aboutsummaryrefslogtreecommitdiff
path: root/deps/v8/src/regexp
diff options
context:
space:
mode:
authorMichaël Zasso <targos@protonmail.com>2019-08-16 11:32:46 +0200
committerMichaël Zasso <targos@protonmail.com>2019-08-19 09:25:23 +0200
commite31f0a7d25668d3c1531294d2ef44a9f3bde4ef4 (patch)
tree6c6bed9804be9df6162b2483f0a56f371f66464d /deps/v8/src/regexp
parentec16fdae540adaf710b1a86c620170b2880088f0 (diff)
downloadandroid-node-v8-e31f0a7d25668d3c1531294d2ef44a9f3bde4ef4.tar.gz
android-node-v8-e31f0a7d25668d3c1531294d2ef44a9f3bde4ef4.tar.bz2
android-node-v8-e31f0a7d25668d3c1531294d2ef44a9f3bde4ef4.zip
deps: update V8 to 7.7.299.4
PR-URL: https://github.com/nodejs/node/pull/28918 Reviewed-By: Colin Ihrig <cjihrig@gmail.com> Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl> Reviewed-By: Jiawen Geng <technicalcute@gmail.com> Reviewed-By: Rich Trott <rtrott@gmail.com>
Diffstat (limited to 'deps/v8/src/regexp')
-rw-r--r--deps/v8/src/regexp/OWNERS2
-rw-r--r--deps/v8/src/regexp/jsregexp-inl.h86
-rw-r--r--deps/v8/src/regexp/jsregexp.cc7055
-rw-r--r--deps/v8/src/regexp/jsregexp.h1548
-rw-r--r--deps/v8/src/regexp/regexp-ast.h24
-rw-r--r--deps/v8/src/regexp/regexp-bytecode-generator-inl.h (renamed from deps/v8/src/regexp/regexp-macro-assembler-irregexp-inl.h)24
-rw-r--r--deps/v8/src/regexp/regexp-bytecode-generator.cc (renamed from deps/v8/src/regexp/regexp-macro-assembler-irregexp.cc)221
-rw-r--r--deps/v8/src/regexp/regexp-bytecode-generator.h (renamed from deps/v8/src/regexp/regexp-macro-assembler-irregexp.h)39
-rw-r--r--deps/v8/src/regexp/regexp-bytecodes.h (renamed from deps/v8/src/regexp/bytecodes-irregexp.h)11
-rw-r--r--deps/v8/src/regexp/regexp-compiler-tonode.cc1678
-rw-r--r--deps/v8/src/regexp/regexp-compiler.cc3551
-rw-r--r--deps/v8/src/regexp/regexp-compiler.h657
-rw-r--r--deps/v8/src/regexp/regexp-dotprinter.cc244
-rw-r--r--deps/v8/src/regexp/regexp-dotprinter.h23
-rw-r--r--deps/v8/src/regexp/regexp-interpreter.cc (renamed from deps/v8/src/regexp/interpreter-irregexp.cc)269
-rw-r--r--deps/v8/src/regexp/regexp-interpreter.h (renamed from deps/v8/src/regexp/interpreter-irregexp.h)18
-rw-r--r--deps/v8/src/regexp/regexp-macro-assembler-arch.h30
-rw-r--r--deps/v8/src/regexp/regexp-macro-assembler.cc9
-rw-r--r--deps/v8/src/regexp/regexp-macro-assembler.h8
-rw-r--r--deps/v8/src/regexp/regexp-nodes.h675
-rw-r--r--deps/v8/src/regexp/regexp-parser.cc66
-rw-r--r--deps/v8/src/regexp/regexp-parser.h10
-rw-r--r--deps/v8/src/regexp/regexp-utils.cc6
-rw-r--r--deps/v8/src/regexp/regexp.cc1018
-rw-r--r--deps/v8/src/regexp/regexp.h177
25 files changed, 8365 insertions, 9084 deletions
diff --git a/deps/v8/src/regexp/OWNERS b/deps/v8/src/regexp/OWNERS
index 7f916e12ea..250c8c6b88 100644
--- a/deps/v8/src/regexp/OWNERS
+++ b/deps/v8/src/regexp/OWNERS
@@ -1,5 +1,3 @@
-set noparent
-
jgruber@chromium.org
yangguo@chromium.org
diff --git a/deps/v8/src/regexp/jsregexp-inl.h b/deps/v8/src/regexp/jsregexp-inl.h
deleted file mode 100644
index b542add17b..0000000000
--- a/deps/v8/src/regexp/jsregexp-inl.h
+++ /dev/null
@@ -1,86 +0,0 @@
-// Copyright 2013 the V8 project authors. All rights reserved.
-// Use of this source code is governed by a BSD-style license that can be
-// found in the LICENSE file.
-
-
-#ifndef V8_REGEXP_JSREGEXP_INL_H_
-#define V8_REGEXP_JSREGEXP_INL_H_
-
-#include "src/objects/js-regexp-inl.h"
-#include "src/objects/objects.h"
-#include "src/regexp/jsregexp.h"
-#include "src/utils/allocation.h"
-
-namespace v8 {
-namespace internal {
-
-
-RegExpImpl::GlobalCache::~GlobalCache() {
- // Deallocate the register array if we allocated it in the constructor
- // (as opposed to using the existing jsregexp_static_offsets_vector).
- if (register_array_size_ > Isolate::kJSRegexpStaticOffsetsVectorSize) {
- DeleteArray(register_array_);
- }
-}
-
-
-int32_t* RegExpImpl::GlobalCache::FetchNext() {
- current_match_index_++;
- if (current_match_index_ >= num_matches_) {
- // Current batch of results exhausted.
- // Fail if last batch was not even fully filled.
- if (num_matches_ < max_matches_) {
- num_matches_ = 0; // Signal failed match.
- return nullptr;
- }
-
- int32_t* last_match =
- &register_array_[(current_match_index_ - 1) * registers_per_match_];
- int last_end_index = last_match[1];
-
- if (regexp_->TypeTag() == JSRegExp::ATOM) {
- num_matches_ =
- RegExpImpl::AtomExecRaw(isolate_, regexp_, subject_, last_end_index,
- register_array_, register_array_size_);
- } else {
- int last_start_index = last_match[0];
- if (last_start_index == last_end_index) {
- // Zero-length match. Advance by one code point.
- last_end_index = AdvanceZeroLength(last_end_index);
- }
- if (last_end_index > subject_->length()) {
- num_matches_ = 0; // Signal failed match.
- return nullptr;
- }
- num_matches_ = RegExpImpl::IrregexpExecRaw(
- isolate_, regexp_, subject_, last_end_index, register_array_,
- register_array_size_);
- }
-
- if (num_matches_ <= 0) return nullptr;
- current_match_index_ = 0;
- return register_array_;
- } else {
- return &register_array_[current_match_index_ * registers_per_match_];
- }
-}
-
-
-int32_t* RegExpImpl::GlobalCache::LastSuccessfulMatch() {
- int index = current_match_index_ * registers_per_match_;
- if (num_matches_ == 0) {
- // After a failed match we shift back by one result.
- index -= registers_per_match_;
- }
- return &register_array_[index];
-}
-
-RegExpEngine::CompilationResult::CompilationResult(Isolate* isolate,
- const char* error_message)
- : error_message(error_message),
- code(ReadOnlyRoots(isolate).the_hole_value()) {}
-
-} // namespace internal
-} // namespace v8
-
-#endif // V8_REGEXP_JSREGEXP_INL_H_
diff --git a/deps/v8/src/regexp/jsregexp.cc b/deps/v8/src/regexp/jsregexp.cc
deleted file mode 100644
index a6f3a5ebcb..0000000000
--- a/deps/v8/src/regexp/jsregexp.cc
+++ /dev/null
@@ -1,7055 +0,0 @@
-// Copyright 2012 the V8 project authors. All rights reserved.
-// Use of this source code is governed by a BSD-style license that can be
-// found in the LICENSE file.
-
-#include "src/regexp/jsregexp.h"
-
-#include <memory>
-#include <vector>
-
-#include "src/base/platform/platform.h"
-#include "src/codegen/compilation-cache.h"
-#include "src/diagnostics/code-tracer.h"
-#include "src/execution/execution.h"
-#include "src/execution/isolate-inl.h"
-#include "src/execution/message-template.h"
-#include "src/heap/factory.h"
-#include "src/heap/heap-inl.h"
-#include "src/objects/elements.h"
-#include "src/regexp/interpreter-irregexp.h"
-#include "src/regexp/jsregexp-inl.h"
-#include "src/regexp/regexp-macro-assembler-irregexp.h"
-#include "src/regexp/regexp-macro-assembler-tracer.h"
-#include "src/regexp/regexp-macro-assembler.h"
-#include "src/regexp/regexp-parser.h"
-#include "src/regexp/regexp-stack.h"
-#ifdef V8_INTL_SUPPORT
-#include "src/regexp/special-case.h"
-#endif // V8_INTL_SUPPORT
-#include "src/runtime/runtime.h"
-#include "src/strings/string-search.h"
-#include "src/strings/unicode-decoder.h"
-#include "src/strings/unicode-inl.h"
-#include "src/utils/ostreams.h"
-#include "src/utils/splay-tree-inl.h"
-#include "src/zone/zone-list-inl.h"
-
-#ifdef V8_INTL_SUPPORT
-#include "unicode/locid.h"
-#include "unicode/uniset.h"
-#include "unicode/utypes.h"
-#endif // V8_INTL_SUPPORT
-
-#if V8_TARGET_ARCH_IA32
-#include "src/regexp/ia32/regexp-macro-assembler-ia32.h"
-#elif V8_TARGET_ARCH_X64
-#include "src/regexp/x64/regexp-macro-assembler-x64.h"
-#elif V8_TARGET_ARCH_ARM64
-#include "src/regexp/arm64/regexp-macro-assembler-arm64.h"
-#elif V8_TARGET_ARCH_ARM
-#include "src/regexp/arm/regexp-macro-assembler-arm.h"
-#elif V8_TARGET_ARCH_PPC
-#include "src/regexp/ppc/regexp-macro-assembler-ppc.h"
-#elif V8_TARGET_ARCH_S390
-#include "src/regexp/s390/regexp-macro-assembler-s390.h"
-#elif V8_TARGET_ARCH_MIPS
-#include "src/regexp/mips/regexp-macro-assembler-mips.h"
-#elif V8_TARGET_ARCH_MIPS64
-#include "src/regexp/mips64/regexp-macro-assembler-mips64.h"
-#else
-#error Unsupported target architecture.
-#endif
-
-namespace v8 {
-namespace internal {
-
-V8_WARN_UNUSED_RESULT
-static inline MaybeHandle<Object> ThrowRegExpException(
- Isolate* isolate, Handle<JSRegExp> re, Handle<String> pattern,
- Handle<String> error_text) {
- THROW_NEW_ERROR(isolate, NewSyntaxError(MessageTemplate::kMalformedRegExp,
- pattern, error_text),
- Object);
-}
-
-inline void ThrowRegExpException(Isolate* isolate, Handle<JSRegExp> re,
- Handle<String> error_text) {
- USE(ThrowRegExpException(isolate, re, Handle<String>(re->Pattern(), isolate),
- error_text));
-}
-
-
-ContainedInLattice AddRange(ContainedInLattice containment,
- const int* ranges,
- int ranges_length,
- Interval new_range) {
- DCHECK_EQ(1, ranges_length & 1);
- DCHECK_EQ(String::kMaxCodePoint + 1, ranges[ranges_length - 1]);
- if (containment == kLatticeUnknown) return containment;
- bool inside = false;
- int last = 0;
- for (int i = 0; i < ranges_length; inside = !inside, last = ranges[i], i++) {
- // Consider the range from last to ranges[i].
- // We haven't got to the new range yet.
- if (ranges[i] <= new_range.from()) continue;
- // New range is wholly inside last-ranges[i]. Note that new_range.to() is
- // inclusive, but the values in ranges are not.
- if (last <= new_range.from() && new_range.to() < ranges[i]) {
- return Combine(containment, inside ? kLatticeIn : kLatticeOut);
- }
- return kLatticeUnknown;
- }
- return containment;
-}
-
-// More makes code generation slower, less makes V8 benchmark score lower.
-const int kMaxLookaheadForBoyerMoore = 8;
-// In a 3-character pattern you can maximally step forwards 3 characters
-// at a time, which is not always enough to pay for the extra logic.
-const int kPatternTooShortForBoyerMoore = 2;
-
-// Identifies the sort of regexps where the regexp engine is faster
-// than the code used for atom matches.
-static bool HasFewDifferentCharacters(Handle<String> pattern) {
- int length = Min(kMaxLookaheadForBoyerMoore, pattern->length());
- if (length <= kPatternTooShortForBoyerMoore) return false;
- const int kMod = 128;
- bool character_found[kMod];
- int different = 0;
- memset(&character_found[0], 0, sizeof(character_found));
- for (int i = 0; i < length; i++) {
- int ch = (pattern->Get(i) & (kMod - 1));
- if (!character_found[ch]) {
- character_found[ch] = true;
- different++;
- // We declare a regexp low-alphabet if it has at least 3 times as many
- // characters as it has different characters.
- if (different * 3 > length) return false;
- }
- }
- return true;
-}
-
-// Generic RegExp methods. Dispatches to implementation specific methods.
-
-MaybeHandle<Object> RegExpImpl::Compile(Isolate* isolate, Handle<JSRegExp> re,
- Handle<String> pattern,
- JSRegExp::Flags flags) {
- DCHECK(pattern->IsFlat());
-
- Zone zone(isolate->allocator(), ZONE_NAME);
- CompilationCache* compilation_cache = isolate->compilation_cache();
- MaybeHandle<FixedArray> maybe_cached =
- compilation_cache->LookupRegExp(pattern, flags);
- Handle<FixedArray> cached;
- if (maybe_cached.ToHandle(&cached)) {
- re->set_data(*cached);
- return re;
- }
-
- PostponeInterruptsScope postpone(isolate);
- RegExpCompileData parse_result;
- FlatStringReader reader(isolate, pattern);
- DCHECK(!isolate->has_pending_exception());
- if (!RegExpParser::ParseRegExp(isolate, &zone, &reader, flags,
- &parse_result)) {
- // Throw an exception if we fail to parse the pattern.
- return ThrowRegExpException(isolate, re, pattern, parse_result.error);
- }
-
- bool has_been_compiled = false;
-
- if (parse_result.simple && !IgnoreCase(flags) && !IsSticky(flags) &&
- !HasFewDifferentCharacters(pattern)) {
- // Parse-tree is a single atom that is equal to the pattern.
- AtomCompile(isolate, re, pattern, flags, pattern);
- has_been_compiled = true;
- } else if (parse_result.tree->IsAtom() && !IsSticky(flags) &&
- parse_result.capture_count == 0) {
- RegExpAtom* atom = parse_result.tree->AsAtom();
- Vector<const uc16> atom_pattern = atom->data();
- Handle<String> atom_string;
- ASSIGN_RETURN_ON_EXCEPTION(
- isolate, atom_string,
- isolate->factory()->NewStringFromTwoByte(atom_pattern), Object);
- if (!IgnoreCase(atom->flags()) && !HasFewDifferentCharacters(atom_string)) {
- AtomCompile(isolate, re, pattern, flags, atom_string);
- has_been_compiled = true;
- }
- }
- if (!has_been_compiled) {
- IrregexpInitialize(isolate, re, pattern, flags, parse_result.capture_count);
- }
- DCHECK(re->data().IsFixedArray());
- // Compilation succeeded so the data is set on the regexp
- // and we can store it in the cache.
- Handle<FixedArray> data(FixedArray::cast(re->data()), isolate);
- compilation_cache->PutRegExp(pattern, flags, data);
-
- return re;
-}
-
-MaybeHandle<Object> RegExpImpl::Exec(Isolate* isolate, Handle<JSRegExp> regexp,
- Handle<String> subject, int index,
- Handle<RegExpMatchInfo> last_match_info) {
- switch (regexp->TypeTag()) {
- case JSRegExp::ATOM:
- return AtomExec(isolate, regexp, subject, index, last_match_info);
- case JSRegExp::IRREGEXP: {
- return IrregexpExec(isolate, regexp, subject, index, last_match_info);
- }
- default:
- UNREACHABLE();
- }
-}
-
-
-// RegExp Atom implementation: Simple string search using indexOf.
-
-void RegExpImpl::AtomCompile(Isolate* isolate, Handle<JSRegExp> re,
- Handle<String> pattern, JSRegExp::Flags flags,
- Handle<String> match_pattern) {
- isolate->factory()->SetRegExpAtomData(re, JSRegExp::ATOM, pattern, flags,
- match_pattern);
-}
-
-static void SetAtomLastCapture(Isolate* isolate,
- Handle<RegExpMatchInfo> last_match_info,
- String subject, int from, int to) {
- SealHandleScope shs(isolate);
- last_match_info->SetNumberOfCaptureRegisters(2);
- last_match_info->SetLastSubject(subject);
- last_match_info->SetLastInput(subject);
- last_match_info->SetCapture(0, from);
- last_match_info->SetCapture(1, to);
-}
-
-int RegExpImpl::AtomExecRaw(Isolate* isolate, Handle<JSRegExp> regexp,
- Handle<String> subject, int index, int32_t* output,
- int output_size) {
- DCHECK_LE(0, index);
- DCHECK_LE(index, subject->length());
-
- subject = String::Flatten(isolate, subject);
- DisallowHeapAllocation no_gc; // ensure vectors stay valid
-
- String needle = String::cast(regexp->DataAt(JSRegExp::kAtomPatternIndex));
- int needle_len = needle.length();
- DCHECK(needle.IsFlat());
- DCHECK_LT(0, needle_len);
-
- if (index + needle_len > subject->length()) {
- return RegExpImpl::RE_FAILURE;
- }
-
- for (int i = 0; i < output_size; i += 2) {
- String::FlatContent needle_content = needle.GetFlatContent(no_gc);
- String::FlatContent subject_content = subject->GetFlatContent(no_gc);
- DCHECK(needle_content.IsFlat());
- DCHECK(subject_content.IsFlat());
- // dispatch on type of strings
- index =
- (needle_content.IsOneByte()
- ? (subject_content.IsOneByte()
- ? SearchString(isolate, subject_content.ToOneByteVector(),
- needle_content.ToOneByteVector(), index)
- : SearchString(isolate, subject_content.ToUC16Vector(),
- needle_content.ToOneByteVector(), index))
- : (subject_content.IsOneByte()
- ? SearchString(isolate, subject_content.ToOneByteVector(),
- needle_content.ToUC16Vector(), index)
- : SearchString(isolate, subject_content.ToUC16Vector(),
- needle_content.ToUC16Vector(), index)));
- if (index == -1) {
- return i / 2; // Return number of matches.
- } else {
- output[i] = index;
- output[i+1] = index + needle_len;
- index += needle_len;
- }
- }
- return output_size / 2;
-}
-
-Handle<Object> RegExpImpl::AtomExec(Isolate* isolate, Handle<JSRegExp> re,
- Handle<String> subject, int index,
- Handle<RegExpMatchInfo> last_match_info) {
- static const int kNumRegisters = 2;
- STATIC_ASSERT(kNumRegisters <= Isolate::kJSRegexpStaticOffsetsVectorSize);
- int32_t* output_registers = isolate->jsregexp_static_offsets_vector();
-
- int res =
- AtomExecRaw(isolate, re, subject, index, output_registers, kNumRegisters);
-
- if (res == RegExpImpl::RE_FAILURE) return isolate->factory()->null_value();
-
- DCHECK_EQ(res, RegExpImpl::RE_SUCCESS);
- SealHandleScope shs(isolate);
- SetAtomLastCapture(isolate, last_match_info, *subject, output_registers[0],
- output_registers[1]);
- return last_match_info;
-}
-
-
-// Irregexp implementation.
-
-// Ensures that the regexp object contains a compiled version of the
-// source for either one-byte or two-byte subject strings.
-// If the compiled version doesn't already exist, it is compiled
-// from the source pattern.
-// If compilation fails, an exception is thrown and this function
-// returns false.
-bool RegExpImpl::EnsureCompiledIrregexp(Isolate* isolate, Handle<JSRegExp> re,
- Handle<String> sample_subject,
- bool is_one_byte) {
- Object compiled_code = re->DataAt(JSRegExp::code_index(is_one_byte));
- if (compiled_code != Smi::FromInt(JSRegExp::kUninitializedValue)) {
- DCHECK(FLAG_regexp_interpret_all ? compiled_code.IsByteArray()
- : compiled_code.IsCode());
- return true;
- }
- return CompileIrregexp(isolate, re, sample_subject, is_one_byte);
-}
-
-bool RegExpImpl::CompileIrregexp(Isolate* isolate, Handle<JSRegExp> re,
- Handle<String> sample_subject,
- bool is_one_byte) {
- // Compile the RegExp.
- Zone zone(isolate->allocator(), ZONE_NAME);
- PostponeInterruptsScope postpone(isolate);
-#ifdef DEBUG
- Object entry = re->DataAt(JSRegExp::code_index(is_one_byte));
- // When arriving here entry can only be a smi representing an uncompiled
- // regexp.
- DCHECK(entry.IsSmi());
- int entry_value = Smi::ToInt(entry);
- DCHECK_EQ(JSRegExp::kUninitializedValue, entry_value);
-#endif
-
- JSRegExp::Flags flags = re->GetFlags();
-
- Handle<String> pattern(re->Pattern(), isolate);
- pattern = String::Flatten(isolate, pattern);
- RegExpCompileData compile_data;
- FlatStringReader reader(isolate, pattern);
- 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(isolate, re, pattern, compile_data.error));
- return false;
- }
- RegExpEngine::CompilationResult result =
- RegExpEngine::Compile(isolate, &zone, &compile_data, flags, pattern,
- sample_subject, is_one_byte);
- if (result.error_message != nullptr) {
- // Unable to compile regexp.
- if (FLAG_correctness_fuzzer_suppressions &&
- strncmp(result.error_message, "Stack overflow", 15) == 0) {
- FATAL("Aborting on stack overflow");
- }
- Handle<String> error_message = isolate->factory()->NewStringFromUtf8(
- CStrVector(result.error_message)).ToHandleChecked();
- ThrowRegExpException(isolate, re, error_message);
- return false;
- }
-
- Handle<FixedArray> data =
- Handle<FixedArray>(FixedArray::cast(re->data()), isolate);
- data->set(JSRegExp::code_index(is_one_byte), result.code);
- SetIrregexpCaptureNameMap(*data, compile_data.capture_name_map);
- int register_max = IrregexpMaxRegisterCount(*data);
- if (result.num_registers > register_max) {
- SetIrregexpMaxRegisterCount(*data, result.num_registers);
- }
-
- return true;
-}
-
-int RegExpImpl::IrregexpMaxRegisterCount(FixedArray re) {
- return Smi::cast(re.get(JSRegExp::kIrregexpMaxRegisterCountIndex)).value();
-}
-
-void RegExpImpl::SetIrregexpMaxRegisterCount(FixedArray re, int value) {
- re.set(JSRegExp::kIrregexpMaxRegisterCountIndex, Smi::FromInt(value));
-}
-
-void RegExpImpl::SetIrregexpCaptureNameMap(FixedArray re,
- Handle<FixedArray> value) {
- if (value.is_null()) {
- re.set(JSRegExp::kIrregexpCaptureNameMapIndex, Smi::kZero);
- } else {
- re.set(JSRegExp::kIrregexpCaptureNameMapIndex, *value);
- }
-}
-
-int RegExpImpl::IrregexpNumberOfCaptures(FixedArray re) {
- return Smi::ToInt(re.get(JSRegExp::kIrregexpCaptureCountIndex));
-}
-
-int RegExpImpl::IrregexpNumberOfRegisters(FixedArray re) {
- return Smi::ToInt(re.get(JSRegExp::kIrregexpMaxRegisterCountIndex));
-}
-
-ByteArray RegExpImpl::IrregexpByteCode(FixedArray re, bool is_one_byte) {
- return ByteArray::cast(re.get(JSRegExp::code_index(is_one_byte)));
-}
-
-Code RegExpImpl::IrregexpNativeCode(FixedArray re, bool is_one_byte) {
- return Code::cast(re.get(JSRegExp::code_index(is_one_byte)));
-}
-
-void RegExpImpl::IrregexpInitialize(Isolate* isolate, Handle<JSRegExp> re,
- Handle<String> pattern,
- JSRegExp::Flags flags, int capture_count) {
- // Initialize compiled code entries to null.
- isolate->factory()->SetRegExpIrregexpData(re, JSRegExp::IRREGEXP, pattern,
- flags, capture_count);
-}
-
-int RegExpImpl::IrregexpPrepare(Isolate* isolate, Handle<JSRegExp> regexp,
- Handle<String> subject) {
- DCHECK(subject->IsFlat());
-
- // Check representation of the underlying storage.
- bool is_one_byte = String::IsOneByteRepresentationUnderneath(*subject);
- if (!EnsureCompiledIrregexp(isolate, regexp, subject, is_one_byte)) return -1;
-
- if (FLAG_regexp_interpret_all) {
- // Byte-code regexp needs space allocated for all its registers.
- // The result captures are copied to the start of the registers array
- // if the match succeeds. This way those registers are not clobbered
- // when we set the last match info from last successful match.
- return IrregexpNumberOfRegisters(FixedArray::cast(regexp->data())) +
- (IrregexpNumberOfCaptures(FixedArray::cast(regexp->data())) + 1) * 2;
- } else {
- // Native regexp only needs room to output captures. Registers are handled
- // internally.
- return (IrregexpNumberOfCaptures(FixedArray::cast(regexp->data())) + 1) * 2;
- }
-}
-
-int RegExpImpl::IrregexpExecRaw(Isolate* isolate, Handle<JSRegExp> regexp,
- Handle<String> subject, int index,
- int32_t* output, int output_size) {
- Handle<FixedArray> irregexp(FixedArray::cast(regexp->data()), isolate);
-
- DCHECK_LE(0, index);
- DCHECK_LE(index, subject->length());
- DCHECK(subject->IsFlat());
-
- bool is_one_byte = String::IsOneByteRepresentationUnderneath(*subject);
-
- if (!FLAG_regexp_interpret_all) {
- DCHECK(output_size >= (IrregexpNumberOfCaptures(*irregexp) + 1) * 2);
- do {
- EnsureCompiledIrregexp(isolate, regexp, subject, is_one_byte);
- Handle<Code> code(IrregexpNativeCode(*irregexp, is_one_byte), isolate);
- // The stack is used to allocate registers for the compiled regexp code.
- // This means that in case of failure, the output registers array is left
- // untouched and contains the capture results from the previous successful
- // match. We can use that to set the last match info lazily.
- int res = NativeRegExpMacroAssembler::Match(code, subject, output,
- output_size, index, isolate);
- if (res != NativeRegExpMacroAssembler::RETRY) {
- DCHECK(res != NativeRegExpMacroAssembler::EXCEPTION ||
- isolate->has_pending_exception());
- STATIC_ASSERT(static_cast<int>(NativeRegExpMacroAssembler::SUCCESS) ==
- RE_SUCCESS);
- STATIC_ASSERT(static_cast<int>(NativeRegExpMacroAssembler::FAILURE) ==
- RE_FAILURE);
- STATIC_ASSERT(static_cast<int>(NativeRegExpMacroAssembler::EXCEPTION) ==
- RE_EXCEPTION);
- return res;
- }
- // If result is RETRY, the string has changed representation, and we
- // must restart from scratch.
- // In this case, it means we must make sure we are prepared to handle
- // the, potentially, different subject (the string can switch between
- // being internal and external, and even between being Latin1 and UC16,
- // but the characters are always the same).
- IrregexpPrepare(isolate, regexp, subject);
- is_one_byte = String::IsOneByteRepresentationUnderneath(*subject);
- } while (true);
- UNREACHABLE();
- } else {
- DCHECK(FLAG_regexp_interpret_all);
- DCHECK(output_size >= IrregexpNumberOfRegisters(*irregexp));
- // We must have done EnsureCompiledIrregexp, so we can get the number of
- // registers.
- int number_of_capture_registers =
- (IrregexpNumberOfCaptures(*irregexp) + 1) * 2;
- int32_t* raw_output = &output[number_of_capture_registers];
-
- do {
- // We do not touch the actual capture result registers until we know there
- // has been a match so that we can use those capture results to set the
- // last match info.
- for (int i = number_of_capture_registers - 1; i >= 0; i--) {
- raw_output[i] = -1;
- }
- Handle<ByteArray> byte_codes(IrregexpByteCode(*irregexp, is_one_byte),
- isolate);
-
- IrregexpInterpreter::Result result = IrregexpInterpreter::Match(
- isolate, byte_codes, subject, raw_output, index);
- DCHECK_IMPLIES(result == IrregexpInterpreter::EXCEPTION,
- isolate->has_pending_exception());
-
- switch (result) {
- case IrregexpInterpreter::SUCCESS:
- // Copy capture results to the start of the registers array.
- MemCopy(output, raw_output,
- number_of_capture_registers * sizeof(int32_t));
- return result;
- case IrregexpInterpreter::EXCEPTION:
- case IrregexpInterpreter::FAILURE:
- return result;
- case IrregexpInterpreter::RETRY:
- // The string has changed representation, and we must restart the
- // match.
- is_one_byte = String::IsOneByteRepresentationUnderneath(*subject);
- EnsureCompiledIrregexp(isolate, regexp, subject, is_one_byte);
- break;
- }
- } while (true);
- UNREACHABLE();
- }
-}
-
-MaybeHandle<Object> RegExpImpl::IrregexpExec(
- Isolate* isolate, Handle<JSRegExp> regexp, Handle<String> subject,
- int previous_index, Handle<RegExpMatchInfo> last_match_info) {
- DCHECK_EQ(regexp->TypeTag(), JSRegExp::IRREGEXP);
-
- subject = String::Flatten(isolate, subject);
-
- // Prepare space for the return values.
-#ifdef DEBUG
- if (FLAG_regexp_interpret_all && FLAG_trace_regexp_bytecodes) {
- String pattern = regexp->Pattern();
- PrintF("\n\nRegexp match: /%s/\n\n", pattern.ToCString().get());
- PrintF("\n\nSubject string: '%s'\n\n", subject->ToCString().get());
- }
-#endif
- int required_registers =
- RegExpImpl::IrregexpPrepare(isolate, regexp, subject);
- if (required_registers < 0) {
- // Compiling failed with an exception.
- DCHECK(isolate->has_pending_exception());
- return MaybeHandle<Object>();
- }
-
- int32_t* output_registers = nullptr;
- if (required_registers > Isolate::kJSRegexpStaticOffsetsVectorSize) {
- output_registers = NewArray<int32_t>(required_registers);
- }
- std::unique_ptr<int32_t[]> auto_release(output_registers);
- if (output_registers == nullptr) {
- output_registers = isolate->jsregexp_static_offsets_vector();
- }
-
- int res =
- RegExpImpl::IrregexpExecRaw(isolate, regexp, subject, previous_index,
- output_registers, required_registers);
- if (res == RE_SUCCESS) {
- int capture_count =
- IrregexpNumberOfCaptures(FixedArray::cast(regexp->data()));
- return SetLastMatchInfo(isolate, last_match_info, subject, capture_count,
- output_registers);
- }
- if (res == RE_EXCEPTION) {
- DCHECK(isolate->has_pending_exception());
- return MaybeHandle<Object>();
- }
- DCHECK(res == RE_FAILURE);
- return isolate->factory()->null_value();
-}
-
-Handle<RegExpMatchInfo> RegExpImpl::SetLastMatchInfo(
- Isolate* isolate, Handle<RegExpMatchInfo> last_match_info,
- Handle<String> subject, int capture_count, int32_t* match) {
- // This is the only place where match infos can grow. If, after executing the
- // regexp, RegExpExecStub finds that the match info is too small, it restarts
- // execution in RegExpImpl::Exec, which finally grows the match info right
- // here.
-
- int capture_register_count = (capture_count + 1) * 2;
- Handle<RegExpMatchInfo> result = RegExpMatchInfo::ReserveCaptures(
- isolate, last_match_info, capture_register_count);
- result->SetNumberOfCaptureRegisters(capture_register_count);
-
- if (*result != *last_match_info) {
- if (*last_match_info == *isolate->regexp_last_match_info()) {
- // This inner condition is only needed for special situations like the
- // regexp fuzzer, where we pass our own custom RegExpMatchInfo to
- // RegExpImpl::Exec; there actually want to bypass the Isolate's match
- // info and execute the regexp without side effects.
- isolate->native_context()->set_regexp_last_match_info(*result);
- }
- }
-
- DisallowHeapAllocation no_allocation;
- if (match != nullptr) {
- for (int i = 0; i < capture_register_count; i += 2) {
- result->SetCapture(i, match[i]);
- result->SetCapture(i + 1, match[i + 1]);
- }
- }
- result->SetLastSubject(*subject);
- result->SetLastInput(*subject);
- return result;
-}
-
-RegExpImpl::GlobalCache::GlobalCache(Handle<JSRegExp> regexp,
- Handle<String> subject, Isolate* isolate)
- : register_array_(nullptr),
- register_array_size_(0),
- regexp_(regexp),
- subject_(subject),
- isolate_(isolate) {
- bool interpreted = FLAG_regexp_interpret_all;
-
- if (regexp_->TypeTag() == JSRegExp::ATOM) {
- static const int kAtomRegistersPerMatch = 2;
- registers_per_match_ = kAtomRegistersPerMatch;
- // There is no distinction between interpreted and native for atom regexps.
- interpreted = false;
- } else {
- registers_per_match_ =
- RegExpImpl::IrregexpPrepare(isolate_, regexp_, subject_);
- if (registers_per_match_ < 0) {
- num_matches_ = -1; // Signal exception.
- return;
- }
- }
-
- DCHECK(IsGlobal(regexp->GetFlags()));
- if (!interpreted) {
- register_array_size_ =
- Max(registers_per_match_, Isolate::kJSRegexpStaticOffsetsVectorSize);
- max_matches_ = register_array_size_ / registers_per_match_;
- } else {
- // Global loop in interpreted regexp is not implemented. We choose
- // the size of the offsets vector so that it can only store one match.
- register_array_size_ = registers_per_match_;
- max_matches_ = 1;
- }
-
- if (register_array_size_ > Isolate::kJSRegexpStaticOffsetsVectorSize) {
- register_array_ = NewArray<int32_t>(register_array_size_);
- } else {
- register_array_ = isolate->jsregexp_static_offsets_vector();
- }
-
- // Set state so that fetching the results the first time triggers a call
- // to the compiled regexp.
- current_match_index_ = max_matches_ - 1;
- num_matches_ = max_matches_;
- DCHECK_LE(2, registers_per_match_); // Each match has at least one capture.
- DCHECK_GE(register_array_size_, registers_per_match_);
- int32_t* last_match =
- &register_array_[current_match_index_ * registers_per_match_];
- last_match[0] = -1;
- last_match[1] = 0;
-}
-
-int RegExpImpl::GlobalCache::AdvanceZeroLength(int last_index) {
- if (IsUnicode(regexp_->GetFlags()) && 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.
-//
-// The Irregexp regular expression engine is intended to be a complete
-// implementation of ECMAScript regular expressions. It generates either
-// bytecodes or native code.
-
-// The Irregexp regexp engine is structured in three steps.
-// 1) The parser generates an abstract syntax tree. See ast.cc.
-// 2) From the AST a node network is created. The nodes are all
-// subclasses of RegExpNode. The nodes represent states when
-// executing a regular expression. Several optimizations are
-// performed on the node network.
-// 3) From the nodes we generate either byte codes or native code
-// that can actually execute the regular expression (perform
-// the search). The code generation step is described in more
-// detail below.
-
-// Code generation.
-//
-// The nodes are divided into four main categories.
-// * Choice nodes
-// These represent places where the regular expression can
-// match in more than one way. For example on entry to an
-// alternation (foo|bar) or a repetition (*, +, ? or {}).
-// * Action nodes
-// These represent places where some action should be
-// performed. Examples include recording the current position
-// in the input string to a register (in order to implement
-// captures) or other actions on register for example in order
-// to implement the counters needed for {} repetitions.
-// * Matching nodes
-// These attempt to match some element part of the input string.
-// Examples of elements include character classes, plain strings
-// or back references.
-// * End nodes
-// These are used to implement the actions required on finding
-// a successful match or failing to find a match.
-//
-// The code generated (whether as byte codes or native code) maintains
-// some state as it runs. This consists of the following elements:
-//
-// * The capture registers. Used for string captures.
-// * Other registers. Used for counters etc.
-// * The current position.
-// * The stack of backtracking information. Used when a matching node
-// fails to find a match and needs to try an alternative.
-//
-// Conceptual regular expression execution model:
-//
-// There is a simple conceptual model of regular expression execution
-// which will be presented first. The actual code generated is a more
-// efficient simulation of the simple conceptual model:
-//
-// * Choice nodes are implemented as follows:
-// For each choice except the last {
-// push current position
-// push backtrack code location
-// <generate code to test for choice>
-// backtrack code location:
-// pop current position
-// }
-// <generate code to test for last choice>
-//
-// * Actions nodes are generated as follows
-// <push affected registers on backtrack stack>
-// <generate code to perform action>
-// push backtrack code location
-// <generate code to test for following nodes>
-// backtrack code location:
-// <pop affected registers to restore their state>
-// <pop backtrack location from stack and go to it>
-//
-// * Matching nodes are generated as follows:
-// if input string matches at current position
-// update current position
-// <generate code to test for following nodes>
-// else
-// <pop backtrack location from stack and go to it>
-//
-// Thus it can be seen that the current position is saved and restored
-// by the choice nodes, whereas the registers are saved and restored by
-// by the action nodes that manipulate them.
-//
-// The other interesting aspect of this model is that nodes are generated
-// at the point where they are needed by a recursive call to Emit(). If
-// the node has already been code generated then the Emit() call will
-// generate a jump to the previously generated code instead. In order to
-// limit recursion it is possible for the Emit() function to put the node
-// on a work list for later generation and instead generate a jump. The
-// destination of the jump is resolved later when the code is generated.
-//
-// Actual regular expression code generation.
-//
-// Code generation is actually more complicated than the above. In order
-// to improve the efficiency of the generated code some optimizations are
-// performed
-//
-// * Choice nodes have 1-character lookahead.
-// A choice node looks at the following character and eliminates some of
-// the choices immediately based on that character. This is not yet
-// implemented.
-// * Simple greedy loops store reduced backtracking information.
-// A quantifier like /.*foo/m will greedily match the whole input. It will
-// then need to backtrack to a point where it can match "foo". The naive
-// implementation of this would push each character position onto the
-// backtracking stack, then pop them off one by one. This would use space
-// proportional to the length of the input string. However since the "."
-// can only match in one way and always has a constant length (in this case
-// of 1) it suffices to store the current position on the top of the stack
-// once. Matching now becomes merely incrementing the current position and
-// backtracking becomes decrementing the current position and checking the
-// result against the stored current position. This is faster and saves
-// space.
-// * The current state is virtualized.
-// This is used to defer expensive operations until it is clear that they
-// are needed and to generate code for a node more than once, allowing
-// specialized an efficient versions of the code to be created. This is
-// explained in the section below.
-//
-// Execution state virtualization.
-//
-// Instead of emitting code, nodes that manipulate the state can record their
-// manipulation in an object called the Trace. The Trace object can record a
-// current position offset, an optional backtrack code location on the top of
-// the virtualized backtrack stack and some register changes. When a node is
-// to be emitted it can flush the Trace or update it. Flushing the Trace
-// will emit code to bring the actual state into line with the virtual state.
-// Avoiding flushing the state can postpone some work (e.g. updates of capture
-// registers). Postponing work can save time when executing the regular
-// expression since it may be found that the work never has to be done as a
-// failure to match can occur. In addition it is much faster to jump to a
-// known backtrack code location than it is to pop an unknown backtrack
-// location from the stack and jump there.
-//
-// The virtual state found in the Trace affects code generation. For example
-// the virtual state contains the difference between the actual current
-// position and the virtual current position, and matching code needs to use
-// this offset to attempt a match in the correct location of the input
-// string. Therefore code generated for a non-trivial trace is specialized
-// to that trace. The code generator therefore has the ability to generate
-// code for each node several times. In order to limit the size of the
-// generated code there is an arbitrary limit on how many specialized sets of
-// code may be generated for a given node. If the limit is reached, the
-// trace is flushed and a generic version of the code for a node is emitted.
-// This is subsequently used for that node. The code emitted for non-generic
-// trace is not recorded in the node and so it cannot currently be reused in
-// the event that code generation is requested for an identical trace.
-
-
-void RegExpTree::AppendToText(RegExpText* text, Zone* zone) {
- UNREACHABLE();
-}
-
-
-void RegExpAtom::AppendToText(RegExpText* text, Zone* zone) {
- text->AddElement(TextElement::Atom(this), zone);
-}
-
-
-void RegExpCharacterClass::AppendToText(RegExpText* text, Zone* zone) {
- text->AddElement(TextElement::CharClass(this), zone);
-}
-
-
-void RegExpText::AppendToText(RegExpText* text, Zone* zone) {
- for (int i = 0; i < elements()->length(); i++)
- text->AddElement(elements()->at(i), zone);
-}
-
-
-TextElement TextElement::Atom(RegExpAtom* atom) {
- return TextElement(ATOM, atom);
-}
-
-
-TextElement TextElement::CharClass(RegExpCharacterClass* char_class) {
- return TextElement(CHAR_CLASS, char_class);
-}
-
-
-int TextElement::length() const {
- switch (text_type()) {
- case ATOM:
- return atom()->length();
-
- case CHAR_CLASS:
- return 1;
- }
- UNREACHABLE();
-}
-
-
-DispatchTable* ChoiceNode::GetTable(bool ignore_case) {
- if (table_ == nullptr) {
- table_ = new(zone()) DispatchTable(zone());
- DispatchTableConstructor cons(table_, ignore_case, zone());
- cons.BuildTable(this);
- }
- return table_;
-}
-
-
-class FrequencyCollator {
- public:
- FrequencyCollator() : total_samples_(0) {
- for (int i = 0; i < RegExpMacroAssembler::kTableSize; i++) {
- frequencies_[i] = CharacterFrequency(i);
- }
- }
-
- void CountCharacter(int character) {
- int index = (character & RegExpMacroAssembler::kTableMask);
- frequencies_[index].Increment();
- total_samples_++;
- }
-
- // Does not measure in percent, but rather per-128 (the table size from the
- // regexp macro assembler).
- int Frequency(int in_character) {
- DCHECK((in_character & RegExpMacroAssembler::kTableMask) == in_character);
- if (total_samples_ < 1) return 1; // Division by zero.
- int freq_in_per128 =
- (frequencies_[in_character].counter() * 128) / total_samples_;
- return freq_in_per128;
- }
-
- private:
- class CharacterFrequency {
- public:
- CharacterFrequency() : counter_(0), character_(-1) { }
- explicit CharacterFrequency(int character)
- : counter_(0), character_(character) { }
-
- void Increment() { counter_++; }
- int counter() { return counter_; }
- int character() { return character_; }
-
- private:
- int counter_;
- int character_;
- };
-
-
- private:
- CharacterFrequency frequencies_[RegExpMacroAssembler::kTableSize];
- int total_samples_;
-};
-
-
-class RegExpCompiler {
- public:
- RegExpCompiler(Isolate* isolate, Zone* zone, int capture_count,
- bool is_one_byte);
-
- int AllocateRegister() {
- if (next_register_ >= RegExpMacroAssembler::kMaxRegister) {
- reg_exp_too_big_ = true;
- return next_register_;
- }
- 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(Isolate* isolate,
- RegExpMacroAssembler* assembler,
- RegExpNode* start, int capture_count,
- Handle<String> pattern);
-
- inline void AddWork(RegExpNode* node) {
- if (!node->on_work_list() && !node->label()->is_bound()) {
- node->set_on_work_list(true);
- work_list_->push_back(node);
- }
- }
-
- static const int kImplementationOffset = 0;
- static const int kNumberOfRegistersOffset = 0;
- static const int kCodeOffset = 1;
-
- RegExpMacroAssembler* macro_assembler() { return macro_assembler_; }
- EndNode* accept() { return accept_; }
-
- static const int kMaxRecursion = 100;
- inline int recursion_depth() { return recursion_depth_; }
- inline void IncrementRecursionDepth() { recursion_depth_++; }
- inline void DecrementRecursionDepth() { recursion_depth_--; }
-
- void SetRegExpTooBig() { reg_exp_too_big_ = true; }
-
- inline bool one_byte() { return one_byte_; }
- inline bool optimize() { return optimize_; }
- inline void set_optimize(bool value) { optimize_ = value; }
- inline bool limiting_recursion() { return limiting_recursion_; }
- inline void set_limiting_recursion(bool value) {
- limiting_recursion_ = value;
- }
- bool read_backward() { return read_backward_; }
- void set_read_backward(bool value) { read_backward_ = value; }
- FrequencyCollator* frequency_collator() { return &frequency_collator_; }
-
- int current_expansion_factor() { return current_expansion_factor_; }
- void set_current_expansion_factor(int value) {
- current_expansion_factor_ = value;
- }
-
- Isolate* isolate() const { return isolate_; }
- Zone* zone() const { return zone_; }
-
- static const int kNoRegister = -1;
-
- private:
- EndNode* accept_;
- int next_register_;
- int unicode_lookaround_stack_register_;
- int unicode_lookaround_position_register_;
- std::vector<RegExpNode*>* work_list_;
- int recursion_depth_;
- RegExpMacroAssembler* macro_assembler_;
- bool one_byte_;
- bool reg_exp_too_big_;
- bool limiting_recursion_;
- bool optimize_;
- bool read_backward_;
- int current_expansion_factor_;
- FrequencyCollator frequency_collator_;
- Isolate* isolate_;
- Zone* zone_;
-};
-
-
-class RecursionCheck {
- public:
- explicit RecursionCheck(RegExpCompiler* compiler) : compiler_(compiler) {
- compiler->IncrementRecursionDepth();
- }
- ~RecursionCheck() { compiler_->DecrementRecursionDepth(); }
- private:
- RegExpCompiler* compiler_;
-};
-
-
-static RegExpEngine::CompilationResult IrregexpRegExpTooBig(Isolate* isolate) {
- return RegExpEngine::CompilationResult(isolate, "RegExp too big");
-}
-
-
-// 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 one_byte)
- : next_register_(2 * (capture_count + 1)),
- unicode_lookaround_stack_register_(kNoRegister),
- unicode_lookaround_position_register_(kNoRegister),
- work_list_(nullptr),
- recursion_depth_(0),
- one_byte_(one_byte),
- reg_exp_too_big_(false),
- limiting_recursion_(false),
- optimize_(FLAG_regexp_optimization),
- read_backward_(false),
- current_expansion_factor_(1),
- frequency_collator_(),
- isolate_(isolate),
- zone_(zone) {
- accept_ = new(zone) EndNode(EndNode::ACCEPT, zone);
- DCHECK_GE(RegExpMacroAssembler::kMaxRegister, next_register_ - 1);
-}
-
-RegExpEngine::CompilationResult RegExpCompiler::Assemble(
- Isolate* isolate, RegExpMacroAssembler* macro_assembler, RegExpNode* start,
- int capture_count, Handle<String> pattern) {
-#ifdef DEBUG
- if (FLAG_trace_regexp_assembler)
- macro_assembler_ = new RegExpMacroAssemblerTracer(isolate, macro_assembler);
- else
-#endif
- macro_assembler_ = macro_assembler;
-
- std::vector<RegExpNode*> work_list;
- work_list_ = &work_list;
- Label fail;
- macro_assembler_->PushBacktrack(&fail);
- Trace new_trace;
- start->Emit(this, &new_trace);
- macro_assembler_->Bind(&fail);
- macro_assembler_->Fail();
- while (!work_list.empty()) {
- RegExpNode* node = work_list.back();
- work_list.pop_back();
- node->set_on_work_list(false);
- if (!node->label()->is_bound()) node->Emit(this, &new_trace);
- }
- if (reg_exp_too_big_) {
- macro_assembler_->AbortedCodeGeneration();
- return IrregexpRegExpTooBig(isolate_);
- }
-
- Handle<HeapObject> code = macro_assembler_->GetCode(pattern);
- isolate->IncreaseTotalRegexpCodeGenerated(code->Size());
- work_list_ = nullptr;
-#ifdef ENABLE_DISASSEMBLER
- if (FLAG_print_code && !FLAG_regexp_interpret_all) {
- CodeTracer::Scope trace_scope(isolate->GetCodeTracer());
- OFStream os(trace_scope.file());
- Handle<Code>::cast(code)->Disassemble(pattern->ToCString().get(), os);
- }
-#endif
-#ifdef DEBUG
- if (FLAG_trace_regexp_assembler) {
- delete macro_assembler_;
- }
-#endif
- return RegExpEngine::CompilationResult(*code, next_register_);
-}
-
-
-bool Trace::DeferredAction::Mentions(int that) {
- if (action_type() == ActionNode::CLEAR_CAPTURES) {
- Interval range = static_cast<DeferredClearCaptures*>(this)->range();
- return range.Contains(that);
- } else {
- return reg() == that;
- }
-}
-
-
-bool Trace::mentions_reg(int reg) {
- for (DeferredAction* action = actions_; action != nullptr;
- action = action->next()) {
- if (action->Mentions(reg))
- return true;
- }
- return false;
-}
-
-
-bool Trace::GetStoredPosition(int reg, int* cp_offset) {
- DCHECK_EQ(0, *cp_offset);
- for (DeferredAction* action = actions_; action != nullptr;
- action = action->next()) {
- if (action->Mentions(reg)) {
- if (action->action_type() == ActionNode::STORE_POSITION) {
- *cp_offset = static_cast<DeferredCapture*>(action)->cp_offset();
- return true;
- } else {
- return false;
- }
- }
- }
- return false;
-}
-
-
-int Trace::FindAffectedRegisters(OutSet* affected_registers,
- Zone* zone) {
- int max_register = RegExpCompiler::kNoRegister;
- for (DeferredAction* action = actions_; action != nullptr;
- action = action->next()) {
- if (action->action_type() == ActionNode::CLEAR_CAPTURES) {
- Interval range = static_cast<DeferredClearCaptures*>(action)->range();
- for (int i = range.from(); i <= range.to(); i++)
- affected_registers->Set(i, zone);
- if (range.to() > max_register) max_register = range.to();
- } else {
- affected_registers->Set(action->reg(), zone);
- if (action->reg() > max_register) max_register = action->reg();
- }
- }
- return max_register;
-}
-
-
-void Trace::RestoreAffectedRegisters(RegExpMacroAssembler* assembler,
- int max_register,
- const OutSet& registers_to_pop,
- const OutSet& registers_to_clear) {
- for (int reg = max_register; reg >= 0; reg--) {
- if (registers_to_pop.Get(reg)) {
- assembler->PopRegister(reg);
- } else if (registers_to_clear.Get(reg)) {
- int clear_to = reg;
- while (reg > 0 && registers_to_clear.Get(reg - 1)) {
- reg--;
- }
- assembler->ClearRegisters(reg, clear_to);
- }
- }
-}
-
-
-void Trace::PerformDeferredActions(RegExpMacroAssembler* assembler,
- int max_register,
- const OutSet& affected_registers,
- OutSet* registers_to_pop,
- OutSet* registers_to_clear,
- Zone* zone) {
- // The "+1" is to avoid a push_limit of zero if stack_limit_slack() is 1.
- const int push_limit = (assembler->stack_limit_slack() + 1) / 2;
-
- // Count pushes performed to force a stack limit check occasionally.
- int pushes = 0;
-
- for (int reg = 0; reg <= max_register; reg++) {
- if (!affected_registers.Get(reg)) {
- continue;
- }
-
- // The chronologically first deferred action in the trace
- // is used to infer the action needed to restore a register
- // to its previous state (or not, if it's safe to ignore it).
- enum DeferredActionUndoType { IGNORE, RESTORE, CLEAR };
- DeferredActionUndoType undo_action = IGNORE;
-
- int value = 0;
- bool absolute = false;
- bool clear = false;
- static const int kNoStore = kMinInt;
- int store_position = kNoStore;
- // This is a little tricky because we are scanning the actions in reverse
- // historical order (newest first).
- for (DeferredAction* action = actions_; action != nullptr;
- action = action->next()) {
- if (action->Mentions(reg)) {
- switch (action->action_type()) {
- case ActionNode::SET_REGISTER: {
- Trace::DeferredSetRegister* psr =
- static_cast<Trace::DeferredSetRegister*>(action);
- if (!absolute) {
- value += psr->value();
- absolute = true;
- }
- // SET_REGISTER is currently only used for newly introduced loop
- // counters. They can have a significant previous value if they
- // occur in a loop. TODO(lrn): Propagate this information, so
- // we can set undo_action to IGNORE if we know there is no value to
- // restore.
- undo_action = RESTORE;
- DCHECK_EQ(store_position, kNoStore);
- DCHECK(!clear);
- break;
- }
- case ActionNode::INCREMENT_REGISTER:
- if (!absolute) {
- value++;
- }
- DCHECK_EQ(store_position, kNoStore);
- DCHECK(!clear);
- undo_action = RESTORE;
- break;
- case ActionNode::STORE_POSITION: {
- Trace::DeferredCapture* pc =
- static_cast<Trace::DeferredCapture*>(action);
- if (!clear && store_position == kNoStore) {
- store_position = pc->cp_offset();
- }
-
- // For captures we know that stores and clears alternate.
