 Advance() {
52     if (buffer_cursor_ < buffer_end_ || ReadBlock()) {
53       pos_++;
54       return static_cast<uc32>(*(buffer_cursor_++));
55     }
56     // Note: currently the following increment is necessary to avoid a
57     // parser problem! The scanner treats the final kEndOfInput as
58     // a code unit with a position, and does math relative to that
59     // position.
60     pos_++;
61 
62     return kEndOfInput;
63   }
64 
65   // Return the current position in the code unit stream.
66   // Starts at zero.
pos()67   inline unsigned pos() const { return pos_; }
68 
69   // Skips forward past the next code_unit_count UTF-16 code units
70   // in the input, or until the end of input if that comes sooner.
71   // Returns the number of code units actually skipped. If less
72   // than code_unit_count,
SeekForward(unsigned code_unit_count)73   inline unsigned SeekForward(unsigned code_unit_count) {
74     unsigned buffered_chars =
75         static_cast<unsigned>(buffer_end_ - buffer_cursor_);
76     if (code_unit_count <= buffered_chars) {
77       buffer_cursor_ += code_unit_count;
78       pos_ += code_unit_count;
79       return code_unit_count;
80     }
81     return SlowSeekForward(code_unit_count);
82   }
83 
84   // Pushes back the most recently read UTF-16 code unit (or negative
85   // value if at end of input), i.e., the value returned by the most recent
86   // call to Advance.
87   // Must not be used right after calling SeekForward.
88   virtual void PushBack(int32_t code_unit) = 0;
89 
90  protected:
91   static const uc32 kEndOfInput = -1;
92 
93   // Ensures that the buffer_cursor_ points to the code_unit at
94   // position pos_ of the input, if possible. If the position
95   // is at or after the end of the input, return false. If there
96   // are more code_units available, return true.
97   virtual bool ReadBlock() = 0;
98   virtual unsigned SlowSeekForward(unsigned code_unit_count) = 0;
99 
100   const uint16_t* buffer_cursor_;
101   const uint16_t* buffer_end_;
102   unsigned pos_;
103 };
104 
105 
106 // ---------------------------------------------------------------------
107 // Caching predicates used by scanners.
108 
109 class UnicodeCache {
110  public:
UnicodeCache()111   UnicodeCache() {}
112   typedef unibrow::Utf8Decoder<512> Utf8Decoder;
113 
utf8_decoder()114   StaticResource<Utf8Decoder>* utf8_decoder() {
115     return &utf8_decoder_;
116   }
117 
IsIdentifierStart(unibrow::uchar c)118   bool IsIdentifierStart(unibrow::uchar c) { return kIsIdentifierStart.get(c); }
IsIdentifierPart(unibrow::uchar c)119   bool IsIdentifierPart(unibrow::uchar c) { return kIsIdentifierPart.get(c); }
IsLineTerminator(unibrow::uchar c)120   bool IsLineTerminator(unibrow::uchar c) { return kIsLineTerminator.get(c); }
IsWhiteSpace(unibrow::uchar c)121   bool IsWhiteSpace(unibrow::uchar c) { return kIsWhiteSpace.get(c); }
IsWhiteSpaceOrLineTerminator(unibrow::uchar c)122   bool IsWhiteSpaceOrLineTerminator(unibrow::uchar c) {
123     return kIsWhiteSpaceOrLineTerminator.get(c);
124   }
125 
126  private:
127   unibrow::Predicate<IdentifierStart, 128> kIsIdentifierStart;
128   unibrow::Predicate<IdentifierPart, 128> kIsIdentifierPart;
129   unibrow::Predicate<unibrow::LineTerminator, 128> kIsLineTerminator;
130   unibrow::Predicate<WhiteSpace, 128> kIsWhiteSpace;
131   unibrow::Predicate<WhiteSpaceOrLineTerminator, 128>
132       kIsWhiteSpaceOrLineTerminator;
133   StaticResource<Utf8Decoder> utf8_decoder_;
134 
135   DISALLOW_COPY_AND_ASSIGN(UnicodeCache);
136 };
137 
138 
139 // ---------------------------------------------------------------------
140 // DuplicateFinder discovers duplicate symbols.
