1// Copyright 2012 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#include "src/regexp/jsregexp.h"
6
7#include "src/ast/ast.h"
8#include "src/base/platform/platform.h"
9#include "src/compilation-cache.h"
10#include "src/compiler.h"
11#include "src/execution.h"
12#include "src/factory.h"
13#include "src/isolate-inl.h"
14#include "src/messages.h"
15#include "src/ostreams.h"
16#include "src/regexp/interpreter-irregexp.h"
17#include "src/regexp/jsregexp-inl.h"
18#include "src/regexp/regexp-macro-assembler.h"
19#include "src/regexp/regexp-macro-assembler-irregexp.h"
20#include "src/regexp/regexp-macro-assembler-tracer.h"
21#include "src/regexp/regexp-parser.h"
22#include "src/regexp/regexp-stack.h"
23#include "src/runtime/runtime.h"
24#include "src/splay-tree-inl.h"
25#include "src/string-search.h"
26#include "src/unicode-decoder.h"
27
28#ifdef V8_I18N_SUPPORT
29#include "unicode/uset.h"
30#include "unicode/utypes.h"
31#endif  // V8_I18N_SUPPORT
32
33#ifndef V8_INTERPRETED_REGEXP
34#if V8_TARGET_ARCH_IA32
35#include "src/regexp/ia32/regexp-macro-assembler-ia32.h"
36#elif V8_TARGET_ARCH_X64
37#include "src/regexp/x64/regexp-macro-assembler-x64.h"
38#elif V8_TARGET_ARCH_ARM64
39#include "src/regexp/arm64/regexp-macro-assembler-arm64.h"
40#elif V8_TARGET_ARCH_ARM
41#include "src/regexp/arm/regexp-macro-assembler-arm.h"
42#elif V8_TARGET_ARCH_PPC
43#include "src/regexp/ppc/regexp-macro-assembler-ppc.h"
44#elif V8_TARGET_ARCH_S390
45#include "src/regexp/s390/regexp-macro-assembler-s390.h"
46#elif V8_TARGET_ARCH_MIPS
47#include "src/regexp/mips/regexp-macro-assembler-mips.h"
48#elif V8_TARGET_ARCH_MIPS64
49#include "src/regexp/mips64/regexp-macro-assembler-mips64.h"
50#elif V8_TARGET_ARCH_X87
51#include "src/regexp/x87/regexp-macro-assembler-x87.h"
52#else
53#error Unsupported target architecture.
54#endif
55#endif
56
57
58namespace v8 {
59namespace internal {
60
61MUST_USE_RESULT
62static inline MaybeHandle<Object> ThrowRegExpException(
63    Handle<JSRegExp> re, Handle<String> pattern, Handle<String> error_text) {
64  Isolate* isolate = re->GetIsolate();
65  THROW_NEW_ERROR(isolate, NewSyntaxError(MessageTemplate::kMalformedRegExp,
66                                          pattern, error_text),
67                  Object);
68}
69
70
71inline void ThrowRegExpException(Handle<JSRegExp> re,
72                                 Handle<String> error_text) {
73  USE(ThrowRegExpException(re, Handle<String>(re->Pattern()), error_text));
74}
75
76
77ContainedInLattice AddRange(ContainedInLattice containment,
78                            const int* ranges,
79                            int ranges_length,
80                            Interval new_range) {
81  DCHECK((ranges_length & 1) == 1);
82  DCHECK(ranges[ranges_length - 1] == String::kMaxCodePoint + 1);
83  if (containment == kLatticeUnknown) return containment;
84  bool inside = false;
85  int last = 0;
86  for (int i = 0; i < ranges_length; inside = !inside, last = ranges[i], i++) {
87    // Consider the range from last to ranges[i].
88    // We haven't got to the new range yet.
89    if (ranges[i] <= new_range.from()) continue;
90    // New range is wholly inside last-ranges[i].  Note that new_range.to() is
91    // inclusive, but the values in ranges are not.
92    if (last <= new_range.from() && new_range.to() < ranges[i]) {
93      return Combine(containment, inside ? kLatticeIn : kLatticeOut);
94    }
95    return kLatticeUnknown;
96  }
97  return containment;
98}
99
100
101// More makes code generation slower, less makes V8 benchmark score lower.
102const int kMaxLookaheadForBoyerMoore = 8;
103// In a 3-character pattern you can maximally step forwards 3 characters
104// at a time, which is not always enough to pay for the extra logic.
105const int kPatternTooShortForBoyerMoore = 2;
106
107
108// Identifies the sort of regexps where the regexp engine is faster
109// than the code used for atom matches.
110static bool HasFewDifferentCharacters(Handle<String> pattern) {
111  int length = Min(kMaxLookaheadForBoyerMoore, pattern->length());
112  if (length <= kPatternTooShortForBoyerMoore) return false;
113  const int kMod = 128;
114  bool character_found[kMod];
115  int different = 0;
116  memset(&character_found[0], 0, sizeof(character_found));
117  for (int i = 0; i < length; i++) {
118    int ch = (pattern->Get(i) & (kMod - 1));
119    if (!character_found[ch]) {
120      character_found[ch] = true;
121      different++;
122      // We declare a regexp low-alphabet if it has at least 3 times as many
123      // characters as it has different characters.
124      if (different * 3 > length) return false;
125    }
126  }
127  return true;
128}
129
130
131// Generic RegExp methods. Dispatches to implementation specific methods.
132
133
134MaybeHandle<Object> RegExpImpl::Compile(Handle<JSRegExp> re,
135                                        Handle<String> pattern,
136                                        JSRegExp::Flags flags) {
137  Isolate* isolate = re->GetIsolate();
138  Zone zone(isolate->allocator());
139  CompilationCache* compilation_cache = isolate->compilation_cache();
140  MaybeHandle<FixedArray> maybe_cached =
141      compilation_cache->LookupRegExp(pattern, flags);
142  Handle<FixedArray> cached;
143  bool in_cache = maybe_cached.ToHandle(&cached);
144  LOG(isolate, RegExpCompileEvent(re, in_cache));
145
146  Handle<Object> result;
147  if (in_cache) {
148    re->set_data(*cached);
149    return re;
150  }
151  pattern = String::Flatten(pattern);
152  PostponeInterruptsScope postpone(isolate);
153  RegExpCompileData parse_result;
154  FlatStringReader reader(isolate, pattern);
155  if (!RegExpParser::ParseRegExp(re->GetIsolate(), &zone, &reader, flags,
156                                 &parse_result)) {
157    // Throw an exception if we fail to parse the pattern.
158    return ThrowRegExpException(re, pattern, parse_result.error);
159  }
160
161  bool has_been_compiled = false;
162
163  if (parse_result.simple && !(flags & JSRegExp::kIgnoreCase) &&
164      !(flags & JSRegExp::kSticky) && !HasFewDifferentCharacters(pattern)) {
165    // Parse-tree is a single atom that is equal to the pattern.
166    AtomCompile(re, pattern, flags, pattern);
167    has_been_compiled = true;
168  } else if (parse_result.tree->IsAtom() && !(flags & JSRegExp::kIgnoreCase) &&
169             !(flags & JSRegExp::kSticky) && parse_result.capture_count == 0) {
170    RegExpAtom* atom = parse_result.tree->AsAtom();
171    Vector<const uc16> atom_pattern = atom->data();
172    Handle<String> atom_string;
173    ASSIGN_RETURN_ON_EXCEPTION(
174        isolate, atom_string,
175        isolate->factory()->NewStringFromTwoByte(atom_pattern),
176        Object);
177    if (!HasFewDifferentCharacters(atom_string)) {
178      AtomCompile(re, pattern, flags, atom_string);
179      has_been_compiled = true;
180    }
181  }
182  if (!has_been_compiled) {
183    IrregexpInitialize(re, pattern, flags, parse_result.capture_count);
184  }
185  DCHECK(re->data()->IsFixedArray());
186  // Compilation succeeded so the data is set on the regexp
187  // and we can store it in the cache.
188  Handle<FixedArray> data(FixedArray::cast(re->data()));
189  compilation_cache->PutRegExp(pattern, flags, data);
190
191  return re;
192}
193
194
195MaybeHandle<Object> RegExpImpl::Exec(Handle<JSRegExp> regexp,
196                                     Handle<String> subject,
197                                     int index,
198                                     Handle<JSArray> last_match_info) {
199  switch (regexp->TypeTag()) {
200    case JSRegExp::ATOM:
201      return AtomExec(regexp, subject, index, last_match_info);
202    case JSRegExp::IRREGEXP: {
203      return IrregexpExec(regexp, subject, index, last_match_info);
204    }
205    default:
206      UNREACHABLE();
207      return MaybeHandle<Object>();
208  }
209}
210
211
212// RegExp Atom implementation: Simple string search using indexOf.
213
214
215void RegExpImpl::AtomCompile(Handle<JSRegExp> re,
216                             Handle<String> pattern,
217                             JSRegExp::Flags flags,
218                             Handle<String> match_pattern) {
219  re->GetIsolate()->factory()->SetRegExpAtomData(re,
220                                                 JSRegExp::ATOM,
221                                                 pattern,
222                                                 flags,
223                                                 match_pattern);
224}
225
226
227static void SetAtomLastCapture(FixedArray* array,
228                               String* subject,
229                               int from,
230                               int to) {
231  SealHandleScope shs(array->GetIsolate());
232  RegExpImpl::SetLastCaptureCount(array, 2);
233  RegExpImpl::SetLastSubject(array, subject);
234  RegExpImpl::SetLastInput(array, subject);
235  RegExpImpl::SetCapture(array, 0, from);
236  RegExpImpl::SetCapture(array, 1, to);
237}
238
239
240int RegExpImpl::AtomExecRaw(Handle<JSRegExp> regexp,
241                            Handle<String> subject,
242                            int index,
243                            int32_t* output,
244                            int output_size) {
245  Isolate* isolate = regexp->GetIsolate();
246
247  DCHECK(0 <= index);
248  DCHECK(index <= subject->length());
249
250  subject = String::Flatten(subject);
251  DisallowHeapAllocation no_gc;  // ensure vectors stay valid
252
253  String* needle = String::cast(regexp->DataAt(JSRegExp::kAtomPatternIndex));
254  int needle_len = needle->length();
255  DCHECK(needle->IsFlat());
256  DCHECK_LT(0, needle_len);
257
258  if (index + needle_len > subject->length()) {
259    return RegExpImpl::RE_FAILURE;
260  }
261
262  for (int i = 0; i < output_size; i += 2) {
263    String::FlatContent needle_content = needle->GetFlatContent();
264    String::FlatContent subject_content = subject->GetFlatContent();
265    DCHECK(needle_content.IsFlat());
266    DCHECK(subject_content.IsFlat());
267    // dispatch on type of strings
268    index =
269        (needle_content.IsOneByte()
270             ? (subject_content.IsOneByte()
271                    ? SearchString(isolate, subject_content.ToOneByteVector(),
272                                   needle_content.ToOneByteVector(), index)
273                    : SearchString(isolate, subject_content.ToUC16Vector(),
274                                   needle_content.ToOneByteVector(), index))
275             : (subject_content.IsOneByte()
276                    ? SearchString(isolate, subject_content.ToOneByteVector(),
277                                   needle_content.ToUC16Vector(), index)
278                    : SearchString(isolate, subject_content.ToUC16Vector(),
279                                   needle_content.ToUC16Vector(), index)));
280    if (index == -1) {
281      return i / 2;  // Return number of matches.
282    } else {
283      output[i] = index;
284      output[i+1] = index + needle_len;
285      index += needle_len;
286    }
287  }
288  return output_size / 2;
289}
290
291
292Handle<Object> RegExpImpl::AtomExec(Handle<JSRegExp> re,
293                                    Handle<String> subject,
294                                    int index,
295                                    Handle<JSArray> last_match_info) {
296  Isolate* isolate = re->GetIsolate();
297
298  static const int kNumRegisters = 2;
299  STATIC_ASSERT(kNumRegisters <= Isolate::kJSRegexpStaticOffsetsVectorSize);
300  int32_t* output_registers = isolate->jsregexp_static_offsets_vector();
301
302  int res = AtomExecRaw(re, subject, index, output_registers, kNumRegisters);
303
304  if (res == RegExpImpl::RE_FAILURE) return isolate->factory()->null_value();
305
306  DCHECK_EQ(res, RegExpImpl::RE_SUCCESS);
307  SealHandleScope shs(isolate);
308  FixedArray* array = FixedArray::cast(last_match_info->elements());
309  SetAtomLastCapture(array, *subject, output_registers[0], output_registers[1]);
310  return last_match_info;
311}
312
313
314// Irregexp implementation.
315
316// Ensures that the regexp object contains a compiled version of the
317// source for either one-byte or two-byte subject strings.
318// If the compiled version doesn't already exist, it is compiled
319// from the source pattern.
320// If compilation fails, an exception is thrown and this function
321// returns false.
322bool RegExpImpl::EnsureCompiledIrregexp(Handle<JSRegExp> re,
323                                        Handle<String> sample_subject,
324                                        bool is_one_byte) {
325  Object* compiled_code = re->DataAt(JSRegExp::code_index(is_one_byte));
326#ifdef V8_INTERPRETED_REGEXP
327  if (compiled_code->IsByteArray()) return true;
328#else  // V8_INTERPRETED_REGEXP (RegExp native code)
329  if (compiled_code->IsCode()) return true;
330#endif
331  // We could potentially have marked this as flushable, but have kept
332  // a saved version if we did not flush it yet.
333  Object* saved_code = re->DataAt(JSRegExp::saved_code_index(is_one_byte));
334  if (saved_code->IsCode()) {
335    // Reinstate the code in the original place.
336    re->SetDataAt(JSRegExp::code_index(is_one_byte), saved_code);
337    DCHECK(compiled_code->IsSmi());
338    return true;
339  }
340  return CompileIrregexp(re, sample_subject, is_one_byte);
341}
342
343
344bool RegExpImpl::CompileIrregexp(Handle<JSRegExp> re,
345                                 Handle<String> sample_subject,
346                                 bool is_one_byte) {
347  // Compile the RegExp.
348  Isolate* isolate = re->GetIsolate();
349  Zone zone(isolate->allocator());
350  PostponeInterruptsScope postpone(isolate);
351  // If we had a compilation error the last time this is saved at the
352  // saved code index.
353  Object* entry = re->DataAt(JSRegExp::code_index(is_one_byte));
354  // When arriving here entry can only be a smi, either representing an
355  // uncompiled regexp, a previous compilation error, or code that has
356  // been flushed.
357  DCHECK(entry->IsSmi());
358  int entry_value = Smi::cast(entry)->value();
359  DCHECK(entry_value == JSRegExp::kUninitializedValue ||
360         entry_value == JSRegExp::kCompilationErrorValue ||
361         (entry_value < JSRegExp::kCodeAgeMask && entry_value >= 0));
362
363  if (entry_value == JSRegExp::kCompilationErrorValue) {
364    // A previous compilation failed and threw an error which we store in
365    // the saved code index (we store the error message, not the actual
366    // error). Recreate the error object and throw it.
367    Object* error_string = re->DataAt(JSRegExp::saved_code_index(is_one_byte));
368    DCHECK(error_string->IsString());
369    Handle<String> error_message(String::cast(error_string));
370    ThrowRegExpException(re, error_message);
371    return false;
372  }
373
374  JSRegExp::Flags flags = re->GetFlags();
375
376  Handle<String> pattern(re->Pattern());
377  pattern = String::Flatten(pattern);
378  RegExpCompileData compile_data;
379  FlatStringReader reader(isolate, pattern);
380  if (!RegExpParser::ParseRegExp(isolate, &zone, &reader, flags,
381                                 &compile_data)) {
382    // Throw an exception if we fail to parse the pattern.
383    // THIS SHOULD NOT HAPPEN. We already pre-parsed it successfully once.
384    USE(ThrowRegExpException(re, pattern, compile_data.error));
385    return false;
386  }
387  RegExpEngine::CompilationResult result =
388      RegExpEngine::Compile(isolate, &zone, &compile_data, flags, pattern,
389                            sample_subject, is_one_byte);
390  if (result.error_message != NULL) {
391    // Unable to compile regexp.
392    Handle<String> error_message = isolate->factory()->NewStringFromUtf8(
393        CStrVector(result.error_message)).ToHandleChecked();
394    ThrowRegExpException(re, error_message);
395    return false;
396  }
397
398  Handle<FixedArray> data = Handle<FixedArray>(FixedArray::cast(re->data()));
399  data->set(JSRegExp::code_index(is_one_byte), result.code);
400  SetIrregexpCaptureNameMap(*data, compile_data.capture_name_map);
401  int register_max = IrregexpMaxRegisterCount(*data);
402  if (result.num_registers > register_max) {
403    SetIrregexpMaxRegisterCount(*data, result.num_registers);
404  }
405
406  return true;
407}
408
409
410int RegExpImpl::IrregexpMaxRegisterCount(FixedArray* re) {
411  return Smi::cast(
412      re->get(JSRegExp::kIrregexpMaxRegisterCountIndex))->value();
413}
414
415
416void RegExpImpl::SetIrregexpMaxRegisterCount(FixedArray* re, int value) {
417  re->set(JSRegExp::kIrregexpMaxRegisterCountIndex, Smi::FromInt(value));
418}
419
420void RegExpImpl::SetIrregexpCaptureNameMap(FixedArray* re,
421                                           Handle<FixedArray> value) {
422  if (value.is_null()) {
423    re->set(JSRegExp::kIrregexpCaptureNameMapIndex, Smi::FromInt(0));
424  } else {
425    re->set(JSRegExp::kIrregexpCaptureNameMapIndex, *value);
426  }
427}
428
429int RegExpImpl::IrregexpNumberOfCaptures(FixedArray* re) {
430  return Smi::cast(re->get(JSRegExp::kIrregexpCaptureCountIndex))->value();
431}
432
433
434int RegExpImpl::IrregexpNumberOfRegisters(FixedArray* re) {
435  return Smi::cast(re->get(JSRegExp::kIrregexpMaxRegisterCountIndex))->value();
436}
437
438
439ByteArray* RegExpImpl::IrregexpByteCode(FixedArray* re, bool is_one_byte) {
440  return ByteArray::cast(re->get(JSRegExp::code_index(is_one_byte)));
441}
442
443
444Code* RegExpImpl::IrregexpNativeCode(FixedArray* re, bool is_one_byte) {
445  return Code::cast(re->get(JSRegExp::code_index(is_one_byte)));
446}
447
448
449void RegExpImpl::IrregexpInitialize(Handle<JSRegExp> re,
450                                    Handle<String> pattern,
451                                    JSRegExp::Flags flags,
452                                    int capture_count) {
453  // Initialize compiled code entries to null.
454  re->GetIsolate()->factory()->SetRegExpIrregexpData(re,
455                                                     JSRegExp::IRREGEXP,
456                                                     pattern,
457                                                     flags,
458                                                     capture_count);
459}
460
461
462int RegExpImpl::IrregexpPrepare(Handle<JSRegExp> regexp,
463                                Handle<String> subject) {
464  subject = String::Flatten(subject);
465
466  // Check representation of the underlying storage.
467  bool is_one_byte = subject->IsOneByteRepresentationUnderneath();
468  if (!EnsureCompiledIrregexp(regexp, subject, is_one_byte)) return -1;
469
470#ifdef V8_INTERPRETED_REGEXP
471  // Byte-code regexp needs space allocated for all its registers.
472  // The result captures are copied to the start of the registers array
473  // if the match succeeds.  This way those registers are not clobbered
474  // when we set the last match info from last successful match.
475  return IrregexpNumberOfRegisters(FixedArray::cast(regexp->data())) +
476         (IrregexpNumberOfCaptures(FixedArray::cast(regexp->data())) + 1) * 2;
477#else  // V8_INTERPRETED_REGEXP
478  // Native regexp only needs room to output captures. Registers are handled
479  // internally.
480  return (IrregexpNumberOfCaptures(FixedArray::cast(regexp->data())) + 1) * 2;
481#endif  // V8_INTERPRETED_REGEXP
482}
483
484
485int RegExpImpl::IrregexpExecRaw(Handle<JSRegExp> regexp,
486                                Handle<String> subject,
487                                int index,
488                                int32_t* output,
489                                int output_size) {
490  Isolate* isolate = regexp->GetIsolate();
491
492  Handle<FixedArray> irregexp(FixedArray::cast(regexp->data()), isolate);
493
494  DCHECK(index >= 0);
495  DCHECK(index <= subject->length());
496  DCHECK(subject->IsFlat());
497
498  bool is_one_byte = subject->IsOneByteRepresentationUnderneath();
499
500#ifndef V8_INTERPRETED_REGEXP
501  DCHECK(output_size >= (IrregexpNumberOfCaptures(*irregexp) + 1) * 2);
502  do {
503    EnsureCompiledIrregexp(regexp, subject, is_one_byte);
504    Handle<Code> code(IrregexpNativeCode(*irregexp, is_one_byte), isolate);
505    // The stack is used to allocate registers for the compiled regexp code.
506    // This means that in case of failure, the output registers array is left
507    // untouched and contains the capture results from the previous successful
508    // match.  We can use that to set the last match info lazily.
509    NativeRegExpMacroAssembler::Result res =
510        NativeRegExpMacroAssembler::Match(code,
511                                          subject,
512                                          output,
513                                          output_size,
514                                          index,
515                                          isolate);
516    if (res != NativeRegExpMacroAssembler::RETRY) {
517      DCHECK(res != NativeRegExpMacroAssembler::EXCEPTION ||
518             isolate->has_pending_exception());
519      STATIC_ASSERT(
520          static_cast<int>(NativeRegExpMacroAssembler::SUCCESS) == RE_SUCCESS);
521      STATIC_ASSERT(
522          static_cast<int>(NativeRegExpMacroAssembler::FAILURE) == RE_FAILURE);
523      STATIC_ASSERT(static_cast<int>(NativeRegExpMacroAssembler::EXCEPTION)
524                    == RE_EXCEPTION);
525      return static_cast<IrregexpResult>(res);
526    }
527    // If result is RETRY, the string has changed representation, and we
528    // must restart from scratch.
529    // In this case, it means we must make sure we are prepared to handle
530    // the, potentially, different subject (the string can switch between
531    // being internal and external, and even between being Latin1 and UC16,
532    // but the characters are always the same).
533    IrregexpPrepare(regexp, subject);
534    is_one_byte = subject->IsOneByteRepresentationUnderneath();
535  } while (true);
536  UNREACHABLE();
537  return RE_EXCEPTION;
538#else  // V8_INTERPRETED_REGEXP
539
540  DCHECK(output_size >= IrregexpNumberOfRegisters(*irregexp));
541  // We must have done EnsureCompiledIrregexp, so we can get the number of
542  // registers.
543  int number_of_capture_registers =
544      (IrregexpNumberOfCaptures(*irregexp) + 1) * 2;
545  int32_t* raw_output = &output[number_of_capture_registers];
546  // We do not touch the actual capture result registers until we know there
547  // has been a match so that we can use those capture results to set the
548  // last match info.
549  for (int i = number_of_capture_registers - 1; i >= 0; i--) {
550    raw_output[i] = -1;
551  }
552  Handle<ByteArray> byte_codes(IrregexpByteCode(*irregexp, is_one_byte),
553                               isolate);
554
555  IrregexpResult result = IrregexpInterpreter::Match(isolate,
556                                                     byte_codes,
557                                                     subject,
558                                                     raw_output,
559                                                     index);
560  if (result == RE_SUCCESS) {
561    // Copy capture results to the start of the registers array.
562    MemCopy(output, raw_output, number_of_capture_registers * sizeof(int32_t));
563  }
564  if (result == RE_EXCEPTION) {
565    DCHECK(!isolate->has_pending_exception());
566    isolate->StackOverflow();
567  }
568  return result;
569#endif  // V8_INTERPRETED_REGEXP
570}
571
572
573MaybeHandle<Object> RegExpImpl::IrregexpExec(Handle<JSRegExp> regexp,
574                                             Handle<String> subject,
575                                             int previous_index,
576                                             Handle<JSArray> last_match_info) {
577  Isolate* isolate = regexp->GetIsolate();
578  DCHECK_EQ(regexp->TypeTag(), JSRegExp::IRREGEXP);
579
580  // Prepare space for the return values.
581#if defined(V8_INTERPRETED_REGEXP) && defined(DEBUG)
582  if (FLAG_trace_regexp_bytecodes) {
583    String* pattern = regexp->Pattern();
584    PrintF("\n\nRegexp match:   /%s/\n\n", pattern->ToCString().get());
585    PrintF("\n\nSubject string: '%s'\n\n", subject->ToCString().get());
586  }
587#endif
588  int required_registers = RegExpImpl::IrregexpPrepare(regexp, subject);
589  if (required_registers < 0) {
590    // Compiling failed with an exception.
591    DCHECK(isolate->has_pending_exception());
592    return MaybeHandle<Object>();
593  }
594
595  int32_t* output_registers = NULL;
596  if (required_registers > Isolate::kJSRegexpStaticOffsetsVectorSize) {
597    output_registers = NewArray<int32_t>(required_registers);
598  }
599  base::SmartArrayPointer<int32_t> auto_release(output_registers);
600  if (output_registers == NULL) {
601    output_registers = isolate->jsregexp_static_offsets_vector();
602  }
603
604  int res = RegExpImpl::IrregexpExecRaw(
605      regexp, subject, previous_index, output_registers, required_registers);
606  if (res == RE_SUCCESS) {
607    int capture_count =
608        IrregexpNumberOfCaptures(FixedArray::cast(regexp->data()));
609    return SetLastMatchInfo(
610        last_match_info, subject, capture_count, output_registers);
611  }
612  if (res == RE_EXCEPTION) {
613    DCHECK(isolate->has_pending_exception());
614    return MaybeHandle<Object>();
615  }
616  DCHECK(res == RE_FAILURE);
617  return isolate->factory()->null_value();
618}
619
620
621static void EnsureSize(Handle<JSArray> array, uint32_t minimum_size) {
622  if (static_cast<uint32_t>(array->elements()->length()) < minimum_size) {
623    JSArray::SetLength(array, minimum_size);
624  }
625}
626
627
628Handle<JSArray> RegExpImpl::SetLastMatchInfo(Handle<JSArray> last_match_info,
629                                             Handle<String> subject,
630                                             int capture_count,
631                                             int32_t* match) {
632  DCHECK(last_match_info->HasFastObjectElements());
633  int capture_register_count = (capture_count + 1) * 2;
634  EnsureSize(last_match_info, capture_register_count + kLastMatchOverhead);
635  DisallowHeapAllocation no_allocation;
636  FixedArray* array = FixedArray::cast(last_match_info->elements());
637  if (match != NULL) {
638    for (int i = 0; i < capture_register_count; i += 2) {
639      SetCapture(array, i, match[i]);
640      SetCapture(array, i + 1, match[i + 1]);
641    }
642  }
643  SetLastCaptureCount(array, capture_register_count);
644  SetLastSubject(array, *subject);
645  SetLastInput(array, *subject);
646  return last_match_info;
647}
648
649
650RegExpImpl::GlobalCache::GlobalCache(Handle<JSRegExp> regexp,
651                                     Handle<String> subject,
652                                     Isolate* isolate)
653  : register_array_(NULL),
654    register_array_size_(0),
655    regexp_(regexp),
656    subject_(subject) {
657#ifdef V8_INTERPRETED_REGEXP
658  bool interpreted = true;
659#else
660  bool interpreted = false;
661#endif  // V8_INTERPRETED_REGEXP
662
663  if (regexp_->TypeTag() == JSRegExp::ATOM) {
664    static const int kAtomRegistersPerMatch = 2;
665    registers_per_match_ = kAtomRegistersPerMatch;
666    // There is no distinction between interpreted and native for atom regexps.
667    interpreted = false;
668  } else {
669    registers_per_match_ = RegExpImpl::IrregexpPrepare(regexp_, subject_);
670    if (registers_per_match_ < 0) {
671      num_matches_ = -1;  // Signal exception.
672      return;
673    }
674  }
675
676  DCHECK_NE(0, regexp->GetFlags() & JSRegExp::kGlobal);
677  if (!interpreted) {
678    register_array_size_ =
679        Max(registers_per_match_, Isolate::kJSRegexpStaticOffsetsVectorSize);
680    max_matches_ = register_array_size_ / registers_per_match_;
681  } else {
682    // Global loop in interpreted regexp is not implemented.  We choose
683    // the size of the offsets vector so that it can only store one match.
684    register_array_size_ = registers_per_match_;
685    max_matches_ = 1;
686  }
687
688  if (register_array_size_ > Isolate::kJSRegexpStaticOffsetsVectorSize) {
689    register_array_ = NewArray<int32_t>(register_array_size_);
690  } else {
691    register_array_ = isolate->jsregexp_static_offsets_vector();
692  }
693
694  // Set state so that fetching the results the first time triggers a call
695  // to the compiled regexp.
696  current_match_index_ = max_matches_ - 1;
697  num_matches_ = max_matches_;
698  DCHECK(registers_per_match_ >= 2);  // Each match has at least one capture.
699  DCHECK_GE(register_array_size_, registers_per_match_);
700  int32_t* last_match =
701      &register_array_[current_match_index_ * registers_per_match_];
702  last_match[0] = -1;
703  last_match[1] = 0;
704}
705
706int RegExpImpl::GlobalCache::AdvanceZeroLength(int last_index) {
707  if ((regexp_->GetFlags() & JSRegExp::kUnicode) != 0 &&
708      last_index + 1 < subject_->length() &&
709      unibrow::Utf16::IsLeadSurrogate(subject_->Get(last_index)) &&
710      unibrow::Utf16::IsTrailSurrogate(subject_->Get(last_index + 1))) {
711    // Advance over the surrogate pair.
712    return last_index + 2;
713  }
714  return last_index + 1;
715}
716
717// -------------------------------------------------------------------
718// Implementation of the Irregexp regular expression engine.
719//
720// The Irregexp regular expression engine is intended to be a complete
721// implementation of ECMAScript regular expressions.  It generates either
722// bytecodes or native code.
723
724//   The Irregexp regexp engine is structured in three steps.
725//   1) The parser generates an abstract syntax tree.  See ast.cc.
726//   2) From the AST a node network is created.  The nodes are all
727//      subclasses of RegExpNode.  The nodes represent states when
728//      executing a regular expression.  Several optimizations are
729//      performed on the node network.
730//   3) From the nodes we generate either byte codes or native code
731//      that can actually execute the regular expression (perform
732//      the search).  The code generation step is described in more
733//      detail below.
734
735// Code generation.
736//
737//   The nodes are divided into four main categories.
738//   * Choice nodes
739//        These represent places where the regular expression can
740//        match in more than one way.  For example on entry to an
741//        alternation (foo|bar) or a repetition (*, +, ? or {}).
742//   * Action nodes
743//        These represent places where some action should be
744//        performed.  Examples include recording the current position
745//        in the input string to a register (in order to implement
746//        captures) or other actions on register for example in order
747//        to implement the counters needed for {} repetitions.
748//   * Matching nodes
749//        These attempt to match some element part of the input string.
750//        Examples of elements include character classes, plain strings
751//        or back references.
752//   * End nodes
753//        These are used to implement the actions required on finding
754//        a successful match or failing to find a match.
755//
756//   The code generated (whether as byte codes or native code) maintains
757//   some state as it runs.  This consists of the following elements:
758//
759//   * The capture registers.  Used for string captures.
760//   * Other registers.  Used for counters etc.
761//   * The current position.
762//   * The stack of backtracking information.  Used when a matching node
763//     fails to find a match and needs to try an alternative.
764//
765// Conceptual regular expression execution model:
766//
767//   There is a simple conceptual model of regular expression execution
768//   which will be presented first.  The actual code generated is a more
769//   efficient simulation of the simple conceptual model:
770//
771//   * Choice nodes are implemented as follows:
772//     For each choice except the last {
773//       push current position
774//       push backtrack code location
775//       <generate code to test for choice>
776//       backtrack code location:
777//       pop current position
778//     }
779//     <generate code to test for last choice>
780//
781//   * Actions nodes are generated as follows
782//     <push affected registers on backtrack stack>
783//     <generate code to perform action>
784//     push backtrack code location
785//     <generate code to test for following nodes>
786//     backtrack code location:
787//     <pop affected registers to restore their state>
788//     <pop backtrack location from stack and go to it>
789//
790//   * Matching nodes are generated as follows:
791//     if input string matches at current position
792//       update current position
793//       <generate code to test for following nodes>
794//     else
795//       <pop backtrack location from stack and go to it>
796//
797//   Thus it can be seen that the current position is saved and restored
798//   by the choice nodes, whereas the registers are saved and restored by
799//   by the action nodes that manipulate them.
800//
801//   The other interesting aspect of this model is that nodes are generated
802//   at the point where they are needed by a recursive call to Emit().  If
803//   the node has already been code generated then the Emit() call will
804//   generate a jump to the previously generated code instead.  In order to
805//   limit recursion it is possible for the Emit() function to put the node
806//   on a work list for later generation and instead generate a jump.  The
807//   destination of the jump is resolved later when the code is generated.
808//
809// Actual regular expression code generation.
810//
811//   Code generation is actually more complicated than the above.  In order
812//   to improve the efficiency of the generated code some optimizations are
813//   performed
814//
815//   * Choice nodes have 1-character lookahead.
816//     A choice node looks at the following character and eliminates some of
817//     the choices immediately based on that character.  This is not yet
818//     implemented.
819//   * Simple greedy loops store reduced backtracking information.
820//     A quantifier like /.*foo/m will greedily match the whole input.  It will
821//     then need to backtrack to a point where it can match "foo".  The naive
822//     implementation of this would push each character position onto the
823//     backtracking stack, then pop them off one by one.  This would use space
824//     proportional to the length of the input string.  However since the "."
825//     can only match in one way and always has a constant length (in this case
826//     of 1) it suffices to store the current position on the top of the stack
827//     once.  Matching now becomes merely incrementing the current position and
828//     backtracking becomes decrementing the current position and checking the
829//     result against the stored current position.  This is faster and saves
830//     space.
831//   * The current state is virtualized.
832//     This is used to defer expensive operations until it is clear that they
833//     are needed and to generate code for a node more than once, allowing
834//     specialized an efficient versions of the code to be created. This is
835//     explained in the section below.
836//
837// Execution state virtualization.
838//
839//   Instead of emitting code, nodes that manipulate the state can record their
840//   manipulation in an object called the Trace.  The Trace object can record a
841//   current position offset, an optional backtrack code location on the top of
842//   the virtualized backtrack stack and some register changes.  When a node is
843//   to be emitted it can flush the Trace or update it.  Flushing the Trace
844//   will emit code to bring the actual state into line with the virtual state.
845//   Avoiding flushing the state can postpone some work (e.g. updates of capture
846//   registers).  Postponing work can save time when executing the regular
847//   expression since it may be found that the work never has to be done as a
848//   failure to match can occur.  In addition it is much faster to jump to a
849//   known backtrack code location than it is to pop an unknown backtrack
850//   location from the stack and jump there.
