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/base/platform/platform.h"
9 #include "src/compilation-cache.h"
10 #include "src/compiler.h"
11 #include "src/execution.h"
12 #include "src/factory.h"
13 #include "src/jsregexp-inl.h"
14 #include "src/jsregexp.h"
15 #include "src/ostreams.h"
16 #include "src/parser.h"
17 #include "src/regexp-macro-assembler.h"
18 #include "src/regexp-macro-assembler-irregexp.h"
19 #include "src/regexp-macro-assembler-tracer.h"
20 #include "src/regexp-stack.h"
21 #include "src/runtime.h"
22 #include "src/string-search.h"
23 
24 #ifndef V8_INTERPRETED_REGEXP
25 #if V8_TARGET_ARCH_IA32
26 #include "src/ia32/regexp-macro-assembler-ia32.h"  // NOLINT
27 #elif V8_TARGET_ARCH_X64
28 #include "src/x64/regexp-macro-assembler-x64.h"  // NOLINT
29 #elif V8_TARGET_ARCH_ARM64
30 #include "src/arm64/regexp-macro-assembler-arm64.h"  // NOLINT
31 #elif V8_TARGET_ARCH_ARM
32 #include "src/arm/regexp-macro-assembler-arm.h"  // NOLINT
33 #elif V8_TARGET_ARCH_MIPS
34 #include "src/mips/regexp-macro-assembler-mips.h"  // NOLINT
35 #elif V8_TARGET_ARCH_MIPS64
36 #include "src/mips64/regexp-macro-assembler-mips64.h"  // NOLINT
37 #elif V8_TARGET_ARCH_X87
38 #include "src/x87/regexp-macro-assembler-x87.h"  // NOLINT
39 #else
40 #error Unsupported target architecture.
41 #endif
42 #endif
43 
44 #include "src/interpreter-irregexp.h"
45 
46 
47 namespace v8 {
48 namespace internal {
49 
CreateRegExpLiteral(Handle<JSFunction> constructor,Handle<String> pattern,Handle<String> flags)50 MaybeHandle<Object> RegExpImpl::CreateRegExpLiteral(
51     Handle<JSFunction> constructor,
52     Handle<String> pattern,
53     Handle<String> flags) {
54   // Call the construct code with 2 arguments.
55   Handle<Object> argv[] = { pattern, flags };
56   return Execution::New(constructor, arraysize(argv), argv);
57 }
58 
59 
RegExpFlagsFromString(Handle<String> str)60 static JSRegExp::Flags RegExpFlagsFromString(Handle<String> str) {
61   int flags = JSRegExp::NONE;
62   for (int i = 0; i < str->length(); i++) {
63     switch (str->Get(i)) {
64       case 'i':
65         flags |= JSRegExp::IGNORE_CASE;
66         break;
67       case 'g':
68         flags |= JSRegExp::GLOBAL;
69         break;
70       case 'm':
71         flags |= JSRegExp::MULTILINE;
72         break;
73       case 'y':
74         if (FLAG_harmony_regexps) flags |= JSRegExp::STICKY;
75         break;
76     }
77   }
78   return JSRegExp::Flags(flags);
79 }
80 
81 
82 MUST_USE_RESULT
ThrowRegExpException(Handle<JSRegExp> re,Handle<String> pattern,Handle<String> error_text,const char * message)83 static inline MaybeHandle<Object> ThrowRegExpException(
84     Handle<JSRegExp> re,
85     Handle<String> pattern,
86     Handle<String> error_text,
87     const char* message) {
88   Isolate* isolate = re->GetIsolate();
89   Factory* factory = isolate->factory();
90   Handle<FixedArray> elements = factory->NewFixedArray(2);
91   elements->set(0, *pattern);
92   elements->set(1, *error_text);
93   Handle<JSArray> array = factory->NewJSArrayWithElements(elements);
94   Handle<Object> regexp_err;
95   THROW_NEW_ERROR(isolate, NewSyntaxError(message, array), Object);
96 }
97 
98 
AddRange(ContainedInLattice containment,const int * ranges,int ranges_length,Interval new_range)99 ContainedInLattice AddRange(ContainedInLattice containment,
100                             const int* ranges,
101                             int ranges_length,
102                             Interval new_range) {
103   DCHECK((ranges_length & 1) == 1);
104   DCHECK(ranges[ranges_length - 1] == String::kMaxUtf16CodeUnit + 1);
105   if (containment == kLatticeUnknown) return containment;
106   bool inside = false;
107   int last = 0;
108   for (int i = 0; i < ranges_length; inside = !inside, last = ranges[i], i++) {
109     // Consider the range from last to ranges[i].
110     // We haven't got to the new range yet.
111     if (ranges[i] <= new_range.from()) continue;
112     // New range is wholly inside last-ranges[i].  Note that new_range.to() is
113     // inclusive, but the values in ranges are not.
114     if (last <= new_range.from() && new_range.to() < ranges[i]) {
115       return Combine(containment, inside ? kLatticeIn : kLatticeOut);
116     }
117     return kLatticeUnknown;
118   }
119   return containment;
120 }
121 
122 
123 // More makes code generation slower, less makes V8 benchmark score lower.
124 const int kMaxLookaheadForBoyerMoore = 8;
125 // In a 3-character pattern you can maximally step forwards 3 characters
126 // at a time, which is not always enough to pay for the extra logic.
127 const int kPatternTooShortForBoyerMoore = 2;
128 
129 
130 // Identifies the sort of regexps where the regexp engine is faster
131 // than the code used for atom matches.
HasFewDifferentCharacters(Handle<String> pattern)132 static bool HasFewDifferentCharacters(Handle<String> pattern) {
133   int length = Min(kMaxLookaheadForBoyerMoore, pattern->length());
134   if (length <= kPatternTooShortForBoyerMoore) return false;
135   const int kMod = 128;
136   bool character_found[kMod];
137   int different = 0;
138   memset(&character_found[0], 0, sizeof(character_found));
139   for (int i = 0; i < length; i++) {
140     int ch = (pattern->Get(i) & (kMod - 1));
141     if (!character_found[ch]) {
142       character_found[ch] = true;
143       different++;
144       // We declare a regexp low-alphabet if it has at least 3 times as many
145       // characters as it has different characters.
146       if (different * 3 > length) return false;
147     }
148   }
149   return true;
150 }
151 
152 
153 // Generic RegExp methods. Dispatches to implementation specific methods.
154 
155 
Compile(Handle<JSRegExp> re,Handle<String> pattern,Handle<String> flag_str)156 MaybeHandle<Object> RegExpImpl::Compile(Handle<JSRegExp> re,
157                                         Handle<String> pattern,
158                                         Handle<String> flag_str) {
159   Isolate* isolate = re->GetIsolate();
160   Zone zone(isolate);
161   JSRegExp::Flags flags = RegExpFlagsFromString(flag_str);
162   CompilationCache* compilation_cache = isolate->compilation_cache();
163   MaybeHandle<FixedArray> maybe_cached =
164       compilation_cache->LookupRegExp(pattern, flags);
165   Handle<FixedArray> cached;
166   bool in_cache = maybe_cached.ToHandle(&cached);
167   LOG(isolate, RegExpCompileEvent(re, in_cache));
168 
169   Handle<Object> result;
170   if (in_cache) {
171     re->set_data(*cached);
172     return re;
173   }
174   pattern = String::Flatten(pattern);
175   PostponeInterruptsScope postpone(isolate);
176   RegExpCompileData parse_result;
177   FlatStringReader reader(isolate, pattern);
178   if (!RegExpParser::ParseRegExp(&reader, flags.is_multiline(),
179                                  &parse_result, &zone)) {
180     // Throw an exception if we fail to parse the pattern.
181     return ThrowRegExpException(re,
182                                 pattern,
183                                 parse_result.error,
184                                 "malformed_regexp");
185   }
186 
187   bool has_been_compiled = false;
188 
189   if (parse_result.simple &&
190       !flags.is_ignore_case() &&
191       !flags.is_sticky() &&
192       !HasFewDifferentCharacters(pattern)) {
193     // Parse-tree is a single atom that is equal to the pattern.
194     AtomCompile(re, pattern, flags, pattern);
195     has_been_compiled = true;
196   } else if (parse_result.tree->IsAtom() &&
197       !flags.is_ignore_case() &&
198       !flags.is_sticky() &&
199       parse_result.capture_count == 0) {
200     RegExpAtom* atom = parse_result.tree->AsAtom();
201     Vector<const uc16> atom_pattern = atom->data();
202     Handle<String> atom_string;
203     ASSIGN_RETURN_ON_EXCEPTION(
204         isolate, atom_string,
205         isolate->factory()->NewStringFromTwoByte(atom_pattern),
206         Object);
207     if (!HasFewDifferentCharacters(atom_string)) {
208       AtomCompile(re, pattern, flags, atom_string);
209       has_been_compiled = true;
210     }
211   }
212   if (!has_been_compiled) {
213     IrregexpInitialize(re, pattern, flags, parse_result.capture_count);
214   }
215   DCHECK(re->data()->IsFixedArray());
216   // Compilation succeeded so the data is set on the regexp
217   // and we can store it in the cache.
218   Handle<FixedArray> data(FixedArray::cast(re->data()));
219   compilation_cache->PutRegExp(pattern, flags, data);
220 
221   return re;
222 }
223 
224 
Exec(Handle<JSRegExp> regexp,Handle<String> subject,int index,Handle<JSArray> last_match_info)225 MaybeHandle<Object> RegExpImpl::Exec(Handle<JSRegExp> regexp,
226                                      Handle<String> subject,
227                                      int index,
228                                      Handle<JSArray> last_match_info) {
229   switch (regexp->TypeTag()) {
230     case JSRegExp::ATOM:
231       return AtomExec(regexp, subject, index, last_match_info);
232     case JSRegExp::IRREGEXP: {
233       return IrregexpExec(regexp, subject, index, last_match_info);
234     }
235     default:
236       UNREACHABLE();
237       return MaybeHandle<Object>();
238   }
239 }
240 
241 
242 // RegExp Atom implementation: Simple string search using indexOf.
243 
244 
AtomCompile(Handle<JSRegExp> re,Handle<String> pattern,JSRegExp::Flags flags,Handle<String> match_pattern)245 void RegExpImpl::AtomCompile(Handle<JSRegExp> re,
246                              Handle<String> pattern,
247                              JSRegExp::Flags flags,
248                              Handle<String> match_pattern) {
249   re->GetIsolate()->factory()->SetRegExpAtomData(re,
250                                                  JSRegExp::ATOM,
251                                                  pattern,
252                                                  flags,
253                                                  match_pattern);
254 }
255 
256 
SetAtomLastCapture(FixedArray * array,String * subject,int from,int to)257 static void SetAtomLastCapture(FixedArray* array,
258                                String* subject,
259                                int from,
260                                int to) {
261   SealHandleScope shs(array->GetIsolate());
262   RegExpImpl::SetLastCaptureCount(array, 2);
263   RegExpImpl::SetLastSubject(array, subject);
264   RegExpImpl::SetLastInput(array, subject);
265   RegExpImpl::SetCapture(array, 0, from);
266   RegExpImpl::SetCapture(array, 1, to);
267 }
268 
269 
AtomExecRaw(Handle<JSRegExp> regexp,Handle<String> subject,int index,int32_t * output,int output_size)270 int RegExpImpl::AtomExecRaw(Handle<JSRegExp> regexp,
271                             Handle<String> subject,
272                             int index,
273                             int32_t* output,
274                             int output_size) {
275   Isolate* isolate = regexp->GetIsolate();
276 
277   DCHECK(0 <= index);
278   DCHECK(index <= subject->length());
279 
280   subject = String::Flatten(subject);
281   DisallowHeapAllocation no_gc;  // ensure vectors stay valid
282 
283   String* needle = String::cast(regexp->DataAt(JSRegExp::kAtomPatternIndex));
284   int needle_len = needle->length();
285   DCHECK(needle->IsFlat());
286   DCHECK_LT(0, needle_len);
287 
288   if (index + needle_len > subject->length()) {
289     return RegExpImpl::RE_FAILURE;
290   }
291 
292   for (int i = 0; i < output_size; i += 2) {
293     String::FlatContent needle_content = needle->GetFlatContent();
294     String::FlatContent subject_content = subject->GetFlatContent();
295     DCHECK(needle_content.IsFlat());
296     DCHECK(subject_content.IsFlat());
297     // dispatch on type of strings
298     index =
299         (needle_content.IsOneByte()
300              ? (subject_content.IsOneByte()
301                     ? SearchString(isolate, subject_content.ToOneByteVector(),
302                                    needle_content.ToOneByteVector(), index)
303                     : SearchString(isolate, subject_content.ToUC16Vector(),
304                                    needle_content.ToOneByteVector(), index))
305              : (subject_content.IsOneByte()
306                     ? SearchString(isolate, subject_content.ToOneByteVector(),
307                                    needle_content.ToUC16Vector(), index)
308                     : SearchString(isolate, subject_content.ToUC16Vector(),
309                                    needle_content.ToUC16Vector(), 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   DCHECK_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 one-byte or two-byte subject 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_one_byte)352 bool RegExpImpl::EnsureCompiledIrregexp(Handle<JSRegExp> re,
353                                         Handle<String> sample_subject,
354                                         bool is_one_byte) {
355   Object* compiled_code = re->DataAt(JSRegExp::code_index(is_one_byte));
356 #ifdef V8_INTERPRETED_REGEXP
357   if (compiled_code->IsByteArray()) return true;
358 #else  // V8_INTERPRETED_REGEXP (RegExp native code)
359   if (compiled_code->IsCode()) return true;
360 #endif
361   // We could potentially have marked this as flushable, but have kept
362   // a saved version if we did not flush it yet.
363   Object* saved_code = re->DataAt(JSRegExp::saved_code_index(is_one_byte));
364   if (saved_code->IsCode()) {
365     // Reinstate the code in the original place.
366     re->SetDataAt(JSRegExp::code_index(is_one_byte), saved_code);
367     DCHECK(compiled_code->IsSmi());
368     return true;
369   }
370   return CompileIrregexp(re, sample_subject, is_one_byte);
371 }
372 
373 
CreateRegExpErrorObjectAndThrow(Handle<JSRegExp> re,Handle<String> error_message,Isolate * isolate)374 static void CreateRegExpErrorObjectAndThrow(Handle<JSRegExp> re,
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> error;
383   MaybeHandle<Object> maybe_error =
384       factory->NewSyntaxError("malformed_regexp", array);
385   if (maybe_error.ToHandle(&error)) isolate->Throw(*error);
386 }
387 
388 
CompileIrregexp(Handle<JSRegExp> re,Handle<String> sample_subject,bool is_one_byte)389 bool RegExpImpl::CompileIrregexp(Handle<JSRegExp> re,
390                                  Handle<String> sample_subject,
391                                  bool is_one_byte) {
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_one_byte));
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   DCHECK(entry->IsSmi());
403   int entry_value = Smi::cast(entry)->value();
404   DCHECK(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_one_byte));
413     DCHECK(error_string->IsString());
414     Handle<String> error_message(String::cast(error_string));
415     CreateRegExpErrorObjectAndThrow(re, 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 = RegExpEngine::Compile(
437       &compile_data, flags.is_ignore_case(), flags.is_global(),
438       flags.is_multiline(), flags.is_sticky(), pattern, sample_subject,
439       is_one_byte, &zone);
440   if (result.error_message != NULL) {
441     // Unable to compile regexp.
442     Handle<String> error_message = isolate->factory()->NewStringFromUtf8(
443         CStrVector(result.error_message)).ToHandleChecked();
444     CreateRegExpErrorObjectAndThrow(re, error_message, isolate);
445     return false;
446   }
447 
448   Handle<FixedArray> data = Handle<FixedArray>(FixedArray::cast(re->data()));
449   data->set(JSRegExp::code_index(is_one_byte), result.code);
450   int register_max = IrregexpMaxRegisterCount(*data);
451   if (result.num_registers > register_max) {
452     SetIrregexpMaxRegisterCount(*data, result.num_registers);
453   }
454 
455   return true;
456 }
457 
458 
IrregexpMaxRegisterCount(FixedArray * re)459 int RegExpImpl::IrregexpMaxRegisterCount(FixedArray* re) {
460   return Smi::cast(
461       re->get(JSRegExp::kIrregexpMaxRegisterCountIndex))->value();
462 }
463 
464 
SetIrregexpMaxRegisterCount(FixedArray * re,int value)465 void RegExpImpl::SetIrregexpMaxRegisterCount(FixedArray* re, int value) {
466   re->set(JSRegExp::kIrregexpMaxRegisterCountIndex, Smi::FromInt(value));
467 }
468 
469 
IrregexpNumberOfCaptures(FixedArray * re)470 int RegExpImpl::IrregexpNumberOfCaptures(FixedArray* re) {
471   return Smi::cast(re->get(JSRegExp::kIrregexpCaptureCountIndex))->value();
472 }
473 
474 
IrregexpNumberOfRegisters(FixedArray * re)475 int RegExpImpl::IrregexpNumberOfRegisters(FixedArray* re) {
476   return Smi::cast(re->get(JSRegExp::kIrregexpMaxRegisterCountIndex))->value();
477 }
478 
479 
IrregexpByteCode(FixedArray * re,bool is_one_byte)480 ByteArray* RegExpImpl::IrregexpByteCode(FixedArray* re, bool is_one_byte) {
481   return ByteArray::cast(re->get(JSRegExp::code_index(is_one_byte)));
482 }
483 
484 
IrregexpNativeCode(FixedArray * re,bool is_one_byte)485 Code* RegExpImpl::IrregexpNativeCode(FixedArray* re, bool is_one_byte) {
486   return Code::cast(re->get(JSRegExp::code_index(is_one_byte)));
487 }
488 
489 
IrregexpInitialize(Handle<JSRegExp> re,Handle<String> pattern,JSRegExp::Flags flags,int capture_count)490 void RegExpImpl::IrregexpInitialize(Handle<JSRegExp> re,
491                                     Handle<String> pattern,
492                                     JSRegExp::Flags flags,
493                                     int capture_count) {
494   // Initialize compiled code entries to null.
495   re->GetIsolate()->factory()->SetRegExpIrregexpData(re,
496                                                      JSRegExp::IRREGEXP,
497                                                      pattern,
498                                                      flags,
499                                                      capture_count);
500 }
501 
502 
IrregexpPrepare(Handle<JSRegExp> regexp,Handle<String> subject)503 int RegExpImpl::IrregexpPrepare(Handle<JSRegExp> regexp,
504                                 Handle<String> subject) {
505   subject = String::Flatten(subject);
506 
507   // Check representation of the underlying storage.
508   bool is_one_byte = subject->IsOneByteRepresentationUnderneath();
509   if (!EnsureCompiledIrregexp(regexp, subject, is_one_byte)) return -1;
510 
511 #ifdef V8_INTERPRETED_REGEXP
512   // Byte-code regexp needs space allocated for all its registers.
513   // The result captures are copied to the start of the registers array
514   // if the match succeeds.  This way those registers are not clobbered
515   // when we set the last match info from last successful match.
516   return IrregexpNumberOfRegisters(FixedArray::cast(regexp->data())) +
517          (IrregexpNumberOfCaptures(FixedArray::cast(regexp->data())) + 1) * 2;
518 #else  // V8_INTERPRETED_REGEXP
519   // Native regexp only needs room to output captures. Registers are handled
520   // internally.
521   return (IrregexpNumberOfCaptures(FixedArray::cast(regexp->data())) + 1) * 2;
522 #endif  // V8_INTERPRETED_REGEXP
523 }
524 
525 
IrregexpExecRaw(Handle<JSRegExp> regexp,Handle<String> subject,int index,int32_t * output,int output_size)526 int RegExpImpl::IrregexpExecRaw(Handle<JSRegExp> regexp,
527                                 Handle<String> subject,
528                                 int index,
529                                 int32_t* output,
530                                 int output_size) {
531   Isolate* isolate = regexp->GetIsolate();
532 
533   Handle<FixedArray> irregexp(FixedArray::cast(regexp->data()), isolate);
534 
535   DCHECK(index >= 0);
536   DCHECK(index <= subject->length());
537   DCHECK(subject->IsFlat());
538 
539   bool is_one_byte = subject->IsOneByteRepresentationUnderneath();
540 
541 #ifndef V8_INTERPRETED_REGEXP
542   DCHECK(output_size >= (IrregexpNumberOfCaptures(*irregexp) + 1) * 2);
543   do {
544     EnsureCompiledIrregexp(regexp, subject, is_one_byte);
545     Handle<Code> code(IrregexpNativeCode(*irregexp, is_one_byte), isolate);
546     // The stack is used to allocate registers for the compiled regexp code.
547     // This means that in case of failure, the output registers array is left
548     // untouched and contains the capture results from the previous successful
549     // match.  We can use that to set the last match info lazily.
550     NativeRegExpMacroAssembler::Result res =
551         NativeRegExpMacroAssembler::Match(code,
552                                           subject,
553                                           output,
554                                           output_size,
555                                           index,
556                                           isolate);
557     if (res != NativeRegExpMacroAssembler::RETRY) {
558       DCHECK(res != NativeRegExpMacroAssembler::EXCEPTION ||
559              isolate->has_pending_exception());
560       STATIC_ASSERT(
561           static_cast<int>(NativeRegExpMacroAssembler::SUCCESS) == RE_SUCCESS);
562       STATIC_ASSERT(
563           static_cast<int>(NativeRegExpMacroAssembler::FAILURE) == RE_FAILURE);
564       STATIC_ASSERT(static_cast<int>(NativeRegExpMacroAssembler::EXCEPTION)
565                     == RE_EXCEPTION);
566       return static_cast<IrregexpResult>(res);
567     }
568     // If result is RETRY, the string has changed representation, and we
569     // must restart from scratch.
570     // In this case, it means we must make sure we are prepared to handle
571     // the, potentially, different subject (the string can switch between
572     // being internal and external, and even between being Latin1 and UC16,
573     // but the characters are always the same).
574     IrregexpPrepare(regexp, subject);
575     is_one_byte = subject->IsOneByteRepresentationUnderneath();
576   } while (true);
577   UNREACHABLE();
578   return RE_EXCEPTION;
579 #else  // V8_INTERPRETED_REGEXP
580 
581   DCHECK(output_size >= IrregexpNumberOfRegisters(*irregexp));
582   // We must have done EnsureCompiledIrregexp, so we can get the number of
583   // registers.
584   int number_of_capture_registers =
585       (IrregexpNumberOfCaptures(*irregexp) + 1) * 2;
586   int32_t* raw_output = &output[number_of_capture_registers];
587   // We do not touch the actual capture result registers until we know there
588   // has been a match so that we can use those capture results to set the
589   // last match info.
590   for (int i = number_of_capture_registers - 1; i >= 0; i--) {
591     raw_output[i] = -1;
592   }
593   Handle<ByteArray> byte_codes(IrregexpByteCode(*irregexp, is_one_byte),
594                                isolate);
595 
596   IrregexpResult result = IrregexpInterpreter::Match(isolate,
597                                                      byte_codes,
598                                                      subject,
599                                                      raw_output,
600                                                      index);
601   if (result == RE_SUCCESS) {
602     // Copy capture results to the start of the registers array.
603     MemCopy(output, raw_output, number_of_capture_registers * sizeof(int32_t));
604   }
605   if (result == RE_EXCEPTION) {
606     DCHECK(!isolate->has_pending_exception());
607     isolate->StackOverflow();
608   }
609   return result;
610 #endif  // V8_INTERPRETED_REGEXP
611 }
612 
613 
IrregexpExec(Handle<JSRegExp> regexp,Handle<String> subject,int previous_index,Handle<JSArray> last_match_info)614 MaybeHandle<Object> RegExpImpl::IrregexpExec(Handle<JSRegExp> regexp,
615                                              Handle<String> subject,
616                                              int previous_index,
617                                              Handle<JSArray> last_match_info) {
618   Isolate* isolate = regexp->GetIsolate();
619   DCHECK_EQ(regexp->TypeTag(), JSRegExp::IRREGEXP);
620 
621   // Prepare space for the return values.
622 #if defined(V8_INTERPRETED_REGEXP) && defined(DEBUG)
623   if (FLAG_trace_regexp_bytecodes) {
624     String* pattern = regexp->Pattern();
625     PrintF("\n\nRegexp match:   /%s/\n\n", pattern->ToCString().get());
626     PrintF("\n\nSubject string: '%s'\n\n", subject->ToCString().get());
627   }
628 #endif
629   int required_registers = RegExpImpl::IrregexpPrepare(regexp, subject);
630   if (required_registers < 0) {
631     // Compiling failed with an exception.
632     DCHECK(isolate->has_pending_exception());
633     return MaybeHandle<Object>();
634   }
635 
636   int32_t* output_registers = NULL;
637   if (required_registers > Isolate::kJSRegexpStaticOffsetsVectorSize) {
638     output_registers = NewArray<int32_t>(required_registers);
639   }
640   SmartArrayPointer<int32_t> auto_release(output_registers);
641   if (output_registers == NULL) {
642     output_registers = isolate->jsregexp_static_offsets_vector();
643   }
644 
645   int res = RegExpImpl::IrregexpExecRaw(
646       regexp, subject, previous_index, output_registers, required_registers);
647   if (res == RE_SUCCESS) {
648     int capture_count =
649         IrregexpNumberOfCaptures(FixedArray::cast(regexp->data()));
650     return SetLastMatchInfo(
651         last_match_info, subject, capture_count, output_registers);
652   }
653   if (res == RE_EXCEPTION) {
654     DCHECK(isolate->has_pending_exception());
655     return MaybeHandle<Object>();
656   }
657   DCHECK(res == RE_FAILURE);
658   return isolate->factory()->null_value();
659 }
660 
661 
SetLastMatchInfo(Handle<JSArray> last_match_info,Handle<String> subject,int capture_count,int32_t * match)662 Handle<JSArray> RegExpImpl::SetLastMatchInfo(Handle<JSArray> last_match_info,
663                                              Handle<String> subject,
664                                              int capture_count,
665                                              int32_t* match) {
666   DCHECK(last_match_info->HasFastObjectElements());
667   int capture_register_count = (capture_count + 1) * 2;
668   JSArray::EnsureSize(last_match_info,
669                       capture_register_count + kLastMatchOverhead);
670   DisallowHeapAllocation no_allocation;
671   FixedArray* array = FixedArray::cast(last_match_info->elements());
672   if (match != NULL) {
673     for (int i = 0; i < capture_register_count; i += 2) {
674       SetCapture(array, i, match[i]);
675       SetCapture(array, i + 1, match[i + 1]);
676     }
677   }
678   SetLastCaptureCount(array, capture_register_count);
679   SetLastSubject(array, *subject);
680   SetLastInput(array, *subject);
681   return last_match_info;
682 }
683 
684 
GlobalCache(Handle<JSRegExp> regexp,Handle<String> subject,bool is_global,Isolate * isolate)685 RegExpImpl::GlobalCache::GlobalCache(Handle<JSRegExp> regexp,
686                                      Handle<String> subject,
687                                      bool is_global,
688                                      Isolate* isolate)
689   : register_array_(NULL),
690     register_array_size_(0),
691     regexp_(regexp),
692     subject_(subject) {
693 #ifdef V8_INTERPRETED_REGEXP
694   bool interpreted = true;
695 #else
696   bool interpreted = false;
697 #endif  // V8_INTERPRETED_REGEXP
698 
699   if (regexp_->TypeTag() == JSRegExp::ATOM) {
700     static const int kAtomRegistersPerMatch = 2;
701     registers_per_match_ = kAtomRegistersPerMatch;
702     // There is no distinction between interpreted and native for atom regexps.
703     interpreted = false;
704   } else {
705     registers_per_match_ = RegExpImpl::IrregexpPrepare(regexp_, subject_);
706     if (registers_per_match_ < 0) {
707       num_matches_ = -1;  // Signal exception.
708       return;
709     }
710   }
711 
712   if (is_global && !interpreted) {
713     register_array_size_ =
714         Max(registers_per_match_, Isolate::kJSRegexpStaticOffsetsVectorSize);
715     max_matches_ = register_array_size_ / registers_per_match_;
716   } else {
717     // Global loop in interpreted regexp is not implemented.  We choose
718     // the size of the offsets vector so that it can only store one match.
719     register_array_size_ = registers_per_match_;
720     max_matches_ = 1;
721   }
722 
723   if (register_array_size_ > Isolate::kJSRegexpStaticOffsetsVectorSize) {
724     register_array_ = NewArray<int32_t>(register_array_size_);
725   } else {
726     register_array_ = isolate->jsregexp_static_offsets_vector();
727   }
728 
729   // Set state so that fetching the results the first time triggers a call
730   // to the compiled regexp.
731   current_match_index_ = max_matches_ - 1;
732   num_matches_ = max_matches_;
733   DCHECK(registers_per_match_ >= 2);  // Each match has at least one capture.
734   DCHECK_GE(register_array_size_, registers_per_match_);
735   int32_t* last_match =
736       &register_array_[current_match_index_ * registers_per_match_];
737   last_match[0] = -1;
738   last_match[1] = 0;
739 }
740 
741 
742 // -------------------------------------------------------------------
743 // Implementation of the Irregexp regular expression engine.
744 //
745 // The Irregexp regular expression engine is intended to be a complete
746 // implementation of ECMAScript regular expressions.  It generates either
747 // bytecodes or native code.
748 
749 //   The Irregexp regexp engine is structured in three steps.
750 //   1) The parser generates an abstract syntax tree.  See ast.cc.
751 //   2) From the AST a node network is created.  The nodes are all
752 //      subclasses of RegExpNode.  The nodes represent states when
753 //      executing a regular expression.  Several optimizations are
754 //      performed on the node network.
755 //   3) From the nodes we generate either byte codes or native code
756 //      that can actually execute the regular expression (perform
757 //      the search).  The code generation step is described in more
758 //      detail below.
759 
760 // Code generation.
761 //
762 //   The nodes are divided into four main categories.
763 //   * Choice nodes
764 //        These represent places where the regular expression can
765 //        match in more than one way.  For example on entry to an
766 //        alternation (foo|bar) or a repetition (*, +, ? or {}).
767 //   * Action nodes
768 //        These represent places where some action should be
769 //        performed.  Examples include recording the current position
770 //        in the input string to a register (in order to implement
771 //        captures) or other actions on register for example in order
772 //        to implement the counters needed for {} repetitions.
773 //   * Matching nodes
774 //        These attempt to match some element part of the input string.
775 //        Examples of elements include character classes, plain strings
776 //        or back references.
777 //   * End nodes
778 //        These are used to implement the actions required on finding
779 //        a successful match or failing to find a match.
780 //
781 //   The code generated (whether as byte codes or native code) maintains
782 //   some state as it runs.  This consists of the following elements:
783 //
784 //   * The capture registers.  Used for string captures.
785 //   * Other registers.  Used for counters etc.
786 //   * The current position.
787 //   * The stack of backtracking information.  Used when a matching node
788 //     fails to find a match and needs to try an alternative.
789 //
790 // Conceptual regular expression execution model:
791 //
792 //   There is a simple conceptual model of regular expression execution
793 //   which will be presented first.  The actual code generated is a more
794 //   efficient simulation of the simple conceptual model:
795 //
796 //   * Choice nodes are implemented as follows:
797 //     For each choice except the last {
798 //       push current position
799 //       push backtrack code location
800 //       <generate code to test for choice>
801 //       backtrack code location:
802 //       pop current position
803 //     }
804 //     <generate code to test for last choice>
805 //
806 //   * Actions nodes are generated as follows
807 //     <push affected registers on backtrack stack>
808 //     <generate code to perform action>
809 //     push backtrack code location
810 //     <generate code to test for following nodes>
811 //     backtrack code location:
812 //     <pop affected registers to restore their state>
813 //     <pop backtrack location from stack and go to it>
814 //
815 //   * Matching nodes are generated as follows:
816 //     if input string matches at current position
817 //       update current position
818 //       <generate code to test for following nodes>
819 //     else
820 //       <pop backtrack location from stack and go to it>
821 //
822 //   Thus it can be seen that the current position is saved and restored
823 //   by the choice nodes, whereas the registers are saved and restored by
824 //   by the action nodes that manipulate them.
825 //
826 //   The other interesting aspect of this model is that nodes are generated
827 //   at the point where they are needed by a recursive call to Emit().  If
828 //   the node has already been code generated then the Emit() call will
829 //   generate a jump to the previously generated code instead.  In order to
830 //   limit recursion it is possible for the Emit() function to put the node
831 //   on a work list for later generation and instead generate a jump.  The
832 //   destination of the jump is resolved later when the code is generated.
