Revert "Revert "Upgrade to 5.0.71.48"" DO NOT MERGE
This reverts commit f2e3994fa5148cc3d9946666f0b0596290192b0e,
and updates the x64 makefile properly so it doesn't break that
build.
FPIIM-449
Change-Id: Ib83e35bfbae6af627451c926a9650ec57c045605
(cherry picked from commit 109988c7ccb6f3fd1a58574fa3dfb88beaef6632)
diff --git a/src/regexp/jsregexp.cc b/src/regexp/jsregexp.cc
index 34d20fe..80f48ca 100644
--- a/src/regexp/jsregexp.cc
+++ b/src/regexp/jsregexp.cc
@@ -25,6 +25,11 @@
#include "src/string-search.h"
#include "src/unicode-decoder.h"
+#ifdef V8_I18N_SUPPORT
+#include "unicode/uset.h"
+#include "unicode/utypes.h"
+#endif // V8_I18N_SUPPORT
+
#ifndef V8_INTERPRETED_REGEXP
#if V8_TARGET_ARCH_IA32
#include "src/regexp/ia32/regexp-macro-assembler-ia32.h"
@@ -72,7 +77,7 @@
int ranges_length,
Interval new_range) {
DCHECK((ranges_length & 1) == 1);
- DCHECK(ranges[ranges_length - 1] == String::kMaxUtf16CodeUnit + 1);
+ DCHECK(ranges[ranges_length - 1] == String::kMaxCodePoint + 1);
if (containment == kLatticeUnknown) return containment;
bool inside = false;
int last = 0;
@@ -145,9 +150,8 @@
PostponeInterruptsScope postpone(isolate);
RegExpCompileData parse_result;
FlatStringReader reader(isolate, pattern);
- if (!RegExpParser::ParseRegExp(re->GetIsolate(), &zone, &reader,
- flags & JSRegExp::kMultiline,
- flags & JSRegExp::kUnicode, &parse_result)) {
+ if (!RegExpParser::ParseRegExp(re->GetIsolate(), &zone, &reader, flags,
+ &parse_result)) {
// Throw an exception if we fail to parse the pattern.
return ThrowRegExpException(re, pattern, parse_result.error);
}
@@ -371,18 +375,16 @@
pattern = String::Flatten(pattern);
RegExpCompileData compile_data;
FlatStringReader reader(isolate, pattern);
- if (!RegExpParser::ParseRegExp(isolate, &zone, &reader,
- flags & JSRegExp::kMultiline,
- flags & JSRegExp::kUnicode, &compile_data)) {
+ if (!RegExpParser::ParseRegExp(isolate, &zone, &reader, flags,
+ &compile_data)) {
// Throw an exception if we fail to parse the pattern.
// THIS SHOULD NOT HAPPEN. We already pre-parsed it successfully once.
USE(ThrowRegExpException(re, pattern, compile_data.error));
return false;
}
- RegExpEngine::CompilationResult result = RegExpEngine::Compile(
- isolate, &zone, &compile_data, flags & JSRegExp::kIgnoreCase,
- flags & JSRegExp::kGlobal, flags & JSRegExp::kMultiline,
- flags & JSRegExp::kSticky, pattern, sample_subject, is_one_byte);
+ RegExpEngine::CompilationResult result =
+ RegExpEngine::Compile(isolate, &zone, &compile_data, flags, pattern,
+ sample_subject, is_one_byte);
if (result.error_message != NULL) {
// Unable to compile regexp.
Handle<String> error_message = isolate->factory()->NewStringFromUtf8(
@@ -636,7 +638,6 @@
RegExpImpl::GlobalCache::GlobalCache(Handle<JSRegExp> regexp,
Handle<String> subject,
- bool is_global,
Isolate* isolate)
: register_array_(NULL),
register_array_size_(0),
@@ -661,7 +662,8 @@
}
}
- if (is_global && !interpreted) {
+ DCHECK_NE(0, regexp->GetFlags() & JSRegExp::kGlobal);
+ if (!interpreted) {
register_array_size_ =
Max(registers_per_match_, Isolate::kJSRegexpStaticOffsetsVectorSize);
max_matches_ = register_array_size_ / registers_per_match_;
@@ -690,6 +692,16 @@
last_match[1] = 0;
}
+int RegExpImpl::GlobalCache::AdvanceZeroLength(int last_index) {
+ if ((regexp_->GetFlags() & JSRegExp::kUnicode) != 0 &&
+ last_index + 1 < subject_->length() &&
+ unibrow::Utf16::IsLeadSurrogate(subject_->Get(last_index)) &&
+ unibrow::Utf16::IsTrailSurrogate(subject_->Get(last_index + 1))) {
+ // Advance over the surrogate pair.
+ return last_index + 2;
+ }
+ return last_index + 1;
+}
// -------------------------------------------------------------------
// Implementation of the Irregexp regular expression engine.
@@ -945,7 +957,7 @@
class RegExpCompiler {
public:
RegExpCompiler(Isolate* isolate, Zone* zone, int capture_count,
- bool ignore_case, bool is_one_byte);
+ JSRegExp::Flags flags, bool is_one_byte);
int AllocateRegister() {
if (next_register_ >= RegExpMacroAssembler::kMaxRegister) {
@@ -955,6 +967,22 @@
return next_register_++;
}
+ // Lookarounds to match lone surrogates for unicode character class matches
+ // are never nested. We can therefore reuse registers.
+ int UnicodeLookaroundStackRegister() {
+ if (unicode_lookaround_stack_register_ == kNoRegister) {
+ unicode_lookaround_stack_register_ = AllocateRegister();
+ }
+ return unicode_lookaround_stack_register_;
+ }
+
+ int UnicodeLookaroundPositionRegister() {
+ if (unicode_lookaround_position_register_ == kNoRegister) {
+ unicode_lookaround_position_register_ = AllocateRegister();
+ }
+ return unicode_lookaround_position_register_;
+ }
+
RegExpEngine::CompilationResult Assemble(RegExpMacroAssembler* assembler,
RegExpNode* start,
int capture_count,
@@ -981,7 +1009,8 @@
void SetRegExpTooBig() { reg_exp_too_big_ = true; }
- inline bool ignore_case() { return ignore_case_; }
+ inline bool ignore_case() { return (flags_ & JSRegExp::kIgnoreCase) != 0; }
+ inline bool unicode() { return (flags_ & JSRegExp::kUnicode) != 0; }
inline bool one_byte() { return one_byte_; }
inline bool optimize() { return optimize_; }
inline void set_optimize(bool value) { optimize_ = value; }
@@ -1006,10 +1035,12 @@
private:
EndNode* accept_;
int next_register_;
+ int unicode_lookaround_stack_register_;
+ int unicode_lookaround_position_register_;
List<RegExpNode*>* work_list_;
int recursion_depth_;
RegExpMacroAssembler* macro_assembler_;
- bool ignore_case_;
+ JSRegExp::Flags flags_;
bool one_byte_;
bool reg_exp_too_big_;
bool limiting_recursion_;
@@ -1041,11 +1072,13 @@
// Attempts to compile the regexp using an Irregexp code generator. Returns
// a fixed array or a null handle depending on whether it succeeded.
