• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // Copyright 2011 the V8 project authors. All rights reserved.
2 // Redistribution and use in source and binary forms, with or without
3 // modification, are permitted provided that the following conditions are
4 // met:
5 //
6 //     * Redistributions of source code must retain the above copyright
7 //       notice, this list of conditions and the following disclaimer.
8 //     * Redistributions in binary form must reproduce the above
9 //       copyright notice, this list of conditions and the following
10 //       disclaimer in the documentation and/or other materials provided
11 //       with the distribution.
12 //     * Neither the name of Google Inc. nor the names of its
13 //       contributors may be used to endorse or promote products derived
14 //       from this software without specific prior written permission.
15 //
16 // THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
17 // "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
18 // LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
19 // A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
20 // OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
21 // SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
22 // LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23 // DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24 // THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25 // (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
26 // OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27 
28 #include "v8.h"
29 
30 #include "ast.h"
31 #include "compiler.h"
32 #include "execution.h"
33 #include "factory.h"
34 #include "jsregexp.h"
35 #include "platform.h"
36 #include "string-search.h"
37 #include "runtime.h"
38 #include "compilation-cache.h"
39 #include "string-stream.h"
40 #include "parser.h"
41 #include "regexp-macro-assembler.h"
42 #include "regexp-macro-assembler-tracer.h"
43 #include "regexp-macro-assembler-irregexp.h"
44 #include "regexp-stack.h"
45 
46 #ifndef V8_INTERPRETED_REGEXP
47 #if V8_TARGET_ARCH_IA32
48 #include "ia32/regexp-macro-assembler-ia32.h"
49 #elif V8_TARGET_ARCH_X64
50 #include "x64/regexp-macro-assembler-x64.h"
51 #elif V8_TARGET_ARCH_ARM
52 #include "arm/regexp-macro-assembler-arm.h"
53 #elif V8_TARGET_ARCH_MIPS
54 #include "mips/regexp-macro-assembler-mips.h"
55 #else
56 #error Unsupported target architecture.
57 #endif
58 #endif
59 
60 #include "interpreter-irregexp.h"
61 
62 
63 namespace v8 {
64 namespace internal {
65 
CreateRegExpLiteral(Handle<JSFunction> constructor,Handle<String> pattern,Handle<String> flags,bool * has_pending_exception)66 Handle<Object> RegExpImpl::CreateRegExpLiteral(Handle<JSFunction> constructor,
67                                                Handle<String> pattern,
68                                                Handle<String> flags,
69                                                bool* has_pending_exception) {
70   // Call the construct code with 2 arguments.
71   Handle<Object> argv[] = { pattern, flags };
72   return Execution::New(constructor, ARRAY_SIZE(argv), argv,
73                         has_pending_exception);
74 }
75 
76 
RegExpFlagsFromString(Handle<String> str)77 static JSRegExp::Flags RegExpFlagsFromString(Handle<String> str) {
78   int flags = JSRegExp::NONE;
79   for (int i = 0; i < str->length(); i++) {
80     switch (str->Get(i)) {
81       case 'i':
82         flags |= JSRegExp::IGNORE_CASE;
83         break;
84       case 'g':
85         flags |= JSRegExp::GLOBAL;
86         break;
87       case 'm':
88         flags |= JSRegExp::MULTILINE;
89         break;
90     }
91   }
92   return JSRegExp::Flags(flags);
93 }
94 
95 
ThrowRegExpException(Handle<JSRegExp> re,Handle<String> pattern,Handle<String> error_text,const char * message)96 static inline void ThrowRegExpException(Handle<JSRegExp> re,
97                                         Handle<String> pattern,
98                                         Handle<String> error_text,
99                                         const char* message) {
100   Isolate* isolate = re->GetIsolate();
101   Factory* factory = isolate->factory();
102   Handle<FixedArray> elements = factory->NewFixedArray(2);
103   elements->set(0, *pattern);
104   elements->set(1, *error_text);
105   Handle<JSArray> array = factory->NewJSArrayWithElements(elements);
106   Handle<Object> regexp_err = factory->NewSyntaxError(message, array);
107   isolate->Throw(*regexp_err);
108 }
109 
110 
111 // Generic RegExp methods. Dispatches to implementation specific methods.
112 
113 
Compile(Handle<JSRegExp> re,Handle<String> pattern,Handle<String> flag_str)114 Handle<Object> RegExpImpl::Compile(Handle<JSRegExp> re,
115                                    Handle<String> pattern,
116                                    Handle<String> flag_str) {
117   Isolate* isolate = re->GetIsolate();
118   JSRegExp::Flags flags = RegExpFlagsFromString(flag_str);
119   CompilationCache* compilation_cache = isolate->compilation_cache();
120   Handle<FixedArray> cached = compilation_cache->LookupRegExp(pattern, flags);
121   bool in_cache = !cached.is_null();
122   LOG(isolate, RegExpCompileEvent(re, in_cache));
123 
124   Handle<Object> result;
125   if (in_cache) {
126     re->set_data(*cached);
127     return re;
128   }
129   pattern = FlattenGetString(pattern);
130   ZoneScope zone_scope(isolate, DELETE_ON_EXIT);
131   PostponeInterruptsScope postpone(isolate);
132   RegExpCompileData parse_result;
133   FlatStringReader reader(isolate, pattern);
134   if (!RegExpParser::ParseRegExp(&reader, flags.is_multiline(),
135                                  &parse_result)) {
136     // Throw an exception if we fail to parse the pattern.
137     ThrowRegExpException(re,
138                          pattern,
139                          parse_result.error,
140                          "malformed_regexp");
141     return Handle<Object>::null();
142   }
143 
144   if (parse_result.simple && !flags.is_ignore_case()) {
145     // Parse-tree is a single atom that is equal to the pattern.
146     AtomCompile(re, pattern, flags, pattern);
147   } else if (parse_result.tree->IsAtom() &&
148       !flags.is_ignore_case() &&
149       parse_result.capture_count == 0) {
150     RegExpAtom* atom = parse_result.tree->AsAtom();
151     Vector<const uc16> atom_pattern = atom->data();
152     Handle<String> atom_string =
153         isolate->factory()->NewStringFromTwoByte(atom_pattern);
154     AtomCompile(re, pattern, flags, atom_string);
155   } else {
156     IrregexpInitialize(re, pattern, flags, parse_result.capture_count);
157   }
158   ASSERT(re->data()->IsFixedArray());
159   // Compilation succeeded so the data is set on the regexp
160   // and we can store it in the cache.
161   Handle<FixedArray> data(FixedArray::cast(re->data()));
162   compilation_cache->PutRegExp(pattern, flags, data);
163 
164   return re;
165 }
166 
167 
Exec(Handle<JSRegExp> regexp,Handle<String> subject,int index,Handle<JSArray> last_match_info)168 Handle<Object> RegExpImpl::Exec(Handle<JSRegExp> regexp,
169                                 Handle<String> subject,
170                                 int index,
171                                 Handle<JSArray> last_match_info) {
172   switch (regexp->TypeTag()) {
173     case JSRegExp::ATOM:
174       return AtomExec(regexp, subject, index, last_match_info);
175     case JSRegExp::IRREGEXP: {
176       Handle<Object> result =
177           IrregexpExec(regexp, subject, index, last_match_info);
178       ASSERT(!result.is_null() ||
179              regexp->GetIsolate()->has_pending_exception());
180       return result;
181     }
182     default:
183       UNREACHABLE();
184       return Handle<Object>::null();
185   }
186 }
187 
188 
189 // RegExp Atom implementation: Simple string search using indexOf.
190 
191 
AtomCompile(Handle<JSRegExp> re,Handle<String> pattern,JSRegExp::Flags flags,Handle<String> match_pattern)192 void RegExpImpl::AtomCompile(Handle<JSRegExp> re,
193                              Handle<String> pattern,
194                              JSRegExp::Flags flags,
195                              Handle<String> match_pattern) {
196   re->GetIsolate()->factory()->SetRegExpAtomData(re,
197                                                  JSRegExp::ATOM,
198                                                  pattern,
199                                                  flags,
200                                                  match_pattern);
201 }
202 
203 
SetAtomLastCapture(FixedArray * array,String * subject,int from,int to)204 static void SetAtomLastCapture(FixedArray* array,
205                                String* subject,
206                                int from,
207                                int to) {
208   NoHandleAllocation no_handles;
209   RegExpImpl::SetLastCaptureCount(array, 2);
210   RegExpImpl::SetLastSubject(array, subject);
211   RegExpImpl::SetLastInput(array, subject);
212   RegExpImpl::SetCapture(array, 0, from);
213   RegExpImpl::SetCapture(array, 1, to);
214 }
215 
216 
AtomExec(Handle<JSRegExp> re,Handle<String> subject,int index,Handle<JSArray> last_match_info)217 Handle<Object> RegExpImpl::AtomExec(Handle<JSRegExp> re,
218                                     Handle<String> subject,
219                                     int index,
220                                     Handle<JSArray> last_match_info) {
221   Isolate* isolate = re->GetIsolate();
222 
223   ASSERT(0 <= index);
224   ASSERT(index <= subject->length());
225 
226   if (!subject->IsFlat()) FlattenString(subject);
227   AssertNoAllocation no_heap_allocation;  // ensure vectors stay valid
228 
229   String* needle = String::cast(re->DataAt(JSRegExp::kAtomPatternIndex));
230   int needle_len = needle->length();
231   ASSERT(needle->IsFlat());
232 
233   if (needle_len != 0) {
234     if (index + needle_len > subject->length()) {
235       return isolate->factory()->null_value();
236     }
237 
238     String::FlatContent needle_content = needle->GetFlatContent();
239     String::FlatContent subject_content = subject->GetFlatContent();
240     ASSERT(needle_content.IsFlat());
241     ASSERT(subject_content.IsFlat());
242     // dispatch on type of strings
243     index = (needle_content.IsAscii()
244              ? (subject_content.IsAscii()
245                 ? SearchString(isolate,
246                                subject_content.ToAsciiVector(),
247                                needle_content.ToAsciiVector(),
248                                index)
249                 : SearchString(isolate,
250                                subject_content.ToUC16Vector(),
251                                needle_content.ToAsciiVector(),
252                                index))
253              : (subject_content.IsAscii()
254                 ? SearchString(isolate,
255                                subject_content.ToAsciiVector(),
256                                needle_content.ToUC16Vector(),
257                                index)
258                 : SearchString(isolate,
259                                subject_content.ToUC16Vector(),
260                                needle_content.ToUC16Vector(),
261                                index)));
262     if (index == -1) return isolate->factory()->null_value();
263   }
264   ASSERT(last_match_info->HasFastElements());
265 
266   {
267     NoHandleAllocation no_handles;
268     FixedArray* array = FixedArray::cast(last_match_info->elements());
269     SetAtomLastCapture(array, *subject, index, index + needle_len);
270   }
271   return last_match_info;
272 }
273 
274 
275 // Irregexp implementation.
276 
277 // Ensures that the regexp object contains a compiled version of the
278 // source for either ASCII or non-ASCII strings.
279 // If the compiled version doesn't already exist, it is compiled
280 // from the source pattern.
281 // If compilation fails, an exception is thrown and this function
282 // returns false.
EnsureCompiledIrregexp(Handle<JSRegExp> re,bool is_ascii)283 bool RegExpImpl::EnsureCompiledIrregexp(Handle<JSRegExp> re, bool is_ascii) {
284   Object* compiled_code = re->DataAt(JSRegExp::code_index(is_ascii));
285 #ifdef V8_INTERPRETED_REGEXP
286   if (compiled_code->IsByteArray()) return true;
287 #else  // V8_INTERPRETED_REGEXP (RegExp native code)
288   if (compiled_code->IsCode()) return true;
289 #endif
290   // We could potentially have marked this as flushable, but have kept
291   // a saved version if we did not flush it yet.
292   Object* saved_code = re->DataAt(JSRegExp::saved_code_index(is_ascii));
293   if (saved_code->IsCode()) {
294     // Reinstate the code in the original place.
295     re->SetDataAt(JSRegExp::code_index(is_ascii), saved_code);
296     ASSERT(compiled_code->IsSmi());
297     return true;
298   }
299   return CompileIrregexp(re, is_ascii);
300 }
301 
302 
CreateRegExpErrorObjectAndThrow(Handle<JSRegExp> re,bool is_ascii,Handle<String> error_message,Isolate * isolate)303 static bool CreateRegExpErrorObjectAndThrow(Handle<JSRegExp> re,
304                                             bool is_ascii,
305                                             Handle<String> error_message,
306                                             Isolate* isolate) {
307   Factory* factory = isolate->factory();
308   Handle<FixedArray> elements = factory->NewFixedArray(2);
309   elements->set(0, re->Pattern());
310   elements->set(1, *error_message);
311   Handle<JSArray> array = factory->NewJSArrayWithElements(elements);
312   Handle<Object> regexp_err =
313       factory->NewSyntaxError("malformed_regexp", array);
314   isolate->Throw(*regexp_err);
315   return false;
316 }
317 
318 
CompileIrregexp(Handle<JSRegExp> re,bool is_ascii)319 bool RegExpImpl::CompileIrregexp(Handle<JSRegExp> re, bool is_ascii) {
320   // Compile the RegExp.
321   Isolate* isolate = re->GetIsolate();
322   ZoneScope zone_scope(isolate, DELETE_ON_EXIT);
323   PostponeInterruptsScope postpone(isolate);
324   // If we had a compilation error the last time this is saved at the
325   // saved code index.
326   Object* entry = re->DataAt(JSRegExp::code_index(is_ascii));
327   // When arriving here entry can only be a smi, either representing an
328   // uncompiled regexp, a previous compilation error, or code that has
329   // been flushed.
330   ASSERT(entry->IsSmi());
331   int entry_value = Smi::cast(entry)->value();
332   ASSERT(entry_value == JSRegExp::kUninitializedValue ||
333          entry_value == JSRegExp::kCompilationErrorValue ||
334          (entry_value < JSRegExp::kCodeAgeMask && entry_value >= 0));
335 
336   if (entry_value == JSRegExp::kCompilationErrorValue) {
337     // A previous compilation failed and threw an error which we store in
338     // the saved code index (we store the error message, not the actual
339     // error). Recreate the error object and throw it.
340     Object* error_string = re->DataAt(JSRegExp::saved_code_index(is_ascii));
341     ASSERT(error_string->IsString());
342     Handle<String> error_message(String::cast(error_string));
343     CreateRegExpErrorObjectAndThrow(re, is_ascii, error_message, isolate);
344     return false;
345   }
346 
347   JSRegExp::Flags flags = re->GetFlags();
348 
349   Handle<String> pattern(re->Pattern());
350   if (!pattern->IsFlat()) FlattenString(pattern);
351   RegExpCompileData compile_data;
352   FlatStringReader reader(isolate, pattern);
353   if (!RegExpParser::ParseRegExp(&reader, flags.is_multiline(),
354                                  &compile_data)) {
355     // Throw an exception if we fail to parse the pattern.
356     // THIS SHOULD NOT HAPPEN. We already pre-parsed it successfully once.
357     ThrowRegExpException(re,
358                          pattern,
359                          compile_data.error,
360                          "malformed_regexp");
361     return false;
362   }
363   RegExpEngine::CompilationResult result =
364       RegExpEngine::Compile(&compile_data,
365                             flags.is_ignore_case(),
366                             flags.is_multiline(),
367                             pattern,
368                             is_ascii);
369   if (result.error_message != NULL) {
370     // Unable to compile regexp.
371     Handle<String> error_message =
372         isolate->factory()->NewStringFromUtf8(CStrVector(result.error_message));
373     CreateRegExpErrorObjectAndThrow(re, is_ascii, error_message, isolate);
374     return false;
375   }
376 
377   Handle<FixedArray> data = Handle<FixedArray>(FixedArray::cast(re->data()));
378   data->set(JSRegExp::code_index(is_ascii), result.code);
379   int register_max = IrregexpMaxRegisterCount(*data);
380   if (result.num_registers > register_max) {
381     SetIrregexpMaxRegisterCount(*data, result.num_registers);
382   }
383 
384   return true;
385 }
386 
387 
IrregexpMaxRegisterCount(FixedArray * re)388 int RegExpImpl::IrregexpMaxRegisterCount(FixedArray* re) {
389   return Smi::cast(
390       re->get(JSRegExp::kIrregexpMaxRegisterCountIndex))->value();
391 }
392 
393 
SetIrregexpMaxRegisterCount(FixedArray * re,int value)394 void RegExpImpl::SetIrregexpMaxRegisterCount(FixedArray* re, int value) {
395   re->set(JSRegExp::kIrregexpMaxRegisterCountIndex, Smi::FromInt(value));
396 }
397 
398 
IrregexpNumberOfCaptures(FixedArray * re)399 int RegExpImpl::IrregexpNumberOfCaptures(FixedArray* re) {
400   return Smi::cast(re->get(JSRegExp::kIrregexpCaptureCountIndex))->value();
401 }
402 
403 
IrregexpNumberOfRegisters(FixedArray * re)404 int RegExpImpl::IrregexpNumberOfRegisters(FixedArray* re) {
405   return Smi::cast(re->get(JSRegExp::kIrregexpMaxRegisterCountIndex))->value();
406 }
407 
408 
IrregexpByteCode(FixedArray * re,bool is_ascii)409 ByteArray* RegExpImpl::IrregexpByteCode(FixedArray* re, bool is_ascii) {
410   return ByteArray::cast(re->get(JSRegExp::code_index(is_ascii)));
411 }
412 
413 
IrregexpNativeCode(FixedArray * re,bool is_ascii)414 Code* RegExpImpl::IrregexpNativeCode(FixedArray* re, bool is_ascii) {
415   return Code::cast(re->get(JSRegExp::code_index(is_ascii)));
416 }
417 
418 
IrregexpInitialize(Handle<JSRegExp> re,Handle<String> pattern,JSRegExp::Flags flags,int capture_count)419 void RegExpImpl::IrregexpInitialize(Handle<JSRegExp> re,
420                                     Handle<String> pattern,
421                                     JSRegExp::Flags flags,
422                                     int capture_count) {
423   // Initialize compiled code entries to null.
424   re->GetIsolate()->factory()->SetRegExpIrregexpData(re,
425                                                      JSRegExp::IRREGEXP,
426                                                      pattern,
427                                                      flags,
428                                                      capture_count);
429 }
430 
431 
IrregexpPrepare(Handle<JSRegExp> regexp,Handle<String> subject)432 int RegExpImpl::IrregexpPrepare(Handle<JSRegExp> regexp,
433                                 Handle<String> subject) {
434   if (!subject->IsFlat()) FlattenString(subject);
435 
436   // Check the asciiness of the underlying storage.
437   bool is_ascii = subject->IsAsciiRepresentationUnderneath();
438   if (!EnsureCompiledIrregexp(regexp, is_ascii)) return -1;
439 
440 #ifdef V8_INTERPRETED_REGEXP
441   // Byte-code regexp needs space allocated for all its registers.
442   return IrregexpNumberOfRegisters(FixedArray::cast(regexp->data()));
443 #else  // V8_INTERPRETED_REGEXP
444   // Native regexp only needs room to output captures. Registers are handled
445   // internally.
446   return (IrregexpNumberOfCaptures(FixedArray::cast(regexp->data())) + 1) * 2;
447 #endif  // V8_INTERPRETED_REGEXP
448 }
449 
450 
IrregexpExecOnce(Handle<JSRegExp> regexp,Handle<String> subject,int index,Vector<int> output)451 RegExpImpl::IrregexpResult RegExpImpl::IrregexpExecOnce(
452     Handle<JSRegExp> regexp,
453     Handle<String> subject,
454     int index,
455     Vector<int> output) {
456   Isolate* isolate = regexp->GetIsolate();
457 
458   Handle<FixedArray> irregexp(FixedArray::cast(regexp->data()), isolate);
459 
460   ASSERT(index >= 0);
461   ASSERT(index <= subject->length());
462   ASSERT(subject->IsFlat());
463 
464   bool is_ascii = subject->IsAsciiRepresentationUnderneath();
465 
466 #ifndef V8_INTERPRETED_REGEXP
467   ASSERT(output.length() >= (IrregexpNumberOfCaptures(*irregexp) + 1) * 2);
468   do {
469     EnsureCompiledIrregexp(regexp, is_ascii);
470     Handle<Code> code(IrregexpNativeCode(*irregexp, is_ascii), isolate);
471     NativeRegExpMacroAssembler::Result res =
472         NativeRegExpMacroAssembler::Match(code,
473                                           subject,
474                                           output.start(),
475                                           output.length(),
476                                           index,
477                                           isolate);
478     if (res != NativeRegExpMacroAssembler::RETRY) {
479       ASSERT(res != NativeRegExpMacroAssembler::EXCEPTION ||
480              isolate->has_pending_exception());
481       STATIC_ASSERT(
482           static_cast<int>(NativeRegExpMacroAssembler::SUCCESS) == RE_SUCCESS);
483       STATIC_ASSERT(
484           static_cast<int>(NativeRegExpMacroAssembler::FAILURE) == RE_FAILURE);
485       STATIC_ASSERT(static_cast<int>(NativeRegExpMacroAssembler::EXCEPTION)
486                     == RE_EXCEPTION);
487       return static_cast<IrregexpResult>(res);
488     }
489     // If result is RETRY, the string has changed representation, and we
490     // must restart from scratch.
491     // In this case, it means we must make sure we are prepared to handle
492     // the, potentially, different subject (the string can switch between
493     // being internal and external, and even between being ASCII and UC16,
494     // but the characters are always the same).
495     IrregexpPrepare(regexp, subject);
496     is_ascii = subject->IsAsciiRepresentationUnderneath();
497   } while (true);
498   UNREACHABLE();
499   return RE_EXCEPTION;
500 #else  // V8_INTERPRETED_REGEXP
501 
502   ASSERT(output.length() >= IrregexpNumberOfRegisters(*irregexp));
503   // We must have done EnsureCompiledIrregexp, so we can get the number of
504   // registers.
505   int* register_vector = output.start();
506   int number_of_capture_registers =
507       (IrregexpNumberOfCaptures(*irregexp) + 1) * 2;
508   for (int i = number_of_capture_registers - 1; i >= 0; i--) {
509     register_vector[i] = -1;
510   }
511   Handle<ByteArray> byte_codes(IrregexpByteCode(*irregexp, is_ascii), isolate);
512 
513   IrregexpResult result = IrregexpInterpreter::Match(isolate,
514                                                      byte_codes,
515                                                      subject,
516                                                      register_vector,
517                                                      index);
518   if (result == RE_EXCEPTION) {
519     ASSERT(!isolate->has_pending_exception());
520     isolate->StackOverflow();
521   }
522   return result;
523 #endif  // V8_INTERPRETED_REGEXP
524 }
525 
526 
IrregexpExec(Handle<JSRegExp> jsregexp,Handle<String> subject,int previous_index,Handle<JSArray> last_match_info)527 Handle<Object> RegExpImpl::IrregexpExec(Handle<JSRegExp> jsregexp,
528                                         Handle<String> subject,
529                                         int previous_index,
530                                         Handle<JSArray> last_match_info) {
531   Isolate* isolate = jsregexp->GetIsolate();
532   ASSERT_EQ(jsregexp->TypeTag(), JSRegExp::IRREGEXP);
533 
534   // Prepare space for the return values.
535 #ifdef V8_INTERPRETED_REGEXP
536 #ifdef DEBUG
537   if (FLAG_trace_regexp_bytecodes) {
538     String* pattern = jsregexp->Pattern();
539     PrintF("\n\nRegexp match:   /%s/\n\n", *(pattern->ToCString()));
540     PrintF("\n\nSubject string: '%s'\n\n", *(subject->ToCString()));
541   }
542 #endif
543 #endif
544   int required_registers = RegExpImpl::IrregexpPrepare(jsregexp, subject);
545   if (required_registers < 0) {
546     // Compiling failed with an exception.
547     ASSERT(isolate->has_pending_exception());
548     return Handle<Object>::null();
549   }
550 
551   OffsetsVector registers(required_registers, isolate);
552 
553   IrregexpResult res = RegExpImpl::IrregexpExecOnce(
554       jsregexp, subject, previous_index, Vector<int>(registers.vector(),
555                                                      registers.length()));
556   if (res == RE_SUCCESS) {
557     int capture_register_count =
558         (IrregexpNumberOfCaptures(FixedArray::cast(jsregexp->data())) + 1) * 2;
559     last_match_info->EnsureSize(capture_register_count + kLastMatchOverhead);
560     AssertNoAllocation no_gc;
561     int* register_vector = registers.vector();
562     FixedArray* array = FixedArray::cast(last_match_info->elements());
563     for (int i = 0; i < capture_register_count; i += 2) {
564       SetCapture(array, i, register_vector[i]);
565       SetCapture(array, i + 1, register_vector[i + 1]);
566     }
567     SetLastCaptureCount(array, capture_register_count);
568     SetLastSubject(array, *subject);
569     SetLastInput(array, *subject);
570     return last_match_info;
571   }
572   if (res == RE_EXCEPTION) {
573     ASSERT(isolate->has_pending_exception());
574     return Handle<Object>::null();
575   }
576   ASSERT(res == RE_FAILURE);
577   return isolate->factory()->null_value();
578 }
579 
580 
581 // -------------------------------------------------------------------
582 // Implementation of the Irregexp regular expression engine.
583 //
584 // The Irregexp regular expression engine is intended to be a complete
585 // implementation of ECMAScript regular expressions.  It generates either
586 // bytecodes or native code.
587 
588 //   The Irregexp regexp engine is structured in three steps.
589 //   1) The parser generates an abstract syntax tree.  See ast.cc.
590 //   2) From the AST a node network is created.  The nodes are all
591 //      subclasses of RegExpNode.  The nodes represent states when
592 //      executing a regular expression.  Several optimizations are
593 //      performed on the node network.
594 //   3) From the nodes we generate either byte codes or native code
595 //      that can actually execute the regular expression (perform
596 //      the search).  The code generation step is described in more
597 //      detail below.
598 
599 // Code generation.
600 //
601 //   The nodes are divided into four main categories.
602 //   * Choice nodes
603 //        These represent places where the regular expression can
604 //        match in more than one way.  For example on entry to an
605 //        alternation (foo|bar) or a repetition (*, +, ? or {}).
606 //   * Action nodes
607 //        These represent places where some action should be
608 //        performed.  Examples include recording the current position
609 //        in the input string to a register (in order to implement
610 //        captures) or other actions on register for example in order
611 //        to implement the counters needed for {} repetitions.
612 //   * Matching nodes
613 //        These attempt to match some element part of the input string.
614 //        Examples of elements include character classes, plain strings
615 //        or back references.
616 //   * End nodes
617 //        These are used to implement the actions required on finding
618 //        a successful match or failing to find a match.
619 //
620 //   The code generated (whether as byte codes or native code) maintains
621 //   some state as it runs.  This consists of the following elements:
622 //
623 //   * The capture registers.  Used for string captures.
624 //   * Other registers.  Used for counters etc.
625 //   * The current position.
626 //   * The stack of backtracking information.  Used when a matching node
627 //     fails to find a match and needs to try an alternative.
628 //
629 // Conceptual regular expression execution model:
630 //
631 //   There is a simple conceptual model of regular expression execution
632 //   which will be presented first.  The actual code generated is a more
633 //   efficient simulation of the simple conceptual model:
634 //
635 //   * Choice nodes are implemented as follows:
636 //     For each choice except the last {
637 //       push current position
638 //       push backtrack code location
639 //       <generate code to test for choice>
640 //       backtrack code location:
641 //       pop current position
642 //     }
643 //     <generate code to test for last choice>
644 //
645 //   * Actions nodes are generated as follows
646 //     <push affected registers on backtrack stack>
647 //     <generate code to perform action>
648 //     push backtrack code location
649 //     <generate code to test for following nodes>
650 //     backtrack code location:
651 //     <pop affected registers to restore their state>
652 //     <pop backtrack location from stack and go to it>
653 //
654 //   * Matching nodes are generated as follows:
655 //     if input string matches at current position
656 //       update current position
657 //       <generate code to test for following nodes>
658 //     else
659 //       <pop backtrack location from stack and go to it>
660 //
661 //   Thus it can be seen that the current position is saved and restored
662 //   by the choice nodes, whereas the registers are saved and restored by
663 //   by the action nodes that manipulate them.