833 //
834 // Actual regular expression code generation.
835 //
836 //   Code generation is actually more complicated than the above.  In order
837 //   to improve the efficiency of the generated code some optimizations are
838 //   performed
839 //
840 //   * Choice nodes have 1-character lookahead.
841 //     A choice node looks at the following character and eliminates some of
842 //     the choices immediately based on that character.  This is not yet
843 //     implemented.
844 //   * Simple greedy loops store reduced backtracking information.
845 //     A quantifier like /.*foo/m will greedily match the whole input.  It will
846 //     then need to backtrack to a point where it can match "foo".  The naive
847 //     implementation of this would push each character position onto the
848 //     backtracking stack, then pop them off one by one.  This would use space
849 //     proportional to the length of the input string.  However since the "."
850 //     can only match in one way and always has a constant length (in this case
851 //     of 1) it suffices to store the current position on the top of the stack
852 //     once.  Matching now becomes merely incrementing the current position and
853 //     backtracking becomes decrementing the current position and checking the
854 //     result against the stored current position.  This is faster and saves
855 //     space.
856 //   * The current state is virtualized.
857 //     This is used to defer expensive operations until it is clear that they
858 //     are needed and to generate code for a node more than once, allowing
859 //     specialized an efficient versions of the code to be created. This is
860 //     explained in the section below.
861 //
862 // Execution state virtualization.
863 //
864 //   Instead of emitting code, nodes that manipulate the state can record their
865 //   manipulation in an object called the Trace.  The Trace object can record a
866 //   current position offset, an optional backtrack code location on the top of
867 //   the virtualized backtrack stack and some register changes.  When a node is
868 //   to be emitted it can flush the Trace or update it.  Flushing the Trace
869 //   will emit code to bring the actual state into line with the virtual state.
870 //   Avoiding flushing the state can postpone some work (e.g. updates of capture
871 //   registers).  Postponing work can save time when executing the regular
872 //   expression since it may be found that the work never has to be done as a
873 //   failure to match can occur.  In addition it is much faster to jump to a
874 //   known backtrack code location than it is to pop an unknown backtrack
875 //   location from the stack and jump there.
876 //
877 //   The virtual state found in the Trace affects code generation.  For example
878 //   the virtual state contains the difference between the actual current
879 //   position and the virtual current position, and matching code needs to use
880 //   this offset to attempt a match in the correct location of the input
881 //   string.  Therefore code generated for a non-trivial trace is specialized
882 //   to that trace.  The code generator therefore has the ability to generate
883 //   code for each node several times.  In order to limit the size of the
884 //   generated code there is an arbitrary limit on how many specialized sets of
885 //   code may be generated for a given node.  If the limit is reached, the
886 //   trace is flushed and a generic version of the code for a node is emitted.
887 //   This is subsequently used for that node.  The code emitted for non-generic
888 //   trace is not recorded in the node and so it cannot currently be reused in
889 //   the event that code generation is requested for an identical trace.
890 
891 
AppendToText(RegExpText * text,Zone * zone)892 void RegExpTree::AppendToText(RegExpText* text, Zone* zone) {
893   UNREACHABLE();
894 }
895 
896 
AppendToText(RegExpText * text,Zone * zone)897 void RegExpAtom::AppendToText(RegExpText* text, Zone* zone) {
898   text->AddElement(TextElement::Atom(this), zone);
899 }
900 
901 
AppendToText(RegExpText * text,Zone * zone)902 void RegExpCharacterClass::AppendToText(RegExpText* text, Zone* zone) {
903   text->AddElement(TextElement::CharClass(this), zone);
904 }
905 
906 
AppendToText(RegExpText * text,Zone * zone)907 void RegExpText::AppendToText(RegExpText* text, Zone* zone) {
908   for (int i = 0; i < elements()->length(); i++)
909     text->AddElement(elements()->at(i), zone);
910 }
911 
912 
Atom(RegExpAtom * atom)913 TextElement TextElement::Atom(RegExpAtom* atom) {
914   return TextElement(ATOM, atom);
915 }
916 
917 
CharClass(RegExpCharacterClass * char_class)918 TextElement TextElement::CharClass(RegExpCharacterClass* char_class) {
919   return TextElement(CHAR_CLASS, char_class);
920 }
921 
922 
length() const923 int TextElement::length() const {
924   switch (text_type()) {
925     case ATOM:
926       return atom()->length();
927 
928     case CHAR_CLASS:
929       return 1;
930   }
931   UNREACHABLE();
932   return 0;
933 }
934 
935 
GetTable(bool ignore_case)936 DispatchTable* ChoiceNode::GetTable(bool ignore_case) {
937   if (table_ == NULL) {
938     table_ = new(zone()) DispatchTable(zone());
939     DispatchTableConstructor cons(table_, ignore_case, zone());
940     cons.BuildTable(this);
941   }
942   return table_;
943 }
944 
945 
946 class FrequencyCollator {
947  public:
FrequencyCollator()948   FrequencyCollator() : total_samples_(0) {
949     for (int i = 0; i < RegExpMacroAssembler::kTableSize; i++) {
950       frequencies_[i] = CharacterFrequency(i);
951     }
952   }
953 
CountCharacter(int character)954   void CountCharacter(int character) {
955     int index = (character & RegExpMacroAssembler::kTableMask);
956     frequencies_[index].Increment();
957     total_samples_++;
958   }
959 
960   // Does not measure in percent, but rather per-128 (the table size from the
961   // regexp macro assembler).
Frequency(int in_character)962   int Frequency(int in_character) {
963     DCHECK((in_character & RegExpMacroAssembler::kTableMask) == in_character);
964     if (total_samples_ < 1) return 1;  // Division by zero.
965     int freq_in_per128 =
966         (frequencies_[in_character].counter() * 128) / total_samples_;
967     return freq_in_per128;
968   }
969 
970  private:
971   class CharacterFrequency {
972    public:
CharacterFrequency()973     CharacterFrequency() : counter_(0), character_(-1) { }
CharacterFrequency(int character)974     explicit CharacterFrequency(int character)
975         : counter_(0), character_(character) { }
976 
Increment()977     void Increment() { counter_++; }
counter()978     int counter() { return counter_; }
character()979     int character() { return character_; }
980 
981    private:
982     int counter_;
983     int character_;
984   };
985 
986 
987  private:
988   CharacterFrequency frequencies_[RegExpMacroAssembler::kTableSize];
989   int total_samples_;
990 };
991 
992 
993 class RegExpCompiler {
994  public:
995   RegExpCompiler(int capture_count, bool ignore_case, bool is_one_byte,
996                  Zone* zone);
997 
AllocateRegister()998   int AllocateRegister() {
999     if (next_register_ >= RegExpMacroAssembler::kMaxRegister) {
1000       reg_exp_too_big_ = true;
1001       return next_register_;
1002     }
1003     return next_register_++;
1004   }
1005 
1006   RegExpEngine::CompilationResult Assemble(RegExpMacroAssembler* assembler,
1007                                            RegExpNode* start,
1008                                            int capture_count,
1009                                            Handle<String> pattern);
1010 
AddWork(RegExpNode * node)1011   inline void AddWork(RegExpNode* node) { work_list_->Add(node); }
1012 
1013   static const int kImplementationOffset = 0;
1014   static const int kNumberOfRegistersOffset = 0;
1015   static const int kCodeOffset = 1;
1016 
macro_assembler()1017   RegExpMacroAssembler* macro_assembler() { return macro_assembler_; }
accept()1018   EndNode* accept() { return accept_; }
1019 
1020   static const int kMaxRecursion = 100;
recursion_depth()1021   inline int recursion_depth() { return recursion_depth_; }
IncrementRecursionDepth()1022   inline void IncrementRecursionDepth() { recursion_depth_++; }
DecrementRecursionDepth()1023   inline void DecrementRecursionDepth() { recursion_depth_--; }
1024 
SetRegExpTooBig()1025   void SetRegExpTooBig() { reg_exp_too_big_ = true; }
1026 
ignore_case()1027   inline bool ignore_case() { return ignore_case_; }
one_byte()1028   inline bool one_byte() { return one_byte_; }
frequency_collator()1029   FrequencyCollator* frequency_collator() { return &frequency_collator_; }
1030 
current_expansion_factor()1031   int current_expansion_factor() { return current_expansion_factor_; }
set_current_expansion_factor(int value)1032   void set_current_expansion_factor(int value) {
1033     current_expansion_factor_ = value;
1034   }
1035 
zone() const1036   Zone* zone() const { return zone_; }
1037 
1038   static const int kNoRegister = -1;
1039 
1040  private:
1041   EndNode* accept_;
1042   int next_register_;
1043   List<RegExpNode*>* work_list_;
1044   int recursion_depth_;
1045   RegExpMacroAssembler* macro_assembler_;
1046   bool ignore_case_;
1047   bool one_byte_;
1048   bool reg_exp_too_big_;
1049   int current_expansion_factor_;
1050   FrequencyCollator frequency_collator_;
1051   Zone* zone_;
1052 };
1053 
1054 
1055 class RecursionCheck {
1056  public:
RecursionCheck(RegExpCompiler * compiler)1057   explicit RecursionCheck(RegExpCompiler* compiler) : compiler_(compiler) {
1058     compiler->IncrementRecursionDepth();
1059   }
~RecursionCheck()1060   ~RecursionCheck() { compiler_->DecrementRecursionDepth(); }
1061  private:
1062   RegExpCompiler* compiler_;
1063 };
1064 
1065 
IrregexpRegExpTooBig(Isolate * isolate)1066 static RegExpEngine::CompilationResult IrregexpRegExpTooBig(Isolate* isolate) {
1067   return RegExpEngine::CompilationResult(isolate, "RegExp too big");
1068 }
1069 
1070 
1071 // Attempts to compile the regexp using an Irregexp code generator.  Returns
1072 // a fixed array or a null handle depending on whether it succeeded.
RegExpCompiler(int capture_count,bool ignore_case,bool one_byte,Zone * zone)1073 RegExpCompiler::RegExpCompiler(int capture_count, bool ignore_case,
1074                                bool one_byte, Zone* zone)
1075     : next_register_(2 * (capture_count + 1)),
1076       work_list_(NULL),
1077       recursion_depth_(0),
1078       ignore_case_(ignore_case),
1079       one_byte_(one_byte),
1080       reg_exp_too_big_(false),
1081       current_expansion_factor_(1),
1082       frequency_collator_(),
1083       zone_(zone) {
1084   accept_ = new(zone) EndNode(EndNode::ACCEPT, zone);
1085   DCHECK(next_register_ - 1 <= RegExpMacroAssembler::kMaxRegister);
1086 }
1087 
1088 
Assemble(RegExpMacroAssembler * macro_assembler,RegExpNode * start,int capture_count,Handle<String> pattern)1089 RegExpEngine::CompilationResult RegExpCompiler::Assemble(
1090     RegExpMacroAssembler* macro_assembler,
1091     RegExpNode* start,
1092     int capture_count,
1093     Handle<String> pattern) {
1094   Heap* heap = pattern->GetHeap();
1095 
1096   bool use_slow_safe_regexp_compiler = false;
1097   if (heap->total_regexp_code_generated() >
1098           RegExpImpl::kRegWxpCompiledLimit &&
1099       heap->isolate()->memory_allocator()->SizeExecutable() >
1100           RegExpImpl::kRegExpExecutableMemoryLimit) {
1101     use_slow_safe_regexp_compiler = true;
1102   }
1103 
1104   macro_assembler->set_slow_safe(use_slow_safe_regexp_compiler);
1105 
1106 #ifdef DEBUG
1107   if (FLAG_trace_regexp_assembler)
1108     macro_assembler_ = new RegExpMacroAssemblerTracer(macro_assembler);
1109   else
1110 #endif
1111     macro_assembler_ = macro_assembler;
1112 
1113   List <RegExpNode*> work_list(0);
1114   work_list_ = &work_list;
1115   Label fail;
1116   macro_assembler_->PushBacktrack(&fail);
1117   Trace new_trace;
1118   start->Emit(this, &new_trace);
1119   macro_assembler_->Bind(&fail);
1120   macro_assembler_->Fail();
1121   while (!work_list.is_empty()) {
1122     work_list.RemoveLast()->Emit(this, &new_trace);
1123   }
1124   if (reg_exp_too_big_) return IrregexpRegExpTooBig(zone_->isolate());
1125 
1126   Handle<HeapObject> code = macro_assembler_->GetCode(pattern);
1127   heap->IncreaseTotalRegexpCodeGenerated(code->Size());
1128   work_list_ = NULL;
1129 #ifdef DEBUG
1130   if (FLAG_print_code) {
1131     CodeTracer::Scope trace_scope(heap->isolate()->GetCodeTracer());
1132     OFStream os(trace_scope.file());
1133     Handle<Code>::cast(code)->Disassemble(pattern->ToCString().get(), os);
1134   }
1135   if (FLAG_trace_regexp_assembler) {
1136     delete macro_assembler_;
1137   }
1138 #endif
1139   return RegExpEngine::CompilationResult(*code, next_register_);
1140 }
1141 
1142 
Mentions(int that)1143 bool Trace::DeferredAction::Mentions(int that) {
1144   if (action_type() == ActionNode::CLEAR_CAPTURES) {
1145     Interval range = static_cast<DeferredClearCaptures*>(this)->range();
1146     return range.Contains(that);
1147   } else {
1148     return reg() == that;
1149   }
1150 }
1151 
1152 
mentions_reg(int reg)1153 bool Trace::mentions_reg(int reg) {
1154   for (DeferredAction* action = actions_;
1155        action != NULL;
1156        action = action->next()) {
1157     if (action->Mentions(reg))
1158       return true;
1159   }
1160   return false;
1161 }
1162 
1163 
GetStoredPosition(int reg,int * cp_offset)1164 bool Trace::GetStoredPosition(int reg, int* cp_offset) {
1165   DCHECK_EQ(0, *cp_offset);
1166   for (DeferredAction* action = actions_;
1167        action != NULL;
1168        action = action->next()) {
1169     if (action->Mentions(reg)) {
1170       if (action->action_type() == ActionNode::STORE_POSITION) {
1171         *cp_offset = static_cast<DeferredCapture*>(action)->cp_offset();
1172         return true;
1173       } else {
1174         return false;
1175       }
1176     }
1177   }
1178   return false;
1179 }
1180 
1181 
FindAffectedRegisters(OutSet * affected_registers,Zone * zone)1182 int Trace::FindAffectedRegisters(OutSet* affected_registers,
1183                                  Zone* zone) {
1184   int max_register = RegExpCompiler::kNoRegister;
1185   for (DeferredAction* action = actions_;
1186        action != NULL;
1187        action = action->next()) {
1188     if (action->action_type() == ActionNode::CLEAR_CAPTURES) {
1189       Interval range = static_cast<DeferredClearCaptures*>(action)->range();
1190       for (int i = range.from(); i <= range.to(); i++)
1191         affected_registers->Set(i, zone);
1192       if (range.to() > max_register) max_register = range.to();
1193     } else {
1194       affected_registers->Set(action->reg(), zone);
1195       if (action->reg() > max_register) max_register = action->reg();
1196     }
1197   }
1198   return max_register;
1199 }
1200 
1201 
RestoreAffectedRegisters(RegExpMacroAssembler * assembler,int max_register,const OutSet & registers_to_pop,const OutSet & registers_to_clear)1202 void Trace::RestoreAffectedRegisters(RegExpMacroAssembler* assembler,
1203                                      int max_register,
1204                                      const OutSet& registers_to_pop,
1205                                      const OutSet& registers_to_clear) {
1206   for (int reg = max_register; reg >= 0; reg--) {
1207     if (registers_to_pop.Get(reg)) {
1208       assembler->PopRegister(reg);
1209     } else if (registers_to_clear.Get(reg)) {
1210       int clear_to = reg;
1211       while (reg > 0 && registers_to_clear.Get(reg - 1)) {
1212         reg--;
1213       }
1214       assembler->ClearRegisters(reg, clear_to);
1215     }
1216   }
1217 }
1218 
1219 
PerformDeferredActions(RegExpMacroAssembler * assembler,int max_register,const OutSet & affected_registers,OutSet * registers_to_pop,OutSet * registers_to_clear,Zone * zone)1220 void Trace::PerformDeferredActions(RegExpMacroAssembler* assembler,
1221                                    int max_register,
1222                                    const OutSet& affected_registers,
1223                                    OutSet* registers_to_pop,
1224                                    OutSet* registers_to_clear,
1225                                    Zone* zone) {
1226   // The "+1" is to avoid a push_limit of zero if stack_limit_slack() is 1.
1227   const int push_limit = (assembler->stack_limit_slack() + 1) / 2;
1228 
1229   // Count pushes performed to force a stack limit check occasionally.
1230   int pushes = 0;
1231 
1232   for (int reg = 0; reg <= max_register; reg++) {
1233     if (!affected_registers.Get(reg)) {
1234       continue;
1235     }
1236 
1237     // The chronologically first deferred action in the trace
1238     // is used to infer the action needed to restore a register
1239     // to its previous state (or not, if it's safe to ignore it).
1240     enum DeferredActionUndoType { IGNORE, RESTORE, CLEAR };
1241     DeferredActionUndoType undo_action = IGNORE;
1242 
1243     int value = 0;
1244     bool absolute = false;
1245     bool clear = false;
1246     int store_position = -1;
1247     // This is a little tricky because we are scanning the actions in reverse
1248     // historical order (newest first).
1249     for (DeferredAction* action = actions_;
1250          action != NULL;
1251          action = action->next()) {
1252       if (action->Mentions(reg)) {
1253         switch (action->action_type()) {
1254           case ActionNode::SET_REGISTER: {
1255             Trace::DeferredSetRegister* psr =
1256                 static_cast<Trace::DeferredSetRegister*>(action);
1257             if (!absolute) {
1258               value += psr->value();
1259               absolute = true;
1260             }
1261             // SET_REGISTER is currently only used for newly introduced loop
1262             // counters. They can have a significant previous value if they
1263             // occour in a loop. TODO(lrn): Propagate this information, so
1264             // we can set undo_action to IGNORE if we know there is no value to
1265             // restore.
1266             undo_action = RESTORE;
1267             DCHECK_EQ(store_position, -1);
1268             DCHECK(!clear);
1269             break;
1270           }
1271           case ActionNode::INCREMENT_REGISTER:
1272             if (!absolute) {
1273               value++;
1274             }
1275             DCHECK_EQ(store_position, -1);
1276             DCHECK(!clear);
1277             undo_action = RESTORE;
1278             break;
1279           case ActionNode::STORE_POSITION: {
1280             Trace::DeferredCapture* pc =
1281                 static_cast<Trace::DeferredCapture*>(action);
1282             if (!clear && store_position == -1) {
1283               store_position = pc->cp_offset();
1284             }
1285 
1286             // For captures we know that stores and clears alternate.
1287             // Other register, are never cleared, and if the occur
1288             // inside a loop, they might be assigned more than once.
1289             if (reg <= 1) {
1290               // Registers zero and one, aka "capture zero", is
1291               // always set correctly if we succeed. There is no
1292               // need to undo a setting on backtrack, because we
1293               // will set it again or fail.
1294               undo_action = IGNORE;
1295             } else {
1296               undo_action = pc->is_capture() ? CLEAR : RESTORE;
1297             }
1298             DCHECK(!absolute);
1299             DCHECK_EQ(value, 0);
1300             break;
1301           }
1302           case ActionNode::CLEAR_CAPTURES: {
1303             // Since we're scanning in reverse order, if we've already
1304             // set the position we have to ignore historically earlier
1305             // clearing operations.
1306             if (store_position == -1) {
1307               clear = true;
1308             }
1309             undo_action = RESTORE;
1310             DCHECK(!absolute);
1311             DCHECK_EQ(value, 0);
1312             break;
1313           }
1314           default:
1315             UNREACHABLE();
1316             break;
1317         }
1318       }
1319     }
1320     // Prepare for the undo-action (e.g., push if it's going to be popped).
1321     if (undo_action == RESTORE) {
1322       pushes++;
1323       RegExpMacroAssembler::StackCheckFlag stack_check =
1324           RegExpMacroAssembler::kNoStackLimitCheck;
1325       if (pushes == push_limit) {
1326         stack_check = RegExpMacroAssembler::kCheckStackLimit;
1327         pushes = 0;
1328       }
1329 
1330       assembler->PushRegister(reg, stack_check);
1331       registers_to_pop->Set(reg, zone);
1332     } else if (undo_action == CLEAR) {
1333       registers_to_clear->Set(reg, zone);
1334     }
1335     // Perform the chronologically last action (or accumulated increment)
1336     // for the register.
1337     if (store_position != -1) {
1338       assembler->WriteCurrentPositionToRegister(reg, store_position);
1339     } else if (clear) {
1340       assembler->ClearRegisters(reg, reg);
1341     } else if (absolute) {
1342       assembler->SetRegister(reg, value);
1343     } else if (value != 0) {
1344       assembler->AdvanceRegister(reg, value);
1345     }
1346   }
1347 }
1348 
1349 
1350 // This is called as we come into a loop choice node and some other tricky
1351 // nodes.  It normalizes the state of the code generator to ensure we can
1352 // generate generic code.
Flush(RegExpCompiler * compiler,RegExpNode * successor)1353 void Trace::Flush(RegExpCompiler* compiler, RegExpNode* successor) {
1354   RegExpMacroAssembler* assembler = compiler->macro_assembler();
1355 
1356   DCHECK(!is_trivial());
1357 
1358   if (actions_ == NULL && backtrack() == NULL) {
1359     // Here we just have some deferred cp advances to fix and we are back to
1360     // a normal situation.  We may also have to forget some information gained
1361     // through a quick check that was already performed.
1362     if (cp_offset_ != 0) assembler->AdvanceCurrentPosition(cp_offset_);
1363     // Create a new trivial state and generate the node with that.
1364     Trace new_state;
1365     successor->Emit(compiler, &new_state);
1366     return;
1367   }
1368 
1369   // Generate deferred actions here along with code to undo them again.
1370   OutSet affected_registers;
1371 
1372   if (backtrack() != NULL) {
1373     // Here we have a concrete backtrack location.  These are set up by choice
1374     // nodes and so they indicate that we have a deferred save of the current
1375     // position which we may need to emit here.
1376     assembler->PushCurrentPosition();
1377   }
1378 
1379   int max_register = FindAffectedRegisters(&affected_registers,
1380                                            compiler->zone());
1381   OutSet registers_to_pop;
1382   OutSet registers_to_clear;
1383   PerformDeferredActions(assembler,
1384                          max_register,
1385                          affected_registers,
1386                          &registers_to_pop,
1387                          &registers_to_clear,
1388                          compiler->zone());
1389   if (cp_offset_ != 0) {
1390     assembler->AdvanceCurrentPosition(cp_offset_);
1391   }
1392 
1393   // Create a new trivial state and generate the node with that.
1394   Label undo;
1395   assembler->PushBacktrack(&undo);
1396   Trace new_state;
1397   successor->Emit(compiler, &new_state);
1398 
1399   // On backtrack we need to restore state.
1400   assembler->Bind(&undo);
1401   RestoreAffectedRegisters(assembler,
1402                            max_register,
1403                            registers_to_pop,
1404                            registers_to_clear);
1405   if (backtrack() == NULL) {
1406     assembler->Backtrack();
1407   } else {
1408     assembler->PopCurrentPosition();
1409     assembler->GoTo(backtrack());
1410   }
1411 }
1412 
1413 
Emit(RegExpCompiler * compiler,Trace * trace)1414 void NegativeSubmatchSuccess::Emit(RegExpCompiler* compiler, Trace* trace) {
1415   RegExpMacroAssembler* assembler = compiler->macro_assembler();
1416 
1417   // Omit flushing the trace. We discard the entire stack frame anyway.
1418 
1419   if (!label()->is_bound()) {
1420     // We are completely independent of the trace, since we ignore it,
1421     // so this code can be used as the generic version.
1422     assembler->Bind(label());
1423   }
1424 
1425   // Throw away everything on the backtrack stack since the start
1426   // of the negative submatch and restore the character position.
1427   assembler->ReadCurrentPositionFromRegister(current_position_register_);
1428   assembler->ReadStackPointerFromRegister(stack_pointer_register_);
1429   if (clear_capture_count_ > 0) {
1430     // Clear any captures that might have been performed during the success
1431     // of the body of the negative look-ahead.
1432     int clear_capture_end = clear_capture_start_ + clear_capture_count_ - 1;
1433     assembler->ClearRegisters(clear_capture_start_, clear_capture_end);
1434   }
1435   // Now that we have unwound the stack we find at the top of the stack the
1436   // backtrack that the BeginSubmatch node got.
1437   assembler->Backtrack();
1438 }
1439 
1440 
Emit(RegExpCompiler * compiler,Trace * trace)1441 void EndNode::Emit(RegExpCompiler* compiler, Trace* trace) {
1442   if (!trace->is_trivial()) {
1443     trace->Flush(compiler, this);
1444     return;
1445   }
1446   RegExpMacroAssembler* assembler = compiler->macro_assembler();
1447   if (!label()->is_bound()) {
1448     assembler->Bind(label());
1449   }
1450   switch (action_) {
1451     case ACCEPT:
1452       assembler->Succeed();
1453       return;
1454     case BACKTRACK:
1455       assembler->GoTo(trace->backtrack());
1456       return;
1457     case NEGATIVE_SUBMATCH_SUCCESS:
1458       // This case is handled in a different virtual method.
1459       UNREACHABLE();
1460   }
1461   UNIMPLEMENTED();
1462 }
1463 
1464 
AddGuard(Guard * guard,Zone * zone)1465 void GuardedAlternative::AddGuard(Guard* guard, Zone* zone) {
1466   if (guards_ == NULL)
1467     guards_ = new(zone) ZoneList<Guard*>(1, zone);
1468   guards_->Add(guard, zone);
1469 }
1470 
1471 
SetRegister(int reg,int val,RegExpNode * on_success)1472 ActionNode* ActionNode::SetRegister(int reg,
1473                                     int val,
1474                                     RegExpNode* on_success) {
1475   ActionNode* result =
1476       new(on_success->zone()) ActionNode(SET_REGISTER, on_success);
1477   result->data_.u_store_register.reg = reg;
1478   result->data_.u_store_register.value = val;
1479   return result;
1480 }
1481 
1482 
IncrementRegister(int reg,RegExpNode * on_success)1483 ActionNode* ActionNode::IncrementRegister(int reg, RegExpNode* on_success) {
1484   ActionNode* result =
1485       new(on_success->zone()) ActionNode(INCREMENT_REGISTER, on_success);
1486   result->data_.u_increment_register.reg = reg;
1487   return result;
1488 }
1489 
1490 
StorePosition(int reg,bool is_capture,RegExpNode * on_success)1491 ActionNode* ActionNode::StorePosition(int reg,
1492                                       bool is_capture,
1493                                       RegExpNode* on_success) {
1494   ActionNode* result =
1495       new(on_success->zone()) ActionNode(STORE_POSITION, on_success);
1496   result->data_.u_position_register.reg = reg;
1497   result->data_.u_position_register.is_capture = is_capture;
1498   return result;
1499 }
1500 
1501 
ClearCaptures(Interval range,RegExpNode * on_success)1502 ActionNode* ActionNode::ClearCaptures(Interval range,
1503                                       RegExpNode* on_success) {
1504   ActionNode* result =
1505       new(on_success->zone()) ActionNode(CLEAR_CAPTURES, on_success);
1506   result->data_.u_clear_captures.range_from = range.from();
1507   result->data_.u_clear_captures.range_to = range.to();
1508   return result;
1509 }
1510 
1511 
BeginSubmatch(int stack_reg,int position_reg,RegExpNode * on_success)1512 ActionNode* ActionNode::BeginSubmatch(int stack_reg,
1513                                       int position_reg,
1514                                       RegExpNode* on_success) {
1515   ActionNode* result =
1516       new(on_success->zone()) ActionNode(BEGIN_SUBMATCH, on_success);
1517   result->data_.u_submatch.stack_pointer_register = stack_reg;
1518   result->data_.u_submatch.current_position_register = position_reg;
1519   return result;
1520 }
1521 
1522 
PositiveSubmatchSuccess(int stack_reg,int position_reg,int clear_register_count,int clear_register_from,RegExpNode * on_success)1523 ActionNode* ActionNode::PositiveSubmatchSuccess(int stack_reg,
1524                                                 int position_reg,
1525                                                 int clear_register_count,
1526                                                 int clear_register_from,
1527                                                 RegExpNode* on_success) {
1528   ActionNode* result =
1529       new(on_success->zone()) ActionNode(POSITIVE_SUBMATCH_SUCCESS, on_success);
1530   result->data_.u_submatch.stack_pointer_register = stack_reg;
1531   result->data_.u_submatch.current_position_register = position_reg;
1532   result->data_.u_submatch.clear_register_count = clear_register_count;
1533   result->data_.u_submatch.clear_register_from = clear_register_from;
1534   return result;
1535 }
1536 
1537 
EmptyMatchCheck(int start_register,int repetition_register,int repetition_limit,RegExpNode * on_success)1538 ActionNode* ActionNode::EmptyMatchCheck(int start_register,
1539                                         int repetition_register,
1540                                         int repetition_limit,
1541                                         RegExpNode* on_success) {
1542   ActionNode* result =
1543       new(on_success->zone()) ActionNode(EMPTY_MATCH_CHECK, on_success);
1544   result->data_.u_empty_match_check.start_register = start_register;
1545   result->data_.u_empty_match_check.repetition_register = repetition_register;
1546   result->data_.u_empty_match_check.repetition_limit = repetition_limit;
1547   return result;
1548 }
1549 
1550 
1551 #define DEFINE_ACCEPT(Type)                                          \
1552   void Type##Node::Accept(NodeVisitor* visitor) {                    \
1553     visitor->Visit##Type(this);                                      \
1554   }
FOR_EACH_NODE_TYPE(DEFINE_ACCEPT)1555 FOR_EACH_NODE_TYPE(DEFINE_ACCEPT)
1556 #undef DEFINE_ACCEPT
1557 
1558 
1559 void LoopChoiceNode::Accept(NodeVisitor* visitor) {
1560   visitor->VisitLoopChoice(this);
1561 }
1562 
1563 
1564 // -------------------------------------------------------------------
1565 // Emit code.
1566 
1567 
GenerateGuard(RegExpMacroAssembler * macro_assembler,Guard * guard,Trace * trace)1568 void ChoiceNode::GenerateGuard(RegExpMacroAssembler* macro_assembler,
1569                                Guard* guard,
1570                                Trace* trace) {
1571   switch (guard->op()) {
1572     case Guard::LT:
1573       DCHECK(!trace->mentions_reg(guard->reg()));
1574       macro_assembler->IfRegisterGE(guard->reg(),
1575                                     guard->value(),
1576                                     trace->backtrack());
1577       break;
1578     case Guard::GEQ:
1579       DCHECK(!trace->mentions_reg(guard->reg()));
1580       macro_assembler->IfRegisterLT(guard->reg(),
1581                                     guard->value(),
1582                                     trace->backtrack());
1583       break;
1584   }
1585 }
1586 
1587 
1588 // Returns the number of characters in the equivalence class, omitting those
1589 // that cannot occur in the source string because it is ASCII.
GetCaseIndependentLetters(Isolate * isolate,uc16 character,bool one_byte_subject,unibrow::uchar * letters)1590 static int GetCaseIndependentLetters(Isolate* isolate, uc16 character,
1591                                      bool one_byte_subject,
1592                                      unibrow::uchar* letters) {
1593   int length =
1594       isolate->jsregexp_uncanonicalize()->get(character, '\0', letters);
1595   // Unibrow returns 0 or 1 for characters where case independence is
1596   // trivial.
1597   if (length == 0) {
1598     letters[0] = character;
1599     length = 1;
1600   }
1601   if (!one_byte_subject || character <= String::kMaxOneByteCharCode) {
1602     return length;
1603   }
1604 
1605   // The standard requires that non-ASCII characters cannot have ASCII
1606   // character codes in their equivalence class.
1607   // TODO(dcarney): issue 3550 this is not actually true for Latin1 anymore,
1608   // is it?  For example, \u00C5 is equivalent to \u212B.