851//
852//   The virtual state found in the Trace affects code generation.  For example
853//   the virtual state contains the difference between the actual current
854//   position and the virtual current position, and matching code needs to use
855//   this offset to attempt a match in the correct location of the input
856//   string.  Therefore code generated for a non-trivial trace is specialized
857//   to that trace.  The code generator therefore has the ability to generate
858//   code for each node several times.  In order to limit the size of the
859//   generated code there is an arbitrary limit on how many specialized sets of
860//   code may be generated for a given node.  If the limit is reached, the
861//   trace is flushed and a generic version of the code for a node is emitted.
862//   This is subsequently used for that node.  The code emitted for non-generic
863//   trace is not recorded in the node and so it cannot currently be reused in
864//   the event that code generation is requested for an identical trace.
865
866
867void RegExpTree::AppendToText(RegExpText* text, Zone* zone) {
868  UNREACHABLE();
869}
870
871
872void RegExpAtom::AppendToText(RegExpText* text, Zone* zone) {
873  text->AddElement(TextElement::Atom(this), zone);
874}
875
876
877void RegExpCharacterClass::AppendToText(RegExpText* text, Zone* zone) {
878  text->AddElement(TextElement::CharClass(this), zone);
879}
880
881
882void RegExpText::AppendToText(RegExpText* text, Zone* zone) {
883  for (int i = 0; i < elements()->length(); i++)
884    text->AddElement(elements()->at(i), zone);
885}
886
887
888TextElement TextElement::Atom(RegExpAtom* atom) {
889  return TextElement(ATOM, atom);
890}
891
892
893TextElement TextElement::CharClass(RegExpCharacterClass* char_class) {
894  return TextElement(CHAR_CLASS, char_class);
895}
896
897
898int TextElement::length() const {
899  switch (text_type()) {
900    case ATOM:
901      return atom()->length();
902
903    case CHAR_CLASS:
904      return 1;
905  }
906  UNREACHABLE();
907  return 0;
908}
909
910
911DispatchTable* ChoiceNode::GetTable(bool ignore_case) {
912  if (table_ == NULL) {
913    table_ = new(zone()) DispatchTable(zone());
914    DispatchTableConstructor cons(table_, ignore_case, zone());
915    cons.BuildTable(this);
916  }
917  return table_;
918}
919
920
921class FrequencyCollator {
922 public:
923  FrequencyCollator() : total_samples_(0) {
924    for (int i = 0; i < RegExpMacroAssembler::kTableSize; i++) {
925      frequencies_[i] = CharacterFrequency(i);
926    }
927  }
928
929  void CountCharacter(int character) {
930    int index = (character & RegExpMacroAssembler::kTableMask);
931    frequencies_[index].Increment();
932    total_samples_++;
933  }
934
935  // Does not measure in percent, but rather per-128 (the table size from the
936  // regexp macro assembler).
937  int Frequency(int in_character) {
938    DCHECK((in_character & RegExpMacroAssembler::kTableMask) == in_character);
939    if (total_samples_ < 1) return 1;  // Division by zero.
940    int freq_in_per128 =
941        (frequencies_[in_character].counter() * 128) / total_samples_;
942    return freq_in_per128;
943  }
944
945 private:
946  class CharacterFrequency {
947   public:
948    CharacterFrequency() : counter_(0), character_(-1) { }
949    explicit CharacterFrequency(int character)
950        : counter_(0), character_(character) { }
951
952    void Increment() { counter_++; }
953    int counter() { return counter_; }
954    int character() { return character_; }
955
956   private:
957    int counter_;
958    int character_;
959  };
960
961
962 private:
963  CharacterFrequency frequencies_[RegExpMacroAssembler::kTableSize];
964  int total_samples_;
965};
966
967
968class RegExpCompiler {
969 public:
970  RegExpCompiler(Isolate* isolate, Zone* zone, int capture_count,
971                 JSRegExp::Flags flags, bool is_one_byte);
972
973  int AllocateRegister() {
974    if (next_register_ >= RegExpMacroAssembler::kMaxRegister) {
975      reg_exp_too_big_ = true;
976      return next_register_;
977    }
978    return next_register_++;
979  }
980
981  // Lookarounds to match lone surrogates for unicode character class matches
982  // are never nested. We can therefore reuse registers.
983  int UnicodeLookaroundStackRegister() {
984    if (unicode_lookaround_stack_register_ == kNoRegister) {
985      unicode_lookaround_stack_register_ = AllocateRegister();
986    }
987    return unicode_lookaround_stack_register_;
988  }
989
990  int UnicodeLookaroundPositionRegister() {
991    if (unicode_lookaround_position_register_ == kNoRegister) {
992      unicode_lookaround_position_register_ = AllocateRegister();
993    }
994    return unicode_lookaround_position_register_;
995  }
996
997  RegExpEngine::CompilationResult Assemble(RegExpMacroAssembler* assembler,
998                                           RegExpNode* start,
999                                           int capture_count,
1000                                           Handle<String> pattern);
1001
1002  inline void AddWork(RegExpNode* node) {
1003    if (!node->on_work_list() && !node->label()->is_bound()) {
1004      node->set_on_work_list(true);
1005      work_list_->Add(node);
1006    }
1007  }
1008
1009  static const int kImplementationOffset = 0;
1010  static const int kNumberOfRegistersOffset = 0;
1011  static const int kCodeOffset = 1;
1012
1013  RegExpMacroAssembler* macro_assembler() { return macro_assembler_; }
1014  EndNode* accept() { return accept_; }
1015
1016  static const int kMaxRecursion = 100;
1017  inline int recursion_depth() { return recursion_depth_; }
1018  inline void IncrementRecursionDepth() { recursion_depth_++; }
1019  inline void DecrementRecursionDepth() { recursion_depth_--; }
1020
1021  void SetRegExpTooBig() { reg_exp_too_big_ = true; }
1022
1023  inline bool ignore_case() { return (flags_ & JSRegExp::kIgnoreCase) != 0; }
1024  inline bool unicode() { return (flags_ & JSRegExp::kUnicode) != 0; }
1025  inline bool one_byte() { return one_byte_; }
1026  inline bool optimize() { return optimize_; }
1027  inline void set_optimize(bool value) { optimize_ = value; }
1028  inline bool limiting_recursion() { return limiting_recursion_; }
1029  inline void set_limiting_recursion(bool value) {
1030    limiting_recursion_ = value;
1031  }
1032  bool read_backward() { return read_backward_; }
1033  void set_read_backward(bool value) { read_backward_ = value; }
1034  FrequencyCollator* frequency_collator() { return &frequency_collator_; }
1035
1036  int current_expansion_factor() { return current_expansion_factor_; }
1037  void set_current_expansion_factor(int value) {
1038    current_expansion_factor_ = value;
1039  }
1040
1041  Isolate* isolate() const { return isolate_; }
1042  Zone* zone() const { return zone_; }
1043
1044  static const int kNoRegister = -1;
1045
1046 private:
1047  EndNode* accept_;
1048  int next_register_;
1049  int unicode_lookaround_stack_register_;
1050  int unicode_lookaround_position_register_;
1051  List<RegExpNode*>* work_list_;
1052  int recursion_depth_;
1053  RegExpMacroAssembler* macro_assembler_;
1054  JSRegExp::Flags flags_;
1055  bool one_byte_;
1056  bool reg_exp_too_big_;
1057  bool limiting_recursion_;
1058  bool optimize_;
1059  bool read_backward_;
1060  int current_expansion_factor_;
1061  FrequencyCollator frequency_collator_;
1062  Isolate* isolate_;
1063  Zone* zone_;
1064};
1065
1066
1067class RecursionCheck {
1068 public:
1069  explicit RecursionCheck(RegExpCompiler* compiler) : compiler_(compiler) {
1070    compiler->IncrementRecursionDepth();
1071  }
1072  ~RecursionCheck() { compiler_->DecrementRecursionDepth(); }
1073 private:
1074  RegExpCompiler* compiler_;
1075};
1076
1077
1078static RegExpEngine::CompilationResult IrregexpRegExpTooBig(Isolate* isolate) {
1079  return RegExpEngine::CompilationResult(isolate, "RegExp too big");
1080}
1081
1082
1083// Attempts to compile the regexp using an Irregexp code generator.  Returns
1084// a fixed array or a null handle depending on whether it succeeded.
1085RegExpCompiler::RegExpCompiler(Isolate* isolate, Zone* zone, int capture_count,
1086                               JSRegExp::Flags flags, bool one_byte)
1087    : next_register_(2 * (capture_count + 1)),
1088      unicode_lookaround_stack_register_(kNoRegister),
1089      unicode_lookaround_position_register_(kNoRegister),
1090      work_list_(NULL),
1091      recursion_depth_(0),
1092      flags_(flags),
1093      one_byte_(one_byte),
1094      reg_exp_too_big_(false),
1095      limiting_recursion_(false),
1096      optimize_(FLAG_regexp_optimization),
1097      read_backward_(false),
1098      current_expansion_factor_(1),
1099      frequency_collator_(),
1100      isolate_(isolate),
1101      zone_(zone) {
1102  accept_ = new(zone) EndNode(EndNode::ACCEPT, zone);
1103  DCHECK(next_register_ - 1 <= RegExpMacroAssembler::kMaxRegister);
1104}
1105
1106
1107RegExpEngine::CompilationResult RegExpCompiler::Assemble(
1108    RegExpMacroAssembler* macro_assembler,
1109    RegExpNode* start,
1110    int capture_count,
1111    Handle<String> pattern) {
1112  Heap* heap = pattern->GetHeap();
1113
1114#ifdef DEBUG
1115  if (FLAG_trace_regexp_assembler)
1116    macro_assembler_ =
1117        new RegExpMacroAssemblerTracer(isolate(), macro_assembler);
1118  else
1119#endif
1120    macro_assembler_ = macro_assembler;
1121
1122  List <RegExpNode*> work_list(0);
1123  work_list_ = &work_list;
1124  Label fail;
1125  macro_assembler_->PushBacktrack(&fail);
1126  Trace new_trace;
1127  start->Emit(this, &new_trace);
1128  macro_assembler_->Bind(&fail);
1129  macro_assembler_->Fail();
1130  while (!work_list.is_empty()) {
1131    RegExpNode* node = work_list.RemoveLast();
1132    node->set_on_work_list(false);
1133    if (!node->label()->is_bound()) node->Emit(this, &new_trace);
1134  }
1135  if (reg_exp_too_big_) {
1136    macro_assembler_->AbortedCodeGeneration();
1137    return IrregexpRegExpTooBig(isolate_);
1138  }
1139
1140  Handle<HeapObject> code = macro_assembler_->GetCode(pattern);
1141  heap->IncreaseTotalRegexpCodeGenerated(code->Size());
1142  work_list_ = NULL;
1143#ifdef ENABLE_DISASSEMBLER
1144  if (FLAG_print_code) {
1145    CodeTracer::Scope trace_scope(heap->isolate()->GetCodeTracer());
1146    OFStream os(trace_scope.file());
1147    Handle<Code>::cast(code)->Disassemble(pattern->ToCString().get(), os);
1148  }
1149#endif
1150#ifdef DEBUG
1151  if (FLAG_trace_regexp_assembler) {
1152    delete macro_assembler_;
1153  }
1154#endif
1155  return RegExpEngine::CompilationResult(*code, next_register_);
1156}
1157
1158
1159bool Trace::DeferredAction::Mentions(int that) {
1160  if (action_type() == ActionNode::CLEAR_CAPTURES) {
1161    Interval range = static_cast<DeferredClearCaptures*>(this)->range();
1162    return range.Contains(that);
1163  } else {
1164    return reg() == that;
1165  }
1166}
1167
1168
1169bool Trace::mentions_reg(int reg) {
1170  for (DeferredAction* action = actions_;
1171       action != NULL;
1172       action = action->next()) {
1173    if (action->Mentions(reg))
1174      return true;
1175  }
1176  return false;
1177}
1178
1179
1180bool Trace::GetStoredPosition(int reg, int* cp_offset) {
1181  DCHECK_EQ(0, *cp_offset);
1182  for (DeferredAction* action = actions_;
1183       action != NULL;
1184       action = action->next()) {
1185    if (action->Mentions(reg)) {
1186      if (action->action_type() == ActionNode::STORE_POSITION) {
1187        *cp_offset = static_cast<DeferredCapture*>(action)->cp_offset();
1188        return true;
1189      } else {
1190        return false;
1191      }
1192    }
1193  }
1194  return false;
1195}
1196
1197
1198int Trace::FindAffectedRegisters(OutSet* affected_registers,
1199                                 Zone* zone) {
1200  int max_register = RegExpCompiler::kNoRegister;
1201  for (DeferredAction* action = actions_;
1202       action != NULL;
1203       action = action->next()) {
1204    if (action->action_type() == ActionNode::CLEAR_CAPTURES) {
1205      Interval range = static_cast<DeferredClearCaptures*>(action)->range();
1206      for (int i = range.from(); i <= range.to(); i++)
1207        affected_registers->Set(i, zone);
1208      if (range.to() > max_register) max_register = range.to();
1209    } else {
1210      affected_registers->Set(action->reg(), zone);
1211      if (action->reg() > max_register) max_register = action->reg();
1212    }
1213  }
1214  return max_register;
1215}
1216
1217
1218void Trace::RestoreAffectedRegisters(RegExpMacroAssembler* assembler,
1219                                     int max_register,
1220                                     const OutSet& registers_to_pop,
1221                                     const OutSet& registers_to_clear) {
1222  for (int reg = max_register; reg >= 0; reg--) {
1223    if (registers_to_pop.Get(reg)) {
1224      assembler->PopRegister(reg);
1225    } else if (registers_to_clear.Get(reg)) {
1226      int clear_to = reg;
1227      while (reg > 0 && registers_to_clear.Get(reg - 1)) {
1228        reg--;
1229      }
1230      assembler->ClearRegisters(reg, clear_to);
1231    }
1232  }
1233}
1234
1235
1236void Trace::PerformDeferredActions(RegExpMacroAssembler* assembler,
1237                                   int max_register,
1238                                   const OutSet& affected_registers,
1239                                   OutSet* registers_to_pop,
1240                                   OutSet* registers_to_clear,
1241                                   Zone* zone) {
1242  // The "+1" is to avoid a push_limit of zero if stack_limit_slack() is 1.
1243  const int push_limit = (assembler->stack_limit_slack() + 1) / 2;
1244
1245  // Count pushes performed to force a stack limit check occasionally.
1246  int pushes = 0;
1247
1248  for (int reg = 0; reg <= max_register; reg++) {
1249    if (!affected_registers.Get(reg)) {
1250      continue;
1251    }
1252
1253    // The chronologically first deferred action in the trace
1254    // is used to infer the action needed to restore a register
1255    // to its previous state (or not, if it's safe to ignore it).
1256    enum DeferredActionUndoType { IGNORE, RESTORE, CLEAR };
1257    DeferredActionUndoType undo_action = IGNORE;
1258
1259    int value = 0;
1260    bool absolute = false;
1261    bool clear = false;
1262    static const int kNoStore = kMinInt;
1263    int store_position = kNoStore;
1264    // This is a little tricky because we are scanning the actions in reverse
1265    // historical order (newest first).
1266    for (DeferredAction* action = actions_;
1267         action != NULL;
1268         action = action->next()) {
1269      if (action->Mentions(reg)) {
1270        switch (action->action_type()) {
1271          case ActionNode::SET_REGISTER: {
1272            Trace::DeferredSetRegister* psr =
1273                static_cast<Trace::DeferredSetRegister*>(action);
1274            if (!absolute) {
1275              value += psr->value();
1276              absolute = true;
1277            }
1278            // SET_REGISTER is currently only used for newly introduced loop
1279            // counters. They can have a significant previous value if they
1280            // occour in a loop. TODO(lrn): Propagate this information, so
1281            // we can set undo_action to IGNORE if we know there is no value to
1282            // restore.
1283            undo_action = RESTORE;
1284            DCHECK_EQ(store_position, kNoStore);
1285            DCHECK(!clear);
1286            break;
1287          }
1288          case ActionNode::INCREMENT_REGISTER:
1289            if (!absolute) {
1290              value++;
1291            }
1292            DCHECK_EQ(store_position, kNoStore);
1293            DCHECK(!clear);
1294            undo_action = RESTORE;
1295            break;
1296          case ActionNode::STORE_POSITION: {
1297            Trace::DeferredCapture* pc =
1298                static_cast<Trace::DeferredCapture*>(action);
1299            if (!clear && store_position == kNoStore) {
1300              store_position = pc->cp_offset();
1301            }
1302
1303            // For captures we know that stores and clears alternate.
1304            // Other register, are never cleared, and if the occur
1305            // inside a loop, they might be assigned more than once.
1306            if (reg <= 1) {
1307              // Registers zero and one, aka "capture zero", is
1308              // always set correctly if we succeed. There is no
1309              // need to undo a setting on backtrack, because we
1310              // will set it again or fail.
1311              undo_action = IGNORE;
1312            } else {
1313              undo_action = pc->is_capture() ? CLEAR : RESTORE;
1314            }
1315            DCHECK(!absolute);
1316            DCHECK_EQ(value, 0);
1317            break;
1318          }
1319          case ActionNode::CLEAR_CAPTURES: {
1320            // Since we're scanning in reverse order, if we've already
1321            // set the position we have to ignore historically earlier
1322            // clearing operations.
1323            if (store_position == kNoStore) {
1324              clear = true;
1325            }
1326            undo_action = RESTORE;
1327            DCHECK(!absolute);
1328            DCHECK_EQ(value, 0);
1329            break;
1330          }
1331          default:
1332            UNREACHABLE();
1333            break;
1334        }
1335      }
1336    }
1337    // Prepare for the undo-action (e.g., push if it's going to be popped).
1338    if (undo_action == RESTORE) {
1339      pushes++;
1340      RegExpMacroAssembler::StackCheckFlag stack_check =
1341          RegExpMacroAssembler::kNoStackLimitCheck;
1342      if (pushes == push_limit) {
1343        stack_check = RegExpMacroAssembler::kCheckStackLimit;
1344        pushes = 0;
1345      }
1346
1347      assembler->PushRegister(reg, stack_check);
1348      registers_to_pop->Set(reg, zone);
1349    } else if (undo_action == CLEAR) {
1350      registers_to_clear->Set(reg, zone);
1351    }
1352    // Perform the chronologically last action (or accumulated increment)
1353    // for the register.
1354    if (store_position != kNoStore) {
1355      assembler->WriteCurrentPositionToRegister(reg, store_position);
1356    } else if (clear) {
1357      assembler->ClearRegisters(reg, reg);
1358    } else if (absolute) {
1359      assembler->SetRegister(reg, value);
1360    } else if (value != 0) {
1361      assembler->AdvanceRegister(reg, value);
1362    }
1363  }
1364}
1365
1366
1367// This is called as we come into a loop choice node and some other tricky
1368// nodes.  It normalizes the state of the code generator to ensure we can
1369// generate generic code.
1370void Trace::Flush(RegExpCompiler* compiler, RegExpNode* successor) {
1371  RegExpMacroAssembler* assembler = compiler->macro_assembler();
1372
1373  DCHECK(!is_trivial());
1374
1375  if (actions_ == NULL && backtrack() == NULL) {
1376    // Here we just have some deferred cp advances to fix and we are back to
1377    // a normal situation.  We may also have to forget some information gained
1378    // through a quick check that was already performed.
1379    if (cp_offset_ != 0) assembler->AdvanceCurrentPosition(cp_offset_);
1380    // Create a new trivial state and generate the node with that.
1381    Trace new_state;
1382    successor->Emit(compiler, &new_state);
1383    return;
1384  }
1385
1386  // Generate deferred actions here along with code to undo them again.
1387  OutSet affected_registers;
1388
1389  if (backtrack() != NULL) {
1390    // Here we have a concrete backtrack location.  These are set up by choice
1391    // nodes and so they indicate that we have a deferred save of the current
1392    // position which we may need to emit here.
1393    assembler->PushCurrentPosition();
1394  }
1395
1396  int max_register = FindAffectedRegisters(&affected_registers,
1397                                           compiler->zone());
1398  OutSet registers_to_pop;
1399  OutSet registers_to_clear;
1400  PerformDeferredActions(assembler,
1401                         max_register,
1402                         affected_registers,
1403                         &registers_to_pop,
1404                         &registers_to_clear,
1405                         compiler->zone());
1406  if (cp_offset_ != 0) {
1407    assembler->AdvanceCurrentPosition(cp_offset_);
1408  }
1409
1410  // Create a new trivial state and generate the node with that.
1411  Label undo;
1412  assembler->PushBacktrack(&undo);
1413  if (successor->KeepRecursing(compiler)) {
1414    Trace new_state;
1415    successor->Emit(compiler, &new_state);
1416  } else {
1417    compiler->AddWork(successor);
1418    assembler->GoTo(successor->label());
1419  }
1420
1421  // On backtrack we need to restore state.
1422  assembler->Bind(&undo);
1423  RestoreAffectedRegisters(assembler,
1424                           max_register,
1425                           registers_to_pop,
1426                           registers_to_clear);
1427  if (backtrack() == NULL) {
1428    assembler->Backtrack();
1429  } else {
1430    assembler->PopCurrentPosition();
1431    assembler->GoTo(backtrack());
1432  }
1433}
1434
1435
1436void NegativeSubmatchSuccess::Emit(RegExpCompiler* compiler, Trace* trace) {
1437  RegExpMacroAssembler* assembler = compiler->macro_assembler();
1438
1439  // Omit flushing the trace. We discard the entire stack frame anyway.
1440
1441  if (!label()->is_bound()) {
1442    // We are completely independent of the trace, since we ignore it,
1443    // so this code can be used as the generic version.
1444    assembler->Bind(label());
1445  }
1446
1447  // Throw away everything on the backtrack stack since the start
1448  // of the negative submatch and restore the character position.
1449  assembler->ReadCurrentPositionFromRegister(current_position_register_);
1450  assembler->ReadStackPointerFromRegister(stack_pointer_register_);
1451  if (clear_capture_count_ > 0) {
1452    // Clear any captures that might have been performed during the success
1453    // of the body of the negative look-ahead.
1454    int clear_capture_end = clear_capture_start_ + clear_capture_count_ - 1;
1455    assembler->ClearRegisters(clear_capture_start_, clear_capture_end);
1456  }
1457  // Now that we have unwound the stack we find at the top of the stack the
1458  // backtrack that the BeginSubmatch node got.
1459  assembler->Backtrack();
1460}
1461
1462
1463void EndNode::Emit(RegExpCompiler* compiler, Trace* trace) {
1464  if (!trace->is_trivial()) {
1465    trace->Flush(compiler, this);
1466    return;
1467  }
1468  RegExpMacroAssembler* assembler = compiler->macro_assembler();
1469  if (!label()->is_bound()) {
1470    assembler->Bind(label());
1471  }
1472  switch (action_) {
1473    case ACCEPT:
1474      assembler->Succeed();
1475      return;
1476    case BACKTRACK:
1477      assembler->GoTo(trace->backtrack());
1478      return;
1479    case NEGATIVE_SUBMATCH_SUCCESS:
1480      // This case is handled in a different virtual method.
1481      UNREACHABLE();
1482  }
1483  UNIMPLEMENTED();
1484}
1485
1486
1487void GuardedAlternative::AddGuard(Guard* guard, Zone* zone) {
1488  if (guards_ == NULL)
1489    guards_ = new(zone) ZoneList<Guard*>(1, zone);
1490  guards_->Add(guard, zone);
1491}
1492
1493
1494ActionNode* ActionNode::SetRegister(int reg,
1495                                    int val,
1496                                    RegExpNode* on_success) {
1497  ActionNode* result =
1498      new(on_success->zone()) ActionNode(SET_REGISTER, on_success);
1499  result->data_.u_store_register.reg = reg;
1500  result->data_.u_store_register.value = val;
1501  return result;
1502}
1503
1504
1505ActionNode* ActionNode::IncrementRegister(int reg, RegExpNode* on_success) {
1506  ActionNode* result =
1507      new(on_success->zone()) ActionNode(INCREMENT_REGISTER, on_success);
1508  result->data_.u_increment_register.reg = reg;
1509  return result;
1510}
1511
1512
1513ActionNode* ActionNode::StorePosition(int reg,
1514                                      bool is_capture,
1515                                      RegExpNode* on_success) {
1516  ActionNode* result =
1517      new(on_success->zone()) ActionNode(STORE_POSITION, on_success);
1518  result->data_.u_position_register.reg = reg;
1519  result->data_.u_position_register.is_capture = is_capture;
1520  return result;
1521}
1522
1523
1524ActionNode* ActionNode::ClearCaptures(Interval range,
1525                                      RegExpNode* on_success) {
1526  ActionNode* result =
1527      new(on_success->zone()) ActionNode(CLEAR_CAPTURES, on_success);
1528  result->data_.u_clear_captures.range_from = range.from();
1529  result->data_.u_clear_captures.range_to = range.to();
1530  return result;
1531}
1532
1533
1534ActionNode* ActionNode::BeginSubmatch(int stack_reg,
1535                                      int position_reg,
1536                                      RegExpNode* on_success) {
1537  ActionNode* result =
1538      new(on_success->zone()) ActionNode(BEGIN_SUBMATCH, on_success);
1539  result->data_.u_submatch.stack_pointer_register = stack_reg;
1540  result->data_.u_submatch.current_position_register = position_reg;
1541  return result;
1542}
1543
1544
1545ActionNode* ActionNode::PositiveSubmatchSuccess(int stack_reg,
1546                                                int position_reg,
1547                                                int clear_register_count,
1548                                                int clear_register_from,
1549                                                RegExpNode* on_success) {
1550  ActionNode* result =
1551      new(on_success->zone()) ActionNode(POSITIVE_SUBMATCH_SUCCESS, on_success);
1552  result->data_.u_submatch.stack_pointer_register = stack_reg;
1553  result->data_.u_submatch.current_position_register = position_reg;
1554  result->data_.u_submatch.clear_register_count = clear_register_count;
1555  result->data_.u_submatch.clear_register_from = clear_register_from;
1556  return result;
1557}
1558
1559
1560ActionNode* ActionNode::EmptyMatchCheck(int start_register,
1561                                        int repetition_register,
1562                                        int repetition_limit,
1563                                        RegExpNode* on_success) {
1564  ActionNode* result =
1565      new(on_success->zone()) ActionNode(EMPTY_MATCH_CHECK, on_success);
1566  result->data_.u_empty_match_check.start_register = start_register;
1567  result->data_.u_empty_match_check.repetition_register = repetition_register;
1568  result->data_.u_empty_match_check.repetition_limit = repetition_limit;
1569  return result;
1570}
1571
1572
1573#define DEFINE_ACCEPT(Type)                                          \
1574  void Type##Node::Accept(NodeVisitor* visitor) {                    \
1575    visitor->Visit##Type(this);                                      \
1576  }
1577FOR_EACH_NODE_TYPE(DEFINE_ACCEPT)
1578#undef DEFINE_ACCEPT
1579
1580
1581void LoopChoiceNode::Accept(NodeVisitor* visitor) {
1582  visitor->VisitLoopChoice(this);
1583}
1584
1585
1586// -------------------------------------------------------------------
1587// Emit code.
1588
1589
1590void ChoiceNode::GenerateGuard(RegExpMacroAssembler* macro_assembler,
1591                               Guard* guard,
1592                               Trace* trace) {
1593  switch (guard->op()) {
1594    case Guard::LT:
1595      DCHECK(!trace->mentions_reg(guard->reg()));
1596      macro_assembler->IfRegisterGE(guard->reg(),
1597                                    guard->value(),
1598                                    trace->backtrack());
1599      break;
1600    case Guard::GEQ:
1601      DCHECK(!trace->mentions_reg(guard->reg()));
1602      macro_assembler->IfRegisterLT(guard->reg(),
1603                                    guard->value(),
1604                                    trace->backtrack());
1605      break;
1606  }
1607}
1608
1609
1610// Returns the number of characters in the equivalence class, omitting those
1611// that cannot occur in the source string because it is Latin1.
1612static int GetCaseIndependentLetters(Isolate* isolate, uc16 character,
1613                                     bool one_byte_subject,
1614                                     unibrow::uchar* letters) {
1615  int length =
1616      isolate->jsregexp_uncanonicalize()->get(character, '\0', letters);
1617  // Unibrow returns 0 or 1 for characters where case independence is
1618  // trivial.
1619  if (length == 0) {
1620    letters[0] = character;
1621    length = 1;
1622  }
1623
1624  if (one_byte_subject) {
1625    int new_length = 0;
1626    for (int i = 0; i < length; i++) {
1627      if (letters[i] <= String::kMaxOneByteCharCode) {
1628        letters[new_length++] = letters[i];
1629      }
1630    }
1631    length = new_length;
1632  }
1633
1634  return length;
1635}
1636
1637
1638static inline bool EmitSimpleCharacter(Isolate* isolate,
1639                                       RegExpCompiler* compiler,
1640                                       uc16 c,
1641                                       Label* on_failure,
1642                                       int cp_offset,
1643                                       bool check,
1644                                       bool preloaded) {
1645  RegExpMacroAssembler* assembler = compiler->macro_assembler();
1646  bool bound_checked = false;
1647  if (!preloaded) {
1648    assembler->LoadCurrentCharacter(
1649        cp_offset,
1650        on_failure,
1651        check);
1652    bound_checked = true;
1653  }
1654  assembler->CheckNotCharacter(c, on_failure);
1655  return bound_checked;
1656}
1657
1658
1659// Only emits non-letters (things that don't have case).  Only used for case
1660// independent matches.
1661static inline bool EmitAtomNonLetter(Isolate* isolate,
1662                                     RegExpCompiler* compiler,
1663                                     uc16 c,
1664                                     Label* on_failure,
1665                                     int cp_offset,
1666                                     bool check,
1667                                     bool preloaded) {
1668  RegExpMacroAssembler* macro_assembler = compiler->macro_assembler();
1669  bool one_byte = compiler->one_byte();
1670  unibrow::uchar chars[unibrow::Ecma262UnCanonicalize::kMaxWidth];
1671  int length = GetCaseIndependentLetters(isolate, c, one_byte, chars);
1672  if (length < 1) {
1673    // This can't match.  Must be an one-byte subject and a non-one-byte
1674    // character.  We do not need to do anything since the one-byte pass
1675    // already handled this.
1676    return false;  // Bounds not checked.
1677  }
1678  bool checked = false;
1679  // We handle the length > 1 case in a later pass.
1680  if (length == 1) {
1681    if (one_byte && c > String::kMaxOneByteCharCodeU) {
1682      // Can't match - see above.
1683      return false;  // Bounds not checked.
1684    }
1685    if (!preloaded) {
1686      macro_assembler->LoadCurrentCharacter(cp_offset, on_failure, check);
1687      checked = check;
1688    }
1689    macro_assembler->CheckNotCharacter(c, on_failure);
1690  }
1691  return checked;
1692}
1693
1694
1695static bool ShortCutEmitCharacterPair(RegExpMacroAssembler* macro_assembler,
1696                                      bool one_byte, uc16 c1, uc16 c2,
1697                                      Label* on_failure) {
1698  uc16 char_mask;
1699  if (one_byte) {
1700    char_mask = String::kMaxOneByteCharCode;
1701  } else {
1702    char_mask = String::kMaxUtf16CodeUnit;
1703  }
1704  uc16 exor = c1 ^ c2;
1705  // Check whether exor has only one bit set.
1706  if (((exor - 1) & exor) == 0) {
1707    // If c1 and c2 differ only by one bit.
1708    // Ecma262UnCanonicalize always gives the highest number last.
1709    DCHECK(c2 > c1);
1710    uc16 mask = char_mask ^ exor;
1711    macro_assembler->CheckNotCharacterAfterAnd(c1, mask, on_failure);
1712    return true;
1713  }
1714  DCHECK(c2 > c1);
1715  uc16 diff = c2 - c1;
1716  if (((diff - 1) & diff) == 0 && c1 >= diff) {
1717    // If the characters differ by 2^n but don't differ by one bit then
1718    // subtract the difference from the found character, then do the or
1719    // trick.  We avoid the theoretical case where negative numbers are
1720    // involved in order to simplify code generation.
1721    uc16 mask = char_mask ^ diff;
1722    macro_assembler->CheckNotCharacterAfterMinusAnd(c1 - diff,
1723                                                    diff,
1724                                                    mask,
1725                                                    on_failure);
1726    return true;
1727  }
1728  return false;
1729}
1730
1731
1732typedef bool EmitCharacterFunction(Isolate* isolate,
1733                                   RegExpCompiler* compiler,
1734                                   uc16 c,
1735                                   Label* on_failure,
1736                                   int cp_offset,
1737                                   bool check,
1738                                   bool preloaded);
1739
1740// Only emits letters (things that have case).  Only used for case independent
1741// matches.
1742static inline bool EmitAtomLetter(Isolate* isolate,
1743                                  RegExpCompiler* compiler,
1744                                  uc16 c,
1745                                  Label* on_failure,
1746                                  int cp_offset,
1747                                  bool check,
1748                                  bool preloaded) {
1749  RegExpMacroAssembler* macro_assembler = compiler->macro_assembler();
1750  bool one_byte = compiler->one_byte();
1751  unibrow::uchar chars[unibrow::Ecma262UnCanonicalize::kMaxWidth];
1752  int length = GetCaseIndependentLetters(isolate, c, one_byte, chars);
1753  if (length <= 1) return false;
1754  // We may not need to check against the end of the input string
1755  // if this character lies before a character that matched.
1756  if (!preloaded) {
1757    macro_assembler->LoadCurrentCharacter(cp_offset, on_failure, check);
1758  }
1759  Label ok;
1760  DCHECK(unibrow::Ecma262UnCanonicalize::kMaxWidth == 4);
1761  switch (length) {
1762    case 2: {
1763      if (ShortCutEmitCharacterPair(macro_assembler, one_byte, chars[0],
1764                                    chars[1], on_failure)) {
1765      } else {
1766        macro_assembler->CheckCharacter(chars[0], &ok);
1767        macro_assembler->CheckNotCharacter(chars[1], on_failure);
1768        macro_assembler->Bind(&ok);
1769      }
1770      break;
1771    }
1772    case 4:
1773      macro_assembler->CheckCharacter(chars[3], &ok);
1774      // Fall through!
1775    case 3:
1776      macro_assembler->CheckCharacter(chars[0], &ok);
1777      macro_assembler->CheckCharacter(chars[1], &ok);
1778      macro_assembler->CheckNotCharacter(chars[2], on_failure);
1779      macro_assembler->Bind(&ok);
1780      break;
1781    default:
1782      UNREACHABLE();
1783      break;
1784  }
1785  return true;
1786}
1787
1788
1789static void EmitBoundaryTest(RegExpMacroAssembler* masm,
1790                             int border,
1791                             Label* fall_through,
1792                             Label* above_or_equal,
1793                             Label* below) {
1794  if (below != fall_through) {
1795    masm->CheckCharacterLT(border, below);
1796    if (above_or_equal != fall_through) masm->GoTo(above_or_equal);
1797  } else {
1798    masm->CheckCharacterGT(border - 1, above_or_equal);
1799  }
1800}
1801
1802
1803static void EmitDoubleBoundaryTest(RegExpMacroAssembler* masm,
1804                                   int first,
1805                                   int last,
1806                                   Label* fall_through,
1807                                   Label* in_range,
1808                                   Label* out_of_range) {
1809  if (in_range == fall_through) {
1810    if (first == last) {
1811      masm->CheckNotCharacter(first, out_of_range);
1812    } else {
1813      masm->CheckCharacterNotInRange(first, last, out_of_range);
1814    }
1815  } else {
1816    if (first == last) {
1817      masm->CheckCharacter(first, in_range);
1818    } else {
1819      masm->CheckCharacterInRange(first, last, in_range);
1820    }
1821    if (out_of_range != fall_through) masm->GoTo(out_of_range);
1822  }
1823}
1824
1825
1826// even_label is for ranges[i] to ranges[i + 1] where i - start_index is even.