- // Other register, are never cleared, and if the occur
- // inside a loop, they might be assigned more than once.
- if (reg <= 1) {
- // Registers zero and one, aka "capture zero", is
- // always set correctly if we succeed. There is no
- // need to undo a setting on backtrack, because we
- // will set it again or fail.
- undo_action = IGNORE;
- } else {
- undo_action = pc->is_capture() ? CLEAR : RESTORE;
- }
- DCHECK(!absolute);
- DCHECK_EQ(value, 0);
- break;
- }
- case ActionNode::CLEAR_CAPTURES: {
- // Since we're scanning in reverse order, if we've already
- // set the position we have to ignore historically earlier
- // clearing operations.
- if (store_position == kNoStore) {
- clear = true;
- }
- undo_action = RESTORE;
- DCHECK(!absolute);
- DCHECK_EQ(value, 0);
- break;
- }
- default:
- UNREACHABLE();
- break;
- }
- }
- }
- // Prepare for the undo-action (e.g., push if it's going to be popped).
- if (undo_action == RESTORE) {
- pushes++;
- RegExpMacroAssembler::StackCheckFlag stack_check =
- RegExpMacroAssembler::kNoStackLimitCheck;
- if (pushes == push_limit) {
- stack_check = RegExpMacroAssembler::kCheckStackLimit;
- pushes = 0;
- }
-
- assembler->PushRegister(reg, stack_check);
- registers_to_pop->Set(reg, zone);
- } else if (undo_action == CLEAR) {
- registers_to_clear->Set(reg, zone);
- }
- // Perform the chronologically last action (or accumulated increment)
- // for the register.
- if (store_position != kNoStore) {
- assembler->WriteCurrentPositionToRegister(reg, store_position);
- } else if (clear) {
- assembler->ClearRegisters(reg, reg);
- } else if (absolute) {
- assembler->SetRegister(reg, value);
- } else if (value != 0) {
- assembler->AdvanceRegister(reg, value);
- }
- }
-}
-
-
-// This is called as we come into a loop choice node and some other tricky
-// nodes. It normalizes the state of the code generator to ensure we can
-// generate generic code.
-void Trace::Flush(RegExpCompiler* compiler, RegExpNode* successor) {
- RegExpMacroAssembler* assembler = compiler->macro_assembler();
-
- DCHECK(!is_trivial());
-
- if (actions_ == nullptr && backtrack() == nullptr) {
- // Here we just have some deferred cp advances to fix and we are back to
- // a normal situation. We may also have to forget some information gained
- // through a quick check that was already performed.
- if (cp_offset_ != 0) assembler->AdvanceCurrentPosition(cp_offset_);
- // Create a new trivial state and generate the node with that.
- Trace new_state;
- successor->Emit(compiler, &new_state);
- return;
- }
-
- // Generate deferred actions here along with code to undo them again.
- OutSet affected_registers;
-
- if (backtrack() != nullptr) {
- // Here we have a concrete backtrack location. These are set up by choice
- // nodes and so they indicate that we have a deferred save of the current
- // position which we may need to emit here.
- assembler->PushCurrentPosition();
- }
-
- int max_register = FindAffectedRegisters(&affected_registers,
- compiler->zone());
- OutSet registers_to_pop;
- OutSet registers_to_clear;
- PerformDeferredActions(assembler,
- max_register,
- affected_registers,
- &registers_to_pop,
- &registers_to_clear,
- compiler->zone());
- if (cp_offset_ != 0) {
- assembler->AdvanceCurrentPosition(cp_offset_);
- }
-
- // Create a new trivial state and generate the node with that.
- Label undo;
- assembler->PushBacktrack(&undo);
- if (successor->KeepRecursing(compiler)) {
- Trace new_state;
- successor->Emit(compiler, &new_state);
- } else {
- compiler->AddWork(successor);
- assembler->GoTo(successor->label());
- }
-
- // On backtrack we need to restore state.
- assembler->Bind(&undo);
- RestoreAffectedRegisters(assembler,
- max_register,
- registers_to_pop,
- registers_to_clear);
- if (backtrack() == nullptr) {
- assembler->Backtrack();
- } else {
- assembler->PopCurrentPosition();
- assembler->GoTo(backtrack());
- }
-}
-
-
-void NegativeSubmatchSuccess::Emit(RegExpCompiler* compiler, Trace* trace) {
- RegExpMacroAssembler* assembler = compiler->macro_assembler();
-
- // Omit flushing the trace. We discard the entire stack frame anyway.
-
- if (!label()->is_bound()) {
- // We are completely independent of the trace, since we ignore it,
- // so this code can be used as the generic version.
- assembler->Bind(label());
- }
-
- // Throw away everything on the backtrack stack since the start
- // of the negative submatch and restore the character position.
- assembler->ReadCurrentPositionFromRegister(current_position_register_);
- assembler->ReadStackPointerFromRegister(stack_pointer_register_);
- if (clear_capture_count_ > 0) {
- // Clear any captures that might have been performed during the success
- // of the body of the negative look-ahead.
- int clear_capture_end = clear_capture_start_ + clear_capture_count_ - 1;
- assembler->ClearRegisters(clear_capture_start_, clear_capture_end);
- }
- // Now that we have unwound the stack we find at the top of the stack the
- // backtrack that the BeginSubmatch node got.
- assembler->Backtrack();
-}
-
-
-void EndNode::Emit(RegExpCompiler* compiler, Trace* trace) {
- if (!trace->is_trivial()) {
- trace->Flush(compiler, this);
- return;
- }
- RegExpMacroAssembler* assembler = compiler->macro_assembler();
- if (!label()->is_bound()) {
- assembler->Bind(label());
- }
- switch (action_) {
- case ACCEPT:
- assembler->Succeed();
- return;
- case BACKTRACK:
- assembler->GoTo(trace->backtrack());
- return;
- case NEGATIVE_SUBMATCH_SUCCESS:
- // This case is handled in a different virtual method.
- UNREACHABLE();
- }
- UNIMPLEMENTED();
-}
-
-
-void GuardedAlternative::AddGuard(Guard* guard, Zone* zone) {
- if (guards_ == nullptr) guards_ = new (zone) ZoneList<Guard*>(1, zone);
- guards_->Add(guard, zone);
-}
-
-
-ActionNode* ActionNode::SetRegister(int reg,
- int val,
- RegExpNode* on_success) {
- ActionNode* result =
- new(on_success->zone()) ActionNode(SET_REGISTER, on_success);
- result->data_.u_store_register.reg = reg;
- result->data_.u_store_register.value = val;
- return result;
-}
-
-
-ActionNode* ActionNode::IncrementRegister(int reg, RegExpNode* on_success) {
- ActionNode* result =
- new(on_success->zone()) ActionNode(INCREMENT_REGISTER, on_success);
- result->data_.u_increment_register.reg = reg;
- return result;
-}
-
-
-ActionNode* ActionNode::StorePosition(int reg,
- bool is_capture,
- RegExpNode* on_success) {
- ActionNode* result =
- new(on_success->zone()) ActionNode(STORE_POSITION, on_success);
- result->data_.u_position_register.reg = reg;
- result->data_.u_position_register.is_capture = is_capture;
- return result;
-}
-
-
-ActionNode* ActionNode::ClearCaptures(Interval range,
- RegExpNode* on_success) {
- ActionNode* result =
- new(on_success->zone()) ActionNode(CLEAR_CAPTURES, on_success);
- result->data_.u_clear_captures.range_from = range.from();
- result->data_.u_clear_captures.range_to = range.to();
- return result;
-}
-
-
-ActionNode* ActionNode::BeginSubmatch(int stack_reg,
- int position_reg,
- RegExpNode* on_success) {
- ActionNode* result =
- new(on_success->zone()) ActionNode(BEGIN_SUBMATCH, on_success);
- result->data_.u_submatch.stack_pointer_register = stack_reg;
- result->data_.u_submatch.current_position_register = position_reg;
- return result;
-}
-
-
-ActionNode* ActionNode::PositiveSubmatchSuccess(int stack_reg,
- int position_reg,
- int clear_register_count,
- int clear_register_from,
- RegExpNode* on_success) {
- ActionNode* result =
- new(on_success->zone()) ActionNode(POSITIVE_SUBMATCH_SUCCESS, on_success);
- result->data_.u_submatch.stack_pointer_register = stack_reg;
- result->data_.u_submatch.current_position_register = position_reg;
- result->data_.u_submatch.clear_register_count = clear_register_count;
- result->data_.u_submatch.clear_register_from = clear_register_from;
- return result;
-}
-
-
-ActionNode* ActionNode::EmptyMatchCheck(int start_register,
- int repetition_register,
- int repetition_limit,
- RegExpNode* on_success) {
- ActionNode* result =
- new(on_success->zone()) ActionNode(EMPTY_MATCH_CHECK, on_success);
- result->data_.u_empty_match_check.start_register = start_register;
- result->data_.u_empty_match_check.repetition_register = repetition_register;
- result->data_.u_empty_match_check.repetition_limit = repetition_limit;
- return result;
-}
-
-
-#define DEFINE_ACCEPT(Type) \
- void Type##Node::Accept(NodeVisitor* visitor) { \
- visitor->Visit##Type(this); \
- }
-FOR_EACH_NODE_TYPE(DEFINE_ACCEPT)
-#undef DEFINE_ACCEPT
-
-
-void LoopChoiceNode::Accept(NodeVisitor* visitor) {
- visitor->VisitLoopChoice(this);
-}
-
-
-// -------------------------------------------------------------------
-// Emit code.
-
-
-void ChoiceNode::GenerateGuard(RegExpMacroAssembler* macro_assembler,
- Guard* guard,
- Trace* trace) {
- switch (guard->op()) {
- case Guard::LT:
- DCHECK(!trace->mentions_reg(guard->reg()));
- macro_assembler->IfRegisterGE(guard->reg(),
- guard->value(),
- trace->backtrack());
- break;
- case Guard::GEQ:
- DCHECK(!trace->mentions_reg(guard->reg()));
- macro_assembler->IfRegisterLT(guard->reg(),
- guard->value(),
- trace->backtrack());
- break;
- }
-}
-
-
-// Returns the number of characters in the equivalence class, omitting those
-// that cannot occur in the source string because it is Latin1.
-static int GetCaseIndependentLetters(Isolate* isolate, uc16 character,
- bool one_byte_subject,
- unibrow::uchar* letters,
- int letter_length) {
-#ifdef V8_INTL_SUPPORT
- icu::UnicodeSet set;
- set.add(character);
- set = set.closeOver(USET_CASE_INSENSITIVE);
- int32_t range_count = set.getRangeCount();
- int items = 0;
- for (int32_t i = 0; i < range_count; i++) {
- UChar32 start = set.getRangeStart(i);
- UChar32 end = set.getRangeEnd(i);
- CHECK(end - start + items <= letter_length);
- while (start <= end) {
- if (one_byte_subject && start > String::kMaxOneByteCharCode) break;
- letters[items++] = (unibrow::uchar)(start);
- start++;
- }
- }
- return items;
-#else
- int length =
- isolate->jsregexp_uncanonicalize()->get(character, '\0', letters);
- // Unibrow returns 0 or 1 for characters where case independence is
- // trivial.
- if (length == 0) {
- letters[0] = character;
- length = 1;
- }
-
- if (one_byte_subject) {
- int new_length = 0;
- for (int i = 0; i < length; i++) {
- if (letters[i] <= String::kMaxOneByteCharCode) {
- letters[new_length++] = letters[i];
- }
- }
- length = new_length;
- }
-
- return length;
-#endif // V8_INTL_SUPPORT
-}
-
-static inline bool EmitSimpleCharacter(Isolate* isolate,
- RegExpCompiler* compiler,
- uc16 c,
- Label* on_failure,
- int cp_offset,
- bool check,
- bool preloaded) {
- RegExpMacroAssembler* assembler = compiler->macro_assembler();
- bool bound_checked = false;
- if (!preloaded) {
- assembler->LoadCurrentCharacter(
- cp_offset,
- on_failure,
- check);
- bound_checked = true;
- }
- assembler->CheckNotCharacter(c, on_failure);
- return bound_checked;
-}
-
-
-// Only emits non-letters (things that don't have case). Only used for case
-// independent matches.
-static inline bool EmitAtomNonLetter(Isolate* isolate,
- RegExpCompiler* compiler,
- uc16 c,
- Label* on_failure,
- int cp_offset,
- bool check,
- bool preloaded) {
- RegExpMacroAssembler* macro_assembler = compiler->macro_assembler();
- bool one_byte = compiler->one_byte();
- unibrow::uchar chars[4];
- int length = GetCaseIndependentLetters(isolate, c, one_byte, chars, 4);
- if (length < 1) {
- // This can't match. Must be an one-byte subject and a non-one-byte
- // character. We do not need to do anything since the one-byte pass
- // already handled this.
- return false; // Bounds not checked.
- }
- bool checked = false;
- // We handle the length > 1 case in a later pass.
- if (length == 1) {
- if (one_byte && c > String::kMaxOneByteCharCodeU) {
- // Can't match - see above.
- return false; // Bounds not checked.
- }
- if (!preloaded) {
- macro_assembler->LoadCurrentCharacter(cp_offset, on_failure, check);
- checked = check;
- }
- macro_assembler->CheckNotCharacter(c, on_failure);
- }
- return checked;
-}
-
-
-static bool ShortCutEmitCharacterPair(RegExpMacroAssembler* macro_assembler,
- bool one_byte, uc16 c1, uc16 c2,
- Label* on_failure) {
- uc16 char_mask;
- if (one_byte) {
- char_mask = String::kMaxOneByteCharCode;
- } else {
- char_mask = String::kMaxUtf16CodeUnit;
- }
- uc16 exor = c1 ^ c2;
- // Check whether exor has only one bit set.
- if (((exor - 1) & exor) == 0) {
- // If c1 and c2 differ only by one bit.
- // Ecma262UnCanonicalize always gives the highest number last.
- DCHECK(c2 > c1);
- uc16 mask = char_mask ^ exor;
- macro_assembler->CheckNotCharacterAfterAnd(c1, mask, on_failure);
- return true;
- }
- DCHECK(c2 > c1);
- uc16 diff = c2 - c1;
- if (((diff - 1) & diff) == 0 && c1 >= diff) {
- // If the characters differ by 2^n but don't differ by one bit then
- // subtract the difference from the found character, then do the or
- // trick. We avoid the theoretical case where negative numbers are
- // involved in order to simplify code generation.
- uc16 mask = char_mask ^ diff;
- macro_assembler->CheckNotCharacterAfterMinusAnd(c1 - diff,
- diff,
- mask,
- on_failure);
- return true;
- }
- return false;
-}
-
-using EmitCharacterFunction = bool(Isolate* isolate, RegExpCompiler* compiler,
- uc16 c, Label* on_failure, int cp_offset,
- bool check, bool preloaded);
-
-// Only emits letters (things that have case). Only used for case independent
-// matches.
-static inline bool EmitAtomLetter(Isolate* isolate,
- RegExpCompiler* compiler,
- uc16 c,
- Label* on_failure,
- int cp_offset,
- bool check,
- bool preloaded) {
- RegExpMacroAssembler* macro_assembler = compiler->macro_assembler();
- bool one_byte = compiler->one_byte();
- unibrow::uchar chars[4];
- int length = GetCaseIndependentLetters(isolate, c, one_byte, chars, 4);
- if (length <= 1) return false;
- // We may not need to check against the end of the input string
- // if this character lies before a character that matched.
- if (!preloaded) {
- macro_assembler->LoadCurrentCharacter(cp_offset, on_failure, check);
- }
- Label ok;
- switch (length) {
- case 2: {
- if (ShortCutEmitCharacterPair(macro_assembler, one_byte, chars[0],
- chars[1], on_failure)) {
- } else {
- macro_assembler->CheckCharacter(chars[0], &ok);
- macro_assembler->CheckNotCharacter(chars[1], on_failure);
- macro_assembler->Bind(&ok);
- }
- break;
- }
- case 4:
- macro_assembler->CheckCharacter(chars[3], &ok);
- V8_FALLTHROUGH;
- case 3:
- macro_assembler->CheckCharacter(chars[0], &ok);
- macro_assembler->CheckCharacter(chars[1], &ok);
- macro_assembler->CheckNotCharacter(chars[2], on_failure);
- macro_assembler->Bind(&ok);
- break;
- default:
- UNREACHABLE();
- }
- return true;
-}
-
-
-static void EmitBoundaryTest(RegExpMacroAssembler* masm,
- int border,
- Label* fall_through,
- Label* above_or_equal,
- Label* below) {
- if (below != fall_through) {
- masm->CheckCharacterLT(border, below);
- if (above_or_equal != fall_through) masm->GoTo(above_or_equal);
- } else {
- masm->CheckCharacterGT(border - 1, above_or_equal);
- }
-}
-
-
-static void EmitDoubleBoundaryTest(RegExpMacroAssembler* masm,
- int first,
- int last,
- Label* fall_through,
- Label* in_range,
- Label* out_of_range) {
- if (in_range == fall_through) {
- if (first == last) {
- masm->CheckNotCharacter(first, out_of_range);
- } else {
- masm->CheckCharacterNotInRange(first, last, out_of_range);
- }
- } else {
- if (first == last) {
- masm->CheckCharacter(first, in_range);
- } else {
- masm->CheckCharacterInRange(first, last, in_range);
- }
- if (out_of_range != fall_through) masm->GoTo(out_of_range);
- }
-}
-
-
-// even_label is for ranges[i] to ranges[i + 1] where i - start_index is even.
-// odd_label is for ranges[i] to ranges[i + 1] where i - start_index is odd.
-static void EmitUseLookupTable(
- RegExpMacroAssembler* masm,
- ZoneList<int>* ranges,
- int start_index,
- int end_index,
- int min_char,
- Label* fall_through,
- Label* even_label,
- Label* odd_label) {
- static const int kSize = RegExpMacroAssembler::kTableSize;
- static const int kMask = RegExpMacroAssembler::kTableMask;
-
- int base = (min_char & ~kMask);
- USE(base);
-
- // Assert that everything is on one kTableSize page.
- for (int i = start_index; i <= end_index; i++) {
- DCHECK_EQ(ranges->at(i) & ~kMask, base);
- }
- DCHECK(start_index == 0 || (ranges->at(start_index - 1) & ~kMask) <= base);
-
- char templ[kSize];
- Label* on_bit_set;
- Label* on_bit_clear;
- int bit;
- if (even_label == fall_through) {
- on_bit_set = odd_label;
- on_bit_clear = even_label;
- bit = 1;
- } else {
- on_bit_set = even_label;
- on_bit_clear = odd_label;
- bit = 0;
- }
- for (int i = 0; i < (ranges->at(start_index) & kMask) && i < kSize; i++) {
- templ[i] = bit;
- }
- int j = 0;
- bit ^= 1;
- for (int i = start_index; i < end_index; i++) {
- for (j = (ranges->at(i) & kMask); j < (ranges->at(i + 1) & kMask); j++) {
- templ[j] = bit;
- }
- bit ^= 1;
- }
- for (int i = j; i < kSize; i++) {
- templ[i] = bit;
- }
- Factory* factory = masm->isolate()->factory();
- // TODO(erikcorry): Cache these.
- Handle<ByteArray> ba = factory->NewByteArray(kSize, AllocationType::kOld);
- for (int i = 0; i < kSize; i++) {
- ba->set(i, templ[i]);
- }
- masm->CheckBitInTable(ba, on_bit_set);
- if (on_bit_clear != fall_through) masm->GoTo(on_bit_clear);
-}
-
-
-static void CutOutRange(RegExpMacroAssembler* masm,
- ZoneList<int>* ranges,
- int start_index,
- int end_index,
- int cut_index,
- Label* even_label,
- Label* odd_label) {
- bool odd = (((cut_index - start_index) & 1) == 1);
- Label* in_range_label = odd ? odd_label : even_label;
- Label dummy;
- EmitDoubleBoundaryTest(masm,
- ranges->at(cut_index),
- ranges->at(cut_index + 1) - 1,
- &dummy,
- in_range_label,
- &dummy);
- DCHECK(!dummy.is_linked());
- // Cut out the single range by rewriting the array. This creates a new
- // range that is a merger of the two ranges on either side of the one we
- // are cutting out. The oddity of the labels is preserved.
- for (int j = cut_index; j > start_index; j--) {
- ranges->at(j) = ranges->at(j - 1);
- }
- for (int j = cut_index + 1; j < end_index; j++) {
- ranges->at(j) = ranges->at(j + 1);
- }
-}
-
-
-// Unicode case. Split the search space into kSize spaces that are handled
-// with recursion.
-static void SplitSearchSpace(ZoneList<int>* ranges,
- int start_index,
- int end_index,
- int* new_start_index,
- int* new_end_index,
- int* border) {
- static const int kSize = RegExpMacroAssembler::kTableSize;
- static const int kMask = RegExpMacroAssembler::kTableMask;
-
- int first = ranges->at(start_index);
- int last = ranges->at(end_index) - 1;
-
- *new_start_index = start_index;
- *border = (ranges->at(start_index) & ~kMask) + kSize;
- while (*new_start_index < end_index) {
- if (ranges->at(*new_start_index) > *border) break;
- (*new_start_index)++;
- }
- // new_start_index is the index of the first edge that is beyond the
- // current kSize space.
-
- // For very large search spaces we do a binary chop search of the non-Latin1
- // space instead of just going to the end of the current kSize space. The
- // heuristics are complicated a little by the fact that any 128-character
- // encoding space can be quickly tested with a table lookup, so we don't
- // wish to do binary chop search at a smaller granularity than that. A
- // 128-character space can take up a lot of space in the ranges array if,
- // for example, we only want to match every second character (eg. the lower
- // case characters on some Unicode pages).
- int binary_chop_index = (end_index + start_index) / 2;
- // The first test ensures that we get to the code that handles the Latin1
- // range with a single not-taken branch, speeding up this important
- // character range (even non-Latin1 charset-based text has spaces and
- // punctuation).
- if (*border - 1 > String::kMaxOneByteCharCode && // Latin1 case.
- end_index - start_index > (*new_start_index - start_index) * 2 &&
- last - first > kSize * 2 && binary_chop_index > *new_start_index &&
- ranges->at(binary_chop_index) >= first + 2 * kSize) {
- int scan_forward_for_section_border = binary_chop_index;;
- int new_border = (ranges->at(binary_chop_index) | kMask) + 1;
-
- while (scan_forward_for_section_border < end_index) {
- if (ranges->at(scan_forward_for_section_border) > new_border) {
- *new_start_index = scan_forward_for_section_border;
- *border = new_border;
- break;
- }
- scan_forward_for_section_border++;
- }
- }
-
- DCHECK(*new_start_index > start_index);
- *new_end_index = *new_start_index - 1;
- if (ranges->at(*new_end_index) == *border) {
- (*new_end_index)--;
- }
- if (*border >= ranges->at(end_index)) {
- *border = ranges->at(end_index);
- *new_start_index = end_index; // Won't be used.
- *new_end_index = end_index - 1;
- }
-}
-
-// Gets a series of segment boundaries representing a character class. If the
-// character is in the range between an even and an odd boundary (counting from
-// start_index) then go to even_label, otherwise go to odd_label. We already
-// know that the character is in the range of min_char to max_char inclusive.
-// Either label can be nullptr 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, 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;
-
- DCHECK_LT(min_char, first);
-
- // Just need to test if the character is before or on-or-after
- // a particular character.
- if (start_index == end_index) {
- EmitBoundaryTest(masm, first, fall_through, even_label, odd_label);
- return;
- }
-
- // Another almost trivial case: There is one interval in the middle that is
- // different from the end intervals.
- if (start_index + 1 == end_index) {
- EmitDoubleBoundaryTest(
- masm, first, last, fall_through, even_label, odd_label);
- return;
- }
-
- // It's not worth using table lookup if there are very few intervals in the
- // character class.
- if (end_index - start_index <= 6) {
- // It is faster to test for individual characters, so we look for those
- // first, then try arbitrary ranges in the second round.
- static int kNoCutIndex = -1;
- int cut = kNoCutIndex;
- for (int i = start_index; i < end_index; i++) {
- if (ranges->at(i) == ranges->at(i + 1) - 1) {
- cut = i;
- break;
- }
- }
- if (cut == kNoCutIndex) cut = start_index;
- CutOutRange(
- masm, ranges, start_index, end_index, cut, even_label, odd_label);
- DCHECK_GE(end_index - start_index, 2);
- GenerateBranches(masm,
- ranges,
- start_index + 1,
- end_index - 1,
- min_char,
- max_char,
- fall_through,
- even_label,
- odd_label);
- return;
- }
-
- // If there are a lot of intervals in the regexp, then we will use tables to
- // determine whether the character is inside or outside the character class.
- static const int kBits = RegExpMacroAssembler::kTableSizeBits;
-
- if ((max_char >> kBits) == (min_char >> kBits)) {
- EmitUseLookupTable(masm,
- ranges,
- start_index,
- end_index,
- min_char,
- fall_through,
- even_label,
- odd_label);
- return;
- }
-
- if ((min_char >> kBits) != (first >> kBits)) {
- masm->CheckCharacterLT(first, odd_label);
- GenerateBranches(masm,
- ranges,
- start_index + 1,
- end_index,
- first,
- max_char,
- fall_through,
- odd_label,
- even_label);
- return;
- }
-
- int new_start_index = 0;
- int new_end_index = 0;
- int border = 0;
-
- SplitSearchSpace(ranges,
- start_index,
- end_index,
- &new_start_index,
- &new_end_index,
- &border);
-
- Label handle_rest;
- Label* above = &handle_rest;
- if (border == last + 1) {
- // We didn't find any section that started after the limit, so everything
- // above the border is one of the terminal labels.
- above = (end_index & 1) != (start_index & 1) ? odd_label : even_label;
- DCHECK(new_end_index == end_index - 1);
- }
-
- DCHECK_LE(start_index, new_end_index);
- DCHECK_LE(new_start_index, end_index);
- DCHECK_LT(start_index, new_start_index);
- DCHECK_LT(new_end_index, end_index);
- DCHECK(new_end_index + 1 == new_start_index ||
- (new_end_index + 2 == new_start_index &&
- border == ranges->at(new_end_index + 1)));
- DCHECK_LT(min_char, border - 1);
- DCHECK_LT(border, max_char);
- DCHECK_LT(ranges->at(new_end_index), border);
- DCHECK(border < ranges->at(new_start_index) ||
- (border == ranges->at(new_start_index) &&
- new_start_index == end_index &&
- new_end_index == end_index - 1 &&
- border == last + 1));
- DCHECK(new_start_index == 0 || border >= ranges->at(new_start_index - 1));
-
- masm->CheckCharacterGT(border - 1, above);
- Label dummy;
- GenerateBranches(masm,
- ranges,
- start_index,
- new_end_index,
- min_char,
- border - 1,
- &dummy,
- even_label,
- odd_label);
- if (handle_rest.is_linked()) {
- masm->Bind(&handle_rest);
- bool flip = (new_start_index & 1) != (start_index & 1);
- GenerateBranches(masm,
- ranges,
- new_start_index,
- end_index,
- border,
- max_char,
- &dummy,
- flip ? odd_label : even_label,
- flip ? even_label : odd_label);
- }
-}
-
-
-static void EmitCharClass(RegExpMacroAssembler* macro_assembler,
- RegExpCharacterClass* cc, bool one_byte,
- Label* on_failure, int cp_offset, bool check_offset,
- bool preloaded, Zone* zone) {
- ZoneList<CharacterRange>* ranges = cc->ranges(zone);
- CharacterRange::Canonicalize(ranges);
-
- int max_char;
- if (one_byte) {
- max_char = String::kMaxOneByteCharCode;
- } else {
- max_char = String::kMaxUtf16CodeUnit;
- }
-
- int range_count = ranges->length();
-
- int last_valid_range = range_count - 1;
- while (last_valid_range >= 0) {
- CharacterRange& range = ranges->at(last_valid_range);
- if (range.from() <= max_char) {
- break;
- }
- last_valid_range--;
- }
-
- if (last_valid_range < 0) {
- if (!cc->is_negated()) {
- macro_assembler->GoTo(on_failure);
- }
- if (check_offset) {
- macro_assembler->CheckPosition(cp_offset, on_failure);
- }
- return;
- }
-
- if (last_valid_range == 0 &&
- ranges->at(0).IsEverything(max_char)) {
- if (cc->is_negated()) {
- macro_assembler->GoTo(on_failure);
- } else {
- // 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)) {
- return;
- }
-
-
- // A new list with ascending entries. Each entry is a code unit
- // where there is a boundary between code units that are part of
- // the class and code units that are not. Normally we insert an
- // entry at zero which goes to the failure label, but if there
- // was already one there we fall through for success on that entry.
- // Subsequent entries have alternating meaning (success/failure).
- ZoneList<int>* range_boundaries =
- new(zone) ZoneList<int>(last_valid_range, zone);
-
- bool zeroth_entry_is_failure = !cc->is_negated();
-
- for (int i = 0; i <= last_valid_range; i++) {
- CharacterRange& range = ranges->at(i);
- if (range.from() == 0) {
- DCHECK_EQ(i, 0);
- zeroth_entry_is_failure = !zeroth_entry_is_failure;
- } else {
- range_boundaries->Add(range.from(), zone);
- }
- range_boundaries->Add(range.to() + 1, zone);
- }
- int end_index = range_boundaries->length() - 1;
- if (range_boundaries->at(end_index) > max_char) {
- end_index--;
- }
-
- Label fall_through;
- GenerateBranches(macro_assembler,
- range_boundaries,
- 0, // start_index.
- end_index,
- 0, // min_char.
- max_char,
- &fall_through,
- zeroth_entry_is_failure ? &fall_through : on_failure,
- zeroth_entry_is_failure ? on_failure : &fall_through);
- macro_assembler->Bind(&fall_through);
-}
-
-RegExpNode::~RegExpNode() = default;
-
-RegExpNode::LimitResult RegExpNode::LimitVersions(RegExpCompiler* compiler,
- Trace* trace) {
- // If we are generating a greedy loop then don't stop and don't reuse code.
- if (trace->stop_node() != nullptr) {
- return CONTINUE;
- }
-
- RegExpMacroAssembler* macro_assembler = compiler->macro_assembler();
- if (trace->is_trivial()) {
- if (label_.is_bound() || on_work_list() || !KeepRecursing(compiler)) {
- // If a generic version is already scheduled to be generated or we have
- // recursed too deeply then just generate a jump to that code.
- macro_assembler->GoTo(&label_);
- // This will queue it up for generation of a generic version if it hasn't
- // already been queued.
- compiler->AddWork(this);
- return DONE;
- }
- // Generate generic version of the node and bind the label for later use.
- macro_assembler->Bind(&label_);
- return CONTINUE;
- }
-
- // We are being asked to make a non-generic version. Keep track of how many
- // non-generic versions we generate so as not to overdo it.
- trace_count_++;
- if (KeepRecursing(compiler) && compiler->optimize() &&
- trace_count_ < kMaxCopiesCodeGenerated) {
- return CONTINUE;
- }
-
- // If we get here code has been generated for this node too many times or
- // recursion is too deep. Time to switch to a generic version. The code for
- // generic versions above can handle deep recursion properly.
- bool was_limiting = compiler->limiting_recursion();
- compiler->set_limiting_recursion(true);
- trace->Flush(compiler, this);
- compiler->set_limiting_recursion(was_limiting);
- return DONE;
-}
-
-
-bool RegExpNode::KeepRecursing(RegExpCompiler* compiler) {
- return !compiler->limiting_recursion() &&
- compiler->recursion_depth() <= RegExpCompiler::kMaxRecursion;
-}
-
-
-int ActionNode::EatsAtLeast(int still_to_find,
- int budget,
- bool not_at_start) {
- if (budget <= 0) return 0;
- if (action_type_ == POSITIVE_SUBMATCH_SUCCESS) return 0; // Rewinds input!
- return on_success()->EatsAtLeast(still_to_find,
- budget - 1,
- not_at_start);
-}
-
-
-void ActionNode::FillInBMInfo(Isolate* isolate, int offset, int budget,
- BoyerMooreLookahead* bm, bool not_at_start) {
- if (action_type_ != POSITIVE_SUBMATCH_SUCCESS) {
- on_success()->FillInBMInfo(isolate, offset, budget - 1, bm, not_at_start);
- }
- SaveBMInfo(bm, not_at_start, offset);
-}
-
-
-int AssertionNode::EatsAtLeast(int still_to_find,
- int budget,
- bool not_at_start) {
- if (budget <= 0) return 0;
- // If we know we are not at the start and we are asked "how many characters
- // will you match if you succeed?" then we can answer anything since false
- // implies false. So lets just return the max answer (still_to_find) since
- // that won't prevent us from preloading a lot of characters for the other
- // branches in the node graph.
- if (assertion_type() == AT_START && not_at_start) return still_to_find;
- return on_success()->EatsAtLeast(still_to_find,
- budget - 1,
- not_at_start);
-}
-
-
-void AssertionNode::FillInBMInfo(Isolate* isolate, int offset, int budget,
- BoyerMooreLookahead* bm, bool not_at_start) {
- // Match the behaviour of EatsAtLeast on this node.
- if (assertion_type() == AT_START && not_at_start) return;
- on_success()->FillInBMInfo(isolate, offset, budget - 1, bm, not_at_start);
- SaveBMInfo(bm, not_at_start, offset);
-}
-
-
-int BackReferenceNode::EatsAtLeast(int still_to_find,
- int budget,
- bool not_at_start) {
- if (read_backward()) return 0;
- if (budget <= 0) return 0;
- return on_success()->EatsAtLeast(still_to_find,
- budget - 1,
- not_at_start);
-}
-
-
-int TextNode::EatsAtLeast(int still_to_find,
- int budget,
- bool not_at_start) {
- if (read_backward()) return 0;
- int answer = Length();
- if (answer >= still_to_find) return answer;
- if (budget <= 0) return answer;
- // We are not at start after this node so we set the last argument to 'true'.
- return answer + on_success()->EatsAtLeast(still_to_find - answer,
- budget - 1,
- true);
-}
-
-
-int NegativeLookaroundChoiceNode::EatsAtLeast(int still_to_find, int budget,
- bool not_at_start) {
- if (budget <= 0) return 0;
- // Alternative 0 is the negative lookahead, alternative 1 is what comes
- // afterwards.
- RegExpNode* node = alternatives_->at(1).node();
- return node->EatsAtLeast(still_to_find, budget - 1, not_at_start);
-}
-
-
-void NegativeLookaroundChoiceNode::GetQuickCheckDetails(
- QuickCheckDetails* details, RegExpCompiler* compiler, int filled_in,
- bool not_at_start) {
- // Alternative 0 is the negative lookahead, alternative 1 is what comes
- // afterwards.
- RegExpNode* node = alternatives_->at(1).node();
- return node->GetQuickCheckDetails(details, compiler, filled_in, not_at_start);
-}
-
-
-int ChoiceNode::EatsAtLeastHelper(int still_to_find,
- int budget,
- RegExpNode* ignore_this_node,
- bool not_at_start) {
- if (budget <= 0) return 0;
- int min = 100;
- int choice_count = alternatives_->length();
- budget = (budget - 1) / choice_count;
- for (int i = 0; i < choice_count; i++) {
- RegExpNode* node = alternatives_->at(i).node();
- if (node == ignore_this_node) continue;
- int node_eats_at_least =
- node->EatsAtLeast(still_to_find, budget, not_at_start);
- if (node_eats_at_least < min) min = node_eats_at_least;
- if (min == 0) return 0;
- }
- return min;
-}
-
-
-int LoopChoiceNode::EatsAtLeast(int still_to_find,
- int budget,
- bool not_at_start) {
- return EatsAtLeastHelper(still_to_find,
- budget - 1,
- loop_node_,
- not_at_start);
-}
-
-
-int ChoiceNode::EatsAtLeast(int still_to_find,
- int budget,
- bool not_at_start) {
- return EatsAtLeastHelper(still_to_find, budget, nullptr, not_at_start);
-}
-
-
-// Takes the left-most 1-bit and smears it out, setting all bits to its right.
-static inline uint32_t SmearBitsRight(uint32_t v) {
- v |= v >> 1;
- v |= v >> 2;
- v |= v >> 4;
- v |= v >> 8;
- v |= v >> 16;
- return v;
-}
-
-
-bool QuickCheckDetails::Rationalize(bool asc) {
- bool found_useful_op = false;
- uint32_t char_mask;
- if (asc) {
- char_mask = String::kMaxOneByteCharCode;
- } else {
- char_mask = String::kMaxUtf16CodeUnit;
- }
- mask_ = 0;
- value_ = 0;
- int char_shift = 0;
- for (int i = 0; i < characters_; i++) {
- Position* pos = &positions_[i];
- if ((pos->mask & String::kMaxOneByteCharCode) != 0) {
- found_useful_op = true;
- }
- mask_ |= (pos->mask & char_mask) << char_shift;
- value_ |= (pos->value & char_mask) << char_shift;
- char_shift += asc ? 8 : 16;
- }
- return found_useful_op;
-}
-
-
-bool RegExpNode::EmitQuickCheck(RegExpCompiler* compiler,
- Trace* bounds_check_trace,
- Trace* trace,
- bool preload_has_checked_bounds,
- Label* on_possible_success,
- QuickCheckDetails* details,
- bool fall_through_on_failure) {
- if (details->characters() == 0) return false;
- GetQuickCheckDetails(
- details, compiler, 0, trace->at_start() == Trace::FALSE_VALUE);
- if (details->cannot_match()) return false;
- if (!details->Rationalize(compiler->one_byte())) return false;
- DCHECK(details->characters() == 1 ||
- compiler->macro_assembler()->CanReadUnaligned());
- uint32_t mask = details->mask();
- uint32_t value = details->value();
-
- RegExpMacroAssembler* assembler = compiler->macro_assembler();
-
- if (trace->characters_preloaded() != details->characters()) {
- DCHECK(trace->cp_offset() == bounds_check_trace->cp_offset());
- // We are attempting to preload the minimum number of characters
- // any choice would eat, so if the bounds check fails, then none of the
- // choices can succeed, so we can just immediately backtrack, rather
- // than go to the next choice.
- assembler->LoadCurrentCharacter(trace->cp_offset(),
- bounds_check_trace->backtrack(),
- !preload_has_checked_bounds,
- details->characters());
- }
-
-
- bool need_mask = true;
-
- if (details->characters() == 1) {
- // If number of characters preloaded is 1 then we used a byte or 16 bit
- // load so the value is already masked down.
- uint32_t char_mask;
- if (compiler->one_byte()) {
- char_mask = String::kMaxOneByteCharCode;
- } else {
- char_mask = String::kMaxUtf16CodeUnit;
- }
- if ((mask & char_mask) == char_mask) need_mask = false;
- mask &= char_mask;
- } 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 & kTwoByteMask) == kTwoByteMask) need_mask = false;
- } else if (details->characters() == 1 && !compiler->one_byte()) {
- if ((mask & kTwoByteMask) == kTwoByteMask) need_mask = false;
- } else {
- if (mask == kFourByteMask) need_mask = false;
- }
- }
-
- if (fall_through_on_failure) {
- if (need_mask) {
- assembler->CheckCharacterAfterAnd(value, mask, on_possible_success);
- } else {
- assembler->CheckCharacter(value, on_possible_success);
- }
- } else {
- if (need_mask) {
- assembler->CheckNotCharacterAfterAnd(value, mask, trace->backtrack());
- } else {
- assembler->CheckNotCharacter(value, trace->backtrack());
- }
- }
- return true;
-}
-
-
-// Here is the meat of GetQuickCheckDetails (see also the comment on the
-// super-class in the .h file).
-//
-// We iterate along the text object, building up for each character a
-// mask and value that can be used to test for a quick failure to match.
-// The masks and values for the positions will be combined into a single
-// machine word for the current character width in order to be used in
-// generating a quick check.
-void TextNode::GetQuickCheckDetails(QuickCheckDetails* details,
- RegExpCompiler* compiler,
- int characters_filled_in,
- bool not_at_start) {
- // Do not collect any quick check details if the text node reads backward,
- // since it reads in the opposite direction than we use for quick checks.
- if (read_backward()) return;
- Isolate* isolate = compiler->macro_assembler()->isolate();
- DCHECK(characters_filled_in < details->characters());
- int characters = details->characters();
- int char_mask;
- if (compiler->one_byte()) {
- char_mask = String::kMaxOneByteCharCode;
- } else {
- char_mask = String::kMaxUtf16CodeUnit;
- }
- for (int k = 0; k < elements()->length(); k++) {
- TextElement elm = elements()->at(k);
- if (elm.text_type() == TextElement::ATOM) {
- Vector<const uc16> quarks = elm.atom()->data();
- for (int i = 0; i < characters && i < quarks.length(); i++) {
- QuickCheckDetails::Position* pos =
- details->positions(characters_filled_in);
- uc16 c = quarks[i];
- if (elm.atom()->ignore_case()) {
- unibrow::uchar chars[4];
- int length = GetCaseIndependentLetters(
- isolate, c, compiler->one_byte(), chars, 4);
- if (length == 0) {
- // This can happen because all case variants are non-Latin1, but we
- // know the input is Latin1.
- details->set_cannot_match();
- pos->determines_perfectly = false;
- return;
- }
- if (length == 1) {
- // This letter has no case equivalents, so it's nice and simple
- // and the mask-compare will determine definitely whether we have
- // a match at this character position.
- pos->mask = char_mask;
- pos->value = c;
- pos->determines_perfectly = true;
- } else {
- uint32_t common_bits = char_mask;
- uint32_t bits = chars[0];
- for (int j = 1; j < length; j++) {
- uint32_t differing_bits = ((chars[j] & common_bits) ^ bits);
- common_bits ^= differing_bits;
- bits &= common_bits;
- }
- // If length is 2 and common bits has only one zero in it then
- // our mask and compare instruction will determine definitely
- // whether we have a match at this character position. Otherwise
- // it can only be an approximate check.
- uint32_t one_zero = (common_bits | ~char_mask);
- if (length == 2 && ((~one_zero) & ((~one_zero) - 1)) == 0) {
- pos->determines_perfectly = true;
- }
- pos->mask = common_bits;
- pos->value = bits;
- }
- } else {
- // Don't ignore case. Nice simple case where the mask-compare will
- // determine definitely whether we have a match at this character
- // position.
- if (c > char_mask) {
- details->set_cannot_match();
- pos->determines_perfectly = false;
- return;
- }
- pos->mask = char_mask;
- pos->value = c;
- pos->determines_perfectly = true;
- }
- characters_filled_in++;
- DCHECK(characters_filled_in <= details->characters());
- if (characters_filled_in == details->characters()) {
- return;
- }
- }
- } else {
- QuickCheckDetails::Position* pos =
- details->positions(characters_filled_in);
- RegExpCharacterClass* tree = elm.char_class();
- ZoneList<CharacterRange>* ranges = tree->ranges(zone());
- DCHECK(!ranges->is_empty());
- if (tree->is_negated()) {
- // A quick check uses multi-character mask and compare. There is no
- // useful way to incorporate a negative char class into this scheme
- // so we just conservatively create a mask and value that will always
- // succeed.
- pos->mask = 0;
- pos->value = 0;
- } else {
- int first_range = 0;
- while (ranges->at(first_range).from() > char_mask) {
- first_range++;
- if (first_range == ranges->length()) {
- details->set_cannot_match();
- pos->determines_perfectly = false;
- return;
- }
- }
- CharacterRange range = ranges->at(first_range);
- uc16 from = range.from();
- uc16 to = range.to();
- if (to > char_mask) {
- to = char_mask;
- }
- uint32_t differing_bits = (from ^ to);
- // A mask and compare is only perfect if the differing bits form a
- // number like 00011111 with one single block of trailing 1s.
- if ((differing_bits & (differing_bits + 1)) == 0 &&
- from + differing_bits == to) {
- pos->determines_perfectly = true;
- }
- uint32_t common_bits = ~SmearBitsRight(differing_bits);
- uint32_t bits = (from & common_bits);
- for (int i = first_range + 1; i < ranges->length(); i++) {
- CharacterRange range = ranges->at(i);
- uc16 from = range.from();
- uc16 to = range.to();
- if (from > char_mask) continue;
- if (to > char_mask) to = char_mask;
- // Here we are combining more ranges into the mask and compare
- // value. With each new range the mask becomes more sparse and
- // so the chances of a false positive rise. A character class
- // with multiple ranges is assumed never to be equivalent to a
- // mask and compare operation.
- pos->determines_perfectly = false;
- uint32_t new_common_bits = (from ^ to);
- new_common_bits = ~SmearBitsRight(new_common_bits);
- common_bits &= new_common_bits;
- bits &= new_common_bits;
- uint32_t differing_bits = (from & common_bits) ^ bits;
- common_bits ^= differing_bits;
- bits &= common_bits;
- }
- pos->mask = common_bits;
- pos->value = bits;
- }
- characters_filled_in++;
- DCHECK(characters_filled_in <= details->characters());
- if (characters_filled_in == details->characters()) {
- return;
- }
- }
- }
- DCHECK(characters_filled_in != details->characters());
- if (!details->cannot_match()) {
- on_success()-> GetQuickCheckDetails(details,
- compiler,
- characters_filled_in,
- true);
- }
-}
-
-
-void QuickCheckDetails::Clear() {
- for (int i = 0; i < characters_; i++) {
- positions_[i].mask = 0;
- positions_[i].value = 0;
- positions_[i].determines_perfectly = false;
- }
- characters_ = 0;
-}
-
-
-void QuickCheckDetails::Advance(int by, bool one_byte) {
- if (by >= characters_ || by < 0) {
- DCHECK_IMPLIES(by < 0, characters_ == 0);
- Clear();
- return;
- }
- DCHECK_LE(characters_ - by, 4);
- DCHECK_LE(characters_, 4);
- for (int i = 0; i < characters_ - by; i++) {
- positions_[i] = positions_[by + i];
- }
- for (int i = characters_ - by; i < characters_; i++) {
- positions_[i].mask = 0;
- positions_[i].value = 0;
- positions_[i].determines_perfectly = false;
- }
- characters_ -= by;
- // We could change mask_ and value_ here but we would never advance unless
- // they had already been used in a check and they won't be used again because
- // it would gain us nothing. So there's no point.
-}
-
-
-void QuickCheckDetails::Merge(QuickCheckDetails* other, int from_index) {
- DCHECK(characters_ == other->characters_);
- if (other->cannot_match_) {
- return;
- }
- if (cannot_match_) {
- *this = *other;
- return;
- }
- for (int i = from_index; i < characters_; i++) {
- QuickCheckDetails::Position* pos = positions(i);
- QuickCheckDetails::Position* other_pos = other->positions(i);
- if (pos->mask != other_pos->mask ||
- pos->value != other_pos->value ||
- !other_pos->determines_perfectly) {
- // Our mask-compare operation will be approximate unless we have the
- // exact same operation on both sides of the alternation.
- pos->determines_perfectly = false;
- }
- pos->mask &= other_pos->mask;
- pos->value &= pos->mask;
- other_pos->value &= pos->mask;
- uc16 differing_bits = (pos->value ^ other_pos->value);
- pos->mask &= ~differing_bits;
- pos->value &= pos->mask;
- }
-}
-
-
-class VisitMarker {
- public:
- explicit VisitMarker(NodeInfo* info) : info_(info) {
- DCHECK(!info->visited);
- info->visited = true;
- }
- ~VisitMarker() {
- info_->visited = false;
- }
- private:
- NodeInfo* info_;
-};
-
-RegExpNode* SeqRegExpNode::FilterOneByte(int depth) {
- if (info()->replacement_calculated) return replacement();
- if (depth < 0) return this;
- DCHECK(!info()->visited);
- VisitMarker marker(info());
- return FilterSuccessor(depth - 1);
-}
-
-RegExpNode* SeqRegExpNode::FilterSuccessor(int depth) {
- RegExpNode* next = on_success_->FilterOneByte(depth - 1);
- if (next == nullptr) return set_replacement(nullptr);
- on_success_ = next;
- return set_replacement(this);
-}
-
-// We need to check for the following characters: 0x39C 0x3BC 0x178.
-static inline bool RangeContainsLatin1Equivalents(CharacterRange range) {
- // TODO(dcarney): this could be a lot more efficient.
- return range.Contains(0x039C) || range.Contains(0x03BC) ||
- range.Contains(0x0178);
-}
-
-
-static bool RangesContainLatin1Equivalents(ZoneList<CharacterRange>* ranges) {
- for (int i = 0; i < ranges->length(); i++) {
- // TODO(dcarney): this could be a lot more efficient.
- if (RangeContainsLatin1Equivalents(ranges->at(i))) return true;
- }
- return false;
-}
-
-RegExpNode* TextNode::FilterOneByte(int depth) {
- if (info()->replacement_calculated) return replacement();
- if (depth < 0) return this;
- DCHECK(!info()->visited);
- VisitMarker marker(info());
- int element_count = elements()->length();
- for (int i = 0; i < element_count; i++) {
- TextElement elm = elements()->at(i);
- if (elm.text_type() == TextElement::ATOM) {
- Vector<const uc16> quarks = elm.atom()->data();
- for (int j = 0; j < quarks.length(); j++) {
- uint16_t c = quarks[j];
- if (elm.atom()->ignore_case()) {
- c = unibrow::Latin1::TryConvertToLatin1(c);
- }
- if (c > unibrow::Latin1::kMaxChar) return set_replacement(nullptr);
- // Replace quark in case we converted to Latin-1.
- uint16_t* writable_quarks = const_cast<uint16_t*>(quarks.begin());
- writable_quarks[j] = c;
- }
- } else {
- DCHECK(elm.text_type() == TextElement::CHAR_CLASS);
- RegExpCharacterClass* cc = elm.char_class();
- ZoneList<CharacterRange>* ranges = cc->ranges(zone());
- 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()) {
- if (range_count != 0 &&
- ranges->at(0).from() == 0 &&
- ranges->at(0).to() >= String::kMaxOneByteCharCode) {
- // This will be handled in a later filter.
- if (IgnoreCase(cc->flags()) && RangesContainLatin1Equivalents(ranges))
- continue;
- return set_replacement(nullptr);
- }
- } else {
- if (range_count == 0 ||
- ranges->at(0).from() > String::kMaxOneByteCharCode) {
- // This will be handled in a later filter.
- if (IgnoreCase(cc->flags()) && RangesContainLatin1Equivalents(ranges))
- continue;
- return set_replacement(nullptr);
- }
- }
- }
- }
- return FilterSuccessor(depth - 1);
-}
-
-RegExpNode* LoopChoiceNode::FilterOneByte(int depth) {
- if (info()->replacement_calculated) return replacement();
- if (depth < 0) return this;
- if (info()->visited) return this;
- {
- VisitMarker marker(info());
-
- RegExpNode* continue_replacement = continue_node_->FilterOneByte(depth - 1);
- // If we can't continue after the loop then there is no sense in doing the
- // loop.
- if (continue_replacement == nullptr) return set_replacement(nullptr);
- }
-
- return ChoiceNode::FilterOneByte(depth - 1);
-}
-
-RegExpNode* ChoiceNode::FilterOneByte(int depth) {
- if (info()->replacement_calculated) return replacement();
- if (depth < 0) return this;
- if (info()->visited) return this;
- VisitMarker marker(info());
- int choice_count = alternatives_->length();
-
- for (int i = 0; i < choice_count; i++) {
- GuardedAlternative alternative = alternatives_->at(i);
- if (alternative.guards() != nullptr &&
- alternative.guards()->length() != 0) {
- set_replacement(this);
- return this;
- }
- }
-
- int surviving = 0;
- RegExpNode* survivor = nullptr;
- for (int i = 0; i < choice_count; i++) {
- GuardedAlternative alternative = alternatives_->at(i);
- RegExpNode* replacement = alternative.node()->FilterOneByte(depth - 1);
- DCHECK(replacement != this); // No missing EMPTY_MATCH_CHECK.
- if (replacement != nullptr) {
- alternatives_->at(i).set_node(replacement);
- surviving++;
- survivor = replacement;
- }
- }
- if (surviving < 2) return set_replacement(survivor);
-
- set_replacement(this);
- if (surviving == choice_count) {
- return this;
- }
- // Only some of the nodes survived the filtering. We need to rebuild the
- // alternatives list.
- ZoneList<GuardedAlternative>* new_alternatives =
- new(zone()) ZoneList<GuardedAlternative>(surviving, zone());
- for (int i = 0; i < choice_count; i++) {
- RegExpNode* replacement =
- alternatives_->at(i).node()->FilterOneByte(depth - 1);
- if (replacement != nullptr) {
- alternatives_->at(i).set_node(replacement);
- new_alternatives->Add(alternatives_->at(i), zone());
- }
- }
- alternatives_ = new_alternatives;
- return this;
-}
-
-RegExpNode* NegativeLookaroundChoiceNode::FilterOneByte(int depth) {
- if (info()->replacement_calculated) return replacement();
- if (depth < 0) return this;
- if (info()->visited) return this;
- VisitMarker marker(info());
- // Alternative 0 is the negative lookahead, alternative 1 is what comes
- // afterwards.
