1
// Copyright 2012 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
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.
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.
32
#include "execution.h"
36
#include "string-search.h"
38
#include "compilation-cache.h"
39
#include "string-stream.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"
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"
56
#error Unsupported target architecture.
60
#include "interpreter-irregexp.h"
66
Handle<Object> RegExpImpl::CreateRegExpLiteral(Handle<JSFunction> constructor,
67
Handle<String> pattern,
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);
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)) {
82
flags |= JSRegExp::IGNORE_CASE;
85
flags |= JSRegExp::GLOBAL;
88
flags |= JSRegExp::MULTILINE;
92
return JSRegExp::Flags(flags);
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);
111
ContainedInLattice AddRange(ContainedInLattice containment,
114
Interval new_range) {
115
ASSERT((ranges_length & 1) == 1);
116
ASSERT(ranges[ranges_length - 1] == String::kMaxUtf16CodeUnit + 1);
117
if (containment == kLatticeUnknown) return containment;
120
for (int i = 0; i < ranges_length; inside = !inside, last = ranges[i], i++) {
121
// Consider the range from last to ranges[i].
122
// We haven't got to the new range yet.
123
if (ranges[i] <= new_range.from()) continue;
124
// New range is wholly inside last-ranges[i]. Note that new_range.to() is
125
// inclusive, but the values in ranges are not.
126
if (last <= new_range.from() && new_range.to() < ranges[i]) {
127
return Combine(containment, inside ? kLatticeIn : kLatticeOut);
129
return kLatticeUnknown;
135
// More makes code generation slower, less makes V8 benchmark score lower.
136
const int kMaxLookaheadForBoyerMoore = 8;
137
// In a 3-character pattern you can maximally step forwards 3 characters
138
// at a time, which is not always enough to pay for the extra logic.
139
const int kPatternTooShortForBoyerMoore = 2;
142
// Identifies the sort of regexps where the regexp engine is faster
143
// than the code used for atom matches.
144
static bool HasFewDifferentCharacters(Handle<String> pattern) {
145
int length = Min(kMaxLookaheadForBoyerMoore, pattern->length());
146
if (length <= kPatternTooShortForBoyerMoore) return false;
147
const int kMod = 128;
148
bool character_found[kMod];
150
memset(&character_found[0], 0, sizeof(character_found));
151
for (int i = 0; i < length; i++) {
152
int ch = (pattern->Get(i) & (kMod - 1));
153
if (!character_found[ch]) {
154
character_found[ch] = true;
156
// We declare a regexp low-alphabet if it has at least 3 times as many
157
// characters as it has different characters.
158
if (different * 3 > length) return false;
165
// Generic RegExp methods. Dispatches to implementation specific methods.
168
Handle<Object> RegExpImpl::Compile(Handle<JSRegExp> re,
169
Handle<String> pattern,
170
Handle<String> flag_str,
172
ZoneScope zone_scope(zone, DELETE_ON_EXIT);
173
Isolate* isolate = re->GetIsolate();
174
JSRegExp::Flags flags = RegExpFlagsFromString(flag_str);
175
CompilationCache* compilation_cache = isolate->compilation_cache();
176
Handle<FixedArray> cached = compilation_cache->LookupRegExp(pattern, flags);
177
bool in_cache = !cached.is_null();
178
LOG(isolate, RegExpCompileEvent(re, in_cache));
180
Handle<Object> result;
182
re->set_data(*cached);
185
pattern = FlattenGetString(pattern);
186
PostponeInterruptsScope postpone(isolate);
187
RegExpCompileData parse_result;
188
FlatStringReader reader(isolate, pattern);
189
if (!RegExpParser::ParseRegExp(&reader, flags.is_multiline(),
190
&parse_result, zone)) {
191
// Throw an exception if we fail to parse the pattern.
192
ThrowRegExpException(re,
196
return Handle<Object>::null();
199
bool has_been_compiled = false;
201
if (parse_result.simple &&
202
!flags.is_ignore_case() &&
203
!HasFewDifferentCharacters(pattern)) {
204
// Parse-tree is a single atom that is equal to the pattern.
205
AtomCompile(re, pattern, flags, pattern);
206
has_been_compiled = true;
207
} else if (parse_result.tree->IsAtom() &&
208
!flags.is_ignore_case() &&
209
parse_result.capture_count == 0) {
210
RegExpAtom* atom = parse_result.tree->AsAtom();
211
Vector<const uc16> atom_pattern = atom->data();
212
Handle<String> atom_string =
213
isolate->factory()->NewStringFromTwoByte(atom_pattern);
214
if (!HasFewDifferentCharacters(atom_string)) {
215
AtomCompile(re, pattern, flags, atom_string);
216
has_been_compiled = true;
219
if (!has_been_compiled) {
220
IrregexpInitialize(re, pattern, flags, parse_result.capture_count);
222
ASSERT(re->data()->IsFixedArray());
223
// Compilation succeeded so the data is set on the regexp
224
// and we can store it in the cache.
225
Handle<FixedArray> data(FixedArray::cast(re->data()));
226
compilation_cache->PutRegExp(pattern, flags, data);
232
Handle<Object> RegExpImpl::Exec(Handle<JSRegExp> regexp,
233
Handle<String> subject,
235
Handle<JSArray> last_match_info) {
236
switch (regexp->TypeTag()) {
238
return AtomExec(regexp, subject, index, last_match_info);
239
case JSRegExp::IRREGEXP: {
240
Handle<Object> result =
241
IrregexpExec(regexp, subject, index, last_match_info);
242
ASSERT(!result.is_null() ||
243
regexp->GetIsolate()->has_pending_exception());
248
return Handle<Object>::null();
253
// RegExp Atom implementation: Simple string search using indexOf.
256
void RegExpImpl::AtomCompile(Handle<JSRegExp> re,
257
Handle<String> pattern,
258
JSRegExp::Flags flags,
259
Handle<String> match_pattern) {
260
re->GetIsolate()->factory()->SetRegExpAtomData(re,
268
static void SetAtomLastCapture(FixedArray* array,
272
NoHandleAllocation no_handles;
273
RegExpImpl::SetLastCaptureCount(array, 2);
274
RegExpImpl::SetLastSubject(array, subject);
275
RegExpImpl::SetLastInput(array, subject);
276
RegExpImpl::SetCapture(array, 0, from);
277
RegExpImpl::SetCapture(array, 1, to);
281
Handle<Object> RegExpImpl::AtomExec(Handle<JSRegExp> re,
282
Handle<String> subject,
284
Handle<JSArray> last_match_info) {
285
Isolate* isolate = re->GetIsolate();
288
ASSERT(index <= subject->length());
290
if (!subject->IsFlat()) FlattenString(subject);
291
AssertNoAllocation no_heap_allocation; // ensure vectors stay valid
293
String* needle = String::cast(re->DataAt(JSRegExp::kAtomPatternIndex));
294
int needle_len = needle->length();
295
ASSERT(needle->IsFlat());
297
if (needle_len != 0) {
298
if (index + needle_len > subject->length()) {
299
return isolate->factory()->null_value();
302
String::FlatContent needle_content = needle->GetFlatContent();
303
String::FlatContent subject_content = subject->GetFlatContent();
304
ASSERT(needle_content.IsFlat());
305
ASSERT(subject_content.IsFlat());
306
// dispatch on type of strings
307
index = (needle_content.IsAscii()
308
? (subject_content.IsAscii()
309
? SearchString(isolate,
310
subject_content.ToAsciiVector(),
311
needle_content.ToAsciiVector(),
313
: SearchString(isolate,
314
subject_content.ToUC16Vector(),
315
needle_content.ToAsciiVector(),
317
: (subject_content.IsAscii()
318
? SearchString(isolate,
319
subject_content.ToAsciiVector(),
320
needle_content.ToUC16Vector(),
322
: SearchString(isolate,
323
subject_content.ToUC16Vector(),
324
needle_content.ToUC16Vector(),
326
if (index == -1) return isolate->factory()->null_value();
328
ASSERT(last_match_info->HasFastObjectElements());
331
NoHandleAllocation no_handles;
332
FixedArray* array = FixedArray::cast(last_match_info->elements());
333
SetAtomLastCapture(array, *subject, index, index + needle_len);
335
return last_match_info;
339
// Irregexp implementation.
341
// Ensures that the regexp object contains a compiled version of the
342
// source for either ASCII or non-ASCII strings.
343
// If the compiled version doesn't already exist, it is compiled
344
// from the source pattern.
345
// If compilation fails, an exception is thrown and this function
347
bool RegExpImpl::EnsureCompiledIrregexp(
348
Handle<JSRegExp> re, Handle<String> sample_subject, bool is_ascii) {
349
Object* compiled_code = re->DataAt(JSRegExp::code_index(is_ascii));
350
#ifdef V8_INTERPRETED_REGEXP
351
if (compiled_code->IsByteArray()) return true;
352
#else // V8_INTERPRETED_REGEXP (RegExp native code)
353
if (compiled_code->IsCode()) return true;
355
// We could potentially have marked this as flushable, but have kept
356
// a saved version if we did not flush it yet.
357
Object* saved_code = re->DataAt(JSRegExp::saved_code_index(is_ascii));
358
if (saved_code->IsCode()) {
359
// Reinstate the code in the original place.
360
re->SetDataAt(JSRegExp::code_index(is_ascii), saved_code);
361
ASSERT(compiled_code->IsSmi());
364
return CompileIrregexp(re, sample_subject, is_ascii);
368
static bool CreateRegExpErrorObjectAndThrow(Handle<JSRegExp> re,
370
Handle<String> error_message,
372
Factory* factory = isolate->factory();
373
Handle<FixedArray> elements = factory->NewFixedArray(2);
374
elements->set(0, re->Pattern());
375
elements->set(1, *error_message);
376
Handle<JSArray> array = factory->NewJSArrayWithElements(elements);
377
Handle<Object> regexp_err =
378
factory->NewSyntaxError("malformed_regexp", array);
379
isolate->Throw(*regexp_err);
384
bool RegExpImpl::CompileIrregexp(Handle<JSRegExp> re,
385
Handle<String> sample_subject,
387
// Compile the RegExp.
388
Isolate* isolate = re->GetIsolate();
389
ZoneScope zone_scope(isolate->runtime_zone(), DELETE_ON_EXIT);
390
PostponeInterruptsScope postpone(isolate);
391
// If we had a compilation error the last time this is saved at the
393
Object* entry = re->DataAt(JSRegExp::code_index(is_ascii));
394
// When arriving here entry can only be a smi, either representing an
395
// uncompiled regexp, a previous compilation error, or code that has
397
ASSERT(entry->IsSmi());
398
int entry_value = Smi::cast(entry)->value();
399
ASSERT(entry_value == JSRegExp::kUninitializedValue ||
400
entry_value == JSRegExp::kCompilationErrorValue ||
401
(entry_value < JSRegExp::kCodeAgeMask && entry_value >= 0));
403
if (entry_value == JSRegExp::kCompilationErrorValue) {
404
// A previous compilation failed and threw an error which we store in
405
// the saved code index (we store the error message, not the actual
406
// error). Recreate the error object and throw it.
407
Object* error_string = re->DataAt(JSRegExp::saved_code_index(is_ascii));
408
ASSERT(error_string->IsString());
409
Handle<String> error_message(String::cast(error_string));
410
CreateRegExpErrorObjectAndThrow(re, is_ascii, error_message, isolate);
414
JSRegExp::Flags flags = re->GetFlags();
416
Handle<String> pattern(re->Pattern());
417
if (!pattern->IsFlat()) FlattenString(pattern);
418
RegExpCompileData compile_data;
419
FlatStringReader reader(isolate, pattern);
420
Zone* zone = isolate->runtime_zone();
421
if (!RegExpParser::ParseRegExp(&reader, flags.is_multiline(),
424
// Throw an exception if we fail to parse the pattern.
425
// THIS SHOULD NOT HAPPEN. We already pre-parsed it successfully once.
426
ThrowRegExpException(re,
432
RegExpEngine::CompilationResult result =
433
RegExpEngine::Compile(&compile_data,
434
flags.is_ignore_case(),
436
flags.is_multiline(),
441
if (result.error_message != NULL) {
442
// Unable to compile regexp.
443
Handle<String> error_message =
444
isolate->factory()->NewStringFromUtf8(CStrVector(result.error_message));
445
CreateRegExpErrorObjectAndThrow(re, is_ascii, error_message, isolate);
449
Handle<FixedArray> data = Handle<FixedArray>(FixedArray::cast(re->data()));
450
data->set(JSRegExp::code_index(is_ascii), result.code);
451
int register_max = IrregexpMaxRegisterCount(*data);
452
if (result.num_registers > register_max) {
453
SetIrregexpMaxRegisterCount(*data, result.num_registers);
460
int RegExpImpl::IrregexpMaxRegisterCount(FixedArray* re) {
462
re->get(JSRegExp::kIrregexpMaxRegisterCountIndex))->value();
466
void RegExpImpl::SetIrregexpMaxRegisterCount(FixedArray* re, int value) {
467
re->set(JSRegExp::kIrregexpMaxRegisterCountIndex, Smi::FromInt(value));
471
int RegExpImpl::IrregexpNumberOfCaptures(FixedArray* re) {
472
return Smi::cast(re->get(JSRegExp::kIrregexpCaptureCountIndex))->value();
476
int RegExpImpl::IrregexpNumberOfRegisters(FixedArray* re) {
477
return Smi::cast(re->get(JSRegExp::kIrregexpMaxRegisterCountIndex))->value();
481
ByteArray* RegExpImpl::IrregexpByteCode(FixedArray* re, bool is_ascii) {
482
return ByteArray::cast(re->get(JSRegExp::code_index(is_ascii)));
486
Code* RegExpImpl::IrregexpNativeCode(FixedArray* re, bool is_ascii) {
487
return Code::cast(re->get(JSRegExp::code_index(is_ascii)));
491
void RegExpImpl::IrregexpInitialize(Handle<JSRegExp> re,
492
Handle<String> pattern,
493
JSRegExp::Flags flags,
495
// Initialize compiled code entries to null.
496
re->GetIsolate()->factory()->SetRegExpIrregexpData(re,
504
int RegExpImpl::IrregexpPrepare(Handle<JSRegExp> regexp,
505
Handle<String> subject) {
506
if (!subject->IsFlat()) FlattenString(subject);
508
// Check the asciiness of the underlying storage.
509
bool is_ascii = subject->IsAsciiRepresentationUnderneath();
510
if (!EnsureCompiledIrregexp(regexp, subject, is_ascii)) return -1;
512
#ifdef V8_INTERPRETED_REGEXP
513
// Byte-code regexp needs space allocated for all its registers.
514
return IrregexpNumberOfRegisters(FixedArray::cast(regexp->data()));
515
#else // V8_INTERPRETED_REGEXP
516
// Native regexp only needs room to output captures. Registers are handled
518
return (IrregexpNumberOfCaptures(FixedArray::cast(regexp->data())) + 1) * 2;
519
#endif // V8_INTERPRETED_REGEXP
523
int RegExpImpl::GlobalOffsetsVectorSize(Handle<JSRegExp> regexp,
524
int registers_per_match,
526
#ifdef V8_INTERPRETED_REGEXP
527
// Global loop in interpreted regexp is not implemented. Therefore we choose
528
// the size of the offsets vector so that it can only store one match.
530
return registers_per_match;
531
#else // V8_INTERPRETED_REGEXP
532
int size = Max(registers_per_match, OffsetsVector::kStaticOffsetsVectorSize);
533
*max_matches = size / registers_per_match;
535
#endif // V8_INTERPRETED_REGEXP
539
int RegExpImpl::IrregexpExecRaw(
540
Handle<JSRegExp> regexp,
541
Handle<String> subject,
543
Vector<int> output) {
544
Isolate* isolate = regexp->GetIsolate();
546
Handle<FixedArray> irregexp(FixedArray::cast(regexp->data()), isolate);
549
ASSERT(index <= subject->length());
550
ASSERT(subject->IsFlat());
552
bool is_ascii = subject->IsAsciiRepresentationUnderneath();
554
#ifndef V8_INTERPRETED_REGEXP
555
ASSERT(output.length() >= (IrregexpNumberOfCaptures(*irregexp) + 1) * 2);
557
EnsureCompiledIrregexp(regexp, subject, is_ascii);
558
Handle<Code> code(IrregexpNativeCode(*irregexp, is_ascii), isolate);
559
NativeRegExpMacroAssembler::Result res =
560
NativeRegExpMacroAssembler::Match(code,
566
if (res != NativeRegExpMacroAssembler::RETRY) {
567
ASSERT(res != NativeRegExpMacroAssembler::EXCEPTION ||
568
isolate->has_pending_exception());
570
static_cast<int>(NativeRegExpMacroAssembler::SUCCESS) == RE_SUCCESS);
572
static_cast<int>(NativeRegExpMacroAssembler::FAILURE) == RE_FAILURE);
573
STATIC_ASSERT(static_cast<int>(NativeRegExpMacroAssembler::EXCEPTION)
575
return static_cast<IrregexpResult>(res);
577
// If result is RETRY, the string has changed representation, and we
578
// must restart from scratch.
579
// In this case, it means we must make sure we are prepared to handle
580
// the, potentially, different subject (the string can switch between
581
// being internal and external, and even between being ASCII and UC16,
582
// but the characters are always the same).
583
IrregexpPrepare(regexp, subject);
584
is_ascii = subject->IsAsciiRepresentationUnderneath();
588
#else // V8_INTERPRETED_REGEXP
590
ASSERT(output.length() >= IrregexpNumberOfRegisters(*irregexp));
591
// We must have done EnsureCompiledIrregexp, so we can get the number of
593
int* register_vector = output.start();
594
int number_of_capture_registers =
595
(IrregexpNumberOfCaptures(*irregexp) + 1) * 2;
596
for (int i = number_of_capture_registers - 1; i >= 0; i--) {
597
register_vector[i] = -1;
599
Handle<ByteArray> byte_codes(IrregexpByteCode(*irregexp, is_ascii), isolate);
601
IrregexpResult result = IrregexpInterpreter::Match(isolate,
606
if (result == RE_EXCEPTION) {
607
ASSERT(!isolate->has_pending_exception());
608
isolate->StackOverflow();
611
#endif // V8_INTERPRETED_REGEXP
615
Handle<Object> RegExpImpl::IrregexpExec(Handle<JSRegExp> jsregexp,
616
Handle<String> subject,
618
Handle<JSArray> last_match_info) {
619
Isolate* isolate = jsregexp->GetIsolate();
620
ASSERT_EQ(jsregexp->TypeTag(), JSRegExp::IRREGEXP);
622
// Prepare space for the return values.
623
#ifdef V8_INTERPRETED_REGEXP
625
if (FLAG_trace_regexp_bytecodes) {
626
String* pattern = jsregexp->Pattern();
627
PrintF("\n\nRegexp match: /%s/\n\n", *(pattern->ToCString()));
628
PrintF("\n\nSubject string: '%s'\n\n", *(subject->ToCString()));
632
int required_registers = RegExpImpl::IrregexpPrepare(jsregexp, subject);
633
if (required_registers < 0) {
634
// Compiling failed with an exception.
635
ASSERT(isolate->has_pending_exception());
636
return Handle<Object>::null();
639
OffsetsVector registers(required_registers, isolate);
641
int res = RegExpImpl::IrregexpExecRaw(jsregexp, subject, previous_index,
642
Vector<int>(registers.vector(),
643
registers.length()));
644
if (res == RE_SUCCESS) {
645
int capture_register_count =
646
(IrregexpNumberOfCaptures(FixedArray::cast(jsregexp->data())) + 1) * 2;
647
last_match_info->EnsureSize(capture_register_count + kLastMatchOverhead);
648
AssertNoAllocation no_gc;
649
int* register_vector = registers.vector();
650
FixedArray* array = FixedArray::cast(last_match_info->elements());
651
for (int i = 0; i < capture_register_count; i += 2) {
652
SetCapture(array, i, register_vector[i]);
653
SetCapture(array, i + 1, register_vector[i + 1]);
655
SetLastCaptureCount(array, capture_register_count);
656
SetLastSubject(array, *subject);
657
SetLastInput(array, *subject);
658
return last_match_info;
660
if (res == RE_EXCEPTION) {
661
ASSERT(isolate->has_pending_exception());
662
return Handle<Object>::null();
664
ASSERT(res == RE_FAILURE);
665
return isolate->factory()->null_value();
669
// -------------------------------------------------------------------
670
// Implementation of the Irregexp regular expression engine.
672
// The Irregexp regular expression engine is intended to be a complete
673
// implementation of ECMAScript regular expressions. It generates either
674
// bytecodes or native code.
676
// The Irregexp regexp engine is structured in three steps.
677
// 1) The parser generates an abstract syntax tree. See ast.cc.
678
// 2) From the AST a node network is created. The nodes are all
679
// subclasses of RegExpNode. The nodes represent states when
680
// executing a regular expression. Several optimizations are
681
// performed on the node network.
682
// 3) From the nodes we generate either byte codes or native code
683
// that can actually execute the regular expression (perform
684
// the search). The code generation step is described in more
689
// The nodes are divided into four main categories.
691
// These represent places where the regular expression can
692
// match in more than one way. For example on entry to an
693
// alternation (foo|bar) or a repetition (*, +, ? or {}).
695
// These represent places where some action should be
696
// performed. Examples include recording the current position
697
// in the input string to a register (in order to implement
698
// captures) or other actions on register for example in order
699
// to implement the counters needed for {} repetitions.
701
// These attempt to match some element part of the input string.
702
// Examples of elements include character classes, plain strings
703
// or back references.
705
// These are used to implement the actions required on finding
706
// a successful match or failing to find a match.
708
// The code generated (whether as byte codes or native code) maintains
709
// some state as it runs. This consists of the following elements:
711
// * The capture registers. Used for string captures.
712
// * Other registers. Used for counters etc.
713
// * The current position.
714
// * The stack of backtracking information. Used when a matching node
715
// fails to find a match and needs to try an alternative.
717
// Conceptual regular expression execution model:
719
// There is a simple conceptual model of regular expression execution
720
// which will be presented first. The actual code generated is a more
721
// efficient simulation of the simple conceptual model:
723
// * Choice nodes are implemented as follows:
724
// For each choice except the last {
725
// push current position
726
// push backtrack code location
727
// <generate code to test for choice>
728
// backtrack code location:
729
// pop current position
731
// <generate code to test for last choice>
733
// * Actions nodes are generated as follows
734
// <push affected registers on backtrack stack>
735
// <generate code to perform action>
736
// push backtrack code location
737
// <generate code to test for following nodes>
738
// backtrack code location:
739
// <pop affected registers to restore their state>
740
// <pop backtrack location from stack and go to it>
742
// * Matching nodes are generated as follows:
743
// if input string matches at current position
744
// update current position
745
// <generate code to test for following nodes>
747
// <pop backtrack location from stack and go to it>
749
// Thus it can be seen that the current position is saved and restored
750
// by the choice nodes, whereas the registers are saved and restored by
751
// by the action nodes that manipulate them.
753
// The other interesting aspect of this model is that nodes are generated
754
// at the point where they are needed by a recursive call to Emit(). If
755
// the node has already been code generated then the Emit() call will
756
// generate a jump to the previously generated code instead. In order to
757
// limit recursion it is possible for the Emit() function to put the node
758
// on a work list for later generation and instead generate a jump. The
759
// destination of the jump is resolved later when the code is generated.
761
// Actual regular expression code generation.
763
// Code generation is actually more complicated than the above. In order
764
// to improve the efficiency of the generated code some optimizations are
767
// * Choice nodes have 1-character lookahead.
768
// A choice node looks at the following character and eliminates some of
769
// the choices immediately based on that character. This is not yet
771
// * Simple greedy loops store reduced backtracking information.
772
// A quantifier like /.*foo/m will greedily match the whole input. It will
773
// then need to backtrack to a point where it can match "foo". The naive
774
// implementation of this would push each character position onto the
775
// backtracking stack, then pop them off one by one. This would use space
776
// proportional to the length of the input string. However since the "."
777
// can only match in one way and always has a constant length (in this case
778
// of 1) it suffices to store the current position on the top of the stack
779
// once. Matching now becomes merely incrementing the current position and
780
// backtracking becomes decrementing the current position and checking the
781
// result against the stored current position. This is faster and saves
783
// * The current state is virtualized.