141 
142 class DuplicateFinder {
143  public:
DuplicateFinder(UnicodeCache * constants)144   explicit DuplicateFinder(UnicodeCache* constants)
145       : unicode_constants_(constants),
146         backing_store_(16),
147         map_(&Match) { }
148 
149   int AddOneByteSymbol(Vector<const uint8_t> key, int value);
150   int AddTwoByteSymbol(Vector<const uint16_t> key, int value);
151   // Add a a number literal by converting it (if necessary)
152   // to the string that ToString(ToNumber(literal)) would generate.
153   // and then adding that string with AddAsciiSymbol.
154   // This string is the actual value used as key in an object literal,
155   // and the one that must be different from the other keys.
156   int AddNumber(Vector<const uint8_t> key, int value);
157 
158  private:
159   int AddSymbol(Vector<const uint8_t> key, bool is_one_byte, int value);
160   // Backs up the key and its length in the backing store.
161   // The backup is stored with a base 127 encoding of the
162   // length (plus a bit saying whether the string is one byte),
163   // followed by the bytes of the key.
164   uint8_t* BackupKey(Vector<const uint8_t> key, bool is_one_byte);
165 
166   // Compare two encoded keys (both pointing into the backing store)
167   // for having the same base-127 encoded lengths and ASCII-ness,
168   // and then having the same 'length' bytes following.
169   static bool Match(void* first, void* second);
170   // Creates a hash from a sequence of bytes.
171   static uint32_t Hash(Vector<const uint8_t> key, bool is_one_byte);
172   // Checks whether a string containing a JS number is its canonical
173   // form.
174   static bool IsNumberCanonical(Vector<const uint8_t> key);
175 
176   // Size of buffer. Sufficient for using it to call DoubleToCString in
177   // from conversions.h.
178   static const int kBufferSize = 100;
179 
180   UnicodeCache* unicode_constants_;
181   // Backing store used to store strings used as hashmap keys.
182   SequenceCollector<unsigned char> backing_store_;
183   HashMap map_;
184   // Buffer used for string->number->canonical string conversions.
185   char number_buffer_[kBufferSize];
186 };
187 
188 
189 // ----------------------------------------------------------------------------
190 // LiteralBuffer -  Collector of chars of literals.
191 
192 class LiteralBuffer {
193  public:
LiteralBuffer()194   LiteralBuffer() : is_one_byte_(true), position_(0), backing_store_() { }
195 
~LiteralBuffer()196   ~LiteralBuffer() {
197     if (backing_store_.length() > 0) {
198       backing_store_.Dispose();
199     }
200   }
201 
INLINE(void AddChar (uint32_t code_unit))202   INLINE(void AddChar(uint32_t code_unit)) {
203     if (position_ >= backing_store_.length()) ExpandBuffer();
204     if (is_one_byte_) {
205       if (code_unit <= unibrow::Latin1::kMaxChar) {
206         backing_store_[position_] = static_cast<byte>(code_unit);
207         position_ += kOneByteSize;
208         return;
209       }
210       ConvertToTwoByte();
211     }
212     ASSERT(code_unit < 0x10000u);
213     *reinterpret_cast<uint16_t*>(&backing_store_[position_]) = code_unit;
214     position_ += kUC16Size;
215   }
216 
is_one_byte()217   bool is_one_byte() { return is_one_byte_; }
218 
is_contextual_keyword(Vector<const char> keyword)219   bool is_contextual_keyword(Vector<const char> keyword) {
220     return is_one_byte() && keyword.length() == position_ &&
221         (memcmp(keyword.start(), backing_store_.start(), position_) == 0);
222   }
223 
two_byte_literal()224   Vector<const uint16_t> two_byte_literal() {
225     ASSERT(!is_one_byte_);
226     ASSERT((position_ & 0x1) == 0);
227     return Vector<const uint16_t>(
228         reinterpret_cast<const uint16_t*>(backing_store_.start()),
229         position_ >> 1);
230   }