- RegExpNode* node = alternatives_->at(1).node();
- RegExpNode* replacement = node->FilterOneByte(depth - 1);
- if (replacement == nullptr) return set_replacement(nullptr);
- alternatives_->at(1).set_node(replacement);
-
- RegExpNode* neg_node = alternatives_->at(0).node();
- RegExpNode* neg_replacement = neg_node->FilterOneByte(depth - 1);
- // If the negative lookahead is always going to fail then
- // we don't need to check it.
- if (neg_replacement == nullptr) return set_replacement(replacement);
- alternatives_->at(0).set_node(neg_replacement);
- return set_replacement(this);
-}
-
-
-void LoopChoiceNode::GetQuickCheckDetails(QuickCheckDetails* details,
- RegExpCompiler* compiler,
- int characters_filled_in,
- bool not_at_start) {
- if (body_can_be_zero_length_ || info()->visited) return;
- VisitMarker marker(info());
- return ChoiceNode::GetQuickCheckDetails(details,
- compiler,
- characters_filled_in,
- not_at_start);
-}
-
-
-void LoopChoiceNode::FillInBMInfo(Isolate* isolate, int offset, int budget,
- BoyerMooreLookahead* bm, bool not_at_start) {
- if (body_can_be_zero_length_ || budget <= 0) {
- bm->SetRest(offset);
- SaveBMInfo(bm, not_at_start, offset);
- return;
- }
- ChoiceNode::FillInBMInfo(isolate, offset, budget - 1, bm, not_at_start);
- SaveBMInfo(bm, not_at_start, offset);
-}
-
-
-void ChoiceNode::GetQuickCheckDetails(QuickCheckDetails* details,
- RegExpCompiler* compiler,
- int characters_filled_in,
- bool not_at_start) {
- not_at_start = (not_at_start || not_at_start_);
- int choice_count = alternatives_->length();
- DCHECK_LT(0, choice_count);
- alternatives_->at(0).node()->GetQuickCheckDetails(details,
- compiler,
- characters_filled_in,
- not_at_start);
- for (int i = 1; i < choice_count; i++) {
- QuickCheckDetails new_details(details->characters());
- RegExpNode* node = alternatives_->at(i).node();
- node->GetQuickCheckDetails(&new_details, compiler,
- characters_filled_in,
- not_at_start);
- // Here we merge the quick match details of the two branches.
- details->Merge(&new_details, characters_filled_in);
- }
-}
-
-
-// Check for [0-9A-Z_a-z].
-static void EmitWordCheck(RegExpMacroAssembler* assembler,
- Label* word,
- Label* non_word,
- bool fall_through_on_word) {
- if (assembler->CheckSpecialCharacterClass(
- fall_through_on_word ? 'w' : 'W',
- fall_through_on_word ? non_word : word)) {
- // Optimized implementation available.
- return;
- }
- assembler->CheckCharacterGT('z', non_word);
- assembler->CheckCharacterLT('0', non_word);
- assembler->CheckCharacterGT('a' - 1, word);
- assembler->CheckCharacterLT('9' + 1, word);
- assembler->CheckCharacterLT('A', non_word);
- assembler->CheckCharacterLT('Z' + 1, word);
- if (fall_through_on_word) {
- assembler->CheckNotCharacter('_', non_word);
- } else {
- assembler->CheckCharacter('_', word);
- }
-}
-
-
-// Emit the code to check for a ^ in multiline mode (1-character lookbehind
-// that matches newline or the start of input).
-static void EmitHat(RegExpCompiler* compiler,
- RegExpNode* on_success,
- Trace* trace) {
- RegExpMacroAssembler* assembler = compiler->macro_assembler();
- // We will be loading the previous character into the current character
- // register.
- Trace new_trace(*trace);
- new_trace.InvalidateCurrentCharacter();
-
- Label ok;
- if (new_trace.cp_offset() == 0) {
- // The start of input counts as a newline in this context, so skip to
- // ok if we are at the start.
- assembler->CheckAtStart(&ok);
- }
- // We already checked that we are not at the start of input so it must be
- // OK to load the previous character.
- assembler->LoadCurrentCharacter(new_trace.cp_offset() -1,
- new_trace.backtrack(),
- false);
- if (!assembler->CheckSpecialCharacterClass('n',
- new_trace.backtrack())) {
- // Newline means \n, \r, 0x2028 or 0x2029.
- if (!compiler->one_byte()) {
- assembler->CheckCharacterAfterAnd(0x2028, 0xFFFE, &ok);
- }
- assembler->CheckCharacter('\n', &ok);
- assembler->CheckNotCharacter('\r', new_trace.backtrack());
- }
- assembler->Bind(&ok);
- on_success->Emit(compiler, &new_trace);
-}
-
-
-// Emit the code to handle \b and \B (word-boundary or non-word-boundary).
-void AssertionNode::EmitBoundaryCheck(RegExpCompiler* compiler, Trace* trace) {
- RegExpMacroAssembler* assembler = compiler->macro_assembler();
- Isolate* isolate = assembler->isolate();
- Trace::TriBool next_is_word_character = Trace::UNKNOWN;
- bool not_at_start = (trace->at_start() == Trace::FALSE_VALUE);
- BoyerMooreLookahead* lookahead = bm_info(not_at_start);
- if (lookahead == nullptr) {
- int eats_at_least =
- Min(kMaxLookaheadForBoyerMoore, EatsAtLeast(kMaxLookaheadForBoyerMoore,
- kRecursionBudget,
- not_at_start));
- if (eats_at_least >= 1) {
- BoyerMooreLookahead* bm =
- new(zone()) BoyerMooreLookahead(eats_at_least, compiler, zone());
- FillInBMInfo(isolate, 0, kRecursionBudget, bm, not_at_start);
- if (bm->at(0)->is_non_word())
- next_is_word_character = Trace::FALSE_VALUE;
- if (bm->at(0)->is_word()) next_is_word_character = Trace::TRUE_VALUE;
- }
- } else {
- if (lookahead->at(0)->is_non_word())
- next_is_word_character = Trace::FALSE_VALUE;
- if (lookahead->at(0)->is_word())
- next_is_word_character = Trace::TRUE_VALUE;
- }
- bool at_boundary = (assertion_type_ == AssertionNode::AT_BOUNDARY);
- if (next_is_word_character == Trace::UNKNOWN) {
- Label before_non_word;
- Label before_word;
- if (trace->characters_preloaded() != 1) {
- assembler->LoadCurrentCharacter(trace->cp_offset(), &before_non_word);
- }
- // Fall through on non-word.
- EmitWordCheck(assembler, &before_word, &before_non_word, false);
- // Next character is not a word character.
- assembler->Bind(&before_non_word);
- Label ok;
- BacktrackIfPrevious(compiler, trace, at_boundary ? kIsNonWord : kIsWord);
- assembler->GoTo(&ok);
-
- assembler->Bind(&before_word);
- BacktrackIfPrevious(compiler, trace, at_boundary ? kIsWord : kIsNonWord);
- assembler->Bind(&ok);
- } else if (next_is_word_character == Trace::TRUE_VALUE) {
- BacktrackIfPrevious(compiler, trace, at_boundary ? kIsWord : kIsNonWord);
- } else {
- DCHECK(next_is_word_character == Trace::FALSE_VALUE);
- BacktrackIfPrevious(compiler, trace, at_boundary ? kIsNonWord : kIsWord);
- }
-}
-
-
-void AssertionNode::BacktrackIfPrevious(
- RegExpCompiler* compiler,
- Trace* trace,
- AssertionNode::IfPrevious backtrack_if_previous) {
- RegExpMacroAssembler* assembler = compiler->macro_assembler();
- Trace new_trace(*trace);
- new_trace.InvalidateCurrentCharacter();
-
- Label fall_through, dummy;
-
- Label* non_word = backtrack_if_previous == kIsNonWord ?
- new_trace.backtrack() :
- &fall_through;
- Label* word = backtrack_if_previous == kIsNonWord ?
- &fall_through :
- new_trace.backtrack();
-
- if (new_trace.cp_offset() == 0) {
- // The start of input counts as a non-word character, so the question is
- // decided if we are at the start.
- assembler->CheckAtStart(non_word);
- }
- // We already checked that we are not at the start of input so it must be
- // OK to load the previous character.
- assembler->LoadCurrentCharacter(new_trace.cp_offset() - 1, &dummy, false);
- EmitWordCheck(assembler, word, non_word, backtrack_if_previous == kIsNonWord);
-
- assembler->Bind(&fall_through);
- on_success()->Emit(compiler, &new_trace);
-}
-
-
-void AssertionNode::GetQuickCheckDetails(QuickCheckDetails* details,
- RegExpCompiler* compiler,
- int filled_in,
- bool not_at_start) {
- if (assertion_type_ == AT_START && not_at_start) {
- details->set_cannot_match();
- return;
- }
- return on_success()->GetQuickCheckDetails(details,
- compiler,
- filled_in,
- not_at_start);
-}
-
-
-void AssertionNode::Emit(RegExpCompiler* compiler, Trace* trace) {
- RegExpMacroAssembler* assembler = compiler->macro_assembler();
- switch (assertion_type_) {
- case AT_END: {
- Label ok;
- assembler->CheckPosition(trace->cp_offset(), &ok);
- assembler->GoTo(trace->backtrack());
- assembler->Bind(&ok);
- break;
- }
- case AT_START: {
- if (trace->at_start() == Trace::FALSE_VALUE) {
- assembler->GoTo(trace->backtrack());
- return;
- }
- if (trace->at_start() == Trace::UNKNOWN) {
- assembler->CheckNotAtStart(trace->cp_offset(), trace->backtrack());
- Trace at_start_trace = *trace;
- at_start_trace.set_at_start(Trace::TRUE_VALUE);
- on_success()->Emit(compiler, &at_start_trace);
- return;
- }
- }
- break;
- case AFTER_NEWLINE:
- EmitHat(compiler, on_success(), trace);
- return;
- case AT_BOUNDARY:
- case AT_NON_BOUNDARY: {
- EmitBoundaryCheck(compiler, trace);
- return;
- }
- }
- on_success()->Emit(compiler, trace);
-}
-
-
-static bool DeterminedAlready(QuickCheckDetails* quick_check, int offset) {
- if (quick_check == nullptr) return false;
- if (offset >= quick_check->characters()) return false;
- return quick_check->positions(offset)->determines_perfectly;
-}
-
-
-static void UpdateBoundsCheck(int index, int* checked_up_to) {
- if (index > *checked_up_to) {
- *checked_up_to = index;
- }
-}
-
-
-// We call this repeatedly to generate code for each pass over the text node.
-// The passes are in increasing order of difficulty because we hope one
-// of the first passes will fail in which case we are saved the work of the
-// later passes. for example for the case independent regexp /%[asdfghjkl]a/
-// we will check the '%' in the first pass, the case independent 'a' in the
-// second pass and the character class in the last pass.
-//
-// The passes are done from right to left, so for example to test for /bar/
-// we will first test for an 'r' with offset 2, then an 'a' with offset 1
-// and then a 'b' with offset 0. This means we can avoid the end-of-input
-// bounds check most of the time. In the example we only need to check for
-// end-of-input when loading the putative 'r'.
-//
-// A slight complication involves the fact that the first character may already
-// be fetched into a register by the previous node. In this case we want to
-// do the test for that character first. We do this in separate passes. The
-// 'preloaded' argument indicates that we are doing such a 'pass'. If such a
-// pass has been performed then subsequent passes will have true in
-// first_element_checked to indicate that that character does not need to be
-// checked again.
-//
-// In addition to all this we are passed a Trace, which can
-// contain an AlternativeGeneration object. In this AlternativeGeneration
-// object we can see details of any quick check that was already passed in
-// order to get to the code we are now generating. The quick check can involve
-// loading characters, which means we do not need to recheck the bounds
-// up to the limit the quick check already checked. In addition the quick
-// check can have involved a mask and compare operation which may simplify
-// or obviate the need for further checks at some character positions.
-void TextNode::TextEmitPass(RegExpCompiler* compiler,
- TextEmitPassType pass,
- bool preloaded,
- Trace* trace,
- bool first_element_checked,
- int* checked_up_to) {
- RegExpMacroAssembler* assembler = compiler->macro_assembler();
- Isolate* isolate = assembler->isolate();
- bool one_byte = compiler->one_byte();
- Label* backtrack = trace->backtrack();
- QuickCheckDetails* quick_check = trace->quick_check_performed();
- int element_count = elements()->length();
- int backward_offset = read_backward() ? -Length() : 0;
- for (int i = preloaded ? 0 : element_count - 1; i >= 0; i--) {
- TextElement elm = elements()->at(i);
- int cp_offset = trace->cp_offset() + elm.cp_offset() + backward_offset;
- if (elm.text_type() == TextElement::ATOM) {
- if (SkipPass(pass, elm.atom()->ignore_case())) continue;
- Vector<const uc16> quarks = elm.atom()->data();
- for (int j = preloaded ? 0 : quarks.length() - 1; j >= 0; j--) {
- if (first_element_checked && i == 0 && j == 0) continue;
- if (DeterminedAlready(quick_check, elm.cp_offset() + j)) continue;
- EmitCharacterFunction* emit_function = nullptr;
- uc16 quark = quarks[j];
- if (elm.atom()->ignore_case()) {
- // Everywhere else we assume that a non-Latin-1 character cannot match
- // a Latin-1 character. Avoid the cases where this is assumption is
- // invalid by using the Latin1 equivalent instead.
- quark = unibrow::Latin1::TryConvertToLatin1(quark);
- }
- switch (pass) {
- case NON_LATIN1_MATCH:
- DCHECK(one_byte);
- if (quark > String::kMaxOneByteCharCode) {
- assembler->GoTo(backtrack);
- return;
- }
- break;
- case NON_LETTER_CHARACTER_MATCH:
- emit_function = &EmitAtomNonLetter;
- break;
- case SIMPLE_CHARACTER_MATCH:
- emit_function = &EmitSimpleCharacter;
- break;
- case CASE_CHARACTER_MATCH:
- emit_function = &EmitAtomLetter;
- break;
- default:
- break;
- }
- if (emit_function != nullptr) {
- bool bounds_check = *checked_up_to < cp_offset + j || read_backward();
- bool bound_checked =
- emit_function(isolate, compiler, quark, backtrack, cp_offset + j,
- bounds_check, preloaded);
- if (bound_checked) UpdateBoundsCheck(cp_offset + j, checked_up_to);
- }
- }
- } else {
- DCHECK_EQ(TextElement::CHAR_CLASS, elm.text_type());
- if (pass == CHARACTER_CLASS_MATCH) {
- if (first_element_checked && i == 0) continue;
- if (DeterminedAlready(quick_check, elm.cp_offset())) continue;
- RegExpCharacterClass* cc = elm.char_class();
- bool bounds_check = *checked_up_to < cp_offset || read_backward();
- EmitCharClass(assembler, cc, one_byte, backtrack, cp_offset,
- bounds_check, preloaded, zone());
- UpdateBoundsCheck(cp_offset, checked_up_to);
- }
- }
- }
-}
-
-
-int TextNode::Length() {
- TextElement elm = elements()->last();
- DCHECK_LE(0, elm.cp_offset());
- return elm.cp_offset() + elm.length();
-}
-
-bool TextNode::SkipPass(TextEmitPassType pass, bool ignore_case) {
- if (ignore_case) {
- return pass == SIMPLE_CHARACTER_MATCH;
- } else {
- return pass == NON_LETTER_CHARACTER_MATCH || pass == CASE_CHARACTER_MATCH;
- }
-}
-
-TextNode* TextNode::CreateForCharacterRanges(Zone* zone,
- ZoneList<CharacterRange>* ranges,
- bool read_backward,
- RegExpNode* on_success,
- JSRegExp::Flags flags) {
- DCHECK_NOT_NULL(ranges);
- ZoneList<TextElement>* elms = new (zone) ZoneList<TextElement>(1, zone);
- elms->Add(TextElement::CharClass(
- new (zone) RegExpCharacterClass(zone, ranges, flags)),
- 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,
- JSRegExp::Flags flags) {
- 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(zone, lead_ranges, flags)),
- zone);
- elms->Add(TextElement::CharClass(
- new (zone) RegExpCharacterClass(zone, trail_ranges, flags)),
- 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
-// pass from left to right. Instead we pass over the text node several times,
-// emitting code for some character positions every time. See the comment on
-// TextEmitPass for details.
-void TextNode::Emit(RegExpCompiler* compiler, Trace* trace) {
- LimitResult limit_result = LimitVersions(compiler, trace);
- if (limit_result == DONE) return;
- DCHECK(limit_result == CONTINUE);
-
- if (trace->cp_offset() + Length() > RegExpMacroAssembler::kMaxCPOffset) {
- compiler->SetRegExpTooBig();
- return;
- }
-
- if (compiler->one_byte()) {
- int dummy = 0;
- TextEmitPass(compiler, NON_LATIN1_MATCH, false, trace, false, &dummy);
- }
-
- bool first_elt_done = false;
- int bound_checked_to = trace->cp_offset() - 1;
- bound_checked_to += trace->bound_checked_up_to();
-
- // If a character is preloaded into the current character register then
- // check that now.
- if (trace->characters_preloaded() == 1) {
- for (int pass = kFirstRealPass; pass <= kLastPass; pass++) {
- TextEmitPass(compiler, static_cast<TextEmitPassType>(pass), true, trace,
- false, &bound_checked_to);
- }
- first_elt_done = true;
- }
-
- for (int pass = kFirstRealPass; pass <= kLastPass; pass++) {
- TextEmitPass(compiler, static_cast<TextEmitPassType>(pass), false, trace,
- first_elt_done, &bound_checked_to);
- }
-
- Trace successor_trace(*trace);
- // If we advance backward, we may end up at the start.
- successor_trace.AdvanceCurrentPositionInTrace(
- read_backward() ? -Length() : Length(), compiler);
- successor_trace.set_at_start(read_backward() ? Trace::UNKNOWN
- : Trace::FALSE_VALUE);
- RecursionCheck rc(compiler);
- on_success()->Emit(compiler, &successor_trace);
-}
-
-
-void Trace::InvalidateCurrentCharacter() {
- characters_preloaded_ = 0;
-}
-
-
-void Trace::AdvanceCurrentPositionInTrace(int by, RegExpCompiler* compiler) {
- // We don't have an instruction for shifting the current character register
- // down or for using a shifted value for anything so lets just forget that
- // we preloaded any characters into it.
- characters_preloaded_ = 0;
- // Adjust the offsets of the quick check performed information. This
- // information is used to find out what we already determined about the
- // characters by means of mask and compare.
- quick_check_performed_.Advance(by, compiler->one_byte());
- cp_offset_ += by;
- if (cp_offset_ > RegExpMacroAssembler::kMaxCPOffset) {
- compiler->SetRegExpTooBig();
- cp_offset_ = 0;
- }
- bound_checked_up_to_ = Max(0, bound_checked_up_to_ - by);
-}
-
-
-void TextNode::MakeCaseIndependent(Isolate* isolate, bool is_one_byte) {
- int element_count = elements()->length();
- for (int i = 0; i < element_count; i++) {
- TextElement elm = elements()->at(i);
- if (elm.text_type() == TextElement::CHAR_CLASS) {
- RegExpCharacterClass* cc = elm.char_class();
-#ifdef V8_INTL_SUPPORT
- bool case_equivalents_already_added =
- NeedsUnicodeCaseEquivalents(cc->flags());
-#else
- bool case_equivalents_already_added = false;
-#endif
- if (IgnoreCase(cc->flags()) && !case_equivalents_already_added) {
- // None of the standard character classes is different in the case
- // 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());
- CharacterRange::AddCaseEquivalents(isolate, zone(), ranges,
- is_one_byte);
- }
- }
- }
-}
-
-
-int TextNode::GreedyLoopTextLength() { return Length(); }
-
-
-RegExpNode* TextNode::GetSuccessorOfOmnivorousTextNode(
- RegExpCompiler* compiler) {
- if (read_backward()) return nullptr;
- if (elements()->length() != 1) return nullptr;
- TextElement elm = elements()->at(0);
- if (elm.text_type() != TextElement::CHAR_CLASS) return nullptr;
- RegExpCharacterClass* node = elm.char_class();
- ZoneList<CharacterRange>* ranges = node->ranges(zone());
- CharacterRange::Canonicalize(ranges);
- if (node->is_negated()) {
- return ranges->length() == 0 ? on_success() : nullptr;
- }
- if (ranges->length() != 1) return nullptr;
- uint32_t max_char;
- if (compiler->one_byte()) {
- max_char = String::kMaxOneByteCharCode;
- } else {
- max_char = String::kMaxUtf16CodeUnit;
- }
- return ranges->at(0).IsEverything(max_char) ? on_success() : nullptr;
-}
-
-
-// Finds the fixed match length of a sequence of nodes that goes from
-// this alternative and back to this choice node. If there are variable
-// length nodes or other complications in the way then return a sentinel
-// value indicating that a greedy loop cannot be constructed.
-int ChoiceNode::GreedyLoopTextLengthForAlternative(
- GuardedAlternative* alternative) {
- int length = 0;
- RegExpNode* node = alternative->node();
- // Later we will generate code for all these text nodes using recursion
- // so we have to limit the max number.
- int recursion_depth = 0;
- while (node != this) {
- if (recursion_depth++ > RegExpCompiler::kMaxRecursion) {
- return kNodeIsTooComplexForGreedyLoops;
- }
- int node_length = node->GreedyLoopTextLength();
- if (node_length == kNodeIsTooComplexForGreedyLoops) {
- return kNodeIsTooComplexForGreedyLoops;
- }
- length += node_length;
- SeqRegExpNode* seq_node = static_cast<SeqRegExpNode*>(node);
- node = seq_node->on_success();
- }
- return read_backward() ? -length : length;
-}
-
-
-void LoopChoiceNode::AddLoopAlternative(GuardedAlternative alt) {
- DCHECK_NULL(loop_node_);
- AddAlternative(alt);
- loop_node_ = alt.node();
-}
-
-
-void LoopChoiceNode::AddContinueAlternative(GuardedAlternative alt) {
- DCHECK_NULL(continue_node_);
- AddAlternative(alt);
- continue_node_ = alt.node();
-}
-
-
-void LoopChoiceNode::Emit(RegExpCompiler* compiler, Trace* trace) {
- RegExpMacroAssembler* macro_assembler = compiler->macro_assembler();
- if (trace->stop_node() == this) {
- // Back edge of greedy optimized loop node graph.
- int text_length =
- GreedyLoopTextLengthForAlternative(&(alternatives_->at(0)));
- DCHECK_NE(kNodeIsTooComplexForGreedyLoops, text_length);
- // Update the counter-based backtracking info on the stack. This is an
- // optimization for greedy loops (see below).
- DCHECK(trace->cp_offset() == text_length);
- macro_assembler->AdvanceCurrentPosition(text_length);
- macro_assembler->GoTo(trace->loop_label());
- return;
- }
- DCHECK_NULL(trace->stop_node());
- if (!trace->is_trivial()) {
- trace->Flush(compiler, this);
- return;
- }
- ChoiceNode::Emit(compiler, trace);
-}
-
-
-int ChoiceNode::CalculatePreloadCharacters(RegExpCompiler* compiler,
- int eats_at_least) {
- int preload_characters = Min(4, eats_at_least);
- DCHECK_LE(preload_characters, 4);
- if (compiler->macro_assembler()->CanReadUnaligned()) {
- bool one_byte = compiler->one_byte();
- if (one_byte) {
- // We can't preload 3 characters because there is no machine instruction
- // to do that. We can't just load 4 because we could be reading
- // beyond the end of the string, which could cause a memory fault.
- if (preload_characters == 3) preload_characters = 2;
- } else {
- if (preload_characters > 2) preload_characters = 2;
- }
- } else {
- if (preload_characters > 1) preload_characters = 1;
- }
- return preload_characters;
-}
-
-
-// This class is used when generating the alternatives in a choice node. It
-// records the way the alternative is being code generated.
-class AlternativeGeneration: public Malloced {
- public:
- AlternativeGeneration()
- : possible_success(),
- expects_preload(false),
- after(),
- quick_check_details() { }
- Label possible_success;
- bool expects_preload;
- Label after;
- QuickCheckDetails quick_check_details;
-};
-
-
-// Creates a list of AlternativeGenerations. If the list has a reasonable
-// size then it is on the stack, otherwise the excess is on the heap.
-class AlternativeGenerationList {
- public:
- AlternativeGenerationList(int count, Zone* zone)
- : alt_gens_(count, zone) {
- for (int i = 0; i < count && i < kAFew; i++) {
- alt_gens_.Add(a_few_alt_gens_ + i, zone);
- }
- for (int i = kAFew; i < count; i++) {
- alt_gens_.Add(new AlternativeGeneration(), zone);
- }
- }
- ~AlternativeGenerationList() {
- for (int i = kAFew; i < alt_gens_.length(); i++) {
- delete alt_gens_[i];
- alt_gens_[i] = nullptr;
- }
- }
-
- AlternativeGeneration* at(int i) {
- return alt_gens_[i];
- }
-
- private:
- static const int kAFew = 10;
- ZoneList<AlternativeGeneration*> alt_gens_;
- AlternativeGeneration a_few_alt_gens_[kAFew];
-};
-
-
-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, 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, kRangeEndMarker};
-static const int kWordRangeCount = arraysize(kWordRanges);
-static const int kDigitRanges[] = {'0', '9' + 1, kRangeEndMarker};
-static const int kDigitRangeCount = arraysize(kDigitRanges);
-static const int kSurrogateRanges[] = {
- kLeadSurrogateStart, kLeadSurrogateStart + 1, kRangeEndMarker};
-static const int kSurrogateRangeCount = arraysize(kSurrogateRanges);
-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));
-}
-
-
-void BoyerMoorePositionInfo::SetInterval(const Interval& interval) {
- s_ = AddRange(s_, kSpaceRanges, kSpaceRangeCount, interval);
- w_ = AddRange(w_, kWordRanges, kWordRangeCount, interval);
- d_ = AddRange(d_, kDigitRanges, kDigitRangeCount, interval);
- surrogate_ =
- AddRange(surrogate_, kSurrogateRanges, kSurrogateRangeCount, interval);
- if (interval.to() - interval.from() >= kMapSize - 1) {
- if (map_count_ != kMapSize) {
- map_count_ = kMapSize;
- for (int i = 0; i < kMapSize; i++) map_->at(i) = true;
- }
- return;
- }
- for (int i = interval.from(); i <= interval.to(); i++) {
- int mod_character = (i & kMask);
- if (!map_->at(mod_character)) {
- map_count_++;
- map_->at(mod_character) = true;
- }
- if (map_count_ == kMapSize) return;
- }
-}
-
-
-void BoyerMoorePositionInfo::SetAll() {
- s_ = w_ = d_ = kLatticeUnknown;
- if (map_count_ != kMapSize) {
- map_count_ = kMapSize;
- for (int i = 0; i < kMapSize; i++) map_->at(i) = true;
- }
-}
-
-
-BoyerMooreLookahead::BoyerMooreLookahead(
- int length, RegExpCompiler* compiler, Zone* zone)
- : length_(length),
- compiler_(compiler) {
- if (compiler->one_byte()) {
- max_char_ = String::kMaxOneByteCharCode;
- } else {
- max_char_ = String::kMaxUtf16CodeUnit;
- }
- bitmaps_ = new(zone) ZoneList<BoyerMoorePositionInfo*>(length, zone);
- for (int i = 0; i < length; i++) {
- bitmaps_->Add(new(zone) BoyerMoorePositionInfo(zone), zone);
- }
-}
-
-
-// Find the longest range of lookahead that has the fewest number of different
-// characters that can occur at a given position. Since we are optimizing two
-// different parameters at once this is a tradeoff.
-bool BoyerMooreLookahead::FindWorthwhileInterval(int* from, int* to) {
- int biggest_points = 0;
- // If more than 32 characters out of 128 can occur it is unlikely that we can
- // be lucky enough to step forwards much of the time.
- const int kMaxMax = 32;
- for (int max_number_of_chars = 4;
- max_number_of_chars < kMaxMax;
- max_number_of_chars *= 2) {
- biggest_points =
- FindBestInterval(max_number_of_chars, biggest_points, from, to);
- }
- if (biggest_points == 0) return false;
- return true;
-}
-
-
-// Find the highest-points range between 0 and length_ where the character
-// information is not too vague. 'Too vague' means that there are more than
-// max_number_of_chars that can occur at this position. Calculates the number
-// of points as the product of width-of-the-range and
-// probability-of-finding-one-of-the-characters, where the probability is
-// calculated using the frequency distribution of the sample subject string.
-int BoyerMooreLookahead::FindBestInterval(
- int max_number_of_chars, int old_biggest_points, int* from, int* to) {
- int biggest_points = old_biggest_points;
- static const int kSize = RegExpMacroAssembler::kTableSize;
- for (int i = 0; i < length_; ) {
- while (i < length_ && Count(i) > max_number_of_chars) i++;
- if (i == length_) break;
- int remembered_from = i;
- bool union_map[kSize];
- for (int j = 0; j < kSize; j++) union_map[j] = false;
- while (i < length_ && Count(i) <= max_number_of_chars) {
- BoyerMoorePositionInfo* map = bitmaps_->at(i);
- for (int j = 0; j < kSize; j++) union_map[j] |= map->at(j);
- i++;
- }
- int frequency = 0;
- for (int j = 0; j < kSize; j++) {
- if (union_map[j]) {
- // Add 1 to the frequency to give a small per-character boost for
- // the cases where our sampling is not good enough and many
- // characters have a frequency of zero. This means the frequency
- // can theoretically be up to 2*kSize though we treat it mostly as
- // a fraction of kSize.
- frequency += compiler_->frequency_collator()->Frequency(j) + 1;
- }
- }
- // We use the probability of skipping times the distance we are skipping to
- // judge the effectiveness of this. Actually we have a cut-off: By
- // dividing by 2 we switch off the skipping if the probability of skipping
- // is less than 50%. This is because the multibyte mask-and-compare
- // skipping in quickcheck is more likely to do well on this case.
- bool in_quickcheck_range =
- ((i - remembered_from < 4) ||
- (compiler_->one_byte() ? remembered_from <= 4 : remembered_from <= 2));
- // Called 'probability' but it is only a rough estimate and can actually
- // be outside the 0-kSize range.
- int probability = (in_quickcheck_range ? kSize / 2 : kSize) - frequency;
- int points = (i - remembered_from) * probability;
- if (points > biggest_points) {
- *from = remembered_from;
- *to = i - 1;
- biggest_points = points;
- }
- }
- return biggest_points;
-}
-
-
-// Take all the characters that will not prevent a successful match if they
-// occur in the subject string in the range between min_lookahead and
-// max_lookahead (inclusive) measured from the current position. If the
-// character at max_lookahead offset is not one of these characters, then we
-// can safely skip forwards by the number of characters in the range.
-int BoyerMooreLookahead::GetSkipTable(int min_lookahead,
- int max_lookahead,
- Handle<ByteArray> boolean_skip_table) {
- const int kSize = RegExpMacroAssembler::kTableSize;
-
- const int kSkipArrayEntry = 0;
- const int kDontSkipArrayEntry = 1;
-
- for (int i = 0; i < kSize; i++) {
- boolean_skip_table->set(i, kSkipArrayEntry);
- }
- int skip = max_lookahead + 1 - min_lookahead;
-
- for (int i = max_lookahead; i >= min_lookahead; i--) {
- BoyerMoorePositionInfo* map = bitmaps_->at(i);
- for (int j = 0; j < kSize; j++) {
- if (map->at(j)) {
- boolean_skip_table->set(j, kDontSkipArrayEntry);
- }
- }
- }
-
- return skip;
-}
-
-
-// See comment above on the implementation of GetSkipTable.
-void BoyerMooreLookahead::EmitSkipInstructions(RegExpMacroAssembler* masm) {
- const int kSize = RegExpMacroAssembler::kTableSize;
-
- int min_lookahead = 0;
- int max_lookahead = 0;
-
- if (!FindWorthwhileInterval(&min_lookahead, &max_lookahead)) return;
-
- bool found_single_character = false;
- int single_character = 0;
- for (int i = max_lookahead; i >= min_lookahead; i--) {
- BoyerMoorePositionInfo* map = bitmaps_->at(i);
- if (map->map_count() > 1 ||
- (found_single_character && map->map_count() != 0)) {
- found_single_character = false;
- break;
- }
- for (int j = 0; j < kSize; j++) {
- if (map->at(j)) {
- found_single_character = true;
- single_character = j;
- break;
- }
- }
- }
-
- int lookahead_width = max_lookahead + 1 - min_lookahead;
-
- if (found_single_character && lookahead_width == 1 && max_lookahead < 3) {
- // The mask-compare can probably handle this better.
- return;
- }
-
- if (found_single_character) {
- Label cont, again;
- masm->Bind(&again);
- masm->LoadCurrentCharacter(max_lookahead, &cont, true);
- if (max_char_ > kSize) {
- masm->CheckCharacterAfterAnd(single_character,
- RegExpMacroAssembler::kTableMask,
- &cont);
- } else {
- masm->CheckCharacter(single_character, &cont);
- }
- masm->AdvanceCurrentPosition(lookahead_width);
- masm->GoTo(&again);
- masm->Bind(&cont);
- return;
- }
-
- Factory* factory = masm->isolate()->factory();
- Handle<ByteArray> boolean_skip_table =
- factory->NewByteArray(kSize, AllocationType::kOld);
- int skip_distance = GetSkipTable(
- min_lookahead, max_lookahead, boolean_skip_table);
- DCHECK_NE(0, skip_distance);
-
- Label cont, again;
- masm->Bind(&again);
- masm->LoadCurrentCharacter(max_lookahead, &cont, true);
- masm->CheckBitInTable(boolean_skip_table, &cont);
- masm->AdvanceCurrentPosition(skip_distance);
- masm->GoTo(&again);
- masm->Bind(&cont);
-}
-
-
-/* Code generation for choice nodes.
- *
- * We generate quick checks that do a mask and compare to eliminate a
- * choice. If the quick check succeeds then it jumps to the continuation to
- * do slow checks and check subsequent nodes. If it fails (the common case)
- * it falls through to the next choice.
- *
- * Here is the desired flow graph. Nodes directly below each other imply
- * fallthrough. Alternatives 1 and 2 have quick checks. Alternative
- * 3 doesn't have a quick check so we have to call the slow check.
- * Nodes are marked Qn for quick checks and Sn for slow checks. The entire
- * regexp continuation is generated directly after the Sn node, up to the
- * next GoTo if we decide to reuse some already generated code. Some
- * nodes expect preload_characters to be preloaded into the current
- * character register. R nodes do this preloading. Vertices are marked
- * F for failures and S for success (possible success in the case of quick
- * nodes). L, V, < and > are used as arrow heads.
- *
- * ----------> R
- * |
- * V
- * Q1 -----> S1
- * | S /
- * F| /
- * | F/
- * | /
- * | R
- * | /
- * V L
- * Q2 -----> S2
- * | S /
- * F| /
- * | F/
- * | /
- * | R
- * | /
- * V L
- * S3
- * |
- * F|
- * |
- * R
- * |
- * backtrack V
- * <----------Q4
- * \ F |
- * \ |S
- * \ F V
- * \-----S4
- *
- * For greedy loops we push the current position, then generate the code that
- * eats the input specially in EmitGreedyLoop. The other choice (the
- * continuation) is generated by the normal code in EmitChoices, and steps back
- * in the input to the starting position when it fails to match. The loop code
- * looks like this (U is the unwind code that steps back in the greedy loop).
- *
- * _____
- * / \
- * V |
- * ----------> S1 |
- * /| |
- * / |S |
- * F/ \_____/
- * /
- * |<-----
- * | \
- * V |S
- * Q2 ---> U----->backtrack
- * | F /
- * S| /
- * V F /
- * S2--/
- */
-
-GreedyLoopState::GreedyLoopState(bool not_at_start) {
- counter_backtrack_trace_.set_backtrack(&label_);
- if (not_at_start) counter_backtrack_trace_.set_at_start(Trace::FALSE_VALUE);
-}
-
-
-void ChoiceNode::AssertGuardsMentionRegisters(Trace* trace) {
-#ifdef DEBUG
- int choice_count = alternatives_->length();
- for (int i = 0; i < choice_count - 1; i++) {
- GuardedAlternative alternative = alternatives_->at(i);
- ZoneList<Guard*>* guards = alternative.guards();
- int guard_count = (guards == nullptr) ? 0 : guards->length();
- for (int j = 0; j < guard_count; j++) {
- DCHECK(!trace->mentions_reg(guards->at(j)->reg()));
- }
- }
-#endif
-}
-
-
-void ChoiceNode::SetUpPreLoad(RegExpCompiler* compiler,
- Trace* current_trace,
- PreloadState* state) {
- if (state->eats_at_least_ == PreloadState::kEatsAtLeastNotYetInitialized) {
- // Save some time by looking at most one machine word ahead.
- state->eats_at_least_ =
- EatsAtLeast(compiler->one_byte() ? 4 : 2, kRecursionBudget,
- current_trace->at_start() == Trace::FALSE_VALUE);
- }
- state->preload_characters_ =
- CalculatePreloadCharacters(compiler, state->eats_at_least_);
-
- state->preload_is_current_ =
- (current_trace->characters_preloaded() == state->preload_characters_);
- state->preload_has_checked_bounds_ = state->preload_is_current_;
-}
-
-
-void ChoiceNode::Emit(RegExpCompiler* compiler, Trace* trace) {
- int choice_count = alternatives_->length();
-
- if (choice_count == 1 && alternatives_->at(0).guards() == nullptr) {
- alternatives_->at(0).node()->Emit(compiler, trace);
- return;
- }
-
- AssertGuardsMentionRegisters(trace);
-
- LimitResult limit_result = LimitVersions(compiler, trace);
- if (limit_result == DONE) return;
- DCHECK(limit_result == CONTINUE);
-
- // For loop nodes we already flushed (see LoopChoiceNode::Emit), but for
- // other choice nodes we only flush if we are out of code size budget.
- if (trace->flush_budget() == 0 && trace->actions() != nullptr) {
- trace->Flush(compiler, this);
- return;
- }
-
- RecursionCheck rc(compiler);
-
- PreloadState preload;
- preload.init();
- GreedyLoopState greedy_loop_state(not_at_start());
-
- int text_length = GreedyLoopTextLengthForAlternative(&alternatives_->at(0));
- AlternativeGenerationList alt_gens(choice_count, zone());
-
- if (choice_count > 1 && text_length != kNodeIsTooComplexForGreedyLoops) {
- trace = EmitGreedyLoop(compiler,
- trace,
- &alt_gens,
- &preload,
- &greedy_loop_state,
- text_length);
- } else {
- // TODO(erikcorry): Delete this. We don't need this label, but it makes us
- // match the traces produced pre-cleanup.
- Label second_choice;
- compiler->macro_assembler()->Bind(&second_choice);
-
- preload.eats_at_least_ = EmitOptimizedUnanchoredSearch(compiler, trace);
-
- EmitChoices(compiler,
- &alt_gens,
- 0,
- trace,
- &preload);
- }
-
- // At this point we need to generate slow checks for the alternatives where
- // the quick check was inlined. We can recognize these because the associated
- // label was bound.
- int new_flush_budget = trace->flush_budget() / choice_count;
- for (int i = 0; i < choice_count; i++) {
- AlternativeGeneration* alt_gen = alt_gens.at(i);
- Trace new_trace(*trace);
- // If there are actions to be flushed we have to limit how many times
- // they are flushed. Take the budget of the parent trace and distribute
- // it fairly amongst the children.
- if (new_trace.actions() != nullptr) {
- new_trace.set_flush_budget(new_flush_budget);
- }
- bool next_expects_preload =
- i == choice_count - 1 ? false : alt_gens.at(i + 1)->expects_preload;
- EmitOutOfLineContinuation(compiler,
- &new_trace,
- alternatives_->at(i),
- alt_gen,
- preload.preload_characters_,
- next_expects_preload);
- }
-}
-
-
-Trace* ChoiceNode::EmitGreedyLoop(RegExpCompiler* compiler,
- Trace* trace,
- AlternativeGenerationList* alt_gens,
- PreloadState* preload,
- GreedyLoopState* greedy_loop_state,
- int text_length) {
- RegExpMacroAssembler* macro_assembler = compiler->macro_assembler();
- // Here we have special handling for greedy loops containing only text nodes
- // and other simple nodes. These are handled by pushing the current
- // position on the stack and then incrementing the current position each
- // time around the switch. On backtrack we decrement the current position
- // and check it against the pushed value. This avoids pushing backtrack
- // information for each iteration of the loop, which could take up a lot of
- // space.
- DCHECK(trace->stop_node() == nullptr);
- macro_assembler->PushCurrentPosition();
- Label greedy_match_failed;
- Trace greedy_match_trace;
- if (not_at_start()) greedy_match_trace.set_at_start(Trace::FALSE_VALUE);
- greedy_match_trace.set_backtrack(&greedy_match_failed);
- Label loop_label;
- macro_assembler->Bind(&loop_label);
- greedy_match_trace.set_stop_node(this);
- greedy_match_trace.set_loop_label(&loop_label);
- alternatives_->at(0).node()->Emit(compiler, &greedy_match_trace);
- macro_assembler->Bind(&greedy_match_failed);
-
- Label second_choice; // For use in greedy matches.
- macro_assembler->Bind(&second_choice);
-
- Trace* new_trace = greedy_loop_state->counter_backtrack_trace();
-
- EmitChoices(compiler,
- alt_gens,
- 1,
- new_trace,
- preload);
-
- macro_assembler->Bind(greedy_loop_state->label());
- // If we have unwound to the bottom then backtrack.
- macro_assembler->CheckGreedyLoop(trace->backtrack());
- // Otherwise try the second priority at an earlier position.
- macro_assembler->AdvanceCurrentPosition(-text_length);
- macro_assembler->GoTo(&second_choice);
- return new_trace;
-}
-
-int ChoiceNode::EmitOptimizedUnanchoredSearch(RegExpCompiler* compiler,
- Trace* trace) {
- int eats_at_least = PreloadState::kEatsAtLeastNotYetInitialized;
- if (alternatives_->length() != 2) return eats_at_least;
-
- GuardedAlternative alt1 = alternatives_->at(1);
- if (alt1.guards() != nullptr && alt1.guards()->length() != 0) {
- return eats_at_least;
- }
- RegExpNode* eats_anything_node = alt1.node();
- if (eats_anything_node->GetSuccessorOfOmnivorousTextNode(compiler) != this) {
- return eats_at_least;
- }
-
- // Really we should be creating a new trace when we execute this function,
- // but there is no need, because the code it generates cannot backtrack, and
- // we always arrive here with a trivial trace (since it's the entry to a
- // loop. That also implies that there are no preloaded characters, which is
- // good, because it means we won't be violating any assumptions by
- // overwriting those characters with new load instructions.
- DCHECK(trace->is_trivial());
-
- RegExpMacroAssembler* macro_assembler = compiler->macro_assembler();
- Isolate* isolate = macro_assembler->isolate();
- // At this point we know that we are at a non-greedy loop that will eat
- // any character one at a time. Any non-anchored regexp has such a
- // loop prepended to it in order to find where it starts. We look for
- // a pattern of the form ...abc... where we can look 6 characters ahead
- // and step forwards 3 if the character is not one of abc. Abc need
- // not be atoms, they can be any reasonably limited character class or
- // small alternation.
- BoyerMooreLookahead* bm = bm_info(false);
- if (bm == nullptr) {
- eats_at_least = Min(kMaxLookaheadForBoyerMoore,
- EatsAtLeast(kMaxLookaheadForBoyerMoore,
- kRecursionBudget,
- false));
- if (eats_at_least >= 1) {
- bm = new(zone()) BoyerMooreLookahead(eats_at_least,
- compiler,
- zone());
- GuardedAlternative alt0 = alternatives_->at(0);
- alt0.node()->FillInBMInfo(isolate, 0, kRecursionBudget, bm, false);
- }
- }
- if (bm != nullptr) {
- bm->EmitSkipInstructions(macro_assembler);
- }
- return eats_at_least;
-}
-
-
-void ChoiceNode::EmitChoices(RegExpCompiler* compiler,
- AlternativeGenerationList* alt_gens,
- int first_choice,
- Trace* trace,
- PreloadState* preload) {
- RegExpMacroAssembler* macro_assembler = compiler->macro_assembler();
- SetUpPreLoad(compiler, trace, preload);
-
- // For now we just call all choices one after the other. The idea ultimately
- // is to use the Dispatch table to try only the relevant ones.
- int choice_count = alternatives_->length();
-
- int new_flush_budget = trace->flush_budget() / choice_count;
-
- for (int i = first_choice; i < choice_count; i++) {
- bool is_last = i == choice_count - 1;
- bool fall_through_on_failure = !is_last;
- GuardedAlternative alternative = alternatives_->at(i);
- AlternativeGeneration* alt_gen = alt_gens->at(i);
- alt_gen->quick_check_details.set_characters(preload->preload_characters_);
- ZoneList<Guard*>* guards = alternative.guards();
- int guard_count = (guards == nullptr) ? 0 : guards->length();
- Trace new_trace(*trace);
- new_trace.set_characters_preloaded(preload->preload_is_current_ ?
- preload->preload_characters_ :
- 0);
- if (preload->preload_has_checked_bounds_) {
- new_trace.set_bound_checked_up_to(preload->preload_characters_);
- }
- new_trace.quick_check_performed()->Clear();
- if (not_at_start_) new_trace.set_at_start(Trace::FALSE_VALUE);
- if (!is_last) {
- new_trace.set_backtrack(&alt_gen->after);
- }
- alt_gen->expects_preload = preload->preload_is_current_;
- bool generate_full_check_inline = false;
- if (compiler->optimize() &&
- try_to_emit_quick_check_for_alternative(i == 0) &&
- alternative.node()->EmitQuickCheck(
- compiler, trace, &new_trace, preload->preload_has_checked_bounds_,
- &alt_gen->possible_success, &alt_gen->quick_check_details,
- fall_through_on_failure)) {
- // Quick check was generated for this choice.
- preload->preload_is_current_ = true;
- preload->preload_has_checked_bounds_ = true;
- // If we generated the quick check to fall through on possible success,
- // we now need to generate the full check inline.
- if (!fall_through_on_failure) {
- macro_assembler->Bind(&alt_gen->possible_success);
- new_trace.set_quick_check_performed(&alt_gen->quick_check_details);
- new_trace.set_characters_preloaded(preload->preload_characters_);
- new_trace.set_bound_checked_up_to(preload->preload_characters_);
- generate_full_check_inline = true;
- }
- } else if (alt_gen->quick_check_details.cannot_match()) {
- if (!fall_through_on_failure) {
- macro_assembler->GoTo(trace->backtrack());
- }
- continue;
- } else {
- // No quick check was generated. Put the full code here.
- // If this is not the first choice then there could be slow checks from
- // previous cases that go here when they fail. There's no reason to
- // insist that they preload characters since the slow check we are about
- // to generate probably can't use it.
- if (i != first_choice) {
- alt_gen->expects_preload = false;
- new_trace.InvalidateCurrentCharacter();
- }
- generate_full_check_inline = true;
- }
- if (generate_full_check_inline) {
- if (new_trace.actions() != nullptr) {
- new_trace.set_flush_budget(new_flush_budget);
- }
- for (int j = 0; j < guard_count; j++) {
- GenerateGuard(macro_assembler, guards->at(j), &new_trace);
- }
- alternative.node()->Emit(compiler, &new_trace);
- preload->preload_is_current_ = false;
- }
- macro_assembler->Bind(&alt_gen->after);
- }
-}
-
-
-void ChoiceNode::EmitOutOfLineContinuation(RegExpCompiler* compiler,
- Trace* trace,
- GuardedAlternative alternative,
- AlternativeGeneration* alt_gen,
- int preload_characters,
- bool next_expects_preload) {
- if (!alt_gen->possible_success.is_linked()) return;
-
- RegExpMacroAssembler* macro_assembler = compiler->macro_assembler();
- macro_assembler->Bind(&alt_gen->possible_success);
- Trace out_of_line_trace(*trace);
- out_of_line_trace.set_characters_preloaded(preload_characters);
- out_of_line_trace.set_quick_check_performed(&alt_gen->quick_check_details);
- if (not_at_start_) out_of_line_trace.set_at_start(Trace::FALSE_VALUE);
- ZoneList<Guard*>* guards = alternative.guards();
- int guard_count = (guards == nullptr) ? 0 : guards->length();
- if (next_expects_preload) {
- Label reload_current_char;
- out_of_line_trace.set_backtrack(&reload_current_char);
- for (int j = 0; j < guard_count; j++) {
- GenerateGuard(macro_assembler, guards->at(j), &out_of_line_trace);
- }
- alternative.node()->Emit(compiler, &out_of_line_trace);
- macro_assembler->Bind(&reload_current_char);
- // Reload the current character, since the next quick check expects that.
- // We don't need to check bounds here because we only get into this
- // code through a quick check which already did the checked load.
- macro_assembler->LoadCurrentCharacter(trace->cp_offset(), nullptr, false,
- preload_characters);
- macro_assembler->GoTo(&(alt_gen->after));
- } else {
- out_of_line_trace.set_backtrack(&(alt_gen->after));
- for (int j = 0; j < guard_count; j++) {
- GenerateGuard(macro_assembler, guards->at(j), &out_of_line_trace);
- }
- alternative.node()->Emit(compiler, &out_of_line_trace);
- }
-}
-
-
-void ActionNode::Emit(RegExpCompiler* compiler, Trace* trace) {
- RegExpMacroAssembler* assembler = compiler->macro_assembler();
- LimitResult limit_result = LimitVersions(compiler, trace);
- if (limit_result == DONE) return;
- DCHECK(limit_result == CONTINUE);
-
- RecursionCheck rc(compiler);
-
- switch (action_type_) {
- case STORE_POSITION: {
- Trace::DeferredCapture
- new_capture(data_.u_position_register.reg,
- data_.u_position_register.is_capture,
- trace);
- Trace new_trace = *trace;
- new_trace.add_action(&new_capture);
- on_success()->Emit(compiler, &new_trace);
- break;
- }
- case INCREMENT_REGISTER: {
- Trace::DeferredIncrementRegister
- new_increment(data_.u_increment_register.reg);
- Trace new_trace = *trace;
- new_trace.add_action(&new_increment);
- on_success()->Emit(compiler, &new_trace);
- break;
- }
- case SET_REGISTER: {
- Trace::DeferredSetRegister
- new_set(data_.u_store_register.reg, data_.u_store_register.value);
- Trace new_trace = *trace;
- new_trace.add_action(&new_set);
- on_success()->Emit(compiler, &new_trace);
- break;
- }
- case CLEAR_CAPTURES: {
- Trace::DeferredClearCaptures
- new_capture(Interval(data_.u_clear_captures.range_from,
- data_.u_clear_captures.range_to));
- Trace new_trace = *trace;
- new_trace.add_action(&new_capture);
- on_success()->Emit(compiler, &new_trace);
- break;
- }
- case BEGIN_SUBMATCH:
- if (!trace->is_trivial()) {
- trace->Flush(compiler, this);
- } else {
- assembler->WriteCurrentPositionToRegister(
- data_.u_submatch.current_position_register, 0);
- assembler->WriteStackPointerToRegister(
- data_.u_submatch.stack_pointer_register);
- on_success()->Emit(compiler, trace);
- }
- break;
- case EMPTY_MATCH_CHECK: {
- int start_pos_reg = data_.u_empty_match_check.start_register;
- int stored_pos = 0;
- int rep_reg = data_.u_empty_match_check.repetition_register;
- bool has_minimum = (rep_reg != RegExpCompiler::kNoRegister);
- bool know_dist = trace->GetStoredPosition(start_pos_reg, &stored_pos);
- if (know_dist && !has_minimum && stored_pos == trace->cp_offset()) {
- // If we know we haven't advanced and there is no minimum we
- // can just backtrack immediately.
- assembler->GoTo(trace->backtrack());
- } else if (know_dist && stored_pos < trace->cp_offset()) {
- // If we know we've advanced we can generate the continuation
- // immediately.
- on_success()->Emit(compiler, trace);
- } else if (!trace->is_trivial()) {
- trace->Flush(compiler, this);
- } else {
- Label skip_empty_check;
- // If we have a minimum number of repetitions we check the current
- // number first and skip the empty check if it's not enough.
- if (has_minimum) {
- int limit = data_.u_empty_match_check.repetition_limit;
- assembler->IfRegisterLT(rep_reg, limit, &skip_empty_check);
- }
- // If the match is empty we bail out, otherwise we fall through
- // to the on-success continuation.