784
// This is used to defer expensive operations until it is clear that they
785
// are needed and to generate code for a node more than once, allowing
786
// specialized an efficient versions of the code to be created. This is
787
// explained in the section below.
789
// Execution state virtualization.
791
// Instead of emitting code, nodes that manipulate the state can record their
792
// manipulation in an object called the Trace. The Trace object can record a
793
// current position offset, an optional backtrack code location on the top of
794
// the virtualized backtrack stack and some register changes. When a node is
795
// to be emitted it can flush the Trace or update it. Flushing the Trace
796
// will emit code to bring the actual state into line with the virtual state.
797
// Avoiding flushing the state can postpone some work (e.g. updates of capture
798
// registers). Postponing work can save time when executing the regular
799
// expression since it may be found that the work never has to be done as a
800
// failure to match can occur. In addition it is much faster to jump to a
801
// known backtrack code location than it is to pop an unknown backtrack
802
// location from the stack and jump there.
804
// The virtual state found in the Trace affects code generation. For example
805
// the virtual state contains the difference between the actual current
806
// position and the virtual current position, and matching code needs to use
807
// this offset to attempt a match in the correct location of the input
808
// string. Therefore code generated for a non-trivial trace is specialized
809
// to that trace. The code generator therefore has the ability to generate
810
// code for each node several times. In order to limit the size of the
811
// generated code there is an arbitrary limit on how many specialized sets of
812
// code may be generated for a given node. If the limit is reached, the
813
// trace is flushed and a generic version of the code for a node is emitted.
814
// This is subsequently used for that node. The code emitted for non-generic
815
// trace is not recorded in the node and so it cannot currently be reused in
816
// the event that code generation is requested for an identical trace.
819
void RegExpTree::AppendToText(RegExpText* text, Zone* zone) {
824
void RegExpAtom::AppendToText(RegExpText* text, Zone* zone) {
825
text->AddElement(TextElement::Atom(this), zone);
829
void RegExpCharacterClass::AppendToText(RegExpText* text, Zone* zone) {
830
text->AddElement(TextElement::CharClass(this), zone);
834
void RegExpText::AppendToText(RegExpText* text, Zone* zone) {
835
for (int i = 0; i < elements()->length(); i++)
836
text->AddElement(elements()->at(i), zone);
840
TextElement TextElement::Atom(RegExpAtom* atom) {
841
TextElement result = TextElement(ATOM);
842
result.data.u_atom = atom;
847
TextElement TextElement::CharClass(
848
RegExpCharacterClass* char_class) {
849
TextElement result = TextElement(CHAR_CLASS);
850
result.data.u_char_class = char_class;
855
int TextElement::length() {
857
return data.u_atom->length();
859
ASSERT(type == CHAR_CLASS);
865
DispatchTable* ChoiceNode::GetTable(bool ignore_case) {
866
if (table_ == NULL) {
867
table_ = new(zone()) DispatchTable(zone());
868
DispatchTableConstructor cons(table_, ignore_case, zone());
869
cons.BuildTable(this);
875
class FrequencyCollator {
877
FrequencyCollator() : total_samples_(0) {
878
for (int i = 0; i < RegExpMacroAssembler::kTableSize; i++) {
879
frequencies_[i] = CharacterFrequency(i);
883
void CountCharacter(int character) {
884
int index = (character & RegExpMacroAssembler::kTableMask);
885
frequencies_[index].Increment();
889
// Does not measure in percent, but rather per-128 (the table size from the
890
// regexp macro assembler).
891
int Frequency(int in_character) {
892
ASSERT((in_character & RegExpMacroAssembler::kTableMask) == in_character);
893
if (total_samples_ < 1) return 1; // Division by zero.
895
(frequencies_[in_character].counter() * 128) / total_samples_;
896
return freq_in_per128;
900
class CharacterFrequency {
902
CharacterFrequency() : counter_(0), character_(-1) { }
903
explicit CharacterFrequency(int character)
904
: counter_(0), character_(character) { }
906
void Increment() { counter_++; }
907
int counter() { return counter_; }
908
int character() { return character_; }
917
CharacterFrequency frequencies_[RegExpMacroAssembler::kTableSize];
922
class RegExpCompiler {
924
RegExpCompiler(int capture_count, bool ignore_case, bool is_ascii,
927
int AllocateRegister() {
928
if (next_register_ >= RegExpMacroAssembler::kMaxRegister) {
929
reg_exp_too_big_ = true;
930
return next_register_;
932
return next_register_++;
935
RegExpEngine::CompilationResult Assemble(RegExpMacroAssembler* assembler,
938
Handle<String> pattern);
940
inline void AddWork(RegExpNode* node) { work_list_->Add(node); }
942
static const int kImplementationOffset = 0;
943
static const int kNumberOfRegistersOffset = 0;
944
static const int kCodeOffset = 1;
946
RegExpMacroAssembler* macro_assembler() { return macro_assembler_; }
947
EndNode* accept() { return accept_; }
949
static const int kMaxRecursion = 100;
950
inline int recursion_depth() { return recursion_depth_; }
951
inline void IncrementRecursionDepth() { recursion_depth_++; }
952
inline void DecrementRecursionDepth() { recursion_depth_--; }
954
void SetRegExpTooBig() { reg_exp_too_big_ = true; }
956
inline bool ignore_case() { return ignore_case_; }
957
inline bool ascii() { return ascii_; }
958
FrequencyCollator* frequency_collator() { return &frequency_collator_; }
960
int current_expansion_factor() { return current_expansion_factor_; }
961
void set_current_expansion_factor(int value) {
962
current_expansion_factor_ = value;
965
Zone* zone() const { return zone_; }
967
static const int kNoRegister = -1;
972
List<RegExpNode*>* work_list_;
973
int recursion_depth_;
974
RegExpMacroAssembler* macro_assembler_;
977
bool reg_exp_too_big_;
978
int current_expansion_factor_;
979
FrequencyCollator frequency_collator_;
984
class RecursionCheck {
986
explicit RecursionCheck(RegExpCompiler* compiler) : compiler_(compiler) {
987
compiler->IncrementRecursionDepth();
989
~RecursionCheck() { compiler_->DecrementRecursionDepth(); }
991
RegExpCompiler* compiler_;
995
static RegExpEngine::CompilationResult IrregexpRegExpTooBig() {
996
return RegExpEngine::CompilationResult("RegExp too big");
1000
// Attempts to compile the regexp using an Irregexp code generator. Returns
1001
// a fixed array or a null handle depending on whether it succeeded.
1002
RegExpCompiler::RegExpCompiler(int capture_count, bool ignore_case, bool ascii,
1004
: next_register_(2 * (capture_count + 1)),
1006
recursion_depth_(0),
1007
ignore_case_(ignore_case),
1009
reg_exp_too_big_(false),
1010
current_expansion_factor_(1),
1011
frequency_collator_(),
1013
accept_ = new(zone) EndNode(EndNode::ACCEPT, zone);
1014
ASSERT(next_register_ - 1 <= RegExpMacroAssembler::kMaxRegister);
1018
RegExpEngine::CompilationResult RegExpCompiler::Assemble(
1019
RegExpMacroAssembler* macro_assembler,
1022
Handle<String> pattern) {
1023
Heap* heap = pattern->GetHeap();
1025
bool use_slow_safe_regexp_compiler = false;
1026
if (heap->total_regexp_code_generated() >
1027
RegExpImpl::kRegWxpCompiledLimit &&
1028
heap->isolate()->memory_allocator()->SizeExecutable() >
1029
RegExpImpl::kRegExpExecutableMemoryLimit) {
1030
use_slow_safe_regexp_compiler = true;
1033
macro_assembler->set_slow_safe(use_slow_safe_regexp_compiler);
1036
if (FLAG_trace_regexp_assembler)
1037
macro_assembler_ = new RegExpMacroAssemblerTracer(macro_assembler);
1040
macro_assembler_ = macro_assembler;
1042
List <RegExpNode*> work_list(0);
1043
work_list_ = &work_list;
1045
macro_assembler_->PushBacktrack(&fail);
1047
start->Emit(this, &new_trace);
1048
macro_assembler_->Bind(&fail);
1049
macro_assembler_->Fail();
1050
while (!work_list.is_empty()) {
1051
work_list.RemoveLast()->Emit(this, &new_trace);
1053
if (reg_exp_too_big_) return IrregexpRegExpTooBig();
1055
Handle<HeapObject> code = macro_assembler_->GetCode(pattern);
1056
heap->IncreaseTotalRegexpCodeGenerated(code->Size());
1059
if (FLAG_print_code) {
1060
Handle<Code>::cast(code)->Disassemble(*pattern->ToCString());
1062
if (FLAG_trace_regexp_assembler) {
1063
delete macro_assembler_;
1066
return RegExpEngine::CompilationResult(*code, next_register_);
1070
bool Trace::DeferredAction::Mentions(int that) {
1071
if (type() == ActionNode::CLEAR_CAPTURES) {
1072
Interval range = static_cast<DeferredClearCaptures*>(this)->range();
1073
return range.Contains(that);
1075
return reg() == that;
1080
bool Trace::mentions_reg(int reg) {
1081
for (DeferredAction* action = actions_;
1083
action = action->next()) {
1084
if (action->Mentions(reg))
1091
bool Trace::GetStoredPosition(int reg, int* cp_offset) {
1092
ASSERT_EQ(0, *cp_offset);
1093
for (DeferredAction* action = actions_;
1095
action = action->next()) {
1096
if (action->Mentions(reg)) {
1097
if (action->type() == ActionNode::STORE_POSITION) {
1098
*cp_offset = static_cast<DeferredCapture*>(action)->cp_offset();
1109
int Trace::FindAffectedRegisters(OutSet* affected_registers,
1111
int max_register = RegExpCompiler::kNoRegister;
1112
for (DeferredAction* action = actions_;
1114
action = action->next()) {
1115
if (action->type() == ActionNode::CLEAR_CAPTURES) {
1116
Interval range = static_cast<DeferredClearCaptures*>(action)->range();
1117
for (int i = range.from(); i <= range.to(); i++)
1118
affected_registers->Set(i, zone);
1119
if (range.to() > max_register) max_register = range.to();
1121
affected_registers->Set(action->reg(), zone);
1122
if (action->reg() > max_register) max_register = action->reg();
1125
return max_register;
1129
void Trace::RestoreAffectedRegisters(RegExpMacroAssembler* assembler,
1131
OutSet& registers_to_pop,
1132
OutSet& registers_to_clear) {
1133
for (int reg = max_register; reg >= 0; reg--) {
1134
if (registers_to_pop.Get(reg)) assembler->PopRegister(reg);
1135
else if (registers_to_clear.Get(reg)) {
1137
while (reg > 0 && registers_to_clear.Get(reg - 1)) {
1140
assembler->ClearRegisters(reg, clear_to);
1146
void Trace::PerformDeferredActions(RegExpMacroAssembler* assembler,
1148
OutSet& affected_registers,
1149
OutSet* registers_to_pop,
1150
OutSet* registers_to_clear,
1152
// The "+1" is to avoid a push_limit of zero if stack_limit_slack() is 1.
1153
const int push_limit = (assembler->stack_limit_slack() + 1) / 2;
1155
// Count pushes performed to force a stack limit check occasionally.
1158
for (int reg = 0; reg <= max_register; reg++) {
1159
if (!affected_registers.Get(reg)) {
1163
// The chronologically first deferred action in the trace
1164
// is used to infer the action needed to restore a register
1165
// to its previous state (or not, if it's safe to ignore it).
1166
enum DeferredActionUndoType { IGNORE, RESTORE, CLEAR };
1167
DeferredActionUndoType undo_action = IGNORE;
1170
bool absolute = false;
1172
int store_position = -1;
1173
// This is a little tricky because we are scanning the actions in reverse
1174
// historical order (newest first).
1175
for (DeferredAction* action = actions_;
1177
action = action->next()) {
1178
if (action->Mentions(reg)) {
1179
switch (action->type()) {
1180
case ActionNode::SET_REGISTER: {
1181
Trace::DeferredSetRegister* psr =
1182
static_cast<Trace::DeferredSetRegister*>(action);
1184
value += psr->value();
1187
// SET_REGISTER is currently only used for newly introduced loop
1188
// counters. They can have a significant previous value if they
1189
// occour in a loop. TODO(lrn): Propagate this information, so
1190
// we can set undo_action to IGNORE if we know there is no value to
1192
undo_action = RESTORE;
1193
ASSERT_EQ(store_position, -1);
1197
case ActionNode::INCREMENT_REGISTER:
1201
ASSERT_EQ(store_position, -1);
1203
undo_action = RESTORE;
1205
case ActionNode::STORE_POSITION: {
1206
Trace::DeferredCapture* pc =
1207
static_cast<Trace::DeferredCapture*>(action);
1208
if (!clear && store_position == -1) {
1209
store_position = pc->cp_offset();
1212
// For captures we know that stores and clears alternate.
1213
// Other register, are never cleared, and if the occur
1214
// inside a loop, they might be assigned more than once.
1216
// Registers zero and one, aka "capture zero", is
1217
// always set correctly if we succeed. There is no
1218
// need to undo a setting on backtrack, because we
1219
// will set it again or fail.
1220
undo_action = IGNORE;
1222
undo_action = pc->is_capture() ? CLEAR : RESTORE;
1225
ASSERT_EQ(value, 0);
1228
case ActionNode::CLEAR_CAPTURES: {
1229
// Since we're scanning in reverse order, if we've already
1230
// set the position we have to ignore historically earlier
1231
// clearing operations.
1232
if (store_position == -1) {
1235
undo_action = RESTORE;
1237
ASSERT_EQ(value, 0);
1246
// Prepare for the undo-action (e.g., push if it's going to be popped).
1247
if (undo_action == RESTORE) {
1249
RegExpMacroAssembler::StackCheckFlag stack_check =
1250
RegExpMacroAssembler::kNoStackLimitCheck;
1251
if (pushes == push_limit) {
1252
stack_check = RegExpMacroAssembler::kCheckStackLimit;
1256
assembler->PushRegister(reg, stack_check);
1257
registers_to_pop->Set(reg, zone);
1258
} else if (undo_action == CLEAR) {
1259
registers_to_clear->Set(reg, zone);
1261
// Perform the chronologically last action (or accumulated increment)
1262
// for the register.
1263
if (store_position != -1) {
1264
assembler->WriteCurrentPositionToRegister(reg, store_position);
1266
assembler->ClearRegisters(reg, reg);
1267
} else if (absolute) {
1268
assembler->SetRegister(reg, value);
1269
} else if (value != 0) {
1270
assembler->AdvanceRegister(reg, value);
1276
// This is called as we come into a loop choice node and some other tricky
1277
// nodes. It normalizes the state of the code generator to ensure we can
1278
// generate generic code.
1279
void Trace::Flush(RegExpCompiler* compiler, RegExpNode* successor) {
1280
RegExpMacroAssembler* assembler = compiler->macro_assembler();
1282
ASSERT(!is_trivial());
1284
if (actions_ == NULL && backtrack() == NULL) {
1285
// Here we just have some deferred cp advances to fix and we are back to
1286
// a normal situation. We may also have to forget some information gained
1287
// through a quick check that was already performed.
1288
if (cp_offset_ != 0) assembler->AdvanceCurrentPosition(cp_offset_);
1289
// Create a new trivial state and generate the node with that.
1291
successor->Emit(compiler, &new_state);
1295
// Generate deferred actions here along with code to undo them again.
1296
OutSet affected_registers;
1298
if (backtrack() != NULL) {
1299
// Here we have a concrete backtrack location. These are set up by choice
1300
// nodes and so they indicate that we have a deferred save of the current
1301
// position which we may need to emit here.
1302
assembler->PushCurrentPosition();
1305
int max_register = FindAffectedRegisters(&affected_registers,
1307
OutSet registers_to_pop;
1308
OutSet registers_to_clear;
1309
PerformDeferredActions(assembler,
1313
®isters_to_clear,
1315
if (cp_offset_ != 0) {
1316
assembler->AdvanceCurrentPosition(cp_offset_);
1319
// Create a new trivial state and generate the node with that.
1321
assembler->PushBacktrack(&undo);
1323
successor->Emit(compiler, &new_state);
1325
// On backtrack we need to restore state.
1326
assembler->Bind(&undo);
1327
RestoreAffectedRegisters(assembler,
1330
registers_to_clear);
1331
if (backtrack() == NULL) {
1332
assembler->Backtrack();
1334
assembler->PopCurrentPosition();
1335
assembler->GoTo(backtrack());
1340
void NegativeSubmatchSuccess::Emit(RegExpCompiler* compiler, Trace* trace) {
1341
RegExpMacroAssembler* assembler = compiler->macro_assembler();
1343
// Omit flushing the trace. We discard the entire stack frame anyway.
1345
if (!label()->is_bound()) {
1346
// We are completely independent of the trace, since we ignore it,
1347
// so this code can be used as the generic version.
1348
assembler->Bind(label());
1351
// Throw away everything on the backtrack stack since the start
1352
// of the negative submatch and restore the character position.
1353
assembler->ReadCurrentPositionFromRegister(current_position_register_);
1354
assembler->ReadStackPointerFromRegister(stack_pointer_register_);
1355
if (clear_capture_count_ > 0) {
1356
// Clear any captures that might have been performed during the success
1357
// of the body of the negative look-ahead.
1358
int clear_capture_end = clear_capture_start_ + clear_capture_count_ - 1;
1359
assembler->ClearRegisters(clear_capture_start_, clear_capture_end);
1361
// Now that we have unwound the stack we find at the top of the stack the
1362
// backtrack that the BeginSubmatch node got.
1363
assembler->Backtrack();
1367
void EndNode::Emit(RegExpCompiler* compiler, Trace* trace) {
1368
if (!trace->is_trivial()) {
1369
trace->Flush(compiler, this);
1372
RegExpMacroAssembler* assembler = compiler->macro_assembler();
1373
if (!label()->is_bound()) {
1374
assembler->Bind(label());
1378
assembler->Succeed();
1381
assembler->GoTo(trace->backtrack());
1383
case NEGATIVE_SUBMATCH_SUCCESS:
1384
// This case is handled in a different virtual method.
1391
void GuardedAlternative::AddGuard(Guard* guard, Zone* zone) {
1392
if (guards_ == NULL)
1393
guards_ = new(zone) ZoneList<Guard*>(1, zone);
1394
guards_->Add(guard, zone);
1398
ActionNode* ActionNode::SetRegister(int reg,
1400
RegExpNode* on_success) {
1401
ActionNode* result =
1402
new(on_success->zone()) ActionNode(SET_REGISTER, on_success);
1403
result->data_.u_store_register.reg = reg;
1404
result->data_.u_store_register.value = val;
1409
ActionNode* ActionNode::IncrementRegister(int reg, RegExpNode* on_success) {
1410
ActionNode* result =
1411
new(on_success->zone()) ActionNode(INCREMENT_REGISTER, on_success);
1412
result->data_.u_increment_register.reg = reg;
1417
ActionNode* ActionNode::StorePosition(int reg,
1419
RegExpNode* on_success) {
1420
ActionNode* result =
1421
new(on_success->zone()) ActionNode(STORE_POSITION, on_success);
1422
result->data_.u_position_register.reg = reg;
1423
result->data_.u_position_register.is_capture = is_capture;
1428
ActionNode* ActionNode::ClearCaptures(Interval range,
1429
RegExpNode* on_success) {
1430
ActionNode* result =
1431
new(on_success->zone()) ActionNode(CLEAR_CAPTURES, on_success);
1432
result->data_.u_clear_captures.range_from = range.from();
1433
result->data_.u_clear_captures.range_to = range.to();
1438
ActionNode* ActionNode::BeginSubmatch(int stack_reg,
1440
RegExpNode* on_success) {
1441
ActionNode* result =
1442
new(on_success->zone()) ActionNode(BEGIN_SUBMATCH, on_success);
1443
result->data_.u_submatch.stack_pointer_register = stack_reg;
1444
result->data_.u_submatch.current_position_register = position_reg;
1449
ActionNode* ActionNode::PositiveSubmatchSuccess(int stack_reg,
1451
int clear_register_count,
1452
int clear_register_from,
1453
RegExpNode* on_success) {
1454
ActionNode* result =
1455
new(on_success->zone()) ActionNode(POSITIVE_SUBMATCH_SUCCESS, on_success);
1456
result->data_.u_submatch.stack_pointer_register = stack_reg;
1457
result->data_.u_submatch.current_position_register = position_reg;
1458
result->data_.u_submatch.clear_register_count = clear_register_count;
1459
result->data_.u_submatch.clear_register_from = clear_register_from;
1464
ActionNode* ActionNode::EmptyMatchCheck(int start_register,
1465
int repetition_register,
1466
int repetition_limit,
1467
RegExpNode* on_success) {
1468
ActionNode* result =
1469
new(on_success->zone()) ActionNode(EMPTY_MATCH_CHECK, on_success);
1470
result->data_.u_empty_match_check.start_register = start_register;
1471
result->data_.u_empty_match_check.repetition_register = repetition_register;
1472
result->data_.u_empty_match_check.repetition_limit = repetition_limit;
1477
#define DEFINE_ACCEPT(Type) \
1478
void Type##Node::Accept(NodeVisitor* visitor) { \
1479
visitor->Visit##Type(this); \
1481
FOR_EACH_NODE_TYPE(DEFINE_ACCEPT)
1482
#undef DEFINE_ACCEPT
1485
void LoopChoiceNode::Accept(NodeVisitor* visitor) {
1486
visitor->VisitLoopChoice(this);
1490
// -------------------------------------------------------------------
1494
void ChoiceNode::GenerateGuard(RegExpMacroAssembler* macro_assembler,
1497
switch (guard->op()) {
1499
ASSERT(!trace->mentions_reg(guard->reg()));
1500
macro_assembler->IfRegisterGE(guard->reg(),
1502
trace->backtrack());
1505
ASSERT(!trace->mentions_reg(guard->reg()));
1506
macro_assembler->IfRegisterLT(guard->reg(),
1508
trace->backtrack());
1514
// Returns the number of characters in the equivalence class, omitting those
1515
// that cannot occur in the source string because it is ASCII.
1516
static int GetCaseIndependentLetters(Isolate* isolate,
1519
unibrow::uchar* letters) {
1521
isolate->jsregexp_uncanonicalize()->get(character, '\0', letters);
1522
// Unibrow returns 0 or 1 for characters where case independence is
1525
letters[0] = character;
1528
if (!ascii_subject || character <= String::kMaxAsciiCharCode) {
1531
// The standard requires that non-ASCII characters cannot have ASCII
1532
// character codes in their equivalence class.
1537
static inline bool EmitSimpleCharacter(Isolate* isolate,
1538
RegExpCompiler* compiler,
1544
RegExpMacroAssembler* assembler = compiler->macro_assembler();
1545
bool bound_checked = false;
1547
assembler->LoadCurrentCharacter(
1551
bound_checked = true;
1553
assembler->CheckNotCharacter(c, on_failure);
1554
return bound_checked;
1558
// Only emits non-letters (things that don't have case). Only used for case
1559
// independent matches.
1560
static inline bool EmitAtomNonLetter(Isolate* isolate,
1561
RegExpCompiler* compiler,
1567
RegExpMacroAssembler* macro_assembler = compiler->macro_assembler();
1568
bool ascii = compiler->ascii();
1569
unibrow::uchar chars[unibrow::Ecma262UnCanonicalize::kMaxWidth];
1570
int length = GetCaseIndependentLetters(isolate, c, ascii, chars);
1572
// This can't match. Must be an ASCII subject and a non-ASCII character.
1573
// We do not need to do anything since the ASCII pass already handled this.
1574
return false; // Bounds not checked.
1576
bool checked = false;
1577
// We handle the length > 1 case in a later pass.
1579
if (ascii && c > String::kMaxAsciiCharCodeU) {
1580
// Can't match - see above.
1581
return false; // Bounds not checked.
1584
macro_assembler->LoadCurrentCharacter(cp_offset, on_failure, check);
1587
macro_assembler->CheckNotCharacter(c, on_failure);
1593
static bool ShortCutEmitCharacterPair(RegExpMacroAssembler* macro_assembler,
1597
Label* on_failure) {
1600
char_mask = String::kMaxAsciiCharCode;
1602
char_mask = String::kMaxUtf16CodeUnit;
1604
uc16 exor = c1 ^ c2;
1605
// Check whether exor has only one bit set.