1609   return 0;
1610 }
1611 
1612 
EmitSimpleCharacter(Isolate * isolate,RegExpCompiler * compiler,uc16 c,Label * on_failure,int cp_offset,bool check,bool preloaded)1613 static inline bool EmitSimpleCharacter(Isolate* isolate,
1614                                        RegExpCompiler* compiler,
1615                                        uc16 c,
1616                                        Label* on_failure,
1617                                        int cp_offset,
1618                                        bool check,
1619                                        bool preloaded) {
1620   RegExpMacroAssembler* assembler = compiler->macro_assembler();
1621   bool bound_checked = false;
1622   if (!preloaded) {
1623     assembler->LoadCurrentCharacter(
1624         cp_offset,
1625         on_failure,
1626         check);
1627     bound_checked = true;
1628   }
1629   assembler->CheckNotCharacter(c, on_failure);
1630   return bound_checked;
1631 }
1632 
1633 
1634 // Only emits non-letters (things that don't have case).  Only used for case
1635 // independent matches.
EmitAtomNonLetter(Isolate * isolate,RegExpCompiler * compiler,uc16 c,Label * on_failure,int cp_offset,bool check,bool preloaded)1636 static inline bool EmitAtomNonLetter(Isolate* isolate,
1637                                      RegExpCompiler* compiler,
1638                                      uc16 c,
1639                                      Label* on_failure,
1640                                      int cp_offset,
1641                                      bool check,
1642                                      bool preloaded) {
1643   RegExpMacroAssembler* macro_assembler = compiler->macro_assembler();
1644   bool one_byte = compiler->one_byte();
1645   unibrow::uchar chars[unibrow::Ecma262UnCanonicalize::kMaxWidth];
1646   int length = GetCaseIndependentLetters(isolate, c, one_byte, chars);
1647   if (length < 1) {
1648     // This can't match.  Must be an one-byte subject and a non-one-byte
1649     // character.  We do not need to do anything since the one-byte pass
1650     // already handled this.
1651     return false;  // Bounds not checked.
1652   }
1653   bool checked = false;
1654   // We handle the length > 1 case in a later pass.
1655   if (length == 1) {
1656     if (one_byte && c > String::kMaxOneByteCharCodeU) {
1657       // Can't match - see above.
1658       return false;  // Bounds not checked.
1659     }
1660     if (!preloaded) {
1661       macro_assembler->LoadCurrentCharacter(cp_offset, on_failure, check);
1662       checked = check;
1663     }
1664     macro_assembler->CheckNotCharacter(c, on_failure);
1665   }
1666   return checked;
1667 }
1668 
1669 
ShortCutEmitCharacterPair(RegExpMacroAssembler * macro_assembler,bool one_byte,uc16 c1,uc16 c2,Label * on_failure)1670 static bool ShortCutEmitCharacterPair(RegExpMacroAssembler* macro_assembler,
1671                                       bool one_byte, uc16 c1, uc16 c2,
1672                                       Label* on_failure) {
1673   uc16 char_mask;
1674   if (one_byte) {
1675     char_mask = String::kMaxOneByteCharCode;
1676   } else {
1677     char_mask = String::kMaxUtf16CodeUnit;
1678   }
1679   uc16 exor = c1 ^ c2;
1680   // Check whether exor has only one bit set.
1681   if (((exor - 1) & exor) == 0) {
1682     // If c1 and c2 differ only by one bit.
1683     // Ecma262UnCanonicalize always gives the highest number last.
1684     DCHECK(c2 > c1);
1685     uc16 mask = char_mask ^ exor;
1686     macro_assembler->CheckNotCharacterAfterAnd(c1, mask, on_failure);
1687     return true;
1688   }
1689   DCHECK(c2 > c1);
1690   uc16 diff = c2 - c1;
1691   if (((diff - 1) & diff) == 0 && c1 >= diff) {
1692     // If the characters differ by 2^n but don't differ by one bit then
1693     // subtract the difference from the found character, then do the or
1694     // trick.  We avoid the theoretical case where negative numbers are
1695     // involved in order to simplify code generation.
1696     uc16 mask = char_mask ^ diff;
1697     macro_assembler->CheckNotCharacterAfterMinusAnd(c1 - diff,
1698                                                     diff,
1699                                                     mask,
1700                                                     on_failure);
1701     return true;
1702   }
1703   return false;
1704 }
1705 
1706 
1707 typedef bool EmitCharacterFunction(Isolate* isolate,
1708                                    RegExpCompiler* compiler,
1709                                    uc16 c,
1710                                    Label* on_failure,
1711                                    int cp_offset,
1712                                    bool check,
1713                                    bool preloaded);
1714 
1715 // Only emits letters (things that have case).  Only used for case independent
1716 // matches.
EmitAtomLetter(Isolate * isolate,RegExpCompiler * compiler,uc16 c,Label * on_failure,int cp_offset,bool check,bool preloaded)1717 static inline bool EmitAtomLetter(Isolate* isolate,
1718                                   RegExpCompiler* compiler,
1719                                   uc16 c,
1720                                   Label* on_failure,
1721                                   int cp_offset,
1722                                   bool check,
1723                                   bool preloaded) {
1724   RegExpMacroAssembler* macro_assembler = compiler->macro_assembler();
1725   bool one_byte = compiler->one_byte();
1726   unibrow::uchar chars[unibrow::Ecma262UnCanonicalize::kMaxWidth];
1727   int length = GetCaseIndependentLetters(isolate, c, one_byte, chars);
1728   if (length <= 1) return false;
1729   // We may not need to check against the end of the input string
1730   // if this character lies before a character that matched.
1731   if (!preloaded) {
1732     macro_assembler->LoadCurrentCharacter(cp_offset, on_failure, check);
1733   }
1734   Label ok;
1735   DCHECK(unibrow::Ecma262UnCanonicalize::kMaxWidth == 4);
1736   switch (length) {
1737     case 2: {
1738       if (ShortCutEmitCharacterPair(macro_assembler, one_byte, chars[0],
1739                                     chars[1], on_failure)) {
1740       } else {
1741         macro_assembler->CheckCharacter(chars[0], &ok);
1742         macro_assembler->CheckNotCharacter(chars[1], on_failure);
1743         macro_assembler->Bind(&ok);
1744       }
1745       break;
1746     }
1747     case 4:
1748       macro_assembler->CheckCharacter(chars[3], &ok);
1749       // Fall through!
1750     case 3:
1751       macro_assembler->CheckCharacter(chars[0], &ok);
1752       macro_assembler->CheckCharacter(chars[1], &ok);
1753       macro_assembler->CheckNotCharacter(chars[2], on_failure);
1754       macro_assembler->Bind(&ok);
1755       break;
1756     default:
1757       UNREACHABLE();
1758       break;
1759   }
1760   return true;
1761 }
1762 
1763 
EmitBoundaryTest(RegExpMacroAssembler * masm,int border,Label * fall_through,Label * above_or_equal,Label * below)1764 static void EmitBoundaryTest(RegExpMacroAssembler* masm,
1765                              int border,
1766                              Label* fall_through,
1767                              Label* above_or_equal,
1768                              Label* below) {
1769   if (below != fall_through) {
1770     masm->CheckCharacterLT(border, below);
1771     if (above_or_equal != fall_through) masm->GoTo(above_or_equal);
1772   } else {
1773     masm->CheckCharacterGT(border - 1, above_or_equal);
1774   }
1775 }
1776 
1777 
EmitDoubleBoundaryTest(RegExpMacroAssembler * masm,int first,int last,Label * fall_through,Label * in_range,Label * out_of_range)1778 static void EmitDoubleBoundaryTest(RegExpMacroAssembler* masm,
1779                                    int first,
1780                                    int last,
1781                                    Label* fall_through,
1782                                    Label* in_range,
1783                                    Label* out_of_range) {
1784   if (in_range == fall_through) {
1785     if (first == last) {
1786       masm->CheckNotCharacter(first, out_of_range);
1787     } else {
1788       masm->CheckCharacterNotInRange(first, last, out_of_range);
1789     }
1790   } else {
1791     if (first == last) {
1792       masm->CheckCharacter(first, in_range);
1793     } else {
1794       masm->CheckCharacterInRange(first, last, in_range);
1795     }
1796     if (out_of_range != fall_through) masm->GoTo(out_of_range);
1797   }
1798 }
1799 
1800 
1801 // even_label is for ranges[i] to ranges[i + 1] where i - start_index is even.
1802 // 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)1803 static void EmitUseLookupTable(
1804     RegExpMacroAssembler* masm,
1805     ZoneList<int>* ranges,
1806     int start_index,
1807     int end_index,
1808     int min_char,
1809     Label* fall_through,
1810     Label* even_label,
1811     Label* odd_label) {
1812   static const int kSize = RegExpMacroAssembler::kTableSize;
1813   static const int kMask = RegExpMacroAssembler::kTableMask;
1814 
1815   int base = (min_char & ~kMask);
1816   USE(base);
1817 
1818   // Assert that everything is on one kTableSize page.
1819   for (int i = start_index; i <= end_index; i++) {
1820     DCHECK_EQ(ranges->at(i) & ~kMask, base);
1821   }
1822   DCHECK(start_index == 0 || (ranges->at(start_index - 1) & ~kMask) <= base);
1823 
1824   char templ[kSize];
1825   Label* on_bit_set;
1826   Label* on_bit_clear;
1827   int bit;
1828   if (even_label == fall_through) {
1829     on_bit_set = odd_label;
1830     on_bit_clear = even_label;
1831     bit = 1;
1832   } else {
1833     on_bit_set = even_label;
1834     on_bit_clear = odd_label;
1835     bit = 0;
1836   }
1837   for (int i = 0; i < (ranges->at(start_index) & kMask) && i < kSize; i++) {
1838     templ[i] = bit;
1839   }
1840   int j = 0;
1841   bit ^= 1;
1842   for (int i = start_index; i < end_index; i++) {
1843     for (j = (ranges->at(i) & kMask); j < (ranges->at(i + 1) & kMask); j++) {
1844       templ[j] = bit;
1845     }
1846     bit ^= 1;
1847   }
1848   for (int i = j; i < kSize; i++) {
1849     templ[i] = bit;
1850   }
1851   Factory* factory = masm->zone()->isolate()->factory();
1852   // TODO(erikcorry): Cache these.
1853   Handle<ByteArray> ba = factory->NewByteArray(kSize, TENURED);
1854   for (int i = 0; i < kSize; i++) {
1855     ba->set(i, templ[i]);
1856   }
1857   masm->CheckBitInTable(ba, on_bit_set);
1858   if (on_bit_clear != fall_through) masm->GoTo(on_bit_clear);
1859 }
1860 
1861 
CutOutRange(RegExpMacroAssembler * masm,ZoneList<int> * ranges,int start_index,int end_index,int cut_index,Label * even_label,Label * odd_label)1862 static void CutOutRange(RegExpMacroAssembler* masm,
1863                         ZoneList<int>* ranges,
1864                         int start_index,
1865                         int end_index,
1866                         int cut_index,
1867                         Label* even_label,
1868                         Label* odd_label) {
1869   bool odd = (((cut_index - start_index) & 1) == 1);
1870   Label* in_range_label = odd ? odd_label : even_label;
1871   Label dummy;
1872   EmitDoubleBoundaryTest(masm,
1873                          ranges->at(cut_index),
1874                          ranges->at(cut_index + 1) - 1,
1875                          &dummy,
1876                          in_range_label,
1877                          &dummy);
1878   DCHECK(!dummy.is_linked());
1879   // Cut out the single range by rewriting the array.  This creates a new
1880   // range that is a merger of the two ranges on either side of the one we
1881   // are cutting out.  The oddity of the labels is preserved.
1882   for (int j = cut_index; j > start_index; j--) {
1883     ranges->at(j) = ranges->at(j - 1);
1884   }
1885   for (int j = cut_index + 1; j < end_index; j++) {
1886     ranges->at(j) = ranges->at(j + 1);
1887   }
1888 }
1889 
1890 
1891 // Unicode case.  Split the search space into kSize spaces that are handled
1892 // with recursion.
SplitSearchSpace(ZoneList<int> * ranges,int start_index,int end_index,int * new_start_index,int * new_end_index,int * border)1893 static void SplitSearchSpace(ZoneList<int>* ranges,
1894                              int start_index,
1895                              int end_index,
1896                              int* new_start_index,
1897                              int* new_end_index,
1898                              int* border) {
1899   static const int kSize = RegExpMacroAssembler::kTableSize;
1900   static const int kMask = RegExpMacroAssembler::kTableMask;
1901 
1902   int first = ranges->at(start_index);
1903   int last = ranges->at(end_index) - 1;
1904 
1905   *new_start_index = start_index;
1906   *border = (ranges->at(start_index) & ~kMask) + kSize;
1907   while (*new_start_index < end_index) {
1908     if (ranges->at(*new_start_index) > *border) break;
1909     (*new_start_index)++;
1910   }
1911   // new_start_index is the index of the first edge that is beyond the
1912   // current kSize space.
1913 
1914   // For very large search spaces we do a binary chop search of the non-Latin1
1915   // space instead of just going to the end of the current kSize space.  The
1916   // heuristics are complicated a little by the fact that any 128-character
1917   // encoding space can be quickly tested with a table lookup, so we don't
1918   // wish to do binary chop search at a smaller granularity than that.  A
1919   // 128-character space can take up a lot of space in the ranges array if,
1920   // for example, we only want to match every second character (eg. the lower
1921   // case characters on some Unicode pages).
1922   int binary_chop_index = (end_index + start_index) / 2;
1923   // The first test ensures that we get to the code that handles the Latin1
1924   // range with a single not-taken branch, speeding up this important
1925   // character range (even non-Latin1 charset-based text has spaces and
1926   // punctuation).
1927   if (*border - 1 > String::kMaxOneByteCharCode &&  // Latin1 case.
1928       end_index - start_index > (*new_start_index - start_index) * 2 &&
1929       last - first > kSize * 2 && binary_chop_index > *new_start_index &&
1930       ranges->at(binary_chop_index) >= first + 2 * kSize) {
1931     int scan_forward_for_section_border = binary_chop_index;;
1932     int new_border = (ranges->at(binary_chop_index) | kMask) + 1;
1933 
1934     while (scan_forward_for_section_border < end_index) {
1935       if (ranges->at(scan_forward_for_section_border) > new_border) {
1936         *new_start_index = scan_forward_for_section_border;
1937         *border = new_border;
1938         break;
1939       }
1940       scan_forward_for_section_border++;
1941     }
1942   }
1943 
1944   DCHECK(*new_start_index > start_index);
1945   *new_end_index = *new_start_index - 1;
1946   if (ranges->at(*new_end_index) == *border) {
1947     (*new_end_index)--;
1948   }
1949   if (*border >= ranges->at(end_index)) {
1950     *border = ranges->at(end_index);
1951     *new_start_index = end_index;  // Won't be used.
1952     *new_end_index = end_index - 1;
1953   }
1954 }
1955 
1956 
1957 // Gets a series of segment boundaries representing a character class.  If the
1958 // character is in the range between an even and an odd boundary (counting from
1959 // start_index) then go to even_label, otherwise go to odd_label.  We already
1960 // know that the character is in the range of min_char to max_char inclusive.
1961 // Either label can be NULL indicating backtracking.  Either label can also be
1962 // 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)1963 static void GenerateBranches(RegExpMacroAssembler* masm,
1964                              ZoneList<int>* ranges,
1965                              int start_index,
1966                              int end_index,
1967                              uc16 min_char,
1968                              uc16 max_char,
1969                              Label* fall_through,
1970                              Label* even_label,
1971                              Label* odd_label) {
1972   int first = ranges->at(start_index);
1973   int last = ranges->at(end_index) - 1;
1974 
1975   DCHECK_LT(min_char, first);
1976 
1977   // Just need to test if the character is before or on-or-after
1978   // a particular character.
1979   if (start_index == end_index) {
1980     EmitBoundaryTest(masm, first, fall_through, even_label, odd_label);
1981     return;
1982   }
1983 
1984   // Another almost trivial case:  There is one interval in the middle that is
1985   // different from the end intervals.
1986   if (start_index + 1 == end_index) {
1987     EmitDoubleBoundaryTest(
1988         masm, first, last, fall_through, even_label, odd_label);
1989     return;
1990   }
1991 
1992   // It's not worth using table lookup if there are very few intervals in the
1993   // character class.
1994   if (end_index - start_index <= 6) {
1995     // It is faster to test for individual characters, so we look for those
1996     // first, then try arbitrary ranges in the second round.
1997     static int kNoCutIndex = -1;
1998     int cut = kNoCutIndex;
1999     for (int i = start_index; i < end_index; i++) {
2000       if (ranges->at(i) == ranges->at(i + 1) - 1) {
2001         cut = i;
2002         break;
2003       }
2004     }
2005     if (cut == kNoCutIndex) cut = start_index;
2006     CutOutRange(
2007         masm, ranges, start_index, end_index, cut, even_label, odd_label);
2008     DCHECK_GE(end_index - start_index, 2);
2009     GenerateBranches(masm,
2010                      ranges,
2011                      start_index + 1,
2012                      end_index - 1,
2013                      min_char,
2014                      max_char,
2015                      fall_through,
2016                      even_label,
2017                      odd_label);
2018     return;
2019   }
2020 
2021   // If there are a lot of intervals in the regexp, then we will use tables to
2022   // determine whether the character is inside or outside the character class.
2023   static const int kBits = RegExpMacroAssembler::kTableSizeBits;
2024 
2025   if ((max_char >> kBits) == (min_char >> kBits)) {
2026     EmitUseLookupTable(masm,
2027                        ranges,
2028                        start_index,
2029                        end_index,
2030                        min_char,
2031                        fall_through,
2032                        even_label,
2033                        odd_label);
2034     return;
2035   }
2036 
2037   if ((min_char >> kBits) != (first >> kBits)) {
2038     masm->CheckCharacterLT(first, odd_label);
2039     GenerateBranches(masm,
2040                      ranges,
2041                      start_index + 1,
2042                      end_index,
2043                      first,
2044                      max_char,
2045                      fall_through,
2046                      odd_label,
2047                      even_label);
2048     return;
2049   }
2050 
2051   int new_start_index = 0;
2052   int new_end_index = 0;
2053   int border = 0;
2054 
2055   SplitSearchSpace(ranges,
2056                    start_index,
2057                    end_index,
2058                    &new_start_index,
2059                    &new_end_index,
2060                    &border);
2061 
2062   Label handle_rest;
2063   Label* above = &handle_rest;
2064   if (border == last + 1) {
2065     // We didn't find any section that started after the limit, so everything
2066     // above the border is one of the terminal labels.
2067     above = (end_index & 1) != (start_index & 1) ? odd_label : even_label;
2068     DCHECK(new_end_index == end_index - 1);
2069   }
2070 
2071   DCHECK_LE(start_index, new_end_index);
2072   DCHECK_LE(new_start_index, end_index);
2073   DCHECK_LT(start_index, new_start_index);
2074   DCHECK_LT(new_end_index, end_index);
2075   DCHECK(new_end_index + 1 == new_start_index ||
2076          (new_end_index + 2 == new_start_index &&
2077           border == ranges->at(new_end_index + 1)));
2078   DCHECK_LT(min_char, border - 1);
2079   DCHECK_LT(border, max_char);
2080   DCHECK_LT(ranges->at(new_end_index), border);
2081   DCHECK(border < ranges->at(new_start_index) ||
2082          (border == ranges->at(new_start_index) &&
2083           new_start_index == end_index &&
2084           new_end_index == end_index - 1 &&
2085           border == last + 1));
2086   DCHECK(new_start_index == 0 || border >= ranges->at(new_start_index - 1));
2087 
2088   masm->CheckCharacterGT(border - 1, above);
2089   Label dummy;
2090   GenerateBranches(masm,
2091                    ranges,
2092                    start_index,
2093                    new_end_index,
2094                    min_char,
2095                    border - 1,
2096                    &dummy,
2097                    even_label,
2098                    odd_label);
2099   if (handle_rest.is_linked()) {
2100     masm->Bind(&handle_rest);
2101     bool flip = (new_start_index & 1) != (start_index & 1);
2102     GenerateBranches(masm,
2103                      ranges,
2104                      new_start_index,
2105                      end_index,
2106                      border,
2107                      max_char,
2108                      &dummy,
2109                      flip ? odd_label : even_label,
2110                      flip ? even_label : odd_label);
2111   }
2112 }
2113 
2114 
EmitCharClass(RegExpMacroAssembler * macro_assembler,RegExpCharacterClass * cc,bool one_byte,Label * on_failure,int cp_offset,bool check_offset,bool preloaded,Zone * zone)2115 static void EmitCharClass(RegExpMacroAssembler* macro_assembler,
2116                           RegExpCharacterClass* cc, bool one_byte,
2117                           Label* on_failure, int cp_offset, bool check_offset,
2118                           bool preloaded, Zone* zone) {
2119   ZoneList<CharacterRange>* ranges = cc->ranges(zone);
2120   if (!CharacterRange::IsCanonical(ranges)) {
2121     CharacterRange::Canonicalize(ranges);
2122   }
2123 
2124   int max_char;
2125   if (one_byte) {
2126     max_char = String::kMaxOneByteCharCode;
2127   } else {
2128     max_char = String::kMaxUtf16CodeUnit;
2129   }
2130 
2131   int range_count = ranges->length();
2132 
2133   int last_valid_range = range_count - 1;
2134   while (last_valid_range >= 0) {
2135     CharacterRange& range = ranges->at(last_valid_range);
2136     if (range.from() <= max_char) {
2137       break;
2138     }
2139     last_valid_range--;
2140   }
2141 
2142   if (last_valid_range < 0) {
2143     if (!cc->is_negated()) {
2144       macro_assembler->GoTo(on_failure);
2145     }
2146     if (check_offset) {
2147       macro_assembler->CheckPosition(cp_offset, on_failure);
2148     }
2149     return;
2150   }
2151 
2152   if (last_valid_range == 0 &&
2153       ranges->at(0).IsEverything(max_char)) {
2154     if (cc->is_negated()) {
2155       macro_assembler->GoTo(on_failure);
2156     } else {
2157       // This is a common case hit by non-anchored expressions.
2158       if (check_offset) {
2159         macro_assembler->CheckPosition(cp_offset, on_failure);
2160       }
2161     }
2162     return;
2163   }
2164   if (last_valid_range == 0 &&
2165       !cc->is_negated() &&
2166       ranges->at(0).IsEverything(max_char)) {
2167     // This is a common case hit by non-anchored expressions.
2168     if (check_offset) {
2169       macro_assembler->CheckPosition(cp_offset, on_failure);
2170     }
2171     return;
2172   }
2173 
2174   if (!preloaded) {
2175     macro_assembler->LoadCurrentCharacter(cp_offset, on_failure, check_offset);
2176   }
2177 
2178   if (cc->is_standard(zone) &&
2179         macro_assembler->CheckSpecialCharacterClass(cc->standard_type(),
2180                                                     on_failure)) {
2181       return;
2182   }
2183 
2184 
2185   // A new list with ascending entries.  Each entry is a code unit
2186   // where there is a boundary between code units that are part of
2187   // the class and code units that are not.  Normally we insert an
2188   // entry at zero which goes to the failure label, but if there
2189   // was already one there we fall through for success on that entry.
2190   // Subsequent entries have alternating meaning (success/failure).
2191   ZoneList<int>* range_boundaries =
2192       new(zone) ZoneList<int>(last_valid_range, zone);
2193 
2194   bool zeroth_entry_is_failure = !cc->is_negated();
2195 
2196   for (int i = 0; i <= last_valid_range; i++) {
2197     CharacterRange& range = ranges->at(i);
2198     if (range.from() == 0) {
2199       DCHECK_EQ(i, 0);
2200       zeroth_entry_is_failure = !zeroth_entry_is_failure;
2201     } else {
2202       range_boundaries->Add(range.from(), zone);
2203     }
2204     range_boundaries->Add(range.to() + 1, zone);
2205   }
2206   int end_index = range_boundaries->length() - 1;
2207   if (range_boundaries->at(end_index) > max_char) {
2208     end_index--;
2209   }
2210 
2211   Label fall_through;
2212   GenerateBranches(macro_assembler,
2213                    range_boundaries,
2214                    0,  // start_index.
2215                    end_index,
2216                    0,  // min_char.
2217                    max_char,
2218                    &fall_through,
2219                    zeroth_entry_is_failure ? &fall_through : on_failure,
2220                    zeroth_entry_is_failure ? on_failure : &fall_through);
2221   macro_assembler->Bind(&fall_through);
2222 }
2223 
2224 
~RegExpNode()2225 RegExpNode::~RegExpNode() {
2226 }
2227 
2228 
LimitVersions(RegExpCompiler * compiler,Trace * trace)2229 RegExpNode::LimitResult RegExpNode::LimitVersions(RegExpCompiler* compiler,
2230                                                   Trace* trace) {
2231   // If we are generating a greedy loop then don't stop and don't reuse code.
2232   if (trace->stop_node() != NULL) {
2233     return CONTINUE;
2234   }
2235 
2236   RegExpMacroAssembler* macro_assembler = compiler->macro_assembler();
2237   if (trace->is_trivial()) {
2238     if (label_.is_bound()) {
2239       // We are being asked to generate a generic version, but that's already
2240       // been done so just go to it.
2241       macro_assembler->GoTo(&label_);
2242       return DONE;
2243     }
2244     if (compiler->recursion_depth() >= RegExpCompiler::kMaxRecursion) {
2245       // To avoid too deep recursion we push the node to the work queue and just
2246       // generate a goto here.
2247       compiler->AddWork(this);
2248       macro_assembler->GoTo(&label_);
2249       return DONE;
2250     }
2251     // Generate generic version of the node and bind the label for later use.
2252     macro_assembler->Bind(&label_);
2253     return CONTINUE;
2254   }
2255 
2256   // We are being asked to make a non-generic version.  Keep track of how many
2257   // non-generic versions we generate so as not to overdo it.
2258   trace_count_++;
2259   if (FLAG_regexp_optimization &&
2260       trace_count_ < kMaxCopiesCodeGenerated &&
2261       compiler->recursion_depth() <= RegExpCompiler::kMaxRecursion) {
2262     return CONTINUE;
2263   }
2264 
2265   // If we get here code has been generated for this node too many times or
2266   // recursion is too deep.  Time to switch to a generic version.  The code for
2267   // generic versions above can handle deep recursion properly.
2268   trace->Flush(compiler, this);
2269   return DONE;
2270 }
2271 
2272 
EatsAtLeast(int still_to_find,int budget,bool not_at_start)2273 int ActionNode::EatsAtLeast(int still_to_find,
2274                             int budget,
2275                             bool not_at_start) {
2276   if (budget <= 0) return 0;
2277   if (action_type_ == POSITIVE_SUBMATCH_SUCCESS) return 0;  // Rewinds input!
2278   return on_success()->EatsAtLeast(still_to_find,
2279                                    budget - 1,
2280                                    not_at_start);
2281 }
2282 
2283 
FillInBMInfo(int offset,int budget,BoyerMooreLookahead * bm,bool not_at_start)2284 void ActionNode::FillInBMInfo(int offset,
2285                               int budget,
2286                               BoyerMooreLookahead* bm,
2287                               bool not_at_start) {
2288   if (action_type_ == BEGIN_SUBMATCH) {
2289     bm->SetRest(offset);
2290   } else if (action_type_ != POSITIVE_SUBMATCH_SUCCESS) {
2291     on_success()->FillInBMInfo(offset, budget - 1, bm, not_at_start);
2292   }
2293   SaveBMInfo(bm, not_at_start, offset);
2294 }
2295 
2296 
EatsAtLeast(int still_to_find,int budget,bool not_at_start)2297 int AssertionNode::EatsAtLeast(int still_to_find,
2298                                int budget,
2299                                bool not_at_start) {
2300   if (budget <= 0) return 0;
2301   // If we know we are not at the start and we are asked "how many characters
2302   // will you match if you succeed?" then we can answer anything since false
2303   // implies false.  So lets just return the max answer (still_to_find) since
2304   // that won't prevent us from preloading a lot of characters for the other
2305   // branches in the node graph.
2306   if (assertion_type() == AT_START && not_at_start) return still_to_find;
2307   return on_success()->EatsAtLeast(still_to_find,
2308                                    budget - 1,
2309                                    not_at_start);
2310 }
2311 
2312 
FillInBMInfo(int offset,int budget,BoyerMooreLookahead * bm,bool not_at_start)2313 void AssertionNode::FillInBMInfo(int offset,
2314                                  int budget,
2315                                  BoyerMooreLookahead* bm,
2316                                  bool not_at_start) {
2317   // Match the behaviour of EatsAtLeast on this node.
2318   if (assertion_type() == AT_START && not_at_start) return;
2319   on_success()->FillInBMInfo(offset, budget - 1, bm, not_at_start);
2320   SaveBMInfo(bm, not_at_start, offset);
2321 }
2322 
2323 
EatsAtLeast(int still_to_find,int budget,bool not_at_start)2324 int BackReferenceNode::EatsAtLeast(int still_to_find,
2325                                    int budget,
2326                                    bool not_at_start) {
2327   if (budget <= 0) return 0;
2328   return on_success()->EatsAtLeast(still_to_find,
2329                                    budget - 1,
2330                                    not_at_start);
2331 }
2332 
2333 
EatsAtLeast(int still_to_find,int budget,bool not_at_start)2334 int TextNode::EatsAtLeast(int still_to_find,
2335                           int budget,
2336                           bool not_at_start) {
2337   int answer = Length();
2338   if (answer >= still_to_find) return answer;
2339   if (budget <= 0) return answer;
2340   // We are not at start after this node so we set the last argument to 'true'.
2341   return answer + on_success()->EatsAtLeast(still_to_find - answer,
2342                                             budget - 1,
2343                                             true);
2344 }
2345 
2346 
EatsAtLeast(int still_to_find,int budget,bool not_at_start)2347 int NegativeLookaheadChoiceNode::EatsAtLeast(int still_to_find,
2348                                              int budget,
2349                                              bool not_at_start) {
2350   if (budget <= 0) return 0;
2351   // Alternative 0 is the negative lookahead, alternative 1 is what comes
2352   // afterwards.
2353   RegExpNode* node = alternatives_->at(1).node();
2354   return node->EatsAtLeast(still_to_find, budget - 1, not_at_start);
2355 }
2356 
2357 
GetQuickCheckDetails(QuickCheckDetails * details,RegExpCompiler * compiler,int filled_in,bool not_at_start)2358 void NegativeLookaheadChoiceNode::GetQuickCheckDetails(
2359     QuickCheckDetails* details,
2360     RegExpCompiler* compiler,
2361     int filled_in,
2362     bool not_at_start) {
2363   // Alternative 0 is the negative lookahead, alternative 1 is what comes
2364   // afterwards.
2365   RegExpNode* node = alternatives_->at(1).node();
2366   return node->GetQuickCheckDetails(details, compiler, filled_in, not_at_start);
2367 }
2368 
2369 
EatsAtLeastHelper(int still_to_find,int budget,RegExpNode * ignore_this_node,bool not_at_start)2370 int ChoiceNode::EatsAtLeastHelper(int still_to_find,
2371                                   int budget,
2372                                   RegExpNode* ignore_this_node,
2373                                   bool not_at_start) {
2374   if (budget <= 0) return 0;
2375   int min = 100;
2376   int choice_count = alternatives_->length();
2377   budget = (budget - 1) / choice_count;
2378   for (int i = 0; i < choice_count; i++) {
2379     RegExpNode* node = alternatives_->at(i).node();
2380     if (node == ignore_this_node) continue;
2381     int node_eats_at_least =
2382         node->EatsAtLeast(still_to_find, budget, not_at_start);
2383     if (node_eats_at_least < min) min = node_eats_at_least;
2384     if (min == 0) return 0;
2385   }
2386   return min;
2387 }
2388 
2389 
EatsAtLeast(int still_to_find,int budget,bool not_at_start)2390 int LoopChoiceNode::EatsAtLeast(int still_to_find,
2391                                 int budget,
2392                                 bool not_at_start) {
2393   return EatsAtLeastHelper(still_to_find,
2394                            budget - 1,
2395                            loop_node_,
2396                            not_at_start);
2397 }
2398 
2399 
EatsAtLeast(int still_to_find,int budget,bool not_at_start)2400 int ChoiceNode::EatsAtLeast(int still_to_find,
2401                             int budget,
2402                             bool not_at_start) {
2403   return EatsAtLeastHelper(still_to_find,
2404                            budget,
2405                            NULL,
2406                            not_at_start);
2407 }
2408 
2409 
2410 // Takes the left-most 1-bit and smears it out, setting all bits to its right.