- assembler->IfRegisterEqPos(data_.u_empty_match_check.start_register,
- trace->backtrack());
- assembler->Bind(&skip_empty_check);
- on_success()->Emit(compiler, trace);
- }
- break;
- }
- case POSITIVE_SUBMATCH_SUCCESS: {
- if (!trace->is_trivial()) {
- trace->Flush(compiler, this);
- return;
- }
- assembler->ReadCurrentPositionFromRegister(
- data_.u_submatch.current_position_register);
- assembler->ReadStackPointerFromRegister(
- data_.u_submatch.stack_pointer_register);
- int clear_register_count = data_.u_submatch.clear_register_count;
- if (clear_register_count == 0) {
- on_success()->Emit(compiler, trace);
- return;
- }
- int clear_registers_from = data_.u_submatch.clear_register_from;
- Label clear_registers_backtrack;
- Trace new_trace = *trace;
- new_trace.set_backtrack(&clear_registers_backtrack);
- on_success()->Emit(compiler, &new_trace);
-
- assembler->Bind(&clear_registers_backtrack);
- int clear_registers_to = clear_registers_from + clear_register_count - 1;
- assembler->ClearRegisters(clear_registers_from, clear_registers_to);
-
- DCHECK(trace->backtrack() == nullptr);
- assembler->Backtrack();
- return;
- }
- default:
- UNREACHABLE();
- }
-}
-
-
-void BackReferenceNode::Emit(RegExpCompiler* compiler, Trace* trace) {
- RegExpMacroAssembler* assembler = compiler->macro_assembler();
- if (!trace->is_trivial()) {
- trace->Flush(compiler, this);
- return;
- }
-
- LimitResult limit_result = LimitVersions(compiler, trace);
- if (limit_result == DONE) return;
- DCHECK(limit_result == CONTINUE);
-
- RecursionCheck rc(compiler);
-
- DCHECK_EQ(start_reg_ + 1, end_reg_);
- if (IgnoreCase(flags_)) {
- assembler->CheckNotBackReferenceIgnoreCase(
- start_reg_, read_backward(), IsUnicode(flags_), 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 (IsUnicode(flags_) && !compiler->one_byte()) {
- assembler->CheckNotInSurrogatePair(trace->cp_offset(), trace->backtrack());
- }
- on_success()->Emit(compiler, trace);
-}
-
-
-// -------------------------------------------------------------------
-// Dot/dotty output
-
-
-#ifdef DEBUG
-
-
-class DotPrinter: public NodeVisitor {
- public:
- DotPrinter(std::ostream& os, bool ignore_case) // NOLINT
- : os_(os),
- ignore_case_(ignore_case) {}
- void PrintNode(const char* label, RegExpNode* node);
- void Visit(RegExpNode* node);
- void PrintAttributes(RegExpNode* from);
- void PrintOnFailure(RegExpNode* from, RegExpNode* to);
-#define DECLARE_VISIT(Type) \
- virtual void Visit##Type(Type##Node* that);
-FOR_EACH_NODE_TYPE(DECLARE_VISIT)
-#undef DECLARE_VISIT
- private:
- std::ostream& os_;
- bool ignore_case_;
-};
-
-
-void DotPrinter::PrintNode(const char* label, RegExpNode* node) {
- os_ << "digraph G {\n graph [label=\"";
- for (int i = 0; label[i]; i++) {
- switch (label[i]) {
- case '\\':
- os_ << "\\\\";
- break;
- case '"':
- os_ << "\"";
- break;
- default:
- os_ << label[i];
- break;
- }
- }
- os_ << "\"];\n";
- Visit(node);
- os_ << "}" << std::endl;
-}
-
-
-void DotPrinter::Visit(RegExpNode* node) {
- if (node->info()->visited) return;
- node->info()->visited = true;
- node->Accept(this);
-}
-
-
-void DotPrinter::PrintOnFailure(RegExpNode* from, RegExpNode* on_failure) {
- os_ << " n" << from << " -> n" << on_failure << " [style=dotted];\n";
- Visit(on_failure);
-}
-
-
-class TableEntryBodyPrinter {
- public:
- TableEntryBodyPrinter(std::ostream& os, ChoiceNode* choice) // NOLINT
- : os_(os),
- choice_(choice) {}
- void Call(uc16 from, DispatchTable::Entry entry) {
- OutSet* out_set = entry.out_set();
- for (unsigned i = 0; i < OutSet::kFirstLimit; i++) {
- if (out_set->Get(i)) {
- os_ << " n" << choice() << ":s" << from << "o" << i << " -> n"
- << choice()->alternatives()->at(i).node() << ";\n";
- }
- }
- }
- private:
- ChoiceNode* choice() { return choice_; }
- std::ostream& os_;
- ChoiceNode* choice_;
-};
-
-
-class TableEntryHeaderPrinter {
- public:
- explicit TableEntryHeaderPrinter(std::ostream& os) // NOLINT
- : first_(true),
- os_(os) {}
- void Call(uc16 from, DispatchTable::Entry entry) {
- if (first_) {
- first_ = false;
- } else {
- os_ << "|";
- }
- os_ << "{\\" << AsUC16(from) << "-\\" << AsUC16(entry.to()) << "|{";
- OutSet* out_set = entry.out_set();
- int priority = 0;
- for (unsigned i = 0; i < OutSet::kFirstLimit; i++) {
- if (out_set->Get(i)) {
- if (priority > 0) os_ << "|";
- os_ << "<s" << from << "o" << i << "> " << priority;
- priority++;
- }
- }
- os_ << "}}";
- }
-
- private:
- bool first_;
- std::ostream& os_;
-};
-
-
-class AttributePrinter {
- public:
- explicit AttributePrinter(std::ostream& os) // NOLINT
- : os_(os),
- first_(true) {}
- void PrintSeparator() {
- if (first_) {
- first_ = false;
- } else {
- os_ << "|";
- }
- }
- void PrintBit(const char* name, bool value) {
- if (!value) return;
- PrintSeparator();
- os_ << "{" << name << "}";
- }
- void PrintPositive(const char* name, int value) {
- if (value < 0) return;
- PrintSeparator();
- os_ << "{" << name << "|" << value << "}";
- }
-
- private:
- std::ostream& os_;
- bool first_;
-};
-
-
-void DotPrinter::PrintAttributes(RegExpNode* that) {
- os_ << " a" << that << " [shape=Mrecord, color=grey, fontcolor=grey, "
- << "margin=0.1, fontsize=10, label=\"{";
- AttributePrinter printer(os_);
- NodeInfo* info = that->info();
- printer.PrintBit("NI", info->follows_newline_interest);
- printer.PrintBit("WI", info->follows_word_interest);
- printer.PrintBit("SI", info->follows_start_interest);
- Label* label = that->label();
- if (label->is_bound())
- printer.PrintPositive("@", label->pos());
- os_ << "}\"];\n"
- << " a" << that << " -> n" << that
- << " [style=dashed, color=grey, arrowhead=none];\n";
-}
-
-
-static const bool kPrintDispatchTable = false;
-void DotPrinter::VisitChoice(ChoiceNode* that) {
- if (kPrintDispatchTable) {
- os_ << " n" << that << " [shape=Mrecord, label=\"";
- TableEntryHeaderPrinter header_printer(os_);
- that->GetTable(ignore_case_)->ForEach(&header_printer);
- os_ << "\"]\n";
- PrintAttributes(that);
- TableEntryBodyPrinter body_printer(os_, that);
- that->GetTable(ignore_case_)->ForEach(&body_printer);
- } else {
- os_ << " n" << that << " [shape=Mrecord, label=\"?\"];\n";
- for (int i = 0; i < that->alternatives()->length(); i++) {
- GuardedAlternative alt = that->alternatives()->at(i);
- os_ << " n" << that << " -> n" << alt.node();
- }
- }
- for (int i = 0; i < that->alternatives()->length(); i++) {
- GuardedAlternative alt = that->alternatives()->at(i);
- alt.node()->Accept(this);
- }
-}
-
-
-void DotPrinter::VisitText(TextNode* that) {
- Zone* zone = that->zone();
- os_ << " n" << that << " [label=\"";
- for (int i = 0; i < that->elements()->length(); i++) {
- if (i > 0) os_ << " ";
- TextElement elm = that->elements()->at(i);
- switch (elm.text_type()) {
- case TextElement::ATOM: {
- Vector<const uc16> data = elm.atom()->data();
- for (int i = 0; i < data.length(); i++) {
- os_ << static_cast<char>(data[i]);
- }
- break;
- }
- case TextElement::CHAR_CLASS: {
- RegExpCharacterClass* node = elm.char_class();
- os_ << "[";
- if (node->is_negated()) os_ << "^";
- for (int j = 0; j < node->ranges(zone)->length(); j++) {
- CharacterRange range = node->ranges(zone)->at(j);
- os_ << AsUC16(range.from()) << "-" << AsUC16(range.to());
- }
- os_ << "]";
- break;
- }
- default:
- UNREACHABLE();
- }
- }
- os_ << "\", shape=box, peripheries=2];\n";
- PrintAttributes(that);
- os_ << " n" << that << " -> n" << that->on_success() << ";\n";
- Visit(that->on_success());
-}
-
-
-void DotPrinter::VisitBackReference(BackReferenceNode* that) {
- os_ << " n" << that << " [label=\"$" << that->start_register() << "..$"
- << that->end_register() << "\", shape=doubleoctagon];\n";
- PrintAttributes(that);
- os_ << " n" << that << " -> n" << that->on_success() << ";\n";
- Visit(that->on_success());
-}
-
-
-void DotPrinter::VisitEnd(EndNode* that) {
- os_ << " n" << that << " [style=bold, shape=point];\n";
- PrintAttributes(that);
-}
-
-
-void DotPrinter::VisitAssertion(AssertionNode* that) {
- os_ << " n" << that << " [";
- switch (that->assertion_type()) {
- case AssertionNode::AT_END:
- os_ << "label=\"$\", shape=septagon";
- break;
- case AssertionNode::AT_START:
- os_ << "label=\"^\", shape=septagon";
- break;
- case AssertionNode::AT_BOUNDARY:
- os_ << "label=\"\\b\", shape=septagon";
- break;
- case AssertionNode::AT_NON_BOUNDARY:
- os_ << "label=\"\\B\", shape=septagon";
- break;
- case AssertionNode::AFTER_NEWLINE:
- os_ << "label=\"(?<=\\n)\", shape=septagon";
- break;
- }
- os_ << "];\n";
- PrintAttributes(that);
- RegExpNode* successor = that->on_success();
- os_ << " n" << that << " -> n" << successor << ";\n";
- Visit(successor);
-}
-
-
-void DotPrinter::VisitAction(ActionNode* that) {
- os_ << " n" << that << " [";
- switch (that->action_type_) {
- case ActionNode::SET_REGISTER:
- os_ << "label=\"$" << that->data_.u_store_register.reg
- << ":=" << that->data_.u_store_register.value << "\", shape=octagon";
- break;
- case ActionNode::INCREMENT_REGISTER:
- os_ << "label=\"$" << that->data_.u_increment_register.reg
- << "++\", shape=octagon";
- break;
- case ActionNode::STORE_POSITION:
- os_ << "label=\"$" << that->data_.u_position_register.reg
- << ":=$pos\", shape=octagon";
- break;
- case ActionNode::BEGIN_SUBMATCH:
- os_ << "label=\"$" << that->data_.u_submatch.current_position_register
- << ":=$pos,begin\", shape=septagon";
- break;
- case ActionNode::POSITIVE_SUBMATCH_SUCCESS:
- os_ << "label=\"escape\", shape=septagon";
- break;
- case ActionNode::EMPTY_MATCH_CHECK:
- os_ << "label=\"$" << that->data_.u_empty_match_check.start_register
- << "=$pos?,$" << that->data_.u_empty_match_check.repetition_register
- << "<" << that->data_.u_empty_match_check.repetition_limit
- << "?\", shape=septagon";
- break;
- case ActionNode::CLEAR_CAPTURES: {
- os_ << "label=\"clear $" << that->data_.u_clear_captures.range_from
- << " to $" << that->data_.u_clear_captures.range_to
- << "\", shape=septagon";
- break;
- }
- }
- os_ << "];\n";
- PrintAttributes(that);
- RegExpNode* successor = that->on_success();
- os_ << " n" << that << " -> n" << successor << ";\n";
- Visit(successor);
-}
-
-
-class DispatchTableDumper {
- public:
- explicit DispatchTableDumper(std::ostream& os) : os_(os) {}
- void Call(uc16 key, DispatchTable::Entry entry);
- private:
- std::ostream& os_;
-};
-
-
-void DispatchTableDumper::Call(uc16 key, DispatchTable::Entry entry) {
- os_ << "[" << AsUC16(key) << "-" << AsUC16(entry.to()) << "]: {";
- OutSet* set = entry.out_set();
- bool first = true;
- for (unsigned i = 0; i < OutSet::kFirstLimit; i++) {
- if (set->Get(i)) {
- if (first) {
- first = false;
- } else {
- os_ << ", ";
- }
- os_ << i;
- }
- }
- os_ << "}\n";
-}
-
-
-void DispatchTable::Dump() {
- OFStream os(stderr);
- DispatchTableDumper dumper(os);
- tree()->ForEach(&dumper);
-}
-
-
-void RegExpEngine::DotPrint(const char* label,
- RegExpNode* node,
- bool ignore_case) {
- StdoutStream os;
- DotPrinter printer(os, ignore_case);
- printer.PrintNode(label, node);
-}
-
-
-#endif // DEBUG
-
-
-// -------------------------------------------------------------------
-// Tree to graph conversion
-
-RegExpNode* RegExpAtom::ToNode(RegExpCompiler* compiler,
- RegExpNode* on_success) {
- ZoneList<TextElement>* elms =
- new(compiler->zone()) ZoneList<TextElement>(1, compiler->zone());
- elms->Add(TextElement::Atom(this), compiler->zone());
- return new (compiler->zone())
- TextNode(elms, compiler->read_backward(), on_success);
-}
-
-
-RegExpNode* RegExpText::ToNode(RegExpCompiler* compiler,
- RegExpNode* on_success) {
- return new (compiler->zone())
- TextNode(elements(), compiler->read_backward(), on_success);
-}
-
-
-static bool CompareInverseRanges(ZoneList<CharacterRange>* ranges,
- const int* special_class,
- int length) {
- length--; // Remove final marker.
- DCHECK_EQ(kRangeEndMarker, special_class[length]);
- DCHECK_NE(0, ranges->length());
- DCHECK_NE(0, length);
- DCHECK_NE(0, special_class[0]);
- if (ranges->length() != (length >> 1) + 1) {
- return false;
- }
- CharacterRange range = ranges->at(0);
- if (range.from() != 0) {
- return false;
- }
- for (int i = 0; i < length; i += 2) {
- if (special_class[i] != (range.to() + 1)) {
- return false;
- }
- range = ranges->at((i >> 1) + 1);
- if (special_class[i+1] != range.from()) {
- return false;
- }
- }
- if (range.to() != String::kMaxCodePoint) {
- return false;
- }
- return true;
-}
-
-
-static bool CompareRanges(ZoneList<CharacterRange>* ranges,
- const int* special_class,
- int length) {
- length--; // Remove final marker.
- DCHECK_EQ(kRangeEndMarker, special_class[length]);
- if (ranges->length() * 2 != length) {
- return false;
- }
- for (int i = 0; i < length; i += 2) {
- CharacterRange range = ranges->at(i >> 1);
- if (range.from() != special_class[i] ||
- range.to() != special_class[i + 1] - 1) {
- return false;
- }
- }
- return true;
-}
-
-
-bool RegExpCharacterClass::is_standard(Zone* zone) {
- // TODO(lrn): Remove need for this function, by not throwing away information
- // along the way.
- if (is_negated()) {
- return false;
- }
- if (set_.is_standard()) {
- return true;
- }
- if (CompareRanges(set_.ranges(zone), kSpaceRanges, kSpaceRangeCount)) {
- set_.set_standard_set_type('s');
- return true;
- }
- if (CompareInverseRanges(set_.ranges(zone), kSpaceRanges, kSpaceRangeCount)) {
- set_.set_standard_set_type('S');
- return true;
- }
- if (CompareInverseRanges(set_.ranges(zone),
- kLineTerminatorRanges,
- kLineTerminatorRangeCount)) {
- set_.set_standard_set_type('.');
- return true;
- }
- if (CompareRanges(set_.ranges(zone),
- kLineTerminatorRanges,
- kLineTerminatorRangeCount)) {
- set_.set_standard_set_type('n');
- return true;
- }
- if (CompareRanges(set_.ranges(zone), kWordRanges, kWordRangeCount)) {
- set_.set_standard_set_type('w');
- return true;
- }
- if (CompareInverseRanges(set_.ranges(zone), kWordRanges, kWordRangeCount)) {
- set_.set_standard_set_type('W');
- return true;
- }
- return false;
-}
-
-
-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 = nullptr;
- 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 == nullptr)
- *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;
- JSRegExp::Flags default_flags = JSRegExp::Flags();
- result->AddAlternative(GuardedAlternative(TextNode::CreateForCharacterRanges(
- compiler->zone(), bmp, compiler->read_backward(), on_success,
- default_flags)));
-}
-
-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->one_byte());
- Zone* zone = compiler->zone();
- JSRegExp::Flags default_flags = JSRegExp::Flags();
- 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, default_flags)));
- } 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, default_flags)));
- 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, default_flags)));
- 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, default_flags)));
- }
- }
- }
-}
-
-RegExpNode* NegativeLookaroundAgainstReadDirectionAndMatch(
- RegExpCompiler* compiler, ZoneList<CharacterRange>* lookbehind,
- ZoneList<CharacterRange>* match, RegExpNode* on_success, bool read_backward,
- JSRegExp::Flags flags) {
- Zone* zone = compiler->zone();
- RegExpNode* match_node = TextNode::CreateForCharacterRanges(
- zone, match, read_backward, on_success, flags);
- 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(), flags);
- return lookaround.ForMatch(negative_match);
-}
-
-RegExpNode* MatchAndNegativeLookaroundInReadDirection(
- RegExpCompiler* compiler, ZoneList<CharacterRange>* match,
- ZoneList<CharacterRange>* lookahead, RegExpNode* on_success,
- bool read_backward, JSRegExp::Flags flags) {
- 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(), flags);
- return TextNode::CreateForCharacterRanges(
- zone, match, read_backward, lookaround.ForMatch(negative_match), flags);
-}
-
-void AddLoneLeadSurrogates(RegExpCompiler* compiler, ChoiceNode* result,
- RegExpNode* on_success,
- UnicodeRangeSplitter* splitter) {
- JSRegExp::Flags default_flags = JSRegExp::Flags();
- 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,
- default_flags);
- } 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,
- default_flags);
- }
- result->AddAlternative(GuardedAlternative(match));
-}
-
-void AddLoneTrailSurrogates(RegExpCompiler* compiler, ChoiceNode* result,
- RegExpNode* on_success,
- UnicodeRangeSplitter* splitter) {
- JSRegExp::Flags default_flags = JSRegExp::Flags();
- 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,
- default_flags);
- } 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,
- default_flags);
- }
- 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));
- JSRegExp::Flags default_flags = JSRegExp::Flags();
- return TextNode::CreateForCharacterRanges(zone, range, false, on_success,
- default_flags);
-}
-
-void AddUnicodeCaseEquivalents(ZoneList<CharacterRange>* ranges, Zone* zone) {
-#ifdef V8_INTL_SUPPORT
- DCHECK(CharacterRange::IsCanonical(ranges));
-
- // Micro-optimization to avoid passing large ranges to UnicodeSet::closeOver.
- // See also https://crbug.com/v8/6727.
- // TODO(jgruber): This only covers the special case of the {0,0x10FFFF} range,
- // which we use frequently internally. But large ranges can also easily be
- // created by the user. We might want to have a more general caching mechanism
- // for such ranges.
- if (ranges->length() == 1 && ranges->at(0).IsEverything(kNonBmpEnd)) return;
-
- // Use ICU to compute the case fold closure over the ranges.
- icu::UnicodeSet set;
- for (int i = 0; i < ranges->length(); i++) {
- set.add(ranges->at(i).from(), ranges->at(i).to());
- }
- ranges->Clear();
- set.closeOver(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.
- set.removeAllStrings();
- for (int i = 0; i < set.getRangeCount(); i++) {
- ranges->Add(CharacterRange::Range(set.getRangeStart(i), set.getRangeEnd(i)),
- zone);
- }
- // No errors and everything we collected have been ranges.
- CharacterRange::Canonicalize(ranges);
-#endif // V8_INTL_SUPPORT
-}
-
-
-RegExpNode* RegExpCharacterClass::ToNode(RegExpCompiler* compiler,
- RegExpNode* on_success) {
- set_.Canonicalize();
- Zone* zone = compiler->zone();
- ZoneList<CharacterRange>* ranges = this->ranges(zone);
- if (NeedsUnicodeCaseEquivalents(flags_)) {
- AddUnicodeCaseEquivalents(ranges, zone);
- }
- if (IsUnicode(flags_) && !compiler->one_byte() &&
- !contains_split_surrogate()) {
- if (is_negated()) {
- ZoneList<CharacterRange>* negated =
- new (zone) ZoneList<CharacterRange>(2, zone);
- CharacterRange::Negate(ranges, negated, zone);
- ranges = negated;
- }
- if (ranges->length() == 0) {
- JSRegExp::Flags default_flags;
- RegExpCharacterClass* fail =
- new (zone) RegExpCharacterClass(zone, ranges, default_flags);
- return new (zone) TextNode(fail, compiler->read_backward(), on_success);
- }
- 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);
- }
-}
-
-
-int CompareFirstChar(RegExpTree* const* a, RegExpTree* const* b) {
- RegExpAtom* atom1 = (*a)->AsAtom();
- RegExpAtom* atom2 = (*b)->AsAtom();
- uc16 character1 = atom1->data().at(0);
- uc16 character2 = atom2->data().at(0);
- if (character1 < character2) return -1;
- if (character1 > character2) return 1;
- return 0;
-}
-
-#ifdef V8_INTL_SUPPORT
-
-// Case Insensitve comparesion
-int CompareFirstCharCaseInsensitve(RegExpTree* const* a, RegExpTree* const* b) {
- RegExpAtom* atom1 = (*a)->AsAtom();
- RegExpAtom* atom2 = (*b)->AsAtom();
- icu::UnicodeString character1(atom1->data().at(0));
- return character1.caseCompare(atom2->data().at(0), U_FOLD_CASE_DEFAULT);
-}
-
-#else
-
-static unibrow::uchar Canonical(
- unibrow::Mapping<unibrow::Ecma262Canonicalize>* canonicalize,
- unibrow::uchar c) {
- unibrow::uchar chars[unibrow::Ecma262Canonicalize::kMaxWidth];
- int length = canonicalize->get(c, '\0', chars);
- DCHECK_LE(length, 1);
- unibrow::uchar canonical = c;
- if (length == 1) canonical = chars[0];
- return canonical;
-}
-
-int CompareFirstCharCaseIndependent(
- unibrow::Mapping<unibrow::Ecma262Canonicalize>* canonicalize,
- RegExpTree* const* a, RegExpTree* const* b) {
- RegExpAtom* atom1 = (*a)->AsAtom();
- RegExpAtom* atom2 = (*b)->AsAtom();
- unibrow::uchar character1 = atom1->data().at(0);
- unibrow::uchar character2 = atom2->data().at(0);
- if (character1 == character2) return 0;
- if (character1 >= 'a' || character2 >= 'a') {
- character1 = Canonical(canonicalize, character1);
- character2 = Canonical(canonicalize, character2);
- }
- return static_cast<int>(character1) - static_cast<int>(character2);
-}
-#endif // V8_INTL_SUPPORT
-
-// We can stable sort runs of atoms, since the order does not matter if they
-// start with different characters.
-// Returns true if any consecutive atoms were found.
-bool RegExpDisjunction::SortConsecutiveAtoms(RegExpCompiler* compiler) {
- ZoneList<RegExpTree*>* alternatives = this->alternatives();
- int length = alternatives->length();
- bool found_consecutive_atoms = false;
- for (int i = 0; i < length; i++) {
- while (i < length) {
- RegExpTree* alternative = alternatives->at(i);
- if (alternative->IsAtom()) break;
- i++;
- }
- // i is length or it is the index of an atom.
- if (i == length) break;
- int first_atom = i;
- JSRegExp::Flags flags = alternatives->at(i)->AsAtom()->flags();
- i++;
- while (i < length) {
- RegExpTree* alternative = alternatives->at(i);
- if (!alternative->IsAtom()) break;
- if (alternative->AsAtom()->flags() != flags) break;
- i++;
- }
- // Sort atoms to get ones with common prefixes together.
- // This step is more tricky if we are in a case-independent regexp,
- // because it would change /is|I/ to /I|is/, and order matters when
- // the regexp parts don't match only disjoint starting points. To fix
- // this we have a version of CompareFirstChar that uses case-
- // independent character classes for comparison.
- DCHECK_LT(first_atom, alternatives->length());
- DCHECK_LE(i, alternatives->length());
- DCHECK_LE(first_atom, i);
- if (IgnoreCase(flags)) {
-#ifdef V8_INTL_SUPPORT
- alternatives->StableSort(CompareFirstCharCaseInsensitve, first_atom,
- i - first_atom);
-#else
- unibrow::Mapping<unibrow::Ecma262Canonicalize>* canonicalize =
- compiler->isolate()->regexp_macro_assembler_canonicalize();
- auto compare_closure =
- [canonicalize](RegExpTree* const* a, RegExpTree* const* b) {
- return CompareFirstCharCaseIndependent(canonicalize, a, b);
- };
- alternatives->StableSort(compare_closure, first_atom, i - first_atom);
-#endif // V8_INTL_SUPPORT
- } else {
- alternatives->StableSort(CompareFirstChar, first_atom, i - first_atom);
- }
- if (i - first_atom > 1) found_consecutive_atoms = true;
- }
- return found_consecutive_atoms;
-}
-
-
-// Optimizes ab|ac|az to a(?:b|c|d).
-void RegExpDisjunction::RationalizeConsecutiveAtoms(RegExpCompiler* compiler) {
- Zone* zone = compiler->zone();
- ZoneList<RegExpTree*>* alternatives = this->alternatives();
- int length = alternatives->length();
-
- int write_posn = 0;
- int i = 0;
- while (i < length) {
- RegExpTree* alternative = alternatives->at(i);
- if (!alternative->IsAtom()) {
- alternatives->at(write_posn++) = alternatives->at(i);
- i++;
- continue;
- }
- RegExpAtom* const atom = alternative->AsAtom();
- JSRegExp::Flags flags = atom->flags();
-#ifdef V8_INTL_SUPPORT
- icu::UnicodeString common_prefix(atom->data().at(0));
-#else
- unibrow::uchar common_prefix = atom->data().at(0);
-#endif // V8_INTL_SUPPORT
- int first_with_prefix = i;
- int prefix_length = atom->length();
- i++;
- while (i < length) {
- alternative = alternatives->at(i);
- if (!alternative->IsAtom()) break;
- RegExpAtom* const atom = alternative->AsAtom();
- if (atom->flags() != flags) break;
-#ifdef V8_INTL_SUPPORT
- icu::UnicodeString new_prefix(atom->data().at(0));
- if (new_prefix != common_prefix) {
- if (!IgnoreCase(flags)) break;
- if (common_prefix.caseCompare(new_prefix, U_FOLD_CASE_DEFAULT) != 0)
- break;
- }
-#else
- unibrow::uchar new_prefix = atom->data().at(0);
- if (new_prefix != common_prefix) {
- if (!IgnoreCase(flags)) break;
- unibrow::Mapping<unibrow::Ecma262Canonicalize>* canonicalize =
- compiler->isolate()->regexp_macro_assembler_canonicalize();
- new_prefix = Canonical(canonicalize, new_prefix);
- common_prefix = Canonical(canonicalize, common_prefix);
- if (new_prefix != common_prefix) break;
- }
-#endif // V8_INTL_SUPPORT
- prefix_length = Min(prefix_length, atom->length());
- i++;
- }
- if (i > first_with_prefix + 2) {
- // Found worthwhile run of alternatives with common prefix of at least one
- // character. The sorting function above did not sort on more than one
- // character for reasons of correctness, but there may still be a longer
- // common prefix if the terms were similar or presorted in the input.
- // Find out how long the common prefix is.
- int run_length = i - first_with_prefix;
- RegExpAtom* const atom = alternatives->at(first_with_prefix)->AsAtom();
- for (int j = 1; j < run_length && prefix_length > 1; j++) {
- RegExpAtom* old_atom =
- alternatives->at(j + first_with_prefix)->AsAtom();
- for (int k = 1; k < prefix_length; k++) {
- if (atom->data().at(k) != old_atom->data().at(k)) {
- prefix_length = k;
- break;
- }
- }
- }
- RegExpAtom* prefix = new (zone)
- RegExpAtom(atom->data().SubVector(0, prefix_length), flags);
- ZoneList<RegExpTree*>* pair = new (zone) ZoneList<RegExpTree*>(2, zone);
- pair->Add(prefix, zone);
- ZoneList<RegExpTree*>* suffixes =
- new (zone) ZoneList<RegExpTree*>(run_length, zone);
- for (int j = 0; j < run_length; j++) {
- RegExpAtom* old_atom =
- alternatives->at(j + first_with_prefix)->AsAtom();
- int len = old_atom->length();
- if (len == prefix_length) {
- suffixes->Add(new (zone) RegExpEmpty(), zone);
- } else {
- RegExpTree* suffix = new (zone) RegExpAtom(
- old_atom->data().SubVector(prefix_length, old_atom->length()),
- flags);
- suffixes->Add(suffix, zone);
- }
- }
- pair->Add(new (zone) RegExpDisjunction(suffixes), zone);
- alternatives->at(write_posn++) = new (zone) RegExpAlternative(pair);
- } else {
- // Just copy any non-worthwhile alternatives.
- for (int j = first_with_prefix; j < i; j++) {
- alternatives->at(write_posn++) = alternatives->at(j);
- }
- }
- }
- alternatives->Rewind(write_posn); // Trim end of array.
-}
-
-
-// Optimizes b|c|z to [bcz].
-void RegExpDisjunction::FixSingleCharacterDisjunctions(
- RegExpCompiler* compiler) {
- Zone* zone = compiler->zone();
- ZoneList<RegExpTree*>* alternatives = this->alternatives();
- int length = alternatives->length();
-
- int write_posn = 0;
- int i = 0;
- while (i < length) {
- RegExpTree* alternative = alternatives->at(i);
- if (!alternative->IsAtom()) {
- alternatives->at(write_posn++) = alternatives->at(i);
- i++;
- continue;
- }
- RegExpAtom* const atom = alternative->AsAtom();
- if (atom->length() != 1) {
- alternatives->at(write_posn++) = alternatives->at(i);
- i++;
- continue;
- }
- JSRegExp::Flags flags = atom->flags();
- DCHECK_IMPLIES(IsUnicode(flags),
- !unibrow::Utf16::IsLeadSurrogate(atom->data().at(0)));
- bool contains_trail_surrogate =
- unibrow::Utf16::IsTrailSurrogate(atom->data().at(0));
- int first_in_run = i;
- i++;
- // Find a run of single-character atom alternatives that have identical
- // flags (case independence and unicode-ness).
- while (i < length) {
- alternative = alternatives->at(i);
- if (!alternative->IsAtom()) break;
- RegExpAtom* const atom = alternative->AsAtom();
- if (atom->length() != 1) break;
- if (atom->flags() != flags) break;
- DCHECK_IMPLIES(IsUnicode(flags),
- !unibrow::Utf16::IsLeadSurrogate(atom->data().at(0)));
- contains_trail_surrogate |=
- unibrow::Utf16::IsTrailSurrogate(atom->data().at(0));
- i++;
- }
- if (i > first_in_run + 1) {
- // Found non-trivial run of single-character alternatives.
- int run_length = i - first_in_run;
- ZoneList<CharacterRange>* ranges =
- new (zone) ZoneList<CharacterRange>(2, zone);
- for (int j = 0; j < run_length; j++) {
- RegExpAtom* old_atom = alternatives->at(j + first_in_run)->AsAtom();
- DCHECK_EQ(old_atom->length(), 1);
- ranges->Add(CharacterRange::Singleton(old_atom->data().at(0)), zone);
- }
- RegExpCharacterClass::CharacterClassFlags character_class_flags;
- if (IsUnicode(flags) && contains_trail_surrogate) {
- character_class_flags = RegExpCharacterClass::CONTAINS_SPLIT_SURROGATE;
- }
- alternatives->at(write_posn++) = new (zone)
- RegExpCharacterClass(zone, ranges, flags, character_class_flags);
- } else {
- // Just copy any trivial alternatives.
- for (int j = first_in_run; j < i; j++) {
- alternatives->at(write_posn++) = alternatives->at(j);
- }
- }
- }
- alternatives->Rewind(write_posn); // Trim end of array.
-}
-
-
-RegExpNode* RegExpDisjunction::ToNode(RegExpCompiler* compiler,
- RegExpNode* on_success) {
- ZoneList<RegExpTree*>* alternatives = this->alternatives();
-
- if (alternatives->length() > 2) {
- bool found_consecutive_atoms = SortConsecutiveAtoms(compiler);
- if (found_consecutive_atoms) RationalizeConsecutiveAtoms(compiler);
- FixSingleCharacterDisjunctions(compiler);
- if (alternatives->length() == 1) {
- return alternatives->at(0)->ToNode(compiler, on_success);
- }
- }
-
- int length = alternatives->length();
-
- ChoiceNode* result =
- new(compiler->zone()) ChoiceNode(length, compiler->zone());
- for (int i = 0; i < length; i++) {
- GuardedAlternative alternative(alternatives->at(i)->ToNode(compiler,
- on_success));
- result->AddAlternative(alternative);
- }
- return result;
-}
-
-
-RegExpNode* RegExpQuantifier::ToNode(RegExpCompiler* compiler,
- RegExpNode* on_success) {
- return ToNode(min(),
- max(),
- is_greedy(),
- body(),
- compiler,
- on_success);
-}
-
-
-// Scoped object to keep track of how much we unroll quantifier loops in the
-// regexp graph generator.
-class RegExpExpansionLimiter {
- public:
- static const int kMaxExpansionFactor = 6;
- RegExpExpansionLimiter(RegExpCompiler* compiler, int factor)
- : compiler_(compiler),
- saved_expansion_factor_(compiler->current_expansion_factor()),
- ok_to_expand_(saved_expansion_factor_ <= kMaxExpansionFactor) {
- DCHECK_LT(0, factor);
- if (ok_to_expand_) {
- if (factor > kMaxExpansionFactor) {
- // Avoid integer overflow of the current expansion factor.
- ok_to_expand_ = false;
- compiler->set_current_expansion_factor(kMaxExpansionFactor + 1);
- } else {
- int new_factor = saved_expansion_factor_ * factor;
- ok_to_expand_ = (new_factor <= kMaxExpansionFactor);
- compiler->set_current_expansion_factor(new_factor);
- }
- }
- }
-
- ~RegExpExpansionLimiter() {
- compiler_->set_current_expansion_factor(saved_expansion_factor_);
- }
-
- bool ok_to_expand() { return ok_to_expand_; }
-
- private:
- RegExpCompiler* compiler_;
- int saved_expansion_factor_;
- bool ok_to_expand_;
-
- DISALLOW_IMPLICIT_CONSTRUCTORS(RegExpExpansionLimiter);
-};
-
-
-RegExpNode* RegExpQuantifier::ToNode(int min,
- int max,
- bool is_greedy,
- RegExpTree* body,
- RegExpCompiler* compiler,
- RegExpNode* on_success,
- bool not_at_start) {
- // x{f, t} becomes this:
- //
- // (r++)<-.
- // | `
- // | (x)
- // v ^
- // (r=0)-->(?)---/ [if r < t]
- // |
- // [if r >= f] \----> ...
- //
-
- // 15.10.2.5 RepeatMatcher algorithm.
- // The parser has already eliminated the case where max is 0. In the case
- // where max_match is zero the parser has removed the quantifier if min was
- // > 0 and removed the atom if min was 0. See AddQuantifierToAtom.
-
- // If we know that we cannot match zero length then things are a little
- // simpler since we don't need to make the special zero length match check
- // from step 2.1. If the min and max are small we can unroll a little in
- // this case.
- static const int kMaxUnrolledMinMatches = 3; // Unroll (foo)+ and (foo){3,}
- static const int kMaxUnrolledMaxMatches = 3; // Unroll (foo)? and (foo){x,3}
- if (max == 0) return on_success; // This can happen due to recursion.
- bool body_can_be_empty = (body->min_match() == 0);
- int body_start_reg = RegExpCompiler::kNoRegister;
- Interval capture_registers = body->CaptureRegisters();
- bool needs_capture_clearing = !capture_registers.is_empty();
- Zone* zone = compiler->zone();
-
- if (body_can_be_empty) {
- body_start_reg = compiler->AllocateRegister();
- } else if (compiler->optimize() && !needs_capture_clearing) {
- // Only unroll if there are no captures and the body can't be
- // empty.
- {
- RegExpExpansionLimiter limiter(
- compiler, min + ((max != min) ? 1 : 0));
- if (min > 0 && min <= kMaxUnrolledMinMatches && limiter.ok_to_expand()) {
- int new_max = (max == kInfinity) ? max : max - min;
- // Recurse once to get the loop or optional matches after the fixed
- // ones.
- RegExpNode* answer = ToNode(
- 0, new_max, is_greedy, body, compiler, on_success, true);
- // Unroll the forced matches from 0 to min. This can cause chains of
- // TextNodes (which the parser does not generate). These should be
- // combined if it turns out they hinder good code generation.
- for (int i = 0; i < min; i++) {
- answer = body->ToNode(compiler, answer);
- }
- return answer;
- }
- }
- if (max <= kMaxUnrolledMaxMatches && min == 0) {
- DCHECK_LT(0, max); // Due to the 'if' above.
- RegExpExpansionLimiter limiter(compiler, max);
- if (limiter.ok_to_expand()) {
- // Unroll the optional matches up to max.
- RegExpNode* answer = on_success;
- for (int i = 0; i < max; i++) {
- ChoiceNode* alternation = new(zone) ChoiceNode(2, zone);
- if (is_greedy) {
- alternation->AddAlternative(
- GuardedAlternative(body->ToNode(compiler, answer)));
- alternation->AddAlternative(GuardedAlternative(on_success));
- } else {
- alternation->AddAlternative(GuardedAlternative(on_success));
- alternation->AddAlternative(
- GuardedAlternative(body->ToNode(compiler, answer)));
- }
- answer = alternation;
- if (not_at_start && !compiler->read_backward()) {
- alternation->set_not_at_start();
- }
- }
- return answer;
- }
- }
- }
- bool has_min = min > 0;
- bool has_max = max < RegExpTree::kInfinity;
- bool needs_counter = has_min || has_max;
- int reg_ctr = needs_counter
- ? compiler->AllocateRegister()
- : RegExpCompiler::kNoRegister;
- LoopChoiceNode* center = new (zone)
- LoopChoiceNode(body->min_match() == 0, compiler->read_backward(), zone);
- if (not_at_start && !compiler->read_backward()) center->set_not_at_start();
- RegExpNode* loop_return = needs_counter
- ? static_cast<RegExpNode*>(ActionNode::IncrementRegister(reg_ctr, center))
- : static_cast<RegExpNode*>(center);
- if (body_can_be_empty) {
- // If the body can be empty we need to check if it was and then
- // backtrack.
- loop_return = ActionNode::EmptyMatchCheck(body_start_reg,
- reg_ctr,
- min,
- loop_return);
- }
- RegExpNode* body_node = body->ToNode(compiler, loop_return);
- if (body_can_be_empty) {
- // If the body can be empty we need to store the start position
- // so we can bail out if it was empty.
- body_node = ActionNode::StorePosition(body_start_reg, false, body_node);
- }
- if (needs_capture_clearing) {
- // Before entering the body of this loop we need to clear captures.
- body_node = ActionNode::ClearCaptures(capture_registers, body_node);
- }
- GuardedAlternative body_alt(body_node);
- if (has_max) {
- Guard* body_guard =
- new(zone) Guard(reg_ctr, Guard::LT, max);
- body_alt.AddGuard(body_guard, zone);
- }
- GuardedAlternative rest_alt(on_success);
- if (has_min) {
- Guard* rest_guard = new(compiler->zone()) Guard(reg_ctr, Guard::GEQ, min);
- rest_alt.AddGuard(rest_guard, zone);
- }
- if (is_greedy) {
- center->AddLoopAlternative(body_alt);
- center->AddContinueAlternative(rest_alt);
- } else {
- center->AddContinueAlternative(rest_alt);
- center->AddLoopAlternative(body_alt);
- }
- if (needs_counter) {
- return ActionNode::SetRegister(reg_ctr, 0, center);
- } else {
- return center;
- }
-}
-
-namespace {
-// Desugar \b to (?<=\w)(?=\W)|(?<=\W)(?=\w) and
-// \B to (?<=\w)(?=\w)|(?<=\W)(?=\W)
-RegExpNode* BoundaryAssertionAsLookaround(RegExpCompiler* compiler,
- RegExpNode* on_success,
- RegExpAssertion::AssertionType type,
- JSRegExp::Flags flags) {
- DCHECK(NeedsUnicodeCaseEquivalents(flags));
- Zone* zone = compiler->zone();
- ZoneList<CharacterRange>* word_range =
- new (zone) ZoneList<CharacterRange>(2, zone);
- CharacterRange::AddClassEscape('w', word_range, true, zone);
- int stack_register = compiler->UnicodeLookaroundStackRegister();
- int position_register = compiler->UnicodeLookaroundPositionRegister();
- ChoiceNode* result = new (zone) ChoiceNode(2, zone);
- // Add two choices. The (non-)boundary could start with a word or
- // a non-word-character.
- for (int i = 0; i < 2; i++) {
- bool lookbehind_for_word = i == 0;
- bool lookahead_for_word =
- (type == RegExpAssertion::BOUNDARY) ^ lookbehind_for_word;
- // Look to the left.
- RegExpLookaround::Builder lookbehind(lookbehind_for_word, on_success,
- stack_register, position_register);
- RegExpNode* backward = TextNode::CreateForCharacterRanges(
- zone, word_range, true, lookbehind.on_match_success(), flags);
- // Look to the right.
- RegExpLookaround::Builder lookahead(lookahead_for_word,
- lookbehind.ForMatch(backward),
- stack_register, position_register);
- RegExpNode* forward = TextNode::CreateForCharacterRanges(
- zone, word_range, false, lookahead.on_match_success(), flags);
- result->AddAlternative(GuardedAlternative(lookahead.ForMatch(forward)));
- }
- return result;
-}
-} // anonymous namespace
-
-RegExpNode* RegExpAssertion::ToNode(RegExpCompiler* compiler,
- RegExpNode* on_success) {
- NodeInfo info;
- Zone* zone = compiler->zone();
-
- switch (assertion_type()) {
- case START_OF_LINE:
- return AssertionNode::AfterNewline(on_success);
- case START_OF_INPUT:
- return AssertionNode::AtStart(on_success);
- case BOUNDARY:
- return NeedsUnicodeCaseEquivalents(flags_)
- ? BoundaryAssertionAsLookaround(compiler, on_success, BOUNDARY,
- flags_)
- : AssertionNode::AtBoundary(on_success);
- case NON_BOUNDARY:
- return NeedsUnicodeCaseEquivalents(flags_)
- ? BoundaryAssertionAsLookaround(compiler, on_success,
- NON_BOUNDARY, flags_)
- : AssertionNode::AtNonBoundary(on_success);
- case END_OF_INPUT:
- return AssertionNode::AtEnd(on_success);
- case END_OF_LINE: {
- // Compile $ in multiline regexps as an alternation with a positive
- // lookahead in one side and an end-of-input on the other side.
- // We need two registers for the lookahead.
- int stack_pointer_register = compiler->AllocateRegister();
- int position_register = compiler->AllocateRegister();
- // The ChoiceNode to distinguish between a newline and end-of-input.
- ChoiceNode* result = new(zone) ChoiceNode(2, zone);
- // Create a newline atom.
- ZoneList<CharacterRange>* newline_ranges =
- new(zone) ZoneList<CharacterRange>(3, zone);
- CharacterRange::AddClassEscape('n', newline_ranges, false, zone);
- JSRegExp::Flags default_flags = JSRegExp::Flags();
- RegExpCharacterClass* newline_atom =
- new (zone) RegExpCharacterClass('n', default_flags);
- TextNode* newline_matcher = new (zone) TextNode(
- newline_atom, false, ActionNode::PositiveSubmatchSuccess(
- stack_pointer_register, position_register,
- 0, // No captures inside.
- -1, // Ignored if no captures.
- on_success));
- // Create an end-of-input matcher.
- RegExpNode* end_of_line = ActionNode::BeginSubmatch(
- stack_pointer_register,
- position_register,
- newline_matcher);
- // Add the two alternatives to the ChoiceNode.
- GuardedAlternative eol_alternative(end_of_line);
- result->AddAlternative(eol_alternative);
- GuardedAlternative end_alternative(AssertionNode::AtEnd(on_success));
- result->AddAlternative(end_alternative);
- return result;
- }
- default:
- UNREACHABLE();
- }
- return on_success;
-}
-
-
-RegExpNode* RegExpBackReference::ToNode(RegExpCompiler* compiler,
- RegExpNode* on_success) {
- return new (compiler->zone())
- BackReferenceNode(RegExpCapture::StartRegister(index()),
- RegExpCapture::EndRegister(index()), flags_,
- compiler->read_backward(), on_success);
-}
-
-
-RegExpNode* RegExpEmpty::ToNode(RegExpCompiler* compiler,
- RegExpNode* on_success) {
- return on_success;
-}
-
-
-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();
- int position_register = compiler->AllocateRegister();
-
- const int registers_per_capture = 2;
- const int register_of_first_capture = 2;
- int register_count = capture_count_ * registers_per_capture;
- int register_start =
- register_of_first_capture + capture_from_ * registers_per_capture;
-
- RegExpNode* result;
- bool was_reading_backward = compiler->read_backward();
- compiler->set_read_backward(type() == LOOKBEHIND);
- 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;
-}
-
-
-RegExpNode* RegExpCapture::ToNode(RegExpCompiler* compiler,
- RegExpNode* on_success) {
- return ToNode(body(), index(), compiler, on_success);
-}
-
-
-RegExpNode* RegExpCapture::ToNode(RegExpTree* body,
- int index,
- RegExpCompiler* compiler,
- RegExpNode* on_success) {
- DCHECK_NOT_NULL(body);
- int start_reg = RegExpCapture::StartRegister(index);
- int end_reg = RegExpCapture::EndRegister(index);
- if (compiler->read_backward()) std::swap(start_reg, end_reg);
- RegExpNode* store_end = ActionNode::StorePosition(end_reg, true, on_success);
- RegExpNode* body_node = body->ToNode(compiler, store_end);
- return ActionNode::StorePosition(start_reg, true, body_node);
-}
-
-
-RegExpNode* RegExpAlternative::ToNode(RegExpCompiler* compiler,
- RegExpNode* on_success) {
- ZoneList<RegExpTree*>* children = nodes();
- RegExpNode* current = on_success;
- if (compiler->read_backward()) {
- for (int i = 0; i < children->length(); i++) {
- current = children->at(i)->ToNode(compiler, current);
- }
- } else {
- for (int i = children->length() - 1; i >= 0; i--) {
- current = children->at(i)->ToNode(compiler, current);
- }
- }
- return current;
-}
-
-
-static void AddClass(const int* elmv,
- int elmc,
- ZoneList<CharacterRange>* ranges,
- Zone* zone) {
- elmc--;
- DCHECK_EQ(kRangeEndMarker, elmv[elmc]);
- for (int i = 0; i < elmc; i += 2) {
- DCHECK(elmv[i] < elmv[i + 1]);
- ranges->Add(CharacterRange::Range(elmv[i], elmv[i + 1] - 1), zone);
- }
-}
-
-
-static void AddClassNegated(const int *elmv,
- int elmc,
- ZoneList<CharacterRange>* ranges,
- Zone* zone) {
- elmc--;
- DCHECK_EQ(kRangeEndMarker, elmv[elmc]);
- DCHECK_NE(0x0000, elmv[0]);
- DCHECK_NE(String::kMaxCodePoint, elmv[elmc - 1]);
- 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::Range(last, elmv[i] - 1), zone);
- last = elmv[i + 1];
- }
- ranges->Add(CharacterRange::Range(last, String::kMaxCodePoint), zone);
-}
-
-void CharacterRange::AddClassEscape(char type, ZoneList<CharacterRange>* ranges,
- bool add_unicode_case_equivalents,
- Zone* zone) {
- if (add_unicode_case_equivalents && (type == 'w' || type == 'W')) {
- // See #sec-runtime-semantics-wordcharacters-abstract-operation
- // In case of unicode and ignore_case, we need to create the closure over
- // case equivalent characters before negating.
- ZoneList<CharacterRange>* new_ranges =
- new (zone) ZoneList<CharacterRange>(2, zone);
- AddClass(kWordRanges, kWordRangeCount, new_ranges, zone);
- AddUnicodeCaseEquivalents(new_ranges, zone);
- if (type == 'W') {
- ZoneList<CharacterRange>* negated =
- new (zone) ZoneList<CharacterRange>(2, zone);
- CharacterRange::Negate(new_ranges, negated, zone);
- new_ranges = negated;
- }
- ranges->AddAll(*new_ranges, zone);
- return;
- }
- AddClassEscape(type, ranges, zone);
-}
-
-void CharacterRange::AddClassEscape(char type, ZoneList<CharacterRange>* ranges,
- Zone* zone) {
- switch (type) {
- case 's':
- AddClass(kSpaceRanges, kSpaceRangeCount, ranges, zone);
- break;
- case 'S':
- AddClassNegated(kSpaceRanges, kSpaceRangeCount, ranges, zone);
- break;
- case 'w':
- AddClass(kWordRanges, kWordRangeCount, ranges, zone);
- break;
- case 'W':
- AddClassNegated(kWordRanges, kWordRangeCount, ranges, zone);
- break;
- case 'd':
- AddClass(kDigitRanges, kDigitRangeCount, ranges, zone);
- break;
- case 'D':
- AddClassNegated(kDigitRanges, kDigitRangeCount, ranges, zone);
- break;
- case '.':
- AddClassNegated(kLineTerminatorRanges,
- kLineTerminatorRangeCount,
- ranges,
- zone);
- break;
- // This is not a character range as defined by the spec but a
- // convenient shorthand for a character class that matches any
- // character.
- case '*':
- ranges->Add(CharacterRange::Everything(), zone);
- break;
- // This is the set of characters matched by the $ and ^ symbols
- // in multiline mode.
- case 'n':
- AddClass(kLineTerminatorRanges,
- kLineTerminatorRangeCount,
- ranges,
- zone);
- break;
- default:
- UNREACHABLE();
- }
-}
-
-
-Vector<const int> CharacterRange::GetWordBounds() {
- return Vector<const int>(kWordRanges, kWordRangeCount - 1);
-}
-
-#ifdef V8_INTL_SUPPORT
-struct IgnoreSet {
- IgnoreSet() : set(BuildIgnoreSet()) {}
- const icu::UnicodeSet set;
-};
-
-struct SpecialAddSet {
- SpecialAddSet() : set(BuildSpecialAddSet()) {}
- const icu::UnicodeSet set;
-};
-
-icu::UnicodeSet BuildAsciiAToZSet() {
- icu::UnicodeSet set('a', 'z');
- set.add('A', 'Z');
- set.freeze();
- return set;
-}
-
-struct AsciiAToZSet {
- AsciiAToZSet() : set(BuildAsciiAToZSet()) {}
- const icu::UnicodeSet set;
-};
-
-static base::LazyInstance<IgnoreSet>::type ignore_set =
- LAZY_INSTANCE_INITIALIZER;
-
-static base::LazyInstance<SpecialAddSet>::type special_add_set =
- LAZY_INSTANCE_INITIALIZER;
-
-static base::LazyInstance<AsciiAToZSet>::type ascii_a_to_z_set =
- LAZY_INSTANCE_INITIALIZER;
-#endif // V8_INTL_SUPPORT
-
-// static
-void CharacterRange::AddCaseEquivalents(Isolate* isolate, Zone* zone,
- ZoneList<CharacterRange>* ranges,
- bool is_one_byte) {
- CharacterRange::Canonicalize(ranges);
- int range_count = ranges->length();
-#ifdef V8_INTL_SUPPORT
- icu::UnicodeSet others;
- for (int i = 0; i < range_count; i++) {
- CharacterRange range = ranges->at(i);
- uc32 from = range.from();
- if (from > String::kMaxUtf16CodeUnit) continue;
- uc32 to = Min(range.to(), String::kMaxUtf16CodeUnit);
- // Nothing to be done for surrogates.
- if (from >= kLeadSurrogateStart && to <= kTrailSurrogateEnd) continue;
- if (is_one_byte && !RangeContainsLatin1Equivalents(range)) {
- if (from > String::kMaxOneByteCharCode) continue;
- if (to > String::kMaxOneByteCharCode) to = String::kMaxOneByteCharCode;
- }
- others.add(from, to);
- }
-
- // Set of characters already added to ranges that do not need to be added
- // again.
- icu::UnicodeSet already_added(others);
-
- // Set of characters in ranges that are in the 52 ASCII characters [a-zA-Z].
- icu::UnicodeSet in_ascii_a_to_z(others);
- in_ascii_a_to_z.retainAll(ascii_a_to_z_set.Pointer()->set);
-
- // Remove all chars in [a-zA-Z] from others.
- others.removeAll(in_ascii_a_to_z);
-
- // Set of characters in ranges that are overlapping with special add set.
- icu::UnicodeSet in_special_add(others);
- in_special_add.retainAll(special_add_set.Pointer()->set);
-
- others.removeAll(in_special_add);
-
- // Ignore all chars in ignore set.
- others.removeAll(ignore_set.Pointer()->set);
-
- // For most of the chars in ranges that is still in others, find the case
- // equivlant set by calling closeOver(USET_CASE_INSENSITIVE).