1606
if (((exor - 1) & exor) == 0) {
1607
// If c1 and c2 differ only by one bit.
1608
// Ecma262UnCanonicalize always gives the highest number last.
1610
uc16 mask = char_mask ^ exor;
1611
macro_assembler->CheckNotCharacterAfterAnd(c1, mask, on_failure);
1615
uc16 diff = c2 - c1;
1616
if (((diff - 1) & diff) == 0 && c1 >= diff) {
1617
// If the characters differ by 2^n but don't differ by one bit then
1618
// subtract the difference from the found character, then do the or
1619
// trick. We avoid the theoretical case where negative numbers are
1620
// involved in order to simplify code generation.
1621
uc16 mask = char_mask ^ diff;
1622
macro_assembler->CheckNotCharacterAfterMinusAnd(c1 - diff,
1632
typedef bool EmitCharacterFunction(Isolate* isolate,
1633
RegExpCompiler* compiler,
1640
// Only emits letters (things that have case). Only used for case independent
1642
static inline bool EmitAtomLetter(Isolate* isolate,
1643
RegExpCompiler* compiler,
1649
RegExpMacroAssembler* macro_assembler = compiler->macro_assembler();
1650
bool ascii = compiler->ascii();
1651
unibrow::uchar chars[unibrow::Ecma262UnCanonicalize::kMaxWidth];
1652
int length = GetCaseIndependentLetters(isolate, c, ascii, chars);
1653
if (length <= 1) return false;
1654
// We may not need to check against the end of the input string
1655
// if this character lies before a character that matched.
1657
macro_assembler->LoadCurrentCharacter(cp_offset, on_failure, check);
1660
ASSERT(unibrow::Ecma262UnCanonicalize::kMaxWidth == 4);
1663
if (ShortCutEmitCharacterPair(macro_assembler,
1669
macro_assembler->CheckCharacter(chars[0], &ok);
1670
macro_assembler->CheckNotCharacter(chars[1], on_failure);
1671
macro_assembler->Bind(&ok);
1676
macro_assembler->CheckCharacter(chars[3], &ok);
1679
macro_assembler->CheckCharacter(chars[0], &ok);
1680
macro_assembler->CheckCharacter(chars[1], &ok);
1681
macro_assembler->CheckNotCharacter(chars[2], on_failure);
1682
macro_assembler->Bind(&ok);
1692
static void EmitBoundaryTest(RegExpMacroAssembler* masm,
1694
Label* fall_through,
1695
Label* above_or_equal,
1697
if (below != fall_through) {
1698
masm->CheckCharacterLT(border, below);
1699
if (above_or_equal != fall_through) masm->GoTo(above_or_equal);
1701
masm->CheckCharacterGT(border - 1, above_or_equal);
1706
static void EmitDoubleBoundaryTest(RegExpMacroAssembler* masm,
1709
Label* fall_through,
1711
Label* out_of_range) {
1712
if (in_range == fall_through) {
1713
if (first == last) {
1714
masm->CheckNotCharacter(first, out_of_range);
1716
masm->CheckCharacterNotInRange(first, last, out_of_range);
1719
if (first == last) {
1720
masm->CheckCharacter(first, in_range);
1722
masm->CheckCharacterInRange(first, last, in_range);
1724
if (out_of_range != fall_through) masm->GoTo(out_of_range);
1729
// even_label is for ranges[i] to ranges[i + 1] where i - start_index is even.
1730
// odd_label is for ranges[i] to ranges[i + 1] where i - start_index is odd.
1731
static void EmitUseLookupTable(
1732
RegExpMacroAssembler* masm,
1733
ZoneList<int>* ranges,
1737
Label* fall_through,
1740
static const int kSize = RegExpMacroAssembler::kTableSize;
1741
static const int kMask = RegExpMacroAssembler::kTableMask;
1743
int base = (min_char & ~kMask);
1746
// Assert that everything is on one kTableSize page.
1747
for (int i = start_index; i <= end_index; i++) {
1748
ASSERT_EQ(ranges->at(i) & ~kMask, base);
1750
ASSERT(start_index == 0 || (ranges->at(start_index - 1) & ~kMask) <= base);
1754
Label* on_bit_clear;
1756
if (even_label == fall_through) {
1757
on_bit_set = odd_label;
1758
on_bit_clear = even_label;
1761
on_bit_set = even_label;
1762
on_bit_clear = odd_label;
1765
for (int i = 0; i < (ranges->at(start_index) & kMask) && i < kSize; i++) {
1770
for (int i = start_index; i < end_index; i++) {
1771
for (j = (ranges->at(i) & kMask); j < (ranges->at(i + 1) & kMask); j++) {
1776
for (int i = j; i < kSize; i++) {
1779
// TODO(erikcorry): Cache these.
1780
Handle<ByteArray> ba = FACTORY->NewByteArray(kSize, TENURED);
1781
for (int i = 0; i < kSize; i++) {
1782
ba->set(i, templ[i]);
1784
masm->CheckBitInTable(ba, on_bit_set);
1785
if (on_bit_clear != fall_through) masm->GoTo(on_bit_clear);
1789
static void CutOutRange(RegExpMacroAssembler* masm,
1790
ZoneList<int>* ranges,
1796
bool odd = (((cut_index - start_index) & 1) == 1);
1797
Label* in_range_label = odd ? odd_label : even_label;
1799
EmitDoubleBoundaryTest(masm,
1800
ranges->at(cut_index),
1801
ranges->at(cut_index + 1) - 1,
1805
ASSERT(!dummy.is_linked());
1806
// Cut out the single range by rewriting the array. This creates a new
1807
// range that is a merger of the two ranges on either side of the one we
1808
// are cutting out. The oddity of the labels is preserved.
1809
for (int j = cut_index; j > start_index; j--) {
1810
ranges->at(j) = ranges->at(j - 1);
1812
for (int j = cut_index + 1; j < end_index; j++) {
1813
ranges->at(j) = ranges->at(j + 1);
1818
// Unicode case. Split the search space into kSize spaces that are handled
1820
static void SplitSearchSpace(ZoneList<int>* ranges,
1823
int* new_start_index,
1826
static const int kSize = RegExpMacroAssembler::kTableSize;
1827
static const int kMask = RegExpMacroAssembler::kTableMask;
1829
int first = ranges->at(start_index);
1830
int last = ranges->at(end_index) - 1;
1832
*new_start_index = start_index;
1833
*border = (ranges->at(start_index) & ~kMask) + kSize;
1834
while (*new_start_index < end_index) {
1835
if (ranges->at(*new_start_index) > *border) break;
1836
(*new_start_index)++;
1838
// new_start_index is the index of the first edge that is beyond the
1839
// current kSize space.
1841
// For very large search spaces we do a binary chop search of the non-ASCII
1842
// space instead of just going to the end of the current kSize space. The
1843
// heuristics are complicated a little by the fact that any 128-character
1844
// encoding space can be quickly tested with a table lookup, so we don't
1845
// wish to do binary chop search at a smaller granularity than that. A
1846
// 128-character space can take up a lot of space in the ranges array if,
1847
// for example, we only want to match every second character (eg. the lower
1848
// case characters on some Unicode pages).
1849
int binary_chop_index = (end_index + start_index) / 2;
1850
// The first test ensures that we get to the code that handles the ASCII
1851
// range with a single not-taken branch, speeding up this important
1852
// character range (even non-ASCII charset-based text has spaces and
1854
if (*border - 1 > String::kMaxAsciiCharCode && // ASCII case.
1855
end_index - start_index > (*new_start_index - start_index) * 2 &&
1856
last - first > kSize * 2 &&
1857
binary_chop_index > *new_start_index &&
1858
ranges->at(binary_chop_index) >= first + 2 * kSize) {
1859
int scan_forward_for_section_border = binary_chop_index;;
1860
int new_border = (ranges->at(binary_chop_index) | kMask) + 1;
1862
while (scan_forward_for_section_border < end_index) {
1863
if (ranges->at(scan_forward_for_section_border) > new_border) {
1864
*new_start_index = scan_forward_for_section_border;
1865
*border = new_border;
1868
scan_forward_for_section_border++;
1872
ASSERT(*new_start_index > start_index);
1873
*new_end_index = *new_start_index - 1;
1874
if (ranges->at(*new_end_index) == *border) {
1877
if (*border >= ranges->at(end_index)) {
1878
*border = ranges->at(end_index);
1879
*new_start_index = end_index; // Won't be used.
1880
*new_end_index = end_index - 1;
1885
// Gets a series of segment boundaries representing a character class. If the
1886
// character is in the range between an even and an odd boundary (counting from
1887
// start_index) then go to even_label, otherwise go to odd_label. We already
1888
// know that the character is in the range of min_char to max_char inclusive.
1889
// Either label can be NULL indicating backtracking. Either label can also be
1890
// equal to the fall_through label.
1891
static void GenerateBranches(RegExpMacroAssembler* masm,
1892
ZoneList<int>* ranges,
1897
Label* fall_through,
1900
int first = ranges->at(start_index);
1901
int last = ranges->at(end_index) - 1;
1903
ASSERT_LT(min_char, first);
1905
// Just need to test if the character is before or on-or-after
1906
// a particular character.
1907
if (start_index == end_index) {
1908
EmitBoundaryTest(masm, first, fall_through, even_label, odd_label);
1912
// Another almost trivial case: There is one interval in the middle that is
1913
// different from the end intervals.
1914
if (start_index + 1 == end_index) {
1915
EmitDoubleBoundaryTest(
1916
masm, first, last, fall_through, even_label, odd_label);
1920
// It's not worth using table lookup if there are very few intervals in the
1922
if (end_index - start_index <= 6) {
1923
// It is faster to test for individual characters, so we look for those
1924
// first, then try arbitrary ranges in the second round.
1925
static int kNoCutIndex = -1;
1926
int cut = kNoCutIndex;
1927
for (int i = start_index; i < end_index; i++) {
1928
if (ranges->at(i) == ranges->at(i + 1) - 1) {
1933
if (cut == kNoCutIndex) cut = start_index;
1935
masm, ranges, start_index, end_index, cut, even_label, odd_label);
1936
ASSERT_GE(end_index - start_index, 2);
1937
GenerateBranches(masm,
1949
// If there are a lot of intervals in the regexp, then we will use tables to
1950
// determine whether the character is inside or outside the character class.
1951
static const int kBits = RegExpMacroAssembler::kTableSizeBits;
1953
if ((max_char >> kBits) == (min_char >> kBits)) {
1954
EmitUseLookupTable(masm,
1965
if ((min_char >> kBits) != (first >> kBits)) {
1966
masm->CheckCharacterLT(first, odd_label);
1967
GenerateBranches(masm,
1979
int new_start_index = 0;
1980
int new_end_index = 0;
1983
SplitSearchSpace(ranges,
1991
Label* above = &handle_rest;
1992
if (border == last + 1) {
1993
// We didn't find any section that started after the limit, so everything
1994
// above the border is one of the terminal labels.
1995
above = (end_index & 1) != (start_index & 1) ? odd_label : even_label;
1996
ASSERT(new_end_index == end_index - 1);
1999
ASSERT_LE(start_index, new_end_index);
2000
ASSERT_LE(new_start_index, end_index);
2001
ASSERT_LT(start_index, new_start_index);
2002
ASSERT_LT(new_end_index, end_index);
2003
ASSERT(new_end_index + 1 == new_start_index ||
2004
(new_end_index + 2 == new_start_index &&
2005
border == ranges->at(new_end_index + 1)));
2006
ASSERT_LT(min_char, border - 1);
2007
ASSERT_LT(border, max_char);
2008
ASSERT_LT(ranges->at(new_end_index), border);
2009
ASSERT(border < ranges->at(new_start_index) ||
2010
(border == ranges->at(new_start_index) &&
2011
new_start_index == end_index &&
2012
new_end_index == end_index - 1 &&
2013
border == last + 1));
2014
ASSERT(new_start_index == 0 || border >= ranges->at(new_start_index - 1));
2016
masm->CheckCharacterGT(border - 1, above);
2018
GenerateBranches(masm,
2027
if (handle_rest.is_linked()) {
2028
masm->Bind(&handle_rest);
2029
bool flip = (new_start_index & 1) != (start_index & 1);
2030
GenerateBranches(masm,
2037
flip ? odd_label : even_label,
2038
flip ? even_label : odd_label);
2043
static void EmitCharClass(RegExpMacroAssembler* macro_assembler,
2044
RegExpCharacterClass* cc,
2051
ZoneList<CharacterRange>* ranges = cc->ranges(zone);
2052
if (!CharacterRange::IsCanonical(ranges)) {
2053
CharacterRange::Canonicalize(ranges);
2058
max_char = String::kMaxAsciiCharCode;
2060
max_char = String::kMaxUtf16CodeUnit;
2063
int range_count = ranges->length();
2065
int last_valid_range = range_count - 1;
2066
while (last_valid_range >= 0) {
2067
CharacterRange& range = ranges->at(last_valid_range);
2068
if (range.from() <= max_char) {
2074
if (last_valid_range < 0) {
2075
if (!cc->is_negated()) {
2076
macro_assembler->GoTo(on_failure);
2079
macro_assembler->CheckPosition(cp_offset, on_failure);
2084
if (last_valid_range == 0 &&
2085
ranges->at(0).IsEverything(max_char)) {
2086
if (cc->is_negated()) {
2087
macro_assembler->GoTo(on_failure);
2089
// This is a common case hit by non-anchored expressions.
2091
macro_assembler->CheckPosition(cp_offset, on_failure);
2096
if (last_valid_range == 0 &&
2097
!cc->is_negated() &&
2098
ranges->at(0).IsEverything(max_char)) {
2099
// This is a common case hit by non-anchored expressions.
2101
macro_assembler->CheckPosition(cp_offset, on_failure);
2107
macro_assembler->LoadCurrentCharacter(cp_offset, on_failure, check_offset);
2110
if (cc->is_standard(zone) &&
2111
macro_assembler->CheckSpecialCharacterClass(cc->standard_type(),
2117
// A new list with ascending entries. Each entry is a code unit
2118
// where there is a boundary between code units that are part of
2119
// the class and code units that are not. Normally we insert an
2120
// entry at zero which goes to the failure label, but if there
2121
// was already one there we fall through for success on that entry.
2122
// Subsequent entries have alternating meaning (success/failure).
2123
ZoneList<int>* range_boundaries =
2124
new(zone) ZoneList<int>(last_valid_range, zone);
2126
bool zeroth_entry_is_failure = !cc->is_negated();
2128
for (int i = 0; i <= last_valid_range; i++) {
2129
CharacterRange& range = ranges->at(i);
2130
if (range.from() == 0) {
2132
zeroth_entry_is_failure = !zeroth_entry_is_failure;
2134
range_boundaries->Add(range.from(), zone);
2136
range_boundaries->Add(range.to() + 1, zone);
2138
int end_index = range_boundaries->length() - 1;
2139
if (range_boundaries->at(end_index) > max_char) {
2144
GenerateBranches(macro_assembler,
2151
zeroth_entry_is_failure ? &fall_through : on_failure,
2152
zeroth_entry_is_failure ? on_failure : &fall_through);
2153
macro_assembler->Bind(&fall_through);
2157
RegExpNode::~RegExpNode() {
2161
RegExpNode::LimitResult RegExpNode::LimitVersions(RegExpCompiler* compiler,
2163
// If we are generating a greedy loop then don't stop and don't reuse code.
2164
if (trace->stop_node() != NULL) {
2168
RegExpMacroAssembler* macro_assembler = compiler->macro_assembler();
2169
if (trace->is_trivial()) {
2170
if (label_.is_bound()) {
2171
// We are being asked to generate a generic version, but that's already
2172
// been done so just go to it.
2173
macro_assembler->GoTo(&label_);
2176
if (compiler->recursion_depth() >= RegExpCompiler::kMaxRecursion) {
2177
// To avoid too deep recursion we push the node to the work queue and just
2178
// generate a goto here.
2179
compiler->AddWork(this);
2180
macro_assembler->GoTo(&label_);
2183
// Generate generic version of the node and bind the label for later use.
2184
macro_assembler->Bind(&label_);
2188
// We are being asked to make a non-generic version. Keep track of how many
2189
// non-generic versions we generate so as not to overdo it.
2191
if (FLAG_regexp_optimization &&
2192
trace_count_ < kMaxCopiesCodeGenerated &&
2193
compiler->recursion_depth() <= RegExpCompiler::kMaxRecursion) {
2197
// If we get here code has been generated for this node too many times or
2198
// recursion is too deep. Time to switch to a generic version. The code for
2199
// generic versions above can handle deep recursion properly.
2200
trace->Flush(compiler, this);
2205
int ActionNode::EatsAtLeast(int still_to_find,
2206
int recursion_depth,
2207
bool not_at_start) {
2208
if (recursion_depth > RegExpCompiler::kMaxRecursion) return 0;
2209
if (type_ == POSITIVE_SUBMATCH_SUCCESS) return 0; // Rewinds input!
2210
return on_success()->EatsAtLeast(still_to_find,
2211
recursion_depth + 1,
2216
void ActionNode::FillInBMInfo(int offset,
2217
int recursion_depth,
2219
BoyerMooreLookahead* bm,
2220
bool not_at_start) {
2221
if (type_ == BEGIN_SUBMATCH) {
2222
bm->SetRest(offset);
2223
} else if (type_ != POSITIVE_SUBMATCH_SUCCESS) {
2224
on_success()->FillInBMInfo(
2225
offset, recursion_depth + 1, budget - 1, bm, not_at_start);
2227
SaveBMInfo(bm, not_at_start, offset);
2231
int AssertionNode::EatsAtLeast(int still_to_find,
2232
int recursion_depth,
2233
bool not_at_start) {
2234
if (recursion_depth > RegExpCompiler::kMaxRecursion) return 0;
2235
// If we know we are not at the start and we are asked "how many characters
2236
// will you match if you succeed?" then we can answer anything since false
2237
// implies false. So lets just return the max answer (still_to_find) since
2238
// that won't prevent us from preloading a lot of characters for the other
2239
// branches in the node graph.
2240
if (type() == AT_START && not_at_start) return still_to_find;
2241
return on_success()->EatsAtLeast(still_to_find,
2242
recursion_depth + 1,
2247
void AssertionNode::FillInBMInfo(int offset,
2248
int recursion_depth,
2250
BoyerMooreLookahead* bm,
2251
bool not_at_start) {
2252
// Match the behaviour of EatsAtLeast on this node.
2253
if (type() == AT_START && not_at_start) return;
2254
on_success()->FillInBMInfo(
2255
offset, recursion_depth + 1, budget - 1, bm, not_at_start);
2256
SaveBMInfo(bm, not_at_start, offset);
2260
int BackReferenceNode::EatsAtLeast(int still_to_find,
2261
int recursion_depth,
2262
bool not_at_start) {
2263
if (recursion_depth > RegExpCompiler::kMaxRecursion) return 0;
2264
return on_success()->EatsAtLeast(still_to_find,
2265
recursion_depth + 1,
2270
int TextNode::EatsAtLeast(int still_to_find,
2271
int recursion_depth,
2272
bool not_at_start) {
2273
int answer = Length();
2274
if (answer >= still_to_find) return answer;
2275
if (recursion_depth > RegExpCompiler::kMaxRecursion) return answer;
2276
// We are not at start after this node so we set the last argument to 'true'.
2277
return answer + on_success()->EatsAtLeast(still_to_find - answer,
2278
recursion_depth + 1,
2283
int NegativeLookaheadChoiceNode::EatsAtLeast(int still_to_find,
2284
int recursion_depth,
2285
bool not_at_start) {
2286
if (recursion_depth > RegExpCompiler::kMaxRecursion) return 0;
2287
// Alternative 0 is the negative lookahead, alternative 1 is what comes
2289
RegExpNode* node = alternatives_->at(1).node();
2290
return node->EatsAtLeast(still_to_find, recursion_depth + 1, not_at_start);
2294
void NegativeLookaheadChoiceNode::GetQuickCheckDetails(
2295
QuickCheckDetails* details,
2296
RegExpCompiler* compiler,
2298
bool not_at_start) {
2299
// Alternative 0 is the negative lookahead, alternative 1 is what comes
2301
RegExpNode* node = alternatives_->at(1).node();
2302
return node->GetQuickCheckDetails(details, compiler, filled_in, not_at_start);
2306
int ChoiceNode::EatsAtLeastHelper(int still_to_find,
2307
int recursion_depth,
2308
RegExpNode* ignore_this_node,
2309
bool not_at_start) {
2310
if (recursion_depth > RegExpCompiler::kMaxRecursion) return 0;
2312
int choice_count = alternatives_->length();
2313
for (int i = 0; i < choice_count; i++) {
2314
RegExpNode* node = alternatives_->at(i).node();
2315
if (node == ignore_this_node) continue;
2316
int node_eats_at_least = node->EatsAtLeast(still_to_find,
2317
recursion_depth + 1,
2319
if (node_eats_at_least < min) min = node_eats_at_least;
2325
int LoopChoiceNode::EatsAtLeast(int still_to_find,
2326
int recursion_depth,
2327
bool not_at_start) {
2328
return EatsAtLeastHelper(still_to_find,
2335
int ChoiceNode::EatsAtLeast(int still_to_find,
2336
int recursion_depth,
2337
bool not_at_start) {
2338
return EatsAtLeastHelper(still_to_find,
2345
// Takes the left-most 1-bit and smears it out, setting all bits to its right.
2346
static inline uint32_t SmearBitsRight(uint32_t v) {
2356
bool QuickCheckDetails::Rationalize(bool asc) {
2357
bool found_useful_op = false;
2360
char_mask = String::kMaxAsciiCharCode;
2362
char_mask = String::kMaxUtf16CodeUnit;
2367
for (int i = 0; i < characters_; i++) {
2368
Position* pos = &positions_[i];
2369
if ((pos->mask & String::kMaxAsciiCharCode) != 0) {
2370
found_useful_op = true;
2372
mask_ |= (pos->mask & char_mask) << char_shift;
2373
value_ |= (pos->value & char_mask) << char_shift;
2374
char_shift += asc ? 8 : 16;
2376
return found_useful_op;
2380
bool RegExpNode::EmitQuickCheck(RegExpCompiler* compiler,
2382
bool preload_has_checked_bounds,
2383
Label* on_possible_success,
2384
QuickCheckDetails* details,
2385
bool fall_through_on_failure) {
2386
if (details->characters() == 0) return false;
2387
GetQuickCheckDetails(details, compiler, 0, trace->at_start() == Trace::FALSE);
2388
if (details->cannot_match()) return false;
2389
if (!details->Rationalize(compiler->ascii())) return false;
2390
ASSERT(details->characters() == 1 ||
2391
compiler->macro_assembler()->CanReadUnaligned());
2392
uint32_t mask = details->mask();
2393
uint32_t value = details->value();
2395
RegExpMacroAssembler* assembler = compiler->macro_assembler();
2397
if (trace->characters_preloaded() != details->characters()) {
2398
assembler->LoadCurrentCharacter(trace->cp_offset(),
2400
!preload_has_checked_bounds,
2401
details->characters());
2405
bool need_mask = true;
2407
if (details->characters() == 1) {
2408
// If number of characters preloaded is 1 then we used a byte or 16 bit
2409
// load so the value is already masked down.
2411
if (compiler->ascii()) {
2412
char_mask = String::kMaxAsciiCharCode;
2414
char_mask = String::kMaxUtf16CodeUnit;
2416
if ((mask & char_mask) == char_mask) need_mask = false;
2419
// For 2-character preloads in ASCII mode or 1-character preloads in
2420
// TWO_BYTE mode we also use a 16 bit load with zero extend.
2421
if (details->characters() == 2 && compiler->ascii()) {
2422
if ((mask & 0x7f7f) == 0x7f7f) need_mask = false;
2423
} else if (details->characters() == 1 && !compiler->ascii()) {
2424
if ((mask & 0xffff) == 0xffff) need_mask = false;
2426
if (mask == 0xffffffff) need_mask = false;
2430
if (fall_through_on_failure) {
2432
assembler->CheckCharacterAfterAnd(value, mask, on_possible_success);
2434
assembler->CheckCharacter(value, on_possible_success);
2438
assembler->CheckNotCharacterAfterAnd(value, mask, trace->backtrack());
2440
assembler->CheckNotCharacter(value, trace->backtrack());
2447
// Here is the meat of GetQuickCheckDetails (see also the comment on the
2448
// super-class in the .h file).
