• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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/v8.h"
6 
7 #include "src/ast.h"
8 #include "src/compiler.h"
9 #include "src/execution.h"
10 #include "src/factory.h"
11 #include "src/jsregexp.h"
12 #include "src/jsregexp-inl.h"
13 #include "src/platform.h"
14 #include "src/string-search.h"
15 #include "src/runtime.h"
16 #include "src/compilation-cache.h"
17 #include "src/string-stream.h"
18 #include "src/parser.h"
19 #include "src/regexp-macro-assembler.h"
20 #include "src/regexp-macro-assembler-tracer.h"
21 #include "src/regexp-macro-assembler-irregexp.h"
22 #include "src/regexp-stack.h"
23 
24 #ifndef V8_INTERPRETED_REGEXP
25 #if V8_TARGET_ARCH_IA32
26 #include "src/ia32/regexp-macro-assembler-ia32.h"
27 #elif V8_TARGET_ARCH_X64
28 #include "src/x64/regexp-macro-assembler-x64.h"
29 #elif V8_TARGET_ARCH_ARM64
30 #include "src/arm64/regexp-macro-assembler-arm64.h"
31 #elif V8_TARGET_ARCH_ARM
32 #include "src/arm/regexp-macro-assembler-arm.h"
33 #elif V8_TARGET_ARCH_MIPS
34 #include "src/mips/regexp-macro-assembler-mips.h"
35 #elif V8_TARGET_ARCH_X87
36 #include "src/x87/regexp-macro-assembler-x87.h"
37 #else
38 #error Unsupported target architecture.
39 #endif
40 #endif
41 
42 #include "src/interpreter-irregexp.h"
43 
44 
45 namespace v8 {
46 namespace internal {
47 
CreateRegExpLiteral(Handle<JSFunction> constructor,Handle<String> pattern,Handle<String> flags)48 MaybeHandle<Object> RegExpImpl::CreateRegExpLiteral(
49     Handle<JSFunction> constructor,
50     Handle<String> pattern,
51     Handle<String> flags) {
52   // Call the construct code with 2 arguments.
53   Handle<Object> argv[] = { pattern, flags };
54   return Execution::New(constructor, ARRAY_SIZE(argv), argv);
55 }
56 
57 
RegExpFlagsFromString(Handle<String> str)58 static JSRegExp::Flags RegExpFlagsFromString(Handle<String> str) {
59   int flags = JSRegExp::NONE;
60   for (int i = 0; i < str->length(); i++) {
61     switch (str->Get(i)) {
62       case 'i':
63         flags |= JSRegExp::IGNORE_CASE;
64         break;
65       case 'g':
66         flags |= JSRegExp::GLOBAL;
67         break;
68       case 'm':
69         flags |= JSRegExp::MULTILINE;
70         break;
71     }
72   }
73   return JSRegExp::Flags(flags);
74 }
75 
76 
77 MUST_USE_RESULT
ThrowRegExpException(Handle<JSRegExp> re,Handle<String> pattern,Handle<String> error_text,const char * message)78 static inline MaybeHandle<Object> ThrowRegExpException(
79     Handle<JSRegExp> re,
80     Handle<String> pattern,
81     Handle<String> error_text,
82     const char* message) {
83   Isolate* isolate = re->GetIsolate();
84   Factory* factory = isolate->factory();
85   Handle<FixedArray> elements = factory->NewFixedArray(2);
86   elements->set(0, *pattern);
87   elements->set(1, *error_text);
88   Handle<JSArray> array = factory->NewJSArrayWithElements(elements);
89   Handle<Object> regexp_err = factory->NewSyntaxError(message, array);
90   return isolate->Throw<Object>(regexp_err);
91 }
92 
93 
AddRange(ContainedInLattice containment,const int * ranges,int ranges_length,Interval new_range)94 ContainedInLattice AddRange(ContainedInLattice containment,
95                             const int* ranges,
96                             int ranges_length,
97                             Interval new_range) {
98   ASSERT((ranges_length & 1) == 1);
99   ASSERT(ranges[ranges_length - 1] == String::kMaxUtf16CodeUnit + 1);
100   if (containment == kLatticeUnknown) return containment;
101   bool inside = false;
102   int last = 0;
103   for (int i = 0; i < ranges_length; inside = !inside, last = ranges[i], i++) {
104     // Consider the range from last to ranges[i].
105     // We haven't got to the new range yet.
106     if (ranges[i] <= new_range.from()) continue;
107     // New range is wholly inside last-ranges[i].  Note that new_range.to() is
108     // inclusive, but the values in ranges are not.
109     if (last <= new_range.from() && new_range.to() < ranges[i]) {
110       return Combine(containment, inside ? kLatticeIn : kLatticeOut);
111     }
112     return kLatticeUnknown;
113   }
114   return containment;
115 }
116 
117 
118 // More makes code generation slower, less makes V8 benchmark score lower.
119 const int kMaxLookaheadForBoyerMoore = 8;
120 // In a 3-character pattern you can maximally step forwards 3 characters
121 // at a time, which is not always enough to pay for the extra logic.
122 const int kPatternTooShortForBoyerMoore = 2;
123 
124 
125 // Identifies the sort of regexps where the regexp engine is faster
126 // than the code used for atom matches.
HasFewDifferentCharacters(Handle<String> pattern)127 static bool HasFewDifferentCharacters(Handle<String> pattern) {
128   int length = Min(kMaxLookaheadForBoyerMoore, pattern->length());
129   if (length <= kPatternTooShortForBoyerMoore) return false;
130   const int kMod = 128;
131   bool character_found[kMod];
132   int different = 0;
133   memset(&character_found[0], 0, sizeof(character_found));
134   for (int i = 0; i < length; i++) {
135     int ch = (pattern->Get(i) & (kMod - 1));
136     if (!character_found[ch]) {
137       character_found[ch] = true;
138       different++;
139       // We declare a regexp low-alphabet if it has at least 3 times as many
140       // characters as it has different characters.
141       if (different * 3 > length) return false;
142     }
143   }
144   return true;
145 }
146 
147 
148 // Generic RegExp methods. Dispatches to implementation specific methods.
149 
150 
Compile(Handle<JSRegExp> re,Handle<String> pattern,Handle<String> flag_str)151 MaybeHandle<Object> RegExpImpl::Compile(Handle<JSRegExp> re,
152                                         Handle<String> pattern,
153                                         Handle<String> flag_str) {
154   Isolate* isolate = re->GetIsolate();
155   Zone zone(isolate);
156   JSRegExp::Flags flags = RegExpFlagsFromString(flag_str);
157   CompilationCache* compilation_cache = isolate->compilation_cache();
158   MaybeHandle<FixedArray> maybe_cached =
159       compilation_cache->LookupRegExp(pattern, flags);
160   Handle<FixedArray> cached;
161   bool in_cache = maybe_cached.ToHandle(&cached);
162   LOG(isolate, RegExpCompileEvent(re, in_cache));
163 
164   Handle<Object> result;
165   if (in_cache) {
166     re->set_data(*cached);
167     return re;
168   }
169   pattern = String::Flatten(pattern);
170   PostponeInterruptsScope postpone(isolate);
171   RegExpCompileData parse_result;
172   FlatStringReader reader(isolate, pattern);
173   if (!RegExpParser::ParseRegExp(&reader, flags.is_multiline(),
174                                  &parse_result, &zone)) {
175     // Throw an exception if we fail to parse the pattern.
176     return ThrowRegExpException(re,
177                                 pattern,
178                                 parse_result.error,
179                                 "malformed_regexp");
180   }
181 
182   bool has_been_compiled = false;
183 
184   if (parse_result.simple &&
185       !flags.is_ignore_case() &&
186       !HasFewDifferentCharacters(pattern)) {
187     // Parse-tree is a single atom that is equal to the pattern.
188     AtomCompile(re, pattern, flags, pattern);
189     has_been_compiled = true;
190   } else if (parse_result.tree->IsAtom() &&
191       !flags.is_ignore_case() &&
192       parse_result.capture_count == 0) {
193     RegExpAtom* atom = parse_result.tree->AsAtom();
194     Vector<const uc16> atom_pattern = atom->data();
195     Handle<String> atom_string;
196     ASSIGN_RETURN_ON_EXCEPTION(
197         isolate, atom_string,
198         isolate->factory()->NewStringFromTwoByte(atom_pattern),
199         Object);
200     if (!HasFewDifferentCharacters(atom_string)) {
201       AtomCompile(re, pattern, flags, atom_string);
202       has_been_compiled = true;
203     }
204   }
205   if (!has_been_compiled) {
206     IrregexpInitialize(re, pattern, flags, parse_result.capture_count);
207   }
208   ASSERT(re->data()->IsFixedArray());
209   // Compilation succeeded so the data is set on the regexp
210   // and we can store it in the cache.
211   Handle<FixedArray> data(FixedArray::cast(re->data()));
212   compilation_cache->PutRegExp(pattern, flags, data);
213 
214   return re;
215 }
216 
217 
Exec(Handle<JSRegExp> regexp,Handle<String> subject,int index,Handle<JSArray> last_match_info)218 MaybeHandle<Object> RegExpImpl::Exec(Handle<JSRegExp> regexp,
219                                      Handle<String> subject,
220                                      int index,
221                                      Handle<JSArray> last_match_info) {
222   switch (regexp->TypeTag()) {
223     case JSRegExp::ATOM:
224       return AtomExec(regexp, subject, index, last_match_info);
225     case JSRegExp::IRREGEXP: {
226       return IrregexpExec(regexp, subject, index, last_match_info);
227     }
228     default:
229       UNREACHABLE();
230       return MaybeHandle<Object>();
231   }
232 }
233 
234 
235 // RegExp Atom implementation: Simple string search using indexOf.
236 
237 
AtomCompile(Handle<JSRegExp> re,Handle<String> pattern,JSRegExp::Flags flags,Handle<String> match_pattern)238 void RegExpImpl::AtomCompile(Handle<JSRegExp> re,
239                              Handle<String> pattern,
240                              JSRegExp::Flags flags,
241                              Handle<String> match_pattern) {
242   re->GetIsolate()->factory()->SetRegExpAtomData(re,
243                                                  JSRegExp::ATOM,
244                                                  pattern,
245                                                  flags,
246                                                  match_pattern);
247 }
248 
249 
SetAtomLastCapture(FixedArray * array,String * subject,int from,int to)250 static void SetAtomLastCapture(FixedArray* array,
251                                String* subject,
252                                int from,
253                                int to) {
254   SealHandleScope shs(array->GetIsolate());
255   RegExpImpl::SetLastCaptureCount(array, 2);
256   RegExpImpl::SetLastSubject(array, subject);
257   RegExpImpl::SetLastInput(array, subject);
258   RegExpImpl::SetCapture(array, 0, from);
259   RegExpImpl::SetCapture(array, 1, to);
260 }
261 
262 
AtomExecRaw(Handle<JSRegExp> regexp,Handle<String> subject,int index,int32_t * output,int output_size)263 int RegExpImpl::AtomExecRaw(Handle<JSRegExp> regexp,
264                             Handle<String> subject,
265                             int index,
266                             int32_t* output,
267                             int output_size) {
268   Isolate* isolate = regexp->GetIsolate();
269 
270   ASSERT(0 <= index);
271   ASSERT(index <= subject->length());
272 
273   subject = String::Flatten(subject);
274   DisallowHeapAllocation no_gc;  // ensure vectors stay valid
275 
276   String* needle = String::cast(regexp->DataAt(JSRegExp::kAtomPatternIndex));
277   int needle_len = needle->length();
278   ASSERT(needle->IsFlat());
279   ASSERT_LT(0, needle_len);
280 
281   if (index + needle_len > subject->length()) {
282     return RegExpImpl::RE_FAILURE;
283   }
284 
285   for (int i = 0; i < output_size; i += 2) {
286     String::FlatContent needle_content = needle->GetFlatContent();
287     String::FlatContent subject_content = subject->GetFlatContent();
288     ASSERT(needle_content.IsFlat());
289     ASSERT(subject_content.IsFlat());
290     // dispatch on type of strings
291     index = (needle_content.IsAscii()
292              ? (subject_content.IsAscii()
293                 ? SearchString(isolate,
294                                subject_content.ToOneByteVector(),
295                                needle_content.ToOneByteVector(),
296                                index)
297                 : SearchString(isolate,
298                                subject_content.ToUC16Vector(),
299                                needle_content.ToOneByteVector(),
300                                index))
301              : (subject_content.IsAscii()
302                 ? SearchString(isolate,
303                                subject_content.ToOneByteVector(),
304                                needle_content.ToUC16Vector(),
305                                index)
306                 : SearchString(isolate,
307                                subject_content.ToUC16Vector(),
308                                needle_content.ToUC16Vector(),
309                                index)));
310     if (index == -1) {
311       return i / 2;  // Return number of matches.
312     } else {
313       output[i] = index;
314       output[i+1] = index + needle_len;
315       index += needle_len;
316     }
317   }
318   return output_size / 2;
319 }
320 
321 
AtomExec(Handle<JSRegExp> re,Handle<String> subject,int index,Handle<JSArray> last_match_info)322 Handle<Object> RegExpImpl::AtomExec(Handle<JSRegExp> re,
323                                     Handle<String> subject,
324                                     int index,
325                                     Handle<JSArray> last_match_info) {
326   Isolate* isolate = re->GetIsolate();
327 
328   static const int kNumRegisters = 2;
329   STATIC_ASSERT(kNumRegisters <= Isolate::kJSRegexpStaticOffsetsVectorSize);
330   int32_t* output_registers = isolate->jsregexp_static_offsets_vector();
331 
332   int res = AtomExecRaw(re, subject, index, output_registers, kNumRegisters);
333 
334   if (res == RegExpImpl::RE_FAILURE) return isolate->factory()->null_value();
335 
336   ASSERT_EQ(res, RegExpImpl::RE_SUCCESS);
337   SealHandleScope shs(isolate);
338   FixedArray* array = FixedArray::cast(last_match_info->elements());
339   SetAtomLastCapture(array, *subject, output_registers[0], output_registers[1]);
340   return last_match_info;
341 }
342 
343 
344 // Irregexp implementation.
345 
346 // Ensures that the regexp object contains a compiled version of the
347 // source for either ASCII or non-ASCII strings.
348 // If the compiled version doesn't already exist, it is compiled
349 // from the source pattern.
350 // If compilation fails, an exception is thrown and this function
351 // returns false.
EnsureCompiledIrregexp(Handle<JSRegExp> re,Handle<String> sample_subject,bool is_ascii)352 bool RegExpImpl::EnsureCompiledIrregexp(
353     Handle<JSRegExp> re, Handle<String> sample_subject, bool is_ascii) {
354   Object* compiled_code = re->DataAt(JSRegExp::code_index(is_ascii));
355 #ifdef V8_INTERPRETED_REGEXP
356   if (compiled_code->IsByteArray()) return true;
357 #else  // V8_INTERPRETED_REGEXP (RegExp native code)
358   if (compiled_code->IsCode()) return true;
359 #endif
360   // We could potentially have marked this as flushable, but have kept
361   // a saved version if we did not flush it yet.
362   Object* saved_code = re->DataAt(JSRegExp::saved_code_index(is_ascii));
363   if (saved_code->IsCode()) {
364     // Reinstate the code in the original place.
365     re->SetDataAt(JSRegExp::code_index(is_ascii), saved_code);
366     ASSERT(compiled_code->IsSmi());
367     return true;
368   }
369   return CompileIrregexp(re, sample_subject, is_ascii);
370 }
371 
372 
CreateRegExpErrorObjectAndThrow(Handle<JSRegExp> re,bool is_ascii,Handle<String> error_message,Isolate * isolate)373 static bool CreateRegExpErrorObjectAndThrow(Handle<JSRegExp> re,
374                                             bool is_ascii,
375                                             Handle<String> error_message,
376                                             Isolate* isolate) {
377   Factory* factory = isolate->factory();
378   Handle<FixedArray> elements = factory->NewFixedArray(2);
379   elements->set(0, re->Pattern());
380   elements->set(1, *error_message);
381   Handle<JSArray> array = factory->NewJSArrayWithElements(elements);
382   Handle<Object> regexp_err =
383       factory->NewSyntaxError("malformed_regexp", array);
384   isolate->Throw(*regexp_err);
385   return false;
386 }
387 
388 
CompileIrregexp(Handle<JSRegExp> re,Handle<String> sample_subject,bool is_ascii)389 bool RegExpImpl::CompileIrregexp(Handle<JSRegExp> re,
390                                  Handle<String> sample_subject,
391                                  bool is_ascii) {
392   // Compile the RegExp.
393   Isolate* isolate = re->GetIsolate();
394   Zone zone(isolate);
395   PostponeInterruptsScope postpone(isolate);
396   // If we had a compilation error the last time this is saved at the
397   // saved code index.
398   Object* entry = re->DataAt(JSRegExp::code_index(is_ascii));
399   // When arriving here entry can only be a smi, either representing an
400   // uncompiled regexp, a previous compilation error, or code that has
401   // been flushed.
402   ASSERT(entry->IsSmi());
403   int entry_value = Smi::cast(entry)->value();
404   ASSERT(entry_value == JSRegExp::kUninitializedValue ||
405          entry_value == JSRegExp::kCompilationErrorValue ||
406          (entry_value < JSRegExp::kCodeAgeMask && entry_value >= 0));
407 
408   if (entry_value == JSRegExp::kCompilationErrorValue) {
409     // A previous compilation failed and threw an error which we store in
410     // the saved code index (we store the error message, not the actual
411     // error). Recreate the error object and throw it.
412     Object* error_string = re->DataAt(JSRegExp::saved_code_index(is_ascii));
413     ASSERT(error_string->IsString());
414     Handle<String> error_message(String::cast(error_string));
415     CreateRegExpErrorObjectAndThrow(re, is_ascii, error_message, isolate);
416     return false;
417   }
418 
419   JSRegExp::Flags flags = re->GetFlags();
420 
421   Handle<String> pattern(re->Pattern());
422   pattern = String::Flatten(pattern);
423   RegExpCompileData compile_data;
424   FlatStringReader reader(isolate, pattern);
425   if (!RegExpParser::ParseRegExp(&reader, flags.is_multiline(),
426                                  &compile_data,
427                                  &zone)) {
428     // Throw an exception if we fail to parse the pattern.
429     // THIS SHOULD NOT HAPPEN. We already pre-parsed it successfully once.
430     USE(ThrowRegExpException(re,
431                              pattern,
432                              compile_data.error,
433                              "malformed_regexp"));
434     return false;
435   }
436   RegExpEngine::CompilationResult result =
437       RegExpEngine::Compile(&compile_data,
438                             flags.is_ignore_case(),
439                             flags.is_global(),
440                             flags.is_multiline(),
441                             pattern,
442                             sample_subject,
443                             is_ascii,
444                             &zone);
445   if (result.error_message != NULL) {
446     // Unable to compile regexp.
447     Handle<String> error_message = isolate->factory()->NewStringFromUtf8(
448         CStrVector(result.error_message)).ToHandleChecked();
449     CreateRegExpErrorObjectAndThrow(re, is_ascii, error_message, isolate);
450     return false;
451   }
452 
453   Handle<FixedArray> data = Handle<FixedArray>(FixedArray::cast(re->data()));
454   data->set(JSRegExp::code_index(is_ascii), result.code);
455   int register_max = IrregexpMaxRegisterCount(*data);
456   if (result.num_registers > register_max) {
457     SetIrregexpMaxRegisterCount(*data, result.num_registers);
458   }
459 
460   return true;
461 }
462 
463 
IrregexpMaxRegisterCount(FixedArray * re)464 int RegExpImpl::IrregexpMaxRegisterCount(FixedArray* re) {
465   return Smi::cast(
466       re->get(JSRegExp::kIrregexpMaxRegisterCountIndex))->value();
467 }
468 
469 
SetIrregexpMaxRegisterCount(FixedArray * re,int value)470 void RegExpImpl::SetIrregexpMaxRegisterCount(FixedArray* re, int value) {
471   re->set(JSRegExp::kIrregexpMaxRegisterCountIndex, Smi::FromInt(value));
472 }
473 
474 
IrregexpNumberOfCaptures(FixedArray * re)475 int RegExpImpl::IrregexpNumberOfCaptures(FixedArray* re) {
476   return Smi::cast(re->get(JSRegExp::kIrregexpCaptureCountIndex))->value();
477 }
478 
479 
IrregexpNumberOfRegisters(FixedArray * re)480 int RegExpImpl::IrregexpNumberOfRegisters(FixedArray* re) {
481   return Smi::cast(re->get(JSRegExp::kIrregexpMaxRegisterCountIndex))->value();
482 }
483 
484 
IrregexpByteCode(FixedArray * re,bool is_ascii)485 ByteArray* RegExpImpl::IrregexpByteCode(FixedArray* re, bool is_ascii) {
486   return ByteArray::cast(re->get(JSRegExp::code_index(is_ascii)));
487 }
488 
489 
IrregexpNativeCode(FixedArray * re,bool is_ascii)490 Code* RegExpImpl::IrregexpNativeCode(FixedArray* re, bool is_ascii) {
491   return Code::cast(re->get(JSRegExp::code_index(is_ascii)));
492 }
493 
494 
IrregexpInitialize(Handle<JSRegExp> re,Handle<String> pattern,JSRegExp::Flags flags,int capture_count)495 void RegExpImpl::IrregexpInitialize(Handle<JSRegExp> re,
496                                     Handle<String> pattern,
497                                     JSRegExp::Flags flags,
498                                     int capture_count) {
499   // Initialize compiled code entries to null.
500   re->GetIsolate()->factory()->SetRegExpIrregexpData(re,
501                                                      JSRegExp::IRREGEXP,
502                                                      pattern,
503                                                      flags,
504                                                      capture_count);
505 }
506 
507 
IrregexpPrepare(Handle<JSRegExp> regexp,Handle<String> subject)508 int RegExpImpl::IrregexpPrepare(Handle<JSRegExp> regexp,
509                                 Handle<String> subject) {
510   subject = String::Flatten(subject);
511 
512   // Check the asciiness of the underlying storage.
513   bool is_ascii = subject->IsOneByteRepresentationUnderneath();
514   if (!EnsureCompiledIrregexp(regexp, subject, is_ascii)) return -1;
515 
516 #ifdef V8_INTERPRETED_REGEXP
517   // Byte-code regexp needs space allocated for all its registers.
518   // The result captures are copied to the start of the registers array
519   // if the match succeeds.  This way those registers are not clobbered
520   // when we set the last match info from last successful match.
521   return IrregexpNumberOfRegisters(FixedArray::cast(regexp->data())) +
522          (IrregexpNumberOfCaptures(FixedArray::cast(regexp->data())) + 1) * 2;
523 #else  // V8_INTERPRETED_REGEXP
524   // Native regexp only needs room to output captures. Registers are handled
525   // internally.
526   return (IrregexpNumberOfCaptures(FixedArray::cast(regexp->data())) + 1) * 2;
527 #endif  // V8_INTERPRETED_REGEXP
528 }
529 
530 
IrregexpExecRaw(Handle<JSRegExp> regexp,Handle<String> subject,int index,int32_t * output,int output_size)531 int RegExpImpl::IrregexpExecRaw(Handle<JSRegExp> regexp,
532                                 Handle<String> subject,
533                                 int index,
534                                 int32_t* output,
535                                 int output_size) {
536   Isolate* isolate = regexp->GetIsolate();
537 
538   Handle<FixedArray> irregexp(FixedArray::cast(regexp->data()), isolate);
539 
540   ASSERT(index >= 0);
541   ASSERT(index <= subject->length());
542   ASSERT(subject->IsFlat());
543 
544   bool is_ascii = subject->IsOneByteRepresentationUnderneath();
545 
546 #ifndef V8_INTERPRETED_REGEXP
547   ASSERT(output_size >= (IrregexpNumberOfCaptures(*irregexp) + 1) * 2);
548   do {
549     EnsureCompiledIrregexp(regexp, subject, is_ascii);
550     Handle<Code> code(IrregexpNativeCode(*irregexp, is_ascii), isolate);
551     // The stack is used to allocate registers for the compiled regexp code.
552     // This means that in case of failure, the output registers array is left
553     // untouched and contains the capture results from the previous successful
554     // match.  We can use that to set the last match info lazily.
555     NativeRegExpMacroAssembler::Result res =
556         NativeRegExpMacroAssembler::Match(code,
557                                           subject,
558                                           output,
559                                           output_size,
560                                           index,
561                                           isolate);
562     if (res != NativeRegExpMacroAssembler::RETRY) {
563       ASSERT(res != NativeRegExpMacroAssembler::EXCEPTION ||
564              isolate->has_pending_exception());
565       STATIC_ASSERT(
566           static_cast<int>(NativeRegExpMacroAssembler::SUCCESS) == RE_SUCCESS);
567       STATIC_ASSERT(
568           static_cast<int>(NativeRegExpMacroAssembler::FAILURE) == RE_FAILURE);
569       STATIC_ASSERT(static_cast<int>(NativeRegExpMacroAssembler::EXCEPTION)
570                     == RE_EXCEPTION);
571       return static_cast<IrregexpResult>(res);
572     }
573     // If result is RETRY, the string has changed representation, and we
574     // must restart from scratch.
575     // In this case, it means we must make sure we are prepared to handle
576     // the, potentially, different subject (the string can switch between
577     // being internal and external, and even between being ASCII and UC16,
578     // but the characters are always the same).
579     IrregexpPrepare(regexp, subject);
580     is_ascii = subject->IsOneByteRepresentationUnderneath();
581   } while (true);
582   UNREACHABLE();
583   return RE_EXCEPTION;
584 #else  // V8_INTERPRETED_REGEXP
585 
586   ASSERT(output_size >= IrregexpNumberOfRegisters(*irregexp));
587   // We must have done EnsureCompiledIrregexp, so we can get the number of
588   // registers.
589   int number_of_capture_registers =
590       (IrregexpNumberOfCaptures(*irregexp) + 1) * 2;
591   int32_t* raw_output = &output[number_of_capture_registers];
592   // We do not touch the actual capture result registers until we know there
593   // has been a match so that we can use those capture results to set the
594   // last match info.
595   for (int i = number_of_capture_registers - 1; i >= 0; i--) {
596     raw_output[i] = -1;
597   }
598   Handle<ByteArray> byte_codes(IrregexpByteCode(*irregexp, is_ascii), isolate);
599 
600   IrregexpResult result = IrregexpInterpreter::Match(isolate,
601                                                      byte_codes,
602                                                      subject,
603                                                      raw_output,
604                                                      index);
605   if (result == RE_SUCCESS) {
606     // Copy capture results to the start of the registers array.
607     MemCopy(output, raw_output, number_of_capture_registers * sizeof(int32_t));
608   }
609   if (result == RE_EXCEPTION) {
610     ASSERT(!isolate->has_pending_exception());
611     isolate->StackOverflow();
612   }
613   return result;
614 #endif  // V8_INTERPRETED_REGEXP
615 }
616 
617 
IrregexpExec(Handle<JSRegExp> regexp,Handle<String> subject,int previous_index,Handle<JSArray> last_match_info)618 MaybeHandle<Object> RegExpImpl::IrregexpExec(Handle<JSRegExp> regexp,
619                                              Handle<String> subject,
620                                              int previous_index,
621                                              Handle<JSArray> last_match_info) {
622   Isolate* isolate = regexp->GetIsolate();
623   ASSERT_EQ(regexp->TypeTag(), JSRegExp::IRREGEXP);
624 
625   // Prepare space for the return values.
626 #if defined(V8_INTERPRETED_REGEXP) && defined(DEBUG)
627   if (FLAG_trace_regexp_bytecodes) {
628     String* pattern = regexp->Pattern();
629     PrintF("\n\nRegexp match:   /%s/\n\n", pattern->ToCString().get());
630     PrintF("\n\nSubject string: '%s'\n\n", subject->ToCString().get());
631   }
632 #endif
633   int required_registers = RegExpImpl::IrregexpPrepare(regexp, subject);
634   if (required_registers < 0) {
635     // Compiling failed with an exception.
636     ASSERT(isolate->has_pending_exception());
637     return MaybeHandle<Object>();
638   }
639 
640   int32_t* output_registers = NULL;
641   if (required_registers > Isolate::kJSRegexpStaticOffsetsVectorSize) {
642     output_registers = NewArray<int32_t>(required_registers);
643   }
644   SmartArrayPointer<int32_t> auto_release(output_registers);
645   if (output_registers == NULL) {
646     output_registers = isolate->jsregexp_static_offsets_vector();
647   }
648 
649   int res = RegExpImpl::IrregexpExecRaw(
650       regexp, subject, previous_index, output_registers, required_registers);
651   if (res == RE_SUCCESS) {
652     int capture_count =
653         IrregexpNumberOfCaptures(FixedArray::cast(regexp->data()));
654     return SetLastMatchInfo(
655         last_match_info, subject, capture_count, output_registers);
656   }
657   if (res == RE_EXCEPTION) {
658     ASSERT(isolate->has_pending_exception());
659     return MaybeHandle<Object>();
660   }
661   ASSERT(res == RE_FAILURE);
662   return isolate->factory()->null_value();
663 }
664 
665 
SetLastMatchInfo(Handle<JSArray> last_match_info,Handle<String> subject,int capture_count,int32_t * match)666 Handle<JSArray> RegExpImpl::SetLastMatchInfo(Handle<JSArray> last_match_info,
667                                              Handle<String> subject,
668                                              int capture_count,
669                                              int32_t* match) {
670   ASSERT(last_match_info->HasFastObjectElements());
671   int capture_register_count = (capture_count + 1) * 2;
672   JSArray::EnsureSize(last_match_info,
673                       capture_register_count + kLastMatchOverhead);
674   DisallowHeapAllocation no_allocation;
675   FixedArray* array = FixedArray::cast(last_match_info->elements());
676   if (match != NULL) {
677     for (int i = 0; i < capture_register_count; i += 2) {
678       SetCapture(array, i, match[i]);
679       SetCapture(array, i + 1, match[i + 1]);
680     }
681   }
682   SetLastCaptureCount(array, capture_register_count);
683   SetLastSubject(array, *subject);
684   SetLastInput(array, *subject);
685   return last_match_info;
686 }
687 
688 
GlobalCache(Handle<JSRegExp> regexp,Handle<String> subject,bool is_global,Isolate * isolate)689 RegExpImpl::GlobalCache::GlobalCache(Handle<JSRegExp> regexp,
690                                      Handle<String> subject,
691                                      bool is_global,
692                                      Isolate* isolate)
693   : register_array_(NULL),
694     register_array_size_(0),
695     regexp_(regexp),
696     subject_(subject) {
697 #ifdef V8_INTERPRETED_REGEXP
698   bool interpreted = true;
699 #else
700   bool interpreted = false;
701 #endif  // V8_INTERPRETED_REGEXP
702 
703   if (regexp_->TypeTag() == JSRegExp::ATOM) {
704     static const int kAtomRegistersPerMatch = 2;
705     registers_per_match_ = kAtomRegistersPerMatch;
706     // There is no distinction between interpreted and native for atom regexps.
707     interpreted = false;
708   } else {
709     registers_per_match_ = RegExpImpl::IrregexpPrepare(regexp_, subject_);
710     if (registers_per_match_ < 0) {
711       num_matches_ = -1;  // Signal exception.
712       return;
713     }
714   }
715 
716   if (is_global && !interpreted) {
717     register_array_size_ =
718         Max(registers_per_match_, Isolate::kJSRegexpStaticOffsetsVectorSize);
719     max_matches_ = register_array_size_ / registers_per_match_;
720   } else {
721     // Global loop in interpreted regexp is not implemented.  We choose
722     // the size of the offsets vector so that it can only store one match.
723     register_array_size_ = registers_per_match_;
724     max_matches_ = 1;
725   }
726 
727   if (register_array_size_ > Isolate::kJSRegexpStaticOffsetsVectorSize) {
728     register_array_ = NewArray<int32_t>(register_array_size_);
729   } else {
730     register_array_ = isolate->jsregexp_static_offsets_vector();
731   }
732 
733   // Set state so that fetching the results the first time triggers a call
734   // to the compiled regexp.