231 
one_byte_literal()232   Vector<const uint8_t> one_byte_literal() {
233     ASSERT(is_one_byte_);
234     return Vector<const uint8_t>(
235         reinterpret_cast<const uint8_t*>(backing_store_.start()),
236         position_);
237   }
238 
length()239   int length() {
240     return is_one_byte_ ? position_ : (position_ >> 1);
241   }
242 
Reset()243   void Reset() {
244     position_ = 0;
245     is_one_byte_ = true;
246   }
247 
248  private:
249   static const int kInitialCapacity = 16;
250   static const int kGrowthFactory = 4;
251   static const int kMinConversionSlack = 256;
252   static const int kMaxGrowth = 1 * MB;
NewCapacity(int min_capacity)253   inline int NewCapacity(int min_capacity) {
254     int capacity = Max(min_capacity, backing_store_.length());
255     int new_capacity = Min(capacity * kGrowthFactory, capacity + kMaxGrowth);
256     return new_capacity;
257   }
258 
ExpandBuffer()259   void ExpandBuffer() {
260     Vector<byte> new_store = Vector<byte>::New(NewCapacity(kInitialCapacity));
261     MemCopy(new_store.start(), backing_store_.start(), position_);
262     backing_store_.Dispose();
263     backing_store_ = new_store;
264   }
265 
ConvertToTwoByte()266   void ConvertToTwoByte() {
267     ASSERT(is_one_byte_);
268     Vector<byte> new_store;
269     int new_content_size = position_ * kUC16Size;
270     if (new_content_size >= backing_store_.length()) {
271       // Ensure room for all currently read code units as UC16 as well
272       // as the code unit about to be stored.
273       new_store = Vector<byte>::New(NewCapacity(new_content_size));
274     } else {
275       new_store = backing_store_;
276     }
277     uint8_t* src = backing_store_.start();
278     uint16_t* dst = reinterpret_cast<uint16_t*>(new_store.start());
279     for (int i = position_ - 1; i >= 0; i--) {
280       dst[i] = src[i];
281     }
282     if (new_store.start() != backing_store_.start()) {
283       backing_store_.Dispose();
284       backing_store_ = new_store;
285     }
286     position_ = new_content_size;
287     is_one_byte_ = false;
288   }
289 
290   bool is_one_byte_;
291   int position_;
292   Vector<byte> backing_store_;
293 
294   DISALLOW_COPY_AND_ASSIGN(LiteralBuffer);
295 };
296 
297 
298 // ----------------------------------------------------------------------------
299 // JavaScript Scanner.
300 
301 class Scanner {
302  public:
303   // Scoped helper for literal recording. Automatically drops the literal
304   // if aborting the scanning before it's complete.
305   class LiteralScope {
306    public:
LiteralScope(Scanner * self)307     explicit LiteralScope(Scanner* self)
308         : scanner_(self), complete_(false) {
309       scanner_->StartLiteral();
310     }
~LiteralScope()311      ~LiteralScope() {
312        if (!complete_) scanner_->DropLiteral();
313      }
Complete()314     void Complete() {
315       scanner_->TerminateLiteral();
316       complete_ = true;
317     }
318 
319    private:
320     Scanner* scanner_;
321     bool complete_;
322   };
323 
324   // Representation of an interval of source positions.
325   struct Location {
LocationLocation326     Location(int b, int e) : beg_pos(b), end_pos(e) { }
LocationLocation327     Location() : beg_pos(0), end_pos(0) { }
328 
IsValidLocation329     bool IsValid() const {
330       return beg_pos >= 0 && end_pos >= beg_pos;
331     }
332 
invalidLocation333     static Location invalid() { return Location(-1, -1); }
334 
335     int beg_pos;
336     int end_pos;
337   };
338 
339   // -1 is outside of the range of any real source code.
340   static const int kNoOctalLocation = -1;
341 
342   explicit Scanner(UnicodeCache* scanner_contants);
343 
344   void Initialize(Utf16CharacterStream* source);
345 
346   // Returns the next token and advances input.