- others.closeOver(USET_CASE_INSENSITIVE);
-
- // Because closeOver(USET_CASE_INSENSITIVE) may add ASCII [a-zA-Z] to others,
- // but ECMA262 "i" mode won't consider that, remove them from others.
- // Ex: U+017F add 'S' and 's' to others.
- others.removeAll(ascii_a_to_z_set.Pointer()->set);
-
- // Special handling for in_ascii_a_to_z.
- for (int32_t i = 0; i < in_ascii_a_to_z.getRangeCount(); i++) {
- UChar32 start = in_ascii_a_to_z.getRangeStart(i);
- UChar32 end = in_ascii_a_to_z.getRangeEnd(i);
- // Check if it is uppercase A-Z by checking bit 6.
- if (start & 0x0020) {
- // Add the lowercases
- others.add(start & 0x005F, end & 0x005F);
- } else {
- // Add the uppercases
- others.add(start | 0x0020, end | 0x0020);
- }
- }
-
- // Special handling for chars in "Special Add" set.
- for (int32_t i = 0; i < in_special_add.getRangeCount(); i++) {
- UChar32 end = in_special_add.getRangeEnd(i);
- for (UChar32 ch = in_special_add.getRangeStart(i); ch <= end; ch++) {
- // Add the uppercase of this character if itself is not an uppercase
- // character.
- // Note: The if condiction cannot be u_islower(ch) because ch could be
- // neither uppercase nor lowercase but Mn.
- if (!u_isupper(ch)) {
- others.add(u_toupper(ch));
- }
- icu::UnicodeSet candidates(ch, ch);
- candidates.closeOver(USET_CASE_INSENSITIVE);
- for (int32_t j = 0; j < candidates.getRangeCount(); j++) {
- UChar32 end2 = candidates.getRangeEnd(j);
- for (UChar32 ch2 = candidates.getRangeStart(j); ch2 <= end2; ch2++) {
- // Add character that is not uppercase to others.
- if (!u_isupper(ch2)) {
- others.add(ch2);
- }
- }
- }
- }
- }
-
- // Remove all characters which already in the ranges.
- others.removeAll(already_added);
-
- // Add others to the ranges
- for (int32_t i = 0; i < others.getRangeCount(); i++) {
- UChar32 from = others.getRangeStart(i);
- UChar32 to = others.getRangeEnd(i);
- if (from == to) {
- ranges->Add(CharacterRange::Singleton(from), zone);
- } else {
- ranges->Add(CharacterRange::Range(from, to), zone);
- }
- }
-#else
- for (int i = 0; i < range_count; i++) {
- CharacterRange range = ranges->at(i);
- uc32 bottom = range.from();
- if (bottom > String::kMaxUtf16CodeUnit) continue;
- uc32 top = Min(range.to(), String::kMaxUtf16CodeUnit);
- // Nothing to be done for surrogates.
- if (bottom >= kLeadSurrogateStart && top <= kTrailSurrogateEnd) continue;
- if (is_one_byte && !RangeContainsLatin1Equivalents(range)) {
- if (bottom > String::kMaxOneByteCharCode) continue;
- 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);
- }
- }
- } 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;
- }
- }
- }
-#endif // V8_INTL_SUPPORT
-}
-
-bool CharacterRange::IsCanonical(ZoneList<CharacterRange>* ranges) {
- DCHECK_NOT_NULL(ranges);
- int n = ranges->length();
- if (n <= 1) return true;
- int max = ranges->at(0).to();
- for (int i = 1; i < n; i++) {
- CharacterRange next_range = ranges->at(i);
- if (next_range.from() <= max + 1) return false;
- max = next_range.to();
- }
- return true;
-}
-
-
-ZoneList<CharacterRange>* CharacterSet::ranges(Zone* zone) {
- if (ranges_ == nullptr) {
- ranges_ = new(zone) ZoneList<CharacterRange>(2, zone);
- CharacterRange::AddClassEscape(standard_set_type_, ranges_, false, zone);
- }
- return ranges_;
-}
-
-
-// Move a number of elements in a zonelist to another position
-// in the same list. Handles overlapping source and target areas.
-static void MoveRanges(ZoneList<CharacterRange>* list,
- int from,
- int to,
- int count) {
- // Ranges are potentially overlapping.
- if (from < to) {
- for (int i = count - 1; i >= 0; i--) {
- list->at(to + i) = list->at(from + i);
- }
- } else {
- for (int i = 0; i < count; i++) {
- list->at(to + i) = list->at(from + i);
- }
- }
-}
-
-
-static int InsertRangeInCanonicalList(ZoneList<CharacterRange>* list,
- int count,
- CharacterRange insert) {
- // Inserts a range into list[0..count[, which must be sorted
- // by from value and non-overlapping and non-adjacent, using at most
- // 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.
- uc32 from = insert.from();
- uc32 to = insert.to();
- int start_pos = 0;
- int end_pos = count;
- for (int i = count - 1; i >= 0; i--) {
- CharacterRange current = list->at(i);
- if (current.from() > to + 1) {
- end_pos = i;
- } else if (current.to() + 1 < from) {
- start_pos = i + 1;
- break;
- }
- }
-
- // Inserted range overlaps, or is adjacent to, ranges at positions
- // [start_pos..end_pos[. Ranges before start_pos or at or after end_pos are
- // not affected by the insertion.
- // If start_pos == end_pos, the range must be inserted before start_pos.
- // if start_pos < end_pos, the entire range from start_pos to end_pos
- // must be merged with the insert range.
-
- if (start_pos == end_pos) {
- // Insert between existing ranges at position start_pos.
- if (start_pos < count) {
- MoveRanges(list, start_pos, start_pos + 1, count - start_pos);
- }
- list->at(start_pos) = insert;
- return count + 1;
- }
- if (start_pos + 1 == end_pos) {
- // Replace single existing range at position start_pos.
- 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::Range(new_from, new_to);
- return count;
- }
- // Replace a number of existing ranges from start_pos to end_pos - 1.
- // Move the remaining ranges down.
-
- int new_from = Min(list->at(start_pos).from(), from);
- int new_to = Max(list->at(end_pos - 1).to(), to);
- if (end_pos < count) {
- MoveRanges(list, end_pos, start_pos + 1, count - end_pos);
- }
- list->at(start_pos) = CharacterRange::Range(new_from, new_to);
- return count - (end_pos - start_pos) + 1;
-}
-
-
-void CharacterSet::Canonicalize() {
- // Special/default classes are always considered canonical. The result
- // of calling ranges() will be sorted.
- if (ranges_ == nullptr) return;
- CharacterRange::Canonicalize(ranges_);
-}
-
-
-void CharacterRange::Canonicalize(ZoneList<CharacterRange>* character_ranges) {
- if (character_ranges->length() <= 1) return;
- // Check whether ranges are already canonical (increasing, non-overlapping,
- // non-adjacent).
- int n = character_ranges->length();
- int max = character_ranges->at(0).to();
- int i = 1;
- while (i < n) {
- CharacterRange current = character_ranges->at(i);
- if (current.from() <= max + 1) {
- break;
- }
- max = current.to();
- i++;
- }
- // Canonical until the i'th range. If that's all of them, we are done.
- if (i == n) return;
-
- // The ranges at index i and forward are not canonicalized. Make them so by
- // doing the equivalent of insertion sort (inserting each into the previous
- // list, in order).
- // Notice that inserting a range can reduce the number of ranges in the
- // result due to combining of adjacent and overlapping ranges.
- int read = i; // Range to insert.
- int num_canonical = i; // Length of canonicalized part of list.
- do {
- num_canonical = InsertRangeInCanonicalList(character_ranges,
- num_canonical,
- character_ranges->at(read));
- read++;
- } while (read < n);
- character_ranges->Rewind(num_canonical);
-
- DCHECK(CharacterRange::IsCanonical(character_ranges));
-}
-
-
-void CharacterRange::Negate(ZoneList<CharacterRange>* ranges,
- ZoneList<CharacterRange>* negated_ranges,
- Zone* zone) {
- DCHECK(CharacterRange::IsCanonical(ranges));
- DCHECK_EQ(0, negated_ranges->length());
- int range_count = ranges->length();
- uc32 from = 0;
- int i = 0;
- if (range_count > 0 && ranges->at(0).from() == 0) {
- from = ranges->at(0).to() + 1;
- i = 1;
- }
- while (i < range_count) {
- CharacterRange range = ranges->at(i);
- negated_ranges->Add(CharacterRange::Range(from, range.from() - 1), zone);
- from = range.to() + 1;
- i++;
- }
- if (from < String::kMaxCodePoint) {
- negated_ranges->Add(CharacterRange::Range(from, String::kMaxCodePoint),
- zone);
- }
-}
-
-
-// -------------------------------------------------------------------
-// Splay tree
-
-
-OutSet* OutSet::Extend(unsigned value, Zone* zone) {
- if (Get(value))
- return this;
- if (successors(zone) != nullptr) {
- for (int i = 0; i < successors(zone)->length(); i++) {
- OutSet* successor = successors(zone)->at(i);
- if (successor->Get(value))
- return successor;
- }
- } else {
- successors_ = new(zone) ZoneList<OutSet*>(2, zone);
- }
- OutSet* result = new(zone) OutSet(first_, remaining_);
- result->Set(value, zone);
- successors(zone)->Add(result, zone);
- return result;
-}
-
-
-void OutSet::Set(unsigned value, Zone *zone) {
- if (value < kFirstLimit) {
- first_ |= (1 << value);
- } else {
- if (remaining_ == nullptr)
- remaining_ = new(zone) ZoneList<unsigned>(1, zone);
- if (remaining_->is_empty() || !remaining_->Contains(value))
- remaining_->Add(value, zone);
- }
-}
-
-
-bool OutSet::Get(unsigned value) const {
- if (value < kFirstLimit) {
- return (first_ & (1 << value)) != 0;
- } else if (remaining_ == nullptr) {
- return false;
- } else {
- return remaining_->Contains(value);
- }
-}
-
-
-const uc32 DispatchTable::Config::kNoKey = unibrow::Utf8::kBadChar;
-
-
-void DispatchTable::AddRange(CharacterRange full_range, int value,
- Zone* zone) {
- CharacterRange current = full_range;
- if (tree()->is_empty()) {
- // If this is the first range we just insert into the table.
- ZoneSplayTree<Config>::Locator loc;
- bool inserted = tree()->Insert(current.from(), &loc);
- DCHECK(inserted);
- USE(inserted);
- loc.set_value(Entry(current.from(), current.to(),
- empty()->Extend(value, zone)));
- return;
- }
- // First see if there is a range to the left of this one that
- // overlaps.
- ZoneSplayTree<Config>::Locator loc;
- if (tree()->FindGreatestLessThan(current.from(), &loc)) {
- Entry* entry = &loc.value();
- // If we've found a range that overlaps with this one, and it
- // starts strictly to the left of this one, we have to fix it
- // because the following code only handles ranges that start on
- // or after the start point of the range we're adding.
- 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 =
- 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());
- // The right part is the one that overlaps. We add this part
- // to the map and let the next step deal with merging it with
- // the range we're adding.
- ZoneSplayTree<Config>::Locator loc;
- bool inserted = tree()->Insert(right.from(), &loc);
- DCHECK(inserted);
- USE(inserted);
- loc.set_value(Entry(right.from(),
- right.to(),
- entry->out_set()));
- }
- }
- while (current.is_valid()) {
- if (tree()->FindLeastGreaterThan(current.from(), &loc) &&
- (loc.value().from() <= current.to()) &&
- (loc.value().to() >= current.from())) {
- Entry* entry = &loc.value();
- // We have overlap. If there is space between the start point of
- // the range we're adding and where the overlapping range starts
- // then we have to add a range covering just that space.
- if (current.from() < entry->from()) {
- ZoneSplayTree<Config>::Locator ins;
- bool inserted = tree()->Insert(current.from(), &ins);
- DCHECK(inserted);
- USE(inserted);
- ins.set_value(Entry(current.from(),
- entry->from() - 1,
- empty()->Extend(value, zone)));
- current.set_from(entry->from());
- }
- DCHECK_EQ(current.from(), entry->from());
- // If the overlapping range extends beyond the one we want to add
- // we have to snap the right part off and add it separately.
- if (entry->to() > current.to()) {
- ZoneSplayTree<Config>::Locator ins;
- bool inserted = tree()->Insert(current.to() + 1, &ins);
- DCHECK(inserted);
- USE(inserted);
- ins.set_value(Entry(current.to() + 1,
- entry->to(),
- entry->out_set()));
- entry->set_to(current.to());
- }
- DCHECK(entry->to() <= current.to());
- // The overlapping range is now completely contained by the range
- // 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);
- DCHECK(entry->to() + 1 > current.from());
- current.set_from(entry->to() + 1);
- } else {
- // There is no overlap so we can just add the range
- ZoneSplayTree<Config>::Locator ins;
- bool inserted = tree()->Insert(current.from(), &ins);
- DCHECK(inserted);
- USE(inserted);
- ins.set_value(Entry(current.from(),
- current.to(),
- empty()->Extend(value, zone)));
- break;
- }
- }
-}
-
-
-OutSet* DispatchTable::Get(uc32 value) {
- ZoneSplayTree<Config>::Locator loc;
- if (!tree()->FindGreatestLessThan(value, &loc))
- return empty();
- Entry* entry = &loc.value();
- if (value <= entry->to())
- return entry->out_set();
- else
- return empty();
-}
-
-
-// -------------------------------------------------------------------
-// Analysis
-
-
-void Analysis::EnsureAnalyzed(RegExpNode* that) {
- StackLimitCheck check(isolate());
- if (check.HasOverflowed()) {
- fail("Stack overflow");
- return;
- }
- if (that->info()->been_analyzed || that->info()->being_analyzed)
- return;
- that->info()->being_analyzed = true;
- that->Accept(this);
- that->info()->being_analyzed = false;
- that->info()->been_analyzed = true;
-}
-
-
-void Analysis::VisitEnd(EndNode* that) {
- // nothing to do
-}
-
-
-void TextNode::CalculateOffsets() {
- int element_count = elements()->length();
- // Set up the offsets of the elements relative to the start. This is a fixed
- // quantity since a TextNode can only contain fixed-width things.
- int cp_offset = 0;
- for (int i = 0; i < element_count; i++) {
- TextElement& elm = elements()->at(i);
- elm.set_cp_offset(cp_offset);
- cp_offset += elm.length();
- }
-}
-
-
-void Analysis::VisitText(TextNode* that) {
- that->MakeCaseIndependent(isolate(), is_one_byte_);
- EnsureAnalyzed(that->on_success());
- if (!has_failed()) {
- that->CalculateOffsets();
- }
-}
-
-
-void Analysis::VisitAction(ActionNode* that) {
- RegExpNode* target = that->on_success();
- EnsureAnalyzed(target);
- if (!has_failed()) {
- // If the next node is interested in what it follows then this node
- // has to be interested too so it can pass the information on.
- that->info()->AddFromFollowing(target->info());
- }
-}
-
-
-void Analysis::VisitChoice(ChoiceNode* that) {
- NodeInfo* info = that->info();
- for (int i = 0; i < that->alternatives()->length(); i++) {
- RegExpNode* node = that->alternatives()->at(i).node();
- EnsureAnalyzed(node);
- if (has_failed()) return;
- // Anything the following nodes need to know has to be known by
- // this node also, so it can pass it on.
- info->AddFromFollowing(node->info());
- }
-}
-
-
-void Analysis::VisitLoopChoice(LoopChoiceNode* that) {
- NodeInfo* info = that->info();
- for (int i = 0; i < that->alternatives()->length(); i++) {
- RegExpNode* node = that->alternatives()->at(i).node();
- if (node != that->loop_node()) {
- EnsureAnalyzed(node);
- if (has_failed()) return;
- info->AddFromFollowing(node->info());
- }
- }
- // Check the loop last since it may need the value of this node
- // to get a correct result.
- EnsureAnalyzed(that->loop_node());
- if (!has_failed()) {
- info->AddFromFollowing(that->loop_node()->info());
- }
-}
-
-
-void Analysis::VisitBackReference(BackReferenceNode* that) {
- EnsureAnalyzed(that->on_success());
-}
-
-
-void Analysis::VisitAssertion(AssertionNode* that) {
- EnsureAnalyzed(that->on_success());
-}
-
-
-void BackReferenceNode::FillInBMInfo(Isolate* isolate, int offset, int budget,
- BoyerMooreLookahead* bm,
- bool not_at_start) {
- // Working out the set of characters that a backreference can match is too
- // hard, so we just say that any character can match.
- bm->SetRest(offset);
- SaveBMInfo(bm, not_at_start, offset);
-}
-
-
-STATIC_ASSERT(BoyerMoorePositionInfo::kMapSize ==
- RegExpMacroAssembler::kTableSize);
-
-
-void ChoiceNode::FillInBMInfo(Isolate* isolate, int offset, int budget,
- BoyerMooreLookahead* bm, bool not_at_start) {
- ZoneList<GuardedAlternative>* alts = alternatives();
- budget = (budget - 1) / alts->length();
- for (int i = 0; i < alts->length(); i++) {
- GuardedAlternative& alt = alts->at(i);
- if (alt.guards() != nullptr && alt.guards()->length() != 0) {
- bm->SetRest(offset); // Give up trying to fill in info.
- SaveBMInfo(bm, not_at_start, offset);
- return;
- }
- alt.node()->FillInBMInfo(isolate, offset, budget, bm, not_at_start);
- }
- SaveBMInfo(bm, not_at_start, offset);
-}
-
-
-void TextNode::FillInBMInfo(Isolate* isolate, int initial_offset, int budget,
- BoyerMooreLookahead* bm, bool not_at_start) {
- if (initial_offset >= bm->length()) return;
- int offset = initial_offset;
- int max_char = bm->max_char();
- for (int i = 0; i < elements()->length(); i++) {
- if (offset >= bm->length()) {
- if (initial_offset == 0) set_bm_info(not_at_start, bm);
- return;
- }
- TextElement text = elements()->at(i);
- if (text.text_type() == TextElement::ATOM) {
- RegExpAtom* atom = text.atom();
- for (int j = 0; j < atom->length(); j++, offset++) {
- if (offset >= bm->length()) {
- if (initial_offset == 0) set_bm_info(not_at_start, bm);
- return;
- }
- uc16 character = atom->data()[j];
- if (IgnoreCase(atom->flags())) {
- unibrow::uchar chars[4];
- int length = GetCaseIndependentLetters(
- isolate, character, bm->max_char() == String::kMaxOneByteCharCode,
- chars, 4);
- for (int j = 0; j < length; j++) {
- bm->Set(offset, chars[j]);
- }
- } else {
- if (character <= max_char) bm->Set(offset, character);
- }
- }
- } else {
- DCHECK_EQ(TextElement::CHAR_CLASS, text.text_type());
- RegExpCharacterClass* char_class = text.char_class();
- ZoneList<CharacterRange>* ranges = char_class->ranges(zone());
- if (char_class->is_negated()) {
- bm->SetAll(offset);
- } else {
- for (int k = 0; k < ranges->length(); k++) {
- CharacterRange& range = ranges->at(k);
- if (range.from() > max_char) continue;
- int to = Min(max_char, static_cast<int>(range.to()));
- bm->SetInterval(offset, Interval(range.from(), to));
- }
- }
- offset++;
- }
- }
- if (offset >= bm->length()) {
- if (initial_offset == 0) set_bm_info(not_at_start, bm);
- return;
- }
- on_success()->FillInBMInfo(isolate, offset, budget - 1, bm,
- true); // Not at start after a text node.
- if (initial_offset == 0) set_bm_info(not_at_start, bm);
-}
-
-
-// -------------------------------------------------------------------
-// Dispatch table construction
-
-
-void DispatchTableConstructor::VisitEnd(EndNode* that) {
- AddRange(CharacterRange::Everything());
-}
-
-
-void DispatchTableConstructor::BuildTable(ChoiceNode* node) {
- node->set_being_calculated(true);
- ZoneList<GuardedAlternative>* alternatives = node->alternatives();
- for (int i = 0; i < alternatives->length(); i++) {
- set_choice_index(i);
- alternatives->at(i).node()->Accept(this);
- }
- node->set_being_calculated(false);
-}
-
-
-class AddDispatchRange {
- public:
- explicit AddDispatchRange(DispatchTableConstructor* constructor)
- : constructor_(constructor) { }
- void Call(uc32 from, DispatchTable::Entry entry);
- private:
- DispatchTableConstructor* constructor_;
-};
-
-
-void AddDispatchRange::Call(uc32 from, DispatchTable::Entry entry) {
- constructor_->AddRange(CharacterRange::Range(from, entry.to()));
-}
-
-
-void DispatchTableConstructor::VisitChoice(ChoiceNode* node) {
- if (node->being_calculated())
- return;
- DispatchTable* table = node->GetTable(ignore_case_);
- AddDispatchRange adder(this);
- table->ForEach(&adder);
-}
-
-
-void DispatchTableConstructor::VisitBackReference(BackReferenceNode* that) {
- // TODO(160): Find the node that we refer back to and propagate its start
- // set back to here. For now we just accept anything.
- AddRange(CharacterRange::Everything());
-}
-
-
-void DispatchTableConstructor::VisitAssertion(AssertionNode* that) {
- RegExpNode* target = that->on_success();
- target->Accept(this);
-}
-
-
-static int CompareRangeByFrom(const CharacterRange* a,
- const CharacterRange* b) {
- return Compare<uc16>(a->from(), b->from());
-}
-
-
-void DispatchTableConstructor::AddInverse(ZoneList<CharacterRange>* ranges) {
- ranges->Sort(CompareRangeByFrom);
- uc16 last = 0;
- for (int i = 0; i < ranges->length(); i++) {
- CharacterRange range = ranges->at(i);
- if (last < range.from())
- AddRange(CharacterRange::Range(last, range.from() - 1));
- if (range.to() >= last) {
- if (range.to() == String::kMaxCodePoint) {
- return;
- } else {
- last = range.to() + 1;
- }
- }
- }
- AddRange(CharacterRange::Range(last, String::kMaxCodePoint));
-}
-
-
-void DispatchTableConstructor::VisitText(TextNode* that) {
- TextElement elm = that->elements()->at(0);
- switch (elm.text_type()) {
- case TextElement::ATOM: {
- uc16 c = elm.atom()->data()[0];
- AddRange(CharacterRange::Range(c, c));
- break;
- }
- case TextElement::CHAR_CLASS: {
- RegExpCharacterClass* tree = elm.char_class();
- ZoneList<CharacterRange>* ranges = tree->ranges(that->zone());
- if (tree->is_negated()) {
- AddInverse(ranges);
- } else {
- for (int i = 0; i < ranges->length(); i++)
- AddRange(ranges->at(i));
- }
- break;
- }
- default: {
- UNIMPLEMENTED();
- }
- }
-}
-
-
-void DispatchTableConstructor::VisitAction(ActionNode* that) {
- RegExpNode* target = that->on_success();
- target->Accept(this);
-}
-
-RegExpNode* OptionallyStepBackToLeadSurrogate(RegExpCompiler* compiler,
- RegExpNode* on_success,
- JSRegExp::Flags flags) {
- // 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, flags);
- RegExpLookaround::Builder builder(true, step_back, stack_register,
- position_register);
- RegExpNode* match_trail = TextNode::CreateForCharacterRanges(
- zone, trail_surrogates, false, builder.on_match_success(), flags);
-
- 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,
- 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);
- }
- bool is_sticky = IsSticky(flags);
- bool is_global = IsGlobal(flags);
- bool is_unicode = IsUnicode(flags);
- RegExpCompiler compiler(isolate, zone, data->capture_count, is_one_byte);
-
- if (compiler.optimize())
- compiler.set_optimize(!TooMuchRegExpCode(isolate, pattern));
-
- // Sample some characters from the middle of the string.
- static const int kSampleSize = 128;
-
- sample_subject = String::Flatten(isolate, sample_subject);
- int chars_sampled = 0;
- int half_way = (sample_subject->length() - kSampleSize) / 2;
- for (int i = Max(0, half_way);
- i < sample_subject->length() && chars_sampled < kSampleSize;
- i++, chars_sampled++) {
- compiler.frequency_collator()->CountCharacter(sample_subject->Get(i));
- }
-
- // Wrap the body of the regexp in capture #0.
- RegExpNode* captured_body = RegExpCapture::ToNode(data->tree,
- 0,
- &compiler,
- compiler.accept());
- RegExpNode* node = captured_body;
- bool is_end_anchored = data->tree->IsAnchoredAtEnd();
- bool is_start_anchored = data->tree->IsAnchoredAtStart();
- int max_length = data->tree->max_match();
- if (!is_start_anchored && !is_sticky) {
- // Add a .*? at the beginning, outside the body capture, unless
- // this expression is anchored at the beginning or sticky.
- JSRegExp::Flags default_flags = JSRegExp::Flags();
- RegExpNode* loop_node = RegExpQuantifier::ToNode(
- 0, RegExpTree::kInfinity, false,
- new (zone) RegExpCharacterClass('*', default_flags), &compiler,
- captured_body, data->contains_anchor);
-
- if (data->contains_anchor) {
- // Unroll loop once, to take care of the case that might start
- // at the start of input.
- ChoiceNode* first_step_node = new(zone) ChoiceNode(2, zone);
- first_step_node->AddAlternative(GuardedAlternative(captured_body));
- first_step_node->AddAlternative(GuardedAlternative(new (zone) TextNode(
- new (zone) RegExpCharacterClass('*', default_flags), false,
- loop_node)));
- node = first_step_node;
- } else {
- node = loop_node;
- }
- }
- if (is_one_byte) {
- node = node->FilterOneByte(RegExpCompiler::kMaxRecursion);
- // Do it again to propagate the new nodes to places where they were not
- // put because they had not been calculated yet.
- if (node != nullptr) {
- node = node->FilterOneByte(RegExpCompiler::kMaxRecursion);
- }
- } else if (is_unicode && (is_global || is_sticky)) {
- node = OptionallyStepBackToLeadSurrogate(&compiler, node, flags);
- }
-
- if (node == nullptr) node = new (zone) EndNode(EndNode::BACKTRACK, zone);
- data->node = node;
- Analysis analysis(isolate, is_one_byte);
- analysis.EnsureAnalyzed(node);
- if (analysis.has_failed()) {
- const char* error_message = analysis.error_message();
- return CompilationResult(isolate, error_message);
- }
-
- // Create the correct assembler for the architecture.
- std::unique_ptr<RegExpMacroAssembler> macro_assembler;
- if (!FLAG_regexp_interpret_all) {
- // Native regexp implementation.
- DCHECK(!FLAG_jitless);
-
- NativeRegExpMacroAssembler::Mode mode =
- is_one_byte ? NativeRegExpMacroAssembler::LATIN1
- : NativeRegExpMacroAssembler::UC16;
-
-#if V8_TARGET_ARCH_IA32
- macro_assembler.reset(new RegExpMacroAssemblerIA32(
- isolate, zone, mode, (data->capture_count + 1) * 2));
-#elif V8_TARGET_ARCH_X64
- macro_assembler.reset(new RegExpMacroAssemblerX64(
- isolate, zone, mode, (data->capture_count + 1) * 2));
-#elif V8_TARGET_ARCH_ARM
- macro_assembler.reset(new RegExpMacroAssemblerARM(
- isolate, zone, mode, (data->capture_count + 1) * 2));
-#elif V8_TARGET_ARCH_ARM64
- macro_assembler.reset(new RegExpMacroAssemblerARM64(
- isolate, zone, mode, (data->capture_count + 1) * 2));
-#elif V8_TARGET_ARCH_S390
- macro_assembler.reset(new RegExpMacroAssemblerS390(
- isolate, zone, mode, (data->capture_count + 1) * 2));
-#elif V8_TARGET_ARCH_PPC
- macro_assembler.reset(new RegExpMacroAssemblerPPC(
- isolate, zone, mode, (data->capture_count + 1) * 2));
-#elif V8_TARGET_ARCH_MIPS
- macro_assembler.reset(new RegExpMacroAssemblerMIPS(
- isolate, zone, mode, (data->capture_count + 1) * 2));
-#elif V8_TARGET_ARCH_MIPS64
- macro_assembler.reset(new RegExpMacroAssemblerMIPS(
- isolate, zone, mode, (data->capture_count + 1) * 2));
-#else
-#error "Unsupported architecture"
-#endif
- } else {
- DCHECK(FLAG_regexp_interpret_all);
-
- // Interpreted regexp implementation.
- macro_assembler.reset(new RegExpMacroAssemblerIrregexp(isolate, zone));
- }
-
- macro_assembler->set_slow_safe(TooMuchRegExpCode(isolate, pattern));
-
- // Inserted here, instead of in Assembler, because it depends on information
- // in the AST that isn't replicated in the Node structure.
- static const int kMaxBacksearchLimit = 1024;
- if (is_end_anchored && !is_start_anchored && !is_sticky &&
- max_length < kMaxBacksearchLimit) {
- macro_assembler->SetCurrentPositionFromEnd(max_length);
- }
-
- if (is_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(isolate, macro_assembler.get(), node,
- data->capture_count, pattern);
-}
-
-bool RegExpEngine::TooMuchRegExpCode(Isolate* isolate, Handle<String> pattern) {
- Heap* heap = isolate->heap();
- bool too_much = pattern->length() > RegExpImpl::kRegExpTooLargeToOptimize;
- if (isolate->total_regexp_code_generated() >
- RegExpImpl::kRegExpCompiledLimit &&
- heap->CommittedMemoryExecutable() >
- RegExpImpl::kRegExpExecutableMemoryLimit) {
- too_much = true;
- }
- return too_much;
-}
-
-Object RegExpResultsCache::Lookup(Heap* heap, String key_string,
- Object key_pattern,
- FixedArray* last_match_cache,
- ResultsCacheType type) {
- FixedArray cache;
- if (!key_string.IsInternalizedString()) return Smi::kZero;
- if (type == STRING_SPLIT_SUBSTRINGS) {
- DCHECK(key_pattern.IsString());
- if (!key_pattern.IsInternalizedString()) return Smi::kZero;
- cache = heap->string_split_cache();
- } else {
- DCHECK(type == REGEXP_MULTIPLE_INDICES);
- DCHECK(key_pattern.IsFixedArray());
- cache = heap->regexp_multiple_cache();
- }
-
- uint32_t hash = key_string.Hash();
- uint32_t index = ((hash & (kRegExpResultsCacheSize - 1)) &
- ~(kArrayEntriesPerCacheEntry - 1));
- if (cache.get(index + kStringOffset) != key_string ||
- cache.get(index + kPatternOffset) != key_pattern) {
- index =
- ((index + kArrayEntriesPerCacheEntry) & (kRegExpResultsCacheSize - 1));
- if (cache.get(index + kStringOffset) != key_string ||
- cache.get(index + kPatternOffset) != key_pattern) {
- return Smi::kZero;
- }
- }
-
- *last_match_cache = FixedArray::cast(cache.get(index + kLastMatchOffset));
- return cache.get(index + kArrayOffset);
-}
-
-void RegExpResultsCache::Enter(Isolate* isolate, Handle<String> key_string,
- Handle<Object> key_pattern,
- Handle<FixedArray> value_array,
- Handle<FixedArray> last_match_cache,
- ResultsCacheType type) {
- Factory* factory = isolate->factory();
- Handle<FixedArray> cache;
- if (!key_string->IsInternalizedString()) return;
- if (type == STRING_SPLIT_SUBSTRINGS) {
- DCHECK(key_pattern->IsString());
- if (!key_pattern->IsInternalizedString()) return;
- cache = factory->string_split_cache();
- } else {
- DCHECK(type == REGEXP_MULTIPLE_INDICES);
- DCHECK(key_pattern->IsFixedArray());
- cache = factory->regexp_multiple_cache();
- }
-
- uint32_t hash = key_string->Hash();
- uint32_t index = ((hash & (kRegExpResultsCacheSize - 1)) &
- ~(kArrayEntriesPerCacheEntry - 1));
- if (cache->get(index + kStringOffset) == Smi::kZero) {
- cache->set(index + kStringOffset, *key_string);
- cache->set(index + kPatternOffset, *key_pattern);
- cache->set(index + kArrayOffset, *value_array);
- cache->set(index + kLastMatchOffset, *last_match_cache);
- } else {
- uint32_t index2 =
- ((index + kArrayEntriesPerCacheEntry) & (kRegExpResultsCacheSize - 1));
- if (cache->get(index2 + kStringOffset) == Smi::kZero) {
- cache->set(index2 + kStringOffset, *key_string);
- cache->set(index2 + kPatternOffset, *key_pattern);
- cache->set(index2 + kArrayOffset, *value_array);
- cache->set(index2 + kLastMatchOffset, *last_match_cache);
- } else {
- cache->set(index2 + kStringOffset, Smi::kZero);
- cache->set(index2 + kPatternOffset, Smi::kZero);
- cache->set(index2 + kArrayOffset, Smi::kZero);
- cache->set(index2 + kLastMatchOffset, Smi::kZero);
- cache->set(index + kStringOffset, *key_string);
- cache->set(index + kPatternOffset, *key_pattern);
- cache->set(index + kArrayOffset, *value_array);
- cache->set(index + kLastMatchOffset, *last_match_cache);
- }
- }
- // If the array is a reasonably short list of substrings, convert it into a
- // list of internalized strings.
- if (type == STRING_SPLIT_SUBSTRINGS && value_array->length() < 100) {
- for (int i = 0; i < value_array->length(); i++) {
- Handle<String> str(String::cast(value_array->get(i)), isolate);
- Handle<String> internalized_str = factory->InternalizeString(str);
- value_array->set(i, *internalized_str);
- }
- }
- // Convert backing store to a copy-on-write array.
- value_array->set_map_no_write_barrier(
- ReadOnlyRoots(isolate).fixed_cow_array_map());
-}
-
-void RegExpResultsCache::Clear(FixedArray cache) {
- for (int i = 0; i < kRegExpResultsCacheSize; i++) {
- cache.set(i, Smi::kZero);
- }
-}
-
-} // namespace internal
-} // namespace v8
diff --git a/deps/v8/src/regexp/jsregexp.h b/deps/v8/src/regexp/jsregexp.h
deleted file mode 100644
index 832c7e3aa5..0000000000
--- a/deps/v8/src/regexp/jsregexp.h
+++ /dev/null
@@ -1,1548 +0,0 @@
-// Copyright 2012 the V8 project authors. All rights reserved.
-// Use of this source code is governed by a BSD-style license that can be
-// found in the LICENSE file.
-
-#ifndef V8_REGEXP_JSREGEXP_H_
-#define V8_REGEXP_JSREGEXP_H_
-
-#include "src/execution/isolate.h"
-#include "src/objects/js-regexp.h"
-#include "src/regexp/regexp-ast.h"
-#include "src/regexp/regexp-macro-assembler.h"
-#include "src/utils/allocation.h"
-#include "src/zone/zone-splay-tree.h"
-
-namespace v8 {
-namespace internal {
-
-class NodeVisitor;
-class RegExpCompiler;
-class RegExpMacroAssembler;
-class RegExpNode;
-class RegExpTree;
-class BoyerMooreLookahead;
-
-inline bool IgnoreCase(JSRegExp::Flags flags) {
- return (flags & JSRegExp::kIgnoreCase) != 0;
-}
-
-inline bool IsUnicode(JSRegExp::Flags flags) {
- return (flags & JSRegExp::kUnicode) != 0;
-}
-
-inline bool IsSticky(JSRegExp::Flags flags) {
- return (flags & JSRegExp::kSticky) != 0;
-}
-
-inline bool IsGlobal(JSRegExp::Flags flags) {
- return (flags & JSRegExp::kGlobal) != 0;
-}
-
-inline bool DotAll(JSRegExp::Flags flags) {
- return (flags & JSRegExp::kDotAll) != 0;
-}
-
-inline bool Multiline(JSRegExp::Flags flags) {
- return (flags & JSRegExp::kMultiline) != 0;
-}
-
-inline bool NeedsUnicodeCaseEquivalents(JSRegExp::Flags flags) {
- // Both unicode and ignore_case flags are set. We need to use ICU to find
- // the closure over case equivalents.
- return IsUnicode(flags) && IgnoreCase(flags);
-}
-
-class RegExpImpl {
- public:
- // Whether the irregexp engine generates native code or interpreter bytecode.
- static bool UsesNativeRegExp() { return !FLAG_regexp_interpret_all; }
-
- // Returns a string representation of a regular expression.
- // Implements RegExp.prototype.toString, see ECMA-262 section 15.10.6.4.
- // This function calls the garbage collector if necessary.
- static Handle<String> ToString(Handle<Object> value);
-
- // Parses the RegExp pattern and prepares the JSRegExp object with
- // generic data and choice of implementation - as well as what
- // the implementation wants to store in the data field.
- // Returns false if compilation fails.
- V8_WARN_UNUSED_RESULT static MaybeHandle<Object> Compile(
- Isolate* isolate, Handle<JSRegExp> re, Handle<String> pattern,
- JSRegExp::Flags flags);
-
- // See ECMA-262 section 15.10.6.2.
- // This function calls the garbage collector if necessary.
- V8_EXPORT_PRIVATE V8_WARN_UNUSED_RESULT static MaybeHandle<Object> Exec(
- Isolate* isolate, Handle<JSRegExp> regexp, Handle<String> subject,
- int index, Handle<RegExpMatchInfo> last_match_info);
-
- // Prepares a JSRegExp object with Irregexp-specific data.
- static void IrregexpInitialize(Isolate* isolate, Handle<JSRegExp> re,
- Handle<String> pattern, JSRegExp::Flags flags,
- int capture_register_count);
-
- static void AtomCompile(Isolate* isolate, Handle<JSRegExp> re,
- Handle<String> pattern, JSRegExp::Flags flags,
- Handle<String> match_pattern);
-
- static int AtomExecRaw(Isolate* isolate, Handle<JSRegExp> regexp,
- Handle<String> subject, int index, int32_t* output,
- int output_size);
-
- static Handle<Object> AtomExec(Isolate* isolate, Handle<JSRegExp> regexp,
- Handle<String> subject, int index,
- Handle<RegExpMatchInfo> last_match_info);
-
- enum IrregexpResult { RE_FAILURE = 0, RE_SUCCESS = 1, RE_EXCEPTION = -1 };
-
- // Prepare a RegExp for being executed one or more times (using
- // IrregexpExecOnce) on the subject.
- // This ensures that the regexp is compiled for the subject, and that
- // the subject is flat.
- // Returns the number of integer spaces required by IrregexpExecOnce
- // as its "registers" argument. If the regexp cannot be compiled,
- // an exception is set as pending, and this function returns negative.
- static int IrregexpPrepare(Isolate* isolate, Handle<JSRegExp> regexp,
- Handle<String> subject);
-
- // Execute a regular expression on the subject, starting from index.
- // If matching succeeds, return the number of matches. This can be larger
- // than one in the case of global regular expressions.
- // The captures and subcaptures are stored into the registers vector.
- // If matching fails, returns RE_FAILURE.
- // If execution fails, sets a pending exception and returns RE_EXCEPTION.
- static int IrregexpExecRaw(Isolate* isolate, Handle<JSRegExp> regexp,
- Handle<String> subject, int index, int32_t* output,
- int output_size);
-
- // Execute an Irregexp bytecode pattern.
- // On a successful match, the result is a JSArray containing
- // captured positions. On a failure, the result is the null value.
- // Returns an empty handle in case of an exception.
- V8_WARN_UNUSED_RESULT static MaybeHandle<Object> IrregexpExec(
- Isolate* isolate, Handle<JSRegExp> regexp, Handle<String> subject,
- int index, Handle<RegExpMatchInfo> last_match_info);
-
- // Set last match info. If match is nullptr, then setting captures is
- // omitted.
- static Handle<RegExpMatchInfo> SetLastMatchInfo(
- Isolate* isolate, Handle<RegExpMatchInfo> last_match_info,
- Handle<String> subject, int capture_count, int32_t* match);
-
- class GlobalCache {
- public:
- GlobalCache(Handle<JSRegExp> regexp,
- Handle<String> subject,
- Isolate* isolate);
-
- V8_INLINE ~GlobalCache();
-
- // Fetch the next entry in the cache for global regexp match results.
- // This does not set the last match info. Upon failure, nullptr is
- // returned. The cause can be checked with Result(). The previous result is
- // still in available in memory when a failure happens.
- V8_INLINE int32_t* FetchNext();
-
- V8_INLINE int32_t* LastSuccessfulMatch();
-
- V8_INLINE bool HasException() { return num_matches_ < 0; }
-
- private:
- int AdvanceZeroLength(int last_index);
-
- int num_matches_;
- int max_matches_;
- int current_match_index_;
- int registers_per_match_;
- // Pointer to the last set of captures.
- int32_t* register_array_;
- int register_array_size_;
- Handle<JSRegExp> regexp_;
- Handle<String> subject_;
- Isolate* isolate_;
- };
-
- // For acting on the JSRegExp data FixedArray.
- static int IrregexpMaxRegisterCount(FixedArray re);
- static void SetIrregexpMaxRegisterCount(FixedArray re, int value);
- static void SetIrregexpCaptureNameMap(FixedArray re,
- Handle<FixedArray> value);
- static int IrregexpNumberOfCaptures(FixedArray re);
- static int IrregexpNumberOfRegisters(FixedArray re);
- static ByteArray IrregexpByteCode(FixedArray re, bool is_one_byte);
- static Code IrregexpNativeCode(FixedArray re, bool is_one_byte);
-
- // Limit the space regexps take up on the heap. In order to limit this we
- // would like to keep track of the amount of regexp code on the heap. This
- // is not tracked, however. As a conservative approximation we track the
- // total regexp code compiled including code that has subsequently been freed
- // and the total executable memory at any point.
- static const size_t kRegExpExecutableMemoryLimit = 16 * MB;
- static const size_t kRegExpCompiledLimit = 1 * MB;
- static const int kRegExpTooLargeToOptimize = 20 * KB;
-
- private:
- static bool CompileIrregexp(Isolate* isolate, Handle<JSRegExp> re,
- Handle<String> sample_subject, bool is_one_byte);
- static inline bool EnsureCompiledIrregexp(Isolate* isolate,
- Handle<JSRegExp> re,
- Handle<String> sample_subject,
- bool is_one_byte);
-};
-
-
-// Represents the location of one element relative to the intersection of
-// two sets. Corresponds to the four areas of a Venn diagram.
-enum ElementInSetsRelation {
- kInsideNone = 0,
- kInsideFirst = 1,
- kInsideSecond = 2,
- kInsideBoth = 3
-};
-
-
-// A set of unsigned integers that behaves especially well on small
-// integers (< 32). May do zone-allocation.
-class OutSet: public ZoneObject {
- public:
- OutSet() : first_(0), remaining_(nullptr), successors_(nullptr) {}
- OutSet* Extend(unsigned value, Zone* zone);
- V8_EXPORT_PRIVATE bool Get(unsigned value) const;
- static const unsigned kFirstLimit = 32;
-
- private:
- // Destructively set a value in this set. In most cases you want
- // to use Extend instead to ensure that only one instance exists
- // that contains the same values.
- void Set(unsigned value, Zone* zone);
-
- // The successors are a list of sets that contain the same values
- // as this set and the one more value that is not present in this
- // set.
- ZoneList<OutSet*>* successors(Zone* zone) { return successors_; }
-
- OutSet(uint32_t first, ZoneList<unsigned>* remaining)
- : first_(first), remaining_(remaining), successors_(nullptr) {}
- uint32_t first_;
- ZoneList<unsigned>* remaining_;
- ZoneList<OutSet*>* successors_;
- friend class Trace;
-};
-
-
-// A mapping from integers, specified as ranges, to a set of integers.
-// Used for mapping character ranges to choices.
-class DispatchTable : public ZoneObject {
- public:
- explicit DispatchTable(Zone* zone) : tree_(zone) { }
-
- class Entry {
- public:
- Entry() : from_(0), to_(0), out_set_(nullptr) {}
- Entry(uc32 from, uc32 to, OutSet* out_set)
- : from_(from), to_(to), out_set_(out_set) {
- DCHECK(from <= to);
- }
- uc32 from() { return from_; }
- uc32 to() { return to_; }
- void set_to(uc32 value) { to_ = value; }
- void AddValue(int value, Zone* zone) {
- out_set_ = out_set_->Extend(value, zone);
- }
- OutSet* out_set() { return out_set_; }
- private:
- uc32 from_;
- uc32 to_;
- OutSet* out_set_;
- };
-
- class Config {
- public:
- using Key = uc32;
- using Value = Entry;
- static const uc32 kNoKey;
- static const Entry NoValue() { return Value(); }
- static inline int Compare(uc32 a, uc32 b) {
- if (a == b)
- return 0;
- else if (a < b)
- return -1;
- else
- return 1;
- }
- };
-
- V8_EXPORT_PRIVATE void AddRange(CharacterRange range, int value, Zone* zone);
- V8_EXPORT_PRIVATE OutSet* Get(uc32 value);
- void Dump();
-
- template <typename Callback>
- void ForEach(Callback* callback) {
- return tree()->ForEach(callback);
- }
-
- private:
- // There can't be a static empty set since it allocates its
- // successors in a zone and caches them.
- OutSet* empty() { return &empty_; }
- OutSet empty_;
- ZoneSplayTree<Config>* tree() { return &tree_; }
- ZoneSplayTree<Config> tree_;
-};
-
-
-// Categorizes character ranges into BMP, non-BMP, lead, and trail surrogates.
-class UnicodeRangeSplitter {
- public:
- V8_EXPORT_PRIVATE UnicodeRangeSplitter(Zone* zone,
- ZoneList<CharacterRange>* base);
- void Call(uc32 from, DispatchTable::Entry entry);
-
- ZoneList<CharacterRange>* bmp() { return bmp_; }
- ZoneList<CharacterRange>* lead_surrogates() { return lead_surrogates_; }
- ZoneList<CharacterRange>* trail_surrogates() { return trail_surrogates_; }
- ZoneList<CharacterRange>* non_bmp() const { return non_bmp_; }
-
- private:
- static const int kBase = 0;
- // Separate ranges into
- static const int kBmpCodePoints = 1;
- static const int kLeadSurrogates = 2;
- static const int kTrailSurrogates = 3;
- static const int kNonBmpCodePoints = 4;
-
- Zone* zone_;
- DispatchTable table_;
- ZoneList<CharacterRange>* bmp_;
- ZoneList<CharacterRange>* lead_surrogates_;
- ZoneList<CharacterRange>* trail_surrogates_;
- ZoneList<CharacterRange>* non_bmp_;
-};
-
-#define FOR_EACH_NODE_TYPE(VISIT) \
- VISIT(End) \
- VISIT(Action) \
- VISIT(Choice) \
- VISIT(BackReference) \
- VISIT(Assertion) \
- VISIT(Text)
-
-
-class Trace;
-struct PreloadState;
-class GreedyLoopState;
-class AlternativeGenerationList;
-
-struct NodeInfo {
- NodeInfo()
- : being_analyzed(false),
- been_analyzed(false),
- follows_word_interest(false),
- follows_newline_interest(false),
- follows_start_interest(false),
- at_end(false),
- visited(false),
- replacement_calculated(false) { }
-
- // Returns true if the interests and assumptions of this node
- // matches the given one.
- bool Matches(NodeInfo* that) {
- return (at_end == that->at_end) &&
- (follows_word_interest == that->follows_word_interest) &&
- (follows_newline_interest == that->follows_newline_interest) &&
- (follows_start_interest == that->follows_start_interest);
- }
-
- // Updates the interests of this node given the interests of the
- // node preceding it.
- void AddFromPreceding(NodeInfo* that) {
- at_end |= that->at_end;
- follows_word_interest |= that->follows_word_interest;
- follows_newline_interest |= that->follows_newline_interest;
- follows_start_interest |= that->follows_start_interest;
- }
-
- bool HasLookbehind() {
- return follows_word_interest ||
- follows_newline_interest ||
- follows_start_interest;
- }
-
- // Sets the interests of this node to include the interests of the
- // following node.
- void AddFromFollowing(NodeInfo* that) {
- follows_word_interest |= that->follows_word_interest;
- follows_newline_interest |= that->follows_newline_interest;
- follows_start_interest |= that->follows_start_interest;
- }
-
- void ResetCompilationState() {
- being_analyzed = false;
- been_analyzed = false;
- }
-
- bool being_analyzed: 1;
- bool been_analyzed: 1;
-
- // These bits are set of this node has to know what the preceding
- // character was.
- bool follows_word_interest: 1;
- bool follows_newline_interest: 1;
- bool follows_start_interest: 1;
-
- bool at_end: 1;
- bool visited: 1;
- bool replacement_calculated: 1;
-};
-
-
-// Details of a quick mask-compare check that can look ahead in the
-// input stream.
-class QuickCheckDetails {
- public:
- QuickCheckDetails()
- : characters_(0),
- mask_(0),
- value_(0),
- cannot_match_(false) { }
- explicit QuickCheckDetails(int characters)
- : characters_(characters),
- mask_(0),
- value_(0),
- cannot_match_(false) { }
- bool Rationalize(bool one_byte);
- // Merge in the information from another branch of an alternation.
- void Merge(QuickCheckDetails* other, int from_index);
- // Advance the current position by some amount.
- void Advance(int by, bool one_byte);
- void Clear();
- bool cannot_match() { return cannot_match_; }
- void set_cannot_match() { cannot_match_ = true; }
- struct Position {
- Position() : mask(0), value(0), determines_perfectly(false) { }
- uc16 mask;
- uc16 value;
- bool determines_perfectly;
- };
- int characters() { return characters_; }
- void set_characters(int characters) { characters_ = characters; }
- Position* positions(int index) {
- DCHECK_LE(0, index);
- DCHECK_GT(characters_, index);
- return positions_ + index;
- }
- uint32_t mask() { return mask_; }
- uint32_t value() { return value_; }
-
- private:
- // How many characters do we have quick check information from. This is
- // the same for all branches of a choice node.
- int characters_;
- Position positions_[4];
- // These values are the condensate of the above array after Rationalize().
- uint32_t mask_;
- uint32_t value_;
- // If set to true, there is no way this quick check can match at all.
- // E.g., if it requires to be at the start of the input, and isn't.
- bool cannot_match_;
-};
-
-
-extern int kUninitializedRegExpNodePlaceHolder;
-
-
-class RegExpNode: public ZoneObject {
- public:
- explicit RegExpNode(Zone* zone)
- : replacement_(nullptr),
- on_work_list_(false),
- trace_count_(0),
- zone_(zone) {
- bm_info_[0] = bm_info_[1] = nullptr;
- }
- virtual ~RegExpNode();
- virtual void Accept(NodeVisitor* visitor) = 0;
- // Generates a goto to this node or actually generates the code at this point.
- virtual void Emit(RegExpCompiler* compiler, Trace* trace) = 0;
- // How many characters must this node consume at a minimum in order to
- // succeed. If we have found at least 'still_to_find' characters that
- // must be consumed there is no need to ask any following nodes whether
- // they are sure to eat any more characters. The not_at_start argument is
- // used to indicate that we know we are not at the start of the input. In
- // this case anchored branches will always fail and can be ignored when
- // determining how many characters are consumed on success.
- virtual int EatsAtLeast(int still_to_find, int budget, bool not_at_start) = 0;
- // Emits some quick code that checks whether the preloaded characters match.
- // Falls through on certain failure, jumps to the label on possible success.
- // If the node cannot make a quick check it does nothing and returns false.
- bool EmitQuickCheck(RegExpCompiler* compiler,
- Trace* bounds_check_trace,
- Trace* trace,
- bool preload_has_checked_bounds,
- Label* on_possible_success,
- QuickCheckDetails* details_return,
- bool fall_through_on_failure);
- // For a given number of characters this returns a mask and a value. The
- // next n characters are anded with the mask and compared with the value.
- // A comparison failure indicates the node cannot match the next n characters.
- // A comparison success indicates the node may match.
- virtual void GetQuickCheckDetails(QuickCheckDetails* details,
- RegExpCompiler* compiler,
- int characters_filled_in,
- bool not_at_start) = 0;
- static const int kNodeIsTooComplexForGreedyLoops = kMinInt;
- virtual int GreedyLoopTextLength() { return kNodeIsTooComplexForGreedyLoops; }
- // Only returns the successor for a text node of length 1 that matches any
- // character and that has no guards on it.
- virtual RegExpNode* GetSuccessorOfOmnivorousTextNode(
- RegExpCompiler* compiler) {
- return nullptr;
- }
-
- // Collects information on the possible code units (mod 128) that can match if
- // we look forward. This is used for a Boyer-Moore-like string searching
- // implementation. TODO(erikcorry): This should share more code with
- // EatsAtLeast, GetQuickCheckDetails. The budget argument is used to limit
- // the number of nodes we are willing to look at in order to create this data.
- static const int kRecursionBudget = 200;
- bool KeepRecursing(RegExpCompiler* compiler);
- virtual void FillInBMInfo(Isolate* isolate, int offset, int budget,
- BoyerMooreLookahead* bm, bool not_at_start) {
- UNREACHABLE();
- }
-
- // If we know that the input is one-byte then there are some nodes that can
- // never match. This method returns a node that can be substituted for
- // itself, or nullptr if the node can never match.
- virtual RegExpNode* FilterOneByte(int depth) { return this; }
- // Helper for FilterOneByte.
- RegExpNode* replacement() {
- DCHECK(info()->replacement_calculated);
- return replacement_;
- }
- RegExpNode* set_replacement(RegExpNode* replacement) {
- info()->replacement_calculated = true;
- replacement_ = replacement;
- return replacement; // For convenience.