735   current_match_index_ = max_matches_ - 1;
736   num_matches_ = max_matches_;
737   ASSERT(registers_per_match_ >= 2);  // Each match has at least one capture.
738   ASSERT_GE(register_array_size_, registers_per_match_);
739   int32_t* last_match =
740       &register_array_[current_match_index_ * registers_per_match_];
741   last_match[0] = -1;
742   last_match[1] = 0;
743 }
744 
745 
746 // -------------------------------------------------------------------
747 // Implementation of the Irregexp regular expression engine.
748 //
749 // The Irregexp regular expression engine is intended to be a complete
750 // implementation of ECMAScript regular expressions.  It generates either
751 // bytecodes or native code.
752 
753 //   The Irregexp regexp engine is structured in three steps.
754 //   1) The parser generates an abstract syntax tree.  See ast.cc.
755 //   2) From the AST a node network is created.  The nodes are all
756 //      subclasses of RegExpNode.  The nodes represent states when
757 //      executing a regular expression.  Several optimizations are
758 //      performed on the node network.
759 //   3) From the nodes we generate either byte codes or native code
760 //      that can actually execute the regular expression (perform
761 //      the search).  The code generation step is described in more
762 //      detail below.
763 
764 // Code generation.
765 //
766 //   The nodes are divided into four main categories.
767 //   * Choice nodes
768 //        These represent places where the regular expression can
769 //        match in more than one way.  For example on entry to an
770 //        alternation (foo|bar) or a repetition (*, +, ? or {}).
771 //   * Action nodes
772 //        These represent places where some action should be
773 //        performed.  Examples include recording the current position
774 //        in the input string to a register (in order to implement
775 //        captures) or other actions on register for example in order
776 //        to implement the counters needed for {} repetitions.
777 //   * Matching nodes
778 //        These attempt to match some element part of the input string.
779 //        Examples of elements include character classes, plain strings
780 //        or back references.
781 //   * End nodes
782 //        These are used to implement the actions required on finding
783 //        a successful match or failing to find a match.
784 //
785 //   The code generated (whether as byte codes or native code) maintains
786 //   some state as it runs.  This consists of the following elements:
787 //
788 //   * The capture registers.  Used for string captures.
789 //   * Other registers.  Used for counters etc.
790 //   * The current position.
791 //   * The stack of backtracking information.  Used when a matching node
792 //     fails to find a match and needs to try an alternative.
793 //
794 // Conceptual regular expression execution model:
795 //
796 //   There is a simple conceptual model of regular expression execution
797 //   which will be presented first.  The actual code generated is a more
798 //   efficient simulation of the simple conceptual model:
799 //
800 //   * Choice nodes are implemented as follows:
801 //     For each choice except the last {
802 //       push current position
803 //       push backtrack code location
804 //       <generate code to test for choice>
805 //       backtrack code location:
806 //       pop current position
807 //     }
808 //     <generate code to test for last choice>
809 //
810 //   * Actions nodes are generated as follows
811 //     <push affected registers on backtrack stack>
812 //     <generate code to perform action>
813 //     push backtrack code location
814 //     <generate code to test for following nodes>
815 //     backtrack code location:
816 //     <pop affected registers to restore their state>
817 //     <pop backtrack location from stack and go to it>
818 //
819 //   * Matching nodes are generated as follows:
820 //     if input string matches at current position
821 //       update current position
822 //       <generate code to test for following nodes>
823 //     else
824 //       <pop backtrack location from stack and go to it>
825 //
826 //   Thus it can be seen that the current position is saved and restored
827 //   by the choice nodes, whereas the registers are saved and restored by
828 //   by the action nodes that manipulate them.
829 //
830 //   The other interesting aspect of this model is that nodes are generated
831 //   at the point where they are needed by a recursive call to Emit().  If
832 //   the node has already been code generated then the Emit() call will
833 //   generate a jump to the previously generated code instead.  In order to
834 //   limit recursion it is possible for the Emit() function to put the node
835 //   on a work list for later generation and instead generate a jump.  The
836 //   destination of the jump is resolved later when the code is generated.
837 //
838 // Actual regular expression code generation.
839 //
840 //   Code generation is actually more complicated than the above.  In order
841 //   to improve the efficiency of the generated code some optimizations are
842 //   performed
843 //
844 //   * Choice nodes have 1-character lookahead.
845 //     A choice node looks at the following character and eliminates some of
846 //     the choices immediately based on that character.  This is not yet
847 //     implemented.
848 //   * Simple greedy loops store reduced backtracking information.
849 //     A quantifier like /.*foo/m will greedily match the whole input.  It will
850 //     then need to backtrack to a point where it can match "foo".  The naive
851 //     implementation of this would push each character position onto the
852 //     backtracking stack, then pop them off one by one.  This would use space
853 //     proportional to the length of the input string.  However since the "."
854 //     can only match in one way and always has a constant length (in this case
855 //     of 1) it suffices to store the current position on the top of the stack
856 //     once.  Matching now becomes merely incrementing the current position and
857 //     backtracking becomes decrementing the current position and checking the
858 //     result against the stored current position.  This is faster and saves
859 //     space.
860 //   * The current state is virtualized.
861 //     This is used to defer expensive operations until it is clear that they
862 //     are needed and to generate code for a node more than once, allowing
863 //     specialized an efficient versions of the code to be created. This is
864 //     explained in the section below.
865 //
866 // Execution state virtualization.
867 //
868 //   Instead of emitting code, nodes that manipulate the state can record their
869 //   manipulation in an object called the Trace.  The Trace object can record a
870 //   current position offset, an optional backtrack code location on the top of
871 //   the virtualized backtrack stack and some register changes.  When a node is
872 //   to be emitted it can flush the Trace or update it.  Flushing the Trace
873 //   will emit code to bring the actual state into line with the virtual state.
874 //   Avoiding flushing the state can postpone some work (e.g. updates of capture
875 //   registers).  Postponing work can save time when executing the regular
876 //   expression since it may be found that the work never has to be done as a
877 //   failure to match can occur.  In addition it is much faster to jump to a
878 //   known backtrack code location than it is to pop an unknown backtrack
879 //   location from the stack and jump there.
880 //
881 //   The virtual state found in the Trace affects code generation.  For example
882 //   the virtual state contains the difference between the actual current
883 //   position and the virtual current position, and matching code needs to use
884 //   this offset to attempt a match in the correct location of the input
885 //   string.  Therefore code generated for a non-trivial trace is specialized
886 //   to that trace.  The code generator therefore has the ability to generate
887 //   code for each node several times.  In order to limit the size of the
888 //   generated code there is an arbitrary limit on how many specialized sets of
889 //   code may be generated for a given node.  If the limit is reached, the
890 //   trace is flushed and a generic version of the code for a node is emitted.
891 //   This is subsequently used for that node.  The code emitted for non-generic
892 //   trace is not recorded in the node and so it cannot currently be reused in
893 //   the event that code generation is requested for an identical trace.
894 
895 
AppendToText(RegExpText * text,Zone * zone)896 void RegExpTree::AppendToText(RegExpText* text, Zone* zone) {
897   UNREACHABLE();
898 }
899 
900 
AppendToText(RegExpText * text,Zone * zone)901 void RegExpAtom::AppendToText(RegExpText* text, Zone* zone) {
902   text->AddElement(TextElement::Atom(this), zone);
903 }
904 
905 
AppendToText(RegExpText * text,Zone * zone)906 void RegExpCharacterClass::AppendToText(RegExpText* text, Zone* zone) {
907   text->AddElement(TextElement::CharClass(this), zone);
908 }
909 
910 
AppendToText(RegExpText * text,Zone * zone)911 void RegExpText::AppendToText(RegExpText* text, Zone* zone) {
912   for (int i = 0; i < elements()->length(); i++)
913     text->AddElement(elements()->at(i), zone);
914 }
915 
916 
Atom(RegExpAtom * atom)917 TextElement TextElement::Atom(RegExpAtom* atom) {
918   return TextElement(ATOM, atom);
919 }
920 
921 
CharClass(RegExpCharacterClass * char_class)922 TextElement TextElement::CharClass(RegExpCharacterClass* char_class) {
923   return TextElement(CHAR_CLASS, char_class);
924 }
925 
926 
length() const927 int TextElement::length() const {
928   switch (text_type()) {
929     case ATOM:
930       return atom()->length();
931 
932     case CHAR_CLASS:
933       return 1;
934   }
935   UNREACHABLE();
936   return 0;
937 }
938 
939 
GetTable(bool ignore_case)940 DispatchTable* ChoiceNode::GetTable(bool ignore_case) {
941   if (table_ == NULL) {
942     table_ = new(zone()) DispatchTable(zone());
943     DispatchTableConstructor cons(table_, ignore_case, zone());
944     cons.BuildTable(this);
945   }
946   return table_;
947 }
948 
949 
950 class FrequencyCollator {
951  public:
FrequencyCollator()952   FrequencyCollator() : total_samples_(0) {
953     for (int i = 0; i < RegExpMacroAssembler::kTableSize; i++) {
954       frequencies_[i] = CharacterFrequency(i);
955     }
956   }
957 
CountCharacter(int character)958   void CountCharacter(int character) {
959     int index = (character & RegExpMacroAssembler::kTableMask);
960     frequencies_[index].Increment();
961     total_samples_++;
962   }
963 
964   // Does not measure in percent, but rather per-128 (the table size from the
965   // regexp macro assembler).
Frequency(int in_character)966   int Frequency(int in_character) {
967     ASSERT((in_character & RegExpMacroAssembler::kTableMask) == in_character);
968     if (total_samples_ < 1) return 1;  // Division by zero.
969     int freq_in_per128 =
970         (frequencies_[in_character].counter() * 128) / total_samples_;
971     return freq_in_per128;
972   }
973 
974  private:
975   class CharacterFrequency {
976    public:
CharacterFrequency()977     CharacterFrequency() : counter_(0), character_(-1) { }
CharacterFrequency(int character)978     explicit CharacterFrequency(int character)
979         : counter_(0), character_(character) { }
980 
Increment()981     void Increment() { counter_++; }
counter()982     int counter() { return counter_; }
character()983     int character() { return character_; }
984 
985    private:
986     int counter_;
987     int character_;
988   };
989 
990 
991  private:
992   CharacterFrequency frequencies_[RegExpMacroAssembler::kTableSize];
993   int total_samples_;
994 };
995 
996 
997 class RegExpCompiler {
998  public:
999   RegExpCompiler(int capture_count, bool ignore_case, bool is_ascii,
1000                  Zone* zone);
1001 
AllocateRegister()1002   int AllocateRegister() {
1003     if (next_register_ >= RegExpMacroAssembler::kMaxRegister) {
1004       reg_exp_too_big_ = true;
1005       return next_register_;
1006     }
1007     return next_register_++;
1008   }
1009 
1010   RegExpEngine::CompilationResult Assemble(RegExpMacroAssembler* assembler,
1011                                            RegExpNode* start,
1012                                            int capture_count,
1013                                            Handle<String> pattern);
1014 
AddWork(RegExpNode * node)1015   inline void AddWork(RegExpNode* node) { work_list_->Add(node); }
1016 
1017   static const int kImplementationOffset = 0;
1018   static const int kNumberOfRegistersOffset = 0;
1019   static const int kCodeOffset = 1;
1020 
macro_assembler()1021   RegExpMacroAssembler* macro_assembler() { return macro_assembler_; }
accept()1022   EndNode* accept() { return accept_; }
1023 
1024   static const int kMaxRecursion = 100;
recursion_depth()1025   inline int recursion_depth() { return recursion_depth_; }
IncrementRecursionDepth()1026   inline void IncrementRecursionDepth() { recursion_depth_++; }
DecrementRecursionDepth()1027   inline void DecrementRecursionDepth() { recursion_depth_--; }
1028 
SetRegExpTooBig()1029   void SetRegExpTooBig() { reg_exp_too_big_ = true; }
1030 
ignore_case()1031   inline bool ignore_case() { return ignore_case_; }
ascii()1032   inline bool ascii() { return ascii_; }
frequency_collator()1033   FrequencyCollator* frequency_collator() { return &frequency_collator_; }
1034 
current_expansion_factor()1035   int current_expansion_factor() { return current_expansion_factor_; }
set_current_expansion_factor(int value)1036   void set_current_expansion_factor(int value) {
1037     current_expansion_factor_ = value;
1038   }
1039 
zone() const1040   Zone* zone() const { return zone_; }
1041 
1042   static const int kNoRegister = -1;
1043 
1044  private:
1045   EndNode* accept_;
1046   int next_register_;
1047   List<RegExpNode*>* work_list_;
1048   int recursion_depth_;
1049   RegExpMacroAssembler* macro_assembler_;
1050   bool ignore_case_;
1051   bool ascii_;
1052   bool reg_exp_too_big_;
1053   int current_expansion_factor_;
1054   FrequencyCollator frequency_collator_;
1055   Zone* zone_;
1056 };
1057 
1058 
1059 class RecursionCheck {
1060  public:
RecursionCheck(RegExpCompiler * compiler)1061   explicit RecursionCheck(RegExpCompiler* compiler) : compiler_(compiler) {
1062     compiler->IncrementRecursionDepth();
1063   }
~RecursionCheck()1064   ~RecursionCheck() { compiler_->DecrementRecursionDepth(); }
1065  private:
1066   RegExpCompiler* compiler_;
1067 };
1068 
1069 
IrregexpRegExpTooBig(Isolate * isolate)1070 static RegExpEngine::CompilationResult IrregexpRegExpTooBig(Isolate* isolate) {
1071   return RegExpEngine::CompilationResult(isolate, "RegExp too big");
1072 }
1073 
1074 
1075 // Attempts to compile the regexp using an Irregexp code generator.  Returns
1076 // a fixed array or a null handle depending on whether it succeeded.
RegExpCompiler(int capture_count,bool ignore_case,bool ascii,Zone * zone)1077 RegExpCompiler::RegExpCompiler(int capture_count, bool ignore_case, bool ascii,
1078                                Zone* zone)
1079     : next_register_(2 * (capture_count + 1)),
1080       work_list_(NULL),
1081       recursion_depth_(0),
1082       ignore_case_(ignore_case),
1083       ascii_(ascii),
1084       reg_exp_too_big_(false),
1085       current_expansion_factor_(1),
1086       frequency_collator_(),
1087       zone_(zone) {
1088   accept_ = new(zone) EndNode(EndNode::ACCEPT, zone);
1089   ASSERT(next_register_ - 1 <= RegExpMacroAssembler::kMaxRegister);
1090 }
1091 
1092 
Assemble(RegExpMacroAssembler * macro_assembler,RegExpNode * start,int capture_count,Handle<String> pattern)1093 RegExpEngine::CompilationResult RegExpCompiler::Assemble(
1094     RegExpMacroAssembler* macro_assembler,
1095     RegExpNode* start,
1096     int capture_count,
1097     Handle<String> pattern) {
1098   Heap* heap = pattern->GetHeap();
1099 
1100   bool use_slow_safe_regexp_compiler = false;
1101   if (heap->total_regexp_code_generated() >
1102           RegExpImpl::kRegWxpCompiledLimit &&
1103       heap->isolate()->memory_allocator()->SizeExecutable() >
1104           RegExpImpl::kRegExpExecutableMemoryLimit) {
1105     use_slow_safe_regexp_compiler = true;
1106   }
1107 
1108   macro_assembler->set_slow_safe(use_slow_safe_regexp_compiler);
1109 
1110 #ifdef DEBUG
1111   if (FLAG_trace_regexp_assembler)
1112     macro_assembler_ = new RegExpMacroAssemblerTracer(macro_assembler);
1113   else
1114 #endif
1115     macro_assembler_ = macro_assembler;
1116 
1117   List <RegExpNode*> work_list(0);
1118   work_list_ = &work_list;
1119   Label fail;
1120   macro_assembler_->PushBacktrack(&fail);
1121   Trace new_trace;
1122   start->Emit(this, &new_trace);
1123   macro_assembler_->Bind(&fail);
1124   macro_assembler_->Fail();
1125   while (!work_list.is_empty()) {
1126     work_list.RemoveLast()->Emit(this, &new_trace);
1127   }
1128   if (reg_exp_too_big_) return IrregexpRegExpTooBig(zone_->isolate());
1129 
1130   Handle<HeapObject> code = macro_assembler_->GetCode(pattern);
1131   heap->IncreaseTotalRegexpCodeGenerated(code->Size());
1132   work_list_ = NULL;
1133 #ifdef DEBUG
1134   if (FLAG_print_code) {
1135     CodeTracer::Scope trace_scope(heap->isolate()->GetCodeTracer());
1136     Handle<Code>::cast(code)->Disassemble(pattern->ToCString().get(),
1137                                           trace_scope.file());
1138   }
1139   if (FLAG_trace_regexp_assembler) {
1140     delete macro_assembler_;
1141   }
1142 #endif
1143   return RegExpEngine::CompilationResult(*code, next_register_);
1144 }
1145 
1146 
Mentions(int that)1147 bool Trace::DeferredAction::Mentions(int that) {
1148   if (action_type() == ActionNode::CLEAR_CAPTURES) {
1149     Interval range = static_cast<DeferredClearCaptures*>(this)->range();
1150     return range.Contains(that);
1151   } else {
1152     return reg() == that;
1153   }
1154 }
1155 
1156 
mentions_reg(int reg)1157 bool Trace::mentions_reg(int reg) {
1158   for (DeferredAction* action = actions_;
1159        action != NULL;
1160        action = action->next()) {
1161     if (action->Mentions(reg))
1162       return true;
1163   }
1164   return false;
1165 }
1166 
1167 
GetStoredPosition(int reg,int * cp_offset)1168 bool Trace::GetStoredPosition(int reg, int* cp_offset) {
1169   ASSERT_EQ(0, *cp_offset);
1170   for (DeferredAction* action = actions_;
1171        action != NULL;
1172        action = action->next()) {
1173     if (action->Mentions(reg)) {
1174       if (action->action_type() == ActionNode::STORE_POSITION) {
1175         *cp_offset = static_cast<DeferredCapture*>(action)->cp_offset();
1176         return true;
1177       } else {
1178         return false;
1179       }
1180     }
1181   }
1182   return false;
1183 }
1184 
1185 
FindAffectedRegisters(OutSet * affected_registers,Zone * zone)1186 int Trace::FindAffectedRegisters(OutSet* affected_registers,
1187                                  Zone* zone) {
1188   int max_register = RegExpCompiler::kNoRegister;
1189   for (DeferredAction* action = actions_;
1190        action != NULL;
1191        action = action->next()) {
1192     if (action->action_type() == ActionNode::CLEAR_CAPTURES) {
1193       Interval range = static_cast<DeferredClearCaptures*>(action)->range();
1194       for (int i = range.from(); i <= range.to(); i++)
1195         affected_registers->Set(i, zone);
1196       if (range.to() > max_register) max_register = range.to();
1197     } else {
1198       affected_registers->Set(action->reg(), zone);
1199       if (action->reg() > max_register) max_register = action->reg();
1200     }
1201   }
1202   return max_register;
1203 }
1204 
1205 
RestoreAffectedRegisters(RegExpMacroAssembler * assembler,int max_register,const OutSet & registers_to_pop,const OutSet & registers_to_clear)1206 void Trace::RestoreAffectedRegisters(RegExpMacroAssembler* assembler,
1207                                      int max_register,
1208                                      const OutSet& registers_to_pop,
1209                                      const OutSet& registers_to_clear) {
1210   for (int reg = max_register; reg >= 0; reg--) {
1211     if (registers_to_pop.Get(reg)) {
1212       assembler->PopRegister(reg);
1213     } else if (registers_to_clear.Get(reg)) {
1214       int clear_to = reg;
1215       while (reg > 0 && registers_to_clear.Get(reg - 1)) {
1216         reg--;
1217       }
1218       assembler->ClearRegisters(reg, clear_to);
1219     }
1220   }
1221 }
1222 
1223 
PerformDeferredActions(RegExpMacroAssembler * assembler,int max_register,const OutSet & affected_registers,OutSet * registers_to_pop,OutSet * registers_to_clear,Zone * zone)1224 void Trace::PerformDeferredActions(RegExpMacroAssembler* assembler,
1225                                    int max_register,
1226                                    const OutSet& affected_registers,
1227                                    OutSet* registers_to_pop,
1228                                    OutSet* registers_to_clear,
1229                                    Zone* zone) {
1230   // The "+1" is to avoid a push_limit of zero if stack_limit_slack() is 1.
1231   const int push_limit = (assembler->stack_limit_slack() + 1) / 2;
1232 
1233   // Count pushes performed to force a stack limit check occasionally.
1234   int pushes = 0;
1235 
1236   for (int reg = 0; reg <= max_register; reg++) {
1237     if (!affected_registers.Get(reg)) {
1238       continue;
1239     }
1240 
1241     // The chronologically first deferred action in the trace
1242     // is used to infer the action needed to restore a register
1243     // to its previous state (or not, if it's safe to ignore it).
1244     enum DeferredActionUndoType { IGNORE, RESTORE, CLEAR };
1245     DeferredActionUndoType undo_action = IGNORE;
1246 
1247     int value = 0;
1248     bool absolute = false;
1249     bool clear = false;
1250     int store_position = -1;
1251     // This is a little tricky because we are scanning the actions in reverse
1252     // historical order (newest first).
1253     for (DeferredAction* action = actions_;
1254          action != NULL;
1255          action = action->next()) {
1256       if (action->Mentions(reg)) {
1257         switch (action->action_type()) {
1258           case ActionNode::SET_REGISTER: {
1259             Trace::DeferredSetRegister* psr =
1260                 static_cast<Trace::DeferredSetRegister*>(action);
1261             if (!absolute) {
1262               value += psr->value();
1263               absolute = true;
1264             }
1265             // SET_REGISTER is currently only used for newly introduced loop
1266             // counters. They can have a significant previous value if they
1267             // occour in a loop. TODO(lrn): Propagate this information, so
1268             // we can set undo_action to IGNORE if we know there is no value to
1269             // restore.
1270             undo_action = RESTORE;
1271             ASSERT_EQ(store_position, -1);
1272             ASSERT(!clear);
1273             break;
1274           }
1275           case ActionNode::INCREMENT_REGISTER:
1276             if (!absolute) {
1277               value++;
1278             }
1279             ASSERT_EQ(store_position, -1);
1280             ASSERT(!clear);
1281             undo_action = RESTORE;
1282             break;
1283           case ActionNode::STORE_POSITION: {
1284             Trace::DeferredCapture* pc =
1285                 static_cast<Trace::DeferredCapture*>(action);
1286             if (!clear && store_position == -1) {
1287               store_position = pc->cp_offset();
1288             }
1289 
1290             // For captures we know that stores and clears alternate.
1291             // Other register, are never cleared, and if the occur
1292             // inside a loop, they might be assigned more than once.
1293             if (reg <= 1) {
1294               // Registers zero and one, aka "capture zero", is
1295               // always set correctly if we succeed. There is no
1296               // need to undo a setting on backtrack, because we
1297               // will set it again or fail.
1298               undo_action = IGNORE;
1299             } else {
1300               undo_action = pc->is_capture() ? CLEAR : RESTORE;
1301             }
1302             ASSERT(!absolute);
1303             ASSERT_EQ(value, 0);
1304             break;
1305           }
1306           case ActionNode::CLEAR_CAPTURES: {
1307             // Since we're scanning in reverse order, if we've already
1308             // set the position we have to ignore historically earlier
1309             // clearing operations.
1310             if (store_position == -1) {
1311               clear = true;
1312             }
1313             undo_action = RESTORE;
1314             ASSERT(!absolute);
1315             ASSERT_EQ(value, 0);
1316             break;
1317           }
1318           default:
1319             UNREACHABLE();
1320             break;
1321         }
1322       }
1323     }
1324     // Prepare for the undo-action (e.g., push if it's going to be popped).
1325     if (undo_action == RESTORE) {
1326       pushes++;
1327       RegExpMacroAssembler::StackCheckFlag stack_check =
1328           RegExpMacroAssembler::kNoStackLimitCheck;
1329       if (pushes == push_limit) {
1330         stack_check = RegExpMacroAssembler::kCheckStackLimit;
1331         pushes = 0;
1332       }
1333 
1334       assembler->PushRegister(reg, stack_check);
1335       registers_to_pop->Set(reg, zone);
1336     } else if (undo_action == CLEAR) {
1337       registers_to_clear->Set(reg, zone);
1338     }
1339     // Perform the chronologically last action (or accumulated increment)
1340     // for the register.
1341     if (store_position != -1) {
1342       assembler->WriteCurrentPositionToRegister(reg, store_position);
1343     } else if (clear) {
1344       assembler->ClearRegisters(reg, reg);
1345     } else if (absolute) {
1346       assembler->SetRegister(reg, value);
1347     } else if (value != 0) {
1348       assembler->AdvanceRegister(reg, value);
1349     }
1350   }
1351 }
1352 
1353 
1354 // This is called as we come into a loop choice node and some other tricky
1355 // nodes.  It normalizes the state of the code generator to ensure we can
1356 // generate generic code.
Flush(RegExpCompiler * compiler,RegExpNode * successor)1357 void Trace::Flush(RegExpCompiler* compiler, RegExpNode* successor) {
1358   RegExpMacroAssembler* assembler = compiler->macro_assembler();
1359 
1360   ASSERT(!is_trivial());
1361 
1362   if (actions_ == NULL && backtrack() == NULL) {
1363     // Here we just have some deferred cp advances to fix and we are back to
1364     // a normal situation.  We may also have to forget some information gained
1365     // through a quick check that was already performed.
1366     if (cp_offset_ != 0) assembler->AdvanceCurrentPosition(cp_offset_);
1367     // Create a new trivial state and generate the node with that.
1368     Trace new_state;
1369     successor->Emit(compiler, &new_state);
1370     return;
1371   }
1372 
1373   // Generate deferred actions here along with code to undo them again.
1374   OutSet affected_registers;
1375 
1376   if (backtrack() != NULL) {
1377     // Here we have a concrete backtrack location.  These are set up by choice
1378     // nodes and so they indicate that we have a deferred save of the current
1379     // position which we may need to emit here.
1380     assembler->PushCurrentPosition();
1381   }
1382 
1383   int max_register = FindAffectedRegisters(&affected_registers,
1384                                            compiler->zone());
1385   OutSet registers_to_pop;
1386   OutSet registers_to_clear;
1387   PerformDeferredActions(assembler,
1388                          max_register,
1389                          affected_registers,
1390                          &registers_to_pop,
1391                          &registers_to_clear,
1392                          compiler->zone());
1393   if (cp_offset_ != 0) {
1394     assembler->AdvanceCurrentPosition(cp_offset_);
1395   }
1396 
1397   // Create a new trivial state and generate the node with that.
1398   Label undo;
1399   assembler->PushBacktrack(&undo);
1400   Trace new_state;
1401   successor->Emit(compiler, &new_state);
1402 
1403   // On backtrack we need to restore state.
1404   assembler->Bind(&undo);
1405   RestoreAffectedRegisters(assembler,
1406                            max_register,
1407                            registers_to_pop,
1408                            registers_to_clear);
1409   if (backtrack() == NULL) {
1410     assembler->Backtrack();
1411   } else {
1412     assembler->PopCurrentPosition();
1413     assembler->GoTo(backtrack());
1414   }
1415 }
1416 
1417 
Emit(RegExpCompiler * compiler,Trace * trace)1418 void NegativeSubmatchSuccess::Emit(RegExpCompiler* compiler, Trace* trace) {
1419   RegExpMacroAssembler* assembler = compiler->macro_assembler();
1420 
1421   // Omit flushing the trace. We discard the entire stack frame anyway.
1422 
1423   if (!label()->is_bound()) {
1424     // We are completely independent of the trace, since we ignore it,
1425     // so this code can be used as the generic version.
1426     assembler->Bind(label());
1427   }
1428 
1429   // Throw away everything on the backtrack stack since the start
1430   // of the negative submatch and restore the character position.
1431   assembler->ReadCurrentPositionFromRegister(current_position_register_);
1432   assembler->ReadStackPointerFromRegister(stack_pointer_register_);
1433   if (clear_capture_count_ > 0) {
1434     // Clear any captures that might have been performed during the success
1435     // of the body of the negative look-ahead.
1436     int clear_capture_end = clear_capture_start_ + clear_capture_count_ - 1;
1437     assembler->ClearRegisters(clear_capture_start_, clear_capture_end);
1438   }
1439   // Now that we have unwound the stack we find at the top of the stack the
1440   // backtrack that the BeginSubmatch node got.
1441   assembler->Backtrack();
1442 }
1443 
1444 
Emit(RegExpCompiler * compiler,Trace * trace)1445 void EndNode::Emit(RegExpCompiler* compiler, Trace* trace) {
1446   if (!trace->is_trivial()) {
1447     trace->Flush(compiler, this);
1448     return;
1449   }
1450   RegExpMacroAssembler* assembler = compiler->macro_assembler();
1451   if (!label()->is_bound()) {
1452     assembler->Bind(label());
1453   }
1454   switch (action_) {
1455     case ACCEPT:
1456       assembler->Succeed();
1457       return;
1458     case BACKTRACK:
1459       assembler->GoTo(trace->backtrack());
1460       return;
1461     case NEGATIVE_SUBMATCH_SUCCESS:
1462       // This case is handled in a different virtual method.
1463       UNREACHABLE();
1464   }
1465   UNIMPLEMENTED();
1466 }
1467 
1468 
AddGuard(Guard * guard,Zone * zone)1469 void GuardedAlternative::AddGuard(Guard* guard, Zone* zone) {
1470   if (guards_ == NULL)
1471     guards_ = new(zone) ZoneList<Guard*>(1, zone);
1472   guards_->Add(guard, zone);
1473 }
1474 
1475 
SetRegister(int reg,int val,RegExpNode * on_success)1476 ActionNode* ActionNode::SetRegister(int reg,
1477                                     int val,
1478                                     RegExpNode* on_success) {
1479   ActionNode* result =
1480       new(on_success->zone()) ActionNode(SET_REGISTER, on_success);
1481   result->data_.u_store_register.reg = reg;
1482   result->data_.u_store_register.value = val;
1483   return result;
1484 }
1485 
1486 
IncrementRegister(int reg,RegExpNode * on_success)1487 ActionNode* ActionNode::IncrementRegister(int reg, RegExpNode* on_success) {
1488   ActionNode* result =
1489       new(on_success->zone()) ActionNode(INCREMENT_REGISTER, on_success);
1490   result->data_.u_increment_register.reg = reg;
1491   return result;
1492 }
1493 
1494 
StorePosition(int reg,bool is_capture,RegExpNode * on_success)1495 ActionNode* ActionNode::StorePosition(int reg,
1496                                       bool is_capture,
1497                                       RegExpNode* on_success) {
1498   ActionNode* result =
1499       new(on_success->zone()) ActionNode(STORE_POSITION, on_success);
1500   result->data_.u_position_register.reg = reg;
1501   result->data_.u_position_register.is_capture = is_capture;
1502   return result;
1503 }
1504 
1505 
ClearCaptures(Interval range,RegExpNode * on_success)1506 ActionNode* ActionNode::ClearCaptures(Interval range,
1507                                       RegExpNode* on_success) {
1508   ActionNode* result =
1509       new(on_success->zone()) ActionNode(CLEAR_CAPTURES, on_success);
1510   result->data_.u_clear_captures.range_from = range.from();
1511   result->data_.u_clear_captures.range_to = range.to();
1512   return result;
1513 }
1514 
1515 
BeginSubmatch(int stack_reg,int position_reg,RegExpNode * on_success)1516 ActionNode* ActionNode::BeginSubmatch(int stack_reg,
1517                                       int position_reg,
1518                                       RegExpNode* on_success) {
1519   ActionNode* result =
1520       new(on_success->zone()) ActionNode(BEGIN_SUBMATCH, on_success);
1521   result->data_.u_submatch.stack_pointer_register = stack_reg;
1522   result->data_.u_submatch.current_position_register = position_reg;
1523   return result;
1524 }
1525 
1526 
PositiveSubmatchSuccess(int stack_reg,int position_reg,int clear_register_count,int clear_register_from,RegExpNode * on_success)1527 ActionNode* ActionNode::PositiveSubmatchSuccess(int stack_reg,
1528                                                 int position_reg,
1529                                                 int clear_register_count,
1530                                                 int clear_register_from,
1531                                                 RegExpNode* on_success) {
1532   ActionNode* result =
1533       new(on_success->zone()) ActionNode(POSITIVE_SUBMATCH_SUCCESS, on_success);
1534   result->data_.u_submatch.stack_pointer_register = stack_reg;
1535   result->data_.u_submatch.current_position_register = position_reg;
1536   result->data_.u_submatch.clear_register_count = clear_register_count;
1537   result->data_.u_submatch.clear_register_from = clear_register_from;
1538   return result;
1539 }
1540 
1541 
EmptyMatchCheck(int start_register,int repetition_register,int repetition_limit,RegExpNode * on_success)1542 ActionNode* ActionNode::EmptyMatchCheck(int start_register,
1543                                         int repetition_register,
1544                                         int repetition_limit,
1545                                         RegExpNode* on_success) {
1546   ActionNode* result =
1547       new(on_success->zone()) ActionNode(EMPTY_MATCH_CHECK, on_success);
1548   result->data_.u_empty_match_check.start_register = start_register;
1549   result->data_.u_empty_match_check.repetition_register = repetition_register;
1550   result->data_.u_empty_match_check.repetition_limit = repetition_limit;
1551   return result;
1552 }
1553 
1554 
1555 #define DEFINE_ACCEPT(Type)                                          \
1556   void Type##Node::Accept(NodeVisitor* visitor) {                    \
1557     visitor->Visit##Type(this);                                      \
1558   }
FOR_EACH_NODE_TYPE(DEFINE_ACCEPT)1559 FOR_EACH_NODE_TYPE(DEFINE_ACCEPT)
1560 #undef DEFINE_ACCEPT
1561 
1562 
1563 void LoopChoiceNode::Accept(NodeVisitor* visitor) {
1564   visitor->VisitLoopChoice(this);
1565 }
1566 
1567 
1568 // -------------------------------------------------------------------
1569 // Emit code.