1827// odd_label is for ranges[i] to ranges[i + 1] where i - start_index is odd.
1828static void EmitUseLookupTable(
1829    RegExpMacroAssembler* masm,
1830    ZoneList<int>* ranges,
1831    int start_index,
1832    int end_index,
1833    int min_char,
1834    Label* fall_through,
1835    Label* even_label,
1836    Label* odd_label) {
1837  static const int kSize = RegExpMacroAssembler::kTableSize;
1838  static const int kMask = RegExpMacroAssembler::kTableMask;
1839
1840  int base = (min_char & ~kMask);
1841  USE(base);
1842
1843  // Assert that everything is on one kTableSize page.
1844  for (int i = start_index; i <= end_index; i++) {
1845    DCHECK_EQ(ranges->at(i) & ~kMask, base);
1846  }
1847  DCHECK(start_index == 0 || (ranges->at(start_index - 1) & ~kMask) <= base);
1848
1849  char templ[kSize];
1850  Label* on_bit_set;
1851  Label* on_bit_clear;
1852  int bit;
1853  if (even_label == fall_through) {
1854    on_bit_set = odd_label;
1855    on_bit_clear = even_label;
1856    bit = 1;
1857  } else {
1858    on_bit_set = even_label;
1859    on_bit_clear = odd_label;
1860    bit = 0;
1861  }
1862  for (int i = 0; i < (ranges->at(start_index) & kMask) && i < kSize; i++) {
1863    templ[i] = bit;
1864  }
1865  int j = 0;
1866  bit ^= 1;
1867  for (int i = start_index; i < end_index; i++) {
1868    for (j = (ranges->at(i) & kMask); j < (ranges->at(i + 1) & kMask); j++) {
1869      templ[j] = bit;
1870    }
1871    bit ^= 1;
1872  }
1873  for (int i = j; i < kSize; i++) {
1874    templ[i] = bit;
1875  }
1876  Factory* factory = masm->isolate()->factory();
1877  // TODO(erikcorry): Cache these.
1878  Handle<ByteArray> ba = factory->NewByteArray(kSize, TENURED);
1879  for (int i = 0; i < kSize; i++) {
1880    ba->set(i, templ[i]);
1881  }
1882  masm->CheckBitInTable(ba, on_bit_set);
1883  if (on_bit_clear != fall_through) masm->GoTo(on_bit_clear);
1884}
1885
1886
1887static void CutOutRange(RegExpMacroAssembler* masm,
1888                        ZoneList<int>* ranges,
1889                        int start_index,
1890                        int end_index,
1891                        int cut_index,
1892                        Label* even_label,
1893                        Label* odd_label) {
1894  bool odd = (((cut_index - start_index) & 1) == 1);
1895  Label* in_range_label = odd ? odd_label : even_label;
1896  Label dummy;
1897  EmitDoubleBoundaryTest(masm,
1898                         ranges->at(cut_index),
1899                         ranges->at(cut_index + 1) - 1,
1900                         &dummy,
1901                         in_range_label,
1902                         &dummy);
1903  DCHECK(!dummy.is_linked());
1904  // Cut out the single range by rewriting the array.  This creates a new
1905  // range that is a merger of the two ranges on either side of the one we
1906  // are cutting out.  The oddity of the labels is preserved.
1907  for (int j = cut_index; j > start_index; j--) {
1908    ranges->at(j) = ranges->at(j - 1);
1909  }
1910  for (int j = cut_index + 1; j < end_index; j++) {
1911    ranges->at(j) = ranges->at(j + 1);
1912  }
1913}
1914
1915
1916// Unicode case.  Split the search space into kSize spaces that are handled
1917// with recursion.
1918static void SplitSearchSpace(ZoneList<int>* ranges,
1919                             int start_index,
1920                             int end_index,
1921                             int* new_start_index,
1922                             int* new_end_index,
1923                             int* border) {
1924  static const int kSize = RegExpMacroAssembler::kTableSize;
1925  static const int kMask = RegExpMacroAssembler::kTableMask;
1926
1927  int first = ranges->at(start_index);
1928  int last = ranges->at(end_index) - 1;
1929
1930  *new_start_index = start_index;
1931  *border = (ranges->at(start_index) & ~kMask) + kSize;
1932  while (*new_start_index < end_index) {
1933    if (ranges->at(*new_start_index) > *border) break;
1934    (*new_start_index)++;
1935  }
1936  // new_start_index is the index of the first edge that is beyond the
1937  // current kSize space.
1938
1939  // For very large search spaces we do a binary chop search of the non-Latin1
1940  // space instead of just going to the end of the current kSize space.  The
1941  // heuristics are complicated a little by the fact that any 128-character
1942  // encoding space can be quickly tested with a table lookup, so we don't
1943  // wish to do binary chop search at a smaller granularity than that.  A
1944  // 128-character space can take up a lot of space in the ranges array if,
1945  // for example, we only want to match every second character (eg. the lower
1946  // case characters on some Unicode pages).
1947  int binary_chop_index = (end_index + start_index) / 2;
1948  // The first test ensures that we get to the code that handles the Latin1
1949  // range with a single not-taken branch, speeding up this important
1950  // character range (even non-Latin1 charset-based text has spaces and
1951  // punctuation).
1952  if (*border - 1 > String::kMaxOneByteCharCode &&  // Latin1 case.
1953      end_index - start_index > (*new_start_index - start_index) * 2 &&
1954      last - first > kSize * 2 && binary_chop_index > *new_start_index &&
1955      ranges->at(binary_chop_index) >= first + 2 * kSize) {
1956    int scan_forward_for_section_border = binary_chop_index;;
1957    int new_border = (ranges->at(binary_chop_index) | kMask) + 1;
1958
1959    while (scan_forward_for_section_border < end_index) {
1960      if (ranges->at(scan_forward_for_section_border) > new_border) {
1961        *new_start_index = scan_forward_for_section_border;
1962        *border = new_border;
1963        break;
1964      }
1965      scan_forward_for_section_border++;
1966    }
1967  }
1968
1969  DCHECK(*new_start_index > start_index);
1970  *new_end_index = *new_start_index - 1;
1971  if (ranges->at(*new_end_index) == *border) {
1972    (*new_end_index)--;
1973  }
1974  if (*border >= ranges->at(end_index)) {
1975    *border = ranges->at(end_index);
1976    *new_start_index = end_index;  // Won't be used.
1977    *new_end_index = end_index - 1;
1978  }
1979}
1980
1981
1982// Gets a series of segment boundaries representing a character class.  If the
1983// character is in the range between an even and an odd boundary (counting from
1984// start_index) then go to even_label, otherwise go to odd_label.  We already
1985// know that the character is in the range of min_char to max_char inclusive.
1986// Either label can be NULL indicating backtracking.  Either label can also be
1987// equal to the fall_through label.
1988static void GenerateBranches(RegExpMacroAssembler* masm, ZoneList<int>* ranges,
1989                             int start_index, int end_index, uc32 min_char,
1990                             uc32 max_char, Label* fall_through,
1991                             Label* even_label, Label* odd_label) {
1992  DCHECK_LE(min_char, String::kMaxUtf16CodeUnit);
1993  DCHECK_LE(max_char, String::kMaxUtf16CodeUnit);
1994
1995  int first = ranges->at(start_index);
1996  int last = ranges->at(end_index) - 1;
1997
1998  DCHECK_LT(min_char, first);
1999
2000  // Just need to test if the character is before or on-or-after
2001  // a particular character.
2002  if (start_index == end_index) {
2003    EmitBoundaryTest(masm, first, fall_through, even_label, odd_label);
2004    return;
2005  }
2006
2007  // Another almost trivial case:  There is one interval in the middle that is
2008  // different from the end intervals.
2009  if (start_index + 1 == end_index) {
2010    EmitDoubleBoundaryTest(
2011        masm, first, last, fall_through, even_label, odd_label);
2012    return;
2013  }
2014
2015  // It's not worth using table lookup if there are very few intervals in the
2016  // character class.
2017  if (end_index - start_index <= 6) {
2018    // It is faster to test for individual characters, so we look for those
2019    // first, then try arbitrary ranges in the second round.
2020    static int kNoCutIndex = -1;
2021    int cut = kNoCutIndex;
2022    for (int i = start_index; i < end_index; i++) {
2023      if (ranges->at(i) == ranges->at(i + 1) - 1) {
2024        cut = i;
2025        break;
2026      }
2027    }
2028    if (cut == kNoCutIndex) cut = start_index;
2029    CutOutRange(
2030        masm, ranges, start_index, end_index, cut, even_label, odd_label);
2031    DCHECK_GE(end_index - start_index, 2);
2032    GenerateBranches(masm,
2033                     ranges,
2034                     start_index + 1,
2035                     end_index - 1,
2036                     min_char,
2037                     max_char,
2038                     fall_through,
2039                     even_label,
2040                     odd_label);
2041    return;
2042  }
2043
2044  // If there are a lot of intervals in the regexp, then we will use tables to
2045  // determine whether the character is inside or outside the character class.
2046  static const int kBits = RegExpMacroAssembler::kTableSizeBits;
2047
2048  if ((max_char >> kBits) == (min_char >> kBits)) {
2049    EmitUseLookupTable(masm,
2050                       ranges,
2051                       start_index,
2052                       end_index,
2053                       min_char,
2054                       fall_through,
2055                       even_label,
2056                       odd_label);
2057    return;
2058  }
2059
2060  if ((min_char >> kBits) != (first >> kBits)) {
2061    masm->CheckCharacterLT(first, odd_label);
2062    GenerateBranches(masm,
2063                     ranges,
2064                     start_index + 1,
2065                     end_index,
2066                     first,
2067                     max_char,
2068                     fall_through,
2069                     odd_label,
2070                     even_label);
2071    return;
2072  }
2073
2074  int new_start_index = 0;
2075  int new_end_index = 0;
2076  int border = 0;
2077
2078  SplitSearchSpace(ranges,
2079                   start_index,
2080                   end_index,
2081                   &new_start_index,
2082                   &new_end_index,
2083                   &border);
2084
2085  Label handle_rest;
2086  Label* above = &handle_rest;
2087  if (border == last + 1) {
2088    // We didn't find any section that started after the limit, so everything
2089    // above the border is one of the terminal labels.
2090    above = (end_index & 1) != (start_index & 1) ? odd_label : even_label;
2091    DCHECK(new_end_index == end_index - 1);
2092  }
2093
2094  DCHECK_LE(start_index, new_end_index);
2095  DCHECK_LE(new_start_index, end_index);
2096  DCHECK_LT(start_index, new_start_index);
2097  DCHECK_LT(new_end_index, end_index);
2098  DCHECK(new_end_index + 1 == new_start_index ||
2099         (new_end_index + 2 == new_start_index &&
2100          border == ranges->at(new_end_index + 1)));
2101  DCHECK_LT(min_char, border - 1);
2102  DCHECK_LT(border, max_char);
2103  DCHECK_LT(ranges->at(new_end_index), border);
2104  DCHECK(border < ranges->at(new_start_index) ||
2105         (border == ranges->at(new_start_index) &&
2106          new_start_index == end_index &&
2107          new_end_index == end_index - 1 &&
2108          border == last + 1));
2109  DCHECK(new_start_index == 0 || border >= ranges->at(new_start_index - 1));
2110
2111  masm->CheckCharacterGT(border - 1, above);
2112  Label dummy;
2113  GenerateBranches(masm,
2114                   ranges,
2115                   start_index,
2116                   new_end_index,
2117                   min_char,
2118                   border - 1,
2119                   &dummy,
2120                   even_label,
2121                   odd_label);
2122  if (handle_rest.is_linked()) {
2123    masm->Bind(&handle_rest);
2124    bool flip = (new_start_index & 1) != (start_index & 1);
2125    GenerateBranches(masm,
2126                     ranges,
2127                     new_start_index,
2128                     end_index,
2129                     border,
2130                     max_char,
2131                     &dummy,
2132                     flip ? odd_label : even_label,
2133                     flip ? even_label : odd_label);
2134  }
2135}
2136
2137
2138static void EmitCharClass(RegExpMacroAssembler* macro_assembler,
2139                          RegExpCharacterClass* cc, bool one_byte,
2140                          Label* on_failure, int cp_offset, bool check_offset,
2141                          bool preloaded, Zone* zone) {
2142  ZoneList<CharacterRange>* ranges = cc->ranges(zone);
2143  CharacterRange::Canonicalize(ranges);
2144
2145  int max_char;
2146  if (one_byte) {
2147    max_char = String::kMaxOneByteCharCode;
2148  } else {
2149    max_char = String::kMaxUtf16CodeUnit;
2150  }
2151
2152  int range_count = ranges->length();
2153
2154  int last_valid_range = range_count - 1;
2155  while (last_valid_range >= 0) {
2156    CharacterRange& range = ranges->at(last_valid_range);
2157    if (range.from() <= max_char) {
2158      break;
2159    }
2160    last_valid_range--;
2161  }
2162
2163  if (last_valid_range < 0) {
2164    if (!cc->is_negated()) {
2165      macro_assembler->GoTo(on_failure);
2166    }
2167    if (check_offset) {
2168      macro_assembler->CheckPosition(cp_offset, on_failure);
2169    }
2170    return;
2171  }
2172
2173  if (last_valid_range == 0 &&
2174      ranges->at(0).IsEverything(max_char)) {
2175    if (cc->is_negated()) {
2176      macro_assembler->GoTo(on_failure);
2177    } else {
2178      // This is a common case hit by non-anchored expressions.
2179      if (check_offset) {
2180        macro_assembler->CheckPosition(cp_offset, on_failure);
2181      }
2182    }
2183    return;
2184  }
2185
2186  if (!preloaded) {
2187    macro_assembler->LoadCurrentCharacter(cp_offset, on_failure, check_offset);
2188  }
2189
2190  if (cc->is_standard(zone) &&
2191      macro_assembler->CheckSpecialCharacterClass(cc->standard_type(),
2192                                                  on_failure)) {
2193      return;
2194  }
2195
2196
2197  // A new list with ascending entries.  Each entry is a code unit
2198  // where there is a boundary between code units that are part of
2199  // the class and code units that are not.  Normally we insert an
2200  // entry at zero which goes to the failure label, but if there
2201  // was already one there we fall through for success on that entry.
2202  // Subsequent entries have alternating meaning (success/failure).
2203  ZoneList<int>* range_boundaries =
2204      new(zone) ZoneList<int>(last_valid_range, zone);
2205
2206  bool zeroth_entry_is_failure = !cc->is_negated();
2207
2208  for (int i = 0; i <= last_valid_range; i++) {
2209    CharacterRange& range = ranges->at(i);
2210    if (range.from() == 0) {
2211      DCHECK_EQ(i, 0);
2212      zeroth_entry_is_failure = !zeroth_entry_is_failure;
2213    } else {
2214      range_boundaries->Add(range.from(), zone);
2215    }
2216    range_boundaries->Add(range.to() + 1, zone);
2217  }
2218  int end_index = range_boundaries->length() - 1;
2219  if (range_boundaries->at(end_index) > max_char) {
2220    end_index--;
2221  }
2222
2223  Label fall_through;
2224  GenerateBranches(macro_assembler,
2225                   range_boundaries,
2226                   0,  // start_index.
2227                   end_index,
2228                   0,  // min_char.
2229                   max_char,
2230                   &fall_through,
2231                   zeroth_entry_is_failure ? &fall_through : on_failure,
2232                   zeroth_entry_is_failure ? on_failure : &fall_through);
2233  macro_assembler->Bind(&fall_through);
2234}
2235
2236
2237RegExpNode::~RegExpNode() {
2238}
2239
2240
2241RegExpNode::LimitResult RegExpNode::LimitVersions(RegExpCompiler* compiler,
2242                                                  Trace* trace) {
2243  // If we are generating a greedy loop then don't stop and don't reuse code.
2244  if (trace->stop_node() != NULL) {
2245    return CONTINUE;
2246  }
2247
2248  RegExpMacroAssembler* macro_assembler = compiler->macro_assembler();
2249  if (trace->is_trivial()) {
2250    if (label_.is_bound() || on_work_list() || !KeepRecursing(compiler)) {
2251      // If a generic version is already scheduled to be generated or we have
2252      // recursed too deeply then just generate a jump to that code.
2253      macro_assembler->GoTo(&label_);
2254      // This will queue it up for generation of a generic version if it hasn't
2255      // already been queued.
2256      compiler->AddWork(this);
2257      return DONE;
2258    }
2259    // Generate generic version of the node and bind the label for later use.
2260    macro_assembler->Bind(&label_);
2261    return CONTINUE;
2262  }
2263
2264  // We are being asked to make a non-generic version.  Keep track of how many
2265  // non-generic versions we generate so as not to overdo it.
2266  trace_count_++;
2267  if (KeepRecursing(compiler) && compiler->optimize() &&
2268      trace_count_ < kMaxCopiesCodeGenerated) {
2269    return CONTINUE;
2270  }
2271
2272  // If we get here code has been generated for this node too many times or
2273  // recursion is too deep.  Time to switch to a generic version.  The code for
2274  // generic versions above can handle deep recursion properly.
2275  bool was_limiting = compiler->limiting_recursion();
2276  compiler->set_limiting_recursion(true);
2277  trace->Flush(compiler, this);
2278  compiler->set_limiting_recursion(was_limiting);
2279  return DONE;
2280}
2281
2282
2283bool RegExpNode::KeepRecursing(RegExpCompiler* compiler) {
2284  return !compiler->limiting_recursion() &&
2285         compiler->recursion_depth() <= RegExpCompiler::kMaxRecursion;
2286}
2287
2288
2289int ActionNode::EatsAtLeast(int still_to_find,
2290                            int budget,
2291                            bool not_at_start) {
2292  if (budget <= 0) return 0;
2293  if (action_type_ == POSITIVE_SUBMATCH_SUCCESS) return 0;  // Rewinds input!
2294  return on_success()->EatsAtLeast(still_to_find,
2295                                   budget - 1,
2296                                   not_at_start);
2297}
2298
2299
2300void ActionNode::FillInBMInfo(Isolate* isolate, int offset, int budget,
2301                              BoyerMooreLookahead* bm, bool not_at_start) {
2302  if (action_type_ == BEGIN_SUBMATCH) {
2303    bm->SetRest(offset);
2304  } else if (action_type_ != POSITIVE_SUBMATCH_SUCCESS) {
2305    on_success()->FillInBMInfo(isolate, offset, budget - 1, bm, not_at_start);
2306  }
2307  SaveBMInfo(bm, not_at_start, offset);
2308}
2309
2310
2311int AssertionNode::EatsAtLeast(int still_to_find,
2312                               int budget,
2313                               bool not_at_start) {
2314  if (budget <= 0) return 0;
2315  // If we know we are not at the start and we are asked "how many characters
2316  // will you match if you succeed?" then we can answer anything since false
2317  // implies false.  So lets just return the max answer (still_to_find) since
2318  // that won't prevent us from preloading a lot of characters for the other
2319  // branches in the node graph.
2320  if (assertion_type() == AT_START && not_at_start) return still_to_find;
2321  return on_success()->EatsAtLeast(still_to_find,
2322                                   budget - 1,
2323                                   not_at_start);
2324}
2325
2326
2327void AssertionNode::FillInBMInfo(Isolate* isolate, int offset, int budget,
2328                                 BoyerMooreLookahead* bm, bool not_at_start) {
2329  // Match the behaviour of EatsAtLeast on this node.
2330  if (assertion_type() == AT_START && not_at_start) return;
2331  on_success()->FillInBMInfo(isolate, offset, budget - 1, bm, not_at_start);
2332  SaveBMInfo(bm, not_at_start, offset);
2333}
2334
2335
2336int BackReferenceNode::EatsAtLeast(int still_to_find,
2337                                   int budget,
2338                                   bool not_at_start) {
2339  if (read_backward()) return 0;
2340  if (budget <= 0) return 0;
2341  return on_success()->EatsAtLeast(still_to_find,
2342                                   budget - 1,
2343                                   not_at_start);
2344}
2345
2346
2347int TextNode::EatsAtLeast(int still_to_find,
2348                          int budget,
2349                          bool not_at_start) {
2350  if (read_backward()) return 0;
2351  int answer = Length();
2352  if (answer >= still_to_find) return answer;
2353  if (budget <= 0) return answer;
2354  // We are not at start after this node so we set the last argument to 'true'.
2355  return answer + on_success()->EatsAtLeast(still_to_find - answer,
2356                                            budget - 1,
2357                                            true);
2358}
2359
2360
2361int NegativeLookaroundChoiceNode::EatsAtLeast(int still_to_find, int budget,
2362                                              bool not_at_start) {
2363  if (budget <= 0) return 0;
2364  // Alternative 0 is the negative lookahead, alternative 1 is what comes
2365  // afterwards.
2366  RegExpNode* node = alternatives_->at(1).node();
2367  return node->EatsAtLeast(still_to_find, budget - 1, not_at_start);
2368}
2369
2370
2371void NegativeLookaroundChoiceNode::GetQuickCheckDetails(
2372    QuickCheckDetails* details, RegExpCompiler* compiler, int filled_in,
2373    bool not_at_start) {
2374  // Alternative 0 is the negative lookahead, alternative 1 is what comes
2375  // afterwards.
2376  RegExpNode* node = alternatives_->at(1).node();
2377  return node->GetQuickCheckDetails(details, compiler, filled_in, not_at_start);
2378}
2379
2380
2381int ChoiceNode::EatsAtLeastHelper(int still_to_find,
2382                                  int budget,
2383                                  RegExpNode* ignore_this_node,
2384                                  bool not_at_start) {
2385  if (budget <= 0) return 0;
2386  int min = 100;
2387  int choice_count = alternatives_->length();
2388  budget = (budget - 1) / choice_count;
2389  for (int i = 0; i < choice_count; i++) {
2390    RegExpNode* node = alternatives_->at(i).node();
2391    if (node == ignore_this_node) continue;
2392    int node_eats_at_least =
2393        node->EatsAtLeast(still_to_find, budget, not_at_start);
2394    if (node_eats_at_least < min) min = node_eats_at_least;
2395    if (min == 0) return 0;
2396  }
2397  return min;
2398}
2399
2400
2401int LoopChoiceNode::EatsAtLeast(int still_to_find,
2402                                int budget,
2403                                bool not_at_start) {
2404  return EatsAtLeastHelper(still_to_find,
2405                           budget - 1,
2406                           loop_node_,
2407                           not_at_start);
2408}
2409
2410
2411int ChoiceNode::EatsAtLeast(int still_to_find,
2412                            int budget,
2413                            bool not_at_start) {
2414  return EatsAtLeastHelper(still_to_find,
2415                           budget,
2416                           NULL,
2417                           not_at_start);
2418}
2419
2420
2421// Takes the left-most 1-bit and smears it out, setting all bits to its right.
2422static inline uint32_t SmearBitsRight(uint32_t v) {
2423  v |= v >> 1;
2424  v |= v >> 2;
2425  v |= v >> 4;
2426  v |= v >> 8;
2427  v |= v >> 16;
2428  return v;
2429}
2430
2431
2432bool QuickCheckDetails::Rationalize(bool asc) {
2433  bool found_useful_op = false;
2434  uint32_t char_mask;
2435  if (asc) {
2436    char_mask = String::kMaxOneByteCharCode;
2437  } else {
2438    char_mask = String::kMaxUtf16CodeUnit;
2439  }
2440  mask_ = 0;
2441  value_ = 0;
2442  int char_shift = 0;
2443  for (int i = 0; i < characters_; i++) {
2444    Position* pos = &positions_[i];
2445    if ((pos->mask & String::kMaxOneByteCharCode) != 0) {
2446      found_useful_op = true;
2447    }
2448    mask_ |= (pos->mask & char_mask) << char_shift;
2449    value_ |= (pos->value & char_mask) << char_shift;
2450    char_shift += asc ? 8 : 16;
2451  }
2452  return found_useful_op;
2453}
2454
2455
2456bool RegExpNode::EmitQuickCheck(RegExpCompiler* compiler,
2457                                Trace* bounds_check_trace,
2458                                Trace* trace,
2459                                bool preload_has_checked_bounds,
2460                                Label* on_possible_success,
2461                                QuickCheckDetails* details,
2462                                bool fall_through_on_failure) {
2463  if (details->characters() == 0) return false;
2464  GetQuickCheckDetails(
2465      details, compiler, 0, trace->at_start() == Trace::FALSE_VALUE);
2466  if (details->cannot_match()) return false;
2467  if (!details->Rationalize(compiler->one_byte())) return false;
2468  DCHECK(details->characters() == 1 ||
2469         compiler->macro_assembler()->CanReadUnaligned());
2470  uint32_t mask = details->mask();
2471  uint32_t value = details->value();
2472
2473  RegExpMacroAssembler* assembler = compiler->macro_assembler();
2474
2475  if (trace->characters_preloaded() != details->characters()) {
2476    DCHECK(trace->cp_offset() == bounds_check_trace->cp_offset());
2477    // We are attempting to preload the minimum number of characters
2478    // any choice would eat, so if the bounds check fails, then none of the
2479    // choices can succeed, so we can just immediately backtrack, rather
2480    // than go to the next choice.
2481    assembler->LoadCurrentCharacter(trace->cp_offset(),
2482                                    bounds_check_trace->backtrack(),
2483                                    !preload_has_checked_bounds,
2484                                    details->characters());
2485  }
2486
2487
2488  bool need_mask = true;
2489
2490  if (details->characters() == 1) {
2491    // If number of characters preloaded is 1 then we used a byte or 16 bit
2492    // load so the value is already masked down.
2493    uint32_t char_mask;
2494    if (compiler->one_byte()) {
2495      char_mask = String::kMaxOneByteCharCode;
2496    } else {
2497      char_mask = String::kMaxUtf16CodeUnit;
2498    }
2499    if ((mask & char_mask) == char_mask) need_mask = false;
2500    mask &= char_mask;
2501  } else {
2502    // For 2-character preloads in one-byte mode or 1-character preloads in
2503    // two-byte mode we also use a 16 bit load with zero extend.
2504    static const uint32_t kTwoByteMask = 0xffff;
2505    static const uint32_t kFourByteMask = 0xffffffff;
2506    if (details->characters() == 2 && compiler->one_byte()) {
2507      if ((mask & kTwoByteMask) == kTwoByteMask) need_mask = false;
2508    } else if (details->characters() == 1 && !compiler->one_byte()) {
2509      if ((mask & kTwoByteMask) == kTwoByteMask) need_mask = false;
2510    } else {
2511      if (mask == kFourByteMask) need_mask = false;
2512    }
2513  }
2514
2515  if (fall_through_on_failure) {
2516    if (need_mask) {
2517      assembler->CheckCharacterAfterAnd(value, mask, on_possible_success);
2518    } else {
2519      assembler->CheckCharacter(value, on_possible_success);
2520    }
2521  } else {
2522    if (need_mask) {
2523      assembler->CheckNotCharacterAfterAnd(value, mask, trace->backtrack());
2524    } else {
2525      assembler->CheckNotCharacter(value, trace->backtrack());
2526    }
2527  }
2528  return true;
2529}
2530
2531
2532// Here is the meat of GetQuickCheckDetails (see also the comment on the
2533// super-class in the .h file).
2534//
2535// We iterate along the text object, building up for each character a
2536// mask and value that can be used to test for a quick failure to match.
2537// The masks and values for the positions will be combined into a single
2538// machine word for the current character width in order to be used in
2539// generating a quick check.
2540void TextNode::GetQuickCheckDetails(QuickCheckDetails* details,
2541                                    RegExpCompiler* compiler,
2542                                    int characters_filled_in,
2543                                    bool not_at_start) {
2544  // Do not collect any quick check details if the text node reads backward,
2545  // since it reads in the opposite direction than we use for quick checks.
2546  if (read_backward()) return;
2547  Isolate* isolate = compiler->macro_assembler()->isolate();
2548  DCHECK(characters_filled_in < details->characters());
2549  int characters = details->characters();
2550  int char_mask;
2551  if (compiler->one_byte()) {
2552    char_mask = String::kMaxOneByteCharCode;
2553  } else {
2554    char_mask = String::kMaxUtf16CodeUnit;
2555  }
2556  for (int k = 0; k < elements()->length(); k++) {
2557    TextElement elm = elements()->at(k);
2558    if (elm.text_type() == TextElement::ATOM) {
2559      Vector<const uc16> quarks = elm.atom()->data();
2560      for (int i = 0; i < characters && i < quarks.length(); i++) {
2561        QuickCheckDetails::Position* pos =
2562            details->positions(characters_filled_in);
2563        uc16 c = quarks[i];
2564        if (compiler->ignore_case()) {
2565          unibrow::uchar chars[unibrow::Ecma262UnCanonicalize::kMaxWidth];
2566          int length = GetCaseIndependentLetters(isolate, c,
2567                                                 compiler->one_byte(), chars);
2568          if (length == 0) {
2569            // This can happen because all case variants are non-Latin1, but we
2570            // know the input is Latin1.
2571            details->set_cannot_match();
2572            pos->determines_perfectly = false;
2573            return;
2574          }
2575          if (length == 1) {
2576            // This letter has no case equivalents, so it's nice and simple
2577            // and the mask-compare will determine definitely whether we have
2578            // a match at this character position.
2579            pos->mask = char_mask;
2580            pos->value = c;
2581            pos->determines_perfectly = true;
2582          } else {
2583            uint32_t common_bits = char_mask;
2584            uint32_t bits = chars[0];
2585            for (int j = 1; j < length; j++) {
2586              uint32_t differing_bits = ((chars[j] & common_bits) ^ bits);
2587              common_bits ^= differing_bits;
2588              bits &= common_bits;
2589            }
2590            // If length is 2 and common bits has only one zero in it then
2591            // our mask and compare instruction will determine definitely
2592            // whether we have a match at this character position.  Otherwise
2593            // it can only be an approximate check.
2594            uint32_t one_zero = (common_bits | ~char_mask);
2595            if (length == 2 && ((~one_zero) & ((~one_zero) - 1)) == 0) {
2596              pos->determines_perfectly = true;
2597            }
2598            pos->mask = common_bits;
2599            pos->value = bits;
2600          }
2601        } else {
2602          // Don't ignore case.  Nice simple case where the mask-compare will
2603          // determine definitely whether we have a match at this character
2604          // position.
2605          if (c > char_mask) {
2606            details->set_cannot_match();
2607            pos->determines_perfectly = false;
2608            return;
2609          }
2610          pos->mask = char_mask;
2611          pos->value = c;
2612          pos->determines_perfectly = true;
2613        }
2614        characters_filled_in++;
2615        DCHECK(characters_filled_in <= details->characters());
2616        if (characters_filled_in == details->characters()) {
2617          return;
2618        }
2619      }
2620    } else {
2621      QuickCheckDetails::Position* pos =
2622          details->positions(characters_filled_in);
2623      RegExpCharacterClass* tree = elm.char_class();
2624      ZoneList<CharacterRange>* ranges = tree->ranges(zone());
2625      if (tree->is_negated()) {
2626        // A quick check uses multi-character mask and compare.  There is no
2627        // useful way to incorporate a negative char class into this scheme
2628        // so we just conservatively create a mask and value that will always
2629        // succeed.
2630        pos->mask = 0;
2631        pos->value = 0;
2632      } else {
2633        int first_range = 0;
2634        while (ranges->at(first_range).from() > char_mask) {
2635          first_range++;
2636          if (first_range == ranges->length()) {
2637            details->set_cannot_match();
2638            pos->determines_perfectly = false;
2639            return;
2640          }
2641        }
2642        CharacterRange range = ranges->at(first_range);
2643        uc16 from = range.from();
2644        uc16 to = range.to();
2645        if (to > char_mask) {
2646          to = char_mask;
2647        }
2648        uint32_t differing_bits = (from ^ to);
2649        // A mask and compare is only perfect if the differing bits form a
2650        // number like 00011111 with one single block of trailing 1s.
2651        if ((differing_bits & (differing_bits + 1)) == 0 &&
2652             from + differing_bits == to) {
2653          pos->determines_perfectly = true;
2654        }
2655        uint32_t common_bits = ~SmearBitsRight(differing_bits);
2656        uint32_t bits = (from & common_bits);
2657        for (int i = first_range + 1; i < ranges->length(); i++) {
2658          CharacterRange range = ranges->at(i);
2659          uc16 from = range.from();
2660          uc16 to = range.to();
2661          if (from > char_mask) continue;
2662          if (to > char_mask) to = char_mask;
2663          // Here we are combining more ranges into the mask and compare
2664          // value.  With each new range the mask becomes more sparse and
2665          // so the chances of a false positive rise.  A character class
2666          // with multiple ranges is assumed never to be equivalent to a
2667          // mask and compare operation.
2668          pos->determines_perfectly = false;
2669          uint32_t new_common_bits = (from ^ to);
2670          new_common_bits = ~SmearBitsRight(new_common_bits);
2671          common_bits &= new_common_bits;
2672          bits &= new_common_bits;
2673          uint32_t differing_bits = (from & common_bits) ^ bits;
2674          common_bits ^= differing_bits;
2675          bits &= common_bits;
2676        }
2677        pos->mask = common_bits;
2678        pos->value = bits;
2679      }
2680      characters_filled_in++;
2681      DCHECK(characters_filled_in <= details->characters());
2682      if (characters_filled_in == details->characters()) {
2683        return;
2684      }
2685    }
2686  }
2687  DCHECK(characters_filled_in != details->characters());
2688  if (!details->cannot_match()) {
2689    on_success()-> GetQuickCheckDetails(details,
2690                                        compiler,
2691                                        characters_filled_in,
2692                                        true);
2693  }
2694}
2695
2696
2697void QuickCheckDetails::Clear() {
2698  for (int i = 0; i < characters_; i++) {
2699    positions_[i].mask = 0;
2700    positions_[i].value = 0;
2701    positions_[i].determines_perfectly = false;
2702  }
2703  characters_ = 0;
2704}
2705
2706
2707void QuickCheckDetails::Advance(int by, bool one_byte) {
2708  if (by >= characters_ || by < 0) {
2709    DCHECK_IMPLIES(by < 0, characters_ == 0);
2710    Clear();
2711    return;
2712  }
2713  DCHECK_LE(characters_ - by, 4);
2714  DCHECK_LE(characters_, 4);
2715  for (int i = 0; i < characters_ - by; i++) {
2716    positions_[i] = positions_[by + i];
2717  }
2718  for (int i = characters_ - by; i < characters_; i++) {
2719    positions_[i].mask = 0;
2720    positions_[i].value = 0;
2721    positions_[i].determines_perfectly = false;
2722  }
2723  characters_ -= by;
2724  // We could change mask_ and value_ here but we would never advance unless
2725  // they had already been used in a check and they won't be used again because
2726  // it would gain us nothing.  So there's no point.