2450
// We iterate along the text object, building up for each character a
2451
// mask and value that can be used to test for a quick failure to match.
2452
// The masks and values for the positions will be combined into a single
2453
// machine word for the current character width in order to be used in
2454
// generating a quick check.
2455
void TextNode::GetQuickCheckDetails(QuickCheckDetails* details,
2456
RegExpCompiler* compiler,
2457
int characters_filled_in,
2458
bool not_at_start) {
2459
Isolate* isolate = Isolate::Current();
2460
ASSERT(characters_filled_in < details->characters());
2461
int characters = details->characters();
2463
if (compiler->ascii()) {
2464
char_mask = String::kMaxAsciiCharCode;
2466
char_mask = String::kMaxUtf16CodeUnit;
2468
for (int k = 0; k < elms_->length(); k++) {
2469
TextElement elm = elms_->at(k);
2470
if (elm.type == TextElement::ATOM) {
2471
Vector<const uc16> quarks = elm.data.u_atom->data();
2472
for (int i = 0; i < characters && i < quarks.length(); i++) {
2473
QuickCheckDetails::Position* pos =
2474
details->positions(characters_filled_in);
2476
if (c > char_mask) {
2477
// If we expect a non-ASCII character from an ASCII string,
2478
// there is no way we can match. Not even case independent
2479
// matching can turn an ASCII character into non-ASCII or
2481
details->set_cannot_match();
2482
pos->determines_perfectly = false;
2485
if (compiler->ignore_case()) {
2486
unibrow::uchar chars[unibrow::Ecma262UnCanonicalize::kMaxWidth];
2487
int length = GetCaseIndependentLetters(isolate, c, compiler->ascii(),
2489
ASSERT(length != 0); // Can only happen if c > char_mask (see above).
2491
// This letter has no case equivalents, so it's nice and simple
2492
// and the mask-compare will determine definitely whether we have
2493
// a match at this character position.
2494
pos->mask = char_mask;
2496
pos->determines_perfectly = true;
2498
uint32_t common_bits = char_mask;
2499
uint32_t bits = chars[0];
2500
for (int j = 1; j < length; j++) {
2501
uint32_t differing_bits = ((chars[j] & common_bits) ^ bits);
2502
common_bits ^= differing_bits;
2503
bits &= common_bits;
2505
// If length is 2 and common bits has only one zero in it then
2506
// our mask and compare instruction will determine definitely
2507
// whether we have a match at this character position. Otherwise
2508
// it can only be an approximate check.
2509
uint32_t one_zero = (common_bits | ~char_mask);
2510
if (length == 2 && ((~one_zero) & ((~one_zero) - 1)) == 0) {
2511
pos->determines_perfectly = true;
2513
pos->mask = common_bits;
2517
// Don't ignore case. Nice simple case where the mask-compare will
2518
// determine definitely whether we have a match at this character
2520
pos->mask = char_mask;
2522
pos->determines_perfectly = true;
2524
characters_filled_in++;
2525
ASSERT(characters_filled_in <= details->characters());
2526
if (characters_filled_in == details->characters()) {
2531
QuickCheckDetails::Position* pos =
2532
details->positions(characters_filled_in);
2533
RegExpCharacterClass* tree = elm.data.u_char_class;
2534
ZoneList<CharacterRange>* ranges = tree->ranges(zone());
2535
if (tree->is_negated()) {
2536
// A quick check uses multi-character mask and compare. There is no
2537
// useful way to incorporate a negative char class into this scheme
2538
// so we just conservatively create a mask and value that will always
2543
int first_range = 0;
2544
while (ranges->at(first_range).from() > char_mask) {
2546
if (first_range == ranges->length()) {
2547
details->set_cannot_match();
2548
pos->determines_perfectly = false;
2552
CharacterRange range = ranges->at(first_range);
2553
uc16 from = range.from();
2554
uc16 to = range.to();
2555
if (to > char_mask) {
2558
uint32_t differing_bits = (from ^ to);
2559
// A mask and compare is only perfect if the differing bits form a
2560
// number like 00011111 with one single block of trailing 1s.
2561
if ((differing_bits & (differing_bits + 1)) == 0 &&
2562
from + differing_bits == to) {
2563
pos->determines_perfectly = true;
2565
uint32_t common_bits = ~SmearBitsRight(differing_bits);
2566
uint32_t bits = (from & common_bits);
2567
for (int i = first_range + 1; i < ranges->length(); i++) {
2568
CharacterRange range = ranges->at(i);
2569
uc16 from = range.from();
2570
uc16 to = range.to();
2571
if (from > char_mask) continue;
2572
if (to > char_mask) to = char_mask;
2573
// Here we are combining more ranges into the mask and compare
2574
// value. With each new range the mask becomes more sparse and
2575
// so the chances of a false positive rise. A character class
2576
// with multiple ranges is assumed never to be equivalent to a
2577
// mask and compare operation.
2578
pos->determines_perfectly = false;
2579
uint32_t new_common_bits = (from ^ to);
2580
new_common_bits = ~SmearBitsRight(new_common_bits);
2581
common_bits &= new_common_bits;
2582
bits &= new_common_bits;
2583
uint32_t differing_bits = (from & common_bits) ^ bits;
2584
common_bits ^= differing_bits;
2585
bits &= common_bits;
2587
pos->mask = common_bits;
2590
characters_filled_in++;
2591
ASSERT(characters_filled_in <= details->characters());
2592
if (characters_filled_in == details->characters()) {
2597
ASSERT(characters_filled_in != details->characters());
2598
if (!details->cannot_match()) {
2599
on_success()-> GetQuickCheckDetails(details,
2601
characters_filled_in,
2607
void QuickCheckDetails::Clear() {
2608
for (int i = 0; i < characters_; i++) {
2609
positions_[i].mask = 0;
2610
positions_[i].value = 0;
2611
positions_[i].determines_perfectly = false;
2617
void QuickCheckDetails::Advance(int by, bool ascii) {
2619
if (by >= characters_) {
2623
for (int i = 0; i < characters_ - by; i++) {
2624
positions_[i] = positions_[by + i];
2626
for (int i = characters_ - by; i < characters_; i++) {
2627
positions_[i].mask = 0;
2628
positions_[i].value = 0;
2629
positions_[i].determines_perfectly = false;
2632
// We could change mask_ and value_ here but we would never advance unless
2633
// they had already been used in a check and they won't be used again because
2634
// it would gain us nothing. So there's no point.
2638
void QuickCheckDetails::Merge(QuickCheckDetails* other, int from_index) {
2639
ASSERT(characters_ == other->characters_);
2640
if (other->cannot_match_) {
2643
if (cannot_match_) {
2647
for (int i = from_index; i < characters_; i++) {
2648
QuickCheckDetails::Position* pos = positions(i);
2649
QuickCheckDetails::Position* other_pos = other->positions(i);
2650
if (pos->mask != other_pos->mask ||
2651
pos->value != other_pos->value ||
2652
!other_pos->determines_perfectly) {
2653
// Our mask-compare operation will be approximate unless we have the
2654
// exact same operation on both sides of the alternation.
2655
pos->determines_perfectly = false;
2657
pos->mask &= other_pos->mask;
2658
pos->value &= pos->mask;
2659
other_pos->value &= pos->mask;
2660
uc16 differing_bits = (pos->value ^ other_pos->value);
2661
pos->mask &= ~differing_bits;
2662
pos->value &= pos->mask;
2669
explicit VisitMarker(NodeInfo* info) : info_(info) {
2670
ASSERT(!info->visited);
2671
info->visited = true;
2674
info_->visited = false;
2681
RegExpNode* SeqRegExpNode::FilterASCII(int depth) {
2682
if (info()->replacement_calculated) return replacement();
2683
if (depth < 0) return this;
2684
ASSERT(!info()->visited);
2685
VisitMarker marker(info());
2686
return FilterSuccessor(depth - 1);
2690
RegExpNode* SeqRegExpNode::FilterSuccessor(int depth) {
2691
RegExpNode* next = on_success_->FilterASCII(depth - 1);
2692
if (next == NULL) return set_replacement(NULL);
2694
return set_replacement(this);
2698
RegExpNode* TextNode::FilterASCII(int depth) {
2699
if (info()->replacement_calculated) return replacement();
2700
if (depth < 0) return this;
2701
ASSERT(!info()->visited);
2702
VisitMarker marker(info());
2703
int element_count = elms_->length();
2704
for (int i = 0; i < element_count; i++) {
2705
TextElement elm = elms_->at(i);
2706
if (elm.type == TextElement::ATOM) {
2707
Vector<const uc16> quarks = elm.data.u_atom->data();
2708
for (int j = 0; j < quarks.length(); j++) {
2709
// We don't need special handling for case independence
2710
// because of the rule that case independence cannot make
2711
// a non-ASCII character match an ASCII character.
2712
if (quarks[j] > String::kMaxAsciiCharCode) {
2713
return set_replacement(NULL);
2717
ASSERT(elm.type == TextElement::CHAR_CLASS);
2718
RegExpCharacterClass* cc = elm.data.u_char_class;
2719
ZoneList<CharacterRange>* ranges = cc->ranges(zone());
2720
if (!CharacterRange::IsCanonical(ranges)) {
2721
CharacterRange::Canonicalize(ranges);
2723
// Now they are in order so we only need to look at the first.
2724
int range_count = ranges->length();
2725
if (cc->is_negated()) {
2726
if (range_count != 0 &&
2727
ranges->at(0).from() == 0 &&
2728
ranges->at(0).to() >= String::kMaxAsciiCharCode) {
2729
return set_replacement(NULL);
2732
if (range_count == 0 ||
2733
ranges->at(0).from() > String::kMaxAsciiCharCode) {
2734
return set_replacement(NULL);
2739
return FilterSuccessor(depth - 1);
2743
RegExpNode* LoopChoiceNode::FilterASCII(int depth) {
2744
if (info()->replacement_calculated) return replacement();
2745
if (depth < 0) return this;
2746
if (info()->visited) return this;
2748
VisitMarker marker(info());
2750
RegExpNode* continue_replacement = continue_node_->FilterASCII(depth - 1);
2751
// If we can't continue after the loop then there is no sense in doing the
2753
if (continue_replacement == NULL) return set_replacement(NULL);
2756
return ChoiceNode::FilterASCII(depth - 1);
2760
RegExpNode* ChoiceNode::FilterASCII(int depth) {
2761
if (info()->replacement_calculated) return replacement();
2762
if (depth < 0) return this;
2763
if (info()->visited) return this;
2764
VisitMarker marker(info());
2765
int choice_count = alternatives_->length();
2767
for (int i = 0; i < choice_count; i++) {
2768
GuardedAlternative alternative = alternatives_->at(i);
2769
if (alternative.guards() != NULL && alternative.guards()->length() != 0) {
2770
set_replacement(this);
2776
RegExpNode* survivor = NULL;
2777
for (int i = 0; i < choice_count; i++) {
2778
GuardedAlternative alternative = alternatives_->at(i);
2779
RegExpNode* replacement = alternative.node()->FilterASCII(depth - 1);
2780
ASSERT(replacement != this); // No missing EMPTY_MATCH_CHECK.
2781
if (replacement != NULL) {
2782
alternatives_->at(i).set_node(replacement);
2784
survivor = replacement;
2787
if (surviving < 2) return set_replacement(survivor);
2789
set_replacement(this);
2790
if (surviving == choice_count) {
2793
// Only some of the nodes survived the filtering. We need to rebuild the
2794
// alternatives list.
2795
ZoneList<GuardedAlternative>* new_alternatives =
2796
new(zone()) ZoneList<GuardedAlternative>(surviving, zone());
2797
for (int i = 0; i < choice_count; i++) {
2798
RegExpNode* replacement =
2799
alternatives_->at(i).node()->FilterASCII(depth - 1);
2800
if (replacement != NULL) {
2801
alternatives_->at(i).set_node(replacement);
2802
new_alternatives->Add(alternatives_->at(i), zone());
2805
alternatives_ = new_alternatives;
2810
RegExpNode* NegativeLookaheadChoiceNode::FilterASCII(int depth) {
2811
if (info()->replacement_calculated) return replacement();
2812
if (depth < 0) return this;
2813
if (info()->visited) return this;
2814
VisitMarker marker(info());
2815
// Alternative 0 is the negative lookahead, alternative 1 is what comes
2817
RegExpNode* node = alternatives_->at(1).node();
2818
RegExpNode* replacement = node->FilterASCII(depth - 1);
2819
if (replacement == NULL) return set_replacement(NULL);
2820
alternatives_->at(1).set_node(replacement);
2822
RegExpNode* neg_node = alternatives_->at(0).node();
2823
RegExpNode* neg_replacement = neg_node->FilterASCII(depth - 1);
2824
// If the negative lookahead is always going to fail then
2825
// we don't need to check it.
2826
if (neg_replacement == NULL) return set_replacement(replacement);
2827
alternatives_->at(0).set_node(neg_replacement);
2828
return set_replacement(this);
2832
void LoopChoiceNode::GetQuickCheckDetails(QuickCheckDetails* details,
2833
RegExpCompiler* compiler,
2834
int characters_filled_in,
2835
bool not_at_start) {
2836
if (body_can_be_zero_length_ || info()->visited) return;
2837
VisitMarker marker(info());
2838
return ChoiceNode::GetQuickCheckDetails(details,
2840
characters_filled_in,
2845
void LoopChoiceNode::FillInBMInfo(int offset,
2846
int recursion_depth,
2848
BoyerMooreLookahead* bm,
2849
bool not_at_start) {
2850
if (body_can_be_zero_length_ ||
2851
recursion_depth > RegExpCompiler::kMaxRecursion ||
2853
bm->SetRest(offset);
2854
SaveBMInfo(bm, not_at_start, offset);
2857
ChoiceNode::FillInBMInfo(
2858
offset, recursion_depth + 1, budget - 1, bm, not_at_start);
2859
SaveBMInfo(bm, not_at_start, offset);
2863
void ChoiceNode::GetQuickCheckDetails(QuickCheckDetails* details,
2864
RegExpCompiler* compiler,
2865
int characters_filled_in,
2866
bool not_at_start) {
2867
not_at_start = (not_at_start || not_at_start_);
2868
int choice_count = alternatives_->length();
2869
ASSERT(choice_count > 0);
2870
alternatives_->at(0).node()->GetQuickCheckDetails(details,
2872
characters_filled_in,
2874
for (int i = 1; i < choice_count; i++) {
2875
QuickCheckDetails new_details(details->characters());
2876
RegExpNode* node = alternatives_->at(i).node();
2877
node->GetQuickCheckDetails(&new_details, compiler,
2878
characters_filled_in,
2880
// Here we merge the quick match details of the two branches.
2881
details->Merge(&new_details, characters_filled_in);
2886
// Check for [0-9A-Z_a-z].
2887
static void EmitWordCheck(RegExpMacroAssembler* assembler,
2890
bool fall_through_on_word) {
2891
if (assembler->CheckSpecialCharacterClass(
2892
fall_through_on_word ? 'w' : 'W',
2893
fall_through_on_word ? non_word : word)) {
2894
// Optimized implementation available.
2897
assembler->CheckCharacterGT('z', non_word);
2898
assembler->CheckCharacterLT('0', non_word);
2899
assembler->CheckCharacterGT('a' - 1, word);
2900
assembler->CheckCharacterLT('9' + 1, word);
2901
assembler->CheckCharacterLT('A', non_word);
2902
assembler->CheckCharacterLT('Z' + 1, word);
2903
if (fall_through_on_word) {
2904
assembler->CheckNotCharacter('_', non_word);
2906
assembler->CheckCharacter('_', word);
2911
// Emit the code to check for a ^ in multiline mode (1-character lookbehind
2912
// that matches newline or the start of input).
2913
static void EmitHat(RegExpCompiler* compiler,
2914
RegExpNode* on_success,
2916
RegExpMacroAssembler* assembler = compiler->macro_assembler();
2917
// We will be loading the previous character into the current character
2919
Trace new_trace(*trace);
2920
new_trace.InvalidateCurrentCharacter();
2923
if (new_trace.cp_offset() == 0) {
2924
// The start of input counts as a newline in this context, so skip to
2925
// ok if we are at the start.
2926
assembler->CheckAtStart(&ok);
2928
// We already checked that we are not at the start of input so it must be
2929
// OK to load the previous character.
2930
assembler->LoadCurrentCharacter(new_trace.cp_offset() -1,
2931
new_trace.backtrack(),
2933
if (!assembler->CheckSpecialCharacterClass('n',
2934
new_trace.backtrack())) {
2935
// Newline means \n, \r, 0x2028 or 0x2029.
2936
if (!compiler->ascii()) {
2937
assembler->CheckCharacterAfterAnd(0x2028, 0xfffe, &ok);
2939
assembler->CheckCharacter('\n', &ok);
2940
assembler->CheckNotCharacter('\r', new_trace.backtrack());
2942
assembler->Bind(&ok);
2943
on_success->Emit(compiler, &new_trace);
2947
// Emit the code to handle \b and \B (word-boundary or non-word-boundary).
2948
void AssertionNode::EmitBoundaryCheck(RegExpCompiler* compiler, Trace* trace) {
2949
RegExpMacroAssembler* assembler = compiler->macro_assembler();
2950
Trace::TriBool next_is_word_character = Trace::UNKNOWN;
2951
bool not_at_start = (trace->at_start() == Trace::FALSE);
2952
BoyerMooreLookahead* lookahead = bm_info(not_at_start);
2953
if (lookahead == NULL) {
2955
Min(kMaxLookaheadForBoyerMoore,
2956
EatsAtLeast(kMaxLookaheadForBoyerMoore, 0, not_at_start));
2957
if (eats_at_least >= 1) {
2958
BoyerMooreLookahead* bm =
2959
new(zone()) BoyerMooreLookahead(eats_at_least, compiler, zone());
2960
FillInBMInfo(0, 0, kFillInBMBudget, bm, not_at_start);
2961
if (bm->at(0)->is_non_word()) next_is_word_character = Trace::FALSE;
2962
if (bm->at(0)->is_word()) next_is_word_character = Trace::TRUE;
2965
if (lookahead->at(0)->is_non_word()) next_is_word_character = Trace::FALSE;
2966
if (lookahead->at(0)->is_word()) next_is_word_character = Trace::TRUE;
2968
bool at_boundary = (type_ == AssertionNode::AT_BOUNDARY);
2969
if (next_is_word_character == Trace::UNKNOWN) {
2970
Label before_non_word;
2972
if (trace->characters_preloaded() != 1) {
2973
assembler->LoadCurrentCharacter(trace->cp_offset(), &before_non_word);
2975
// Fall through on non-word.
2976
EmitWordCheck(assembler, &before_word, &before_non_word, false);
2977
// Next character is not a word character.
2978
assembler->Bind(&before_non_word);
2980
BacktrackIfPrevious(compiler, trace, at_boundary ? kIsNonWord : kIsWord);
2981
assembler->GoTo(&ok);
2983
assembler->Bind(&before_word);
2984
BacktrackIfPrevious(compiler, trace, at_boundary ? kIsWord : kIsNonWord);
2985
assembler->Bind(&ok);
2986
} else if (next_is_word_character == Trace::TRUE) {
2987
BacktrackIfPrevious(compiler, trace, at_boundary ? kIsWord : kIsNonWord);
2989
ASSERT(next_is_word_character == Trace::FALSE);
2990
BacktrackIfPrevious(compiler, trace, at_boundary ? kIsNonWord : kIsWord);
2995
void AssertionNode::BacktrackIfPrevious(
2996
RegExpCompiler* compiler,
2998
AssertionNode::IfPrevious backtrack_if_previous) {
2999
RegExpMacroAssembler* assembler = compiler->macro_assembler();
3000
Trace new_trace(*trace);
3001
new_trace.InvalidateCurrentCharacter();
3003
Label fall_through, dummy;
3005
Label* non_word = backtrack_if_previous == kIsNonWord ?
3006
new_trace.backtrack() :
3008
Label* word = backtrack_if_previous == kIsNonWord ?
3010
new_trace.backtrack();
3012
if (new_trace.cp_offset() == 0) {
3013
// The start of input counts as a non-word character, so the question is
3014
// decided if we are at the start.
3015
assembler->CheckAtStart(non_word);
3017
// We already checked that we are not at the start of input so it must be
3018
// OK to load the previous character.
3019
assembler->LoadCurrentCharacter(new_trace.cp_offset() - 1, &dummy, false);
3020
EmitWordCheck(assembler, word, non_word, backtrack_if_previous == kIsNonWord);
3022
assembler->Bind(&fall_through);
3023
on_success()->Emit(compiler, &new_trace);
3027
void AssertionNode::GetQuickCheckDetails(QuickCheckDetails* details,
3028
RegExpCompiler* compiler,
3030
bool not_at_start) {
3031
if (type_ == AT_START && not_at_start) {
3032
details->set_cannot_match();
3035
return on_success()->GetQuickCheckDetails(details,
3042
void AssertionNode::Emit(RegExpCompiler* compiler, Trace* trace) {
3043
RegExpMacroAssembler* assembler = compiler->macro_assembler();
3047
assembler->CheckPosition(trace->cp_offset(), &ok);
3048
assembler->GoTo(trace->backtrack());
3049
assembler->Bind(&ok);
3053
if (trace->at_start() == Trace::FALSE) {
3054
assembler->GoTo(trace->backtrack());
3057
if (trace->at_start() == Trace::UNKNOWN) {
3058
assembler->CheckNotAtStart(trace->backtrack());
3059
Trace at_start_trace = *trace;
3060
at_start_trace.set_at_start(true);
3061
on_success()->Emit(compiler, &at_start_trace);
3067
EmitHat(compiler, on_success(), trace);
3070
case AT_NON_BOUNDARY: {
3071
EmitBoundaryCheck(compiler, trace);
3075
on_success()->Emit(compiler, trace);
3079
static bool DeterminedAlready(QuickCheckDetails* quick_check, int offset) {
3080
if (quick_check == NULL) return false;
3081
if (offset >= quick_check->characters()) return false;
3082
return quick_check->positions(offset)->determines_perfectly;
3086
static void UpdateBoundsCheck(int index, int* checked_up_to) {
3087
if (index > *checked_up_to) {
3088
*checked_up_to = index;
3093
// We call this repeatedly to generate code for each pass over the text node.
3094
// The passes are in increasing order of difficulty because we hope one
3095
// of the first passes will fail in which case we are saved the work of the
3096
// later passes. for example for the case independent regexp /%[asdfghjkl]a/
3097
// we will check the '%' in the first pass, the case independent 'a' in the
3098
// second pass and the character class in the last pass.
3100
// The passes are done from right to left, so for example to test for /bar/
3101
// we will first test for an 'r' with offset 2, then an 'a' with offset 1
3102
// and then a 'b' with offset 0. This means we can avoid the end-of-input
3103
// bounds check most of the time. In the example we only need to check for
3104
// end-of-input when loading the putative 'r'.
3106
// A slight complication involves the fact that the first character may already
3107
// be fetched into a register by the previous node. In this case we want to
3108
// do the test for that character first. We do this in separate passes. The
3109
// 'preloaded' argument indicates that we are doing such a 'pass'. If such a
3110
// pass has been performed then subsequent passes will have true in
3111
// first_element_checked to indicate that that character does not need to be
3114
// In addition to all this we are passed a Trace, which can
3115
// contain an AlternativeGeneration object. In this AlternativeGeneration
3116
// object we can see details of any quick check that was already passed in
3117
// order to get to the code we are now generating. The quick check can involve
3118
// loading characters, which means we do not need to recheck the bounds
3119
// up to the limit the quick check already checked. In addition the quick
3120
// check can have involved a mask and compare operation which may simplify
3121
// or obviate the need for further checks at some character positions.