1570 
1571 
GenerateGuard(RegExpMacroAssembler * macro_assembler,Guard * guard,Trace * trace)1572 void ChoiceNode::GenerateGuard(RegExpMacroAssembler* macro_assembler,
1573                                Guard* guard,
1574                                Trace* trace) {
1575   switch (guard->op()) {
1576     case Guard::LT:
1577       ASSERT(!trace->mentions_reg(guard->reg()));
1578       macro_assembler->IfRegisterGE(guard->reg(),
1579                                     guard->value(),
1580                                     trace->backtrack());
1581       break;
1582     case Guard::GEQ:
1583       ASSERT(!trace->mentions_reg(guard->reg()));
1584       macro_assembler->IfRegisterLT(guard->reg(),
1585                                     guard->value(),
1586                                     trace->backtrack());
1587       break;
1588   }
1589 }
1590 
1591 
1592 // Returns the number of characters in the equivalence class, omitting those
1593 // that cannot occur in the source string because it is ASCII.
GetCaseIndependentLetters(Isolate * isolate,uc16 character,bool ascii_subject,unibrow::uchar * letters)1594 static int GetCaseIndependentLetters(Isolate* isolate,
1595                                      uc16 character,
1596                                      bool ascii_subject,
1597                                      unibrow::uchar* letters) {
1598   int length =
1599       isolate->jsregexp_uncanonicalize()->get(character, '\0', letters);
1600   // Unibrow returns 0 or 1 for characters where case independence is
1601   // trivial.
1602   if (length == 0) {
1603     letters[0] = character;
1604     length = 1;
1605   }
1606   if (!ascii_subject || character <= String::kMaxOneByteCharCode) {
1607     return length;
1608   }
1609   // The standard requires that non-ASCII characters cannot have ASCII
1610   // character codes in their equivalence class.
1611   return 0;
1612 }
1613 
1614 
EmitSimpleCharacter(Isolate * isolate,RegExpCompiler * compiler,uc16 c,Label * on_failure,int cp_offset,bool check,bool preloaded)1615 static inline bool EmitSimpleCharacter(Isolate* isolate,
1616                                        RegExpCompiler* compiler,
1617                                        uc16 c,
1618                                        Label* on_failure,
1619                                        int cp_offset,
1620                                        bool check,
1621                                        bool preloaded) {
1622   RegExpMacroAssembler* assembler = compiler->macro_assembler();
1623   bool bound_checked = false;
1624   if (!preloaded) {
1625     assembler->LoadCurrentCharacter(
1626         cp_offset,
1627         on_failure,
1628         check);
1629     bound_checked = true;
1630   }
1631   assembler->CheckNotCharacter(c, on_failure);
1632   return bound_checked;
1633 }
1634 
1635 
1636 // Only emits non-letters (things that don't have case).  Only used for case
1637 // independent matches.
EmitAtomNonLetter(Isolate * isolate,RegExpCompiler * compiler,uc16 c,Label * on_failure,int cp_offset,bool check,bool preloaded)1638 static inline bool EmitAtomNonLetter(Isolate* isolate,
1639                                      RegExpCompiler* compiler,
1640                                      uc16 c,
1641                                      Label* on_failure,
1642                                      int cp_offset,
1643                                      bool check,
1644                                      bool preloaded) {
1645   RegExpMacroAssembler* macro_assembler = compiler->macro_assembler();
1646   bool ascii = compiler->ascii();
1647   unibrow::uchar chars[unibrow::Ecma262UnCanonicalize::kMaxWidth];
1648   int length = GetCaseIndependentLetters(isolate, c, ascii, chars);
1649   if (length < 1) {
1650     // This can't match.  Must be an ASCII subject and a non-ASCII character.
1651     // We do not need to do anything since the ASCII pass already handled this.
1652     return false;  // Bounds not checked.
1653   }
1654   bool checked = false;
1655   // We handle the length > 1 case in a later pass.
1656   if (length == 1) {
1657     if (ascii && c > String::kMaxOneByteCharCodeU) {
1658       // Can't match - see above.
1659       return false;  // Bounds not checked.
1660     }
1661     if (!preloaded) {
1662       macro_assembler->LoadCurrentCharacter(cp_offset, on_failure, check);
1663       checked = check;
1664     }
1665     macro_assembler->CheckNotCharacter(c, on_failure);
1666   }
1667   return checked;
1668 }
1669 
1670 
ShortCutEmitCharacterPair(RegExpMacroAssembler * macro_assembler,bool ascii,uc16 c1,uc16 c2,Label * on_failure)1671 static bool ShortCutEmitCharacterPair(RegExpMacroAssembler* macro_assembler,
1672                                       bool ascii,
1673                                       uc16 c1,
1674                                       uc16 c2,
1675                                       Label* on_failure) {
1676   uc16 char_mask;
1677   if (ascii) {
1678     char_mask = String::kMaxOneByteCharCode;
1679   } else {
1680     char_mask = String::kMaxUtf16CodeUnit;
1681   }
1682   uc16 exor = c1 ^ c2;
1683   // Check whether exor has only one bit set.
1684   if (((exor - 1) & exor) == 0) {
1685     // If c1 and c2 differ only by one bit.
1686     // Ecma262UnCanonicalize always gives the highest number last.
1687     ASSERT(c2 > c1);
1688     uc16 mask = char_mask ^ exor;
1689     macro_assembler->CheckNotCharacterAfterAnd(c1, mask, on_failure);
1690     return true;
1691   }
1692   ASSERT(c2 > c1);
1693   uc16 diff = c2 - c1;
1694   if (((diff - 1) & diff) == 0 && c1 >= diff) {
1695     // If the characters differ by 2^n but don't differ by one bit then
1696     // subtract the difference from the found character, then do the or
1697     // trick.  We avoid the theoretical case where negative numbers are
1698     // involved in order to simplify code generation.
1699     uc16 mask = char_mask ^ diff;
1700     macro_assembler->CheckNotCharacterAfterMinusAnd(c1 - diff,
1701                                                     diff,
1702                                                     mask,
1703                                                     on_failure);
1704     return true;
1705   }
1706   return false;
1707 }
1708 
1709 
1710 typedef bool EmitCharacterFunction(Isolate* isolate,
1711                                    RegExpCompiler* compiler,
1712                                    uc16 c,
1713                                    Label* on_failure,
1714                                    int cp_offset,
1715                                    bool check,
1716                                    bool preloaded);
1717 
1718 // Only emits letters (things that have case).  Only used for case independent
1719 // matches.
EmitAtomLetter(Isolate * isolate,RegExpCompiler * compiler,uc16 c,Label * on_failure,int cp_offset,bool check,bool preloaded)1720 static inline bool EmitAtomLetter(Isolate* isolate,
1721                                   RegExpCompiler* compiler,
1722                                   uc16 c,
1723                                   Label* on_failure,
1724                                   int cp_offset,
1725                                   bool check,
1726                                   bool preloaded) {
1727   RegExpMacroAssembler* macro_assembler = compiler->macro_assembler();
1728   bool ascii = compiler->ascii();
1729   unibrow::uchar chars[unibrow::Ecma262UnCanonicalize::kMaxWidth];
1730   int length = GetCaseIndependentLetters(isolate, c, ascii, chars);
1731   if (length <= 1) return false;
1732   // We may not need to check against the end of the input string
1733   // if this character lies before a character that matched.
1734   if (!preloaded) {
1735     macro_assembler->LoadCurrentCharacter(cp_offset, on_failure, check);
1736   }
1737   Label ok;
1738   ASSERT(unibrow::Ecma262UnCanonicalize::kMaxWidth == 4);
1739   switch (length) {
1740     case 2: {
1741       if (ShortCutEmitCharacterPair(macro_assembler,
1742                                     ascii,
1743                                     chars[0],
1744                                     chars[1],
1745                                     on_failure)) {
1746       } else {
1747         macro_assembler->CheckCharacter(chars[0], &ok);
1748         macro_assembler->CheckNotCharacter(chars[1], on_failure);
1749         macro_assembler->Bind(&ok);
1750       }
1751       break;
1752     }
1753     case 4:
1754       macro_assembler->CheckCharacter(chars[3], &ok);
1755       // Fall through!
1756     case 3:
1757       macro_assembler->CheckCharacter(chars[0], &ok);
1758       macro_assembler->CheckCharacter(chars[1], &ok);
1759       macro_assembler->CheckNotCharacter(chars[2], on_failure);
1760       macro_assembler->Bind(&ok);
1761       break;
1762     default:
1763       UNREACHABLE();
1764       break;
1765   }
1766   return true;
1767 }
1768 
1769 
EmitBoundaryTest(RegExpMacroAssembler * masm,int border,Label * fall_through,Label * above_or_equal,Label * below)1770 static void EmitBoundaryTest(RegExpMacroAssembler* masm,
1771                              int border,
1772                              Label* fall_through,
1773                              Label* above_or_equal,
1774                              Label* below) {
1775   if (below != fall_through) {
1776     masm->CheckCharacterLT(border, below);
1777     if (above_or_equal != fall_through) masm->GoTo(above_or_equal);
1778   } else {
1779     masm->CheckCharacterGT(border - 1, above_or_equal);
1780   }
1781 }
1782 
1783 
EmitDoubleBoundaryTest(RegExpMacroAssembler * masm,int first,int last,Label * fall_through,Label * in_range,Label * out_of_range)1784 static void EmitDoubleBoundaryTest(RegExpMacroAssembler* masm,
1785                                    int first,
1786                                    int last,
1787                                    Label* fall_through,
1788                                    Label* in_range,
1789                                    Label* out_of_range) {
1790   if (in_range == fall_through) {
1791     if (first == last) {
1792       masm->CheckNotCharacter(first, out_of_range);
1793     } else {
1794       masm->CheckCharacterNotInRange(first, last, out_of_range);
1795     }
1796   } else {
1797     if (first == last) {
1798       masm->CheckCharacter(first, in_range);
1799     } else {
1800       masm->CheckCharacterInRange(first, last, in_range);
1801     }
1802     if (out_of_range != fall_through) masm->GoTo(out_of_range);
1803   }
1804 }
1805 
1806 
1807 // even_label is for ranges[i] to ranges[i + 1] where i - start_index is even.
1808 // odd_label is for ranges[i] to ranges[i + 1] where i - start_index is odd.
EmitUseLookupTable(RegExpMacroAssembler * masm,ZoneList<int> * ranges,int start_index,int end_index,int min_char,Label * fall_through,Label * even_label,Label * odd_label)1809 static void EmitUseLookupTable(
1810     RegExpMacroAssembler* masm,
1811     ZoneList<int>* ranges,
1812     int start_index,
1813     int end_index,
1814     int min_char,
1815     Label* fall_through,
1816     Label* even_label,
1817     Label* odd_label) {
1818   static const int kSize = RegExpMacroAssembler::kTableSize;
1819   static const int kMask = RegExpMacroAssembler::kTableMask;
1820 
1821   int base = (min_char & ~kMask);
1822   USE(base);
1823 
1824   // Assert that everything is on one kTableSize page.
1825   for (int i = start_index; i <= end_index; i++) {
1826     ASSERT_EQ(ranges->at(i) & ~kMask, base);
1827   }
1828   ASSERT(start_index == 0 || (ranges->at(start_index - 1) & ~kMask) <= base);
1829 
1830   char templ[kSize];
1831   Label* on_bit_set;
1832   Label* on_bit_clear;
1833   int bit;
1834   if (even_label == fall_through) {
1835     on_bit_set = odd_label;
1836     on_bit_clear = even_label;
1837     bit = 1;
1838   } else {
1839     on_bit_set = even_label;
1840     on_bit_clear = odd_label;
1841     bit = 0;
1842   }
1843   for (int i = 0; i < (ranges->at(start_index) & kMask) && i < kSize; i++) {
1844     templ[i] = bit;
1845   }
1846   int j = 0;
1847   bit ^= 1;
1848   for (int i = start_index; i < end_index; i++) {
1849     for (j = (ranges->at(i) & kMask); j < (ranges->at(i + 1) & kMask); j++) {
1850       templ[j] = bit;
1851     }
1852     bit ^= 1;
1853   }
1854   for (int i = j; i < kSize; i++) {
1855     templ[i] = bit;
1856   }
1857   Factory* factory = masm->zone()->isolate()->factory();
1858   // TODO(erikcorry): Cache these.
1859   Handle<ByteArray> ba = factory->NewByteArray(kSize, TENURED);
1860   for (int i = 0; i < kSize; i++) {
1861     ba->set(i, templ[i]);
1862   }
1863   masm->CheckBitInTable(ba, on_bit_set);
1864   if (on_bit_clear != fall_through) masm->GoTo(on_bit_clear);
1865 }
1866 
1867 
CutOutRange(RegExpMacroAssembler * masm,ZoneList<int> * ranges,int start_index,int end_index,int cut_index,Label * even_label,Label * odd_label)1868 static void CutOutRange(RegExpMacroAssembler* masm,
1869                         ZoneList<int>* ranges,
1870                         int start_index,
1871                         int end_index,
1872                         int cut_index,
1873                         Label* even_label,
1874                         Label* odd_label) {
1875   bool odd = (((cut_index - start_index) & 1) == 1);
1876   Label* in_range_label = odd ? odd_label : even_label;
1877   Label dummy;
1878   EmitDoubleBoundaryTest(masm,
1879                          ranges->at(cut_index),
1880                          ranges->at(cut_index + 1) - 1,
1881                          &dummy,
1882                          in_range_label,
1883                          &dummy);
1884   ASSERT(!dummy.is_linked());
1885   // Cut out the single range by rewriting the array.  This creates a new
1886   // range that is a merger of the two ranges on either side of the one we
1887   // are cutting out.  The oddity of the labels is preserved.
1888   for (int j = cut_index; j > start_index; j--) {
1889     ranges->at(j) = ranges->at(j - 1);
1890   }
1891   for (int j = cut_index + 1; j < end_index; j++) {
1892     ranges->at(j) = ranges->at(j + 1);
1893   }
1894 }
1895 
1896 
1897 // Unicode case.  Split the search space into kSize spaces that are handled
1898 // with recursion.
SplitSearchSpace(ZoneList<int> * ranges,int start_index,int end_index,int * new_start_index,int * new_end_index,int * border)1899 static void SplitSearchSpace(ZoneList<int>* ranges,
1900                              int start_index,
1901                              int end_index,
1902                              int* new_start_index,
1903                              int* new_end_index,
1904                              int* border) {
1905   static const int kSize = RegExpMacroAssembler::kTableSize;
1906   static const int kMask = RegExpMacroAssembler::kTableMask;
1907 
1908   int first = ranges->at(start_index);
1909   int last = ranges->at(end_index) - 1;
1910 
1911   *new_start_index = start_index;
1912   *border = (ranges->at(start_index) & ~kMask) + kSize;
1913   while (*new_start_index < end_index) {
1914     if (ranges->at(*new_start_index) > *border) break;
1915     (*new_start_index)++;
1916   }
1917   // new_start_index is the index of the first edge that is beyond the
1918   // current kSize space.
1919 
1920   // For very large search spaces we do a binary chop search of the non-ASCII
1921   // space instead of just going to the end of the current kSize space.  The
1922   // heuristics are complicated a little by the fact that any 128-character
1923   // encoding space can be quickly tested with a table lookup, so we don't
1924   // wish to do binary chop search at a smaller granularity than that.  A
1925   // 128-character space can take up a lot of space in the ranges array if,
1926   // for example, we only want to match every second character (eg. the lower
1927   // case characters on some Unicode pages).
1928   int binary_chop_index = (end_index + start_index) / 2;
1929   // The first test ensures that we get to the code that handles the ASCII
1930   // range with a single not-taken branch, speeding up this important
1931   // character range (even non-ASCII charset-based text has spaces and
1932   // punctuation).
1933   if (*border - 1 > String::kMaxOneByteCharCode &&  // ASCII case.
1934       end_index - start_index > (*new_start_index - start_index) * 2 &&
1935       last - first > kSize * 2 &&
1936       binary_chop_index > *new_start_index &&
1937       ranges->at(binary_chop_index) >= first + 2 * kSize) {
1938     int scan_forward_for_section_border = binary_chop_index;;
1939     int new_border = (ranges->at(binary_chop_index) | kMask) + 1;
1940 
1941     while (scan_forward_for_section_border < end_index) {
1942       if (ranges->at(scan_forward_for_section_border) > new_border) {
1943         *new_start_index = scan_forward_for_section_border;
1944         *border = new_border;
1945         break;
1946       }
1947       scan_forward_for_section_border++;
1948     }
1949   }
1950 
1951   ASSERT(*new_start_index > start_index);
1952   *new_end_index = *new_start_index - 1;
1953   if (ranges->at(*new_end_index) == *border) {
1954     (*new_end_index)--;
1955   }
1956   if (*border >= ranges->at(end_index)) {
1957     *border = ranges->at(end_index);
1958     *new_start_index = end_index;  // Won't be used.
1959     *new_end_index = end_index - 1;
1960   }
1961 }
1962 
1963 
1964 // Gets a series of segment boundaries representing a character class.  If the
1965 // character is in the range between an even and an odd boundary (counting from
1966 // start_index) then go to even_label, otherwise go to odd_label.  We already
1967 // know that the character is in the range of min_char to max_char inclusive.
1968 // Either label can be NULL indicating backtracking.  Either label can also be
1969 // equal to the fall_through label.
GenerateBranches(RegExpMacroAssembler * masm,ZoneList<int> * ranges,int start_index,int end_index,uc16 min_char,uc16 max_char,Label * fall_through,Label * even_label,Label * odd_label)1970 static void GenerateBranches(RegExpMacroAssembler* masm,
1971                              ZoneList<int>* ranges,
1972                              int start_index,
1973                              int end_index,
1974                              uc16 min_char,
1975                              uc16 max_char,
1976                              Label* fall_through,
1977                              Label* even_label,
1978                              Label* odd_label) {
1979   int first = ranges->at(start_index);
1980   int last = ranges->at(end_index) - 1;
1981 
1982   ASSERT_LT(min_char, first);
1983 
1984   // Just need to test if the character is before or on-or-after
1985   // a particular character.
1986   if (start_index == end_index) {
1987     EmitBoundaryTest(masm, first, fall_through, even_label, odd_label);
1988     return;
1989   }
1990 
1991   // Another almost trivial case:  There is one interval in the middle that is
1992   // different from the end intervals.
1993   if (start_index + 1 == end_index) {
1994     EmitDoubleBoundaryTest(
1995         masm, first, last, fall_through, even_label, odd_label);
1996     return;
1997   }
1998 
1999   // It's not worth using table lookup if there are very few intervals in the
2000   // character class.
2001   if (end_index - start_index <= 6) {
2002     // It is faster to test for individual characters, so we look for those
2003     // first, then try arbitrary ranges in the second round.
2004     static int kNoCutIndex = -1;
2005     int cut = kNoCutIndex;
2006     for (int i = start_index; i < end_index; i++) {
2007       if (ranges->at(i) == ranges->at(i + 1) - 1) {
2008         cut = i;
2009         break;
2010       }
2011     }
2012     if (cut == kNoCutIndex) cut = start_index;
2013     CutOutRange(
2014         masm, ranges, start_index, end_index, cut, even_label, odd_label);
2015     ASSERT_GE(end_index - start_index, 2);
2016     GenerateBranches(masm,
2017                      ranges,
2018                      start_index + 1,
2019                      end_index - 1,
2020                      min_char,
2021                      max_char,
2022                      fall_through,
2023                      even_label,
2024                      odd_label);
2025     return;
2026   }
2027 
2028   // If there are a lot of intervals in the regexp, then we will use tables to
2029   // determine whether the character is inside or outside the character class.
2030   static const int kBits = RegExpMacroAssembler::kTableSizeBits;
2031 
2032   if ((max_char >> kBits) == (min_char >> kBits)) {
2033     EmitUseLookupTable(masm,
2034                        ranges,
2035                        start_index,
2036                        end_index,
2037                        min_char,
2038                        fall_through,
2039                        even_label,
2040                        odd_label);
2041     return;
2042   }
2043 
2044   if ((min_char >> kBits) != (first >> kBits)) {
2045     masm->CheckCharacterLT(first, odd_label);
2046     GenerateBranches(masm,
2047                      ranges,
2048                      start_index + 1,
2049                      end_index,
2050                      first,
2051                      max_char,
2052                      fall_through,
2053                      odd_label,
2054                      even_label);
2055     return;
2056   }
2057 
2058   int new_start_index = 0;
2059   int new_end_index = 0;
2060   int border = 0;
2061 
2062   SplitSearchSpace(ranges,
2063                    start_index,
2064                    end_index,
2065                    &new_start_index,
2066                    &new_end_index,
2067                    &border);
2068 
2069   Label handle_rest;
2070   Label* above = &handle_rest;
2071   if (border == last + 1) {
2072     // We didn't find any section that started after the limit, so everything
2073     // above the border is one of the terminal labels.
2074     above = (end_index & 1) != (start_index & 1) ? odd_label : even_label;
2075     ASSERT(new_end_index == end_index - 1);
2076   }
2077 
2078   ASSERT_LE(start_index, new_end_index);
2079   ASSERT_LE(new_start_index, end_index);
2080   ASSERT_LT(start_index, new_start_index);
2081   ASSERT_LT(new_end_index, end_index);
2082   ASSERT(new_end_index + 1 == new_start_index ||
2083          (new_end_index + 2 == new_start_index &&
2084           border == ranges->at(new_end_index + 1)));
2085   ASSERT_LT(min_char, border - 1);
2086   ASSERT_LT(border, max_char);
2087   ASSERT_LT(ranges->at(new_end_index), border);
2088   ASSERT(border < ranges->at(new_start_index) ||
2089          (border == ranges->at(new_start_index) &&
2090           new_start_index == end_index &&
2091           new_end_index == end_index - 1 &&
2092           border == last + 1));
2093   ASSERT(new_start_index == 0 || border >= ranges->at(new_start_index - 1));
2094 
2095   masm->CheckCharacterGT(border - 1, above);
2096   Label dummy;
2097   GenerateBranches(masm,
2098                    ranges,
2099                    start_index,
2100                    new_end_index,
2101                    min_char,
2102                    border - 1,
2103                    &dummy,
2104                    even_label,
2105                    odd_label);
2106   if (handle_rest.is_linked()) {
2107     masm->Bind(&handle_rest);
2108     bool flip = (new_start_index & 1) != (start_index & 1);
2109     GenerateBranches(masm,
2110                      ranges,
2111                      new_start_index,
2112                      end_index,
2113                      border,
2114                      max_char,
2115                      &dummy,
2116                      flip ? odd_label : even_label,
2117                      flip ? even_label : odd_label);
2118   }
2119 }
2120 
2121 
EmitCharClass(RegExpMacroAssembler * macro_assembler,RegExpCharacterClass * cc,bool ascii,Label * on_failure,int cp_offset,bool check_offset,bool preloaded,Zone * zone)2122 static void EmitCharClass(RegExpMacroAssembler* macro_assembler,
2123                           RegExpCharacterClass* cc,
2124                           bool ascii,
2125                           Label* on_failure,
2126                           int cp_offset,
2127                           bool check_offset,
2128                           bool preloaded,
2129                           Zone* zone) {
2130   ZoneList<CharacterRange>* ranges = cc->ranges(zone);
2131   if (!CharacterRange::IsCanonical(ranges)) {
2132     CharacterRange::Canonicalize(ranges);
2133   }
2134 
2135   int max_char;
2136   if (ascii) {
2137     max_char = String::kMaxOneByteCharCode;
2138   } else {
2139     max_char = String::kMaxUtf16CodeUnit;
2140   }
2141 
2142   int range_count = ranges->length();
2143 
2144   int last_valid_range = range_count - 1;
2145   while (last_valid_range >= 0) {
2146     CharacterRange& range = ranges->at(last_valid_range);
2147     if (range.from() <= max_char) {
2148       break;
2149     }
2150     last_valid_range--;
2151   }
2152 
2153   if (last_valid_range < 0) {
2154     if (!cc->is_negated()) {
2155       macro_assembler->GoTo(on_failure);
2156     }
2157     if (check_offset) {
2158       macro_assembler->CheckPosition(cp_offset, on_failure);
2159     }
2160     return;
2161   }
2162 
2163   if (last_valid_range == 0 &&
2164       ranges->at(0).IsEverything(max_char)) {
2165     if (cc->is_negated()) {
2166       macro_assembler->GoTo(on_failure);
2167     } else {
2168       // This is a common case hit by non-anchored expressions.
2169       if (check_offset) {
2170         macro_assembler->CheckPosition(cp_offset, on_failure);
2171       }
2172     }
2173     return;
2174   }
2175   if (last_valid_range == 0 &&
2176       !cc->is_negated() &&
2177       ranges->at(0).IsEverything(max_char)) {
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     return;
2183   }
2184 
2185   if (!preloaded) {
2186     macro_assembler->LoadCurrentCharacter(cp_offset, on_failure, check_offset);
2187   }
2188 
2189   if (cc->is_standard(zone) &&
2190         macro_assembler->CheckSpecialCharacterClass(cc->standard_type(),
2191                                                     on_failure)) {
2192       return;
2193   }
2194 
2195 
2196   // A new list with ascending entries.  Each entry is a code unit
2197   // where there is a boundary between code units that are part of
2198   // the class and code units that are not.  Normally we insert an
2199   // entry at zero which goes to the failure label, but if there
2200   // was already one there we fall through for success on that entry.
2201   // Subsequent entries have alternating meaning (success/failure).
2202   ZoneList<int>* range_boundaries =
2203       new(zone) ZoneList<int>(last_valid_range, zone);
2204 
2205   bool zeroth_entry_is_failure = !cc->is_negated();
2206 
2207   for (int i = 0; i <= last_valid_range; i++) {
2208     CharacterRange& range = ranges->at(i);
2209     if (range.from() == 0) {
2210       ASSERT_EQ(i, 0);
2211       zeroth_entry_is_failure = !zeroth_entry_is_failure;
2212     } else {
2213       range_boundaries->Add(range.from(), zone);
2214     }
2215     range_boundaries->Add(range.to() + 1, zone);
2216   }
2217   int end_index = range_boundaries->length() - 1;
2218   if (range_boundaries->at(end_index) > max_char) {
2219     end_index--;
2220   }
2221 
2222   Label fall_through;
2223   GenerateBranches(macro_assembler,
2224                    range_boundaries,
2225                    0,  // start_index.
2226                    end_index,
2227                    0,  // min_char.
2228                    max_char,
2229                    &fall_through,
2230                    zeroth_entry_is_failure ? &fall_through : on_failure,
2231                    zeroth_entry_is_failure ? on_failure : &fall_through);
2232   macro_assembler->Bind(&fall_through);
2233 }
2234 
2235 
~RegExpNode()2236 RegExpNode::~RegExpNode() {
2237 }
2238 
2239 
LimitVersions(RegExpCompiler * compiler,Trace * trace)2240 RegExpNode::LimitResult RegExpNode::LimitVersions(RegExpCompiler* compiler,
2241                                                   Trace* trace) {
2242   // If we are generating a greedy loop then don't stop and don't reuse code.
2243   if (trace->stop_node() != NULL) {
2244     return CONTINUE;
2245   }
2246 
2247   RegExpMacroAssembler* macro_assembler = compiler->macro_assembler();
2248   if (trace->is_trivial()) {
2249     if (label_.is_bound()) {
2250       // We are being asked to generate a generic version, but that's already
2251       // been done so just go to it.
2252       macro_assembler->GoTo(&label_);
2253       return DONE;
2254     }
2255     if (compiler->recursion_depth() >= RegExpCompiler::kMaxRecursion) {
2256       // To avoid too deep recursion we push the node to the work queue and just
2257       // generate a goto here.
2258       compiler->AddWork(this);
2259       macro_assembler->GoTo(&label_);
2260       return DONE;
2261     }
2262     // Generate generic version of the node and bind the label for later use.
2263     macro_assembler->Bind(&label_);
2264     return CONTINUE;
2265   }
2266 
2267   // We are being asked to make a non-generic version.  Keep track of how many
2268   // non-generic versions we generate so as not to overdo it.
2269   trace_count_++;
2270   if (FLAG_regexp_optimization &&
2271       trace_count_ < kMaxCopiesCodeGenerated &&
2272       compiler->recursion_depth() <= RegExpCompiler::kMaxRecursion) {
2273     return CONTINUE;
2274   }
2275 
2276   // If we get here code has been generated for this node too many times or
2277   // recursion is too deep.  Time to switch to a generic version.  The code for
2278   // generic versions above can handle deep recursion properly.
2279   trace->Flush(compiler, this);
2280   return DONE;
2281 }
2282 
2283 
EatsAtLeast(int still_to_find,int budget,bool not_at_start)2284 int ActionNode::EatsAtLeast(int still_to_find,
2285                             int budget,
2286                             bool not_at_start) {
2287   if (budget <= 0) return 0;
2288   if (action_type_ == POSITIVE_SUBMATCH_SUCCESS) return 0;  // Rewinds input!
2289   return on_success()->EatsAtLeast(still_to_find,
2290                                    budget - 1,
2291                                    not_at_start);
2292 }
2293 
2294 
FillInBMInfo(int offset,int budget,BoyerMooreLookahead * bm,bool not_at_start)2295 void ActionNode::FillInBMInfo(int offset,
2296                               int budget,
2297                               BoyerMooreLookahead* bm,
2298                               bool not_at_start) {
2299   if (action_type_ == BEGIN_SUBMATCH) {
2300     bm->SetRest(offset);
2301   } else if (action_type_ != POSITIVE_SUBMATCH_SUCCESS) {
2302     on_success()->FillInBMInfo(offset, budget - 1, bm, not_at_start);
2303   }
2304   SaveBMInfo(bm, not_at_start, offset);
2305 }
2306 
2307 
EatsAtLeast(int still_to_find,int budget,bool not_at_start)2308 int AssertionNode::EatsAtLeast(int still_to_find,
2309                                int budget,
2310                                bool not_at_start) {
2311   if (budget <= 0) return 0;
2312   // If we know we are not at the start and we are asked "how many characters
2313   // will you match if you succeed?" then we can answer anything since false
2314   // implies false.  So lets just return the max answer (still_to_find) since
2315   // that won't prevent us from preloading a lot of characters for the other
2316   // branches in the node graph.