664 //
665 //   The other interesting aspect of this model is that nodes are generated
666 //   at the point where they are needed by a recursive call to Emit().  If
667 //   the node has already been code generated then the Emit() call will
668 //   generate a jump to the previously generated code instead.  In order to
669 //   limit recursion it is possible for the Emit() function to put the node
670 //   on a work list for later generation and instead generate a jump.  The
671 //   destination of the jump is resolved later when the code is generated.
672 //
673 // Actual regular expression code generation.
674 //
675 //   Code generation is actually more complicated than the above.  In order
676 //   to improve the efficiency of the generated code some optimizations are
677 //   performed
678 //
679 //   * Choice nodes have 1-character lookahead.
680 //     A choice node looks at the following character and eliminates some of
681 //     the choices immediately based on that character.  This is not yet
682 //     implemented.
683 //   * Simple greedy loops store reduced backtracking information.
684 //     A quantifier like /.*foo/m will greedily match the whole input.  It will
685 //     then need to backtrack to a point where it can match "foo".  The naive
686 //     implementation of this would push each character position onto the
687 //     backtracking stack, then pop them off one by one.  This would use space
688 //     proportional to the length of the input string.  However since the "."
689 //     can only match in one way and always has a constant length (in this case
690 //     of 1) it suffices to store the current position on the top of the stack
691 //     once.  Matching now becomes merely incrementing the current position and
692 //     backtracking becomes decrementing the current position and checking the
693 //     result against the stored current position.  This is faster and saves
694 //     space.
695 //   * The current state is virtualized.
696 //     This is used to defer expensive operations until it is clear that they
697 //     are needed and to generate code for a node more than once, allowing
698 //     specialized an efficient versions of the code to be created. This is
699 //     explained in the section below.
700 //
701 // Execution state virtualization.
702 //
703 //   Instead of emitting code, nodes that manipulate the state can record their
704 //   manipulation in an object called the Trace.  The Trace object can record a
705 //   current position offset, an optional backtrack code location on the top of
706 //   the virtualized backtrack stack and some register changes.  When a node is
707 //   to be emitted it can flush the Trace or update it.  Flushing the Trace
708 //   will emit code to bring the actual state into line with the virtual state.
709 //   Avoiding flushing the state can postpone some work (e.g. updates of capture
710 //   registers).  Postponing work can save time when executing the regular
711 //   expression since it may be found that the work never has to be done as a
712 //   failure to match can occur.  In addition it is much faster to jump to a
713 //   known backtrack code location than it is to pop an unknown backtrack
714 //   location from the stack and jump there.
715 //
716 //   The virtual state found in the Trace affects code generation.  For example
717 //   the virtual state contains the difference between the actual current
718 //   position and the virtual current position, and matching code needs to use
719 //   this offset to attempt a match in the correct location of the input
720 //   string.  Therefore code generated for a non-trivial trace is specialized
721 //   to that trace.  The code generator therefore has the ability to generate
722 //   code for each node several times.  In order to limit the size of the
723 //   generated code there is an arbitrary limit on how many specialized sets of
724 //   code may be generated for a given node.  If the limit is reached, the
725 //   trace is flushed and a generic version of the code for a node is emitted.
726 //   This is subsequently used for that node.  The code emitted for non-generic
727 //   trace is not recorded in the node and so it cannot currently be reused in
728 //   the event that code generation is requested for an identical trace.
729 
730 
AppendToText(RegExpText * text)731 void RegExpTree::AppendToText(RegExpText* text) {
732   UNREACHABLE();
733 }
734 
735 
AppendToText(RegExpText * text)736 void RegExpAtom::AppendToText(RegExpText* text) {
737   text->AddElement(TextElement::Atom(this));
738 }
739 
740 
AppendToText(RegExpText * text)741 void RegExpCharacterClass::AppendToText(RegExpText* text) {
742   text->AddElement(TextElement::CharClass(this));
743 }
744 
745 
AppendToText(RegExpText * text)746 void RegExpText::AppendToText(RegExpText* text) {
747   for (int i = 0; i < elements()->length(); i++)
748     text->AddElement(elements()->at(i));
749 }
750 
751 
Atom(RegExpAtom * atom)752 TextElement TextElement::Atom(RegExpAtom* atom) {
753   TextElement result = TextElement(ATOM);
754   result.data.u_atom = atom;
755   return result;
756 }
757 
758 
CharClass(RegExpCharacterClass * char_class)759 TextElement TextElement::CharClass(
760       RegExpCharacterClass* char_class) {
761   TextElement result = TextElement(CHAR_CLASS);
762   result.data.u_char_class = char_class;
763   return result;
764 }
765 
766 
length()767 int TextElement::length() {
768   if (type == ATOM) {
769     return data.u_atom->length();
770   } else {
771     ASSERT(type == CHAR_CLASS);
772     return 1;
773   }
774 }
775 
776 
GetTable(bool ignore_case)777 DispatchTable* ChoiceNode::GetTable(bool ignore_case) {
778   if (table_ == NULL) {
779     table_ = new DispatchTable();
780     DispatchTableConstructor cons(table_, ignore_case);
781     cons.BuildTable(this);
782   }
783   return table_;
784 }
785 
786 
787 class RegExpCompiler {
788  public:
789   RegExpCompiler(int capture_count, bool ignore_case, bool is_ascii);
790 
AllocateRegister()791   int AllocateRegister() {
792     if (next_register_ >= RegExpMacroAssembler::kMaxRegister) {
793       reg_exp_too_big_ = true;
794       return next_register_;
795     }
796     return next_register_++;
797   }
798 
799   RegExpEngine::CompilationResult Assemble(RegExpMacroAssembler* assembler,
800                                            RegExpNode* start,
801                                            int capture_count,
802                                            Handle<String> pattern);
803 
AddWork(RegExpNode * node)804   inline void AddWork(RegExpNode* node) { work_list_->Add(node); }
805 
806   static const int kImplementationOffset = 0;
807   static const int kNumberOfRegistersOffset = 0;
808   static const int kCodeOffset = 1;
809 
macro_assembler()810   RegExpMacroAssembler* macro_assembler() { return macro_assembler_; }
accept()811   EndNode* accept() { return accept_; }
812 
813   static const int kMaxRecursion = 100;
recursion_depth()814   inline int recursion_depth() { return recursion_depth_; }
IncrementRecursionDepth()815   inline void IncrementRecursionDepth() { recursion_depth_++; }
DecrementRecursionDepth()816   inline void DecrementRecursionDepth() { recursion_depth_--; }
817 
SetRegExpTooBig()818   void SetRegExpTooBig() { reg_exp_too_big_ = true; }
819 
ignore_case()820   inline bool ignore_case() { return ignore_case_; }
ascii()821   inline bool ascii() { return ascii_; }
822 
current_expansion_factor()823   int current_expansion_factor() { return current_expansion_factor_; }
set_current_expansion_factor(int value)824   void set_current_expansion_factor(int value) {
825     current_expansion_factor_ = value;
826   }
827 
828   static const int kNoRegister = -1;
829 
830  private:
831   EndNode* accept_;
832   int next_register_;
833   List<RegExpNode*>* work_list_;
834   int recursion_depth_;
835   RegExpMacroAssembler* macro_assembler_;
836   bool ignore_case_;
837   bool ascii_;
838   bool reg_exp_too_big_;
839   int current_expansion_factor_;
840 };
841 
842 
843 class RecursionCheck {
844  public:
RecursionCheck(RegExpCompiler * compiler)845   explicit RecursionCheck(RegExpCompiler* compiler) : compiler_(compiler) {
846     compiler->IncrementRecursionDepth();
847   }
~RecursionCheck()848   ~RecursionCheck() { compiler_->DecrementRecursionDepth(); }
849  private:
850   RegExpCompiler* compiler_;
851 };
852 
853 
IrregexpRegExpTooBig()854 static RegExpEngine::CompilationResult IrregexpRegExpTooBig() {
855   return RegExpEngine::CompilationResult("RegExp too big");
856 }
857 
858 
859 // Attempts to compile the regexp using an Irregexp code generator.  Returns
860 // a fixed array or a null handle depending on whether it succeeded.
RegExpCompiler(int capture_count,bool ignore_case,bool ascii)861 RegExpCompiler::RegExpCompiler(int capture_count, bool ignore_case, bool ascii)
862     : next_register_(2 * (capture_count + 1)),
863       work_list_(NULL),
864       recursion_depth_(0),
865       ignore_case_(ignore_case),
866       ascii_(ascii),
867       reg_exp_too_big_(false),
868       current_expansion_factor_(1) {
869   accept_ = new EndNode(EndNode::ACCEPT);
870   ASSERT(next_register_ - 1 <= RegExpMacroAssembler::kMaxRegister);
871 }
872 
873 
Assemble(RegExpMacroAssembler * macro_assembler,RegExpNode * start,int capture_count,Handle<String> pattern)874 RegExpEngine::CompilationResult RegExpCompiler::Assemble(
875     RegExpMacroAssembler* macro_assembler,
876     RegExpNode* start,
877     int capture_count,
878     Handle<String> pattern) {
879   Heap* heap = pattern->GetHeap();
880 
881   bool use_slow_safe_regexp_compiler = false;
882   if (heap->total_regexp_code_generated() >
883           RegExpImpl::kRegWxpCompiledLimit &&
884       heap->isolate()->memory_allocator()->SizeExecutable() >
885           RegExpImpl::kRegExpExecutableMemoryLimit) {
886     use_slow_safe_regexp_compiler = true;
887   }
888 
889   macro_assembler->set_slow_safe(use_slow_safe_regexp_compiler);
890 
891 #ifdef DEBUG
892   if (FLAG_trace_regexp_assembler)
893     macro_assembler_ = new RegExpMacroAssemblerTracer(macro_assembler);
894   else
895 #endif
896     macro_assembler_ = macro_assembler;
897 
898   List <RegExpNode*> work_list(0);
899   work_list_ = &work_list;
900   Label fail;
901   macro_assembler_->PushBacktrack(&fail);
902   Trace new_trace;
903   start->Emit(this, &new_trace);
904   macro_assembler_->Bind(&fail);
905   macro_assembler_->Fail();
906   while (!work_list.is_empty()) {
907     work_list.RemoveLast()->Emit(this, &new_trace);
908   }
909   if (reg_exp_too_big_) return IrregexpRegExpTooBig();
910 
911   Handle<HeapObject> code = macro_assembler_->GetCode(pattern);
912   heap->IncreaseTotalRegexpCodeGenerated(code->Size());
913   work_list_ = NULL;
914 #ifdef DEBUG
915   if (FLAG_print_code) {
916     Handle<Code>::cast(code)->Disassemble(*pattern->ToCString());
917   }
918   if (FLAG_trace_regexp_assembler) {
919     delete macro_assembler_;
920   }
921 #endif
922   return RegExpEngine::CompilationResult(*code, next_register_);
923 }
924 
925 
Mentions(int that)926 bool Trace::DeferredAction::Mentions(int that) {
927   if (type() == ActionNode::CLEAR_CAPTURES) {
928     Interval range = static_cast<DeferredClearCaptures*>(this)->range();
929     return range.Contains(that);
930   } else {
931     return reg() == that;
932   }
933 }
934 
935 
mentions_reg(int reg)936 bool Trace::mentions_reg(int reg) {
937   for (DeferredAction* action = actions_;
938        action != NULL;
939        action = action->next()) {
940     if (action->Mentions(reg))
941       return true;
942   }
943   return false;
944 }
945 
946 
GetStoredPosition(int reg,int * cp_offset)947 bool Trace::GetStoredPosition(int reg, int* cp_offset) {
948   ASSERT_EQ(0, *cp_offset);
949   for (DeferredAction* action = actions_;
950        action != NULL;
951        action = action->next()) {
952     if (action->Mentions(reg)) {
953       if (action->type() == ActionNode::STORE_POSITION) {
954         *cp_offset = static_cast<DeferredCapture*>(action)->cp_offset();
955         return true;
956       } else {
957         return false;
958       }
959     }
960   }
961   return false;
962 }
963 
964 
FindAffectedRegisters(OutSet * affected_registers)965 int Trace::FindAffectedRegisters(OutSet* affected_registers) {
966   int max_register = RegExpCompiler::kNoRegister;
967   for (DeferredAction* action = actions_;
968        action != NULL;
969        action = action->next()) {
970     if (action->type() == ActionNode::CLEAR_CAPTURES) {
971       Interval range = static_cast<DeferredClearCaptures*>(action)->range();
972       for (int i = range.from(); i <= range.to(); i++)
973         affected_registers->Set(i);
974       if (range.to() > max_register) max_register = range.to();
975     } else {
976       affected_registers->Set(action->reg());
977       if (action->reg() > max_register) max_register = action->reg();
978     }
979   }
980   return max_register;
981 }
982 
983 
RestoreAffectedRegisters(RegExpMacroAssembler * assembler,int max_register,OutSet & registers_to_pop,OutSet & registers_to_clear)984 void Trace::RestoreAffectedRegisters(RegExpMacroAssembler* assembler,
985                                      int max_register,
986                                      OutSet& registers_to_pop,
987                                      OutSet& registers_to_clear) {
988   for (int reg = max_register; reg >= 0; reg--) {
989     if (registers_to_pop.Get(reg)) assembler->PopRegister(reg);
990     else if (registers_to_clear.Get(reg)) {
991       int clear_to = reg;
992       while (reg > 0 && registers_to_clear.Get(reg - 1)) {
993         reg--;
994       }
995       assembler->ClearRegisters(reg, clear_to);
996     }
997   }
998 }
999 
1000 
PerformDeferredActions(RegExpMacroAssembler * assembler,int max_register,OutSet & affected_registers,OutSet * registers_to_pop,OutSet * registers_to_clear)1001 void Trace::PerformDeferredActions(RegExpMacroAssembler* assembler,
1002                                    int max_register,
1003                                    OutSet& affected_registers,
1004                                    OutSet* registers_to_pop,
1005                                    OutSet* registers_to_clear) {
1006   // The "+1" is to avoid a push_limit of zero if stack_limit_slack() is 1.
1007   const int push_limit = (assembler->stack_limit_slack() + 1) / 2;
1008 
1009   // Count pushes performed to force a stack limit check occasionally.
1010   int pushes = 0;
1011 
1012   for (int reg = 0; reg <= max_register; reg++) {
1013     if (!affected_registers.Get(reg)) {
1014       continue;
1015     }
1016 
1017     // The chronologically first deferred action in the trace
1018     // is used to infer the action needed to restore a register
1019     // to its previous state (or not, if it's safe to ignore it).
1020     enum DeferredActionUndoType { IGNORE, RESTORE, CLEAR };
1021     DeferredActionUndoType undo_action = IGNORE;
1022 
1023     int value = 0;
1024     bool absolute = false;
1025     bool clear = false;
1026     int store_position = -1;
1027     // This is a little tricky because we are scanning the actions in reverse
1028     // historical order (newest first).
1029     for (DeferredAction* action = actions_;
1030          action != NULL;
1031          action = action->next()) {
1032       if (action->Mentions(reg)) {
1033         switch (action->type()) {
1034           case ActionNode::SET_REGISTER: {
1035             Trace::DeferredSetRegister* psr =
1036                 static_cast<Trace::DeferredSetRegister*>(action);
1037             if (!absolute) {
1038               value += psr->value();
1039               absolute = true;
1040             }
1041             // SET_REGISTER is currently only used for newly introduced loop
1042             // counters. They can have a significant previous value if they
1043             // occour in a loop. TODO(lrn): Propagate this information, so
1044             // we can set undo_action to IGNORE if we know there is no value to
1045             // restore.
1046             undo_action = RESTORE;
1047             ASSERT_EQ(store_position, -1);
1048             ASSERT(!clear);
1049             break;
1050           }
1051           case ActionNode::INCREMENT_REGISTER:
1052             if (!absolute) {
1053               value++;
1054             }
1055             ASSERT_EQ(store_position, -1);
1056             ASSERT(!clear);
1057             undo_action = RESTORE;
1058             break;
1059           case ActionNode::STORE_POSITION: {
1060             Trace::DeferredCapture* pc =
1061                 static_cast<Trace::DeferredCapture*>(action);
1062             if (!clear && store_position == -1) {
1063               store_position = pc->cp_offset();
1064             }
1065 
1066             // For captures we know that stores and clears alternate.
1067             // Other register, are never cleared, and if the occur
1068             // inside a loop, they might be assigned more than once.
1069             if (reg <= 1) {
1070               // Registers zero and one, aka "capture zero", is
1071               // always set correctly if we succeed. There is no
1072               // need to undo a setting on backtrack, because we
1073               // will set it again or fail.
1074               undo_action = IGNORE;
1075             } else {
1076               undo_action = pc->is_capture() ? CLEAR : RESTORE;
1077             }
1078             ASSERT(!absolute);
1079             ASSERT_EQ(value, 0);
1080             break;
1081           }
1082           case ActionNode::CLEAR_CAPTURES: {
1083             // Since we're scanning in reverse order, if we've already
1084             // set the position we have to ignore historically earlier
1085             // clearing operations.
1086             if (store_position == -1) {
1087               clear = true;
1088             }
1089             undo_action = RESTORE;
1090             ASSERT(!absolute);
1091             ASSERT_EQ(value, 0);
1092             break;
1093           }
1094           default:
1095             UNREACHABLE();
1096             break;
1097         }
1098       }
1099     }
1100     // Prepare for the undo-action (e.g., push if it's going to be popped).
1101     if (undo_action == RESTORE) {
1102       pushes++;
1103       RegExpMacroAssembler::StackCheckFlag stack_check =
1104           RegExpMacroAssembler::kNoStackLimitCheck;
1105       if (pushes == push_limit) {
1106         stack_check = RegExpMacroAssembler::kCheckStackLimit;
1107         pushes = 0;
1108       }
1109 
1110       assembler->PushRegister(reg, stack_check);
1111       registers_to_pop->Set(reg);
1112     } else if (undo_action == CLEAR) {
1113       registers_to_clear->Set(reg);
1114     }
1115     // Perform the chronologically last action (or accumulated increment)
1116     // for the register.
1117     if (store_position != -1) {
1118       assembler->WriteCurrentPositionToRegister(reg, store_position);
1119     } else if (clear) {
1120       assembler->ClearRegisters(reg, reg);
1121     } else if (absolute) {
1122       assembler->SetRegister(reg, value);
1123     } else if (value != 0) {
1124       assembler->AdvanceRegister(reg, value);
1125     }
1126   }
1127 }
1128 
1129 
1130 // This is called as we come into a loop choice node and some other tricky
1131 // nodes.  It normalizes the state of the code generator to ensure we can
1132 // generate generic code.
Flush(RegExpCompiler * compiler,RegExpNode * successor)1133 void Trace::Flush(RegExpCompiler* compiler, RegExpNode* successor) {
1134   RegExpMacroAssembler* assembler = compiler->macro_assembler();
1135 
1136   ASSERT(!is_trivial());
1137 
1138   if (actions_ == NULL && backtrack() == NULL) {
1139     // Here we just have some deferred cp advances to fix and we are back to
1140     // a normal situation.  We may also have to forget some information gained
1141     // through a quick check that was already performed.
1142     if (cp_offset_ != 0) assembler->AdvanceCurrentPosition(cp_offset_);
1143     // Create a new trivial state and generate the node with that.
1144     Trace new_state;
1145     successor->Emit(compiler, &new_state);
1146     return;
1147   }
1148 
1149   // Generate deferred actions here along with code to undo them again.
1150   OutSet affected_registers;
1151 
1152   if (backtrack() != NULL) {
1153     // Here we have a concrete backtrack location.  These are set up by choice
1154     // nodes and so they indicate that we have a deferred save of the current
1155     // position which we may need to emit here.
1156     assembler->PushCurrentPosition();
1157   }
1158 
1159   int max_register = FindAffectedRegisters(&affected_registers);
1160   OutSet registers_to_pop;
1161   OutSet registers_to_clear;
1162   PerformDeferredActions(assembler,
1163                          max_register,
1164                          affected_registers,
1165                          &registers_to_pop,
1166                          &registers_to_clear);
1167   if (cp_offset_ != 0) {
1168     assembler->AdvanceCurrentPosition(cp_offset_);
1169   }
1170 
1171   // Create a new trivial state and generate the node with that.
1172   Label undo;
1173   assembler->PushBacktrack(&undo);
1174   Trace new_state;
1175   successor->Emit(compiler, &new_state);
1176 
1177   // On backtrack we need to restore state.
1178   assembler->Bind(&undo);
1179   RestoreAffectedRegisters(assembler,
1180                            max_register,
1181                            registers_to_pop,
1182                            registers_to_clear);
1183   if (backtrack() == NULL) {
1184     assembler->Backtrack();
1185   } else {
1186     assembler->PopCurrentPosition();
1187     assembler->GoTo(backtrack());
1188   }
1189 }
1190 
1191 
Emit(RegExpCompiler * compiler,Trace * trace)1192 void NegativeSubmatchSuccess::Emit(RegExpCompiler* compiler, Trace* trace) {
1193   RegExpMacroAssembler* assembler = compiler->macro_assembler();
1194 
1195   // Omit flushing the trace. We discard the entire stack frame anyway.
1196 
1197   if (!label()->is_bound()) {
1198     // We are completely independent of the trace, since we ignore it,
1199     // so this code can be used as the generic version.
1200     assembler->Bind(label());
1201   }
1202 
1203   // Throw away everything on the backtrack stack since the start
1204   // of the negative submatch and restore the character position.
1205   assembler->ReadCurrentPositionFromRegister(current_position_register_);
1206   assembler->ReadStackPointerFromRegister(stack_pointer_register_);
1207   if (clear_capture_count_ > 0) {
1208     // Clear any captures that might have been performed during the success
1209     // of the body of the negative look-ahead.
1210     int clear_capture_end = clear_capture_start_ + clear_capture_count_ - 1;
1211     assembler->ClearRegisters(clear_capture_start_, clear_capture_end);
1212   }
1213   // Now that we have unwound the stack we find at the top of the stack the
1214   // backtrack that the BeginSubmatch node got.
1215   assembler->Backtrack();
1216 }
1217 
1218 
Emit(RegExpCompiler * compiler,Trace * trace)1219 void EndNode::Emit(RegExpCompiler* compiler, Trace* trace) {
1220   if (!trace->is_trivial()) {
1221     trace->Flush(compiler, this);
1222     return;
1223   }
1224   RegExpMacroAssembler* assembler = compiler->macro_assembler();
1225   if (!label()->is_bound()) {
1226     assembler->Bind(label());
1227   }
1228   switch (action_) {
1229     case ACCEPT:
1230       assembler->Succeed();
1231       return;
1232     case BACKTRACK:
1233       assembler->GoTo(trace->backtrack());
1234       return;
1235     case NEGATIVE_SUBMATCH_SUCCESS:
1236       // This case is handled in a different virtual method.
1237       UNREACHABLE();
1238   }
1239   UNIMPLEMENTED();
1240 }
1241 
1242 
AddGuard(Guard * guard)1243 void GuardedAlternative::AddGuard(Guard* guard) {
1244   if (guards_ == NULL)
1245     guards_ = new ZoneList<Guard*>(1);
1246   guards_->Add(guard);
1247 }
1248 
1249 
SetRegister(int reg,int val,RegExpNode * on_success)1250 ActionNode* ActionNode::SetRegister(int reg,
1251                                     int val,
1252                                     RegExpNode* on_success) {
1253   ActionNode* result = new ActionNode(SET_REGISTER, on_success);
1254   result->data_.u_store_register.reg = reg;
1255   result->data_.u_store_register.value = val;
1256   return result;
1257 }
1258 
1259 
IncrementRegister(int reg,RegExpNode * on_success)1260 ActionNode* ActionNode::IncrementRegister(int reg, RegExpNode* on_success) {
1261   ActionNode* result = new ActionNode(INCREMENT_REGISTER, on_success);
1262   result->data_.u_increment_register.reg = reg;
1263   return result;
1264 }
1265 
1266 
StorePosition(int reg,bool is_capture,RegExpNode * on_success)1267 ActionNode* ActionNode::StorePosition(int reg,
1268                                       bool is_capture,
1269                                       RegExpNode* on_success) {
1270   ActionNode* result = new ActionNode(STORE_POSITION, on_success);
1271   result->data_.u_position_register.reg = reg;
1272   result->data_.u_position_register.is_capture = is_capture;
1273   return result;
1274 }
1275 
1276 
ClearCaptures(Interval range,RegExpNode * on_success)1277 ActionNode* ActionNode::ClearCaptures(Interval range,
1278                                       RegExpNode* on_success) {
1279   ActionNode* result = new ActionNode(CLEAR_CAPTURES, on_success);
1280   result->data_.u_clear_captures.range_from = range.from();
1281   result->data_.u_clear_captures.range_to = range.to();
1282   return result;
1283 }
1284 
1285 
BeginSubmatch(int stack_reg,int position_reg,RegExpNode * on_success)1286 ActionNode* ActionNode::BeginSubmatch(int stack_reg,
1287                                       int position_reg,
1288                                       RegExpNode* on_success) {
1289   ActionNode* result = new ActionNode(BEGIN_SUBMATCH, on_success);
1290   result->data_.u_submatch.stack_pointer_register = stack_reg;
1291   result->data_.u_submatch.current_position_register = position_reg;
1292   return result;
1293 }
1294 
1295 
PositiveSubmatchSuccess(int stack_reg,int position_reg,int clear_register_count,int clear_register_from,RegExpNode * on_success)1296 ActionNode* ActionNode::PositiveSubmatchSuccess(int stack_reg,
1297                                                 int position_reg,
1298                                                 int clear_register_count,
1299                                                 int clear_register_from,
1300                                                 RegExpNode* on_success) {
1301   ActionNode* result = new ActionNode(POSITIVE_SUBMATCH_SUCCESS, on_success);
1302   result->data_.u_submatch.stack_pointer_register = stack_reg;
1303   result->data_.u_submatch.current_position_register = position_reg;
1304   result->data_.u_submatch.clear_register_count = clear_register_count;
1305   result->data_.u_submatch.clear_register_from = clear_register_from;
1306   return result;
1307 }
1308 
1309 
EmptyMatchCheck(int start_register,int repetition_register,int repetition_limit,RegExpNode * on_success)1310 ActionNode* ActionNode::EmptyMatchCheck(int start_register,
1311                                         int repetition_register,
1312                                         int repetition_limit,
1313                                         RegExpNode* on_success) {
1314   ActionNode* result = new ActionNode(EMPTY_MATCH_CHECK, on_success);
1315   result->data_.u_empty_match_check.start_register = start_register;
1316   result->data_.u_empty_match_check.repetition_register = repetition_register;
1317   result->data_.u_empty_match_check.repetition_limit = repetition_limit;
1318   return result;
1319 }
1320 
1321 
1322 #define DEFINE_ACCEPT(Type)                                          \
1323   void Type##Node::Accept(NodeVisitor* visitor) {                    \
1324     visitor->Visit##Type(this);                                      \
1325   }
FOR_EACH_NODE_TYPE(DEFINE_ACCEPT)1326 FOR_EACH_NODE_TYPE(DEFINE_ACCEPT)
1327 #undef DEFINE_ACCEPT
1328 
1329 
1330 void LoopChoiceNode::Accept(NodeVisitor* visitor) {
1331   visitor->VisitLoopChoice(this);
1332 }
1333 
1334 
1335 // -------------------------------------------------------------------
1336 // Emit code.