3122
void TextNode::TextEmitPass(RegExpCompiler* compiler,
3123
TextEmitPassType pass,
3126
bool first_element_checked,
3127
int* checked_up_to) {
3128
Isolate* isolate = Isolate::Current();
3129
RegExpMacroAssembler* assembler = compiler->macro_assembler();
3130
bool ascii = compiler->ascii();
3131
Label* backtrack = trace->backtrack();
3132
QuickCheckDetails* quick_check = trace->quick_check_performed();
3133
int element_count = elms_->length();
3134
for (int i = preloaded ? 0 : element_count - 1; i >= 0; i--) {
3135
TextElement elm = elms_->at(i);
3136
int cp_offset = trace->cp_offset() + elm.cp_offset;
3137
if (elm.type == TextElement::ATOM) {
3138
Vector<const uc16> quarks = elm.data.u_atom->data();
3139
for (int j = preloaded ? 0 : quarks.length() - 1; j >= 0; j--) {
3140
if (first_element_checked && i == 0 && j == 0) continue;
3141
if (DeterminedAlready(quick_check, elm.cp_offset + j)) continue;
3142
EmitCharacterFunction* emit_function = NULL;
3144
case NON_ASCII_MATCH:
3146
if (quarks[j] > String::kMaxAsciiCharCode) {
3147
assembler->GoTo(backtrack);
3151
case NON_LETTER_CHARACTER_MATCH:
3152
emit_function = &EmitAtomNonLetter;
3154
case SIMPLE_CHARACTER_MATCH:
3155
emit_function = &EmitSimpleCharacter;
3157
case CASE_CHARACTER_MATCH:
3158
emit_function = &EmitAtomLetter;
3163
if (emit_function != NULL) {
3164
bool bound_checked = emit_function(isolate,
3169
*checked_up_to < cp_offset + j,
3171
if (bound_checked) UpdateBoundsCheck(cp_offset + j, checked_up_to);
3175
ASSERT_EQ(elm.type, TextElement::CHAR_CLASS);
3176
if (pass == CHARACTER_CLASS_MATCH) {
3177
if (first_element_checked && i == 0) continue;
3178
if (DeterminedAlready(quick_check, elm.cp_offset)) continue;
3179
RegExpCharacterClass* cc = elm.data.u_char_class;
3180
EmitCharClass(assembler,
3185
*checked_up_to < cp_offset,
3188
UpdateBoundsCheck(cp_offset, checked_up_to);
3195
int TextNode::Length() {
3196
TextElement elm = elms_->last();
3197
ASSERT(elm.cp_offset >= 0);
3198
if (elm.type == TextElement::ATOM) {
3199
return elm.cp_offset + elm.data.u_atom->data().length();
3201
return elm.cp_offset + 1;
3206
bool TextNode::SkipPass(int int_pass, bool ignore_case) {
3207
TextEmitPassType pass = static_cast<TextEmitPassType>(int_pass);
3209
return pass == SIMPLE_CHARACTER_MATCH;
3211
return pass == NON_LETTER_CHARACTER_MATCH || pass == CASE_CHARACTER_MATCH;
3216
// This generates the code to match a text node. A text node can contain
3217
// straight character sequences (possibly to be matched in a case-independent
3218
// way) and character classes. For efficiency we do not do this in a single
3219
// pass from left to right. Instead we pass over the text node several times,
3220
// emitting code for some character positions every time. See the comment on
3221
// TextEmitPass for details.
3222
void TextNode::Emit(RegExpCompiler* compiler, Trace* trace) {
3223
LimitResult limit_result = LimitVersions(compiler, trace);
3224
if (limit_result == DONE) return;
3225
ASSERT(limit_result == CONTINUE);
3227
if (trace->cp_offset() + Length() > RegExpMacroAssembler::kMaxCPOffset) {
3228
compiler->SetRegExpTooBig();
3232
if (compiler->ascii()) {
3234
TextEmitPass(compiler, NON_ASCII_MATCH, false, trace, false, &dummy);
3237
bool first_elt_done = false;
3238
int bound_checked_to = trace->cp_offset() - 1;
3239
bound_checked_to += trace->bound_checked_up_to();
3241
// If a character is preloaded into the current character register then
3243
if (trace->characters_preloaded() == 1) {
3244
for (int pass = kFirstRealPass; pass <= kLastPass; pass++) {
3245
if (!SkipPass(pass, compiler->ignore_case())) {
3246
TextEmitPass(compiler,
3247
static_cast<TextEmitPassType>(pass),
3254
first_elt_done = true;
3257
for (int pass = kFirstRealPass; pass <= kLastPass; pass++) {
3258
if (!SkipPass(pass, compiler->ignore_case())) {
3259
TextEmitPass(compiler,
3260
static_cast<TextEmitPassType>(pass),
3268
Trace successor_trace(*trace);
3269
successor_trace.set_at_start(false);
3270
successor_trace.AdvanceCurrentPositionInTrace(Length(), compiler);
3271
RecursionCheck rc(compiler);
3272
on_success()->Emit(compiler, &successor_trace);
3276
void Trace::InvalidateCurrentCharacter() {
3277
characters_preloaded_ = 0;
3281
void Trace::AdvanceCurrentPositionInTrace(int by, RegExpCompiler* compiler) {
3283
// We don't have an instruction for shifting the current character register
3284
// down or for using a shifted value for anything so lets just forget that
3285
// we preloaded any characters into it.
3286
characters_preloaded_ = 0;
3287
// Adjust the offsets of the quick check performed information. This
3288
// information is used to find out what we already determined about the
3289
// characters by means of mask and compare.
3290
quick_check_performed_.Advance(by, compiler->ascii());
3292
if (cp_offset_ > RegExpMacroAssembler::kMaxCPOffset) {
3293
compiler->SetRegExpTooBig();
3296
bound_checked_up_to_ = Max(0, bound_checked_up_to_ - by);
3300
void TextNode::MakeCaseIndependent(bool is_ascii) {
3301
int element_count = elms_->length();
3302
for (int i = 0; i < element_count; i++) {
3303
TextElement elm = elms_->at(i);
3304
if (elm.type == TextElement::CHAR_CLASS) {
3305
RegExpCharacterClass* cc = elm.data.u_char_class;
3306
// None of the standard character classes is different in the case
3307
// independent case and it slows us down if we don't know that.
3308
if (cc->is_standard(zone())) continue;
3309
ZoneList<CharacterRange>* ranges = cc->ranges(zone());
3310
int range_count = ranges->length();
3311
for (int j = 0; j < range_count; j++) {
3312
ranges->at(j).AddCaseEquivalents(ranges, is_ascii, zone());
3319
int TextNode::GreedyLoopTextLength() {
3320
TextElement elm = elms_->at(elms_->length() - 1);
3321
if (elm.type == TextElement::CHAR_CLASS) {
3322
return elm.cp_offset + 1;
3324
return elm.cp_offset + elm.data.u_atom->data().length();
3329
RegExpNode* TextNode::GetSuccessorOfOmnivorousTextNode(
3330
RegExpCompiler* compiler) {
3331
if (elms_->length() != 1) return NULL;
3332
TextElement elm = elms_->at(0);
3333
if (elm.type != TextElement::CHAR_CLASS) return NULL;
3334
RegExpCharacterClass* node = elm.data.u_char_class;
3335
ZoneList<CharacterRange>* ranges = node->ranges(zone());
3336
if (!CharacterRange::IsCanonical(ranges)) {
3337
CharacterRange::Canonicalize(ranges);
3339
if (node->is_negated()) {
3340
return ranges->length() == 0 ? on_success() : NULL;
3342
if (ranges->length() != 1) return NULL;
3344
if (compiler->ascii()) {
3345
max_char = String::kMaxAsciiCharCode;
3347
max_char = String::kMaxUtf16CodeUnit;
3349
return ranges->at(0).IsEverything(max_char) ? on_success() : NULL;
3353
// Finds the fixed match length of a sequence of nodes that goes from
3354
// this alternative and back to this choice node. If there are variable
3355
// length nodes or other complications in the way then return a sentinel
3356
// value indicating that a greedy loop cannot be constructed.
3357
int ChoiceNode::GreedyLoopTextLengthForAlternative(
3358
GuardedAlternative* alternative) {
3360
RegExpNode* node = alternative->node();
3361
// Later we will generate code for all these text nodes using recursion
3362
// so we have to limit the max number.
3363
int recursion_depth = 0;
3364
while (node != this) {
3365
if (recursion_depth++ > RegExpCompiler::kMaxRecursion) {
3366
return kNodeIsTooComplexForGreedyLoops;
3368
int node_length = node->GreedyLoopTextLength();
3369
if (node_length == kNodeIsTooComplexForGreedyLoops) {
3370
return kNodeIsTooComplexForGreedyLoops;
3372
length += node_length;
3373
SeqRegExpNode* seq_node = static_cast<SeqRegExpNode*>(node);
3374
node = seq_node->on_success();
3380
void LoopChoiceNode::AddLoopAlternative(GuardedAlternative alt) {
3381
ASSERT_EQ(loop_node_, NULL);
3382
AddAlternative(alt);
3383
loop_node_ = alt.node();
3387
void LoopChoiceNode::AddContinueAlternative(GuardedAlternative alt) {
3388
ASSERT_EQ(continue_node_, NULL);
3389
AddAlternative(alt);
3390
continue_node_ = alt.node();
3394
void LoopChoiceNode::Emit(RegExpCompiler* compiler, Trace* trace) {
3395
RegExpMacroAssembler* macro_assembler = compiler->macro_assembler();
3396
if (trace->stop_node() == this) {
3398
GreedyLoopTextLengthForAlternative(&(alternatives_->at(0)));
3399
ASSERT(text_length != kNodeIsTooComplexForGreedyLoops);
3400
// Update the counter-based backtracking info on the stack. This is an
3401
// optimization for greedy loops (see below).
3402
ASSERT(trace->cp_offset() == text_length);
3403
macro_assembler->AdvanceCurrentPosition(text_length);
3404
macro_assembler->GoTo(trace->loop_label());
3407
ASSERT(trace->stop_node() == NULL);
3408
if (!trace->is_trivial()) {
3409
trace->Flush(compiler, this);
3412
ChoiceNode::Emit(compiler, trace);
3416
int ChoiceNode::CalculatePreloadCharacters(RegExpCompiler* compiler,
3417
int eats_at_least) {
3418
int preload_characters = Min(4, eats_at_least);
3419
if (compiler->macro_assembler()->CanReadUnaligned()) {
3420
bool ascii = compiler->ascii();
3422
if (preload_characters > 4) preload_characters = 4;
3423
// We can't preload 3 characters because there is no machine instruction
3424
// to do that. We can't just load 4 because we could be reading
3425
// beyond the end of the string, which could cause a memory fault.
3426
if (preload_characters == 3) preload_characters = 2;
3428
if (preload_characters > 2) preload_characters = 2;
3431
if (preload_characters > 1) preload_characters = 1;
3433
return preload_characters;
3437
// This class is used when generating the alternatives in a choice node. It
3438
// records the way the alternative is being code generated.
3439
class AlternativeGeneration: public Malloced {
3441
AlternativeGeneration()
3442
: possible_success(),
3443
expects_preload(false),
3445
quick_check_details() { }
3446
Label possible_success;
3447
bool expects_preload;
3449
QuickCheckDetails quick_check_details;
3453
// Creates a list of AlternativeGenerations. If the list has a reasonable
3454
// size then it is on the stack, otherwise the excess is on the heap.
3455
class AlternativeGenerationList {
3457
AlternativeGenerationList(int count, Zone* zone)
3458
: alt_gens_(count, zone) {
3459
for (int i = 0; i < count && i < kAFew; i++) {
3460
alt_gens_.Add(a_few_alt_gens_ + i, zone);
3462
for (int i = kAFew; i < count; i++) {
3463
alt_gens_.Add(new AlternativeGeneration(), zone);
3466
~AlternativeGenerationList() {
3467
for (int i = kAFew; i < alt_gens_.length(); i++) {
3468
delete alt_gens_[i];
3469
alt_gens_[i] = NULL;
3473
AlternativeGeneration* at(int i) {
3474
return alt_gens_[i];
3478
static const int kAFew = 10;
3479
ZoneList<AlternativeGeneration*> alt_gens_;
3480
AlternativeGeneration a_few_alt_gens_[kAFew];
3484
// The '2' variant is has inclusive from and exclusive to.
3485
static const int kSpaceRanges[] = { '\t', '\r' + 1, ' ', ' ' + 1, 0x00A0,
3486
0x00A1, 0x1680, 0x1681, 0x180E, 0x180F, 0x2000, 0x200B, 0x2028, 0x202A,
3487
0x202F, 0x2030, 0x205F, 0x2060, 0x3000, 0x3001, 0xFEFF, 0xFF00, 0x10000 };
3488
static const int kSpaceRangeCount = ARRAY_SIZE(kSpaceRanges);
3490
static const int kWordRanges[] = {
3491
'0', '9' + 1, 'A', 'Z' + 1, '_', '_' + 1, 'a', 'z' + 1, 0x10000 };
3492
static const int kWordRangeCount = ARRAY_SIZE(kWordRanges);
3493
static const int kDigitRanges[] = { '0', '9' + 1, 0x10000 };
3494
static const int kDigitRangeCount = ARRAY_SIZE(kDigitRanges);
3495
static const int kSurrogateRanges[] = { 0xd800, 0xe000, 0x10000 };
3496
static const int kSurrogateRangeCount = ARRAY_SIZE(kSurrogateRanges);
3497
static const int kLineTerminatorRanges[] = { 0x000A, 0x000B, 0x000D, 0x000E,
3498
0x2028, 0x202A, 0x10000 };
3499
static const int kLineTerminatorRangeCount = ARRAY_SIZE(kLineTerminatorRanges);
3502
void BoyerMoorePositionInfo::Set(int character) {
3503
SetInterval(Interval(character, character));
3507
void BoyerMoorePositionInfo::SetInterval(const Interval& interval) {
3508
s_ = AddRange(s_, kSpaceRanges, kSpaceRangeCount, interval);
3509
w_ = AddRange(w_, kWordRanges, kWordRangeCount, interval);
3510
d_ = AddRange(d_, kDigitRanges, kDigitRangeCount, interval);
3512
AddRange(surrogate_, kSurrogateRanges, kSurrogateRangeCount, interval);
3513
if (interval.to() - interval.from() >= kMapSize - 1) {
3514
if (map_count_ != kMapSize) {
3515
map_count_ = kMapSize;
3516
for (int i = 0; i < kMapSize; i++) map_->at(i) = true;
3520
for (int i = interval.from(); i <= interval.to(); i++) {
3521
int mod_character = (i & kMask);
3522
if (!map_->at(mod_character)) {
3524
map_->at(mod_character) = true;
3526
if (map_count_ == kMapSize) return;
3531
void BoyerMoorePositionInfo::SetAll() {
3532
s_ = w_ = d_ = kLatticeUnknown;
3533
if (map_count_ != kMapSize) {
3534
map_count_ = kMapSize;
3535
for (int i = 0; i < kMapSize; i++) map_->at(i) = true;
3540
BoyerMooreLookahead::BoyerMooreLookahead(
3541
int length, RegExpCompiler* compiler, Zone* zone)
3543
compiler_(compiler) {
3544
if (compiler->ascii()) {
3545
max_char_ = String::kMaxAsciiCharCode;
3547
max_char_ = String::kMaxUtf16CodeUnit;
3549
bitmaps_ = new(zone) ZoneList<BoyerMoorePositionInfo*>(length, zone);
3550
for (int i = 0; i < length; i++) {
3551
bitmaps_->Add(new(zone) BoyerMoorePositionInfo(zone), zone);
3556
// Find the longest range of lookahead that has the fewest number of different
3557
// characters that can occur at a given position. Since we are optimizing two
3558
// different parameters at once this is a tradeoff.
3559
bool BoyerMooreLookahead::FindWorthwhileInterval(int* from, int* to) {
3560
int biggest_points = 0;
3561
// If more than 32 characters out of 128 can occur it is unlikely that we can
3562
// be lucky enough to step forwards much of the time.
3563
const int kMaxMax = 32;
3564
for (int max_number_of_chars = 4;
3565
max_number_of_chars < kMaxMax;
3566
max_number_of_chars *= 2) {
3568
FindBestInterval(max_number_of_chars, biggest_points, from, to);
3570
if (biggest_points == 0) return false;
3575
// Find the highest-points range between 0 and length_ where the character
3576
// information is not too vague. 'Too vague' means that there are more than
3577
// max_number_of_chars that can occur at this position. Calculates the number
3578
// of points as the product of width-of-the-range and
3579
// probability-of-finding-one-of-the-characters, where the probability is
3580
// calculated using the frequency distribution of the sample subject string.
3581
int BoyerMooreLookahead::FindBestInterval(
3582
int max_number_of_chars, int old_biggest_points, int* from, int* to) {
3583
int biggest_points = old_biggest_points;
3584
static const int kSize = RegExpMacroAssembler::kTableSize;
3585
for (int i = 0; i < length_; ) {
3586
while (i < length_ && Count(i) > max_number_of_chars) i++;
3587
if (i == length_) break;
3588
int remembered_from = i;
3589
bool union_map[kSize];
3590
for (int j = 0; j < kSize; j++) union_map[j] = false;
3591
while (i < length_ && Count(i) <= max_number_of_chars) {
3592
BoyerMoorePositionInfo* map = bitmaps_->at(i);
3593
for (int j = 0; j < kSize; j++) union_map[j] |= map->at(j);
3597
for (int j = 0; j < kSize; j++) {
3599
// Add 1 to the frequency to give a small per-character boost for
3600
// the cases where our sampling is not good enough and many
3601
// characters have a frequency of zero. This means the frequency
3602
// can theoretically be up to 2*kSize though we treat it mostly as
3603
// a fraction of kSize.
3604
frequency += compiler_->frequency_collator()->Frequency(j) + 1;
3607
// We use the probability of skipping times the distance we are skipping to
3608
// judge the effectiveness of this. Actually we have a cut-off: By
3609
// dividing by 2 we switch off the skipping if the probability of skipping
3610
// is less than 50%. This is because the multibyte mask-and-compare
3611
// skipping in quickcheck is more likely to do well on this case.
3612
bool in_quickcheck_range = ((i - remembered_from < 4) ||
3613
(compiler_->ascii() ? remembered_from <= 4 : remembered_from <= 2));
3614
// Called 'probability' but it is only a rough estimate and can actually
3615
// be outside the 0-kSize range.
3616
int probability = (in_quickcheck_range ? kSize / 2 : kSize) - frequency;
3617
int points = (i - remembered_from) * probability;
3618
if (points > biggest_points) {
3619
*from = remembered_from;
3621
biggest_points = points;
3624
return biggest_points;
3628
// Take all the characters that will not prevent a successful match if they
3629
// occur in the subject string in the range between min_lookahead and
3630
// max_lookahead (inclusive) measured from the current position. If the
3631
// character at max_lookahead offset is not one of these characters, then we
3632
// can safely skip forwards by the number of characters in the range.
3633
int BoyerMooreLookahead::GetSkipTable(int min_lookahead,
3635
Handle<ByteArray> boolean_skip_table) {
3636
const int kSize = RegExpMacroAssembler::kTableSize;
3638
const int kSkipArrayEntry = 0;
3639
const int kDontSkipArrayEntry = 1;
3641
for (int i = 0; i < kSize; i++) {
3642
boolean_skip_table->set(i, kSkipArrayEntry);
3644
int skip = max_lookahead + 1 - min_lookahead;
3646
for (int i = max_lookahead; i >= min_lookahead; i--) {
3647
BoyerMoorePositionInfo* map = bitmaps_->at(i);
3648
for (int j = 0; j < kSize; j++) {
3650
boolean_skip_table->set(j, kDontSkipArrayEntry);
3659
// See comment above on the implementation of GetSkipTable.
3660
bool BoyerMooreLookahead::EmitSkipInstructions(RegExpMacroAssembler* masm) {
3661
const int kSize = RegExpMacroAssembler::kTableSize;
3663
int min_lookahead = 0;
3664
int max_lookahead = 0;
3666
if (!FindWorthwhileInterval(&min_lookahead, &max_lookahead)) return false;
3668
bool found_single_character = false;
3669
int single_character = 0;
3670
for (int i = max_lookahead; i >= min_lookahead; i--) {
3671
BoyerMoorePositionInfo* map = bitmaps_->at(i);
3672
if (map->map_count() > 1 ||
3673
(found_single_character && map->map_count() != 0)) {
3674
found_single_character = false;
3677
for (int j = 0; j < kSize; j++) {
3679
found_single_character = true;
3680
single_character = j;
3686
int lookahead_width = max_lookahead + 1 - min_lookahead;
3688
if (found_single_character && lookahead_width == 1 && max_lookahead < 3) {
3689
// The mask-compare can probably handle this better.
3693
if (found_single_character) {
3696
masm->LoadCurrentCharacter(max_lookahead, &cont, true);
3697
if (max_char_ > kSize) {
3698
masm->CheckCharacterAfterAnd(single_character,
3699
RegExpMacroAssembler::kTableMask,
3702
masm->CheckCharacter(single_character, &cont);
3704
masm->AdvanceCurrentPosition(lookahead_width);
3710
Handle<ByteArray> boolean_skip_table =
3711
FACTORY->NewByteArray(kSize, TENURED);
3712
int skip_distance = GetSkipTable(
3713
min_lookahead, max_lookahead, boolean_skip_table);
3714
ASSERT(skip_distance != 0);
3718
masm->LoadCurrentCharacter(max_lookahead, &cont, true);
3719
masm->CheckBitInTable(boolean_skip_table, &cont);
3720
masm->AdvanceCurrentPosition(skip_distance);
3728
/* Code generation for choice nodes.
3730
* We generate quick checks that do a mask and compare to eliminate a
3731
* choice. If the quick check succeeds then it jumps to the continuation to
3732
* do slow checks and check subsequent nodes. If it fails (the common case)
3733
* it falls through to the next choice.
3735
* Here is the desired flow graph. Nodes directly below each other imply
3736
* fallthrough. Alternatives 1 and 2 have quick checks. Alternative
3737
* 3 doesn't have a quick check so we have to call the slow check.
3738
* Nodes are marked Qn for quick checks and Sn for slow checks. The entire
3739
* regexp continuation is generated directly after the Sn node, up to the
3740
* next GoTo if we decide to reuse some already generated code. Some
3741
* nodes expect preload_characters to be preloaded into the current
3742
* character register. R nodes do this preloading. Vertices are marked
3743
* F for failures and S for success (possible success in the case of quick
3744
* nodes). L, V, < and > are used as arrow heads.
3778
* For greedy loops we reverse our expectation and expect to match rather
3779
* than fail. Therefore we want the loop code to look like this (U is the
3780
* unwind code that steps back in the greedy loop). The following alternatives
3781
* look the same as above.
3806
void ChoiceNode::Emit(RegExpCompiler* compiler, Trace* trace) {
3807
RegExpMacroAssembler* macro_assembler = compiler->macro_assembler();
3808
int choice_count = alternatives_->length();
3810
for (int i = 0; i < choice_count - 1; i++) {
3811
GuardedAlternative alternative = alternatives_->at(i);
3812
ZoneList<Guard*>* guards = alternative.guards();
3813
int guard_count = (guards == NULL) ? 0 : guards->length();
3814
for (int j = 0; j < guard_count; j++) {
3815
ASSERT(!trace->mentions_reg(guards->at(j)->reg()));
3820
LimitResult limit_result = LimitVersions(compiler, trace);
3821
if (limit_result == DONE) return;
3822
ASSERT(limit_result == CONTINUE);
3824
int new_flush_budget = trace->flush_budget() / choice_count;
3825
if (trace->flush_budget() == 0 && trace->actions() != NULL) {
3826
trace->Flush(compiler, this);
3830
RecursionCheck rc(compiler);
3832
Trace* current_trace = trace;
3834
int text_length = GreedyLoopTextLengthForAlternative(&(alternatives_->at(0)));
3835
bool greedy_loop = false;
3836
Label greedy_loop_label;
3837
Trace counter_backtrack_trace;
3838
counter_backtrack_trace.set_backtrack(&greedy_loop_label);
3839
if (not_at_start()) counter_backtrack_trace.set_at_start(false);
3841
if (choice_count > 1 && text_length != kNodeIsTooComplexForGreedyLoops) {
3842
// Here we have special handling for greedy loops containing only text nodes
3843
// and other simple nodes. These are handled by pushing the current
3844
// position on the stack and then incrementing the current position each
3845
// time around the switch. On backtrack we decrement the current position
3846
// and check it against the pushed value. This avoids pushing backtrack
3847
// information for each iteration of the loop, which could take up a lot of
3850
ASSERT(trace->stop_node() == NULL);
3851
macro_assembler->PushCurrentPosition();
3852
current_trace = &counter_backtrack_trace;
3853
Label greedy_match_failed;
3854
Trace greedy_match_trace;
3855
if (not_at_start()) greedy_match_trace.set_at_start(false);
3856
greedy_match_trace.set_backtrack(&greedy_match_failed);
3858
macro_assembler->Bind(&loop_label);
3859
greedy_match_trace.set_stop_node(this);
3860
greedy_match_trace.set_loop_label(&loop_label);
3861
alternatives_->at(0).node()->Emit(compiler, &greedy_match_trace);
3862
macro_assembler->Bind(&greedy_match_failed);
3865
Label second_choice; // For use in greedy matches.