2317   if (assertion_type() == AT_START && not_at_start) return still_to_find;
2318   return on_success()->EatsAtLeast(still_to_find,
2319                                    budget - 1,
2320                                    not_at_start);
2321 }
2322 
2323 
FillInBMInfo(int offset,int budget,BoyerMooreLookahead * bm,bool not_at_start)2324 void AssertionNode::FillInBMInfo(int offset,
2325                                  int budget,
2326                                  BoyerMooreLookahead* bm,
2327                                  bool not_at_start) {
2328   // Match the behaviour of EatsAtLeast on this node.
2329   if (assertion_type() == AT_START && not_at_start) return;
2330   on_success()->FillInBMInfo(offset, budget - 1, bm, not_at_start);
2331   SaveBMInfo(bm, not_at_start, offset);
2332 }
2333 
2334 
EatsAtLeast(int still_to_find,int budget,bool not_at_start)2335 int BackReferenceNode::EatsAtLeast(int still_to_find,
2336                                    int budget,
2337                                    bool not_at_start) {
2338   if (budget <= 0) return 0;
2339   return on_success()->EatsAtLeast(still_to_find,
2340                                    budget - 1,
2341                                    not_at_start);
2342 }
2343 
2344 
EatsAtLeast(int still_to_find,int budget,bool not_at_start)2345 int TextNode::EatsAtLeast(int still_to_find,
2346                           int budget,
2347                           bool not_at_start) {
2348   int answer = Length();
2349   if (answer >= still_to_find) return answer;
2350   if (budget <= 0) return answer;
2351   // We are not at start after this node so we set the last argument to 'true'.
2352   return answer + on_success()->EatsAtLeast(still_to_find - answer,
2353                                             budget - 1,
2354                                             true);
2355 }
2356 
2357 
EatsAtLeast(int still_to_find,int budget,bool not_at_start)2358 int NegativeLookaheadChoiceNode::EatsAtLeast(int still_to_find,
2359                                              int budget,
2360                                              bool not_at_start) {
2361   if (budget <= 0) return 0;
2362   // Alternative 0 is the negative lookahead, alternative 1 is what comes
2363   // afterwards.
2364   RegExpNode* node = alternatives_->at(1).node();
2365   return node->EatsAtLeast(still_to_find, budget - 1, not_at_start);
2366 }
2367 
2368 
GetQuickCheckDetails(QuickCheckDetails * details,RegExpCompiler * compiler,int filled_in,bool not_at_start)2369 void NegativeLookaheadChoiceNode::GetQuickCheckDetails(
2370     QuickCheckDetails* details,
2371     RegExpCompiler* compiler,
2372     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 
EatsAtLeastHelper(int still_to_find,int budget,RegExpNode * ignore_this_node,bool not_at_start)2381 int 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 
EatsAtLeast(int still_to_find,int budget,bool not_at_start)2401 int 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 
EatsAtLeast(int still_to_find,int budget,bool not_at_start)2411 int 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.
SmearBitsRight(uint32_t v)2422 static 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 
Rationalize(bool asc)2432 bool 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 
EmitQuickCheck(RegExpCompiler * compiler,Trace * trace,bool preload_has_checked_bounds,Label * on_possible_success,QuickCheckDetails * details,bool fall_through_on_failure)2456 bool RegExpNode::EmitQuickCheck(RegExpCompiler* compiler,
2457                                 Trace* trace,
2458                                 bool preload_has_checked_bounds,
2459                                 Label* on_possible_success,
2460                                 QuickCheckDetails* details,
2461                                 bool fall_through_on_failure) {
2462   if (details->characters() == 0) return false;
2463   GetQuickCheckDetails(
2464       details, compiler, 0, trace->at_start() == Trace::FALSE_VALUE);
2465   if (details->cannot_match()) return false;
2466   if (!details->Rationalize(compiler->ascii())) return false;
2467   ASSERT(details->characters() == 1 ||
2468          compiler->macro_assembler()->CanReadUnaligned());
2469   uint32_t mask = details->mask();
2470   uint32_t value = details->value();
2471 
2472   RegExpMacroAssembler* assembler = compiler->macro_assembler();
2473 
2474   if (trace->characters_preloaded() != details->characters()) {
2475     assembler->LoadCurrentCharacter(trace->cp_offset(),
2476                                     trace->backtrack(),
2477                                     !preload_has_checked_bounds,
2478                                     details->characters());
2479   }
2480 
2481 
2482   bool need_mask = true;
2483 
2484   if (details->characters() == 1) {
2485     // If number of characters preloaded is 1 then we used a byte or 16 bit
2486     // load so the value is already masked down.
2487     uint32_t char_mask;
2488     if (compiler->ascii()) {
2489       char_mask = String::kMaxOneByteCharCode;
2490     } else {
2491       char_mask = String::kMaxUtf16CodeUnit;
2492     }
2493     if ((mask & char_mask) == char_mask) need_mask = false;
2494     mask &= char_mask;
2495   } else {
2496     // For 2-character preloads in ASCII mode or 1-character preloads in
2497     // TWO_BYTE mode we also use a 16 bit load with zero extend.
2498     if (details->characters() == 2 && compiler->ascii()) {
2499       if ((mask & 0xffff) == 0xffff) need_mask = false;
2500     } else if (details->characters() == 1 && !compiler->ascii()) {
2501       if ((mask & 0xffff) == 0xffff) need_mask = false;
2502     } else {
2503       if (mask == 0xffffffff) need_mask = false;
2504     }
2505   }
2506 
2507   if (fall_through_on_failure) {
2508     if (need_mask) {
2509       assembler->CheckCharacterAfterAnd(value, mask, on_possible_success);
2510     } else {
2511       assembler->CheckCharacter(value, on_possible_success);
2512     }
2513   } else {
2514     if (need_mask) {
2515       assembler->CheckNotCharacterAfterAnd(value, mask, trace->backtrack());
2516     } else {
2517       assembler->CheckNotCharacter(value, trace->backtrack());
2518     }
2519   }
2520   return true;
2521 }
2522 
2523 
2524 // Here is the meat of GetQuickCheckDetails (see also the comment on the
2525 // super-class in the .h file).
2526 //
2527 // We iterate along the text object, building up for each character a
2528 // mask and value that can be used to test for a quick failure to match.
2529 // The masks and values for the positions will be combined into a single
2530 // machine word for the current character width in order to be used in
2531 // generating a quick check.
GetQuickCheckDetails(QuickCheckDetails * details,RegExpCompiler * compiler,int characters_filled_in,bool not_at_start)2532 void TextNode::GetQuickCheckDetails(QuickCheckDetails* details,
2533                                     RegExpCompiler* compiler,
2534                                     int characters_filled_in,
2535                                     bool not_at_start) {
2536   Isolate* isolate = compiler->macro_assembler()->zone()->isolate();
2537   ASSERT(characters_filled_in < details->characters());
2538   int characters = details->characters();
2539   int char_mask;
2540   if (compiler->ascii()) {
2541     char_mask = String::kMaxOneByteCharCode;
2542   } else {
2543     char_mask = String::kMaxUtf16CodeUnit;
2544   }
2545   for (int k = 0; k < elms_->length(); k++) {
2546     TextElement elm = elms_->at(k);
2547     if (elm.text_type() == TextElement::ATOM) {
2548       Vector<const uc16> quarks = elm.atom()->data();
2549       for (int i = 0; i < characters && i < quarks.length(); i++) {
2550         QuickCheckDetails::Position* pos =
2551             details->positions(characters_filled_in);
2552         uc16 c = quarks[i];
2553         if (c > char_mask) {
2554           // If we expect a non-ASCII character from an ASCII string,
2555           // there is no way we can match. Not even case independent
2556           // matching can turn an ASCII character into non-ASCII or
2557           // vice versa.
2558           details->set_cannot_match();
2559           pos->determines_perfectly = false;
2560           return;
2561         }
2562         if (compiler->ignore_case()) {
2563           unibrow::uchar chars[unibrow::Ecma262UnCanonicalize::kMaxWidth];
2564           int length = GetCaseIndependentLetters(isolate, c, compiler->ascii(),
2565                                                  chars);
2566           ASSERT(length != 0);  // Can only happen if c > char_mask (see above).
2567           if (length == 1) {
2568             // This letter has no case equivalents, so it's nice and simple
2569             // and the mask-compare will determine definitely whether we have
2570             // a match at this character position.
2571             pos->mask = char_mask;
2572             pos->value = c;
2573             pos->determines_perfectly = true;
2574           } else {
2575             uint32_t common_bits = char_mask;
2576             uint32_t bits = chars[0];
2577             for (int j = 1; j < length; j++) {
2578               uint32_t differing_bits = ((chars[j] & common_bits) ^ bits);
2579               common_bits ^= differing_bits;
2580               bits &= common_bits;
2581             }
2582             // If length is 2 and common bits has only one zero in it then
2583             // our mask and compare instruction will determine definitely
2584             // whether we have a match at this character position.  Otherwise
2585             // it can only be an approximate check.
2586             uint32_t one_zero = (common_bits | ~char_mask);
2587             if (length == 2 && ((~one_zero) & ((~one_zero) - 1)) == 0) {
2588               pos->determines_perfectly = true;
2589             }
2590             pos->mask = common_bits;
2591             pos->value = bits;
2592           }
2593         } else {
2594           // Don't ignore case.  Nice simple case where the mask-compare will
2595           // determine definitely whether we have a match at this character
2596           // position.
2597           pos->mask = char_mask;
2598           pos->value = c;
2599           pos->determines_perfectly = true;
2600         }
2601         characters_filled_in++;
2602         ASSERT(characters_filled_in <= details->characters());
2603         if (characters_filled_in == details->characters()) {
2604           return;
2605         }
2606       }
2607     } else {
2608       QuickCheckDetails::Position* pos =
2609           details->positions(characters_filled_in);
2610       RegExpCharacterClass* tree = elm.char_class();
2611       ZoneList<CharacterRange>* ranges = tree->ranges(zone());
2612       if (tree->is_negated()) {
2613         // A quick check uses multi-character mask and compare.  There is no
2614         // useful way to incorporate a negative char class into this scheme
2615         // so we just conservatively create a mask and value that will always
2616         // succeed.
2617         pos->mask = 0;
2618         pos->value = 0;
2619       } else {
2620         int first_range = 0;
2621         while (ranges->at(first_range).from() > char_mask) {
2622           first_range++;
2623           if (first_range == ranges->length()) {
2624             details->set_cannot_match();
2625             pos->determines_perfectly = false;
2626             return;
2627           }
2628         }
2629         CharacterRange range = ranges->at(first_range);
2630         uc16 from = range.from();
2631         uc16 to = range.to();
2632         if (to > char_mask) {
2633           to = char_mask;
2634         }
2635         uint32_t differing_bits = (from ^ to);
2636         // A mask and compare is only perfect if the differing bits form a
2637         // number like 00011111 with one single block of trailing 1s.
2638         if ((differing_bits & (differing_bits + 1)) == 0 &&
2639              from + differing_bits == to) {
2640           pos->determines_perfectly = true;
2641         }
2642         uint32_t common_bits = ~SmearBitsRight(differing_bits);
2643         uint32_t bits = (from & common_bits);
2644         for (int i = first_range + 1; i < ranges->length(); i++) {
2645           CharacterRange range = ranges->at(i);
2646           uc16 from = range.from();
2647           uc16 to = range.to();
2648           if (from > char_mask) continue;
2649           if (to > char_mask) to = char_mask;
2650           // Here we are combining more ranges into the mask and compare
2651           // value.  With each new range the mask becomes more sparse and
2652           // so the chances of a false positive rise.  A character class
2653           // with multiple ranges is assumed never to be equivalent to a
2654           // mask and compare operation.
2655           pos->determines_perfectly = false;
2656           uint32_t new_common_bits = (from ^ to);
2657           new_common_bits = ~SmearBitsRight(new_common_bits);
2658           common_bits &= new_common_bits;
2659           bits &= new_common_bits;
2660           uint32_t differing_bits = (from & common_bits) ^ bits;
2661           common_bits ^= differing_bits;
2662           bits &= common_bits;
2663         }
2664         pos->mask = common_bits;
2665         pos->value = bits;
2666       }
2667       characters_filled_in++;
2668       ASSERT(characters_filled_in <= details->characters());
2669       if (characters_filled_in == details->characters()) {
2670         return;
2671       }
2672     }
2673   }
2674   ASSERT(characters_filled_in != details->characters());
2675   if (!details->cannot_match()) {
2676     on_success()-> GetQuickCheckDetails(details,
2677                                         compiler,
2678                                         characters_filled_in,
2679                                         true);
2680   }
2681 }
2682 
2683 
Clear()2684 void QuickCheckDetails::Clear() {
2685   for (int i = 0; i < characters_; i++) {
2686     positions_[i].mask = 0;
2687     positions_[i].value = 0;
2688     positions_[i].determines_perfectly = false;
2689   }
2690   characters_ = 0;
2691 }
2692 
2693 
Advance(int by,bool ascii)2694 void QuickCheckDetails::Advance(int by, bool ascii) {
2695   ASSERT(by >= 0);
2696   if (by >= characters_) {
2697     Clear();
2698     return;
2699   }
2700   for (int i = 0; i < characters_ - by; i++) {
2701     positions_[i] = positions_[by + i];
2702   }
2703   for (int i = characters_ - by; i < characters_; i++) {
2704     positions_[i].mask = 0;
2705     positions_[i].value = 0;
2706     positions_[i].determines_perfectly = false;
2707   }
2708   characters_ -= by;
2709   // We could change mask_ and value_ here but we would never advance unless
2710   // they had already been used in a check and they won't be used again because
2711   // it would gain us nothing.  So there's no point.
2712 }
2713 
2714 
Merge(QuickCheckDetails * other,int from_index)2715 void QuickCheckDetails::Merge(QuickCheckDetails* other, int from_index) {
2716   ASSERT(characters_ == other->characters_);
2717   if (other->cannot_match_) {
2718     return;
2719   }
2720   if (cannot_match_) {
2721     *this = *other;
2722     return;
2723   }
2724   for (int i = from_index; i < characters_; i++) {
2725     QuickCheckDetails::Position* pos = positions(i);
2726     QuickCheckDetails::Position* other_pos = other->positions(i);
2727     if (pos->mask != other_pos->mask ||
2728         pos->value != other_pos->value ||
2729         !other_pos->determines_perfectly) {
2730       // Our mask-compare operation will be approximate unless we have the
2731       // exact same operation on both sides of the alternation.
2732       pos->determines_perfectly = false;
2733     }
2734     pos->mask &= other_pos->mask;
2735     pos->value &= pos->mask;
2736     other_pos->value &= pos->mask;
2737     uc16 differing_bits = (pos->value ^ other_pos->value);
2738     pos->mask &= ~differing_bits;
2739     pos->value &= pos->mask;
2740   }
2741 }
2742 
2743 
2744 class VisitMarker {
2745  public:
VisitMarker(NodeInfo * info)2746   explicit VisitMarker(NodeInfo* info) : info_(info) {
2747     ASSERT(!info->visited);
2748     info->visited = true;
2749   }
~VisitMarker()2750   ~VisitMarker() {
2751     info_->visited = false;
2752   }
2753  private:
2754   NodeInfo* info_;
2755 };
2756 
2757 
FilterASCII(int depth,bool ignore_case)2758 RegExpNode* SeqRegExpNode::FilterASCII(int depth, bool ignore_case) {
2759   if (info()->replacement_calculated) return replacement();
2760   if (depth < 0) return this;
2761   ASSERT(!info()->visited);
2762   VisitMarker marker(info());
2763   return FilterSuccessor(depth - 1, ignore_case);
2764 }
2765 
2766 
FilterSuccessor(int depth,bool ignore_case)2767 RegExpNode* SeqRegExpNode::FilterSuccessor(int depth, bool ignore_case) {
2768   RegExpNode* next = on_success_->FilterASCII(depth - 1, ignore_case);
2769   if (next == NULL) return set_replacement(NULL);
2770   on_success_ = next;
2771   return set_replacement(this);
2772 }
2773 
2774 
2775 // We need to check for the following characters: 0x39c 0x3bc 0x178.
RangeContainsLatin1Equivalents(CharacterRange range)2776 static inline bool RangeContainsLatin1Equivalents(CharacterRange range) {
2777   // TODO(dcarney): this could be a lot more efficient.
2778   return range.Contains(0x39c) ||
2779       range.Contains(0x3bc) || range.Contains(0x178);
2780 }
2781 
2782 
RangesContainLatin1Equivalents(ZoneList<CharacterRange> * ranges)2783 static bool RangesContainLatin1Equivalents(ZoneList<CharacterRange>* ranges) {
2784   for (int i = 0; i < ranges->length(); i++) {
2785     // TODO(dcarney): this could be a lot more efficient.
2786     if (RangeContainsLatin1Equivalents(ranges->at(i))) return true;
2787   }
2788   return false;
2789 }
2790 
2791 
FilterASCII(int depth,bool ignore_case)2792 RegExpNode* TextNode::FilterASCII(int depth, bool ignore_case) {
2793   if (info()->replacement_calculated) return replacement();
2794   if (depth < 0) return this;
2795   ASSERT(!info()->visited);
2796   VisitMarker marker(info());
2797   int element_count = elms_->length();
2798   for (int i = 0; i < element_count; i++) {
2799     TextElement elm = elms_->at(i);
2800     if (elm.text_type() == TextElement::ATOM) {
2801       Vector<const uc16> quarks = elm.atom()->data();
2802       for (int j = 0; j < quarks.length(); j++) {
2803         uint16_t c = quarks[j];
2804         if (c <= String::kMaxOneByteCharCode) continue;
2805         if (!ignore_case) return set_replacement(NULL);
2806         // Here, we need to check for characters whose upper and lower cases
2807         // are outside the Latin-1 range.
2808         uint16_t converted = unibrow::Latin1::ConvertNonLatin1ToLatin1(c);
2809         // Character is outside Latin-1 completely
2810         if (converted == 0) return set_replacement(NULL);
2811         // Convert quark to Latin-1 in place.
2812         uint16_t* copy = const_cast<uint16_t*>(quarks.start());
2813         copy[j] = converted;
2814       }
2815     } else {
2816       ASSERT(elm.text_type() == TextElement::CHAR_CLASS);
2817       RegExpCharacterClass* cc = elm.char_class();
2818       ZoneList<CharacterRange>* ranges = cc->ranges(zone());
2819       if (!CharacterRange::IsCanonical(ranges)) {
2820         CharacterRange::Canonicalize(ranges);
2821       }
2822       // Now they are in order so we only need to look at the first.
2823       int range_count = ranges->length();
2824       if (cc->is_negated()) {
2825         if (range_count != 0 &&
2826             ranges->at(0).from() == 0 &&
2827             ranges->at(0).to() >= String::kMaxOneByteCharCode) {
2828           // This will be handled in a later filter.
2829           if (ignore_case && RangesContainLatin1Equivalents(ranges)) continue;
2830           return set_replacement(NULL);
2831         }
2832       } else {
2833         if (range_count == 0 ||
2834             ranges->at(0).from() > String::kMaxOneByteCharCode) {
2835           // This will be handled in a later filter.
2836           if (ignore_case && RangesContainLatin1Equivalents(ranges)) continue;
2837           return set_replacement(NULL);
2838         }
2839       }
2840     }
2841   }
2842   return FilterSuccessor(depth - 1, ignore_case);
2843 }
2844 
2845 
FilterASCII(int depth,bool ignore_case)2846 RegExpNode* LoopChoiceNode::FilterASCII(int depth, bool ignore_case) {
2847   if (info()->replacement_calculated) return replacement();
2848   if (depth < 0) return this;
2849   if (info()->visited) return this;
2850   {
2851     VisitMarker marker(info());
2852 
2853     RegExpNode* continue_replacement =
2854         continue_node_->FilterASCII(depth - 1, ignore_case);
2855     // If we can't continue after the loop then there is no sense in doing the
2856     // loop.
2857     if (continue_replacement == NULL) return set_replacement(NULL);
2858   }
2859 
2860   return ChoiceNode::FilterASCII(depth - 1, ignore_case);
2861 }
2862 
2863 
FilterASCII(int depth,bool ignore_case)2864 RegExpNode* ChoiceNode::FilterASCII(int depth, bool ignore_case) {
2865   if (info()->replacement_calculated) return replacement();
2866   if (depth < 0) return this;
2867   if (info()->visited) return this;
2868   VisitMarker marker(info());
2869   int choice_count = alternatives_->length();
2870 
2871   for (int i = 0; i < choice_count; i++) {
2872     GuardedAlternative alternative = alternatives_->at(i);
2873     if (alternative.guards() != NULL && alternative.guards()->length() != 0) {
2874       set_replacement(this);
2875       return this;
2876     }
2877   }
2878 
2879   int surviving = 0;
2880   RegExpNode* survivor = NULL;
2881   for (int i = 0; i < choice_count; i++) {
2882     GuardedAlternative alternative = alternatives_->at(i);
2883     RegExpNode* replacement =
2884         alternative.node()->FilterASCII(depth - 1, ignore_case);
2885     ASSERT(replacement != this);  // No missing EMPTY_MATCH_CHECK.
2886     if (replacement != NULL) {
2887       alternatives_->at(i).set_node(replacement);
2888       surviving++;
2889       survivor = replacement;
2890     }
2891   }
2892   if (surviving < 2) return set_replacement(survivor);
2893 
2894   set_replacement(this);
2895   if (surviving == choice_count) {
2896     return this;
2897   }
2898   // Only some of the nodes survived the filtering.  We need to rebuild the
2899   // alternatives list.
2900   ZoneList<GuardedAlternative>* new_alternatives =
2901       new(zone()) ZoneList<GuardedAlternative>(surviving, zone());
2902   for (int i = 0; i < choice_count; i++) {
2903     RegExpNode* replacement =
2904         alternatives_->at(i).node()->FilterASCII(depth - 1, ignore_case);
2905     if (replacement != NULL) {
2906       alternatives_->at(i).set_node(replacement);
2907       new_alternatives->Add(alternatives_->at(i), zone());
2908     }
2909   }
2910   alternatives_ = new_alternatives;
2911   return this;
2912 }
2913 
2914 
FilterASCII(int depth,bool ignore_case)2915 RegExpNode* NegativeLookaheadChoiceNode::FilterASCII(int depth,
2916                                                      bool ignore_case) {
2917   if (info()->replacement_calculated) return replacement();
2918   if (depth < 0) return this;
2919   if (info()->visited) return this;
2920   VisitMarker marker(info());
2921   // Alternative 0 is the negative lookahead, alternative 1 is what comes
2922   // afterwards.
2923   RegExpNode* node = alternatives_->at(1).node();
2924   RegExpNode* replacement = node->FilterASCII(depth - 1, ignore_case);
2925   if (replacement == NULL) return set_replacement(NULL);
2926   alternatives_->at(1).set_node(replacement);
2927 
2928   RegExpNode* neg_node = alternatives_->at(0).node();
2929   RegExpNode* neg_replacement = neg_node->FilterASCII(depth - 1, ignore_case);
2930   // If the negative lookahead is always going to fail then
2931   // we don't need to check it.
2932   if (neg_replacement == NULL) return set_replacement(replacement);
2933   alternatives_->at(0).set_node(neg_replacement);
2934   return set_replacement(this);
2935 }
2936 
2937 
GetQuickCheckDetails(QuickCheckDetails * details,RegExpCompiler * compiler,int characters_filled_in,bool not_at_start)2938 void LoopChoiceNode::GetQuickCheckDetails(QuickCheckDetails* details,
2939                                           RegExpCompiler* compiler,
2940                                           int characters_filled_in,
2941                                           bool not_at_start) {
2942   if (body_can_be_zero_length_ || info()->visited) return;
2943   VisitMarker marker(info());
2944   return ChoiceNode::GetQuickCheckDetails(details,
2945                                           compiler,
2946                                           characters_filled_in,
2947                                           not_at_start);
2948 }
2949 
2950 
FillInBMInfo(int offset,int budget,BoyerMooreLookahead * bm,bool not_at_start)2951 void LoopChoiceNode::FillInBMInfo(int offset,
2952                                   int budget,
2953                                   BoyerMooreLookahead* bm,
2954                                   bool not_at_start) {
2955   if (body_can_be_zero_length_ || budget <= 0) {
2956     bm->SetRest(offset);
2957     SaveBMInfo(bm, not_at_start, offset);
2958     return;
2959   }
2960   ChoiceNode::FillInBMInfo(offset, budget - 1, bm, not_at_start);
2961   SaveBMInfo(bm, not_at_start, offset);
2962 }
2963 
2964 
GetQuickCheckDetails(QuickCheckDetails * details,RegExpCompiler * compiler,int characters_filled_in,bool not_at_start)2965 void ChoiceNode::GetQuickCheckDetails(QuickCheckDetails* details,
2966                                       RegExpCompiler* compiler,
2967                                       int characters_filled_in,
2968                                       bool not_at_start) {
2969   not_at_start = (not_at_start || not_at_start_);
2970   int choice_count = alternatives_->length();
2971   ASSERT(choice_count > 0);
2972   alternatives_->at(0).node()->GetQuickCheckDetails(details,
2973                                                     compiler,
2974                                                     characters_filled_in,
2975                                                     not_at_start);
2976   for (int i = 1; i < choice_count; i++) {
2977     QuickCheckDetails new_details(details->characters());
2978     RegExpNode* node = alternatives_->at(i).node();
2979     node->GetQuickCheckDetails(&new_details, compiler,
2980                                characters_filled_in,
2981                                not_at_start);
2982     // Here we merge the quick match details of the two branches.
2983     details->Merge(&new_details, characters_filled_in);
2984   }
2985 }
2986 
2987 
2988 // Check for [0-9A-Z_a-z].
EmitWordCheck(RegExpMacroAssembler * assembler,Label * word,Label * non_word,bool fall_through_on_word)2989 static void EmitWordCheck(RegExpMacroAssembler* assembler,
2990                           Label* word,
2991                           Label* non_word,
2992                           bool fall_through_on_word) {
2993   if (assembler->CheckSpecialCharacterClass(
2994           fall_through_on_word ? 'w' : 'W',
2995           fall_through_on_word ? non_word : word)) {
2996     // Optimized implementation available.
2997     return;
2998   }
2999   assembler->CheckCharacterGT('z', non_word);
3000   assembler->CheckCharacterLT('0', non_word);
3001   assembler->CheckCharacterGT('a' - 1, word);
3002   assembler->CheckCharacterLT('9' + 1, word);
3003   assembler->CheckCharacterLT('A', non_word);
3004   assembler->CheckCharacterLT('Z' + 1, word);
3005   if (fall_through_on_word) {
3006     assembler->CheckNotCharacter('_', non_word);
3007   } else {
3008     assembler->CheckCharacter('_', word);
3009   }
3010 }
3011 
3012 
3013 // Emit the code to check for a ^ in multiline mode (1-character lookbehind
3014 // that matches newline or the start of input).
EmitHat(RegExpCompiler * compiler,RegExpNode * on_success,Trace * trace)3015 static void EmitHat(RegExpCompiler* compiler,
3016                     RegExpNode* on_success,
3017                     Trace* trace) {
3018   RegExpMacroAssembler* assembler = compiler->macro_assembler();
3019   // We will be loading the previous character into the current character
3020   // register.
3021   Trace new_trace(*trace);
3022   new_trace.InvalidateCurrentCharacter();
3023 
3024   Label ok;
3025   if (new_trace.cp_offset() == 0) {
3026     // The start of input counts as a newline in this context, so skip to
3027     // ok if we are at the start.
3028     assembler->CheckAtStart(&ok);
3029   }
3030   // We already checked that we are not at the start of input so it must be
3031   // OK to load the previous character.
3032   assembler->LoadCurrentCharacter(new_trace.cp_offset() -1,
3033                                   new_trace.backtrack(),
3034                                   false);
3035   if (!assembler->CheckSpecialCharacterClass('n',
3036                                              new_trace.backtrack())) {
3037     // Newline means \n, \r, 0x2028 or 0x2029.
3038     if (!compiler->ascii()) {
3039       assembler->CheckCharacterAfterAnd(0x2028, 0xfffe, &ok);
3040     }
3041     assembler->CheckCharacter('\n', &ok);
3042     assembler->CheckNotCharacter('\r', new_trace.backtrack());
3043   }
3044   assembler->Bind(&ok);
3045   on_success->Emit(compiler, &new_trace);
3046 }
3047 
3048 
3049 // Emit the code to handle \b and \B (word-boundary or non-word-boundary).
EmitBoundaryCheck(RegExpCompiler * compiler,Trace * trace)3050 void AssertionNode::EmitBoundaryCheck(RegExpCompiler* compiler, Trace* trace) {
3051   RegExpMacroAssembler* assembler = compiler->macro_assembler();
3052   Trace::TriBool next_is_word_character = Trace::UNKNOWN;
3053   bool not_at_start = (trace->at_start() == Trace::FALSE_VALUE);
3054   BoyerMooreLookahead* lookahead = bm_info(not_at_start);
3055   if (lookahead == NULL) {
3056     int eats_at_least =
3057         Min(kMaxLookaheadForBoyerMoore, EatsAtLeast(kMaxLookaheadForBoyerMoore,
3058                                                     kRecursionBudget,
3059                                                     not_at_start));
3060     if (eats_at_least >= 1) {
3061       BoyerMooreLookahead* bm =
3062           new(zone()) BoyerMooreLookahead(eats_at_least, compiler, zone());
3063       FillInBMInfo(0, kRecursionBudget, bm, not_at_start);
3064       if (bm->at(0)->is_non_word())
3065         next_is_word_character = Trace::FALSE_VALUE;
3066       if (bm->at(0)->is_word()) next_is_word_character = Trace::TRUE_VALUE;
3067     }
3068   } else {
3069     if (lookahead->at(0)->is_non_word())
3070       next_is_word_character = Trace::FALSE_VALUE;
3071     if (lookahead->at(0)->is_word())
3072       next_is_word_character = Trace::TRUE_VALUE;
3073   }
3074   bool at_boundary = (assertion_type_ == AssertionNode::AT_BOUNDARY);
3075   if (next_is_word_character == Trace::UNKNOWN) {
3076     Label before_non_word;
3077     Label before_word;
3078     if (trace->characters_preloaded() != 1) {
3079       assembler->LoadCurrentCharacter(trace->cp_offset(), &before_non_word);
3080     }
3081     // Fall through on non-word.
3082     EmitWordCheck(assembler, &before_word, &before_non_word, false);
3083     // Next character is not a word character.
3084     assembler->Bind(&before_non_word);
3085     Label ok;
3086     BacktrackIfPrevious(compiler, trace, at_boundary ? kIsNonWord : kIsWord);
3087     assembler->GoTo(&ok);
3088 
3089     assembler->Bind(&before_word);
3090     BacktrackIfPrevious(compiler, trace, at_boundary ? kIsWord : kIsNonWord);
3091     assembler->Bind(&ok);
3092   } else if (next_is_word_character == Trace::TRUE_VALUE) {
3093     BacktrackIfPrevious(compiler, trace, at_boundary ? kIsWord : kIsNonWord);
3094   } else {
3095     ASSERT(next_is_word_character == Trace::FALSE_VALUE);
3096     BacktrackIfPrevious(compiler, trace, at_boundary ? kIsNonWord : kIsWord);
3097   }
3098 }
3099 
3100 
BacktrackIfPrevious(RegExpCompiler * compiler,Trace * trace,AssertionNode::IfPrevious backtrack_if_previous)3101 void AssertionNode::BacktrackIfPrevious(
3102     RegExpCompiler* compiler,
3103     Trace* trace,
3104     AssertionNode::IfPrevious backtrack_if_previous) {
3105   RegExpMacroAssembler* assembler = compiler->macro_assembler();
3106   Trace new_trace(*trace);
3107   new_trace.InvalidateCurrentCharacter();
3108 
3109   Label fall_through, dummy;
3110 
3111   Label* non_word = backtrack_if_previous == kIsNonWord ?
3112                     new_trace.backtrack() :
3113                     &fall_through;
3114   Label* word = backtrack_if_previous == kIsNonWord ?
3115                 &fall_through :
3116                 new_trace.backtrack();
3117 
3118   if (new_trace.cp_offset() == 0) {
3119     // The start of input counts as a non-word character, so the question is
3120     // decided if we are at the start.