1337 
1338 
GenerateGuard(RegExpMacroAssembler * macro_assembler,Guard * guard,Trace * trace)1339 void ChoiceNode::GenerateGuard(RegExpMacroAssembler* macro_assembler,
1340                                Guard* guard,
1341                                Trace* trace) {
1342   switch (guard->op()) {
1343     case Guard::LT:
1344       ASSERT(!trace->mentions_reg(guard->reg()));
1345       macro_assembler->IfRegisterGE(guard->reg(),
1346                                     guard->value(),
1347                                     trace->backtrack());
1348       break;
1349     case Guard::GEQ:
1350       ASSERT(!trace->mentions_reg(guard->reg()));
1351       macro_assembler->IfRegisterLT(guard->reg(),
1352                                     guard->value(),
1353                                     trace->backtrack());
1354       break;
1355   }
1356 }
1357 
1358 
1359 // Returns the number of characters in the equivalence class, omitting those
1360 // that cannot occur in the source string because it is ASCII.
GetCaseIndependentLetters(Isolate * isolate,uc16 character,bool ascii_subject,unibrow::uchar * letters)1361 static int GetCaseIndependentLetters(Isolate* isolate,
1362                                      uc16 character,
1363                                      bool ascii_subject,
1364                                      unibrow::uchar* letters) {
1365   int length =
1366       isolate->jsregexp_uncanonicalize()->get(character, '\0', letters);
1367   // Unibrow returns 0 or 1 for characters where case independence is
1368   // trivial.
1369   if (length == 0) {
1370     letters[0] = character;
1371     length = 1;
1372   }
1373   if (!ascii_subject || character <= String::kMaxAsciiCharCode) {
1374     return length;
1375   }
1376   // The standard requires that non-ASCII characters cannot have ASCII
1377   // character codes in their equivalence class.
1378   return 0;
1379 }
1380 
1381 
EmitSimpleCharacter(Isolate * isolate,RegExpCompiler * compiler,uc16 c,Label * on_failure,int cp_offset,bool check,bool preloaded)1382 static inline bool EmitSimpleCharacter(Isolate* isolate,
1383                                        RegExpCompiler* compiler,
1384                                        uc16 c,
1385                                        Label* on_failure,
1386                                        int cp_offset,
1387                                        bool check,
1388                                        bool preloaded) {
1389   RegExpMacroAssembler* assembler = compiler->macro_assembler();
1390   bool bound_checked = false;
1391   if (!preloaded) {
1392     assembler->LoadCurrentCharacter(
1393         cp_offset,
1394         on_failure,
1395         check);
1396     bound_checked = true;
1397   }
1398   assembler->CheckNotCharacter(c, on_failure);
1399   return bound_checked;
1400 }
1401 
1402 
1403 // Only emits non-letters (things that don't have case).  Only used for case
1404 // independent matches.
EmitAtomNonLetter(Isolate * isolate,RegExpCompiler * compiler,uc16 c,Label * on_failure,int cp_offset,bool check,bool preloaded)1405 static inline bool EmitAtomNonLetter(Isolate* isolate,
1406                                      RegExpCompiler* compiler,
1407                                      uc16 c,
1408                                      Label* on_failure,
1409                                      int cp_offset,
1410                                      bool check,
1411                                      bool preloaded) {
1412   RegExpMacroAssembler* macro_assembler = compiler->macro_assembler();
1413   bool ascii = compiler->ascii();
1414   unibrow::uchar chars[unibrow::Ecma262UnCanonicalize::kMaxWidth];
1415   int length = GetCaseIndependentLetters(isolate, c, ascii, chars);
1416   if (length < 1) {
1417     // This can't match.  Must be an ASCII subject and a non-ASCII character.
1418     // We do not need to do anything since the ASCII pass already handled this.
1419     return false;  // Bounds not checked.
1420   }
1421   bool checked = false;
1422   // We handle the length > 1 case in a later pass.
1423   if (length == 1) {
1424     if (ascii && c > String::kMaxAsciiCharCodeU) {
1425       // Can't match - see above.
1426       return false;  // Bounds not checked.
1427     }
1428     if (!preloaded) {
1429       macro_assembler->LoadCurrentCharacter(cp_offset, on_failure, check);
1430       checked = check;
1431     }
1432     macro_assembler->CheckNotCharacter(c, on_failure);
1433   }
1434   return checked;
1435 }
1436 
1437 
ShortCutEmitCharacterPair(RegExpMacroAssembler * macro_assembler,bool ascii,uc16 c1,uc16 c2,Label * on_failure)1438 static bool ShortCutEmitCharacterPair(RegExpMacroAssembler* macro_assembler,
1439                                       bool ascii,
1440                                       uc16 c1,
1441                                       uc16 c2,
1442                                       Label* on_failure) {
1443   uc16 char_mask;
1444   if (ascii) {
1445     char_mask = String::kMaxAsciiCharCode;
1446   } else {
1447     char_mask = String::kMaxUtf16CodeUnit;
1448   }
1449   uc16 exor = c1 ^ c2;
1450   // Check whether exor has only one bit set.
1451   if (((exor - 1) & exor) == 0) {
1452     // If c1 and c2 differ only by one bit.
1453     // Ecma262UnCanonicalize always gives the highest number last.
1454     ASSERT(c2 > c1);
1455     uc16 mask = char_mask ^ exor;
1456     macro_assembler->CheckNotCharacterAfterAnd(c1, mask, on_failure);
1457     return true;
1458   }
1459   ASSERT(c2 > c1);
1460   uc16 diff = c2 - c1;
1461   if (((diff - 1) & diff) == 0 && c1 >= diff) {
1462     // If the characters differ by 2^n but don't differ by one bit then
1463     // subtract the difference from the found character, then do the or
1464     // trick.  We avoid the theoretical case where negative numbers are
1465     // involved in order to simplify code generation.
1466     uc16 mask = char_mask ^ diff;
1467     macro_assembler->CheckNotCharacterAfterMinusAnd(c1 - diff,
1468                                                     diff,
1469                                                     mask,
1470                                                     on_failure);
1471     return true;
1472   }
1473   return false;
1474 }
1475 
1476 
1477 typedef bool EmitCharacterFunction(Isolate* isolate,
1478                                    RegExpCompiler* compiler,
1479                                    uc16 c,
1480                                    Label* on_failure,
1481                                    int cp_offset,
1482                                    bool check,
1483                                    bool preloaded);
1484 
1485 // Only emits letters (things that have case).  Only used for case independent
1486 // matches.
EmitAtomLetter(Isolate * isolate,RegExpCompiler * compiler,uc16 c,Label * on_failure,int cp_offset,bool check,bool preloaded)1487 static inline bool EmitAtomLetter(Isolate* isolate,
1488                                   RegExpCompiler* compiler,
1489                                   uc16 c,
1490                                   Label* on_failure,
1491                                   int cp_offset,
1492                                   bool check,
1493                                   bool preloaded) {
1494   RegExpMacroAssembler* macro_assembler = compiler->macro_assembler();
1495   bool ascii = compiler->ascii();
1496   unibrow::uchar chars[unibrow::Ecma262UnCanonicalize::kMaxWidth];
1497   int length = GetCaseIndependentLetters(isolate, c, ascii, chars);
1498   if (length <= 1) return false;
1499   // We may not need to check against the end of the input string
1500   // if this character lies before a character that matched.
1501   if (!preloaded) {
1502     macro_assembler->LoadCurrentCharacter(cp_offset, on_failure, check);
1503   }
1504   Label ok;
1505   ASSERT(unibrow::Ecma262UnCanonicalize::kMaxWidth == 4);
1506   switch (length) {
1507     case 2: {
1508       if (ShortCutEmitCharacterPair(macro_assembler,
1509                                     ascii,
1510                                     chars[0],
1511                                     chars[1],
1512                                     on_failure)) {
1513       } else {
1514         macro_assembler->CheckCharacter(chars[0], &ok);
1515         macro_assembler->CheckNotCharacter(chars[1], on_failure);
1516         macro_assembler->Bind(&ok);
1517       }
1518       break;
1519     }
1520     case 4:
1521       macro_assembler->CheckCharacter(chars[3], &ok);
1522       // Fall through!
1523     case 3:
1524       macro_assembler->CheckCharacter(chars[0], &ok);
1525       macro_assembler->CheckCharacter(chars[1], &ok);
1526       macro_assembler->CheckNotCharacter(chars[2], on_failure);
1527       macro_assembler->Bind(&ok);
1528       break;
1529     default:
1530       UNREACHABLE();
1531       break;
1532   }
1533   return true;
1534 }
1535 
1536 
EmitCharClass(RegExpMacroAssembler * macro_assembler,RegExpCharacterClass * cc,bool ascii,Label * on_failure,int cp_offset,bool check_offset,bool preloaded)1537 static void EmitCharClass(RegExpMacroAssembler* macro_assembler,
1538                           RegExpCharacterClass* cc,
1539                           bool ascii,
1540                           Label* on_failure,
1541                           int cp_offset,
1542                           bool check_offset,
1543                           bool preloaded) {
1544   ZoneList<CharacterRange>* ranges = cc->ranges();
1545   int max_char;
1546   if (ascii) {
1547     max_char = String::kMaxAsciiCharCode;
1548   } else {
1549     max_char = String::kMaxUtf16CodeUnit;
1550   }
1551 
1552   Label success;
1553 
1554   Label* char_is_in_class =
1555       cc->is_negated() ? on_failure : &success;
1556 
1557   int range_count = ranges->length();
1558 
1559   int last_valid_range = range_count - 1;
1560   while (last_valid_range >= 0) {
1561     CharacterRange& range = ranges->at(last_valid_range);
1562     if (range.from() <= max_char) {
1563       break;
1564     }
1565     last_valid_range--;
1566   }
1567 
1568   if (last_valid_range < 0) {
1569     if (!cc->is_negated()) {
1570       // TODO(plesner): We can remove this when the node level does our
1571       // ASCII optimizations for us.
1572       macro_assembler->GoTo(on_failure);
1573     }
1574     if (check_offset) {
1575       macro_assembler->CheckPosition(cp_offset, on_failure);
1576     }
1577     return;
1578   }
1579 
1580   if (last_valid_range == 0 &&
1581       !cc->is_negated() &&
1582       ranges->at(0).IsEverything(max_char)) {
1583     // This is a common case hit by non-anchored expressions.
1584     if (check_offset) {
1585       macro_assembler->CheckPosition(cp_offset, on_failure);
1586     }
1587     return;
1588   }
1589 
1590   if (!preloaded) {
1591     macro_assembler->LoadCurrentCharacter(cp_offset, on_failure, check_offset);
1592   }
1593 
1594   if (cc->is_standard() &&
1595         macro_assembler->CheckSpecialCharacterClass(cc->standard_type(),
1596                                                     on_failure)) {
1597       return;
1598   }
1599 
1600   for (int i = 0; i < last_valid_range; i++) {
1601     CharacterRange& range = ranges->at(i);
1602     Label next_range;
1603     uc16 from = range.from();
1604     uc16 to = range.to();
1605     if (from > max_char) {
1606       continue;
1607     }
1608     if (to > max_char) to = max_char;
1609     if (to == from) {
1610       macro_assembler->CheckCharacter(to, char_is_in_class);
1611     } else {
1612       if (from != 0) {
1613         macro_assembler->CheckCharacterLT(from, &next_range);
1614       }
1615       if (to != max_char) {
1616         macro_assembler->CheckCharacterLT(to + 1, char_is_in_class);
1617       } else {
1618         macro_assembler->GoTo(char_is_in_class);
1619       }
1620     }
1621     macro_assembler->Bind(&next_range);
1622   }
1623 
1624   CharacterRange& range = ranges->at(last_valid_range);
1625   uc16 from = range.from();
1626   uc16 to = range.to();
1627 
1628   if (to > max_char) to = max_char;
1629   ASSERT(to >= from);
1630 
1631   if (to == from) {
1632     if (cc->is_negated()) {
1633       macro_assembler->CheckCharacter(to, on_failure);
1634     } else {
1635       macro_assembler->CheckNotCharacter(to, on_failure);
1636     }
1637   } else {
1638     if (from != 0) {
1639       if (cc->is_negated()) {
1640         macro_assembler->CheckCharacterLT(from, &success);
1641       } else {
1642         macro_assembler->CheckCharacterLT(from, on_failure);
1643       }
1644     }
1645     if (to != String::kMaxUtf16CodeUnit) {
1646       if (cc->is_negated()) {
1647         macro_assembler->CheckCharacterLT(to + 1, on_failure);
1648       } else {
1649         macro_assembler->CheckCharacterGT(to, on_failure);
1650       }
1651     } else {
1652       if (cc->is_negated()) {
1653         macro_assembler->GoTo(on_failure);
1654       }
1655     }
1656   }
1657   macro_assembler->Bind(&success);
1658 }
1659 
1660 
~RegExpNode()1661 RegExpNode::~RegExpNode() {
1662 }
1663 
1664 
LimitVersions(RegExpCompiler * compiler,Trace * trace)1665 RegExpNode::LimitResult RegExpNode::LimitVersions(RegExpCompiler* compiler,
1666                                                   Trace* trace) {
1667   // If we are generating a greedy loop then don't stop and don't reuse code.
1668   if (trace->stop_node() != NULL) {
1669     return CONTINUE;
1670   }
1671 
1672   RegExpMacroAssembler* macro_assembler = compiler->macro_assembler();
1673   if (trace->is_trivial()) {
1674     if (label_.is_bound()) {
1675       // We are being asked to generate a generic version, but that's already
1676       // been done so just go to it.
1677       macro_assembler->GoTo(&label_);
1678       return DONE;
1679     }
1680     if (compiler->recursion_depth() >= RegExpCompiler::kMaxRecursion) {
1681       // To avoid too deep recursion we push the node to the work queue and just
1682       // generate a goto here.
1683       compiler->AddWork(this);
1684       macro_assembler->GoTo(&label_);
1685       return DONE;
1686     }
1687     // Generate generic version of the node and bind the label for later use.
1688     macro_assembler->Bind(&label_);
1689     return CONTINUE;
1690   }
1691 
1692   // We are being asked to make a non-generic version.  Keep track of how many
1693   // non-generic versions we generate so as not to overdo it.
1694   trace_count_++;
1695   if (FLAG_regexp_optimization &&
1696       trace_count_ < kMaxCopiesCodeGenerated &&
1697       compiler->recursion_depth() <= RegExpCompiler::kMaxRecursion) {
1698     return CONTINUE;
1699   }
1700 
1701   // If we get here code has been generated for this node too many times or
1702   // recursion is too deep.  Time to switch to a generic version.  The code for
1703   // generic versions above can handle deep recursion properly.
1704   trace->Flush(compiler, this);
1705   return DONE;
1706 }
1707 
1708 
EatsAtLeast(int still_to_find,int recursion_depth,bool not_at_start)1709 int ActionNode::EatsAtLeast(int still_to_find,
1710                             int recursion_depth,
1711                             bool not_at_start) {
1712   if (recursion_depth > RegExpCompiler::kMaxRecursion) return 0;
1713   if (type_ == POSITIVE_SUBMATCH_SUCCESS) return 0;  // Rewinds input!
1714   return on_success()->EatsAtLeast(still_to_find,
1715                                    recursion_depth + 1,
1716                                    not_at_start);
1717 }
1718 
1719 
EatsAtLeast(int still_to_find,int recursion_depth,bool not_at_start)1720 int AssertionNode::EatsAtLeast(int still_to_find,
1721                                int recursion_depth,
1722                                bool not_at_start) {
1723   if (recursion_depth > RegExpCompiler::kMaxRecursion) return 0;
1724   // If we know we are not at the start and we are asked "how many characters
1725   // will you match if you succeed?" then we can answer anything since false
1726   // implies false.  So lets just return the max answer (still_to_find) since
1727   // that won't prevent us from preloading a lot of characters for the other
1728   // branches in the node graph.
1729   if (type() == AT_START && not_at_start) return still_to_find;
1730   return on_success()->EatsAtLeast(still_to_find,
1731                                    recursion_depth + 1,
1732                                    not_at_start);
1733 }
1734 
1735 
EatsAtLeast(int still_to_find,int recursion_depth,bool not_at_start)1736 int BackReferenceNode::EatsAtLeast(int still_to_find,
1737                                    int recursion_depth,
1738                                    bool not_at_start) {
1739   if (recursion_depth > RegExpCompiler::kMaxRecursion) return 0;
1740   return on_success()->EatsAtLeast(still_to_find,
1741                                    recursion_depth + 1,
1742                                    not_at_start);
1743 }
1744 
1745 
EatsAtLeast(int still_to_find,int recursion_depth,bool not_at_start)1746 int TextNode::EatsAtLeast(int still_to_find,
1747                           int recursion_depth,
1748                           bool not_at_start) {
1749   int answer = Length();
1750   if (answer >= still_to_find) return answer;
1751   if (recursion_depth > RegExpCompiler::kMaxRecursion) return answer;
1752   // We are not at start after this node so we set the last argument to 'true'.
1753   return answer + on_success()->EatsAtLeast(still_to_find - answer,
1754                                             recursion_depth + 1,
1755                                             true);
1756 }
1757 
1758 
EatsAtLeast(int still_to_find,int recursion_depth,bool not_at_start)1759 int NegativeLookaheadChoiceNode::EatsAtLeast(int still_to_find,
1760                                              int recursion_depth,
1761                                              bool not_at_start) {
1762   if (recursion_depth > RegExpCompiler::kMaxRecursion) return 0;
1763   // Alternative 0 is the negative lookahead, alternative 1 is what comes
1764   // afterwards.
1765   RegExpNode* node = alternatives_->at(1).node();
1766   return node->EatsAtLeast(still_to_find, recursion_depth + 1, not_at_start);
1767 }
1768 
1769 
GetQuickCheckDetails(QuickCheckDetails * details,RegExpCompiler * compiler,int filled_in,bool not_at_start)1770 void NegativeLookaheadChoiceNode::GetQuickCheckDetails(
1771     QuickCheckDetails* details,
1772     RegExpCompiler* compiler,
1773     int filled_in,
1774     bool not_at_start) {
1775   // Alternative 0 is the negative lookahead, alternative 1 is what comes
1776   // afterwards.
1777   RegExpNode* node = alternatives_->at(1).node();
1778   return node->GetQuickCheckDetails(details, compiler, filled_in, not_at_start);
1779 }
1780 
1781 
EatsAtLeastHelper(int still_to_find,int recursion_depth,RegExpNode * ignore_this_node,bool not_at_start)1782 int ChoiceNode::EatsAtLeastHelper(int still_to_find,
1783                                   int recursion_depth,
1784                                   RegExpNode* ignore_this_node,
1785                                   bool not_at_start) {
1786   if (recursion_depth > RegExpCompiler::kMaxRecursion) return 0;
1787   int min = 100;
1788   int choice_count = alternatives_->length();
1789   for (int i = 0; i < choice_count; i++) {
1790     RegExpNode* node = alternatives_->at(i).node();
1791     if (node == ignore_this_node) continue;
1792     int node_eats_at_least = node->EatsAtLeast(still_to_find,
1793                                                recursion_depth + 1,
1794                                                not_at_start);
1795     if (node_eats_at_least < min) min = node_eats_at_least;
1796   }
1797   return min;
1798 }
1799 
1800 
EatsAtLeast(int still_to_find,int recursion_depth,bool not_at_start)1801 int LoopChoiceNode::EatsAtLeast(int still_to_find,
1802                                 int recursion_depth,
1803                                 bool not_at_start) {
1804   return EatsAtLeastHelper(still_to_find,
1805                            recursion_depth,
1806                            loop_node_,
1807                            not_at_start);
1808 }
1809 
1810 
EatsAtLeast(int still_to_find,int recursion_depth,bool not_at_start)1811 int ChoiceNode::EatsAtLeast(int still_to_find,
1812                             int recursion_depth,
1813                             bool not_at_start) {
1814   return EatsAtLeastHelper(still_to_find,
1815                            recursion_depth,
1816                            NULL,
1817                            not_at_start);
1818 }
1819 
1820 
1821 // Takes the left-most 1-bit and smears it out, setting all bits to its right.
SmearBitsRight(uint32_t v)1822 static inline uint32_t SmearBitsRight(uint32_t v) {
1823   v |= v >> 1;
1824   v |= v >> 2;
1825   v |= v >> 4;
1826   v |= v >> 8;
1827   v |= v >> 16;
1828   return v;
1829 }
1830 
1831 
Rationalize(bool asc)1832 bool QuickCheckDetails::Rationalize(bool asc) {
1833   bool found_useful_op = false;
1834   uint32_t char_mask;
1835   if (asc) {
1836     char_mask = String::kMaxAsciiCharCode;
1837   } else {
1838     char_mask = String::kMaxUtf16CodeUnit;
1839   }
1840   mask_ = 0;
1841   value_ = 0;
1842   int char_shift = 0;
1843   for (int i = 0; i < characters_; i++) {
1844     Position* pos = &positions_[i];
1845     if ((pos->mask & String::kMaxAsciiCharCode) != 0) {
1846       found_useful_op = true;
1847     }
1848     mask_ |= (pos->mask & char_mask) << char_shift;
1849     value_ |= (pos->value & char_mask) << char_shift;
1850     char_shift += asc ? 8 : 16;
1851   }
1852   return found_useful_op;
1853 }
1854 
1855 
EmitQuickCheck(RegExpCompiler * compiler,Trace * trace,bool preload_has_checked_bounds,Label * on_possible_success,QuickCheckDetails * details,bool fall_through_on_failure)1856 bool RegExpNode::EmitQuickCheck(RegExpCompiler* compiler,
1857                                 Trace* trace,
1858                                 bool preload_has_checked_bounds,
1859                                 Label* on_possible_success,
1860                                 QuickCheckDetails* details,
1861                                 bool fall_through_on_failure) {
1862   if (details->characters() == 0) return false;
1863   GetQuickCheckDetails(details, compiler, 0, trace->at_start() == Trace::FALSE);
1864   if (details->cannot_match()) return false;
1865   if (!details->Rationalize(compiler->ascii())) return false;
1866   ASSERT(details->characters() == 1 ||
1867          compiler->macro_assembler()->CanReadUnaligned());
1868   uint32_t mask = details->mask();
1869   uint32_t value = details->value();
1870 
1871   RegExpMacroAssembler* assembler = compiler->macro_assembler();
1872 
1873   if (trace->characters_preloaded() != details->characters()) {
1874     assembler->LoadCurrentCharacter(trace->cp_offset(),
1875                                     trace->backtrack(),
1876                                     !preload_has_checked_bounds,
1877                                     details->characters());
1878   }
1879 
1880 
1881   bool need_mask = true;
1882 
1883   if (details->characters() == 1) {
1884     // If number of characters preloaded is 1 then we used a byte or 16 bit
1885     // load so the value is already masked down.
1886     uint32_t char_mask;
1887     if (compiler->ascii()) {
1888       char_mask = String::kMaxAsciiCharCode;
1889     } else {
1890       char_mask = String::kMaxUtf16CodeUnit;
1891     }
1892     if ((mask & char_mask) == char_mask) need_mask = false;
1893     mask &= char_mask;
1894   } else {
1895     // For 2-character preloads in ASCII mode or 1-character preloads in
1896     // TWO_BYTE mode we also use a 16 bit load with zero extend.
1897     if (details->characters() == 2 && compiler->ascii()) {
1898       if ((mask & 0x7f7f) == 0x7f7f) need_mask = false;
1899     } else if (details->characters() == 1 && !compiler->ascii()) {
1900       if ((mask & 0xffff) == 0xffff) need_mask = false;
1901     } else {
1902       if (mask == 0xffffffff) need_mask = false;
1903     }
1904   }
1905 
1906   if (fall_through_on_failure) {
1907     if (need_mask) {
1908       assembler->CheckCharacterAfterAnd(value, mask, on_possible_success);
1909     } else {
1910       assembler->CheckCharacter(value, on_possible_success);
1911     }
1912   } else {
1913     if (need_mask) {
1914       assembler->CheckNotCharacterAfterAnd(value, mask, trace->backtrack());
1915     } else {
1916       assembler->CheckNotCharacter(value, trace->backtrack());
1917     }
1918   }
1919   return true;
1920 }
1921 
1922 
1923 // Here is the meat of GetQuickCheckDetails (see also the comment on the
1924 // super-class in the .h file).
1925 //
1926 // We iterate along the text object, building up for each character a
1927 // mask and value that can be used to test for a quick failure to match.
1928 // The masks and values for the positions will be combined into a single
1929 // machine word for the current character width in order to be used in
1930 // generating a quick check.
GetQuickCheckDetails(QuickCheckDetails * details,RegExpCompiler * compiler,int characters_filled_in,bool not_at_start)1931 void TextNode::GetQuickCheckDetails(QuickCheckDetails* details,
1932                                     RegExpCompiler* compiler,
1933                                     int characters_filled_in,
1934                                     bool not_at_start) {
1935   Isolate* isolate = Isolate::Current();
1936   ASSERT(characters_filled_in < details->characters());
1937   int characters = details->characters();
1938   int char_mask;
1939   if (compiler->ascii()) {
1940     char_mask = String::kMaxAsciiCharCode;
1941   } else {
1942     char_mask = String::kMaxUtf16CodeUnit;
1943   }
1944   for (int k = 0; k < elms_->length(); k++) {
1945     TextElement elm = elms_->at(k);
1946     if (elm.type == TextElement::ATOM) {
1947       Vector<const uc16> quarks = elm.data.u_atom->data();
1948       for (int i = 0; i < characters && i < quarks.length(); i++) {
1949         QuickCheckDetails::Position* pos =
1950             details->positions(characters_filled_in);
1951         uc16 c = quarks[i];
1952         if (c > char_mask) {
1953           // If we expect a non-ASCII character from an ASCII string,
1954           // there is no way we can match. Not even case independent
1955           // matching can turn an ASCII character into non-ASCII or
1956           // vice versa.
1957           details->set_cannot_match();
1958           pos->determines_perfectly = false;
1959           return;
1960         }
1961         if (compiler->ignore_case()) {
1962           unibrow::uchar chars[unibrow::Ecma262UnCanonicalize::kMaxWidth];
1963           int length = GetCaseIndependentLetters(isolate, c, compiler->ascii(),
1964                                                  chars);
1965           ASSERT(length != 0);  // Can only happen if c > char_mask (see above).
1966           if (length == 1) {
1967             // This letter has no case equivalents, so it's nice and simple
1968             // and the mask-compare will determine definitely whether we have
1969             // a match at this character position.
1970             pos->mask = char_mask;
1971             pos->value = c;
1972             pos->determines_perfectly = true;
1973           } else {
1974             uint32_t common_bits = char_mask;
1975             uint32_t bits = chars[0];
1976             for (int j = 1; j < length; j++) {
1977               uint32_t differing_bits = ((chars[j] & common_bits) ^ bits);
1978               common_bits ^= differing_bits;
1979               bits &= common_bits;
1980             }
1981             // If length is 2 and common bits has only one zero in it then
1982             // our mask and compare instruction will determine definitely
1983             // whether we have a match at this character position.  Otherwise
1984             // it can only be an approximate check.
1985             uint32_t one_zero = (common_bits | ~char_mask);
1986             if (length == 2 && ((~one_zero) & ((~one_zero) - 1)) == 0) {
1987               pos->determines_perfectly = true;
1988             }
1989             pos->mask = common_bits;
1990             pos->value = bits;
1991           }
1992         } else {
1993           // Don't ignore case.  Nice simple case where the mask-compare will
1994           // determine definitely whether we have a match at this character
1995           // position.
1996           pos->mask = char_mask;
1997           pos->value = c;
1998           pos->determines_perfectly = true;
1999         }
2000         characters_filled_in++;
2001         ASSERT(characters_filled_in <= details->characters());
2002         if (characters_filled_in == details->characters()) {
2003           return;
2004         }
2005       }
2006     } else {
2007       QuickCheckDetails::Position* pos =
2008           details->positions(characters_filled_in);
2009       RegExpCharacterClass* tree = elm.data.u_char_class;
2010       ZoneList<CharacterRange>* ranges = tree->ranges();
2011       if (tree->is_negated()) {
2012         // A quick check uses multi-character mask and compare.  There is no
2013         // useful way to incorporate a negative char class into this scheme
2014         // so we just conservatively create a mask and value that will always
2015         // succeed.