3866
macro_assembler->Bind(&second_choice);
3868
int first_normal_choice = greedy_loop ? 1 : 0;
3870
bool not_at_start = current_trace->at_start() == Trace::FALSE;
3871
const int kEatsAtLeastNotYetInitialized = -1;
3872
int eats_at_least = kEatsAtLeastNotYetInitialized;
3874
bool skip_was_emitted = false;
3876
if (!greedy_loop && choice_count == 2) {
3877
GuardedAlternative alt1 = alternatives_->at(1);
3878
if (alt1.guards() == NULL || alt1.guards()->length() == 0) {
3879
RegExpNode* eats_anything_node = alt1.node();
3880
if (eats_anything_node->GetSuccessorOfOmnivorousTextNode(compiler) ==
3882
// At this point we know that we are at a non-greedy loop that will eat
3883
// any character one at a time. Any non-anchored regexp has such a
3884
// loop prepended to it in order to find where it starts. We look for
3885
// a pattern of the form ...abc... where we can look 6 characters ahead
3886
// and step forwards 3 if the character is not one of abc. Abc need
3887
// not be atoms, they can be any reasonably limited character class or
3888
// small alternation.
3889
ASSERT(trace->is_trivial()); // This is the case on LoopChoiceNodes.
3890
BoyerMooreLookahead* lookahead = bm_info(not_at_start);
3891
if (lookahead == NULL) {
3893
Min(kMaxLookaheadForBoyerMoore,
3894
EatsAtLeast(kMaxLookaheadForBoyerMoore, 0, not_at_start));
3895
if (eats_at_least >= 1) {
3896
BoyerMooreLookahead* bm =
3897
new(zone()) BoyerMooreLookahead(eats_at_least,
3900
GuardedAlternative alt0 = alternatives_->at(0);
3901
alt0.node()->FillInBMInfo(0, 0, kFillInBMBudget, bm, not_at_start);
3902
skip_was_emitted = bm->EmitSkipInstructions(macro_assembler);
3905
skip_was_emitted = lookahead->EmitSkipInstructions(macro_assembler);
3911
if (eats_at_least == kEatsAtLeastNotYetInitialized) {
3912
// Save some time by looking at most one machine word ahead.
3913
eats_at_least = EatsAtLeast(compiler->ascii() ? 4 : 2, 0, not_at_start);
3915
int preload_characters = CalculatePreloadCharacters(compiler, eats_at_least);
3917
bool preload_is_current = !skip_was_emitted &&
3918
(current_trace->characters_preloaded() == preload_characters);
3919
bool preload_has_checked_bounds = preload_is_current;
3921
AlternativeGenerationList alt_gens(choice_count, zone());
3923
// For now we just call all choices one after the other. The idea ultimately
3924
// is to use the Dispatch table to try only the relevant ones.
3925
for (int i = first_normal_choice; i < choice_count; i++) {
3926
GuardedAlternative alternative = alternatives_->at(i);
3927
AlternativeGeneration* alt_gen = alt_gens.at(i);
3928
alt_gen->quick_check_details.set_characters(preload_characters);
3929
ZoneList<Guard*>* guards = alternative.guards();
3930
int guard_count = (guards == NULL) ? 0 : guards->length();
3931
Trace new_trace(*current_trace);
3932
new_trace.set_characters_preloaded(preload_is_current ?
3933
preload_characters :
3935
if (preload_has_checked_bounds) {
3936
new_trace.set_bound_checked_up_to(preload_characters);
3938
new_trace.quick_check_performed()->Clear();
3939
if (not_at_start_) new_trace.set_at_start(Trace::FALSE);
3940
alt_gen->expects_preload = preload_is_current;
3941
bool generate_full_check_inline = false;
3942
if (FLAG_regexp_optimization &&
3943
try_to_emit_quick_check_for_alternative(i) &&
3944
alternative.node()->EmitQuickCheck(compiler,
3946
preload_has_checked_bounds,
3947
&alt_gen->possible_success,
3948
&alt_gen->quick_check_details,
3949
i < choice_count - 1)) {
3950
// Quick check was generated for this choice.
3951
preload_is_current = true;
3952
preload_has_checked_bounds = true;
3953
// On the last choice in the ChoiceNode we generated the quick
3954
// check to fall through on possible success. So now we need to
3955
// generate the full check inline.
3956
if (i == choice_count - 1) {
3957
macro_assembler->Bind(&alt_gen->possible_success);
3958
new_trace.set_quick_check_performed(&alt_gen->quick_check_details);
3959
new_trace.set_characters_preloaded(preload_characters);
3960
new_trace.set_bound_checked_up_to(preload_characters);
3961
generate_full_check_inline = true;
3963
} else if (alt_gen->quick_check_details.cannot_match()) {
3964
if (i == choice_count - 1 && !greedy_loop) {
3965
macro_assembler->GoTo(trace->backtrack());
3969
// No quick check was generated. Put the full code here.
3970
// If this is not the first choice then there could be slow checks from
3971
// previous cases that go here when they fail. There's no reason to
3972
// insist that they preload characters since the slow check we are about
3973
// to generate probably can't use it.
3974
if (i != first_normal_choice) {
3975
alt_gen->expects_preload = false;
3976
new_trace.InvalidateCurrentCharacter();
3978
if (i < choice_count - 1) {
3979
new_trace.set_backtrack(&alt_gen->after);
3981
generate_full_check_inline = true;
3983
if (generate_full_check_inline) {
3984
if (new_trace.actions() != NULL) {
3985
new_trace.set_flush_budget(new_flush_budget);
3987
for (int j = 0; j < guard_count; j++) {
3988
GenerateGuard(macro_assembler, guards->at(j), &new_trace);
3990
alternative.node()->Emit(compiler, &new_trace);
3991
preload_is_current = false;
3993
macro_assembler->Bind(&alt_gen->after);
3996
macro_assembler->Bind(&greedy_loop_label);
3997
// If we have unwound to the bottom then backtrack.
3998
macro_assembler->CheckGreedyLoop(trace->backtrack());
3999
// Otherwise try the second priority at an earlier position.
4000
macro_assembler->AdvanceCurrentPosition(-text_length);
4001
macro_assembler->GoTo(&second_choice);
4004
// At this point we need to generate slow checks for the alternatives where
4005
// the quick check was inlined. We can recognize these because the associated
4007
for (int i = first_normal_choice; i < choice_count - 1; i++) {
4008
AlternativeGeneration* alt_gen = alt_gens.at(i);
4009
Trace new_trace(*current_trace);
4010
// If there are actions to be flushed we have to limit how many times
4011
// they are flushed. Take the budget of the parent trace and distribute
4012
// it fairly amongst the children.
4013
if (new_trace.actions() != NULL) {
4014
new_trace.set_flush_budget(new_flush_budget);
4016
EmitOutOfLineContinuation(compiler,
4018
alternatives_->at(i),
4021
alt_gens.at(i + 1)->expects_preload);
4026
void ChoiceNode::EmitOutOfLineContinuation(RegExpCompiler* compiler,
4028
GuardedAlternative alternative,
4029
AlternativeGeneration* alt_gen,
4030
int preload_characters,
4031
bool next_expects_preload) {
4032
if (!alt_gen->possible_success.is_linked()) return;
4034
RegExpMacroAssembler* macro_assembler = compiler->macro_assembler();
4035
macro_assembler->Bind(&alt_gen->possible_success);
4036
Trace out_of_line_trace(*trace);
4037
out_of_line_trace.set_characters_preloaded(preload_characters);
4038
out_of_line_trace.set_quick_check_performed(&alt_gen->quick_check_details);
4039
if (not_at_start_) out_of_line_trace.set_at_start(Trace::FALSE);
4040
ZoneList<Guard*>* guards = alternative.guards();
4041
int guard_count = (guards == NULL) ? 0 : guards->length();
4042
if (next_expects_preload) {
4043
Label reload_current_char;
4044
out_of_line_trace.set_backtrack(&reload_current_char);
4045
for (int j = 0; j < guard_count; j++) {
4046
GenerateGuard(macro_assembler, guards->at(j), &out_of_line_trace);
4048
alternative.node()->Emit(compiler, &out_of_line_trace);
4049
macro_assembler->Bind(&reload_current_char);
4050
// Reload the current character, since the next quick check expects that.
4051
// We don't need to check bounds here because we only get into this
4052
// code through a quick check which already did the checked load.
4053
macro_assembler->LoadCurrentCharacter(trace->cp_offset(),
4056
preload_characters);
4057
macro_assembler->GoTo(&(alt_gen->after));
4059
out_of_line_trace.set_backtrack(&(alt_gen->after));
4060
for (int j = 0; j < guard_count; j++) {
4061
GenerateGuard(macro_assembler, guards->at(j), &out_of_line_trace);
4063
alternative.node()->Emit(compiler, &out_of_line_trace);
4068
void ActionNode::Emit(RegExpCompiler* compiler, Trace* trace) {
4069
RegExpMacroAssembler* assembler = compiler->macro_assembler();
4070
LimitResult limit_result = LimitVersions(compiler, trace);
4071
if (limit_result == DONE) return;
4072
ASSERT(limit_result == CONTINUE);
4074
RecursionCheck rc(compiler);
4077
case STORE_POSITION: {
4078
Trace::DeferredCapture
4079
new_capture(data_.u_position_register.reg,
4080
data_.u_position_register.is_capture,
4082
Trace new_trace = *trace;
4083
new_trace.add_action(&new_capture);
4084
on_success()->Emit(compiler, &new_trace);
4087
case INCREMENT_REGISTER: {
4088
Trace::DeferredIncrementRegister
4089
new_increment(data_.u_increment_register.reg);
4090
Trace new_trace = *trace;
4091
new_trace.add_action(&new_increment);
4092
on_success()->Emit(compiler, &new_trace);
4095
case SET_REGISTER: {
4096
Trace::DeferredSetRegister
4097
new_set(data_.u_store_register.reg, data_.u_store_register.value);
4098
Trace new_trace = *trace;
4099
new_trace.add_action(&new_set);
4100
on_success()->Emit(compiler, &new_trace);
4103
case CLEAR_CAPTURES: {
4104
Trace::DeferredClearCaptures
4105
new_capture(Interval(data_.u_clear_captures.range_from,
4106
data_.u_clear_captures.range_to));
4107
Trace new_trace = *trace;
4108
new_trace.add_action(&new_capture);
4109
on_success()->Emit(compiler, &new_trace);
4112
case BEGIN_SUBMATCH:
4113
if (!trace->is_trivial()) {
4114
trace->Flush(compiler, this);
4116
assembler->WriteCurrentPositionToRegister(
4117
data_.u_submatch.current_position_register, 0);
4118
assembler->WriteStackPointerToRegister(
4119
data_.u_submatch.stack_pointer_register);
4120
on_success()->Emit(compiler, trace);
4123
case EMPTY_MATCH_CHECK: {
4124
int start_pos_reg = data_.u_empty_match_check.start_register;
4126
int rep_reg = data_.u_empty_match_check.repetition_register;
4127
bool has_minimum = (rep_reg != RegExpCompiler::kNoRegister);
4128
bool know_dist = trace->GetStoredPosition(start_pos_reg, &stored_pos);
4129
if (know_dist && !has_minimum && stored_pos == trace->cp_offset()) {
4130
// If we know we haven't advanced and there is no minimum we
4131
// can just backtrack immediately.
4132
assembler->GoTo(trace->backtrack());
4133
} else if (know_dist && stored_pos < trace->cp_offset()) {
4134
// If we know we've advanced we can generate the continuation
4136
on_success()->Emit(compiler, trace);
4137
} else if (!trace->is_trivial()) {
4138
trace->Flush(compiler, this);
4140
Label skip_empty_check;
4141
// If we have a minimum number of repetitions we check the current
4142
// number first and skip the empty check if it's not enough.
4144
int limit = data_.u_empty_match_check.repetition_limit;
4145
assembler->IfRegisterLT(rep_reg, limit, &skip_empty_check);
4147
// If the match is empty we bail out, otherwise we fall through
4148
// to the on-success continuation.
4149
assembler->IfRegisterEqPos(data_.u_empty_match_check.start_register,
4150
trace->backtrack());
4151
assembler->Bind(&skip_empty_check);
4152
on_success()->Emit(compiler, trace);
4156
case POSITIVE_SUBMATCH_SUCCESS: {
4157
if (!trace->is_trivial()) {
4158
trace->Flush(compiler, this);
4161
assembler->ReadCurrentPositionFromRegister(
4162
data_.u_submatch.current_position_register);
4163
assembler->ReadStackPointerFromRegister(
4164
data_.u_submatch.stack_pointer_register);
4165
int clear_register_count = data_.u_submatch.clear_register_count;
4166
if (clear_register_count == 0) {
4167
on_success()->Emit(compiler, trace);
4170
int clear_registers_from = data_.u_submatch.clear_register_from;
4171
Label clear_registers_backtrack;
4172
Trace new_trace = *trace;
4173
new_trace.set_backtrack(&clear_registers_backtrack);
4174
on_success()->Emit(compiler, &new_trace);
4176
assembler->Bind(&clear_registers_backtrack);
4177
int clear_registers_to = clear_registers_from + clear_register_count - 1;
4178
assembler->ClearRegisters(clear_registers_from, clear_registers_to);
4180
ASSERT(trace->backtrack() == NULL);
4181
assembler->Backtrack();
4190
void BackReferenceNode::Emit(RegExpCompiler* compiler, Trace* trace) {
4191
RegExpMacroAssembler* assembler = compiler->macro_assembler();
4192
if (!trace->is_trivial()) {
4193
trace->Flush(compiler, this);
4197
LimitResult limit_result = LimitVersions(compiler, trace);
4198
if (limit_result == DONE) return;
4199
ASSERT(limit_result == CONTINUE);
4201
RecursionCheck rc(compiler);
4203
ASSERT_EQ(start_reg_ + 1, end_reg_);
4204
if (compiler->ignore_case()) {
4205
assembler->CheckNotBackReferenceIgnoreCase(start_reg_,
4206
trace->backtrack());
4208
assembler->CheckNotBackReference(start_reg_, trace->backtrack());
4210
on_success()->Emit(compiler, trace);
4214
// -------------------------------------------------------------------
4221
class DotPrinter: public NodeVisitor {
4223
explicit DotPrinter(bool ignore_case)
4224
: ignore_case_(ignore_case),
4225
stream_(&alloc_) { }
4226
void PrintNode(const char* label, RegExpNode* node);
4227
void Visit(RegExpNode* node);
4228
void PrintAttributes(RegExpNode* from);
4229
StringStream* stream() { return &stream_; }
4230
void PrintOnFailure(RegExpNode* from, RegExpNode* to);
4231
#define DECLARE_VISIT(Type) \
4232
virtual void Visit##Type(Type##Node* that);
4233
FOR_EACH_NODE_TYPE(DECLARE_VISIT)
4234
#undef DECLARE_VISIT
4237
HeapStringAllocator alloc_;
4238
StringStream stream_;
4242
void DotPrinter::PrintNode(const char* label, RegExpNode* node) {
4243
stream()->Add("digraph G {\n graph [label=\"");
4244
for (int i = 0; label[i]; i++) {
4247
stream()->Add("\\\\");
4250
stream()->Add("\"");
4253
stream()->Put(label[i]);
4257
stream()->Add("\"];\n");
4259
stream()->Add("}\n");
4260
printf("%s", *(stream()->ToCString()));
4264
void DotPrinter::Visit(RegExpNode* node) {
4265
if (node->info()->visited) return;
4266
node->info()->visited = true;
4271
void DotPrinter::PrintOnFailure(RegExpNode* from, RegExpNode* on_failure) {
4272
stream()->Add(" n%p -> n%p [style=dotted];\n", from, on_failure);
4277
class TableEntryBodyPrinter {
4279
TableEntryBodyPrinter(StringStream* stream, ChoiceNode* choice)
4280
: stream_(stream), choice_(choice) { }
4281
void Call(uc16 from, DispatchTable::Entry entry) {
4282
OutSet* out_set = entry.out_set();
4283
for (unsigned i = 0; i < OutSet::kFirstLimit; i++) {
4284
if (out_set->Get(i)) {
4285
stream()->Add(" n%p:s%io%i -> n%p;\n",
4289
choice()->alternatives()->at(i).node());
4294
StringStream* stream() { return stream_; }
4295
ChoiceNode* choice() { return choice_; }
4296
StringStream* stream_;
4297
ChoiceNode* choice_;
4301
class TableEntryHeaderPrinter {
4303
explicit TableEntryHeaderPrinter(StringStream* stream)
4304
: first_(true), stream_(stream) { }
4305
void Call(uc16 from, DispatchTable::Entry entry) {
4311
stream()->Add("{\\%k-\\%k|{", from, entry.to());
4312
OutSet* out_set = entry.out_set();
4314
for (unsigned i = 0; i < OutSet::kFirstLimit; i++) {
4315
if (out_set->Get(i)) {
4316
if (priority > 0) stream()->Add("|");
4317
stream()->Add("<s%io%i> %i", from, i, priority);
4321
stream()->Add("}}");
4326
StringStream* stream() { return stream_; }
4327
StringStream* stream_;
4331
class AttributePrinter {
4333
explicit AttributePrinter(DotPrinter* out)
4334
: out_(out), first_(true) { }
4335
void PrintSeparator() {
4339
out_->stream()->Add("|");
4342
void PrintBit(const char* name, bool value) {
4345
out_->stream()->Add("{%s}", name);
4347
void PrintPositive(const char* name, int value) {
4348
if (value < 0) return;
4350
out_->stream()->Add("{%s|%x}", name, value);
4358
void DotPrinter::PrintAttributes(RegExpNode* that) {
4359
stream()->Add(" a%p [shape=Mrecord, color=grey, fontcolor=grey, "
4360
"margin=0.1, fontsize=10, label=\"{",
4362
AttributePrinter printer(this);
4363
NodeInfo* info = that->info();
4364
printer.PrintBit("NI", info->follows_newline_interest);
4365
printer.PrintBit("WI", info->follows_word_interest);
4366
printer.PrintBit("SI", info->follows_start_interest);
4367
Label* label = that->label();
4368
if (label->is_bound())
4369
printer.PrintPositive("@", label->pos());
4370
stream()->Add("}\"];\n");
4371
stream()->Add(" a%p -> n%p [style=dashed, color=grey, "
4372
"arrowhead=none];\n", that, that);
4376
static const bool kPrintDispatchTable = false;
4377
void DotPrinter::VisitChoice(ChoiceNode* that) {
4378
if (kPrintDispatchTable) {
4379
stream()->Add(" n%p [shape=Mrecord, label=\"", that);
4380
TableEntryHeaderPrinter header_printer(stream());
4381
that->GetTable(ignore_case_)->ForEach(&header_printer);
4382
stream()->Add("\"]\n", that);
4383
PrintAttributes(that);
4384
TableEntryBodyPrinter body_printer(stream(), that);
4385
that->GetTable(ignore_case_)->ForEach(&body_printer);
4387
stream()->Add(" n%p [shape=Mrecord, label=\"?\"];\n", that);
4388
for (int i = 0; i < that->alternatives()->length(); i++) {
4389
GuardedAlternative alt = that->alternatives()->at(i);
4390
stream()->Add(" n%p -> n%p;\n", that, alt.node());
4393
for (int i = 0; i < that->alternatives()->length(); i++) {
4394
GuardedAlternative alt = that->alternatives()->at(i);
4395
alt.node()->Accept(this);
4400
void DotPrinter::VisitText(TextNode* that) {
4401
Zone* zone = that->zone();
4402
stream()->Add(" n%p [label=\"", that);
4403
for (int i = 0; i < that->elements()->length(); i++) {
4404
if (i > 0) stream()->Add(" ");
4405
TextElement elm = that->elements()->at(i);
4407
case TextElement::ATOM: {
4408
stream()->Add("'%w'", elm.data.u_atom->data());
4411
case TextElement::CHAR_CLASS: {
4412
RegExpCharacterClass* node = elm.data.u_char_class;
4414
if (node->is_negated())
4416
for (int j = 0; j < node->ranges(zone)->length(); j++) {
4417
CharacterRange range = node->ranges(zone)->at(j);
4418
stream()->Add("%k-%k", range.from(), range.to());
4427
stream()->Add("\", shape=box, peripheries=2];\n");
4428
PrintAttributes(that);
4429
stream()->Add(" n%p -> n%p;\n", that, that->on_success());
4430
Visit(that->on_success());
4434
void DotPrinter::VisitBackReference(BackReferenceNode* that) {
4435
stream()->Add(" n%p [label=\"$%i..$%i\", shape=doubleoctagon];\n",
4437
that->start_register(),
4438
that->end_register());
4439
PrintAttributes(that);
4440
stream()->Add(" n%p -> n%p;\n", that, that->on_success());
4441
Visit(that->on_success());
4445
void DotPrinter::VisitEnd(EndNode* that) {
4446
stream()->Add(" n%p [style=bold, shape=point];\n", that);
4447
PrintAttributes(that);
4451
void DotPrinter::VisitAssertion(AssertionNode* that) {
4452
stream()->Add(" n%p [", that);
4453
switch (that->type()) {
4454
case AssertionNode::AT_END:
4455
stream()->Add("label=\"$\", shape=septagon");
4457
case AssertionNode::AT_START:
4458
stream()->Add("label=\"^\", shape=septagon");
4460
case AssertionNode::AT_BOUNDARY:
4461
stream()->Add("label=\"\\b\", shape=septagon");
4463
case AssertionNode::AT_NON_BOUNDARY:
4464
stream()->Add("label=\"\\B\", shape=septagon");
4466
case AssertionNode::AFTER_NEWLINE:
4467
stream()->Add("label=\"(?<=\\n)\", shape=septagon");
4470
stream()->Add("];\n");
4471
PrintAttributes(that);
4472
RegExpNode* successor = that->on_success();
4473
stream()->Add(" n%p -> n%p;\n", that, successor);
4478
void DotPrinter::VisitAction(ActionNode* that) {
4479
stream()->Add(" n%p [", that);
4480
switch (that->type_) {
4481
case ActionNode::SET_REGISTER:
4482
stream()->Add("label=\"$%i:=%i\", shape=octagon",
4483
that->data_.u_store_register.reg,
4484
that->data_.u_store_register.value);
4486
case ActionNode::INCREMENT_REGISTER:
4487
stream()->Add("label=\"$%i++\", shape=octagon",
4488
that->data_.u_increment_register.reg);
4490
case ActionNode::STORE_POSITION:
4491
stream()->Add("label=\"$%i:=$pos\", shape=octagon",
4492
that->data_.u_position_register.reg);
4494
case ActionNode::BEGIN_SUBMATCH:
4495
stream()->Add("label=\"$%i:=$pos,begin\", shape=septagon",
4496
that->data_.u_submatch.current_position_register);
4498
case ActionNode::POSITIVE_SUBMATCH_SUCCESS:
4499
stream()->Add("label=\"escape\", shape=septagon");
4501
case ActionNode::EMPTY_MATCH_CHECK:
4502
stream()->Add("label=\"$%i=$pos?,$%i<%i?\", shape=septagon",
4503
that->data_.u_empty_match_check.start_register,
4504
that->data_.u_empty_match_check.repetition_register,
4505
that->data_.u_empty_match_check.repetition_limit);
4507
case ActionNode::CLEAR_CAPTURES: {
4508
stream()->Add("label=\"clear $%i to $%i\", shape=septagon",
4509
that->data_.u_clear_captures.range_from,
4510
that->data_.u_clear_captures.range_to);
4514
stream()->Add("];\n");
4515
PrintAttributes(that);
4516
RegExpNode* successor = that->on_success();
4517
stream()->Add(" n%p -> n%p;\n", that, successor);
4522
class DispatchTableDumper {
4524
explicit DispatchTableDumper(StringStream* stream) : stream_(stream) { }
4525
void Call(uc16 key, DispatchTable::Entry entry);
4526
StringStream* stream() { return stream_; }
4528
StringStream* stream_;
4532
void DispatchTableDumper::Call(uc16 key, DispatchTable::Entry entry) {
4533
stream()->Add("[%k-%k]: {", key, entry.to());
4534
OutSet* set = entry.out_set();
4536
for (unsigned i = 0; i < OutSet::kFirstLimit; i++) {
4541
stream()->Add(", ");
4543
stream()->Add("%i", i);
4546
stream()->Add("}\n");
4550
void DispatchTable::Dump() {
4551
HeapStringAllocator alloc;
4552
StringStream stream(&alloc);
4553
DispatchTableDumper dumper(&stream);
4554
tree()->ForEach(&dumper);
4555
OS::PrintError("%s", *stream.ToCString());
4559
void RegExpEngine::DotPrint(const char* label,
4562
DotPrinter printer(ignore_case);
4563
printer.PrintNode(label, node);
4570
// -------------------------------------------------------------------
4571
// Tree to graph conversion
4573
RegExpNode* RegExpAtom::ToNode(RegExpCompiler* compiler,
4574
RegExpNode* on_success) {
4575
ZoneList<TextElement>* elms =
4576
new(compiler->zone()) ZoneList<TextElement>(1, compiler->zone());
4577
elms->Add(TextElement::Atom(this), compiler->zone());
4578
return new(compiler->zone()) TextNode(elms, on_success);
4582
RegExpNode* RegExpText::ToNode(RegExpCompiler* compiler,
4583
RegExpNode* on_success) {
4584
return new(compiler->zone()) TextNode(elements(), on_success);
4588
static bool CompareInverseRanges(ZoneList<CharacterRange>* ranges,
4589
const int* special_class,
4591
length--; // Remove final 0x10000.