3121     assembler->CheckAtStart(non_word);
3122   }
3123   // We already checked that we are not at the start of input so it must be
3124   // OK to load the previous character.
3125   assembler->LoadCurrentCharacter(new_trace.cp_offset() - 1, &dummy, false);
3126   EmitWordCheck(assembler, word, non_word, backtrack_if_previous == kIsNonWord);
3127 
3128   assembler->Bind(&fall_through);
3129   on_success()->Emit(compiler, &new_trace);
3130 }
3131 
3132 
GetQuickCheckDetails(QuickCheckDetails * details,RegExpCompiler * compiler,int filled_in,bool not_at_start)3133 void AssertionNode::GetQuickCheckDetails(QuickCheckDetails* details,
3134                                          RegExpCompiler* compiler,
3135                                          int filled_in,
3136                                          bool not_at_start) {
3137   if (assertion_type_ == AT_START && not_at_start) {
3138     details->set_cannot_match();
3139     return;
3140   }
3141   return on_success()->GetQuickCheckDetails(details,
3142                                             compiler,
3143                                             filled_in,
3144                                             not_at_start);
3145 }
3146 
3147 
Emit(RegExpCompiler * compiler,Trace * trace)3148 void AssertionNode::Emit(RegExpCompiler* compiler, Trace* trace) {
3149   RegExpMacroAssembler* assembler = compiler->macro_assembler();
3150   switch (assertion_type_) {
3151     case AT_END: {
3152       Label ok;
3153       assembler->CheckPosition(trace->cp_offset(), &ok);
3154       assembler->GoTo(trace->backtrack());
3155       assembler->Bind(&ok);
3156       break;
3157     }
3158     case AT_START: {
3159       if (trace->at_start() == Trace::FALSE_VALUE) {
3160         assembler->GoTo(trace->backtrack());
3161         return;
3162       }
3163       if (trace->at_start() == Trace::UNKNOWN) {
3164         assembler->CheckNotAtStart(trace->backtrack());
3165         Trace at_start_trace = *trace;
3166         at_start_trace.set_at_start(true);
3167         on_success()->Emit(compiler, &at_start_trace);
3168         return;
3169       }
3170     }
3171     break;
3172     case AFTER_NEWLINE:
3173       EmitHat(compiler, on_success(), trace);
3174       return;
3175     case AT_BOUNDARY:
3176     case AT_NON_BOUNDARY: {
3177       EmitBoundaryCheck(compiler, trace);
3178       return;
3179     }
3180   }
3181   on_success()->Emit(compiler, trace);
3182 }
3183 
3184 
DeterminedAlready(QuickCheckDetails * quick_check,int offset)3185 static bool DeterminedAlready(QuickCheckDetails* quick_check, int offset) {
3186   if (quick_check == NULL) return false;
3187   if (offset >= quick_check->characters()) return false;
3188   return quick_check->positions(offset)->determines_perfectly;
3189 }
3190 
3191 
UpdateBoundsCheck(int index,int * checked_up_to)3192 static void UpdateBoundsCheck(int index, int* checked_up_to) {
3193   if (index > *checked_up_to) {
3194     *checked_up_to = index;
3195   }
3196 }
3197 
3198 
3199 // We call this repeatedly to generate code for each pass over the text node.
3200 // The passes are in increasing order of difficulty because we hope one
3201 // of the first passes will fail in which case we are saved the work of the
3202 // later passes.  for example for the case independent regexp /%[asdfghjkl]a/
3203 // we will check the '%' in the first pass, the case independent 'a' in the
3204 // second pass and the character class in the last pass.
3205 //
3206 // The passes are done from right to left, so for example to test for /bar/
3207 // we will first test for an 'r' with offset 2, then an 'a' with offset 1
3208 // and then a 'b' with offset 0.  This means we can avoid the end-of-input
3209 // bounds check most of the time.  In the example we only need to check for
3210 // end-of-input when loading the putative 'r'.
3211 //
3212 // A slight complication involves the fact that the first character may already
3213 // be fetched into a register by the previous node.  In this case we want to
3214 // do the test for that character first.  We do this in separate passes.  The
3215 // 'preloaded' argument indicates that we are doing such a 'pass'.  If such a
3216 // pass has been performed then subsequent passes will have true in
3217 // first_element_checked to indicate that that character does not need to be
3218 // checked again.
3219 //
3220 // In addition to all this we are passed a Trace, which can
3221 // contain an AlternativeGeneration object.  In this AlternativeGeneration
3222 // object we can see details of any quick check that was already passed in
3223 // order to get to the code we are now generating.  The quick check can involve
3224 // loading characters, which means we do not need to recheck the bounds
3225 // up to the limit the quick check already checked.  In addition the quick
3226 // check can have involved a mask and compare operation which may simplify
3227 // or obviate the need for further checks at some character positions.
TextEmitPass(RegExpCompiler * compiler,TextEmitPassType pass,bool preloaded,Trace * trace,bool first_element_checked,int * checked_up_to)3228 void TextNode::TextEmitPass(RegExpCompiler* compiler,
3229                             TextEmitPassType pass,
3230                             bool preloaded,
3231                             Trace* trace,
3232                             bool first_element_checked,
3233                             int* checked_up_to) {
3234   RegExpMacroAssembler* assembler = compiler->macro_assembler();
3235   Isolate* isolate = assembler->zone()->isolate();
3236   bool ascii = compiler->ascii();
3237   Label* backtrack = trace->backtrack();
3238   QuickCheckDetails* quick_check = trace->quick_check_performed();
3239   int element_count = elms_->length();
3240   for (int i = preloaded ? 0 : element_count - 1; i >= 0; i--) {
3241     TextElement elm = elms_->at(i);
3242     int cp_offset = trace->cp_offset() + elm.cp_offset();
3243     if (elm.text_type() == TextElement::ATOM) {
3244       Vector<const uc16> quarks = elm.atom()->data();
3245       for (int j = preloaded ? 0 : quarks.length() - 1; j >= 0; j--) {
3246         if (first_element_checked && i == 0 && j == 0) continue;
3247         if (DeterminedAlready(quick_check, elm.cp_offset() + j)) continue;
3248         EmitCharacterFunction* emit_function = NULL;
3249         switch (pass) {
3250           case NON_ASCII_MATCH:
3251             ASSERT(ascii);
3252             if (quarks[j] > String::kMaxOneByteCharCode) {
3253               assembler->GoTo(backtrack);
3254               return;
3255             }
3256             break;
3257           case NON_LETTER_CHARACTER_MATCH:
3258             emit_function = &EmitAtomNonLetter;
3259             break;
3260           case SIMPLE_CHARACTER_MATCH:
3261             emit_function = &EmitSimpleCharacter;
3262             break;
3263           case CASE_CHARACTER_MATCH:
3264             emit_function = &EmitAtomLetter;
3265             break;
3266           default:
3267             break;
3268         }
3269         if (emit_function != NULL) {
3270           bool bound_checked = emit_function(isolate,
3271                                              compiler,
3272                                              quarks[j],
3273                                              backtrack,
3274                                              cp_offset + j,
3275                                              *checked_up_to < cp_offset + j,
3276                                              preloaded);
3277           if (bound_checked) UpdateBoundsCheck(cp_offset + j, checked_up_to);
3278         }
3279       }
3280     } else {
3281       ASSERT_EQ(TextElement::CHAR_CLASS, elm.text_type());
3282       if (pass == CHARACTER_CLASS_MATCH) {
3283         if (first_element_checked && i == 0) continue;
3284         if (DeterminedAlready(quick_check, elm.cp_offset())) continue;
3285         RegExpCharacterClass* cc = elm.char_class();
3286         EmitCharClass(assembler,
3287                       cc,
3288                       ascii,
3289                       backtrack,
3290                       cp_offset,
3291                       *checked_up_to < cp_offset,
3292                       preloaded,
3293                       zone());
3294         UpdateBoundsCheck(cp_offset, checked_up_to);
3295       }
3296     }
3297   }
3298 }
3299 
3300 
Length()3301 int TextNode::Length() {
3302   TextElement elm = elms_->last();
3303   ASSERT(elm.cp_offset() >= 0);
3304   return elm.cp_offset() + elm.length();
3305 }
3306 
3307 
SkipPass(int int_pass,bool ignore_case)3308 bool TextNode::SkipPass(int int_pass, bool ignore_case) {
3309   TextEmitPassType pass = static_cast<TextEmitPassType>(int_pass);
3310   if (ignore_case) {
3311     return pass == SIMPLE_CHARACTER_MATCH;
3312   } else {
3313     return pass == NON_LETTER_CHARACTER_MATCH || pass == CASE_CHARACTER_MATCH;
3314   }
3315 }
3316 
3317 
3318 // This generates the code to match a text node.  A text node can contain
3319 // straight character sequences (possibly to be matched in a case-independent
3320 // way) and character classes.  For efficiency we do not do this in a single
3321 // pass from left to right.  Instead we pass over the text node several times,
3322 // emitting code for some character positions every time.  See the comment on
3323 // TextEmitPass for details.
Emit(RegExpCompiler * compiler,Trace * trace)3324 void TextNode::Emit(RegExpCompiler* compiler, Trace* trace) {
3325   LimitResult limit_result = LimitVersions(compiler, trace);
3326   if (limit_result == DONE) return;
3327   ASSERT(limit_result == CONTINUE);
3328 
3329   if (trace->cp_offset() + Length() > RegExpMacroAssembler::kMaxCPOffset) {
3330     compiler->SetRegExpTooBig();
3331     return;
3332   }
3333 
3334   if (compiler->ascii()) {
3335     int dummy = 0;
3336     TextEmitPass(compiler, NON_ASCII_MATCH, false, trace, false, &dummy);
3337   }
3338 
3339   bool first_elt_done = false;
3340   int bound_checked_to = trace->cp_offset() - 1;
3341   bound_checked_to += trace->bound_checked_up_to();
3342 
3343   // If a character is preloaded into the current character register then
3344   // check that now.
3345   if (trace->characters_preloaded() == 1) {
3346     for (int pass = kFirstRealPass; pass <= kLastPass; pass++) {
3347       if (!SkipPass(pass, compiler->ignore_case())) {
3348         TextEmitPass(compiler,
3349                      static_cast<TextEmitPassType>(pass),
3350                      true,
3351                      trace,
3352                      false,
3353                      &bound_checked_to);
3354       }
3355     }
3356     first_elt_done = true;
3357   }
3358 
3359   for (int pass = kFirstRealPass; pass <= kLastPass; pass++) {
3360     if (!SkipPass(pass, compiler->ignore_case())) {
3361       TextEmitPass(compiler,
3362                    static_cast<TextEmitPassType>(pass),
3363                    false,
3364                    trace,
3365                    first_elt_done,
3366                    &bound_checked_to);
3367     }
3368   }
3369 
3370   Trace successor_trace(*trace);
3371   successor_trace.set_at_start(false);
3372   successor_trace.AdvanceCurrentPositionInTrace(Length(), compiler);
3373   RecursionCheck rc(compiler);
3374   on_success()->Emit(compiler, &successor_trace);
3375 }
3376 
3377 
InvalidateCurrentCharacter()3378 void Trace::InvalidateCurrentCharacter() {
3379   characters_preloaded_ = 0;
3380 }
3381 
3382 
AdvanceCurrentPositionInTrace(int by,RegExpCompiler * compiler)3383 void Trace::AdvanceCurrentPositionInTrace(int by, RegExpCompiler* compiler) {
3384   ASSERT(by > 0);
3385   // We don't have an instruction for shifting the current character register
3386   // down or for using a shifted value for anything so lets just forget that
3387   // we preloaded any characters into it.
3388   characters_preloaded_ = 0;
3389   // Adjust the offsets of the quick check performed information.  This
3390   // information is used to find out what we already determined about the
3391   // characters by means of mask and compare.
3392   quick_check_performed_.Advance(by, compiler->ascii());
3393   cp_offset_ += by;
3394   if (cp_offset_ > RegExpMacroAssembler::kMaxCPOffset) {
3395     compiler->SetRegExpTooBig();
3396     cp_offset_ = 0;
3397   }
3398   bound_checked_up_to_ = Max(0, bound_checked_up_to_ - by);
3399 }
3400 
3401 
MakeCaseIndependent(bool is_ascii)3402 void TextNode::MakeCaseIndependent(bool is_ascii) {
3403   int element_count = elms_->length();
3404   for (int i = 0; i < element_count; i++) {
3405     TextElement elm = elms_->at(i);
3406     if (elm.text_type() == TextElement::CHAR_CLASS) {
3407       RegExpCharacterClass* cc = elm.char_class();
3408       // None of the standard character classes is different in the case
3409       // independent case and it slows us down if we don't know that.
3410       if (cc->is_standard(zone())) continue;
3411       ZoneList<CharacterRange>* ranges = cc->ranges(zone());
3412       int range_count = ranges->length();
3413       for (int j = 0; j < range_count; j++) {
3414         ranges->at(j).AddCaseEquivalents(ranges, is_ascii, zone());
3415       }
3416     }
3417   }
3418 }
3419 
3420 
GreedyLoopTextLength()3421 int TextNode::GreedyLoopTextLength() {
3422   TextElement elm = elms_->at(elms_->length() - 1);
3423   return elm.cp_offset() + elm.length();
3424 }
3425 
3426 
GetSuccessorOfOmnivorousTextNode(RegExpCompiler * compiler)3427 RegExpNode* TextNode::GetSuccessorOfOmnivorousTextNode(
3428     RegExpCompiler* compiler) {
3429   if (elms_->length() != 1) return NULL;
3430   TextElement elm = elms_->at(0);
3431   if (elm.text_type() != TextElement::CHAR_CLASS) return NULL;
3432   RegExpCharacterClass* node = elm.char_class();
3433   ZoneList<CharacterRange>* ranges = node->ranges(zone());
3434   if (!CharacterRange::IsCanonical(ranges)) {
3435     CharacterRange::Canonicalize(ranges);
3436   }
3437   if (node->is_negated()) {
3438     return ranges->length() == 0 ? on_success() : NULL;
3439   }
3440   if (ranges->length() != 1) return NULL;
3441   uint32_t max_char;
3442   if (compiler->ascii()) {
3443     max_char = String::kMaxOneByteCharCode;
3444   } else {
3445     max_char = String::kMaxUtf16CodeUnit;
3446   }
3447   return ranges->at(0).IsEverything(max_char) ? on_success() : NULL;
3448 }
3449 
3450 
3451 // Finds the fixed match length of a sequence of nodes that goes from
3452 // this alternative and back to this choice node.  If there are variable
3453 // length nodes or other complications in the way then return a sentinel
3454 // value indicating that a greedy loop cannot be constructed.
GreedyLoopTextLengthForAlternative(GuardedAlternative * alternative)3455 int ChoiceNode::GreedyLoopTextLengthForAlternative(
3456     GuardedAlternative* alternative) {
3457   int length = 0;
3458   RegExpNode* node = alternative->node();
3459   // Later we will generate code for all these text nodes using recursion
3460   // so we have to limit the max number.
3461   int recursion_depth = 0;
3462   while (node != this) {
3463     if (recursion_depth++ > RegExpCompiler::kMaxRecursion) {
3464       return kNodeIsTooComplexForGreedyLoops;
3465     }
3466     int node_length = node->GreedyLoopTextLength();
3467     if (node_length == kNodeIsTooComplexForGreedyLoops) {
3468       return kNodeIsTooComplexForGreedyLoops;
3469     }
3470     length += node_length;
3471     SeqRegExpNode* seq_node = static_cast<SeqRegExpNode*>(node);
3472     node = seq_node->on_success();
3473   }
3474   return length;
3475 }
3476 
3477 
AddLoopAlternative(GuardedAlternative alt)3478 void LoopChoiceNode::AddLoopAlternative(GuardedAlternative alt) {
3479   ASSERT_EQ(loop_node_, NULL);
3480   AddAlternative(alt);
3481   loop_node_ = alt.node();
3482 }
3483 
3484 
AddContinueAlternative(GuardedAlternative alt)3485 void LoopChoiceNode::AddContinueAlternative(GuardedAlternative alt) {
3486   ASSERT_EQ(continue_node_, NULL);
3487   AddAlternative(alt);
3488   continue_node_ = alt.node();
3489 }
3490 
3491 
Emit(RegExpCompiler * compiler,Trace * trace)3492 void LoopChoiceNode::Emit(RegExpCompiler* compiler, Trace* trace) {
3493   RegExpMacroAssembler* macro_assembler = compiler->macro_assembler();
3494   if (trace->stop_node() == this) {
3495     int text_length =
3496         GreedyLoopTextLengthForAlternative(&(alternatives_->at(0)));
3497     ASSERT(text_length != kNodeIsTooComplexForGreedyLoops);
3498     // Update the counter-based backtracking info on the stack.  This is an
3499     // optimization for greedy loops (see below).
3500     ASSERT(trace->cp_offset() == text_length);
3501     macro_assembler->AdvanceCurrentPosition(text_length);
3502     macro_assembler->GoTo(trace->loop_label());
3503     return;
3504   }
3505   ASSERT(trace->stop_node() == NULL);
3506   if (!trace->is_trivial()) {
3507     trace->Flush(compiler, this);
3508     return;
3509   }
3510   ChoiceNode::Emit(compiler, trace);
3511 }
3512 
3513 
CalculatePreloadCharacters(RegExpCompiler * compiler,int eats_at_least)3514 int ChoiceNode::CalculatePreloadCharacters(RegExpCompiler* compiler,
3515                                            int eats_at_least) {
3516   int preload_characters = Min(4, eats_at_least);
3517   if (compiler->macro_assembler()->CanReadUnaligned()) {
3518     bool ascii = compiler->ascii();
3519     if (ascii) {
3520       if (preload_characters > 4) preload_characters = 4;
3521       // We can't preload 3 characters because there is no machine instruction
3522       // to do that.  We can't just load 4 because we could be reading
3523       // beyond the end of the string, which could cause a memory fault.
3524       if (preload_characters == 3) preload_characters = 2;
3525     } else {
3526       if (preload_characters > 2) preload_characters = 2;
3527     }
3528   } else {
3529     if (preload_characters > 1) preload_characters = 1;
3530   }
3531   return preload_characters;
3532 }
3533 
3534 
3535 // This class is used when generating the alternatives in a choice node.  It
3536 // records the way the alternative is being code generated.
3537 class AlternativeGeneration: public Malloced {
3538  public:
AlternativeGeneration()3539   AlternativeGeneration()
3540       : possible_success(),
3541         expects_preload(false),
3542         after(),
3543         quick_check_details() { }
3544   Label possible_success;
3545   bool expects_preload;
3546   Label after;
3547   QuickCheckDetails quick_check_details;
3548 };
3549 
3550 
3551 // Creates a list of AlternativeGenerations.  If the list has a reasonable
3552 // size then it is on the stack, otherwise the excess is on the heap.
3553 class AlternativeGenerationList {
3554  public:
AlternativeGenerationList(int count,Zone * zone)3555   AlternativeGenerationList(int count, Zone* zone)
3556       : alt_gens_(count, zone) {
3557     for (int i = 0; i < count && i < kAFew; i++) {
3558       alt_gens_.Add(a_few_alt_gens_ + i, zone);
3559     }
3560     for (int i = kAFew; i < count; i++) {
3561       alt_gens_.Add(new AlternativeGeneration(), zone);
3562     }
3563   }
~AlternativeGenerationList()3564   ~AlternativeGenerationList() {
3565     for (int i = kAFew; i < alt_gens_.length(); i++) {
3566       delete alt_gens_[i];
3567       alt_gens_[i] = NULL;
3568     }
3569   }
3570 
at(int i)3571   AlternativeGeneration* at(int i) {
3572     return alt_gens_[i];
3573   }
3574 
3575  private:
3576   static const int kAFew = 10;
3577   ZoneList<AlternativeGeneration*> alt_gens_;
3578   AlternativeGeneration a_few_alt_gens_[kAFew];
3579 };
3580 
3581 
3582 // The '2' variant is has inclusive from and exclusive to.
3583 // This covers \s as defined in ECMA-262 5.1, 15.10.2.12,
3584 // which include WhiteSpace (7.2) or LineTerminator (7.3) values.
3585 static const int kSpaceRanges[] = { '\t', '\r' + 1, ' ', ' ' + 1,
3586     0x00A0, 0x00A1, 0x1680, 0x1681, 0x180E, 0x180F, 0x2000, 0x200B,
3587     0x2028, 0x202A, 0x202F, 0x2030, 0x205F, 0x2060, 0x3000, 0x3001,
3588     0xFEFF, 0xFF00, 0x10000 };
3589 static const int kSpaceRangeCount = ARRAY_SIZE(kSpaceRanges);
3590 
3591 static const int kWordRanges[] = {
3592     '0', '9' + 1, 'A', 'Z' + 1, '_', '_' + 1, 'a', 'z' + 1, 0x10000 };
3593 static const int kWordRangeCount = ARRAY_SIZE(kWordRanges);
3594 static const int kDigitRanges[] = { '0', '9' + 1, 0x10000 };
3595 static const int kDigitRangeCount = ARRAY_SIZE(kDigitRanges);
3596 static const int kSurrogateRanges[] = { 0xd800, 0xe000, 0x10000 };
3597 static const int kSurrogateRangeCount = ARRAY_SIZE(kSurrogateRanges);
3598 static const int kLineTerminatorRanges[] = { 0x000A, 0x000B, 0x000D, 0x000E,
3599     0x2028, 0x202A, 0x10000 };
3600 static const int kLineTerminatorRangeCount = ARRAY_SIZE(kLineTerminatorRanges);
3601 
3602 
Set(int character)3603 void BoyerMoorePositionInfo::Set(int character) {
3604   SetInterval(Interval(character, character));
3605 }
3606 
3607 
SetInterval(const Interval & interval)3608 void BoyerMoorePositionInfo::SetInterval(const Interval& interval) {
3609   s_ = AddRange(s_, kSpaceRanges, kSpaceRangeCount, interval);
3610   w_ = AddRange(w_, kWordRanges, kWordRangeCount, interval);
3611   d_ = AddRange(d_, kDigitRanges, kDigitRangeCount, interval);
3612   surrogate_ =
3613       AddRange(surrogate_, kSurrogateRanges, kSurrogateRangeCount, interval);
3614   if (interval.to() - interval.from() >= kMapSize - 1) {
3615     if (map_count_ != kMapSize) {
3616       map_count_ = kMapSize;
3617       for (int i = 0; i < kMapSize; i++) map_->at(i) = true;
3618     }
3619     return;
3620   }
3621   for (int i = interval.from(); i <= interval.to(); i++) {
3622     int mod_character = (i & kMask);
3623     if (!map_->at(mod_character)) {
3624       map_count_++;
3625       map_->at(mod_character) = true;
3626     }
3627     if (map_count_ == kMapSize) return;
3628   }
3629 }
3630 
3631 
SetAll()3632 void BoyerMoorePositionInfo::SetAll() {
3633   s_ = w_ = d_ = kLatticeUnknown;
3634   if (map_count_ != kMapSize) {
3635     map_count_ = kMapSize;
3636     for (int i = 0; i < kMapSize; i++) map_->at(i) = true;
3637   }
3638 }
3639 
3640 
BoyerMooreLookahead(int length,RegExpCompiler * compiler,Zone * zone)3641 BoyerMooreLookahead::BoyerMooreLookahead(
3642     int length, RegExpCompiler* compiler, Zone* zone)
3643     : length_(length),
3644       compiler_(compiler) {
3645   if (compiler->ascii()) {
3646     max_char_ = String::kMaxOneByteCharCode;
3647   } else {
3648     max_char_ = String::kMaxUtf16CodeUnit;
3649   }
3650   bitmaps_ = new(zone) ZoneList<BoyerMoorePositionInfo*>(length, zone);
3651   for (int i = 0; i < length; i++) {
3652     bitmaps_->Add(new(zone) BoyerMoorePositionInfo(zone), zone);
3653   }
3654 }
3655 
3656 
3657 // Find the longest range of lookahead that has the fewest number of different
3658 // characters that can occur at a given position.  Since we are optimizing two
3659 // different parameters at once this is a tradeoff.
FindWorthwhileInterval(int * from,int * to)3660 bool BoyerMooreLookahead::FindWorthwhileInterval(int* from, int* to) {
3661   int biggest_points = 0;
3662   // If more than 32 characters out of 128 can occur it is unlikely that we can
3663   // be lucky enough to step forwards much of the time.
3664   const int kMaxMax = 32;
3665   for (int max_number_of_chars = 4;
3666        max_number_of_chars < kMaxMax;
3667        max_number_of_chars *= 2) {
3668     biggest_points =
3669         FindBestInterval(max_number_of_chars, biggest_points, from, to);
3670   }
3671   if (biggest_points == 0) return false;
3672   return true;
3673 }
3674 
3675 
3676 // Find the highest-points range between 0 and length_ where the character
3677 // information is not too vague.  'Too vague' means that there are more than
3678 // max_number_of_chars that can occur at this position.  Calculates the number
3679 // of points as the product of width-of-the-range and
3680 // probability-of-finding-one-of-the-characters, where the probability is
3681 // calculated using the frequency distribution of the sample subject string.
FindBestInterval(int max_number_of_chars,int old_biggest_points,int * from,int * to)3682 int BoyerMooreLookahead::FindBestInterval(
3683     int max_number_of_chars, int old_biggest_points, int* from, int* to) {
3684   int biggest_points = old_biggest_points;
3685   static const int kSize = RegExpMacroAssembler::kTableSize;
3686   for (int i = 0; i < length_; ) {
3687     while (i < length_ && Count(i) > max_number_of_chars) i++;
3688     if (i == length_) break;
3689     int remembered_from = i;
3690     bool union_map[kSize];
3691     for (int j = 0; j < kSize; j++) union_map[j] = false;
3692     while (i < length_ && Count(i) <= max_number_of_chars) {
3693       BoyerMoorePositionInfo* map = bitmaps_->at(i);
3694       for (int j = 0; j < kSize; j++) union_map[j] |= map->at(j);
3695       i++;
3696     }
3697     int frequency = 0;
3698     for (int j = 0; j < kSize; j++) {
3699       if (union_map[j]) {
3700         // Add 1 to the frequency to give a small per-character boost for
3701         // the cases where our sampling is not good enough and many
3702         // characters have a frequency of zero.  This means the frequency
3703         // can theoretically be up to 2*kSize though we treat it mostly as
3704         // a fraction of kSize.
3705         frequency += compiler_->frequency_collator()->Frequency(j) + 1;
3706       }
3707     }
3708     // We use the probability of skipping times the distance we are skipping to
3709     // judge the effectiveness of this.  Actually we have a cut-off:  By
3710     // dividing by 2 we switch off the skipping if the probability of skipping
3711     // is less than 50%.  This is because the multibyte mask-and-compare
3712     // skipping in quickcheck is more likely to do well on this case.
3713     bool in_quickcheck_range = ((i - remembered_from < 4) ||
3714         (compiler_->ascii() ? remembered_from <= 4 : remembered_from <= 2));
3715     // Called 'probability' but it is only a rough estimate and can actually
3716     // be outside the 0-kSize range.
3717     int probability = (in_quickcheck_range ? kSize / 2 : kSize) - frequency;
3718     int points = (i - remembered_from) * probability;
3719     if (points > biggest_points) {
3720       *from = remembered_from;
3721       *to = i - 1;
3722       biggest_points = points;
3723     }
3724   }
3725   return biggest_points;
3726 }
3727 
3728 
3729 // Take all the characters that will not prevent a successful match if they
3730 // occur in the subject string in the range between min_lookahead and
3731 // max_lookahead (inclusive) measured from the current position.  If the
3732 // character at max_lookahead offset is not one of these characters, then we
3733 // can safely skip forwards by the number of characters in the range.
GetSkipTable(int min_lookahead,int max_lookahead,Handle<ByteArray> boolean_skip_table)3734 int BoyerMooreLookahead::GetSkipTable(int min_lookahead,
3735                                       int max_lookahead,
3736                                       Handle<ByteArray> boolean_skip_table) {
3737   const int kSize = RegExpMacroAssembler::kTableSize;
3738 
3739   const int kSkipArrayEntry = 0;
3740   const int kDontSkipArrayEntry = 1;
3741 
3742   for (int i = 0; i < kSize; i++) {
3743     boolean_skip_table->set(i, kSkipArrayEntry);
3744   }
3745   int skip = max_lookahead + 1 - min_lookahead;
3746 
3747   for (int i = max_lookahead; i >= min_lookahead; i--) {
3748     BoyerMoorePositionInfo* map = bitmaps_->at(i);
3749     for (int j = 0; j < kSize; j++) {
3750       if (map->at(j)) {
3751         boolean_skip_table->set(j, kDontSkipArrayEntry);
3752       }
3753     }
3754   }
3755 
3756   return skip;
3757 }
3758 
3759 
3760 // See comment above on the implementation of GetSkipTable.
EmitSkipInstructions(RegExpMacroAssembler * masm)3761 bool BoyerMooreLookahead::EmitSkipInstructions(RegExpMacroAssembler* masm) {
3762   const int kSize = RegExpMacroAssembler::kTableSize;
3763 
3764   int min_lookahead = 0;
3765   int max_lookahead = 0;
3766 
3767   if (!FindWorthwhileInterval(&min_lookahead, &max_lookahead)) return false;
3768 
3769   bool found_single_character = false;
3770   int single_character = 0;
3771   for (int i = max_lookahead; i >= min_lookahead; i--) {
3772     BoyerMoorePositionInfo* map = bitmaps_->at(i);
3773     if (map->map_count() > 1 ||
3774         (found_single_character && map->map_count() != 0)) {
3775       found_single_character = false;
3776       break;
3777     }
3778     for (int j = 0; j < kSize; j++) {
3779       if (map->at(j)) {
3780         found_single_character = true;
3781         single_character = j;
3782         break;
3783       }
3784     }
3785   }
3786 
3787   int lookahead_width = max_lookahead + 1 - min_lookahead;
3788 
3789   if (found_single_character && lookahead_width == 1 && max_lookahead < 3) {
3790     // The mask-compare can probably handle this better.
3791     return false;
3792   }
3793 
3794   if (found_single_character) {
3795     Label cont, again;
3796     masm->Bind(&again);
3797     masm->LoadCurrentCharacter(max_lookahead, &cont, true);
3798     if (max_char_ > kSize) {
3799       masm->CheckCharacterAfterAnd(single_character,
3800                                    RegExpMacroAssembler::kTableMask,
3801                                    &cont);
3802     } else {
3803       masm->CheckCharacter(single_character, &cont);
3804     }
3805     masm->AdvanceCurrentPosition(lookahead_width);
3806     masm->GoTo(&again);
3807     masm->Bind(&cont);
3808     return true;
3809   }
3810 
3811   Factory* factory = masm->zone()->isolate()->factory();
3812   Handle<ByteArray> boolean_skip_table = factory->NewByteArray(kSize, TENURED);
3813   int skip_distance = GetSkipTable(
3814       min_lookahead, max_lookahead, boolean_skip_table);
3815   ASSERT(skip_distance != 0);
3816 
3817   Label cont, again;
3818   masm->Bind(&again);
3819   masm->LoadCurrentCharacter(max_lookahead, &cont, true);
3820   masm->CheckBitInTable(boolean_skip_table, &cont);
3821   masm->AdvanceCurrentPosition(skip_distance);
3822   masm->GoTo(&again);
3823   masm->Bind(&cont);
3824 
3825   return true;
3826 }
3827 
3828 
3829 /* Code generation for choice nodes.
3830  *
3831  * We generate quick checks that do a mask and compare to eliminate a
3832  * choice.  If the quick check succeeds then it jumps to the continuation to
3833  * do slow checks and check subsequent nodes.  If it fails (the common case)
3834  * it falls through to the next choice.
3835  *
3836  * Here is the desired flow graph.  Nodes directly below each other imply
3837  * fallthrough.  Alternatives 1 and 2 have quick checks.  Alternative
3838  * 3 doesn't have a quick check so we have to call the slow check.
3839  * Nodes are marked Qn for quick checks and Sn for slow checks.  The entire
3840  * regexp continuation is generated directly after the Sn node, up to the
3841  * next GoTo if we decide to reuse some already generated code.  Some
3842  * nodes expect preload_characters to be preloaded into the current
3843  * character register.  R nodes do this preloading.  Vertices are marked
3844  * F for failures and S for success (possible success in the case of quick
3845  * nodes).  L, V, < and > are used as arrow heads.