347   Token::Value Next();
348   // Returns the current token again.
current_token()349   Token::Value current_token() { return current_.token; }
350   // Returns the location information for the current token
351   // (the token last returned by Next()).
location()352   Location location() const { return current_.location; }
353 
354   // Similar functions for the upcoming token.
355 
356   // One token look-ahead (past the token returned by Next()).
peek()357   Token::Value peek() const { return next_.token; }
358 
peek_location()359   Location peek_location() const { return next_.location; }
360 
literal_contains_escapes()361   bool literal_contains_escapes() const {
362     Location location = current_.location;
363     int source_length = (location.end_pos - location.beg_pos);
364     if (current_.token == Token::STRING) {
365       // Subtract delimiters.
366       source_length -= 2;
367     }
368     return current_.literal_chars->length() != source_length;
369   }
is_literal_contextual_keyword(Vector<const char> keyword)370   bool is_literal_contextual_keyword(Vector<const char> keyword) {
371     ASSERT_NOT_NULL(current_.literal_chars);
372     return current_.literal_chars->is_contextual_keyword(keyword);
373   }
is_next_contextual_keyword(Vector<const char> keyword)374   bool is_next_contextual_keyword(Vector<const char> keyword) {
375     ASSERT_NOT_NULL(next_.literal_chars);
376     return next_.literal_chars->is_contextual_keyword(keyword);
377   }
378 
379   Handle<String> AllocateNextLiteralString(Isolate* isolate,
380                                            PretenureFlag tenured);
381   Handle<String> AllocateInternalizedString(Isolate* isolate);
382 
383   double DoubleValue();
UnescapedLiteralMatches(const char * data,int length)384   bool UnescapedLiteralMatches(const char* data, int length) {
385     if (is_literal_one_byte() &&
386         literal_length() == length &&
387         !literal_contains_escapes()) {
388       const char* token =
389           reinterpret_cast<const char*>(literal_one_byte_string().start());
390       return !strncmp(token, data, length);
391     }
392     return false;
393   }
IsGetOrSet(bool * is_get,bool * is_set)394   void IsGetOrSet(bool* is_get, bool* is_set) {
395     if (is_literal_one_byte() &&
396         literal_length() == 3 &&
397         !literal_contains_escapes()) {
398       const char* token =
399           reinterpret_cast<const char*>(literal_one_byte_string().start());
400       *is_get = strncmp(token, "get", 3) == 0;
401       *is_set = !*is_get && strncmp(token, "set", 3) == 0;
402     }
403   }
404 
405   int FindNumber(DuplicateFinder* finder, int value);
406   int FindSymbol(DuplicateFinder* finder, int value);
407 
unicode_cache()408   UnicodeCache* unicode_cache() { return unicode_cache_; }
409 
410   // Returns the location of the last seen octal literal.
octal_position()411   Location octal_position() const { return octal_pos_; }
clear_octal_position()412   void clear_octal_position() { octal_pos_ = Location::invalid(); }
413 
414   // Seek forward to the given position.  This operation does not
415   // work in general, for instance when there are pushed back
416   // characters, but works for seeking forward until simple delimiter
417   // tokens, which is what it is used for.
418   void SeekForward(int pos);
419 
HarmonyScoping()420   bool HarmonyScoping() const {
421     return harmony_scoping_;
422   }
SetHarmonyScoping(bool scoping)423   void SetHarmonyScoping(bool scoping) {
424     harmony_scoping_ = scoping;
425   }
HarmonyModules()426   bool HarmonyModules() const {
427     return harmony_modules_;
428   }
SetHarmonyModules(bool modules)429   void SetHarmonyModules(bool modules) {
430     harmony_modules_ = modules;
431   }
HarmonyNumericLiterals()432   bool HarmonyNumericLiterals() const {
433     return harmony_numeric_literals_;
434   }
SetHarmonyNumericLiterals(bool numeric_literals)435   void SetHarmonyNumericLiterals(bool numeric_literals) {
436     harmony_numeric_literals_ = numeric_literals;
437   }
438 
439   // Returns true if there was a line terminator before the peek'ed token,
440   // possibly inside a multi-line comment.