RegExpCompiler::RegExpCompiler(Isolate* isolate, Zone* zone, int capture_count,
- bool ignore_case, bool one_byte)
+ JSRegExp::Flags flags, bool one_byte)
: next_register_(2 * (capture_count + 1)),
+ unicode_lookaround_stack_register_(kNoRegister),
+ unicode_lookaround_position_register_(kNoRegister),
work_list_(NULL),
recursion_depth_(0),
- ignore_case_(ignore_case),
+ flags_(flags),
one_byte_(one_byte),
reg_exp_too_big_(false),
limiting_recursion_(false),
@@ -1941,15 +1974,13 @@
// know that the character is in the range of min_char to max_char inclusive.
// Either label can be NULL indicating backtracking. Either label can also be
// equal to the fall_through label.
-static void GenerateBranches(RegExpMacroAssembler* masm,
- ZoneList<int>* ranges,
- int start_index,
- int end_index,
- uc16 min_char,
- uc16 max_char,
- Label* fall_through,
- Label* even_label,
- Label* odd_label) {
+static void GenerateBranches(RegExpMacroAssembler* masm, ZoneList<int>* ranges,
+ int start_index, int end_index, uc32 min_char,
+ uc32 max_char, Label* fall_through,
+ Label* even_label, Label* odd_label) {
+ DCHECK_LE(min_char, String::kMaxUtf16CodeUnit);
+ DCHECK_LE(max_char, String::kMaxUtf16CodeUnit);
+
int first = ranges->at(start_index);
int last = ranges->at(end_index) - 1;
@@ -2098,9 +2129,7 @@
Label* on_failure, int cp_offset, bool check_offset,
bool preloaded, Zone* zone) {
ZoneList<CharacterRange>* ranges = cc->ranges(zone);
- if (!CharacterRange::IsCanonical(ranges)) {
- CharacterRange::Canonicalize(ranges);
- }
+ CharacterRange::Canonicalize(ranges);
int max_char;
if (one_byte) {
@@ -2142,23 +2171,14 @@
}
return;
}
- if (last_valid_range == 0 &&
- !cc->is_negated() &&
- ranges->at(0).IsEverything(max_char)) {
- // This is a common case hit by non-anchored expressions.
- if (check_offset) {
- macro_assembler->CheckPosition(cp_offset, on_failure);
- }
- return;
- }
if (!preloaded) {
macro_assembler->LoadCurrentCharacter(cp_offset, on_failure, check_offset);
}
if (cc->is_standard(zone) &&
- macro_assembler->CheckSpecialCharacterClass(cc->standard_type(),
- on_failure)) {
+ macro_assembler->CheckSpecialCharacterClass(cc->standard_type(),
+ on_failure)) {
return;
}
@@ -2470,12 +2490,14 @@
} else {
// For 2-character preloads in one-byte mode or 1-character preloads in
// two-byte mode we also use a 16 bit load with zero extend.
+ static const uint32_t kTwoByteMask = 0xffff;
+ static const uint32_t kFourByteMask = 0xffffffff;
if (details->characters() == 2 && compiler->one_byte()) {
- if ((mask & 0xffff) == 0xffff) need_mask = false;
+ if ((mask & kTwoByteMask) == kTwoByteMask) need_mask = false;
} else if (details->characters() == 1 && !compiler->one_byte()) {
- if ((mask & 0xffff) == 0xffff) need_mask = false;
+ if ((mask & kTwoByteMask) == kTwoByteMask) need_mask = false;
} else {
- if (mask == 0xffffffff) need_mask = false;
+ if (mask == kFourByteMask) need_mask = false;
}
}
@@ -2798,9 +2820,7 @@
DCHECK(elm.text_type() == TextElement::CHAR_CLASS);
RegExpCharacterClass* cc = elm.char_class();
ZoneList<CharacterRange>* ranges = cc->ranges(zone());
- if (!CharacterRange::IsCanonical(ranges)) {
- CharacterRange::Canonicalize(ranges);
- }
+ CharacterRange::Canonicalize(ranges);
// Now they are in order so we only need to look at the first.
int range_count = ranges->length();
if (cc->is_negated()) {
@@ -3289,6 +3309,36 @@
}
+TextNode* TextNode::CreateForCharacterRanges(Zone* zone,
+ ZoneList<CharacterRange>* ranges,
+ bool read_backward,
+ RegExpNode* on_success) {
+ DCHECK_NOT_NULL(ranges);
+ ZoneList<TextElement>* elms = new (zone) ZoneList<TextElement>(1, zone);
+ elms->Add(
+ TextElement::CharClass(new (zone) RegExpCharacterClass(ranges, false)),
+ zone);
+ return new (zone) TextNode(elms, read_backward, on_success);
+}
+
+
+TextNode* TextNode::CreateForSurrogatePair(Zone* zone, CharacterRange lead,
+ CharacterRange trail,
+ bool read_backward,
+ RegExpNode* on_success) {
+ ZoneList<CharacterRange>* lead_ranges = CharacterRange::List(zone, lead);
+ ZoneList<CharacterRange>* trail_ranges = CharacterRange::List(zone, trail);
+ ZoneList<TextElement>* elms = new (zone) ZoneList<TextElement>(2, zone);
+ elms->Add(TextElement::CharClass(
+ new (zone) RegExpCharacterClass(lead_ranges, false)),
+ zone);
+ elms->Add(TextElement::CharClass(
+ new (zone) RegExpCharacterClass(trail_ranges, false)),
+ zone);
+ return new (zone) TextNode(elms, read_backward, on_success);
+}
+
+
// This generates the code to match a text node. A text node can contain
// straight character sequences (possibly to be matched in a case-independent
// way) and character classes. For efficiency we do not do this in a single
@@ -3385,10 +3435,7 @@
// independent case and it slows us down if we don't know that.
if (cc->is_standard(zone())) continue;
ZoneList<CharacterRange>* ranges = cc->ranges(zone());
- int range_count = ranges->length();
- for (int j = 0; j < range_count; j++) {
- ranges->at(j).AddCaseEquivalents(isolate, zone(), ranges, is_one_byte);
- }
+ CharacterRange::AddCaseEquivalents(isolate, zone(), ranges, is_one_byte);
}
}
}
@@ -3405,9 +3452,7 @@
if (elm.text_type() != TextElement::CHAR_CLASS) return NULL;
RegExpCharacterClass* node = elm.char_class();
ZoneList<CharacterRange>* ranges = node->ranges(zone());
- if (!CharacterRange::IsCanonical(ranges)) {
- CharacterRange::Canonicalize(ranges);
- }
+ CharacterRange::Canonicalize(ranges);
if (node->is_negated()) {
return ranges->length() == 0 ? on_success() : NULL;
}
@@ -3554,27 +3599,29 @@
};
+static const uc32 kRangeEndMarker = 0x110000;
+
// The '2' variant is has inclusive from and exclusive to.
// This covers \s as defined in ECMA-262 5.1, 15.10.2.12,
// which include WhiteSpace (7.2) or LineTerminator (7.3) values.