3846  *
3847  * ----------> R
3848  *             |
3849  *             V
3850  *            Q1 -----> S1
3851  *             |   S   /
3852  *            F|      /
3853  *             |    F/
3854  *             |    /
3855  *             |   R
3856  *             |  /
3857  *             V L
3858  *            Q2 -----> S2
3859  *             |   S   /
3860  *            F|      /
3861  *             |    F/
3862  *             |    /
3863  *             |   R
3864  *             |  /
3865  *             V L
3866  *            S3
3867  *             |
3868  *            F|
3869  *             |
3870  *             R
3871  *             |
3872  * backtrack   V
3873  * <----------Q4
3874  *   \    F    |
3875  *    \        |S
3876  *     \   F   V
3877  *      \-----S4
3878  *
3879  * For greedy loops we reverse our expectation and expect to match rather
3880  * than fail. Therefore we want the loop code to look like this (U is the
3881  * unwind code that steps back in the greedy loop).  The following alternatives
3882  * look the same as above.
3883  *              _____
3884  *             /     \
3885  *             V     |
3886  * ----------> S1    |
3887  *            /|     |
3888  *           / |S    |
3889  *         F/  \_____/
3890  *         /
3891  *        |<-----------
3892  *        |            \
3893  *        V             \
3894  *        Q2 ---> S2     \
3895  *        |  S   /       |
3896  *       F|     /        |
3897  *        |   F/         |
3898  *        |   /          |
3899  *        |  R           |
3900  *        | /            |
3901  *   F    VL             |
3902  * <------U              |
3903  * back   |S             |
3904  *        \______________/
3905  */
3906 
Emit(RegExpCompiler * compiler,Trace * trace)3907 void ChoiceNode::Emit(RegExpCompiler* compiler, Trace* trace) {
3908   RegExpMacroAssembler* macro_assembler = compiler->macro_assembler();
3909   int choice_count = alternatives_->length();
3910 #ifdef DEBUG
3911   for (int i = 0; i < choice_count - 1; i++) {
3912     GuardedAlternative alternative = alternatives_->at(i);
3913     ZoneList<Guard*>* guards = alternative.guards();
3914     int guard_count = (guards == NULL) ? 0 : guards->length();
3915     for (int j = 0; j < guard_count; j++) {
3916       ASSERT(!trace->mentions_reg(guards->at(j)->reg()));
3917     }
3918   }
3919 #endif
3920 
3921   LimitResult limit_result = LimitVersions(compiler, trace);
3922   if (limit_result == DONE) return;
3923   ASSERT(limit_result == CONTINUE);
3924 
3925   int new_flush_budget = trace->flush_budget() / choice_count;
3926   if (trace->flush_budget() == 0 && trace->actions() != NULL) {
3927     trace->Flush(compiler, this);
3928     return;
3929   }
3930 
3931   RecursionCheck rc(compiler);
3932 
3933   Trace* current_trace = trace;
3934 
3935   int text_length = GreedyLoopTextLengthForAlternative(&(alternatives_->at(0)));
3936   bool greedy_loop = false;
3937   Label greedy_loop_label;
3938   Trace counter_backtrack_trace;
3939   counter_backtrack_trace.set_backtrack(&greedy_loop_label);
3940   if (not_at_start()) counter_backtrack_trace.set_at_start(false);
3941 
3942   if (choice_count > 1 && text_length != kNodeIsTooComplexForGreedyLoops) {
3943     // Here we have special handling for greedy loops containing only text nodes
3944     // and other simple nodes.  These are handled by pushing the current
3945     // position on the stack and then incrementing the current position each
3946     // time around the switch.  On backtrack we decrement the current position
3947     // and check it against the pushed value.  This avoids pushing backtrack
3948     // information for each iteration of the loop, which could take up a lot of
3949     // space.
3950     greedy_loop = true;
3951     ASSERT(trace->stop_node() == NULL);
3952     macro_assembler->PushCurrentPosition();
3953     current_trace = &counter_backtrack_trace;
3954     Label greedy_match_failed;
3955     Trace greedy_match_trace;
3956     if (not_at_start()) greedy_match_trace.set_at_start(false);
3957     greedy_match_trace.set_backtrack(&greedy_match_failed);
3958     Label loop_label;
3959     macro_assembler->Bind(&loop_label);
3960     greedy_match_trace.set_stop_node(this);
3961     greedy_match_trace.set_loop_label(&loop_label);
3962     alternatives_->at(0).node()->Emit(compiler, &greedy_match_trace);
3963     macro_assembler->Bind(&greedy_match_failed);
3964   }
3965 
3966   Label second_choice;  // For use in greedy matches.
3967   macro_assembler->Bind(&second_choice);
3968 
3969   int first_normal_choice = greedy_loop ? 1 : 0;
3970 
3971   bool not_at_start = current_trace->at_start() == Trace::FALSE_VALUE;
3972   const int kEatsAtLeastNotYetInitialized = -1;
3973   int eats_at_least = kEatsAtLeastNotYetInitialized;
3974 
3975   bool skip_was_emitted = false;
3976 
3977   if (!greedy_loop && choice_count == 2) {
3978     GuardedAlternative alt1 = alternatives_->at(1);
3979     if (alt1.guards() == NULL || alt1.guards()->length() == 0) {
3980       RegExpNode* eats_anything_node = alt1.node();
3981       if (eats_anything_node->GetSuccessorOfOmnivorousTextNode(compiler) ==
3982           this) {
3983         // At this point we know that we are at a non-greedy loop that will eat
3984         // any character one at a time.  Any non-anchored regexp has such a
3985         // loop prepended to it in order to find where it starts.  We look for
3986         // a pattern of the form ...abc... where we can look 6 characters ahead
3987         // and step forwards 3 if the character is not one of abc.  Abc need
3988         // not be atoms, they can be any reasonably limited character class or
3989         // small alternation.
3990         ASSERT(trace->is_trivial());  // This is the case on LoopChoiceNodes.
3991         BoyerMooreLookahead* lookahead = bm_info(not_at_start);
3992         if (lookahead == NULL) {
3993           eats_at_least = Min(kMaxLookaheadForBoyerMoore,
3994                               EatsAtLeast(kMaxLookaheadForBoyerMoore,
3995                                           kRecursionBudget,
3996                                           not_at_start));
3997           if (eats_at_least >= 1) {
3998             BoyerMooreLookahead* bm =
3999                 new(zone()) BoyerMooreLookahead(eats_at_least,
4000                                                 compiler,
4001                                                 zone());
4002             GuardedAlternative alt0 = alternatives_->at(0);
4003             alt0.node()->FillInBMInfo(0, kRecursionBudget, bm, not_at_start);
4004             skip_was_emitted = bm->EmitSkipInstructions(macro_assembler);
4005           }
4006         } else {
4007           skip_was_emitted = lookahead->EmitSkipInstructions(macro_assembler);
4008         }
4009       }
4010     }
4011   }
4012 
4013   if (eats_at_least == kEatsAtLeastNotYetInitialized) {
4014     // Save some time by looking at most one machine word ahead.
4015     eats_at_least =
4016         EatsAtLeast(compiler->ascii() ? 4 : 2, kRecursionBudget, not_at_start);
4017   }
4018   int preload_characters = CalculatePreloadCharacters(compiler, eats_at_least);
4019 
4020   bool preload_is_current = !skip_was_emitted &&
4021       (current_trace->characters_preloaded() == preload_characters);
4022   bool preload_has_checked_bounds = preload_is_current;
4023 
4024   AlternativeGenerationList alt_gens(choice_count, zone());
4025 
4026   // For now we just call all choices one after the other.  The idea ultimately
4027   // is to use the Dispatch table to try only the relevant ones.
4028   for (int i = first_normal_choice; i < choice_count; i++) {
4029     GuardedAlternative alternative = alternatives_->at(i);
4030     AlternativeGeneration* alt_gen = alt_gens.at(i);
4031     alt_gen->quick_check_details.set_characters(preload_characters);
4032     ZoneList<Guard*>* guards = alternative.guards();
4033     int guard_count = (guards == NULL) ? 0 : guards->length();
4034     Trace new_trace(*current_trace);
4035     new_trace.set_characters_preloaded(preload_is_current ?
4036                                          preload_characters :
4037                                          0);
4038     if (preload_has_checked_bounds) {
4039       new_trace.set_bound_checked_up_to(preload_characters);
4040     }
4041     new_trace.quick_check_performed()->Clear();
4042     if (not_at_start_) new_trace.set_at_start(Trace::FALSE_VALUE);
4043     alt_gen->expects_preload = preload_is_current;
4044     bool generate_full_check_inline = false;
4045     if (FLAG_regexp_optimization &&
4046         try_to_emit_quick_check_for_alternative(i) &&
4047         alternative.node()->EmitQuickCheck(compiler,
4048                                            &new_trace,
4049                                            preload_has_checked_bounds,
4050                                            &alt_gen->possible_success,
4051                                            &alt_gen->quick_check_details,
4052                                            i < choice_count - 1)) {
4053       // Quick check was generated for this choice.
4054       preload_is_current = true;
4055       preload_has_checked_bounds = true;
4056       // On the last choice in the ChoiceNode we generated the quick
4057       // check to fall through on possible success.  So now we need to
4058       // generate the full check inline.
4059       if (i == choice_count - 1) {
4060         macro_assembler->Bind(&alt_gen->possible_success);
4061         new_trace.set_quick_check_performed(&alt_gen->quick_check_details);
4062         new_trace.set_characters_preloaded(preload_characters);
4063         new_trace.set_bound_checked_up_to(preload_characters);
4064         generate_full_check_inline = true;
4065       }
4066     } else if (alt_gen->quick_check_details.cannot_match()) {
4067       if (i == choice_count - 1 && !greedy_loop) {
4068         macro_assembler->GoTo(trace->backtrack());
4069       }
4070       continue;
4071     } else {
4072       // No quick check was generated.  Put the full code here.
4073       // If this is not the first choice then there could be slow checks from
4074       // previous cases that go here when they fail.  There's no reason to
4075       // insist that they preload characters since the slow check we are about
4076       // to generate probably can't use it.
4077       if (i != first_normal_choice) {
4078         alt_gen->expects_preload = false;
4079         new_trace.InvalidateCurrentCharacter();
4080       }
4081       if (i < choice_count - 1) {
4082         new_trace.set_backtrack(&alt_gen->after);
4083       }
4084       generate_full_check_inline = true;
4085     }
4086     if (generate_full_check_inline) {
4087       if (new_trace.actions() != NULL) {
4088         new_trace.set_flush_budget(new_flush_budget);
4089       }
4090       for (int j = 0; j < guard_count; j++) {
4091         GenerateGuard(macro_assembler, guards->at(j), &new_trace);
4092       }
4093       alternative.node()->Emit(compiler, &new_trace);
4094       preload_is_current = false;
4095     }
4096     macro_assembler->Bind(&alt_gen->after);
4097   }
4098   if (greedy_loop) {
4099     macro_assembler->Bind(&greedy_loop_label);
4100     // If we have unwound to the bottom then backtrack.
4101     macro_assembler->CheckGreedyLoop(trace->backtrack());
4102     // Otherwise try the second priority at an earlier position.
4103     macro_assembler->AdvanceCurrentPosition(-text_length);
4104     macro_assembler->GoTo(&second_choice);
4105   }
4106 
4107   // At this point we need to generate slow checks for the alternatives where
4108   // the quick check was inlined.  We can recognize these because the associated
4109   // label was bound.
4110   for (int i = first_normal_choice; i < choice_count - 1; i++) {
4111     AlternativeGeneration* alt_gen = alt_gens.at(i);
4112     Trace new_trace(*current_trace);
4113     // If there are actions to be flushed we have to limit how many times
4114     // they are flushed.  Take the budget of the parent trace and distribute
4115     // it fairly amongst the children.
4116     if (new_trace.actions() != NULL) {
4117       new_trace.set_flush_budget(new_flush_budget);
4118     }
4119     EmitOutOfLineContinuation(compiler,
4120                               &new_trace,
4121                               alternatives_->at(i),
4122                               alt_gen,
4123                               preload_characters,
4124                               alt_gens.at(i + 1)->expects_preload);
4125   }
4126 }
4127 
4128 
EmitOutOfLineContinuation(RegExpCompiler * compiler,Trace * trace,GuardedAlternative alternative,AlternativeGeneration * alt_gen,int preload_characters,bool next_expects_preload)4129 void ChoiceNode::EmitOutOfLineContinuation(RegExpCompiler* compiler,
4130                                            Trace* trace,
4131                                            GuardedAlternative alternative,
4132                                            AlternativeGeneration* alt_gen,
4133                                            int preload_characters,
4134                                            bool next_expects_preload) {
4135   if (!alt_gen->possible_success.is_linked()) return;
4136 
4137   RegExpMacroAssembler* macro_assembler = compiler->macro_assembler();
4138   macro_assembler->Bind(&alt_gen->possible_success);
4139   Trace out_of_line_trace(*trace);
4140   out_of_line_trace.set_characters_preloaded(preload_characters);
4141   out_of_line_trace.set_quick_check_performed(&alt_gen->quick_check_details);
4142   if (not_at_start_) out_of_line_trace.set_at_start(Trace::FALSE_VALUE);
4143   ZoneList<Guard*>* guards = alternative.guards();
4144   int guard_count = (guards == NULL) ? 0 : guards->length();
4145   if (next_expects_preload) {
4146     Label reload_current_char;
4147     out_of_line_trace.set_backtrack(&reload_current_char);
4148     for (int j = 0; j < guard_count; j++) {
4149       GenerateGuard(macro_assembler, guards->at(j), &out_of_line_trace);
4150     }
4151     alternative.node()->Emit(compiler, &out_of_line_trace);
4152     macro_assembler->Bind(&reload_current_char);
4153     // Reload the current character, since the next quick check expects that.
4154     // We don't need to check bounds here because we only get into this
4155     // code through a quick check which already did the checked load.
4156     macro_assembler->LoadCurrentCharacter(trace->cp_offset(),
4157                                           NULL,
4158                                           false,
4159                                           preload_characters);
4160     macro_assembler->GoTo(&(alt_gen->after));
4161   } else {
4162     out_of_line_trace.set_backtrack(&(alt_gen->after));
4163     for (int j = 0; j < guard_count; j++) {
4164       GenerateGuard(macro_assembler, guards->at(j), &out_of_line_trace);
4165     }
4166     alternative.node()->Emit(compiler, &out_of_line_trace);
4167   }
4168 }
4169 
4170 
Emit(RegExpCompiler * compiler,Trace * trace)4171 void ActionNode::Emit(RegExpCompiler* compiler, Trace* trace) {
4172   RegExpMacroAssembler* assembler = compiler->macro_assembler();
4173   LimitResult limit_result = LimitVersions(compiler, trace);
4174   if (limit_result == DONE) return;
4175   ASSERT(limit_result == CONTINUE);
4176 
4177   RecursionCheck rc(compiler);
4178 
4179   switch (action_type_) {
4180     case STORE_POSITION: {
4181       Trace::DeferredCapture
4182           new_capture(data_.u_position_register.reg,
4183                       data_.u_position_register.is_capture,
4184                       trace);
4185       Trace new_trace = *trace;
4186       new_trace.add_action(&new_capture);
4187       on_success()->Emit(compiler, &new_trace);
4188       break;
4189     }
4190     case INCREMENT_REGISTER: {
4191       Trace::DeferredIncrementRegister
4192           new_increment(data_.u_increment_register.reg);
4193       Trace new_trace = *trace;
4194       new_trace.add_action(&new_increment);
4195       on_success()->Emit(compiler, &new_trace);
4196       break;
4197     }
4198     case SET_REGISTER: {
4199       Trace::DeferredSetRegister
4200           new_set(data_.u_store_register.reg, data_.u_store_register.value);
4201       Trace new_trace = *trace;
4202       new_trace.add_action(&new_set);
4203       on_success()->Emit(compiler, &new_trace);
4204       break;
4205     }
4206     case CLEAR_CAPTURES: {
4207       Trace::DeferredClearCaptures
4208         new_capture(Interval(data_.u_clear_captures.range_from,
4209                              data_.u_clear_captures.range_to));
4210       Trace new_trace = *trace;
4211       new_trace.add_action(&new_capture);
4212       on_success()->Emit(compiler, &new_trace);
4213       break;
4214     }
4215     case BEGIN_SUBMATCH:
4216       if (!trace->is_trivial()) {
4217         trace->Flush(compiler, this);
4218       } else {
4219         assembler->WriteCurrentPositionToRegister(
4220             data_.u_submatch.current_position_register, 0);
4221         assembler->WriteStackPointerToRegister(
4222             data_.u_submatch.stack_pointer_register);
4223         on_success()->Emit(compiler, trace);
4224       }
4225       break;
4226     case EMPTY_MATCH_CHECK: {
4227       int start_pos_reg = data_.u_empty_match_check.start_register;
4228       int stored_pos = 0;
4229       int rep_reg = data_.u_empty_match_check.repetition_register;
4230       bool has_minimum = (rep_reg != RegExpCompiler::kNoRegister);
4231       bool know_dist = trace->GetStoredPosition(start_pos_reg, &stored_pos);
4232       if (know_dist && !has_minimum && stored_pos == trace->cp_offset()) {
4233         // If we know we haven't advanced and there is no minimum we
4234         // can just backtrack immediately.
4235         assembler->GoTo(trace->backtrack());
4236       } else if (know_dist && stored_pos < trace->cp_offset()) {
4237         // If we know we've advanced we can generate the continuation
4238         // immediately.
4239         on_success()->Emit(compiler, trace);
4240       } else if (!trace->is_trivial()) {
4241         trace->Flush(compiler, this);
4242       } else {
4243         Label skip_empty_check;
4244         // If we have a minimum number of repetitions we check the current
4245         // number first and skip the empty check if it's not enough.
4246         if (has_minimum) {
4247           int limit = data_.u_empty_match_check.repetition_limit;
4248           assembler->IfRegisterLT(rep_reg, limit, &skip_empty_check);
4249         }
4250         // If the match is empty we bail out, otherwise we fall through
4251         // to the on-success continuation.
4252         assembler->IfRegisterEqPos(data_.u_empty_match_check.start_register,
4253                                    trace->backtrack());
4254         assembler->Bind(&skip_empty_check);
4255         on_success()->Emit(compiler, trace);
4256       }
4257       break;
4258     }
4259     case POSITIVE_SUBMATCH_SUCCESS: {
4260       if (!trace->is_trivial()) {
4261         trace->Flush(compiler, this);
4262         return;
4263       }
4264       assembler->ReadCurrentPositionFromRegister(
4265           data_.u_submatch.current_position_register);
4266       assembler->ReadStackPointerFromRegister(
4267           data_.u_submatch.stack_pointer_register);
4268       int clear_register_count = data_.u_submatch.clear_register_count;
4269       if (clear_register_count == 0) {
4270         on_success()->Emit(compiler, trace);
4271         return;
4272       }
4273       int clear_registers_from = data_.u_submatch.clear_register_from;
4274       Label clear_registers_backtrack;
4275       Trace new_trace = *trace;
4276       new_trace.set_backtrack(&clear_registers_backtrack);
4277       on_success()->Emit(compiler, &new_trace);
4278 
4279       assembler->Bind(&clear_registers_backtrack);
4280       int clear_registers_to = clear_registers_from + clear_register_count - 1;
4281       assembler->ClearRegisters(clear_registers_from, clear_registers_to);
4282 
4283       ASSERT(trace->backtrack() == NULL);
4284       assembler->Backtrack();
4285       return;
4286     }
4287     default:
4288       UNREACHABLE();
4289   }
4290 }
4291 
4292 
Emit(RegExpCompiler * compiler,Trace * trace)4293 void BackReferenceNode::Emit(RegExpCompiler* compiler, Trace* trace) {
4294   RegExpMacroAssembler* assembler = compiler->macro_assembler();
4295   if (!trace->is_trivial()) {
4296     trace->Flush(compiler, this);
4297     return;
4298   }
4299 
4300   LimitResult limit_result = LimitVersions(compiler, trace);
4301   if (limit_result == DONE) return;
4302   ASSERT(limit_result == CONTINUE);
4303 
4304   RecursionCheck rc(compiler);
4305 
4306   ASSERT_EQ(start_reg_ + 1, end_reg_);
4307   if (compiler->ignore_case()) {
4308     assembler->CheckNotBackReferenceIgnoreCase(start_reg_,
4309                                                trace->backtrack());
4310   } else {
4311     assembler->CheckNotBackReference(start_reg_, trace->backtrack());
4312   }
4313   on_success()->Emit(compiler, trace);
4314 }
4315 
4316 
4317 // -------------------------------------------------------------------
4318 // Dot/dotty output
4319 
4320 
4321 #ifdef DEBUG
4322 
4323 
4324 class DotPrinter: public NodeVisitor {
4325  public:
DotPrinter(bool ignore_case)4326   explicit DotPrinter(bool ignore_case)
4327       : ignore_case_(ignore_case),
4328         stream_(&alloc_) { }
4329   void PrintNode(const char* label, RegExpNode* node);
4330   void Visit(RegExpNode* node);
4331   void PrintAttributes(RegExpNode* from);
stream()4332   StringStream* stream() { return &stream_; }
4333   void PrintOnFailure(RegExpNode* from, RegExpNode* to);
4334 #define DECLARE_VISIT(Type)                                          \
4335   virtual void Visit##Type(Type##Node* that);
4336 FOR_EACH_NODE_TYPE(DECLARE_VISIT)
4337 #undef DECLARE_VISIT
4338  private:
4339   bool ignore_case_;
4340   HeapStringAllocator alloc_;
4341   StringStream stream_;
4342 };
4343 
4344 
PrintNode(const char * label,RegExpNode * node)4345 void DotPrinter::PrintNode(const char* label, RegExpNode* node) {
4346   stream()->Add("digraph G {\n  graph [label=\"");
4347   for (int i = 0; label[i]; i++) {
4348     switch (label[i]) {
4349       case '\\':
4350         stream()->Add("\\\\");
4351         break;
4352       case '"':
4353         stream()->Add("\"");
4354         break;
4355       default:
4356         stream()->Put(label[i]);
4357         break;
4358     }
4359   }
4360   stream()->Add("\"];\n");
4361   Visit(node);
4362   stream()->Add("}\n");
4363   printf("%s", stream()->ToCString().get());
4364 }
4365 
4366 
Visit(RegExpNode * node)4367 void DotPrinter::Visit(RegExpNode* node) {
4368   if (node->info()->visited) return;
4369   node->info()->visited = true;
4370   node->Accept(this);
4371 }
4372 
4373 
PrintOnFailure(RegExpNode * from,RegExpNode * on_failure)4374 void DotPrinter::PrintOnFailure(RegExpNode* from, RegExpNode* on_failure) {
4375   stream()->Add("  n%p -> n%p [style=dotted];\n", from, on_failure);
4376   Visit(on_failure);
4377 }
4378 
4379 
4380 class TableEntryBodyPrinter {
4381  public:
TableEntryBodyPrinter(StringStream * stream,ChoiceNode * choice)4382   TableEntryBodyPrinter(StringStream* stream, ChoiceNode* choice)
4383       : stream_(stream), choice_(choice) { }
Call(uc16 from,DispatchTable::Entry entry)4384   void Call(uc16 from, DispatchTable::Entry entry) {
4385     OutSet* out_set = entry.out_set();
4386     for (unsigned i = 0; i < OutSet::kFirstLimit; i++) {
4387       if (out_set->Get(i)) {
4388         stream()->Add("    n%p:s%io%i -> n%p;\n",
4389                       choice(),
4390                       from,
4391                       i,
4392                       choice()->alternatives()->at(i).node());
4393       }
4394     }
4395   }
4396  private:
stream()4397   StringStream* stream() { return stream_; }
choice()4398   ChoiceNode* choice() { return choice_; }
4399   StringStream* stream_;
4400   ChoiceNode* choice_;
4401 };
4402 
4403 
4404 class TableEntryHeaderPrinter {
4405  public:
TableEntryHeaderPrinter(StringStream * stream)4406   explicit TableEntryHeaderPrinter(StringStream* stream)
4407       : first_(true), stream_(stream) { }
Call(uc16 from,DispatchTable::Entry entry)4408   void Call(uc16 from, DispatchTable::Entry entry) {
4409     if (first_) {
4410       first_ = false;
4411     } else {
4412       stream()->Add("|");
4413     }
4414     stream()->Add("{\\%k-\\%k|{", from, entry.to());
4415     OutSet* out_set = entry.out_set();
4416     int priority = 0;
4417     for (unsigned i = 0; i < OutSet::kFirstLimit; i++) {
4418       if (out_set->Get(i)) {
4419         if (priority > 0) stream()->Add("|");
4420         stream()->Add("<s%io%i> %i", from, i, priority);
4421         priority++;
4422       }
4423     }
4424     stream()->Add("}}");
4425   }
4426 
4427  private:
4428   bool first_;
stream()4429   StringStream* stream() { return stream_; }
4430   StringStream* stream_;
4431 };
4432 
4433 
4434 class AttributePrinter {
4435  public:
AttributePrinter(DotPrinter * out)4436   explicit AttributePrinter(DotPrinter* out)
4437       : out_(out), first_(true) { }
PrintSeparator()4438   void PrintSeparator() {
4439     if (first_) {
4440       first_ = false;
4441     } else {
4442       out_->stream()->Add("|");
4443     }
4444   }
PrintBit(const char * name,bool value)4445   void PrintBit(const char* name, bool value) {
4446     if (!value) return;
4447     PrintSeparator();
4448     out_->stream()->Add("{%s}", name);
4449   }
PrintPositive(const char * name,int value)4450   void PrintPositive(const char* name, int value) {
4451     if (value < 0) return;
4452     PrintSeparator();
4453     out_->stream()->Add("{%s|%x}", name, value);
4454   }
4455  private:
4456   DotPrinter* out_;
4457   bool first_;
4458 };
4459 
4460 
PrintAttributes(RegExpNode * that)4461 void DotPrinter::PrintAttributes(RegExpNode* that) {
4462   stream()->Add("  a%p [shape=Mrecord, color=grey, fontcolor=grey, "
4463                 "margin=0.1, fontsize=10, label=\"{",
4464                 that);
4465   AttributePrinter printer(this);
4466   NodeInfo* info = that->info();
4467   printer.PrintBit("NI", info->follows_newline_interest);
4468   printer.PrintBit("WI", info->follows_word_interest);
4469   printer.PrintBit("SI", info->follows_start_interest);
4470   Label* label = that->label();
4471   if (label->is_bound())
4472     printer.PrintPositive("@", label->pos());
4473   stream()->Add("}\"];\n");
4474   stream()->Add("  a%p -> n%p [style=dashed, color=grey, "
4475                 "arrowhead=none];\n", that, that);
4476 }
4477 
4478 
4479 static const bool kPrintDispatchTable = false;
VisitChoice(ChoiceNode * that)4480 void DotPrinter::VisitChoice(ChoiceNode* that) {
4481   if (kPrintDispatchTable) {
4482     stream()->Add("  n%p [shape=Mrecord, label=\"", that);
4483     TableEntryHeaderPrinter header_printer(stream());
4484     that->GetTable(ignore_case_)->ForEach(&header_printer);
4485     stream()->Add("\"]\n", that);
4486     PrintAttributes(that);
4487     TableEntryBodyPrinter body_printer(stream(), that);
4488     that->GetTable(ignore_case_)->ForEach(&body_printer);
4489   } else {
4490     stream()->Add("  n%p [shape=Mrecord, label=\"?\"];\n", that);
4491     for (int i = 0; i < that->alternatives()->length(); i++) {
4492       GuardedAlternative alt = that->alternatives()->at(i);
4493       stream()->Add("  n%p -> n%p;\n", that, alt.node());
4494     }
4495   }
4496   for (int i = 0; i < that->alternatives()->length(); i++) {
4497     GuardedAlternative alt = that->alternatives()->at(i);
4498     alt.node()->Accept(this);
4499   }
4500 }
4501 
4502 
VisitText(TextNode * that)4503 void DotPrinter::VisitText(TextNode* that) {
4504   Zone* zone = that->zone();
4505   stream()->Add("  n%p [label=\"", that);
4506   for (int i = 0; i < that->elements()->length(); i++) {
4507     if (i > 0) stream()->Add(" ");
4508     TextElement elm = that->elements()->at(i);
4509     switch (elm.text_type()) {
4510       case TextElement::ATOM: {
4511         stream()->Add("'%w'", elm.atom()->data());
4512         break;
4513       }
4514       case TextElement::CHAR_CLASS: {
4515         RegExpCharacterClass* node = elm.char_class();
4516         stream()->Add("[");
4517         if (node->is_negated())
4518           stream()->Add("^");
4519         for (int j = 0; j < node->ranges(zone)->length(); j++) {
4520           CharacterRange range = node->ranges(zone)->at(j);
4521           stream()->Add("%k-%k", range.from(), range.to());
4522         }
4523         stream()->Add("]");
4524         break;
4525       }
4526       default:
4527         UNREACHABLE();
4528     }
4529   }
4530   stream()->Add("\", shape=box, peripheries=2];\n");
4531   PrintAttributes(that);
4532   stream()->Add("  n%p -> n%p;\n", that, that->on_success());
4533   Visit(that->on_success());
4534 }
4535 
4536 
VisitBackReference(BackReferenceNode * that)4537 void DotPrinter::VisitBackReference(BackReferenceNode* that) {
4538   stream()->Add("  n%p [label=\"$%i..$%i\", shape=doubleoctagon];\n",
4539                 that,
4540                 that->start_register(),
4541                 that->end_register());
4542   PrintAttributes(that);
4543   stream()->Add("  n%p -> n%p;\n", that, that->on_success());
4544   Visit(that->on_success());
4545 }
4546 
4547 
VisitEnd(EndNode * that)4548 void DotPrinter::VisitEnd(EndNode* that) {
4549   stream()->Add("  n%p [style=bold, shape=point];\n", that);
4550   PrintAttributes(that);
4551 }
4552 
4553 
VisitAssertion(AssertionNode * that)4554 void DotPrinter::VisitAssertion(AssertionNode* that) {
4555   stream()->Add("  n%p [", that);
4556   switch (that->assertion_type()) {
4557     case AssertionNode::AT_END:
4558       stream()->Add("label=\"$\", shape=septagon");
4559       break;
4560     case AssertionNode::AT_START:
4561       stream()->Add("label=\"^\", shape=septagon");
4562       break;
4563     case AssertionNode::AT_BOUNDARY:
4564       stream()->Add("label=\"\\b\", shape=septagon");
4565       break;
4566     case AssertionNode::AT_NON_BOUNDARY:
4567       stream()->Add("label=\"\\B\", shape=septagon");
4568       break;
4569     case AssertionNode::AFTER_NEWLINE:
4570       stream()->Add("label=\"(?<=\\n)\", shape=septagon");
4571       break;
4572   }
4573   stream()->Add("];\n");
4574   PrintAttributes(that);
4575   RegExpNode* successor = that->on_success();
4576   stream()->Add("  n%p -> n%p;\n", that, successor);
4577   Visit(successor);
4578 }
4579 
4580 
VisitAction(ActionNode * that)4581 void DotPrinter::VisitAction(ActionNode* that) {
4582   stream()->Add("  n%p [", that);
4583   switch (that->action_type_) {
4584     case ActionNode::SET_REGISTER:
4585       stream()->Add("label=\"$%i:=%i\", shape=octagon",
4586                     that->data_.u_store_register.reg,
4587                     that->data_.u_store_register.value);
4588       break;
4589     case ActionNode::INCREMENT_REGISTER:
4590       stream()->Add("label=\"$%i++\", shape=octagon",
4591                     that->data_.u_increment_register.reg);
4592       break;
4593     case ActionNode::STORE_POSITION:
4594       stream()->Add("label=\"$%i:=$pos\", shape=octagon",
4595                     that->data_.u_position_register.reg);
4596       break;
4597     case ActionNode::BEGIN_SUBMATCH:
4598       stream()->Add("label=\"$%i:=$pos,begin\", shape=septagon",
4599                     that->data_.u_submatch.current_position_register);
4600       break;
4601     case ActionNode::POSITIVE_SUBMATCH_SUCCESS:
4602       stream()->Add("label=\"escape\", shape=septagon");
4603       break;
4604     case ActionNode::EMPTY_MATCH_CHECK:
4605       stream()->Add("label=\"$%i=$pos?,$%i<%i?\", shape=septagon",
4606                     that->data_.u_empty_match_check.start_register,
4607                     that->data_.u_empty_match_check.repetition_register,
4608                     that->data_.u_empty_match_check.repetition_limit);
4609       break;
4610     case ActionNode::CLEAR_CAPTURES: {
4611       stream()->Add("label=\"clear $%i to $%i\", shape=septagon",
4612                     that->data_.u_clear_captures.range_from,
4613                     that->data_.u_clear_captures.range_to);
4614       break;
4615     }
4616   }
4617   stream()->Add("];\n");
4618   PrintAttributes(that);
4619   RegExpNode* successor = that->on_success();
4620   stream()->Add("  n%p -> n%p;\n", that, successor);
4621   Visit(successor);
4622 }
4623 
4624 
4625 class DispatchTableDumper {
4626  public:
DispatchTableDumper(StringStream * stream)4627   explicit DispatchTableDumper(StringStream* stream) : stream_(stream) { }
4628   void Call(uc16 key, DispatchTable::Entry entry);
stream()4629   StringStream* stream() { return stream_; }
4630  private:
4631   StringStream* stream_;
4632 };
4633 
4634 
Call(uc16 key,DispatchTable::Entry entry)4635 void DispatchTableDumper::Call(uc16 key, DispatchTable::Entry entry) {
4636   stream()->Add("[%k-%k]: {", key, entry.to());
4637   OutSet* set = entry.out_set();
4638   bool first = true;
4639   for (unsigned i = 0; i < OutSet::kFirstLimit; i++) {
4640     if (set->Get(i)) {
4641       if (first) {
4642         first = false;
4643       } else {
4644         stream()->Add(", ");
4645       }
4646       stream()->Add("%i", i);
4647     }
4648   }
4649   stream()->Add("}\n");
4650 }
4651 
4652 
Dump()4653 void DispatchTable::Dump() {
4654   HeapStringAllocator alloc;
4655   StringStream stream(&alloc);
4656   DispatchTableDumper dumper(&stream);
4657   tree()->ForEach(&dumper);
4658   OS::PrintError("%s", stream.ToCString().get());
4659 }
4660 
4661 
DotPrint(const char * label,RegExpNode * node,bool ignore_case)4662 void RegExpEngine::DotPrint(const char* label,
4663                             RegExpNode* node,
4664                             bool ignore_case) {
4665   DotPrinter printer(ignore_case);
4666   printer.PrintNode(label, node);
4667 }
4668 
4669 
4670 #endif  // DEBUG
4671 
4672 
4673 // -------------------------------------------------------------------
4674 // Tree to graph conversion
4675 
ToNode(RegExpCompiler * compiler,RegExpNode * on_success)4676 RegExpNode* RegExpAtom::ToNode(RegExpCompiler* compiler,
4677                                RegExpNode* on_success) {
4678   ZoneList<TextElement>* elms =
4679       new(compiler->zone()) ZoneList<TextElement>(1, compiler->zone());
4680   elms->Add(TextElement::Atom(this), compiler->zone());
4681   return new(compiler->zone()) TextNode(elms, on_success);
4682 }
4683 
4684 
ToNode(RegExpCompiler * compiler,RegExpNode * on_success)4685 RegExpNode* RegExpText::ToNode(RegExpCompiler* compiler,
4686                                RegExpNode* on_success) {
4687   return new(compiler->zone()) TextNode(elements(), on_success);
4688 }
4689 
4690 
CompareInverseRanges(ZoneList<CharacterRange> * ranges,const int * special_class,int length)4691 static bool CompareInverseRanges(ZoneList<CharacterRange>* ranges,
4692                                  const int* special_class,
4693                                  int length) {
4694   length--;  // Remove final 0x10000.