2727}
2728
2729
2730void QuickCheckDetails::Merge(QuickCheckDetails* other, int from_index) {
2731  DCHECK(characters_ == other->characters_);
2732  if (other->cannot_match_) {
2733    return;
2734  }
2735  if (cannot_match_) {
2736    *this = *other;
2737    return;
2738  }
2739  for (int i = from_index; i < characters_; i++) {
2740    QuickCheckDetails::Position* pos = positions(i);
2741    QuickCheckDetails::Position* other_pos = other->positions(i);
2742    if (pos->mask != other_pos->mask ||
2743        pos->value != other_pos->value ||
2744        !other_pos->determines_perfectly) {
2745      // Our mask-compare operation will be approximate unless we have the
2746      // exact same operation on both sides of the alternation.
2747      pos->determines_perfectly = false;
2748    }
2749    pos->mask &= other_pos->mask;
2750    pos->value &= pos->mask;
2751    other_pos->value &= pos->mask;
2752    uc16 differing_bits = (pos->value ^ other_pos->value);
2753    pos->mask &= ~differing_bits;
2754    pos->value &= pos->mask;
2755  }
2756}
2757
2758
2759class VisitMarker {
2760 public:
2761  explicit VisitMarker(NodeInfo* info) : info_(info) {
2762    DCHECK(!info->visited);
2763    info->visited = true;
2764  }
2765  ~VisitMarker() {
2766    info_->visited = false;
2767  }
2768 private:
2769  NodeInfo* info_;
2770};
2771
2772
2773RegExpNode* SeqRegExpNode::FilterOneByte(int depth, bool ignore_case) {
2774  if (info()->replacement_calculated) return replacement();
2775  if (depth < 0) return this;
2776  DCHECK(!info()->visited);
2777  VisitMarker marker(info());
2778  return FilterSuccessor(depth - 1, ignore_case);
2779}
2780
2781
2782RegExpNode* SeqRegExpNode::FilterSuccessor(int depth, bool ignore_case) {
2783  RegExpNode* next = on_success_->FilterOneByte(depth - 1, ignore_case);
2784  if (next == NULL) return set_replacement(NULL);
2785  on_success_ = next;
2786  return set_replacement(this);
2787}
2788
2789
2790// We need to check for the following characters: 0x39c 0x3bc 0x178.
2791static inline bool RangeContainsLatin1Equivalents(CharacterRange range) {
2792  // TODO(dcarney): this could be a lot more efficient.
2793  return range.Contains(0x39c) ||
2794      range.Contains(0x3bc) || range.Contains(0x178);
2795}
2796
2797
2798static bool RangesContainLatin1Equivalents(ZoneList<CharacterRange>* ranges) {
2799  for (int i = 0; i < ranges->length(); i++) {
2800    // TODO(dcarney): this could be a lot more efficient.
2801    if (RangeContainsLatin1Equivalents(ranges->at(i))) return true;
2802  }
2803  return false;
2804}
2805
2806
2807RegExpNode* TextNode::FilterOneByte(int depth, bool ignore_case) {
2808  if (info()->replacement_calculated) return replacement();
2809  if (depth < 0) return this;
2810  DCHECK(!info()->visited);
2811  VisitMarker marker(info());
2812  int element_count = elements()->length();
2813  for (int i = 0; i < element_count; i++) {
2814    TextElement elm = elements()->at(i);
2815    if (elm.text_type() == TextElement::ATOM) {
2816      Vector<const uc16> quarks = elm.atom()->data();
2817      for (int j = 0; j < quarks.length(); j++) {
2818        uint16_t c = quarks[j];
2819        if (c <= String::kMaxOneByteCharCode) continue;
2820        if (!ignore_case) return set_replacement(NULL);
2821        // Here, we need to check for characters whose upper and lower cases
2822        // are outside the Latin-1 range.
2823        uint16_t converted = unibrow::Latin1::ConvertNonLatin1ToLatin1(c);
2824        // Character is outside Latin-1 completely
2825        if (converted == 0) return set_replacement(NULL);
2826        // Convert quark to Latin-1 in place.
2827        uint16_t* copy = const_cast<uint16_t*>(quarks.start());
2828        copy[j] = converted;
2829      }
2830    } else {
2831      DCHECK(elm.text_type() == TextElement::CHAR_CLASS);
2832      RegExpCharacterClass* cc = elm.char_class();
2833      ZoneList<CharacterRange>* ranges = cc->ranges(zone());
2834      CharacterRange::Canonicalize(ranges);
2835      // Now they are in order so we only need to look at the first.
2836      int range_count = ranges->length();
2837      if (cc->is_negated()) {
2838        if (range_count != 0 &&
2839            ranges->at(0).from() == 0 &&
2840            ranges->at(0).to() >= String::kMaxOneByteCharCode) {
2841          // This will be handled in a later filter.
2842          if (ignore_case && RangesContainLatin1Equivalents(ranges)) continue;
2843          return set_replacement(NULL);
2844        }
2845      } else {
2846        if (range_count == 0 ||
2847            ranges->at(0).from() > String::kMaxOneByteCharCode) {
2848          // This will be handled in a later filter.
2849          if (ignore_case && RangesContainLatin1Equivalents(ranges)) continue;
2850          return set_replacement(NULL);
2851        }
2852      }
2853    }
2854  }
2855  return FilterSuccessor(depth - 1, ignore_case);
2856}
2857
2858
2859RegExpNode* LoopChoiceNode::FilterOneByte(int depth, bool ignore_case) {
2860  if (info()->replacement_calculated) return replacement();
2861  if (depth < 0) return this;
2862  if (info()->visited) return this;
2863  {
2864    VisitMarker marker(info());
2865
2866    RegExpNode* continue_replacement =
2867        continue_node_->FilterOneByte(depth - 1, ignore_case);
2868    // If we can't continue after the loop then there is no sense in doing the
2869    // loop.
2870    if (continue_replacement == NULL) return set_replacement(NULL);
2871  }
2872
2873  return ChoiceNode::FilterOneByte(depth - 1, ignore_case);
2874}
2875
2876
2877RegExpNode* ChoiceNode::FilterOneByte(int depth, bool ignore_case) {
2878  if (info()->replacement_calculated) return replacement();
2879  if (depth < 0) return this;
2880  if (info()->visited) return this;
2881  VisitMarker marker(info());
2882  int choice_count = alternatives_->length();
2883
2884  for (int i = 0; i < choice_count; i++) {
2885    GuardedAlternative alternative = alternatives_->at(i);
2886    if (alternative.guards() != NULL && alternative.guards()->length() != 0) {
2887      set_replacement(this);
2888      return this;
2889    }
2890  }
2891
2892  int surviving = 0;
2893  RegExpNode* survivor = NULL;
2894  for (int i = 0; i < choice_count; i++) {
2895    GuardedAlternative alternative = alternatives_->at(i);
2896    RegExpNode* replacement =
2897        alternative.node()->FilterOneByte(depth - 1, ignore_case);
2898    DCHECK(replacement != this);  // No missing EMPTY_MATCH_CHECK.
2899    if (replacement != NULL) {
2900      alternatives_->at(i).set_node(replacement);
2901      surviving++;
2902      survivor = replacement;
2903    }
2904  }
2905  if (surviving < 2) return set_replacement(survivor);
2906
2907  set_replacement(this);
2908  if (surviving == choice_count) {
2909    return this;
2910  }
2911  // Only some of the nodes survived the filtering.  We need to rebuild the
2912  // alternatives list.
2913  ZoneList<GuardedAlternative>* new_alternatives =
2914      new(zone()) ZoneList<GuardedAlternative>(surviving, zone());
2915  for (int i = 0; i < choice_count; i++) {
2916    RegExpNode* replacement =
2917        alternatives_->at(i).node()->FilterOneByte(depth - 1, ignore_case);
2918    if (replacement != NULL) {
2919      alternatives_->at(i).set_node(replacement);
2920      new_alternatives->Add(alternatives_->at(i), zone());
2921    }
2922  }
2923  alternatives_ = new_alternatives;
2924  return this;
2925}
2926
2927
2928RegExpNode* NegativeLookaroundChoiceNode::FilterOneByte(int depth,
2929                                                        bool ignore_case) {
2930  if (info()->replacement_calculated) return replacement();
2931  if (depth < 0) return this;
2932  if (info()->visited) return this;
2933  VisitMarker marker(info());
2934  // Alternative 0 is the negative lookahead, alternative 1 is what comes
2935  // afterwards.
2936  RegExpNode* node = alternatives_->at(1).node();
2937  RegExpNode* replacement = node->FilterOneByte(depth - 1, ignore_case);
2938  if (replacement == NULL) return set_replacement(NULL);
2939  alternatives_->at(1).set_node(replacement);
2940
2941  RegExpNode* neg_node = alternatives_->at(0).node();
2942  RegExpNode* neg_replacement = neg_node->FilterOneByte(depth - 1, ignore_case);
2943  // If the negative lookahead is always going to fail then
2944  // we don't need to check it.
2945  if (neg_replacement == NULL) return set_replacement(replacement);
2946  alternatives_->at(0).set_node(neg_replacement);
2947  return set_replacement(this);
2948}
2949
2950
2951void LoopChoiceNode::GetQuickCheckDetails(QuickCheckDetails* details,
2952                                          RegExpCompiler* compiler,
2953                                          int characters_filled_in,
2954                                          bool not_at_start) {
2955  if (body_can_be_zero_length_ || info()->visited) return;
2956  VisitMarker marker(info());
2957  return ChoiceNode::GetQuickCheckDetails(details,
2958                                          compiler,
2959                                          characters_filled_in,
2960                                          not_at_start);
2961}
2962
2963
2964void LoopChoiceNode::FillInBMInfo(Isolate* isolate, int offset, int budget,
2965                                  BoyerMooreLookahead* bm, bool not_at_start) {
2966  if (body_can_be_zero_length_ || budget <= 0) {
2967    bm->SetRest(offset);
2968    SaveBMInfo(bm, not_at_start, offset);
2969    return;
2970  }
2971  ChoiceNode::FillInBMInfo(isolate, offset, budget - 1, bm, not_at_start);
2972  SaveBMInfo(bm, not_at_start, offset);
2973}
2974
2975
2976void ChoiceNode::GetQuickCheckDetails(QuickCheckDetails* details,
2977                                      RegExpCompiler* compiler,
2978                                      int characters_filled_in,
2979                                      bool not_at_start) {
2980  not_at_start = (not_at_start || not_at_start_);
2981  int choice_count = alternatives_->length();
2982  DCHECK(choice_count > 0);
2983  alternatives_->at(0).node()->GetQuickCheckDetails(details,
2984                                                    compiler,
2985                                                    characters_filled_in,
2986                                                    not_at_start);
2987  for (int i = 1; i < choice_count; i++) {
2988    QuickCheckDetails new_details(details->characters());
2989    RegExpNode* node = alternatives_->at(i).node();
2990    node->GetQuickCheckDetails(&new_details, compiler,
2991                               characters_filled_in,
2992                               not_at_start);
2993    // Here we merge the quick match details of the two branches.
2994    details->Merge(&new_details, characters_filled_in);
2995  }
2996}
2997
2998
2999// Check for [0-9A-Z_a-z].
3000static void EmitWordCheck(RegExpMacroAssembler* assembler,
3001                          Label* word,
3002                          Label* non_word,
3003                          bool fall_through_on_word) {
3004  if (assembler->CheckSpecialCharacterClass(
3005          fall_through_on_word ? 'w' : 'W',
3006          fall_through_on_word ? non_word : word)) {
3007    // Optimized implementation available.
3008    return;
3009  }
3010  assembler->CheckCharacterGT('z', non_word);
3011  assembler->CheckCharacterLT('0', non_word);
3012  assembler->CheckCharacterGT('a' - 1, word);
3013  assembler->CheckCharacterLT('9' + 1, word);
3014  assembler->CheckCharacterLT('A', non_word);
3015  assembler->CheckCharacterLT('Z' + 1, word);
3016  if (fall_through_on_word) {
3017    assembler->CheckNotCharacter('_', non_word);
3018  } else {
3019    assembler->CheckCharacter('_', word);
3020  }
3021}
3022
3023
3024// Emit the code to check for a ^ in multiline mode (1-character lookbehind
3025// that matches newline or the start of input).
3026static void EmitHat(RegExpCompiler* compiler,
3027                    RegExpNode* on_success,
3028                    Trace* trace) {
3029  RegExpMacroAssembler* assembler = compiler->macro_assembler();
3030  // We will be loading the previous character into the current character
3031  // register.
3032  Trace new_trace(*trace);
3033  new_trace.InvalidateCurrentCharacter();
3034
3035  Label ok;
3036  if (new_trace.cp_offset() == 0) {
3037    // The start of input counts as a newline in this context, so skip to
3038    // ok if we are at the start.
3039    assembler->CheckAtStart(&ok);
3040  }
3041  // We already checked that we are not at the start of input so it must be
3042  // OK to load the previous character.
3043  assembler->LoadCurrentCharacter(new_trace.cp_offset() -1,
3044                                  new_trace.backtrack(),
3045                                  false);
3046  if (!assembler->CheckSpecialCharacterClass('n',
3047                                             new_trace.backtrack())) {
3048    // Newline means \n, \r, 0x2028 or 0x2029.
3049    if (!compiler->one_byte()) {
3050      assembler->CheckCharacterAfterAnd(0x2028, 0xfffe, &ok);
3051    }
3052    assembler->CheckCharacter('\n', &ok);
3053    assembler->CheckNotCharacter('\r', new_trace.backtrack());
3054  }
3055  assembler->Bind(&ok);
3056  on_success->Emit(compiler, &new_trace);
3057}
3058
3059
3060// Emit the code to handle \b and \B (word-boundary or non-word-boundary).
3061void AssertionNode::EmitBoundaryCheck(RegExpCompiler* compiler, Trace* trace) {
3062  RegExpMacroAssembler* assembler = compiler->macro_assembler();
3063  Isolate* isolate = assembler->isolate();
3064  Trace::TriBool next_is_word_character = Trace::UNKNOWN;
3065  bool not_at_start = (trace->at_start() == Trace::FALSE_VALUE);
3066  BoyerMooreLookahead* lookahead = bm_info(not_at_start);
3067  if (lookahead == NULL) {
3068    int eats_at_least =
3069        Min(kMaxLookaheadForBoyerMoore, EatsAtLeast(kMaxLookaheadForBoyerMoore,
3070                                                    kRecursionBudget,
3071                                                    not_at_start));
3072    if (eats_at_least >= 1) {
3073      BoyerMooreLookahead* bm =
3074          new(zone()) BoyerMooreLookahead(eats_at_least, compiler, zone());
3075      FillInBMInfo(isolate, 0, kRecursionBudget, bm, not_at_start);
3076      if (bm->at(0)->is_non_word())
3077        next_is_word_character = Trace::FALSE_VALUE;
3078      if (bm->at(0)->is_word()) next_is_word_character = Trace::TRUE_VALUE;
3079    }
3080  } else {
3081    if (lookahead->at(0)->is_non_word())
3082      next_is_word_character = Trace::FALSE_VALUE;
3083    if (lookahead->at(0)->is_word())
3084      next_is_word_character = Trace::TRUE_VALUE;
3085  }
3086  bool at_boundary = (assertion_type_ == AssertionNode::AT_BOUNDARY);
3087  if (next_is_word_character == Trace::UNKNOWN) {
3088    Label before_non_word;
3089    Label before_word;
3090    if (trace->characters_preloaded() != 1) {
3091      assembler->LoadCurrentCharacter(trace->cp_offset(), &before_non_word);
3092    }
3093    // Fall through on non-word.
3094    EmitWordCheck(assembler, &before_word, &before_non_word, false);
3095    // Next character is not a word character.
3096    assembler->Bind(&before_non_word);
3097    Label ok;
3098    BacktrackIfPrevious(compiler, trace, at_boundary ? kIsNonWord : kIsWord);
3099    assembler->GoTo(&ok);
3100
3101    assembler->Bind(&before_word);
3102    BacktrackIfPrevious(compiler, trace, at_boundary ? kIsWord : kIsNonWord);
3103    assembler->Bind(&ok);
3104  } else if (next_is_word_character == Trace::TRUE_VALUE) {
3105    BacktrackIfPrevious(compiler, trace, at_boundary ? kIsWord : kIsNonWord);
3106  } else {
3107    DCHECK(next_is_word_character == Trace::FALSE_VALUE);
3108    BacktrackIfPrevious(compiler, trace, at_boundary ? kIsNonWord : kIsWord);
3109  }
3110}
3111
3112
3113void AssertionNode::BacktrackIfPrevious(
3114    RegExpCompiler* compiler,
3115    Trace* trace,
3116    AssertionNode::IfPrevious backtrack_if_previous) {
3117  RegExpMacroAssembler* assembler = compiler->macro_assembler();
3118  Trace new_trace(*trace);
3119  new_trace.InvalidateCurrentCharacter();
3120
3121  Label fall_through, dummy;
3122
3123  Label* non_word = backtrack_if_previous == kIsNonWord ?
3124                    new_trace.backtrack() :
3125                    &fall_through;
3126  Label* word = backtrack_if_previous == kIsNonWord ?
3127                &fall_through :
3128                new_trace.backtrack();
3129
3130  if (new_trace.cp_offset() == 0) {
3131    // The start of input counts as a non-word character, so the question is
3132    // decided if we are at the start.
3133    assembler->CheckAtStart(non_word);
3134  }
3135  // We already checked that we are not at the start of input so it must be
3136  // OK to load the previous character.
3137  assembler->LoadCurrentCharacter(new_trace.cp_offset() - 1, &dummy, false);
3138  EmitWordCheck(assembler, word, non_word, backtrack_if_previous == kIsNonWord);
3139
3140  assembler->Bind(&fall_through);
3141  on_success()->Emit(compiler, &new_trace);
3142}
3143
3144
3145void AssertionNode::GetQuickCheckDetails(QuickCheckDetails* details,
3146                                         RegExpCompiler* compiler,
3147                                         int filled_in,
3148                                         bool not_at_start) {
3149  if (assertion_type_ == AT_START && not_at_start) {
3150    details->set_cannot_match();
3151    return;
3152  }
3153  return on_success()->GetQuickCheckDetails(details,
3154                                            compiler,
3155                                            filled_in,
3156                                            not_at_start);
3157}
3158
3159
3160void AssertionNode::Emit(RegExpCompiler* compiler, Trace* trace) {
3161  RegExpMacroAssembler* assembler = compiler->macro_assembler();
3162  switch (assertion_type_) {
3163    case AT_END: {
3164      Label ok;
3165      assembler->CheckPosition(trace->cp_offset(), &ok);
3166      assembler->GoTo(trace->backtrack());
3167      assembler->Bind(&ok);
3168      break;
3169    }
3170    case AT_START: {
3171      if (trace->at_start() == Trace::FALSE_VALUE) {
3172        assembler->GoTo(trace->backtrack());
3173        return;
3174      }
3175      if (trace->at_start() == Trace::UNKNOWN) {
3176        assembler->CheckNotAtStart(trace->cp_offset(), trace->backtrack());
3177        Trace at_start_trace = *trace;
3178        at_start_trace.set_at_start(Trace::TRUE_VALUE);
3179        on_success()->Emit(compiler, &at_start_trace);
3180        return;
3181      }
3182    }
3183    break;
3184    case AFTER_NEWLINE:
3185      EmitHat(compiler, on_success(), trace);
3186      return;
3187    case AT_BOUNDARY:
3188    case AT_NON_BOUNDARY: {
3189      EmitBoundaryCheck(compiler, trace);
3190      return;
3191    }
3192  }
3193  on_success()->Emit(compiler, trace);
3194}
3195
3196
3197static bool DeterminedAlready(QuickCheckDetails* quick_check, int offset) {
3198  if (quick_check == NULL) return false;
3199  if (offset >= quick_check->characters()) return false;
3200  return quick_check->positions(offset)->determines_perfectly;
3201}
3202
3203
3204static void UpdateBoundsCheck(int index, int* checked_up_to) {
3205  if (index > *checked_up_to) {
3206    *checked_up_to = index;
3207  }
3208}
3209
3210
3211// We call this repeatedly to generate code for each pass over the text node.
3212// The passes are in increasing order of difficulty because we hope one
3213// of the first passes will fail in which case we are saved the work of the
3214// later passes.  for example for the case independent regexp /%[asdfghjkl]a/
3215// we will check the '%' in the first pass, the case independent 'a' in the
3216// second pass and the character class in the last pass.
3217//
3218// The passes are done from right to left, so for example to test for /bar/
3219// we will first test for an 'r' with offset 2, then an 'a' with offset 1
3220// and then a 'b' with offset 0.  This means we can avoid the end-of-input
3221// bounds check most of the time.  In the example we only need to check for
3222// end-of-input when loading the putative 'r'.
3223//
3224// A slight complication involves the fact that the first character may already
3225// be fetched into a register by the previous node.  In this case we want to
3226// do the test for that character first.  We do this in separate passes.  The
3227// 'preloaded' argument indicates that we are doing such a 'pass'.  If such a
3228// pass has been performed then subsequent passes will have true in
3229// first_element_checked to indicate that that character does not need to be
3230// checked again.
3231//
3232// In addition to all this we are passed a Trace, which can
3233// contain an AlternativeGeneration object.  In this AlternativeGeneration
3234// object we can see details of any quick check that was already passed in
3235// order to get to the code we are now generating.  The quick check can involve
3236// loading characters, which means we do not need to recheck the bounds
3237// up to the limit the quick check already checked.  In addition the quick
3238// check can have involved a mask and compare operation which may simplify
3239// or obviate the need for further checks at some character positions.
3240void TextNode::TextEmitPass(RegExpCompiler* compiler,
3241                            TextEmitPassType pass,
3242                            bool preloaded,
3243                            Trace* trace,
3244                            bool first_element_checked,
3245                            int* checked_up_to) {
3246  RegExpMacroAssembler* assembler = compiler->macro_assembler();
3247  Isolate* isolate = assembler->isolate();
3248  bool one_byte = compiler->one_byte();
3249  Label* backtrack = trace->backtrack();
3250  QuickCheckDetails* quick_check = trace->quick_check_performed();
3251  int element_count = elements()->length();
3252  int backward_offset = read_backward() ? -Length() : 0;
3253  for (int i = preloaded ? 0 : element_count - 1; i >= 0; i--) {
3254    TextElement elm = elements()->at(i);
3255    int cp_offset = trace->cp_offset() + elm.cp_offset() + backward_offset;
3256    if (elm.text_type() == TextElement::ATOM) {
3257      Vector<const uc16> quarks = elm.atom()->data();
3258      for (int j = preloaded ? 0 : quarks.length() - 1; j >= 0; j--) {
3259        if (first_element_checked && i == 0 && j == 0) continue;
3260        if (DeterminedAlready(quick_check, elm.cp_offset() + j)) continue;
3261        EmitCharacterFunction* emit_function = NULL;
3262        switch (pass) {
3263          case NON_LATIN1_MATCH:
3264            DCHECK(one_byte);
3265            if (quarks[j] > String::kMaxOneByteCharCode) {
3266              assembler->GoTo(backtrack);
3267              return;
3268            }
3269            break;
3270          case NON_LETTER_CHARACTER_MATCH:
3271            emit_function = &EmitAtomNonLetter;
3272            break;
3273          case SIMPLE_CHARACTER_MATCH:
3274            emit_function = &EmitSimpleCharacter;
3275            break;
3276          case CASE_CHARACTER_MATCH:
3277            emit_function = &EmitAtomLetter;
3278            break;
3279          default:
3280            break;
3281        }
3282        if (emit_function != NULL) {
3283          bool bounds_check = *checked_up_to < cp_offset + j || read_backward();
3284          bool bound_checked =
3285              emit_function(isolate, compiler, quarks[j], backtrack,
3286                            cp_offset + j, bounds_check, preloaded);
3287          if (bound_checked) UpdateBoundsCheck(cp_offset + j, checked_up_to);
3288        }
3289      }
3290    } else {
3291      DCHECK_EQ(TextElement::CHAR_CLASS, elm.text_type());
3292      if (pass == CHARACTER_CLASS_MATCH) {
3293        if (first_element_checked && i == 0) continue;
3294        if (DeterminedAlready(quick_check, elm.cp_offset())) continue;
3295        RegExpCharacterClass* cc = elm.char_class();
3296        bool bounds_check = *checked_up_to < cp_offset || read_backward();
3297        EmitCharClass(assembler, cc, one_byte, backtrack, cp_offset,
3298                      bounds_check, preloaded, zone());
3299        UpdateBoundsCheck(cp_offset, checked_up_to);
3300      }
3301    }
3302  }
3303}
3304
3305
3306int TextNode::Length() {
3307  TextElement elm = elements()->last();
3308  DCHECK(elm.cp_offset() >= 0);
3309  return elm.cp_offset() + elm.length();
3310}
3311
3312
3313bool TextNode::SkipPass(int int_pass, bool ignore_case) {
3314  TextEmitPassType pass = static_cast<TextEmitPassType>(int_pass);
3315  if (ignore_case) {
3316    return pass == SIMPLE_CHARACTER_MATCH;
3317  } else {
3318    return pass == NON_LETTER_CHARACTER_MATCH || pass == CASE_CHARACTER_MATCH;
3319  }
3320}
3321
3322
3323TextNode* TextNode::CreateForCharacterRanges(Zone* zone,
3324                                             ZoneList<CharacterRange>* ranges,
3325                                             bool read_backward,
3326                                             RegExpNode* on_success) {
3327  DCHECK_NOT_NULL(ranges);
3328  ZoneList<TextElement>* elms = new (zone) ZoneList<TextElement>(1, zone);
3329  elms->Add(
3330      TextElement::CharClass(new (zone) RegExpCharacterClass(ranges, false)),
3331      zone);
3332  return new (zone) TextNode(elms, read_backward, on_success);
3333}
3334
3335
3336TextNode* TextNode::CreateForSurrogatePair(Zone* zone, CharacterRange lead,
3337                                           CharacterRange trail,
3338                                           bool read_backward,
3339                                           RegExpNode* on_success) {
3340  ZoneList<CharacterRange>* lead_ranges = CharacterRange::List(zone, lead);
3341  ZoneList<CharacterRange>* trail_ranges = CharacterRange::List(zone, trail);
3342  ZoneList<TextElement>* elms = new (zone) ZoneList<TextElement>(2, zone);
3343  elms->Add(TextElement::CharClass(
3344                new (zone) RegExpCharacterClass(lead_ranges, false)),
3345            zone);
3346  elms->Add(TextElement::CharClass(
3347                new (zone) RegExpCharacterClass(trail_ranges, false)),
3348            zone);
3349  return new (zone) TextNode(elms, read_backward, on_success);
3350}
3351
3352
3353// This generates the code to match a text node.  A text node can contain
3354// straight character sequences (possibly to be matched in a case-independent
3355// way) and character classes.  For efficiency we do not do this in a single
3356// pass from left to right.  Instead we pass over the text node several times,
3357// emitting code for some character positions every time.  See the comment on
3358// TextEmitPass for details.
3359void TextNode::Emit(RegExpCompiler* compiler, Trace* trace) {
3360  LimitResult limit_result = LimitVersions(compiler, trace);
3361  if (limit_result == DONE) return;
3362  DCHECK(limit_result == CONTINUE);
3363
3364  if (trace->cp_offset() + Length() > RegExpMacroAssembler::kMaxCPOffset) {
3365    compiler->SetRegExpTooBig();
3366    return;
3367  }
3368
3369  if (compiler->one_byte()) {
3370    int dummy = 0;
3371    TextEmitPass(compiler, NON_LATIN1_MATCH, false, trace, false, &dummy);
3372  }
3373
3374  bool first_elt_done = false;
3375  int bound_checked_to = trace->cp_offset() - 1;
3376  bound_checked_to += trace->bound_checked_up_to();
3377
3378  // If a character is preloaded into the current character register then
3379  // check that now.
3380  if (trace->characters_preloaded() == 1) {
3381    for (int pass = kFirstRealPass; pass <= kLastPass; pass++) {
3382      if (!SkipPass(pass, compiler->ignore_case())) {
3383        TextEmitPass(compiler,
3384                     static_cast<TextEmitPassType>(pass),
3385                     true,
3386                     trace,
3387                     false,
3388                     &bound_checked_to);
3389      }
3390    }
3391    first_elt_done = true;
3392  }
3393
3394  for (int pass = kFirstRealPass; pass <= kLastPass; pass++) {
3395    if (!SkipPass(pass, compiler->ignore_case())) {
3396      TextEmitPass(compiler,
3397                   static_cast<TextEmitPassType>(pass),
3398                   false,
3399                   trace,
3400                   first_elt_done,
3401                   &bound_checked_to);
3402    }
3403  }
3404
3405  Trace successor_trace(*trace);
3406  // If we advance backward, we may end up at the start.
3407  successor_trace.AdvanceCurrentPositionInTrace(
3408      read_backward() ? -Length() : Length(), compiler);
3409  successor_trace.set_at_start(read_backward() ? Trace::UNKNOWN
3410                                               : Trace::FALSE_VALUE);
3411  RecursionCheck rc(compiler);
3412  on_success()->Emit(compiler, &successor_trace);
3413}
3414
3415
3416void Trace::InvalidateCurrentCharacter() {
3417  characters_preloaded_ = 0;
3418}
3419
3420
3421void Trace::AdvanceCurrentPositionInTrace(int by, RegExpCompiler* compiler) {
3422  // We don't have an instruction for shifting the current character register
3423  // down or for using a shifted value for anything so lets just forget that
3424  // we preloaded any characters into it.
3425  characters_preloaded_ = 0;
3426  // Adjust the offsets of the quick check performed information.  This
3427  // information is used to find out what we already determined about the
3428  // characters by means of mask and compare.
3429  quick_check_performed_.Advance(by, compiler->one_byte());
3430  cp_offset_ += by;
3431  if (cp_offset_ > RegExpMacroAssembler::kMaxCPOffset) {
3432    compiler->SetRegExpTooBig();
3433    cp_offset_ = 0;
3434  }
3435  bound_checked_up_to_ = Max(0, bound_checked_up_to_ - by);
3436}
3437
3438
3439void TextNode::MakeCaseIndependent(Isolate* isolate, bool is_one_byte) {
3440  int element_count = elements()->length();
3441  for (int i = 0; i < element_count; i++) {
3442    TextElement elm = elements()->at(i);
3443    if (elm.text_type() == TextElement::CHAR_CLASS) {
3444      RegExpCharacterClass* cc = elm.char_class();
3445      // None of the standard character classes is different in the case
3446      // independent case and it slows us down if we don't know that.
3447      if (cc->is_standard(zone())) continue;
3448      ZoneList<CharacterRange>* ranges = cc->ranges(zone());
3449      CharacterRange::AddCaseEquivalents(isolate, zone(), ranges, is_one_byte);
3450    }
3451  }
3452}
3453
3454
3455int TextNode::GreedyLoopTextLength() { return Length(); }
3456
3457
3458RegExpNode* TextNode::GetSuccessorOfOmnivorousTextNode(
3459    RegExpCompiler* compiler) {
3460  if (read_backward()) return NULL;
3461  if (elements()->length() != 1) return NULL;
3462  TextElement elm = elements()->at(0);
3463  if (elm.text_type() != TextElement::CHAR_CLASS) return NULL;
3464  RegExpCharacterClass* node = elm.char_class();
3465  ZoneList<CharacterRange>* ranges = node->ranges(zone());
3466  CharacterRange::Canonicalize(ranges);
3467  if (node->is_negated()) {
3468    return ranges->length() == 0 ? on_success() : NULL;
3469  }
3470  if (ranges->length() != 1) return NULL;
3471  uint32_t max_char;
3472  if (compiler->one_byte()) {
3473    max_char = String::kMaxOneByteCharCode;
3474  } else {
3475    max_char = String::kMaxUtf16CodeUnit;
3476  }
3477  return ranges->at(0).IsEverything(max_char) ? on_success() : NULL;
3478}
3479
3480
3481// Finds the fixed match length of a sequence of nodes that goes from
3482// this alternative and back to this choice node.  If there are variable
3483// length nodes or other complications in the way then return a sentinel
3484// value indicating that a greedy loop cannot be constructed.
3485int ChoiceNode::GreedyLoopTextLengthForAlternative(
3486    GuardedAlternative* alternative) {
3487  int length = 0;
3488  RegExpNode* node = alternative->node();
3489  // Later we will generate code for all these text nodes using recursion
3490  // so we have to limit the max number.
3491  int recursion_depth = 0;
3492  while (node != this) {
3493    if (recursion_depth++ > RegExpCompiler::kMaxRecursion) {
3494      return kNodeIsTooComplexForGreedyLoops;
3495    }
3496    int node_length = node->GreedyLoopTextLength();
3497    if (node_length == kNodeIsTooComplexForGreedyLoops) {
3498      return kNodeIsTooComplexForGreedyLoops;
3499    }
3500    length += node_length;
3501    SeqRegExpNode* seq_node = static_cast<SeqRegExpNode*>(node);
3502    node = seq_node->on_success();
3503  }
3504  return read_backward() ? -length : length;
3505}
3506
3507
3508void LoopChoiceNode::AddLoopAlternative(GuardedAlternative alt) {
3509  DCHECK_NULL(loop_node_);
3510  AddAlternative(alt);
3511  loop_node_ = alt.node();
3512}
3513
3514
3515void LoopChoiceNode::AddContinueAlternative(GuardedAlternative alt) {
3516  DCHECK_NULL(continue_node_);
3517  AddAlternative(alt);
3518  continue_node_ = alt.node();
3519}
3520
3521
3522void LoopChoiceNode::Emit(RegExpCompiler* compiler, Trace* trace) {
3523  RegExpMacroAssembler* macro_assembler = compiler->macro_assembler();
3524  if (trace->stop_node() == this) {
3525    // Back edge of greedy optimized loop node graph.
3526    int text_length =
3527        GreedyLoopTextLengthForAlternative(&(alternatives_->at(0)));
3528    DCHECK(text_length != kNodeIsTooComplexForGreedyLoops);
3529    // Update the counter-based backtracking info on the stack.  This is an
3530    // optimization for greedy loops (see below).