SmearBitsRight(uint32_t v)2411 static inline uint32_t SmearBitsRight(uint32_t v) {
2412   v |= v >> 1;
2413   v |= v >> 2;
2414   v |= v >> 4;
2415   v |= v >> 8;
2416   v |= v >> 16;
2417   return v;
2418 }
2419 
2420 
Rationalize(bool asc)2421 bool QuickCheckDetails::Rationalize(bool asc) {
2422   bool found_useful_op = false;
2423   uint32_t char_mask;
2424   if (asc) {
2425     char_mask = String::kMaxOneByteCharCode;
2426   } else {
2427     char_mask = String::kMaxUtf16CodeUnit;
2428   }
2429   mask_ = 0;
2430   value_ = 0;
2431   int char_shift = 0;
2432   for (int i = 0; i < characters_; i++) {
2433     Position* pos = &positions_[i];
2434     if ((pos->mask & String::kMaxOneByteCharCode) != 0) {
2435       found_useful_op = true;
2436     }
2437     mask_ |= (pos->mask & char_mask) << char_shift;
2438     value_ |= (pos->value & char_mask) << char_shift;
2439     char_shift += asc ? 8 : 16;
2440   }
2441   return found_useful_op;
2442 }
2443 
2444 
EmitQuickCheck(RegExpCompiler * compiler,Trace * bounds_check_trace,Trace * trace,bool preload_has_checked_bounds,Label * on_possible_success,QuickCheckDetails * details,bool fall_through_on_failure)2445 bool RegExpNode::EmitQuickCheck(RegExpCompiler* compiler,
2446                                 Trace* bounds_check_trace,
2447                                 Trace* trace,
2448                                 bool preload_has_checked_bounds,
2449                                 Label* on_possible_success,
2450                                 QuickCheckDetails* details,
2451                                 bool fall_through_on_failure) {
2452   if (details->characters() == 0) return false;
2453   GetQuickCheckDetails(
2454       details, compiler, 0, trace->at_start() == Trace::FALSE_VALUE);
2455   if (details->cannot_match()) return false;
2456   if (!details->Rationalize(compiler->one_byte())) return false;
2457   DCHECK(details->characters() == 1 ||
2458          compiler->macro_assembler()->CanReadUnaligned());
2459   uint32_t mask = details->mask();
2460   uint32_t value = details->value();
2461 
2462   RegExpMacroAssembler* assembler = compiler->macro_assembler();
2463 
2464   if (trace->characters_preloaded() != details->characters()) {
2465     DCHECK(trace->cp_offset() == bounds_check_trace->cp_offset());
2466     // We are attempting to preload the minimum number of characters
2467     // any choice would eat, so if the bounds check fails, then none of the
2468     // choices can succeed, so we can just immediately backtrack, rather
2469     // than go to the next choice.
2470     assembler->LoadCurrentCharacter(trace->cp_offset(),
2471                                     bounds_check_trace->backtrack(),
2472                                     !preload_has_checked_bounds,
2473                                     details->characters());
2474   }
2475 
2476 
2477   bool need_mask = true;
2478 
2479   if (details->characters() == 1) {
2480     // If number of characters preloaded is 1 then we used a byte or 16 bit
2481     // load so the value is already masked down.
2482     uint32_t char_mask;
2483     if (compiler->one_byte()) {
2484       char_mask = String::kMaxOneByteCharCode;
2485     } else {
2486       char_mask = String::kMaxUtf16CodeUnit;
2487     }
2488     if ((mask & char_mask) == char_mask) need_mask = false;
2489     mask &= char_mask;
2490   } else {
2491     // For 2-character preloads in one-byte mode or 1-character preloads in
2492     // two-byte mode we also use a 16 bit load with zero extend.
2493     if (details->characters() == 2 && compiler->one_byte()) {
2494       if ((mask & 0xffff) == 0xffff) need_mask = false;
2495     } else if (details->characters() == 1 && !compiler->one_byte()) {
2496       if ((mask & 0xffff) == 0xffff) need_mask = false;
2497     } else {
2498       if (mask == 0xffffffff) need_mask = false;
2499     }
2500   }
2501 
2502   if (fall_through_on_failure) {
2503     if (need_mask) {
2504       assembler->CheckCharacterAfterAnd(value, mask, on_possible_success);
2505     } else {
2506       assembler->CheckCharacter(value, on_possible_success);
2507     }
2508   } else {
2509     if (need_mask) {
2510       assembler->CheckNotCharacterAfterAnd(value, mask, trace->backtrack());
2511     } else {
2512       assembler->CheckNotCharacter(value, trace->backtrack());
2513     }
2514   }
2515   return true;
2516 }
2517 
2518 
2519 // Here is the meat of GetQuickCheckDetails (see also the comment on the
2520 // super-class in the .h file).
2521 //
2522 // We iterate along the text object, building up for each character a
2523 // mask and value that can be used to test for a quick failure to match.
2524 // The masks and values for the positions will be combined into a single
2525 // machine word for the current character width in order to be used in
2526 // generating a quick check.
GetQuickCheckDetails(QuickCheckDetails * details,RegExpCompiler * compiler,int characters_filled_in,bool not_at_start)2527 void TextNode::GetQuickCheckDetails(QuickCheckDetails* details,
2528                                     RegExpCompiler* compiler,
2529                                     int characters_filled_in,
2530                                     bool not_at_start) {
2531   Isolate* isolate = compiler->macro_assembler()->zone()->isolate();
2532   DCHECK(characters_filled_in < details->characters());
2533   int characters = details->characters();
2534   int char_mask;
2535   if (compiler->one_byte()) {
2536     char_mask = String::kMaxOneByteCharCode;
2537   } else {
2538     char_mask = String::kMaxUtf16CodeUnit;
2539   }
2540   for (int k = 0; k < elms_->length(); k++) {
2541     TextElement elm = elms_->at(k);
2542     if (elm.text_type() == TextElement::ATOM) {
2543       Vector<const uc16> quarks = elm.atom()->data();
2544       for (int i = 0; i < characters && i < quarks.length(); i++) {
2545         QuickCheckDetails::Position* pos =
2546             details->positions(characters_filled_in);
2547         uc16 c = quarks[i];
2548         if (c > char_mask) {
2549           // If we expect a non-Latin1 character from an one-byte string,
2550           // there is no way we can match. Not even case-independent
2551           // matching can turn an Latin1 character into non-Latin1 or
2552           // vice versa.
2553           // TODO(dcarney): issue 3550.  Verify that this works as expected.
2554           // For example, \u0178 is uppercase of \u00ff (y-umlaut).
2555           details->set_cannot_match();
2556           pos->determines_perfectly = false;
2557           return;
2558         }
2559         if (compiler->ignore_case()) {
2560           unibrow::uchar chars[unibrow::Ecma262UnCanonicalize::kMaxWidth];
2561           int length = GetCaseIndependentLetters(isolate, c,
2562                                                  compiler->one_byte(), chars);
2563           DCHECK(length != 0);  // Can only happen if c > char_mask (see above).
2564           if (length == 1) {
2565             // This letter has no case equivalents, so it's nice and simple
2566             // and the mask-compare will determine definitely whether we have
2567             // a match at this character position.
2568             pos->mask = char_mask;
2569             pos->value = c;
2570             pos->determines_perfectly = true;
2571           } else {
2572             uint32_t common_bits = char_mask;
2573             uint32_t bits = chars[0];
2574             for (int j = 1; j < length; j++) {
2575               uint32_t differing_bits = ((chars[j] & common_bits) ^ bits);
2576               common_bits ^= differing_bits;
2577               bits &= common_bits;
2578             }
2579             // If length is 2 and common bits has only one zero in it then
2580             // our mask and compare instruction will determine definitely
2581             // whether we have a match at this character position.  Otherwise
2582             // it can only be an approximate check.
2583             uint32_t one_zero = (common_bits | ~char_mask);
2584             if (length == 2 && ((~one_zero) & ((~one_zero) - 1)) == 0) {
2585               pos->determines_perfectly = true;
2586             }
2587             pos->mask = common_bits;
2588             pos->value = bits;
2589           }
2590         } else {
2591           // Don't ignore case.  Nice simple case where the mask-compare will
2592           // determine definitely whether we have a match at this character
2593           // position.
2594           pos->mask = char_mask;
2595           pos->value = c;
2596           pos->determines_perfectly = true;
2597         }
2598         characters_filled_in++;
2599         DCHECK(characters_filled_in <= details->characters());
2600         if (characters_filled_in == details->characters()) {
2601           return;
2602         }
2603       }
2604     } else {
2605       QuickCheckDetails::Position* pos =
2606           details->positions(characters_filled_in);
2607       RegExpCharacterClass* tree = elm.char_class();
2608       ZoneList<CharacterRange>* ranges = tree->ranges(zone());
2609       if (tree->is_negated()) {
2610         // A quick check uses multi-character mask and compare.  There is no
2611         // useful way to incorporate a negative char class into this scheme
2612         // so we just conservatively create a mask and value that will always
2613         // succeed.
2614         pos->mask = 0;
2615         pos->value = 0;
2616       } else {
2617         int first_range = 0;
2618         while (ranges->at(first_range).from() > char_mask) {
2619           first_range++;
2620           if (first_range == ranges->length()) {
2621             details->set_cannot_match();
2622             pos->determines_perfectly = false;
2623             return;
2624           }
2625         }
2626         CharacterRange range = ranges->at(first_range);
2627         uc16 from = range.from();
2628         uc16 to = range.to();
2629         if (to > char_mask) {
2630           to = char_mask;
2631         }
2632         uint32_t differing_bits = (from ^ to);
2633         // A mask and compare is only perfect if the differing bits form a
2634         // number like 00011111 with one single block of trailing 1s.
2635         if ((differing_bits & (differing_bits + 1)) == 0 &&
2636              from + differing_bits == to) {
2637           pos->determines_perfectly = true;
2638         }
2639         uint32_t common_bits = ~SmearBitsRight(differing_bits);
2640         uint32_t bits = (from & common_bits);
2641         for (int i = first_range + 1; i < ranges->length(); i++) {
2642           CharacterRange range = ranges->at(i);
2643           uc16 from = range.from();
2644           uc16 to = range.to();
2645           if (from > char_mask) continue;
2646           if (to > char_mask) to = char_mask;
2647           // Here we are combining more ranges into the mask and compare
2648           // value.  With each new range the mask becomes more sparse and
2649           // so the chances of a false positive rise.  A character class
2650           // with multiple ranges is assumed never to be equivalent to a
2651           // mask and compare operation.
2652           pos->determines_perfectly = false;
2653           uint32_t new_common_bits = (from ^ to);
2654           new_common_bits = ~SmearBitsRight(new_common_bits);
2655           common_bits &= new_common_bits;
2656           bits &= new_common_bits;
2657           uint32_t differing_bits = (from & common_bits) ^ bits;
2658           common_bits ^= differing_bits;
2659           bits &= common_bits;
2660         }
2661         pos->mask = common_bits;
2662         pos->value = bits;
2663       }
2664       characters_filled_in++;
2665       DCHECK(characters_filled_in <= details->characters());
2666       if (characters_filled_in == details->characters()) {
2667         return;
2668       }
2669     }
2670   }
2671   DCHECK(characters_filled_in != details->characters());
2672   if (!details->cannot_match()) {
2673     on_success()-> GetQuickCheckDetails(details,
2674                                         compiler,
2675                                         characters_filled_in,
2676                                         true);
2677   }
2678 }
2679 
2680 
Clear()2681 void QuickCheckDetails::Clear() {
2682   for (int i = 0; i < characters_; i++) {
2683     positions_[i].mask = 0;
2684     positions_[i].value = 0;
2685     positions_[i].determines_perfectly = false;
2686   }
2687   characters_ = 0;
2688 }
2689 
2690 
Advance(int by,bool one_byte)2691 void QuickCheckDetails::Advance(int by, bool one_byte) {
2692   DCHECK(by >= 0);
2693   if (by >= characters_) {
2694     Clear();
2695     return;
2696   }
2697   for (int i = 0; i < characters_ - by; i++) {
2698     positions_[i] = positions_[by + i];
2699   }
2700   for (int i = characters_ - by; i < characters_; i++) {
2701     positions_[i].mask = 0;
2702     positions_[i].value = 0;
2703     positions_[i].determines_perfectly = false;
2704   }
2705   characters_ -= by;
2706   // We could change mask_ and value_ here but we would never advance unless
2707   // they had already been used in a check and they won't be used again because
2708   // it would gain us nothing.  So there's no point.
2709 }
2710 
2711 
Merge(QuickCheckDetails * other,int from_index)2712 void QuickCheckDetails::Merge(QuickCheckDetails* other, int from_index) {
2713   DCHECK(characters_ == other->characters_);
2714   if (other->cannot_match_) {
2715     return;
2716   }
2717   if (cannot_match_) {
2718     *this = *other;
2719     return;
2720   }
2721   for (int i = from_index; i < characters_; i++) {
2722     QuickCheckDetails::Position* pos = positions(i);
2723     QuickCheckDetails::Position* other_pos = other->positions(i);
2724     if (pos->mask != other_pos->mask ||
2725         pos->value != other_pos->value ||
2726         !other_pos->determines_perfectly) {
2727       // Our mask-compare operation will be approximate unless we have the
2728       // exact same operation on both sides of the alternation.
2729       pos->determines_perfectly = false;
2730     }
2731     pos->mask &= other_pos->mask;
2732     pos->value &= pos->mask;
2733     other_pos->value &= pos->mask;
2734     uc16 differing_bits = (pos->value ^ other_pos->value);
2735     pos->mask &= ~differing_bits;
2736     pos->value &= pos->mask;
2737   }
2738 }
2739 
2740 
2741 class VisitMarker {
2742  public:
VisitMarker(NodeInfo * info)2743   explicit VisitMarker(NodeInfo* info) : info_(info) {
2744     DCHECK(!info->visited);
2745     info->visited = true;
2746   }
~VisitMarker()2747   ~VisitMarker() {
2748     info_->visited = false;
2749   }
2750  private:
2751   NodeInfo* info_;
2752 };
2753 
2754 
FilterOneByte(int depth,bool ignore_case)2755 RegExpNode* SeqRegExpNode::FilterOneByte(int depth, bool ignore_case) {
2756   if (info()->replacement_calculated) return replacement();
2757   if (depth < 0) return this;
2758   DCHECK(!info()->visited);
2759   VisitMarker marker(info());
2760   return FilterSuccessor(depth - 1, ignore_case);
2761 }
2762 
2763 
FilterSuccessor(int depth,bool ignore_case)2764 RegExpNode* SeqRegExpNode::FilterSuccessor(int depth, bool ignore_case) {
2765   RegExpNode* next = on_success_->FilterOneByte(depth - 1, ignore_case);
2766   if (next == NULL) return set_replacement(NULL);
2767   on_success_ = next;
2768   return set_replacement(this);
2769 }
2770 
2771 
2772 // We need to check for the following characters: 0x39c 0x3bc 0x178.
RangeContainsLatin1Equivalents(CharacterRange range)2773 static inline bool RangeContainsLatin1Equivalents(CharacterRange range) {
2774   // TODO(dcarney): this could be a lot more efficient.
2775   return range.Contains(0x39c) ||
2776       range.Contains(0x3bc) || range.Contains(0x178);
2777 }
2778 
2779 
RangesContainLatin1Equivalents(ZoneList<CharacterRange> * ranges)2780 static bool RangesContainLatin1Equivalents(ZoneList<CharacterRange>* ranges) {
2781   for (int i = 0; i < ranges->length(); i++) {
2782     // TODO(dcarney): this could be a lot more efficient.
2783     if (RangeContainsLatin1Equivalents(ranges->at(i))) return true;
2784   }
2785   return false;
2786 }
2787 
2788 
FilterOneByte(int depth,bool ignore_case)2789 RegExpNode* TextNode::FilterOneByte(int depth, bool ignore_case) {
2790   if (info()->replacement_calculated) return replacement();
2791   if (depth < 0) return this;
2792   DCHECK(!info()->visited);
2793   VisitMarker marker(info());
2794   int element_count = elms_->length();
2795   for (int i = 0; i < element_count; i++) {
2796     TextElement elm = elms_->at(i);
2797     if (elm.text_type() == TextElement::ATOM) {
2798       Vector<const uc16> quarks = elm.atom()->data();
2799       for (int j = 0; j < quarks.length(); j++) {
2800         uint16_t c = quarks[j];
2801         if (c <= String::kMaxOneByteCharCode) continue;
2802         if (!ignore_case) return set_replacement(NULL);
2803         // Here, we need to check for characters whose upper and lower cases
2804         // are outside the Latin-1 range.
2805         uint16_t converted = unibrow::Latin1::ConvertNonLatin1ToLatin1(c);
2806         // Character is outside Latin-1 completely
2807         if (converted == 0) return set_replacement(NULL);
2808         // Convert quark to Latin-1 in place.
2809         uint16_t* copy = const_cast<uint16_t*>(quarks.start());
2810         copy[j] = converted;
2811       }
2812     } else {
2813       DCHECK(elm.text_type() == TextElement::CHAR_CLASS);
2814       RegExpCharacterClass* cc = elm.char_class();
2815       ZoneList<CharacterRange>* ranges = cc->ranges(zone());
2816       if (!CharacterRange::IsCanonical(ranges)) {
2817         CharacterRange::Canonicalize(ranges);
2818       }
2819       // Now they are in order so we only need to look at the first.
2820       int range_count = ranges->length();
2821       if (cc->is_negated()) {
2822         if (range_count != 0 &&
2823             ranges->at(0).from() == 0 &&
2824             ranges->at(0).to() >= String::kMaxOneByteCharCode) {
2825           // This will be handled in a later filter.
2826           if (ignore_case && RangesContainLatin1Equivalents(ranges)) continue;
2827           return set_replacement(NULL);
2828         }
2829       } else {
2830         if (range_count == 0 ||
2831             ranges->at(0).from() > String::kMaxOneByteCharCode) {
2832           // This will be handled in a later filter.
2833           if (ignore_case && RangesContainLatin1Equivalents(ranges)) continue;
2834           return set_replacement(NULL);
2835         }
2836       }
2837     }
2838   }
2839   return FilterSuccessor(depth - 1, ignore_case);
2840 }
2841 
2842 
FilterOneByte(int depth,bool ignore_case)2843 RegExpNode* LoopChoiceNode::FilterOneByte(int depth, bool ignore_case) {
2844   if (info()->replacement_calculated) return replacement();
2845   if (depth < 0) return this;
2846   if (info()->visited) return this;
2847   {
2848     VisitMarker marker(info());
2849 
2850     RegExpNode* continue_replacement =
2851         continue_node_->FilterOneByte(depth - 1, ignore_case);
2852     // If we can't continue after the loop then there is no sense in doing the
2853     // loop.
2854     if (continue_replacement == NULL) return set_replacement(NULL);
2855   }
2856 
2857   return ChoiceNode::FilterOneByte(depth - 1, ignore_case);
2858 }
2859 
2860 
FilterOneByte(int depth,bool ignore_case)2861 RegExpNode* ChoiceNode::FilterOneByte(int depth, bool ignore_case) {
2862   if (info()->replacement_calculated) return replacement();
2863   if (depth < 0) return this;
2864   if (info()->visited) return this;
2865   VisitMarker marker(info());
2866   int choice_count = alternatives_->length();
2867 
2868   for (int i = 0; i < choice_count; i++) {
2869     GuardedAlternative alternative = alternatives_->at(i);
2870     if (alternative.guards() != NULL && alternative.guards()->length() != 0) {
2871       set_replacement(this);
2872       return this;
2873     }
2874   }
2875 
2876   int surviving = 0;
2877   RegExpNode* survivor = NULL;
2878   for (int i = 0; i < choice_count; i++) {
2879     GuardedAlternative alternative = alternatives_->at(i);
2880     RegExpNode* replacement =
2881         alternative.node()->FilterOneByte(depth - 1, ignore_case);
2882     DCHECK(replacement != this);  // No missing EMPTY_MATCH_CHECK.
2883     if (replacement != NULL) {
2884       alternatives_->at(i).set_node(replacement);
2885       surviving++;
2886       survivor = replacement;
2887     }
2888   }
2889   if (surviving < 2) return set_replacement(survivor);
2890 
2891   set_replacement(this);
2892   if (surviving == choice_count) {
2893     return this;
2894   }
2895   // Only some of the nodes survived the filtering.  We need to rebuild the
2896   // alternatives list.
2897   ZoneList<GuardedAlternative>* new_alternatives =
2898       new(zone()) ZoneList<GuardedAlternative>(surviving, zone());
2899   for (int i = 0; i < choice_count; i++) {
2900     RegExpNode* replacement =
2901         alternatives_->at(i).node()->FilterOneByte(depth - 1, ignore_case);
2902     if (replacement != NULL) {
2903       alternatives_->at(i).set_node(replacement);
2904       new_alternatives->Add(alternatives_->at(i), zone());
2905     }
2906   }
2907   alternatives_ = new_alternatives;
2908   return this;
2909 }
2910 
2911 
FilterOneByte(int depth,bool ignore_case)2912 RegExpNode* NegativeLookaheadChoiceNode::FilterOneByte(int depth,
2913                                                        bool ignore_case) {
2914   if (info()->replacement_calculated) return replacement();
2915   if (depth < 0) return this;
2916   if (info()->visited) return this;
2917   VisitMarker marker(info());
2918   // Alternative 0 is the negative lookahead, alternative 1 is what comes
2919   // afterwards.
2920   RegExpNode* node = alternatives_->at(1).node();
2921   RegExpNode* replacement = node->FilterOneByte(depth - 1, ignore_case);
2922   if (replacement == NULL) return set_replacement(NULL);
2923   alternatives_->at(1).set_node(replacement);
2924 
2925   RegExpNode* neg_node = alternatives_->at(0).node();
2926   RegExpNode* neg_replacement = neg_node->FilterOneByte(depth - 1, ignore_case);
2927   // If the negative lookahead is always going to fail then
2928   // we don't need to check it.
2929   if (neg_replacement == NULL) return set_replacement(replacement);
2930   alternatives_->at(0).set_node(neg_replacement);
2931   return set_replacement(this);
2932 }
2933 
2934 
GetQuickCheckDetails(QuickCheckDetails * details,RegExpCompiler * compiler,int characters_filled_in,bool not_at_start)2935 void LoopChoiceNode::GetQuickCheckDetails(QuickCheckDetails* details,
2936                                           RegExpCompiler* compiler,
2937                                           int characters_filled_in,
2938                                           bool not_at_start) {
2939   if (body_can_be_zero_length_ || info()->visited) return;
2940   VisitMarker marker(info());
2941   return ChoiceNode::GetQuickCheckDetails(details,
2942                                           compiler,
2943                                           characters_filled_in,
2944                                           not_at_start);
2945 }
2946 
2947 
FillInBMInfo(int offset,int budget,BoyerMooreLookahead * bm,bool not_at_start)2948 void LoopChoiceNode::FillInBMInfo(int offset,
2949                                   int budget,
2950                                   BoyerMooreLookahead* bm,
2951                                   bool not_at_start) {
2952   if (body_can_be_zero_length_ || budget <= 0) {
2953     bm->SetRest(offset);
2954     SaveBMInfo(bm, not_at_start, offset);
2955     return;
2956   }
2957   ChoiceNode::FillInBMInfo(offset, budget - 1, bm, not_at_start);
2958   SaveBMInfo(bm, not_at_start, offset);
2959 }
2960 
2961 
GetQuickCheckDetails(QuickCheckDetails * details,RegExpCompiler * compiler,int characters_filled_in,bool not_at_start)2962 void ChoiceNode::GetQuickCheckDetails(QuickCheckDetails* details,
2963                                       RegExpCompiler* compiler,
2964                                       int characters_filled_in,
2965                                       bool not_at_start) {
2966   not_at_start = (not_at_start || not_at_start_);
2967   int choice_count = alternatives_->length();
2968   DCHECK(choice_count > 0);
2969   alternatives_->at(0).node()->GetQuickCheckDetails(details,
2970                                                     compiler,
2971                                                     characters_filled_in,
2972                                                     not_at_start);
2973   for (int i = 1; i < choice_count; i++) {
2974     QuickCheckDetails new_details(details->characters());
2975     RegExpNode* node = alternatives_->at(i).node();
2976     node->GetQuickCheckDetails(&new_details, compiler,
2977                                characters_filled_in,
2978                                not_at_start);
2979     // Here we merge the quick match details of the two branches.
2980     details->Merge(&new_details, characters_filled_in);
2981   }
2982 }
2983 
2984 
2985 // Check for [0-9A-Z_a-z].
EmitWordCheck(RegExpMacroAssembler * assembler,Label * word,Label * non_word,bool fall_through_on_word)2986 static void EmitWordCheck(RegExpMacroAssembler* assembler,
2987                           Label* word,
2988                           Label* non_word,
2989                           bool fall_through_on_word) {
2990   if (assembler->CheckSpecialCharacterClass(
2991           fall_through_on_word ? 'w' : 'W',
2992           fall_through_on_word ? non_word : word)) {
2993     // Optimized implementation available.
2994     return;
2995   }
2996   assembler->CheckCharacterGT('z', non_word);
2997   assembler->CheckCharacterLT('0', non_word);
2998   assembler->CheckCharacterGT('a' - 1, word);
2999   assembler->CheckCharacterLT('9' + 1, word);
3000   assembler->CheckCharacterLT('A', non_word);
3001   assembler->CheckCharacterLT('Z' + 1, word);
3002   if (fall_through_on_word) {
3003     assembler->CheckNotCharacter('_', non_word);
3004   } else {
3005     assembler->CheckCharacter('_', word);
3006   }
3007 }
3008 
3009 
3010 // Emit the code to check for a ^ in multiline mode (1-character lookbehind
3011 // that matches newline or the start of input).
EmitHat(RegExpCompiler * compiler,RegExpNode * on_success,Trace * trace)3012 static void EmitHat(RegExpCompiler* compiler,
3013                     RegExpNode* on_success,
3014                     Trace* trace) {
3015   RegExpMacroAssembler* assembler = compiler->macro_assembler();
3016   // We will be loading the previous character into the current character
3017   // register.
3018   Trace new_trace(*trace);
3019   new_trace.InvalidateCurrentCharacter();
3020 
3021   Label ok;
3022   if (new_trace.cp_offset() == 0) {
3023     // The start of input counts as a newline in this context, so skip to
3024     // ok if we are at the start.
3025     assembler->CheckAtStart(&ok);
3026   }
3027   // We already checked that we are not at the start of input so it must be
3028   // OK to load the previous character.
3029   assembler->LoadCurrentCharacter(new_trace.cp_offset() -1,
3030                                   new_trace.backtrack(),
3031                                   false);
3032   if (!assembler->CheckSpecialCharacterClass('n',
3033                                              new_trace.backtrack())) {
3034     // Newline means \n, \r, 0x2028 or 0x2029.
3035     if (!compiler->one_byte()) {
3036       assembler->CheckCharacterAfterAnd(0x2028, 0xfffe, &ok);
3037     }
3038     assembler->CheckCharacter('\n', &ok);
3039     assembler->CheckNotCharacter('\r', new_trace.backtrack());
3040   }
3041   assembler->Bind(&ok);
3042   on_success->Emit(compiler, &new_trace);
3043 }
3044 
3045 
3046 // Emit the code to handle \b and \B (word-boundary or non-word-boundary).
EmitBoundaryCheck(RegExpCompiler * compiler,Trace * trace)3047 void AssertionNode::EmitBoundaryCheck(RegExpCompiler* compiler, Trace* trace) {
3048   RegExpMacroAssembler* assembler = compiler->macro_assembler();
3049   Trace::TriBool next_is_word_character = Trace::UNKNOWN;
3050   bool not_at_start = (trace->at_start() == Trace::FALSE_VALUE);
3051   BoyerMooreLookahead* lookahead = bm_info(not_at_start);
3052   if (lookahead == NULL) {
3053     int eats_at_least =
3054         Min(kMaxLookaheadForBoyerMoore, EatsAtLeast(kMaxLookaheadForBoyerMoore,
3055                                                     kRecursionBudget,
3056                                                     not_at_start));
3057     if (eats_at_least >= 1) {
3058       BoyerMooreLookahead* bm =
3059           new(zone()) BoyerMooreLookahead(eats_at_least, compiler, zone());
3060       FillInBMInfo(0, kRecursionBudget, bm, not_at_start);
3061       if (bm->at(0)->is_non_word())
3062         next_is_word_character = Trace::FALSE_VALUE;
3063       if (bm->at(0)->is_word()) next_is_word_character = Trace::TRUE_VALUE;
3064     }
3065   } else {
3066     if (lookahead->at(0)->is_non_word())
3067       next_is_word_character = Trace::FALSE_VALUE;
3068     if (lookahead->at(0)->is_word())
3069       next_is_word_character = Trace::TRUE_VALUE;
3070   }
3071   bool at_boundary = (assertion_type_ == AssertionNode::AT_BOUNDARY);
3072   if (next_is_word_character == Trace::UNKNOWN) {
3073     Label before_non_word;
3074     Label before_word;
3075     if (trace->characters_preloaded() != 1) {
3076       assembler->LoadCurrentCharacter(trace->cp_offset(), &before_non_word);
3077     }
3078     // Fall through on non-word.
3079     EmitWordCheck(assembler, &before_word, &before_non_word, false);
3080     // Next character is not a word character.
3081     assembler->Bind(&before_non_word);
3082     Label ok;
3083     BacktrackIfPrevious(compiler, trace, at_boundary ? kIsNonWord : kIsWord);
3084     assembler->GoTo(&ok);
3085 
3086     assembler->Bind(&before_word);
3087     BacktrackIfPrevious(compiler, trace, at_boundary ? kIsWord : kIsNonWord);
3088     assembler->Bind(&ok);
3089   } else if (next_is_word_character == Trace::TRUE_VALUE) {
3090     BacktrackIfPrevious(compiler, trace, at_boundary ? kIsWord : kIsNonWord);
3091   } else {
3092     DCHECK(next_is_word_character == Trace::FALSE_VALUE);
3093     BacktrackIfPrevious(compiler, trace, at_boundary ? kIsNonWord : kIsWord);
3094   }
3095 }
3096 
3097 
BacktrackIfPrevious(RegExpCompiler * compiler,Trace * trace,AssertionNode::IfPrevious backtrack_if_previous)3098 void AssertionNode::BacktrackIfPrevious(
3099     RegExpCompiler* compiler,
3100     Trace* trace,
3101     AssertionNode::IfPrevious backtrack_if_previous) {
3102   RegExpMacroAssembler* assembler = compiler->macro_assembler();
3103   Trace new_trace(*trace);
3104   new_trace.InvalidateCurrentCharacter();
3105 
3106   Label fall_through, dummy;
3107 
3108   Label* non_word = backtrack_if_previous == kIsNonWord ?
3109                     new_trace.backtrack() :
3110                     &fall_through;
3111   Label* word = backtrack_if_previous == kIsNonWord ?
3112                 &fall_through :
3113                 new_trace.backtrack();
3114 
3115   if (new_trace.cp_offset() == 0) {
3116     // The start of input counts as a non-word character, so the question is
3117     // decided if we are at the start.
3118     assembler->CheckAtStart(non_word);
3119   }
3120   // We already checked that we are not at the start of input so it must be
3121   // OK to load the previous character.