- }
-
- // We want to avoid recalculating the lookahead info, so we store it on the
- // node. Only info that is for this node is stored. We can tell that the
- // info is for this node when offset == 0, so the information is calculated
- // relative to this node.
- void SaveBMInfo(BoyerMooreLookahead* bm, bool not_at_start, int offset) {
- if (offset == 0) set_bm_info(not_at_start, bm);
- }
-
- Label* label() { return &label_; }
- // If non-generic code is generated for a node (i.e. the node is not at the
- // start of the trace) then it cannot be reused. This variable sets a limit
- // on how often we allow that to happen before we insist on starting a new
- // trace and generating generic code for a node that can be reused by flushing
- // the deferred actions in the current trace and generating a goto.
- static const int kMaxCopiesCodeGenerated = 10;
-
- bool on_work_list() { return on_work_list_; }
- void set_on_work_list(bool value) { on_work_list_ = value; }
-
- NodeInfo* info() { return &info_; }
-
- BoyerMooreLookahead* bm_info(bool not_at_start) {
- return bm_info_[not_at_start ? 1 : 0];
- }
-
- Zone* zone() const { return zone_; }
-
- protected:
- enum LimitResult { DONE, CONTINUE };
- RegExpNode* replacement_;
-
- LimitResult LimitVersions(RegExpCompiler* compiler, Trace* trace);
-
- void set_bm_info(bool not_at_start, BoyerMooreLookahead* bm) {
- bm_info_[not_at_start ? 1 : 0] = bm;
- }
-
- private:
- static const int kFirstCharBudget = 10;
- Label label_;
- bool on_work_list_;
- NodeInfo info_;
- // This variable keeps track of how many times code has been generated for
- // this node (in different traces). We don't keep track of where the
- // generated code is located unless the code is generated at the start of
- // a trace, in which case it is generic and can be reused by flushing the
- // deferred operations in the current trace and generating a goto.
- int trace_count_;
- BoyerMooreLookahead* bm_info_[2];
-
- Zone* zone_;
-};
-
-
-class SeqRegExpNode: public RegExpNode {
- public:
- explicit SeqRegExpNode(RegExpNode* on_success)
- : RegExpNode(on_success->zone()), on_success_(on_success) { }
- RegExpNode* on_success() { return on_success_; }
- void set_on_success(RegExpNode* node) { on_success_ = node; }
- RegExpNode* FilterOneByte(int depth) override;
- void FillInBMInfo(Isolate* isolate, int offset, int budget,
- BoyerMooreLookahead* bm, bool not_at_start) override {
- on_success_->FillInBMInfo(isolate, offset, budget - 1, bm, not_at_start);
- if (offset == 0) set_bm_info(not_at_start, bm);
- }
-
- protected:
- RegExpNode* FilterSuccessor(int depth);
-
- private:
- RegExpNode* on_success_;
-};
-
-
-class ActionNode: public SeqRegExpNode {
- public:
- enum ActionType {
- SET_REGISTER,
- INCREMENT_REGISTER,
- STORE_POSITION,
- BEGIN_SUBMATCH,
- POSITIVE_SUBMATCH_SUCCESS,
- EMPTY_MATCH_CHECK,
- CLEAR_CAPTURES
- };
- static ActionNode* SetRegister(int reg, int val, RegExpNode* on_success);
- static ActionNode* IncrementRegister(int reg, RegExpNode* on_success);
- static ActionNode* StorePosition(int reg,
- bool is_capture,
- RegExpNode* on_success);
- static ActionNode* ClearCaptures(Interval range, RegExpNode* on_success);
- static ActionNode* BeginSubmatch(int stack_pointer_reg,
- int position_reg,
- RegExpNode* on_success);
- static ActionNode* PositiveSubmatchSuccess(int stack_pointer_reg,
- int restore_reg,
- int clear_capture_count,
- int clear_capture_from,
- RegExpNode* on_success);
- static ActionNode* EmptyMatchCheck(int start_register,
- int repetition_register,
- int repetition_limit,
- RegExpNode* on_success);
- void Accept(NodeVisitor* visitor) override;
- void Emit(RegExpCompiler* compiler, Trace* trace) override;
- int EatsAtLeast(int still_to_find, int budget, bool not_at_start) override;
- void GetQuickCheckDetails(QuickCheckDetails* details,
- RegExpCompiler* compiler, int filled_in,
- bool not_at_start) override {
- return on_success()->GetQuickCheckDetails(
- details, compiler, filled_in, not_at_start);
- }
- void FillInBMInfo(Isolate* isolate, int offset, int budget,
- BoyerMooreLookahead* bm, bool not_at_start) override;
- ActionType action_type() { return action_type_; }
- // TODO(erikcorry): We should allow some action nodes in greedy loops.
- int GreedyLoopTextLength() override {
- return kNodeIsTooComplexForGreedyLoops;
- }
-
- private:
- union {
- struct {
- int reg;
- int value;
- } u_store_register;
- struct {
- int reg;
- } u_increment_register;
- struct {
- int reg;
- bool is_capture;
- } u_position_register;
- struct {
- int stack_pointer_register;
- int current_position_register;
- int clear_register_count;
- int clear_register_from;
- } u_submatch;
- struct {
- int start_register;
- int repetition_register;
- int repetition_limit;
- } u_empty_match_check;
- struct {
- int range_from;
- int range_to;
- } u_clear_captures;
- } data_;
- ActionNode(ActionType action_type, RegExpNode* on_success)
- : SeqRegExpNode(on_success),
- action_type_(action_type) { }
- ActionType action_type_;
- friend class DotPrinter;
-};
-
-
-class TextNode: public SeqRegExpNode {
- public:
- TextNode(ZoneList<TextElement>* elms, bool read_backward,
- RegExpNode* on_success)
- : SeqRegExpNode(on_success), elms_(elms), read_backward_(read_backward) {}
- TextNode(RegExpCharacterClass* that, bool read_backward,
- RegExpNode* on_success)
- : SeqRegExpNode(on_success),
- elms_(new (zone()) ZoneList<TextElement>(1, zone())),
- read_backward_(read_backward) {
- elms_->Add(TextElement::CharClass(that), zone());
- }
- // Create TextNode for a single character class for the given ranges.
- static TextNode* CreateForCharacterRanges(Zone* zone,
- ZoneList<CharacterRange>* ranges,
- bool read_backward,
- RegExpNode* on_success,
- JSRegExp::Flags flags);
- // Create TextNode for a surrogate pair with a range given for the
- // lead and the trail surrogate each.
- static TextNode* CreateForSurrogatePair(Zone* zone, CharacterRange lead,
- CharacterRange trail,
- bool read_backward,
- RegExpNode* on_success,
- JSRegExp::Flags flags);
- void Accept(NodeVisitor* visitor) override;
- void Emit(RegExpCompiler* compiler, Trace* trace) override;
- int EatsAtLeast(int still_to_find, int budget, bool not_at_start) override;
- void GetQuickCheckDetails(QuickCheckDetails* details,
- RegExpCompiler* compiler, int characters_filled_in,
- bool not_at_start) override;
- ZoneList<TextElement>* elements() { return elms_; }
- bool read_backward() { return read_backward_; }
- void MakeCaseIndependent(Isolate* isolate, bool is_one_byte);
- int GreedyLoopTextLength() override;
- RegExpNode* GetSuccessorOfOmnivorousTextNode(
- RegExpCompiler* compiler) override;
- void FillInBMInfo(Isolate* isolate, int offset, int budget,
- BoyerMooreLookahead* bm, bool not_at_start) override;
- void CalculateOffsets();
- RegExpNode* FilterOneByte(int depth) override;
-
- private:
- enum TextEmitPassType {
- NON_LATIN1_MATCH, // Check for characters that can't match.
- SIMPLE_CHARACTER_MATCH, // Case-dependent single character check.
- NON_LETTER_CHARACTER_MATCH, // Check characters that have no case equivs.
- CASE_CHARACTER_MATCH, // Case-independent single character check.
- CHARACTER_CLASS_MATCH // Character class.
- };
- static bool SkipPass(TextEmitPassType pass, bool ignore_case);
- static const int kFirstRealPass = SIMPLE_CHARACTER_MATCH;
- static const int kLastPass = CHARACTER_CLASS_MATCH;
- void TextEmitPass(RegExpCompiler* compiler,
- TextEmitPassType pass,
- bool preloaded,
- Trace* trace,
- bool first_element_checked,
- int* checked_up_to);
- int Length();
- ZoneList<TextElement>* elms_;
- bool read_backward_;
-};
-
-
-class AssertionNode: public SeqRegExpNode {
- public:
- enum AssertionType {
- AT_END,
- AT_START,
- AT_BOUNDARY,
- AT_NON_BOUNDARY,
- AFTER_NEWLINE
- };
- static AssertionNode* AtEnd(RegExpNode* on_success) {
- return new(on_success->zone()) AssertionNode(AT_END, on_success);
- }
- static AssertionNode* AtStart(RegExpNode* on_success) {
- return new(on_success->zone()) AssertionNode(AT_START, on_success);
- }
- static AssertionNode* AtBoundary(RegExpNode* on_success) {
- return new(on_success->zone()) AssertionNode(AT_BOUNDARY, on_success);
- }
- static AssertionNode* AtNonBoundary(RegExpNode* on_success) {
- return new(on_success->zone()) AssertionNode(AT_NON_BOUNDARY, on_success);
- }
- static AssertionNode* AfterNewline(RegExpNode* on_success) {
- return new(on_success->zone()) AssertionNode(AFTER_NEWLINE, on_success);
- }
- void Accept(NodeVisitor* visitor) override;
- void Emit(RegExpCompiler* compiler, Trace* trace) override;
- int EatsAtLeast(int still_to_find, int budget, bool not_at_start) override;
- void GetQuickCheckDetails(QuickCheckDetails* details,
- RegExpCompiler* compiler, int filled_in,
- bool not_at_start) override;
- void FillInBMInfo(Isolate* isolate, int offset, int budget,
- BoyerMooreLookahead* bm, bool not_at_start) override;
- AssertionType assertion_type() { return assertion_type_; }
-
- private:
- void EmitBoundaryCheck(RegExpCompiler* compiler, Trace* trace);
- enum IfPrevious { kIsNonWord, kIsWord };
- void BacktrackIfPrevious(RegExpCompiler* compiler,
- Trace* trace,
- IfPrevious backtrack_if_previous);
- AssertionNode(AssertionType t, RegExpNode* on_success)
- : SeqRegExpNode(on_success), assertion_type_(t) { }
- AssertionType assertion_type_;
-};
-
-
-class BackReferenceNode: public SeqRegExpNode {
- public:
- BackReferenceNode(int start_reg, int end_reg, JSRegExp::Flags flags,
- bool read_backward, RegExpNode* on_success)
- : SeqRegExpNode(on_success),
- start_reg_(start_reg),
- end_reg_(end_reg),
- flags_(flags),
- read_backward_(read_backward) {}
- void Accept(NodeVisitor* visitor) override;
- int start_register() { return start_reg_; }
- int end_register() { return end_reg_; }
- bool read_backward() { return read_backward_; }
- void Emit(RegExpCompiler* compiler, Trace* trace) override;
- int EatsAtLeast(int still_to_find, int recursion_depth,
- bool not_at_start) override;
- void GetQuickCheckDetails(QuickCheckDetails* details,
- RegExpCompiler* compiler, int characters_filled_in,
- bool not_at_start) override {
- return;
- }
- void FillInBMInfo(Isolate* isolate, int offset, int budget,
- BoyerMooreLookahead* bm, bool not_at_start) override;
-
- private:
- int start_reg_;
- int end_reg_;
- JSRegExp::Flags flags_;
- bool read_backward_;
-};
-
-
-class EndNode: public RegExpNode {
- public:
- enum Action { ACCEPT, BACKTRACK, NEGATIVE_SUBMATCH_SUCCESS };
- EndNode(Action action, Zone* zone) : RegExpNode(zone), action_(action) {}
- void Accept(NodeVisitor* visitor) override;
- void Emit(RegExpCompiler* compiler, Trace* trace) override;
- int EatsAtLeast(int still_to_find, int recursion_depth,
- bool not_at_start) override {
- return 0;
- }
- void GetQuickCheckDetails(QuickCheckDetails* details,
- RegExpCompiler* compiler, int characters_filled_in,
- bool not_at_start) override {
- // Returning 0 from EatsAtLeast should ensure we never get here.
- UNREACHABLE();
- }
- void FillInBMInfo(Isolate* isolate, int offset, int budget,
- BoyerMooreLookahead* bm, bool not_at_start) override {
- // Returning 0 from EatsAtLeast should ensure we never get here.
- UNREACHABLE();
- }
-
- private:
- Action action_;
-};
-
-
-class NegativeSubmatchSuccess: public EndNode {
- public:
- NegativeSubmatchSuccess(int stack_pointer_reg,
- int position_reg,
- int clear_capture_count,
- int clear_capture_start,
- Zone* zone)
- : EndNode(NEGATIVE_SUBMATCH_SUCCESS, zone),
- stack_pointer_register_(stack_pointer_reg),
- current_position_register_(position_reg),
- clear_capture_count_(clear_capture_count),
- clear_capture_start_(clear_capture_start) { }
- void Emit(RegExpCompiler* compiler, Trace* trace) override;
-
- private:
- int stack_pointer_register_;
- int current_position_register_;
- int clear_capture_count_;
- int clear_capture_start_;
-};
-
-
-class Guard: public ZoneObject {
- public:
- enum Relation { LT, GEQ };
- Guard(int reg, Relation op, int value)
- : reg_(reg),
- op_(op),
- value_(value) { }
- int reg() { return reg_; }
- Relation op() { return op_; }
- int value() { return value_; }
-
- private:
- int reg_;
- Relation op_;
- int value_;
-};
-
-
-class GuardedAlternative {
- public:
- explicit GuardedAlternative(RegExpNode* node)
- : node_(node), guards_(nullptr) {}
- void AddGuard(Guard* guard, Zone* zone);
- RegExpNode* node() { return node_; }
- void set_node(RegExpNode* node) { node_ = node; }
- ZoneList<Guard*>* guards() { return guards_; }
-
- private:
- RegExpNode* node_;
- ZoneList<Guard*>* guards_;
-};
-
-
-class AlternativeGeneration;
-
-
-class ChoiceNode: public RegExpNode {
- public:
- explicit ChoiceNode(int expected_size, Zone* zone)
- : RegExpNode(zone),
- alternatives_(new (zone)
- ZoneList<GuardedAlternative>(expected_size, zone)),
- table_(nullptr),
- not_at_start_(false),
- being_calculated_(false) {}
- void Accept(NodeVisitor* visitor) override;
- void AddAlternative(GuardedAlternative node) {
- alternatives()->Add(node, zone());
- }
- ZoneList<GuardedAlternative>* alternatives() { return alternatives_; }
- DispatchTable* GetTable(bool ignore_case);
- void Emit(RegExpCompiler* compiler, Trace* trace) override;
- int EatsAtLeast(int still_to_find, int budget, bool not_at_start) override;
- int EatsAtLeastHelper(int still_to_find,
- int budget,
- RegExpNode* ignore_this_node,
- bool not_at_start);
- void GetQuickCheckDetails(QuickCheckDetails* details,
- RegExpCompiler* compiler, int characters_filled_in,
- bool not_at_start) override;
- void FillInBMInfo(Isolate* isolate, int offset, int budget,
- BoyerMooreLookahead* bm, bool not_at_start) override;
-
- bool being_calculated() { return being_calculated_; }
- bool not_at_start() { return not_at_start_; }
- void set_not_at_start() { not_at_start_ = true; }
- void set_being_calculated(bool b) { being_calculated_ = b; }
- virtual bool try_to_emit_quick_check_for_alternative(bool is_first) {
- return true;
- }
- RegExpNode* FilterOneByte(int depth) override;
- virtual bool read_backward() { return false; }
-
- protected:
- int GreedyLoopTextLengthForAlternative(GuardedAlternative* alternative);
- ZoneList<GuardedAlternative>* alternatives_;
-
- private:
- friend class DispatchTableConstructor;
- friend class Analysis;
- void GenerateGuard(RegExpMacroAssembler* macro_assembler,
- Guard* guard,
- Trace* trace);
- int CalculatePreloadCharacters(RegExpCompiler* compiler, int eats_at_least);
- void EmitOutOfLineContinuation(RegExpCompiler* compiler,
- Trace* trace,
- GuardedAlternative alternative,
- AlternativeGeneration* alt_gen,
- int preload_characters,
- bool next_expects_preload);
- void SetUpPreLoad(RegExpCompiler* compiler,
- Trace* current_trace,
- PreloadState* preloads);
- void AssertGuardsMentionRegisters(Trace* trace);
- int EmitOptimizedUnanchoredSearch(RegExpCompiler* compiler, Trace* trace);
- Trace* EmitGreedyLoop(RegExpCompiler* compiler,
- Trace* trace,
- AlternativeGenerationList* alt_gens,
- PreloadState* preloads,
- GreedyLoopState* greedy_loop_state,
- int text_length);
- void EmitChoices(RegExpCompiler* compiler,
- AlternativeGenerationList* alt_gens,
- int first_choice,
- Trace* trace,
- PreloadState* preloads);
- DispatchTable* table_;
- // If true, this node is never checked at the start of the input.
- // Allows a new trace to start with at_start() set to false.
- bool not_at_start_;
- bool being_calculated_;
-};
-
-
-class NegativeLookaroundChoiceNode : public ChoiceNode {
- public:
- explicit NegativeLookaroundChoiceNode(GuardedAlternative this_must_fail,
- GuardedAlternative then_do_this,
- Zone* zone)
- : ChoiceNode(2, zone) {
- AddAlternative(this_must_fail);
- AddAlternative(then_do_this);
- }
- int EatsAtLeast(int still_to_find, int budget, bool not_at_start) override;
- void GetQuickCheckDetails(QuickCheckDetails* details,
- RegExpCompiler* compiler, int characters_filled_in,
- bool not_at_start) override;
- void FillInBMInfo(Isolate* isolate, int offset, int budget,
- BoyerMooreLookahead* bm, bool not_at_start) override {
- alternatives_->at(1).node()->FillInBMInfo(isolate, offset, budget - 1, bm,
- not_at_start);
- if (offset == 0) set_bm_info(not_at_start, bm);
- }
- // For a negative lookahead we don't emit the quick check for the
- // alternative that is expected to fail. This is because quick check code
- // starts by loading enough characters for the alternative that takes fewest
- // characters, but on a negative lookahead the negative branch did not take
- // part in that calculation (EatsAtLeast) so the assumptions don't hold.
- bool try_to_emit_quick_check_for_alternative(bool is_first) override {
- return !is_first;
- }
- RegExpNode* FilterOneByte(int depth) override;
-};
-
-
-class LoopChoiceNode: public ChoiceNode {
- public:
- LoopChoiceNode(bool body_can_be_zero_length, bool read_backward, Zone* zone)
- : ChoiceNode(2, zone),
- loop_node_(nullptr),
- continue_node_(nullptr),
- body_can_be_zero_length_(body_can_be_zero_length),
- read_backward_(read_backward) {}
- void AddLoopAlternative(GuardedAlternative alt);
- void AddContinueAlternative(GuardedAlternative alt);
- void Emit(RegExpCompiler* compiler, Trace* trace) override;
- int EatsAtLeast(int still_to_find, int budget, bool not_at_start) override;
- void GetQuickCheckDetails(QuickCheckDetails* details,
- RegExpCompiler* compiler, int characters_filled_in,
- bool not_at_start) override;
- void FillInBMInfo(Isolate* isolate, int offset, int budget,
- BoyerMooreLookahead* bm, bool not_at_start) override;
- RegExpNode* loop_node() { return loop_node_; }
- RegExpNode* continue_node() { return continue_node_; }
- bool body_can_be_zero_length() { return body_can_be_zero_length_; }
- bool read_backward() override { return read_backward_; }
- void Accept(NodeVisitor* visitor) override;
- RegExpNode* FilterOneByte(int depth) override;
-
- private:
- // AddAlternative is made private for loop nodes because alternatives
- // should not be added freely, we need to keep track of which node
- // goes back to the node itself.
- void AddAlternative(GuardedAlternative node) {
- ChoiceNode::AddAlternative(node);
- }
-
- RegExpNode* loop_node_;
- RegExpNode* continue_node_;
- bool body_can_be_zero_length_;
- bool read_backward_;
-};
-
-
-// Improve the speed that we scan for an initial point where a non-anchored
-// regexp can match by using a Boyer-Moore-like table. This is done by
-// identifying non-greedy non-capturing loops in the nodes that eat any
-// character one at a time. For example in the middle of the regexp
-// /foo[\s\S]*?bar/ we find such a loop. There is also such a loop implicitly
-// inserted at the start of any non-anchored regexp.
-//
-// When we have found such a loop we look ahead in the nodes to find the set of
-// characters that can come at given distances. For example for the regexp
-// /.?foo/ we know that there are at least 3 characters ahead of us, and the
-// sets of characters that can occur are [any, [f, o], [o]]. We find a range in
-// the lookahead info where the set of characters is reasonably constrained. In
-// our example this is from index 1 to 2 (0 is not constrained). We can now
-// look 3 characters ahead and if we don't find one of [f, o] (the union of
-// [f, o] and [o]) then we can skip forwards by the range size (in this case 2).
-//
-// For Unicode input strings we do the same, but modulo 128.
-//
-// We also look at the first string fed to the regexp and use that to get a hint
-// of the character frequencies in the inputs. This affects the assessment of
-// whether the set of characters is 'reasonably constrained'.
-//
-// We also have another lookahead mechanism (called quick check in the code),
-// which uses a wide load of multiple characters followed by a mask and compare
-// to determine whether a match is possible at this point.
-enum ContainedInLattice {
- kNotYet = 0,
- kLatticeIn = 1,
- kLatticeOut = 2,
- kLatticeUnknown = 3 // Can also mean both in and out.
-};
-
-
-inline ContainedInLattice Combine(ContainedInLattice a, ContainedInLattice b) {
- return static_cast<ContainedInLattice>(a | b);
-}
-
-
-ContainedInLattice AddRange(ContainedInLattice a,
- const int* ranges,
- int ranges_size,
- Interval new_range);
-
-
-class BoyerMoorePositionInfo : public ZoneObject {
- public:
- explicit BoyerMoorePositionInfo(Zone* zone)
- : map_(new(zone) ZoneList<bool>(kMapSize, zone)),
- map_count_(0),
- w_(kNotYet),
- s_(kNotYet),
- d_(kNotYet),
- surrogate_(kNotYet) {
- for (int i = 0; i < kMapSize; i++) {
- map_->Add(false, zone);
- }
- }
-
- bool& at(int i) { return map_->at(i); }
-
- static const int kMapSize = 128;
- static const int kMask = kMapSize - 1;
-
- int map_count() const { return map_count_; }
-
- void Set(int character);
- void SetInterval(const Interval& interval);
- void SetAll();
- bool is_non_word() { return w_ == kLatticeOut; }
- bool is_word() { return w_ == kLatticeIn; }
-
- private:
- ZoneList<bool>* map_;
- int map_count_; // Number of set bits in the map.
- ContainedInLattice w_; // The \w character class.
- ContainedInLattice s_; // The \s character class.
- ContainedInLattice d_; // The \d character class.
- ContainedInLattice surrogate_; // Surrogate UTF-16 code units.
-};
-
-
-class BoyerMooreLookahead : public ZoneObject {
- public:
- BoyerMooreLookahead(int length, RegExpCompiler* compiler, Zone* zone);
-
- int length() { return length_; }
- int max_char() { return max_char_; }
- RegExpCompiler* compiler() { return compiler_; }
-
- int Count(int map_number) {
- return bitmaps_->at(map_number)->map_count();
- }
-
- BoyerMoorePositionInfo* at(int i) { return bitmaps_->at(i); }
-
- void Set(int map_number, int character) {
- if (character > max_char_) return;
- BoyerMoorePositionInfo* info = bitmaps_->at(map_number);
- info->Set(character);
- }
-
- void SetInterval(int map_number, const Interval& interval) {
- if (interval.from() > max_char_) return;
- BoyerMoorePositionInfo* info = bitmaps_->at(map_number);
- if (interval.to() > max_char_) {
- info->SetInterval(Interval(interval.from(), max_char_));
- } else {
- info->SetInterval(interval);
- }
- }
-
- void SetAll(int map_number) {
- bitmaps_->at(map_number)->SetAll();
- }
-
- void SetRest(int from_map) {
- for (int i = from_map; i < length_; i++) SetAll(i);
- }
- void EmitSkipInstructions(RegExpMacroAssembler* masm);
-
- private:
- // This is the value obtained by EatsAtLeast. If we do not have at least this
- // many characters left in the sample string then the match is bound to fail.
- // Therefore it is OK to read a character this far ahead of the current match
- // point.
- int length_;
- RegExpCompiler* compiler_;
- // 0xff for Latin1, 0xffff for UTF-16.
- int max_char_;
- ZoneList<BoyerMoorePositionInfo*>* bitmaps_;
-
- int GetSkipTable(int min_lookahead,
- int max_lookahead,
- Handle<ByteArray> boolean_skip_table);
- bool FindWorthwhileInterval(int* from, int* to);
- int FindBestInterval(
- int max_number_of_chars, int old_biggest_points, int* from, int* to);
-};
-
-
-// There are many ways to generate code for a node. This class encapsulates
-// the current way we should be generating. In other words it encapsulates
-// the current state of the code generator. The effect of this is that we
-// generate code for paths that the matcher can take through the regular
-// expression. A given node in the regexp can be code-generated several times
-// as it can be part of several traces. For example for the regexp:
-// /foo(bar|ip)baz/ the code to match baz will be generated twice, once as part
-// of the foo-bar-baz trace and once as part of the foo-ip-baz trace. The code
-// to match foo is generated only once (the traces have a common prefix). The
-// code to store the capture is deferred and generated (twice) after the places
-// where baz has been matched.
-class Trace {
- public:
- // A value for a property that is either known to be true, know to be false,
- // or not known.
- enum TriBool {
- UNKNOWN = -1, FALSE_VALUE = 0, TRUE_VALUE = 1
- };
-
- class DeferredAction {
- public:
- DeferredAction(ActionNode::ActionType action_type, int reg)
- : action_type_(action_type), reg_(reg), next_(nullptr) {}
- DeferredAction* next() { return next_; }
- bool Mentions(int reg);
- int reg() { return reg_; }
- ActionNode::ActionType action_type() { return action_type_; }
- private:
- ActionNode::ActionType action_type_;
- int reg_;
- DeferredAction* next_;
- friend class Trace;
- };
-
- class DeferredCapture : public DeferredAction {
- public:
- DeferredCapture(int reg, bool is_capture, Trace* trace)
- : DeferredAction(ActionNode::STORE_POSITION, reg),
- cp_offset_(trace->cp_offset()),
- is_capture_(is_capture) { }
- int cp_offset() { return cp_offset_; }
- bool is_capture() { return is_capture_; }
- private:
- int cp_offset_;
- bool is_capture_;
- void set_cp_offset(int cp_offset) { cp_offset_ = cp_offset; }
- };
-
- class DeferredSetRegister : public DeferredAction {
- public:
- DeferredSetRegister(int reg, int value)
- : DeferredAction(ActionNode::SET_REGISTER, reg),
- value_(value) { }
- int value() { return value_; }
- private:
- int value_;
- };
-
- class DeferredClearCaptures : public DeferredAction {
- public:
- explicit DeferredClearCaptures(Interval range)
- : DeferredAction(ActionNode::CLEAR_CAPTURES, -1),
- range_(range) { }
- Interval range() { return range_; }
- private:
- Interval range_;
- };
-
- class DeferredIncrementRegister : public DeferredAction {
- public:
- explicit DeferredIncrementRegister(int reg)
- : DeferredAction(ActionNode::INCREMENT_REGISTER, reg) { }
- };
-
- Trace()
- : cp_offset_(0),
- actions_(nullptr),
- backtrack_(nullptr),
- stop_node_(nullptr),
- loop_label_(nullptr),
- characters_preloaded_(0),
- bound_checked_up_to_(0),
- flush_budget_(100),
- at_start_(UNKNOWN) {}
-
- // End the trace. This involves flushing the deferred actions in the trace
- // and pushing a backtrack location onto the backtrack stack. Once this is
- // done we can start a new trace or go to one that has already been
- // generated.
- void Flush(RegExpCompiler* compiler, RegExpNode* successor);
- int cp_offset() { return cp_offset_; }
- DeferredAction* actions() { return actions_; }
- // A trivial trace is one that has no deferred actions or other state that
- // affects the assumptions used when generating code. There is no recorded
- // backtrack location in a trivial trace, so with a trivial trace we will
- // generate code that, on a failure to match, gets the backtrack location
- // from the backtrack stack rather than using a direct jump instruction. We
- // always start code generation with a trivial trace and non-trivial traces
- // are created as we emit code for nodes or add to the list of deferred
- // actions in the trace. The location of the code generated for a node using
- // a trivial trace is recorded in a label in the node so that gotos can be
- // generated to that code.
- bool is_trivial() {
- return backtrack_ == nullptr && actions_ == nullptr && cp_offset_ == 0 &&
- characters_preloaded_ == 0 && bound_checked_up_to_ == 0 &&
- quick_check_performed_.characters() == 0 && at_start_ == UNKNOWN;
- }
- TriBool at_start() { return at_start_; }
- void set_at_start(TriBool at_start) { at_start_ = at_start; }
- Label* backtrack() { return backtrack_; }
- Label* loop_label() { return loop_label_; }
- RegExpNode* stop_node() { return stop_node_; }
- int characters_preloaded() { return characters_preloaded_; }
- int bound_checked_up_to() { return bound_checked_up_to_; }
- int flush_budget() { return flush_budget_; }
- QuickCheckDetails* quick_check_performed() { return &quick_check_performed_; }
- bool mentions_reg(int reg);
- // Returns true if a deferred position store exists to the specified
- // register and stores the offset in the out-parameter. Otherwise
- // returns false.
- bool GetStoredPosition(int reg, int* cp_offset);
- // These set methods and AdvanceCurrentPositionInTrace should be used only on
- // new traces - the intention is that traces are immutable after creation.
- void add_action(DeferredAction* new_action) {
- DCHECK(new_action->next_ == nullptr);
- new_action->next_ = actions_;
- actions_ = new_action;
- }
- void set_backtrack(Label* backtrack) { backtrack_ = backtrack; }
- void set_stop_node(RegExpNode* node) { stop_node_ = node; }
- void set_loop_label(Label* label) { loop_label_ = label; }
- void set_characters_preloaded(int count) { characters_preloaded_ = count; }
- void set_bound_checked_up_to(int to) { bound_checked_up_to_ = to; }
- void set_flush_budget(int to) { flush_budget_ = to; }
- void set_quick_check_performed(QuickCheckDetails* d) {
- quick_check_performed_ = *d;
- }
- void InvalidateCurrentCharacter();
- void AdvanceCurrentPositionInTrace(int by, RegExpCompiler* compiler);
-
- private:
- int FindAffectedRegisters(OutSet* affected_registers, Zone* zone);
- void PerformDeferredActions(RegExpMacroAssembler* macro,
- int max_register,
- const OutSet& affected_registers,
- OutSet* registers_to_pop,
- OutSet* registers_to_clear,
- Zone* zone);
- void RestoreAffectedRegisters(RegExpMacroAssembler* macro,
- int max_register,
- const OutSet& registers_to_pop,
- const OutSet& registers_to_clear);
- int cp_offset_;
- DeferredAction* actions_;
- Label* backtrack_;
- RegExpNode* stop_node_;
- Label* loop_label_;
- int characters_preloaded_;
- int bound_checked_up_to_;
- QuickCheckDetails quick_check_performed_;
- int flush_budget_;
- TriBool at_start_;
-};
-
-
-class GreedyLoopState {
- public:
- explicit GreedyLoopState(bool not_at_start);
-
- Label* label() { return &label_; }
- Trace* counter_backtrack_trace() { return &counter_backtrack_trace_; }
-
- private:
- Label label_;
- Trace counter_backtrack_trace_;
-};
-
-
-struct PreloadState {
- static const int kEatsAtLeastNotYetInitialized = -1;
- bool preload_is_current_;
- bool preload_has_checked_bounds_;
- int preload_characters_;
- int eats_at_least_;
- void init() {
- eats_at_least_ = kEatsAtLeastNotYetInitialized;
- }
-};
-
-
-class NodeVisitor {
- public:
- virtual ~NodeVisitor() = default;
-#define DECLARE_VISIT(Type) \
- virtual void Visit##Type(Type##Node* that) = 0;
-FOR_EACH_NODE_TYPE(DECLARE_VISIT)
-#undef DECLARE_VISIT
- virtual void VisitLoopChoice(LoopChoiceNode* that) { VisitChoice(that); }
-};
-
-
-// Node visitor used to add the start set of the alternatives to the
-// dispatch table of a choice node.
-class V8_EXPORT_PRIVATE DispatchTableConstructor : public NodeVisitor {
- public:
- DispatchTableConstructor(DispatchTable* table, bool ignore_case,
- Zone* zone)
- : table_(table),
- choice_index_(-1),
- ignore_case_(ignore_case),
- zone_(zone) { }
-
- void BuildTable(ChoiceNode* node);
-
- void AddRange(CharacterRange range) {
- table()->AddRange(range, choice_index_, zone_);
- }
-
- void AddInverse(ZoneList<CharacterRange>* ranges);
-
-#define DECLARE_VISIT(Type) \
- virtual void Visit##Type(Type##Node* that);
-FOR_EACH_NODE_TYPE(DECLARE_VISIT)
-#undef DECLARE_VISIT
-
- DispatchTable* table() { return table_; }
- void set_choice_index(int value) { choice_index_ = value; }
-
- protected:
- DispatchTable* table_;
- int choice_index_;
- bool ignore_case_;
- Zone* zone_;
-};
-
-// Assertion propagation moves information about assertions such as
-// \b to the affected nodes. For instance, in /.\b./ information must
-// be propagated to the first '.' that whatever follows needs to know
-// if it matched a word or a non-word, and to the second '.' that it
-// has to check if it succeeds a word or non-word. In this case the
-// result will be something like:
-//
-// +-------+ +------------+
-// | . | | . |
-// +-------+ ---> +------------+
-// | word? | | check word |
-// +-------+ +------------+
-class Analysis: public NodeVisitor {
- public:
- Analysis(Isolate* isolate, bool is_one_byte)
- : isolate_(isolate), is_one_byte_(is_one_byte), error_message_(nullptr) {}
- void EnsureAnalyzed(RegExpNode* node);
-
-#define DECLARE_VISIT(Type) void Visit##Type(Type##Node* that) override;
- FOR_EACH_NODE_TYPE(DECLARE_VISIT)
-#undef DECLARE_VISIT
- void VisitLoopChoice(LoopChoiceNode* that) override;
-
- bool has_failed() { return error_message_ != nullptr; }
- const char* error_message() {
- DCHECK(error_message_ != nullptr);
- return error_message_;
- }
- void fail(const char* error_message) {
- error_message_ = error_message;
- }
-
- Isolate* isolate() const { return isolate_; }
-
- private:
- Isolate* isolate_;
- bool is_one_byte_;
- const char* error_message_;
-
- DISALLOW_IMPLICIT_CONSTRUCTORS(Analysis);
-};
-
-
-struct RegExpCompileData {
- RegExpCompileData()
- : tree(nullptr),
- node(nullptr),
- simple(true),
- contains_anchor(false),
- capture_count(0) {}
- RegExpTree* tree;
- RegExpNode* node;
- bool simple;
- bool contains_anchor;
- Handle<FixedArray> capture_name_map;
- Handle<String> error;
- int capture_count;
-};
-
-
-class RegExpEngine: public AllStatic {
- public:
- struct CompilationResult {
- inline CompilationResult(Isolate* isolate, const char* error_message);
- CompilationResult(Object code, int registers)
- : code(code), num_registers(registers) {}
- const char* const error_message = nullptr;
- Object const code;
- int const num_registers = 0;
- };
-
- V8_EXPORT_PRIVATE static CompilationResult Compile(
- Isolate* isolate, Zone* zone, RegExpCompileData* input,
- JSRegExp::Flags flags, Handle<String> pattern,
- Handle<String> sample_subject, bool is_one_byte);
-
- static bool TooMuchRegExpCode(Isolate* isolate, Handle<String> pattern);
-
- V8_EXPORT_PRIVATE static void DotPrint(const char* label, RegExpNode* node,
- bool ignore_case);
-};
-
-
-class RegExpResultsCache : public AllStatic {
- public:
- enum ResultsCacheType { REGEXP_MULTIPLE_INDICES, STRING_SPLIT_SUBSTRINGS };
-
- // Attempt to retrieve a cached result. On failure, 0 is returned as a Smi.
- // On success, the returned result is guaranteed to be a COW-array.
- static Object Lookup(Heap* heap, String key_string, Object key_pattern,
- FixedArray* last_match_out, ResultsCacheType type);
- // Attempt to add value_array to the cache specified by type. On success,
- // value_array is turned into a COW-array.
- static void Enter(Isolate* isolate, Handle<String> key_string,
- Handle<Object> key_pattern, Handle<FixedArray> value_array,
- Handle<FixedArray> last_match_cache, ResultsCacheType type);
- static void Clear(FixedArray cache);
- static const int kRegExpResultsCacheSize = 0x100;
-
- private:
- static const int kArrayEntriesPerCacheEntry = 4;
- static const int kStringOffset = 0;
- static const int kPatternOffset = 1;
- static const int kArrayOffset = 2;
- static const int kLastMatchOffset = 3;
-};
-
-} // namespace internal
-} // namespace v8
-
-#endif // V8_REGEXP_JSREGEXP_H_
diff --git a/deps/v8/src/regexp/regexp-ast.h b/deps/v8/src/regexp/regexp-ast.h
index 1fa9f7a35b..aab67cad15 100644
--- a/deps/v8/src/regexp/regexp-ast.h
+++ b/deps/v8/src/regexp/regexp-ast.h
@@ -50,7 +50,7 @@ class RegExpVisitor {
// A simple closed interval.
class Interval {
public:
- Interval() : from_(kNone), to_(kNone) {}
+ Interval() : from_(kNone), to_(kNone - 1) {} // '- 1' for branchless size().
Interval(int from, int to) : from_(from), to_(to) {}
Interval Union(Interval that) {
if (that.from_ == kNone)
@@ -60,12 +60,16 @@ class Interval {
else
return Interval(Min(from_, that.from_), Max(to_, that.to_));
}
+
bool Contains(int value) { return (from_ <= value) && (value <= to_); }
bool is_empty() { return from_ == kNone; }
int from() const { return from_; }
int to() const { return to_; }
+ int size() const { return to_ - from_ + 1; }
+
static Interval Empty() { return Interval(); }
- static const int kNone = -1;
+
+ static constexpr int kNone = -1;
private:
int from_;
@@ -268,12 +272,13 @@ class RegExpAlternative final : public RegExpTree {
class RegExpAssertion final : public RegExpTree {
public:
enum AssertionType {
- START_OF_LINE,
- START_OF_INPUT,
- END_OF_LINE,
- END_OF_INPUT,
- BOUNDARY,
- NON_BOUNDARY
+ START_OF_LINE = 0,
+ START_OF_INPUT = 1,
+ END_OF_LINE = 2,
+ END_OF_INPUT = 3,
+ BOUNDARY = 4,
+ NON_BOUNDARY = 5,
+ LAST_TYPE = NON_BOUNDARY,
};
RegExpAssertion(AssertionType type, JSRegExp::Flags flags)
: assertion_type_(type), flags_(flags) {}
@@ -285,7 +290,8 @@ class RegExpAssertion final : public RegExpTree {
bool IsAnchoredAtEnd() override;
int min_match() override { return 0; }
int max_match() override { return 0; }
- AssertionType assertion_type() { return assertion_type_; }
+ AssertionType assertion_type() const { return assertion_type_; }
+ JSRegExp::Flags flags() const { return flags_; }
private:
const AssertionType assertion_type_;
diff --git a/deps/v8/src/regexp/regexp-macro-assembler-irregexp-inl.h b/deps/v8/src/regexp/regexp-bytecode-generator-inl.h
index cda48aa00b..bd906fea15 100644
--- a/deps/v8/src/regexp/regexp-macro-assembler-irregexp-inl.h
+++ b/deps/v8/src/regexp/regexp-bytecode-generator-inl.h
@@ -2,30 +2,28 @@
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.
-#ifndef V8_REGEXP_REGEXP_MACRO_ASSEMBLER_IRREGEXP_INL_H_
-#define V8_REGEXP_REGEXP_MACRO_ASSEMBLER_IRREGEXP_INL_H_
+#ifndef V8_REGEXP_REGEXP_BYTECODE_GENERATOR_INL_H_
+#define V8_REGEXP_REGEXP_BYTECODE_GENERATOR_INL_H_
-#include "src/regexp/regexp-macro-assembler-irregexp.h"
+#include "src/regexp/regexp-bytecode-generator.h"
#include "src/ast/ast.h"
-#include "src/regexp/bytecodes-irregexp.h"
+#include "src/regexp/regexp-bytecodes.h"
namespace v8 {
namespace internal {
-void RegExpMacroAssemblerIrregexp::Emit(uint32_t byte,
- uint32_t twenty_four_bits) {
+void RegExpBytecodeGenerator::Emit(uint32_t byte, uint32_t twenty_four_bits) {
uint32_t word = ((twenty_four_bits << BYTECODE_SHIFT) | byte);
DCHECK(pc_ <= buffer_.length());
- if (pc_ + 3 >= buffer_.length()) {
+ if (pc_ + 3 >= buffer_.length()) {
Expand();
}
*reinterpret_cast<uint32_t*>(buffer_.begin() + pc_) = word;
pc_ += 4;
}
-
-void RegExpMacroAssemblerIrregexp::Emit16(uint32_t word) {
+void RegExpBytecodeGenerator::Emit16(uint32_t word) {
DCHECK(pc_ <= buffer_.length());
if (pc_ + 1 >= buffer_.length()) {
Expand();
@@ -34,8 +32,7 @@ void RegExpMacroAssemblerIrregexp::Emit16(uint32_t word) {
pc_ += 2;
}
-
-void RegExpMacroAssemblerIrregexp::Emit8(uint32_t word) {
+void RegExpBytecodeGenerator::Emit8(uint32_t word) {
DCHECK(pc_ <= buffer_.length());
if (pc_ == buffer_.length()) {
Expand();
@@ -44,8 +41,7 @@ void RegExpMacroAssemblerIrregexp::Emit8(uint32_t word) {
pc_ += 1;
}
-
-void RegExpMacroAssemblerIrregexp::Emit32(uint32_t word) {
+void RegExpBytecodeGenerator::Emit32(uint32_t word) {
DCHECK(pc_ <= buffer_.length());
if (pc_ + 3 >= buffer_.length()) {
Expand();
@@ -57,4 +53,4 @@ void RegExpMacroAssemblerIrregexp::Emit32(uint32_t word) {
} // namespace internal
} // namespace v8
-#endif // V8_REGEXP_REGEXP_MACRO_ASSEMBLER_IRREGEXP_INL_H_
+#endif // V8_REGEXP_REGEXP_BYTECODE_GENERATOR_INL_H_
diff --git a/deps/v8/src/regexp/regexp-macro-assembler-irregexp.cc b/deps/v8/src/regexp/regexp-bytecode-generator.cc
index 712f00e509..ee3b4015d5 100644
--- a/deps/v8/src/regexp/regexp-macro-assembler-irregexp.cc
+++ b/deps/v8/src/regexp/regexp-bytecode-generator.cc
@@ -2,39 +2,35 @@
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.
-#include "src/regexp/regexp-macro-assembler-irregexp.h"
+#include "src/regexp/regexp-bytecode-generator.h"
#include "src/ast/ast.h"
#include "src/objects/objects-inl.h"
-#include "src/regexp/bytecodes-irregexp.h"
-#include "src/regexp/regexp-macro-assembler-irregexp-inl.h"
+#include "src/regexp/regexp-bytecode-generator-inl.h"
+#include "src/regexp/regexp-bytecodes.h"
#include "src/regexp/regexp-macro-assembler.h"
namespace v8 {
namespace internal {
-RegExpMacroAssemblerIrregexp::RegExpMacroAssemblerIrregexp(Isolate* isolate,
- Zone* zone)
+RegExpBytecodeGenerator::RegExpBytecodeGenerator(Isolate* isolate, Zone* zone)
: RegExpMacroAssembler(isolate, zone),
buffer_(Vector<byte>::New(1024)),
pc_(0),
- own_buffer_(true),
advance_current_end_(kInvalidPC),
isolate_(isolate) {}
-RegExpMacroAssemblerIrregexp::~RegExpMacroAssemblerIrregexp() {
+RegExpBytecodeGenerator::~RegExpBytecodeGenerator() {
if (backtrack_.is_linked()) backtrack_.Unuse();
- if (own_buffer_) buffer_.Dispose();
+ buffer_.Dispose();
}
-
-RegExpMacroAssemblerIrregexp::IrregexpImplementation
-RegExpMacroAssemblerIrregexp::Implementation() {
+RegExpBytecodeGenerator::IrregexpImplementation
+RegExpBytecodeGenerator::Implementation() {
return kBytecodeImplementation;
}
-
-void RegExpMacroAssemblerIrregexp::Bind(Label* l) {
+void RegExpBytecodeGenerator::Bind(Label* l) {
advance_current_end_ = kInvalidPC;
DCHECK(!l->is_bound());
if (l->is_linked()) {
@@ -48,8 +44,7 @@ void RegExpMacroAssemblerIrregexp::Bind(Label* l) {
l->bind_to(pc_);
}
-
-void RegExpMacroAssemblerIrregexp::EmitOrLink(Label* l) {
+void RegExpBytecodeGenerator::EmitOrLink(Label* l) {
if (l == nullptr) l = &backtrack_;
if (l->is_bound()) {
Emit32(l->pos());
@@ -63,102 +58,79 @@ void RegExpMacroAssemblerIrregexp::EmitOrLink(Label* l) {
}
}
-
-void RegExpMacroAssemblerIrregexp::PopRegister(int register_index) {
+void RegExpBytecodeGenerator::PopRegister(int register_index) {
DCHECK_LE(0, register_index);
DCHECK_GE(kMaxRegister, register_index);
Emit(BC_POP_REGISTER, register_index);
}
-
-void RegExpMacroAssemblerIrregexp::PushRegister(
- int register_index,
- StackCheckFlag check_stack_limit) {
+void RegExpBytecodeGenerator::PushRegister(int register_index,
+ StackCheckFlag check_stack_limit) {
DCHECK_LE(0, register_index);
DCHECK_GE(kMaxRegister, register_index);
Emit(BC_PUSH_REGISTER, register_index);
}
-
-void RegExpMacroAssemblerIrregexp::WriteCurrentPositionToRegister(
- int register_index, int cp_offset) {
+void RegExpBytecodeGenerator::WriteCurrentPositionToRegister(int register_index,
+ int cp_offset) {
DCHECK_LE(0, register_index);
DCHECK_GE(kMaxRegister, register_index);
Emit(BC_SET_REGISTER_TO_CP, register_index);
Emit32(cp_offset); // Current position offset.
}
-
-void RegExpMacroAssemblerIrregexp::ClearRegisters(int reg_from, int reg_to) {
+void RegExpBytecodeGenerator::ClearRegisters(int reg_from, int reg_to) {
DCHECK(reg_from <= reg_to);
for (int reg = reg_from; reg <= reg_to; reg++) {
SetRegister(reg, -1);
}
}
-
-void RegExpMacroAssemblerIrregexp::ReadCurrentPositionFromRegister(
+void RegExpBytecodeGenerator::ReadCurrentPositionFromRegister(
int register_index) {
DCHECK_LE(0, register_index);
DCHECK_GE(kMaxRegister, register_index);
Emit(BC_SET_CP_TO_REGISTER, register_index);
}
-
-void RegExpMacroAssemblerIrregexp::WriteStackPointerToRegister(
- int register_index) {
+void RegExpBytecodeGenerator::WriteStackPointerToRegister(int register_index) {
DCHECK_LE(0, register_index);
DCHECK_GE(kMaxRegister, register_index);
Emit(BC_SET_REGISTER_TO_SP, register_index);
}
-
-void RegExpMacroAssemblerIrregexp::ReadStackPointerFromRegister(
- int register_index) {
+void RegExpBytecodeGenerator::ReadStackPointerFromRegister(int register_index) {
DCHECK_LE(0, register_index);
DCHECK_GE(kMaxRegister, register_index);
Emit(BC_SET_SP_TO_REGISTER, register_index);
}
-
-void RegExpMacroAssemblerIrregexp::SetCurrentPositionFromEnd(int by) {
+void RegExpBytecodeGenerator::SetCurrentPositionFromEnd(int by) {
DCHECK(is_uint24(by));
Emit(BC_SET_CURRENT_POSITION_FROM_END, by);
}
-
-void RegExpMacroAssemblerIrregexp::SetRegister(int register_index, int to) {
+void RegExpBytecodeGenerator::SetRegister(int register_index, int to) {
DCHECK_LE(0, register_index);
DCHECK_GE(kMaxRegister, register_index);
Emit(BC_SET_REGISTER, register_index);
Emit32(to);
}
-
-void RegExpMacroAssemblerIrregexp::AdvanceRegister(int register_index, int by) {
+void RegExpBytecodeGenerator::AdvanceRegister(int register_index, int by) {
DCHECK_LE(0, register_index);
DCHECK_GE(kMaxRegister, register_index);
Emit(BC_ADVANCE_REGISTER, register_index);
Emit32(by);
}
+void RegExpBytecodeGenerator::PopCurrentPosition() { Emit(BC_POP_CP, 0); }
-void RegExpMacroAssemblerIrregexp::PopCurrentPosition() {
- Emit(BC_POP_CP, 0);
-}
+void RegExpBytecodeGenerator::PushCurrentPosition() { Emit(BC_PUSH_CP, 0); }
+void RegExpBytecodeGenerator::Backtrack() { Emit(BC_POP_BT, 0); }
-void RegExpMacroAssemblerIrregexp::PushCurrentPosition() {
- Emit(BC_PUSH_CP, 0);
-}
-
-
-void RegExpMacroAssemblerIrregexp::Backtrack() {
- Emit(BC_POP_BT, 0);
-}
-
-
-void RegExpMacroAssemblerIrregexp::GoTo(Label* l) {
+void RegExpBytecodeGenerator::GoTo(Label* l) {
if (advance_current_end_ == pc_) {
// Combine advance current and goto.
pc_ = advance_current_start_;
@@ -172,25 +144,19 @@ void RegExpMacroAssemblerIrregexp::GoTo(Label* l) {
}
}
-
-void RegExpMacroAssemblerIrregexp::PushBacktrack(Label* l) {
+void RegExpBytecodeGenerator::PushBacktrack(Label* l) {
Emit(BC_PUSH_BT, 0);
EmitOrLink(l);
}
-
-bool RegExpMacroAssemblerIrregexp::Succeed() {
+bool RegExpBytecodeGenerator::Succeed() {
Emit(BC_SUCCEED, 0);
return false; // Restart matching for global regexp not supported.