3531    DCHECK(trace->cp_offset() == text_length);
3532    macro_assembler->AdvanceCurrentPosition(text_length);
3533    macro_assembler->GoTo(trace->loop_label());
3534    return;
3535  }
3536  DCHECK_NULL(trace->stop_node());
3537  if (!trace->is_trivial()) {
3538    trace->Flush(compiler, this);
3539    return;
3540  }
3541  ChoiceNode::Emit(compiler, trace);
3542}
3543
3544
3545int ChoiceNode::CalculatePreloadCharacters(RegExpCompiler* compiler,
3546                                           int eats_at_least) {
3547  int preload_characters = Min(4, eats_at_least);
3548  if (compiler->macro_assembler()->CanReadUnaligned()) {
3549    bool one_byte = compiler->one_byte();
3550    if (one_byte) {
3551      if (preload_characters > 4) preload_characters = 4;
3552      // We can't preload 3 characters because there is no machine instruction
3553      // to do that.  We can't just load 4 because we could be reading
3554      // beyond the end of the string, which could cause a memory fault.
3555      if (preload_characters == 3) preload_characters = 2;
3556    } else {
3557      if (preload_characters > 2) preload_characters = 2;
3558    }
3559  } else {
3560    if (preload_characters > 1) preload_characters = 1;
3561  }
3562  return preload_characters;
3563}
3564
3565
3566// This class is used when generating the alternatives in a choice node.  It
3567// records the way the alternative is being code generated.
3568class AlternativeGeneration: public Malloced {
3569 public:
3570  AlternativeGeneration()
3571      : possible_success(),
3572        expects_preload(false),
3573        after(),
3574        quick_check_details() { }
3575  Label possible_success;
3576  bool expects_preload;
3577  Label after;
3578  QuickCheckDetails quick_check_details;
3579};
3580
3581
3582// Creates a list of AlternativeGenerations.  If the list has a reasonable
3583// size then it is on the stack, otherwise the excess is on the heap.
3584class AlternativeGenerationList {
3585 public:
3586  AlternativeGenerationList(int count, Zone* zone)
3587      : alt_gens_(count, zone) {
3588    for (int i = 0; i < count && i < kAFew; i++) {
3589      alt_gens_.Add(a_few_alt_gens_ + i, zone);
3590    }
3591    for (int i = kAFew; i < count; i++) {
3592      alt_gens_.Add(new AlternativeGeneration(), zone);
3593    }
3594  }
3595  ~AlternativeGenerationList() {
3596    for (int i = kAFew; i < alt_gens_.length(); i++) {
3597      delete alt_gens_[i];
3598      alt_gens_[i] = NULL;
3599    }
3600  }
3601
3602  AlternativeGeneration* at(int i) {
3603    return alt_gens_[i];
3604  }
3605
3606 private:
3607  static const int kAFew = 10;
3608  ZoneList<AlternativeGeneration*> alt_gens_;
3609  AlternativeGeneration a_few_alt_gens_[kAFew];
3610};
3611
3612
3613static const uc32 kRangeEndMarker = 0x110000;
3614
3615// The '2' variant is has inclusive from and exclusive to.
3616// This covers \s as defined in ECMA-262 5.1, 15.10.2.12,
3617// which include WhiteSpace (7.2) or LineTerminator (7.3) values.
3618static const int kSpaceRanges[] = {
3619    '\t',   '\r' + 1, ' ',    ' ' + 1, 0x00A0, 0x00A1, 0x1680,         0x1681,
3620    0x180E, 0x180F,   0x2000, 0x200B,  0x2028, 0x202A, 0x202F,         0x2030,
3621    0x205F, 0x2060,   0x3000, 0x3001,  0xFEFF, 0xFF00, kRangeEndMarker};
3622static const int kSpaceRangeCount = arraysize(kSpaceRanges);
3623
3624static const int kWordRanges[] = {
3625    '0', '9' + 1, 'A', 'Z' + 1, '_', '_' + 1, 'a', 'z' + 1, kRangeEndMarker};
3626static const int kWordRangeCount = arraysize(kWordRanges);
3627static const int kDigitRanges[] = {'0', '9' + 1, kRangeEndMarker};
3628static const int kDigitRangeCount = arraysize(kDigitRanges);
3629static const int kSurrogateRanges[] = {
3630    kLeadSurrogateStart, kLeadSurrogateStart + 1, kRangeEndMarker};
3631static const int kSurrogateRangeCount = arraysize(kSurrogateRanges);
3632static const int kLineTerminatorRanges[] = {
3633    0x000A, 0x000B, 0x000D, 0x000E, 0x2028, 0x202A, kRangeEndMarker};
3634static const int kLineTerminatorRangeCount = arraysize(kLineTerminatorRanges);
3635
3636void BoyerMoorePositionInfo::Set(int character) {
3637  SetInterval(Interval(character, character));
3638}
3639
3640
3641void BoyerMoorePositionInfo::SetInterval(const Interval& interval) {
3642  s_ = AddRange(s_, kSpaceRanges, kSpaceRangeCount, interval);
3643  w_ = AddRange(w_, kWordRanges, kWordRangeCount, interval);
3644  d_ = AddRange(d_, kDigitRanges, kDigitRangeCount, interval);
3645  surrogate_ =
3646      AddRange(surrogate_, kSurrogateRanges, kSurrogateRangeCount, interval);
3647  if (interval.to() - interval.from() >= kMapSize - 1) {
3648    if (map_count_ != kMapSize) {
3649      map_count_ = kMapSize;
3650      for (int i = 0; i < kMapSize; i++) map_->at(i) = true;
3651    }
3652    return;
3653  }
3654  for (int i = interval.from(); i <= interval.to(); i++) {
3655    int mod_character = (i & kMask);
3656    if (!map_->at(mod_character)) {
3657      map_count_++;
3658      map_->at(mod_character) = true;
3659    }
3660    if (map_count_ == kMapSize) return;
3661  }
3662}
3663
3664
3665void BoyerMoorePositionInfo::SetAll() {
3666  s_ = w_ = d_ = kLatticeUnknown;
3667  if (map_count_ != kMapSize) {
3668    map_count_ = kMapSize;
3669    for (int i = 0; i < kMapSize; i++) map_->at(i) = true;
3670  }
3671}
3672
3673
3674BoyerMooreLookahead::BoyerMooreLookahead(
3675    int length, RegExpCompiler* compiler, Zone* zone)
3676    : length_(length),
3677      compiler_(compiler) {
3678  if (compiler->one_byte()) {
3679    max_char_ = String::kMaxOneByteCharCode;
3680  } else {
3681    max_char_ = String::kMaxUtf16CodeUnit;
3682  }
3683  bitmaps_ = new(zone) ZoneList<BoyerMoorePositionInfo*>(length, zone);
3684  for (int i = 0; i < length; i++) {
3685    bitmaps_->Add(new(zone) BoyerMoorePositionInfo(zone), zone);
3686  }
3687}
3688
3689
3690// Find the longest range of lookahead that has the fewest number of different
3691// characters that can occur at a given position.  Since we are optimizing two
3692// different parameters at once this is a tradeoff.
3693bool BoyerMooreLookahead::FindWorthwhileInterval(int* from, int* to) {
3694  int biggest_points = 0;
3695  // If more than 32 characters out of 128 can occur it is unlikely that we can
3696  // be lucky enough to step forwards much of the time.
3697  const int kMaxMax = 32;
3698  for (int max_number_of_chars = 4;
3699       max_number_of_chars < kMaxMax;
3700       max_number_of_chars *= 2) {
3701    biggest_points =
3702        FindBestInterval(max_number_of_chars, biggest_points, from, to);
3703  }
3704  if (biggest_points == 0) return false;
3705  return true;
3706}
3707
3708
3709// Find the highest-points range between 0 and length_ where the character
3710// information is not too vague.  'Too vague' means that there are more than
3711// max_number_of_chars that can occur at this position.  Calculates the number
3712// of points as the product of width-of-the-range and
3713// probability-of-finding-one-of-the-characters, where the probability is
3714// calculated using the frequency distribution of the sample subject string.
3715int BoyerMooreLookahead::FindBestInterval(
3716    int max_number_of_chars, int old_biggest_points, int* from, int* to) {
3717  int biggest_points = old_biggest_points;
3718  static const int kSize = RegExpMacroAssembler::kTableSize;
3719  for (int i = 0; i < length_; ) {
3720    while (i < length_ && Count(i) > max_number_of_chars) i++;
3721    if (i == length_) break;
3722    int remembered_from = i;
3723    bool union_map[kSize];
3724    for (int j = 0; j < kSize; j++) union_map[j] = false;
3725    while (i < length_ && Count(i) <= max_number_of_chars) {
3726      BoyerMoorePositionInfo* map = bitmaps_->at(i);
3727      for (int j = 0; j < kSize; j++) union_map[j] |= map->at(j);
3728      i++;
3729    }
3730    int frequency = 0;
3731    for (int j = 0; j < kSize; j++) {
3732      if (union_map[j]) {
3733        // Add 1 to the frequency to give a small per-character boost for
3734        // the cases where our sampling is not good enough and many
3735        // characters have a frequency of zero.  This means the frequency
3736        // can theoretically be up to 2*kSize though we treat it mostly as
3737        // a fraction of kSize.
3738        frequency += compiler_->frequency_collator()->Frequency(j) + 1;
3739      }
3740    }
3741    // We use the probability of skipping times the distance we are skipping to
3742    // judge the effectiveness of this.  Actually we have a cut-off:  By
3743    // dividing by 2 we switch off the skipping if the probability of skipping
3744    // is less than 50%.  This is because the multibyte mask-and-compare
3745    // skipping in quickcheck is more likely to do well on this case.
3746    bool in_quickcheck_range =
3747        ((i - remembered_from < 4) ||
3748         (compiler_->one_byte() ? remembered_from <= 4 : remembered_from <= 2));
3749    // Called 'probability' but it is only a rough estimate and can actually
3750    // be outside the 0-kSize range.
3751    int probability = (in_quickcheck_range ? kSize / 2 : kSize) - frequency;
3752    int points = (i - remembered_from) * probability;
3753    if (points > biggest_points) {
3754      *from = remembered_from;
3755      *to = i - 1;
3756      biggest_points = points;
3757    }
3758  }
3759  return biggest_points;
3760}
3761
3762
3763// Take all the characters that will not prevent a successful match if they
3764// occur in the subject string in the range between min_lookahead and
3765// max_lookahead (inclusive) measured from the current position.  If the
3766// character at max_lookahead offset is not one of these characters, then we
3767// can safely skip forwards by the number of characters in the range.
3768int BoyerMooreLookahead::GetSkipTable(int min_lookahead,
3769                                      int max_lookahead,
3770                                      Handle<ByteArray> boolean_skip_table) {
3771  const int kSize = RegExpMacroAssembler::kTableSize;
3772
3773  const int kSkipArrayEntry = 0;
3774  const int kDontSkipArrayEntry = 1;
3775
3776  for (int i = 0; i < kSize; i++) {
3777    boolean_skip_table->set(i, kSkipArrayEntry);
3778  }
3779  int skip = max_lookahead + 1 - min_lookahead;
3780
3781  for (int i = max_lookahead; i >= min_lookahead; i--) {
3782    BoyerMoorePositionInfo* map = bitmaps_->at(i);
3783    for (int j = 0; j < kSize; j++) {
3784      if (map->at(j)) {
3785        boolean_skip_table->set(j, kDontSkipArrayEntry);
3786      }
3787    }
3788  }
3789
3790  return skip;
3791}
3792
3793
3794// See comment above on the implementation of GetSkipTable.
3795void BoyerMooreLookahead::EmitSkipInstructions(RegExpMacroAssembler* masm) {
3796  const int kSize = RegExpMacroAssembler::kTableSize;
3797
3798  int min_lookahead = 0;
3799  int max_lookahead = 0;
3800
3801  if (!FindWorthwhileInterval(&min_lookahead, &max_lookahead)) return;
3802
3803  bool found_single_character = false;
3804  int single_character = 0;
3805  for (int i = max_lookahead; i >= min_lookahead; i--) {
3806    BoyerMoorePositionInfo* map = bitmaps_->at(i);
3807    if (map->map_count() > 1 ||
3808        (found_single_character && map->map_count() != 0)) {
3809      found_single_character = false;
3810      break;
3811    }
3812    for (int j = 0; j < kSize; j++) {
3813      if (map->at(j)) {
3814        found_single_character = true;
3815        single_character = j;
3816        break;
3817      }
3818    }
3819  }
3820
3821  int lookahead_width = max_lookahead + 1 - min_lookahead;
3822
3823  if (found_single_character && lookahead_width == 1 && max_lookahead < 3) {
3824    // The mask-compare can probably handle this better.
3825    return;
3826  }
3827
3828  if (found_single_character) {
3829    Label cont, again;
3830    masm->Bind(&again);
3831    masm->LoadCurrentCharacter(max_lookahead, &cont, true);
3832    if (max_char_ > kSize) {
3833      masm->CheckCharacterAfterAnd(single_character,
3834                                   RegExpMacroAssembler::kTableMask,
3835                                   &cont);
3836    } else {
3837      masm->CheckCharacter(single_character, &cont);
3838    }
3839    masm->AdvanceCurrentPosition(lookahead_width);
3840    masm->GoTo(&again);
3841    masm->Bind(&cont);
3842    return;
3843  }
3844
3845  Factory* factory = masm->isolate()->factory();
3846  Handle<ByteArray> boolean_skip_table = factory->NewByteArray(kSize, TENURED);
3847  int skip_distance = GetSkipTable(
3848      min_lookahead, max_lookahead, boolean_skip_table);
3849  DCHECK(skip_distance != 0);
3850
3851  Label cont, again;
3852  masm->Bind(&again);
3853  masm->LoadCurrentCharacter(max_lookahead, &cont, true);
3854  masm->CheckBitInTable(boolean_skip_table, &cont);
3855  masm->AdvanceCurrentPosition(skip_distance);
3856  masm->GoTo(&again);
3857  masm->Bind(&cont);
3858}
3859
3860
3861/* Code generation for choice nodes.
3862 *
3863 * We generate quick checks that do a mask and compare to eliminate a
3864 * choice.  If the quick check succeeds then it jumps to the continuation to
3865 * do slow checks and check subsequent nodes.  If it fails (the common case)
3866 * it falls through to the next choice.
3867 *
3868 * Here is the desired flow graph.  Nodes directly below each other imply
3869 * fallthrough.  Alternatives 1 and 2 have quick checks.  Alternative
3870 * 3 doesn't have a quick check so we have to call the slow check.
3871 * Nodes are marked Qn for quick checks and Sn for slow checks.  The entire
3872 * regexp continuation is generated directly after the Sn node, up to the
3873 * next GoTo if we decide to reuse some already generated code.  Some
3874 * nodes expect preload_characters to be preloaded into the current
3875 * character register.  R nodes do this preloading.  Vertices are marked
3876 * F for failures and S for success (possible success in the case of quick
3877 * nodes).  L, V, < and > are used as arrow heads.
3878 *
3879 * ----------> R
3880 *             |
3881 *             V
3882 *            Q1 -----> S1
3883 *             |   S   /
3884 *            F|      /
3885 *             |    F/
3886 *             |    /
3887 *             |   R
3888 *             |  /
3889 *             V L
3890 *            Q2 -----> S2
3891 *             |   S   /
3892 *            F|      /
3893 *             |    F/
3894 *             |    /
3895 *             |   R
3896 *             |  /
3897 *             V L
3898 *            S3
3899 *             |
3900 *            F|
3901 *             |
3902 *             R
3903 *             |
3904 * backtrack   V
3905 * <----------Q4
3906 *   \    F    |
3907 *    \        |S
3908 *     \   F   V
3909 *      \-----S4
3910 *
3911 * For greedy loops we push the current position, then generate the code that
3912 * eats the input specially in EmitGreedyLoop.  The other choice (the
3913 * continuation) is generated by the normal code in EmitChoices, and steps back
3914 * in the input to the starting position when it fails to match.  The loop code
3915 * looks like this (U is the unwind code that steps back in the greedy loop).
3916 *
3917 *              _____
3918 *             /     \
3919 *             V     |
3920 * ----------> S1    |
3921 *            /|     |
3922 *           / |S    |
3923 *         F/  \_____/
3924 *         /
3925 *        |<-----
3926 *        |      \
3927 *        V       |S
3928 *        Q2 ---> U----->backtrack
3929 *        |  F   /
3930 *       S|     /
3931 *        V  F /
3932 *        S2--/
3933 */
3934
3935GreedyLoopState::GreedyLoopState(bool not_at_start) {
3936  counter_backtrack_trace_.set_backtrack(&label_);
3937  if (not_at_start) counter_backtrack_trace_.set_at_start(Trace::FALSE_VALUE);
3938}
3939
3940
3941void ChoiceNode::AssertGuardsMentionRegisters(Trace* trace) {
3942#ifdef DEBUG
3943  int choice_count = alternatives_->length();
3944  for (int i = 0; i < choice_count - 1; i++) {
3945    GuardedAlternative alternative = alternatives_->at(i);
3946    ZoneList<Guard*>* guards = alternative.guards();
3947    int guard_count = (guards == NULL) ? 0 : guards->length();
3948    for (int j = 0; j < guard_count; j++) {
3949      DCHECK(!trace->mentions_reg(guards->at(j)->reg()));
3950    }
3951  }
3952#endif
3953}
3954
3955
3956void ChoiceNode::SetUpPreLoad(RegExpCompiler* compiler,
3957                              Trace* current_trace,
3958                              PreloadState* state) {
3959    if (state->eats_at_least_ == PreloadState::kEatsAtLeastNotYetInitialized) {
3960      // Save some time by looking at most one machine word ahead.
3961      state->eats_at_least_ =
3962          EatsAtLeast(compiler->one_byte() ? 4 : 2, kRecursionBudget,
3963                      current_trace->at_start() == Trace::FALSE_VALUE);
3964    }
3965    state->preload_characters_ =
3966        CalculatePreloadCharacters(compiler, state->eats_at_least_);
3967
3968    state->preload_is_current_ =
3969        (current_trace->characters_preloaded() == state->preload_characters_);
3970    state->preload_has_checked_bounds_ = state->preload_is_current_;
3971}
3972
3973
3974void ChoiceNode::Emit(RegExpCompiler* compiler, Trace* trace) {
3975  int choice_count = alternatives_->length();
3976
3977  if (choice_count == 1 && alternatives_->at(0).guards() == NULL) {
3978    alternatives_->at(0).node()->Emit(compiler, trace);
3979    return;
3980  }
3981
3982  AssertGuardsMentionRegisters(trace);
3983
3984  LimitResult limit_result = LimitVersions(compiler, trace);
3985  if (limit_result == DONE) return;
3986  DCHECK(limit_result == CONTINUE);
3987
3988  // For loop nodes we already flushed (see LoopChoiceNode::Emit), but for
3989  // other choice nodes we only flush if we are out of code size budget.
3990  if (trace->flush_budget() == 0 && trace->actions() != NULL) {
3991    trace->Flush(compiler, this);
3992    return;
3993  }
3994
3995  RecursionCheck rc(compiler);
3996
3997  PreloadState preload;
3998  preload.init();
3999  GreedyLoopState greedy_loop_state(not_at_start());
4000
4001  int text_length = GreedyLoopTextLengthForAlternative(&alternatives_->at(0));
4002  AlternativeGenerationList alt_gens(choice_count, zone());
4003
4004  if (choice_count > 1 && text_length != kNodeIsTooComplexForGreedyLoops) {
4005    trace = EmitGreedyLoop(compiler,
4006                           trace,
4007                           &alt_gens,
4008                           &preload,
4009                           &greedy_loop_state,
4010                           text_length);
4011  } else {
4012    // TODO(erikcorry): Delete this.  We don't need this label, but it makes us
4013    // match the traces produced pre-cleanup.
4014    Label second_choice;
4015    compiler->macro_assembler()->Bind(&second_choice);
4016
4017    preload.eats_at_least_ = EmitOptimizedUnanchoredSearch(compiler, trace);
4018
4019    EmitChoices(compiler,
4020                &alt_gens,
4021                0,
4022                trace,
4023                &preload);
4024  }
4025
4026  // At this point we need to generate slow checks for the alternatives where
4027  // the quick check was inlined.  We can recognize these because the associated
4028  // label was bound.
4029  int new_flush_budget = trace->flush_budget() / choice_count;
4030  for (int i = 0; i < choice_count; i++) {
4031    AlternativeGeneration* alt_gen = alt_gens.at(i);
4032    Trace new_trace(*trace);
4033    // If there are actions to be flushed we have to limit how many times
4034    // they are flushed.  Take the budget of the parent trace and distribute
4035    // it fairly amongst the children.
4036    if (new_trace.actions() != NULL) {
4037      new_trace.set_flush_budget(new_flush_budget);
4038    }
4039    bool next_expects_preload =
4040        i == choice_count - 1 ? false : alt_gens.at(i + 1)->expects_preload;
4041    EmitOutOfLineContinuation(compiler,
4042                              &new_trace,
4043                              alternatives_->at(i),
4044                              alt_gen,
4045                              preload.preload_characters_,
4046                              next_expects_preload);
4047  }
4048}
4049
4050
4051Trace* ChoiceNode::EmitGreedyLoop(RegExpCompiler* compiler,
4052                                  Trace* trace,
4053                                  AlternativeGenerationList* alt_gens,
4054                                  PreloadState* preload,
4055                                  GreedyLoopState* greedy_loop_state,
4056                                  int text_length) {
4057  RegExpMacroAssembler* macro_assembler = compiler->macro_assembler();
4058  // Here we have special handling for greedy loops containing only text nodes
4059  // and other simple nodes.  These are handled by pushing the current
4060  // position on the stack and then incrementing the current position each
4061  // time around the switch.  On backtrack we decrement the current position
4062  // and check it against the pushed value.  This avoids pushing backtrack
4063  // information for each iteration of the loop, which could take up a lot of
4064  // space.
4065  DCHECK(trace->stop_node() == NULL);
4066  macro_assembler->PushCurrentPosition();
4067  Label greedy_match_failed;
4068  Trace greedy_match_trace;
4069  if (not_at_start()) greedy_match_trace.set_at_start(Trace::FALSE_VALUE);
4070  greedy_match_trace.set_backtrack(&greedy_match_failed);
4071  Label loop_label;
4072  macro_assembler->Bind(&loop_label);
4073  greedy_match_trace.set_stop_node(this);
4074  greedy_match_trace.set_loop_label(&loop_label);
4075  alternatives_->at(0).node()->Emit(compiler, &greedy_match_trace);
4076  macro_assembler->Bind(&greedy_match_failed);
4077
4078  Label second_choice;  // For use in greedy matches.
4079  macro_assembler->Bind(&second_choice);
4080
4081  Trace* new_trace = greedy_loop_state->counter_backtrack_trace();
4082
4083  EmitChoices(compiler,
4084              alt_gens,
4085              1,
4086              new_trace,
4087              preload);
4088
4089  macro_assembler->Bind(greedy_loop_state->label());
4090  // If we have unwound to the bottom then backtrack.
4091  macro_assembler->CheckGreedyLoop(trace->backtrack());
4092  // Otherwise try the second priority at an earlier position.
4093  macro_assembler->AdvanceCurrentPosition(-text_length);
4094  macro_assembler->GoTo(&second_choice);
4095  return new_trace;
4096}
4097
4098int ChoiceNode::EmitOptimizedUnanchoredSearch(RegExpCompiler* compiler,
4099                                              Trace* trace) {
4100  int eats_at_least = PreloadState::kEatsAtLeastNotYetInitialized;
4101  if (alternatives_->length() != 2) return eats_at_least;
4102
4103  GuardedAlternative alt1 = alternatives_->at(1);
4104  if (alt1.guards() != NULL && alt1.guards()->length() != 0) {
4105    return eats_at_least;
4106  }
4107  RegExpNode* eats_anything_node = alt1.node();
4108  if (eats_anything_node->GetSuccessorOfOmnivorousTextNode(compiler) != this) {
4109    return eats_at_least;
4110  }
4111
4112  // Really we should be creating a new trace when we execute this function,
4113  // but there is no need, because the code it generates cannot backtrack, and
4114  // we always arrive here with a trivial trace (since it's the entry to a
4115  // loop.  That also implies that there are no preloaded characters, which is
4116  // good, because it means we won't be violating any assumptions by
4117  // overwriting those characters with new load instructions.
4118  DCHECK(trace->is_trivial());
4119
4120  RegExpMacroAssembler* macro_assembler = compiler->macro_assembler();
4121  Isolate* isolate = macro_assembler->isolate();
4122  // At this point we know that we are at a non-greedy loop that will eat
4123  // any character one at a time.  Any non-anchored regexp has such a
4124  // loop prepended to it in order to find where it starts.  We look for
4125  // a pattern of the form ...abc... where we can look 6 characters ahead
4126  // and step forwards 3 if the character is not one of abc.  Abc need
4127  // not be atoms, they can be any reasonably limited character class or
4128  // small alternation.
4129  BoyerMooreLookahead* bm = bm_info(false);
4130  if (bm == NULL) {
4131    eats_at_least = Min(kMaxLookaheadForBoyerMoore,
4132                        EatsAtLeast(kMaxLookaheadForBoyerMoore,
4133                                    kRecursionBudget,
4134                                    false));
4135    if (eats_at_least >= 1) {
4136      bm = new(zone()) BoyerMooreLookahead(eats_at_least,
4137                                           compiler,
4138                                           zone());
4139      GuardedAlternative alt0 = alternatives_->at(0);
4140      alt0.node()->FillInBMInfo(isolate, 0, kRecursionBudget, bm, false);
4141    }
4142  }
4143  if (bm != NULL) {
4144    bm->EmitSkipInstructions(macro_assembler);
4145  }
4146  return eats_at_least;
4147}
4148
4149
4150void ChoiceNode::EmitChoices(RegExpCompiler* compiler,
4151                             AlternativeGenerationList* alt_gens,
4152                             int first_choice,
4153                             Trace* trace,
4154                             PreloadState* preload) {
4155  RegExpMacroAssembler* macro_assembler = compiler->macro_assembler();
4156  SetUpPreLoad(compiler, trace, preload);
4157
4158  // For now we just call all choices one after the other.  The idea ultimately
4159  // is to use the Dispatch table to try only the relevant ones.
4160  int choice_count = alternatives_->length();
4161
4162  int new_flush_budget = trace->flush_budget() / choice_count;
4163
4164  for (int i = first_choice; i < choice_count; i++) {
4165    bool is_last = i == choice_count - 1;
4166    bool fall_through_on_failure = !is_last;
4167    GuardedAlternative alternative = alternatives_->at(i);
4168    AlternativeGeneration* alt_gen = alt_gens->at(i);
4169    alt_gen->quick_check_details.set_characters(preload->preload_characters_);
4170    ZoneList<Guard*>* guards = alternative.guards();
4171    int guard_count = (guards == NULL) ? 0 : guards->length();
4172    Trace new_trace(*trace);
4173    new_trace.set_characters_preloaded(preload->preload_is_current_ ?
4174                                         preload->preload_characters_ :
4175                                         0);
4176    if (preload->preload_has_checked_bounds_) {
4177      new_trace.set_bound_checked_up_to(preload->preload_characters_);
4178    }
4179    new_trace.quick_check_performed()->Clear();
4180    if (not_at_start_) new_trace.set_at_start(Trace::FALSE_VALUE);
4181    if (!is_last) {
4182      new_trace.set_backtrack(&alt_gen->after);
4183    }
4184    alt_gen->expects_preload = preload->preload_is_current_;
4185    bool generate_full_check_inline = false;
4186    if (compiler->optimize() &&
4187        try_to_emit_quick_check_for_alternative(i == 0) &&
4188        alternative.node()->EmitQuickCheck(
4189            compiler, trace, &new_trace, preload->preload_has_checked_bounds_,
4190            &alt_gen->possible_success, &alt_gen->quick_check_details,
4191            fall_through_on_failure)) {
4192      // Quick check was generated for this choice.
4193      preload->preload_is_current_ = true;
4194      preload->preload_has_checked_bounds_ = true;
4195      // If we generated the quick check to fall through on possible success,
4196      // we now need to generate the full check inline.
4197      if (!fall_through_on_failure) {
4198        macro_assembler->Bind(&alt_gen->possible_success);
4199        new_trace.set_quick_check_performed(&alt_gen->quick_check_details);
4200        new_trace.set_characters_preloaded(preload->preload_characters_);
4201        new_trace.set_bound_checked_up_to(preload->preload_characters_);
4202        generate_full_check_inline = true;
4203      }
4204    } else if (alt_gen->quick_check_details.cannot_match()) {
4205      if (!fall_through_on_failure) {
4206        macro_assembler->GoTo(trace->backtrack());
4207      }
4208      continue;
4209    } else {
4210      // No quick check was generated.  Put the full code here.
4211      // If this is not the first choice then there could be slow checks from
4212      // previous cases that go here when they fail.  There's no reason to
4213      // insist that they preload characters since the slow check we are about
4214      // to generate probably can't use it.
4215      if (i != first_choice) {
4216        alt_gen->expects_preload = false;
4217        new_trace.InvalidateCurrentCharacter();
4218      }
4219      generate_full_check_inline = true;
4220    }
4221    if (generate_full_check_inline) {
4222      if (new_trace.actions() != NULL) {
4223        new_trace.set_flush_budget(new_flush_budget);
4224      }
4225      for (int j = 0; j < guard_count; j++) {
4226        GenerateGuard(macro_assembler, guards->at(j), &new_trace);
4227      }
4228      alternative.node()->Emit(compiler, &new_trace);
4229      preload->preload_is_current_ = false;
4230    }
4231    macro_assembler->Bind(&alt_gen->after);
4232  }
4233}
4234
4235
4236void ChoiceNode::EmitOutOfLineContinuation(RegExpCompiler* compiler,
4237                                           Trace* trace,
4238                                           GuardedAlternative alternative,
4239                                           AlternativeGeneration* alt_gen,
4240                                           int preload_characters,
4241                                           bool next_expects_preload) {
4242  if (!alt_gen->possible_success.is_linked()) return;
4243
4244  RegExpMacroAssembler* macro_assembler = compiler->macro_assembler();
4245  macro_assembler->Bind(&alt_gen->possible_success);
4246  Trace out_of_line_trace(*trace);
4247  out_of_line_trace.set_characters_preloaded(preload_characters);
4248  out_of_line_trace.set_quick_check_performed(&alt_gen->quick_check_details);
4249  if (not_at_start_) out_of_line_trace.set_at_start(Trace::FALSE_VALUE);
4250  ZoneList<Guard*>* guards = alternative.guards();
4251  int guard_count = (guards == NULL) ? 0 : guards->length();
4252  if (next_expects_preload) {
4253    Label reload_current_char;
4254    out_of_line_trace.set_backtrack(&reload_current_char);
4255    for (int j = 0; j < guard_count; j++) {
4256      GenerateGuard(macro_assembler, guards->at(j), &out_of_line_trace);
4257    }
4258    alternative.node()->Emit(compiler, &out_of_line_trace);
4259    macro_assembler->Bind(&reload_current_char);
4260    // Reload the current character, since the next quick check expects that.
4261    // We don't need to check bounds here because we only get into this
4262    // code through a quick check which already did the checked load.
4263    macro_assembler->LoadCurrentCharacter(trace->cp_offset(),
4264                                          NULL,
4265                                          false,
4266                                          preload_characters);
4267    macro_assembler->GoTo(&(alt_gen->after));
4268  } else {
4269    out_of_line_trace.set_backtrack(&(alt_gen->after));
4270    for (int j = 0; j < guard_count; j++) {
4271      GenerateGuard(macro_assembler, guards->at(j), &out_of_line_trace);
4272    }
4273    alternative.node()->Emit(compiler, &out_of_line_trace);
4274  }
4275}
4276
4277
4278void ActionNode::Emit(RegExpCompiler* compiler, Trace* trace) {
4279  RegExpMacroAssembler* assembler = compiler->macro_assembler();
4280  LimitResult limit_result = LimitVersions(compiler, trace);
4281  if (limit_result == DONE) return;
4282  DCHECK(limit_result == CONTINUE);
4283
4284  RecursionCheck rc(compiler);
4285
4286  switch (action_type_) {
4287    case STORE_POSITION: {
4288      Trace::DeferredCapture
4289          new_capture(data_.u_position_register.reg,
4290                      data_.u_position_register.is_capture,
4291                      trace);
4292      Trace new_trace = *trace;
4293      new_trace.add_action(&new_capture);
4294      on_success()->Emit(compiler, &new_trace);
4295      break;
4296    }
4297    case INCREMENT_REGISTER: {
4298      Trace::DeferredIncrementRegister
4299          new_increment(data_.u_increment_register.reg);
4300      Trace new_trace = *trace;
4301      new_trace.add_action(&new_increment);
4302      on_success()->Emit(compiler, &new_trace);
4303      break;
4304    }
4305    case SET_REGISTER: {
4306      Trace::DeferredSetRegister
4307          new_set(data_.u_store_register.reg, data_.u_store_register.value);
4308      Trace new_trace = *trace;
4309      new_trace.add_action(&new_set);
4310      on_success()->Emit(compiler, &new_trace);
4311      break;
4312    }
4313    case CLEAR_CAPTURES: {
4314      Trace::DeferredClearCaptures
4315        new_capture(Interval(data_.u_clear_captures.range_from,
4316                             data_.u_clear_captures.range_to));
4317      Trace new_trace = *trace;
4318      new_trace.add_action(&new_capture);
4319      on_success()->Emit(compiler, &new_trace);
4320      break;
4321    }
4322    case BEGIN_SUBMATCH:
4323      if (!trace->is_trivial()) {
4324        trace->Flush(compiler, this);
4325      } else {
4326        assembler->WriteCurrentPositionToRegister(
4327            data_.u_submatch.current_position_register, 0);
4328        assembler->WriteStackPointerToRegister(
4329            data_.u_submatch.stack_pointer_register);
4330        on_success()->Emit(compiler, trace);
4331      }
4332      break;
4333    case EMPTY_MATCH_CHECK: {
4334      int start_pos_reg = data_.u_empty_match_check.start_register;
4335      int stored_pos = 0;
4336      int rep_reg = data_.u_empty_match_check.repetition_register;
4337      bool has_minimum = (rep_reg != RegExpCompiler::kNoRegister);
4338      bool know_dist = trace->GetStoredPosition(start_pos_reg, &stored_pos);
4339      if (know_dist && !has_minimum && stored_pos == trace->cp_offset()) {
4340        // If we know we haven't advanced and there is no minimum we
4341        // can just backtrack immediately.
4342        assembler->GoTo(trace->backtrack());
4343      } else if (know_dist && stored_pos < trace->cp_offset()) {
4344        // If we know we've advanced we can generate the continuation
4345        // immediately.
4346        on_success()->Emit(compiler, trace);
4347      } else if (!trace->is_trivial()) {
4348        trace->Flush(compiler, this);
4349      } else {
4350        Label skip_empty_check;
4351        // If we have a minimum number of repetitions we check the current
4352        // number first and skip the empty check if it's not enough.
4353        if (has_minimum) {
4354          int limit = data_.u_empty_match_check.repetition_limit;
4355          assembler->IfRegisterLT(rep_reg, limit, &skip_empty_check);
4356        }
4357        // If the match is empty we bail out, otherwise we fall through
4358        // to the on-success continuation.