3122   assembler->LoadCurrentCharacter(new_trace.cp_offset() - 1, &dummy, false);
3123   EmitWordCheck(assembler, word, non_word, backtrack_if_previous == kIsNonWord);
3124 
3125   assembler->Bind(&fall_through);
3126   on_success()->Emit(compiler, &new_trace);
3127 }
3128 
3129 
GetQuickCheckDetails(QuickCheckDetails * details,RegExpCompiler * compiler,int filled_in,bool not_at_start)3130 void AssertionNode::GetQuickCheckDetails(QuickCheckDetails* details,
3131                                          RegExpCompiler* compiler,
3132                                          int filled_in,
3133                                          bool not_at_start) {
3134   if (assertion_type_ == AT_START && not_at_start) {
3135     details->set_cannot_match();
3136     return;
3137   }
3138   return on_success()->GetQuickCheckDetails(details,
3139                                             compiler,
3140                                             filled_in,
3141                                             not_at_start);
3142 }
3143 
3144 
Emit(RegExpCompiler * compiler,Trace * trace)3145 void AssertionNode::Emit(RegExpCompiler* compiler, Trace* trace) {
3146   RegExpMacroAssembler* assembler = compiler->macro_assembler();
3147   switch (assertion_type_) {
3148     case AT_END: {
3149       Label ok;
3150       assembler->CheckPosition(trace->cp_offset(), &ok);
3151       assembler->GoTo(trace->backtrack());
3152       assembler->Bind(&ok);
3153       break;
3154     }
3155     case AT_START: {
3156       if (trace->at_start() == Trace::FALSE_VALUE) {
3157         assembler->GoTo(trace->backtrack());
3158         return;
3159       }
3160       if (trace->at_start() == Trace::UNKNOWN) {
3161         assembler->CheckNotAtStart(trace->backtrack());
3162         Trace at_start_trace = *trace;
3163         at_start_trace.set_at_start(true);
3164         on_success()->Emit(compiler, &at_start_trace);
3165         return;
3166       }
3167     }
3168     break;
3169     case AFTER_NEWLINE:
3170       EmitHat(compiler, on_success(), trace);
3171       return;
3172     case AT_BOUNDARY:
3173     case AT_NON_BOUNDARY: {
3174       EmitBoundaryCheck(compiler, trace);
3175       return;
3176     }
3177   }
3178   on_success()->Emit(compiler, trace);
3179 }
3180 
3181 
DeterminedAlready(QuickCheckDetails * quick_check,int offset)3182 static bool DeterminedAlready(QuickCheckDetails* quick_check, int offset) {
3183   if (quick_check == NULL) return false;
3184   if (offset >= quick_check->characters()) return false;
3185   return quick_check->positions(offset)->determines_perfectly;
3186 }
3187 
3188 
UpdateBoundsCheck(int index,int * checked_up_to)3189 static void UpdateBoundsCheck(int index, int* checked_up_to) {
3190   if (index > *checked_up_to) {
3191     *checked_up_to = index;
3192   }
3193 }
3194 
3195 
3196 // We call this repeatedly to generate code for each pass over the text node.
3197 // The passes are in increasing order of difficulty because we hope one
3198 // of the first passes will fail in which case we are saved the work of the
3199 // later passes.  for example for the case independent regexp /%[asdfghjkl]a/
3200 // we will check the '%' in the first pass, the case independent 'a' in the
3201 // second pass and the character class in the last pass.
3202 //
3203 // The passes are done from right to left, so for example to test for /bar/
3204 // we will first test for an 'r' with offset 2, then an 'a' with offset 1
3205 // and then a 'b' with offset 0.  This means we can avoid the end-of-input
3206 // bounds check most of the time.  In the example we only need to check for
3207 // end-of-input when loading the putative 'r'.
3208 //
3209 // A slight complication involves the fact that the first character may already
3210 // be fetched into a register by the previous node.  In this case we want to
3211 // do the test for that character first.  We do this in separate passes.  The
3212 // 'preloaded' argument indicates that we are doing such a 'pass'.  If such a
3213 // pass has been performed then subsequent passes will have true in
3214 // first_element_checked to indicate that that character does not need to be
3215 // checked again.
3216 //
3217 // In addition to all this we are passed a Trace, which can
3218 // contain an AlternativeGeneration object.  In this AlternativeGeneration
3219 // object we can see details of any quick check that was already passed in
3220 // order to get to the code we are now generating.  The quick check can involve
3221 // loading characters, which means we do not need to recheck the bounds
3222 // up to the limit the quick check already checked.  In addition the quick
3223 // check can have involved a mask and compare operation which may simplify
3224 // 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)3225 void TextNode::TextEmitPass(RegExpCompiler* compiler,
3226                             TextEmitPassType pass,
3227                             bool preloaded,
3228                             Trace* trace,
3229                             bool first_element_checked,
3230                             int* checked_up_to) {
3231   RegExpMacroAssembler* assembler = compiler->macro_assembler();
3232   Isolate* isolate = assembler->zone()->isolate();
3233   bool one_byte = compiler->one_byte();
3234   Label* backtrack = trace->backtrack();
3235   QuickCheckDetails* quick_check = trace->quick_check_performed();
3236   int element_count = elms_->length();
3237   for (int i = preloaded ? 0 : element_count - 1; i >= 0; i--) {
3238     TextElement elm = elms_->at(i);
3239     int cp_offset = trace->cp_offset() + elm.cp_offset();
3240     if (elm.text_type() == TextElement::ATOM) {
3241       Vector<const uc16> quarks = elm.atom()->data();
3242       for (int j = preloaded ? 0 : quarks.length() - 1; j >= 0; j--) {
3243         if (first_element_checked && i == 0 && j == 0) continue;
3244         if (DeterminedAlready(quick_check, elm.cp_offset() + j)) continue;
3245         EmitCharacterFunction* emit_function = NULL;
3246         switch (pass) {
3247           case NON_LATIN1_MATCH:
3248             DCHECK(one_byte);
3249             if (quarks[j] > String::kMaxOneByteCharCode) {
3250               assembler->GoTo(backtrack);
3251               return;
3252             }
3253             break;
3254           case NON_LETTER_CHARACTER_MATCH:
3255             emit_function = &EmitAtomNonLetter;
3256             break;
3257           case SIMPLE_CHARACTER_MATCH:
3258             emit_function = &EmitSimpleCharacter;
3259             break;
3260           case CASE_CHARACTER_MATCH:
3261             emit_function = &EmitAtomLetter;
3262             break;
3263           default:
3264             break;
3265         }
3266         if (emit_function != NULL) {
3267           bool bound_checked = emit_function(isolate,
3268                                              compiler,
3269                                              quarks[j],
3270                                              backtrack,
3271                                              cp_offset + j,
3272                                              *checked_up_to < cp_offset + j,
3273                                              preloaded);
3274           if (bound_checked) UpdateBoundsCheck(cp_offset + j, checked_up_to);
3275         }
3276       }
3277     } else {
3278       DCHECK_EQ(TextElement::CHAR_CLASS, elm.text_type());
3279       if (pass == CHARACTER_CLASS_MATCH) {
3280         if (first_element_checked && i == 0) continue;
3281         if (DeterminedAlready(quick_check, elm.cp_offset())) continue;
3282         RegExpCharacterClass* cc = elm.char_class();
3283         EmitCharClass(assembler, cc, one_byte, backtrack, cp_offset,
3284                       *checked_up_to < cp_offset, preloaded, zone());
3285         UpdateBoundsCheck(cp_offset, checked_up_to);
3286       }
3287     }
3288   }
3289 }
3290 
3291 
Length()3292 int TextNode::Length() {
3293   TextElement elm = elms_->last();
3294   DCHECK(elm.cp_offset() >= 0);
3295   return elm.cp_offset() + elm.length();
3296 }
3297 
3298 
SkipPass(int int_pass,bool ignore_case)3299 bool TextNode::SkipPass(int int_pass, bool ignore_case) {
3300   TextEmitPassType pass = static_cast<TextEmitPassType>(int_pass);
3301   if (ignore_case) {
3302     return pass == SIMPLE_CHARACTER_MATCH;
3303   } else {
3304     return pass == NON_LETTER_CHARACTER_MATCH || pass == CASE_CHARACTER_MATCH;
3305   }
3306 }
3307 
3308 
3309 // This generates the code to match a text node.  A text node can contain
3310 // straight character sequences (possibly to be matched in a case-independent
3311 // way) and character classes.  For efficiency we do not do this in a single
3312 // pass from left to right.  Instead we pass over the text node several times,
3313 // emitting code for some character positions every time.  See the comment on
3314 // TextEmitPass for details.
Emit(RegExpCompiler * compiler,Trace * trace)3315 void TextNode::Emit(RegExpCompiler* compiler, Trace* trace) {
3316   LimitResult limit_result = LimitVersions(compiler, trace);
3317   if (limit_result == DONE) return;
3318   DCHECK(limit_result == CONTINUE);
3319 
3320   if (trace->cp_offset() + Length() > RegExpMacroAssembler::kMaxCPOffset) {
3321     compiler->SetRegExpTooBig();
3322     return;
3323   }
3324 
3325   if (compiler->one_byte()) {
3326     int dummy = 0;
3327     TextEmitPass(compiler, NON_LATIN1_MATCH, false, trace, false, &dummy);
3328   }
3329 
3330   bool first_elt_done = false;
3331   int bound_checked_to = trace->cp_offset() - 1;
3332   bound_checked_to += trace->bound_checked_up_to();
3333 
3334   // If a character is preloaded into the current character register then
3335   // check that now.
3336   if (trace->characters_preloaded() == 1) {
3337     for (int pass = kFirstRealPass; pass <= kLastPass; pass++) {
3338       if (!SkipPass(pass, compiler->ignore_case())) {
3339         TextEmitPass(compiler,
3340                      static_cast<TextEmitPassType>(pass),
3341                      true,
3342                      trace,
3343                      false,
3344                      &bound_checked_to);
3345       }
3346     }
3347     first_elt_done = true;
3348   }
3349 
3350   for (int pass = kFirstRealPass; pass <= kLastPass; pass++) {
3351     if (!SkipPass(pass, compiler->ignore_case())) {
3352       TextEmitPass(compiler,
3353                    static_cast<TextEmitPassType>(pass),
3354                    false,
3355                    trace,
3356                    first_elt_done,
3357                    &bound_checked_to);
3358     }
3359   }
3360 
3361   Trace successor_trace(*trace);
3362   successor_trace.set_at_start(false);
3363   successor_trace.AdvanceCurrentPositionInTrace(Length(), compiler);
3364   RecursionCheck rc(compiler);
3365   on_success()->Emit(compiler, &successor_trace);
3366 }
3367 
3368 
InvalidateCurrentCharacter()3369 void Trace::InvalidateCurrentCharacter() {
3370   characters_preloaded_ = 0;
3371 }
3372 
3373 
AdvanceCurrentPositionInTrace(int by,RegExpCompiler * compiler)3374 void Trace::AdvanceCurrentPositionInTrace(int by, RegExpCompiler* compiler) {
3375   DCHECK(by > 0);
3376   // We don't have an instruction for shifting the current character register
3377   // down or for using a shifted value for anything so lets just forget that
3378   // we preloaded any characters into it.
3379   characters_preloaded_ = 0;
3380   // Adjust the offsets of the quick check performed information.  This
3381   // information is used to find out what we already determined about the
3382   // characters by means of mask and compare.
3383   quick_check_performed_.Advance(by, compiler->one_byte());
3384   cp_offset_ += by;
3385   if (cp_offset_ > RegExpMacroAssembler::kMaxCPOffset) {
3386     compiler->SetRegExpTooBig();
3387     cp_offset_ = 0;
3388   }
3389   bound_checked_up_to_ = Max(0, bound_checked_up_to_ - by);
3390 }
3391 
3392 
MakeCaseIndependent(bool is_one_byte)3393 void TextNode::MakeCaseIndependent(bool is_one_byte) {
3394   int element_count = elms_->length();
3395   for (int i = 0; i < element_count; i++) {
3396     TextElement elm = elms_->at(i);
3397     if (elm.text_type() == TextElement::CHAR_CLASS) {
3398       RegExpCharacterClass* cc = elm.char_class();
3399       // None of the standard character classes is different in the case
3400       // independent case and it slows us down if we don't know that.
3401       if (cc->is_standard(zone())) continue;
3402       ZoneList<CharacterRange>* ranges = cc->ranges(zone());
3403       int range_count = ranges->length();
3404       for (int j = 0; j < range_count; j++) {
3405         ranges->at(j).AddCaseEquivalents(ranges, is_one_byte, zone());
3406       }
3407     }
3408   }
3409 }
3410 
3411 
GreedyLoopTextLength()3412 int TextNode::GreedyLoopTextLength() {
3413   TextElement elm = elms_->at(elms_->length() - 1);
3414   return elm.cp_offset() + elm.length();
3415 }
3416 
3417 
GetSuccessorOfOmnivorousTextNode(RegExpCompiler * compiler)3418 RegExpNode* TextNode::GetSuccessorOfOmnivorousTextNode(
3419     RegExpCompiler* compiler) {
3420   if (elms_->length() != 1) return NULL;
3421   TextElement elm = elms_->at(0);
3422   if (elm.text_type() != TextElement::CHAR_CLASS) return NULL;
3423   RegExpCharacterClass* node = elm.char_class();
3424   ZoneList<CharacterRange>* ranges = node->ranges(zone());
3425   if (!CharacterRange::IsCanonical(ranges)) {
3426     CharacterRange::Canonicalize(ranges);
3427   }
3428   if (node->is_negated()) {
3429     return ranges->length() == 0 ? on_success() : NULL;
3430   }
3431   if (ranges->length() != 1) return NULL;
3432   uint32_t max_char;
3433   if (compiler->one_byte()) {
3434     max_char = String::kMaxOneByteCharCode;
3435   } else {
3436     max_char = String::kMaxUtf16CodeUnit;
3437   }
3438   return ranges->at(0).IsEverything(max_char) ? on_success() : NULL;
3439 }
3440 
3441 
3442 // Finds the fixed match length of a sequence of nodes that goes from
3443 // this alternative and back to this choice node.  If there are variable
3444 // length nodes or other complications in the way then return a sentinel
3445 // value indicating that a greedy loop cannot be constructed.
GreedyLoopTextLengthForAlternative(GuardedAlternative * alternative)3446 int ChoiceNode::GreedyLoopTextLengthForAlternative(
3447     GuardedAlternative* alternative) {
3448   int length = 0;
3449   RegExpNode* node = alternative->node();
3450   // Later we will generate code for all these text nodes using recursion
3451   // so we have to limit the max number.
3452   int recursion_depth = 0;
3453   while (node != this) {
3454     if (recursion_depth++ > RegExpCompiler::kMaxRecursion) {
3455       return kNodeIsTooComplexForGreedyLoops;
3456     }
3457     int node_length = node->GreedyLoopTextLength();
3458     if (node_length == kNodeIsTooComplexForGreedyLoops) {
3459       return kNodeIsTooComplexForGreedyLoops;
3460     }
3461     length += node_length;
3462     SeqRegExpNode* seq_node = static_cast<SeqRegExpNode*>(node);
3463     node = seq_node->on_success();
3464   }
3465   return length;
3466 }
3467 
3468 
AddLoopAlternative(GuardedAlternative alt)3469 void LoopChoiceNode::AddLoopAlternative(GuardedAlternative alt) {
3470   DCHECK_EQ(loop_node_, NULL);
3471   AddAlternative(alt);
3472   loop_node_ = alt.node();
3473 }
3474 
3475 
AddContinueAlternative(GuardedAlternative alt)3476 void LoopChoiceNode::AddContinueAlternative(GuardedAlternative alt) {
3477   DCHECK_EQ(continue_node_, NULL);
3478   AddAlternative(alt);
3479   continue_node_ = alt.node();
3480 }
3481 
3482 
Emit(RegExpCompiler * compiler,Trace * trace)3483 void LoopChoiceNode::Emit(RegExpCompiler* compiler, Trace* trace) {
3484   RegExpMacroAssembler* macro_assembler = compiler->macro_assembler();
3485   if (trace->stop_node() == this) {
3486     // Back edge of greedy optimized loop node graph.
3487     int text_length =
3488         GreedyLoopTextLengthForAlternative(&(alternatives_->at(0)));
3489     DCHECK(text_length != kNodeIsTooComplexForGreedyLoops);
3490     // Update the counter-based backtracking info on the stack.  This is an
3491     // optimization for greedy loops (see below).
3492     DCHECK(trace->cp_offset() == text_length);
3493     macro_assembler->AdvanceCurrentPosition(text_length);
3494     macro_assembler->GoTo(trace->loop_label());
3495     return;
3496   }
3497   DCHECK(trace->stop_node() == NULL);
3498   if (!trace->is_trivial()) {
3499     trace->Flush(compiler, this);
3500     return;
3501   }
3502   ChoiceNode::Emit(compiler, trace);
3503 }
3504 
3505 
CalculatePreloadCharacters(RegExpCompiler * compiler,int eats_at_least)3506 int ChoiceNode::CalculatePreloadCharacters(RegExpCompiler* compiler,
3507                                            int eats_at_least) {
3508   int preload_characters = Min(4, eats_at_least);
3509   if (compiler->macro_assembler()->CanReadUnaligned()) {
3510     bool one_byte = compiler->one_byte();
3511     if (one_byte) {
3512       if (preload_characters > 4) preload_characters = 4;
3513       // We can't preload 3 characters because there is no machine instruction
3514       // to do that.  We can't just load 4 because we could be reading
3515       // beyond the end of the string, which could cause a memory fault.
3516       if (preload_characters == 3) preload_characters = 2;
3517     } else {
3518       if (preload_characters > 2) preload_characters = 2;
3519     }
3520   } else {
3521     if (preload_characters > 1) preload_characters = 1;
3522   }
3523   return preload_characters;
3524 }
3525 
3526 
3527 // This class is used when generating the alternatives in a choice node.  It
3528 // records the way the alternative is being code generated.
3529 class AlternativeGeneration: public Malloced {
3530  public:
AlternativeGeneration()3531   AlternativeGeneration()
3532       : possible_success(),
3533         expects_preload(false),
3534         after(),
3535         quick_check_details() { }
3536   Label possible_success;
3537   bool expects_preload;
3538   Label after;
3539   QuickCheckDetails quick_check_details;
3540 };
3541 
3542 
3543 // Creates a list of AlternativeGenerations.  If the list has a reasonable
3544 // size then it is on the stack, otherwise the excess is on the heap.
3545 class AlternativeGenerationList {
3546  public:
AlternativeGenerationList(int count,Zone * zone)3547   AlternativeGenerationList(int count, Zone* zone)
3548       : alt_gens_(count, zone) {
3549     for (int i = 0; i < count && i < kAFew; i++) {
3550       alt_gens_.Add(a_few_alt_gens_ + i, zone);
3551     }
3552     for (int i = kAFew; i < count; i++) {
3553       alt_gens_.Add(new AlternativeGeneration(), zone);
3554     }
3555   }
~AlternativeGenerationList()3556   ~AlternativeGenerationList() {
3557     for (int i = kAFew; i < alt_gens_.length(); i++) {
3558       delete alt_gens_[i];
3559       alt_gens_[i] = NULL;
3560     }
3561   }
3562 
at(int i)3563   AlternativeGeneration* at(int i) {
3564     return alt_gens_[i];
3565   }
3566 
3567  private:
3568   static const int kAFew = 10;
3569   ZoneList<AlternativeGeneration*> alt_gens_;
3570   AlternativeGeneration a_few_alt_gens_[kAFew];
3571 };
3572 
3573 
3574 // The '2' variant is has inclusive from and exclusive to.
3575 // This covers \s as defined in ECMA-262 5.1, 15.10.2.12,
3576 // which include WhiteSpace (7.2) or LineTerminator (7.3) values.
3577 static const int kSpaceRanges[] = { '\t', '\r' + 1, ' ', ' ' + 1,
3578     0x00A0, 0x00A1, 0x1680, 0x1681, 0x180E, 0x180F, 0x2000, 0x200B,
3579     0x2028, 0x202A, 0x202F, 0x2030, 0x205F, 0x2060, 0x3000, 0x3001,
3580     0xFEFF, 0xFF00, 0x10000 };
3581 static const int kSpaceRangeCount = arraysize(kSpaceRanges);
3582 
3583 static const int kWordRanges[] = {
3584     '0', '9' + 1, 'A', 'Z' + 1, '_', '_' + 1, 'a', 'z' + 1, 0x10000 };
3585 static const int kWordRangeCount = arraysize(kWordRanges);
3586 static const int kDigitRanges[] = { '0', '9' + 1, 0x10000 };
3587 static const int kDigitRangeCount = arraysize(kDigitRanges);
3588 static const int kSurrogateRanges[] = { 0xd800, 0xe000, 0x10000 };
3589 static const int kSurrogateRangeCount = arraysize(kSurrogateRanges);
3590 static const int kLineTerminatorRanges[] = { 0x000A, 0x000B, 0x000D, 0x000E,
3591     0x2028, 0x202A, 0x10000 };
3592 static const int kLineTerminatorRangeCount = arraysize(kLineTerminatorRanges);
3593 
3594 
Set(int character)3595 void BoyerMoorePositionInfo::Set(int character) {
3596   SetInterval(Interval(character, character));
3597 }
3598 
3599 
SetInterval(const Interval & interval)3600 void BoyerMoorePositionInfo::SetInterval(const Interval& interval) {
3601   s_ = AddRange(s_, kSpaceRanges, kSpaceRangeCount, interval);
3602   w_ = AddRange(w_, kWordRanges, kWordRangeCount, interval);
3603   d_ = AddRange(d_, kDigitRanges, kDigitRangeCount, interval);
3604   surrogate_ =
3605       AddRange(surrogate_, kSurrogateRanges, kSurrogateRangeCount, interval);
3606   if (interval.to() - interval.from() >= kMapSize - 1) {
3607     if (map_count_ != kMapSize) {
3608       map_count_ = kMapSize;
3609       for (int i = 0; i < kMapSize; i++) map_->at(i) = true;
3610     }
3611     return;
3612   }
3613   for (int i = interval.from(); i <= interval.to(); i++) {
3614     int mod_character = (i & kMask);
3615     if (!map_->at(mod_character)) {
3616       map_count_++;
3617       map_->at(mod_character) = true;
3618     }
3619     if (map_count_ == kMapSize) return;
3620   }
3621 }
3622 
3623 
SetAll()3624 void BoyerMoorePositionInfo::SetAll() {
3625   s_ = w_ = d_ = kLatticeUnknown;
3626   if (map_count_ != kMapSize) {
3627     map_count_ = kMapSize;
3628     for (int i = 0; i < kMapSize; i++) map_->at(i) = true;
3629   }
3630 }
3631 
3632 
BoyerMooreLookahead(int length,RegExpCompiler * compiler,Zone * zone)3633 BoyerMooreLookahead::BoyerMooreLookahead(
3634     int length, RegExpCompiler* compiler, Zone* zone)
3635     : length_(length),
3636       compiler_(compiler) {
3637   if (compiler->one_byte()) {
3638     max_char_ = String::kMaxOneByteCharCode;
3639   } else {
3640     max_char_ = String::kMaxUtf16CodeUnit;
3641   }
3642   bitmaps_ = new(zone) ZoneList<BoyerMoorePositionInfo*>(length, zone);
3643   for (int i = 0; i < length; i++) {
3644     bitmaps_->Add(new(zone) BoyerMoorePositionInfo(zone), zone);
3645   }
3646 }
3647 
3648 
3649 // Find the longest range of lookahead that has the fewest number of different
3650 // characters that can occur at a given position.  Since we are optimizing two
3651 // different parameters at once this is a tradeoff.
FindWorthwhileInterval(int * from,int * to)3652 bool BoyerMooreLookahead::FindWorthwhileInterval(int* from, int* to) {
3653   int biggest_points = 0;
3654   // If more than 32 characters out of 128 can occur it is unlikely that we can
3655   // be lucky enough to step forwards much of the time.
3656   const int kMaxMax = 32;
3657   for (int max_number_of_chars = 4;
3658        max_number_of_chars < kMaxMax;
3659        max_number_of_chars *= 2) {
3660     biggest_points =
3661         FindBestInterval(max_number_of_chars, biggest_points, from, to);
3662   }
3663   if (biggest_points == 0) return false;
3664   return true;
3665 }
3666 
3667 
3668 // Find the highest-points range between 0 and length_ where the character
3669 // information is not too vague.  'Too vague' means that there are more than
3670 // max_number_of_chars that can occur at this position.  Calculates the number
3671 // of points as the product of width-of-the-range and
3672 // probability-of-finding-one-of-the-characters, where the probability is
3673 // calculated using the frequency distribution of the sample subject string.
FindBestInterval(int max_number_of_chars,int old_biggest_points,int * from,int * to)3674 int BoyerMooreLookahead::FindBestInterval(
3675     int max_number_of_chars, int old_biggest_points, int* from, int* to) {
3676   int biggest_points = old_biggest_points;
3677   static const int kSize = RegExpMacroAssembler::kTableSize;
3678   for (int i = 0; i < length_; ) {
3679     while (i < length_ && Count(i) > max_number_of_chars) i++;
3680     if (i == length_) break;
3681     int remembered_from = i;
3682     bool union_map[kSize];
3683     for (int j = 0; j < kSize; j++) union_map[j] = false;
3684     while (i < length_ && Count(i) <= max_number_of_chars) {
3685       BoyerMoorePositionInfo* map = bitmaps_->at(i);
3686       for (int j = 0; j < kSize; j++) union_map[j] |= map->at(j);
3687       i++;
3688     }
3689     int frequency = 0;
3690     for (int j = 0; j < kSize; j++) {
3691       if (union_map[j]) {
3692         // Add 1 to the frequency to give a small per-character boost for
3693         // the cases where our sampling is not good enough and many
3694         // characters have a frequency of zero.  This means the frequency
3695         // can theoretically be up to 2*kSize though we treat it mostly as
3696         // a fraction of kSize.
3697         frequency += compiler_->frequency_collator()->Frequency(j) + 1;
3698       }
3699     }
3700     // We use the probability of skipping times the distance we are skipping to
3701     // judge the effectiveness of this.  Actually we have a cut-off:  By
3702     // dividing by 2 we switch off the skipping if the probability of skipping
3703     // is less than 50%.  This is because the multibyte mask-and-compare
3704     // skipping in quickcheck is more likely to do well on this case.
3705     bool in_quickcheck_range =
3706         ((i - remembered_from < 4) ||
3707          (compiler_->one_byte() ? remembered_from <= 4 : remembered_from <= 2));
3708     // Called 'probability' but it is only a rough estimate and can actually
3709     // be outside the 0-kSize range.
3710     int probability = (in_quickcheck_range ? kSize / 2 : kSize) - frequency;
3711     int points = (i - remembered_from) * probability;
3712     if (points > biggest_points) {
3713       *from = remembered_from;
3714       *to = i - 1;
3715       biggest_points = points;
3716     }
3717   }
3718   return biggest_points;
3719 }
3720 
3721 
3722 // Take all the characters that will not prevent a successful match if they
3723 // occur in the subject string in the range between min_lookahead and
3724 // max_lookahead (inclusive) measured from the current position.  If the
3725 // character at max_lookahead offset is not one of these characters, then we
3726 // can safely skip forwards by the number of characters in the range.
GetSkipTable(int min_lookahead,int max_lookahead,Handle<ByteArray> boolean_skip_table)3727 int BoyerMooreLookahead::GetSkipTable(int min_lookahead,
3728                                       int max_lookahead,
3729                                       Handle<ByteArray> boolean_skip_table) {
3730   const int kSize = RegExpMacroAssembler::kTableSize;
3731 
3732   const int kSkipArrayEntry = 0;
3733   const int kDontSkipArrayEntry = 1;
3734 
3735   for (int i = 0; i < kSize; i++) {
3736     boolean_skip_table->set(i, kSkipArrayEntry);
3737   }
3738   int skip = max_lookahead + 1 - min_lookahead;
3739 
3740   for (int i = max_lookahead; i >= min_lookahead; i--) {
3741     BoyerMoorePositionInfo* map = bitmaps_->at(i);
3742     for (int j = 0; j < kSize; j++) {
3743       if (map->at(j)) {
3744         boolean_skip_table->set(j, kDontSkipArrayEntry);
3745       }
3746     }
3747   }
3748 
3749   return skip;
3750 }
3751 
3752 
3753 // See comment above on the implementation of GetSkipTable.
EmitSkipInstructions(RegExpMacroAssembler * masm)3754 void BoyerMooreLookahead::EmitSkipInstructions(RegExpMacroAssembler* masm) {
3755   const int kSize = RegExpMacroAssembler::kTableSize;
3756 
3757   int min_lookahead = 0;
3758   int max_lookahead = 0;
3759 
3760   if (!FindWorthwhileInterval(&min_lookahead, &max_lookahead)) return;
3761 
3762   bool found_single_character = false;
3763   int single_character = 0;
3764   for (int i = max_lookahead; i >= min_lookahead; i--) {
3765     BoyerMoorePositionInfo* map = bitmaps_->at(i);
3766     if (map->map_count() > 1 ||
3767         (found_single_character && map->map_count() != 0)) {
3768       found_single_character = false;
3769       break;
3770     }
3771     for (int j = 0; j < kSize; j++) {
3772       if (map->at(j)) {
3773         found_single_character = true;
3774         single_character = j;
3775         break;
3776       }
3777     }
3778   }
3779 
3780   int lookahead_width = max_lookahead + 1 - min_lookahead;
3781 
3782   if (found_single_character && lookahead_width == 1 && max_lookahead < 3) {
3783     // The mask-compare can probably handle this better.
3784     return;
3785   }
3786 
3787   if (found_single_character) {
3788     Label cont, again;
3789     masm->Bind(&again);
3790     masm->LoadCurrentCharacter(max_lookahead, &cont, true);
3791     if (max_char_ > kSize) {
3792       masm->CheckCharacterAfterAnd(single_character,
3793                                    RegExpMacroAssembler::kTableMask,
3794                                    &cont);
3795     } else {
3796       masm->CheckCharacter(single_character, &cont);
3797     }
3798     masm->AdvanceCurrentPosition(lookahead_width);
3799     masm->GoTo(&again);
3800     masm->Bind(&cont);
3801     return;
3802   }
3803 
3804   Factory* factory = masm->zone()->isolate()->factory();
3805   Handle<ByteArray> boolean_skip_table = factory->NewByteArray(kSize, TENURED);
3806   int skip_distance = GetSkipTable(
3807       min_lookahead, max_lookahead, boolean_skip_table);
3808   DCHECK(skip_distance != 0);
3809 
3810   Label cont, again;
3811   masm->Bind(&again);
3812   masm->LoadCurrentCharacter(max_lookahead, &cont, true);
3813   masm->CheckBitInTable(boolean_skip_table, &cont);
3814   masm->AdvanceCurrentPosition(skip_distance);
3815   masm->GoTo(&again);
3816   masm->Bind(&cont);
3817 }
3818 
3819 
3820 /* Code generation for choice nodes.
3821  *
3822  * We generate quick checks that do a mask and compare to eliminate a
3823  * choice.  If the quick check succeeds then it jumps to the continuation to
3824  * do slow checks and check subsequent nodes.  If it fails (the common case)
3825  * it falls through to the next choice.
3826  *
3827  * Here is the desired flow graph.  Nodes directly below each other imply
3828  * fallthrough.  Alternatives 1 and 2 have quick checks.  Alternative
3829  * 3 doesn't have a quick check so we have to call the slow check.
3830  * Nodes are marked Qn for quick checks and Sn for slow checks.  The entire
3831  * regexp continuation is generated directly after the Sn node, up to the
3832  * next GoTo if we decide to reuse some already generated code.  Some
3833  * nodes expect preload_characters to be preloaded into the current
3834  * character register.  R nodes do this preloading.  Vertices are marked
3835  * F for failures and S for success (possible success in the case of quick
3836  * nodes).  L, V, < and > are used as arrow heads.
3837  *
3838  * ----------> R
3839  *             |
3840  *             V
3841  *            Q1 -----> S1
3842  *             |   S   /
3843  *            F|      /
3844  *             |    F/
3845  *             |    /
3846  *             |   R
3847  *             |  /
3848  *             V L
3849  *            Q2 -----> S2
3850  *             |   S   /
3851  *            F|      /
3852  *             |    F/
3853  *             |    /
3854  *             |   R
3855  *             |  /
3856  *             V L
3857  *            S3
3858  *             |
3859  *            F|
3860  *             |
3861  *             R
3862  *             |
3863  * backtrack   V
3864  * <----------Q4
3865  *   \    F    |
3866  *    \        |S
3867  *     \   F   V
3868  *      \-----S4
3869  *
3870  * For greedy loops we push the current position, then generate the code that
3871  * eats the input specially in EmitGreedyLoop.  The other choice (the
3872  * continuation) is generated by the normal code in EmitChoices, and steps back
3873  * in the input to the starting position when it fails to match.  The loop code
3874  * looks like this (U is the unwind code that steps back in the greedy loop).