-static const int kSpaceRanges[] = { '\t', '\r' + 1, ' ', ' ' + 1,
- 0x00A0, 0x00A1, 0x1680, 0x1681, 0x180E, 0x180F, 0x2000, 0x200B,
- 0x2028, 0x202A, 0x202F, 0x2030, 0x205F, 0x2060, 0x3000, 0x3001,
- 0xFEFF, 0xFF00, 0x10000 };
+static const int kSpaceRanges[] = {
+ '\t', '\r' + 1, ' ', ' ' + 1, 0x00A0, 0x00A1, 0x1680, 0x1681,
+ 0x180E, 0x180F, 0x2000, 0x200B, 0x2028, 0x202A, 0x202F, 0x2030,
+ 0x205F, 0x2060, 0x3000, 0x3001, 0xFEFF, 0xFF00, kRangeEndMarker};
static const int kSpaceRangeCount = arraysize(kSpaceRanges);
static const int kWordRanges[] = {
- '0', '9' + 1, 'A', 'Z' + 1, '_', '_' + 1, 'a', 'z' + 1, 0x10000 };
+ '0', '9' + 1, 'A', 'Z' + 1, '_', '_' + 1, 'a', 'z' + 1, kRangeEndMarker};
static const int kWordRangeCount = arraysize(kWordRanges);
-static const int kDigitRanges[] = { '0', '9' + 1, 0x10000 };
+static const int kDigitRanges[] = {'0', '9' + 1, kRangeEndMarker};
static const int kDigitRangeCount = arraysize(kDigitRanges);
-static const int kSurrogateRanges[] = { 0xd800, 0xe000, 0x10000 };
+static const int kSurrogateRanges[] = {
+ kLeadSurrogateStart, kLeadSurrogateStart + 1, kRangeEndMarker};
static const int kSurrogateRangeCount = arraysize(kSurrogateRanges);
-static const int kLineTerminatorRanges[] = { 0x000A, 0x000B, 0x000D, 0x000E,
- 0x2028, 0x202A, 0x10000 };
+static const int kLineTerminatorRanges[] = {
+ 0x000A, 0x000B, 0x000D, 0x000E, 0x2028, 0x202A, kRangeEndMarker};
static const int kLineTerminatorRangeCount = arraysize(kLineTerminatorRanges);
-
void BoyerMoorePositionInfo::Set(int character) {
SetInterval(Interval(character, character));
}
@@ -3916,6 +3963,11 @@
void ChoiceNode::Emit(RegExpCompiler* compiler, Trace* trace) {
int choice_count = alternatives_->length();
+ if (choice_count == 1 && alternatives_->at(0).guards() == NULL) {
+ alternatives_->at(0).node()->Emit(compiler, trace);
+ return;
+ }
+
AssertGuardsMentionRegisters(trace);
LimitResult limit_result = LimitVersions(compiler, trace);
@@ -4349,14 +4401,19 @@
DCHECK_EQ(start_reg_ + 1, end_reg_);
if (compiler->ignore_case()) {
- assembler->CheckNotBackReferenceIgnoreCase(start_reg_, read_backward(),
- trace->backtrack());
+ assembler->CheckNotBackReferenceIgnoreCase(
+ start_reg_, read_backward(), compiler->unicode(), trace->backtrack());
} else {
assembler->CheckNotBackReference(start_reg_, read_backward(),
trace->backtrack());
}
// We are going to advance backward, so we may end up at the start.
if (read_backward()) trace->set_at_start(Trace::UNKNOWN);
+
+ // Check that the back reference does not end inside a surrogate pair.
+ if (compiler->unicode() && !compiler->one_byte()) {
+ assembler->CheckNotInSurrogatePair(trace->cp_offset(), trace->backtrack());
+ }
on_success()->Emit(compiler, trace);
}
@@ -4732,8 +4789,8 @@
static bool CompareInverseRanges(ZoneList<CharacterRange>* ranges,
const int* special_class,
int length) {
- length--; // Remove final 0x10000.
- DCHECK(special_class[length] == 0x10000);
+ length--; // Remove final marker.
+ DCHECK(special_class[length] == kRangeEndMarker);
DCHECK(ranges->length() != 0);
DCHECK(length != 0);
DCHECK(special_class[0] != 0);
@@ -4753,7 +4810,7 @@
return false;
}
}
- if (range.to() != 0xffff) {
+ if (range.to() != String::kMaxCodePoint) {
return false;
}
return true;
@@ -4763,8 +4820,8 @@
static bool CompareRanges(ZoneList<CharacterRange>* ranges,
const int* special_class,
int length) {
- length--; // Remove final 0x10000.
- DCHECK(special_class[length] == 0x10000);
+ length--; // Remove final marker.
+ DCHECK(special_class[length] == kRangeEndMarker);
if (ranges->length() * 2 != length) {
return false;
}
@@ -4820,10 +4877,303 @@
}
+UnicodeRangeSplitter::UnicodeRangeSplitter(Zone* zone,
+ ZoneList<CharacterRange>* base)
+ : zone_(zone),
+ table_(zone),
+ bmp_(nullptr),
+ lead_surrogates_(nullptr),
+ trail_surrogates_(nullptr),
+ non_bmp_(nullptr) {
+ // The unicode range splitter categorizes given character ranges into:
+ // - Code points from the BMP representable by one code unit.
+ // - Code points outside the BMP that need to be split into surrogate pairs.
+ // - Lone lead surrogates.
+ // - Lone trail surrogates.
+ // Lone surrogates are valid code points, even though no actual characters.
+ // They require special matching to make sure we do not split surrogate pairs.
+ // We use the dispatch table to accomplish this. The base range is split up
+ // by the table by the overlay ranges, and the Call callback is used to
+ // filter and collect ranges for each category.
+ for (int i = 0; i < base->length(); i++) {
+ table_.AddRange(base->at(i), kBase, zone_);
+ }
+ // Add overlay ranges.