4359        assembler->IfRegisterEqPos(data_.u_empty_match_check.start_register,
4360                                   trace->backtrack());
4361        assembler->Bind(&skip_empty_check);
4362        on_success()->Emit(compiler, trace);
4363      }
4364      break;
4365    }
4366    case POSITIVE_SUBMATCH_SUCCESS: {
4367      if (!trace->is_trivial()) {
4368        trace->Flush(compiler, this);
4369        return;
4370      }
4371      assembler->ReadCurrentPositionFromRegister(
4372          data_.u_submatch.current_position_register);
4373      assembler->ReadStackPointerFromRegister(
4374          data_.u_submatch.stack_pointer_register);
4375      int clear_register_count = data_.u_submatch.clear_register_count;
4376      if (clear_register_count == 0) {
4377        on_success()->Emit(compiler, trace);
4378        return;
4379      }
4380      int clear_registers_from = data_.u_submatch.clear_register_from;
4381      Label clear_registers_backtrack;
4382      Trace new_trace = *trace;
4383      new_trace.set_backtrack(&clear_registers_backtrack);
4384      on_success()->Emit(compiler, &new_trace);
4385
4386      assembler->Bind(&clear_registers_backtrack);
4387      int clear_registers_to = clear_registers_from + clear_register_count - 1;
4388      assembler->ClearRegisters(clear_registers_from, clear_registers_to);
4389
4390      DCHECK(trace->backtrack() == NULL);
4391      assembler->Backtrack();
4392      return;
4393    }
4394    default:
4395      UNREACHABLE();
4396  }
4397}
4398
4399
4400void BackReferenceNode::Emit(RegExpCompiler* compiler, Trace* trace) {
4401  RegExpMacroAssembler* assembler = compiler->macro_assembler();
4402  if (!trace->is_trivial()) {
4403    trace->Flush(compiler, this);
4404    return;
4405  }
4406
4407  LimitResult limit_result = LimitVersions(compiler, trace);
4408  if (limit_result == DONE) return;
4409  DCHECK(limit_result == CONTINUE);
4410
4411  RecursionCheck rc(compiler);
4412
4413  DCHECK_EQ(start_reg_ + 1, end_reg_);
4414  if (compiler->ignore_case()) {
4415    assembler->CheckNotBackReferenceIgnoreCase(
4416        start_reg_, read_backward(), compiler->unicode(), trace->backtrack());
4417  } else {
4418    assembler->CheckNotBackReference(start_reg_, read_backward(),
4419                                     trace->backtrack());
4420  }
4421  // We are going to advance backward, so we may end up at the start.
4422  if (read_backward()) trace->set_at_start(Trace::UNKNOWN);
4423
4424  // Check that the back reference does not end inside a surrogate pair.
4425  if (compiler->unicode() && !compiler->one_byte()) {
4426    assembler->CheckNotInSurrogatePair(trace->cp_offset(), trace->backtrack());
4427  }
4428  on_success()->Emit(compiler, trace);
4429}
4430
4431
4432// -------------------------------------------------------------------
4433// Dot/dotty output
4434
4435
4436#ifdef DEBUG
4437
4438
4439class DotPrinter: public NodeVisitor {
4440 public:
4441  DotPrinter(std::ostream& os, bool ignore_case)  // NOLINT
4442      : os_(os),
4443        ignore_case_(ignore_case) {}
4444  void PrintNode(const char* label, RegExpNode* node);
4445  void Visit(RegExpNode* node);
4446  void PrintAttributes(RegExpNode* from);
4447  void PrintOnFailure(RegExpNode* from, RegExpNode* to);
4448#define DECLARE_VISIT(Type)                                          \
4449  virtual void Visit##Type(Type##Node* that);
4450FOR_EACH_NODE_TYPE(DECLARE_VISIT)
4451#undef DECLARE_VISIT
4452 private:
4453  std::ostream& os_;
4454  bool ignore_case_;
4455};
4456
4457
4458void DotPrinter::PrintNode(const char* label, RegExpNode* node) {
4459  os_ << "digraph G {\n  graph [label=\"";
4460  for (int i = 0; label[i]; i++) {
4461    switch (label[i]) {
4462      case '\\':
4463        os_ << "\\\\";
4464        break;
4465      case '"':
4466        os_ << "\"";
4467        break;
4468      default:
4469        os_ << label[i];
4470        break;
4471    }
4472  }
4473  os_ << "\"];\n";
4474  Visit(node);
4475  os_ << "}" << std::endl;
4476}
4477
4478
4479void DotPrinter::Visit(RegExpNode* node) {
4480  if (node->info()->visited) return;
4481  node->info()->visited = true;
4482  node->Accept(this);
4483}
4484
4485
4486void DotPrinter::PrintOnFailure(RegExpNode* from, RegExpNode* on_failure) {
4487  os_ << "  n" << from << " -> n" << on_failure << " [style=dotted];\n";
4488  Visit(on_failure);
4489}
4490
4491
4492class TableEntryBodyPrinter {
4493 public:
4494  TableEntryBodyPrinter(std::ostream& os, ChoiceNode* choice)  // NOLINT
4495      : os_(os),
4496        choice_(choice) {}
4497  void Call(uc16 from, DispatchTable::Entry entry) {
4498    OutSet* out_set = entry.out_set();
4499    for (unsigned i = 0; i < OutSet::kFirstLimit; i++) {
4500      if (out_set->Get(i)) {
4501        os_ << "    n" << choice() << ":s" << from << "o" << i << " -> n"
4502            << choice()->alternatives()->at(i).node() << ";\n";
4503      }
4504    }
4505  }
4506 private:
4507  ChoiceNode* choice() { return choice_; }
4508  std::ostream& os_;
4509  ChoiceNode* choice_;
4510};
4511
4512
4513class TableEntryHeaderPrinter {
4514 public:
4515  explicit TableEntryHeaderPrinter(std::ostream& os)  // NOLINT
4516      : first_(true),
4517        os_(os) {}
4518  void Call(uc16 from, DispatchTable::Entry entry) {
4519    if (first_) {
4520      first_ = false;
4521    } else {
4522      os_ << "|";
4523    }
4524    os_ << "{\\" << AsUC16(from) << "-\\" << AsUC16(entry.to()) << "|{";
4525    OutSet* out_set = entry.out_set();
4526    int priority = 0;
4527    for (unsigned i = 0; i < OutSet::kFirstLimit; i++) {
4528      if (out_set->Get(i)) {
4529        if (priority > 0) os_ << "|";
4530        os_ << "<s" << from << "o" << i << "> " << priority;
4531        priority++;
4532      }
4533    }
4534    os_ << "}}";
4535  }
4536
4537 private:
4538  bool first_;
4539  std::ostream& os_;
4540};
4541
4542
4543class AttributePrinter {
4544 public:
4545  explicit AttributePrinter(std::ostream& os)  // NOLINT
4546      : os_(os),
4547        first_(true) {}
4548  void PrintSeparator() {
4549    if (first_) {
4550      first_ = false;
4551    } else {
4552      os_ << "|";
4553    }
4554  }
4555  void PrintBit(const char* name, bool value) {
4556    if (!value) return;
4557    PrintSeparator();
4558    os_ << "{" << name << "}";
4559  }
4560  void PrintPositive(const char* name, int value) {
4561    if (value < 0) return;
4562    PrintSeparator();
4563    os_ << "{" << name << "|" << value << "}";
4564  }
4565
4566 private:
4567  std::ostream& os_;
4568  bool first_;
4569};
4570
4571
4572void DotPrinter::PrintAttributes(RegExpNode* that) {
4573  os_ << "  a" << that << " [shape=Mrecord, color=grey, fontcolor=grey, "
4574      << "margin=0.1, fontsize=10, label=\"{";
4575  AttributePrinter printer(os_);
4576  NodeInfo* info = that->info();
4577  printer.PrintBit("NI", info->follows_newline_interest);
4578  printer.PrintBit("WI", info->follows_word_interest);
4579  printer.PrintBit("SI", info->follows_start_interest);
4580  Label* label = that->label();
4581  if (label->is_bound())
4582    printer.PrintPositive("@", label->pos());
4583  os_ << "}\"];\n"
4584      << "  a" << that << " -> n" << that
4585      << " [style=dashed, color=grey, arrowhead=none];\n";
4586}
4587
4588
4589static const bool kPrintDispatchTable = false;
4590void DotPrinter::VisitChoice(ChoiceNode* that) {
4591  if (kPrintDispatchTable) {
4592    os_ << "  n" << that << " [shape=Mrecord, label=\"";
4593    TableEntryHeaderPrinter header_printer(os_);
4594    that->GetTable(ignore_case_)->ForEach(&header_printer);
4595    os_ << "\"]\n";
4596    PrintAttributes(that);
4597    TableEntryBodyPrinter body_printer(os_, that);
4598    that->GetTable(ignore_case_)->ForEach(&body_printer);
4599  } else {
4600    os_ << "  n" << that << " [shape=Mrecord, label=\"?\"];\n";
4601    for (int i = 0; i < that->alternatives()->length(); i++) {
4602      GuardedAlternative alt = that->alternatives()->at(i);
4603      os_ << "  n" << that << " -> n" << alt.node();
4604    }
4605  }
4606  for (int i = 0; i < that->alternatives()->length(); i++) {
4607    GuardedAlternative alt = that->alternatives()->at(i);
4608    alt.node()->Accept(this);
4609  }
4610}
4611
4612
4613void DotPrinter::VisitText(TextNode* that) {
4614  Zone* zone = that->zone();
4615  os_ << "  n" << that << " [label=\"";
4616  for (int i = 0; i < that->elements()->length(); i++) {
4617    if (i > 0) os_ << " ";
4618    TextElement elm = that->elements()->at(i);
4619    switch (elm.text_type()) {
4620      case TextElement::ATOM: {
4621        Vector<const uc16> data = elm.atom()->data();
4622        for (int i = 0; i < data.length(); i++) {
4623          os_ << static_cast<char>(data[i]);
4624        }
4625        break;
4626      }
4627      case TextElement::CHAR_CLASS: {
4628        RegExpCharacterClass* node = elm.char_class();
4629        os_ << "[";
4630        if (node->is_negated()) os_ << "^";
4631        for (int j = 0; j < node->ranges(zone)->length(); j++) {
4632          CharacterRange range = node->ranges(zone)->at(j);
4633          os_ << AsUC16(range.from()) << "-" << AsUC16(range.to());
4634        }
4635        os_ << "]";
4636        break;
4637      }
4638      default:
4639        UNREACHABLE();
4640    }
4641  }
4642  os_ << "\", shape=box, peripheries=2];\n";
4643  PrintAttributes(that);
4644  os_ << "  n" << that << " -> n" << that->on_success() << ";\n";
4645  Visit(that->on_success());
4646}
4647
4648
4649void DotPrinter::VisitBackReference(BackReferenceNode* that) {
4650  os_ << "  n" << that << " [label=\"$" << that->start_register() << "..$"
4651      << that->end_register() << "\", shape=doubleoctagon];\n";
4652  PrintAttributes(that);
4653  os_ << "  n" << that << " -> n" << that->on_success() << ";\n";
4654  Visit(that->on_success());
4655}
4656
4657
4658void DotPrinter::VisitEnd(EndNode* that) {
4659  os_ << "  n" << that << " [style=bold, shape=point];\n";
4660  PrintAttributes(that);
4661}
4662
4663
4664void DotPrinter::VisitAssertion(AssertionNode* that) {
4665  os_ << "  n" << that << " [";
4666  switch (that->assertion_type()) {
4667    case AssertionNode::AT_END:
4668      os_ << "label=\"$\", shape=septagon";
4669      break;
4670    case AssertionNode::AT_START:
4671      os_ << "label=\"^\", shape=septagon";
4672      break;
4673    case AssertionNode::AT_BOUNDARY:
4674      os_ << "label=\"\\b\", shape=septagon";
4675      break;
4676    case AssertionNode::AT_NON_BOUNDARY:
4677      os_ << "label=\"\\B\", shape=septagon";
4678      break;
4679    case AssertionNode::AFTER_NEWLINE:
4680      os_ << "label=\"(?<=\\n)\", shape=septagon";
4681      break;
4682  }
4683  os_ << "];\n";
4684  PrintAttributes(that);
4685  RegExpNode* successor = that->on_success();
4686  os_ << "  n" << that << " -> n" << successor << ";\n";
4687  Visit(successor);
4688}
4689
4690
4691void DotPrinter::VisitAction(ActionNode* that) {
4692  os_ << "  n" << that << " [";
4693  switch (that->action_type_) {
4694    case ActionNode::SET_REGISTER:
4695      os_ << "label=\"$" << that->data_.u_store_register.reg
4696          << ":=" << that->data_.u_store_register.value << "\", shape=octagon";
4697      break;
4698    case ActionNode::INCREMENT_REGISTER:
4699      os_ << "label=\"$" << that->data_.u_increment_register.reg
4700          << "++\", shape=octagon";
4701      break;
4702    case ActionNode::STORE_POSITION:
4703      os_ << "label=\"$" << that->data_.u_position_register.reg
4704          << ":=$pos\", shape=octagon";
4705      break;
4706    case ActionNode::BEGIN_SUBMATCH:
4707      os_ << "label=\"$" << that->data_.u_submatch.current_position_register
4708          << ":=$pos,begin\", shape=septagon";
4709      break;
4710    case ActionNode::POSITIVE_SUBMATCH_SUCCESS:
4711      os_ << "label=\"escape\", shape=septagon";
4712      break;
4713    case ActionNode::EMPTY_MATCH_CHECK:
4714      os_ << "label=\"$" << that->data_.u_empty_match_check.start_register
4715          << "=$pos?,$" << that->data_.u_empty_match_check.repetition_register
4716          << "<" << that->data_.u_empty_match_check.repetition_limit
4717          << "?\", shape=septagon";
4718      break;
4719    case ActionNode::CLEAR_CAPTURES: {
4720      os_ << "label=\"clear $" << that->data_.u_clear_captures.range_from
4721          << " to $" << that->data_.u_clear_captures.range_to
4722          << "\", shape=septagon";
4723      break;
4724    }
4725  }
4726  os_ << "];\n";
4727  PrintAttributes(that);
4728  RegExpNode* successor = that->on_success();
4729  os_ << "  n" << that << " -> n" << successor << ";\n";
4730  Visit(successor);
4731}
4732
4733
4734class DispatchTableDumper {
4735 public:
4736  explicit DispatchTableDumper(std::ostream& os) : os_(os) {}
4737  void Call(uc16 key, DispatchTable::Entry entry);
4738 private:
4739  std::ostream& os_;
4740};
4741
4742
4743void DispatchTableDumper::Call(uc16 key, DispatchTable::Entry entry) {
4744  os_ << "[" << AsUC16(key) << "-" << AsUC16(entry.to()) << "]: {";
4745  OutSet* set = entry.out_set();
4746  bool first = true;
4747  for (unsigned i = 0; i < OutSet::kFirstLimit; i++) {
4748    if (set->Get(i)) {
4749      if (first) {
4750        first = false;
4751      } else {
4752        os_ << ", ";
4753      }
4754      os_ << i;
4755    }
4756  }
4757  os_ << "}\n";
4758}
4759
4760
4761void DispatchTable::Dump() {
4762  OFStream os(stderr);
4763  DispatchTableDumper dumper(os);
4764  tree()->ForEach(&dumper);
4765}
4766
4767
4768void RegExpEngine::DotPrint(const char* label,
4769                            RegExpNode* node,
4770                            bool ignore_case) {
4771  OFStream os(stdout);
4772  DotPrinter printer(os, ignore_case);
4773  printer.PrintNode(label, node);
4774}
4775
4776
4777#endif  // DEBUG
4778
4779
4780// -------------------------------------------------------------------
4781// Tree to graph conversion
4782
4783RegExpNode* RegExpAtom::ToNode(RegExpCompiler* compiler,
4784                               RegExpNode* on_success) {
4785  ZoneList<TextElement>* elms =
4786      new(compiler->zone()) ZoneList<TextElement>(1, compiler->zone());
4787  elms->Add(TextElement::Atom(this), compiler->zone());
4788  return new (compiler->zone())
4789      TextNode(elms, compiler->read_backward(), on_success);
4790}
4791
4792
4793RegExpNode* RegExpText::ToNode(RegExpCompiler* compiler,
4794                               RegExpNode* on_success) {
4795  return new (compiler->zone())
4796      TextNode(elements(), compiler->read_backward(), on_success);
4797}
4798
4799
4800static bool CompareInverseRanges(ZoneList<CharacterRange>* ranges,
4801                                 const int* special_class,
4802                                 int length) {
4803  length--;  // Remove final marker.
4804  DCHECK(special_class[length] == kRangeEndMarker);
4805  DCHECK(ranges->length() != 0);
4806  DCHECK(length != 0);
4807  DCHECK(special_class[0] != 0);
4808  if (ranges->length() != (length >> 1) + 1) {
4809    return false;
4810  }
4811  CharacterRange range = ranges->at(0);
4812  if (range.from() != 0) {
4813    return false;
4814  }
4815  for (int i = 0; i < length; i += 2) {
4816    if (special_class[i] != (range.to() + 1)) {
4817      return false;
4818    }
4819    range = ranges->at((i >> 1) + 1);
4820    if (special_class[i+1] != range.from()) {
4821      return false;
4822    }
4823  }
4824  if (range.to() != String::kMaxCodePoint) {
4825    return false;
4826  }
4827  return true;
4828}
4829
4830
4831static bool CompareRanges(ZoneList<CharacterRange>* ranges,
4832                          const int* special_class,
4833                          int length) {
4834  length--;  // Remove final marker.
4835  DCHECK(special_class[length] == kRangeEndMarker);
4836  if (ranges->length() * 2 != length) {
4837    return false;
4838  }
4839  for (int i = 0; i < length; i += 2) {
4840    CharacterRange range = ranges->at(i >> 1);
4841    if (range.from() != special_class[i] ||
4842        range.to() != special_class[i + 1] - 1) {
4843      return false;
4844    }
4845  }
4846  return true;
4847}
4848
4849
4850bool RegExpCharacterClass::is_standard(Zone* zone) {
4851  // TODO(lrn): Remove need for this function, by not throwing away information
4852  // along the way.
4853  if (is_negated_) {
4854    return false;
4855  }
4856  if (set_.is_standard()) {
4857    return true;
4858  }
4859  if (CompareRanges(set_.ranges(zone), kSpaceRanges, kSpaceRangeCount)) {
4860    set_.set_standard_set_type('s');
4861    return true;
4862  }
4863  if (CompareInverseRanges(set_.ranges(zone), kSpaceRanges, kSpaceRangeCount)) {
4864    set_.set_standard_set_type('S');
4865    return true;
4866  }
4867  if (CompareInverseRanges(set_.ranges(zone),
4868                           kLineTerminatorRanges,
4869                           kLineTerminatorRangeCount)) {
4870    set_.set_standard_set_type('.');
4871    return true;
4872  }
4873  if (CompareRanges(set_.ranges(zone),
4874                    kLineTerminatorRanges,
4875                    kLineTerminatorRangeCount)) {
4876    set_.set_standard_set_type('n');
4877    return true;
4878  }
4879  if (CompareRanges(set_.ranges(zone), kWordRanges, kWordRangeCount)) {
4880    set_.set_standard_set_type('w');
4881    return true;
4882  }
4883  if (CompareInverseRanges(set_.ranges(zone), kWordRanges, kWordRangeCount)) {
4884    set_.set_standard_set_type('W');
4885    return true;
4886  }
4887  return false;
4888}
4889
4890
4891UnicodeRangeSplitter::UnicodeRangeSplitter(Zone* zone,
4892                                           ZoneList<CharacterRange>* base)
4893    : zone_(zone),
4894      table_(zone),
4895      bmp_(nullptr),
4896      lead_surrogates_(nullptr),
4897      trail_surrogates_(nullptr),
4898      non_bmp_(nullptr) {
4899  // The unicode range splitter categorizes given character ranges into:
4900  // - Code points from the BMP representable by one code unit.
4901  // - Code points outside the BMP that need to be split into surrogate pairs.
4902  // - Lone lead surrogates.
4903  // - Lone trail surrogates.
4904  // Lone surrogates are valid code points, even though no actual characters.
4905  // They require special matching to make sure we do not split surrogate pairs.
4906  // We use the dispatch table to accomplish this. The base range is split up
4907  // by the table by the overlay ranges, and the Call callback is used to
4908  // filter and collect ranges for each category.
4909  for (int i = 0; i < base->length(); i++) {
4910    table_.AddRange(base->at(i), kBase, zone_);
4911  }
4912  // Add overlay ranges.
4913  table_.AddRange(CharacterRange::Range(0, kLeadSurrogateStart - 1),
4914                  kBmpCodePoints, zone_);
4915  table_.AddRange(CharacterRange::Range(kLeadSurrogateStart, kLeadSurrogateEnd),
4916                  kLeadSurrogates, zone_);
4917  table_.AddRange(
4918      CharacterRange::Range(kTrailSurrogateStart, kTrailSurrogateEnd),
4919      kTrailSurrogates, zone_);
4920  table_.AddRange(
4921      CharacterRange::Range(kTrailSurrogateEnd + 1, kNonBmpStart - 1),
4922      kBmpCodePoints, zone_);
4923  table_.AddRange(CharacterRange::Range(kNonBmpStart, kNonBmpEnd),
4924                  kNonBmpCodePoints, zone_);
4925  table_.ForEach(this);
4926}
4927
4928
4929void UnicodeRangeSplitter::Call(uc32 from, DispatchTable::Entry entry) {
4930  OutSet* outset = entry.out_set();
4931  if (!outset->Get(kBase)) return;
4932  ZoneList<CharacterRange>** target = NULL;
4933  if (outset->Get(kBmpCodePoints)) {
4934    target = &bmp_;
4935  } else if (outset->Get(kLeadSurrogates)) {
4936    target = &lead_surrogates_;
4937  } else if (outset->Get(kTrailSurrogates)) {
4938    target = &trail_surrogates_;
4939  } else {
4940    DCHECK(outset->Get(kNonBmpCodePoints));
4941    target = &non_bmp_;
4942  }
4943  if (*target == NULL) *target = new (zone_) ZoneList<CharacterRange>(2, zone_);
4944  (*target)->Add(CharacterRange::Range(entry.from(), entry.to()), zone_);
4945}
4946
4947
4948void AddBmpCharacters(RegExpCompiler* compiler, ChoiceNode* result,
4949                      RegExpNode* on_success, UnicodeRangeSplitter* splitter) {
4950  ZoneList<CharacterRange>* bmp = splitter->bmp();
4951  if (bmp == nullptr) return;
4952  result->AddAlternative(GuardedAlternative(TextNode::CreateForCharacterRanges(
4953      compiler->zone(), bmp, compiler->read_backward(), on_success)));
4954}
4955
4956
4957void AddNonBmpSurrogatePairs(RegExpCompiler* compiler, ChoiceNode* result,
4958                             RegExpNode* on_success,
4959                             UnicodeRangeSplitter* splitter) {
4960  ZoneList<CharacterRange>* non_bmp = splitter->non_bmp();
4961  if (non_bmp == nullptr) return;
4962  DCHECK(compiler->unicode());
4963  DCHECK(!compiler->one_byte());
4964  Zone* zone = compiler->zone();
4965  CharacterRange::Canonicalize(non_bmp);
4966  for (int i = 0; i < non_bmp->length(); i++) {
4967    // Match surrogate pair.
4968    // E.g. [\u10005-\u11005] becomes
4969    //      \ud800[\udc05-\udfff]|
4970    //      [\ud801-\ud803][\udc00-\udfff]|
4971    //      \ud804[\udc00-\udc05]
4972    uc32 from = non_bmp->at(i).from();
4973    uc32 to = non_bmp->at(i).to();
4974    uc16 from_l = unibrow::Utf16::LeadSurrogate(from);
4975    uc16 from_t = unibrow::Utf16::TrailSurrogate(from);
4976    uc16 to_l = unibrow::Utf16::LeadSurrogate(to);
4977    uc16 to_t = unibrow::Utf16::TrailSurrogate(to);
4978    if (from_l == to_l) {
4979      // The lead surrogate is the same.
4980      result->AddAlternative(
4981          GuardedAlternative(TextNode::CreateForSurrogatePair(
4982              zone, CharacterRange::Singleton(from_l),
4983              CharacterRange::Range(from_t, to_t), compiler->read_backward(),
4984              on_success)));
4985    } else {
4986      if (from_t != kTrailSurrogateStart) {
4987        // Add [from_l][from_t-\udfff]
4988        result->AddAlternative(
4989            GuardedAlternative(TextNode::CreateForSurrogatePair(
4990                zone, CharacterRange::Singleton(from_l),
4991                CharacterRange::Range(from_t, kTrailSurrogateEnd),
4992                compiler->read_backward(), on_success)));
4993        from_l++;
4994      }
4995      if (to_t != kTrailSurrogateEnd) {
4996        // Add [to_l][\udc00-to_t]
4997        result->AddAlternative(
4998            GuardedAlternative(TextNode::CreateForSurrogatePair(
4999                zone, CharacterRange::Singleton(to_l),
5000                CharacterRange::Range(kTrailSurrogateStart, to_t),
5001                compiler->read_backward(), on_success)));
5002        to_l--;
5003      }
5004      if (from_l <= to_l) {
5005        // Add [from_l-to_l][\udc00-\udfff]
5006        result->AddAlternative(
5007            GuardedAlternative(TextNode::CreateForSurrogatePair(
5008                zone, CharacterRange::Range(from_l, to_l),
5009                CharacterRange::Range(kTrailSurrogateStart, kTrailSurrogateEnd),
5010                compiler->read_backward(), on_success)));
5011      }
5012    }
5013  }
5014}
5015
5016
5017RegExpNode* NegativeLookaroundAgainstReadDirectionAndMatch(
5018    RegExpCompiler* compiler, ZoneList<CharacterRange>* lookbehind,
5019    ZoneList<CharacterRange>* match, RegExpNode* on_success,
5020    bool read_backward) {
5021  Zone* zone = compiler->zone();
5022  RegExpNode* match_node = TextNode::CreateForCharacterRanges(
5023      zone, match, read_backward, on_success);
5024  int stack_register = compiler->UnicodeLookaroundStackRegister();
5025  int position_register = compiler->UnicodeLookaroundPositionRegister();
5026  RegExpLookaround::Builder lookaround(false, match_node, stack_register,
5027                                       position_register);
5028  RegExpNode* negative_match = TextNode::CreateForCharacterRanges(
5029      zone, lookbehind, !read_backward, lookaround.on_match_success());
5030  return lookaround.ForMatch(negative_match);
5031}
5032
5033
5034RegExpNode* MatchAndNegativeLookaroundInReadDirection(
5035    RegExpCompiler* compiler, ZoneList<CharacterRange>* match,
5036    ZoneList<CharacterRange>* lookahead, RegExpNode* on_success,
5037    bool read_backward) {
5038  Zone* zone = compiler->zone();
5039  int stack_register = compiler->UnicodeLookaroundStackRegister();
5040  int position_register = compiler->UnicodeLookaroundPositionRegister();
5041  RegExpLookaround::Builder lookaround(false, on_success, stack_register,
5042                                       position_register);
5043  RegExpNode* negative_match = TextNode::CreateForCharacterRanges(
5044      zone, lookahead, read_backward, lookaround.on_match_success());
5045  return TextNode::CreateForCharacterRanges(
5046      zone, match, read_backward, lookaround.ForMatch(negative_match));
5047}
5048
5049
5050void AddLoneLeadSurrogates(RegExpCompiler* compiler, ChoiceNode* result,
5051                           RegExpNode* on_success,
5052                           UnicodeRangeSplitter* splitter) {
5053  ZoneList<CharacterRange>* lead_surrogates = splitter->lead_surrogates();
5054  if (lead_surrogates == nullptr) return;
5055  Zone* zone = compiler->zone();
5056  // E.g. \ud801 becomes \ud801(?![\udc00-\udfff]).
5057  ZoneList<CharacterRange>* trail_surrogates = CharacterRange::List(
5058      zone, CharacterRange::Range(kTrailSurrogateStart, kTrailSurrogateEnd));
5059
5060  RegExpNode* match;
5061  if (compiler->read_backward()) {
5062    // Reading backward. Assert that reading forward, there is no trail
5063    // surrogate, and then backward match the lead surrogate.
5064    match = NegativeLookaroundAgainstReadDirectionAndMatch(
5065        compiler, trail_surrogates, lead_surrogates, on_success, true);
5066  } else {
5067    // Reading forward. Forward match the lead surrogate and assert that
5068    // no trail surrogate follows.
5069    match = MatchAndNegativeLookaroundInReadDirection(
5070        compiler, lead_surrogates, trail_surrogates, on_success, false);
5071  }
5072  result->AddAlternative(GuardedAlternative(match));
5073}
5074
5075
5076void AddLoneTrailSurrogates(RegExpCompiler* compiler, ChoiceNode* result,
5077                            RegExpNode* on_success,
5078                            UnicodeRangeSplitter* splitter) {
5079  ZoneList<CharacterRange>* trail_surrogates = splitter->trail_surrogates();
5080  if (trail_surrogates == nullptr) return;
5081  Zone* zone = compiler->zone();
5082  // E.g. \udc01 becomes (?<![\ud800-\udbff])\udc01
5083  ZoneList<CharacterRange>* lead_surrogates = CharacterRange::List(
5084      zone, CharacterRange::Range(kLeadSurrogateStart, kLeadSurrogateEnd));
5085
5086  RegExpNode* match;
5087  if (compiler->read_backward()) {
5088    // Reading backward. Backward match the trail surrogate and assert that no
5089    // lead surrogate precedes it.
5090    match = MatchAndNegativeLookaroundInReadDirection(
5091        compiler, trail_surrogates, lead_surrogates, on_success, true);
5092  } else {
5093    // Reading forward. Assert that reading backward, there is no lead
5094    // surrogate, and then forward match the trail surrogate.
5095    match = NegativeLookaroundAgainstReadDirectionAndMatch(
5096        compiler, lead_surrogates, trail_surrogates, on_success, false);
5097  }
5098  result->AddAlternative(GuardedAlternative(match));
5099}
5100
5101RegExpNode* UnanchoredAdvance(RegExpCompiler* compiler,
5102                              RegExpNode* on_success) {
5103  // This implements ES2015 21.2.5.2.3, AdvanceStringIndex.
5104  DCHECK(!compiler->read_backward());
5105  Zone* zone = compiler->zone();
5106  // Advance any character. If the character happens to be a lead surrogate and
5107  // we advanced into the middle of a surrogate pair, it will work out, as
5108  // nothing will match from there. We will have to advance again, consuming
5109  // the associated trail surrogate.
5110  ZoneList<CharacterRange>* range = CharacterRange::List(
5111      zone, CharacterRange::Range(0, String::kMaxUtf16CodeUnit));
5112  return TextNode::CreateForCharacterRanges(zone, range, false, on_success);
5113}
5114
5115
5116void AddUnicodeCaseEquivalents(RegExpCompiler* compiler,
5117                               ZoneList<CharacterRange>* ranges) {
5118#ifdef V8_I18N_SUPPORT
5119  // Use ICU to compute the case fold closure over the ranges.
5120  DCHECK(compiler->unicode());
5121  DCHECK(compiler->ignore_case());
5122  USet* set = uset_openEmpty();
5123  for (int i = 0; i < ranges->length(); i++) {
5124    uset_addRange(set, ranges->at(i).from(), ranges->at(i).to());
5125  }
5126  ranges->Clear();
5127  uset_closeOver(set, USET_CASE_INSENSITIVE);
5128  // Full case mapping map single characters to multiple characters.
5129  // Those are represented as strings in the set. Remove them so that
5130  // we end up with only simple and common case mappings.
5131  uset_removeAllStrings(set);
5132  int item_count = uset_getItemCount(set);
5133  int item_result = 0;
5134  UErrorCode ec = U_ZERO_ERROR;
5135  Zone* zone = compiler->zone();
5136  for (int i = 0; i < item_count; i++) {
5137    uc32 start = 0;
5138    uc32 end = 0;
5139    item_result += uset_getItem(set, i, &start, &end, nullptr, 0, &ec);
5140    ranges->Add(CharacterRange::Range(start, end), zone);
5141  }
5142  // No errors and everything we collected have been ranges.
5143  DCHECK_EQ(U_ZERO_ERROR, ec);
5144  DCHECK_EQ(0, item_result);
5145  uset_close(set);
5146#else
5147  // Fallback if ICU is not included.
5148  CharacterRange::AddCaseEquivalents(compiler->isolate(), compiler->zone(),
5149                                     ranges, compiler->one_byte());
5150#endif  // V8_I18N_SUPPORT
5151  CharacterRange::Canonicalize(ranges);
5152}
5153
5154
5155RegExpNode* RegExpCharacterClass::ToNode(RegExpCompiler* compiler,
5156                                         RegExpNode* on_success) {
5157  set_.Canonicalize();
5158  Zone* zone = compiler->zone();
5159  ZoneList<CharacterRange>* ranges = this->ranges(zone);
5160  if (compiler->unicode() && compiler->ignore_case()) {
5161    AddUnicodeCaseEquivalents(compiler, ranges);
5162  }
5163  if (compiler->unicode() && !compiler->one_byte()) {
5164    if (is_negated()) {
5165      ZoneList<CharacterRange>* negated =
5166          new (zone) ZoneList<CharacterRange>(2, zone);
5167      CharacterRange::Negate(ranges, negated, zone);
5168      ranges = negated;
5169    }
5170    if (ranges->length() == 0) {
5171      ranges->Add(CharacterRange::Everything(), zone);
5172      RegExpCharacterClass* fail =
5173          new (zone) RegExpCharacterClass(ranges, true);
5174      return new (zone) TextNode(fail, compiler->read_backward(), on_success);
5175    }
5176    if (standard_type() == '*') {
5177      return UnanchoredAdvance(compiler, on_success);
5178    } else {
5179      ChoiceNode* result = new (zone) ChoiceNode(2, zone);
5180      UnicodeRangeSplitter splitter(zone, ranges);
5181      AddBmpCharacters(compiler, result, on_success, &splitter);
5182      AddNonBmpSurrogatePairs(compiler, result, on_success, &splitter);
5183      AddLoneLeadSurrogates(compiler, result, on_success, &splitter);
5184      AddLoneTrailSurrogates(compiler, result, on_success, &splitter);
5185      return result;
5186    }
5187  } else {
5188    return new (zone) TextNode(this, compiler->read_backward(), on_success);
5189  }
5190}
5191
5192
5193int CompareFirstChar(RegExpTree* const* a, RegExpTree* const* b) {
5194  RegExpAtom* atom1 = (*a)->AsAtom();
5195  RegExpAtom* atom2 = (*b)->AsAtom();
5196  uc16 character1 = atom1->data().at(0);
5197  uc16 character2 = atom2->data().at(0);
5198  if (character1 < character2) return -1;
5199  if (character1 > character2) return 1;
5200  return 0;
5201}
5202
5203
5204static unibrow::uchar Canonical(
5205    unibrow::Mapping<unibrow::Ecma262Canonicalize>* canonicalize,
5206    unibrow::uchar c) {
5207  unibrow::uchar chars[unibrow::Ecma262Canonicalize::kMaxWidth];
5208  int length = canonicalize->get(c, '\0', chars);
5209  DCHECK_LE(length, 1);
5210  unibrow::uchar canonical = c;
5211  if (length == 1) canonical = chars[0];
5212  return canonical;
5213}
5214
5215
5216int CompareFirstCharCaseIndependent(
5217    unibrow::Mapping<unibrow::Ecma262Canonicalize>* canonicalize,
5218    RegExpTree* const* a, RegExpTree* const* b) {
5219  RegExpAtom* atom1 = (*a)->AsAtom();
5220  RegExpAtom* atom2 = (*b)->AsAtom();
5221  unibrow::uchar character1 = atom1->data().at(0);
5222  unibrow::uchar character2 = atom2->data().at(0);
5223  if (character1 == character2) return 0;
5224  if (character1 >= 'a' || character2 >= 'a') {
5225    character1 = Canonical(canonicalize, character1);
5226    character2 = Canonical(canonicalize, character2);
5227  }
5228  return static_cast<int>(character1) - static_cast<int>(character2);
5229}
5230
5231
5232// We can stable sort runs of atoms, since the order does not matter if they
5233// start with different characters.