4695   ASSERT(special_class[length] == 0x10000);
4696   ASSERT(ranges->length() != 0);
4697   ASSERT(length != 0);
4698   ASSERT(special_class[0] != 0);
4699   if (ranges->length() != (length >> 1) + 1) {
4700     return false;
4701   }
4702   CharacterRange range = ranges->at(0);
4703   if (range.from() != 0) {
4704     return false;
4705   }
4706   for (int i = 0; i < length; i += 2) {
4707     if (special_class[i] != (range.to() + 1)) {
4708       return false;
4709     }
4710     range = ranges->at((i >> 1) + 1);
4711     if (special_class[i+1] != range.from()) {
4712       return false;
4713     }
4714   }
4715   if (range.to() != 0xffff) {
4716     return false;
4717   }
4718   return true;
4719 }
4720 
4721 
CompareRanges(ZoneList<CharacterRange> * ranges,const int * special_class,int length)4722 static bool CompareRanges(ZoneList<CharacterRange>* ranges,
4723                           const int* special_class,
4724                           int length) {
4725   length--;  // Remove final 0x10000.
4726   ASSERT(special_class[length] == 0x10000);
4727   if (ranges->length() * 2 != length) {
4728     return false;
4729   }
4730   for (int i = 0; i < length; i += 2) {
4731     CharacterRange range = ranges->at(i >> 1);
4732     if (range.from() != special_class[i] ||
4733         range.to() != special_class[i + 1] - 1) {
4734       return false;
4735     }
4736   }
4737   return true;
4738 }
4739 
4740 
is_standard(Zone * zone)4741 bool RegExpCharacterClass::is_standard(Zone* zone) {
4742   // TODO(lrn): Remove need for this function, by not throwing away information
4743   // along the way.
4744   if (is_negated_) {
4745     return false;
4746   }
4747   if (set_.is_standard()) {
4748     return true;
4749   }
4750   if (CompareRanges(set_.ranges(zone), kSpaceRanges, kSpaceRangeCount)) {
4751     set_.set_standard_set_type('s');
4752     return true;
4753   }
4754   if (CompareInverseRanges(set_.ranges(zone), kSpaceRanges, kSpaceRangeCount)) {
4755     set_.set_standard_set_type('S');
4756     return true;
4757   }
4758   if (CompareInverseRanges(set_.ranges(zone),
4759                            kLineTerminatorRanges,
4760                            kLineTerminatorRangeCount)) {
4761     set_.set_standard_set_type('.');
4762     return true;
4763   }
4764   if (CompareRanges(set_.ranges(zone),
4765                     kLineTerminatorRanges,
4766                     kLineTerminatorRangeCount)) {
4767     set_.set_standard_set_type('n');
4768     return true;
4769   }
4770   if (CompareRanges(set_.ranges(zone), kWordRanges, kWordRangeCount)) {
4771     set_.set_standard_set_type('w');
4772     return true;
4773   }
4774   if (CompareInverseRanges(set_.ranges(zone), kWordRanges, kWordRangeCount)) {
4775     set_.set_standard_set_type('W');
4776     return true;
4777   }
4778   return false;
4779 }
4780 
4781 
ToNode(RegExpCompiler * compiler,RegExpNode * on_success)4782 RegExpNode* RegExpCharacterClass::ToNode(RegExpCompiler* compiler,
4783                                          RegExpNode* on_success) {
4784   return new(compiler->zone()) TextNode(this, on_success);
4785 }
4786 
4787 
ToNode(RegExpCompiler * compiler,RegExpNode * on_success)4788 RegExpNode* RegExpDisjunction::ToNode(RegExpCompiler* compiler,
4789                                       RegExpNode* on_success) {
4790   ZoneList<RegExpTree*>* alternatives = this->alternatives();
4791   int length = alternatives->length();
4792   ChoiceNode* result =
4793       new(compiler->zone()) ChoiceNode(length, compiler->zone());
4794   for (int i = 0; i < length; i++) {
4795     GuardedAlternative alternative(alternatives->at(i)->ToNode(compiler,
4796                                                                on_success));
4797     result->AddAlternative(alternative);
4798   }
4799   return result;
4800 }
4801 
4802 
ToNode(RegExpCompiler * compiler,RegExpNode * on_success)4803 RegExpNode* RegExpQuantifier::ToNode(RegExpCompiler* compiler,
4804                                      RegExpNode* on_success) {
4805   return ToNode(min(),
4806                 max(),
4807                 is_greedy(),
4808                 body(),
4809                 compiler,
4810                 on_success);
4811 }
4812 
4813 
4814 // Scoped object to keep track of how much we unroll quantifier loops in the
4815 // regexp graph generator.
4816 class RegExpExpansionLimiter {
4817  public:
4818   static const int kMaxExpansionFactor = 6;
RegExpExpansionLimiter(RegExpCompiler * compiler,int factor)4819   RegExpExpansionLimiter(RegExpCompiler* compiler, int factor)
4820       : compiler_(compiler),
4821         saved_expansion_factor_(compiler->current_expansion_factor()),
4822         ok_to_expand_(saved_expansion_factor_ <= kMaxExpansionFactor) {
4823     ASSERT(factor > 0);
4824     if (ok_to_expand_) {
4825       if (factor > kMaxExpansionFactor) {
4826         // Avoid integer overflow of the current expansion factor.
4827         ok_to_expand_ = false;
4828         compiler->set_current_expansion_factor(kMaxExpansionFactor + 1);
4829       } else {
4830         int new_factor = saved_expansion_factor_ * factor;
4831         ok_to_expand_ = (new_factor <= kMaxExpansionFactor);
4832         compiler->set_current_expansion_factor(new_factor);
4833       }
4834     }
4835   }
4836 
~RegExpExpansionLimiter()4837   ~RegExpExpansionLimiter() {
4838     compiler_->set_current_expansion_factor(saved_expansion_factor_);
4839   }
4840 
ok_to_expand()4841   bool ok_to_expand() { return ok_to_expand_; }
4842 
4843  private:
4844   RegExpCompiler* compiler_;
4845   int saved_expansion_factor_;
4846   bool ok_to_expand_;
4847 
4848   DISALLOW_IMPLICIT_CONSTRUCTORS(RegExpExpansionLimiter);
4849 };
4850 
4851 
ToNode(int min,int max,bool is_greedy,RegExpTree * body,RegExpCompiler * compiler,RegExpNode * on_success,bool not_at_start)4852 RegExpNode* RegExpQuantifier::ToNode(int min,
4853                                      int max,
4854                                      bool is_greedy,
4855                                      RegExpTree* body,
4856                                      RegExpCompiler* compiler,
4857                                      RegExpNode* on_success,
4858                                      bool not_at_start) {
4859   // x{f, t} becomes this:
4860   //
4861   //             (r++)<-.
4862   //               |     `
4863   //               |     (x)
4864   //               v     ^
4865   //      (r=0)-->(?)---/ [if r < t]
4866   //               |
4867   //   [if r >= f] \----> ...
4868   //
4869 
4870   // 15.10.2.5 RepeatMatcher algorithm.
4871   // The parser has already eliminated the case where max is 0.  In the case
4872   // where max_match is zero the parser has removed the quantifier if min was
4873   // > 0 and removed the atom if min was 0.  See AddQuantifierToAtom.
4874 
4875   // If we know that we cannot match zero length then things are a little
4876   // simpler since we don't need to make the special zero length match check
4877   // from step 2.1.  If the min and max are small we can unroll a little in
4878   // this case.
4879   static const int kMaxUnrolledMinMatches = 3;  // Unroll (foo)+ and (foo){3,}
4880   static const int kMaxUnrolledMaxMatches = 3;  // Unroll (foo)? and (foo){x,3}
4881   if (max == 0) return on_success;  // This can happen due to recursion.
4882   bool body_can_be_empty = (body->min_match() == 0);
4883   int body_start_reg = RegExpCompiler::kNoRegister;
4884   Interval capture_registers = body->CaptureRegisters();
4885   bool needs_capture_clearing = !capture_registers.is_empty();
4886   Zone* zone = compiler->zone();
4887 
4888   if (body_can_be_empty) {
4889     body_start_reg = compiler->AllocateRegister();
4890   } else if (FLAG_regexp_optimization && !needs_capture_clearing) {
4891     // Only unroll if there are no captures and the body can't be
4892     // empty.
4893     {
4894       RegExpExpansionLimiter limiter(
4895           compiler, min + ((max != min) ? 1 : 0));
4896       if (min > 0 && min <= kMaxUnrolledMinMatches && limiter.ok_to_expand()) {
4897         int new_max = (max == kInfinity) ? max : max - min;
4898         // Recurse once to get the loop or optional matches after the fixed
4899         // ones.
4900         RegExpNode* answer = ToNode(
4901             0, new_max, is_greedy, body, compiler, on_success, true);
4902         // Unroll the forced matches from 0 to min.  This can cause chains of
4903         // TextNodes (which the parser does not generate).  These should be
4904         // combined if it turns out they hinder good code generation.
4905         for (int i = 0; i < min; i++) {
4906           answer = body->ToNode(compiler, answer);
4907         }
4908         return answer;
4909       }
4910     }
4911     if (max <= kMaxUnrolledMaxMatches && min == 0) {
4912       ASSERT(max > 0);  // Due to the 'if' above.
4913       RegExpExpansionLimiter limiter(compiler, max);
4914       if (limiter.ok_to_expand()) {
4915         // Unroll the optional matches up to max.
4916         RegExpNode* answer = on_success;
4917         for (int i = 0; i < max; i++) {
4918           ChoiceNode* alternation = new(zone) ChoiceNode(2, zone);
4919           if (is_greedy) {
4920             alternation->AddAlternative(
4921                 GuardedAlternative(body->ToNode(compiler, answer)));
4922             alternation->AddAlternative(GuardedAlternative(on_success));
4923           } else {
4924             alternation->AddAlternative(GuardedAlternative(on_success));
4925             alternation->AddAlternative(
4926                 GuardedAlternative(body->ToNode(compiler, answer)));
4927           }
4928           answer = alternation;
4929           if (not_at_start) alternation->set_not_at_start();
4930         }
4931         return answer;
4932       }
4933     }
4934   }
4935   bool has_min = min > 0;
4936   bool has_max = max < RegExpTree::kInfinity;
4937   bool needs_counter = has_min || has_max;
4938   int reg_ctr = needs_counter
4939       ? compiler->AllocateRegister()
4940       : RegExpCompiler::kNoRegister;
4941   LoopChoiceNode* center = new(zone) LoopChoiceNode(body->min_match() == 0,
4942                                                     zone);
4943   if (not_at_start) center->set_not_at_start();
4944   RegExpNode* loop_return = needs_counter
4945       ? static_cast<RegExpNode*>(ActionNode::IncrementRegister(reg_ctr, center))
4946       : static_cast<RegExpNode*>(center);
4947   if (body_can_be_empty) {
4948     // If the body can be empty we need to check if it was and then
4949     // backtrack.
4950     loop_return = ActionNode::EmptyMatchCheck(body_start_reg,
4951                                               reg_ctr,
4952                                               min,
4953                                               loop_return);
4954   }
4955   RegExpNode* body_node = body->ToNode(compiler, loop_return);
4956   if (body_can_be_empty) {
4957     // If the body can be empty we need to store the start position
4958     // so we can bail out if it was empty.
4959     body_node = ActionNode::StorePosition(body_start_reg, false, body_node);
4960   }
4961   if (needs_capture_clearing) {
4962     // Before entering the body of this loop we need to clear captures.
4963     body_node = ActionNode::ClearCaptures(capture_registers, body_node);
4964   }
4965   GuardedAlternative body_alt(body_node);
4966   if (has_max) {
4967     Guard* body_guard =
4968         new(zone) Guard(reg_ctr, Guard::LT, max);
4969     body_alt.AddGuard(body_guard, zone);
4970   }
4971   GuardedAlternative rest_alt(on_success);
4972   if (has_min) {
4973     Guard* rest_guard = new(compiler->zone()) Guard(reg_ctr, Guard::GEQ, min);
4974     rest_alt.AddGuard(rest_guard, zone);
4975   }
4976   if (is_greedy) {
4977     center->AddLoopAlternative(body_alt);
4978     center->AddContinueAlternative(rest_alt);
4979   } else {
4980     center->AddContinueAlternative(rest_alt);
4981     center->AddLoopAlternative(body_alt);
4982   }
4983   if (needs_counter) {
4984     return ActionNode::SetRegister(reg_ctr, 0, center);
4985   } else {
4986     return center;
4987   }
4988 }
4989 
4990 
ToNode(RegExpCompiler * compiler,RegExpNode * on_success)4991 RegExpNode* RegExpAssertion::ToNode(RegExpCompiler* compiler,
4992                                     RegExpNode* on_success) {
4993   NodeInfo info;
4994   Zone* zone = compiler->zone();
4995 
4996   switch (assertion_type()) {
4997     case START_OF_LINE:
4998       return AssertionNode::AfterNewline(on_success);
4999     case START_OF_INPUT:
5000       return AssertionNode::AtStart(on_success);
5001     case BOUNDARY:
5002       return AssertionNode::AtBoundary(on_success);
5003     case NON_BOUNDARY:
5004       return AssertionNode::AtNonBoundary(on_success);
5005     case END_OF_INPUT:
5006       return AssertionNode::AtEnd(on_success);
5007     case END_OF_LINE: {
5008       // Compile $ in multiline regexps as an alternation with a positive
5009       // lookahead in one side and an end-of-input on the other side.
5010       // We need two registers for the lookahead.
5011       int stack_pointer_register = compiler->AllocateRegister();
5012       int position_register = compiler->AllocateRegister();
5013       // The ChoiceNode to distinguish between a newline and end-of-input.
5014       ChoiceNode* result = new(zone) ChoiceNode(2, zone);
5015       // Create a newline atom.
5016       ZoneList<CharacterRange>* newline_ranges =
5017           new(zone) ZoneList<CharacterRange>(3, zone);
5018       CharacterRange::AddClassEscape('n', newline_ranges, zone);
5019       RegExpCharacterClass* newline_atom = new(zone) RegExpCharacterClass('n');
5020       TextNode* newline_matcher = new(zone) TextNode(
5021          newline_atom,
5022          ActionNode::PositiveSubmatchSuccess(stack_pointer_register,
5023                                              position_register,
5024                                              0,  // No captures inside.
5025                                              -1,  // Ignored if no captures.
5026                                              on_success));
5027       // Create an end-of-input matcher.
5028       RegExpNode* end_of_line = ActionNode::BeginSubmatch(
5029           stack_pointer_register,
5030           position_register,
5031           newline_matcher);
5032       // Add the two alternatives to the ChoiceNode.
5033       GuardedAlternative eol_alternative(end_of_line);
5034       result->AddAlternative(eol_alternative);
5035       GuardedAlternative end_alternative(AssertionNode::AtEnd(on_success));
5036       result->AddAlternative(end_alternative);
5037       return result;
5038     }
5039     default:
5040       UNREACHABLE();
5041   }
5042   return on_success;
5043 }
5044 
5045 
ToNode(RegExpCompiler * compiler,RegExpNode * on_success)5046 RegExpNode* RegExpBackReference::ToNode(RegExpCompiler* compiler,
5047                                         RegExpNode* on_success) {
5048   return new(compiler->zone())
5049       BackReferenceNode(RegExpCapture::StartRegister(index()),
5050                         RegExpCapture::EndRegister(index()),
5051                         on_success);
5052 }
5053 
5054 
ToNode(RegExpCompiler * compiler,RegExpNode * on_success)5055 RegExpNode* RegExpEmpty::ToNode(RegExpCompiler* compiler,
5056                                 RegExpNode* on_success) {
5057   return on_success;
5058 }
5059 
5060 
ToNode(RegExpCompiler * compiler,RegExpNode * on_success)5061 RegExpNode* RegExpLookahead::ToNode(RegExpCompiler* compiler,
5062                                     RegExpNode* on_success) {
5063   int stack_pointer_register = compiler->AllocateRegister();
5064   int position_register = compiler->AllocateRegister();
5065 
5066   const int registers_per_capture = 2;
5067   const int register_of_first_capture = 2;
5068   int register_count = capture_count_ * registers_per_capture;
5069   int register_start =
5070     register_of_first_capture + capture_from_ * registers_per_capture;
5071 
5072   RegExpNode* success;
5073   if (is_positive()) {
5074     RegExpNode* node = ActionNode::BeginSubmatch(
5075         stack_pointer_register,
5076         position_register,
5077         body()->ToNode(
5078             compiler,
5079             ActionNode::PositiveSubmatchSuccess(stack_pointer_register,
5080                                                 position_register,
5081                                                 register_count,
5082                                                 register_start,
5083                                                 on_success)));
5084     return node;
5085   } else {
5086     // We use a ChoiceNode for a negative lookahead because it has most of
5087     // the characteristics we need.  It has the body of the lookahead as its
5088     // first alternative and the expression after the lookahead of the second
5089     // alternative.  If the first alternative succeeds then the
5090     // NegativeSubmatchSuccess will unwind the stack including everything the
5091     // choice node set up and backtrack.  If the first alternative fails then
5092     // the second alternative is tried, which is exactly the desired result
5093     // for a negative lookahead.  The NegativeLookaheadChoiceNode is a special
5094     // ChoiceNode that knows to ignore the first exit when calculating quick
5095     // checks.
5096     Zone* zone = compiler->zone();
5097 
5098     GuardedAlternative body_alt(
5099         body()->ToNode(
5100             compiler,
5101             success = new(zone) NegativeSubmatchSuccess(stack_pointer_register,
5102                                                         position_register,
5103                                                         register_count,
5104                                                         register_start,
5105                                                         zone)));
5106     ChoiceNode* choice_node =
5107         new(zone) NegativeLookaheadChoiceNode(body_alt,
5108                                               GuardedAlternative(on_success),
5109                                               zone);
5110     return ActionNode::BeginSubmatch(stack_pointer_register,
5111                                      position_register,
5112                                      choice_node);
5113   }
5114 }
5115 
5116 
ToNode(RegExpCompiler * compiler,RegExpNode * on_success)5117 RegExpNode* RegExpCapture::ToNode(RegExpCompiler* compiler,
5118                                   RegExpNode* on_success) {
5119   return ToNode(body(), index(), compiler, on_success);
5120 }
5121 
5122 
ToNode(RegExpTree * body,int index,RegExpCompiler * compiler,RegExpNode * on_success)5123 RegExpNode* RegExpCapture::ToNode(RegExpTree* body,
5124                                   int index,
5125                                   RegExpCompiler* compiler,
5126                                   RegExpNode* on_success) {
5127   int start_reg = RegExpCapture::StartRegister(index);
5128   int end_reg = RegExpCapture::EndRegister(index);
5129   RegExpNode* store_end = ActionNode::StorePosition(end_reg, true, on_success);
5130   RegExpNode* body_node = body->ToNode(compiler, store_end);
5131   return ActionNode::StorePosition(start_reg, true, body_node);
5132 }
5133 
5134 
ToNode(RegExpCompiler * compiler,RegExpNode * on_success)5135 RegExpNode* RegExpAlternative::ToNode(RegExpCompiler* compiler,
5136                                       RegExpNode* on_success) {
5137   ZoneList<RegExpTree*>* children = nodes();
5138   RegExpNode* current = on_success;
5139   for (int i = children->length() - 1; i >= 0; i--) {
5140     current = children->at(i)->ToNode(compiler, current);
5141   }
5142   return current;
5143 }
5144 
5145 
AddClass(const int * elmv,int elmc,ZoneList<CharacterRange> * ranges,Zone * zone)5146 static void AddClass(const int* elmv,
5147                      int elmc,
5148                      ZoneList<CharacterRange>* ranges,
5149                      Zone* zone) {
5150   elmc--;
5151   ASSERT(elmv[elmc] == 0x10000);
5152   for (int i = 0; i < elmc; i += 2) {
5153     ASSERT(elmv[i] < elmv[i + 1]);
5154     ranges->Add(CharacterRange(elmv[i], elmv[i + 1] - 1), zone);
5155   }
5156 }
5157 
5158 
AddClassNegated(const int * elmv,int elmc,ZoneList<CharacterRange> * ranges,Zone * zone)5159 static void AddClassNegated(const int *elmv,
5160                             int elmc,
5161                             ZoneList<CharacterRange>* ranges,
5162                             Zone* zone) {
5163   elmc--;
5164   ASSERT(elmv[elmc] == 0x10000);
5165   ASSERT(elmv[0] != 0x0000);
5166   ASSERT(elmv[elmc-1] != String::kMaxUtf16CodeUnit);
5167   uc16 last = 0x0000;
5168   for (int i = 0; i < elmc; i += 2) {
5169     ASSERT(last <= elmv[i] - 1);
5170     ASSERT(elmv[i] < elmv[i + 1]);
5171     ranges->Add(CharacterRange(last, elmv[i] - 1), zone);
5172     last = elmv[i + 1];
5173   }
5174   ranges->Add(CharacterRange(last, String::kMaxUtf16CodeUnit), zone);
5175 }
5176 
5177 
AddClassEscape(uc16 type,ZoneList<CharacterRange> * ranges,Zone * zone)5178 void CharacterRange::AddClassEscape(uc16 type,
5179                                     ZoneList<CharacterRange>* ranges,
5180                                     Zone* zone) {
5181   switch (type) {
5182     case 's':
5183       AddClass(kSpaceRanges, kSpaceRangeCount, ranges, zone);
5184       break;
5185     case 'S':
5186       AddClassNegated(kSpaceRanges, kSpaceRangeCount, ranges, zone);
5187       break;
5188     case 'w':
5189       AddClass(kWordRanges, kWordRangeCount, ranges, zone);
5190       break;
5191     case 'W':
5192       AddClassNegated(kWordRanges, kWordRangeCount, ranges, zone);
5193       break;
5194     case 'd':
5195       AddClass(kDigitRanges, kDigitRangeCount, ranges, zone);
5196       break;
5197     case 'D':
5198       AddClassNegated(kDigitRanges, kDigitRangeCount, ranges, zone);
5199       break;
5200     case '.':
5201       AddClassNegated(kLineTerminatorRanges,
5202                       kLineTerminatorRangeCount,
5203                       ranges,
5204                       zone);
5205       break;
5206     // This is not a character range as defined by the spec but a
5207     // convenient shorthand for a character class that matches any
5208     // character.
5209     case '*':
5210       ranges->Add(CharacterRange::Everything(), zone);
5211       break;
5212     // This is the set of characters matched by the $ and ^ symbols
5213     // in multiline mode.
5214     case 'n':
5215       AddClass(kLineTerminatorRanges,
5216                kLineTerminatorRangeCount,
5217                ranges,
5218                zone);
5219       break;
5220     default:
5221       UNREACHABLE();
5222   }
5223 }
5224 
5225 
GetWordBounds()5226 Vector<const int> CharacterRange::GetWordBounds() {
5227   return Vector<const int>(kWordRanges, kWordRangeCount - 1);
5228 }
5229 
5230 
5231 class CharacterRangeSplitter {
5232  public:
CharacterRangeSplitter(ZoneList<CharacterRange> ** included,ZoneList<CharacterRange> ** excluded,Zone * zone)5233   CharacterRangeSplitter(ZoneList<CharacterRange>** included,
5234                          ZoneList<CharacterRange>** excluded,
5235                          Zone* zone)
5236       : included_(included),
5237         excluded_(excluded),
5238         zone_(zone) { }
5239   void Call(uc16 from, DispatchTable::Entry entry);
5240 
5241   static const int kInBase = 0;
5242   static const int kInOverlay = 1;
5243 
5244  private:
5245   ZoneList<CharacterRange>** included_;
5246   ZoneList<CharacterRange>** excluded_;
5247   Zone* zone_;
5248 };
5249 
5250 
Call(uc16 from,DispatchTable::Entry entry)5251 void CharacterRangeSplitter::Call(uc16 from, DispatchTable::Entry entry) {
5252   if (!entry.out_set()->Get(kInBase)) return;
5253   ZoneList<CharacterRange>** target = entry.out_set()->Get(kInOverlay)
5254     ? included_
5255     : excluded_;
5256   if (*target == NULL) *target = new(zone_) ZoneList<CharacterRange>(2, zone_);
5257   (*target)->Add(CharacterRange(entry.from(), entry.to()), zone_);
5258 }
5259 
5260 
Split(ZoneList<CharacterRange> * base,Vector<const int> overlay,ZoneList<CharacterRange> ** included,ZoneList<CharacterRange> ** excluded,Zone * zone)5261 void CharacterRange::Split(ZoneList<CharacterRange>* base,
5262                            Vector<const int> overlay,
5263                            ZoneList<CharacterRange>** included,
5264                            ZoneList<CharacterRange>** excluded,
5265                            Zone* zone) {
5266   ASSERT_EQ(NULL, *included);
5267   ASSERT_EQ(NULL, *excluded);
5268   DispatchTable table(zone);
5269   for (int i = 0; i < base->length(); i++)
5270     table.AddRange(base->at(i), CharacterRangeSplitter::kInBase, zone);
5271   for (int i = 0; i < overlay.length(); i += 2) {
5272     table.AddRange(CharacterRange(overlay[i], overlay[i + 1] - 1),
5273                    CharacterRangeSplitter::kInOverlay, zone);
5274   }
5275   CharacterRangeSplitter callback(included, excluded, zone);
5276   table.ForEach(&callback);
5277 }
5278 
5279 
AddCaseEquivalents(ZoneList<CharacterRange> * ranges,bool is_ascii,Zone * zone)5280 void CharacterRange::AddCaseEquivalents(ZoneList<CharacterRange>* ranges,
5281                                         bool is_ascii,
5282                                         Zone* zone) {
5283   Isolate* isolate = zone->isolate();
5284   uc16 bottom = from();
5285   uc16 top = to();
5286   if (is_ascii && !RangeContainsLatin1Equivalents(*this)) {
5287     if (bottom > String::kMaxOneByteCharCode) return;
5288     if (top > String::kMaxOneByteCharCode) top = String::kMaxOneByteCharCode;
5289   }
5290   unibrow::uchar chars[unibrow::Ecma262UnCanonicalize::kMaxWidth];
5291   if (top == bottom) {
5292     // If this is a singleton we just expand the one character.
5293     int length = isolate->jsregexp_uncanonicalize()->get(bottom, '\0', chars);
5294     for (int i = 0; i < length; i++) {
5295       uc32 chr = chars[i];
5296       if (chr != bottom) {
5297         ranges->Add(CharacterRange::Singleton(chars[i]), zone);
5298       }
5299     }
5300   } else {
5301     // If this is a range we expand the characters block by block,
5302     // expanding contiguous subranges (blocks) one at a time.
5303     // The approach is as follows.  For a given start character we
5304     // look up the remainder of the block that contains it (represented
5305     // by the end point), for instance we find 'z' if the character
5306     // is 'c'.  A block is characterized by the property
5307     // that all characters uncanonicalize in the same way, except that
5308     // each entry in the result is incremented by the distance from the first
5309     // element.  So a-z is a block because 'a' uncanonicalizes to ['a', 'A'] and
5310     // the k'th letter uncanonicalizes to ['a' + k, 'A' + k].
5311     // Once we've found the end point we look up its uncanonicalization
5312     // and produce a range for each element.  For instance for [c-f]
5313     // we look up ['z', 'Z'] and produce [c-f] and [C-F].  We then only
5314     // add a range if it is not already contained in the input, so [c-f]
5315     // will be skipped but [C-F] will be added.  If this range is not
5316     // completely contained in a block we do this for all the blocks
5317     // covered by the range (handling characters that is not in a block
5318     // as a "singleton block").