4592
ASSERT(special_class[length] == 0x10000);
4593
ASSERT(ranges->length() != 0);
4594
ASSERT(length != 0);
4595
ASSERT(special_class[0] != 0);
4596
if (ranges->length() != (length >> 1) + 1) {
4599
CharacterRange range = ranges->at(0);
4600
if (range.from() != 0) {
4603
for (int i = 0; i < length; i += 2) {
4604
if (special_class[i] != (range.to() + 1)) {
4607
range = ranges->at((i >> 1) + 1);
4608
if (special_class[i+1] != range.from()) {
4612
if (range.to() != 0xffff) {
4619
static bool CompareRanges(ZoneList<CharacterRange>* ranges,
4620
const int* special_class,
4622
length--; // Remove final 0x10000.
4623
ASSERT(special_class[length] == 0x10000);
4624
if (ranges->length() * 2 != length) {
4627
for (int i = 0; i < length; i += 2) {
4628
CharacterRange range = ranges->at(i >> 1);
4629
if (range.from() != special_class[i] ||
4630
range.to() != special_class[i + 1] - 1) {
4638
bool RegExpCharacterClass::is_standard(Zone* zone) {
4639
// TODO(lrn): Remove need for this function, by not throwing away information
4644
if (set_.is_standard()) {
4647
if (CompareRanges(set_.ranges(zone), kSpaceRanges, kSpaceRangeCount)) {
4648
set_.set_standard_set_type('s');
4651
if (CompareInverseRanges(set_.ranges(zone), kSpaceRanges, kSpaceRangeCount)) {
4652
set_.set_standard_set_type('S');
4655
if (CompareInverseRanges(set_.ranges(zone),
4656
kLineTerminatorRanges,
4657
kLineTerminatorRangeCount)) {
4658
set_.set_standard_set_type('.');
4661
if (CompareRanges(set_.ranges(zone),
4662
kLineTerminatorRanges,
4663
kLineTerminatorRangeCount)) {
4664
set_.set_standard_set_type('n');
4667
if (CompareRanges(set_.ranges(zone), kWordRanges, kWordRangeCount)) {
4668
set_.set_standard_set_type('w');
4671
if (CompareInverseRanges(set_.ranges(zone), kWordRanges, kWordRangeCount)) {
4672
set_.set_standard_set_type('W');
4679
RegExpNode* RegExpCharacterClass::ToNode(RegExpCompiler* compiler,
4680
RegExpNode* on_success) {
4681
return new(compiler->zone()) TextNode(this, on_success);
4685
RegExpNode* RegExpDisjunction::ToNode(RegExpCompiler* compiler,
4686
RegExpNode* on_success) {
4687
ZoneList<RegExpTree*>* alternatives = this->alternatives();
4688
int length = alternatives->length();
4689
ChoiceNode* result =
4690
new(compiler->zone()) ChoiceNode(length, compiler->zone());
4691
for (int i = 0; i < length; i++) {
4692
GuardedAlternative alternative(alternatives->at(i)->ToNode(compiler,
4694
result->AddAlternative(alternative);
4700
RegExpNode* RegExpQuantifier::ToNode(RegExpCompiler* compiler,
4701
RegExpNode* on_success) {
4702
return ToNode(min(),
4711
// Scoped object to keep track of how much we unroll quantifier loops in the
4712
// regexp graph generator.
4713
class RegExpExpansionLimiter {
4715
static const int kMaxExpansionFactor = 6;
4716
RegExpExpansionLimiter(RegExpCompiler* compiler, int factor)
4717
: compiler_(compiler),
4718
saved_expansion_factor_(compiler->current_expansion_factor()),
4719
ok_to_expand_(saved_expansion_factor_ <= kMaxExpansionFactor) {
4721
if (ok_to_expand_) {
4722
if (factor > kMaxExpansionFactor) {
4723
// Avoid integer overflow of the current expansion factor.
4724
ok_to_expand_ = false;
4725
compiler->set_current_expansion_factor(kMaxExpansionFactor + 1);
4727
int new_factor = saved_expansion_factor_ * factor;
4728
ok_to_expand_ = (new_factor <= kMaxExpansionFactor);
4729
compiler->set_current_expansion_factor(new_factor);
4734
~RegExpExpansionLimiter() {
4735
compiler_->set_current_expansion_factor(saved_expansion_factor_);
4738
bool ok_to_expand() { return ok_to_expand_; }
4741
RegExpCompiler* compiler_;
4742
int saved_expansion_factor_;
4745
DISALLOW_IMPLICIT_CONSTRUCTORS(RegExpExpansionLimiter);
4749
RegExpNode* RegExpQuantifier::ToNode(int min,
4753
RegExpCompiler* compiler,
4754
RegExpNode* on_success,
4755
bool not_at_start) {
4756
// x{f, t} becomes this:
4762
// (r=0)-->(?)---/ [if r < t]
4764
// [if r >= f] \----> ...
4767
// 15.10.2.5 RepeatMatcher algorithm.
4768
// The parser has already eliminated the case where max is 0. In the case
4769
// where max_match is zero the parser has removed the quantifier if min was
4770
// > 0 and removed the atom if min was 0. See AddQuantifierToAtom.
4772
// If we know that we cannot match zero length then things are a little
4773
// simpler since we don't need to make the special zero length match check
4774
// from step 2.1. If the min and max are small we can unroll a little in
4776
static const int kMaxUnrolledMinMatches = 3; // Unroll (foo)+ and (foo){3,}
4777
static const int kMaxUnrolledMaxMatches = 3; // Unroll (foo)? and (foo){x,3}
4778
if (max == 0) return on_success; // This can happen due to recursion.
4779
bool body_can_be_empty = (body->min_match() == 0);
4780
int body_start_reg = RegExpCompiler::kNoRegister;
4781
Interval capture_registers = body->CaptureRegisters();
4782
bool needs_capture_clearing = !capture_registers.is_empty();
4783
Zone* zone = compiler->zone();
4785
if (body_can_be_empty) {
4786
body_start_reg = compiler->AllocateRegister();
4787
} else if (FLAG_regexp_optimization && !needs_capture_clearing) {
4788
// Only unroll if there are no captures and the body can't be
4791
RegExpExpansionLimiter limiter(
4792
compiler, min + ((max != min) ? 1 : 0));
4793
if (min > 0 && min <= kMaxUnrolledMinMatches && limiter.ok_to_expand()) {
4794
int new_max = (max == kInfinity) ? max : max - min;
4795
// Recurse once to get the loop or optional matches after the fixed
4797
RegExpNode* answer = ToNode(
4798
0, new_max, is_greedy, body, compiler, on_success, true);
4799
// Unroll the forced matches from 0 to min. This can cause chains of
4800
// TextNodes (which the parser does not generate). These should be
4801
// combined if it turns out they hinder good code generation.
4802
for (int i = 0; i < min; i++) {
4803
answer = body->ToNode(compiler, answer);
4808
if (max <= kMaxUnrolledMaxMatches && min == 0) {
4809
ASSERT(max > 0); // Due to the 'if' above.
4810
RegExpExpansionLimiter limiter(compiler, max);
4811
if (limiter.ok_to_expand()) {
4812
// Unroll the optional matches up to max.
4813
RegExpNode* answer = on_success;
4814
for (int i = 0; i < max; i++) {
4815
ChoiceNode* alternation = new(zone) ChoiceNode(2, zone);
4817
alternation->AddAlternative(
4818
GuardedAlternative(body->ToNode(compiler, answer)));
4819
alternation->AddAlternative(GuardedAlternative(on_success));
4821
alternation->AddAlternative(GuardedAlternative(on_success));
4822
alternation->AddAlternative(
4823
GuardedAlternative(body->ToNode(compiler, answer)));
4825
answer = alternation;
4826
if (not_at_start) alternation->set_not_at_start();
4832
bool has_min = min > 0;
4833
bool has_max = max < RegExpTree::kInfinity;
4834
bool needs_counter = has_min || has_max;
4835
int reg_ctr = needs_counter
4836
? compiler->AllocateRegister()
4837
: RegExpCompiler::kNoRegister;
4838
LoopChoiceNode* center = new(zone) LoopChoiceNode(body->min_match() == 0,
4840
if (not_at_start) center->set_not_at_start();
4841
RegExpNode* loop_return = needs_counter
4842
? static_cast<RegExpNode*>(ActionNode::IncrementRegister(reg_ctr, center))
4843
: static_cast<RegExpNode*>(center);
4844
if (body_can_be_empty) {
4845
// If the body can be empty we need to check if it was and then
4847
loop_return = ActionNode::EmptyMatchCheck(body_start_reg,
4852
RegExpNode* body_node = body->ToNode(compiler, loop_return);
4853
if (body_can_be_empty) {
4854
// If the body can be empty we need to store the start position
4855
// so we can bail out if it was empty.
4856
body_node = ActionNode::StorePosition(body_start_reg, false, body_node);
4858
if (needs_capture_clearing) {
4859
// Before entering the body of this loop we need to clear captures.
4860
body_node = ActionNode::ClearCaptures(capture_registers, body_node);
4862
GuardedAlternative body_alt(body_node);
4865
new(zone) Guard(reg_ctr, Guard::LT, max);
4866
body_alt.AddGuard(body_guard, zone);
4868
GuardedAlternative rest_alt(on_success);
4870
Guard* rest_guard = new(compiler->zone()) Guard(reg_ctr, Guard::GEQ, min);
4871
rest_alt.AddGuard(rest_guard, zone);
4874
center->AddLoopAlternative(body_alt);
4875
center->AddContinueAlternative(rest_alt);
4877
center->AddContinueAlternative(rest_alt);
4878
center->AddLoopAlternative(body_alt);
4880
if (needs_counter) {
4881
return ActionNode::SetRegister(reg_ctr, 0, center);
4888
RegExpNode* RegExpAssertion::ToNode(RegExpCompiler* compiler,
4889
RegExpNode* on_success) {
4891
Zone* zone = compiler->zone();
4895
return AssertionNode::AfterNewline(on_success);
4896
case START_OF_INPUT:
4897
return AssertionNode::AtStart(on_success);
4899
return AssertionNode::AtBoundary(on_success);
4901
return AssertionNode::AtNonBoundary(on_success);
4903
return AssertionNode::AtEnd(on_success);
4905
// Compile $ in multiline regexps as an alternation with a positive
4906
// lookahead in one side and an end-of-input on the other side.
4907
// We need two registers for the lookahead.
4908
int stack_pointer_register = compiler->AllocateRegister();
4909
int position_register = compiler->AllocateRegister();
4910
// The ChoiceNode to distinguish between a newline and end-of-input.
4911
ChoiceNode* result = new(zone) ChoiceNode(2, zone);
4912
// Create a newline atom.
4913
ZoneList<CharacterRange>* newline_ranges =
4914
new(zone) ZoneList<CharacterRange>(3, zone);
4915
CharacterRange::AddClassEscape('n', newline_ranges, zone);
4916
RegExpCharacterClass* newline_atom = new(zone) RegExpCharacterClass('n');
4917
TextNode* newline_matcher = new(zone) TextNode(
4919
ActionNode::PositiveSubmatchSuccess(stack_pointer_register,
4921
0, // No captures inside.
4922
-1, // Ignored if no captures.
4924
// Create an end-of-input matcher.
4925
RegExpNode* end_of_line = ActionNode::BeginSubmatch(
4926
stack_pointer_register,
4929
// Add the two alternatives to the ChoiceNode.
4930
GuardedAlternative eol_alternative(end_of_line);
4931
result->AddAlternative(eol_alternative);
4932
GuardedAlternative end_alternative(AssertionNode::AtEnd(on_success));
4933
result->AddAlternative(end_alternative);
4943
RegExpNode* RegExpBackReference::ToNode(RegExpCompiler* compiler,
4944
RegExpNode* on_success) {
4945
return new(compiler->zone())
4946
BackReferenceNode(RegExpCapture::StartRegister(index()),
4947
RegExpCapture::EndRegister(index()),
4952
RegExpNode* RegExpEmpty::ToNode(RegExpCompiler* compiler,
4953
RegExpNode* on_success) {
4958
RegExpNode* RegExpLookahead::ToNode(RegExpCompiler* compiler,
4959
RegExpNode* on_success) {
4960
int stack_pointer_register = compiler->AllocateRegister();
4961
int position_register = compiler->AllocateRegister();
4963
const int registers_per_capture = 2;
4964
const int register_of_first_capture = 2;
4965
int register_count = capture_count_ * registers_per_capture;
4966
int register_start =
4967
register_of_first_capture + capture_from_ * registers_per_capture;
4969
RegExpNode* success;
4970
if (is_positive()) {
4971
RegExpNode* node = ActionNode::BeginSubmatch(
4972
stack_pointer_register,
4976
ActionNode::PositiveSubmatchSuccess(stack_pointer_register,
4983
// We use a ChoiceNode for a negative lookahead because it has most of
4984
// the characteristics we need. It has the body of the lookahead as its
4985
// first alternative and the expression after the lookahead of the second
4986
// alternative. If the first alternative succeeds then the
4987
// NegativeSubmatchSuccess will unwind the stack including everything the
4988
// choice node set up and backtrack. If the first alternative fails then
4989
// the second alternative is tried, which is exactly the desired result
4990
// for a negative lookahead. The NegativeLookaheadChoiceNode is a special
4991
// ChoiceNode that knows to ignore the first exit when calculating quick
4993
Zone* zone = compiler->zone();
4995
GuardedAlternative body_alt(
4998
success = new(zone) NegativeSubmatchSuccess(stack_pointer_register,
5003
ChoiceNode* choice_node =
5004
new(zone) NegativeLookaheadChoiceNode(body_alt,
5005
GuardedAlternative(on_success),
5007
return ActionNode::BeginSubmatch(stack_pointer_register,
5014
RegExpNode* RegExpCapture::ToNode(RegExpCompiler* compiler,
5015
RegExpNode* on_success) {
5016
return ToNode(body(), index(), compiler, on_success);
5020
RegExpNode* RegExpCapture::ToNode(RegExpTree* body,
5022
RegExpCompiler* compiler,
5023
RegExpNode* on_success) {
5024
int start_reg = RegExpCapture::StartRegister(index);
5025
int end_reg = RegExpCapture::EndRegister(index);
5026
RegExpNode* store_end = ActionNode::StorePosition(end_reg, true, on_success);
5027
RegExpNode* body_node = body->ToNode(compiler, store_end);
5028
return ActionNode::StorePosition(start_reg, true, body_node);
5032
RegExpNode* RegExpAlternative::ToNode(RegExpCompiler* compiler,
5033
RegExpNode* on_success) {
5034
ZoneList<RegExpTree*>* children = nodes();
5035
RegExpNode* current = on_success;
5036
for (int i = children->length() - 1; i >= 0; i--) {
5037
current = children->at(i)->ToNode(compiler, current);
5043
static void AddClass(const int* elmv,
5045
ZoneList<CharacterRange>* ranges,
5048
ASSERT(elmv[elmc] == 0x10000);
5049
for (int i = 0; i < elmc; i += 2) {
5050
ASSERT(elmv[i] < elmv[i + 1]);
5051
ranges->Add(CharacterRange(elmv[i], elmv[i + 1] - 1), zone);
5056
static void AddClassNegated(const int *elmv,
5058
ZoneList<CharacterRange>* ranges,
5061
ASSERT(elmv[elmc] == 0x10000);
5062
ASSERT(elmv[0] != 0x0000);
5063
ASSERT(elmv[elmc-1] != String::kMaxUtf16CodeUnit);
5065
for (int i = 0; i < elmc; i += 2) {
5066
ASSERT(last <= elmv[i] - 1);
5067
ASSERT(elmv[i] < elmv[i + 1]);
5068
ranges->Add(CharacterRange(last, elmv[i] - 1), zone);
5071
ranges->Add(CharacterRange(last, String::kMaxUtf16CodeUnit), zone);
5075
void CharacterRange::AddClassEscape(uc16 type,
5076
ZoneList<CharacterRange>* ranges,
5080
AddClass(kSpaceRanges, kSpaceRangeCount, ranges, zone);
5083
AddClassNegated(kSpaceRanges, kSpaceRangeCount, ranges, zone);
5086
AddClass(kWordRanges, kWordRangeCount, ranges, zone);
5089
AddClassNegated(kWordRanges, kWordRangeCount, ranges, zone);
5092
AddClass(kDigitRanges, kDigitRangeCount, ranges, zone);
5095
AddClassNegated(kDigitRanges, kDigitRangeCount, ranges, zone);
5098
AddClassNegated(kLineTerminatorRanges,
5099
kLineTerminatorRangeCount,
5103
// This is not a character range as defined by the spec but a
5104
// convenient shorthand for a character class that matches any
5107
ranges->Add(CharacterRange::Everything(), zone);
5109
// This is the set of characters matched by the $ and ^ symbols
5110
// in multiline mode.
5112
AddClass(kLineTerminatorRanges,
5113
kLineTerminatorRangeCount,
5123
Vector<const int> CharacterRange::GetWordBounds() {
5124
return Vector<const int>(kWordRanges, kWordRangeCount - 1);
5128
class CharacterRangeSplitter {
5130
CharacterRangeSplitter(ZoneList<CharacterRange>** included,
5131
ZoneList<CharacterRange>** excluded,
5133
: included_(included),
5134
excluded_(excluded),
5136
void Call(uc16 from, DispatchTable::Entry entry);
5138
static const int kInBase = 0;
5139
static const int kInOverlay = 1;
5142
ZoneList<CharacterRange>** included_;
5143
ZoneList<CharacterRange>** excluded_;
5148
void CharacterRangeSplitter::Call(uc16 from, DispatchTable::Entry entry) {
5149
if (!entry.out_set()->Get(kInBase)) return;
5150
ZoneList<CharacterRange>** target = entry.out_set()->Get(kInOverlay)
5153
if (*target == NULL) *target = new(zone_) ZoneList<CharacterRange>(2, zone_);
5154
(*target)->Add(CharacterRange(entry.from(), entry.to()), zone_);
5158
void CharacterRange::Split(ZoneList<CharacterRange>* base,
5159
Vector<const int> overlay,
5160
ZoneList<CharacterRange>** included,
5161
ZoneList<CharacterRange>** excluded,
5163
ASSERT_EQ(NULL, *included);
5164
ASSERT_EQ(NULL, *excluded);
5165
DispatchTable table(zone);
5166
for (int i = 0; i < base->length(); i++)
5167
table.AddRange(base->at(i), CharacterRangeSplitter::kInBase, zone);
5168
for (int i = 0; i < overlay.length(); i += 2) {
5169
table.AddRange(CharacterRange(overlay[i], overlay[i + 1] - 1),
5170
CharacterRangeSplitter::kInOverlay, zone);
5172
CharacterRangeSplitter callback(included, excluded, zone);
5173
table.ForEach(&callback);
5177
void CharacterRange::AddCaseEquivalents(ZoneList<CharacterRange>* ranges,
5180
Isolate* isolate = Isolate::Current();
5181
uc16 bottom = from();
5184
if (bottom > String::kMaxAsciiCharCode) return;
5185
if (top > String::kMaxAsciiCharCode) top = String::kMaxAsciiCharCode;
5187
unibrow::uchar chars[unibrow::Ecma262UnCanonicalize::kMaxWidth];
5188
if (top == bottom) {
5189
// If this is a singleton we just expand the one character.
5190
int length = isolate->jsregexp_uncanonicalize()->get(bottom, '\0', chars);
5191
for (int i = 0; i < length; i++) {
5192
uc32 chr = chars[i];
5193
if (chr != bottom) {
5194
ranges->Add(CharacterRange::Singleton(chars[i]), zone);
5198
// If this is a range we expand the characters block by block,
5199
// expanding contiguous subranges (blocks) one at a time.
5200
// The approach is as follows. For a given start character we
5201
// look up the remainder of the block that contains it (represented
5202
// by the end point), for instance we find 'z' if the character
5203
// is 'c'. A block is characterized by the property
5204
// that all characters uncanonicalize in the same way, except that
5205
// each entry in the result is incremented by the distance from the first
5206
// element. So a-z is a block because 'a' uncanonicalizes to ['a', 'A'] and
5207
// the k'th letter uncanonicalizes to ['a' + k, 'A' + k].
5208
// Once we've found the end point we look up its uncanonicalization
5209
// and produce a range for each element. For instance for [c-f]
5210
// we look up ['z', 'Z'] and produce [c-f] and [C-F]. We then only
5211
// add a range if it is not already contained in the input, so [c-f]
5212
// will be skipped but [C-F] will be added. If this range is not
5213
// completely contained in a block we do this for all the blocks
5214
// covered by the range (handling characters that is not in a block
5215
// as a "singleton block").
5216
unibrow::uchar range[unibrow::Ecma262UnCanonicalize::kMaxWidth];
5218
while (pos <= top) {
5219
int length = isolate->jsregexp_canonrange()->get(pos, '\0', range);
5224
ASSERT_EQ(1, length);
5225
block_end = range[0];
5227
int end = (block_end > top) ? top : block_end;
5228
length = isolate->jsregexp_uncanonicalize()->get(block_end, '\0', range);
5229
for (int i = 0; i < length; i++) {
5231
uc16 range_from = c - (block_end - pos);
5232
uc16 range_to = c - (block_end - end);
5233
if (!(bottom <= range_from && range_to <= top)) {
5234
ranges->Add(CharacterRange(range_from, range_to), zone);
5243
bool CharacterRange::IsCanonical(ZoneList<CharacterRange>* ranges) {
5244
ASSERT_NOT_NULL(ranges);
5245
int n = ranges->length();
5246
if (n <= 1) return true;
5247
int max = ranges->at(0).to();
5248
for (int i = 1; i < n; i++) {
5249
CharacterRange next_range = ranges->at(i);
5250
if (next_range.from() <= max + 1) return false;
5251
max = next_range.to();
5257
ZoneList<CharacterRange>* CharacterSet::ranges(Zone* zone) {
5258
if (ranges_ == NULL) {
5259
ranges_ = new(zone) ZoneList<CharacterRange>(2, zone);
5260
CharacterRange::AddClassEscape(standard_set_type_, ranges_, zone);
5266
// Move a number of elements in a zonelist to another position
5267
// in the same list. Handles overlapping source and target areas.