+ table_.AddRange(CharacterRange::Range(0, kLeadSurrogateStart - 1),
+ kBmpCodePoints, zone_);
+ table_.AddRange(CharacterRange::Range(kLeadSurrogateStart, kLeadSurrogateEnd),
+ kLeadSurrogates, zone_);
+ table_.AddRange(
+ CharacterRange::Range(kTrailSurrogateStart, kTrailSurrogateEnd),
+ kTrailSurrogates, zone_);
+ table_.AddRange(
+ CharacterRange::Range(kTrailSurrogateEnd + 1, kNonBmpStart - 1),
+ kBmpCodePoints, zone_);
+ table_.AddRange(CharacterRange::Range(kNonBmpStart, kNonBmpEnd),
+ kNonBmpCodePoints, zone_);
+ table_.ForEach(this);
+}
+
+
+void UnicodeRangeSplitter::Call(uc32 from, DispatchTable::Entry entry) {
+ OutSet* outset = entry.out_set();
+ if (!outset->Get(kBase)) return;
+ ZoneList<CharacterRange>** target = NULL;
+ if (outset->Get(kBmpCodePoints)) {
+ target = &bmp_;
+ } else if (outset->Get(kLeadSurrogates)) {
+ target = &lead_surrogates_;
+ } else if (outset->Get(kTrailSurrogates)) {
+ target = &trail_surrogates_;
+ } else {
+ DCHECK(outset->Get(kNonBmpCodePoints));
+ target = &non_bmp_;
+ }
+ if (*target == NULL) *target = new (zone_) ZoneList<CharacterRange>(2, zone_);
+ (*target)->Add(CharacterRange::Range(entry.from(), entry.to()), zone_);
+}
+
+
+void AddBmpCharacters(RegExpCompiler* compiler, ChoiceNode* result,
+ RegExpNode* on_success, UnicodeRangeSplitter* splitter) {
+ ZoneList<CharacterRange>* bmp = splitter->bmp();
+ if (bmp == nullptr) return;
+ result->AddAlternative(GuardedAlternative(TextNode::CreateForCharacterRanges(
+ compiler->zone(), bmp, compiler->read_backward(), on_success)));
+}
+
+
+void AddNonBmpSurrogatePairs(RegExpCompiler* compiler, ChoiceNode* result,
+ RegExpNode* on_success,
+ UnicodeRangeSplitter* splitter) {
+ ZoneList<CharacterRange>* non_bmp = splitter->non_bmp();
+ if (non_bmp == nullptr) return;
+ DCHECK(compiler->unicode());
+ DCHECK(!compiler->one_byte());
+ Zone* zone = compiler->zone();
+ CharacterRange::Canonicalize(non_bmp);
+ for (int i = 0; i < non_bmp->length(); i++) {
+ // Match surrogate pair.
+ // E.g. [\u10005-\u11005] becomes
+ // \ud800[\udc05-\udfff]|
+ // [\ud801-\ud803][\udc00-\udfff]|
+ // \ud804[\udc00-\udc05]
+ uc32 from = non_bmp->at(i).from();
+ uc32 to = non_bmp->at(i).to();
+ uc16 from_l = unibrow::Utf16::LeadSurrogate(from);
+ uc16 from_t = unibrow::Utf16::TrailSurrogate(from);
+ uc16 to_l = unibrow::Utf16::LeadSurrogate(to);
+ uc16 to_t = unibrow::Utf16::TrailSurrogate(to);
+ if (from_l == to_l) {
+ // The lead surrogate is the same.
+ result->AddAlternative(
+ GuardedAlternative(TextNode::CreateForSurrogatePair(
+ zone, CharacterRange::Singleton(from_l),
+ CharacterRange::Range(from_t, to_t), compiler->read_backward(),
+ on_success)));
+ } else {
+ if (from_t != kTrailSurrogateStart) {
+ // Add [from_l][from_t-\udfff]
+ result->AddAlternative(
+ GuardedAlternative(TextNode::CreateForSurrogatePair(
+ zone, CharacterRange::Singleton(from_l),
+ CharacterRange::Range(from_t, kTrailSurrogateEnd),
+ compiler->read_backward(), on_success)));
+ from_l++;
+ }
+ if (to_t != kTrailSurrogateEnd) {
+ // Add [to_l][\udc00-to_t]
+ result->AddAlternative(
+ GuardedAlternative(TextNode::CreateForSurrogatePair(
+ zone, CharacterRange::Singleton(to_l),
+ CharacterRange::Range(kTrailSurrogateStart, to_t),
+ compiler->read_backward(), on_success)));
+ to_l--;
+ }
+ if (from_l <= to_l) {
+ // Add [from_l-to_l][\udc00-\udfff]
+ result->AddAlternative(
+ GuardedAlternative(TextNode::CreateForSurrogatePair(
+ zone, CharacterRange::Range(from_l, to_l),
+ CharacterRange::Range(kTrailSurrogateStart, kTrailSurrogateEnd),
+ compiler->read_backward(), on_success)));
+ }
+ }
+ }
+}
+
+
+RegExpNode* NegativeLookaroundAgainstReadDirectionAndMatch(
+ RegExpCompiler* compiler, ZoneList<CharacterRange>* lookbehind,
+ ZoneList<CharacterRange>* match, RegExpNode* on_success,
+ bool read_backward) {
+ Zone* zone = compiler->zone();
+ RegExpNode* match_node = TextNode::CreateForCharacterRanges(
+ zone, match, read_backward, on_success);
+ int stack_register = compiler->UnicodeLookaroundStackRegister();
+ int position_register = compiler->UnicodeLookaroundPositionRegister();
+ RegExpLookaround::Builder lookaround(false, match_node, stack_register,
+ position_register);
+ RegExpNode* negative_match = TextNode::CreateForCharacterRanges(
+ zone, lookbehind, !read_backward, lookaround.on_match_success());
+ return lookaround.ForMatch(negative_match);
+}
+
+
+RegExpNode* MatchAndNegativeLookaroundInReadDirection(
+ RegExpCompiler* compiler, ZoneList<CharacterRange>* match,
+ ZoneList<CharacterRange>* lookahead, RegExpNode* on_success,
+ bool read_backward) {
+ Zone* zone = compiler->zone();
+ int stack_register = compiler->UnicodeLookaroundStackRegister();
+ int position_register = compiler->UnicodeLookaroundPositionRegister();
+ RegExpLookaround::Builder lookaround(false, on_success, stack_register,
+ position_register);
+ RegExpNode* negative_match = TextNode::CreateForCharacterRanges(
+ zone, lookahead, read_backward, lookaround.on_match_success());
+ return TextNode::CreateForCharacterRanges(
+ zone, match, read_backward, lookaround.ForMatch(negative_match));
+}
+
+
+void AddLoneLeadSurrogates(RegExpCompiler* compiler, ChoiceNode* result,
+ RegExpNode* on_success,
+ UnicodeRangeSplitter* splitter) {
+ ZoneList<CharacterRange>* lead_surrogates = splitter->lead_surrogates();
+ if (lead_surrogates == nullptr) return;
+ Zone* zone = compiler->zone();
+ // E.g. \ud801 becomes \ud801(?![\udc00-\udfff]).
+ ZoneList<CharacterRange>* trail_surrogates = CharacterRange::List(
+ zone, CharacterRange::Range(kTrailSurrogateStart, kTrailSurrogateEnd));
+
+ RegExpNode* match;
+ if (compiler->read_backward()) {
+ // Reading backward. Assert that reading forward, there is no trail
+ // surrogate, and then backward match the lead surrogate.
+ match = NegativeLookaroundAgainstReadDirectionAndMatch(
+ compiler, trail_surrogates, lead_surrogates, on_success, true);
+ } else {
+ // Reading forward. Forward match the lead surrogate and assert that
+ // no trail surrogate follows.