5319     unibrow::uchar range[unibrow::Ecma262UnCanonicalize::kMaxWidth];
5320     int pos = bottom;
5321     while (pos <= top) {
5322       int length = isolate->jsregexp_canonrange()->get(pos, '\0', range);
5323       uc16 block_end;
5324       if (length == 0) {
5325         block_end = pos;
5326       } else {
5327         ASSERT_EQ(1, length);
5328         block_end = range[0];
5329       }
5330       int end = (block_end > top) ? top : block_end;
5331       length = isolate->jsregexp_uncanonicalize()->get(block_end, '\0', range);
5332       for (int i = 0; i < length; i++) {
5333         uc32 c = range[i];
5334         uc16 range_from = c - (block_end - pos);
5335         uc16 range_to = c - (block_end - end);
5336         if (!(bottom <= range_from && range_to <= top)) {
5337           ranges->Add(CharacterRange(range_from, range_to), zone);
5338         }
5339       }
5340       pos = end + 1;
5341     }
5342   }
5343 }
5344 
5345 
IsCanonical(ZoneList<CharacterRange> * ranges)5346 bool CharacterRange::IsCanonical(ZoneList<CharacterRange>* ranges) {
5347   ASSERT_NOT_NULL(ranges);
5348   int n = ranges->length();
5349   if (n <= 1) return true;
5350   int max = ranges->at(0).to();
5351   for (int i = 1; i < n; i++) {
5352     CharacterRange next_range = ranges->at(i);
5353     if (next_range.from() <= max + 1) return false;
5354     max = next_range.to();
5355   }
5356   return true;
5357 }
5358 
5359 
ranges(Zone * zone)5360 ZoneList<CharacterRange>* CharacterSet::ranges(Zone* zone) {
5361   if (ranges_ == NULL) {
5362     ranges_ = new(zone) ZoneList<CharacterRange>(2, zone);
5363     CharacterRange::AddClassEscape(standard_set_type_, ranges_, zone);
5364   }
5365   return ranges_;
5366 }
5367 
5368 
5369 // Move a number of elements in a zonelist to another position
5370 // in the same list. Handles overlapping source and target areas.
MoveRanges(ZoneList<CharacterRange> * list,int from,int to,int count)5371 static void MoveRanges(ZoneList<CharacterRange>* list,
5372                        int from,
5373                        int to,
5374                        int count) {
5375   // Ranges are potentially overlapping.
5376   if (from < to) {
5377     for (int i = count - 1; i >= 0; i--) {
5378       list->at(to + i) = list->at(from + i);
5379     }
5380   } else {
5381     for (int i = 0; i < count; i++) {
5382       list->at(to + i) = list->at(from + i);
5383     }
5384   }
5385 }
5386 
5387 
InsertRangeInCanonicalList(ZoneList<CharacterRange> * list,int count,CharacterRange insert)5388 static int InsertRangeInCanonicalList(ZoneList<CharacterRange>* list,
5389                                       int count,
5390                                       CharacterRange insert) {
5391   // Inserts a range into list[0..count[, which must be sorted
5392   // by from value and non-overlapping and non-adjacent, using at most
5393   // list[0..count] for the result. Returns the number of resulting
5394   // canonicalized ranges. Inserting a range may collapse existing ranges into
5395   // fewer ranges, so the return value can be anything in the range 1..count+1.
5396   uc16 from = insert.from();
5397   uc16 to = insert.to();
5398   int start_pos = 0;
5399   int end_pos = count;
5400   for (int i = count - 1; i >= 0; i--) {
5401     CharacterRange current = list->at(i);
5402     if (current.from() > to + 1) {
5403       end_pos = i;
5404     } else if (current.to() + 1 < from) {
5405       start_pos = i + 1;
5406       break;
5407     }
5408   }
5409 
5410   // Inserted range overlaps, or is adjacent to, ranges at positions
5411   // [start_pos..end_pos[. Ranges before start_pos or at or after end_pos are
5412   // not affected by the insertion.
5413   // If start_pos == end_pos, the range must be inserted before start_pos.
5414   // if start_pos < end_pos, the entire range from start_pos to end_pos
5415   // must be merged with the insert range.
5416 
5417   if (start_pos == end_pos) {
5418     // Insert between existing ranges at position start_pos.
5419     if (start_pos < count) {
5420       MoveRanges(list, start_pos, start_pos + 1, count - start_pos);
5421     }
5422     list->at(start_pos) = insert;
5423     return count + 1;
5424   }
5425   if (start_pos + 1 == end_pos) {
5426     // Replace single existing range at position start_pos.
5427     CharacterRange to_replace = list->at(start_pos);
5428     int new_from = Min(to_replace.from(), from);
5429     int new_to = Max(to_replace.to(), to);
5430     list->at(start_pos) = CharacterRange(new_from, new_to);
5431     return count;
5432   }
5433   // Replace a number of existing ranges from start_pos to end_pos - 1.
5434   // Move the remaining ranges down.
5435 
5436   int new_from = Min(list->at(start_pos).from(), from);
5437   int new_to = Max(list->at(end_pos - 1).to(), to);
5438   if (end_pos < count) {
5439     MoveRanges(list, end_pos, start_pos + 1, count - end_pos);
5440   }
5441   list->at(start_pos) = CharacterRange(new_from, new_to);
5442   return count - (end_pos - start_pos) + 1;
5443 }
5444 
5445 
Canonicalize()5446 void CharacterSet::Canonicalize() {
5447   // Special/default classes are always considered canonical. The result
5448   // of calling ranges() will be sorted.
5449   if (ranges_ == NULL) return;
5450   CharacterRange::Canonicalize(ranges_);
5451 }
5452 
5453 
Canonicalize(ZoneList<CharacterRange> * character_ranges)5454 void CharacterRange::Canonicalize(ZoneList<CharacterRange>* character_ranges) {
5455   if (character_ranges->length() <= 1) return;
5456   // Check whether ranges are already canonical (increasing, non-overlapping,
5457   // non-adjacent).
5458   int n = character_ranges->length();
5459   int max = character_ranges->at(0).to();
5460   int i = 1;
5461   while (i < n) {
5462     CharacterRange current = character_ranges->at(i);
5463     if (current.from() <= max + 1) {
5464       break;
5465     }
5466     max = current.to();
5467     i++;
5468   }
5469   // Canonical until the i'th range. If that's all of them, we are done.
5470   if (i == n) return;
5471 
5472   // The ranges at index i and forward are not canonicalized. Make them so by
5473   // doing the equivalent of insertion sort (inserting each into the previous
5474   // list, in order).
5475   // Notice that inserting a range can reduce the number of ranges in the
5476   // result due to combining of adjacent and overlapping ranges.
5477   int read = i;  // Range to insert.
5478   int num_canonical = i;  // Length of canonicalized part of list.
5479   do {
5480     num_canonical = InsertRangeInCanonicalList(character_ranges,
5481                                                num_canonical,
5482                                                character_ranges->at(read));
5483     read++;
5484   } while (read < n);
5485   character_ranges->Rewind(num_canonical);
5486 
5487   ASSERT(CharacterRange::IsCanonical(character_ranges));
5488 }
5489 
5490 
Negate(ZoneList<CharacterRange> * ranges,ZoneList<CharacterRange> * negated_ranges,Zone * zone)5491 void CharacterRange::Negate(ZoneList<CharacterRange>* ranges,
5492                             ZoneList<CharacterRange>* negated_ranges,
5493                             Zone* zone) {
5494   ASSERT(CharacterRange::IsCanonical(ranges));
5495   ASSERT_EQ(0, negated_ranges->length());
5496   int range_count = ranges->length();
5497   uc16 from = 0;
5498   int i = 0;
5499   if (range_count > 0 && ranges->at(0).from() == 0) {
5500     from = ranges->at(0).to();
5501     i = 1;
5502   }
5503   while (i < range_count) {
5504     CharacterRange range = ranges->at(i);
5505     negated_ranges->Add(CharacterRange(from + 1, range.from() - 1), zone);
5506     from = range.to();
5507     i++;
5508   }
5509   if (from < String::kMaxUtf16CodeUnit) {
5510     negated_ranges->Add(CharacterRange(from + 1, String::kMaxUtf16CodeUnit),
5511                         zone);
5512   }
5513 }
5514 
5515 
5516 // -------------------------------------------------------------------
5517 // Splay tree
5518 
5519 
Extend(unsigned value,Zone * zone)5520 OutSet* OutSet::Extend(unsigned value, Zone* zone) {
5521   if (Get(value))
5522     return this;
5523   if (successors(zone) != NULL) {
5524     for (int i = 0; i < successors(zone)->length(); i++) {
5525       OutSet* successor = successors(zone)->at(i);
5526       if (successor->Get(value))
5527         return successor;
5528     }
5529   } else {
5530     successors_ = new(zone) ZoneList<OutSet*>(2, zone);
5531   }
5532   OutSet* result = new(zone) OutSet(first_, remaining_);
5533   result->Set(value, zone);
5534   successors(zone)->Add(result, zone);
5535   return result;
5536 }
5537 
5538 
Set(unsigned value,Zone * zone)5539 void OutSet::Set(unsigned value, Zone *zone) {
5540   if (value < kFirstLimit) {
5541     first_ |= (1 << value);
5542   } else {
5543     if (remaining_ == NULL)
5544       remaining_ = new(zone) ZoneList<unsigned>(1, zone);
5545     if (remaining_->is_empty() || !remaining_->Contains(value))
5546       remaining_->Add(value, zone);
5547   }
5548 }
5549 
5550 
Get(unsigned value) const5551 bool OutSet::Get(unsigned value) const {
5552   if (value < kFirstLimit) {
5553     return (first_ & (1 << value)) != 0;
5554   } else if (remaining_ == NULL) {
5555     return false;
5556   } else {
5557     return remaining_->Contains(value);
5558   }
5559 }
5560 
5561 
5562 const uc16 DispatchTable::Config::kNoKey = unibrow::Utf8::kBadChar;
5563 
5564 
AddRange(CharacterRange full_range,int value,Zone * zone)5565 void DispatchTable::AddRange(CharacterRange full_range, int value,
5566                              Zone* zone) {
5567   CharacterRange current = full_range;
5568   if (tree()->is_empty()) {
5569     // If this is the first range we just insert into the table.
5570     ZoneSplayTree<Config>::Locator loc;
5571     ASSERT_RESULT(tree()->Insert(current.from(), &loc));
5572     loc.set_value(Entry(current.from(), current.to(),
5573                         empty()->Extend(value, zone)));
5574     return;
5575   }
5576   // First see if there is a range to the left of this one that
5577   // overlaps.
5578   ZoneSplayTree<Config>::Locator loc;
5579   if (tree()->FindGreatestLessThan(current.from(), &loc)) {
5580     Entry* entry = &loc.value();
5581     // If we've found a range that overlaps with this one, and it
5582     // starts strictly to the left of this one, we have to fix it
5583     // because the following code only handles ranges that start on
5584     // or after the start point of the range we're adding.
5585     if (entry->from() < current.from() && entry->to() >= current.from()) {
5586       // Snap the overlapping range in half around the start point of
5587       // the range we're adding.
5588       CharacterRange left(entry->from(), current.from() - 1);
5589       CharacterRange right(current.from(), entry->to());
5590       // The left part of the overlapping range doesn't overlap.
5591       // Truncate the whole entry to be just the left part.
5592       entry->set_to(left.to());
5593       // The right part is the one that overlaps.  We add this part
5594       // to the map and let the next step deal with merging it with
5595       // the range we're adding.
5596       ZoneSplayTree<Config>::Locator loc;
5597       ASSERT_RESULT(tree()->Insert(right.from(), &loc));
5598       loc.set_value(Entry(right.from(),
5599                           right.to(),
5600                           entry->out_set()));
5601     }
5602   }
5603   while (current.is_valid()) {
5604     if (tree()->FindLeastGreaterThan(current.from(), &loc) &&
5605         (loc.value().from() <= current.to()) &&
5606         (loc.value().to() >= current.from())) {
5607       Entry* entry = &loc.value();
5608       // We have overlap.  If there is space between the start point of
5609       // the range we're adding and where the overlapping range starts
5610       // then we have to add a range covering just that space.
5611       if (current.from() < entry->from()) {
5612         ZoneSplayTree<Config>::Locator ins;
5613         ASSERT_RESULT(tree()->Insert(current.from(), &ins));
5614         ins.set_value(Entry(current.from(),
5615                             entry->from() - 1,
5616                             empty()->Extend(value, zone)));
5617         current.set_from(entry->from());
5618       }
5619       ASSERT_EQ(current.from(), entry->from());
5620       // If the overlapping range extends beyond the one we want to add
5621       // we have to snap the right part off and add it separately.
5622       if (entry->to() > current.to()) {
5623         ZoneSplayTree<Config>::Locator ins;
5624         ASSERT_RESULT(tree()->Insert(current.to() + 1, &ins));
5625         ins.set_value(Entry(current.to() + 1,
5626                             entry->to(),
5627                             entry->out_set()));
5628         entry->set_to(current.to());
5629       }
5630       ASSERT(entry->to() <= current.to());
5631       // The overlapping range is now completely contained by the range
5632       // we're adding so we can just update it and move the start point
5633       // of the range we're adding just past it.
5634       entry->AddValue(value, zone);
5635       // Bail out if the last interval ended at 0xFFFF since otherwise
5636       // adding 1 will wrap around to 0.
5637       if (entry->to() == String::kMaxUtf16CodeUnit)
5638         break;
5639       ASSERT(entry->to() + 1 > current.from());
5640       current.set_from(entry->to() + 1);
5641     } else {
5642       // There is no overlap so we can just add the range
5643       ZoneSplayTree<Config>::Locator ins;
5644       ASSERT_RESULT(tree()->Insert(current.from(), &ins));
5645       ins.set_value(Entry(current.from(),
5646                           current.to(),
5647                           empty()->Extend(value, zone)));
5648       break;
5649     }
5650   }
5651 }
5652 
5653 
Get(uc16 value)5654 OutSet* DispatchTable::Get(uc16 value) {
5655   ZoneSplayTree<Config>::Locator loc;
5656   if (!tree()->FindGreatestLessThan(value, &loc))
5657     return empty();
5658   Entry* entry = &loc.value();
5659   if (value <= entry->to())
5660     return entry->out_set();
5661   else
5662     return empty();
5663 }
5664 
5665 
5666 // -------------------------------------------------------------------
5667 // Analysis
5668 
5669 
EnsureAnalyzed(RegExpNode * that)5670 void Analysis::EnsureAnalyzed(RegExpNode* that) {
5671   StackLimitCheck check(that->zone()->isolate());
5672   if (check.HasOverflowed()) {
5673     fail("Stack overflow");
5674     return;
5675   }
5676   if (that->info()->been_analyzed || that->info()->being_analyzed)
5677     return;
5678   that->info()->being_analyzed = true;
5679   that->Accept(this);
5680   that->info()->being_analyzed = false;
5681   that->info()->been_analyzed = true;
5682 }
5683 
5684 
VisitEnd(EndNode * that)5685 void Analysis::VisitEnd(EndNode* that) {
5686   // nothing to do
5687 }
5688 
5689 
CalculateOffsets()5690 void TextNode::CalculateOffsets() {
5691   int element_count = elements()->length();
5692   // Set up the offsets of the elements relative to the start.  This is a fixed
5693   // quantity since a TextNode can only contain fixed-width things.
5694   int cp_offset = 0;
5695   for (int i = 0; i < element_count; i++) {
5696     TextElement& elm = elements()->at(i);
5697     elm.set_cp_offset(cp_offset);
5698     cp_offset += elm.length();
5699   }
5700 }
5701 
5702 
VisitText(TextNode * that)5703 void Analysis::VisitText(TextNode* that) {
5704   if (ignore_case_) {
5705     that->MakeCaseIndependent(is_ascii_);
5706   }
5707   EnsureAnalyzed(that->on_success());
5708   if (!has_failed()) {
5709     that->CalculateOffsets();
5710   }
5711 }
5712 
5713 
VisitAction(ActionNode * that)5714 void Analysis::VisitAction(ActionNode* that) {
5715   RegExpNode* target = that->on_success();
5716   EnsureAnalyzed(target);
5717   if (!has_failed()) {
5718     // If the next node is interested in what it follows then this node
5719     // has to be interested too so it can pass the information on.
5720     that->info()->AddFromFollowing(target->info());
5721   }
5722 }
5723 
5724 
VisitChoice(ChoiceNode * that)5725 void Analysis::VisitChoice(ChoiceNode* that) {
5726   NodeInfo* info = that->info();
5727   for (int i = 0; i < that->alternatives()->length(); i++) {
5728     RegExpNode* node = that->alternatives()->at(i).node();
5729     EnsureAnalyzed(node);
5730     if (has_failed()) return;
5731     // Anything the following nodes need to know has to be known by
5732     // this node also, so it can pass it on.
5733     info->AddFromFollowing(node->info());
5734   }
5735 }
5736 
5737 
VisitLoopChoice(LoopChoiceNode * that)5738 void Analysis::VisitLoopChoice(LoopChoiceNode* that) {
5739   NodeInfo* info = that->info();
5740   for (int i = 0; i < that->alternatives()->length(); i++) {
5741     RegExpNode* node = that->alternatives()->at(i).node();
5742     if (node != that->loop_node()) {
5743       EnsureAnalyzed(node);
5744       if (has_failed()) return;
5745       info->AddFromFollowing(node->info());
5746     }
5747   }
5748   // Check the loop last since it may need the value of this node
5749   // to get a correct result.
5750   EnsureAnalyzed(that->loop_node());
5751   if (!has_failed()) {
5752     info->AddFromFollowing(that->loop_node()->info());
5753   }
5754 }
5755 
5756 
VisitBackReference(BackReferenceNode * that)5757 void Analysis::VisitBackReference(BackReferenceNode* that) {
5758   EnsureAnalyzed(that->on_success());
5759 }
5760 
5761 
VisitAssertion(AssertionNode * that)5762 void Analysis::VisitAssertion(AssertionNode* that) {
5763   EnsureAnalyzed(that->on_success());
5764 }
5765 
5766 
FillInBMInfo(int offset,int budget,BoyerMooreLookahead * bm,bool not_at_start)5767 void BackReferenceNode::FillInBMInfo(int offset,
5768                                      int budget,
5769                                      BoyerMooreLookahead* bm,
5770                                      bool not_at_start) {
5771   // Working out the set of characters that a backreference can match is too
5772   // hard, so we just say that any character can match.
5773   bm->SetRest(offset);
5774   SaveBMInfo(bm, not_at_start, offset);
5775 }
5776 
5777 
5778 STATIC_ASSERT(BoyerMoorePositionInfo::kMapSize ==
5779               RegExpMacroAssembler::kTableSize);
5780 
5781 
FillInBMInfo(int offset,int budget,BoyerMooreLookahead * bm,bool not_at_start)5782 void ChoiceNode::FillInBMInfo(int offset,
5783                               int budget,
5784                               BoyerMooreLookahead* bm,
5785                               bool not_at_start) {
5786   ZoneList<GuardedAlternative>* alts = alternatives();
5787   budget = (budget - 1) / alts->length();
5788   for (int i = 0; i < alts->length(); i++) {
5789     GuardedAlternative& alt = alts->at(i);
5790     if (alt.guards() != NULL && alt.guards()->length() != 0) {
5791       bm->SetRest(offset);  // Give up trying to fill in info.
5792       SaveBMInfo(bm, not_at_start, offset);
5793       return;
5794     }
5795     alt.node()->FillInBMInfo(offset, budget, bm, not_at_start);
5796   }
5797   SaveBMInfo(bm, not_at_start, offset);
5798 }
5799 
5800 
FillInBMInfo(int initial_offset,int budget,BoyerMooreLookahead * bm,bool not_at_start)5801 void TextNode::FillInBMInfo(int initial_offset,
5802                             int budget,
5803                             BoyerMooreLookahead* bm,
5804                             bool not_at_start) {
5805   if (initial_offset >= bm->length()) return;
5806   int offset = initial_offset;
5807   int max_char = bm->max_char();
5808   for (int i = 0; i < elements()->length(); i++) {
5809     if (offset >= bm->length()) {
5810       if (initial_offset == 0) set_bm_info(not_at_start, bm);
5811       return;
5812     }
5813     TextElement text = elements()->at(i);
5814     if (text.text_type() == TextElement::ATOM) {
5815       RegExpAtom* atom = text.atom();
5816       for (int j = 0; j < atom->length(); j++, offset++) {
5817         if (offset >= bm->length()) {
5818           if (initial_offset == 0) set_bm_info(not_at_start, bm);
5819           return;
5820         }
5821         uc16 character = atom->data()[j];
5822         if (bm->compiler()->ignore_case()) {
5823           unibrow::uchar chars[unibrow::Ecma262UnCanonicalize::kMaxWidth];
5824           int length = GetCaseIndependentLetters(
5825               Isolate::Current(),
5826               character,
5827               bm->max_char() == String::kMaxOneByteCharCode,
5828               chars);
5829           for (int j = 0; j < length; j++) {
5830             bm->Set(offset, chars[j]);
5831           }
5832         } else {
5833           if (character <= max_char) bm->Set(offset, character);
5834         }
5835       }
5836     } else {
5837       ASSERT_EQ(TextElement::CHAR_CLASS, text.text_type());
5838       RegExpCharacterClass* char_class = text.char_class();
5839       ZoneList<CharacterRange>* ranges = char_class->ranges(zone());
5840       if (char_class->is_negated()) {
5841         bm->SetAll(offset);
5842       } else {
5843         for (int k = 0; k < ranges->length(); k++) {
5844           CharacterRange& range = ranges->at(k);
5845           if (range.from() > max_char) continue;
5846           int to = Min(max_char, static_cast<int>(range.to()));
5847           bm->SetInterval(offset, Interval(range.from(), to));
5848         }
5849       }
5850       offset++;
5851     }
5852   }
5853   if (offset >= bm->length()) {
5854     if (initial_offset == 0) set_bm_info(not_at_start, bm);
5855     return;
5856   }
5857   on_success()->FillInBMInfo(offset,
5858                              budget - 1,
5859                              bm,
5860                              true);  // Not at start after a text node.
5861   if (initial_offset == 0) set_bm_info(not_at_start, bm);
5862 }
5863 
5864 
5865 // -------------------------------------------------------------------
5866 // Dispatch table construction
5867 
5868 
VisitEnd(EndNode * that)5869 void DispatchTableConstructor::VisitEnd(EndNode* that) {
5870   AddRange(CharacterRange::Everything());
5871 }
5872 
5873 
BuildTable(ChoiceNode * node)5874 void DispatchTableConstructor::BuildTable(ChoiceNode* node) {
5875   node->set_being_calculated(true);
5876   ZoneList<GuardedAlternative>* alternatives = node->alternatives();
5877   for (int i = 0; i < alternatives->length(); i++) {
5878     set_choice_index(i);
5879     alternatives->at(i).node()->Accept(this);
5880   }
5881   node->set_being_calculated(false);
5882 }
5883 
5884 
5885 class AddDispatchRange {
5886  public:
AddDispatchRange(DispatchTableConstructor * constructor)5887   explicit AddDispatchRange(DispatchTableConstructor* constructor)
5888     : constructor_(constructor) { }
5889   void Call(uc32 from, DispatchTable::Entry entry);
5890  private:
5891   DispatchTableConstructor* constructor_;
5892 };
5893 
5894 
Call(uc32 from,DispatchTable::Entry entry)5895 void AddDispatchRange::Call(uc32 from, DispatchTable::Entry entry) {
5896   CharacterRange range(from, entry.to());
5897   constructor_->AddRange(range);
5898 }
5899 
5900 
VisitChoice(ChoiceNode * node)5901 void DispatchTableConstructor::VisitChoice(ChoiceNode* node) {
5902   if (node->being_calculated())
5903     return;
5904   DispatchTable* table = node->GetTable(ignore_case_);
5905   AddDispatchRange adder(this);
5906   table->ForEach(&adder);
5907 }
5908 
5909 
VisitBackReference(BackReferenceNode * that)5910 void DispatchTableConstructor::VisitBackReference(BackReferenceNode* that) {
5911   // TODO(160): Find the node that we refer back to and propagate its start
5912   // set back to here.  For now we just accept anything.
5913   AddRange(CharacterRange::Everything());
5914 }
5915 
5916 
VisitAssertion(AssertionNode * that)5917 void DispatchTableConstructor::VisitAssertion(AssertionNode* that) {
5918   RegExpNode* target = that->on_success();
5919   target->Accept(this);
5920 }
5921 
5922 
CompareRangeByFrom(const CharacterRange * a,const CharacterRange * b)5923 static int CompareRangeByFrom(const CharacterRange* a,
5924                               const CharacterRange* b) {
5925   return Compare<uc16>(a->from(), b->from());
5926 }
5927 
5928 
AddInverse(ZoneList<CharacterRange> * ranges)5929 void DispatchTableConstructor::AddInverse(ZoneList<CharacterRange>* ranges) {
5930   ranges->Sort(CompareRangeByFrom);
5931   uc16 last = 0;
5932   for (int i = 0; i < ranges->length(); i++) {
5933     CharacterRange range = ranges->at(i);
5934     if (last < range.from())
5935       AddRange(CharacterRange(last, range.from() - 1));
5936     if (range.to() >= last) {
5937       if (range.to() == String::kMaxUtf16CodeUnit) {
5938         return;
5939       } else {
5940         last = range.to() + 1;
5941       }
5942     }
5943   }
5944   AddRange(CharacterRange(last, String::kMaxUtf16CodeUnit));
5945 }
5946 
5947 
VisitText(TextNode * that)5948 void DispatchTableConstructor::VisitText(TextNode* that) {
5949   TextElement elm = that->elements()->at(0);
5950   switch (elm.text_type()) {
5951     case TextElement::ATOM: {
5952       uc16 c = elm.atom()->data()[0];
5953       AddRange(CharacterRange(c, c));
5954       break;
5955     }
5956     case TextElement::CHAR_CLASS: {
5957       RegExpCharacterClass* tree = elm.char_class();
5958       ZoneList<CharacterRange>* ranges = tree->ranges(that->zone());
5959       if (tree->is_negated()) {
5960         AddInverse(ranges);
5961       } else {
5962         for (int i = 0; i < ranges->length(); i++)
5963           AddRange(ranges->at(i));
5964       }
5965       break;
5966     }
5967     default: {
5968       UNIMPLEMENTED();
5969     }
5970   }
5971 }
5972 
5973 
VisitAction(ActionNode * that)5974 void DispatchTableConstructor::VisitAction(ActionNode* that) {
5975   RegExpNode* target = that->on_success();
5976   target->Accept(this);
5977 }
5978 
5979 
Compile(RegExpCompileData * data,bool ignore_case,bool is_global,bool is_multiline,Handle<String> pattern,Handle<String> sample_subject,bool is_ascii,Zone * zone)5980 RegExpEngine::CompilationResult RegExpEngine::Compile(
5981     RegExpCompileData* data,
5982     bool ignore_case,
5983     bool is_global,
5984     bool is_multiline,
5985     Handle<String> pattern,
5986     Handle<String> sample_subject,
5987     bool is_ascii,
5988     Zone* zone) {
5989   if ((data->capture_count + 1) * 2 - 1 > RegExpMacroAssembler::kMaxRegister) {
5990     return IrregexpRegExpTooBig(zone->isolate());
5991   }
5992   RegExpCompiler compiler(data->capture_count, ignore_case, is_ascii, zone);
5993 
5994   // Sample some characters from the middle of the string.
5995   static const int kSampleSize = 128;
5996 
5997   sample_subject = String::Flatten(sample_subject);
5998   int chars_sampled = 0;
5999   int half_way = (sample_subject->length() - kSampleSize) / 2;
6000   for (int i = Max(0, half_way);
6001        i < sample_subject->length() && chars_sampled < kSampleSize;
6002        i++, chars_sampled++) {
6003     compiler.frequency_collator()->CountCharacter(sample_subject->Get(i));
6004   }
6005 
6006   // Wrap the body of the regexp in capture #0.
6007   RegExpNode* captured_body = RegExpCapture::ToNode(data->tree,
6008                                                     0,
6009                                                     &compiler,
6010                                                     compiler.accept());
6011   RegExpNode* node = captured_body;
6012   bool is_end_anchored = data->tree->IsAnchoredAtEnd();
6013   bool is_start_anchored = data->tree->IsAnchoredAtStart();
6014   int max_length = data->tree->max_match();
6015   if (!is_start_anchored) {
6016     // Add a .*? at the beginning, outside the body capture, unless
6017     // this expression is anchored at the beginning.
6018     RegExpNode* loop_node =
6019         RegExpQuantifier::ToNode(0,
6020                                  RegExpTree::kInfinity,
6021                                  false,
6022                                  new(zone) RegExpCharacterClass('*'),
6023                                  &compiler,
6024                                  captured_body,
6025                                  data->contains_anchor);
6026 
6027     if (data->contains_anchor) {
6028       // Unroll loop once, to take care of the case that might start
6029       // at the start of input.
6030       ChoiceNode* first_step_node = new(zone) ChoiceNode(2, zone);
6031       first_step_node->AddAlternative(GuardedAlternative(captured_body));
6032       first_step_node->AddAlternative(GuardedAlternative(
6033           new(zone) TextNode(new(zone) RegExpCharacterClass('*'), loop_node)));
6034       node = first_step_node;
6035     } else {
6036       node = loop_node;
6037     }
6038   }
6039   if (is_ascii) {
6040     node = node->FilterASCII(RegExpCompiler::kMaxRecursion, ignore_case);
6041     // Do it again to propagate the new nodes to places where they were not
6042     // put because they had not been calculated yet.
6043     if (node != NULL) {
6044       node = node->FilterASCII(RegExpCompiler::kMaxRecursion, ignore_case);
6045     }
6046   }
6047 
6048   if (node == NULL) node = new(zone) EndNode(EndNode::BACKTRACK, zone);
6049   data->node = node;
6050   Analysis analysis(ignore_case, is_ascii);
6051   analysis.EnsureAnalyzed(node);
6052   if (analysis.has_failed()) {
6053     const char* error_message = analysis.error_message();
6054     return CompilationResult(zone->isolate(), error_message);
6055   }
6056 
6057   // Create the correct assembler for the architecture.
6058 #ifndef V8_INTERPRETED_REGEXP
6059   // Native regexp implementation.
6060 
6061   NativeRegExpMacroAssembler::Mode mode =
6062       is_ascii ? NativeRegExpMacroAssembler::ASCII
6063                : NativeRegExpMacroAssembler::UC16;
6064 
6065 #if V8_TARGET_ARCH_IA32
6066   RegExpMacroAssemblerIA32 macro_assembler(mode, (data->capture_count + 1) * 2,
6067                                            zone);
6068 #elif V8_TARGET_ARCH_X64
6069   RegExpMacroAssemblerX64 macro_assembler(mode, (data->capture_count + 1) * 2,
6070                                           zone);
6071 #elif V8_TARGET_ARCH_ARM
6072   RegExpMacroAssemblerARM macro_assembler(mode, (data->capture_count + 1) * 2,
6073                                           zone);
6074 #elif V8_TARGET_ARCH_ARM64
6075   RegExpMacroAssemblerARM64 macro_assembler(mode, (data->capture_count + 1) * 2,
6076                                             zone);
6077 #elif V8_TARGET_ARCH_MIPS
6078   RegExpMacroAssemblerMIPS macro_assembler(mode, (data->capture_count + 1) * 2,
6079                                            zone);
6080 #elif V8_TARGET_ARCH_X87
6081   RegExpMacroAssemblerX87 macro_assembler(mode, (data->capture_count + 1) * 2,
6082                                           zone);
6083 #else
6084 #error "Unsupported architecture"
6085 #endif
6086 
6087 #else  // V8_INTERPRETED_REGEXP
6088   // Interpreted regexp implementation.
6089   EmbeddedVector<byte, 1024> codes;
6090   RegExpMacroAssemblerIrregexp macro_assembler(codes, zone);
6091 #endif  // V8_INTERPRETED_REGEXP
6092 
6093   // Inserted here, instead of in Assembler, because it depends on information
6094   // in the AST that isn't replicated in the Node structure.
6095   static const int kMaxBacksearchLimit = 1024;
6096   if (is_end_anchored &&
6097       !is_start_anchored &&
6098       max_length < kMaxBacksearchLimit) {
6099     macro_assembler.SetCurrentPositionFromEnd(max_length);
6100   }
6101 
6102   if (is_global) {
6103     macro_assembler.set_global_mode(
6104         (data->tree->min_match() > 0)
6105             ? RegExpMacroAssembler::GLOBAL_NO_ZERO_LENGTH_CHECK
6106             : RegExpMacroAssembler::GLOBAL);
6107   }
6108 
6109   return compiler.Assemble(&macro_assembler,
6110                            node,
6111                            data->capture_count,
6112                            pattern);
6113 }
6114 
6115 
6116 }}  // namespace v8::internal
6117