5234// Returns true if any consecutive atoms were found.
5235bool RegExpDisjunction::SortConsecutiveAtoms(RegExpCompiler* compiler) {
5236  ZoneList<RegExpTree*>* alternatives = this->alternatives();
5237  int length = alternatives->length();
5238  bool found_consecutive_atoms = false;
5239  for (int i = 0; i < length; i++) {
5240    while (i < length) {
5241      RegExpTree* alternative = alternatives->at(i);
5242      if (alternative->IsAtom()) break;
5243      i++;
5244    }
5245    // i is length or it is the index of an atom.
5246    if (i == length) break;
5247    int first_atom = i;
5248    i++;
5249    while (i < length) {
5250      RegExpTree* alternative = alternatives->at(i);
5251      if (!alternative->IsAtom()) break;
5252      i++;
5253    }
5254    // Sort atoms to get ones with common prefixes together.
5255    // This step is more tricky if we are in a case-independent regexp,
5256    // because it would change /is|I/ to /I|is/, and order matters when
5257    // the regexp parts don't match only disjoint starting points. To fix
5258    // this we have a version of CompareFirstChar that uses case-
5259    // independent character classes for comparison.
5260    DCHECK_LT(first_atom, alternatives->length());
5261    DCHECK_LE(i, alternatives->length());
5262    DCHECK_LE(first_atom, i);
5263    if (compiler->ignore_case()) {
5264      unibrow::Mapping<unibrow::Ecma262Canonicalize>* canonicalize =
5265          compiler->isolate()->regexp_macro_assembler_canonicalize();
5266      auto compare_closure =
5267          [canonicalize](RegExpTree* const* a, RegExpTree* const* b) {
5268            return CompareFirstCharCaseIndependent(canonicalize, a, b);
5269          };
5270      alternatives->StableSort(compare_closure, first_atom, i - first_atom);
5271    } else {
5272      alternatives->StableSort(CompareFirstChar, first_atom, i - first_atom);
5273    }
5274    if (i - first_atom > 1) found_consecutive_atoms = true;
5275  }
5276  return found_consecutive_atoms;
5277}
5278
5279
5280// Optimizes ab|ac|az to a(?:b|c|d).
5281void RegExpDisjunction::RationalizeConsecutiveAtoms(RegExpCompiler* compiler) {
5282  Zone* zone = compiler->zone();
5283  ZoneList<RegExpTree*>* alternatives = this->alternatives();
5284  int length = alternatives->length();
5285
5286  int write_posn = 0;
5287  int i = 0;
5288  while (i < length) {
5289    RegExpTree* alternative = alternatives->at(i);
5290    if (!alternative->IsAtom()) {
5291      alternatives->at(write_posn++) = alternatives->at(i);
5292      i++;
5293      continue;
5294    }
5295    RegExpAtom* atom = alternative->AsAtom();
5296    unibrow::uchar common_prefix = atom->data().at(0);
5297    int first_with_prefix = i;
5298    int prefix_length = atom->length();
5299    i++;
5300    while (i < length) {
5301      alternative = alternatives->at(i);
5302      if (!alternative->IsAtom()) break;
5303      atom = alternative->AsAtom();
5304      unibrow::uchar new_prefix = atom->data().at(0);
5305      if (new_prefix != common_prefix) {
5306        if (!compiler->ignore_case()) break;
5307        unibrow::Mapping<unibrow::Ecma262Canonicalize>* canonicalize =
5308            compiler->isolate()->regexp_macro_assembler_canonicalize();
5309        new_prefix = Canonical(canonicalize, new_prefix);
5310        common_prefix = Canonical(canonicalize, common_prefix);
5311        if (new_prefix != common_prefix) break;
5312      }
5313      prefix_length = Min(prefix_length, atom->length());
5314      i++;
5315    }
5316    if (i > first_with_prefix + 2) {
5317      // Found worthwhile run of alternatives with common prefix of at least one
5318      // character.  The sorting function above did not sort on more than one
5319      // character for reasons of correctness, but there may still be a longer
5320      // common prefix if the terms were similar or presorted in the input.
5321      // Find out how long the common prefix is.
5322      int run_length = i - first_with_prefix;
5323      atom = alternatives->at(first_with_prefix)->AsAtom();
5324      for (int j = 1; j < run_length && prefix_length > 1; j++) {
5325        RegExpAtom* old_atom =
5326            alternatives->at(j + first_with_prefix)->AsAtom();
5327        for (int k = 1; k < prefix_length; k++) {
5328          if (atom->data().at(k) != old_atom->data().at(k)) {
5329            prefix_length = k;
5330            break;
5331          }
5332        }
5333      }
5334      RegExpAtom* prefix =
5335          new (zone) RegExpAtom(atom->data().SubVector(0, prefix_length));
5336      ZoneList<RegExpTree*>* pair = new (zone) ZoneList<RegExpTree*>(2, zone);
5337      pair->Add(prefix, zone);
5338      ZoneList<RegExpTree*>* suffixes =
5339          new (zone) ZoneList<RegExpTree*>(run_length, zone);
5340      for (int j = 0; j < run_length; j++) {
5341        RegExpAtom* old_atom =
5342            alternatives->at(j + first_with_prefix)->AsAtom();
5343        int len = old_atom->length();
5344        if (len == prefix_length) {
5345          suffixes->Add(new (zone) RegExpEmpty(), zone);
5346        } else {
5347          RegExpTree* suffix = new (zone) RegExpAtom(
5348              old_atom->data().SubVector(prefix_length, old_atom->length()));
5349          suffixes->Add(suffix, zone);
5350        }
5351      }
5352      pair->Add(new (zone) RegExpDisjunction(suffixes), zone);
5353      alternatives->at(write_posn++) = new (zone) RegExpAlternative(pair);
5354    } else {
5355      // Just copy any non-worthwhile alternatives.
5356      for (int j = first_with_prefix; j < i; j++) {
5357        alternatives->at(write_posn++) = alternatives->at(j);
5358      }
5359    }
5360  }
5361  alternatives->Rewind(write_posn);  // Trim end of array.
5362}
5363
5364
5365// Optimizes b|c|z to [bcz].
5366void RegExpDisjunction::FixSingleCharacterDisjunctions(
5367    RegExpCompiler* compiler) {
5368  Zone* zone = compiler->zone();
5369  ZoneList<RegExpTree*>* alternatives = this->alternatives();
5370  int length = alternatives->length();
5371
5372  int write_posn = 0;
5373  int i = 0;
5374  while (i < length) {
5375    RegExpTree* alternative = alternatives->at(i);
5376    if (!alternative->IsAtom()) {
5377      alternatives->at(write_posn++) = alternatives->at(i);
5378      i++;
5379      continue;
5380    }
5381    RegExpAtom* atom = alternative->AsAtom();
5382    if (atom->length() != 1) {
5383      alternatives->at(write_posn++) = alternatives->at(i);
5384      i++;
5385      continue;
5386    }
5387    int first_in_run = i;
5388    i++;
5389    while (i < length) {
5390      alternative = alternatives->at(i);
5391      if (!alternative->IsAtom()) break;
5392      atom = alternative->AsAtom();
5393      if (atom->length() != 1) break;
5394      i++;
5395    }
5396    if (i > first_in_run + 1) {
5397      // Found non-trivial run of single-character alternatives.
5398      int run_length = i - first_in_run;
5399      ZoneList<CharacterRange>* ranges =
5400          new (zone) ZoneList<CharacterRange>(2, zone);
5401      for (int j = 0; j < run_length; j++) {
5402        RegExpAtom* old_atom = alternatives->at(j + first_in_run)->AsAtom();
5403        DCHECK_EQ(old_atom->length(), 1);
5404        ranges->Add(CharacterRange::Singleton(old_atom->data().at(0)), zone);
5405      }
5406      alternatives->at(write_posn++) =
5407          new (zone) RegExpCharacterClass(ranges, false);
5408    } else {
5409      // Just copy any trivial alternatives.
5410      for (int j = first_in_run; j < i; j++) {
5411        alternatives->at(write_posn++) = alternatives->at(j);
5412      }
5413    }
5414  }
5415  alternatives->Rewind(write_posn);  // Trim end of array.
5416}
5417
5418
5419RegExpNode* RegExpDisjunction::ToNode(RegExpCompiler* compiler,
5420                                      RegExpNode* on_success) {
5421  ZoneList<RegExpTree*>* alternatives = this->alternatives();
5422
5423  if (alternatives->length() > 2) {
5424    bool found_consecutive_atoms = SortConsecutiveAtoms(compiler);
5425    if (found_consecutive_atoms) RationalizeConsecutiveAtoms(compiler);
5426    FixSingleCharacterDisjunctions(compiler);
5427    if (alternatives->length() == 1) {
5428      return alternatives->at(0)->ToNode(compiler, on_success);
5429    }
5430  }
5431
5432  int length = alternatives->length();
5433
5434  ChoiceNode* result =
5435      new(compiler->zone()) ChoiceNode(length, compiler->zone());
5436  for (int i = 0; i < length; i++) {
5437    GuardedAlternative alternative(alternatives->at(i)->ToNode(compiler,
5438                                                               on_success));
5439    result->AddAlternative(alternative);
5440  }
5441  return result;
5442}
5443
5444
5445RegExpNode* RegExpQuantifier::ToNode(RegExpCompiler* compiler,
5446                                     RegExpNode* on_success) {
5447  return ToNode(min(),
5448                max(),
5449                is_greedy(),
5450                body(),
5451                compiler,
5452                on_success);
5453}
5454
5455
5456// Scoped object to keep track of how much we unroll quantifier loops in the
5457// regexp graph generator.
5458class RegExpExpansionLimiter {
5459 public:
5460  static const int kMaxExpansionFactor = 6;
5461  RegExpExpansionLimiter(RegExpCompiler* compiler, int factor)
5462      : compiler_(compiler),
5463        saved_expansion_factor_(compiler->current_expansion_factor()),
5464        ok_to_expand_(saved_expansion_factor_ <= kMaxExpansionFactor) {
5465    DCHECK(factor > 0);
5466    if (ok_to_expand_) {
5467      if (factor > kMaxExpansionFactor) {
5468        // Avoid integer overflow of the current expansion factor.
5469        ok_to_expand_ = false;
5470        compiler->set_current_expansion_factor(kMaxExpansionFactor + 1);
5471      } else {
5472        int new_factor = saved_expansion_factor_ * factor;
5473        ok_to_expand_ = (new_factor <= kMaxExpansionFactor);
5474        compiler->set_current_expansion_factor(new_factor);
5475      }
5476    }
5477  }
5478
5479  ~RegExpExpansionLimiter() {
5480    compiler_->set_current_expansion_factor(saved_expansion_factor_);
5481  }
5482
5483  bool ok_to_expand() { return ok_to_expand_; }
5484
5485 private:
5486  RegExpCompiler* compiler_;
5487  int saved_expansion_factor_;
5488  bool ok_to_expand_;
5489
5490  DISALLOW_IMPLICIT_CONSTRUCTORS(RegExpExpansionLimiter);
5491};
5492
5493
5494RegExpNode* RegExpQuantifier::ToNode(int min,
5495                                     int max,
5496                                     bool is_greedy,
5497                                     RegExpTree* body,
5498                                     RegExpCompiler* compiler,
5499                                     RegExpNode* on_success,
5500                                     bool not_at_start) {
5501  // x{f, t} becomes this:
5502  //
5503  //             (r++)<-.
5504  //               |     `
5505  //               |     (x)
5506  //               v     ^
5507  //      (r=0)-->(?)---/ [if r < t]
5508  //               |
5509  //   [if r >= f] \----> ...
5510  //
5511
5512  // 15.10.2.5 RepeatMatcher algorithm.
5513  // The parser has already eliminated the case where max is 0.  In the case
5514  // where max_match is zero the parser has removed the quantifier if min was
5515  // > 0 and removed the atom if min was 0.  See AddQuantifierToAtom.
5516
5517  // If we know that we cannot match zero length then things are a little
5518  // simpler since we don't need to make the special zero length match check
5519  // from step 2.1.  If the min and max are small we can unroll a little in
5520  // this case.
5521  static const int kMaxUnrolledMinMatches = 3;  // Unroll (foo)+ and (foo){3,}
5522  static const int kMaxUnrolledMaxMatches = 3;  // Unroll (foo)? and (foo){x,3}
5523  if (max == 0) return on_success;  // This can happen due to recursion.
5524  bool body_can_be_empty = (body->min_match() == 0);
5525  int body_start_reg = RegExpCompiler::kNoRegister;
5526  Interval capture_registers = body->CaptureRegisters();
5527  bool needs_capture_clearing = !capture_registers.is_empty();
5528  Zone* zone = compiler->zone();
5529
5530  if (body_can_be_empty) {
5531    body_start_reg = compiler->AllocateRegister();
5532  } else if (compiler->optimize() && !needs_capture_clearing) {
5533    // Only unroll if there are no captures and the body can't be
5534    // empty.
5535    {
5536      RegExpExpansionLimiter limiter(
5537          compiler, min + ((max != min) ? 1 : 0));
5538      if (min > 0 && min <= kMaxUnrolledMinMatches && limiter.ok_to_expand()) {
5539        int new_max = (max == kInfinity) ? max : max - min;
5540        // Recurse once to get the loop or optional matches after the fixed
5541        // ones.
5542        RegExpNode* answer = ToNode(
5543            0, new_max, is_greedy, body, compiler, on_success, true);
5544        // Unroll the forced matches from 0 to min.  This can cause chains of
5545        // TextNodes (which the parser does not generate).  These should be
5546        // combined if it turns out they hinder good code generation.
5547        for (int i = 0; i < min; i++) {
5548          answer = body->ToNode(compiler, answer);
5549        }
5550        return answer;
5551      }
5552    }
5553    if (max <= kMaxUnrolledMaxMatches && min == 0) {
5554      DCHECK(max > 0);  // Due to the 'if' above.
5555      RegExpExpansionLimiter limiter(compiler, max);
5556      if (limiter.ok_to_expand()) {
5557        // Unroll the optional matches up to max.
5558        RegExpNode* answer = on_success;
5559        for (int i = 0; i < max; i++) {
5560          ChoiceNode* alternation = new(zone) ChoiceNode(2, zone);
5561          if (is_greedy) {
5562            alternation->AddAlternative(
5563                GuardedAlternative(body->ToNode(compiler, answer)));
5564            alternation->AddAlternative(GuardedAlternative(on_success));
5565          } else {
5566            alternation->AddAlternative(GuardedAlternative(on_success));
5567            alternation->AddAlternative(
5568                GuardedAlternative(body->ToNode(compiler, answer)));
5569          }
5570          answer = alternation;
5571          if (not_at_start && !compiler->read_backward()) {
5572            alternation->set_not_at_start();
5573          }
5574        }
5575        return answer;
5576      }
5577    }
5578  }
5579  bool has_min = min > 0;
5580  bool has_max = max < RegExpTree::kInfinity;
5581  bool needs_counter = has_min || has_max;
5582  int reg_ctr = needs_counter
5583      ? compiler->AllocateRegister()
5584      : RegExpCompiler::kNoRegister;
5585  LoopChoiceNode* center = new (zone)
5586      LoopChoiceNode(body->min_match() == 0, compiler->read_backward(), zone);
5587  if (not_at_start && !compiler->read_backward()) center->set_not_at_start();
5588  RegExpNode* loop_return = needs_counter
5589      ? static_cast<RegExpNode*>(ActionNode::IncrementRegister(reg_ctr, center))
5590      : static_cast<RegExpNode*>(center);
5591  if (body_can_be_empty) {
5592    // If the body can be empty we need to check if it was and then
5593    // backtrack.
5594    loop_return = ActionNode::EmptyMatchCheck(body_start_reg,
5595                                              reg_ctr,
5596                                              min,
5597                                              loop_return);
5598  }
5599  RegExpNode* body_node = body->ToNode(compiler, loop_return);
5600  if (body_can_be_empty) {
5601    // If the body can be empty we need to store the start position
5602    // so we can bail out if it was empty.
5603    body_node = ActionNode::StorePosition(body_start_reg, false, body_node);
5604  }
5605  if (needs_capture_clearing) {
5606    // Before entering the body of this loop we need to clear captures.
5607    body_node = ActionNode::ClearCaptures(capture_registers, body_node);
5608  }
5609  GuardedAlternative body_alt(body_node);
5610  if (has_max) {
5611    Guard* body_guard =
5612        new(zone) Guard(reg_ctr, Guard::LT, max);
5613    body_alt.AddGuard(body_guard, zone);
5614  }
5615  GuardedAlternative rest_alt(on_success);
5616  if (has_min) {
5617    Guard* rest_guard = new(compiler->zone()) Guard(reg_ctr, Guard::GEQ, min);
5618    rest_alt.AddGuard(rest_guard, zone);
5619  }
5620  if (is_greedy) {
5621    center->AddLoopAlternative(body_alt);
5622    center->AddContinueAlternative(rest_alt);
5623  } else {
5624    center->AddContinueAlternative(rest_alt);
5625    center->AddLoopAlternative(body_alt);
5626  }
5627  if (needs_counter) {
5628    return ActionNode::SetRegister(reg_ctr, 0, center);
5629  } else {
5630    return center;
5631  }
5632}
5633
5634
5635RegExpNode* RegExpAssertion::ToNode(RegExpCompiler* compiler,
5636                                    RegExpNode* on_success) {
5637  NodeInfo info;
5638  Zone* zone = compiler->zone();
5639
5640  switch (assertion_type()) {
5641    case START_OF_LINE:
5642      return AssertionNode::AfterNewline(on_success);
5643    case START_OF_INPUT:
5644      return AssertionNode::AtStart(on_success);
5645    case BOUNDARY:
5646      return AssertionNode::AtBoundary(on_success);
5647    case NON_BOUNDARY:
5648      return AssertionNode::AtNonBoundary(on_success);
5649    case END_OF_INPUT:
5650      return AssertionNode::AtEnd(on_success);
5651    case END_OF_LINE: {
5652      // Compile $ in multiline regexps as an alternation with a positive
5653      // lookahead in one side and an end-of-input on the other side.
5654      // We need two registers for the lookahead.
5655      int stack_pointer_register = compiler->AllocateRegister();
5656      int position_register = compiler->AllocateRegister();
5657      // The ChoiceNode to distinguish between a newline and end-of-input.
5658      ChoiceNode* result = new(zone) ChoiceNode(2, zone);
5659      // Create a newline atom.
5660      ZoneList<CharacterRange>* newline_ranges =
5661          new(zone) ZoneList<CharacterRange>(3, zone);
5662      CharacterRange::AddClassEscape('n', newline_ranges, zone);
5663      RegExpCharacterClass* newline_atom = new (zone) RegExpCharacterClass('n');
5664      TextNode* newline_matcher = new (zone) TextNode(
5665          newline_atom, false, ActionNode::PositiveSubmatchSuccess(
5666                                   stack_pointer_register, position_register,
5667                                   0,   // No captures inside.
5668                                   -1,  // Ignored if no captures.
5669                                   on_success));
5670      // Create an end-of-input matcher.
5671      RegExpNode* end_of_line = ActionNode::BeginSubmatch(
5672          stack_pointer_register,
5673          position_register,
5674          newline_matcher);
5675      // Add the two alternatives to the ChoiceNode.
5676      GuardedAlternative eol_alternative(end_of_line);
5677      result->AddAlternative(eol_alternative);
5678      GuardedAlternative end_alternative(AssertionNode::AtEnd(on_success));
5679      result->AddAlternative(end_alternative);
5680      return result;
5681    }
5682    default:
5683      UNREACHABLE();
5684  }
5685  return on_success;
5686}
5687
5688
5689RegExpNode* RegExpBackReference::ToNode(RegExpCompiler* compiler,
5690                                        RegExpNode* on_success) {
5691  return new (compiler->zone())
5692      BackReferenceNode(RegExpCapture::StartRegister(index()),
5693                        RegExpCapture::EndRegister(index()),
5694                        compiler->read_backward(), on_success);
5695}
5696
5697
5698RegExpNode* RegExpEmpty::ToNode(RegExpCompiler* compiler,
5699                                RegExpNode* on_success) {
5700  return on_success;
5701}
5702
5703
5704RegExpLookaround::Builder::Builder(bool is_positive, RegExpNode* on_success,
5705                                   int stack_pointer_register,
5706                                   int position_register,
5707                                   int capture_register_count,
5708                                   int capture_register_start)
5709    : is_positive_(is_positive),
5710      on_success_(on_success),
5711      stack_pointer_register_(stack_pointer_register),
5712      position_register_(position_register) {
5713  if (is_positive_) {
5714    on_match_success_ = ActionNode::PositiveSubmatchSuccess(
5715        stack_pointer_register, position_register, capture_register_count,
5716        capture_register_start, on_success_);
5717  } else {
5718    Zone* zone = on_success_->zone();
5719    on_match_success_ = new (zone) NegativeSubmatchSuccess(
5720        stack_pointer_register, position_register, capture_register_count,
5721        capture_register_start, zone);
5722  }
5723}
5724
5725
5726RegExpNode* RegExpLookaround::Builder::ForMatch(RegExpNode* match) {
5727  if (is_positive_) {
5728    return ActionNode::BeginSubmatch(stack_pointer_register_,
5729                                     position_register_, match);
5730  } else {
5731    Zone* zone = on_success_->zone();
5732    // We use a ChoiceNode to represent the negative lookaround. The first
5733    // alternative is the negative match. On success, the end node backtracks.
5734    // On failure, the second alternative is tried and leads to success.
5735    // NegativeLookaheadChoiceNode is a special ChoiceNode that ignores the
5736    // first exit when calculating quick checks.
5737    ChoiceNode* choice_node = new (zone) NegativeLookaroundChoiceNode(
5738        GuardedAlternative(match), GuardedAlternative(on_success_), zone);
5739    return ActionNode::BeginSubmatch(stack_pointer_register_,
5740                                     position_register_, choice_node);
5741  }
5742}
5743
5744
5745RegExpNode* RegExpLookaround::ToNode(RegExpCompiler* compiler,
5746                                     RegExpNode* on_success) {
5747  int stack_pointer_register = compiler->AllocateRegister();
5748  int position_register = compiler->AllocateRegister();
5749
5750  const int registers_per_capture = 2;
5751  const int register_of_first_capture = 2;
5752  int register_count = capture_count_ * registers_per_capture;
5753  int register_start =
5754    register_of_first_capture + capture_from_ * registers_per_capture;
5755
5756  RegExpNode* result;
5757  bool was_reading_backward = compiler->read_backward();
5758  compiler->set_read_backward(type() == LOOKBEHIND);
5759  Builder builder(is_positive(), on_success, stack_pointer_register,
5760                  position_register, register_count, register_start);
5761  RegExpNode* match = body_->ToNode(compiler, builder.on_match_success());
5762  result = builder.ForMatch(match);
5763  compiler->set_read_backward(was_reading_backward);
5764  return result;
5765}
5766
5767
5768RegExpNode* RegExpCapture::ToNode(RegExpCompiler* compiler,
5769                                  RegExpNode* on_success) {
5770  return ToNode(body(), index(), compiler, on_success);
5771}
5772
5773
5774RegExpNode* RegExpCapture::ToNode(RegExpTree* body,
5775                                  int index,
5776                                  RegExpCompiler* compiler,
5777                                  RegExpNode* on_success) {
5778  DCHECK_NOT_NULL(body);
5779  int start_reg = RegExpCapture::StartRegister(index);
5780  int end_reg = RegExpCapture::EndRegister(index);
5781  if (compiler->read_backward()) std::swap(start_reg, end_reg);
5782  RegExpNode* store_end = ActionNode::StorePosition(end_reg, true, on_success);
5783  RegExpNode* body_node = body->ToNode(compiler, store_end);
5784  return ActionNode::StorePosition(start_reg, true, body_node);
5785}
5786
5787
5788RegExpNode* RegExpAlternative::ToNode(RegExpCompiler* compiler,
5789                                      RegExpNode* on_success) {
5790  ZoneList<RegExpTree*>* children = nodes();
5791  RegExpNode* current = on_success;
5792  if (compiler->read_backward()) {
5793    for (int i = 0; i < children->length(); i++) {
5794      current = children->at(i)->ToNode(compiler, current);
5795    }
5796  } else {
5797    for (int i = children->length() - 1; i >= 0; i--) {
5798      current = children->at(i)->ToNode(compiler, current);
5799    }
5800  }
5801  return current;
5802}
5803
5804
5805static void AddClass(const int* elmv,
5806                     int elmc,
5807                     ZoneList<CharacterRange>* ranges,
5808                     Zone* zone) {
5809  elmc--;
5810  DCHECK(elmv[elmc] == kRangeEndMarker);
5811  for (int i = 0; i < elmc; i += 2) {
5812    DCHECK(elmv[i] < elmv[i + 1]);
5813    ranges->Add(CharacterRange::Range(elmv[i], elmv[i + 1] - 1), zone);
5814  }
5815}
5816
5817
5818static void AddClassNegated(const int *elmv,
5819                            int elmc,
5820                            ZoneList<CharacterRange>* ranges,
5821                            Zone* zone) {
5822  elmc--;
5823  DCHECK(elmv[elmc] == kRangeEndMarker);
5824  DCHECK(elmv[0] != 0x0000);
5825  DCHECK(elmv[elmc - 1] != String::kMaxCodePoint);
5826  uc16 last = 0x0000;
5827  for (int i = 0; i < elmc; i += 2) {
5828    DCHECK(last <= elmv[i] - 1);
5829    DCHECK(elmv[i] < elmv[i + 1]);
5830    ranges->Add(CharacterRange::Range(last, elmv[i] - 1), zone);
5831    last = elmv[i + 1];
5832  }
5833  ranges->Add(CharacterRange::Range(last, String::kMaxCodePoint), zone);
5834}
5835
5836
5837void CharacterRange::AddClassEscape(uc16 type,
5838                                    ZoneList<CharacterRange>* ranges,
5839                                    Zone* zone) {
5840  switch (type) {
5841    case 's':
5842      AddClass(kSpaceRanges, kSpaceRangeCount, ranges, zone);
5843      break;
5844    case 'S':
5845      AddClassNegated(kSpaceRanges, kSpaceRangeCount, ranges, zone);
5846      break;
5847    case 'w':
5848      AddClass(kWordRanges, kWordRangeCount, ranges, zone);
5849      break;
5850    case 'W':
5851      AddClassNegated(kWordRanges, kWordRangeCount, ranges, zone);
5852      break;
5853    case 'd':
5854      AddClass(kDigitRanges, kDigitRangeCount, ranges, zone);
5855      break;
5856    case 'D':
5857      AddClassNegated(kDigitRanges, kDigitRangeCount, ranges, zone);
5858      break;
5859    case '.':
5860      AddClassNegated(kLineTerminatorRanges,
5861                      kLineTerminatorRangeCount,
5862                      ranges,
5863                      zone);
5864      break;
5865    // This is not a character range as defined by the spec but a
5866    // convenient shorthand for a character class that matches any
5867    // character.
5868    case '*':
5869      ranges->Add(CharacterRange::Everything(), zone);
5870      break;
5871    // This is the set of characters matched by the $ and ^ symbols
5872    // in multiline mode.
5873    case 'n':
5874      AddClass(kLineTerminatorRanges,
5875               kLineTerminatorRangeCount,
5876               ranges,
5877               zone);
5878      break;
5879    default:
5880      UNREACHABLE();
5881  }
5882}
5883
5884
5885Vector<const int> CharacterRange::GetWordBounds() {
5886  return Vector<const int>(kWordRanges, kWordRangeCount - 1);
5887}
5888
5889
5890void CharacterRange::AddCaseEquivalents(Isolate* isolate, Zone* zone,
5891                                        ZoneList<CharacterRange>* ranges,
5892                                        bool is_one_byte) {
5893  CharacterRange::Canonicalize(ranges);
5894  int range_count = ranges->length();
5895  for (int i = 0; i < range_count; i++) {
5896    CharacterRange range = ranges->at(i);
5897    uc32 bottom = range.from();
5898    if (bottom > String::kMaxUtf16CodeUnit) return;
5899    uc32 top = Min(range.to(), String::kMaxUtf16CodeUnit);
5900    // Nothing to be done for surrogates.
5901    if (bottom >= kLeadSurrogateStart && top <= kTrailSurrogateEnd) return;
5902    if (is_one_byte && !RangeContainsLatin1Equivalents(range)) {
5903      if (bottom > String::kMaxOneByteCharCode) return;
5904      if (top > String::kMaxOneByteCharCode) top = String::kMaxOneByteCharCode;
5905    }
5906    unibrow::uchar chars[unibrow::Ecma262UnCanonicalize::kMaxWidth];
5907    if (top == bottom) {
5908      // If this is a singleton we just expand the one character.
5909      int length = isolate->jsregexp_uncanonicalize()->get(bottom, '\0', chars);
5910      for (int i = 0; i < length; i++) {
5911        uc32 chr = chars[i];
5912        if (chr != bottom) {
5913          ranges->Add(CharacterRange::Singleton(chars[i]), zone);
5914        }
5915      }
5916    } else {
5917      // If this is a range we expand the characters block by block, expanding
5918      // contiguous subranges (blocks) one at a time.  The approach is as
5919      // follows.  For a given start character we look up the remainder of the
5920      // block that contains it (represented by the end point), for instance we
5921      // find 'z' if the character is 'c'.  A block is characterized by the
5922      // property that all characters uncanonicalize in the same way, except
5923      // that each entry in the result is incremented by the distance from the
5924      // first element.  So a-z is a block because 'a' uncanonicalizes to ['a',
5925      // 'A'] and the k'th letter uncanonicalizes to ['a' + k, 'A' + k].  Once
5926      // we've found the end point we look up its uncanonicalization and
5927      // produce a range for each element.  For instance for [c-f] we look up
5928      // ['z', 'Z'] and produce [c-f] and [C-F].  We then only add a range if
5929      // it is not already contained in the input, so [c-f] will be skipped but
5930      // [C-F] will be added.  If this range is not completely contained in a
5931      // block we do this for all the blocks covered by the range (handling
5932      // characters that is not in a block as a "singleton block").
5933      unibrow::uchar equivalents[unibrow::Ecma262UnCanonicalize::kMaxWidth];
5934      int pos = bottom;
5935      while (pos <= top) {
5936        int length =
5937            isolate->jsregexp_canonrange()->get(pos, '\0', equivalents);
5938        uc32 block_end;
5939        if (length == 0) {
5940          block_end = pos;
5941        } else {
5942          DCHECK_EQ(1, length);
5943          block_end = equivalents[0];
5944        }
5945        int end = (block_end > top) ? top : block_end;
5946        length = isolate->jsregexp_uncanonicalize()->get(block_end, '\0',
5947                                                         equivalents);
5948        for (int i = 0; i < length; i++) {
5949          uc32 c = equivalents[i];
5950          uc32 range_from = c - (block_end - pos);
5951          uc32 range_to = c - (block_end - end);
5952          if (!(bottom <= range_from && range_to <= top)) {
5953            ranges->Add(CharacterRange::Range(range_from, range_to), zone);
5954          }
5955        }
5956        pos = end + 1;
5957      }
5958    }
5959  }
5960}
5961
5962
5963bool CharacterRange::IsCanonical(ZoneList<CharacterRange>* ranges) {
5964  DCHECK_NOT_NULL(ranges);
5965  int n = ranges->length();
5966  if (n <= 1) return true;
5967  int max = ranges->at(0).to();
5968  for (int i = 1; i < n; i++) {
5969    CharacterRange next_range = ranges->at(i);
5970    if (next_range.from() <= max + 1) return false;
5971    max = next_range.to();
5972  }
5973  return true;
5974}
5975
5976
5977ZoneList<CharacterRange>* CharacterSet::ranges(Zone* zone) {
5978  if (ranges_ == NULL) {
5979    ranges_ = new(zone) ZoneList<CharacterRange>(2, zone);
5980    CharacterRange::AddClassEscape(standard_set_type_, ranges_, zone);
5981  }
5982  return ranges_;
5983}
5984
5985
5986// Move a number of elements in a zonelist to another position
5987// in the same list. Handles overlapping source and target areas.
5988static void MoveRanges(ZoneList<CharacterRange>* list,
5989                       int from,
5990                       int to,
5991                       int count) {
5992  // Ranges are potentially overlapping.
5993  if (from < to) {
5994    for (int i = count - 1; i >= 0; i--) {
5995      list->at(to + i) = list->at(from + i);
5996    }
5997  } else {
5998    for (int i = 0; i < count; i++) {
5999      list->at(to + i) = list->at(from + i);
6000    }
6001  }
6002}
6003
6004
6005static int InsertRangeInCanonicalList(ZoneList<CharacterRange>* list,
6006                                      int count,
6007                                      CharacterRange insert) {
6008  // Inserts a range into list[0..count[, which must be sorted
6009  // by from value and non-overlapping and non-adjacent, using at most
6010  // list[0..count] for the result. Returns the number of resulting
6011  // canonicalized ranges. Inserting a range may collapse existing ranges into
6012  // fewer ranges, so the return value can be anything in the range 1..count+1.
6013  uc32 from = insert.from();
6014  uc32 to = insert.to();
6015  int start_pos = 0;
6016  int end_pos = count;
6017  for (int i = count - 1; i >= 0; i--) {
6018    CharacterRange current = list->at(i);
6019    if (current.from() > to + 1) {
6020      end_pos = i;
6021    } else if (current.to() + 1 < from) {
6022      start_pos = i + 1;
6023      break;
6024    }
6025  }
6026
6027  // Inserted range overlaps, or is adjacent to, ranges at positions
6028  // [start_pos..end_pos[. Ranges before start_pos or at or after end_pos are
6029  // not affected by the insertion.
6030  // If start_pos == end_pos, the range must be inserted before start_pos.
6031  // if start_pos < end_pos, the entire range from start_pos to end_pos
6032  // must be merged with the insert range.
6033
6034  if (start_pos == end_pos) {
6035    // Insert between existing ranges at position start_pos.