2016         pos->mask = 0;
2017         pos->value = 0;
2018       } else {
2019         int first_range = 0;
2020         while (ranges->at(first_range).from() > char_mask) {
2021           first_range++;
2022           if (first_range == ranges->length()) {
2023             details->set_cannot_match();
2024             pos->determines_perfectly = false;
2025             return;
2026           }
2027         }
2028         CharacterRange range = ranges->at(first_range);
2029         uc16 from = range.from();
2030         uc16 to = range.to();
2031         if (to > char_mask) {
2032           to = char_mask;
2033         }
2034         uint32_t differing_bits = (from ^ to);
2035         // A mask and compare is only perfect if the differing bits form a
2036         // number like 00011111 with one single block of trailing 1s.
2037         if ((differing_bits & (differing_bits + 1)) == 0 &&
2038              from + differing_bits == to) {
2039           pos->determines_perfectly = true;
2040         }
2041         uint32_t common_bits = ~SmearBitsRight(differing_bits);
2042         uint32_t bits = (from & common_bits);
2043         for (int i = first_range + 1; i < ranges->length(); i++) {
2044           CharacterRange range = ranges->at(i);
2045           uc16 from = range.from();
2046           uc16 to = range.to();
2047           if (from > char_mask) continue;
2048           if (to > char_mask) to = char_mask;
2049           // Here we are combining more ranges into the mask and compare
2050           // value.  With each new range the mask becomes more sparse and
2051           // so the chances of a false positive rise.  A character class
2052           // with multiple ranges is assumed never to be equivalent to a
2053           // mask and compare operation.
2054           pos->determines_perfectly = false;
2055           uint32_t new_common_bits = (from ^ to);
2056           new_common_bits = ~SmearBitsRight(new_common_bits);
2057           common_bits &= new_common_bits;
2058           bits &= new_common_bits;
2059           uint32_t differing_bits = (from & common_bits) ^ bits;
2060           common_bits ^= differing_bits;
2061           bits &= common_bits;
2062         }
2063         pos->mask = common_bits;
2064         pos->value = bits;
2065       }
2066       characters_filled_in++;
2067       ASSERT(characters_filled_in <= details->characters());
2068       if (characters_filled_in == details->characters()) {
2069         return;
2070       }
2071     }
2072   }
2073   ASSERT(characters_filled_in != details->characters());
2074   on_success()-> GetQuickCheckDetails(details,
2075                                       compiler,
2076                                       characters_filled_in,
2077                                       true);
2078 }
2079 
2080 
Clear()2081 void QuickCheckDetails::Clear() {
2082   for (int i = 0; i < characters_; i++) {
2083     positions_[i].mask = 0;
2084     positions_[i].value = 0;
2085     positions_[i].determines_perfectly = false;
2086   }
2087   characters_ = 0;
2088 }
2089 
2090 
Advance(int by,bool ascii)2091 void QuickCheckDetails::Advance(int by, bool ascii) {
2092   ASSERT(by >= 0);
2093   if (by >= characters_) {
2094     Clear();
2095     return;
2096   }
2097   for (int i = 0; i < characters_ - by; i++) {
2098     positions_[i] = positions_[by + i];
2099   }
2100   for (int i = characters_ - by; i < characters_; i++) {
2101     positions_[i].mask = 0;
2102     positions_[i].value = 0;
2103     positions_[i].determines_perfectly = false;
2104   }
2105   characters_ -= by;
2106   // We could change mask_ and value_ here but we would never advance unless
2107   // they had already been used in a check and they won't be used again because
2108   // it would gain us nothing.  So there's no point.
2109 }
2110 
2111 
Merge(QuickCheckDetails * other,int from_index)2112 void QuickCheckDetails::Merge(QuickCheckDetails* other, int from_index) {
2113   ASSERT(characters_ == other->characters_);
2114   if (other->cannot_match_) {
2115     return;
2116   }
2117   if (cannot_match_) {
2118     *this = *other;
2119     return;
2120   }
2121   for (int i = from_index; i < characters_; i++) {
2122     QuickCheckDetails::Position* pos = positions(i);
2123     QuickCheckDetails::Position* other_pos = other->positions(i);
2124     if (pos->mask != other_pos->mask ||
2125         pos->value != other_pos->value ||
2126         !other_pos->determines_perfectly) {
2127       // Our mask-compare operation will be approximate unless we have the
2128       // exact same operation on both sides of the alternation.
2129       pos->determines_perfectly = false;
2130     }
2131     pos->mask &= other_pos->mask;
2132     pos->value &= pos->mask;
2133     other_pos->value &= pos->mask;
2134     uc16 differing_bits = (pos->value ^ other_pos->value);
2135     pos->mask &= ~differing_bits;
2136     pos->value &= pos->mask;
2137   }
2138 }
2139 
2140 
2141 class VisitMarker {
2142  public:
VisitMarker(NodeInfo * info)2143   explicit VisitMarker(NodeInfo* info) : info_(info) {
2144     ASSERT(!info->visited);
2145     info->visited = true;
2146   }
~VisitMarker()2147   ~VisitMarker() {
2148     info_->visited = false;
2149   }
2150  private:
2151   NodeInfo* info_;
2152 };
2153 
2154 
GetQuickCheckDetails(QuickCheckDetails * details,RegExpCompiler * compiler,int characters_filled_in,bool not_at_start)2155 void LoopChoiceNode::GetQuickCheckDetails(QuickCheckDetails* details,
2156                                           RegExpCompiler* compiler,
2157                                           int characters_filled_in,
2158                                           bool not_at_start) {
2159   if (body_can_be_zero_length_ || info()->visited) return;
2160   VisitMarker marker(info());
2161   return ChoiceNode::GetQuickCheckDetails(details,
2162                                           compiler,
2163                                           characters_filled_in,
2164                                           not_at_start);
2165 }
2166 
2167 
GetQuickCheckDetails(QuickCheckDetails * details,RegExpCompiler * compiler,int characters_filled_in,bool not_at_start)2168 void ChoiceNode::GetQuickCheckDetails(QuickCheckDetails* details,
2169                                       RegExpCompiler* compiler,
2170                                       int characters_filled_in,
2171                                       bool not_at_start) {
2172   not_at_start = (not_at_start || not_at_start_);
2173   int choice_count = alternatives_->length();
2174   ASSERT(choice_count > 0);
2175   alternatives_->at(0).node()->GetQuickCheckDetails(details,
2176                                                     compiler,
2177                                                     characters_filled_in,
2178                                                     not_at_start);
2179   for (int i = 1; i < choice_count; i++) {
2180     QuickCheckDetails new_details(details->characters());
2181     RegExpNode* node = alternatives_->at(i).node();
2182     node->GetQuickCheckDetails(&new_details, compiler,
2183                                characters_filled_in,
2184                                not_at_start);
2185     // Here we merge the quick match details of the two branches.
2186     details->Merge(&new_details, characters_filled_in);
2187   }
2188 }
2189 
2190 
2191 // Check for [0-9A-Z_a-z].
EmitWordCheck(RegExpMacroAssembler * assembler,Label * word,Label * non_word,bool fall_through_on_word)2192 static void EmitWordCheck(RegExpMacroAssembler* assembler,
2193                           Label* word,
2194                           Label* non_word,
2195                           bool fall_through_on_word) {
2196   if (assembler->CheckSpecialCharacterClass(
2197           fall_through_on_word ? 'w' : 'W',
2198           fall_through_on_word ? non_word : word)) {
2199     // Optimized implementation available.
2200     return;
2201   }
2202   assembler->CheckCharacterGT('z', non_word);
2203   assembler->CheckCharacterLT('0', non_word);
2204   assembler->CheckCharacterGT('a' - 1, word);
2205   assembler->CheckCharacterLT('9' + 1, word);
2206   assembler->CheckCharacterLT('A', non_word);
2207   assembler->CheckCharacterLT('Z' + 1, word);
2208   if (fall_through_on_word) {
2209     assembler->CheckNotCharacter('_', non_word);
2210   } else {
2211     assembler->CheckCharacter('_', word);
2212   }
2213 }
2214 
2215 
2216 // Emit the code to check for a ^ in multiline mode (1-character lookbehind
2217 // that matches newline or the start of input).
EmitHat(RegExpCompiler * compiler,RegExpNode * on_success,Trace * trace)2218 static void EmitHat(RegExpCompiler* compiler,
2219                     RegExpNode* on_success,
2220                     Trace* trace) {
2221   RegExpMacroAssembler* assembler = compiler->macro_assembler();
2222   // We will be loading the previous character into the current character
2223   // register.
2224   Trace new_trace(*trace);
2225   new_trace.InvalidateCurrentCharacter();
2226 
2227   Label ok;
2228   if (new_trace.cp_offset() == 0) {
2229     // The start of input counts as a newline in this context, so skip to
2230     // ok if we are at the start.
2231     assembler->CheckAtStart(&ok);
2232   }
2233   // We already checked that we are not at the start of input so it must be
2234   // OK to load the previous character.
2235   assembler->LoadCurrentCharacter(new_trace.cp_offset() -1,
2236                                   new_trace.backtrack(),
2237                                   false);
2238   if (!assembler->CheckSpecialCharacterClass('n',
2239                                              new_trace.backtrack())) {
2240     // Newline means \n, \r, 0x2028 or 0x2029.
2241     if (!compiler->ascii()) {
2242       assembler->CheckCharacterAfterAnd(0x2028, 0xfffe, &ok);
2243     }
2244     assembler->CheckCharacter('\n', &ok);
2245     assembler->CheckNotCharacter('\r', new_trace.backtrack());
2246   }
2247   assembler->Bind(&ok);
2248   on_success->Emit(compiler, &new_trace);
2249 }
2250 
2251 
2252 // Emit the code to handle \b and \B (word-boundary or non-word-boundary)
2253 // when we know whether the next character must be a word character or not.
EmitHalfBoundaryCheck(AssertionNode::AssertionNodeType type,RegExpCompiler * compiler,RegExpNode * on_success,Trace * trace)2254 static void EmitHalfBoundaryCheck(AssertionNode::AssertionNodeType type,
2255                                   RegExpCompiler* compiler,
2256                                   RegExpNode* on_success,
2257                                   Trace* trace) {
2258   RegExpMacroAssembler* assembler = compiler->macro_assembler();
2259   Label done;
2260 
2261   Trace new_trace(*trace);
2262 
2263   bool expect_word_character = (type == AssertionNode::AFTER_WORD_CHARACTER);
2264   Label* on_word = expect_word_character ? &done : new_trace.backtrack();
2265   Label* on_non_word = expect_word_character ? new_trace.backtrack() : &done;
2266 
2267   // Check whether previous character was a word character.
2268   switch (trace->at_start()) {
2269     case Trace::TRUE:
2270       if (expect_word_character) {
2271         assembler->GoTo(on_non_word);
2272       }
2273       break;
2274     case Trace::UNKNOWN:
2275       ASSERT_EQ(0, trace->cp_offset());
2276       assembler->CheckAtStart(on_non_word);
2277       // Fall through.
2278     case Trace::FALSE:
2279       int prev_char_offset = trace->cp_offset() - 1;
2280       assembler->LoadCurrentCharacter(prev_char_offset, NULL, false, 1);
2281       EmitWordCheck(assembler, on_word, on_non_word, expect_word_character);
2282       // We may or may not have loaded the previous character.
2283       new_trace.InvalidateCurrentCharacter();
2284   }
2285 
2286   assembler->Bind(&done);
2287 
2288   on_success->Emit(compiler, &new_trace);
2289 }
2290 
2291 
2292 // Emit the code to handle \b and \B (word-boundary or non-word-boundary).
EmitBoundaryCheck(AssertionNode::AssertionNodeType type,RegExpCompiler * compiler,RegExpNode * on_success,Trace * trace)2293 static void EmitBoundaryCheck(AssertionNode::AssertionNodeType type,
2294                               RegExpCompiler* compiler,
2295                               RegExpNode* on_success,
2296                               Trace* trace) {
2297   RegExpMacroAssembler* assembler = compiler->macro_assembler();
2298   Label before_non_word;
2299   Label before_word;
2300   if (trace->characters_preloaded() != 1) {
2301     assembler->LoadCurrentCharacter(trace->cp_offset(), &before_non_word);
2302   }
2303   // Fall through on non-word.
2304   EmitWordCheck(assembler, &before_word, &before_non_word, false);
2305 
2306   // We will be loading the previous character into the current character
2307   // register.
2308   Trace new_trace(*trace);
2309   new_trace.InvalidateCurrentCharacter();
2310 
2311   Label ok;
2312   Label* boundary;
2313   Label* not_boundary;
2314   if (type == AssertionNode::AT_BOUNDARY) {
2315     boundary = &ok;
2316     not_boundary = new_trace.backtrack();
2317   } else {
2318     not_boundary = &ok;
2319     boundary = new_trace.backtrack();
2320   }
2321 
2322   // Next character is not a word character.
2323   assembler->Bind(&before_non_word);
2324   if (new_trace.cp_offset() == 0) {
2325     // The start of input counts as a non-word character, so the question is
2326     // decided if we are at the start.
2327     assembler->CheckAtStart(not_boundary);
2328   }
2329   // We already checked that we are not at the start of input so it must be
2330   // OK to load the previous character.
2331   assembler->LoadCurrentCharacter(new_trace.cp_offset() - 1,
2332                                   &ok,  // Unused dummy label in this call.
2333                                   false);
2334   // Fall through on non-word.
2335   EmitWordCheck(assembler, boundary, not_boundary, false);
2336   assembler->GoTo(not_boundary);
2337 
2338   // Next character is a word character.
2339   assembler->Bind(&before_word);
2340   if (new_trace.cp_offset() == 0) {
2341     // The start of input counts as a non-word character, so the question is
2342     // decided if we are at the start.
2343     assembler->CheckAtStart(boundary);
2344   }
2345   // We already checked that we are not at the start of input so it must be
2346   // OK to load the previous character.
2347   assembler->LoadCurrentCharacter(new_trace.cp_offset() - 1,
2348                                   &ok,  // Unused dummy label in this call.
2349                                   false);
2350   bool fall_through_on_word = (type == AssertionNode::AT_NON_BOUNDARY);
2351   EmitWordCheck(assembler, not_boundary, boundary, fall_through_on_word);
2352 
2353   assembler->Bind(&ok);
2354 
2355   on_success->Emit(compiler, &new_trace);
2356 }
2357 
2358 
GetQuickCheckDetails(QuickCheckDetails * details,RegExpCompiler * compiler,int filled_in,bool not_at_start)2359 void AssertionNode::GetQuickCheckDetails(QuickCheckDetails* details,
2360                                          RegExpCompiler* compiler,
2361                                          int filled_in,
2362                                          bool not_at_start) {
2363   if (type_ == AT_START && not_at_start) {
2364     details->set_cannot_match();
2365     return;
2366   }
2367   return on_success()->GetQuickCheckDetails(details,
2368                                             compiler,
2369                                             filled_in,
2370                                             not_at_start);
2371 }
2372 
2373 
Emit(RegExpCompiler * compiler,Trace * trace)2374 void AssertionNode::Emit(RegExpCompiler* compiler, Trace* trace) {
2375   RegExpMacroAssembler* assembler = compiler->macro_assembler();
2376   switch (type_) {
2377     case AT_END: {
2378       Label ok;
2379       assembler->CheckPosition(trace->cp_offset(), &ok);
2380       assembler->GoTo(trace->backtrack());
2381       assembler->Bind(&ok);
2382       break;
2383     }
2384     case AT_START: {
2385       if (trace->at_start() == Trace::FALSE) {
2386         assembler->GoTo(trace->backtrack());
2387         return;
2388       }
2389       if (trace->at_start() == Trace::UNKNOWN) {
2390         assembler->CheckNotAtStart(trace->backtrack());
2391         Trace at_start_trace = *trace;
2392         at_start_trace.set_at_start(true);
2393         on_success()->Emit(compiler, &at_start_trace);
2394         return;
2395       }
2396     }
2397     break;
2398     case AFTER_NEWLINE:
2399       EmitHat(compiler, on_success(), trace);
2400       return;
2401     case AT_BOUNDARY:
2402     case AT_NON_BOUNDARY: {
2403       EmitBoundaryCheck(type_, compiler, on_success(), trace);
2404       return;
2405     }
2406     case AFTER_WORD_CHARACTER:
2407     case AFTER_NONWORD_CHARACTER: {
2408       EmitHalfBoundaryCheck(type_, compiler, on_success(), trace);
2409     }
2410   }
2411   on_success()->Emit(compiler, trace);
2412 }
2413 
2414 
DeterminedAlready(QuickCheckDetails * quick_check,int offset)2415 static bool DeterminedAlready(QuickCheckDetails* quick_check, int offset) {
2416   if (quick_check == NULL) return false;
2417   if (offset >= quick_check->characters()) return false;
2418   return quick_check->positions(offset)->determines_perfectly;
2419 }
2420 
2421 
UpdateBoundsCheck(int index,int * checked_up_to)2422 static void UpdateBoundsCheck(int index, int* checked_up_to) {
2423   if (index > *checked_up_to) {
2424     *checked_up_to = index;
2425   }
2426 }
2427 
2428 
2429 // We call this repeatedly to generate code for each pass over the text node.
2430 // The passes are in increasing order of difficulty because we hope one
2431 // of the first passes will fail in which case we are saved the work of the
2432 // later passes.  for example for the case independent regexp /%[asdfghjkl]a/
2433 // we will check the '%' in the first pass, the case independent 'a' in the
2434 // second pass and the character class in the last pass.
2435 //
2436 // The passes are done from right to left, so for example to test for /bar/
2437 // we will first test for an 'r' with offset 2, then an 'a' with offset 1
2438 // and then a 'b' with offset 0.  This means we can avoid the end-of-input
2439 // bounds check most of the time.  In the example we only need to check for
2440 // end-of-input when loading the putative 'r'.
2441 //
2442 // A slight complication involves the fact that the first character may already
2443 // be fetched into a register by the previous node.  In this case we want to
2444 // do the test for that character first.  We do this in separate passes.  The
2445 // 'preloaded' argument indicates that we are doing such a 'pass'.  If such a
2446 // pass has been performed then subsequent passes will have true in
2447 // first_element_checked to indicate that that character does not need to be
2448 // checked again.
2449 //
2450 // In addition to all this we are passed a Trace, which can
2451 // contain an AlternativeGeneration object.  In this AlternativeGeneration
2452 // object we can see details of any quick check that was already passed in
2453 // order to get to the code we are now generating.  The quick check can involve
2454 // loading characters, which means we do not need to recheck the bounds
2455 // up to the limit the quick check already checked.  In addition the quick
2456 // check can have involved a mask and compare operation which may simplify
2457 // 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)2458 void TextNode::TextEmitPass(RegExpCompiler* compiler,
2459                             TextEmitPassType pass,
2460                             bool preloaded,
2461                             Trace* trace,
2462                             bool first_element_checked,
2463                             int* checked_up_to) {
2464   Isolate* isolate = Isolate::Current();
2465   RegExpMacroAssembler* assembler = compiler->macro_assembler();
2466   bool ascii = compiler->ascii();
2467   Label* backtrack = trace->backtrack();
2468   QuickCheckDetails* quick_check = trace->quick_check_performed();
2469   int element_count = elms_->length();
2470   for (int i = preloaded ? 0 : element_count - 1; i >= 0; i--) {
2471     TextElement elm = elms_->at(i);
2472     int cp_offset = trace->cp_offset() + elm.cp_offset;
2473     if (elm.type == TextElement::ATOM) {
2474       Vector<const uc16> quarks = elm.data.u_atom->data();
2475       for (int j = preloaded ? 0 : quarks.length() - 1; j >= 0; j--) {
2476         if (first_element_checked && i == 0 && j == 0) continue;
2477         if (DeterminedAlready(quick_check, elm.cp_offset + j)) continue;
2478         EmitCharacterFunction* emit_function = NULL;
2479         switch (pass) {
2480           case NON_ASCII_MATCH:
2481             ASSERT(ascii);
2482             if (quarks[j] > String::kMaxAsciiCharCode) {
2483               assembler->GoTo(backtrack);
2484               return;
2485             }
2486             break;
2487           case NON_LETTER_CHARACTER_MATCH:
2488             emit_function = &EmitAtomNonLetter;
2489             break;
2490           case SIMPLE_CHARACTER_MATCH:
2491             emit_function = &EmitSimpleCharacter;
2492             break;
2493           case CASE_CHARACTER_MATCH:
2494             emit_function = &EmitAtomLetter;
2495             break;
2496           default:
2497             break;
2498         }
2499         if (emit_function != NULL) {
2500           bool bound_checked = emit_function(isolate,
2501                                              compiler,
2502                                              quarks[j],
2503                                              backtrack,
2504                                              cp_offset + j,
2505                                              *checked_up_to < cp_offset + j,
2506                                              preloaded);
2507           if (bound_checked) UpdateBoundsCheck(cp_offset + j, checked_up_to);
2508         }
2509       }
2510     } else {
2511       ASSERT_EQ(elm.type, TextElement::CHAR_CLASS);
2512       if (pass == CHARACTER_CLASS_MATCH) {
2513         if (first_element_checked && i == 0) continue;
2514         if (DeterminedAlready(quick_check, elm.cp_offset)) continue;
2515         RegExpCharacterClass* cc = elm.data.u_char_class;
2516         EmitCharClass(assembler,
2517                       cc,
2518                       ascii,
2519                       backtrack,
2520                       cp_offset,
2521                       *checked_up_to < cp_offset,
2522                       preloaded);
2523         UpdateBoundsCheck(cp_offset, checked_up_to);
2524       }
2525     }
2526   }
2527 }
2528 
2529 
Length()2530 int TextNode::Length() {
2531   TextElement elm = elms_->last();
2532   ASSERT(elm.cp_offset >= 0);
2533   if (elm.type == TextElement::ATOM) {
2534     return elm.cp_offset + elm.data.u_atom->data().length();
2535   } else {
2536     return elm.cp_offset + 1;
2537   }
2538 }
2539 
2540 
SkipPass(int int_pass,bool ignore_case)2541 bool TextNode::SkipPass(int int_pass, bool ignore_case) {
2542   TextEmitPassType pass = static_cast<TextEmitPassType>(int_pass);
2543   if (ignore_case) {
2544     return pass == SIMPLE_CHARACTER_MATCH;
2545   } else {
2546     return pass == NON_LETTER_CHARACTER_MATCH || pass == CASE_CHARACTER_MATCH;
2547   }
2548 }
2549 
2550 
2551 // This generates the code to match a text node.  A text node can contain
2552 // straight character sequences (possibly to be matched in a case-independent
2553 // way) and character classes.  For efficiency we do not do this in a single
2554 // pass from left to right.  Instead we pass over the text node several times,
2555 // emitting code for some character positions every time.  See the comment on
2556 // TextEmitPass for details.
Emit(RegExpCompiler * compiler,Trace * trace)2557 void TextNode::Emit(RegExpCompiler* compiler, Trace* trace) {
2558   LimitResult limit_result = LimitVersions(compiler, trace);
2559   if (limit_result == DONE) return;
2560   ASSERT(limit_result == CONTINUE);
2561 
2562   if (trace->cp_offset() + Length() > RegExpMacroAssembler::kMaxCPOffset) {
2563     compiler->SetRegExpTooBig();
2564     return;
2565   }
2566 
2567   if (compiler->ascii()) {
2568     int dummy = 0;
2569     TextEmitPass(compiler, NON_ASCII_MATCH, false, trace, false, &dummy);
2570   }
2571 
2572   bool first_elt_done = false;
2573   int bound_checked_to = trace->cp_offset() - 1;
2574   bound_checked_to += trace->bound_checked_up_to();
2575 
2576   // If a character is preloaded into the current character register then
2577   // check that now.
2578   if (trace->characters_preloaded() == 1) {
2579     for (int pass = kFirstRealPass; pass <= kLastPass; pass++) {
2580       if (!SkipPass(pass, compiler->ignore_case())) {
2581         TextEmitPass(compiler,
2582                      static_cast<TextEmitPassType>(pass),
2583                      true,
2584                      trace,
2585                      false,
2586                      &bound_checked_to);
2587       }
2588     }
2589     first_elt_done = true;
2590   }
2591 
2592   for (int pass = kFirstRealPass; pass <= kLastPass; pass++) {
2593     if (!SkipPass(pass, compiler->ignore_case())) {
2594       TextEmitPass(compiler,
2595                    static_cast<TextEmitPassType>(pass),
2596                    false,
2597                    trace,
2598                    first_elt_done,
2599                    &bound_checked_to);
2600     }
2601   }
2602 
2603   Trace successor_trace(*trace);
2604   successor_trace.set_at_start(false);
2605   successor_trace.AdvanceCurrentPositionInTrace(Length(), compiler);
2606   RecursionCheck rc(compiler);
2607   on_success()->Emit(compiler, &successor_trace);
2608 }
2609 
2610 
InvalidateCurrentCharacter()2611 void Trace::InvalidateCurrentCharacter() {
2612   characters_preloaded_ = 0;
2613 }
2614 
2615 
AdvanceCurrentPositionInTrace(int by,RegExpCompiler * compiler)2616 void Trace::AdvanceCurrentPositionInTrace(int by, RegExpCompiler* compiler) {
2617   ASSERT(by > 0);
2618   // We don't have an instruction for shifting the current character register
2619   // down or for using a shifted value for anything so lets just forget that
2620   // we preloaded any characters into it.
2621   characters_preloaded_ = 0;
2622   // Adjust the offsets of the quick check performed information.  This
2623   // information is used to find out what we already determined about the
2624   // characters by means of mask and compare.
2625   quick_check_performed_.Advance(by, compiler->ascii());
2626   cp_offset_ += by;
2627   if (cp_offset_ > RegExpMacroAssembler::kMaxCPOffset) {
2628     compiler->SetRegExpTooBig();
2629     cp_offset_ = 0;
2630   }
2631   bound_checked_up_to_ = Max(0, bound_checked_up_to_ - by);
2632 }
2633 
2634 
MakeCaseIndependent(bool is_ascii)2635 void TextNode::MakeCaseIndependent(bool is_ascii) {
2636   int element_count = elms_->length();
2637   for (int i = 0; i < element_count; i++) {
2638     TextElement elm = elms_->at(i);
2639     if (elm.type == TextElement::CHAR_CLASS) {
2640       RegExpCharacterClass* cc = elm.data.u_char_class;
2641       // None of the standard character classes is different in the case
2642       // independent case and it slows us down if we don't know that.
2643       if (cc->is_standard()) continue;
2644       ZoneList<CharacterRange>* ranges = cc->ranges();
2645       int range_count = ranges->length();
2646       for (int j = 0; j < range_count; j++) {
2647         ranges->at(j).AddCaseEquivalents(ranges, is_ascii);
2648       }
2649     }
2650   }
2651 }
2652 
2653 
GreedyLoopTextLength()2654 int TextNode::GreedyLoopTextLength() {
2655   TextElement elm = elms_->at(elms_->length() - 1);
2656   if (elm.type == TextElement::CHAR_CLASS) {
2657     return elm.cp_offset + 1;
2658   } else {
2659     return elm.cp_offset + elm.data.u_atom->data().length();
2660   }
2661 }
2662 
2663 
2664 // Finds the fixed match length of a sequence of nodes that goes from
2665 // this alternative and back to this choice node.  If there are variable
2666 // length nodes or other complications in the way then return a sentinel
2667 // value indicating that a greedy loop cannot be constructed.
GreedyLoopTextLengthForAlternative(GuardedAlternative * alternative)2668 int ChoiceNode::GreedyLoopTextLengthForAlternative(
2669     GuardedAlternative* alternative) {
2670   int length = 0;
2671   RegExpNode* node = alternative->node();
2672   // Later we will generate code for all these text nodes using recursion
2673   // so we have to limit the max number.