+ match = MatchAndNegativeLookaroundInReadDirection(
+ compiler, lead_surrogates, trail_surrogates, on_success, false);
+ }
+ result->AddAlternative(GuardedAlternative(match));
+}
+
+
+void AddLoneTrailSurrogates(RegExpCompiler* compiler, ChoiceNode* result,
+ RegExpNode* on_success,
+ UnicodeRangeSplitter* splitter) {
+ ZoneList<CharacterRange>* trail_surrogates = splitter->trail_surrogates();
+ if (trail_surrogates == nullptr) return;
+ Zone* zone = compiler->zone();
+ // E.g. \udc01 becomes (?<![\ud800-\udbff])\udc01
+ ZoneList<CharacterRange>* lead_surrogates = CharacterRange::List(
+ zone, CharacterRange::Range(kLeadSurrogateStart, kLeadSurrogateEnd));
+
+ RegExpNode* match;
+ if (compiler->read_backward()) {
+ // Reading backward. Backward match the trail surrogate and assert that no
+ // lead surrogate precedes it.
+ match = MatchAndNegativeLookaroundInReadDirection(
+ compiler, trail_surrogates, lead_surrogates, on_success, true);
+ } else {
+ // Reading forward. Assert that reading backward, there is no lead
+ // surrogate, and then forward match the trail surrogate.
+ match = NegativeLookaroundAgainstReadDirectionAndMatch(
+ compiler, lead_surrogates, trail_surrogates, on_success, false);
+ }
+ result->AddAlternative(GuardedAlternative(match));
+}
+
+RegExpNode* UnanchoredAdvance(RegExpCompiler* compiler,
+ RegExpNode* on_success) {
+ // This implements ES2015 21.2.5.2.3, AdvanceStringIndex.
+ DCHECK(!compiler->read_backward());
+ Zone* zone = compiler->zone();
+ // Advance any character. If the character happens to be a lead surrogate and
+ // we advanced into the middle of a surrogate pair, it will work out, as
+ // nothing will match from there. We will have to advance again, consuming
+ // the associated trail surrogate.
+ ZoneList<CharacterRange>* range = CharacterRange::List(
+ zone, CharacterRange::Range(0, String::kMaxUtf16CodeUnit));
+ return TextNode::CreateForCharacterRanges(zone, range, false, on_success);
+}
+
+
+void AddUnicodeCaseEquivalents(RegExpCompiler* compiler,
+ ZoneList<CharacterRange>* ranges) {
+#ifdef V8_I18N_SUPPORT
+ // Use ICU to compute the case fold closure over the ranges.
+ DCHECK(compiler->unicode());
+ DCHECK(compiler->ignore_case());
+ USet* set = uset_openEmpty();
+ for (int i = 0; i < ranges->length(); i++) {
+ uset_addRange(set, ranges->at(i).from(), ranges->at(i).to());
+ }
+ ranges->Clear();
+ uset_closeOver(set, USET_CASE_INSENSITIVE);
+ // Full case mapping map single characters to multiple characters.
+ // Those are represented as strings in the set. Remove them so that
+ // we end up with only simple and common case mappings.
+ uset_removeAllStrings(set);
+ int item_count = uset_getItemCount(set);
+ int item_result = 0;
+ UErrorCode ec = U_ZERO_ERROR;
+ Zone* zone = compiler->zone();
+ for (int i = 0; i < item_count; i++) {
+ uc32 start = 0;
+ uc32 end = 0;
+ item_result += uset_getItem(set, i, &start, &end, nullptr, 0, &ec);
+ ranges->Add(CharacterRange::Range(start, end), zone);
+ }
+ // No errors and everything we collected have been ranges.
+ DCHECK_EQ(U_ZERO_ERROR, ec);
+ DCHECK_EQ(0, item_result);
+ uset_close(set);
+#else
+ // Fallback if ICU is not included.
+ CharacterRange::AddCaseEquivalents(compiler->isolate(), compiler->zone(),
+ ranges, compiler->one_byte());
+#endif // V8_I18N_SUPPORT
+ CharacterRange::Canonicalize(ranges);
+}
+
+
RegExpNode* RegExpCharacterClass::ToNode(RegExpCompiler* compiler,
RegExpNode* on_success) {
- return new (compiler->zone())
- TextNode(this, compiler->read_backward(), on_success);
+ set_.Canonicalize();
+ Zone* zone = compiler->zone();
+ ZoneList<CharacterRange>* ranges = this->ranges(zone);
+ if (compiler->unicode() && compiler->ignore_case()) {
+ AddUnicodeCaseEquivalents(compiler, ranges);
+ }
+ if (compiler->unicode() && !compiler->one_byte()) {
+ if (is_negated()) {
+ ZoneList<CharacterRange>* negated =
+ new (zone) ZoneList<CharacterRange>(2, zone);
+ CharacterRange::Negate(ranges, negated, zone);
+ ranges = negated;
+ }
+ if (ranges->length() == 0) {
+ // No matches possible.
+ return new (zone) EndNode(EndNode::BACKTRACK, zone);
+ }
+ if (standard_type() == '*') {
+ return UnanchoredAdvance(compiler, on_success);
+ } else {
+ ChoiceNode* result = new (zone) ChoiceNode(2, zone);
+ UnicodeRangeSplitter splitter(zone, ranges);
+ AddBmpCharacters(compiler, result, on_success, &splitter);
+ AddNonBmpSurrogatePairs(compiler, result, on_success, &splitter);
+ AddLoneLeadSurrogates(compiler, result, on_success, &splitter);
+ AddLoneTrailSurrogates(compiler, result, on_success, &splitter);
+ return result;
+ }
+ } else {
+ return new (zone) TextNode(this, compiler->read_backward(), on_success);
+ }
}
@@ -5338,6 +5688,47 @@
}
+RegExpLookaround::Builder::Builder(bool is_positive, RegExpNode* on_success,
+ int stack_pointer_register,
+ int position_register,
+ int capture_register_count,
+ int capture_register_start)
+ : is_positive_(is_positive),
+ on_success_(on_success),
+ stack_pointer_register_(stack_pointer_register),
+ position_register_(position_register) {
+ if (is_positive_) {
+ on_match_success_ = ActionNode::PositiveSubmatchSuccess(
+ stack_pointer_register, position_register, capture_register_count,
+ capture_register_start, on_success_);
+ } else {
+ Zone* zone = on_success_->zone();
+ on_match_success_ = new (zone) NegativeSubmatchSuccess(
+ stack_pointer_register, position_register, capture_register_count,
+ capture_register_start, zone);
+ }
+}
+
+
+RegExpNode* RegExpLookaround::Builder::ForMatch(RegExpNode* match) {
+ if (is_positive_) {
+ return ActionNode::BeginSubmatch(stack_pointer_register_,
+ position_register_, match);
+ } else {
+ Zone* zone = on_success_->zone();
+ // We use a ChoiceNode to represent the negative lookaround. The first
+ // alternative is the negative match. On success, the end node backtracks.
+ // On failure, the second alternative is tried and leads to success.
+ // NegativeLookaheadChoiceNode is a special ChoiceNode that ignores the
+ // first exit when calculating quick checks.