6036    if (start_pos < count) {
6037      MoveRanges(list, start_pos, start_pos + 1, count - start_pos);
6038    }
6039    list->at(start_pos) = insert;
6040    return count + 1;
6041  }
6042  if (start_pos + 1 == end_pos) {
6043    // Replace single existing range at position start_pos.
6044    CharacterRange to_replace = list->at(start_pos);
6045    int new_from = Min(to_replace.from(), from);
6046    int new_to = Max(to_replace.to(), to);
6047    list->at(start_pos) = CharacterRange::Range(new_from, new_to);
6048    return count;
6049  }
6050  // Replace a number of existing ranges from start_pos to end_pos - 1.
6051  // Move the remaining ranges down.
6052
6053  int new_from = Min(list->at(start_pos).from(), from);
6054  int new_to = Max(list->at(end_pos - 1).to(), to);
6055  if (end_pos < count) {
6056    MoveRanges(list, end_pos, start_pos + 1, count - end_pos);
6057  }
6058  list->at(start_pos) = CharacterRange::Range(new_from, new_to);
6059  return count - (end_pos - start_pos) + 1;
6060}
6061
6062
6063void CharacterSet::Canonicalize() {
6064  // Special/default classes are always considered canonical. The result
6065  // of calling ranges() will be sorted.
6066  if (ranges_ == NULL) return;
6067  CharacterRange::Canonicalize(ranges_);
6068}
6069
6070
6071void CharacterRange::Canonicalize(ZoneList<CharacterRange>* character_ranges) {
6072  if (character_ranges->length() <= 1) return;
6073  // Check whether ranges are already canonical (increasing, non-overlapping,
6074  // non-adjacent).
6075  int n = character_ranges->length();
6076  int max = character_ranges->at(0).to();
6077  int i = 1;
6078  while (i < n) {
6079    CharacterRange current = character_ranges->at(i);
6080    if (current.from() <= max + 1) {
6081      break;
6082    }
6083    max = current.to();
6084    i++;
6085  }
6086  // Canonical until the i'th range. If that's all of them, we are done.
6087  if (i == n) return;
6088
6089  // The ranges at index i and forward are not canonicalized. Make them so by
6090  // doing the equivalent of insertion sort (inserting each into the previous
6091  // list, in order).
6092  // Notice that inserting a range can reduce the number of ranges in the
6093  // result due to combining of adjacent and overlapping ranges.
6094  int read = i;  // Range to insert.
6095  int num_canonical = i;  // Length of canonicalized part of list.
6096  do {
6097    num_canonical = InsertRangeInCanonicalList(character_ranges,
6098                                               num_canonical,
6099                                               character_ranges->at(read));
6100    read++;
6101  } while (read < n);
6102  character_ranges->Rewind(num_canonical);
6103
6104  DCHECK(CharacterRange::IsCanonical(character_ranges));
6105}
6106
6107
6108void CharacterRange::Negate(ZoneList<CharacterRange>* ranges,
6109                            ZoneList<CharacterRange>* negated_ranges,
6110                            Zone* zone) {
6111  DCHECK(CharacterRange::IsCanonical(ranges));
6112  DCHECK_EQ(0, negated_ranges->length());
6113  int range_count = ranges->length();
6114  uc32 from = 0;
6115  int i = 0;
6116  if (range_count > 0 && ranges->at(0).from() == 0) {
6117    from = ranges->at(0).to() + 1;
6118    i = 1;
6119  }
6120  while (i < range_count) {
6121    CharacterRange range = ranges->at(i);
6122    negated_ranges->Add(CharacterRange::Range(from, range.from() - 1), zone);
6123    from = range.to() + 1;
6124    i++;
6125  }
6126  if (from < String::kMaxCodePoint) {
6127    negated_ranges->Add(CharacterRange::Range(from, String::kMaxCodePoint),
6128                        zone);
6129  }
6130}
6131
6132
6133// -------------------------------------------------------------------
6134// Splay tree
6135
6136
6137OutSet* OutSet::Extend(unsigned value, Zone* zone) {
6138  if (Get(value))
6139    return this;
6140  if (successors(zone) != NULL) {
6141    for (int i = 0; i < successors(zone)->length(); i++) {
6142      OutSet* successor = successors(zone)->at(i);
6143      if (successor->Get(value))
6144        return successor;
6145    }
6146  } else {
6147    successors_ = new(zone) ZoneList<OutSet*>(2, zone);
6148  }
6149  OutSet* result = new(zone) OutSet(first_, remaining_);
6150  result->Set(value, zone);
6151  successors(zone)->Add(result, zone);
6152  return result;
6153}
6154
6155
6156void OutSet::Set(unsigned value, Zone *zone) {
6157  if (value < kFirstLimit) {
6158    first_ |= (1 << value);
6159  } else {
6160    if (remaining_ == NULL)
6161      remaining_ = new(zone) ZoneList<unsigned>(1, zone);
6162    if (remaining_->is_empty() || !remaining_->Contains(value))
6163      remaining_->Add(value, zone);
6164  }
6165}
6166
6167
6168bool OutSet::Get(unsigned value) const {
6169  if (value < kFirstLimit) {
6170    return (first_ & (1 << value)) != 0;
6171  } else if (remaining_ == NULL) {
6172    return false;
6173  } else {
6174    return remaining_->Contains(value);
6175  }
6176}
6177
6178
6179const uc32 DispatchTable::Config::kNoKey = unibrow::Utf8::kBadChar;
6180
6181
6182void DispatchTable::AddRange(CharacterRange full_range, int value,
6183                             Zone* zone) {
6184  CharacterRange current = full_range;
6185  if (tree()->is_empty()) {
6186    // If this is the first range we just insert into the table.
6187    ZoneSplayTree<Config>::Locator loc;
6188    bool inserted = tree()->Insert(current.from(), &loc);
6189    DCHECK(inserted);
6190    USE(inserted);
6191    loc.set_value(Entry(current.from(), current.to(),
6192                        empty()->Extend(value, zone)));
6193    return;
6194  }
6195  // First see if there is a range to the left of this one that
6196  // overlaps.
6197  ZoneSplayTree<Config>::Locator loc;
6198  if (tree()->FindGreatestLessThan(current.from(), &loc)) {
6199    Entry* entry = &loc.value();
6200    // If we've found a range that overlaps with this one, and it
6201    // starts strictly to the left of this one, we have to fix it
6202    // because the following code only handles ranges that start on
6203    // or after the start point of the range we're adding.
6204    if (entry->from() < current.from() && entry->to() >= current.from()) {
6205      // Snap the overlapping range in half around the start point of
6206      // the range we're adding.
6207      CharacterRange left =
6208          CharacterRange::Range(entry->from(), current.from() - 1);
6209      CharacterRange right = CharacterRange::Range(current.from(), entry->to());
6210      // The left part of the overlapping range doesn't overlap.
6211      // Truncate the whole entry to be just the left part.
6212      entry->set_to(left.to());
6213      // The right part is the one that overlaps.  We add this part
6214      // to the map and let the next step deal with merging it with
6215      // the range we're adding.
6216      ZoneSplayTree<Config>::Locator loc;
6217      bool inserted = tree()->Insert(right.from(), &loc);
6218      DCHECK(inserted);
6219      USE(inserted);
6220      loc.set_value(Entry(right.from(),
6221                          right.to(),
6222                          entry->out_set()));
6223    }
6224  }
6225  while (current.is_valid()) {
6226    if (tree()->FindLeastGreaterThan(current.from(), &loc) &&
6227        (loc.value().from() <= current.to()) &&
6228        (loc.value().to() >= current.from())) {
6229      Entry* entry = &loc.value();
6230      // We have overlap.  If there is space between the start point of
6231      // the range we're adding and where the overlapping range starts
6232      // then we have to add a range covering just that space.
6233      if (current.from() < entry->from()) {
6234        ZoneSplayTree<Config>::Locator ins;
6235        bool inserted = tree()->Insert(current.from(), &ins);
6236        DCHECK(inserted);
6237        USE(inserted);
6238        ins.set_value(Entry(current.from(),
6239                            entry->from() - 1,
6240                            empty()->Extend(value, zone)));
6241        current.set_from(entry->from());
6242      }
6243      DCHECK_EQ(current.from(), entry->from());
6244      // If the overlapping range extends beyond the one we want to add
6245      // we have to snap the right part off and add it separately.
6246      if (entry->to() > current.to()) {
6247        ZoneSplayTree<Config>::Locator ins;
6248        bool inserted = tree()->Insert(current.to() + 1, &ins);
6249        DCHECK(inserted);
6250        USE(inserted);
6251        ins.set_value(Entry(current.to() + 1,
6252                            entry->to(),
6253                            entry->out_set()));
6254        entry->set_to(current.to());
6255      }
6256      DCHECK(entry->to() <= current.to());
6257      // The overlapping range is now completely contained by the range
6258      // we're adding so we can just update it and move the start point
6259      // of the range we're adding just past it.
6260      entry->AddValue(value, zone);
6261      DCHECK(entry->to() + 1 > current.from());
6262      current.set_from(entry->to() + 1);
6263    } else {
6264      // There is no overlap so we can just add the range
6265      ZoneSplayTree<Config>::Locator ins;
6266      bool inserted = tree()->Insert(current.from(), &ins);
6267      DCHECK(inserted);
6268      USE(inserted);
6269      ins.set_value(Entry(current.from(),
6270                          current.to(),
6271                          empty()->Extend(value, zone)));
6272      break;
6273    }
6274  }
6275}
6276
6277
6278OutSet* DispatchTable::Get(uc32 value) {
6279  ZoneSplayTree<Config>::Locator loc;
6280  if (!tree()->FindGreatestLessThan(value, &loc))
6281    return empty();
6282  Entry* entry = &loc.value();
6283  if (value <= entry->to())
6284    return entry->out_set();
6285  else
6286    return empty();
6287}
6288
6289
6290// -------------------------------------------------------------------
6291// Analysis
6292
6293
6294void Analysis::EnsureAnalyzed(RegExpNode* that) {
6295  StackLimitCheck check(isolate());
6296  if (check.HasOverflowed()) {
6297    fail("Stack overflow");
6298    return;
6299  }
6300  if (that->info()->been_analyzed || that->info()->being_analyzed)
6301    return;
6302  that->info()->being_analyzed = true;
6303  that->Accept(this);
6304  that->info()->being_analyzed = false;
6305  that->info()->been_analyzed = true;
6306}
6307
6308
6309void Analysis::VisitEnd(EndNode* that) {
6310  // nothing to do
6311}
6312
6313
6314void TextNode::CalculateOffsets() {
6315  int element_count = elements()->length();
6316  // Set up the offsets of the elements relative to the start.  This is a fixed
6317  // quantity since a TextNode can only contain fixed-width things.
6318  int cp_offset = 0;
6319  for (int i = 0; i < element_count; i++) {
6320    TextElement& elm = elements()->at(i);
6321    elm.set_cp_offset(cp_offset);
6322    cp_offset += elm.length();
6323  }
6324}
6325
6326
6327void Analysis::VisitText(TextNode* that) {
6328  if (ignore_case()) {
6329    that->MakeCaseIndependent(isolate(), is_one_byte_);
6330  }
6331  EnsureAnalyzed(that->on_success());
6332  if (!has_failed()) {
6333    that->CalculateOffsets();
6334  }
6335}
6336
6337
6338void Analysis::VisitAction(ActionNode* that) {
6339  RegExpNode* target = that->on_success();
6340  EnsureAnalyzed(target);
6341  if (!has_failed()) {
6342    // If the next node is interested in what it follows then this node
6343    // has to be interested too so it can pass the information on.
6344    that->info()->AddFromFollowing(target->info());
6345  }
6346}
6347
6348
6349void Analysis::VisitChoice(ChoiceNode* that) {
6350  NodeInfo* info = that->info();
6351  for (int i = 0; i < that->alternatives()->length(); i++) {
6352    RegExpNode* node = that->alternatives()->at(i).node();
6353    EnsureAnalyzed(node);
6354    if (has_failed()) return;
6355    // Anything the following nodes need to know has to be known by
6356    // this node also, so it can pass it on.
6357    info->AddFromFollowing(node->info());
6358  }
6359}
6360
6361
6362void Analysis::VisitLoopChoice(LoopChoiceNode* that) {
6363  NodeInfo* info = that->info();
6364  for (int i = 0; i < that->alternatives()->length(); i++) {
6365    RegExpNode* node = that->alternatives()->at(i).node();
6366    if (node != that->loop_node()) {
6367      EnsureAnalyzed(node);
6368      if (has_failed()) return;
6369      info->AddFromFollowing(node->info());
6370    }
6371  }
6372  // Check the loop last since it may need the value of this node
6373  // to get a correct result.
6374  EnsureAnalyzed(that->loop_node());
6375  if (!has_failed()) {
6376    info->AddFromFollowing(that->loop_node()->info());
6377  }
6378}
6379
6380
6381void Analysis::VisitBackReference(BackReferenceNode* that) {
6382  EnsureAnalyzed(that->on_success());
6383}
6384
6385
6386void Analysis::VisitAssertion(AssertionNode* that) {
6387  EnsureAnalyzed(that->on_success());
6388}
6389
6390
6391void BackReferenceNode::FillInBMInfo(Isolate* isolate, int offset, int budget,
6392                                     BoyerMooreLookahead* bm,
6393                                     bool not_at_start) {
6394  // Working out the set of characters that a backreference can match is too
6395  // hard, so we just say that any character can match.
6396  bm->SetRest(offset);
6397  SaveBMInfo(bm, not_at_start, offset);
6398}
6399
6400
6401STATIC_ASSERT(BoyerMoorePositionInfo::kMapSize ==
6402              RegExpMacroAssembler::kTableSize);
6403
6404
6405void ChoiceNode::FillInBMInfo(Isolate* isolate, int offset, int budget,
6406                              BoyerMooreLookahead* bm, bool not_at_start) {
6407  ZoneList<GuardedAlternative>* alts = alternatives();
6408  budget = (budget - 1) / alts->length();
6409  for (int i = 0; i < alts->length(); i++) {
6410    GuardedAlternative& alt = alts->at(i);
6411    if (alt.guards() != NULL && alt.guards()->length() != 0) {
6412      bm->SetRest(offset);  // Give up trying to fill in info.
6413      SaveBMInfo(bm, not_at_start, offset);
6414      return;
6415    }
6416    alt.node()->FillInBMInfo(isolate, offset, budget, bm, not_at_start);
6417  }
6418  SaveBMInfo(bm, not_at_start, offset);
6419}
6420
6421
6422void TextNode::FillInBMInfo(Isolate* isolate, int initial_offset, int budget,
6423                            BoyerMooreLookahead* bm, bool not_at_start) {
6424  if (initial_offset >= bm->length()) return;
6425  int offset = initial_offset;
6426  int max_char = bm->max_char();
6427  for (int i = 0; i < elements()->length(); i++) {
6428    if (offset >= bm->length()) {
6429      if (initial_offset == 0) set_bm_info(not_at_start, bm);
6430      return;
6431    }
6432    TextElement text = elements()->at(i);
6433    if (text.text_type() == TextElement::ATOM) {
6434      RegExpAtom* atom = text.atom();
6435      for (int j = 0; j < atom->length(); j++, offset++) {
6436        if (offset >= bm->length()) {
6437          if (initial_offset == 0) set_bm_info(not_at_start, bm);
6438          return;
6439        }
6440        uc16 character = atom->data()[j];
6441        if (bm->compiler()->ignore_case()) {
6442          unibrow::uchar chars[unibrow::Ecma262UnCanonicalize::kMaxWidth];
6443          int length = GetCaseIndependentLetters(
6444              isolate, character, bm->max_char() == String::kMaxOneByteCharCode,
6445              chars);
6446          for (int j = 0; j < length; j++) {
6447            bm->Set(offset, chars[j]);
6448          }
6449        } else {
6450          if (character <= max_char) bm->Set(offset, character);
6451        }
6452      }
6453    } else {
6454      DCHECK_EQ(TextElement::CHAR_CLASS, text.text_type());
6455      RegExpCharacterClass* char_class = text.char_class();
6456      ZoneList<CharacterRange>* ranges = char_class->ranges(zone());
6457      if (char_class->is_negated()) {
6458        bm->SetAll(offset);
6459      } else {
6460        for (int k = 0; k < ranges->length(); k++) {
6461          CharacterRange& range = ranges->at(k);
6462          if (range.from() > max_char) continue;
6463          int to = Min(max_char, static_cast<int>(range.to()));
6464          bm->SetInterval(offset, Interval(range.from(), to));
6465        }
6466      }
6467      offset++;
6468    }
6469  }
6470  if (offset >= bm->length()) {
6471    if (initial_offset == 0) set_bm_info(not_at_start, bm);
6472    return;
6473  }
6474  on_success()->FillInBMInfo(isolate, offset, budget - 1, bm,
6475                             true);  // Not at start after a text node.
6476  if (initial_offset == 0) set_bm_info(not_at_start, bm);
6477}
6478
6479
6480// -------------------------------------------------------------------
6481// Dispatch table construction
6482
6483
6484void DispatchTableConstructor::VisitEnd(EndNode* that) {
6485  AddRange(CharacterRange::Everything());
6486}
6487
6488
6489void DispatchTableConstructor::BuildTable(ChoiceNode* node) {
6490  node->set_being_calculated(true);
6491  ZoneList<GuardedAlternative>* alternatives = node->alternatives();
6492  for (int i = 0; i < alternatives->length(); i++) {
6493    set_choice_index(i);
6494    alternatives->at(i).node()->Accept(this);
6495  }
6496  node->set_being_calculated(false);
6497}
6498
6499
6500class AddDispatchRange {
6501 public:
6502  explicit AddDispatchRange(DispatchTableConstructor* constructor)
6503    : constructor_(constructor) { }
6504  void Call(uc32 from, DispatchTable::Entry entry);
6505 private:
6506  DispatchTableConstructor* constructor_;
6507};
6508
6509
6510void AddDispatchRange::Call(uc32 from, DispatchTable::Entry entry) {
6511  constructor_->AddRange(CharacterRange::Range(from, entry.to()));
6512}
6513
6514
6515void DispatchTableConstructor::VisitChoice(ChoiceNode* node) {
6516  if (node->being_calculated())
6517    return;
6518  DispatchTable* table = node->GetTable(ignore_case_);
6519  AddDispatchRange adder(this);
6520  table->ForEach(&adder);
6521}
6522
6523
6524void DispatchTableConstructor::VisitBackReference(BackReferenceNode* that) {
6525  // TODO(160): Find the node that we refer back to and propagate its start
6526  // set back to here.  For now we just accept anything.
6527  AddRange(CharacterRange::Everything());
6528}
6529
6530
6531void DispatchTableConstructor::VisitAssertion(AssertionNode* that) {
6532  RegExpNode* target = that->on_success();
6533  target->Accept(this);
6534}
6535
6536
6537static int CompareRangeByFrom(const CharacterRange* a,
6538                              const CharacterRange* b) {
6539  return Compare<uc16>(a->from(), b->from());
6540}
6541
6542
6543void DispatchTableConstructor::AddInverse(ZoneList<CharacterRange>* ranges) {
6544  ranges->Sort(CompareRangeByFrom);
6545  uc16 last = 0;
6546  for (int i = 0; i < ranges->length(); i++) {
6547    CharacterRange range = ranges->at(i);
6548    if (last < range.from())
6549      AddRange(CharacterRange::Range(last, range.from() - 1));
6550    if (range.to() >= last) {
6551      if (range.to() == String::kMaxCodePoint) {
6552        return;
6553      } else {
6554        last = range.to() + 1;
6555      }
6556    }
6557  }
6558  AddRange(CharacterRange::Range(last, String::kMaxCodePoint));
6559}
6560
6561
6562void DispatchTableConstructor::VisitText(TextNode* that) {
6563  TextElement elm = that->elements()->at(0);
6564  switch (elm.text_type()) {
6565    case TextElement::ATOM: {
6566      uc16 c = elm.atom()->data()[0];
6567      AddRange(CharacterRange::Range(c, c));
6568      break;
6569    }
6570    case TextElement::CHAR_CLASS: {
6571      RegExpCharacterClass* tree = elm.char_class();
6572      ZoneList<CharacterRange>* ranges = tree->ranges(that->zone());
6573      if (tree->is_negated()) {
6574        AddInverse(ranges);
6575      } else {
6576        for (int i = 0; i < ranges->length(); i++)
6577          AddRange(ranges->at(i));
6578      }
6579      break;
6580    }
6581    default: {
6582      UNIMPLEMENTED();
6583    }
6584  }
6585}
6586
6587
6588void DispatchTableConstructor::VisitAction(ActionNode* that) {
6589  RegExpNode* target = that->on_success();
6590  target->Accept(this);
6591}
6592
6593
6594RegExpNode* OptionallyStepBackToLeadSurrogate(RegExpCompiler* compiler,
6595                                              RegExpNode* on_success) {
6596  // If the regexp matching starts within a surrogate pair, step back
6597  // to the lead surrogate and start matching from there.
6598  DCHECK(!compiler->read_backward());
6599  Zone* zone = compiler->zone();
6600  ZoneList<CharacterRange>* lead_surrogates = CharacterRange::List(
6601      zone, CharacterRange::Range(kLeadSurrogateStart, kLeadSurrogateEnd));
6602  ZoneList<CharacterRange>* trail_surrogates = CharacterRange::List(
6603      zone, CharacterRange::Range(kTrailSurrogateStart, kTrailSurrogateEnd));
6604
6605  ChoiceNode* optional_step_back = new (zone) ChoiceNode(2, zone);
6606
6607  int stack_register = compiler->UnicodeLookaroundStackRegister();
6608  int position_register = compiler->UnicodeLookaroundPositionRegister();
6609  RegExpNode* step_back = TextNode::CreateForCharacterRanges(
6610      zone, lead_surrogates, true, on_success);
6611  RegExpLookaround::Builder builder(true, step_back, stack_register,
6612                                    position_register);
6613  RegExpNode* match_trail = TextNode::CreateForCharacterRanges(
6614      zone, trail_surrogates, false, builder.on_match_success());
6615
6616  optional_step_back->AddAlternative(
6617      GuardedAlternative(builder.ForMatch(match_trail)));
6618  optional_step_back->AddAlternative(GuardedAlternative(on_success));
6619
6620  return optional_step_back;
6621}
6622
6623
6624RegExpEngine::CompilationResult RegExpEngine::Compile(
6625    Isolate* isolate, Zone* zone, RegExpCompileData* data,
6626    JSRegExp::Flags flags, Handle<String> pattern,
6627    Handle<String> sample_subject, bool is_one_byte) {
6628  if ((data->capture_count + 1) * 2 - 1 > RegExpMacroAssembler::kMaxRegister) {
6629    return IrregexpRegExpTooBig(isolate);
6630  }
6631  bool ignore_case = flags & JSRegExp::kIgnoreCase;
6632  bool is_sticky = flags & JSRegExp::kSticky;
6633  bool is_global = flags & JSRegExp::kGlobal;
6634  bool is_unicode = flags & JSRegExp::kUnicode;
6635  RegExpCompiler compiler(isolate, zone, data->capture_count, flags,
6636                          is_one_byte);
6637
6638  if (compiler.optimize()) compiler.set_optimize(!TooMuchRegExpCode(pattern));
6639
6640  // Sample some characters from the middle of the string.
6641  static const int kSampleSize = 128;
6642
6643  sample_subject = String::Flatten(sample_subject);
6644  int chars_sampled = 0;
6645  int half_way = (sample_subject->length() - kSampleSize) / 2;
6646  for (int i = Max(0, half_way);
6647       i < sample_subject->length() && chars_sampled < kSampleSize;
6648       i++, chars_sampled++) {
6649    compiler.frequency_collator()->CountCharacter(sample_subject->Get(i));
6650  }
6651
6652  // Wrap the body of the regexp in capture #0.
6653  RegExpNode* captured_body = RegExpCapture::ToNode(data->tree,
6654                                                    0,
6655                                                    &compiler,
6656                                                    compiler.accept());
6657  RegExpNode* node = captured_body;
6658  bool is_end_anchored = data->tree->IsAnchoredAtEnd();
6659  bool is_start_anchored = data->tree->IsAnchoredAtStart();
6660  int max_length = data->tree->max_match();
6661  if (!is_start_anchored && !is_sticky) {
6662    // Add a .*? at the beginning, outside the body capture, unless
6663    // this expression is anchored at the beginning or sticky.
6664    RegExpNode* loop_node = RegExpQuantifier::ToNode(
6665        0, RegExpTree::kInfinity, false, new (zone) RegExpCharacterClass('*'),
6666        &compiler, captured_body, data->contains_anchor);
6667
6668    if (data->contains_anchor) {
6669      // Unroll loop once, to take care of the case that might start
6670      // at the start of input.
6671      ChoiceNode* first_step_node = new(zone) ChoiceNode(2, zone);
6672      first_step_node->AddAlternative(GuardedAlternative(captured_body));
6673      first_step_node->AddAlternative(GuardedAlternative(new (zone) TextNode(
6674          new (zone) RegExpCharacterClass('*'), false, loop_node)));
6675      node = first_step_node;
6676    } else {
6677      node = loop_node;
6678    }
6679  }
6680  if (is_one_byte) {
6681    node = node->FilterOneByte(RegExpCompiler::kMaxRecursion, ignore_case);
6682    // Do it again to propagate the new nodes to places where they were not
6683    // put because they had not been calculated yet.
6684    if (node != NULL) {
6685      node = node->FilterOneByte(RegExpCompiler::kMaxRecursion, ignore_case);
6686    }
6687  } else if (compiler.unicode() && (is_global || is_sticky)) {
6688    node = OptionallyStepBackToLeadSurrogate(&compiler, node);
6689  }
6690
6691  if (node == NULL) node = new(zone) EndNode(EndNode::BACKTRACK, zone);
6692  data->node = node;
6693  Analysis analysis(isolate, flags, is_one_byte);
6694  analysis.EnsureAnalyzed(node);
6695  if (analysis.has_failed()) {
6696    const char* error_message = analysis.error_message();
6697    return CompilationResult(isolate, error_message);
6698  }
6699
6700  // Create the correct assembler for the architecture.
6701#ifndef V8_INTERPRETED_REGEXP
6702  // Native regexp implementation.
6703
6704  NativeRegExpMacroAssembler::Mode mode =
6705      is_one_byte ? NativeRegExpMacroAssembler::LATIN1
6706                  : NativeRegExpMacroAssembler::UC16;
6707
6708#if V8_TARGET_ARCH_IA32
6709  RegExpMacroAssemblerIA32 macro_assembler(isolate, zone, mode,
6710                                           (data->capture_count + 1) * 2);
6711#elif V8_TARGET_ARCH_X64
6712  RegExpMacroAssemblerX64 macro_assembler(isolate, zone, mode,
6713                                          (data->capture_count + 1) * 2);
6714#elif V8_TARGET_ARCH_ARM
6715  RegExpMacroAssemblerARM macro_assembler(isolate, zone, mode,
6716                                          (data->capture_count + 1) * 2);
6717#elif V8_TARGET_ARCH_ARM64
6718  RegExpMacroAssemblerARM64 macro_assembler(isolate, zone, mode,
6719                                            (data->capture_count + 1) * 2);
6720#elif V8_TARGET_ARCH_S390
6721  RegExpMacroAssemblerS390 macro_assembler(isolate, zone, mode,
6722                                           (data->capture_count + 1) * 2);
6723#elif V8_TARGET_ARCH_PPC
6724  RegExpMacroAssemblerPPC macro_assembler(isolate, zone, mode,
6725                                          (data->capture_count + 1) * 2);
6726#elif V8_TARGET_ARCH_MIPS
6727  RegExpMacroAssemblerMIPS macro_assembler(isolate, zone, mode,
6728                                           (data->capture_count + 1) * 2);
6729#elif V8_TARGET_ARCH_MIPS64
6730  RegExpMacroAssemblerMIPS macro_assembler(isolate, zone, mode,
6731                                           (data->capture_count + 1) * 2);
6732#elif V8_TARGET_ARCH_X87
6733  RegExpMacroAssemblerX87 macro_assembler(isolate, zone, mode,
6734                                          (data->capture_count + 1) * 2);
6735#else
6736#error "Unsupported architecture"
6737#endif
6738
6739#else  // V8_INTERPRETED_REGEXP
6740  // Interpreted regexp implementation.
6741  EmbeddedVector<byte, 1024> codes;
6742  RegExpMacroAssemblerIrregexp macro_assembler(isolate, codes, zone);
6743#endif  // V8_INTERPRETED_REGEXP
6744
6745  macro_assembler.set_slow_safe(TooMuchRegExpCode(pattern));
6746
6747  // Inserted here, instead of in Assembler, because it depends on information
6748  // in the AST that isn't replicated in the Node structure.
6749  static const int kMaxBacksearchLimit = 1024;
6750  if (is_end_anchored &&
6751      !is_start_anchored &&
6752      max_length < kMaxBacksearchLimit) {
6753    macro_assembler.SetCurrentPositionFromEnd(max_length);
6754  }
6755
6756  if (is_global) {
6757    RegExpMacroAssembler::GlobalMode mode = RegExpMacroAssembler::GLOBAL;
6758    if (data->tree->min_match() > 0) {
6759      mode = RegExpMacroAssembler::GLOBAL_NO_ZERO_LENGTH_CHECK;
6760    } else if (is_unicode) {
6761      mode = RegExpMacroAssembler::GLOBAL_UNICODE;
6762    }
6763    macro_assembler.set_global_mode(mode);
6764  }
6765
6766  return compiler.Assemble(&macro_assembler,
6767                           node,
6768                           data->capture_count,
6769                           pattern);
6770}
6771
6772
6773bool RegExpEngine::TooMuchRegExpCode(Handle<String> pattern) {
6774  Heap* heap = pattern->GetHeap();
6775  bool too_much = pattern->length() > RegExpImpl::kRegExpTooLargeToOptimize;
6776  if (heap->total_regexp_code_generated() > RegExpImpl::kRegExpCompiledLimit &&
6777      heap->memory_allocator()->SizeExecutable() >
6778          RegExpImpl::kRegExpExecutableMemoryLimit) {
6779    too_much = true;
6780  }
6781  return too_much;
6782}
6783
6784
6785Object* RegExpResultsCache::Lookup(Heap* heap, String* key_string,
6786                                   Object* key_pattern,
6787                                   FixedArray** last_match_cache,
6788                                   ResultsCacheType type) {
6789  FixedArray* cache;
6790  if (!key_string->IsInternalizedString()) return Smi::FromInt(0);
6791  if (type == STRING_SPLIT_SUBSTRINGS) {
6792    DCHECK(key_pattern->IsString());
6793    if (!key_pattern->IsInternalizedString()) return Smi::FromInt(0);
6794    cache = heap->string_split_cache();
6795  } else {
6796    DCHECK(type == REGEXP_MULTIPLE_INDICES);
6797    DCHECK(key_pattern->IsFixedArray());
6798    cache = heap->regexp_multiple_cache();
6799  }
6800
6801  uint32_t hash = key_string->Hash();
6802  uint32_t index = ((hash & (kRegExpResultsCacheSize - 1)) &
6803                    ~(kArrayEntriesPerCacheEntry - 1));
6804  if (cache->get(index + kStringOffset) != key_string ||
6805      cache->get(index + kPatternOffset) != key_pattern) {
6806    index =
6807        ((index + kArrayEntriesPerCacheEntry) & (kRegExpResultsCacheSize - 1));
6808    if (cache->get(index + kStringOffset) != key_string ||
6809        cache->get(index + kPatternOffset) != key_pattern) {
6810      return Smi::FromInt(0);
6811    }
6812  }
6813
6814  *last_match_cache = FixedArray::cast(cache->get(index + kLastMatchOffset));
6815  return cache->get(index + kArrayOffset);
6816}
6817
6818
6819void RegExpResultsCache::Enter(Isolate* isolate, Handle<String> key_string,
6820                               Handle<Object> key_pattern,
6821                               Handle<FixedArray> value_array,
6822                               Handle<FixedArray> last_match_cache,
6823                               ResultsCacheType type) {
6824  Factory* factory = isolate->factory();
6825  Handle<FixedArray> cache;
6826  if (!key_string->IsInternalizedString()) return;
6827  if (type == STRING_SPLIT_SUBSTRINGS) {
6828    DCHECK(key_pattern->IsString());
6829    if (!key_pattern->IsInternalizedString()) return;
6830    cache = factory->string_split_cache();
6831  } else {
6832    DCHECK(type == REGEXP_MULTIPLE_INDICES);
6833    DCHECK(key_pattern->IsFixedArray());
6834    cache = factory->regexp_multiple_cache();
6835  }
6836
6837  uint32_t hash = key_string->Hash();
6838  uint32_t index = ((hash & (kRegExpResultsCacheSize - 1)) &
6839                    ~(kArrayEntriesPerCacheEntry - 1));
6840  if (cache->get(index + kStringOffset) == Smi::FromInt(0)) {
6841    cache->set(index + kStringOffset, *key_string);
6842    cache->set(index + kPatternOffset, *key_pattern);
6843    cache->set(index + kArrayOffset, *value_array);
6844    cache->set(index + kLastMatchOffset, *last_match_cache);
6845  } else {
6846    uint32_t index2 =
6847        ((index + kArrayEntriesPerCacheEntry) & (kRegExpResultsCacheSize - 1));
6848    if (cache->get(index2 + kStringOffset) == Smi::FromInt(0)) {
6849      cache->set(index2 + kStringOffset, *key_string);
6850      cache->set(index2 + kPatternOffset, *key_pattern);
6851      cache->set(index2 + kArrayOffset, *value_array);
6852      cache->set(index2 + kLastMatchOffset, *last_match_cache);
6853    } else {
6854      cache->set(index2 + kStringOffset, Smi::FromInt(0));
6855      cache->set(index2 + kPatternOffset, Smi::FromInt(0));
6856      cache->set(index2 + kArrayOffset, Smi::FromInt(0));
6857      cache->set(index2 + kLastMatchOffset, Smi::FromInt(0));
6858      cache->set(index + kStringOffset, *key_string);
6859      cache->set(index + kPatternOffset, *key_pattern);
6860      cache->set(index + kArrayOffset, *value_array);
6861      cache->set(index + kLastMatchOffset, *last_match_cache);
6862    }
6863  }
6864  // If the array is a reasonably short list of substrings, convert it into a
6865  // list of internalized strings.
6866  if (type == STRING_SPLIT_SUBSTRINGS && value_array->length() < 100) {
6867    for (int i = 0; i < value_array->length(); i++) {
6868      Handle<String> str(String::cast(value_array->get(i)), isolate);
6869      Handle<String> internalized_str = factory->InternalizeString(str);
6870      value_array->set(i, *internalized_str);
6871    }
6872  }
6873  // Convert backing store to a copy-on-write array.
6874  value_array->set_map_no_write_barrier(*factory->fixed_cow_array_map());
6875}
6876
6877
6878void RegExpResultsCache::Clear(FixedArray* cache) {
6879  for (int i = 0; i < kRegExpResultsCacheSize; i++) {
6880    cache->set(i, Smi::FromInt(0));
6881  }
6882}
6883
6884}  // namespace internal
6885}  // namespace v8
6886