3875  *
3876  *              _____
3877  *             /     \
3878  *             V     |
3879  * ----------> S1    |
3880  *            /|     |
3881  *           / |S    |
3882  *         F/  \_____/
3883  *         /
3884  *        |<-----
3885  *        |      \
3886  *        V       |S
3887  *        Q2 ---> U----->backtrack
3888  *        |  F   /
3889  *       S|     /
3890  *        V  F /
3891  *        S2--/
3892  */
3893 
GreedyLoopState(bool not_at_start)3894 GreedyLoopState::GreedyLoopState(bool not_at_start) {
3895   counter_backtrack_trace_.set_backtrack(&label_);
3896   if (not_at_start) counter_backtrack_trace_.set_at_start(false);
3897 }
3898 
3899 
AssertGuardsMentionRegisters(Trace * trace)3900 void ChoiceNode::AssertGuardsMentionRegisters(Trace* trace) {
3901 #ifdef DEBUG
3902   int choice_count = alternatives_->length();
3903   for (int i = 0; i < choice_count - 1; i++) {
3904     GuardedAlternative alternative = alternatives_->at(i);
3905     ZoneList<Guard*>* guards = alternative.guards();
3906     int guard_count = (guards == NULL) ? 0 : guards->length();
3907     for (int j = 0; j < guard_count; j++) {
3908       DCHECK(!trace->mentions_reg(guards->at(j)->reg()));
3909     }
3910   }
3911 #endif
3912 }
3913 
3914 
SetUpPreLoad(RegExpCompiler * compiler,Trace * current_trace,PreloadState * state)3915 void ChoiceNode::SetUpPreLoad(RegExpCompiler* compiler,
3916                               Trace* current_trace,
3917                               PreloadState* state) {
3918     if (state->eats_at_least_ == PreloadState::kEatsAtLeastNotYetInitialized) {
3919       // Save some time by looking at most one machine word ahead.
3920       state->eats_at_least_ =
3921           EatsAtLeast(compiler->one_byte() ? 4 : 2, kRecursionBudget,
3922                       current_trace->at_start() == Trace::FALSE_VALUE);
3923     }
3924     state->preload_characters_ =
3925         CalculatePreloadCharacters(compiler, state->eats_at_least_);
3926 
3927     state->preload_is_current_ =
3928         (current_trace->characters_preloaded() == state->preload_characters_);
3929     state->preload_has_checked_bounds_ = state->preload_is_current_;
3930 }
3931 
3932 
Emit(RegExpCompiler * compiler,Trace * trace)3933 void ChoiceNode::Emit(RegExpCompiler* compiler, Trace* trace) {
3934   int choice_count = alternatives_->length();
3935 
3936   AssertGuardsMentionRegisters(trace);
3937 
3938   LimitResult limit_result = LimitVersions(compiler, trace);
3939   if (limit_result == DONE) return;
3940   DCHECK(limit_result == CONTINUE);
3941 
3942   // For loop nodes we already flushed (see LoopChoiceNode::Emit), but for
3943   // other choice nodes we only flush if we are out of code size budget.
3944   if (trace->flush_budget() == 0 && trace->actions() != NULL) {
3945     trace->Flush(compiler, this);
3946     return;
3947   }
3948 
3949   RecursionCheck rc(compiler);
3950 
3951   PreloadState preload;
3952   preload.init();
3953   GreedyLoopState greedy_loop_state(not_at_start());
3954 
3955   int text_length = GreedyLoopTextLengthForAlternative(&alternatives_->at(0));
3956   AlternativeGenerationList alt_gens(choice_count, zone());
3957 
3958   if (choice_count > 1 && text_length != kNodeIsTooComplexForGreedyLoops) {
3959     trace = EmitGreedyLoop(compiler,
3960                            trace,
3961                            &alt_gens,
3962                            &preload,
3963                            &greedy_loop_state,
3964                            text_length);
3965   } else {
3966     // TODO(erikcorry): Delete this.  We don't need this label, but it makes us
3967     // match the traces produced pre-cleanup.
3968     Label second_choice;
3969     compiler->macro_assembler()->Bind(&second_choice);
3970 
3971     preload.eats_at_least_ = EmitOptimizedUnanchoredSearch(compiler, trace);
3972 
3973     EmitChoices(compiler,
3974                 &alt_gens,
3975                 0,
3976                 trace,
3977                 &preload);
3978   }
3979 
3980   // At this point we need to generate slow checks for the alternatives where
3981   // the quick check was inlined.  We can recognize these because the associated
3982   // label was bound.
3983   int new_flush_budget = trace->flush_budget() / choice_count;
3984   for (int i = 0; i < choice_count; i++) {
3985     AlternativeGeneration* alt_gen = alt_gens.at(i);
3986     Trace new_trace(*trace);
3987     // If there are actions to be flushed we have to limit how many times
3988     // they are flushed.  Take the budget of the parent trace and distribute
3989     // it fairly amongst the children.
3990     if (new_trace.actions() != NULL) {
3991       new_trace.set_flush_budget(new_flush_budget);
3992     }
3993     bool next_expects_preload =
3994         i == choice_count - 1 ? false : alt_gens.at(i + 1)->expects_preload;
3995     EmitOutOfLineContinuation(compiler,
3996                               &new_trace,
3997                               alternatives_->at(i),
3998                               alt_gen,
3999                               preload.preload_characters_,
4000                               next_expects_preload);
4001   }
4002 }
4003 
4004 
EmitGreedyLoop(RegExpCompiler * compiler,Trace * trace,AlternativeGenerationList * alt_gens,PreloadState * preload,GreedyLoopState * greedy_loop_state,int text_length)4005 Trace* ChoiceNode::EmitGreedyLoop(RegExpCompiler* compiler,
4006                                   Trace* trace,
4007                                   AlternativeGenerationList* alt_gens,
4008                                   PreloadState* preload,
4009                                   GreedyLoopState* greedy_loop_state,
4010                                   int text_length) {
4011   RegExpMacroAssembler* macro_assembler = compiler->macro_assembler();
4012   // Here we have special handling for greedy loops containing only text nodes
4013   // and other simple nodes.  These are handled by pushing the current
4014   // position on the stack and then incrementing the current position each
4015   // time around the switch.  On backtrack we decrement the current position
4016   // and check it against the pushed value.  This avoids pushing backtrack
4017   // information for each iteration of the loop, which could take up a lot of
4018   // space.
4019   DCHECK(trace->stop_node() == NULL);
4020   macro_assembler->PushCurrentPosition();
4021   Label greedy_match_failed;
4022   Trace greedy_match_trace;
4023   if (not_at_start()) greedy_match_trace.set_at_start(false);
4024   greedy_match_trace.set_backtrack(&greedy_match_failed);
4025   Label loop_label;
4026   macro_assembler->Bind(&loop_label);
4027   greedy_match_trace.set_stop_node(this);
4028   greedy_match_trace.set_loop_label(&loop_label);
4029   alternatives_->at(0).node()->Emit(compiler, &greedy_match_trace);
4030   macro_assembler->Bind(&greedy_match_failed);
4031 
4032   Label second_choice;  // For use in greedy matches.
4033   macro_assembler->Bind(&second_choice);
4034 
4035   Trace* new_trace = greedy_loop_state->counter_backtrack_trace();
4036 
4037   EmitChoices(compiler,
4038               alt_gens,
4039               1,
4040               new_trace,
4041               preload);
4042 
4043   macro_assembler->Bind(greedy_loop_state->label());
4044   // If we have unwound to the bottom then backtrack.
4045   macro_assembler->CheckGreedyLoop(trace->backtrack());
4046   // Otherwise try the second priority at an earlier position.
4047   macro_assembler->AdvanceCurrentPosition(-text_length);
4048   macro_assembler->GoTo(&second_choice);
4049   return new_trace;
4050 }
4051 
EmitOptimizedUnanchoredSearch(RegExpCompiler * compiler,Trace * trace)4052 int ChoiceNode::EmitOptimizedUnanchoredSearch(RegExpCompiler* compiler,
4053                                               Trace* trace) {
4054   int eats_at_least = PreloadState::kEatsAtLeastNotYetInitialized;
4055   if (alternatives_->length() != 2) return eats_at_least;
4056 
4057   GuardedAlternative alt1 = alternatives_->at(1);
4058   if (alt1.guards() != NULL && alt1.guards()->length() != 0) {
4059     return eats_at_least;
4060   }
4061   RegExpNode* eats_anything_node = alt1.node();
4062   if (eats_anything_node->GetSuccessorOfOmnivorousTextNode(compiler) != this) {
4063     return eats_at_least;
4064   }
4065 
4066   // Really we should be creating a new trace when we execute this function,
4067   // but there is no need, because the code it generates cannot backtrack, and
4068   // we always arrive here with a trivial trace (since it's the entry to a
4069   // loop.  That also implies that there are no preloaded characters, which is
4070   // good, because it means we won't be violating any assumptions by
4071   // overwriting those characters with new load instructions.
4072   DCHECK(trace->is_trivial());
4073 
4074   RegExpMacroAssembler* macro_assembler = compiler->macro_assembler();
4075   // At this point we know that we are at a non-greedy loop that will eat
4076   // any character one at a time.  Any non-anchored regexp has such a
4077   // loop prepended to it in order to find where it starts.  We look for
4078   // a pattern of the form ...abc... where we can look 6 characters ahead
4079   // and step forwards 3 if the character is not one of abc.  Abc need
4080   // not be atoms, they can be any reasonably limited character class or
4081   // small alternation.
4082   BoyerMooreLookahead* bm = bm_info(false);
4083   if (bm == NULL) {
4084     eats_at_least = Min(kMaxLookaheadForBoyerMoore,
4085                         EatsAtLeast(kMaxLookaheadForBoyerMoore,
4086                                     kRecursionBudget,
4087                                     false));
4088     if (eats_at_least >= 1) {
4089       bm = new(zone()) BoyerMooreLookahead(eats_at_least,
4090                                            compiler,
4091                                            zone());
4092       GuardedAlternative alt0 = alternatives_->at(0);
4093       alt0.node()->FillInBMInfo(0, kRecursionBudget, bm, false);
4094     }
4095   }
4096   if (bm != NULL) {
4097     bm->EmitSkipInstructions(macro_assembler);
4098   }
4099   return eats_at_least;
4100 }
4101 
4102 
EmitChoices(RegExpCompiler * compiler,AlternativeGenerationList * alt_gens,int first_choice,Trace * trace,PreloadState * preload)4103 void ChoiceNode::EmitChoices(RegExpCompiler* compiler,
4104                              AlternativeGenerationList* alt_gens,
4105                              int first_choice,
4106                              Trace* trace,
4107                              PreloadState* preload) {
4108   RegExpMacroAssembler* macro_assembler = compiler->macro_assembler();
4109   SetUpPreLoad(compiler, trace, preload);
4110 
4111   // For now we just call all choices one after the other.  The idea ultimately
4112   // is to use the Dispatch table to try only the relevant ones.
4113   int choice_count = alternatives_->length();
4114 
4115   int new_flush_budget = trace->flush_budget() / choice_count;
4116 
4117   for (int i = first_choice; i < choice_count; i++) {
4118     bool is_last = i == choice_count - 1;
4119     bool fall_through_on_failure = !is_last;
4120     GuardedAlternative alternative = alternatives_->at(i);
4121     AlternativeGeneration* alt_gen = alt_gens->at(i);
4122     alt_gen->quick_check_details.set_characters(preload->preload_characters_);
4123     ZoneList<Guard*>* guards = alternative.guards();
4124     int guard_count = (guards == NULL) ? 0 : guards->length();
4125     Trace new_trace(*trace);
4126     new_trace.set_characters_preloaded(preload->preload_is_current_ ?
4127                                          preload->preload_characters_ :
4128                                          0);
4129     if (preload->preload_has_checked_bounds_) {
4130       new_trace.set_bound_checked_up_to(preload->preload_characters_);
4131     }
4132     new_trace.quick_check_performed()->Clear();
4133     if (not_at_start_) new_trace.set_at_start(Trace::FALSE_VALUE);
4134     if (!is_last) {
4135       new_trace.set_backtrack(&alt_gen->after);
4136     }
4137     alt_gen->expects_preload = preload->preload_is_current_;
4138     bool generate_full_check_inline = false;
4139     if (FLAG_regexp_optimization &&
4140         try_to_emit_quick_check_for_alternative(i == 0) &&
4141         alternative.node()->EmitQuickCheck(compiler,
4142                                            trace,
4143                                            &new_trace,
4144                                            preload->preload_has_checked_bounds_,
4145                                            &alt_gen->possible_success,
4146                                            &alt_gen->quick_check_details,
4147                                            fall_through_on_failure)) {
4148       // Quick check was generated for this choice.
4149       preload->preload_is_current_ = true;
4150       preload->preload_has_checked_bounds_ = true;
4151       // If we generated the quick check to fall through on possible success,
4152       // we now need to generate the full check inline.
4153       if (!fall_through_on_failure) {
4154         macro_assembler->Bind(&alt_gen->possible_success);
4155         new_trace.set_quick_check_performed(&alt_gen->quick_check_details);
4156         new_trace.set_characters_preloaded(preload->preload_characters_);
4157         new_trace.set_bound_checked_up_to(preload->preload_characters_);
4158         generate_full_check_inline = true;
4159       }
4160     } else if (alt_gen->quick_check_details.cannot_match()) {
4161       if (!fall_through_on_failure) {
4162         macro_assembler->GoTo(trace->backtrack());
4163       }
4164       continue;
4165     } else {
4166       // No quick check was generated.  Put the full code here.
4167       // If this is not the first choice then there could be slow checks from
4168       // previous cases that go here when they fail.  There's no reason to
4169       // insist that they preload characters since the slow check we are about
4170       // to generate probably can't use it.
4171       if (i != first_choice) {
4172         alt_gen->expects_preload = false;
4173         new_trace.InvalidateCurrentCharacter();
4174       }
4175       generate_full_check_inline = true;
4176     }
4177     if (generate_full_check_inline) {
4178       if (new_trace.actions() != NULL) {
4179         new_trace.set_flush_budget(new_flush_budget);
4180       }
4181       for (int j = 0; j < guard_count; j++) {
4182         GenerateGuard(macro_assembler, guards->at(j), &new_trace);
4183       }
4184       alternative.node()->Emit(compiler, &new_trace);
4185       preload->preload_is_current_ = false;
4186     }
4187     macro_assembler->Bind(&alt_gen->after);
4188   }
4189 }
4190 
4191 
EmitOutOfLineContinuation(RegExpCompiler * compiler,Trace * trace,GuardedAlternative alternative,AlternativeGeneration * alt_gen,int preload_characters,bool next_expects_preload)4192 void ChoiceNode::EmitOutOfLineContinuation(RegExpCompiler* compiler,
4193                                            Trace* trace,
4194                                            GuardedAlternative alternative,
4195                                            AlternativeGeneration* alt_gen,
4196                                            int preload_characters,
4197                                            bool next_expects_preload) {
4198   if (!alt_gen->possible_success.is_linked()) return;
4199 
4200   RegExpMacroAssembler* macro_assembler = compiler->macro_assembler();
4201   macro_assembler->Bind(&alt_gen->possible_success);
4202   Trace out_of_line_trace(*trace);
4203   out_of_line_trace.set_characters_preloaded(preload_characters);
4204   out_of_line_trace.set_quick_check_performed(&alt_gen->quick_check_details);
4205   if (not_at_start_) out_of_line_trace.set_at_start(Trace::FALSE_VALUE);
4206   ZoneList<Guard*>* guards = alternative.guards();
4207   int guard_count = (guards == NULL) ? 0 : guards->length();
4208   if (next_expects_preload) {
4209     Label reload_current_char;
4210     out_of_line_trace.set_backtrack(&reload_current_char);
4211     for (int j = 0; j < guard_count; j++) {
4212       GenerateGuard(macro_assembler, guards->at(j), &out_of_line_trace);
4213     }
4214     alternative.node()->Emit(compiler, &out_of_line_trace);
4215     macro_assembler->Bind(&reload_current_char);
4216     // Reload the current character, since the next quick check expects that.
4217     // We don't need to check bounds here because we only get into this
4218     // code through a quick check which already did the checked load.
4219     macro_assembler->LoadCurrentCharacter(trace->cp_offset(),
4220                                           NULL,
4221                                           false,
4222                                           preload_characters);
4223     macro_assembler->GoTo(&(alt_gen->after));
4224   } else {
4225     out_of_line_trace.set_backtrack(&(alt_gen->after));
4226     for (int j = 0; j < guard_count; j++) {
4227       GenerateGuard(macro_assembler, guards->at(j), &out_of_line_trace);
4228     }
4229     alternative.node()->Emit(compiler, &out_of_line_trace);
4230   }
4231 }
4232 
4233 
Emit(RegExpCompiler * compiler,Trace * trace)4234 void ActionNode::Emit(RegExpCompiler* compiler, Trace* trace) {
4235   RegExpMacroAssembler* assembler = compiler->macro_assembler();
4236   LimitResult limit_result = LimitVersions(compiler, trace);
4237   if (limit_result == DONE) return;
4238   DCHECK(limit_result == CONTINUE);
4239 
4240   RecursionCheck rc(compiler);
4241 
4242   switch (action_type_) {
4243     case STORE_POSITION: {
4244       Trace::DeferredCapture
4245           new_capture(data_.u_position_register.reg,
4246                       data_.u_position_register.is_capture,
4247                       trace);
4248       Trace new_trace = *trace;
4249       new_trace.add_action(&new_capture);
4250       on_success()->Emit(compiler, &new_trace);
4251       break;
4252     }
4253     case INCREMENT_REGISTER: {
4254       Trace::DeferredIncrementRegister
4255           new_increment(data_.u_increment_register.reg);
4256       Trace new_trace = *trace;
4257       new_trace.add_action(&new_increment);
4258       on_success()->Emit(compiler, &new_trace);
4259       break;
4260     }
4261     case SET_REGISTER: {
4262       Trace::DeferredSetRegister
4263           new_set(data_.u_store_register.reg, data_.u_store_register.value);
4264       Trace new_trace = *trace;
4265       new_trace.add_action(&new_set);
4266       on_success()->Emit(compiler, &new_trace);
4267       break;
4268     }
4269     case CLEAR_CAPTURES: {
4270       Trace::DeferredClearCaptures
4271         new_capture(Interval(data_.u_clear_captures.range_from,
4272                              data_.u_clear_captures.range_to));
4273       Trace new_trace = *trace;
4274       new_trace.add_action(&new_capture);
4275       on_success()->Emit(compiler, &new_trace);
4276       break;
4277     }
4278     case BEGIN_SUBMATCH:
4279       if (!trace->is_trivial()) {
4280         trace->Flush(compiler, this);
4281       } else {
4282         assembler->WriteCurrentPositionToRegister(
4283             data_.u_submatch.current_position_register, 0);
4284         assembler->WriteStackPointerToRegister(
4285             data_.u_submatch.stack_pointer_register);
4286         on_success()->Emit(compiler, trace);
4287       }
4288       break;
4289     case EMPTY_MATCH_CHECK: {
4290       int start_pos_reg = data_.u_empty_match_check.start_register;
4291       int stored_pos = 0;
4292       int rep_reg = data_.u_empty_match_check.repetition_register;
4293       bool has_minimum = (rep_reg != RegExpCompiler::kNoRegister);
4294       bool know_dist = trace->GetStoredPosition(start_pos_reg, &stored_pos);
4295       if (know_dist && !has_minimum && stored_pos == trace->cp_offset()) {
4296         // If we know we haven't advanced and there is no minimum we
4297         // can just backtrack immediately.
4298         assembler->GoTo(trace->backtrack());
4299       } else if (know_dist && stored_pos < trace->cp_offset()) {
4300         // If we know we've advanced we can generate the continuation
4301         // immediately.
4302         on_success()->Emit(compiler, trace);
4303       } else if (!trace->is_trivial()) {
4304         trace->Flush(compiler, this);
4305       } else {
4306         Label skip_empty_check;
4307         // If we have a minimum number of repetitions we check the current
4308         // number first and skip the empty check if it's not enough.
4309         if (has_minimum) {
4310           int limit = data_.u_empty_match_check.repetition_limit;
4311           assembler->IfRegisterLT(rep_reg, limit, &skip_empty_check);
4312         }
4313         // If the match is empty we bail out, otherwise we fall through
4314         // to the on-success continuation.
4315         assembler->IfRegisterEqPos(data_.u_empty_match_check.start_register,
4316                                    trace->backtrack());
4317         assembler->Bind(&skip_empty_check);
4318         on_success()->Emit(compiler, trace);
4319       }
4320       break;
4321     }
4322     case POSITIVE_SUBMATCH_SUCCESS: {
4323       if (!trace->is_trivial()) {
4324         trace->Flush(compiler, this);
4325         return;
4326       }
4327       assembler->ReadCurrentPositionFromRegister(
4328           data_.u_submatch.current_position_register);
4329       assembler->ReadStackPointerFromRegister(
4330           data_.u_submatch.stack_pointer_register);
4331       int clear_register_count = data_.u_submatch.clear_register_count;
4332       if (clear_register_count == 0) {
4333         on_success()->Emit(compiler, trace);
4334         return;
4335       }
4336       int clear_registers_from = data_.u_submatch.clear_register_from;
4337       Label clear_registers_backtrack;
4338       Trace new_trace = *trace;
4339       new_trace.set_backtrack(&clear_registers_backtrack);
4340       on_success()->Emit(compiler, &new_trace);
4341 
4342       assembler->Bind(&clear_registers_backtrack);
4343       int clear_registers_to = clear_registers_from + clear_register_count - 1;
4344       assembler->ClearRegisters(clear_registers_from, clear_registers_to);
4345 
4346       DCHECK(trace->backtrack() == NULL);
4347       assembler->Backtrack();
4348       return;
4349     }
4350     default:
4351       UNREACHABLE();
4352   }
4353 }
4354 
4355 
Emit(RegExpCompiler * compiler,Trace * trace)4356 void BackReferenceNode::Emit(RegExpCompiler* compiler, Trace* trace) {
4357   RegExpMacroAssembler* assembler = compiler->macro_assembler();
4358   if (!trace->is_trivial()) {
4359     trace->Flush(compiler, this);
4360     return;
4361   }
4362 
4363   LimitResult limit_result = LimitVersions(compiler, trace);
4364   if (limit_result == DONE) return;
4365   DCHECK(limit_result == CONTINUE);
4366 
4367   RecursionCheck rc(compiler);
4368 
4369   DCHECK_EQ(start_reg_ + 1, end_reg_);
4370   if (compiler->ignore_case()) {
4371     assembler->CheckNotBackReferenceIgnoreCase(start_reg_,
4372                                                trace->backtrack());
4373   } else {
4374     assembler->CheckNotBackReference(start_reg_, trace->backtrack());
4375   }
4376   on_success()->Emit(compiler, trace);
4377 }
4378 
4379 
4380 // -------------------------------------------------------------------
4381 // Dot/dotty output
4382 
4383 
4384 #ifdef DEBUG
4385 
4386 
4387 class DotPrinter: public NodeVisitor {
4388  public:
DotPrinter(OStream & os,bool ignore_case)4389   DotPrinter(OStream& os, bool ignore_case)  // NOLINT
4390       : os_(os),
4391         ignore_case_(ignore_case) {}
4392   void PrintNode(const char* label, RegExpNode* node);
4393   void Visit(RegExpNode* node);
4394   void PrintAttributes(RegExpNode* from);
4395   void PrintOnFailure(RegExpNode* from, RegExpNode* to);
4396 #define DECLARE_VISIT(Type)                                          \
4397   virtual void Visit##Type(Type##Node* that);
4398 FOR_EACH_NODE_TYPE(DECLARE_VISIT)
4399 #undef DECLARE_VISIT
4400  private:
4401   OStream& os_;
4402   bool ignore_case_;
4403 };
4404 
4405 
PrintNode(const char * label,RegExpNode * node)4406 void DotPrinter::PrintNode(const char* label, RegExpNode* node) {
4407   os_ << "digraph G {\n  graph [label=\"";
4408   for (int i = 0; label[i]; i++) {
4409     switch (label[i]) {
4410       case '\\':
4411         os_ << "\\\\";
4412         break;
4413       case '"':
4414         os_ << "\"";
4415         break;
4416       default:
4417         os_ << label[i];
4418         break;
4419     }
4420   }
4421   os_ << "\"];\n";
4422   Visit(node);
4423   os_ << "}" << endl;
4424 }
4425 
4426 
Visit(RegExpNode * node)4427 void DotPrinter::Visit(RegExpNode* node) {
4428   if (node->info()->visited) return;
4429   node->info()->visited = true;
4430   node->Accept(this);
4431 }
4432 
4433 
PrintOnFailure(RegExpNode * from,RegExpNode * on_failure)4434 void DotPrinter::PrintOnFailure(RegExpNode* from, RegExpNode* on_failure) {
4435   os_ << "  n" << from << " -> n" << on_failure << " [style=dotted];\n";
4436   Visit(on_failure);
4437 }
4438 
4439 
4440 class TableEntryBodyPrinter {
4441  public:
TableEntryBodyPrinter(OStream & os,ChoiceNode * choice)4442   TableEntryBodyPrinter(OStream& os, ChoiceNode* choice)  // NOLINT
4443       : os_(os),
4444         choice_(choice) {}
Call(uc16 from,DispatchTable::Entry entry)4445   void Call(uc16 from, DispatchTable::Entry entry) {
4446     OutSet* out_set = entry.out_set();
4447     for (unsigned i = 0; i < OutSet::kFirstLimit; i++) {
4448       if (out_set->Get(i)) {
4449         os_ << "    n" << choice() << ":s" << from << "o" << i << " -> n"
4450             << choice()->alternatives()->at(i).node() << ";\n";
4451       }
4452     }
4453   }
4454  private:
choice()4455   ChoiceNode* choice() { return choice_; }
4456   OStream& os_;
4457   ChoiceNode* choice_;
4458 };
4459 
4460 
4461 class TableEntryHeaderPrinter {
4462  public:
TableEntryHeaderPrinter(OStream & os)4463   explicit TableEntryHeaderPrinter(OStream& os)  // NOLINT
4464       : first_(true),
4465         os_(os) {}
Call(uc16 from,DispatchTable::Entry entry)4466   void Call(uc16 from, DispatchTable::Entry entry) {
4467     if (first_) {
4468       first_ = false;
4469     } else {
4470       os_ << "|";
4471     }
4472     os_ << "{\\" << AsUC16(from) << "-\\" << AsUC16(entry.to()) << "|{";
4473     OutSet* out_set = entry.out_set();
4474     int priority = 0;
4475     for (unsigned i = 0; i < OutSet::kFirstLimit; i++) {
4476       if (out_set->Get(i)) {
4477         if (priority > 0) os_ << "|";
4478         os_ << "<s" << from << "o" << i << "> " << priority;
4479         priority++;
4480       }
4481     }
4482     os_ << "}}";
4483   }
4484 
4485  private:
4486   bool first_;
4487   OStream& os_;
4488 };
4489 
4490 
4491 class AttributePrinter {
4492  public:
AttributePrinter(OStream & os)4493   explicit AttributePrinter(OStream& os)  // NOLINT
4494       : os_(os),
4495         first_(true) {}
PrintSeparator()4496   void PrintSeparator() {
4497     if (first_) {
4498       first_ = false;
4499     } else {
4500       os_ << "|";
4501     }
4502   }
PrintBit(const char * name,bool value)4503   void PrintBit(const char* name, bool value) {
4504     if (!value) return;
4505     PrintSeparator();
4506     os_ << "{" << name << "}";
4507   }
PrintPositive(const char * name,int value)4508   void PrintPositive(const char* name, int value) {
4509     if (value < 0) return;
4510     PrintSeparator();
4511     os_ << "{" << name << "|" << value << "}";
4512   }
4513 
4514  private:
4515   OStream& os_;
4516   bool first_;
4517 };
4518 
4519 
PrintAttributes(RegExpNode * that)4520 void DotPrinter::PrintAttributes(RegExpNode* that) {
4521   os_ << "  a" << that << " [shape=Mrecord, color=grey, fontcolor=grey, "
4522       << "margin=0.1, fontsize=10, label=\"{";
4523   AttributePrinter printer(os_);
4524   NodeInfo* info = that->info();
4525   printer.PrintBit("NI", info->follows_newline_interest);
4526   printer.PrintBit("WI", info->follows_word_interest);
4527   printer.PrintBit("SI", info->follows_start_interest);
4528   Label* label = that->label();
4529   if (label->is_bound())
4530     printer.PrintPositive("@", label->pos());
4531   os_ << "}\"];\n"
4532       << "  a" << that << " -> n" << that
4533       << " [style=dashed, color=grey, arrowhead=none];\n";
4534 }
4535 
4536 
4537 static const bool kPrintDispatchTable = false;
VisitChoice(ChoiceNode * that)4538 void DotPrinter::VisitChoice(ChoiceNode* that) {
4539   if (kPrintDispatchTable) {
4540     os_ << "  n" << that << " [shape=Mrecord, label=\"";
4541     TableEntryHeaderPrinter header_printer(os_);
4542     that->GetTable(ignore_case_)->ForEach(&header_printer);
4543     os_ << "\"]\n";
4544     PrintAttributes(that);
4545     TableEntryBodyPrinter body_printer(os_, that);
4546     that->GetTable(ignore_case_)->ForEach(&body_printer);
4547   } else {
4548     os_ << "  n" << that << " [shape=Mrecord, label=\"?\"];\n";
4549     for (int i = 0; i < that->alternatives()->length(); i++) {
4550       GuardedAlternative alt = that->alternatives()->at(i);
4551       os_ << "  n" << that << " -> n" << alt.node();
4552     }
4553   }
4554   for (int i = 0; i < that->alternatives()->length(); i++) {
4555     GuardedAlternative alt = that->alternatives()->at(i);
4556     alt.node()->Accept(this);
4557   }
4558 }
4559 
4560 
VisitText(TextNode * that)4561 void DotPrinter::VisitText(TextNode* that) {
4562   Zone* zone = that->zone();
4563   os_ << "  n" << that << " [label=\"";
4564   for (int i = 0; i < that->elements()->length(); i++) {
4565     if (i > 0) os_ << " ";
4566     TextElement elm = that->elements()->at(i);
4567     switch (elm.text_type()) {
4568       case TextElement::ATOM: {
4569         Vector<const uc16> data = elm.atom()->data();
4570         for (int i = 0; i < data.length(); i++) {
4571           os_ << static_cast<char>(data[i]);
4572         }
4573         break;
4574       }
4575       case TextElement::CHAR_CLASS: {
4576         RegExpCharacterClass* node = elm.char_class();
4577         os_ << "[";
4578         if (node->is_negated()) os_ << "^";
4579         for (int j = 0; j < node->ranges(zone)->length(); j++) {
4580           CharacterRange range = node->ranges(zone)->at(j);
4581           os_ << AsUC16(range.from()) << "-" << AsUC16(range.to());
4582         }
4583         os_ << "]";
4584         break;
4585       }
4586       default:
4587         UNREACHABLE();
4588     }
4589   }
4590   os_ << "\", shape=box, peripheries=2];\n";
4591   PrintAttributes(that);
4592   os_ << "  n" << that << " -> n" << that->on_success() << ";\n";
4593   Visit(that->on_success());
4594 }
4595 
4596 
VisitBackReference(BackReferenceNode * that)4597 void DotPrinter::VisitBackReference(BackReferenceNode* that) {
4598   os_ << "  n" << that << " [label=\"$" << that->start_register() << "..$"
4599       << that->end_register() << "\", shape=doubleoctagon];\n";
4600   PrintAttributes(that);
4601   os_ << "  n" << that << " -> n" << that->on_success() << ";\n";
4602   Visit(that->on_success());
4603 }
4604 
4605 
VisitEnd(EndNode * that)4606 void DotPrinter::VisitEnd(EndNode* that) {
4607   os_ << "  n" << that << " [style=bold, shape=point];\n";
4608   PrintAttributes(that);
4609 }
4610 
4611 
VisitAssertion(AssertionNode * that)4612 void DotPrinter::VisitAssertion(AssertionNode* that) {
4613   os_ << "  n" << that << " [";
4614   switch (that->assertion_type()) {
4615     case AssertionNode::AT_END:
4616       os_ << "label=\"$\", shape=septagon";
4617       break;
4618     case AssertionNode::AT_START:
4619       os_ << "label=\"^\", shape=septagon";
4620       break;
4621     case AssertionNode::AT_BOUNDARY:
4622       os_ << "label=\"\\b\", shape=septagon";
4623       break;
4624     case AssertionNode::AT_NON_BOUNDARY:
4625       os_ << "label=\"\\B\", shape=septagon";
4626       break;
4627     case AssertionNode::AFTER_NEWLINE:
4628       os_ << "label=\"(?<=\\n)\", shape=septagon";
4629       break;
4630   }
4631   os_ << "];\n";
4632   PrintAttributes(that);
4633   RegExpNode* successor = that->on_success();
4634   os_ << "  n" << that << " -> n" << successor << ";\n";
4635   Visit(successor);
4636 }
4637 
4638 
VisitAction(ActionNode * that)4639 void DotPrinter::VisitAction(ActionNode* that) {
4640   os_ << "  n" << that << " [";
4641   switch (that->action_type_) {
4642     case ActionNode::SET_REGISTER:
4643       os_ << "label=\"$" << that->data_.u_store_register.reg
4644           << ":=" << that->data_.u_store_register.value << "\", shape=octagon";
4645       break;
4646     case ActionNode::INCREMENT_REGISTER:
4647       os_ << "label=\"$" << that->data_.u_increment_register.reg
4648           << "++\", shape=octagon";
4649       break;
4650     case ActionNode::STORE_POSITION:
4651       os_ << "label=\"$" << that->data_.u_position_register.reg
4652           << ":=$pos\", shape=octagon";
4653       break;
4654     case ActionNode::BEGIN_SUBMATCH:
4655       os_ << "label=\"$" << that->data_.u_submatch.current_position_register
4656           << ":=$pos,begin\", shape=septagon";
4657       break;
4658     case ActionNode::POSITIVE_SUBMATCH_SUCCESS:
4659       os_ << "label=\"escape\", shape=septagon";
4660       break;
4661     case ActionNode::EMPTY_MATCH_CHECK:
4662       os_ << "label=\"$" << that->data_.u_empty_match_check.start_register
4663           << "=$pos?,$" << that->data_.u_empty_match_check.repetition_register
4664           << "<" << that->data_.u_empty_match_check.repetition_limit
4665           << "?\", shape=septagon";
4666       break;
4667     case ActionNode::CLEAR_CAPTURES: {
4668       os_ << "label=\"clear $" << that->data_.u_clear_captures.range_from
4669           << " to $" << that->data_.u_clear_captures.range_to
4670           << "\", shape=septagon";
4671       break;
4672     }
4673   }
4674   os_ << "];\n";
4675   PrintAttributes(that);
4676   RegExpNode* successor = that->on_success();
4677   os_ << "  n" << that << " -> n" << successor << ";\n";
4678   Visit(successor);
4679 }
4680 
4681 
4682 class DispatchTableDumper {
4683  public:
DispatchTableDumper(OStream & os)4684   explicit DispatchTableDumper(OStream& os) : os_(os) {}
4685   void Call(uc16 key, DispatchTable::Entry entry);
4686  private:
4687   OStream& os_;
4688 };
4689 
4690 
Call(uc16 key,DispatchTable::Entry entry)4691 void DispatchTableDumper::Call(uc16 key, DispatchTable::Entry entry) {
4692   os_ << "[" << AsUC16(key) << "-" << AsUC16(entry.to()) << "]: {";
4693   OutSet* set = entry.out_set();
4694   bool first = true;
4695   for (unsigned i = 0; i < OutSet::kFirstLimit; i++) {
4696     if (set->Get(i)) {
4697       if (first) {
4698         first = false;
4699       } else {
4700         os_ << ", ";
4701       }
4702       os_ << i;
4703     }
4704   }
4705   os_ << "}\n";
4706 }
4707 
4708 
Dump()4709 void DispatchTable::Dump() {
4710   OFStream os(stderr);
4711   DispatchTableDumper dumper(os);
4712   tree()->ForEach(&dumper);
4713 }
4714 
4715 
DotPrint(const char * label,RegExpNode * node,bool ignore_case)4716 void RegExpEngine::DotPrint(const char* label,
4717                             RegExpNode* node,
4718                             bool ignore_case) {
4719   OFStream os(stdout);
4720   DotPrinter printer(os, ignore_case);
4721   printer.PrintNode(label, node);
4722 }
4723 
4724 
4725 #endif  // DEBUG
4726 
4727 
4728 // -------------------------------------------------------------------
4729 // Tree to graph conversion
4730 
ToNode(RegExpCompiler * compiler,RegExpNode * on_success)4731 RegExpNode* RegExpAtom::ToNode(RegExpCompiler* compiler,
4732                                RegExpNode* on_success) {
4733   ZoneList<TextElement>* elms =
4734       new(compiler->zone()) ZoneList<TextElement>(1, compiler->zone());
4735   elms->Add(TextElement::Atom(this), compiler->zone());
4736   return new(compiler->zone()) TextNode(elms, on_success);
4737 }
4738 
4739 
ToNode(RegExpCompiler * compiler,RegExpNode * on_success)4740 RegExpNode* RegExpText::ToNode(RegExpCompiler* compiler,
4741                                RegExpNode* on_success) {
4742   return new(compiler->zone()) TextNode(elements(), on_success);
4743 }
4744 
4745 
CompareInverseRanges(ZoneList<CharacterRange> * ranges,const int * special_class,int length)4746 static bool CompareInverseRanges(ZoneList<CharacterRange>* ranges,
4747                                  const int* special_class,
4748                                  int length) {
4749   length--;  // Remove final 0x10000.