5268
static void MoveRanges(ZoneList<CharacterRange>* list,
5272
// Ranges are potentially overlapping.
5274
for (int i = count - 1; i >= 0; i--) {
5275
list->at(to + i) = list->at(from + i);
5278
for (int i = 0; i < count; i++) {
5279
list->at(to + i) = list->at(from + i);
5285
static int InsertRangeInCanonicalList(ZoneList<CharacterRange>* list,
5287
CharacterRange insert) {
5288
// Inserts a range into list[0..count[, which must be sorted
5289
// by from value and non-overlapping and non-adjacent, using at most
5290
// list[0..count] for the result. Returns the number of resulting
5291
// canonicalized ranges. Inserting a range may collapse existing ranges into
5292
// fewer ranges, so the return value can be anything in the range 1..count+1.
5293
uc16 from = insert.from();
5294
uc16 to = insert.to();
5296
int end_pos = count;
5297
for (int i = count - 1; i >= 0; i--) {
5298
CharacterRange current = list->at(i);
5299
if (current.from() > to + 1) {
5301
} else if (current.to() + 1 < from) {
5307
// Inserted range overlaps, or is adjacent to, ranges at positions
5308
// [start_pos..end_pos[. Ranges before start_pos or at or after end_pos are
5309
// not affected by the insertion.
5310
// If start_pos == end_pos, the range must be inserted before start_pos.
5311
// if start_pos < end_pos, the entire range from start_pos to end_pos
5312
// must be merged with the insert range.
5314
if (start_pos == end_pos) {
5315
// Insert between existing ranges at position start_pos.
5316
if (start_pos < count) {
5317
MoveRanges(list, start_pos, start_pos + 1, count - start_pos);
5319
list->at(start_pos) = insert;
5322
if (start_pos + 1 == end_pos) {
5323
// Replace single existing range at position start_pos.
5324
CharacterRange to_replace = list->at(start_pos);
5325
int new_from = Min(to_replace.from(), from);
5326
int new_to = Max(to_replace.to(), to);
5327
list->at(start_pos) = CharacterRange(new_from, new_to);
5330
// Replace a number of existing ranges from start_pos to end_pos - 1.
5331
// Move the remaining ranges down.
5333
int new_from = Min(list->at(start_pos).from(), from);
5334
int new_to = Max(list->at(end_pos - 1).to(), to);
5335
if (end_pos < count) {
5336
MoveRanges(list, end_pos, start_pos + 1, count - end_pos);
5338
list->at(start_pos) = CharacterRange(new_from, new_to);
5339
return count - (end_pos - start_pos) + 1;
5343
void CharacterSet::Canonicalize() {
5344
// Special/default classes are always considered canonical. The result
5345
// of calling ranges() will be sorted.
5346
if (ranges_ == NULL) return;
5347
CharacterRange::Canonicalize(ranges_);
5351
void CharacterRange::Canonicalize(ZoneList<CharacterRange>* character_ranges) {
5352
if (character_ranges->length() <= 1) return;
5353
// Check whether ranges are already canonical (increasing, non-overlapping,
5355
int n = character_ranges->length();
5356
int max = character_ranges->at(0).to();
5359
CharacterRange current = character_ranges->at(i);
5360
if (current.from() <= max + 1) {
5366
// Canonical until the i'th range. If that's all of them, we are done.
5369
// The ranges at index i and forward are not canonicalized. Make them so by
5370
// doing the equivalent of insertion sort (inserting each into the previous
5372
// Notice that inserting a range can reduce the number of ranges in the
5373
// result due to combining of adjacent and overlapping ranges.
5374
int read = i; // Range to insert.
5375
int num_canonical = i; // Length of canonicalized part of list.
5377
num_canonical = InsertRangeInCanonicalList(character_ranges,
5379
character_ranges->at(read));
5382
character_ranges->Rewind(num_canonical);
5384
ASSERT(CharacterRange::IsCanonical(character_ranges));
5388
void CharacterRange::Negate(ZoneList<CharacterRange>* ranges,
5389
ZoneList<CharacterRange>* negated_ranges,
5391
ASSERT(CharacterRange::IsCanonical(ranges));
5392
ASSERT_EQ(0, negated_ranges->length());
5393
int range_count = ranges->length();
5396
if (range_count > 0 && ranges->at(0).from() == 0) {
5397
from = ranges->at(0).to();
5400
while (i < range_count) {
5401
CharacterRange range = ranges->at(i);
5402
negated_ranges->Add(CharacterRange(from + 1, range.from() - 1), zone);
5406
if (from < String::kMaxUtf16CodeUnit) {
5407
negated_ranges->Add(CharacterRange(from + 1, String::kMaxUtf16CodeUnit),
5413
// -------------------------------------------------------------------
5417
OutSet* OutSet::Extend(unsigned value, Zone* zone) {
5420
if (successors(zone) != NULL) {
5421
for (int i = 0; i < successors(zone)->length(); i++) {
5422
OutSet* successor = successors(zone)->at(i);
5423
if (successor->Get(value))
5427
successors_ = new(zone) ZoneList<OutSet*>(2, zone);
5429
OutSet* result = new(zone) OutSet(first_, remaining_);
5430
result->Set(value, zone);
5431
successors(zone)->Add(result, zone);
5436
void OutSet::Set(unsigned value, Zone *zone) {
5437
if (value < kFirstLimit) {
5438
first_ |= (1 << value);
5440
if (remaining_ == NULL)
5441
remaining_ = new(zone) ZoneList<unsigned>(1, zone);
5442
if (remaining_->is_empty() || !remaining_->Contains(value))
5443
remaining_->Add(value, zone);
5448
bool OutSet::Get(unsigned value) {
5449
if (value < kFirstLimit) {
5450
return (first_ & (1 << value)) != 0;
5451
} else if (remaining_ == NULL) {
5454
return remaining_->Contains(value);
5459
const uc16 DispatchTable::Config::kNoKey = unibrow::Utf8::kBadChar;
5462
void DispatchTable::AddRange(CharacterRange full_range, int value,
5464
CharacterRange current = full_range;
5465
if (tree()->is_empty()) {
5466
// If this is the first range we just insert into the table.
5467
ZoneSplayTree<Config>::Locator loc;
5468
ASSERT_RESULT(tree()->Insert(current.from(), &loc));
5469
loc.set_value(Entry(current.from(), current.to(),
5470
empty()->Extend(value, zone)));
5473
// First see if there is a range to the left of this one that
5475
ZoneSplayTree<Config>::Locator loc;
5476
if (tree()->FindGreatestLessThan(current.from(), &loc)) {
5477
Entry* entry = &loc.value();
5478
// If we've found a range that overlaps with this one, and it
5479
// starts strictly to the left of this one, we have to fix it
5480
// because the following code only handles ranges that start on
5481
// or after the start point of the range we're adding.
5482
if (entry->from() < current.from() && entry->to() >= current.from()) {
5483
// Snap the overlapping range in half around the start point of
5484
// the range we're adding.
5485
CharacterRange left(entry->from(), current.from() - 1);
5486
CharacterRange right(current.from(), entry->to());
5487
// The left part of the overlapping range doesn't overlap.
5488
// Truncate the whole entry to be just the left part.
5489
entry->set_to(left.to());
5490
// The right part is the one that overlaps. We add this part
5491
// to the map and let the next step deal with merging it with
5492
// the range we're adding.
5493
ZoneSplayTree<Config>::Locator loc;
5494
ASSERT_RESULT(tree()->Insert(right.from(), &loc));
5495
loc.set_value(Entry(right.from(),
5500
while (current.is_valid()) {
5501
if (tree()->FindLeastGreaterThan(current.from(), &loc) &&
5502
(loc.value().from() <= current.to()) &&
5503
(loc.value().to() >= current.from())) {
5504
Entry* entry = &loc.value();
5505
// We have overlap. If there is space between the start point of
5506
// the range we're adding and where the overlapping range starts
5507
// then we have to add a range covering just that space.
5508
if (current.from() < entry->from()) {
5509
ZoneSplayTree<Config>::Locator ins;
5510
ASSERT_RESULT(tree()->Insert(current.from(), &ins));
5511
ins.set_value(Entry(current.from(),
5513
empty()->Extend(value, zone)));
5514
current.set_from(entry->from());
5516
ASSERT_EQ(current.from(), entry->from());
5517
// If the overlapping range extends beyond the one we want to add
5518
// we have to snap the right part off and add it separately.
5519
if (entry->to() > current.to()) {
5520
ZoneSplayTree<Config>::Locator ins;
5521
ASSERT_RESULT(tree()->Insert(current.to() + 1, &ins));
5522
ins.set_value(Entry(current.to() + 1,
5525
entry->set_to(current.to());
5527
ASSERT(entry->to() <= current.to());
5528
// The overlapping range is now completely contained by the range
5529
// we're adding so we can just update it and move the start point
5530
// of the range we're adding just past it.
5531
entry->AddValue(value, zone);
5532
// Bail out if the last interval ended at 0xFFFF since otherwise
5533
// adding 1 will wrap around to 0.
5534
if (entry->to() == String::kMaxUtf16CodeUnit)
5536
ASSERT(entry->to() + 1 > current.from());
5537
current.set_from(entry->to() + 1);
5539
// There is no overlap so we can just add the range
5540
ZoneSplayTree<Config>::Locator ins;
5541
ASSERT_RESULT(tree()->Insert(current.from(), &ins));
5542
ins.set_value(Entry(current.from(),
5544
empty()->Extend(value, zone)));
5551
OutSet* DispatchTable::Get(uc16 value) {
5552
ZoneSplayTree<Config>::Locator loc;
5553
if (!tree()->FindGreatestLessThan(value, &loc))
5555
Entry* entry = &loc.value();
5556
if (value <= entry->to())
5557
return entry->out_set();
5563
// -------------------------------------------------------------------
5567
void Analysis::EnsureAnalyzed(RegExpNode* that) {
5568
StackLimitCheck check(Isolate::Current());
5569
if (check.HasOverflowed()) {
5570
fail("Stack overflow");
5573
if (that->info()->been_analyzed || that->info()->being_analyzed)
5575
that->info()->being_analyzed = true;
5577
that->info()->being_analyzed = false;
5578
that->info()->been_analyzed = true;
5582
void Analysis::VisitEnd(EndNode* that) {
5587
void TextNode::CalculateOffsets() {
5588
int element_count = elements()->length();
5589
// Set up the offsets of the elements relative to the start. This is a fixed
5590
// quantity since a TextNode can only contain fixed-width things.
5592
for (int i = 0; i < element_count; i++) {
5593
TextElement& elm = elements()->at(i);
5594
elm.cp_offset = cp_offset;
5595
if (elm.type == TextElement::ATOM) {
5596
cp_offset += elm.data.u_atom->data().length();
5604
void Analysis::VisitText(TextNode* that) {
5606
that->MakeCaseIndependent(is_ascii_);
5608
EnsureAnalyzed(that->on_success());
5609
if (!has_failed()) {
5610
that->CalculateOffsets();
5615
void Analysis::VisitAction(ActionNode* that) {
5616
RegExpNode* target = that->on_success();
5617
EnsureAnalyzed(target);
5618
if (!has_failed()) {
5619
// If the next node is interested in what it follows then this node
5620
// has to be interested too so it can pass the information on.
5621
that->info()->AddFromFollowing(target->info());
5626
void Analysis::VisitChoice(ChoiceNode* that) {
5627
NodeInfo* info = that->info();
5628
for (int i = 0; i < that->alternatives()->length(); i++) {
5629
RegExpNode* node = that->alternatives()->at(i).node();
5630
EnsureAnalyzed(node);
5631
if (has_failed()) return;
5632
// Anything the following nodes need to know has to be known by
5633
// this node also, so it can pass it on.
5634
info->AddFromFollowing(node->info());
5639
void Analysis::VisitLoopChoice(LoopChoiceNode* that) {
5640
NodeInfo* info = that->info();
5641
for (int i = 0; i < that->alternatives()->length(); i++) {
5642
RegExpNode* node = that->alternatives()->at(i).node();
5643
if (node != that->loop_node()) {
5644
EnsureAnalyzed(node);
5645
if (has_failed()) return;
5646
info->AddFromFollowing(node->info());
5649
// Check the loop last since it may need the value of this node
5650
// to get a correct result.
5651
EnsureAnalyzed(that->loop_node());
5652
if (!has_failed()) {
5653
info->AddFromFollowing(that->loop_node()->info());
5658
void Analysis::VisitBackReference(BackReferenceNode* that) {
5659
EnsureAnalyzed(that->on_success());
5663
void Analysis::VisitAssertion(AssertionNode* that) {
5664
EnsureAnalyzed(that->on_success());
5668
void BackReferenceNode::FillInBMInfo(int offset,
5669
int recursion_depth,
5671
BoyerMooreLookahead* bm,
5672
bool not_at_start) {
5673
// Working out the set of characters that a backreference can match is too
5674
// hard, so we just say that any character can match.
5675
bm->SetRest(offset);
5676
SaveBMInfo(bm, not_at_start, offset);
5680
STATIC_ASSERT(BoyerMoorePositionInfo::kMapSize ==
5681
RegExpMacroAssembler::kTableSize);
5684
void ChoiceNode::FillInBMInfo(int offset,
5685
int recursion_depth,
5687
BoyerMooreLookahead* bm,
5688
bool not_at_start) {
5689
ZoneList<GuardedAlternative>* alts = alternatives();
5690
budget = (budget - 1) / alts->length();
5691
for (int i = 0; i < alts->length(); i++) {
5692
GuardedAlternative& alt = alts->at(i);
5693
if (alt.guards() != NULL && alt.guards()->length() != 0) {
5694
bm->SetRest(offset); // Give up trying to fill in info.
5695
SaveBMInfo(bm, not_at_start, offset);
5698
alt.node()->FillInBMInfo(
5699
offset, recursion_depth + 1, budget, bm, not_at_start);
5701
SaveBMInfo(bm, not_at_start, offset);
5705
void TextNode::FillInBMInfo(int initial_offset,
5706
int recursion_depth,
5708
BoyerMooreLookahead* bm,
5709
bool not_at_start) {
5710
if (initial_offset >= bm->length()) return;
5711
int offset = initial_offset;
5712
int max_char = bm->max_char();
5713
for (int i = 0; i < elements()->length(); i++) {
5714
if (offset >= bm->length()) {
5715
if (initial_offset == 0) set_bm_info(not_at_start, bm);
5718
TextElement text = elements()->at(i);
5719
if (text.type == TextElement::ATOM) {
5720
RegExpAtom* atom = text.data.u_atom;
5721
for (int j = 0; j < atom->length(); j++, offset++) {
5722
if (offset >= bm->length()) {
5723
if (initial_offset == 0) set_bm_info(not_at_start, bm);
5726
uc16 character = atom->data()[j];
5727
if (bm->compiler()->ignore_case()) {
5728
unibrow::uchar chars[unibrow::Ecma262UnCanonicalize::kMaxWidth];
5729
int length = GetCaseIndependentLetters(
5732
bm->max_char() == String::kMaxAsciiCharCode,
5734
for (int j = 0; j < length; j++) {
5735
bm->Set(offset, chars[j]);
5738
if (character <= max_char) bm->Set(offset, character);
5742
ASSERT(text.type == TextElement::CHAR_CLASS);
5743
RegExpCharacterClass* char_class = text.data.u_char_class;
5744
ZoneList<CharacterRange>* ranges = char_class->ranges(zone());
5745
if (char_class->is_negated()) {
5748
for (int k = 0; k < ranges->length(); k++) {
5749
CharacterRange& range = ranges->at(k);
5750
if (range.from() > max_char) continue;
5751
int to = Min(max_char, static_cast<int>(range.to()));
5752
bm->SetInterval(offset, Interval(range.from(), to));
5758
if (offset >= bm->length()) {
5759
if (initial_offset == 0) set_bm_info(not_at_start, bm);
5762
on_success()->FillInBMInfo(offset,
5763
recursion_depth + 1,
5766
true); // Not at start after a text node.
5767
if (initial_offset == 0) set_bm_info(not_at_start, bm);
5771
// -------------------------------------------------------------------
5772
// Dispatch table construction
5775
void DispatchTableConstructor::VisitEnd(EndNode* that) {
5776
AddRange(CharacterRange::Everything());
5780
void DispatchTableConstructor::BuildTable(ChoiceNode* node) {
5781
node->set_being_calculated(true);
5782
ZoneList<GuardedAlternative>* alternatives = node->alternatives();
5783
for (int i = 0; i < alternatives->length(); i++) {
5784
set_choice_index(i);
5785
alternatives->at(i).node()->Accept(this);
5787
node->set_being_calculated(false);
5791
class AddDispatchRange {
5793
explicit AddDispatchRange(DispatchTableConstructor* constructor)
5794
: constructor_(constructor) { }
5795
void Call(uc32 from, DispatchTable::Entry entry);
5797
DispatchTableConstructor* constructor_;
5801
void AddDispatchRange::Call(uc32 from, DispatchTable::Entry entry) {
5802
CharacterRange range(from, entry.to());
5803
constructor_->AddRange(range);
5807
void DispatchTableConstructor::VisitChoice(ChoiceNode* node) {
5808
if (node->being_calculated())
5810
DispatchTable* table = node->GetTable(ignore_case_);
5811
AddDispatchRange adder(this);
5812
table->ForEach(&adder);
5816
void DispatchTableConstructor::VisitBackReference(BackReferenceNode* that) {
5817
// TODO(160): Find the node that we refer back to and propagate its start
5818
// set back to here. For now we just accept anything.
5819
AddRange(CharacterRange::Everything());
5823
void DispatchTableConstructor::VisitAssertion(AssertionNode* that) {
5824
RegExpNode* target = that->on_success();
5825
target->Accept(this);
5829
static int CompareRangeByFrom(const CharacterRange* a,
5830
const CharacterRange* b) {
5831
return Compare<uc16>(a->from(), b->from());
5835
void DispatchTableConstructor::AddInverse(ZoneList<CharacterRange>* ranges) {
5836
ranges->Sort(CompareRangeByFrom);
5838
for (int i = 0; i < ranges->length(); i++) {
5839
CharacterRange range = ranges->at(i);
5840
if (last < range.from())
5841
AddRange(CharacterRange(last, range.from() - 1));
5842
if (range.to() >= last) {
5843
if (range.to() == String::kMaxUtf16CodeUnit) {
5846
last = range.to() + 1;
5850
AddRange(CharacterRange(last, String::kMaxUtf16CodeUnit));
5854
void DispatchTableConstructor::VisitText(TextNode* that) {
5855
TextElement elm = that->elements()->at(0);
5857
case TextElement::ATOM: {
5858
uc16 c = elm.data.u_atom->data()[0];
5859
AddRange(CharacterRange(c, c));
5862
case TextElement::CHAR_CLASS: {
5863
RegExpCharacterClass* tree = elm.data.u_char_class;
5864
ZoneList<CharacterRange>* ranges = tree->ranges(that->zone());
5865
if (tree->is_negated()) {
5868
for (int i = 0; i < ranges->length(); i++)
5869
AddRange(ranges->at(i));
5880
void DispatchTableConstructor::VisitAction(ActionNode* that) {
5881
RegExpNode* target = that->on_success();
5882
target->Accept(this);
5886
RegExpEngine::CompilationResult RegExpEngine::Compile(
5887
RegExpCompileData* data,
5891
Handle<String> pattern,
5892
Handle<String> sample_subject,
5895
if ((data->capture_count + 1) * 2 - 1 > RegExpMacroAssembler::kMaxRegister) {
5896
return IrregexpRegExpTooBig();
5898
RegExpCompiler compiler(data->capture_count, ignore_case, is_ascii, zone);
5900
// Sample some characters from the middle of the string.
5901
static const int kSampleSize = 128;
5903
FlattenString(sample_subject);
5904
int chars_sampled = 0;
5905
int half_way = (sample_subject->length() - kSampleSize) / 2;
5906
for (int i = Max(0, half_way);
5907
i < sample_subject->length() && chars_sampled < kSampleSize;
5908
i++, chars_sampled++) {
5909
compiler.frequency_collator()->CountCharacter(sample_subject->Get(i));
5912
// Wrap the body of the regexp in capture #0.
5913
RegExpNode* captured_body = RegExpCapture::ToNode(data->tree,
5917
RegExpNode* node = captured_body;
5918
bool is_end_anchored = data->tree->IsAnchoredAtEnd();
5919
bool is_start_anchored = data->tree->IsAnchoredAtStart();
5920
int max_length = data->tree->max_match();
5921
if (!is_start_anchored) {
5922
// Add a .*? at the beginning, outside the body capture, unless
5923
// this expression is anchored at the beginning.
5924
RegExpNode* loop_node =
5925
RegExpQuantifier::ToNode(0,
5926
RegExpTree::kInfinity,
5928
new(zone) RegExpCharacterClass('*'),
5931
data->contains_anchor);
5933
if (data->contains_anchor) {
5934
// Unroll loop once, to take care of the case that might start
5935
// at the start of input.
5936
ChoiceNode* first_step_node = new(zone) ChoiceNode(2, zone);
5937
first_step_node->AddAlternative(GuardedAlternative(captured_body));
5938
first_step_node->AddAlternative(GuardedAlternative(
5939
new(zone) TextNode(new(zone) RegExpCharacterClass('*'), loop_node)));
5940
node = first_step_node;
5946
node = node->FilterASCII(RegExpCompiler::kMaxRecursion);
5947
// Do it again to propagate the new nodes to places where they were not
5948
// put because they had not been calculated yet.
5949
if (node != NULL) node = node->FilterASCII(RegExpCompiler::kMaxRecursion);
5952
if (node == NULL) node = new(zone) EndNode(EndNode::BACKTRACK, zone);
5954
Analysis analysis(ignore_case, is_ascii);
5955
analysis.EnsureAnalyzed(node);
5956
if (analysis.has_failed()) {
5957
const char* error_message = analysis.error_message();
5958
return CompilationResult(error_message);
5961
// Create the correct assembler for the architecture.
5962
#ifndef V8_INTERPRETED_REGEXP
5963
// Native regexp implementation.
5965
NativeRegExpMacroAssembler::Mode mode =
5966
is_ascii ? NativeRegExpMacroAssembler::ASCII
5967
: NativeRegExpMacroAssembler::UC16;
5969
#if V8_TARGET_ARCH_IA32
5970
RegExpMacroAssemblerIA32 macro_assembler(mode, (data->capture_count + 1) * 2,
5972
#elif V8_TARGET_ARCH_X64
5973
RegExpMacroAssemblerX64 macro_assembler(mode, (data->capture_count + 1) * 2,
5975
#elif V8_TARGET_ARCH_ARM
5976
RegExpMacroAssemblerARM macro_assembler(mode, (data->capture_count + 1) * 2,
5978
#elif V8_TARGET_ARCH_MIPS
5979
RegExpMacroAssemblerMIPS macro_assembler(mode, (data->capture_count + 1) * 2,
5983
#else // V8_INTERPRETED_REGEXP
5984
// Interpreted regexp implementation.
5985
EmbeddedVector<byte, 1024> codes;
5986
RegExpMacroAssemblerIrregexp macro_assembler(codes, zone);
5987
#endif // V8_INTERPRETED_REGEXP
5989
// Inserted here, instead of in Assembler, because it depends on information
5990
// in the AST that isn't replicated in the Node structure.
5991
static const int kMaxBacksearchLimit = 1024;
5992
if (is_end_anchored &&
5993
!is_start_anchored &&
5994
max_length < kMaxBacksearchLimit) {
5995
macro_assembler.SetCurrentPositionFromEnd(max_length);
5999
macro_assembler.set_global_mode(
6000
(data->tree->min_match() > 0)
6001
? RegExpMacroAssembler::GLOBAL_NO_ZERO_LENGTH_CHECK
6002
: RegExpMacroAssembler::GLOBAL);
6005
return compiler.Assemble(¯o_assembler,
6007
data->capture_count,
6012
}} // namespace v8::internal
added
removed
src/third_party/v8/src/jsregexp.cc