2674   int recursion_depth = 0;
2675   while (node != this) {
2676     if (recursion_depth++ > RegExpCompiler::kMaxRecursion) {
2677       return kNodeIsTooComplexForGreedyLoops;
2678     }
2679     int node_length = node->GreedyLoopTextLength();
2680     if (node_length == kNodeIsTooComplexForGreedyLoops) {
2681       return kNodeIsTooComplexForGreedyLoops;
2682     }
2683     length += node_length;
2684     SeqRegExpNode* seq_node = static_cast<SeqRegExpNode*>(node);
2685     node = seq_node->on_success();
2686   }
2687   return length;
2688 }
2689 
2690 
AddLoopAlternative(GuardedAlternative alt)2691 void LoopChoiceNode::AddLoopAlternative(GuardedAlternative alt) {
2692   ASSERT_EQ(loop_node_, NULL);
2693   AddAlternative(alt);
2694   loop_node_ = alt.node();
2695 }
2696 
2697 
AddContinueAlternative(GuardedAlternative alt)2698 void LoopChoiceNode::AddContinueAlternative(GuardedAlternative alt) {
2699   ASSERT_EQ(continue_node_, NULL);
2700   AddAlternative(alt);
2701   continue_node_ = alt.node();
2702 }
2703 
2704 
Emit(RegExpCompiler * compiler,Trace * trace)2705 void LoopChoiceNode::Emit(RegExpCompiler* compiler, Trace* trace) {
2706   RegExpMacroAssembler* macro_assembler = compiler->macro_assembler();
2707   if (trace->stop_node() == this) {
2708     int text_length =
2709         GreedyLoopTextLengthForAlternative(&(alternatives_->at(0)));
2710     ASSERT(text_length != kNodeIsTooComplexForGreedyLoops);
2711     // Update the counter-based backtracking info on the stack.  This is an
2712     // optimization for greedy loops (see below).
2713     ASSERT(trace->cp_offset() == text_length);
2714     macro_assembler->AdvanceCurrentPosition(text_length);
2715     macro_assembler->GoTo(trace->loop_label());
2716     return;
2717   }
2718   ASSERT(trace->stop_node() == NULL);
2719   if (!trace->is_trivial()) {
2720     trace->Flush(compiler, this);
2721     return;
2722   }
2723   ChoiceNode::Emit(compiler, trace);
2724 }
2725 
2726 
CalculatePreloadCharacters(RegExpCompiler * compiler,bool not_at_start)2727 int ChoiceNode::CalculatePreloadCharacters(RegExpCompiler* compiler,
2728                                            bool not_at_start) {
2729   int preload_characters = EatsAtLeast(4, 0, not_at_start);
2730   if (compiler->macro_assembler()->CanReadUnaligned()) {
2731     bool ascii = compiler->ascii();
2732     if (ascii) {
2733       if (preload_characters > 4) preload_characters = 4;
2734       // We can't preload 3 characters because there is no machine instruction
2735       // to do that.  We can't just load 4 because we could be reading
2736       // beyond the end of the string, which could cause a memory fault.
2737       if (preload_characters == 3) preload_characters = 2;
2738     } else {
2739       if (preload_characters > 2) preload_characters = 2;
2740     }
2741   } else {
2742     if (preload_characters > 1) preload_characters = 1;
2743   }
2744   return preload_characters;
2745 }
2746 
2747 
2748 // This class is used when generating the alternatives in a choice node.  It
2749 // records the way the alternative is being code generated.
2750 class AlternativeGeneration: public Malloced {
2751  public:
AlternativeGeneration()2752   AlternativeGeneration()
2753       : possible_success(),
2754         expects_preload(false),
2755         after(),
2756         quick_check_details() { }
2757   Label possible_success;
2758   bool expects_preload;
2759   Label after;
2760   QuickCheckDetails quick_check_details;
2761 };
2762 
2763 
2764 // Creates a list of AlternativeGenerations.  If the list has a reasonable
2765 // size then it is on the stack, otherwise the excess is on the heap.
2766 class AlternativeGenerationList {
2767  public:
AlternativeGenerationList(int count)2768   explicit AlternativeGenerationList(int count)
2769       : alt_gens_(count) {
2770     for (int i = 0; i < count && i < kAFew; i++) {
2771       alt_gens_.Add(a_few_alt_gens_ + i);
2772     }
2773     for (int i = kAFew; i < count; i++) {
2774       alt_gens_.Add(new AlternativeGeneration());
2775     }
2776   }
~AlternativeGenerationList()2777   ~AlternativeGenerationList() {
2778     for (int i = kAFew; i < alt_gens_.length(); i++) {
2779       delete alt_gens_[i];
2780       alt_gens_[i] = NULL;
2781     }
2782   }
2783 
at(int i)2784   AlternativeGeneration* at(int i) {
2785     return alt_gens_[i];
2786   }
2787 
2788  private:
2789   static const int kAFew = 10;
2790   ZoneList<AlternativeGeneration*> alt_gens_;
2791   AlternativeGeneration a_few_alt_gens_[kAFew];
2792 };
2793 
2794 
2795 /* Code generation for choice nodes.
2796  *
2797  * We generate quick checks that do a mask and compare to eliminate a
2798  * choice.  If the quick check succeeds then it jumps to the continuation to
2799  * do slow checks and check subsequent nodes.  If it fails (the common case)
2800  * it falls through to the next choice.
2801  *
2802  * Here is the desired flow graph.  Nodes directly below each other imply
2803  * fallthrough.  Alternatives 1 and 2 have quick checks.  Alternative
2804  * 3 doesn't have a quick check so we have to call the slow check.
2805  * Nodes are marked Qn for quick checks and Sn for slow checks.  The entire
2806  * regexp continuation is generated directly after the Sn node, up to the
2807  * next GoTo if we decide to reuse some already generated code.  Some
2808  * nodes expect preload_characters to be preloaded into the current
2809  * character register.  R nodes do this preloading.  Vertices are marked
2810  * F for failures and S for success (possible success in the case of quick
2811  * nodes).  L, V, < and > are used as arrow heads.
2812  *
2813  * ----------> R
2814  *             |
2815  *             V
2816  *            Q1 -----> S1
2817  *             |   S   /
2818  *            F|      /
2819  *             |    F/
2820  *             |    /
2821  *             |   R
2822  *             |  /
2823  *             V L
2824  *            Q2 -----> S2
2825  *             |   S   /
2826  *            F|      /
2827  *             |    F/
2828  *             |    /
2829  *             |   R
2830  *             |  /
2831  *             V L
2832  *            S3
2833  *             |
2834  *            F|
2835  *             |
2836  *             R
2837  *             |
2838  * backtrack   V
2839  * <----------Q4
2840  *   \    F    |
2841  *    \        |S
2842  *     \   F   V
2843  *      \-----S4
2844  *
2845  * For greedy loops we reverse our expectation and expect to match rather
2846  * than fail. Therefore we want the loop code to look like this (U is the
2847  * unwind code that steps back in the greedy loop).  The following alternatives
2848  * look the same as above.
2849  *              _____
2850  *             /     \
2851  *             V     |
2852  * ----------> S1    |
2853  *            /|     |
2854  *           / |S    |
2855  *         F/  \_____/
2856  *         /
2857  *        |<-----------
2858  *        |            \
2859  *        V             \
2860  *        Q2 ---> S2     \
2861  *        |  S   /       |
2862  *       F|     /        |
2863  *        |   F/         |
2864  *        |   /          |
2865  *        |  R           |
2866  *        | /            |
2867  *   F    VL             |
2868  * <------U              |
2869  * back   |S             |
2870  *        \______________/
2871  */
2872 
2873 
Emit(RegExpCompiler * compiler,Trace * trace)2874 void ChoiceNode::Emit(RegExpCompiler* compiler, Trace* trace) {
2875   RegExpMacroAssembler* macro_assembler = compiler->macro_assembler();
2876   int choice_count = alternatives_->length();
2877 #ifdef DEBUG
2878   for (int i = 0; i < choice_count - 1; i++) {
2879     GuardedAlternative alternative = alternatives_->at(i);
2880     ZoneList<Guard*>* guards = alternative.guards();
2881     int guard_count = (guards == NULL) ? 0 : guards->length();
2882     for (int j = 0; j < guard_count; j++) {
2883       ASSERT(!trace->mentions_reg(guards->at(j)->reg()));
2884     }
2885   }
2886 #endif
2887 
2888   LimitResult limit_result = LimitVersions(compiler, trace);
2889   if (limit_result == DONE) return;
2890   ASSERT(limit_result == CONTINUE);
2891 
2892   int new_flush_budget = trace->flush_budget() / choice_count;
2893   if (trace->flush_budget() == 0 && trace->actions() != NULL) {
2894     trace->Flush(compiler, this);
2895     return;
2896   }
2897 
2898   RecursionCheck rc(compiler);
2899 
2900   Trace* current_trace = trace;
2901 
2902   int text_length = GreedyLoopTextLengthForAlternative(&(alternatives_->at(0)));
2903   bool greedy_loop = false;
2904   Label greedy_loop_label;
2905   Trace counter_backtrack_trace;
2906   counter_backtrack_trace.set_backtrack(&greedy_loop_label);
2907   if (not_at_start()) counter_backtrack_trace.set_at_start(false);
2908 
2909   if (choice_count > 1 && text_length != kNodeIsTooComplexForGreedyLoops) {
2910     // Here we have special handling for greedy loops containing only text nodes
2911     // and other simple nodes.  These are handled by pushing the current
2912     // position on the stack and then incrementing the current position each
2913     // time around the switch.  On backtrack we decrement the current position
2914     // and check it against the pushed value.  This avoids pushing backtrack
2915     // information for each iteration of the loop, which could take up a lot of
2916     // space.
2917     greedy_loop = true;
2918     ASSERT(trace->stop_node() == NULL);
2919     macro_assembler->PushCurrentPosition();
2920     current_trace = &counter_backtrack_trace;
2921     Label greedy_match_failed;
2922     Trace greedy_match_trace;
2923     if (not_at_start()) greedy_match_trace.set_at_start(false);
2924     greedy_match_trace.set_backtrack(&greedy_match_failed);
2925     Label loop_label;
2926     macro_assembler->Bind(&loop_label);
2927     greedy_match_trace.set_stop_node(this);
2928     greedy_match_trace.set_loop_label(&loop_label);
2929     alternatives_->at(0).node()->Emit(compiler, &greedy_match_trace);
2930     macro_assembler->Bind(&greedy_match_failed);
2931   }
2932 
2933   Label second_choice;  // For use in greedy matches.
2934   macro_assembler->Bind(&second_choice);
2935 
2936   int first_normal_choice = greedy_loop ? 1 : 0;
2937 
2938   int preload_characters =
2939       CalculatePreloadCharacters(compiler,
2940                                  current_trace->at_start() == Trace::FALSE);
2941   bool preload_is_current =
2942       (current_trace->characters_preloaded() == preload_characters);
2943   bool preload_has_checked_bounds = preload_is_current;
2944 
2945   AlternativeGenerationList alt_gens(choice_count);
2946 
2947   // For now we just call all choices one after the other.  The idea ultimately
2948   // is to use the Dispatch table to try only the relevant ones.
2949   for (int i = first_normal_choice; i < choice_count; i++) {
2950     GuardedAlternative alternative = alternatives_->at(i);
2951     AlternativeGeneration* alt_gen = alt_gens.at(i);
2952     alt_gen->quick_check_details.set_characters(preload_characters);
2953     ZoneList<Guard*>* guards = alternative.guards();
2954     int guard_count = (guards == NULL) ? 0 : guards->length();
2955     Trace new_trace(*current_trace);
2956     new_trace.set_characters_preloaded(preload_is_current ?
2957                                          preload_characters :
2958                                          0);
2959     if (preload_has_checked_bounds) {
2960       new_trace.set_bound_checked_up_to(preload_characters);
2961     }
2962     new_trace.quick_check_performed()->Clear();
2963     if (not_at_start_) new_trace.set_at_start(Trace::FALSE);
2964     alt_gen->expects_preload = preload_is_current;
2965     bool generate_full_check_inline = false;
2966     if (FLAG_regexp_optimization &&
2967         try_to_emit_quick_check_for_alternative(i) &&
2968         alternative.node()->EmitQuickCheck(compiler,
2969                                            &new_trace,
2970                                            preload_has_checked_bounds,
2971                                            &alt_gen->possible_success,
2972                                            &alt_gen->quick_check_details,
2973                                            i < choice_count - 1)) {
2974       // Quick check was generated for this choice.
2975       preload_is_current = true;
2976       preload_has_checked_bounds = true;
2977       // On the last choice in the ChoiceNode we generated the quick
2978       // check to fall through on possible success.  So now we need to
2979       // generate the full check inline.
2980       if (i == choice_count - 1) {
2981         macro_assembler->Bind(&alt_gen->possible_success);
2982         new_trace.set_quick_check_performed(&alt_gen->quick_check_details);
2983         new_trace.set_characters_preloaded(preload_characters);
2984         new_trace.set_bound_checked_up_to(preload_characters);
2985         generate_full_check_inline = true;
2986       }
2987     } else if (alt_gen->quick_check_details.cannot_match()) {
2988       if (i == choice_count - 1 && !greedy_loop) {
2989         macro_assembler->GoTo(trace->backtrack());
2990       }
2991       continue;
2992     } else {
2993       // No quick check was generated.  Put the full code here.
2994       // If this is not the first choice then there could be slow checks from
2995       // previous cases that go here when they fail.  There's no reason to
2996       // insist that they preload characters since the slow check we are about
2997       // to generate probably can't use it.
2998       if (i != first_normal_choice) {
2999         alt_gen->expects_preload = false;
3000         new_trace.InvalidateCurrentCharacter();
3001       }
3002       if (i < choice_count - 1) {
3003         new_trace.set_backtrack(&alt_gen->after);
3004       }
3005       generate_full_check_inline = true;
3006     }
3007     if (generate_full_check_inline) {
3008       if (new_trace.actions() != NULL) {
3009         new_trace.set_flush_budget(new_flush_budget);
3010       }
3011       for (int j = 0; j < guard_count; j++) {
3012         GenerateGuard(macro_assembler, guards->at(j), &new_trace);
3013       }
3014       alternative.node()->Emit(compiler, &new_trace);
3015       preload_is_current = false;
3016     }
3017     macro_assembler->Bind(&alt_gen->after);
3018   }
3019   if (greedy_loop) {
3020     macro_assembler->Bind(&greedy_loop_label);
3021     // If we have unwound to the bottom then backtrack.
3022     macro_assembler->CheckGreedyLoop(trace->backtrack());
3023     // Otherwise try the second priority at an earlier position.
3024     macro_assembler->AdvanceCurrentPosition(-text_length);
3025     macro_assembler->GoTo(&second_choice);
3026   }
3027 
3028   // At this point we need to generate slow checks for the alternatives where
3029   // the quick check was inlined.  We can recognize these because the associated
3030   // label was bound.
3031   for (int i = first_normal_choice; i < choice_count - 1; i++) {
3032     AlternativeGeneration* alt_gen = alt_gens.at(i);
3033     Trace new_trace(*current_trace);
3034     // If there are actions to be flushed we have to limit how many times
3035     // they are flushed.  Take the budget of the parent trace and distribute
3036     // it fairly amongst the children.
3037     if (new_trace.actions() != NULL) {
3038       new_trace.set_flush_budget(new_flush_budget);
3039     }
3040     EmitOutOfLineContinuation(compiler,
3041                               &new_trace,
3042                               alternatives_->at(i),
3043                               alt_gen,
3044                               preload_characters,
3045                               alt_gens.at(i + 1)->expects_preload);
3046   }
3047 }
3048 
3049 
EmitOutOfLineContinuation(RegExpCompiler * compiler,Trace * trace,GuardedAlternative alternative,AlternativeGeneration * alt_gen,int preload_characters,bool next_expects_preload)3050 void ChoiceNode::EmitOutOfLineContinuation(RegExpCompiler* compiler,
3051                                            Trace* trace,
3052                                            GuardedAlternative alternative,
3053                                            AlternativeGeneration* alt_gen,
3054                                            int preload_characters,
3055                                            bool next_expects_preload) {
3056   if (!alt_gen->possible_success.is_linked()) return;
3057 
3058   RegExpMacroAssembler* macro_assembler = compiler->macro_assembler();
3059   macro_assembler->Bind(&alt_gen->possible_success);
3060   Trace out_of_line_trace(*trace);
3061   out_of_line_trace.set_characters_preloaded(preload_characters);
3062   out_of_line_trace.set_quick_check_performed(&alt_gen->quick_check_details);
3063   if (not_at_start_) out_of_line_trace.set_at_start(Trace::FALSE);
3064   ZoneList<Guard*>* guards = alternative.guards();
3065   int guard_count = (guards == NULL) ? 0 : guards->length();
3066   if (next_expects_preload) {
3067     Label reload_current_char;
3068     out_of_line_trace.set_backtrack(&reload_current_char);
3069     for (int j = 0; j < guard_count; j++) {
3070       GenerateGuard(macro_assembler, guards->at(j), &out_of_line_trace);
3071     }
3072     alternative.node()->Emit(compiler, &out_of_line_trace);
3073     macro_assembler->Bind(&reload_current_char);
3074     // Reload the current character, since the next quick check expects that.
3075     // We don't need to check bounds here because we only get into this
3076     // code through a quick check which already did the checked load.
3077     macro_assembler->LoadCurrentCharacter(trace->cp_offset(),
3078                                           NULL,
3079                                           false,
3080                                           preload_characters);
3081     macro_assembler->GoTo(&(alt_gen->after));
3082   } else {
3083     out_of_line_trace.set_backtrack(&(alt_gen->after));
3084     for (int j = 0; j < guard_count; j++) {
3085       GenerateGuard(macro_assembler, guards->at(j), &out_of_line_trace);
3086     }
3087     alternative.node()->Emit(compiler, &out_of_line_trace);
3088   }
3089 }
3090 
3091 
Emit(RegExpCompiler * compiler,Trace * trace)3092 void ActionNode::Emit(RegExpCompiler* compiler, Trace* trace) {
3093   RegExpMacroAssembler* assembler = compiler->macro_assembler();
3094   LimitResult limit_result = LimitVersions(compiler, trace);
3095   if (limit_result == DONE) return;
3096   ASSERT(limit_result == CONTINUE);
3097 
3098   RecursionCheck rc(compiler);
3099 
3100   switch (type_) {
3101     case STORE_POSITION: {
3102       Trace::DeferredCapture
3103           new_capture(data_.u_position_register.reg,
3104                       data_.u_position_register.is_capture,
3105                       trace);
3106       Trace new_trace = *trace;
3107       new_trace.add_action(&new_capture);
3108       on_success()->Emit(compiler, &new_trace);
3109       break;
3110     }
3111     case INCREMENT_REGISTER: {
3112       Trace::DeferredIncrementRegister
3113           new_increment(data_.u_increment_register.reg);
3114       Trace new_trace = *trace;
3115       new_trace.add_action(&new_increment);
3116       on_success()->Emit(compiler, &new_trace);
3117       break;
3118     }
3119     case SET_REGISTER: {
3120       Trace::DeferredSetRegister
3121           new_set(data_.u_store_register.reg, data_.u_store_register.value);
3122       Trace new_trace = *trace;
3123       new_trace.add_action(&new_set);
3124       on_success()->Emit(compiler, &new_trace);
3125       break;
3126     }
3127     case CLEAR_CAPTURES: {
3128       Trace::DeferredClearCaptures
3129         new_capture(Interval(data_.u_clear_captures.range_from,
3130                              data_.u_clear_captures.range_to));
3131       Trace new_trace = *trace;
3132       new_trace.add_action(&new_capture);
3133       on_success()->Emit(compiler, &new_trace);
3134       break;
3135     }
3136     case BEGIN_SUBMATCH:
3137       if (!trace->is_trivial()) {
3138         trace->Flush(compiler, this);
3139       } else {
3140         assembler->WriteCurrentPositionToRegister(
3141             data_.u_submatch.current_position_register, 0);
3142         assembler->WriteStackPointerToRegister(
3143             data_.u_submatch.stack_pointer_register);
3144         on_success()->Emit(compiler, trace);
3145       }
3146       break;
3147     case EMPTY_MATCH_CHECK: {
3148       int start_pos_reg = data_.u_empty_match_check.start_register;
3149       int stored_pos = 0;
3150       int rep_reg = data_.u_empty_match_check.repetition_register;
3151       bool has_minimum = (rep_reg != RegExpCompiler::kNoRegister);
3152       bool know_dist = trace->GetStoredPosition(start_pos_reg, &stored_pos);
3153       if (know_dist && !has_minimum && stored_pos == trace->cp_offset()) {
3154         // If we know we haven't advanced and there is no minimum we
3155         // can just backtrack immediately.
3156         assembler->GoTo(trace->backtrack());
3157       } else if (know_dist && stored_pos < trace->cp_offset()) {
3158         // If we know we've advanced we can generate the continuation
3159         // immediately.
3160         on_success()->Emit(compiler, trace);
3161       } else if (!trace->is_trivial()) {
3162         trace->Flush(compiler, this);
3163       } else {
3164         Label skip_empty_check;
3165         // If we have a minimum number of repetitions we check the current
3166         // number first and skip the empty check if it's not enough.
3167         if (has_minimum) {
3168           int limit = data_.u_empty_match_check.repetition_limit;
3169           assembler->IfRegisterLT(rep_reg, limit, &skip_empty_check);
3170         }
3171         // If the match is empty we bail out, otherwise we fall through
3172         // to the on-success continuation.