4750   DCHECK(special_class[length] == 0x10000);
4751   DCHECK(ranges->length() != 0);
4752   DCHECK(length != 0);
4753   DCHECK(special_class[0] != 0);
4754   if (ranges->length() != (length >> 1) + 1) {
4755     return false;
4756   }
4757   CharacterRange range = ranges->at(0);
4758   if (range.from() != 0) {
4759     return false;
4760   }
4761   for (int i = 0; i < length; i += 2) {
4762     if (special_class[i] != (range.to() + 1)) {
4763       return false;
4764     }
4765     range = ranges->at((i >> 1) + 1);
4766     if (special_class[i+1] != range.from()) {
4767       return false;
4768     }
4769   }
4770   if (range.to() != 0xffff) {
4771     return false;
4772   }
4773   return true;
4774 }
4775 
4776 
CompareRanges(ZoneList<CharacterRange> * ranges,const int * special_class,int length)4777 static bool CompareRanges(ZoneList<CharacterRange>* ranges,
4778                           const int* special_class,
4779                           int length) {
4780   length--;  // Remove final 0x10000.
4781   DCHECK(special_class[length] == 0x10000);
4782   if (ranges->length() * 2 != length) {
4783     return false;
4784   }
4785   for (int i = 0; i < length; i += 2) {
4786     CharacterRange range = ranges->at(i >> 1);
4787     if (range.from() != special_class[i] ||
4788         range.to() != special_class[i + 1] - 1) {
4789       return false;
4790     }
4791   }
4792   return true;
4793 }
4794 
4795 
is_standard(Zone * zone)4796 bool RegExpCharacterClass::is_standard(Zone* zone) {
4797   // TODO(lrn): Remove need for this function, by not throwing away information
4798   // along the way.
4799   if (is_negated_) {
4800     return false;
4801   }
4802   if (set_.is_standard()) {
4803     return true;
4804   }
4805   if (CompareRanges(set_.ranges(zone), kSpaceRanges, kSpaceRangeCount)) {
4806     set_.set_standard_set_type('s');
4807     return true;
4808   }
4809   if (CompareInverseRanges(set_.ranges(zone), kSpaceRanges, kSpaceRangeCount)) {
4810     set_.set_standard_set_type('S');
4811     return true;
4812   }
4813   if (CompareInverseRanges(set_.ranges(zone),
4814                            kLineTerminatorRanges,
4815                            kLineTerminatorRangeCount)) {
4816     set_.set_standard_set_type('.');
4817     return true;
4818   }
4819   if (CompareRanges(set_.ranges(zone),
4820                     kLineTerminatorRanges,
4821                     kLineTerminatorRangeCount)) {
4822     set_.set_standard_set_type('n');
4823     return true;
4824   }
4825   if (CompareRanges(set_.ranges(zone), kWordRanges, kWordRangeCount)) {
4826     set_.set_standard_set_type('w');
4827     return true;
4828   }
4829   if (CompareInverseRanges(set_.ranges(zone), kWordRanges, kWordRangeCount)) {
4830     set_.set_standard_set_type('W');
4831     return true;
4832   }
4833   return false;
4834 }
4835 
4836 
ToNode(RegExpCompiler * compiler,RegExpNode * on_success)4837 RegExpNode* RegExpCharacterClass::ToNode(RegExpCompiler* compiler,
4838                                          RegExpNode* on_success) {
4839   return new(compiler->zone()) TextNode(this, on_success);
4840 }
4841 
4842 
ToNode(RegExpCompiler * compiler,RegExpNode * on_success)4843 RegExpNode* RegExpDisjunction::ToNode(RegExpCompiler* compiler,
4844                                       RegExpNode* on_success) {
4845   ZoneList<RegExpTree*>* alternatives = this->alternatives();
4846   int length = alternatives->length();
4847   ChoiceNode* result =
4848       new(compiler->zone()) ChoiceNode(length, compiler->zone());
4849   for (int i = 0; i < length; i++) {
4850     GuardedAlternative alternative(alternatives->at(i)->ToNode(compiler,
4851                                                                on_success));
4852     result->AddAlternative(alternative);
4853   }
4854   return result;
4855 }
4856 
4857 
ToNode(RegExpCompiler * compiler,RegExpNode * on_success)4858 RegExpNode* RegExpQuantifier::ToNode(RegExpCompiler* compiler,
4859                                      RegExpNode* on_success) {
4860   return ToNode(min(),
4861                 max(),
4862                 is_greedy(),
4863                 body(),
4864                 compiler,
4865                 on_success);
4866 }
4867 
4868 
4869 // Scoped object to keep track of how much we unroll quantifier loops in the
4870 // regexp graph generator.
4871 class RegExpExpansionLimiter {
4872  public:
4873   static const int kMaxExpansionFactor = 6;
RegExpExpansionLimiter(RegExpCompiler * compiler,int factor)4874   RegExpExpansionLimiter(RegExpCompiler* compiler, int factor)
4875       : compiler_(compiler),
4876         saved_expansion_factor_(compiler->current_expansion_factor()),
4877         ok_to_expand_(saved_expansion_factor_ <= kMaxExpansionFactor) {
4878     DCHECK(factor > 0);
4879     if (ok_to_expand_) {
4880       if (factor > kMaxExpansionFactor) {
4881         // Avoid integer overflow of the current expansion factor.
4882         ok_to_expand_ = false;
4883         compiler->set_current_expansion_factor(kMaxExpansionFactor + 1);
4884       } else {
4885         int new_factor = saved_expansion_factor_ * factor;
4886         ok_to_expand_ = (new_factor <= kMaxExpansionFactor);
4887         compiler->set_current_expansion_factor(new_factor);
4888       }
4889     }
4890   }
4891 
~RegExpExpansionLimiter()4892   ~RegExpExpansionLimiter() {
4893     compiler_->set_current_expansion_factor(saved_expansion_factor_);
4894   }
4895 
ok_to_expand()4896   bool ok_to_expand() { return ok_to_expand_; }
4897 
4898  private:
4899   RegExpCompiler* compiler_;
4900   int saved_expansion_factor_;
4901   bool ok_to_expand_;
4902 
4903   DISALLOW_IMPLICIT_CONSTRUCTORS(RegExpExpansionLimiter);
4904 };
4905 
4906 
ToNode(int min,int max,bool is_greedy,RegExpTree * body,RegExpCompiler * compiler,RegExpNode * on_success,bool not_at_start)4907 RegExpNode* RegExpQuantifier::ToNode(int min,
4908                                      int max,
4909                                      bool is_greedy,
4910                                      RegExpTree* body,
4911                                      RegExpCompiler* compiler,
4912                                      RegExpNode* on_success,
4913                                      bool not_at_start) {
4914   // x{f, t} becomes this:
4915   //
4916   //             (r++)<-.
4917   //               |     `
4918   //               |     (x)
4919   //               v     ^
4920   //      (r=0)-->(?)---/ [if r < t]
4921   //               |
4922   //   [if r >= f] \----> ...
4923   //
4924 
4925   // 15.10.2.5 RepeatMatcher algorithm.
4926   // The parser has already eliminated the case where max is 0.  In the case
4927   // where max_match is zero the parser has removed the quantifier if min was
4928   // > 0 and removed the atom if min was 0.  See AddQuantifierToAtom.
4929 
4930   // If we know that we cannot match zero length then things are a little
4931   // simpler since we don't need to make the special zero length match check
4932   // from step 2.1.  If the min and max are small we can unroll a little in
4933   // this case.
4934   static const int kMaxUnrolledMinMatches = 3;  // Unroll (foo)+ and (foo){3,}
4935   static const int kMaxUnrolledMaxMatches = 3;  // Unroll (foo)? and (foo){x,3}
4936   if (max == 0) return on_success;  // This can happen due to recursion.
4937   bool body_can_be_empty = (body->min_match() == 0);
4938   int body_start_reg = RegExpCompiler::kNoRegister;
4939   Interval capture_registers = body->CaptureRegisters();
4940   bool needs_capture_clearing = !capture_registers.is_empty();
4941   Zone* zone = compiler->zone();
4942 
4943   if (body_can_be_empty) {
4944     body_start_reg = compiler->AllocateRegister();
4945   } else if (FLAG_regexp_optimization && !needs_capture_clearing) {
4946     // Only unroll if there are no captures and the body can't be
4947     // empty.
4948     {
4949       RegExpExpansionLimiter limiter(
4950           compiler, min + ((max != min) ? 1 : 0));
4951       if (min > 0 && min <= kMaxUnrolledMinMatches && limiter.ok_to_expand()) {
4952         int new_max = (max == kInfinity) ? max : max - min;
4953         // Recurse once to get the loop or optional matches after the fixed
4954         // ones.
4955         RegExpNode* answer = ToNode(
4956             0, new_max, is_greedy, body, compiler, on_success, true);
4957         // Unroll the forced matches from 0 to min.  This can cause chains of
4958         // TextNodes (which the parser does not generate).  These should be
4959         // combined if it turns out they hinder good code generation.
4960         for (int i = 0; i < min; i++) {
4961           answer = body->ToNode(compiler, answer);
4962         }
4963         return answer;
4964       }
4965     }
4966     if (max <= kMaxUnrolledMaxMatches && min == 0) {
4967       DCHECK(max > 0);  // Due to the 'if' above.
4968       RegExpExpansionLimiter limiter(compiler, max);
4969       if (limiter.ok_to_expand()) {
4970         // Unroll the optional matches up to max.
4971         RegExpNode* answer = on_success;
4972         for (int i = 0; i < max; i++) {
4973           ChoiceNode* alternation = new(zone) ChoiceNode(2, zone);
4974           if (is_greedy) {
4975             alternation->AddAlternative(
4976                 GuardedAlternative(body->ToNode(compiler, answer)));
4977             alternation->AddAlternative(GuardedAlternative(on_success));
4978           } else {
4979             alternation->AddAlternative(GuardedAlternative(on_success));
4980             alternation->AddAlternative(
4981                 GuardedAlternative(body->ToNode(compiler, answer)));
4982           }
4983           answer = alternation;
4984           if (not_at_start) alternation->set_not_at_start();
4985         }
4986         return answer;
4987       }
4988     }
4989   }
4990   bool has_min = min > 0;
4991   bool has_max = max < RegExpTree::kInfinity;
4992   bool needs_counter = has_min || has_max;
4993   int reg_ctr = needs_counter
4994       ? compiler->AllocateRegister()
4995       : RegExpCompiler::kNoRegister;
4996   LoopChoiceNode* center = new(zone) LoopChoiceNode(body->min_match() == 0,
4997                                                     zone);
4998   if (not_at_start) center->set_not_at_start();
4999   RegExpNode* loop_return = needs_counter
5000       ? static_cast<RegExpNode*>(ActionNode::IncrementRegister(reg_ctr, center))
5001       : static_cast<RegExpNode*>(center);
5002   if (body_can_be_empty) {
5003     // If the body can be empty we need to check if it was and then
5004     // backtrack.
5005     loop_return = ActionNode::EmptyMatchCheck(body_start_reg,
5006                                               reg_ctr,
5007                                               min,
5008                                               loop_return);
5009   }
5010   RegExpNode* body_node = body->ToNode(compiler, loop_return);
5011   if (body_can_be_empty) {
5012     // If the body can be empty we need to store the start position
5013     // so we can bail out if it was empty.
5014     body_node = ActionNode::StorePosition(body_start_reg, false, body_node);
5015   }
5016   if (needs_capture_clearing) {
5017     // Before entering the body of this loop we need to clear captures.
5018     body_node = ActionNode::ClearCaptures(capture_registers, body_node);
5019   }
5020   GuardedAlternative body_alt(body_node);
5021   if (has_max) {
5022     Guard* body_guard =
5023         new(zone) Guard(reg_ctr, Guard::LT, max);
5024     body_alt.AddGuard(body_guard, zone);
5025   }
5026   GuardedAlternative rest_alt(on_success);
5027   if (has_min) {
5028     Guard* rest_guard = new(compiler->zone()) Guard(reg_ctr, Guard::GEQ, min);
5029     rest_alt.AddGuard(rest_guard, zone);
5030   }
5031   if (is_greedy) {
5032     center->AddLoopAlternative(body_alt);
5033     center->AddContinueAlternative(rest_alt);
5034   } else {
5035     center->AddContinueAlternative(rest_alt);
5036     center->AddLoopAlternative(body_alt);
5037   }
5038   if (needs_counter) {
5039     return ActionNode::SetRegister(reg_ctr, 0, center);
5040   } else {
5041     return center;
5042   }
5043 }
5044 
5045 
ToNode(RegExpCompiler * compiler,RegExpNode * on_success)5046 RegExpNode* RegExpAssertion::ToNode(RegExpCompiler* compiler,
5047                                     RegExpNode* on_success) {
5048   NodeInfo info;
5049   Zone* zone = compiler->zone();
5050 
5051   switch (assertion_type()) {
5052     case START_OF_LINE:
5053       return AssertionNode::AfterNewline(on_success);
5054     case START_OF_INPUT:
5055       return AssertionNode::AtStart(on_success);
5056     case BOUNDARY:
5057       return AssertionNode::AtBoundary(on_success);
5058     case NON_BOUNDARY:
5059       return AssertionNode::AtNonBoundary(on_success);
5060     case END_OF_INPUT:
5061       return AssertionNode::AtEnd(on_success);
5062     case END_OF_LINE: {
5063       // Compile $ in multiline regexps as an alternation with a positive
5064       // lookahead in one side and an end-of-input on the other side.
5065       // We need two registers for the lookahead.
5066       int stack_pointer_register = compiler->AllocateRegister();
5067       int position_register = compiler->AllocateRegister();
5068       // The ChoiceNode to distinguish between a newline and end-of-input.
5069       ChoiceNode* result = new(zone) ChoiceNode(2, zone);
5070       // Create a newline atom.
5071       ZoneList<CharacterRange>* newline_ranges =
5072           new(zone) ZoneList<CharacterRange>(3, zone);
5073       CharacterRange::AddClassEscape('n', newline_ranges, zone);
5074       RegExpCharacterClass* newline_atom = new(zone) RegExpCharacterClass('n');
5075       TextNode* newline_matcher = new(zone) TextNode(
5076          newline_atom,
5077          ActionNode::PositiveSubmatchSuccess(stack_pointer_register,
5078                                              position_register,
5079                                              0,  // No captures inside.
5080                                              -1,  // Ignored if no captures.
5081                                              on_success));
5082       // Create an end-of-input matcher.
5083       RegExpNode* end_of_line = ActionNode::BeginSubmatch(
5084           stack_pointer_register,
5085           position_register,
5086           newline_matcher);
5087       // Add the two alternatives to the ChoiceNode.
5088       GuardedAlternative eol_alternative(end_of_line);
5089       result->AddAlternative(eol_alternative);
5090       GuardedAlternative end_alternative(AssertionNode::AtEnd(on_success));
5091       result->AddAlternative(end_alternative);
5092       return result;
5093     }
5094     default:
5095       UNREACHABLE();
5096   }
5097   return on_success;
5098 }
5099 
5100 
ToNode(RegExpCompiler * compiler,RegExpNode * on_success)5101 RegExpNode* RegExpBackReference::ToNode(RegExpCompiler* compiler,
5102                                         RegExpNode* on_success) {
5103   return new(compiler->zone())
5104       BackReferenceNode(RegExpCapture::StartRegister(index()),
5105                         RegExpCapture::EndRegister(index()),
5106                         on_success);
5107 }
5108 
5109 
ToNode(RegExpCompiler * compiler,RegExpNode * on_success)5110 RegExpNode* RegExpEmpty::ToNode(RegExpCompiler* compiler,
5111                                 RegExpNode* on_success) {
5112   return on_success;
5113 }
5114 
5115 
ToNode(RegExpCompiler * compiler,RegExpNode * on_success)5116 RegExpNode* RegExpLookahead::ToNode(RegExpCompiler* compiler,
5117                                     RegExpNode* on_success) {
5118   int stack_pointer_register = compiler->AllocateRegister();
5119   int position_register = compiler->AllocateRegister();
5120 
5121   const int registers_per_capture = 2;
5122   const int register_of_first_capture = 2;
5123   int register_count = capture_count_ * registers_per_capture;
5124   int register_start =
5125     register_of_first_capture + capture_from_ * registers_per_capture;
5126 
5127   RegExpNode* success;
5128   if (is_positive()) {
5129     RegExpNode* node = ActionNode::BeginSubmatch(
5130         stack_pointer_register,
5131         position_register,
5132         body()->ToNode(
5133             compiler,
5134             ActionNode::PositiveSubmatchSuccess(stack_pointer_register,
5135                                                 position_register,
5136                                                 register_count,
5137                                                 register_start,
5138                                                 on_success)));
5139     return node;
5140   } else {
5141     // We use a ChoiceNode for a negative lookahead because it has most of
5142     // the characteristics we need.  It has the body of the lookahead as its
5143     // first alternative and the expression after the lookahead of the second
5144     // alternative.  If the first alternative succeeds then the
5145     // NegativeSubmatchSuccess will unwind the stack including everything the
5146     // choice node set up and backtrack.  If the first alternative fails then
5147     // the second alternative is tried, which is exactly the desired result
5148     // for a negative lookahead.  The NegativeLookaheadChoiceNode is a special
5149     // ChoiceNode that knows to ignore the first exit when calculating quick
5150     // checks.
5151     Zone* zone = compiler->zone();
5152 
5153     GuardedAlternative body_alt(
5154         body()->ToNode(
5155             compiler,
5156             success = new(zone) NegativeSubmatchSuccess(stack_pointer_register,
5157                                                         position_register,
5158                                                         register_count,
5159                                                         register_start,
5160                                                         zone)));
5161     ChoiceNode* choice_node =
5162         new(zone) NegativeLookaheadChoiceNode(body_alt,
5163                                               GuardedAlternative(on_success),
5164                                               zone);
5165     return ActionNode::BeginSubmatch(stack_pointer_register,
5166                                      position_register,
5167                                      choice_node);
5168   }
5169 }
5170 
5171 
ToNode(RegExpCompiler * compiler,RegExpNode * on_success)5172 RegExpNode* RegExpCapture::ToNode(RegExpCompiler* compiler,
5173                                   RegExpNode* on_success) {
5174   return ToNode(body(), index(), compiler, on_success);
5175 }
5176 
5177 
ToNode(RegExpTree * body,int index,RegExpCompiler * compiler,RegExpNode * on_success)5178 RegExpNode* RegExpCapture::ToNode(RegExpTree* body,
5179                                   int index,
5180                                   RegExpCompiler* compiler,
5181                                   RegExpNode* on_success) {
5182   int start_reg = RegExpCapture::StartRegister(index);
5183   int end_reg = RegExpCapture::EndRegister(index);
5184   RegExpNode* store_end = ActionNode::StorePosition(end_reg, true, on_success);
5185   RegExpNode* body_node = body->ToNode(compiler, store_end);
5186   return ActionNode::StorePosition(start_reg, true, body_node);
5187 }
5188 
5189 
ToNode(RegExpCompiler * compiler,RegExpNode * on_success)5190 RegExpNode* RegExpAlternative::ToNode(RegExpCompiler* compiler,
5191                                       RegExpNode* on_success) {
5192   ZoneList<RegExpTree*>* children = nodes();
5193   RegExpNode* current = on_success;
5194   for (int i = children->length() - 1; i >= 0; i--) {
5195     current = children->at(i)->ToNode(compiler, current);
5196   }
5197   return current;
5198 }
5199 
5200 
AddClass(const int * elmv,int elmc,ZoneList<CharacterRange> * ranges,Zone * zone)5201 static void AddClass(const int* elmv,
5202                      int elmc,
5203                      ZoneList<CharacterRange>* ranges,
5204                      Zone* zone) {
5205   elmc--;
5206   DCHECK(elmv[elmc] == 0x10000);
5207   for (int i = 0; i < elmc; i += 2) {
5208     DCHECK(elmv[i] < elmv[i + 1]);
5209     ranges->Add(CharacterRange(elmv[i], elmv[i + 1] - 1), zone);
5210   }
5211 }
5212 
5213 
AddClassNegated(const int * elmv,int elmc,ZoneList<CharacterRange> * ranges,Zone * zone)5214 static void AddClassNegated(const int *elmv,
5215                             int elmc,
5216                             ZoneList<CharacterRange>* ranges,
5217                             Zone* zone) {
5218   elmc--;
5219   DCHECK(elmv[elmc] == 0x10000);
5220   DCHECK(elmv[0] != 0x0000);
5221   DCHECK(elmv[elmc-1] != String::kMaxUtf16CodeUnit);
5222   uc16 last = 0x0000;
5223   for (int i = 0; i < elmc; i += 2) {
5224     DCHECK(last <= elmv[i] - 1);
5225     DCHECK(elmv[i] < elmv[i + 1]);
5226     ranges->Add(CharacterRange(last, elmv[i] - 1), zone);
5227     last = elmv[i + 1];
5228   }
5229   ranges->Add(CharacterRange(last, String::kMaxUtf16CodeUnit), zone);
5230 }
5231 
5232 
AddClassEscape(uc16 type,ZoneList<CharacterRange> * ranges,Zone * zone)5233 void CharacterRange::AddClassEscape(uc16 type,
5234                                     ZoneList<CharacterRange>* ranges,
5235                                     Zone* zone) {
5236   switch (type) {
5237     case 's':
5238       AddClass(kSpaceRanges, kSpaceRangeCount, ranges, zone);
5239       break;
5240     case 'S':
5241       AddClassNegated(kSpaceRanges, kSpaceRangeCount, ranges, zone);
5242       break;
5243     case 'w':
5244       AddClass(kWordRanges, kWordRangeCount, ranges, zone);
5245       break;
5246     case 'W':
5247       AddClassNegated(kWordRanges, kWordRangeCount, ranges, zone);
5248       break;
5249     case 'd':
5250       AddClass(kDigitRanges, kDigitRangeCount, ranges, zone);
5251       break;
5252     case 'D':
5253       AddClassNegated(kDigitRanges, kDigitRangeCount, ranges, zone);
5254       break;
5255     case '.':
5256       AddClassNegated(kLineTerminatorRanges,
5257                       kLineTerminatorRangeCount,
5258                       ranges,
5259                       zone);
5260       break;
5261     // This is not a character range as defined by the spec but a
5262     // convenient shorthand for a character class that matches any
5263     // character.
5264     case '*':
5265       ranges->Add(CharacterRange::Everything(), zone);
5266       break;
5267     // This is the set of characters matched by the $ and ^ symbols
5268     // in multiline mode.
5269     case 'n':
5270       AddClass(kLineTerminatorRanges,
5271                kLineTerminatorRangeCount,
5272                ranges,
5273                zone);
5274       break;
5275     default:
5276       UNREACHABLE();
5277   }
5278 }
5279 
5280 
GetWordBounds()5281 Vector<const int> CharacterRange::GetWordBounds() {
5282   return Vector<const int>(kWordRanges, kWordRangeCount - 1);
5283 }
5284 
5285 
5286 class CharacterRangeSplitter {
5287  public:
CharacterRangeSplitter(ZoneList<CharacterRange> ** included,ZoneList<CharacterRange> ** excluded,Zone * zone)5288   CharacterRangeSplitter(ZoneList<CharacterRange>** included,
5289                          ZoneList<CharacterRange>** excluded,
5290                          Zone* zone)
5291       : included_(included),
5292         excluded_(excluded),
5293         zone_(zone) { }
5294   void Call(uc16 from, DispatchTable::Entry entry);
5295 
5296   static const int kInBase = 0;
5297   static const int kInOverlay = 1;
5298 
5299  private:
5300   ZoneList<CharacterRange>** included_;
5301   ZoneList<CharacterRange>** excluded_;
5302   Zone* zone_;
5303 };
5304 
5305 
Call(uc16 from,DispatchTable::Entry entry)5306 void CharacterRangeSplitter::Call(uc16 from, DispatchTable::Entry entry) {
5307   if (!entry.out_set()->Get(kInBase)) return;
5308   ZoneList<CharacterRange>** target = entry.out_set()->Get(kInOverlay)
5309     ? included_
5310     : excluded_;
5311   if (*target == NULL) *target = new(zone_) ZoneList<CharacterRange>(2, zone_);
5312   (*target)->Add(CharacterRange(entry.from(), entry.to()), zone_);
5313 }
5314 
5315 
Split(ZoneList<CharacterRange> * base,Vector<const int> overlay,ZoneList<CharacterRange> ** included,ZoneList<CharacterRange> ** excluded,Zone * zone)5316 void CharacterRange::Split(ZoneList<CharacterRange>* base,
5317                            Vector<const int> overlay,
5318                            ZoneList<CharacterRange>** included,
5319                            ZoneList<CharacterRange>** excluded,
5320                            Zone* zone) {
5321   DCHECK_EQ(NULL, *included);
5322   DCHECK_EQ(NULL, *excluded);
5323   DispatchTable table(zone);
5324   for (int i = 0; i < base->length(); i++)
5325     table.AddRange(base->at(i), CharacterRangeSplitter::kInBase, zone);
5326   for (int i = 0; i < overlay.length(); i += 2) {
5327     table.AddRange(CharacterRange(overlay[i], overlay[i + 1] - 1),
5328                    CharacterRangeSplitter::kInOverlay, zone);
5329   }
5330   CharacterRangeSplitter callback(included, excluded, zone);
5331   table.ForEach(&callback);
5332 }
5333 
5334 
AddCaseEquivalents(ZoneList<CharacterRange> * ranges,bool is_one_byte,Zone * zone)5335 void CharacterRange::AddCaseEquivalents(ZoneList<CharacterRange>* ranges,
5336                                         bool is_one_byte, Zone* zone) {
5337   Isolate* isolate = zone->isolate();
5338   uc16 bottom = from();
5339   uc16 top = to();
5340   if (is_one_byte && !RangeContainsLatin1Equivalents(*this)) {
5341     if (bottom > String::kMaxOneByteCharCode) return;
5342     if (top > String::kMaxOneByteCharCode) top = String::kMaxOneByteCharCode;
5343   }
5344   unibrow::uchar chars[unibrow::Ecma262UnCanonicalize::kMaxWidth];
5345   if (top == bottom) {
5346     // If this is a singleton we just expand the one character.
5347     int length = isolate->jsregexp_uncanonicalize()->get(bottom, '\0', chars);
5348     for (int i = 0; i < length; i++) {
5349       uc32 chr = chars[i];
5350       if (chr != bottom) {
5351         ranges->Add(CharacterRange::Singleton(chars[i]), zone);
5352       }
5353     }
5354   } else {
5355     // If this is a range we expand the characters block by block,
5356     // expanding contiguous subranges (blocks) one at a time.
5357     // The approach is as follows.  For a given start character we
5358     // look up the remainder of the block that contains it (represented
5359     // by the end point), for instance we find 'z' if the character
5360     // is 'c'.  A block is characterized by the property
5361     // that all characters uncanonicalize in the same way, except that
5362     // each entry in the result is incremented by the distance from the first
5363     // element.  So a-z is a block because 'a' uncanonicalizes to ['a', 'A'] and
5364     // the k'th letter uncanonicalizes to ['a' + k, 'A' + k].
5365     // Once we've found the end point we look up its uncanonicalization
5366     // and produce a range for each element.  For instance for [c-f]
5367     // we look up ['z', 'Z'] and produce [c-f] and [C-F].  We then only
5368     // add a range if it is not already contained in the input, so [c-f]
5369     // will be skipped but [C-F] will be added.  If this range is not
5370     // completely contained in a block we do this for all the blocks
5371     // covered by the range (handling characters that is not in a block
5372     // as a "singleton block").