+ ChoiceNode* choice_node = new (zone) NegativeLookaroundChoiceNode(
+ GuardedAlternative(match), GuardedAlternative(on_success_), zone);
+ return ActionNode::BeginSubmatch(stack_pointer_register_,
+ position_register_, choice_node);
+ }
+}
+
+
RegExpNode* RegExpLookaround::ToNode(RegExpCompiler* compiler,
RegExpNode* on_success) {
int stack_pointer_register = compiler->AllocateRegister();
@@ -5352,35 +5743,10 @@
RegExpNode* result;
bool was_reading_backward = compiler->read_backward();
compiler->set_read_backward(type() == LOOKBEHIND);
- if (is_positive()) {
- result = ActionNode::BeginSubmatch(
- stack_pointer_register, position_register,
- body()->ToNode(compiler,
- ActionNode::PositiveSubmatchSuccess(
- stack_pointer_register, position_register,
- register_count, register_start, on_success)));
- } else {
- // We use a ChoiceNode for a negative lookahead because it has most of
- // the characteristics we need. It has the body of the lookahead as its
- // first alternative and the expression after the lookahead of the second
- // alternative. If the first alternative succeeds then the
- // NegativeSubmatchSuccess will unwind the stack including everything the
- // choice node set up and backtrack. If the first alternative fails then
- // the second alternative is tried, which is exactly the desired result
- // for a negative lookahead. The NegativeLookaheadChoiceNode is a special
- // ChoiceNode that knows to ignore the first exit when calculating quick
- // checks.
- Zone* zone = compiler->zone();
-
- GuardedAlternative body_alt(
- body()->ToNode(compiler, new (zone) NegativeSubmatchSuccess(
- stack_pointer_register, position_register,
- register_count, register_start, zone)));
- ChoiceNode* choice_node = new (zone) NegativeLookaroundChoiceNode(
- body_alt, GuardedAlternative(on_success), zone);
- result = ActionNode::BeginSubmatch(stack_pointer_register,
- position_register, choice_node);
- }
+ Builder builder(is_positive(), on_success, stack_pointer_register,
+ position_register, register_count, register_start);
+ RegExpNode* match = body_->ToNode(compiler, builder.on_match_success());
+ result = builder.ForMatch(match);
compiler->set_read_backward(was_reading_backward);
return result;
}
@@ -5428,10 +5794,10 @@
ZoneList<CharacterRange>* ranges,
Zone* zone) {
elmc--;
- DCHECK(elmv[elmc] == 0x10000);
+ DCHECK(elmv[elmc] == kRangeEndMarker);
for (int i = 0; i < elmc; i += 2) {
DCHECK(elmv[i] < elmv[i + 1]);
- ranges->Add(CharacterRange(elmv[i], elmv[i + 1] - 1), zone);
+ ranges->Add(CharacterRange::Range(elmv[i], elmv[i + 1] - 1), zone);
}
}
@@ -5441,17 +5807,17 @@
ZoneList<CharacterRange>* ranges,
Zone* zone) {
elmc--;
- DCHECK(elmv[elmc] == 0x10000);
+ DCHECK(elmv[elmc] == kRangeEndMarker);
DCHECK(elmv[0] != 0x0000);
- DCHECK(elmv[elmc-1] != String::kMaxUtf16CodeUnit);
+ DCHECK(elmv[elmc - 1] != String::kMaxCodePoint);
uc16 last = 0x0000;
for (int i = 0; i < elmc; i += 2) {
DCHECK(last <= elmv[i] - 1);
DCHECK(elmv[i] < elmv[i + 1]);
- ranges->Add(CharacterRange(last, elmv[i] - 1), zone);
+ ranges->Add(CharacterRange::Range(last, elmv[i] - 1), zone);
last = elmv[i + 1];
}
- ranges->Add(CharacterRange(last, String::kMaxUtf16CodeUnit), zone);
+ ranges->Add(CharacterRange::Range(last, String::kMaxCodePoint), zone);
}
@@ -5508,115 +5874,73 @@
}
-class CharacterRangeSplitter {
- public:
- CharacterRangeSplitter(ZoneList<CharacterRange>** included,
- ZoneList<CharacterRange>** excluded,
- Zone* zone)
- : included_(included),
- excluded_(excluded),
- zone_(zone) { }
- void Call(uc16 from, DispatchTable::Entry entry);
-
- static const int kInBase = 0;
- static const int kInOverlay = 1;
-
- private:
- ZoneList<CharacterRange>** included_;
- ZoneList<CharacterRange>** excluded_;
- Zone* zone_;
-};
-
-
-void CharacterRangeSplitter::Call(uc16 from, DispatchTable::Entry entry) {
- if (!entry.out_set()->Get(kInBase)) return;
- ZoneList<CharacterRange>** target = entry.out_set()->Get(kInOverlay)
- ? included_
- : excluded_;
- if (*target == NULL) *target = new(zone_) ZoneList<CharacterRange>(2, zone_);
- (*target)->Add(CharacterRange(entry.from(), entry.to()), zone_);
-}
-
-
-void CharacterRange::Split(ZoneList<CharacterRange>* base,
- Vector<const int> overlay,
- ZoneList<CharacterRange>** included,
- ZoneList<CharacterRange>** excluded,
- Zone* zone) {
- DCHECK_NULL(*included);
- DCHECK_NULL(*excluded);
- DispatchTable table(zone);
- for (int i = 0; i < base->length(); i++)
- table.AddRange(base->at(i), CharacterRangeSplitter::kInBase, zone);
- for (int i = 0; i < overlay.length(); i += 2) {
- table.AddRange(CharacterRange(overlay[i], overlay[i + 1] - 1),
- CharacterRangeSplitter::kInOverlay, zone);
- }
- CharacterRangeSplitter callback(included, excluded, zone);
- table.ForEach(&callback);
-}
-
-
void CharacterRange::AddCaseEquivalents(Isolate* isolate, Zone* zone,
ZoneList<CharacterRange>* ranges,
bool is_one_byte) {
- uc16 bottom = from();
- uc16 top = to();
- if (is_one_byte && !RangeContainsLatin1Equivalents(*this)) {
- if (bottom > String::kMaxOneByteCharCode) return;
- if (top > String::kMaxOneByteCharCode) top = String::kMaxOneByteCharCode;
- }
- unibrow::uchar chars[unibrow::Ecma262UnCanonicalize::kMaxWidth];
- if (top == bottom) {
- // If this is a singleton we just expand the one character.
- int length = isolate->jsregexp_uncanonicalize()->get(bottom, '\0', chars);
- for (int i = 0; i < length; i++) {
- uc32 chr = chars[i];
- if (chr != bottom) {
- ranges->Add(CharacterRange::Singleton(chars[i]), zone);
- }
+ int range_count = ranges->length();
+ for (int i = 0; i < range_count; i++) {
+ CharacterRange range = ranges->at(i);
+ uc32 bottom = range.from();
+ if (bottom > String::kMaxUtf16CodeUnit) return;
+ uc32 top = Min(range.to(), String::kMaxUtf16CodeUnit);
+ // Nothing to be done for surrogates.