3173         assembler->IfRegisterEqPos(data_.u_empty_match_check.start_register,
3174                                    trace->backtrack());
3175         assembler->Bind(&skip_empty_check);
3176         on_success()->Emit(compiler, trace);
3177       }
3178       break;
3179     }
3180     case POSITIVE_SUBMATCH_SUCCESS: {
3181       if (!trace->is_trivial()) {
3182         trace->Flush(compiler, this);
3183         return;
3184       }
3185       assembler->ReadCurrentPositionFromRegister(
3186           data_.u_submatch.current_position_register);
3187       assembler->ReadStackPointerFromRegister(
3188           data_.u_submatch.stack_pointer_register);
3189       int clear_register_count = data_.u_submatch.clear_register_count;
3190       if (clear_register_count == 0) {
3191         on_success()->Emit(compiler, trace);
3192         return;
3193       }
3194       int clear_registers_from = data_.u_submatch.clear_register_from;
3195       Label clear_registers_backtrack;
3196       Trace new_trace = *trace;
3197       new_trace.set_backtrack(&clear_registers_backtrack);
3198       on_success()->Emit(compiler, &new_trace);
3199 
3200       assembler->Bind(&clear_registers_backtrack);
3201       int clear_registers_to = clear_registers_from + clear_register_count - 1;
3202       assembler->ClearRegisters(clear_registers_from, clear_registers_to);
3203 
3204       ASSERT(trace->backtrack() == NULL);
3205       assembler->Backtrack();
3206       return;
3207     }
3208     default:
3209       UNREACHABLE();
3210   }
3211 }
3212 
3213 
Emit(RegExpCompiler * compiler,Trace * trace)3214 void BackReferenceNode::Emit(RegExpCompiler* compiler, Trace* trace) {
3215   RegExpMacroAssembler* assembler = compiler->macro_assembler();
3216   if (!trace->is_trivial()) {
3217     trace->Flush(compiler, this);
3218     return;
3219   }
3220 
3221   LimitResult limit_result = LimitVersions(compiler, trace);
3222   if (limit_result == DONE) return;
3223   ASSERT(limit_result == CONTINUE);
3224 
3225   RecursionCheck rc(compiler);
3226 
3227   ASSERT_EQ(start_reg_ + 1, end_reg_);
3228   if (compiler->ignore_case()) {
3229     assembler->CheckNotBackReferenceIgnoreCase(start_reg_,
3230                                                trace->backtrack());
3231   } else {
3232     assembler->CheckNotBackReference(start_reg_, trace->backtrack());
3233   }
3234   on_success()->Emit(compiler, trace);
3235 }
3236 
3237 
3238 // -------------------------------------------------------------------
3239 // Dot/dotty output
3240 
3241 
3242 #ifdef DEBUG
3243 
3244 
3245 class DotPrinter: public NodeVisitor {
3246  public:
DotPrinter(bool ignore_case)3247   explicit DotPrinter(bool ignore_case)
3248       : ignore_case_(ignore_case),
3249         stream_(&alloc_) { }
3250   void PrintNode(const char* label, RegExpNode* node);
3251   void Visit(RegExpNode* node);
3252   void PrintAttributes(RegExpNode* from);
stream()3253   StringStream* stream() { return &stream_; }
3254   void PrintOnFailure(RegExpNode* from, RegExpNode* to);
3255 #define DECLARE_VISIT(Type)                                          \
3256   virtual void Visit##Type(Type##Node* that);
3257 FOR_EACH_NODE_TYPE(DECLARE_VISIT)
3258 #undef DECLARE_VISIT
3259  private:
3260   bool ignore_case_;
3261   HeapStringAllocator alloc_;
3262   StringStream stream_;
3263 };
3264 
3265 
PrintNode(const char * label,RegExpNode * node)3266 void DotPrinter::PrintNode(const char* label, RegExpNode* node) {
3267   stream()->Add("digraph G {\n  graph [label=\"");
3268   for (int i = 0; label[i]; i++) {
3269     switch (label[i]) {
3270       case '\\':
3271         stream()->Add("\\\\");
3272         break;
3273       case '"':
3274         stream()->Add("\"");
3275         break;
3276       default:
3277         stream()->Put(label[i]);
3278         break;
3279     }
3280   }
3281   stream()->Add("\"];\n");
3282   Visit(node);
3283   stream()->Add("}\n");
3284   printf("%s", *(stream()->ToCString()));
3285 }
3286 
3287 
Visit(RegExpNode * node)3288 void DotPrinter::Visit(RegExpNode* node) {
3289   if (node->info()->visited) return;
3290   node->info()->visited = true;
3291   node->Accept(this);
3292 }
3293 
3294 
PrintOnFailure(RegExpNode * from,RegExpNode * on_failure)3295 void DotPrinter::PrintOnFailure(RegExpNode* from, RegExpNode* on_failure) {
3296   stream()->Add("  n%p -> n%p [style=dotted];\n", from, on_failure);
3297   Visit(on_failure);
3298 }
3299 
3300 
3301 class TableEntryBodyPrinter {
3302  public:
TableEntryBodyPrinter(StringStream * stream,ChoiceNode * choice)3303   TableEntryBodyPrinter(StringStream* stream, ChoiceNode* choice)
3304       : stream_(stream), choice_(choice) { }
Call(uc16 from,DispatchTable::Entry entry)3305   void Call(uc16 from, DispatchTable::Entry entry) {
3306     OutSet* out_set = entry.out_set();
3307     for (unsigned i = 0; i < OutSet::kFirstLimit; i++) {
3308       if (out_set->Get(i)) {
3309         stream()->Add("    n%p:s%io%i -> n%p;\n",
3310                       choice(),
3311                       from,
3312                       i,
3313                       choice()->alternatives()->at(i).node());
3314       }
3315     }
3316   }
3317  private:
stream()3318   StringStream* stream() { return stream_; }
choice()3319   ChoiceNode* choice() { return choice_; }
3320   StringStream* stream_;
3321   ChoiceNode* choice_;
3322 };
3323 
3324 
3325 class TableEntryHeaderPrinter {
3326  public:
TableEntryHeaderPrinter(StringStream * stream)3327   explicit TableEntryHeaderPrinter(StringStream* stream)
3328       : first_(true), stream_(stream) { }
Call(uc16 from,DispatchTable::Entry entry)3329   void Call(uc16 from, DispatchTable::Entry entry) {
3330     if (first_) {
3331       first_ = false;
3332     } else {
3333       stream()->Add("|");
3334     }
3335     stream()->Add("{\\%k-\\%k|{", from, entry.to());
3336     OutSet* out_set = entry.out_set();
3337     int priority = 0;
3338     for (unsigned i = 0; i < OutSet::kFirstLimit; i++) {
3339       if (out_set->Get(i)) {
3340         if (priority > 0) stream()->Add("|");
3341         stream()->Add("<s%io%i> %i", from, i, priority);
3342         priority++;
3343       }
3344     }
3345     stream()->Add("}}");
3346   }
3347 
3348  private:
3349   bool first_;
stream()3350   StringStream* stream() { return stream_; }
3351   StringStream* stream_;
3352 };
3353 
3354 
3355 class AttributePrinter {
3356  public:
AttributePrinter(DotPrinter * out)3357   explicit AttributePrinter(DotPrinter* out)
3358       : out_(out), first_(true) { }
PrintSeparator()3359   void PrintSeparator() {
3360     if (first_) {
3361       first_ = false;
3362     } else {
3363       out_->stream()->Add("|");
3364     }
3365   }
PrintBit(const char * name,bool value)3366   void PrintBit(const char* name, bool value) {
3367     if (!value) return;
3368     PrintSeparator();
3369     out_->stream()->Add("{%s}", name);
3370   }
PrintPositive(const char * name,int value)3371   void PrintPositive(const char* name, int value) {
3372     if (value < 0) return;
3373     PrintSeparator();
3374     out_->stream()->Add("{%s|%x}", name, value);
3375   }
3376  private:
3377   DotPrinter* out_;
3378   bool first_;
3379 };
3380 
3381 
PrintAttributes(RegExpNode * that)3382 void DotPrinter::PrintAttributes(RegExpNode* that) {
3383   stream()->Add("  a%p [shape=Mrecord, color=grey, fontcolor=grey, "
3384                 "margin=0.1, fontsize=10, label=\"{",
3385                 that);
3386   AttributePrinter printer(this);
3387   NodeInfo* info = that->info();
3388   printer.PrintBit("NI", info->follows_newline_interest);
3389   printer.PrintBit("WI", info->follows_word_interest);
3390   printer.PrintBit("SI", info->follows_start_interest);
3391   Label* label = that->label();
3392   if (label->is_bound())
3393     printer.PrintPositive("@", label->pos());
3394   stream()->Add("}\"];\n");
3395   stream()->Add("  a%p -> n%p [style=dashed, color=grey, "
3396                 "arrowhead=none];\n", that, that);
3397 }
3398 
3399 
3400 static const bool kPrintDispatchTable = false;
VisitChoice(ChoiceNode * that)3401 void DotPrinter::VisitChoice(ChoiceNode* that) {
3402   if (kPrintDispatchTable) {
3403     stream()->Add("  n%p [shape=Mrecord, label=\"", that);
3404     TableEntryHeaderPrinter header_printer(stream());
3405     that->GetTable(ignore_case_)->ForEach(&header_printer);
3406     stream()->Add("\"]\n", that);
3407     PrintAttributes(that);
3408     TableEntryBodyPrinter body_printer(stream(), that);
3409     that->GetTable(ignore_case_)->ForEach(&body_printer);
3410   } else {
3411     stream()->Add("  n%p [shape=Mrecord, label=\"?\"];\n", that);
3412     for (int i = 0; i < that->alternatives()->length(); i++) {
3413       GuardedAlternative alt = that->alternatives()->at(i);
3414       stream()->Add("  n%p -> n%p;\n", that, alt.node());
3415     }
3416   }
3417   for (int i = 0; i < that->alternatives()->length(); i++) {
3418     GuardedAlternative alt = that->alternatives()->at(i);
3419     alt.node()->Accept(this);
3420   }
3421 }
3422 
3423 
VisitText(TextNode * that)3424 void DotPrinter::VisitText(TextNode* that) {
3425   stream()->Add("  n%p [label=\"", that);
3426   for (int i = 0; i < that->elements()->length(); i++) {
3427     if (i > 0) stream()->Add(" ");
3428     TextElement elm = that->elements()->at(i);
3429     switch (elm.type) {
3430       case TextElement::ATOM: {
3431         stream()->Add("'%w'", elm.data.u_atom->data());
3432         break;
3433       }
3434       case TextElement::CHAR_CLASS: {
3435         RegExpCharacterClass* node = elm.data.u_char_class;
3436         stream()->Add("[");
3437         if (node->is_negated())
3438           stream()->Add("^");
3439         for (int j = 0; j < node->ranges()->length(); j++) {
3440           CharacterRange range = node->ranges()->at(j);
3441           stream()->Add("%k-%k", range.from(), range.to());
3442         }
3443         stream()->Add("]");
3444         break;
3445       }
3446       default:
3447         UNREACHABLE();
3448     }
3449   }
3450   stream()->Add("\", shape=box, peripheries=2];\n");
3451   PrintAttributes(that);
3452   stream()->Add("  n%p -> n%p;\n", that, that->on_success());
3453   Visit(that->on_success());
3454 }
3455 
3456 
VisitBackReference(BackReferenceNode * that)3457 void DotPrinter::VisitBackReference(BackReferenceNode* that) {
3458   stream()->Add("  n%p [label=\"$%i..$%i\", shape=doubleoctagon];\n",
3459                 that,
3460                 that->start_register(),
3461                 that->end_register());
3462   PrintAttributes(that);
3463   stream()->Add("  n%p -> n%p;\n", that, that->on_success());
3464   Visit(that->on_success());
3465 }
3466 
3467 
VisitEnd(EndNode * that)3468 void DotPrinter::VisitEnd(EndNode* that) {
3469   stream()->Add("  n%p [style=bold, shape=point];\n", that);
3470   PrintAttributes(that);
3471 }
3472 
3473 
VisitAssertion(AssertionNode * that)3474 void DotPrinter::VisitAssertion(AssertionNode* that) {
3475   stream()->Add("  n%p [", that);
3476   switch (that->type()) {
3477     case AssertionNode::AT_END:
3478       stream()->Add("label=\"$\", shape=septagon");
3479       break;
3480     case AssertionNode::AT_START:
3481       stream()->Add("label=\"^\", shape=septagon");
3482       break;
3483     case AssertionNode::AT_BOUNDARY:
3484       stream()->Add("label=\"\\b\", shape=septagon");
3485       break;
3486     case AssertionNode::AT_NON_BOUNDARY:
3487       stream()->Add("label=\"\\B\", shape=septagon");
3488       break;
3489     case AssertionNode::AFTER_NEWLINE:
3490       stream()->Add("label=\"(?<=\\n)\", shape=septagon");
3491       break;
3492     case AssertionNode::AFTER_WORD_CHARACTER:
3493       stream()->Add("label=\"(?<=\\w)\", shape=septagon");
3494       break;
3495     case AssertionNode::AFTER_NONWORD_CHARACTER:
3496       stream()->Add("label=\"(?<=\\W)\", shape=septagon");
3497       break;
3498   }
3499   stream()->Add("];\n");
3500   PrintAttributes(that);
3501   RegExpNode* successor = that->on_success();
3502   stream()->Add("  n%p -> n%p;\n", that, successor);
3503   Visit(successor);
3504 }
3505 
3506 
VisitAction(ActionNode * that)3507 void DotPrinter::VisitAction(ActionNode* that) {
3508   stream()->Add("  n%p [", that);
3509   switch (that->type_) {
3510     case ActionNode::SET_REGISTER:
3511       stream()->Add("label=\"$%i:=%i\", shape=octagon",
3512                     that->data_.u_store_register.reg,
3513                     that->data_.u_store_register.value);
3514       break;
3515     case ActionNode::INCREMENT_REGISTER:
3516       stream()->Add("label=\"$%i++\", shape=octagon",
3517                     that->data_.u_increment_register.reg);
3518       break;
3519     case ActionNode::STORE_POSITION:
3520       stream()->Add("label=\"$%i:=$pos\", shape=octagon",
3521                     that->data_.u_position_register.reg);
3522       break;
3523     case ActionNode::BEGIN_SUBMATCH:
3524       stream()->Add("label=\"$%i:=$pos,begin\", shape=septagon",
3525                     that->data_.u_submatch.current_position_register);
3526       break;
3527     case ActionNode::POSITIVE_SUBMATCH_SUCCESS:
3528       stream()->Add("label=\"escape\", shape=septagon");
3529       break;
3530     case ActionNode::EMPTY_MATCH_CHECK:
3531       stream()->Add("label=\"$%i=$pos?,$%i<%i?\", shape=septagon",
3532                     that->data_.u_empty_match_check.start_register,
3533                     that->data_.u_empty_match_check.repetition_register,
3534                     that->data_.u_empty_match_check.repetition_limit);
3535       break;
3536     case ActionNode::CLEAR_CAPTURES: {
3537       stream()->Add("label=\"clear $%i to $%i\", shape=septagon",
3538                     that->data_.u_clear_captures.range_from,
3539                     that->data_.u_clear_captures.range_to);
3540       break;
3541     }
3542   }
3543   stream()->Add("];\n");
3544   PrintAttributes(that);
3545   RegExpNode* successor = that->on_success();
3546   stream()->Add("  n%p -> n%p;\n", that, successor);
3547   Visit(successor);
3548 }
3549 
3550 
3551 class DispatchTableDumper {
3552  public:
DispatchTableDumper(StringStream * stream)3553   explicit DispatchTableDumper(StringStream* stream) : stream_(stream) { }
3554   void Call(uc16 key, DispatchTable::Entry entry);
stream()3555   StringStream* stream() { return stream_; }
3556  private:
3557   StringStream* stream_;
3558 };
3559 
3560 
Call(uc16 key,DispatchTable::Entry entry)3561 void DispatchTableDumper::Call(uc16 key, DispatchTable::Entry entry) {
3562   stream()->Add("[%k-%k]: {", key, entry.to());
3563   OutSet* set = entry.out_set();
3564   bool first = true;
3565   for (unsigned i = 0; i < OutSet::kFirstLimit; i++) {
3566     if (set->Get(i)) {
3567       if (first) {
3568         first = false;
3569       } else {
3570         stream()->Add(", ");
3571       }
3572       stream()->Add("%i", i);
3573     }
3574   }
3575   stream()->Add("}\n");
3576 }
3577 
3578 
Dump()3579 void DispatchTable::Dump() {
3580   HeapStringAllocator alloc;
3581   StringStream stream(&alloc);
3582   DispatchTableDumper dumper(&stream);
3583   tree()->ForEach(&dumper);
3584   OS::PrintError("%s", *stream.ToCString());
3585 }
3586 
3587 
DotPrint(const char * label,RegExpNode * node,bool ignore_case)3588 void RegExpEngine::DotPrint(const char* label,
3589                             RegExpNode* node,
3590                             bool ignore_case) {
3591   DotPrinter printer(ignore_case);
3592   printer.PrintNode(label, node);
3593 }
3594 
3595 
3596 #endif  // DEBUG
3597 
3598 
3599 // -------------------------------------------------------------------
3600 // Tree to graph conversion
3601 
3602 static const uc16 kSpaceRanges[] = { 0x0009, 0x000D, 0x0020, 0x0020, 0x00A0,
3603     0x00A0, 0x1680, 0x1680, 0x180E, 0x180E, 0x2000, 0x200A, 0x2028, 0x2029,
3604     0x202F, 0x202F, 0x205F, 0x205F, 0x3000, 0x3000, 0xFEFF, 0xFEFF };
3605 static const int kSpaceRangeCount = ARRAY_SIZE(kSpaceRanges);
3606 
3607 static const uc16 kWordRanges[] = { '0', '9', 'A', 'Z', '_', '_', 'a', 'z' };
3608 static const int kWordRangeCount = ARRAY_SIZE(kWordRanges);
3609 
3610 static const uc16 kDigitRanges[] = { '0', '9' };
3611 static const int kDigitRangeCount = ARRAY_SIZE(kDigitRanges);
3612 
3613 static const uc16 kLineTerminatorRanges[] = { 0x000A, 0x000A, 0x000D, 0x000D,
3614     0x2028, 0x2029 };
3615 static const int kLineTerminatorRangeCount = ARRAY_SIZE(kLineTerminatorRanges);
3616 
ToNode(RegExpCompiler * compiler,RegExpNode * on_success)3617 RegExpNode* RegExpAtom::ToNode(RegExpCompiler* compiler,
3618                                RegExpNode* on_success) {
3619   ZoneList<TextElement>* elms = new ZoneList<TextElement>(1);
3620   elms->Add(TextElement::Atom(this));
3621   return new TextNode(elms, on_success);
3622 }
3623 
3624 
ToNode(RegExpCompiler * compiler,RegExpNode * on_success)3625 RegExpNode* RegExpText::ToNode(RegExpCompiler* compiler,
3626                                RegExpNode* on_success) {
3627   return new TextNode(elements(), on_success);
3628 }
3629 
CompareInverseRanges(ZoneList<CharacterRange> * ranges,const uc16 * special_class,int length)3630 static bool CompareInverseRanges(ZoneList<CharacterRange>* ranges,
3631                                  const uc16* special_class,
3632                                  int length) {
3633   ASSERT(ranges->length() != 0);
3634   ASSERT(length != 0);
3635   ASSERT(special_class[0] != 0);
3636   if (ranges->length() != (length >> 1) + 1) {
3637     return false;
3638   }
3639   CharacterRange range = ranges->at(0);
3640   if (range.from() != 0) {
3641     return false;
3642   }
3643   for (int i = 0; i < length; i += 2) {
3644     if (special_class[i] != (range.to() + 1)) {
3645       return false;
3646     }
3647     range = ranges->at((i >> 1) + 1);
3648     if (special_class[i+1] != range.from() - 1) {
3649       return false;
3650     }
3651   }
3652   if (range.to() != 0xffff) {
3653     return false;
3654   }
3655   return true;
3656 }
3657 
3658 
CompareRanges(ZoneList<CharacterRange> * ranges,const uc16 * special_class,int length)3659 static bool CompareRanges(ZoneList<CharacterRange>* ranges,
3660                           const uc16* special_class,
3661                           int length) {
3662   if (ranges->length() * 2 != length) {
3663     return false;
3664   }
3665   for (int i = 0; i < length; i += 2) {
3666     CharacterRange range = ranges->at(i >> 1);
3667     if (range.from() != special_class[i] || range.to() != special_class[i+1]) {
3668       return false;
3669     }
3670   }
3671   return true;
3672 }
3673 
3674 
is_standard()3675 bool RegExpCharacterClass::is_standard() {
3676   // TODO(lrn): Remove need for this function, by not throwing away information
3677   // along the way.
3678   if (is_negated_) {
3679     return false;
3680   }
3681   if (set_.is_standard()) {
3682     return true;
3683   }
3684   if (CompareRanges(set_.ranges(), kSpaceRanges, kSpaceRangeCount)) {
3685     set_.set_standard_set_type('s');
3686     return true;
3687   }
3688   if (CompareInverseRanges(set_.ranges(), kSpaceRanges, kSpaceRangeCount)) {
3689     set_.set_standard_set_type('S');
3690     return true;
3691   }
3692   if (CompareInverseRanges(set_.ranges(),
3693                            kLineTerminatorRanges,
3694                            kLineTerminatorRangeCount)) {
3695     set_.set_standard_set_type('.');
3696     return true;
3697   }
3698   if (CompareRanges(set_.ranges(),
3699                     kLineTerminatorRanges,
3700                     kLineTerminatorRangeCount)) {
3701     set_.set_standard_set_type('n');
3702     return true;
3703   }
3704   if (CompareRanges(set_.ranges(), kWordRanges, kWordRangeCount)) {
3705     set_.set_standard_set_type('w');
3706     return true;
3707   }
3708   if (CompareInverseRanges(set_.ranges(), kWordRanges, kWordRangeCount)) {
3709     set_.set_standard_set_type('W');
3710     return true;
3711   }
3712   return false;
3713 }
3714 
3715 
ToNode(RegExpCompiler * compiler,RegExpNode * on_success)3716 RegExpNode* RegExpCharacterClass::ToNode(RegExpCompiler* compiler,
3717                                          RegExpNode* on_success) {
3718   return new TextNode(this, on_success);
3719 }
3720 
3721 
ToNode(RegExpCompiler * compiler,RegExpNode * on_success)3722 RegExpNode* RegExpDisjunction::ToNode(RegExpCompiler* compiler,
3723                                       RegExpNode* on_success) {
3724   ZoneList<RegExpTree*>* alternatives = this->alternatives();
3725   int length = alternatives->length();
3726   ChoiceNode* result = new ChoiceNode(length);
3727   for (int i = 0; i < length; i++) {
3728     GuardedAlternative alternative(alternatives->at(i)->ToNode(compiler,
3729                                                                on_success));
3730     result->AddAlternative(alternative);
3731   }
3732   return result;
3733 }
3734 
3735 
ToNode(RegExpCompiler * compiler,RegExpNode * on_success)3736 RegExpNode* RegExpQuantifier::ToNode(RegExpCompiler* compiler,
3737                                      RegExpNode* on_success) {
3738   return ToNode(min(),
3739                 max(),
3740                 is_greedy(),
3741                 body(),
3742                 compiler,
3743                 on_success);
3744 }
3745 
3746 
3747 // Scoped object to keep track of how much we unroll quantifier loops in the
3748 // regexp graph generator.
3749 class RegExpExpansionLimiter {
3750  public:
3751   static const int kMaxExpansionFactor = 6;
RegExpExpansionLimiter(RegExpCompiler * compiler,int factor)3752   RegExpExpansionLimiter(RegExpCompiler* compiler, int factor)
3753       : compiler_(compiler),
3754         saved_expansion_factor_(compiler->current_expansion_factor()),
3755         ok_to_expand_(saved_expansion_factor_ <= kMaxExpansionFactor) {
3756     ASSERT(factor > 0);
3757     if (ok_to_expand_) {
3758       if (factor > kMaxExpansionFactor) {
3759         // Avoid integer overflow of the current expansion factor.
3760         ok_to_expand_ = false;
3761         compiler->set_current_expansion_factor(kMaxExpansionFactor + 1);
3762       } else {
3763         int new_factor = saved_expansion_factor_ * factor;
3764         ok_to_expand_ = (new_factor <= kMaxExpansionFactor);
3765         compiler->set_current_expansion_factor(new_factor);
3766       }
3767     }
3768   }
3769 
~RegExpExpansionLimiter()3770   ~RegExpExpansionLimiter() {
3771     compiler_->set_current_expansion_factor(saved_expansion_factor_);
3772   }
3773 
ok_to_expand()3774   bool ok_to_expand() { return ok_to_expand_; }
3775 
3776  private:
3777   RegExpCompiler* compiler_;
3778   int saved_expansion_factor_;
3779   bool ok_to_expand_;
3780 
3781   DISALLOW_IMPLICIT_CONSTRUCTORS(RegExpExpansionLimiter);
3782 };
3783 
3784 
ToNode(int min,int max,bool is_greedy,RegExpTree * body,RegExpCompiler * compiler,RegExpNode * on_success,bool not_at_start)3785 RegExpNode* RegExpQuantifier::ToNode(int min,
3786                                      int max,
3787                                      bool is_greedy,
3788                                      RegExpTree* body,
3789                                      RegExpCompiler* compiler,
3790                                      RegExpNode* on_success,
3791                                      bool not_at_start) {
3792   // x{f, t} becomes this:
3793   //
3794   //             (r++)<-.
3795   //               |     `
3796   //               |     (x)
3797   //               v     ^
3798   //      (r=0)-->(?)---/ [if r < t]
3799   //               |
3800   //   [if r >= f] \----> ...
3801   //
3802 
3803   // 15.10.2.5 RepeatMatcher algorithm.
3804   // The parser has already eliminated the case where max is 0.  In the case
3805   // where max_match is zero the parser has removed the quantifier if min was
3806   // > 0 and removed the atom if min was 0.  See AddQuantifierToAtom.
3807 
3808   // If we know that we cannot match zero length then things are a little
3809   // simpler since we don't need to make the special zero length match check
3810   // from step 2.1.  If the min and max are small we can unroll a little in
3811   // this case.
3812   static const int kMaxUnrolledMinMatches = 3;  // Unroll (foo)+ and (foo){3,}
3813   static const int kMaxUnrolledMaxMatches = 3;  // Unroll (foo)? and (foo){x,3}
3814   if (max == 0) return on_success;  // This can happen due to recursion.
3815   bool body_can_be_empty = (body->min_match() == 0);
3816   int body_start_reg = RegExpCompiler::kNoRegister;
3817   Interval capture_registers = body->CaptureRegisters();
3818   bool needs_capture_clearing = !capture_registers.is_empty();
3819   if (body_can_be_empty) {
3820     body_start_reg = compiler->AllocateRegister();
3821   } else if (FLAG_regexp_optimization && !needs_capture_clearing) {
3822     // Only unroll if there are no captures and the body can't be
3823     // empty.
3824     {
3825       RegExpExpansionLimiter limiter(
3826           compiler, min + ((max != min) ? 1 : 0));
3827       if (min > 0 && min <= kMaxUnrolledMinMatches && limiter.ok_to_expand()) {
3828         int new_max = (max == kInfinity) ? max : max - min;
3829         // Recurse once to get the loop or optional matches after the fixed
3830         // ones.
3831         RegExpNode* answer = ToNode(
3832             0, new_max, is_greedy, body, compiler, on_success, true);
3833         // Unroll the forced matches from 0 to min.  This can cause chains of
3834         // TextNodes (which the parser does not generate).  These should be
3835         // combined if it turns out they hinder good code generation.
3836         for (int i = 0; i < min; i++) {
3837           answer = body->ToNode(compiler, answer);
3838         }
3839         return answer;
3840       }
3841     }
3842     if (max <= kMaxUnrolledMaxMatches && min == 0) {
3843       ASSERT(max > 0);  // Due to the 'if' above.
3844       RegExpExpansionLimiter limiter(compiler, max);
3845       if (limiter.ok_to_expand()) {
3846         // Unroll the optional matches up to max.
3847         RegExpNode* answer = on_success;
3848         for (int i = 0; i < max; i++) {
3849           ChoiceNode* alternation = new ChoiceNode(2);
3850           if (is_greedy) {
3851             alternation->AddAlternative(
3852                 GuardedAlternative(body->ToNode(compiler, answer)));
3853             alternation->AddAlternative(GuardedAlternative(on_success));
3854           } else {
3855             alternation->AddAlternative(GuardedAlternative(on_success));
3856             alternation->AddAlternative(
3857                 GuardedAlternative(body->ToNode(compiler, answer)));
3858           }
3859           answer = alternation;
3860           if (not_at_start) alternation->set_not_at_start();
3861         }
3862         return answer;
3863       }
3864     }
3865   }
3866   bool has_min = min > 0;
3867   bool has_max = max < RegExpTree::kInfinity;
3868   bool needs_counter = has_min || has_max;
3869   int reg_ctr = needs_counter
3870       ? compiler->AllocateRegister()
3871       : RegExpCompiler::kNoRegister;
3872   LoopChoiceNode* center = new LoopChoiceNode(body->min_match() == 0);
3873   if (not_at_start) center->set_not_at_start();
3874   RegExpNode* loop_return = needs_counter
3875       ? static_cast<RegExpNode*>(ActionNode::IncrementRegister(reg_ctr, center))
3876       : static_cast<RegExpNode*>(center);
3877   if (body_can_be_empty) {
3878     // If the body can be empty we need to check if it was and then
3879     // backtrack.
3880     loop_return = ActionNode::EmptyMatchCheck(body_start_reg,
3881                                               reg_ctr,
3882                                               min,
3883                                               loop_return);
3884   }
3885   RegExpNode* body_node = body->ToNode(compiler, loop_return);
3886   if (body_can_be_empty) {
3887     // If the body can be empty we need to store the start position
3888     // so we can bail out if it was empty.
3889     body_node = ActionNode::StorePosition(body_start_reg, false, body_node);
3890   }
3891   if (needs_capture_clearing) {
3892     // Before entering the body of this loop we need to clear captures.
3893     body_node = ActionNode::ClearCaptures(capture_registers, body_node);
3894   }
3895   GuardedAlternative body_alt(body_node);
3896   if (has_max) {
3897     Guard* body_guard = new Guard(reg_ctr, Guard::LT, max);
3898     body_alt.AddGuard(body_guard);
3899   }
3900   GuardedAlternative rest_alt(on_success);
3901   if (has_min) {
3902     Guard* rest_guard = new Guard(reg_ctr, Guard::GEQ, min);
3903     rest_alt.AddGuard(rest_guard);
3904   }
3905   if (is_greedy) {
3906     center->AddLoopAlternative(body_alt);
3907     center->AddContinueAlternative(rest_alt);
3908   } else {
3909     center->AddContinueAlternative(rest_alt);
3910     center->AddLoopAlternative(body_alt);
3911   }
3912   if (needs_counter) {
3913     return ActionNode::SetRegister(reg_ctr, 0, center);
3914   } else {
3915     return center;
3916   }
3917 }
3918 
3919 
ToNode(RegExpCompiler * compiler,RegExpNode * on_success)3920 RegExpNode* RegExpAssertion::ToNode(RegExpCompiler* compiler,
3921                                     RegExpNode* on_success) {
3922   NodeInfo info;
3923   switch (type()) {
3924     case START_OF_LINE:
3925       return AssertionNode::AfterNewline(on_success);
3926     case START_OF_INPUT:
3927       return AssertionNode::AtStart(on_success);
3928     case BOUNDARY:
3929       return AssertionNode::AtBoundary(on_success);
3930     case NON_BOUNDARY:
3931       return AssertionNode::AtNonBoundary(on_success);
3932     case END_OF_INPUT:
3933       return AssertionNode::AtEnd(on_success);
3934     case END_OF_LINE: {
3935       // Compile $ in multiline regexps as an alternation with a positive
3936       // lookahead in one side and an end-of-input on the other side.
3937       // We need two registers for the lookahead.
3938       int stack_pointer_register = compiler->AllocateRegister();
3939       int position_register = compiler->AllocateRegister();
3940       // The ChoiceNode to distinguish between a newline and end-of-input.
3941       ChoiceNode* result = new ChoiceNode(2);
3942       // Create a newline atom.
3943       ZoneList<CharacterRange>* newline_ranges =
3944           new ZoneList<CharacterRange>(3);
3945       CharacterRange::AddClassEscape('n', newline_ranges);
3946       RegExpCharacterClass* newline_atom = new RegExpCharacterClass('n');
3947       TextNode* newline_matcher = new TextNode(
3948          newline_atom,
3949          ActionNode::PositiveSubmatchSuccess(stack_pointer_register,
3950                                              position_register,
3951                                              0,  // No captures inside.
3952                                              -1,  // Ignored if no captures.
3953                                              on_success));
3954       // Create an end-of-input matcher.
3955       RegExpNode* end_of_line = ActionNode::BeginSubmatch(
3956           stack_pointer_register,
3957           position_register,
3958           newline_matcher);
3959       // Add the two alternatives to the ChoiceNode.