}
+void RegExpBytecodeGenerator::Fail() { Emit(BC_FAIL, 0); }
-void RegExpMacroAssemblerIrregexp::Fail() {
- Emit(BC_FAIL, 0);
-}
-
-
-void RegExpMacroAssemblerIrregexp::AdvanceCurrentPosition(int by) {
+void RegExpBytecodeGenerator::AdvanceCurrentPosition(int by) {
DCHECK_LE(kMinCPOffset, by);
DCHECK_GE(kMaxCPOffset, by);
advance_current_start_ = pc_;
@@ -199,18 +165,16 @@ void RegExpMacroAssemblerIrregexp::AdvanceCurrentPosition(int by) {
advance_current_end_ = pc_;
}
-
-void RegExpMacroAssemblerIrregexp::CheckGreedyLoop(
- Label* on_tos_equals_current_position) {
+void RegExpBytecodeGenerator::CheckGreedyLoop(
+ Label* on_tos_equals_current_position) {
Emit(BC_CHECK_GREEDY, 0);
EmitOrLink(on_tos_equals_current_position);
}
-
-void RegExpMacroAssemblerIrregexp::LoadCurrentCharacter(int cp_offset,
- Label* on_failure,
- bool check_bounds,
- int characters) {
+void RegExpBytecodeGenerator::LoadCurrentCharacter(int cp_offset,
+ Label* on_failure,
+ bool check_bounds,
+ int characters) {
DCHECK_LE(kMinCPOffset, cp_offset);
DCHECK_GE(kMaxCPOffset, cp_offset);
int bytecode;
@@ -237,22 +201,17 @@ void RegExpMacroAssemblerIrregexp::LoadCurrentCharacter(int cp_offset,
if (check_bounds) EmitOrLink(on_failure);
}
-
-void RegExpMacroAssemblerIrregexp::CheckCharacterLT(uc16 limit,
- Label* on_less) {
+void RegExpBytecodeGenerator::CheckCharacterLT(uc16 limit, Label* on_less) {
Emit(BC_CHECK_LT, limit);
EmitOrLink(on_less);
}
-
-void RegExpMacroAssemblerIrregexp::CheckCharacterGT(uc16 limit,
- Label* on_greater) {
+void RegExpBytecodeGenerator::CheckCharacterGT(uc16 limit, Label* on_greater) {
Emit(BC_CHECK_GT, limit);
EmitOrLink(on_greater);
}
-
-void RegExpMacroAssemblerIrregexp::CheckCharacter(uint32_t c, Label* on_equal) {
+void RegExpBytecodeGenerator::CheckCharacter(uint32_t c, Label* on_equal) {
if (c > MAX_FIRST_ARG) {
Emit(BC_CHECK_4_CHARS, 0);
Emit32(c);
@@ -262,22 +221,19 @@ void RegExpMacroAssemblerIrregexp::CheckCharacter(uint32_t c, Label* on_equal) {
EmitOrLink(on_equal);
}
-
-void RegExpMacroAssemblerIrregexp::CheckAtStart(Label* on_at_start) {
+void RegExpBytecodeGenerator::CheckAtStart(Label* on_at_start) {
Emit(BC_CHECK_AT_START, 0);
EmitOrLink(on_at_start);
}
-
-void RegExpMacroAssemblerIrregexp::CheckNotAtStart(int cp_offset,
- Label* on_not_at_start) {
+void RegExpBytecodeGenerator::CheckNotAtStart(int cp_offset,
+ Label* on_not_at_start) {
Emit(BC_CHECK_NOT_AT_START, cp_offset);
EmitOrLink(on_not_at_start);
}
-
-void RegExpMacroAssemblerIrregexp::CheckNotCharacter(uint32_t c,
- Label* on_not_equal) {
+void RegExpBytecodeGenerator::CheckNotCharacter(uint32_t c,
+ Label* on_not_equal) {
if (c > MAX_FIRST_ARG) {
Emit(BC_CHECK_NOT_4_CHARS, 0);
Emit32(c);
@@ -287,11 +243,8 @@ void RegExpMacroAssemblerIrregexp::CheckNotCharacter(uint32_t c,
EmitOrLink(on_not_equal);
}
-
-void RegExpMacroAssemblerIrregexp::CheckCharacterAfterAnd(
- uint32_t c,
- uint32_t mask,
- Label* on_equal) {
+void RegExpBytecodeGenerator::CheckCharacterAfterAnd(uint32_t c, uint32_t mask,
+ Label* on_equal) {
if (c > MAX_FIRST_ARG) {
Emit(BC_AND_CHECK_4_CHARS, 0);
Emit32(c);
@@ -302,11 +255,9 @@ void RegExpMacroAssemblerIrregexp::CheckCharacterAfterAnd(
EmitOrLink(on_equal);
}
-
-void RegExpMacroAssemblerIrregexp::CheckNotCharacterAfterAnd(
- uint32_t c,
- uint32_t mask,
- Label* on_not_equal) {
+void RegExpBytecodeGenerator::CheckNotCharacterAfterAnd(uint32_t c,
+ uint32_t mask,
+ Label* on_not_equal) {
if (c > MAX_FIRST_ARG) {
Emit(BC_AND_CHECK_NOT_4_CHARS, 0);
Emit32(c);
@@ -317,43 +268,32 @@ void RegExpMacroAssemblerIrregexp::CheckNotCharacterAfterAnd(
EmitOrLink(on_not_equal);
}
-
-void RegExpMacroAssemblerIrregexp::CheckNotCharacterAfterMinusAnd(
- uc16 c,
- uc16 minus,
- uc16 mask,
- Label* on_not_equal) {
+void RegExpBytecodeGenerator::CheckNotCharacterAfterMinusAnd(
+ uc16 c, uc16 minus, uc16 mask, Label* on_not_equal) {
Emit(BC_MINUS_AND_CHECK_NOT_CHAR, c);
Emit16(minus);
Emit16(mask);
EmitOrLink(on_not_equal);
}
-
-void RegExpMacroAssemblerIrregexp::CheckCharacterInRange(
- uc16 from,
- uc16 to,
- Label* on_in_range) {
+void RegExpBytecodeGenerator::CheckCharacterInRange(uc16 from, uc16 to,
+ Label* on_in_range) {
Emit(BC_CHECK_CHAR_IN_RANGE, 0);
Emit16(from);
Emit16(to);
EmitOrLink(on_in_range);
}
-
-void RegExpMacroAssemblerIrregexp::CheckCharacterNotInRange(
- uc16 from,
- uc16 to,
- Label* on_not_in_range) {
+void RegExpBytecodeGenerator::CheckCharacterNotInRange(uc16 from, uc16 to,
+ Label* on_not_in_range) {
Emit(BC_CHECK_CHAR_NOT_IN_RANGE, 0);
Emit16(from);
Emit16(to);
EmitOrLink(on_not_in_range);
}
-
-void RegExpMacroAssemblerIrregexp::CheckBitInTable(
- Handle<ByteArray> table, Label* on_bit_set) {
+void RegExpBytecodeGenerator::CheckBitInTable(Handle<ByteArray> table,
+ Label* on_bit_set) {
Emit(BC_CHECK_BIT_IN_TABLE, 0);
EmitOrLink(on_bit_set);
for (int i = 0; i < kTableSize; i += kBitsPerByte) {
@@ -365,10 +305,9 @@ void RegExpMacroAssemblerIrregexp::CheckBitInTable(
}
}
-
-void RegExpMacroAssemblerIrregexp::CheckNotBackReference(int start_reg,
- bool read_backward,
- Label* on_not_equal) {
+void RegExpBytecodeGenerator::CheckNotBackReference(int start_reg,
+ bool read_backward,
+ Label* on_not_equal) {
DCHECK_LE(0, start_reg);
DCHECK_GE(kMaxRegister, start_reg);
Emit(read_backward ? BC_CHECK_NOT_BACK_REF_BACKWARD : BC_CHECK_NOT_BACK_REF,
@@ -376,8 +315,7 @@ void RegExpMacroAssemblerIrregexp::CheckNotBackReference(int start_reg,
EmitOrLink(on_not_equal);
}
-
-void RegExpMacroAssemblerIrregexp::CheckNotBackReferenceIgnoreCase(
+void RegExpBytecodeGenerator::CheckNotBackReferenceIgnoreCase(
int start_reg, bool read_backward, bool unicode, Label* on_not_equal) {
DCHECK_LE(0, start_reg);
DCHECK_GE(kMaxRegister, start_reg);
@@ -389,10 +327,8 @@ void RegExpMacroAssemblerIrregexp::CheckNotBackReferenceIgnoreCase(
EmitOrLink(on_not_equal);
}
-
-void RegExpMacroAssemblerIrregexp::IfRegisterLT(int register_index,
- int comparand,
- Label* on_less_than) {
+void RegExpBytecodeGenerator::IfRegisterLT(int register_index, int comparand,
+ Label* on_less_than) {
DCHECK_LE(0, register_index);
DCHECK_GE(kMaxRegister, register_index);
Emit(BC_CHECK_REGISTER_LT, register_index);
@@ -400,10 +336,8 @@ void RegExpMacroAssemblerIrregexp::IfRegisterLT(int register_index,
EmitOrLink(on_less_than);
}
-
-void RegExpMacroAssemblerIrregexp::IfRegisterGE(int register_index,
- int comparand,
- Label* on_greater_or_equal) {
+void RegExpBytecodeGenerator::IfRegisterGE(int register_index, int comparand,
+ Label* on_greater_or_equal) {
DCHECK_LE(0, register_index);
DCHECK_GE(kMaxRegister, register_index);
Emit(BC_CHECK_REGISTER_GE, register_index);
@@ -411,18 +345,15 @@ void RegExpMacroAssemblerIrregexp::IfRegisterGE(int register_index,
EmitOrLink(on_greater_or_equal);
}
-
-void RegExpMacroAssemblerIrregexp::IfRegisterEqPos(int register_index,
- Label* on_eq) {
+void RegExpBytecodeGenerator::IfRegisterEqPos(int register_index,
+ Label* on_eq) {
DCHECK_LE(0, register_index);
DCHECK_GE(kMaxRegister, register_index);
Emit(BC_CHECK_REGISTER_EQ_POS, register_index);
EmitOrLink(on_eq);
}
-
-Handle<HeapObject> RegExpMacroAssemblerIrregexp::GetCode(
- Handle<String> source) {
+Handle<HeapObject> RegExpBytecodeGenerator::GetCode(Handle<String> source) {
Bind(&backtrack_);
Emit(BC_POP_BT, 0);
Handle<ByteArray> array = isolate_->factory()->NewByteArray(length());
@@ -430,25 +361,17 @@ Handle<HeapObject> RegExpMacroAssemblerIrregexp::GetCode(
return array;
}
+int RegExpBytecodeGenerator::length() { return pc_; }
-int RegExpMacroAssemblerIrregexp::length() {
- return pc_;
-}
-
-void RegExpMacroAssemblerIrregexp::Copy(byte* a) {
+void RegExpBytecodeGenerator::Copy(byte* a) {
MemCopy(a, buffer_.begin(), length());
}
-
-void RegExpMacroAssemblerIrregexp::Expand() {
- bool old_buffer_was_our_own = own_buffer_;
+void RegExpBytecodeGenerator::Expand() {
Vector<byte> old_buffer = buffer_;
buffer_ = Vector<byte>::New(old_buffer.length() * 2);
- own_buffer_ = true;
MemCopy(buffer_.begin(), old_buffer.begin(), old_buffer.length());
- if (old_buffer_was_our_own) {
- old_buffer.Dispose();
- }
+ old_buffer.Dispose();
}
} // namespace internal
diff --git a/deps/v8/src/regexp/regexp-macro-assembler-irregexp.h b/deps/v8/src/regexp/regexp-bytecode-generator.h
index 9e17dca415..b7207e977c 100644
--- a/deps/v8/src/regexp/regexp-macro-assembler-irregexp.h
+++ b/deps/v8/src/regexp/regexp-bytecode-generator.h
@@ -2,17 +2,16 @@
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.
-#ifndef V8_REGEXP_REGEXP_MACRO_ASSEMBLER_IRREGEXP_H_
-#define V8_REGEXP_REGEXP_MACRO_ASSEMBLER_IRREGEXP_H_
+#ifndef V8_REGEXP_REGEXP_BYTECODE_GENERATOR_H_
+#define V8_REGEXP_REGEXP_BYTECODE_GENERATOR_H_
#include "src/regexp/regexp-macro-assembler.h"
namespace v8 {
namespace internal {
-// A light-weight assembler for the Irregexp byte code.
-class V8_EXPORT_PRIVATE RegExpMacroAssemblerIrregexp
- : public RegExpMacroAssembler {
+// An assembler/generator for the Irregexp byte code.
+class V8_EXPORT_PRIVATE RegExpBytecodeGenerator : public RegExpMacroAssembler {
public:
// Create an assembler. Instructions and relocation information are emitted
// into a buffer, with the instructions starting from the beginning and the
@@ -22,8 +21,8 @@ class V8_EXPORT_PRIVATE RegExpMacroAssemblerIrregexp
// The assembler allocates and grows its own buffer, and buffer_size
// determines the initial buffer size. The buffer is owned by the assembler
// and deallocated upon destruction of the assembler.
- RegExpMacroAssemblerIrregexp(Isolate* isolate, Zone* zone);
- virtual ~RegExpMacroAssemblerIrregexp();
+ RegExpBytecodeGenerator(Isolate* isolate, Zone* zone);
+ virtual ~RegExpBytecodeGenerator();
// The byte-code interpreter checks on each push anyway.
virtual int stack_limit_slack() { return 1; }
virtual bool CanReadUnaligned() { return false; }
@@ -47,13 +46,11 @@ class V8_EXPORT_PRIVATE RegExpMacroAssemblerIrregexp
virtual void ReadCurrentPositionFromRegister(int reg);
virtual void WriteStackPointerToRegister(int reg);
virtual void ReadStackPointerFromRegister(int reg);
- virtual void LoadCurrentCharacter(int cp_offset,
- Label* on_end_of_input,
+ virtual void LoadCurrentCharacter(int cp_offset, Label* on_end_of_input,
bool check_bounds = true,
int characters = 1);
virtual void CheckCharacter(unsigned c, Label* on_equal);
- virtual void CheckCharacterAfterAnd(unsigned c,
- unsigned mask,
+ virtual void CheckCharacterAfterAnd(unsigned c, unsigned mask,
Label* on_equal);
virtual void CheckCharacterGT(uc16 limit, Label* on_greater);
virtual void CheckCharacterLT(uc16 limit, Label* on_less);
@@ -61,18 +58,12 @@ class V8_EXPORT_PRIVATE RegExpMacroAssemblerIrregexp
virtual void CheckAtStart(Label* on_at_start);
virtual void CheckNotAtStart(int cp_offset, Label* on_not_at_start);
virtual void CheckNotCharacter(unsigned c, Label* on_not_equal);
- virtual void CheckNotCharacterAfterAnd(unsigned c,
- unsigned mask,
+ virtual void CheckNotCharacterAfterAnd(unsigned c, unsigned mask,
Label* on_not_equal);
- virtual void CheckNotCharacterAfterMinusAnd(uc16 c,
- uc16 minus,
- uc16 mask,
+ virtual void CheckNotCharacterAfterMinusAnd(uc16 c, uc16 minus, uc16 mask,
Label* on_not_equal);
- virtual void CheckCharacterInRange(uc16 from,
- uc16 to,
- Label* on_in_range);
- virtual void CheckCharacterNotInRange(uc16 from,
- uc16 to,
+ virtual void CheckCharacterInRange(uc16 from, uc16 to, Label* on_in_range);
+ virtual void CheckCharacterNotInRange(uc16 from, uc16 to,
Label* on_not_in_range);
virtual void CheckBitInTable(Handle<ByteArray> table, Label* on_bit_set);
virtual void CheckNotBackReference(int start_reg, bool read_backward,
@@ -103,8 +94,6 @@ class V8_EXPORT_PRIVATE RegExpMacroAssemblerIrregexp
Vector<byte> buffer_;
// The program counter.
int pc_;
- // True if the assembler owns the buffer, false if buffer is external.
- bool own_buffer_;
Label backtrack_;
int advance_current_start_;
@@ -115,10 +104,10 @@ class V8_EXPORT_PRIVATE RegExpMacroAssemblerIrregexp
static const int kInvalidPC = -1;
- DISALLOW_IMPLICIT_CONSTRUCTORS(RegExpMacroAssemblerIrregexp);
+ DISALLOW_IMPLICIT_CONSTRUCTORS(RegExpBytecodeGenerator);
};
} // namespace internal
} // namespace v8
-#endif // V8_REGEXP_REGEXP_MACRO_ASSEMBLER_IRREGEXP_H_
+#endif // V8_REGEXP_REGEXP_BYTECODE_GENERATOR_H_
diff --git a/deps/v8/src/regexp/bytecodes-irregexp.h b/deps/v8/src/regexp/regexp-bytecodes.h
index a27c9a0a2b..8b1468c1bf 100644
--- a/deps/v8/src/regexp/bytecodes-irregexp.h
+++ b/deps/v8/src/regexp/regexp-bytecodes.h
@@ -2,14 +2,12 @@
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.
-
-#ifndef V8_REGEXP_BYTECODES_IRREGEXP_H_
-#define V8_REGEXP_BYTECODES_IRREGEXP_H_
+#ifndef V8_REGEXP_REGEXP_BYTECODES_H_
+#define V8_REGEXP_REGEXP_BYTECODES_H_
namespace v8 {
namespace internal {
-
const int BYTECODE_MASK = 0xff;
// The first argument is packed in with the byte code in one word, but so it
// has 24 bits, but it can be positive and negative so only use 23 bits for
@@ -71,8 +69,7 @@ const int BYTECODE_SHIFT = 8;
V(ADVANCE_CP_AND_GOTO, 50, 8) /* bc8 offset24 addr32 */ \
V(SET_CURRENT_POSITION_FROM_END, 51, 4) /* bc8 idx24 */
-#define DECLARE_BYTECODES(name, code, length) \
- static const int BC_##name = code;
+#define DECLARE_BYTECODES(name, code, length) static const int BC_##name = code;
BYTECODE_ITERATOR(DECLARE_BYTECODES)
#undef DECLARE_BYTECODES
@@ -84,4 +81,4 @@ BYTECODE_ITERATOR(DECLARE_BYTECODE_LENGTH)
} // namespace internal
} // namespace v8
-#endif // V8_REGEXP_BYTECODES_IRREGEXP_H_
+#endif // V8_REGEXP_REGEXP_BYTECODES_H_
diff --git a/deps/v8/src/regexp/regexp-compiler-tonode.cc b/deps/v8/src/regexp/regexp-compiler-tonode.cc
new file mode 100644
index 0000000000..d12c35682e
--- /dev/null
+++ b/deps/v8/src/regexp/regexp-compiler-tonode.cc
@@ -0,0 +1,1678 @@
+// Copyright 2019 the V8 project authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+#include "src/regexp/regexp-compiler.h"
+
+#include "src/execution/isolate.h"
+#include "src/regexp/regexp.h"
+#ifdef V8_INTL_SUPPORT
+#include "src/regexp/special-case.h"
+#endif // V8_INTL_SUPPORT
+#include "src/strings/unicode-inl.h"
+#include "src/zone/zone-list-inl.h"
+
+#ifdef V8_INTL_SUPPORT
+#include "unicode/locid.h"
+#include "unicode/uniset.h"
+#include "unicode/utypes.h"
+#endif // V8_INTL_SUPPORT
+
+namespace v8 {
+namespace internal {
+
+using namespace regexp_compiler_constants; // NOLINT(build/namespaces)
+
+// -------------------------------------------------------------------
+// Tree to graph conversion
+
+RegExpNode* RegExpAtom::ToNode(RegExpCompiler* compiler,
+ RegExpNode* on_success) {
+ ZoneList<TextElement>* elms =
+ new (compiler->zone()) ZoneList<TextElement>(1, compiler->zone());
+ elms->Add(TextElement::Atom(this), compiler->zone());
+ return new (compiler->zone())
+ TextNode(elms, compiler->read_backward(), on_success);
+}
+
+RegExpNode* RegExpText::ToNode(RegExpCompiler* compiler,
+ RegExpNode* on_success) {
+ return new (compiler->zone())
+ TextNode(elements(), compiler->read_backward(), on_success);
+}
+
+static bool CompareInverseRanges(ZoneList<CharacterRange>* ranges,
+ const int* special_class, int length) {
+ length--; // Remove final marker.
+ DCHECK_EQ(kRangeEndMarker, special_class[length]);
+ DCHECK_NE(0, ranges->length());
+ DCHECK_NE(0, length);
+ DCHECK_NE(0, special_class[0]);
+ if (ranges->length() != (length >> 1) + 1) {
+ return false;
+ }
+ CharacterRange range = ranges->at(0);
+ if (range.from() != 0) {
+ return false;
+ }
+ for (int i = 0; i < length; i += 2) {
+ if (special_class[i] != (range.to() + 1)) {
+ return false;
+ }
+ range = ranges->at((i >> 1) + 1);
+ if (special_class[i + 1] != range.from()) {
+ return false;
+ }
+ }
+ if (range.to() != String::kMaxCodePoint) {
+ return false;
+ }
+ return true;
+}
+
+static bool CompareRanges(ZoneList<CharacterRange>* ranges,
+ const int* special_class, int length) {
+ length--; // Remove final marker.
+ DCHECK_EQ(kRangeEndMarker, special_class[length]);
+ if (ranges->length() * 2 != length) {
+ return false;
+ }
+ for (int i = 0; i < length; i += 2) {
+ CharacterRange range = ranges->at(i >> 1);
+ if (range.from() != special_class[i] ||
+ range.to() != special_class[i + 1] - 1) {
+ return false;
+ }
+ }
+ return true;
+}
+
+bool RegExpCharacterClass::is_standard(Zone* zone) {
+ // TODO(lrn): Remove need for this function, by not throwing away information
+ // along the way.
+ if (is_negated()) {
+ return false;
+ }
+ if (set_.is_standard()) {
+ return true;
+ }
+ if (CompareRanges(set_.ranges(zone), kSpaceRanges, kSpaceRangeCount)) {
+ set_.set_standard_set_type('s');
+ return true;
+ }
+ if (CompareInverseRanges(set_.ranges(zone), kSpaceRanges, kSpaceRangeCount)) {
+ set_.set_standard_set_type('S');
+ return true;
+ }
+ if (CompareInverseRanges(set_.ranges(zone), kLineTerminatorRanges,
+ kLineTerminatorRangeCount)) {
+ set_.set_standard_set_type('.');
+ return true;
+ }
+ if (CompareRanges(set_.ranges(zone), kLineTerminatorRanges,
+ kLineTerminatorRangeCount)) {
+ set_.set_standard_set_type('n');
+ return true;
+ }
+ if (CompareRanges(set_.ranges(zone), kWordRanges, kWordRangeCount)) {
+ set_.set_standard_set_type('w');
+ return true;
+ }
+ if (CompareInverseRanges(set_.ranges(zone), kWordRanges, kWordRangeCount)) {
+ set_.set_standard_set_type('W');
+ return true;
+ }
+ return false;
+}
+
+UnicodeRangeSplitter::UnicodeRangeSplitter(ZoneList<CharacterRange>* base) {
+ // 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.
+
+ for (int i = 0; i < base->length(); i++) AddRange(base->at(i));
+}
+
+void UnicodeRangeSplitter::AddRange(CharacterRange range) {
+ static constexpr uc32 kBmp1Start = 0;
+ static constexpr uc32 kBmp1End = kLeadSurrogateStart - 1;
+ static constexpr uc32 kBmp2Start = kTrailSurrogateEnd + 1;
+ static constexpr uc32 kBmp2End = kNonBmpStart - 1;
+
+ // Ends are all inclusive.
+ STATIC_ASSERT(kBmp1Start == 0);
+ STATIC_ASSERT(kBmp1Start < kBmp1End);
+ STATIC_ASSERT(kBmp1End + 1 == kLeadSurrogateStart);
+ STATIC_ASSERT(kLeadSurrogateStart < kLeadSurrogateEnd);
+ STATIC_ASSERT(kLeadSurrogateEnd + 1 == kTrailSurrogateStart);
+ STATIC_ASSERT(kTrailSurrogateStart < kTrailSurrogateEnd);
+ STATIC_ASSERT(kTrailSurrogateEnd + 1 == kBmp2Start);
+ STATIC_ASSERT(kBmp2Start < kBmp2End);
+ STATIC_ASSERT(kBmp2End + 1 == kNonBmpStart);
+ STATIC_ASSERT(kNonBmpStart < kNonBmpEnd);
+
+ static constexpr uc32 kStarts[] = {
+ kBmp1Start, kLeadSurrogateStart, kTrailSurrogateStart,
+ kBmp2Start, kNonBmpStart,
+ };
+
+ static constexpr uc32 kEnds[] = {
+ kBmp1End, kLeadSurrogateEnd, kTrailSurrogateEnd, kBmp2End, kNonBmpEnd,
+ };
+
+ CharacterRangeVector* const kTargets[] = {
+ &bmp_, &lead_surrogates_, &trail_surrogates_, &bmp_, &non_bmp_,
+ };
+
+ static constexpr int kCount = arraysize(kStarts);
+ STATIC_ASSERT(kCount == arraysize(kEnds));
+ STATIC_ASSERT(kCount == arraysize(kTargets));
+
+ for (int i = 0; i < kCount; i++) {
+ if (kStarts[i] > range.to()) break;
+ const uc32 from = std::max(kStarts[i], range.from());
+ const uc32 to = std::min(kEnds[i], range.to());
+ if (from > to) continue;
+ kTargets[i]->emplace_back(CharacterRange::Range(from, to));
+ }
+}
+
+namespace {
+
+// Translates between new and old V8-isms (SmallVector, ZoneList).
+ZoneList<CharacterRange>* ToCanonicalZoneList(
+ const UnicodeRangeSplitter::CharacterRangeVector* v, Zone* zone) {
+ if (v->empty()) return nullptr;
+
+ ZoneList<CharacterRange>* result =
+ new (zone) ZoneList<CharacterRange>(static_cast<int>(v->size()), zone);
+ for (size_t i = 0; i < v->size(); i++) {
+ result->Add(v->at(i), zone);
+ }
+
+ CharacterRange::Canonicalize(result);
+ return result;
+}
+
+void AddBmpCharacters(RegExpCompiler* compiler, ChoiceNode* result,
+ RegExpNode* on_success, UnicodeRangeSplitter* splitter) {
+ ZoneList<CharacterRange>* bmp =
+ ToCanonicalZoneList(splitter->bmp(), compiler->zone());
+ if (bmp == nullptr) return;
+ JSRegExp::Flags default_flags = JSRegExp::Flags();
+ result->AddAlternative(GuardedAlternative(TextNode::CreateForCharacterRanges(
+ compiler->zone(), bmp, compiler->read_backward(), on_success,
+ default_flags)));
+}
+
+void AddNonBmpSurrogatePairs(RegExpCompiler* compiler, ChoiceNode* result,
+ RegExpNode* on_success,
+ UnicodeRangeSplitter* splitter) {
+ ZoneList<CharacterRange>* non_bmp =
+ ToCanonicalZoneList(splitter->non_bmp(), compiler->zone());
+ if (non_bmp == nullptr) return;
+ DCHECK(!compiler->one_byte());
+ Zone* zone = compiler->zone();
+ JSRegExp::Flags default_flags = JSRegExp::Flags();
+ 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, default_flags)));
+ } 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, default_flags)));
+ 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, default_flags)));
+ 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, default_flags)));
+ }
+ }
+ }
+}
+
+RegExpNode* NegativeLookaroundAgainstReadDirectionAndMatch(
+ RegExpCompiler* compiler, ZoneList<CharacterRange>* lookbehind,
+ ZoneList<CharacterRange>* match, RegExpNode* on_success, bool read_backward,
+ JSRegExp::Flags flags) {
+ Zone* zone = compiler->zone();
+ RegExpNode* match_node = TextNode::CreateForCharacterRanges(
+ zone, match, read_backward, on_success, flags);
+ 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(), flags);
+ return lookaround.ForMatch(negative_match);
+}
+
+RegExpNode* MatchAndNegativeLookaroundInReadDirection(
+ RegExpCompiler* compiler, ZoneList<CharacterRange>* match,
+ ZoneList<CharacterRange>* lookahead, RegExpNode* on_success,
+ bool read_backward, JSRegExp::Flags flags) {
+ 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(), flags);
+ return TextNode::CreateForCharacterRanges(
+ zone, match, read_backward, lookaround.ForMatch(negative_match), flags);
+}
+
+void AddLoneLeadSurrogates(RegExpCompiler* compiler, ChoiceNode* result,
+ RegExpNode* on_success,
+ UnicodeRangeSplitter* splitter) {
+ JSRegExp::Flags default_flags = JSRegExp::Flags();
+ ZoneList<CharacterRange>* lead_surrogates =
+ ToCanonicalZoneList(splitter->lead_surrogates(), compiler->zone());
+ 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,
+ default_flags);
+ } 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,
+ default_flags);
+ }
+ result->AddAlternative(GuardedAlternative(match));
+}
+
+void AddLoneTrailSurrogates(RegExpCompiler* compiler, ChoiceNode* result,
+ RegExpNode* on_success,
+ UnicodeRangeSplitter* splitter) {
+ JSRegExp::Flags default_flags = JSRegExp::Flags();
+ ZoneList<CharacterRange>* trail_surrogates =
+ ToCanonicalZoneList(splitter->trail_surrogates(), compiler->zone());
+ 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,
+ default_flags);
+ } 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,
+ default_flags);
+ }
+ 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));
+ JSRegExp::Flags default_flags = JSRegExp::Flags();
+ return TextNode::CreateForCharacterRanges(zone, range, false, on_success,
+ default_flags);
+}
+
+void AddUnicodeCaseEquivalents(ZoneList<CharacterRange>* ranges, Zone* zone) {
+#ifdef V8_INTL_SUPPORT
+ DCHECK(CharacterRange::IsCanonical(ranges));
+
+ // Micro-optimization to avoid passing large ranges to UnicodeSet::closeOver.
+ // See also https://crbug.com/v8/6727.
+ // TODO(jgruber): This only covers the special case of the {0,0x10FFFF} range,
+ // which we use frequently internally. But large ranges can also easily be
+ // created by the user. We might want to have a more general caching mechanism
+ // for such ranges.
+ if (ranges->length() == 1 && ranges->at(0).IsEverything(kNonBmpEnd)) return;
+
+ // Use ICU to compute the case fold closure over the ranges.
+ icu::UnicodeSet set;
+ for (int i = 0; i < ranges->length(); i++) {
+ set.add(ranges->at(i).from(), ranges->at(i).to());
+ }
+ ranges->Clear();
+ set.closeOver(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.
+ set.removeAllStrings();
+ for (int i = 0; i < set.getRangeCount(); i++) {
+ ranges->Add(CharacterRange::Range(set.getRangeStart(i), set.getRangeEnd(i)),
+ zone);
+ }
+ // No errors and everything we collected have been ranges.
+ CharacterRange::Canonicalize(ranges);
+#endif // V8_INTL_SUPPORT
+}
+
+} // namespace
+
+RegExpNode* RegExpCharacterClass::ToNode(RegExpCompiler* compiler,
+ RegExpNode* on_success) {
+ set_.Canonicalize();
+ Zone* zone = compiler->zone();
+ ZoneList<CharacterRange>* ranges = this->ranges(zone);
+ if (NeedsUnicodeCaseEquivalents(flags_)) {
+ AddUnicodeCaseEquivalents(ranges, zone);
+ }
+ if (IsUnicode(flags_) && !compiler->one_byte() &&
+ !contains_split_surrogate()) {
+ if (is_negated()) {
+ ZoneList<CharacterRange>* negated =
+ new (zone) ZoneList<CharacterRange>(2, zone);
+ CharacterRange::Negate(ranges, negated, zone);
+ ranges = negated;
+ }
+ if (ranges->length() == 0) {
+ JSRegExp::Flags default_flags;
+ RegExpCharacterClass* fail =
+ new (zone) RegExpCharacterClass(zone, ranges, default_flags);
+ return new (zone) TextNode(fail, compiler->read_backward(), on_success);
+ }
+ if (standard_type() == '*') {
+ return UnanchoredAdvance(compiler, on_success);
+ } else {
+ ChoiceNode* result = new (zone) ChoiceNode(2, zone);
+ UnicodeRangeSplitter splitter(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);
+ }
+}
+
+int CompareFirstChar(RegExpTree* const* a, RegExpTree* const* b) {
+ RegExpAtom* atom1 = (*a)->AsAtom();
+ RegExpAtom* atom2 = (*b)->AsAtom();
+ uc16 character1 = atom1->data().at(0);
+ uc16 character2 = atom2->data().at(0);
+ if (character1 < character2) return -1;
+ if (character1 > character2) return 1;
+ return 0;
+}
+
+#ifdef V8_INTL_SUPPORT
+
+// Case Insensitve comparesion
+int CompareFirstCharCaseInsensitve(RegExpTree* const* a, RegExpTree* const* b) {
+ RegExpAtom* atom1 = (*a)->AsAtom();
+ RegExpAtom* atom2 = (*b)->AsAtom();
+ icu::UnicodeString character1(atom1->data().at(0));
+ return character1.caseCompare(atom2->data().at(0), U_FOLD_CASE_DEFAULT);
+}
+
+#else
+
+static unibrow::uchar Canonical(
+ unibrow::Mapping<unibrow::Ecma262Canonicalize>* canonicalize,
+ unibrow::uchar c) {
+ unibrow::uchar chars[unibrow::Ecma262Canonicalize::kMaxWidth];
+ int length = canonicalize->get(c, '\0', chars);
+ DCHECK_LE(length, 1);
+ unibrow::uchar canonical = c;
+ if (length == 1) canonical = chars[0];
+ return canonical;
+}
+
+int CompareFirstCharCaseIndependent(
+ unibrow::Mapping<unibrow::Ecma262Canonicalize>* canonicalize,
+ RegExpTree* const* a, RegExpTree* const* b) {
+ RegExpAtom* atom1 = (*a)->AsAtom();
+ RegExpAtom* atom2 = (*b)->AsAtom();
+ unibrow::uchar character1 = atom1->data().at(0);
+ unibrow::uchar character2 = atom2->data().at(0);
+ if (character1 == character2) return 0;
+ if (character1 >= 'a' || character2 >= 'a') {
+ character1 = Canonical(canonicalize, character1);
+ character2 = Canonical(canonicalize, character2);
+ }
+ return static_cast<int>(character1) - static_cast<int>(character2);
+}
+#endif // V8_INTL_SUPPORT
+
+// We can stable sort runs of atoms, since the order does not matter if they
+// start with different characters.
+// Returns true if any consecutive atoms were found.
+bool RegExpDisjunction::SortConsecutiveAtoms(RegExpCompiler* compiler) {
+ ZoneList<RegExpTree*>* alternatives = this->alternatives();
+ int length = alternatives->length();
+ bool found_consecutive_atoms = false;
+ for (int i = 0; i < length; i++) {
+ while (i < length) {
+ RegExpTree* alternative = alternatives->at(i);
+ if (alternative->IsAtom()) break;
+ i++;
+ }
+ // i is length or it is the index of an atom.
+ if (i == length) break;
+ int first_atom = i;
+ JSRegExp::Flags flags = alternatives->at(i)->AsAtom()->flags();
+ i++;
+ while (i < length) {
+ RegExpTree* alternative = alternatives->at(i);
+ if (!alternative->IsAtom()) break;
+ if (alternative->AsAtom()->flags() != flags) break;
+ i++;
+ }
+ // Sort atoms to get ones with common prefixes together.
+ // This step is more tricky if we are in a case-independent regexp,
+ // because it would change /is|I/ to /I|is/, and order matters when
+ // the regexp parts don't match only disjoint starting points. To fix
+ // this we have a version of CompareFirstChar that uses case-
+ // independent character classes for comparison.
+ DCHECK_LT(first_atom, alternatives->length());
+ DCHECK_LE(i, alternatives->length());
+ DCHECK_LE(first_atom, i);
+ if (IgnoreCase(flags)) {
+#ifdef V8_INTL_SUPPORT
+ alternatives->StableSort(CompareFirstCharCaseInsensitve, first_atom,
+ i - first_atom);
+#else
+ unibrow::Mapping<unibrow::Ecma262Canonicalize>* canonicalize =
+ compiler->isolate()->regexp_macro_assembler_canonicalize();
+ auto compare_closure = [canonicalize](RegExpTree* const* a,
+ RegExpTree* const* b) {
+ return CompareFirstCharCaseIndependent(canonicalize, a, b);
+ };
+ alternatives->StableSort(compare_closure, first_atom, i - first_atom);
+#endif // V8_INTL_SUPPORT
+ } else {
+ alternatives->StableSort(CompareFirstChar, first_atom, i - first_atom);
+ }
+ if (i - first_atom > 1) found_consecutive_atoms = true;
+ }
+ return found_consecutive_atoms;
+}
+
+// Optimizes ab|ac|az to a(?:b|c|d).
+void RegExpDisjunction::RationalizeConsecutiveAtoms(RegExpCompiler* compiler) {
+ Zone* zone = compiler->zone();
+ ZoneList<RegExpTree*>* alternatives = this->alternatives();
+ int length = alternatives->length();
+
+ int write_posn = 0;
+ int i = 0;
+ while (i < length) {
+ RegExpTree* alternative = alternatives->at(i);
+ if (!alternative->IsAtom()) {
+ alternatives->at(write_posn++) = alternatives->at(i);
+ i++;
+ continue;
+ }
+ RegExpAtom* const atom = alternative->AsAtom();
+ JSRegExp::Flags flags = atom->flags();
+#ifdef V8_INTL_SUPPORT
+ icu::UnicodeString common_prefix(atom->data().at(0));
+#else
+ unibrow::uchar common_prefix = atom->data().at(0);
+#endif // V8_INTL_SUPPORT
+ int first_with_prefix = i;
+ int prefix_length = atom->length();
+ i++;
+ while (i < length) {
+ alternative = alternatives->at(i);
+ if (!alternative->IsAtom()) break;
+ RegExpAtom* const atom = alternative->AsAtom();
+ if (atom->flags() != flags) break;
+#ifdef V8_INTL_SUPPORT
+ icu::UnicodeString new_prefix(atom->data().at(0));
+ if (new_prefix != common_prefix) {
+ if (!IgnoreCase(flags)) break;
+ if (common_prefix.caseCompare(new_prefix, U_FOLD_CASE_DEFAULT) != 0)
+ break;
+ }
+#else
+ unibrow::uchar new_prefix = atom->data().at(0);
+ if (new_prefix != common_prefix) {
+ if (!IgnoreCase(flags)) break;
+ unibrow::Mapping<unibrow::Ecma262Canonicalize>* canonicalize =
+ compiler->isolate()->regexp_macro_assembler_canonicalize();
+ new_prefix = Canonical(canonicalize, new_prefix);
+ common_prefix = Canonical(canonicalize, common_prefix);
+ if (new_prefix != common_prefix) break;
+ }
+#endif // V8_INTL_SUPPORT
+ prefix_length = Min(prefix_length, atom->length());
+ i++;
+ }
+ if (i > first_with_prefix + 2) {
+ // Found worthwhile run of alternatives with common prefix of at least one
+ // character. The sorting function above did not sort on more than one
+ // character for reasons of correctness, but there may still be a longer
+ // common prefix if the terms were similar or presorted in the input.
+ // Find out how long the common prefix is.
+ int run_length = i - first_with_prefix;
+ RegExpAtom* const atom = alternatives->at(first_with_prefix)->AsAtom();
+ for (int j = 1; j < run_length && prefix_length > 1; j++) {
+ RegExpAtom* old_atom =
+ alternatives->at(j + first_with_prefix)->AsAtom();
+ for (int k = 1; k < prefix_length; k++) {
+ if (atom->data().at(k) != old_atom->data().at(k)) {
+ prefix_length = k;
+ break;
+ }
+ }
+ }
+ RegExpAtom* prefix = new (zone)
+ RegExpAtom(atom->data().SubVector(0, prefix_length), flags);
+ ZoneList<RegExpTree*>* pair = new (zone) ZoneList<RegExpTree*>(2, zone);
+ pair->Add(prefix, zone);
+ ZoneList<RegExpTree*>* suffixes =
+ new (zone) ZoneList<RegExpTree*>(run_length, zone);
+ for (int j = 0; j < run_length; j++) {
+ RegExpAtom* old_atom =
+ alternatives->at(j + first_with_prefix)->AsAtom();
+ int len = old_atom->length();
+ if (len == prefix_length) {
+ suffixes->Add(new (zone) RegExpEmpty(), zone);
+ } else {
+ RegExpTree* suffix = new (zone) RegExpAtom(
+ old_atom->data().SubVector(prefix_length, old_atom->length()),
+ flags);
+ suffixes->Add(suffix, zone);
+ }
+ }
+ pair->Add(new (zone) RegExpDisjunction(suffixes), zone);
+ alternatives->at(write_posn++) = new (zone) RegExpAlternative(pair);
+ } else {
+ // Just copy any non-worthwhile alternatives.
+ for (int j = first_with_prefix; j < i; j++) {
+ alternatives->at(write_posn++) = alternatives->at(j);
+ }
+ }
+ }
+ alternatives->Rewind(write_posn); // Trim end of array.
+}
+
+// Optimizes b|c|z to [bcz].
+void RegExpDisjunction::FixSingleCharacterDisjunctions(
+ RegExpCompiler* compiler) {
+ Zone* zone = compiler->zone();
+ ZoneList<RegExpTree*>* alternatives = this->alternatives();
+ int length = alternatives->length();
+
+ int write_posn = 0;
+ int i = 0;
+ while (i < length) {
+ RegExpTree* alternative = alternatives->at(i);
+ if (!alternative->IsAtom()) {
+ alternatives->at(write_posn++) = alternatives->at(i);
+ i++;
+ continue;
+ }
+ RegExpAtom* const atom = alternative->AsAtom();
+ if (atom->length() != 1) {
+ alternatives->at(write_posn++) = alternatives->at(i);
+ i++;
+ continue;
+ }
+ JSRegExp::Flags flags = atom->flags();
+ DCHECK_IMPLIES(IsUnicode(flags),
+ !unibrow::Utf16::IsLeadSurrogate(atom->data().at(0)));
+ bool contains_trail_surrogate =
+ unibrow::Utf16::IsTrailSurrogate(atom->data().at(0));
+ int first_in_run = i;
+ i++;
+ // Find a run of single-character atom alternatives that have identical
+ // flags (case independence and unicode-ness).
+ while (i < length) {
+ alternative = alternatives->at(i);
+ if (!alternative->IsAtom()) break;
+ RegExpAtom* const atom = alternative->AsAtom();
+ if (atom->length() != 1) break;
+ if (atom->flags() != flags) break;
+ DCHECK_IMPLIES(IsUnicode(flags),
+ !unibrow::Utf16::IsLeadSurrogate(atom->data().at(0)));
+ contains_trail_surrogate |=
+ unibrow::Utf16::IsTrailSurrogate(atom->data().at(0));
+ i++;
+ }
+ if (i > first_in_run + 1) {
+ // Found non-trivial run of single-character alternatives.
+ int run_length = i - first_in_run;
+ ZoneList<CharacterRange>* ranges =
+ new (zone) ZoneList<CharacterRange>(2, zone);
+ for (int j = 0; j < run_length; j++) {
+ RegExpAtom* old_atom = alternatives->at(j + first_in_run)->AsAtom();
+ DCHECK_EQ(old_atom->length(), 1);
+ ranges->Add(CharacterRange::Singleton(old_atom->data().at(0)), zone);
+ }
+ RegExpCharacterClass::CharacterClassFlags character_class_flags;
+ if (IsUnicode(flags) && contains_trail_surrogate) {
+ character_class_flags = RegExpCharacterClass::CONTAINS_SPLIT_SURROGATE;
+ }
+ alternatives->at(write_posn++) = new (zone)
+ RegExpCharacterClass(zone, ranges, flags, character_class_flags);
+ } else {
+ // Just copy any trivial alternatives.
+ for (int j = first_in_run; j < i; j++) {
+ alternatives->at(write_posn++) = alternatives->at(j);
+ }
+ }
+ }
+ alternatives->Rewind(write_posn); // Trim end of array.
+}
+
+RegExpNode* RegExpDisjunction::ToNode(RegExpCompiler* compiler,
+ RegExpNode* on_success) {
+ ZoneList<RegExpTree*>* alternatives = this->alternatives();
+
+ if (alternatives->length() > 2) {
+ bool found_consecutive_atoms = SortConsecutiveAtoms(compiler);
+ if (found_consecutive_atoms) RationalizeConsecutiveAtoms(compiler);
+ FixSingleCharacterDisjunctions(compiler);
+ if (alternatives->length() == 1) {
+ return alternatives->at(0)->ToNode(compiler, on_success);
+ }
+ }
+
+ int length = alternatives->length();
+
+ ChoiceNode* result =
+ new (compiler->zone()) ChoiceNode(length, compiler->zone());
+ for (int i = 0; i < length; i++) {
+ GuardedAlternative alternative(
+ alternatives->at(i)->ToNode(compiler, on_success));
+ result->AddAlternative(alternative);
+ }
+ return result;
+}
+
+RegExpNode* RegExpQuantifier::ToNode(RegExpCompiler* compiler,
+ RegExpNode* on_success) {
+ return ToNode(min(), max(), is_greedy(), body(), compiler, on_success);
+}
+
+namespace {
+// Desugar \b to (?<=\w)(?=\W)|(?<=\W)(?=\w) and
+// \B to (?<=\w)(?=\w)|(?<=\W)(?=\W)
+RegExpNode* BoundaryAssertionAsLookaround(RegExpCompiler* compiler,
+ RegExpNode* on_success,
+ RegExpAssertion::AssertionType type,
+ JSRegExp::Flags flags) {
+ DCHECK(NeedsUnicodeCaseEquivalents(flags));
+ Zone* zone = compiler->zone();
+ ZoneList<CharacterRange>* word_range =
+ new (zone) ZoneList<CharacterRange>(2, zone);
+ CharacterRange::AddClassEscape('w', word_range, true, zone);
+ int stack_register = compiler->UnicodeLookaroundStackRegister();
+ int position_register = compiler->UnicodeLookaroundPositionRegister();
+ ChoiceNode* result = new (zone) ChoiceNode(2, zone);
+ // Add two choices. The (non-)boundary could start with a word or
+ // a non-word-character.
+ for (int i = 0; i < 2; i++) {
+ bool lookbehind_for_word = i == 0;
+ bool lookahead_for_word =
+ (type == RegExpAssertion::BOUNDARY) ^ lookbehind_for_word;
+ // Look to the left.
+ RegExpLookaround::Builder lookbehind(lookbehind_for_word, on_success,
+ stack_register, position_register);
+ RegExpNode* backward = TextNode::CreateForCharacterRanges(
+ zone, word_range, true, lookbehind.on_match_success(), flags);
+ // Look to the right.
+ RegExpLookaround::Builder lookahead(lookahead_for_word,
+ lookbehind.ForMatch(backward),
+ stack_register, position_register);
+ RegExpNode* forward = TextNode::CreateForCharacterRanges(
+ zone, word_range, false, lookahead.on_match_success(), flags);
+ result->AddAlternative(GuardedAlternative(lookahead.ForMatch(forward)));
+ }
+ return result;
+}
+} // anonymous namespace
+
+RegExpNode* RegExpAssertion::ToNode(RegExpCompiler* compiler,
+ RegExpNode* on_success) {
+ NodeInfo info;
+ Zone* zone = compiler->zone();
+
+ switch (assertion_type()) {
+ case START_OF_LINE:
+ return AssertionNode::AfterNewline(on_success);
+ case START_OF_INPUT:
+ return AssertionNode::AtStart(on_success);
+ case BOUNDARY:
+ return NeedsUnicodeCaseEquivalents(flags_)
+ ? BoundaryAssertionAsLookaround(compiler, on_success, BOUNDARY,
+ flags_)
+ : AssertionNode::AtBoundary(on_success);
+ case NON_BOUNDARY:
+ return NeedsUnicodeCaseEquivalents(flags_)
+ ? BoundaryAssertionAsLookaround(compiler, on_success,
+ NON_BOUNDARY, flags_)
+ : AssertionNode::AtNonBoundary(on_success);
+ case END_OF_INPUT:
+ return AssertionNode::AtEnd(on_success);
+ case END_OF_LINE: {
+ // Compile $ in multiline regexps as an alternation with a positive
+ // lookahead in one side and an end-of-input on the other side.
+ // We need two registers for the lookahead.
+ int stack_pointer_register = compiler->AllocateRegister();
+ int position_register = compiler->AllocateRegister();
+ // The ChoiceNode to distinguish between a newline and end-of-input.
+ ChoiceNode* result = new (zone) ChoiceNode(2, zone);
+ // Create a newline atom.
+ ZoneList<CharacterRange>* newline_ranges =
+ new (zone) ZoneList<CharacterRange>(3, zone);
+ CharacterRange::AddClassEscape('n', newline_ranges, false, zone);
+ JSRegExp::Flags default_flags = JSRegExp::Flags();
+ RegExpCharacterClass* newline_atom =
+ new (zone) RegExpCharacterClass('n', default_flags);
+ TextNode* newline_matcher =
+ new (zone) TextNode(newline_atom, false,
+ ActionNode::PositiveSubmatchSuccess(
+ stack_pointer_register, position_register,
+ 0, // No captures inside.
+ -1, // Ignored if no captures.
+ on_success));
+ // Create an end-of-input matcher.
+ RegExpNode* end_of_line = ActionNode::BeginSubmatch(
+ stack_pointer_register, position_register, newline_matcher);
+ // Add the two alternatives to the ChoiceNode.
+ GuardedAlternative eol_alternative(end_of_line);
+ result->AddAlternative(eol_alternative);
+ GuardedAlternative end_alternative(AssertionNode::AtEnd(on_success));
+ result->AddAlternative(end_alternative);
+ return result;
+ }
+ default:
+ UNREACHABLE();
+ }
+ return on_success;
+}
+
+RegExpNode* RegExpBackReference::ToNode(RegExpCompiler* compiler,
+ RegExpNode* on_success) {
+ return new (compiler->zone())
+ BackReferenceNode(RegExpCapture::StartRegister(index()),
+ RegExpCapture::EndRegister(index()), flags_,
+ compiler->read_backward(), on_success);
+}
+
+RegExpNode* RegExpEmpty::ToNode(RegExpCompiler* compiler,
+ RegExpNode* on_success) {
+ return on_success;
+}
+
+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();
+ int position_register = compiler->AllocateRegister();
+
+ const int registers_per_capture = 2;
+ const int register_of_first_capture = 2;
+ int register_count = capture_count_ * registers_per_capture;
+ int register_start =
+ register_of_first_capture + capture_from_ * registers_per_capture;
+
+ RegExpNode* result;
+ bool was_reading_backward = compiler->read_backward();
+ compiler->set_read_backward(type() == LOOKBEHIND);
+ 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;
+}
+
+RegExpNode* RegExpCapture::ToNode(RegExpCompiler* compiler,
+ RegExpNode* on_success) {
+ return ToNode(body(), index(), compiler, on_success);
+}
+
+RegExpNode* RegExpCapture::ToNode(RegExpTree* body, int index,
+ RegExpCompiler* compiler,
+ RegExpNode* on_success) {
+ DCHECK_NOT_NULL(body);
+ int start_reg = RegExpCapture::StartRegister(index);
+ int end_reg = RegExpCapture::EndRegister(index);
+ if (compiler->read_backward()) std::swap(start_reg, end_reg);
+ RegExpNode* store_end = ActionNode::StorePosition(end_reg, true, on_success);
+ RegExpNode* body_node = body->ToNode(compiler, store_end);
+ return ActionNode::StorePosition(start_reg, true, body_node);
+}
+
+namespace {
+
+class AssertionSequenceRewriter final {
+ public:
+ // TODO(jgruber): Consider moving this to a separate AST tree rewriter pass
+ // instead of sprinkling rewrites into the AST->Node conversion process.
+ static void MaybeRewrite(ZoneList<RegExpTree*>* terms, Zone* zone) {
+ AssertionSequenceRewriter rewriter(terms, zone);
+
+ static constexpr int kNoIndex = -1;
+ int from = kNoIndex;
+
+ for (int i = 0; i < terms->length(); i++) {
+ RegExpTree* t = terms->at(i);
+ if (from == kNoIndex && t->IsAssertion()) {
+ from = i; // Start a sequence.
+ } else if (from != kNoIndex && !t->IsAssertion()) {
+ // Terminate and process the sequence.
+ if (i - from > 1) rewriter.Rewrite(from, i);
+ from = kNoIndex;
+ }
+ }
+
+ if (from != kNoIndex && terms->length() - from > 1) {
+ rewriter.Rewrite(from, terms->length());
+ }
+ }
+
+ // All assertions are zero width. A consecutive sequence of assertions is
+ // order-independent. There's two ways we can optimize here:
+ // 1. fold all identical assertions.
+ // 2. if any assertion combinations are known to fail (e.g. \b\B), the entire
+ // sequence fails.
+ void Rewrite(int from, int to) {
+ DCHECK_GT(to, from + 1);
+
+ // Bitfield of all seen assertions.
+ uint32_t seen_assertions = 0;
+ STATIC_ASSERT(RegExpAssertion::LAST_TYPE < kUInt32Size * kBitsPerByte);
+
+ // Flags must match for folding.
+ JSRegExp::Flags flags = terms_->at(from)->AsAssertion()->flags();
+ bool saw_mismatched_flags = false;
+
+ for (int i = from; i < to; i++) {
+ RegExpAssertion* t = terms_->at(i)->AsAssertion();
+ if (t->flags() != flags) saw_mismatched_flags = true;
+ const uint32_t bit = 1 << t->assertion_type();
+
+ if ((seen_assertions & bit) && !saw_mismatched_flags) {
+ // Fold duplicates.
+ terms_->Set(i, new (zone_) RegExpEmpty());
+ }
+
+ seen_assertions |= bit;
+ }
+
+ // Collapse failures.
+ const uint32_t always_fails_mask =
+ 1 << RegExpAssertion::BOUNDARY | 1 << RegExpAssertion::NON_BOUNDARY;
+ if ((seen_assertions & always_fails_mask) == always_fails_mask) {
+ ReplaceSequenceWithFailure(from, to);
+ }
+ }
+
+ void ReplaceSequenceWithFailure(int from, int to) {
+ // Replace the entire sequence with a single node that always fails.
+ // TODO(jgruber): Consider adding an explicit Fail kind. Until then, the
+ // negated '*' (everything) range serves the purpose.
+ ZoneList<CharacterRange>* ranges =
+ new (zone_) ZoneList<CharacterRange>(0, zone_);
+ RegExpCharacterClass* cc =
+ new (zone_) RegExpCharacterClass(zone_, ranges, JSRegExp::Flags());
+ terms_->Set(from, cc);
+
+ // Zero out the rest.