+ if (bottom >= kLeadSurrogateStart && top <= kTrailSurrogateEnd) return;
+ if (is_one_byte && !RangeContainsLatin1Equivalents(range)) {
+ if (bottom > String::kMaxOneByteCharCode) return;
+ if (top > String::kMaxOneByteCharCode) top = String::kMaxOneByteCharCode;
}
- } else {
- // If this is a range we expand the characters block by block,
- // expanding contiguous subranges (blocks) one at a time.
- // The approach is as follows. For a given start character we
- // look up the remainder of the block that contains it (represented
- // by the end point), for instance we find 'z' if the character
- // is 'c'. A block is characterized by the property
- // that all characters uncanonicalize in the same way, except that
- // each entry in the result is incremented by the distance from the first
- // element. So a-z is a block because 'a' uncanonicalizes to ['a', 'A'] and
- // the k'th letter uncanonicalizes to ['a' + k, 'A' + k].
- // Once we've found the end point we look up its uncanonicalization
- // and produce a range for each element. For instance for [c-f]
- // we look up ['z', 'Z'] and produce [c-f] and [C-F]. We then only
- // add a range if it is not already contained in the input, so [c-f]
- // will be skipped but [C-F] will be added. If this range is not
- // completely contained in a block we do this for all the blocks
- // covered by the range (handling characters that is not in a block
- // as a "singleton block").
- unibrow::uchar range[unibrow::Ecma262UnCanonicalize::kMaxWidth];
- int pos = bottom;
- while (pos <= top) {
- int length = isolate->jsregexp_canonrange()->get(pos, '\0', range);
- uc16 block_end;
- if (length == 0) {
- block_end = pos;
- } else {
- DCHECK_EQ(1, length);
- block_end = range[0];
- }
- int end = (block_end > top) ? top : block_end;
- length = isolate->jsregexp_uncanonicalize()->get(block_end, '\0', range);
+ unibrow::uchar chars[unibrow::Ecma262UnCanonicalize::kMaxWidth];
+ if (top == bottom) {
+ // If this is a singleton we just expand the one character.
+ int length = isolate->jsregexp_uncanonicalize()->get(bottom, '\0', chars);
for (int i = 0; i < length; i++) {
- uc32 c = range[i];
- uc16 range_from = c - (block_end - pos);
- uc16 range_to = c - (block_end - end);
- if (!(bottom <= range_from && range_to <= top)) {
- ranges->Add(CharacterRange(range_from, range_to), zone);
+ uc32 chr = chars[i];
+ if (chr != bottom) {
+ ranges->Add(CharacterRange::Singleton(chars[i]), zone);
}
}
- pos = end + 1;
+ } else {
+ // If this is a range we expand the characters block by block, expanding
+ // contiguous subranges (blocks) one at a time. The approach is as
+ // follows. For a given start character we look up the remainder of the
+ // block that contains it (represented by the end point), for instance we
+ // find 'z' if the character is 'c'. A block is characterized by the
+ // property that all characters uncanonicalize in the same way, except
+ // that each entry in the result is incremented by the distance from the
+ // first element. So a-z is a block because 'a' uncanonicalizes to ['a',
+ // 'A'] and the k'th letter uncanonicalizes to ['a' + k, 'A' + k]. Once
+ // we've found the end point we look up its uncanonicalization and
+ // produce a range for each element. For instance for [c-f] we look up
+ // ['z', 'Z'] and produce [c-f] and [C-F]. We then only add a range if
+ // it is not already contained in the input, so [c-f] will be skipped but
+ // [C-F] will be added. If this range is not completely contained in a
+ // block we do this for all the blocks covered by the range (handling
+ // characters that is not in a block as a "singleton block").
+ unibrow::uchar equivalents[unibrow::Ecma262UnCanonicalize::kMaxWidth];
+ int pos = bottom;
+ while (pos <= top) {
+ int length =
+ isolate->jsregexp_canonrange()->get(pos, '\0', equivalents);
+ uc32 block_end;
+ if (length == 0) {
+ block_end = pos;
+ } else {
+ DCHECK_EQ(1, length);
+ block_end = equivalents[0];
+ }
+ int end = (block_end > top) ? top : block_end;
+ length = isolate->jsregexp_uncanonicalize()->get(block_end, '\0',
+ equivalents);
+ for (int i = 0; i < length; i++) {
+ uc32 c = equivalents[i];
+ uc32 range_from = c - (block_end - pos);
+ uc32 range_to = c - (block_end - end);
+ if (!(bottom <= range_from && range_to <= top)) {
+ ranges->Add(CharacterRange::Range(range_from, range_to), zone);
+ }
+ }
+ pos = end + 1;
+ }
}
}
}
@@ -5672,8 +5996,8 @@
// list[0..count] for the result. Returns the number of resulting
// canonicalized ranges. Inserting a range may collapse existing ranges into
// fewer ranges, so the return value can be anything in the range 1..count+1.
- uc16 from = insert.from();
- uc16 to = insert.to();
+ uc32 from = insert.from();
+ uc32 to = insert.to();
int start_pos = 0;
int end_pos = count;
for (int i = count - 1; i >= 0; i--) {
@@ -5706,7 +6030,7 @@
CharacterRange to_replace = list->at(start_pos);
int new_from = Min(to_replace.from(), from);
int new_to = Max(to_replace.to(), to);
- list->at(start_pos) = CharacterRange(new_from, new_to);
+ list->at(start_pos) = CharacterRange::Range(new_from, new_to);
return count;
}
// Replace a number of existing ranges from start_pos to end_pos - 1.
@@ -5717,7 +6041,7 @@
if (end_pos < count) {
MoveRanges(list, end_pos, start_pos + 1, count - end_pos);
}
- list->at(start_pos) = CharacterRange(new_from, new_to);
+ list->at(start_pos) = CharacterRange::Range(new_from, new_to);
return count - (end_pos - start_pos) + 1;
}
@@ -5773,20 +6097,20 @@
DCHECK(CharacterRange::IsCanonical(ranges));
DCHECK_EQ(0, negated_ranges->length());
int range_count = ranges->length();
- uc16 from = 0;
+ uc32 from = 0;
int i = 0;
if (range_count > 0 && ranges->at(0).from() == 0) {
- from = ranges->at(0).to();
+ from = ranges->at(0).to() + 1;
i = 1;
}
while (i < range_count) {
CharacterRange range = ranges->at(i);
- negated_ranges->Add(CharacterRange(from + 1, range.from() - 1), zone);
- from = range.to();
+ negated_ranges->Add(CharacterRange::Range(from, range.from() - 1), zone);
+ from = range.to() + 1;
i++;
}
- if (from < String::kMaxUtf16CodeUnit) {
- negated_ranges->Add(CharacterRange(from + 1, String::kMaxUtf16CodeUnit),
+ if (from < String::kMaxCodePoint) {
+ negated_ranges->Add(CharacterRange::Range(from, String::kMaxCodePoint),
zone);
}
}
@@ -5838,7 +6162,7 @@
}
-const uc16 DispatchTable::Config::kNoKey = unibrow::Utf8::kBadChar;
+const uc32 DispatchTable::Config::kNoKey = unibrow::Utf8::kBadChar;
void DispatchTable::AddRange(CharacterRange full_range, int value,
@@ -5866,8 +6190,9 @@
if (entry->from() < current.from() && entry->to() >= current.from()) {
// Snap the overlapping range in half around the start point of
// the range we're adding.