3960       GuardedAlternative eol_alternative(end_of_line);
3961       result->AddAlternative(eol_alternative);
3962       GuardedAlternative end_alternative(AssertionNode::AtEnd(on_success));
3963       result->AddAlternative(end_alternative);
3964       return result;
3965     }
3966     default:
3967       UNREACHABLE();
3968   }
3969   return on_success;
3970 }
3971 
3972 
ToNode(RegExpCompiler * compiler,RegExpNode * on_success)3973 RegExpNode* RegExpBackReference::ToNode(RegExpCompiler* compiler,
3974                                         RegExpNode* on_success) {
3975   return new BackReferenceNode(RegExpCapture::StartRegister(index()),
3976                                RegExpCapture::EndRegister(index()),
3977                                on_success);
3978 }
3979 
3980 
ToNode(RegExpCompiler * compiler,RegExpNode * on_success)3981 RegExpNode* RegExpEmpty::ToNode(RegExpCompiler* compiler,
3982                                 RegExpNode* on_success) {
3983   return on_success;
3984 }
3985 
3986 
ToNode(RegExpCompiler * compiler,RegExpNode * on_success)3987 RegExpNode* RegExpLookahead::ToNode(RegExpCompiler* compiler,
3988                                     RegExpNode* on_success) {
3989   int stack_pointer_register = compiler->AllocateRegister();
3990   int position_register = compiler->AllocateRegister();
3991 
3992   const int registers_per_capture = 2;
3993   const int register_of_first_capture = 2;
3994   int register_count = capture_count_ * registers_per_capture;
3995   int register_start =
3996     register_of_first_capture + capture_from_ * registers_per_capture;
3997 
3998   RegExpNode* success;
3999   if (is_positive()) {
4000     RegExpNode* node = ActionNode::BeginSubmatch(
4001         stack_pointer_register,
4002         position_register,
4003         body()->ToNode(
4004             compiler,
4005             ActionNode::PositiveSubmatchSuccess(stack_pointer_register,
4006                                                 position_register,
4007                                                 register_count,
4008                                                 register_start,
4009                                                 on_success)));
4010     return node;
4011   } else {
4012     // We use a ChoiceNode for a negative lookahead because it has most of
4013     // the characteristics we need.  It has the body of the lookahead as its
4014     // first alternative and the expression after the lookahead of the second
4015     // alternative.  If the first alternative succeeds then the
4016     // NegativeSubmatchSuccess will unwind the stack including everything the
4017     // choice node set up and backtrack.  If the first alternative fails then
4018     // the second alternative is tried, which is exactly the desired result
4019     // for a negative lookahead.  The NegativeLookaheadChoiceNode is a special
4020     // ChoiceNode that knows to ignore the first exit when calculating quick
4021     // checks.
4022     GuardedAlternative body_alt(
4023         body()->ToNode(
4024             compiler,
4025             success = new NegativeSubmatchSuccess(stack_pointer_register,
4026                                                   position_register,
4027                                                   register_count,
4028                                                   register_start)));
4029     ChoiceNode* choice_node =
4030         new NegativeLookaheadChoiceNode(body_alt,
4031                                         GuardedAlternative(on_success));
4032     return ActionNode::BeginSubmatch(stack_pointer_register,
4033                                      position_register,
4034                                      choice_node);
4035   }
4036 }
4037 
4038 
ToNode(RegExpCompiler * compiler,RegExpNode * on_success)4039 RegExpNode* RegExpCapture::ToNode(RegExpCompiler* compiler,
4040                                   RegExpNode* on_success) {
4041   return ToNode(body(), index(), compiler, on_success);
4042 }
4043 
4044 
ToNode(RegExpTree * body,int index,RegExpCompiler * compiler,RegExpNode * on_success)4045 RegExpNode* RegExpCapture::ToNode(RegExpTree* body,
4046                                   int index,
4047                                   RegExpCompiler* compiler,
4048                                   RegExpNode* on_success) {
4049   int start_reg = RegExpCapture::StartRegister(index);
4050   int end_reg = RegExpCapture::EndRegister(index);
4051   RegExpNode* store_end = ActionNode::StorePosition(end_reg, true, on_success);
4052   RegExpNode* body_node = body->ToNode(compiler, store_end);
4053   return ActionNode::StorePosition(start_reg, true, body_node);
4054 }
4055 
4056 
ToNode(RegExpCompiler * compiler,RegExpNode * on_success)4057 RegExpNode* RegExpAlternative::ToNode(RegExpCompiler* compiler,
4058                                       RegExpNode* on_success) {
4059   ZoneList<RegExpTree*>* children = nodes();
4060   RegExpNode* current = on_success;
4061   for (int i = children->length() - 1; i >= 0; i--) {
4062     current = children->at(i)->ToNode(compiler, current);
4063   }
4064   return current;
4065 }
4066 
4067 
AddClass(const uc16 * elmv,int elmc,ZoneList<CharacterRange> * ranges)4068 static void AddClass(const uc16* elmv,
4069                      int elmc,
4070                      ZoneList<CharacterRange>* ranges) {
4071   for (int i = 0; i < elmc; i += 2) {
4072     ASSERT(elmv[i] <= elmv[i + 1]);
4073     ranges->Add(CharacterRange(elmv[i], elmv[i + 1]));
4074   }
4075 }
4076 
4077 
AddClassNegated(const uc16 * elmv,int elmc,ZoneList<CharacterRange> * ranges)4078 static void AddClassNegated(const uc16 *elmv,
4079                             int elmc,
4080                             ZoneList<CharacterRange>* ranges) {
4081   ASSERT(elmv[0] != 0x0000);
4082   ASSERT(elmv[elmc-1] != String::kMaxUtf16CodeUnit);
4083   uc16 last = 0x0000;
4084   for (int i = 0; i < elmc; i += 2) {
4085     ASSERT(last <= elmv[i] - 1);
4086     ASSERT(elmv[i] <= elmv[i + 1]);
4087     ranges->Add(CharacterRange(last, elmv[i] - 1));
4088     last = elmv[i + 1] + 1;
4089   }
4090   ranges->Add(CharacterRange(last, String::kMaxUtf16CodeUnit));
4091 }
4092 
4093 
AddClassEscape(uc16 type,ZoneList<CharacterRange> * ranges)4094 void CharacterRange::AddClassEscape(uc16 type,
4095                                     ZoneList<CharacterRange>* ranges) {
4096   switch (type) {
4097     case 's':
4098       AddClass(kSpaceRanges, kSpaceRangeCount, ranges);
4099       break;
4100     case 'S':
4101       AddClassNegated(kSpaceRanges, kSpaceRangeCount, ranges);
4102       break;
4103     case 'w':
4104       AddClass(kWordRanges, kWordRangeCount, ranges);
4105       break;
4106     case 'W':
4107       AddClassNegated(kWordRanges, kWordRangeCount, ranges);
4108       break;
4109     case 'd':
4110       AddClass(kDigitRanges, kDigitRangeCount, ranges);
4111       break;
4112     case 'D':
4113       AddClassNegated(kDigitRanges, kDigitRangeCount, ranges);
4114       break;
4115     case '.':
4116       AddClassNegated(kLineTerminatorRanges,
4117                       kLineTerminatorRangeCount,
4118                       ranges);
4119       break;
4120     // This is not a character range as defined by the spec but a
4121     // convenient shorthand for a character class that matches any
4122     // character.
4123     case '*':
4124       ranges->Add(CharacterRange::Everything());
4125       break;
4126     // This is the set of characters matched by the $ and ^ symbols
4127     // in multiline mode.
4128     case 'n':
4129       AddClass(kLineTerminatorRanges,
4130                kLineTerminatorRangeCount,
4131                ranges);
4132       break;
4133     default:
4134       UNREACHABLE();
4135   }
4136 }
4137 
4138 
GetWordBounds()4139 Vector<const uc16> CharacterRange::GetWordBounds() {
4140   return Vector<const uc16>(kWordRanges, kWordRangeCount);
4141 }
4142 
4143 
4144 class CharacterRangeSplitter {
4145  public:
CharacterRangeSplitter(ZoneList<CharacterRange> ** included,ZoneList<CharacterRange> ** excluded)4146   CharacterRangeSplitter(ZoneList<CharacterRange>** included,
4147                           ZoneList<CharacterRange>** excluded)
4148       : included_(included),
4149         excluded_(excluded) { }
4150   void Call(uc16 from, DispatchTable::Entry entry);
4151 
4152   static const int kInBase = 0;
4153   static const int kInOverlay = 1;
4154 
4155  private:
4156   ZoneList<CharacterRange>** included_;
4157   ZoneList<CharacterRange>** excluded_;
4158 };
4159 
4160 
Call(uc16 from,DispatchTable::Entry entry)4161 void CharacterRangeSplitter::Call(uc16 from, DispatchTable::Entry entry) {
4162   if (!entry.out_set()->Get(kInBase)) return;
4163   ZoneList<CharacterRange>** target = entry.out_set()->Get(kInOverlay)
4164     ? included_
4165     : excluded_;
4166   if (*target == NULL) *target = new ZoneList<CharacterRange>(2);
4167   (*target)->Add(CharacterRange(entry.from(), entry.to()));
4168 }
4169 
4170 
Split(ZoneList<CharacterRange> * base,Vector<const uc16> overlay,ZoneList<CharacterRange> ** included,ZoneList<CharacterRange> ** excluded)4171 void CharacterRange::Split(ZoneList<CharacterRange>* base,
4172                            Vector<const uc16> overlay,
4173                            ZoneList<CharacterRange>** included,
4174                            ZoneList<CharacterRange>** excluded) {
4175   ASSERT_EQ(NULL, *included);
4176   ASSERT_EQ(NULL, *excluded);
4177   DispatchTable table;
4178   for (int i = 0; i < base->length(); i++)
4179     table.AddRange(base->at(i), CharacterRangeSplitter::kInBase);
4180   for (int i = 0; i < overlay.length(); i += 2) {
4181     table.AddRange(CharacterRange(overlay[i], overlay[i+1]),
4182                    CharacterRangeSplitter::kInOverlay);
4183   }
4184   CharacterRangeSplitter callback(included, excluded);
4185   table.ForEach(&callback);
4186 }
4187 
4188 
AddCaseEquivalents(ZoneList<CharacterRange> * ranges,bool is_ascii)4189 void CharacterRange::AddCaseEquivalents(ZoneList<CharacterRange>* ranges,
4190                                         bool is_ascii) {
4191   Isolate* isolate = Isolate::Current();
4192   uc16 bottom = from();
4193   uc16 top = to();
4194   if (is_ascii) {
4195     if (bottom > String::kMaxAsciiCharCode) return;
4196     if (top > String::kMaxAsciiCharCode) top = String::kMaxAsciiCharCode;
4197   }
4198   unibrow::uchar chars[unibrow::Ecma262UnCanonicalize::kMaxWidth];
4199   if (top == bottom) {
4200     // If this is a singleton we just expand the one character.
4201     int length = isolate->jsregexp_uncanonicalize()->get(bottom, '\0', chars);
4202     for (int i = 0; i < length; i++) {
4203       uc32 chr = chars[i];
4204       if (chr != bottom) {
4205         ranges->Add(CharacterRange::Singleton(chars[i]));
4206       }
4207     }
4208   } else {
4209     // If this is a range we expand the characters block by block,
4210     // expanding contiguous subranges (blocks) one at a time.
4211     // The approach is as follows.  For a given start character we
4212     // look up the remainder of the block that contains it (represented
4213     // by the end point), for instance we find 'z' if the character
4214     // is 'c'.  A block is characterized by the property
4215     // that all characters uncanonicalize in the same way, except that
4216     // each entry in the result is incremented by the distance from the first
4217     // element.  So a-z is a block because 'a' uncanonicalizes to ['a', 'A'] and
4218     // the k'th letter uncanonicalizes to ['a' + k, 'A' + k].
4219     // Once we've found the end point we look up its uncanonicalization
4220     // and produce a range for each element.  For instance for [c-f]
4221     // we look up ['z', 'Z'] and produce [c-f] and [C-F].  We then only
4222     // add a range if it is not already contained in the input, so [c-f]
4223     // will be skipped but [C-F] will be added.  If this range is not
4224     // completely contained in a block we do this for all the blocks
4225     // covered by the range (handling characters that is not in a block
4226     // as a "singleton block").
4227     unibrow::uchar range[unibrow::Ecma262UnCanonicalize::kMaxWidth];
4228     int pos = bottom;
4229     while (pos < top) {
4230       int length = isolate->jsregexp_canonrange()->get(pos, '\0', range);
4231       uc16 block_end;
4232       if (length == 0) {
4233         block_end = pos;
4234       } else {
4235         ASSERT_EQ(1, length);
4236         block_end = range[0];
4237       }
4238       int end = (block_end > top) ? top : block_end;
4239       length = isolate->jsregexp_uncanonicalize()->get(block_end, '\0', range);
4240       for (int i = 0; i < length; i++) {
4241         uc32 c = range[i];
4242         uc16 range_from = c - (block_end - pos);
4243         uc16 range_to = c - (block_end - end);
4244         if (!(bottom <= range_from && range_to <= top)) {
4245           ranges->Add(CharacterRange(range_from, range_to));
4246         }
4247       }
4248       pos = end + 1;
4249     }
4250   }
4251 }
4252 
4253 
IsCanonical(ZoneList<CharacterRange> * ranges)4254 bool CharacterRange::IsCanonical(ZoneList<CharacterRange>* ranges) {
4255   ASSERT_NOT_NULL(ranges);
4256   int n = ranges->length();
4257   if (n <= 1) return true;
4258   int max = ranges->at(0).to();
4259   for (int i = 1; i < n; i++) {
4260     CharacterRange next_range = ranges->at(i);
4261     if (next_range.from() <= max + 1) return false;
4262     max = next_range.to();
4263   }
4264   return true;
4265 }
4266 
WordCharacterRelation(ZoneList<CharacterRange> * range)4267 SetRelation CharacterRange::WordCharacterRelation(
4268     ZoneList<CharacterRange>* range) {
4269   ASSERT(IsCanonical(range));
4270   int i = 0;  // Word character range index.
4271   int j = 0;  // Argument range index.
4272   ASSERT_NE(0, kWordRangeCount);
4273   SetRelation result;
4274   if (range->length() == 0) {
4275     result.SetElementsInSecondSet();
4276     return result;
4277   }
4278   CharacterRange argument_range = range->at(0);
4279   CharacterRange word_range = CharacterRange(kWordRanges[0], kWordRanges[1]);
4280   while (i < kWordRangeCount && j < range->length()) {
4281     // Check the two ranges for the five cases:
4282     // - no overlap.
4283     // - partial overlap (there are elements in both ranges that isn't
4284     //   in the other, and there are also elements that are in both).
4285     // - argument range entirely inside word range.
4286     // - word range entirely inside argument range.
4287     // - ranges are completely equal.
4288 
4289     // First check for no overlap. The earlier range is not in the other set.
4290     if (argument_range.from() > word_range.to()) {
4291       // Ranges are disjoint. The earlier word range contains elements that
4292       // cannot be in the argument set.
4293       result.SetElementsInSecondSet();
4294     } else if (word_range.from() > argument_range.to()) {
4295       // Ranges are disjoint. The earlier argument range contains elements that
4296       // cannot be in the word set.
4297       result.SetElementsInFirstSet();
4298     } else if (word_range.from() <= argument_range.from() &&
4299                word_range.to() >= argument_range.from()) {
4300       result.SetElementsInBothSets();
4301       // argument range completely inside word range.
4302       if (word_range.from() < argument_range.from() ||
4303           word_range.to() > argument_range.from()) {
4304         result.SetElementsInSecondSet();
4305       }
4306     } else if (word_range.from() >= argument_range.from() &&
4307                word_range.to() <= argument_range.from()) {
4308       result.SetElementsInBothSets();
4309       result.SetElementsInFirstSet();
4310     } else {
4311       // There is overlap, and neither is a subrange of the other
4312       result.SetElementsInFirstSet();
4313       result.SetElementsInSecondSet();
4314       result.SetElementsInBothSets();
4315     }
4316     if (result.NonTrivialIntersection()) {
4317       // The result is as (im)precise as we can possibly make it.
4318       return result;
4319     }
4320     // Progress the range(s) with minimal to-character.
4321     uc16 word_to = word_range.to();
4322     uc16 argument_to = argument_range.to();
4323     if (argument_to <= word_to) {
4324       j++;
4325       if (j < range->length()) {
4326         argument_range = range->at(j);
4327       }
4328     }
4329     if (word_to <= argument_to) {
4330       i += 2;
4331       if (i < kWordRangeCount) {
4332         word_range = CharacterRange(kWordRanges[i], kWordRanges[i + 1]);
4333       }
4334     }
4335   }
4336   // Check if anything wasn't compared in the loop.
4337   if (i < kWordRangeCount) {
4338     // word range contains something not in argument range.
4339     result.SetElementsInSecondSet();
4340   } else if (j < range->length()) {
4341     // Argument range contains something not in word range.
4342     result.SetElementsInFirstSet();
4343   }
4344 
4345   return result;
4346 }
4347 
4348 
ranges()4349 ZoneList<CharacterRange>* CharacterSet::ranges() {
4350   if (ranges_ == NULL) {
4351     ranges_ = new ZoneList<CharacterRange>(2);
4352     CharacterRange::AddClassEscape(standard_set_type_, ranges_);
4353   }
4354   return ranges_;
4355 }
4356 
4357 
4358 // Move a number of elements in a zonelist to another position
4359 // in the same list. Handles overlapping source and target areas.
MoveRanges(ZoneList<CharacterRange> * list,int from,int to,int count)4360 static void MoveRanges(ZoneList<CharacterRange>* list,
4361                        int from,
4362                        int to,
4363                        int count) {
4364   // Ranges are potentially overlapping.
4365   if (from < to) {
4366     for (int i = count - 1; i >= 0; i--) {
4367       list->at(to + i) = list->at(from + i);
4368     }
4369   } else {
4370     for (int i = 0; i < count; i++) {
4371       list->at(to + i) = list->at(from + i);
4372     }
4373   }
4374 }
4375 
4376 
InsertRangeInCanonicalList(ZoneList<CharacterRange> * list,int count,CharacterRange insert)4377 static int InsertRangeInCanonicalList(ZoneList<CharacterRange>* list,
4378                                       int count,
4379                                       CharacterRange insert) {
4380   // Inserts a range into list[0..count[, which must be sorted
4381   // by from value and non-overlapping and non-adjacent, using at most
4382   // list[0..count] for the result. Returns the number of resulting
4383   // canonicalized ranges. Inserting a range may collapse existing ranges into
4384   // fewer ranges, so the return value can be anything in the range 1..count+1.
4385   uc16 from = insert.from();
4386   uc16 to = insert.to();
4387   int start_pos = 0;
4388   int end_pos = count;
4389   for (int i = count - 1; i >= 0; i--) {
4390     CharacterRange current = list->at(i);
4391     if (current.from() > to + 1) {
4392       end_pos = i;
4393     } else if (current.to() + 1 < from) {
4394       start_pos = i + 1;
4395       break;
4396     }
4397   }
4398 
4399   // Inserted range overlaps, or is adjacent to, ranges at positions
4400   // [start_pos..end_pos[. Ranges before start_pos or at or after end_pos are
4401   // not affected by the insertion.
4402   // If start_pos == end_pos, the range must be inserted before start_pos.
4403   // if start_pos < end_pos, the entire range from start_pos to end_pos
4404   // must be merged with the insert range.
4405 
4406   if (start_pos == end_pos) {
4407     // Insert between existing ranges at position start_pos.
4408     if (start_pos < count) {
4409       MoveRanges(list, start_pos, start_pos + 1, count - start_pos);
4410     }
4411     list->at(start_pos) = insert;
4412     return count + 1;
4413   }
4414   if (start_pos + 1 == end_pos) {
4415     // Replace single existing range at position start_pos.
4416     CharacterRange to_replace = list->at(start_pos);
4417     int new_from = Min(to_replace.from(), from);
4418     int new_to = Max(to_replace.to(), to);
4419     list->at(start_pos) = CharacterRange(new_from, new_to);
4420     return count;
4421   }
4422   // Replace a number of existing ranges from start_pos to end_pos - 1.
4423   // Move the remaining ranges down.
4424 
4425   int new_from = Min(list->at(start_pos).from(), from);
4426   int new_to = Max(list->at(end_pos - 1).to(), to);
4427   if (end_pos < count) {
4428     MoveRanges(list, end_pos, start_pos + 1, count - end_pos);
4429   }
4430   list->at(start_pos) = CharacterRange(new_from, new_to);
4431   return count - (end_pos - start_pos) + 1;
4432 }
4433 
4434 
Canonicalize()4435 void CharacterSet::Canonicalize() {
4436   // Special/default classes are always considered canonical. The result
4437   // of calling ranges() will be sorted.
4438   if (ranges_ == NULL) return;
4439   CharacterRange::Canonicalize(ranges_);
4440 }
4441 
4442 
Canonicalize(ZoneList<CharacterRange> * character_ranges)4443 void CharacterRange::Canonicalize(ZoneList<CharacterRange>* character_ranges) {
4444   if (character_ranges->length() <= 1) return;
4445   // Check whether ranges are already canonical (increasing, non-overlapping,
4446   // non-adjacent).
4447   int n = character_ranges->length();
4448   int max = character_ranges->at(0).to();
4449   int i = 1;
4450   while (i < n) {
4451     CharacterRange current = character_ranges->at(i);
4452     if (current.from() <= max + 1) {
4453       break;
4454     }
4455     max = current.to();
4456     i++;
4457   }
4458   // Canonical until the i'th range. If that's all of them, we are done.
4459   if (i == n) return;
4460 
4461   // The ranges at index i and forward are not canonicalized. Make them so by
4462   // doing the equivalent of insertion sort (inserting each into the previous
4463   // list, in order).
4464   // Notice that inserting a range can reduce the number of ranges in the
4465   // result due to combining of adjacent and overlapping ranges.
4466   int read = i;  // Range to insert.
4467   int num_canonical = i;  // Length of canonicalized part of list.
4468   do {
4469     num_canonical = InsertRangeInCanonicalList(character_ranges,
4470                                                num_canonical,
4471                                                character_ranges->at(read));
4472     read++;
4473   } while (read < n);
4474   character_ranges->Rewind(num_canonical);
4475 
4476   ASSERT(CharacterRange::IsCanonical(character_ranges));
4477 }
4478 
4479 
4480 // Utility function for CharacterRange::Merge. Adds a range at the end of
4481 // a canonicalized range list, if necessary merging the range with the last
4482 // range of the list.
AddRangeToSet(ZoneList<CharacterRange> * set,CharacterRange range)4483 static void AddRangeToSet(ZoneList<CharacterRange>* set, CharacterRange range) {
4484   if (set == NULL) return;
4485   ASSERT(set->length() == 0 || set->at(set->length() - 1).to() < range.from());
4486   int n = set->length();
4487   if (n > 0) {
4488     CharacterRange lastRange = set->at(n - 1);
4489     if (lastRange.to() == range.from() - 1) {
4490       set->at(n - 1) = CharacterRange(lastRange.from(), range.to());
4491       return;
4492     }
4493   }
4494   set->Add(range);
4495 }
4496 
4497 
AddRangeToSelectedSet(int selector,ZoneList<CharacterRange> * first_set,ZoneList<CharacterRange> * second_set,ZoneList<CharacterRange> * intersection_set,CharacterRange range)4498 static void AddRangeToSelectedSet(int selector,
4499                                   ZoneList<CharacterRange>* first_set,
4500                                   ZoneList<CharacterRange>* second_set,
4501                                   ZoneList<CharacterRange>* intersection_set,
4502                                   CharacterRange range) {
4503   switch (selector) {
4504     case kInsideFirst:
4505       AddRangeToSet(first_set, range);
4506       break;
4507     case kInsideSecond:
4508       AddRangeToSet(second_set, range);
4509       break;
4510     case kInsideBoth:
4511       AddRangeToSet(intersection_set, range);
4512       break;
4513   }
4514 }
4515 
4516 
4517 
Merge(ZoneList<CharacterRange> * first_set,ZoneList<CharacterRange> * second_set,ZoneList<CharacterRange> * first_set_only_out,ZoneList<CharacterRange> * second_set_only_out,ZoneList<CharacterRange> * both_sets_out)4518 void CharacterRange::Merge(ZoneList<CharacterRange>* first_set,
4519                            ZoneList<CharacterRange>* second_set,
4520                            ZoneList<CharacterRange>* first_set_only_out,
4521                            ZoneList<CharacterRange>* second_set_only_out,
4522                            ZoneList<CharacterRange>* both_sets_out) {
4523   // Inputs are canonicalized.
4524   ASSERT(CharacterRange::IsCanonical(first_set));
4525   ASSERT(CharacterRange::IsCanonical(second_set));
4526   // Outputs are empty, if applicable.
4527   ASSERT(first_set_only_out == NULL || first_set_only_out->length() == 0);
4528   ASSERT(second_set_only_out == NULL || second_set_only_out->length() == 0);
4529   ASSERT(both_sets_out == NULL || both_sets_out->length() == 0);
4530 
4531   // Merge sets by iterating through the lists in order of lowest "from" value,
4532   // and putting intervals into one of three sets.
4533 
4534   if (first_set->length() == 0) {
4535     second_set_only_out->AddAll(*second_set);
4536     return;
4537   }
4538   if (second_set->length() == 0) {
4539     first_set_only_out->AddAll(*first_set);
4540     return;
4541   }
4542   // Indices into input lists.
4543   int i1 = 0;
4544   int i2 = 0;
4545   // Cache length of input lists.
4546   int n1 = first_set->length();
4547   int n2 = second_set->length();
4548   // Current range. May be invalid if state is kInsideNone.
4549   int from = 0;
4550   int to = -1;
4551   // Where current range comes from.
4552   int state = kInsideNone;
4553 
4554   while (i1 < n1 || i2 < n2) {
4555     CharacterRange next_range;
4556     int range_source;
4557     if (i2 == n2 ||
4558         (i1 < n1 && first_set->at(i1).from() < second_set->at(i2).from())) {
4559       // Next smallest element is in first set.
4560       next_range = first_set->at(i1++);
4561       range_source = kInsideFirst;
4562     } else {
4563       // Next smallest element is in second set.
4564       next_range = second_set->at(i2++);
4565       range_source = kInsideSecond;
4566     }
4567     if (to < next_range.from()) {
4568       // Ranges disjoint: |current|  |next|
4569       AddRangeToSelectedSet(state,
4570                             first_set_only_out,
4571                             second_set_only_out,
4572                             both_sets_out,
4573                             CharacterRange(from, to));
4574       from = next_range.from();
4575       to = next_range.to();
4576       state = range_source;
4577     } else {
4578       if (from < next_range.from()) {
4579         AddRangeToSelectedSet(state,
4580                               first_set_only_out,
4581                               second_set_only_out,
4582                               both_sets_out,
4583                               CharacterRange(from, next_range.from()-1));
4584       }
4585       if (to < next_range.to()) {
4586         // Ranges overlap:  |current|
4587         //                       |next|
4588         AddRangeToSelectedSet(state | range_source,
4589                               first_set_only_out,
4590                               second_set_only_out,
4591                               both_sets_out,
4592                               CharacterRange(next_range.from(), to));
4593         from = to + 1;
4594         to = next_range.to();
4595         state = range_source;
4596       } else {
4597         // Range included:    |current| , possibly ending at same character.
4598         //                      |next|
4599         AddRangeToSelectedSet(
4600             state | range_source,
4601             first_set_only_out,
4602             second_set_only_out,
4603             both_sets_out,
4604             CharacterRange(next_range.from(), next_range.to()));
4605         from = next_range.to() + 1;
4606         // If ranges end at same character, both ranges are consumed completely.