HasAnyLineTerminatorBeforeNext()441   bool HasAnyLineTerminatorBeforeNext() const {
442     return has_line_terminator_before_next_ ||
443            has_multiline_comment_before_next_;
444   }
445 
446   // Scans the input as a regular expression pattern, previous
447   // character(s) must be /(=). Returns true if a pattern is scanned.
448   bool ScanRegExpPattern(bool seen_equal);
449   // Returns true if regexp flags are scanned (always since flags can
450   // be empty).
451   bool ScanRegExpFlags();
452 
453  private:
454   // The current and look-ahead token.
455   struct TokenDesc {
456     Token::Value token;
457     Location location;
458     LiteralBuffer* literal_chars;
459   };
460 
461   static const int kCharacterLookaheadBufferSize = 1;
462 
463   // Scans octal escape sequence. Also accepts "\0" decimal escape sequence.
464   uc32 ScanOctalEscape(uc32 c, int length);
465 
466   // Call this after setting source_ to the input.
Init()467   void Init() {
468     // Set c0_ (one character ahead)
469     STATIC_ASSERT(kCharacterLookaheadBufferSize == 1);
470     Advance();
471     // Initialize current_ to not refer to a literal.
472     current_.literal_chars = NULL;
473   }
474 
475   // Literal buffer support
StartLiteral()476   inline void StartLiteral() {
477     LiteralBuffer* free_buffer = (current_.literal_chars == &literal_buffer1_) ?
478             &literal_buffer2_ : &literal_buffer1_;
479     free_buffer->Reset();
480     next_.literal_chars = free_buffer;
481   }
482 
INLINE(void AddLiteralChar (uc32 c))483   INLINE(void AddLiteralChar(uc32 c)) {
484     ASSERT_NOT_NULL(next_.literal_chars);
485     next_.literal_chars->AddChar(c);
486   }
487 
488   // Complete scanning of a literal.
TerminateLiteral()489   inline void TerminateLiteral() {
490     // Does nothing in the current implementation.
491   }
492 
493   // Stops scanning of a literal and drop the collected characters,
494   // e.g., due to an encountered error.
DropLiteral()495   inline void DropLiteral() {
496     next_.literal_chars = NULL;
497   }
498 
AddLiteralCharAdvance()499   inline void AddLiteralCharAdvance() {
500     AddLiteralChar(c0_);
501     Advance();
502   }
503 
504   // Low-level scanning support.
Advance()505   void Advance() { c0_ = source_->Advance(); }
PushBack(uc32 ch)506   void PushBack(uc32 ch) {
507     source_->PushBack(c0_);
508     c0_ = ch;
509   }
510 
Select(Token::Value tok)511   inline Token::Value Select(Token::Value tok) {
512     Advance();
513     return tok;
514   }
515 
Select(uc32 next,Token::Value then,Token::Value else_)516   inline Token::Value Select(uc32 next, Token::Value then, Token::Value else_) {
517     Advance();
518     if (c0_ == next) {
519       Advance();
520       return then;
521     } else {
522       return else_;
523     }
524   }
525 
526   // Returns the literal string, if any, for the current token (the
527   // token last returned by Next()). The string is 0-terminated.
528   // Literal strings are collected for identifiers, strings, and
529   // numbers.
530   // These functions only give the correct result if the literal
531   // was scanned between calls to StartLiteral() and TerminateLiteral().
literal_one_byte_string()532   Vector<const uint8_t> literal_one_byte_string() {
533     ASSERT_NOT_NULL(current_.literal_chars);
534     return current_.literal_chars->one_byte_literal();
535   }
literal_two_byte_string()536   Vector<const uint16_t> literal_two_byte_string() {
537     ASSERT_NOT_NULL(current_.literal_chars);
538     return current_.literal_chars->two_byte_literal();
539   }
is_literal_one_byte()540   bool is_literal_one_byte() {
541     ASSERT_NOT_NULL(current_.literal_chars);
542     return current_.literal_chars->is_one_byte();
543   }
literal_length()544   int literal_length() const {
545     ASSERT_NOT_NULL(current_.literal_chars);
546     return current_.literal_chars->length();
547   }
548   // Returns the literal string for the next token (the token that
549   // would be returned if Next() were called).