- CharacterRange left(entry->from(), current.from() - 1);
- CharacterRange right(current.from(), entry->to());
+ CharacterRange left =
+ CharacterRange::Range(entry->from(), current.from() - 1);
+ CharacterRange right = CharacterRange::Range(current.from(), entry->to());
// The left part of the overlapping range doesn't overlap.
// Truncate the whole entry to be just the left part.
entry->set_to(left.to());
@@ -5919,10 +6244,6 @@
// we're adding so we can just update it and move the start point
// of the range we're adding just past it.
entry->AddValue(value, zone);
- // Bail out if the last interval ended at 0xFFFF since otherwise
- // adding 1 will wrap around to 0.
- if (entry->to() == String::kMaxUtf16CodeUnit)
- break;
DCHECK(entry->to() + 1 > current.from());
current.set_from(entry->to() + 1);
} else {
@@ -5940,7 +6261,7 @@
}
-OutSet* DispatchTable::Get(uc16 value) {
+OutSet* DispatchTable::Get(uc32 value) {
ZoneSplayTree<Config>::Locator loc;
if (!tree()->FindGreatestLessThan(value, &loc))
return empty();
@@ -5990,7 +6311,7 @@
void Analysis::VisitText(TextNode* that) {
- if (ignore_case_) {
+ if (ignore_case()) {
that->MakeCaseIndependent(isolate(), is_one_byte_);
}
EnsureAnalyzed(that->on_success());
@@ -6173,8 +6494,7 @@
void AddDispatchRange::Call(uc32 from, DispatchTable::Entry entry) {
- CharacterRange range(from, entry.to());
- constructor_->AddRange(range);
+ constructor_->AddRange(CharacterRange::Range(from, entry.to()));
}
@@ -6212,16 +6532,16 @@
for (int i = 0; i < ranges->length(); i++) {
CharacterRange range = ranges->at(i);
if (last < range.from())
- AddRange(CharacterRange(last, range.from() - 1));
+ AddRange(CharacterRange::Range(last, range.from() - 1));
if (range.to() >= last) {
- if (range.to() == String::kMaxUtf16CodeUnit) {
+ if (range.to() == String::kMaxCodePoint) {
return;
} else {
last = range.to() + 1;
}
}
}
- AddRange(CharacterRange(last, String::kMaxUtf16CodeUnit));
+ AddRange(CharacterRange::Range(last, String::kMaxCodePoint));
}
@@ -6230,7 +6550,7 @@
switch (elm.text_type()) {
case TextElement::ATOM: {
uc16 c = elm.atom()->data()[0];
- AddRange(CharacterRange(c, c));
+ AddRange(CharacterRange::Range(c, c));
break;
}
case TextElement::CHAR_CLASS: {
@@ -6257,14 +6577,48 @@
}
+RegExpNode* OptionallyStepBackToLeadSurrogate(RegExpCompiler* compiler,
+ RegExpNode* on_success) {
+ // If the regexp matching starts within a surrogate pair, step back
+ // to the lead surrogate and start matching from there.
+ DCHECK(!compiler->read_backward());
+ Zone* zone = compiler->zone();
+ ZoneList<CharacterRange>* lead_surrogates = CharacterRange::List(
+ zone, CharacterRange::Range(kLeadSurrogateStart, kLeadSurrogateEnd));
+ ZoneList<CharacterRange>* trail_surrogates = CharacterRange::List(
+ zone, CharacterRange::Range(kTrailSurrogateStart, kTrailSurrogateEnd));
+
+ ChoiceNode* optional_step_back = new (zone) ChoiceNode(2, zone);
+
+ int stack_register = compiler->UnicodeLookaroundStackRegister();
+ int position_register = compiler->UnicodeLookaroundPositionRegister();
+ RegExpNode* step_back = TextNode::CreateForCharacterRanges(
+ zone, lead_surrogates, true, on_success);
+ RegExpLookaround::Builder builder(true, step_back, stack_register,
+ position_register);
+ RegExpNode* match_trail = TextNode::CreateForCharacterRanges(
+ zone, trail_surrogates, false, builder.on_match_success());
+
+ optional_step_back->AddAlternative(
+ GuardedAlternative(builder.ForMatch(match_trail)));
+ optional_step_back->AddAlternative(GuardedAlternative(on_success));
+
+ return optional_step_back;
+}
+
+
RegExpEngine::CompilationResult RegExpEngine::Compile(
- Isolate* isolate, Zone* zone, RegExpCompileData* data, bool ignore_case,
- bool is_global, bool is_multiline, bool is_sticky, Handle<String> pattern,
+ Isolate* isolate, Zone* zone, RegExpCompileData* data,
+ JSRegExp::Flags flags, Handle<String> pattern,
Handle<String> sample_subject, bool is_one_byte) {
if ((data->capture_count + 1) * 2 - 1 > RegExpMacroAssembler::kMaxRegister) {
return IrregexpRegExpTooBig(isolate);
}
- RegExpCompiler compiler(isolate, zone, data->capture_count, ignore_case,
+ bool ignore_case = flags & JSRegExp::kIgnoreCase;
+ bool is_sticky = flags & JSRegExp::kSticky;
+ bool is_global = flags & JSRegExp::kGlobal;
+ bool is_unicode = flags & JSRegExp::kUnicode;
+ RegExpCompiler compiler(isolate, zone, data->capture_count, flags,
is_one_byte);
if (compiler.optimize()) compiler.set_optimize(!TooMuchRegExpCode(pattern));
@@ -6316,11 +6670,13 @@
if (node != NULL) {
node = node->FilterOneByte(RegExpCompiler::kMaxRecursion, ignore_case);
}
+ } else if (compiler.unicode() && (is_global || is_sticky)) {
+ node = OptionallyStepBackToLeadSurrogate(&compiler, node);
}
if (node == NULL) node = new(zone) EndNode(EndNode::BACKTRACK, zone);
data->node = node;
- Analysis analysis(isolate, ignore_case, is_one_byte);
+ Analysis analysis(isolate, flags, is_one_byte);
analysis.EnsureAnalyzed(node);
if (analysis.has_failed()) {
const char* error_message = analysis.error_message();
@@ -6381,10 +6737,13 @@
}
if (is_global) {
- macro_assembler.set_global_mode(
- (data->tree->min_match() > 0)
- ? RegExpMacroAssembler::GLOBAL_NO_ZERO_LENGTH_CHECK
- : RegExpMacroAssembler::GLOBAL);
+ RegExpMacroAssembler::GlobalMode mode = RegExpMacroAssembler::GLOBAL;
+ if (data->tree->min_match() > 0) {
+ mode = RegExpMacroAssembler::GLOBAL_NO_ZERO_LENGTH_CHECK;
+ } else if (is_unicode) {
+ mode = RegExpMacroAssembler::GLOBAL_UNICODE;
+ }
+ macro_assembler.set_global_mode(mode);
}
return compiler.Assemble(¯o_assembler,