4607         if (next_range.to() == to) state = kInsideNone;
4608       }
4609     }
4610   }
4611   AddRangeToSelectedSet(state,
4612                         first_set_only_out,
4613                         second_set_only_out,
4614                         both_sets_out,
4615                         CharacterRange(from, to));
4616 }
4617 
4618 
Negate(ZoneList<CharacterRange> * ranges,ZoneList<CharacterRange> * negated_ranges)4619 void CharacterRange::Negate(ZoneList<CharacterRange>* ranges,
4620                             ZoneList<CharacterRange>* negated_ranges) {
4621   ASSERT(CharacterRange::IsCanonical(ranges));
4622   ASSERT_EQ(0, negated_ranges->length());
4623   int range_count = ranges->length();
4624   uc16 from = 0;
4625   int i = 0;
4626   if (range_count > 0 && ranges->at(0).from() == 0) {
4627     from = ranges->at(0).to();
4628     i = 1;
4629   }
4630   while (i < range_count) {
4631     CharacterRange range = ranges->at(i);
4632     negated_ranges->Add(CharacterRange(from + 1, range.from() - 1));
4633     from = range.to();
4634     i++;
4635   }
4636   if (from < String::kMaxUtf16CodeUnit) {
4637     negated_ranges->Add(CharacterRange(from + 1, String::kMaxUtf16CodeUnit));
4638   }
4639 }
4640 
4641 
4642 
4643 // -------------------------------------------------------------------
4644 // Interest propagation
4645 
4646 
TryGetSibling(NodeInfo * info)4647 RegExpNode* RegExpNode::TryGetSibling(NodeInfo* info) {
4648   for (int i = 0; i < siblings_.length(); i++) {
4649     RegExpNode* sibling = siblings_.Get(i);
4650     if (sibling->info()->Matches(info))
4651       return sibling;
4652   }
4653   return NULL;
4654 }
4655 
4656 
EnsureSibling(NodeInfo * info,bool * cloned)4657 RegExpNode* RegExpNode::EnsureSibling(NodeInfo* info, bool* cloned) {
4658   ASSERT_EQ(false, *cloned);
4659   siblings_.Ensure(this);
4660   RegExpNode* result = TryGetSibling(info);
4661   if (result != NULL) return result;
4662   result = this->Clone();
4663   NodeInfo* new_info = result->info();
4664   new_info->ResetCompilationState();
4665   new_info->AddFromPreceding(info);
4666   AddSibling(result);
4667   *cloned = true;
4668   return result;
4669 }
4670 
4671 
4672 template <class C>
PropagateToEndpoint(C * node,NodeInfo * info)4673 static RegExpNode* PropagateToEndpoint(C* node, NodeInfo* info) {
4674   NodeInfo full_info(*node->info());
4675   full_info.AddFromPreceding(info);
4676   bool cloned = false;
4677   return RegExpNode::EnsureSibling(node, &full_info, &cloned);
4678 }
4679 
4680 
4681 // -------------------------------------------------------------------
4682 // Splay tree
4683 
4684 
Extend(unsigned value)4685 OutSet* OutSet::Extend(unsigned value) {
4686   if (Get(value))
4687     return this;
4688   if (successors() != NULL) {
4689     for (int i = 0; i < successors()->length(); i++) {
4690       OutSet* successor = successors()->at(i);
4691       if (successor->Get(value))
4692         return successor;
4693     }
4694   } else {
4695     successors_ = new ZoneList<OutSet*>(2);
4696   }
4697   OutSet* result = new OutSet(first_, remaining_);
4698   result->Set(value);
4699   successors()->Add(result);
4700   return result;
4701 }
4702 
4703 
Set(unsigned value)4704 void OutSet::Set(unsigned value) {
4705   if (value < kFirstLimit) {
4706     first_ |= (1 << value);
4707   } else {
4708     if (remaining_ == NULL)
4709       remaining_ = new ZoneList<unsigned>(1);
4710     if (remaining_->is_empty() || !remaining_->Contains(value))
4711       remaining_->Add(value);
4712   }
4713 }
4714 
4715 
Get(unsigned value)4716 bool OutSet::Get(unsigned value) {
4717   if (value < kFirstLimit) {
4718     return (first_ & (1 << value)) != 0;
4719   } else if (remaining_ == NULL) {
4720     return false;
4721   } else {
4722     return remaining_->Contains(value);
4723   }
4724 }
4725 
4726 
4727 const uc16 DispatchTable::Config::kNoKey = unibrow::Utf8::kBadChar;
4728 
4729 
AddRange(CharacterRange full_range,int value)4730 void DispatchTable::AddRange(CharacterRange full_range, int value) {
4731   CharacterRange current = full_range;
4732   if (tree()->is_empty()) {
4733     // If this is the first range we just insert into the table.
4734     ZoneSplayTree<Config>::Locator loc;
4735     ASSERT_RESULT(tree()->Insert(current.from(), &loc));
4736     loc.set_value(Entry(current.from(), current.to(), empty()->Extend(value)));
4737     return;
4738   }
4739   // First see if there is a range to the left of this one that
4740   // overlaps.
4741   ZoneSplayTree<Config>::Locator loc;
4742   if (tree()->FindGreatestLessThan(current.from(), &loc)) {
4743     Entry* entry = &loc.value();
4744     // If we've found a range that overlaps with this one, and it
4745     // starts strictly to the left of this one, we have to fix it
4746     // because the following code only handles ranges that start on
4747     // or after the start point of the range we're adding.
4748     if (entry->from() < current.from() && entry->to() >= current.from()) {
4749       // Snap the overlapping range in half around the start point of
4750       // the range we're adding.
4751       CharacterRange left(entry->from(), current.from() - 1);
4752       CharacterRange right(current.from(), entry->to());
4753       // The left part of the overlapping range doesn't overlap.
4754       // Truncate the whole entry to be just the left part.
4755       entry->set_to(left.to());
4756       // The right part is the one that overlaps.  We add this part
4757       // to the map and let the next step deal with merging it with
4758       // the range we're adding.
4759       ZoneSplayTree<Config>::Locator loc;
4760       ASSERT_RESULT(tree()->Insert(right.from(), &loc));
4761       loc.set_value(Entry(right.from(),
4762                           right.to(),
4763                           entry->out_set()));
4764     }
4765   }
4766   while (current.is_valid()) {
4767     if (tree()->FindLeastGreaterThan(current.from(), &loc) &&
4768         (loc.value().from() <= current.to()) &&
4769         (loc.value().to() >= current.from())) {
4770       Entry* entry = &loc.value();
4771       // We have overlap.  If there is space between the start point of
4772       // the range we're adding and where the overlapping range starts
4773       // then we have to add a range covering just that space.
4774       if (current.from() < entry->from()) {
4775         ZoneSplayTree<Config>::Locator ins;
4776         ASSERT_RESULT(tree()->Insert(current.from(), &ins));
4777         ins.set_value(Entry(current.from(),
4778                             entry->from() - 1,
4779                             empty()->Extend(value)));
4780         current.set_from(entry->from());
4781       }
4782       ASSERT_EQ(current.from(), entry->from());
4783       // If the overlapping range extends beyond the one we want to add
4784       // we have to snap the right part off and add it separately.
4785       if (entry->to() > current.to()) {
4786         ZoneSplayTree<Config>::Locator ins;
4787         ASSERT_RESULT(tree()->Insert(current.to() + 1, &ins));
4788         ins.set_value(Entry(current.to() + 1,
4789                             entry->to(),
4790                             entry->out_set()));
4791         entry->set_to(current.to());
4792       }
4793       ASSERT(entry->to() <= current.to());
4794       // The overlapping range is now completely contained by the range
4795       // we're adding so we can just update it and move the start point
4796       // of the range we're adding just past it.
4797       entry->AddValue(value);
4798       // Bail out if the last interval ended at 0xFFFF since otherwise
4799       // adding 1 will wrap around to 0.
4800       if (entry->to() == String::kMaxUtf16CodeUnit)
4801         break;
4802       ASSERT(entry->to() + 1 > current.from());
4803       current.set_from(entry->to() + 1);
4804     } else {
4805       // There is no overlap so we can just add the range
4806       ZoneSplayTree<Config>::Locator ins;
4807       ASSERT_RESULT(tree()->Insert(current.from(), &ins));
4808       ins.set_value(Entry(current.from(),
4809                           current.to(),
4810                           empty()->Extend(value)));
4811       break;
4812     }
4813   }
4814 }
4815 
4816 
Get(uc16 value)4817 OutSet* DispatchTable::Get(uc16 value) {
4818   ZoneSplayTree<Config>::Locator loc;
4819   if (!tree()->FindGreatestLessThan(value, &loc))
4820     return empty();
4821   Entry* entry = &loc.value();
4822   if (value <= entry->to())
4823     return entry->out_set();
4824   else
4825     return empty();
4826 }
4827 
4828 
4829 // -------------------------------------------------------------------
4830 // Analysis
4831 
4832 
EnsureAnalyzed(RegExpNode * that)4833 void Analysis::EnsureAnalyzed(RegExpNode* that) {
4834   StackLimitCheck check(Isolate::Current());
4835   if (check.HasOverflowed()) {
4836     fail("Stack overflow");
4837     return;
4838   }
4839   if (that->info()->been_analyzed || that->info()->being_analyzed)
4840     return;
4841   that->info()->being_analyzed = true;
4842   that->Accept(this);
4843   that->info()->being_analyzed = false;
4844   that->info()->been_analyzed = true;
4845 }
4846 
4847 
VisitEnd(EndNode * that)4848 void Analysis::VisitEnd(EndNode* that) {
4849   // nothing to do
4850 }
4851 
4852 
CalculateOffsets()4853 void TextNode::CalculateOffsets() {
4854   int element_count = elements()->length();
4855   // Set up the offsets of the elements relative to the start.  This is a fixed
4856   // quantity since a TextNode can only contain fixed-width things.
4857   int cp_offset = 0;
4858   for (int i = 0; i < element_count; i++) {
4859     TextElement& elm = elements()->at(i);
4860     elm.cp_offset = cp_offset;
4861     if (elm.type == TextElement::ATOM) {
4862       cp_offset += elm.data.u_atom->data().length();
4863     } else {
4864       cp_offset++;
4865     }
4866   }
4867 }
4868 
4869 
VisitText(TextNode * that)4870 void Analysis::VisitText(TextNode* that) {
4871   if (ignore_case_) {
4872     that->MakeCaseIndependent(is_ascii_);
4873   }
4874   EnsureAnalyzed(that->on_success());
4875   if (!has_failed()) {
4876     that->CalculateOffsets();
4877   }
4878 }
4879 
4880 
VisitAction(ActionNode * that)4881 void Analysis::VisitAction(ActionNode* that) {
4882   RegExpNode* target = that->on_success();
4883   EnsureAnalyzed(target);
4884   if (!has_failed()) {
4885     // If the next node is interested in what it follows then this node
4886     // has to be interested too so it can pass the information on.
4887     that->info()->AddFromFollowing(target->info());
4888   }
4889 }
4890 
4891 
VisitChoice(ChoiceNode * that)4892 void Analysis::VisitChoice(ChoiceNode* that) {
4893   NodeInfo* info = that->info();
4894   for (int i = 0; i < that->alternatives()->length(); i++) {
4895     RegExpNode* node = that->alternatives()->at(i).node();
4896     EnsureAnalyzed(node);
4897     if (has_failed()) return;
4898     // Anything the following nodes need to know has to be known by
4899     // this node also, so it can pass it on.
4900     info->AddFromFollowing(node->info());
4901   }
4902 }
4903 
4904 
VisitLoopChoice(LoopChoiceNode * that)4905 void Analysis::VisitLoopChoice(LoopChoiceNode* that) {
4906   NodeInfo* info = that->info();
4907   for (int i = 0; i < that->alternatives()->length(); i++) {
4908     RegExpNode* node = that->alternatives()->at(i).node();
4909     if (node != that->loop_node()) {
4910       EnsureAnalyzed(node);
4911       if (has_failed()) return;
4912       info->AddFromFollowing(node->info());
4913     }
4914   }
4915   // Check the loop last since it may need the value of this node
4916   // to get a correct result.
4917   EnsureAnalyzed(that->loop_node());
4918   if (!has_failed()) {
4919     info->AddFromFollowing(that->loop_node()->info());
4920   }
4921 }
4922 
4923 
VisitBackReference(BackReferenceNode * that)4924 void Analysis::VisitBackReference(BackReferenceNode* that) {
4925   EnsureAnalyzed(that->on_success());
4926 }
4927 
4928 
VisitAssertion(AssertionNode * that)4929 void Analysis::VisitAssertion(AssertionNode* that) {
4930   EnsureAnalyzed(that->on_success());
4931   AssertionNode::AssertionNodeType type = that->type();
4932   if (type == AssertionNode::AT_BOUNDARY ||
4933       type == AssertionNode::AT_NON_BOUNDARY) {
4934     // Check if the following character is known to be a word character
4935     // or known to not be a word character.
4936     ZoneList<CharacterRange>* following_chars = that->FirstCharacterSet();
4937 
4938     CharacterRange::Canonicalize(following_chars);
4939 
4940     SetRelation word_relation =
4941         CharacterRange::WordCharacterRelation(following_chars);
4942     if (word_relation.Disjoint()) {
4943       // Includes the case where following_chars is empty (e.g., end-of-input).
4944       // Following character is definitely *not* a word character.
4945       type = (type == AssertionNode::AT_BOUNDARY) ?
4946                  AssertionNode::AFTER_WORD_CHARACTER :
4947                  AssertionNode::AFTER_NONWORD_CHARACTER;
4948       that->set_type(type);
4949     } else if (word_relation.ContainedIn()) {
4950       // Following character is definitely a word character.
4951       type = (type == AssertionNode::AT_BOUNDARY) ?
4952                  AssertionNode::AFTER_NONWORD_CHARACTER :
4953                  AssertionNode::AFTER_WORD_CHARACTER;
4954       that->set_type(type);
4955     }
4956   }
4957 }
4958 
4959 
FirstCharacterSet()4960 ZoneList<CharacterRange>* RegExpNode::FirstCharacterSet() {
4961   if (first_character_set_ == NULL) {
4962     if (ComputeFirstCharacterSet(kFirstCharBudget) < 0) {
4963       // If we can't find an exact solution within the budget, we
4964       // set the value to the set of every character, i.e., all characters
4965       // are possible.
4966       ZoneList<CharacterRange>* all_set = new ZoneList<CharacterRange>(1);
4967       all_set->Add(CharacterRange::Everything());
4968       first_character_set_ = all_set;
4969     }
4970   }
4971   return first_character_set_;
4972 }
4973 
4974 
ComputeFirstCharacterSet(int budget)4975 int RegExpNode::ComputeFirstCharacterSet(int budget) {
4976   // Default behavior is to not be able to determine the first character.
4977   return kComputeFirstCharacterSetFail;
4978 }
4979 
4980 
ComputeFirstCharacterSet(int budget)4981 int LoopChoiceNode::ComputeFirstCharacterSet(int budget) {
4982   budget--;
4983   if (budget >= 0) {
4984     // Find loop min-iteration. It's the value of the guarded choice node
4985     // with a GEQ guard, if any.
4986     int min_repetition = 0;
4987 
4988     for (int i = 0; i <= 1; i++) {
4989       GuardedAlternative alternative = alternatives()->at(i);
4990       ZoneList<Guard*>* guards = alternative.guards();
4991       if (guards != NULL && guards->length() > 0) {
4992         Guard* guard = guards->at(0);
4993         if (guard->op() == Guard::GEQ) {
4994           min_repetition = guard->value();
4995           break;
4996         }
4997       }
4998     }
4999 
5000     budget = loop_node()->ComputeFirstCharacterSet(budget);
5001     if (budget >= 0) {
5002       ZoneList<CharacterRange>* character_set =
5003           loop_node()->first_character_set();
5004       if (body_can_be_zero_length() || min_repetition == 0) {
5005         budget = continue_node()->ComputeFirstCharacterSet(budget);
5006         if (budget < 0) return budget;
5007         ZoneList<CharacterRange>* body_set =
5008             continue_node()->first_character_set();
5009         ZoneList<CharacterRange>* union_set =
5010           new ZoneList<CharacterRange>(Max(character_set->length(),
5011                                            body_set->length()));
5012         CharacterRange::Merge(character_set,
5013                               body_set,
5014                               union_set,
5015                               union_set,
5016                               union_set);
5017         character_set = union_set;
5018       }
5019       set_first_character_set(character_set);
5020     }
5021   }
5022   return budget;
5023 }
5024 
5025 
ComputeFirstCharacterSet(int budget)5026 int NegativeLookaheadChoiceNode::ComputeFirstCharacterSet(int budget) {
5027   budget--;
5028   if (budget >= 0) {
5029     GuardedAlternative successor = this->alternatives()->at(1);
5030     RegExpNode* successor_node = successor.node();
5031     budget = successor_node->ComputeFirstCharacterSet(budget);
5032     if (budget >= 0) {
5033       set_first_character_set(successor_node->first_character_set());
5034     }
5035   }
5036   return budget;
5037 }
5038 
5039 
5040 // The first character set of an EndNode is unknowable. Just use the
5041 // default implementation that fails and returns all characters as possible.
5042 
5043 
ComputeFirstCharacterSet(int budget)5044 int AssertionNode::ComputeFirstCharacterSet(int budget) {
5045   budget -= 1;
5046   if (budget >= 0) {
5047     switch (type_) {
5048       case AT_END: {
5049         set_first_character_set(new ZoneList<CharacterRange>(0));
5050         break;
5051       }
5052       case AT_START:
5053       case AT_BOUNDARY:
5054       case AT_NON_BOUNDARY:
5055       case AFTER_NEWLINE:
5056       case AFTER_NONWORD_CHARACTER:
5057       case AFTER_WORD_CHARACTER: {
5058         ASSERT_NOT_NULL(on_success());
5059         budget = on_success()->ComputeFirstCharacterSet(budget);
5060         if (budget >= 0) {
5061           set_first_character_set(on_success()->first_character_set());
5062         }
5063         break;
5064       }
5065     }
5066   }
5067   return budget;
5068 }
5069 
5070 
ComputeFirstCharacterSet(int budget)5071 int ActionNode::ComputeFirstCharacterSet(int budget) {
5072   if (type_ == POSITIVE_SUBMATCH_SUCCESS) return kComputeFirstCharacterSetFail;
5073   budget--;
5074   if (budget >= 0) {
5075     ASSERT_NOT_NULL(on_success());
5076     budget = on_success()->ComputeFirstCharacterSet(budget);
5077     if (budget >= 0) {
5078       set_first_character_set(on_success()->first_character_set());
5079     }
5080   }
5081   return budget;
5082 }
5083 
5084 
ComputeFirstCharacterSet(int budget)5085 int BackReferenceNode::ComputeFirstCharacterSet(int budget) {
5086   // We don't know anything about the first character of a backreference
5087   // at this point.
5088   // The potential first characters are the first characters of the capture,
5089   // and the first characters of the on_success node, depending on whether the
5090   // capture can be empty and whether it is known to be participating or known
5091   // not to be.
5092   return kComputeFirstCharacterSetFail;
5093 }
5094 
5095 
ComputeFirstCharacterSet(int budget)5096 int TextNode::ComputeFirstCharacterSet(int budget) {
5097   budget--;
5098   if (budget >= 0) {
5099     ASSERT_NE(0, elements()->length());
5100     TextElement text = elements()->at(0);
5101     if (text.type == TextElement::ATOM) {
5102       RegExpAtom* atom = text.data.u_atom;
5103       ASSERT_NE(0, atom->length());
5104       uc16 first_char = atom->data()[0];
5105       ZoneList<CharacterRange>* range = new ZoneList<CharacterRange>(1);
5106       range->Add(CharacterRange(first_char, first_char));
5107       set_first_character_set(range);
5108     } else {
5109       ASSERT(text.type == TextElement::CHAR_CLASS);
5110       RegExpCharacterClass* char_class = text.data.u_char_class;
5111       ZoneList<CharacterRange>* ranges = char_class->ranges();
5112       // TODO(lrn): Canonicalize ranges when they are created
5113       // instead of waiting until now.
5114       CharacterRange::Canonicalize(ranges);
5115       if (char_class->is_negated()) {
5116         int length = ranges->length();
5117         int new_length = length + 1;
5118         if (length > 0) {
5119           if (ranges->at(0).from() == 0) new_length--;
5120           if (ranges->at(length - 1).to() == String::kMaxUtf16CodeUnit) {
5121             new_length--;
5122           }
5123         }
5124         ZoneList<CharacterRange>* negated_ranges =
5125             new ZoneList<CharacterRange>(new_length);
5126         CharacterRange::Negate(ranges, negated_ranges);
5127         set_first_character_set(negated_ranges);
5128       } else {
5129         set_first_character_set(ranges);
5130       }
5131     }
5132   }
5133   return budget;
5134 }
5135 
5136 
5137 
5138 // -------------------------------------------------------------------
5139 // Dispatch table construction
5140 
5141 
VisitEnd(EndNode * that)5142 void DispatchTableConstructor::VisitEnd(EndNode* that) {
5143   AddRange(CharacterRange::Everything());
5144 }
5145 
5146 
BuildTable(ChoiceNode * node)5147 void DispatchTableConstructor::BuildTable(ChoiceNode* node) {
5148   node->set_being_calculated(true);
5149   ZoneList<GuardedAlternative>* alternatives = node->alternatives();
5150   for (int i = 0; i < alternatives->length(); i++) {
5151     set_choice_index(i);
5152     alternatives->at(i).node()->Accept(this);
5153   }
5154   node->set_being_calculated(false);
5155 }
5156 
5157 
5158 class AddDispatchRange {
5159  public:
AddDispatchRange(DispatchTableConstructor * constructor)5160   explicit AddDispatchRange(DispatchTableConstructor* constructor)
5161     : constructor_(constructor) { }
5162   void Call(uc32 from, DispatchTable::Entry entry);
5163  private:
5164   DispatchTableConstructor* constructor_;
5165 };
5166 
5167 
Call(uc32 from,DispatchTable::Entry entry)5168 void AddDispatchRange::Call(uc32 from, DispatchTable::Entry entry) {
5169   CharacterRange range(from, entry.to());
5170   constructor_->AddRange(range);
5171 }
5172 
5173 
VisitChoice(ChoiceNode * node)5174 void DispatchTableConstructor::VisitChoice(ChoiceNode* node) {
5175   if (node->being_calculated())
5176     return;
5177   DispatchTable* table = node->GetTable(ignore_case_);
5178   AddDispatchRange adder(this);
5179   table->ForEach(&adder);
5180 }
5181 
5182 
VisitBackReference(BackReferenceNode * that)5183 void DispatchTableConstructor::VisitBackReference(BackReferenceNode* that) {
5184   // TODO(160): Find the node that we refer back to and propagate its start
5185   // set back to here.  For now we just accept anything.
5186   AddRange(CharacterRange::Everything());
5187 }
5188 
5189 
VisitAssertion(AssertionNode * that)5190 void DispatchTableConstructor::VisitAssertion(AssertionNode* that) {
5191   RegExpNode* target = that->on_success();
5192   target->Accept(this);
5193 }
5194 
5195 
CompareRangeByFrom(const CharacterRange * a,const CharacterRange * b)5196 static int CompareRangeByFrom(const CharacterRange* a,
5197                               const CharacterRange* b) {
5198   return Compare<uc16>(a->from(), b->from());
5199 }
5200 
5201 
AddInverse(ZoneList<CharacterRange> * ranges)5202 void DispatchTableConstructor::AddInverse(ZoneList<CharacterRange>* ranges) {
5203   ranges->Sort(CompareRangeByFrom);
5204   uc16 last = 0;
5205   for (int i = 0; i < ranges->length(); i++) {
5206     CharacterRange range = ranges->at(i);
5207     if (last < range.from())
5208       AddRange(CharacterRange(last, range.from() - 1));
5209     if (range.to() >= last) {
5210       if (range.to() == String::kMaxUtf16CodeUnit) {
5211         return;
5212       } else {
5213         last = range.to() + 1;
5214       }
5215     }
5216   }
5217   AddRange(CharacterRange(last, String::kMaxUtf16CodeUnit));
5218 }
5219 
5220 
VisitText(TextNode * that)5221 void DispatchTableConstructor::VisitText(TextNode* that) {
5222   TextElement elm = that->elements()->at(0);
5223   switch (elm.type) {
5224     case TextElement::ATOM: {
5225       uc16 c = elm.data.u_atom->data()[0];
5226       AddRange(CharacterRange(c, c));
5227       break;
5228     }
5229     case TextElement::CHAR_CLASS: {
5230       RegExpCharacterClass* tree = elm.data.u_char_class;
5231       ZoneList<CharacterRange>* ranges = tree->ranges();
5232       if (tree->is_negated()) {
5233         AddInverse(ranges);
5234       } else {
5235         for (int i = 0; i < ranges->length(); i++)
5236           AddRange(ranges->at(i));
5237       }
5238       break;
5239     }
5240     default: {
5241       UNIMPLEMENTED();
5242     }
5243   }
5244 }
5245 
5246 
VisitAction(ActionNode * that)5247 void DispatchTableConstructor::VisitAction(ActionNode* that) {
5248   RegExpNode* target = that->on_success();
5249   target->Accept(this);
5250 }
5251 
5252 
Compile(RegExpCompileData * data,bool ignore_case,bool is_multiline,Handle<String> pattern,bool is_ascii)5253 RegExpEngine::CompilationResult RegExpEngine::Compile(RegExpCompileData* data,
5254                                                       bool ignore_case,
5255                                                       bool is_multiline,
5256                                                       Handle<String> pattern,
5257                                                       bool is_ascii) {
5258   if ((data->capture_count + 1) * 2 - 1 > RegExpMacroAssembler::kMaxRegister) {
5259     return IrregexpRegExpTooBig();
5260   }
5261   RegExpCompiler compiler(data->capture_count, ignore_case, is_ascii);
5262   // Wrap the body of the regexp in capture #0.
5263   RegExpNode* captured_body = RegExpCapture::ToNode(data->tree,
5264                                                     0,
5265                                                     &compiler,
5266                                                     compiler.accept());
5267   RegExpNode* node = captured_body;
5268   bool is_end_anchored = data->tree->IsAnchoredAtEnd();
5269   bool is_start_anchored = data->tree->IsAnchoredAtStart();
5270   int max_length = data->tree->max_match();
5271   if (!is_start_anchored) {
5272     // Add a .*? at the beginning, outside the body capture, unless
5273     // this expression is anchored at the beginning.
5274     RegExpNode* loop_node =
5275         RegExpQuantifier::ToNode(0,
5276                                  RegExpTree::kInfinity,
5277                                  false,
5278                                  new RegExpCharacterClass('*'),
5279                                  &compiler,
5280                                  captured_body,
5281                                  data->contains_anchor);
5282 
5283     if (data->contains_anchor) {
5284       // Unroll loop once, to take care of the case that might start
5285       // at the start of input.
5286       ChoiceNode* first_step_node = new ChoiceNode(2);
5287       first_step_node->AddAlternative(GuardedAlternative(captured_body));
5288       first_step_node->AddAlternative(GuardedAlternative(
5289           new TextNode(new RegExpCharacterClass('*'), loop_node)));
5290       node = first_step_node;
5291     } else {
5292       node = loop_node;
5293     }
5294   }
5295   data->node = node;
5296   Analysis analysis(ignore_case, is_ascii);
5297   analysis.EnsureAnalyzed(node);
5298   if (analysis.has_failed()) {
5299     const char* error_message = analysis.error_message();
5300     return CompilationResult(error_message);
5301   }
5302 
5303   // Create the correct assembler for the architecture.
5304 #ifndef V8_INTERPRETED_REGEXP
5305   // Native regexp implementation.
5306 
5307   NativeRegExpMacroAssembler::Mode mode =
5308       is_ascii ? NativeRegExpMacroAssembler::ASCII
5309                : NativeRegExpMacroAssembler::UC16;
5310 
5311 #if V8_TARGET_ARCH_IA32
5312   RegExpMacroAssemblerIA32 macro_assembler(mode, (data->capture_count + 1) * 2);
5313 #elif V8_TARGET_ARCH_X64
5314   RegExpMacroAssemblerX64 macro_assembler(mode, (data->capture_count + 1) * 2);
5315 #elif V8_TARGET_ARCH_ARM
5316   RegExpMacroAssemblerARM macro_assembler(mode, (data->capture_count + 1) * 2);
5317 #elif V8_TARGET_ARCH_MIPS
5318   RegExpMacroAssemblerMIPS macro_assembler(mode, (data->capture_count + 1) * 2);
5319 #endif
5320 
5321 #else  // V8_INTERPRETED_REGEXP
5322   // Interpreted regexp implementation.
5323   EmbeddedVector<byte, 1024> codes;
5324   RegExpMacroAssemblerIrregexp macro_assembler(codes);
5325 #endif  // V8_INTERPRETED_REGEXP
5326 
5327   // Inserted here, instead of in Assembler, because it depends on information
5328   // in the AST that isn't replicated in the Node structure.
5329   static const int kMaxBacksearchLimit = 1024;
5330   if (is_end_anchored &&
5331       !is_start_anchored &&
5332       max_length < kMaxBacksearchLimit) {
5333     macro_assembler.SetCurrentPositionFromEnd(max_length);
5334   }
5335 
5336   return compiler.Assemble(&macro_assembler,
5337                            node,
5338                            data->capture_count,
5339                            pattern);
5340 }
5341 
5342 
5343 }}  // namespace v8::internal
5344