next_literal_one_byte_string()550   Vector<const uint8_t> next_literal_one_byte_string() {
551     ASSERT_NOT_NULL(next_.literal_chars);
552     return next_.literal_chars->one_byte_literal();
553   }
next_literal_two_byte_string()554   Vector<const uint16_t> next_literal_two_byte_string() {
555     ASSERT_NOT_NULL(next_.literal_chars);
556     return next_.literal_chars->two_byte_literal();
557   }
is_next_literal_one_byte()558   bool is_next_literal_one_byte() {
559     ASSERT_NOT_NULL(next_.literal_chars);
560     return next_.literal_chars->is_one_byte();
561   }
next_literal_length()562   int next_literal_length() const {
563     ASSERT_NOT_NULL(next_.literal_chars);
564     return next_.literal_chars->length();
565   }
566 
567   uc32 ScanHexNumber(int expected_length);
568 
569   // Scans a single JavaScript token.
570   void Scan();
571 
572   bool SkipWhiteSpace();
573   Token::Value SkipSingleLineComment();
574   Token::Value SkipMultiLineComment();
575   // Scans a possible HTML comment -- begins with '<!'.
576   Token::Value ScanHtmlComment();
577 
578   void ScanDecimalDigits();
579   Token::Value ScanNumber(bool seen_period);
580   Token::Value ScanIdentifierOrKeyword();
581   Token::Value ScanIdentifierSuffix(LiteralScope* literal);
582 
583   Token::Value ScanString();
584 
585   // Scans an escape-sequence which is part of a string and adds the
586   // decoded character to the current literal. Returns true if a pattern
587   // is scanned.
588   bool ScanEscape();
589   // Decodes a Unicode escape-sequence which is part of an identifier.
590   // If the escape sequence cannot be decoded the result is kBadChar.
591   uc32 ScanIdentifierUnicodeEscape();
592   // Scans a Unicode escape-sequence and adds its characters,
593   // uninterpreted, to the current literal. Used for parsing RegExp
594   // flags.
595   bool ScanLiteralUnicodeEscape();
596 
597   // Return the current source position.
source_pos()598   int source_pos() {
599     return source_->pos() - kCharacterLookaheadBufferSize;
600   }
601 
602   UnicodeCache* unicode_cache_;
603 
604   // Buffers collecting literal strings, numbers, etc.
605   LiteralBuffer literal_buffer1_;
606   LiteralBuffer literal_buffer2_;
607 
608   TokenDesc current_;  // desc for current token (as returned by Next())
609   TokenDesc next_;     // desc for next token (one token look-ahead)
610 
611   // Input stream. Must be initialized to an Utf16CharacterStream.
612   Utf16CharacterStream* source_;
613 
614 
615   // Start position of the octal literal last scanned.
616   Location octal_pos_;
617 
618   // One Unicode character look-ahead; c0_ < 0 at the end of the input.
619   uc32 c0_;
620 
621   // Whether there is a line terminator whitespace character after
622   // the current token, and  before the next. Does not count newlines
623   // inside multiline comments.
624   bool has_line_terminator_before_next_;
625   // Whether there is a multi-line comment that contains a
626   // line-terminator after the current token, and before the next.
627   bool has_multiline_comment_before_next_;
628   // Whether we scan 'let' as a keyword for harmony block-scoped let bindings.
629   bool harmony_scoping_;
630   // Whether we scan 'module', 'import', 'export' as keywords.
631   bool harmony_modules_;
632   // Whether we scan 0o777 and 0b111 as numbers.
633   bool harmony_numeric_literals_;
634 };
635 
636 } }  // namespace v8::internal
637 
638 #endif  // V8_SCANNER_H_
639