5373     unibrow::uchar range[unibrow::Ecma262UnCanonicalize::kMaxWidth];
5374     int pos = bottom;
5375     while (pos <= top) {
5376       int length = isolate->jsregexp_canonrange()->get(pos, '\0', range);
5377       uc16 block_end;
5378       if (length == 0) {
5379         block_end = pos;
5380       } else {
5381         DCHECK_EQ(1, length);
5382         block_end = range[0];
5383       }
5384       int end = (block_end > top) ? top : block_end;
5385       length = isolate->jsregexp_uncanonicalize()->get(block_end, '\0', range);
5386       for (int i = 0; i < length; i++) {
5387         uc32 c = range[i];
5388         uc16 range_from = c - (block_end - pos);
5389         uc16 range_to = c - (block_end - end);
5390         if (!(bottom <= range_from && range_to <= top)) {
5391           ranges->Add(CharacterRange(range_from, range_to), zone);
5392         }
5393       }
5394       pos = end + 1;
5395     }
5396   }
5397 }
5398 
5399 
IsCanonical(ZoneList<CharacterRange> * ranges)5400 bool CharacterRange::IsCanonical(ZoneList<CharacterRange>* ranges) {
5401   DCHECK_NOT_NULL(ranges);
5402   int n = ranges->length();
5403   if (n <= 1) return true;
5404   int max = ranges->at(0).to();
5405   for (int i = 1; i < n; i++) {
5406     CharacterRange next_range = ranges->at(i);
5407     if (next_range.from() <= max + 1) return false;
5408     max = next_range.to();
5409   }
5410   return true;
5411 }
5412 
5413 
ranges(Zone * zone)5414 ZoneList<CharacterRange>* CharacterSet::ranges(Zone* zone) {
5415   if (ranges_ == NULL) {
5416     ranges_ = new(zone) ZoneList<CharacterRange>(2, zone);
5417     CharacterRange::AddClassEscape(standard_set_type_, ranges_, zone);
5418   }
5419   return ranges_;
5420 }
5421 
5422 
5423 // Move a number of elements in a zonelist to another position
5424 // in the same list. Handles overlapping source and target areas.
MoveRanges(ZoneList<CharacterRange> * list,int from,int to,int count)5425 static void MoveRanges(ZoneList<CharacterRange>* list,
5426                        int from,
5427                        int to,
5428                        int count) {
5429   // Ranges are potentially overlapping.
5430   if (from < to) {
5431     for (int i = count - 1; i >= 0; i--) {
5432       list->at(to + i) = list->at(from + i);
5433     }
5434   } else {
5435     for (int i = 0; i < count; i++) {
5436       list->at(to + i) = list->at(from + i);
5437     }
5438   }
5439 }
5440 
5441 
InsertRangeInCanonicalList(ZoneList<CharacterRange> * list,int count,CharacterRange insert)5442 static int InsertRangeInCanonicalList(ZoneList<CharacterRange>* list,
5443                                       int count,
5444                                       CharacterRange insert) {
5445   // Inserts a range into list[0..count[, which must be sorted
5446   // by from value and non-overlapping and non-adjacent, using at most
5447   // list[0..count] for the result. Returns the number of resulting
5448   // canonicalized ranges. Inserting a range may collapse existing ranges into
5449   // fewer ranges, so the return value can be anything in the range 1..count+1.
5450   uc16 from = insert.from();
5451   uc16 to = insert.to();
5452   int start_pos = 0;
5453   int end_pos = count;
5454   for (int i = count - 1; i >= 0; i--) {
5455     CharacterRange current = list->at(i);
5456     if (current.from() > to + 1) {
5457       end_pos = i;
5458     } else if (current.to() + 1 < from) {
5459       start_pos = i + 1;
5460       break;
5461     }
5462   }
5463 
5464   // Inserted range overlaps, or is adjacent to, ranges at positions
5465   // [start_pos..end_pos[. Ranges before start_pos or at or after end_pos are
5466   // not affected by the insertion.
5467   // If start_pos == end_pos, the range must be inserted before start_pos.
5468   // if start_pos < end_pos, the entire range from start_pos to end_pos
5469   // must be merged with the insert range.
5470 
5471   if (start_pos == end_pos) {
5472     // Insert between existing ranges at position start_pos.
5473     if (start_pos < count) {
5474       MoveRanges(list, start_pos, start_pos + 1, count - start_pos);
5475     }
5476     list->at(start_pos) = insert;
5477     return count + 1;
5478   }
5479   if (start_pos + 1 == end_pos) {
5480     // Replace single existing range at position start_pos.
5481     CharacterRange to_replace = list->at(start_pos);
5482     int new_from = Min(to_replace.from(), from);
5483     int new_to = Max(to_replace.to(), to);
5484     list->at(start_pos) = CharacterRange(new_from, new_to);
5485     return count;
5486   }
5487   // Replace a number of existing ranges from start_pos to end_pos - 1.
5488   // Move the remaining ranges down.
5489 
5490   int new_from = Min(list->at(start_pos).from(), from);
5491   int new_to = Max(list->at(end_pos - 1).to(), to);
5492   if (end_pos < count) {
5493     MoveRanges(list, end_pos, start_pos + 1, count - end_pos);
5494   }
5495   list->at(start_pos) = CharacterRange(new_from, new_to);
5496   return count - (end_pos - start_pos) + 1;
5497 }
5498 
5499 
Canonicalize()5500 void CharacterSet::Canonicalize() {
5501   // Special/default classes are always considered canonical. The result
5502   // of calling ranges() will be sorted.
5503   if (ranges_ == NULL) return;
5504   CharacterRange::Canonicalize(ranges_);
5505 }
5506 
5507 
Canonicalize(ZoneList<CharacterRange> * character_ranges)5508 void CharacterRange::Canonicalize(ZoneList<CharacterRange>* character_ranges) {
5509   if (character_ranges->length() <= 1) return;
5510   // Check whether ranges are already canonical (increasing, non-overlapping,
5511   // non-adjacent).
5512   int n = character_ranges->length();
5513   int max = character_ranges->at(0).to();
5514   int i = 1;
5515   while (i < n) {
5516     CharacterRange current = character_ranges->at(i);
5517     if (current.from() <= max + 1) {
5518       break;
5519     }
5520     max = current.to();
5521     i++;
5522   }
5523   // Canonical until the i'th range. If that's all of them, we are done.
5524   if (i == n) return;
5525 
5526   // The ranges at index i and forward are not canonicalized. Make them so by
5527   // doing the equivalent of insertion sort (inserting each into the previous
5528   // list, in order).
5529   // Notice that inserting a range can reduce the number of ranges in the
5530   // result due to combining of adjacent and overlapping ranges.
5531   int read = i;  // Range to insert.
5532   int num_canonical = i;  // Length of canonicalized part of list.
5533   do {
5534     num_canonical = InsertRangeInCanonicalList(character_ranges,
5535                                                num_canonical,
5536                                                character_ranges->at(read));
5537     read++;
5538   } while (read < n);
5539   character_ranges->Rewind(num_canonical);
5540 
5541   DCHECK(CharacterRange::IsCanonical(character_ranges));
5542 }
5543 
5544 
Negate(ZoneList<CharacterRange> * ranges,ZoneList<CharacterRange> * negated_ranges,Zone * zone)5545 void CharacterRange::Negate(ZoneList<CharacterRange>* ranges,
5546                             ZoneList<CharacterRange>* negated_ranges,
5547                             Zone* zone) {
5548   DCHECK(CharacterRange::IsCanonical(ranges));
5549   DCHECK_EQ(0, negated_ranges->length());
5550   int range_count = ranges->length();
5551   uc16 from = 0;
5552   int i = 0;
5553   if (range_count > 0 && ranges->at(0).from() == 0) {
5554     from = ranges->at(0).to();
5555     i = 1;
5556   }
5557   while (i < range_count) {
5558     CharacterRange range = ranges->at(i);
5559     negated_ranges->Add(CharacterRange(from + 1, range.from() - 1), zone);
5560     from = range.to();
5561     i++;
5562   }
5563   if (from < String::kMaxUtf16CodeUnit) {
5564     negated_ranges->Add(CharacterRange(from + 1, String::kMaxUtf16CodeUnit),
5565                         zone);
5566   }
5567 }
5568 
5569 
5570 // -------------------------------------------------------------------
5571 // Splay tree
5572 
5573 
Extend(unsigned value,Zone * zone)5574 OutSet* OutSet::Extend(unsigned value, Zone* zone) {
5575   if (Get(value))
5576     return this;
5577   if (successors(zone) != NULL) {
5578     for (int i = 0; i < successors(zone)->length(); i++) {
5579       OutSet* successor = successors(zone)->at(i);
5580       if (successor->Get(value))
5581         return successor;
5582     }
5583   } else {
5584     successors_ = new(zone) ZoneList<OutSet*>(2, zone);
5585   }
5586   OutSet* result = new(zone) OutSet(first_, remaining_);
5587   result->Set(value, zone);
5588   successors(zone)->Add(result, zone);
5589   return result;
5590 }
5591 
5592 
Set(unsigned value,Zone * zone)5593 void OutSet::Set(unsigned value, Zone *zone) {
5594   if (value < kFirstLimit) {
5595     first_ |= (1 << value);
5596   } else {
5597     if (remaining_ == NULL)
5598       remaining_ = new(zone) ZoneList<unsigned>(1, zone);
5599     if (remaining_->is_empty() || !remaining_->Contains(value))
5600       remaining_->Add(value, zone);
5601   }
5602 }
5603 
5604 
Get(unsigned value) const5605 bool OutSet::Get(unsigned value) const {
5606   if (value < kFirstLimit) {
5607     return (first_ & (1 << value)) != 0;
5608   } else if (remaining_ == NULL) {
5609     return false;
5610   } else {
5611     return remaining_->Contains(value);
5612   }
5613 }
5614 
5615 
5616 const uc16 DispatchTable::Config::kNoKey = unibrow::Utf8::kBadChar;
5617 
5618 
AddRange(CharacterRange full_range,int value,Zone * zone)5619 void DispatchTable::AddRange(CharacterRange full_range, int value,
5620                              Zone* zone) {
5621   CharacterRange current = full_range;
5622   if (tree()->is_empty()) {
5623     // If this is the first range we just insert into the table.
5624     ZoneSplayTree<Config>::Locator loc;
5625     DCHECK_RESULT(tree()->Insert(current.from(), &loc));
5626     loc.set_value(Entry(current.from(), current.to(),
5627                         empty()->Extend(value, zone)));
5628     return;
5629   }
5630   // First see if there is a range to the left of this one that
5631   // overlaps.
5632   ZoneSplayTree<Config>::Locator loc;
5633   if (tree()->FindGreatestLessThan(current.from(), &loc)) {
5634     Entry* entry = &loc.value();
5635     // If we've found a range that overlaps with this one, and it
5636     // starts strictly to the left of this one, we have to fix it
5637     // because the following code only handles ranges that start on
5638     // or after the start point of the range we're adding.
5639     if (entry->from() < current.from() && entry->to() >= current.from()) {
5640       // Snap the overlapping range in half around the start point of
5641       // the range we're adding.
5642       CharacterRange left(entry->from(), current.from() - 1);
5643       CharacterRange right(current.from(), entry->to());
5644       // The left part of the overlapping range doesn't overlap.
5645       // Truncate the whole entry to be just the left part.
5646       entry->set_to(left.to());
5647       // The right part is the one that overlaps.  We add this part
5648       // to the map and let the next step deal with merging it with
5649       // the range we're adding.
5650       ZoneSplayTree<Config>::Locator loc;
5651       DCHECK_RESULT(tree()->Insert(right.from(), &loc));
5652       loc.set_value(Entry(right.from(),
5653                           right.to(),
5654                           entry->out_set()));
5655     }
5656   }
5657   while (current.is_valid()) {
5658     if (tree()->FindLeastGreaterThan(current.from(), &loc) &&
5659         (loc.value().from() <= current.to()) &&
5660         (loc.value().to() >= current.from())) {
5661       Entry* entry = &loc.value();
5662       // We have overlap.  If there is space between the start point of
5663       // the range we're adding and where the overlapping range starts
5664       // then we have to add a range covering just that space.
5665       if (current.from() < entry->from()) {
5666         ZoneSplayTree<Config>::Locator ins;
5667         DCHECK_RESULT(tree()->Insert(current.from(), &ins));
5668         ins.set_value(Entry(current.from(),
5669                             entry->from() - 1,
5670                             empty()->Extend(value, zone)));
5671         current.set_from(entry->from());
5672       }
5673       DCHECK_EQ(current.from(), entry->from());
5674       // If the overlapping range extends beyond the one we want to add
5675       // we have to snap the right part off and add it separately.
5676       if (entry->to() > current.to()) {
5677         ZoneSplayTree<Config>::Locator ins;
5678         DCHECK_RESULT(tree()->Insert(current.to() + 1, &ins));
5679         ins.set_value(Entry(current.to() + 1,
5680                             entry->to(),
5681                             entry->out_set()));
5682         entry->set_to(current.to());
5683       }
5684       DCHECK(entry->to() <= current.to());
5685       // The overlapping range is now completely contained by the range
5686       // we're adding so we can just update it and move the start point
5687       // of the range we're adding just past it.
5688       entry->AddValue(value, zone);
5689       // Bail out if the last interval ended at 0xFFFF since otherwise
5690       // adding 1 will wrap around to 0.
5691       if (entry->to() == String::kMaxUtf16CodeUnit)
5692         break;
5693       DCHECK(entry->to() + 1 > current.from());
5694       current.set_from(entry->to() + 1);
5695     } else {
5696       // There is no overlap so we can just add the range
5697       ZoneSplayTree<Config>::Locator ins;
5698       DCHECK_RESULT(tree()->Insert(current.from(), &ins));
5699       ins.set_value(Entry(current.from(),
5700                           current.to(),
5701                           empty()->Extend(value, zone)));
5702       break;
5703     }
5704   }
5705 }
5706 
5707 
Get(uc16 value)5708 OutSet* DispatchTable::Get(uc16 value) {
5709   ZoneSplayTree<Config>::Locator loc;
5710   if (!tree()->FindGreatestLessThan(value, &loc))
5711     return empty();
5712   Entry* entry = &loc.value();
5713   if (value <= entry->to())
5714     return entry->out_set();
5715   else
5716     return empty();
5717 }
5718 
5719 
5720 // -------------------------------------------------------------------
5721 // Analysis
5722 
5723 
EnsureAnalyzed(RegExpNode * that)5724 void Analysis::EnsureAnalyzed(RegExpNode* that) {
5725   StackLimitCheck check(that->zone()->isolate());
5726   if (check.HasOverflowed()) {
5727     fail("Stack overflow");
5728     return;
5729   }
5730   if (that->info()->been_analyzed || that->info()->being_analyzed)
5731     return;
5732   that->info()->being_analyzed = true;
5733   that->Accept(this);
5734   that->info()->being_analyzed = false;
5735   that->info()->been_analyzed = true;
5736 }
5737 
5738 
VisitEnd(EndNode * that)5739 void Analysis::VisitEnd(EndNode* that) {
5740   // nothing to do
5741 }
5742 
5743 
CalculateOffsets()5744 void TextNode::CalculateOffsets() {
5745   int element_count = elements()->length();
5746   // Set up the offsets of the elements relative to the start.  This is a fixed
5747   // quantity since a TextNode can only contain fixed-width things.
5748   int cp_offset = 0;
5749   for (int i = 0; i < element_count; i++) {
5750     TextElement& elm = elements()->at(i);
5751     elm.set_cp_offset(cp_offset);
5752     cp_offset += elm.length();
5753   }
5754 }
5755 
5756 
VisitText(TextNode * that)5757 void Analysis::VisitText(TextNode* that) {
5758   if (ignore_case_) {
5759     that->MakeCaseIndependent(is_one_byte_);
5760   }
5761   EnsureAnalyzed(that->on_success());
5762   if (!has_failed()) {
5763     that->CalculateOffsets();
5764   }
5765 }
5766 
5767 
VisitAction(ActionNode * that)5768 void Analysis::VisitAction(ActionNode* that) {
5769   RegExpNode* target = that->on_success();
5770   EnsureAnalyzed(target);
5771   if (!has_failed()) {
5772     // If the next node is interested in what it follows then this node
5773     // has to be interested too so it can pass the information on.
5774     that->info()->AddFromFollowing(target->info());
5775   }
5776 }
5777 
5778 
VisitChoice(ChoiceNode * that)5779 void Analysis::VisitChoice(ChoiceNode* that) {
5780   NodeInfo* info = that->info();
5781   for (int i = 0; i < that->alternatives()->length(); i++) {
5782     RegExpNode* node = that->alternatives()->at(i).node();
5783     EnsureAnalyzed(node);
5784     if (has_failed()) return;
5785     // Anything the following nodes need to know has to be known by
5786     // this node also, so it can pass it on.
5787     info->AddFromFollowing(node->info());
5788   }
5789 }
5790 
5791 
VisitLoopChoice(LoopChoiceNode * that)5792 void Analysis::VisitLoopChoice(LoopChoiceNode* that) {
5793   NodeInfo* info = that->info();
5794   for (int i = 0; i < that->alternatives()->length(); i++) {
5795     RegExpNode* node = that->alternatives()->at(i).node();
5796     if (node != that->loop_node()) {
5797       EnsureAnalyzed(node);
5798       if (has_failed()) return;
5799       info->AddFromFollowing(node->info());
5800     }
5801   }
5802   // Check the loop last since it may need the value of this node
5803   // to get a correct result.
5804   EnsureAnalyzed(that->loop_node());
5805   if (!has_failed()) {
5806     info->AddFromFollowing(that->loop_node()->info());
5807   }
5808 }
5809 
5810 
VisitBackReference(BackReferenceNode * that)5811 void Analysis::VisitBackReference(BackReferenceNode* that) {
5812   EnsureAnalyzed(that->on_success());
5813 }
5814 
5815 
VisitAssertion(AssertionNode * that)5816 void Analysis::VisitAssertion(AssertionNode* that) {
5817   EnsureAnalyzed(that->on_success());
5818 }
5819 
5820 
FillInBMInfo(int offset,int budget,BoyerMooreLookahead * bm,bool not_at_start)5821 void BackReferenceNode::FillInBMInfo(int offset,
5822                                      int budget,
5823                                      BoyerMooreLookahead* bm,
5824                                      bool not_at_start) {
5825   // Working out the set of characters that a backreference can match is too
5826   // hard, so we just say that any character can match.
5827   bm->SetRest(offset);
5828   SaveBMInfo(bm, not_at_start, offset);
5829 }
5830 
5831 
5832 STATIC_ASSERT(BoyerMoorePositionInfo::kMapSize ==
5833               RegExpMacroAssembler::kTableSize);
5834 
5835 
FillInBMInfo(int offset,int budget,BoyerMooreLookahead * bm,bool not_at_start)5836 void ChoiceNode::FillInBMInfo(int offset,
5837                               int budget,
5838                               BoyerMooreLookahead* bm,
5839                               bool not_at_start) {
5840   ZoneList<GuardedAlternative>* alts = alternatives();
5841   budget = (budget - 1) / alts->length();
5842   for (int i = 0; i < alts->length(); i++) {
5843     GuardedAlternative& alt = alts->at(i);
5844     if (alt.guards() != NULL && alt.guards()->length() != 0) {
5845       bm->SetRest(offset);  // Give up trying to fill in info.
5846       SaveBMInfo(bm, not_at_start, offset);
5847       return;
5848     }
5849     alt.node()->FillInBMInfo(offset, budget, bm, not_at_start);
5850   }
5851   SaveBMInfo(bm, not_at_start, offset);
5852 }
5853 
5854 
FillInBMInfo(int initial_offset,int budget,BoyerMooreLookahead * bm,bool not_at_start)5855 void TextNode::FillInBMInfo(int initial_offset,
5856                             int budget,
5857                             BoyerMooreLookahead* bm,
5858                             bool not_at_start) {
5859   if (initial_offset >= bm->length()) return;
5860   int offset = initial_offset;
5861   int max_char = bm->max_char();
5862   for (int i = 0; i < elements()->length(); i++) {
5863     if (offset >= bm->length()) {
5864       if (initial_offset == 0) set_bm_info(not_at_start, bm);
5865       return;
5866     }
5867     TextElement text = elements()->at(i);
5868     if (text.text_type() == TextElement::ATOM) {
5869       RegExpAtom* atom = text.atom();
5870       for (int j = 0; j < atom->length(); j++, offset++) {
5871         if (offset >= bm->length()) {
5872           if (initial_offset == 0) set_bm_info(not_at_start, bm);
5873           return;
5874         }
5875         uc16 character = atom->data()[j];
5876         if (bm->compiler()->ignore_case()) {
5877           unibrow::uchar chars[unibrow::Ecma262UnCanonicalize::kMaxWidth];
5878           int length = GetCaseIndependentLetters(
5879               Isolate::Current(),
5880               character,
5881               bm->max_char() == String::kMaxOneByteCharCode,
5882               chars);
5883           for (int j = 0; j < length; j++) {
5884             bm->Set(offset, chars[j]);
5885           }
5886         } else {
5887           if (character <= max_char) bm->Set(offset, character);
5888         }
5889       }
5890     } else {
5891       DCHECK_EQ(TextElement::CHAR_CLASS, text.text_type());
5892       RegExpCharacterClass* char_class = text.char_class();
5893       ZoneList<CharacterRange>* ranges = char_class->ranges(zone());
5894       if (char_class->is_negated()) {
5895         bm->SetAll(offset);
5896       } else {
5897         for (int k = 0; k < ranges->length(); k++) {
5898           CharacterRange& range = ranges->at(k);
5899           if (range.from() > max_char) continue;
5900           int to = Min(max_char, static_cast<int>(range.to()));
5901           bm->SetInterval(offset, Interval(range.from(), to));
5902         }
5903       }
5904       offset++;
5905     }
5906   }
5907   if (offset >= bm->length()) {
5908     if (initial_offset == 0) set_bm_info(not_at_start, bm);
5909     return;
5910   }
5911   on_success()->FillInBMInfo(offset,
5912                              budget - 1,
5913                              bm,
5914                              true);  // Not at start after a text node.
5915   if (initial_offset == 0) set_bm_info(not_at_start, bm);
5916 }
5917 
5918 
5919 // -------------------------------------------------------------------
5920 // Dispatch table construction
5921 
5922 
VisitEnd(EndNode * that)5923 void DispatchTableConstructor::VisitEnd(EndNode* that) {
5924   AddRange(CharacterRange::Everything());
5925 }
5926 
5927 
BuildTable(ChoiceNode * node)5928 void DispatchTableConstructor::BuildTable(ChoiceNode* node) {
5929   node->set_being_calculated(true);
5930   ZoneList<GuardedAlternative>* alternatives = node->alternatives();
5931   for (int i = 0; i < alternatives->length(); i++) {
5932     set_choice_index(i);
5933     alternatives->at(i).node()->Accept(this);
5934   }
5935   node->set_being_calculated(false);
5936 }
5937 
5938 
5939 class AddDispatchRange {
5940  public:
AddDispatchRange(DispatchTableConstructor * constructor)5941   explicit AddDispatchRange(DispatchTableConstructor* constructor)
5942     : constructor_(constructor) { }
5943   void Call(uc32 from, DispatchTable::Entry entry);
5944  private:
5945   DispatchTableConstructor* constructor_;
5946 };
5947 
5948 
Call(uc32 from,DispatchTable::Entry entry)5949 void AddDispatchRange::Call(uc32 from, DispatchTable::Entry entry) {
5950   CharacterRange range(from, entry.to());
5951   constructor_->AddRange(range);
5952 }
5953 
5954 
VisitChoice(ChoiceNode * node)5955 void DispatchTableConstructor::VisitChoice(ChoiceNode* node) {
5956   if (node->being_calculated())
5957     return;
5958   DispatchTable* table = node->GetTable(ignore_case_);
5959   AddDispatchRange adder(this);
5960   table->ForEach(&adder);
5961 }
5962 
5963 
VisitBackReference(BackReferenceNode * that)5964 void DispatchTableConstructor::VisitBackReference(BackReferenceNode* that) {
5965   // TODO(160): Find the node that we refer back to and propagate its start
5966   // set back to here.  For now we just accept anything.
5967   AddRange(CharacterRange::Everything());
5968 }
5969 
5970 
VisitAssertion(AssertionNode * that)5971 void DispatchTableConstructor::VisitAssertion(AssertionNode* that) {
5972   RegExpNode* target = that->on_success();
5973   target->Accept(this);
5974 }
5975 
5976 
CompareRangeByFrom(const CharacterRange * a,const CharacterRange * b)5977 static int CompareRangeByFrom(const CharacterRange* a,
5978                               const CharacterRange* b) {
5979   return Compare<uc16>(a->from(), b->from());
5980 }
5981 
5982 
AddInverse(ZoneList<CharacterRange> * ranges)5983 void DispatchTableConstructor::AddInverse(ZoneList<CharacterRange>* ranges) {
5984   ranges->Sort(CompareRangeByFrom);
5985   uc16 last = 0;
5986   for (int i = 0; i < ranges->length(); i++) {
5987     CharacterRange range = ranges->at(i);
5988     if (last < range.from())
5989       AddRange(CharacterRange(last, range.from() - 1));
5990     if (range.to() >= last) {
5991       if (range.to() == String::kMaxUtf16CodeUnit) {
5992         return;
5993       } else {
5994         last = range.to() + 1;
5995       }
5996     }
5997   }
5998   AddRange(CharacterRange(last, String::kMaxUtf16CodeUnit));
5999 }
6000 
6001 
VisitText(TextNode * that)6002 void DispatchTableConstructor::VisitText(TextNode* that) {
6003   TextElement elm = that->elements()->at(0);
6004   switch (elm.text_type()) {
6005     case TextElement::ATOM: {
6006       uc16 c = elm.atom()->data()[0];
6007       AddRange(CharacterRange(c, c));
6008       break;
6009     }
6010     case TextElement::CHAR_CLASS: {
6011       RegExpCharacterClass* tree = elm.char_class();
6012       ZoneList<CharacterRange>* ranges = tree->ranges(that->zone());
6013       if (tree->is_negated()) {
6014         AddInverse(ranges);
6015       } else {
6016         for (int i = 0; i < ranges->length(); i++)
6017           AddRange(ranges->at(i));
6018       }
6019       break;
6020     }
6021     default: {
6022       UNIMPLEMENTED();
6023     }
6024   }
6025 }
6026 
6027 
VisitAction(ActionNode * that)6028 void DispatchTableConstructor::VisitAction(ActionNode* that) {
6029   RegExpNode* target = that->on_success();
6030   target->Accept(this);
6031 }
6032 
6033 
Compile(RegExpCompileData * data,bool ignore_case,bool is_global,bool is_multiline,bool is_sticky,Handle<String> pattern,Handle<String> sample_subject,bool is_one_byte,Zone * zone)6034 RegExpEngine::CompilationResult RegExpEngine::Compile(
6035     RegExpCompileData* data, bool ignore_case, bool is_global,
6036     bool is_multiline, bool is_sticky, Handle<String> pattern,
6037     Handle<String> sample_subject, bool is_one_byte, Zone* zone) {
6038   if ((data->capture_count + 1) * 2 - 1 > RegExpMacroAssembler::kMaxRegister) {
6039     return IrregexpRegExpTooBig(zone->isolate());
6040   }
6041   RegExpCompiler compiler(data->capture_count, ignore_case, is_one_byte, zone);
6042 
6043   // Sample some characters from the middle of the string.
6044   static const int kSampleSize = 128;
6045 
6046   sample_subject = String::Flatten(sample_subject);
6047   int chars_sampled = 0;
6048   int half_way = (sample_subject->length() - kSampleSize) / 2;
6049   for (int i = Max(0, half_way);
6050        i < sample_subject->length() && chars_sampled < kSampleSize;
6051        i++, chars_sampled++) {
6052     compiler.frequency_collator()->CountCharacter(sample_subject->Get(i));
6053   }
6054 
6055   // Wrap the body of the regexp in capture #0.
6056   RegExpNode* captured_body = RegExpCapture::ToNode(data->tree,
6057                                                     0,
6058                                                     &compiler,
6059                                                     compiler.accept());
6060   RegExpNode* node = captured_body;
6061   bool is_end_anchored = data->tree->IsAnchoredAtEnd();
6062   bool is_start_anchored = data->tree->IsAnchoredAtStart();
6063   int max_length = data->tree->max_match();
6064   if (!is_start_anchored && !is_sticky) {
6065     // Add a .*? at the beginning, outside the body capture, unless
6066     // this expression is anchored at the beginning or sticky.
6067     RegExpNode* loop_node =
6068         RegExpQuantifier::ToNode(0,
6069                                  RegExpTree::kInfinity,
6070                                  false,
6071                                  new(zone) RegExpCharacterClass('*'),
6072                                  &compiler,
6073                                  captured_body,
6074                                  data->contains_anchor);
6075 
6076     if (data->contains_anchor) {
6077       // Unroll loop once, to take care of the case that might start
6078       // at the start of input.
6079       ChoiceNode* first_step_node = new(zone) ChoiceNode(2, zone);
6080       first_step_node->AddAlternative(GuardedAlternative(captured_body));
6081       first_step_node->AddAlternative(GuardedAlternative(
6082           new(zone) TextNode(new(zone) RegExpCharacterClass('*'), loop_node)));
6083       node = first_step_node;
6084     } else {
6085       node = loop_node;
6086     }
6087   }
6088   if (is_one_byte) {
6089     node = node->FilterOneByte(RegExpCompiler::kMaxRecursion, ignore_case);
6090     // Do it again to propagate the new nodes to places where they were not
6091     // put because they had not been calculated yet.
6092     if (node != NULL) {
6093       node = node->FilterOneByte(RegExpCompiler::kMaxRecursion, ignore_case);
6094     }
6095   }
6096 
6097   if (node == NULL) node = new(zone) EndNode(EndNode::BACKTRACK, zone);
6098   data->node = node;
6099   Analysis analysis(ignore_case, is_one_byte);
6100   analysis.EnsureAnalyzed(node);
6101   if (analysis.has_failed()) {
6102     const char* error_message = analysis.error_message();
6103     return CompilationResult(zone->isolate(), error_message);
6104   }
6105 
6106   // Create the correct assembler for the architecture.
6107 #ifndef V8_INTERPRETED_REGEXP
6108   // Native regexp implementation.
6109 
6110   NativeRegExpMacroAssembler::Mode mode =
6111       is_one_byte ? NativeRegExpMacroAssembler::LATIN1
6112                   : NativeRegExpMacroAssembler::UC16;
6113 
6114 #if V8_TARGET_ARCH_IA32
6115   RegExpMacroAssemblerIA32 macro_assembler(mode, (data->capture_count + 1) * 2,
6116                                            zone);
6117 #elif V8_TARGET_ARCH_X64
6118   RegExpMacroAssemblerX64 macro_assembler(mode, (data->capture_count + 1) * 2,
6119                                           zone);
6120 #elif V8_TARGET_ARCH_ARM
6121   RegExpMacroAssemblerARM macro_assembler(mode, (data->capture_count + 1) * 2,
6122                                           zone);
6123 #elif V8_TARGET_ARCH_ARM64
6124   RegExpMacroAssemblerARM64 macro_assembler(mode, (data->capture_count + 1) * 2,
6125                                             zone);
6126 #elif V8_TARGET_ARCH_MIPS
6127   RegExpMacroAssemblerMIPS macro_assembler(mode, (data->capture_count + 1) * 2,
6128                                            zone);
6129 #elif V8_TARGET_ARCH_MIPS64
6130   RegExpMacroAssemblerMIPS macro_assembler(mode, (data->capture_count + 1) * 2,
6131                                            zone);
6132 #elif V8_TARGET_ARCH_X87
6133   RegExpMacroAssemblerX87 macro_assembler(mode, (data->capture_count + 1) * 2,
6134                                           zone);
6135 #else
6136 #error "Unsupported architecture"
6137 #endif
6138 
6139 #else  // V8_INTERPRETED_REGEXP
6140   // Interpreted regexp implementation.
6141   EmbeddedVector<byte, 1024> codes;
6142   RegExpMacroAssemblerIrregexp macro_assembler(codes, zone);
6143 #endif  // V8_INTERPRETED_REGEXP
6144 
6145   // Inserted here, instead of in Assembler, because it depends on information
6146   // in the AST that isn't replicated in the Node structure.
6147   static const int kMaxBacksearchLimit = 1024;
6148   if (is_end_anchored &&
6149       !is_start_anchored &&
6150       max_length < kMaxBacksearchLimit) {
6151     macro_assembler.SetCurrentPositionFromEnd(max_length);
6152   }
6153 
6154   if (is_global) {
6155     macro_assembler.set_global_mode(
6156         (data->tree->min_match() > 0)
6157             ? RegExpMacroAssembler::GLOBAL_NO_ZERO_LENGTH_CHECK
6158             : RegExpMacroAssembler::GLOBAL);
6159   }
6160 
6161   return compiler.Assemble(&macro_assembler,
6162                            node,
6163                            data->capture_count,
6164                            pattern);
6165 }
6166 
6167 
6168 }}  // namespace v8::internal
6169