+ RegExpEmpty* empty = new (zone_) RegExpEmpty();
+ for (int i = from + 1; i < to; i++) terms_->Set(i, empty);
+ }
+
+ private:
+ AssertionSequenceRewriter(ZoneList<RegExpTree*>* terms, Zone* zone)
+ : zone_(zone), terms_(terms) {}
+
+ Zone* zone_;
+ ZoneList<RegExpTree*>* terms_;
+};
+
+} // namespace
+
+RegExpNode* RegExpAlternative::ToNode(RegExpCompiler* compiler,
+ RegExpNode* on_success) {
+ ZoneList<RegExpTree*>* children = nodes();
+
+ AssertionSequenceRewriter::MaybeRewrite(children, compiler->zone());
+
+ RegExpNode* current = on_success;
+ if (compiler->read_backward()) {
+ for (int i = 0; i < children->length(); i++) {
+ current = children->at(i)->ToNode(compiler, current);
+ }
+ } else {
+ for (int i = children->length() - 1; i >= 0; i--) {
+ current = children->at(i)->ToNode(compiler, current);
+ }
+ }
+ return current;
+}
+
+static void AddClass(const int* elmv, int elmc,
+ ZoneList<CharacterRange>* ranges, Zone* zone) {
+ elmc--;
+ DCHECK_EQ(kRangeEndMarker, elmv[elmc]);
+ for (int i = 0; i < elmc; i += 2) {
+ DCHECK(elmv[i] < elmv[i + 1]);
+ ranges->Add(CharacterRange::Range(elmv[i], elmv[i + 1] - 1), zone);
+ }
+}
+
+static void AddClassNegated(const int* elmv, int elmc,
+ ZoneList<CharacterRange>* ranges, Zone* zone) {
+ elmc--;
+ DCHECK_EQ(kRangeEndMarker, elmv[elmc]);
+ DCHECK_NE(0x0000, elmv[0]);
+ DCHECK_NE(String::kMaxCodePoint, elmv[elmc - 1]);
+ 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::Range(last, elmv[i] - 1), zone);
+ last = elmv[i + 1];
+ }
+ ranges->Add(CharacterRange::Range(last, String::kMaxCodePoint), zone);
+}
+
+void CharacterRange::AddClassEscape(char type, ZoneList<CharacterRange>* ranges,
+ bool add_unicode_case_equivalents,
+ Zone* zone) {
+ if (add_unicode_case_equivalents && (type == 'w' || type == 'W')) {
+ // See #sec-runtime-semantics-wordcharacters-abstract-operation
+ // In case of unicode and ignore_case, we need to create the closure over
+ // case equivalent characters before negating.
+ ZoneList<CharacterRange>* new_ranges =
+ new (zone) ZoneList<CharacterRange>(2, zone);
+ AddClass(kWordRanges, kWordRangeCount, new_ranges, zone);
+ AddUnicodeCaseEquivalents(new_ranges, zone);
+ if (type == 'W') {
+ ZoneList<CharacterRange>* negated =
+ new (zone) ZoneList<CharacterRange>(2, zone);
+ CharacterRange::Negate(new_ranges, negated, zone);
+ new_ranges = negated;
+ }
+ ranges->AddAll(*new_ranges, zone);
+ return;
+ }
+ AddClassEscape(type, ranges, zone);
+}
+
+void CharacterRange::AddClassEscape(char type, ZoneList<CharacterRange>* ranges,
+ Zone* zone) {
+ switch (type) {
+ case 's':
+ AddClass(kSpaceRanges, kSpaceRangeCount, ranges, zone);
+ break;
+ case 'S':
+ AddClassNegated(kSpaceRanges, kSpaceRangeCount, ranges, zone);
+ break;
+ case 'w':
+ AddClass(kWordRanges, kWordRangeCount, ranges, zone);
+ break;
+ case 'W':
+ AddClassNegated(kWordRanges, kWordRangeCount, ranges, zone);
+ break;
+ case 'd':
+ AddClass(kDigitRanges, kDigitRangeCount, ranges, zone);
+ break;
+ case 'D':
+ AddClassNegated(kDigitRanges, kDigitRangeCount, ranges, zone);
+ break;
+ case '.':
+ AddClassNegated(kLineTerminatorRanges, kLineTerminatorRangeCount, ranges,
+ zone);
+ break;
+ // This is not a character range as defined by the spec but a
+ // convenient shorthand for a character class that matches any
+ // character.
+ case '*':
+ ranges->Add(CharacterRange::Everything(), zone);
+ break;
+ // This is the set of characters matched by the $ and ^ symbols
+ // in multiline mode.
+ case 'n':
+ AddClass(kLineTerminatorRanges, kLineTerminatorRangeCount, ranges, zone);
+ break;
+ default:
+ UNREACHABLE();
+ }
+}
+
+Vector<const int> CharacterRange::GetWordBounds() {
+ return Vector<const int>(kWordRanges, kWordRangeCount - 1);
+}
+
+#ifdef V8_INTL_SUPPORT
+struct IgnoreSet {
+ IgnoreSet() : set(BuildIgnoreSet()) {}
+ const icu::UnicodeSet set;
+};
+
+struct SpecialAddSet {
+ SpecialAddSet() : set(BuildSpecialAddSet()) {}
+ const icu::UnicodeSet set;
+};
+
+icu::UnicodeSet BuildAsciiAToZSet() {
+ icu::UnicodeSet set('a', 'z');
+ set.add('A', 'Z');
+ set.freeze();
+ return set;
+}
+
+struct AsciiAToZSet {
+ AsciiAToZSet() : set(BuildAsciiAToZSet()) {}
+ const icu::UnicodeSet set;
+};
+
+static base::LazyInstance<IgnoreSet>::type ignore_set =
+ LAZY_INSTANCE_INITIALIZER;
+
+static base::LazyInstance<SpecialAddSet>::type special_add_set =
+ LAZY_INSTANCE_INITIALIZER;
+
+static base::LazyInstance<AsciiAToZSet>::type ascii_a_to_z_set =
+ LAZY_INSTANCE_INITIALIZER;
+#endif // V8_INTL_SUPPORT
+
+// static
+void CharacterRange::AddCaseEquivalents(Isolate* isolate, Zone* zone,
+ ZoneList<CharacterRange>* ranges,
+ bool is_one_byte) {
+ CharacterRange::Canonicalize(ranges);
+ int range_count = ranges->length();
+#ifdef V8_INTL_SUPPORT
+ icu::UnicodeSet others;
+ for (int i = 0; i < range_count; i++) {
+ CharacterRange range = ranges->at(i);
+ uc32 from = range.from();
+ if (from > String::kMaxUtf16CodeUnit) continue;
+ uc32 to = Min(range.to(), String::kMaxUtf16CodeUnit);
+ // Nothing to be done for surrogates.
+ if (from >= kLeadSurrogateStart && to <= kTrailSurrogateEnd) continue;
+ if (is_one_byte && !RangeContainsLatin1Equivalents(range)) {
+ if (from > String::kMaxOneByteCharCode) continue;
+ if (to > String::kMaxOneByteCharCode) to = String::kMaxOneByteCharCode;
+ }
+ others.add(from, to);
+ }
+
+ // Set of characters already added to ranges that do not need to be added
+ // again.
+ icu::UnicodeSet already_added(others);
+
+ // Set of characters in ranges that are in the 52 ASCII characters [a-zA-Z].
+ icu::UnicodeSet in_ascii_a_to_z(others);
+ in_ascii_a_to_z.retainAll(ascii_a_to_z_set.Pointer()->set);
+
+ // Remove all chars in [a-zA-Z] from others.
+ others.removeAll(in_ascii_a_to_z);
+
+ // Set of characters in ranges that are overlapping with special add set.
+ icu::UnicodeSet in_special_add(others);
+ in_special_add.retainAll(special_add_set.Pointer()->set);
+
+ others.removeAll(in_special_add);
+
+ // Ignore all chars in ignore set.
+ others.removeAll(ignore_set.Pointer()->set);
+
+ // For most of the chars in ranges that is still in others, find the case
+ // equivlant set by calling closeOver(USET_CASE_INSENSITIVE).
+ others.closeOver(USET_CASE_INSENSITIVE);
+
+ // Because closeOver(USET_CASE_INSENSITIVE) may add ASCII [a-zA-Z] to others,
+ // but ECMA262 "i" mode won't consider that, remove them from others.
+ // Ex: U+017F add 'S' and 's' to others.
+ others.removeAll(ascii_a_to_z_set.Pointer()->set);
+
+ // Special handling for in_ascii_a_to_z.
+ for (int32_t i = 0; i < in_ascii_a_to_z.getRangeCount(); i++) {
+ UChar32 start = in_ascii_a_to_z.getRangeStart(i);
+ UChar32 end = in_ascii_a_to_z.getRangeEnd(i);
+ // Check if it is uppercase A-Z by checking bit 6.
+ if (start & 0x0020) {
+ // Add the lowercases
+ others.add(start & 0x005F, end & 0x005F);
+ } else {
+ // Add the uppercases
+ others.add(start | 0x0020, end | 0x0020);
+ }
+ }
+
+ // Special handling for chars in "Special Add" set.
+ for (int32_t i = 0; i < in_special_add.getRangeCount(); i++) {
+ UChar32 end = in_special_add.getRangeEnd(i);
+ for (UChar32 ch = in_special_add.getRangeStart(i); ch <= end; ch++) {
+ // Add the uppercase of this character if itself is not an uppercase
+ // character.
+ // Note: The if condiction cannot be u_islower(ch) because ch could be
+ // neither uppercase nor lowercase but Mn.
+ if (!u_isupper(ch)) {
+ others.add(u_toupper(ch));
+ }
+ icu::UnicodeSet candidates(ch, ch);
+ candidates.closeOver(USET_CASE_INSENSITIVE);
+ for (int32_t j = 0; j < candidates.getRangeCount(); j++) {
+ UChar32 end2 = candidates.getRangeEnd(j);
+ for (UChar32 ch2 = candidates.getRangeStart(j); ch2 <= end2; ch2++) {
+ // Add character that is not uppercase to others.
+ if (!u_isupper(ch2)) {
+ others.add(ch2);
+ }
+ }
+ }
+ }
+ }
+
+ // Remove all characters which already in the ranges.
+ others.removeAll(already_added);
+
+ // Add others to the ranges
+ for (int32_t i = 0; i < others.getRangeCount(); i++) {
+ UChar32 from = others.getRangeStart(i);
+ UChar32 to = others.getRangeEnd(i);
+ if (from == to) {
+ ranges->Add(CharacterRange::Singleton(from), zone);
+ } else {
+ ranges->Add(CharacterRange::Range(from, to), zone);
+ }
+ }
+#else
+ for (int i = 0; i < range_count; i++) {
+ CharacterRange range = ranges->at(i);
+ uc32 bottom = range.from();
+ if (bottom > String::kMaxUtf16CodeUnit) continue;
+ uc32 top = Min(range.to(), String::kMaxUtf16CodeUnit);
+ // Nothing to be done for surrogates.
+ if (bottom >= kLeadSurrogateStart && top <= kTrailSurrogateEnd) continue;
+ if (is_one_byte && !RangeContainsLatin1Equivalents(range)) {
+ if (bottom > String::kMaxOneByteCharCode) continue;
+ 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);
+ }
+ }
+ } 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;
+ }
+ }
+ }
+#endif // V8_INTL_SUPPORT
+}
+
+bool CharacterRange::IsCanonical(ZoneList<CharacterRange>* ranges) {
+ DCHECK_NOT_NULL(ranges);
+ int n = ranges->length();
+ if (n <= 1) return true;
+ int max = ranges->at(0).to();
+ for (int i = 1; i < n; i++) {
+ CharacterRange next_range = ranges->at(i);
+ if (next_range.from() <= max + 1) return false;
+ max = next_range.to();
+ }
+ return true;
+}
+
+ZoneList<CharacterRange>* CharacterSet::ranges(Zone* zone) {
+ if (ranges_ == nullptr) {
+ ranges_ = new (zone) ZoneList<CharacterRange>(2, zone);
+ CharacterRange::AddClassEscape(standard_set_type_, ranges_, false, zone);
+ }
+ return ranges_;
+}
+
+// Move a number of elements in a zonelist to another position
+// in the same list. Handles overlapping source and target areas.
+static void MoveRanges(ZoneList<CharacterRange>* list, int from, int to,
+ int count) {
+ // Ranges are potentially overlapping.
+ if (from < to) {
+ for (int i = count - 1; i >= 0; i--) {
+ list->at(to + i) = list->at(from + i);
+ }
+ } else {
+ for (int i = 0; i < count; i++) {
+ list->at(to + i) = list->at(from + i);
+ }
+ }
+}
+
+static int InsertRangeInCanonicalList(ZoneList<CharacterRange>* list, int count,
+ CharacterRange insert) {
+ // Inserts a range into list[0..count[, which must be sorted
+ // by from value and non-overlapping and non-adjacent, using at most
+ // 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.
+ uc32 from = insert.from();
+ uc32 to = insert.to();
+ int start_pos = 0;
+ int end_pos = count;
+ for (int i = count - 1; i >= 0; i--) {
+ CharacterRange current = list->at(i);
+ if (current.from() > to + 1) {
+ end_pos = i;
+ } else if (current.to() + 1 < from) {
+ start_pos = i + 1;
+ break;
+ }
+ }
+
+ // Inserted range overlaps, or is adjacent to, ranges at positions
+ // [start_pos..end_pos[. Ranges before start_pos or at or after end_pos are
+ // not affected by the insertion.
+ // If start_pos == end_pos, the range must be inserted before start_pos.
+ // if start_pos < end_pos, the entire range from start_pos to end_pos
+ // must be merged with the insert range.
+
+ if (start_pos == end_pos) {
+ // Insert between existing ranges at position start_pos.
+ if (start_pos < count) {
+ MoveRanges(list, start_pos, start_pos + 1, count - start_pos);
+ }
+ list->at(start_pos) = insert;
+ return count + 1;
+ }
+ if (start_pos + 1 == end_pos) {
+ // Replace single existing range at position start_pos.
+ 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::Range(new_from, new_to);
+ return count;
+ }
+ // Replace a number of existing ranges from start_pos to end_pos - 1.
+ // Move the remaining ranges down.
+
+ int new_from = Min(list->at(start_pos).from(), from);
+ int new_to = Max(list->at(end_pos - 1).to(), to);
+ if (end_pos < count) {
+ MoveRanges(list, end_pos, start_pos + 1, count - end_pos);
+ }
+ list->at(start_pos) = CharacterRange::Range(new_from, new_to);
+ return count - (end_pos - start_pos) + 1;
+}
+
+void CharacterSet::Canonicalize() {
+ // Special/default classes are always considered canonical. The result
+ // of calling ranges() will be sorted.
+ if (ranges_ == nullptr) return;
+ CharacterRange::Canonicalize(ranges_);
+}
+
+void CharacterRange::Canonicalize(ZoneList<CharacterRange>* character_ranges) {
+ if (character_ranges->length() <= 1) return;
+ // Check whether ranges are already canonical (increasing, non-overlapping,
+ // non-adjacent).
+ int n = character_ranges->length();
+ int max = character_ranges->at(0).to();
+ int i = 1;
+ while (i < n) {
+ CharacterRange current = character_ranges->at(i);
+ if (current.from() <= max + 1) {
+ break;
+ }
+ max = current.to();
+ i++;
+ }
+ // Canonical until the i'th range. If that's all of them, we are done.
+ if (i == n) return;
+
+ // The ranges at index i and forward are not canonicalized. Make them so by
+ // doing the equivalent of insertion sort (inserting each into the previous
+ // list, in order).
+ // Notice that inserting a range can reduce the number of ranges in the
+ // result due to combining of adjacent and overlapping ranges.
+ int read = i; // Range to insert.
+ int num_canonical = i; // Length of canonicalized part of list.
+ do {
+ num_canonical = InsertRangeInCanonicalList(character_ranges, num_canonical,
+ character_ranges->at(read));
+ read++;
+ } while (read < n);
+ character_ranges->Rewind(num_canonical);
+
+ DCHECK(CharacterRange::IsCanonical(character_ranges));
+}
+
+void CharacterRange::Negate(ZoneList<CharacterRange>* ranges,
+ ZoneList<CharacterRange>* negated_ranges,
+ Zone* zone) {
+ DCHECK(CharacterRange::IsCanonical(ranges));
+ DCHECK_EQ(0, negated_ranges->length());
+ int range_count = ranges->length();
+ uc32 from = 0;
+ int i = 0;
+ if (range_count > 0 && ranges->at(0).from() == 0) {
+ from = ranges->at(0).to() + 1;
+ i = 1;
+ }
+ while (i < range_count) {
+ CharacterRange range = ranges->at(i);
+ negated_ranges->Add(CharacterRange::Range(from, range.from() - 1), zone);
+ from = range.to() + 1;
+ i++;
+ }
+ if (from < String::kMaxCodePoint) {
+ negated_ranges->Add(CharacterRange::Range(from, String::kMaxCodePoint),
+ zone);
+ }
+}
+
+// Scoped object to keep track of how much we unroll quantifier loops in the
+// regexp graph generator.
+class RegExpExpansionLimiter {
+ public:
+ static const int kMaxExpansionFactor = 6;
+ RegExpExpansionLimiter(RegExpCompiler* compiler, int factor)
+ : compiler_(compiler),
+ saved_expansion_factor_(compiler->current_expansion_factor()),
+ ok_to_expand_(saved_expansion_factor_ <= kMaxExpansionFactor) {
+ DCHECK_LT(0, factor);
+ if (ok_to_expand_) {
+ if (factor > kMaxExpansionFactor) {
+ // Avoid integer overflow of the current expansion factor.
+ ok_to_expand_ = false;
+ compiler->set_current_expansion_factor(kMaxExpansionFactor + 1);
+ } else {
+ int new_factor = saved_expansion_factor_ * factor;
+ ok_to_expand_ = (new_factor <= kMaxExpansionFactor);
+ compiler->set_current_expansion_factor(new_factor);
+ }
+ }
+ }
+
+ ~RegExpExpansionLimiter() {
+ compiler_->set_current_expansion_factor(saved_expansion_factor_);
+ }
+
+ bool ok_to_expand() { return ok_to_expand_; }
+
+ private:
+ RegExpCompiler* compiler_;
+ int saved_expansion_factor_;
+ bool ok_to_expand_;
+
+ DISALLOW_IMPLICIT_CONSTRUCTORS(RegExpExpansionLimiter);
+};
+
+RegExpNode* RegExpQuantifier::ToNode(int min, int max, bool is_greedy,
+ RegExpTree* body, RegExpCompiler* compiler,
+ RegExpNode* on_success,
+ bool not_at_start) {
+ // x{f, t} becomes this:
+ //
+ // (r++)<-.
+ // | `
+ // | (x)
+ // v ^
+ // (r=0)-->(?)---/ [if r < t]
+ // |
+ // [if r >= f] \----> ...
+ //
+
+ // 15.10.2.5 RepeatMatcher algorithm.
+ // The parser has already eliminated the case where max is 0. In the case
+ // where max_match is zero the parser has removed the quantifier if min was
+ // > 0 and removed the atom if min was 0. See AddQuantifierToAtom.
+
+ // If we know that we cannot match zero length then things are a little
+ // simpler since we don't need to make the special zero length match check
+ // from step 2.1. If the min and max are small we can unroll a little in
+ // this case.
+ static const int kMaxUnrolledMinMatches = 3; // Unroll (foo)+ and (foo){3,}
+ static const int kMaxUnrolledMaxMatches = 3; // Unroll (foo)? and (foo){x,3}
+ if (max == 0) return on_success; // This can happen due to recursion.
+ bool body_can_be_empty = (body->min_match() == 0);
+ int body_start_reg = RegExpCompiler::kNoRegister;
+ Interval capture_registers = body->CaptureRegisters();
+ bool needs_capture_clearing = !capture_registers.is_empty();
+ Zone* zone = compiler->zone();
+
+ if (body_can_be_empty) {
+ body_start_reg = compiler->AllocateRegister();
+ } else if (compiler->optimize() && !needs_capture_clearing) {
+ // Only unroll if there are no captures and the body can't be
+ // empty.
+ {
+ RegExpExpansionLimiter limiter(compiler, min + ((max != min) ? 1 : 0));
+ if (min > 0 && min <= kMaxUnrolledMinMatches && limiter.ok_to_expand()) {
+ int new_max = (max == kInfinity) ? max : max - min;
+ // Recurse once to get the loop or optional matches after the fixed
+ // ones.
+ RegExpNode* answer =
+ ToNode(0, new_max, is_greedy, body, compiler, on_success, true);
+ // Unroll the forced matches from 0 to min. This can cause chains of
+ // TextNodes (which the parser does not generate). These should be
+ // combined if it turns out they hinder good code generation.
+ for (int i = 0; i < min; i++) {
+ answer = body->ToNode(compiler, answer);
+ }
+ return answer;
+ }
+ }
+ if (max <= kMaxUnrolledMaxMatches && min == 0) {
+ DCHECK_LT(0, max); // Due to the 'if' above.
+ RegExpExpansionLimiter limiter(compiler, max);
+ if (limiter.ok_to_expand()) {
+ // Unroll the optional matches up to max.
+ RegExpNode* answer = on_success;
+ for (int i = 0; i < max; i++) {
+ ChoiceNode* alternation = new (zone) ChoiceNode(2, zone);
+ if (is_greedy) {
+ alternation->AddAlternative(
+ GuardedAlternative(body->ToNode(compiler, answer)));
+ alternation->AddAlternative(GuardedAlternative(on_success));
+ } else {
+ alternation->AddAlternative(GuardedAlternative(on_success));
+ alternation->AddAlternative(
+ GuardedAlternative(body->ToNode(compiler, answer)));
+ }
+ answer = alternation;
+ if (not_at_start && !compiler->read_backward()) {
+ alternation->set_not_at_start();
+ }
+ }
+ return answer;
+ }
+ }
+ }
+ bool has_min = min > 0;
+ bool has_max = max < RegExpTree::kInfinity;
+ bool needs_counter = has_min || has_max;
+ int reg_ctr = needs_counter ? compiler->AllocateRegister()
+ : RegExpCompiler::kNoRegister;
+ LoopChoiceNode* center = new (zone)
+ LoopChoiceNode(body->min_match() == 0, compiler->read_backward(), zone);
+ if (not_at_start && !compiler->read_backward()) center->set_not_at_start();
+ RegExpNode* loop_return =
+ needs_counter ? static_cast<RegExpNode*>(
+ ActionNode::IncrementRegister(reg_ctr, center))
+ : static_cast<RegExpNode*>(center);
+ if (body_can_be_empty) {
+ // If the body can be empty we need to check if it was and then
+ // backtrack.
+ loop_return =
+ ActionNode::EmptyMatchCheck(body_start_reg, reg_ctr, min, loop_return);
+ }
+ RegExpNode* body_node = body->ToNode(compiler, loop_return);
+ if (body_can_be_empty) {
+ // If the body can be empty we need to store the start position
+ // so we can bail out if it was empty.
+ body_node = ActionNode::StorePosition(body_start_reg, false, body_node);
+ }
+ if (needs_capture_clearing) {
+ // Before entering the body of this loop we need to clear captures.
+ body_node = ActionNode::ClearCaptures(capture_registers, body_node);
+ }
+ GuardedAlternative body_alt(body_node);
+ if (has_max) {
+ Guard* body_guard = new (zone) Guard(reg_ctr, Guard::LT, max);
+ body_alt.AddGuard(body_guard, zone);
+ }
+ GuardedAlternative rest_alt(on_success);
+ if (has_min) {
+ Guard* rest_guard = new (compiler->zone()) Guard(reg_ctr, Guard::GEQ, min);
+ rest_alt.AddGuard(rest_guard, zone);
+ }
+ if (is_greedy) {
+ center->AddLoopAlternative(body_alt);
+ center->AddContinueAlternative(rest_alt);
+ } else {
+ center->AddContinueAlternative(rest_alt);
+ center->AddLoopAlternative(body_alt);
+ }
+ if (needs_counter) {
+ return ActionNode::SetRegister(reg_ctr, 0, center);
+ } else {
+ return center;
+ }
+}
+
+} // namespace internal
+} // namespace v8
diff --git a/deps/v8/src/regexp/regexp-compiler.cc b/deps/v8/src/regexp/regexp-compiler.cc
new file mode 100644
index 0000000000..c643f988c0
--- /dev/null
+++ b/deps/v8/src/regexp/regexp-compiler.cc
@@ -0,0 +1,3551 @@
+// Copyright 2019 the V8 project authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+#include "src/regexp/regexp-compiler.h"
+
+#include "src/diagnostics/code-tracer.h"
+#include "src/execution/isolate.h"
+#include "src/objects/objects-inl.h"
+#include "src/regexp/regexp-macro-assembler-arch.h"
+#include "src/regexp/regexp-macro-assembler-tracer.h"
+#include "src/strings/unicode-inl.h"
+#include "src/utils/ostreams.h"
+#include "src/zone/zone-list-inl.h"
+
+#ifdef V8_INTL_SUPPORT
+#include "unicode/locid.h"
+#include "unicode/uniset.h"
+#include "unicode/utypes.h"
+#endif // V8_INTL_SUPPORT
+
+namespace v8 {
+namespace internal {
+
+using namespace regexp_compiler_constants; // NOLINT(build/namespaces)
+
+// -------------------------------------------------------------------
+// Implementation of the Irregexp regular expression engine.
+//
+// The Irregexp regular expression engine is intended to be a complete
+// implementation of ECMAScript regular expressions. It generates either
+// bytecodes or native code.
+
+// The Irregexp regexp engine is structured in three steps.
+// 1) The parser generates an abstract syntax tree. See ast.cc.
+// 2) From the AST a node network is created. The nodes are all
+// subclasses of RegExpNode. The nodes represent states when
+// executing a regular expression. Several optimizations are
+// performed on the node network.
+// 3) From the nodes we generate either byte codes or native code
+// that can actually execute the regular expression (perform
+// the search). The code generation step is described in more
+// detail below.
+
+// Code generation.
+//
+// The nodes are divided into four main categories.
+// * Choice nodes
+// These represent places where the regular expression can
+// match in more than one way. For example on entry to an
+// alternation (foo|bar) or a repetition (*, +, ? or {}).
+// * Action nodes
+// These represent places where some action should be
+// performed. Examples include recording the current position
+// in the input string to a register (in order to implement
+// captures) or other actions on register for example in order
+// to implement the counters needed for {} repetitions.
+// * Matching nodes
+// These attempt to match some element part of the input string.
+// Examples of elements include character classes, plain strings
+// or back references.
+// * End nodes
+// These are used to implement the actions required on finding
+// a successful match or failing to find a match.
+//
+// The code generated (whether as byte codes or native code) maintains
+// some state as it runs. This consists of the following elements:
+//
+// * The capture registers. Used for string captures.
+// * Other registers. Used for counters etc.
+// * The current position.
+// * The stack of backtracking information. Used when a matching node
+// fails to find a match and needs to try an alternative.
+//
+// Conceptual regular expression execution model:
+//
+// There is a simple conceptual model of regular expression execution
+// which will be presented first. The actual code generated is a more
+// efficient simulation of the simple conceptual model:
+//
+// * Choice nodes are implemented as follows:
+// For each choice except the last {
+// push current position
+// push backtrack code location
+// <generate code to test for choice>
+// backtrack code location:
+// pop current position
+// }
+// <generate code to test for last choice>
+//
+// * Actions nodes are generated as follows
+// <push affected registers on backtrack stack>
+// <generate code to perform action>
+// push backtrack code location
+// <generate code to test for following nodes>
+// backtrack code location:
+// <pop affected registers to restore their state>
+// <pop backtrack location from stack and go to it>
+//
+// * Matching nodes are generated as follows:
+// if input string matches at current position
+// update current position
+// <generate code to test for following nodes>
+// else
+// <pop backtrack location from stack and go to it>
+//
+// Thus it can be seen that the current position is saved and restored
+// by the choice nodes, whereas the registers are saved and restored by
+// by the action nodes that manipulate them.
+//
+// The other interesting aspect of this model is that nodes are generated
+// at the point where they are needed by a recursive call to Emit(). If
+// the node has already been code generated then the Emit() call will
+// generate a jump to the previously generated code instead. In order to
+// limit recursion it is possible for the Emit() function to put the node
+// on a work list for later generation and instead generate a jump. The
+// destination of the jump is resolved later when the code is generated.
+//
+// Actual regular expression code generation.
+//
+// Code generation is actually more complicated than the above. In order
+// to improve the efficiency of the generated code some optimizations are
+// performed
+//
+// * Choice nodes have 1-character lookahead.
+// A choice node looks at the following character and eliminates some of
+// the choices immediately based on that character. This is not yet
+// implemented.
+// * Simple greedy loops store reduced backtracking information.
+// A quantifier like /.*foo/m will greedily match the whole input. It will
+// then need to backtrack to a point where it can match "foo". The naive
+// implementation of this would push each character position onto the
+// backtracking stack, then pop them off one by one. This would use space
+// proportional to the length of the input string. However since the "."
+// can only match in one way and always has a constant length (in this case
+// of 1) it suffices to store the current position on the top of the stack
+// once. Matching now becomes merely incrementing the current position and
+// backtracking becomes decrementing the current position and checking the
+// result against the stored current position. This is faster and saves
+// space.
+// * The current state is virtualized.
+// This is used to defer expensive operations until it is clear that they
+// are needed and to generate code for a node more than once, allowing
+// specialized an efficient versions of the code to be created. This is
+// explained in the section below.
+//
+// Execution state virtualization.
+//
+// Instead of emitting code, nodes that manipulate the state can record their
+// manipulation in an object called the Trace. The Trace object can record a
+// current position offset, an optional backtrack code location on the top of
+// the virtualized backtrack stack and some register changes. When a node is
+// to be emitted it can flush the Trace or update it. Flushing the Trace
+// will emit code to bring the actual state into line with the virtual state.
+// Avoiding flushing the state can postpone some work (e.g. updates of capture
+// registers). Postponing work can save time when executing the regular
+// expression since it may be found that the work never has to be done as a
+// failure to match can occur. In addition it is much faster to jump to a
+// known backtrack code location than it is to pop an unknown backtrack
+// location from the stack and jump there.
+//
+// The virtual state found in the Trace affects code generation. For example
+// the virtual state contains the difference between the actual current
+// position and the virtual current position, and matching code needs to use
+// this offset to attempt a match in the correct location of the input
+// string. Therefore code generated for a non-trivial trace is specialized
+// to that trace. The code generator therefore has the ability to generate
+// code for each node several times. In order to limit the size of the
+// generated code there is an arbitrary limit on how many specialized sets of
+// code may be generated for a given node. If the limit is reached, the
+// trace is flushed and a generic version of the code for a node is emitted.
+// This is subsequently used for that node. The code emitted for non-generic
+// trace is not recorded in the node and so it cannot currently be reused in
+// the event that code generation is requested for an identical trace.
+
+void RegExpTree::AppendToText(RegExpText* text, Zone* zone) { UNREACHABLE(); }
+
+void RegExpAtom::AppendToText(RegExpText* text, Zone* zone) {
+ text->AddElement(TextElement::Atom(this), zone);
+}
+
+void RegExpCharacterClass::AppendToText(RegExpText* text, Zone* zone) {
+ text->AddElement(TextElement::CharClass(this), zone);
+}
+
+void RegExpText::AppendToText(RegExpText* text, Zone* zone) {
+ for (int i = 0; i < elements()->length(); i++)
+ text->AddElement(elements()->at(i), zone);
+}
+
+TextElement TextElement::Atom(RegExpAtom* atom) {
+ return TextElement(ATOM, atom);
+}
+
+TextElement TextElement::CharClass(RegExpCharacterClass* char_class) {
+ return TextElement(CHAR_CLASS, char_class);
+}
+
+int TextElement::length() const {
+ switch (text_type()) {
+ case ATOM:
+ return atom()->length();
+
+ case CHAR_CLASS:
+ return 1;
+ }
+ UNREACHABLE();
+}
+
+class RecursionCheck {
+ public:
+ explicit RecursionCheck(RegExpCompiler* compiler) : compiler_(compiler) {
+ compiler->IncrementRecursionDepth();
+ }
+ ~RecursionCheck() { compiler_->DecrementRecursionDepth(); }
+
+ private:
+ RegExpCompiler* compiler_;
+};
+
+// 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 one_byte)
+ : next_register_(2 * (capture_count + 1)),
+ unicode_lookaround_stack_register_(kNoRegister),
+ unicode_lookaround_position_register_(kNoRegister),
+ work_list_(nullptr),
+ recursion_depth_(0),
+ one_byte_(one_byte),
+ reg_exp_too_big_(false),
+ limiting_recursion_(false),
+ optimize_(FLAG_regexp_optimization),
+ read_backward_(false),
+ current_expansion_factor_(1),
+ frequency_collator_(),
+ isolate_(isolate),
+ zone_(zone) {
+ accept_ = new (zone) EndNode(EndNode::ACCEPT, zone);
+ DCHECK_GE(RegExpMacroAssembler::kMaxRegister, next_register_ - 1);
+}
+
+RegExpCompiler::CompilationResult RegExpCompiler::Assemble(
+ Isolate* isolate, RegExpMacroAssembler* macro_assembler, RegExpNode* start,
+ int capture_count, Handle<String> pattern) {
+#ifdef DEBUG
+ if (FLAG_trace_regexp_assembler)
+ macro_assembler_ = new RegExpMacroAssemblerTracer(isolate, macro_assembler);
+ else
+#endif
+ macro_assembler_ = macro_assembler;
+
+ std::vector<RegExpNode*> work_list;
+ work_list_ = &work_list;
+ Label fail;
+ macro_assembler_->PushBacktrack(&fail);
+ Trace new_trace;
+ start->Emit(this, &new_trace);
+ macro_assembler_->Bind(&fail);
+ macro_assembler_->Fail();
+ while (!work_list.empty()) {
+ RegExpNode* node = work_list.back();
+ work_list.pop_back();
+ node->set_on_work_list(false);
+ if (!node->label()->is_bound()) node->Emit(this, &new_trace);
+ }
+ if (reg_exp_too_big_) {
+ macro_assembler_->AbortedCodeGeneration();
+ return CompilationResult::RegExpTooBig();
+ }
+
+ Handle<HeapObject> code = macro_assembler_->GetCode(pattern);
+ isolate->IncreaseTotalRegexpCodeGenerated(code->Size());
+ work_list_ = nullptr;
+#ifdef ENABLE_DISASSEMBLER
+ if (FLAG_print_code && !FLAG_regexp_interpret_all) {
+ CodeTracer::Scope trace_scope(isolate->GetCodeTracer());
+ OFStream os(trace_scope.file());
+ Handle<Code>::cast(code)->Disassemble(pattern->ToCString().get(), os);
+ }
+#endif
+#ifdef DEBUG
+ if (FLAG_trace_regexp_assembler) {
+ delete macro_assembler_;
+ }
+#endif
+ return {*code, next_register_};
+}
+
+bool Trace::DeferredAction::Mentions(int that) {
+ if (action_type() == ActionNode::CLEAR_CAPTURES) {
+ Interval range = static_cast<DeferredClearCaptures*>(this)->range();
+ return range.Contains(that);
+ } else {
+ return reg() == that;
+ }
+}
+
+bool Trace::mentions_reg(int reg) {
+ for (DeferredAction* action = actions_; action != nullptr;
+ action = action->next()) {
+ if (action->Mentions(reg)) return true;
+ }
+ return false;
+}
+
+bool Trace::GetStoredPosition(int reg, int* cp_offset) {
+ DCHECK_EQ(0, *cp_offset);
+ for (DeferredAction* action = actions_; action != nullptr;
+ action = action->next()) {
+ if (action->Mentions(reg)) {
+ if (action->action_type() == ActionNode::STORE_POSITION) {
+ *cp_offset = static_cast<DeferredCapture*>(action)->cp_offset();
+ return true;
+ } else {
+ return false;
+ }
+ }
+ }
+ return false;
+}
+
+// A (dynamically-sized) set of unsigned integers that behaves especially well
+// on small integers (< kFirstLimit). May do zone-allocation.
+class DynamicBitSet : public ZoneObject {
+ public:
+ V8_EXPORT_PRIVATE bool Get(unsigned value) const {
+ if (value < kFirstLimit) {
+ return (first_ & (1 << value)) != 0;
+ } else if (remaining_ == nullptr) {
+ return false;
+ } else {
+ return remaining_->Contains(value);
+ }
+ }
+
+ // Destructively set a value in this set.
+ void Set(unsigned value, Zone* zone) {
+ if (value < kFirstLimit) {
+ first_ |= (1 << value);
+ } else {
+ if (remaining_ == nullptr)
+ remaining_ = new (zone) ZoneList<unsigned>(1, zone);
+ if (remaining_->is_empty() || !remaining_->Contains(value))
+ remaining_->Add(value, zone);
+ }
+ }
+
+ private:
+ static constexpr unsigned kFirstLimit = 32;
+
+ uint32_t first_ = 0;
+ ZoneList<unsigned>* remaining_ = nullptr;
+};
+
+int Trace::FindAffectedRegisters(DynamicBitSet* affected_registers,
+ Zone* zone) {
+ int max_register = RegExpCompiler::kNoRegister;
+ for (DeferredAction* action = actions_; action != nullptr;
+ action = action->next()) {
+ if (action->action_type() == ActionNode::CLEAR_CAPTURES) {
+ Interval range = static_cast<DeferredClearCaptures*>(action)->range();
+ for (int i = range.from(); i <= range.to(); i++)
+ affected_registers->Set(i, zone);
+ if (range.to() > max_register) max_register = range.to();
+ } else {
+ affected_registers->Set(action->reg(), zone);
+ if (action->reg() > max_register) max_register = action->reg();
+ }
+ }
+ return max_register;
+}
+
+void Trace::RestoreAffectedRegisters(RegExpMacroAssembler* assembler,
+ int max_register,
+ const DynamicBitSet& registers_to_pop,
+ const DynamicBitSet& registers_to_clear) {
+ for (int reg = max_register; reg >= 0; reg--) {
+ if (registers_to_pop.Get(reg)) {
+ assembler->PopRegister(reg);
+ } else if (registers_to_clear.Get(reg)) {
+ int clear_to = reg;
+ while (reg > 0 && registers_to_clear.Get(reg - 1)) {
+ reg--;
+ }
+ assembler->ClearRegisters(reg, clear_to);
+ }
+ }
+}
+
+void Trace::PerformDeferredActions(RegExpMacroAssembler* assembler,
+ int max_register,
+ const DynamicBitSet& affected_registers,
+ DynamicBitSet* registers_to_pop,
+ DynamicBitSet* registers_to_clear,
+ Zone* zone) {
+ // The "+1" is to avoid a push_limit of zero if stack_limit_slack() is 1.
+ const int push_limit = (assembler->stack_limit_slack() + 1) / 2;
+
+ // Count pushes performed to force a stack limit check occasionally.
+ int pushes = 0;
+
+ for (int reg = 0; reg <= max_register; reg++) {
+ if (!affected_registers.Get(reg)) {
+ continue;
+ }
+
+ // The chronologically first deferred action in the trace
+ // is used to infer the action needed to restore a register
+ // to its previous state (or not, if it's safe to ignore it).
+ enum DeferredActionUndoType { IGNORE, RESTORE, CLEAR };
+ DeferredActionUndoType undo_action = IGNORE;
+
+ int value = 0;
+ bool absolute = false;
+ bool clear = false;
+ static const int kNoStore = kMinInt;
+ int store_position = kNoStore;
+ // This is a little tricky because we are scanning the actions in reverse
+ // historical order (newest first).
+ for (DeferredAction* action = actions_; action != nullptr;
+ action = action->next()) {
+ if (action->Mentions(reg)) {
+ switch (action->action_type()) {
+ case ActionNode::SET_REGISTER: {
+ Trace::DeferredSetRegister* psr =
+ static_cast<Trace::DeferredSetRegister*>(action);
+ if (!absolute) {
+ value += psr->value();
+ absolute = true;
+ }
+ // SET_REGISTER is currently only used for newly introduced loop
+ // counters. They can have a significant previous value if they
+ // occur in a loop. TODO(lrn): Propagate this information, so
+ // we can set undo_action to IGNORE if we know there is no value to
+ // restore.
+ undo_action = RESTORE;
+ DCHECK_EQ(store_position, kNoStore);
+ DCHECK(!clear);
+ break;
+ }
+ case ActionNode::INCREMENT_REGISTER:
+ if (!absolute) {
+ value++;
+ }
+ DCHECK_EQ(store_position, kNoStore);
+ DCHECK(!clear);
+ undo_action = RESTORE;
+ break;
+ case ActionNode::STORE_POSITION: {
+ Trace::DeferredCapture* pc =
+ static_cast<Trace::DeferredCapture*>(action);
+ if (!clear && store_position == kNoStore) {
+ store_position = pc->cp_offset();
+ }
+
+ // For captures we know that stores and clears alternate.
+ // Other register, are never cleared, and if the occur
+ // inside a loop, they might be assigned more than once.
+ if (reg <= 1) {
+ // Registers zero and one, aka "capture zero", is
+ // always set correctly if we succeed. There is no
+ // need to undo a setting on backtrack, because we
+ // will set it again or fail.
+ undo_action = IGNORE;
+ } else {
+ undo_action = pc->is_capture() ? CLEAR : RESTORE;
+ }
+ DCHECK(!absolute);
+ DCHECK_EQ(value, 0);
+ break;
+ }
+ case ActionNode::CLEAR_CAPTURES: {
+ // Since we're scanning in reverse order, if we've already
+ // set the position we have to ignore historically earlier
+ // clearing operations.
+ if (store_position == kNoStore) {
+ clear = true;
+ }
+ undo_action = RESTORE;
+ DCHECK(!absolute);
+ DCHECK_EQ(value, 0);
+ break;
+ }
+ default:
+ UNREACHABLE();
+ break;
+ }
+ }
+ }
+ // Prepare for the undo-action (e.g., push if it's going to be popped).
+ if (undo_action == RESTORE) {
+ pushes++;
+ RegExpMacroAssembler::StackCheckFlag stack_check =
+ RegExpMacroAssembler::kNoStackLimitCheck;
+ if (pushes == push_limit) {
+ stack_check = RegExpMacroAssembler::kCheckStackLimit;
+ pushes = 0;
+ }
+
+ assembler->PushRegister(reg, stack_check);
+ registers_to_pop->Set(reg, zone);
+ } else if (undo_action == CLEAR) {
+ registers_to_clear->Set(reg, zone);
+ }
+ // Perform the chronologically last action (or accumulated increment)
+ // for the register.
+ if (store_position != kNoStore) {
+ assembler->WriteCurrentPositionToRegister(reg, store_position);
+ } else if (clear) {
+ assembler->ClearRegisters(reg, reg);
+ } else if (absolute) {
+ assembler->SetRegister(reg, value);
+ } else if (value != 0) {
+ assembler->AdvanceRegister(reg, value);
+ }
+ }
+}
+
+// This is called as we come into a loop choice node and some other tricky
+// nodes. It normalizes the state of the code generator to ensure we can
+// generate generic code.
+void Trace::Flush(RegExpCompiler* compiler, RegExpNode* successor) {
+ RegExpMacroAssembler* assembler = compiler->macro_assembler();
+
+ DCHECK(!is_trivial());
+
+ if (actions_ == nullptr && backtrack() == nullptr) {
+ // Here we just have some deferred cp advances to fix and we are back to
+ // a normal situation. We may also have to forget some information gained
+ // through a quick check that was already performed.
+ if (cp_offset_ != 0) assembler->AdvanceCurrentPosition(cp_offset_);
+ // Create a new trivial state and generate the node with that.
+ Trace new_state;
+ successor->Emit(compiler, &new_state);
+ return;
+ }
+
+ // Generate deferred actions here along with code to undo them again.
+ DynamicBitSet affected_registers;
+
+ if (backtrack() != nullptr) {
+ // Here we have a concrete backtrack location. These are set up by choice
+ // nodes and so they indicate that we have a deferred save of the current
+ // position which we may need to emit here.
+ assembler->PushCurrentPosition();
+ }
+
+ int max_register =
+ FindAffectedRegisters(&affected_registers, compiler->zone());
+ DynamicBitSet registers_to_pop;
+ DynamicBitSet registers_to_clear;
+ PerformDeferredActions(assembler, max_register, affected_registers,
+ &registers_to_pop, &registers_to_clear,
+ compiler->zone());
+ if (cp_offset_ != 0) {
+ assembler->AdvanceCurrentPosition(cp_offset_);
+ }
+
+ // Create a new trivial state and generate the node with that.
+ Label undo;
+ assembler->PushBacktrack(&undo);
+ if (successor->KeepRecursing(compiler)) {
+ Trace new_state;
+ successor->Emit(compiler, &new_state);
+ } else {
+ compiler->AddWork(successor);
+ assembler->GoTo(successor->label());
+ }
+
+ // On backtrack we need to restore state.
+ assembler->Bind(&undo);
+ RestoreAffectedRegisters(assembler, max_register, registers_to_pop,
+ registers_to_clear);
+ if (backtrack() == nullptr) {
+ assembler->Backtrack();
+ } else {
+ assembler->PopCurrentPosition();
+ assembler->GoTo(backtrack());
+ }
+}
+
+void NegativeSubmatchSuccess::Emit(RegExpCompiler* compiler, Trace* trace) {
+ RegExpMacroAssembler* assembler = compiler->macro_assembler();
+
+ // Omit flushing the trace. We discard the entire stack frame anyway.
+
+ if (!label()->is_bound()) {
+ // We are completely independent of the trace, since we ignore it,
+ // so this code can be used as the generic version.
+ assembler->Bind(label());
+ }
+
+ // Throw away everything on the backtrack stack since the start
+ // of the negative submatch and restore the character position.
+ assembler->ReadCurrentPositionFromRegister(current_position_register_);
+ assembler->ReadStackPointerFromRegister(stack_pointer_register_);
+ if (clear_capture_count_ > 0) {
+ // Clear any captures that might have been performed during the success
+ // of the body of the negative look-ahead.
+ int clear_capture_end = clear_capture_start_ + clear_capture_count_ - 1;
+ assembler->ClearRegisters(clear_capture_start_, clear_capture_end);
+ }
+ // Now that we have unwound the stack we find at the top of the stack the
+ // backtrack that the BeginSubmatch node got.
+ assembler->Backtrack();
+}
+
+void EndNode::Emit(RegExpCompiler* compiler, Trace* trace) {
+ if (!trace->is_trivial()) {
+ trace->Flush(compiler, this);
+ return;
+ }
+ RegExpMacroAssembler* assembler = compiler->macro_assembler();
+ if (!label()->is_bound()) {
+ assembler->Bind(label());
+ }
+ switch (action_) {
+ case ACCEPT:
+ assembler->Succeed();
+ return;
+ case BACKTRACK:
+ assembler->GoTo(trace->backtrack());
+ return;
+ case NEGATIVE_SUBMATCH_SUCCESS:
+ // This case is handled in a different virtual method.
+ UNREACHABLE();
+ }
+ UNIMPLEMENTED();
+}
+
+void GuardedAlternative::AddGuard(Guard* guard, Zone* zone) {
+ if (guards_ == nullptr) guards_ = new (zone) ZoneList<Guard*>(1, zone);
+ guards_->Add(guard, zone);
+}
+
+ActionNode* ActionNode::SetRegister(int reg, int val, RegExpNode* on_success) {
+ ActionNode* result =
+ new (on_success->zone()) ActionNode(SET_REGISTER, on_success);
+ result->data_.u_store_register.reg = reg;
+ result->data_.u_store_register.value = val;
+ return result;
+}
+
+ActionNode* ActionNode::IncrementRegister(int reg, RegExpNode* on_success) {
+ ActionNode* result =
+ new (on_success->zone()) ActionNode(INCREMENT_REGISTER, on_success);
+ result->data_.u_increment_register.reg = reg;
+ return result;
+}
+
+ActionNode* ActionNode::StorePosition(int reg, bool is_capture,
+ RegExpNode* on_success) {
+ ActionNode* result =
+ new (on_success->zone()) ActionNode(STORE_POSITION, on_success);
+ result->data_.u_position_register.reg = reg;
+ result->data_.u_position_register.is_capture = is_capture;
+ return result;
+}
+
+ActionNode* ActionNode::ClearCaptures(Interval range, RegExpNode* on_success) {
+ ActionNode* result =
+ new (on_success->zone()) ActionNode(CLEAR_CAPTURES, on_success);
+ result->data_.u_clear_captures.range_from = range.from();
+ result->data_.u_clear_captures.range_to = range.to();
+ return result;
+}
+
+ActionNode* ActionNode::BeginSubmatch(int stack_reg, int position_reg,
+ RegExpNode* on_success) {
+ ActionNode* result =
+ new (on_success->zone()) ActionNode(BEGIN_SUBMATCH, on_success);
+ result->data_.u_submatch.stack_pointer_register = stack_reg;
+ result->data_.u_submatch.current_position_register = position_reg;
+ return result;
+}
+
+ActionNode* ActionNode::PositiveSubmatchSuccess(int stack_reg, int position_reg,
+ int clear_register_count,
+ int clear_register_from,
+ RegExpNode* on_success) {
+ ActionNode* result = new (on_success->zone())
+ ActionNode(POSITIVE_SUBMATCH_SUCCESS, on_success);
+ result->data_.u_submatch.stack_pointer_register = stack_reg;
+ result->data_.u_submatch.current_position_register = position_reg;
+ result->data_.u_submatch.clear_register_count = clear_register_count;
+ result->data_.u_submatch.clear_register_from = clear_register_from;
+ return result;
+}
+
+ActionNode* ActionNode::EmptyMatchCheck(int start_register,
+ int repetition_register,
+ int repetition_limit,
+ RegExpNode* on_success) {
+ ActionNode* result =
+ new (on_success->zone()) ActionNode(EMPTY_MATCH_CHECK, on_success);
+ result->data_.u_empty_match_check.start_register = start_register;
+ result->data_.u_empty_match_check.repetition_register = repetition_register;
+ result->data_.u_empty_match_check.repetition_limit = repetition_limit;
+ return result;
+}
+
+#define DEFINE_ACCEPT(Type) \
+ void Type##Node::Accept(NodeVisitor* visitor) { visitor->Visit##Type(this); }
+FOR_EACH_NODE_TYPE(DEFINE_ACCEPT)
+#undef DEFINE_ACCEPT
+
+void LoopChoiceNode::Accept(NodeVisitor* visitor) {
+ visitor->VisitLoopChoice(this);
+}
+
+// -------------------------------------------------------------------
+// Emit code.
+
+void ChoiceNode::GenerateGuard(RegExpMacroAssembler* macro_assembler,
+ Guard* guard, Trace* trace) {
+ switch (guard->op()) {
+ case Guard::LT:
+ DCHECK(!trace->mentions_reg(guard->reg()));
+ macro_assembler->IfRegisterGE(guard->reg(), guard->value(),
+ trace->backtrack());
+ break;
+ case Guard::GEQ:
+ DCHECK(!trace->mentions_reg(guard->reg()));
+ macro_assembler->IfRegisterLT(guard->reg(), guard->value(),
+ trace->backtrack());
+ break;
+ }
+}
+
+// Returns the number of characters in the equivalence class, omitting those
+// that cannot occur in the source string because it is Latin1.
+static int GetCaseIndependentLetters(Isolate* isolate, uc16 character,
+ bool one_byte_subject,
+ unibrow::uchar* letters,
+ int letter_length) {
+#ifdef V8_INTL_SUPPORT
+ icu::UnicodeSet set;
+ set.add(character);
+ set = set.closeOver(USET_CASE_INSENSITIVE);
+ int32_t range_count = set.getRangeCount();
+ int items = 0;
+ for (int32_t i = 0; i < range_count; i++) {
+ UChar32 start = set.getRangeStart(i);
+ UChar32 end = set.getRangeEnd(i);
+ CHECK(end - start + items <= letter_length);
+ while (start <= end) {
+ if (one_byte_subject && start > String::kMaxOneByteCharCode) break;
+ letters[items++] = (unibrow::uchar)(start);
+ start++;
+ }
+ }
+ return items;
+#else
+ int length =
+ isolate->jsregexp_uncanonicalize()->get(character, '\0', letters);
+ // Unibrow returns 0 or 1 for characters where case independence is
+ // trivial.
+ if (length == 0) {
+ letters[0] = character;
+ length = 1;
+ }
+
+ if (one_byte_subject) {
+ int new_length = 0;
+ for (int i = 0; i < length; i++) {
+ if (letters[i] <= String::kMaxOneByteCharCode) {
+ letters[new_length++] = letters[i];
+ }
+ }
+ length = new_length;
+ }
+
+ return length;
+#endif // V8_INTL_SUPPORT
+}
+
+static inline bool EmitSimpleCharacter(Isolate* isolate,
+ RegExpCompiler* compiler, uc16 c,
+ Label* on_failure, int cp_offset,
+ bool check, bool preloaded) {
+ RegExpMacroAssembler* assembler = compiler->macro_assembler();
+ bool bound_checked = false;
+ if (!preloaded) {
+ assembler->LoadCurrentCharacter(cp_offset, on_failure, check);
+ bound_checked = true;
+ }
+ assembler->CheckNotCharacter(c, on_failure);
+ return bound_checked;
+}
+
+// Only emits non-letters (things that don't have case). Only used for case
+// independent matches.
+static inline bool EmitAtomNonLetter(Isolate* isolate, RegExpCompiler* compiler,
+ uc16 c, Label* on_failure, int cp_offset,
+ bool check, bool preloaded) {
+ RegExpMacroAssembler* macro_assembler = compiler->macro_assembler();
+ bool one_byte = compiler->one_byte();
+ unibrow::uchar chars[4];
+ int length = GetCaseIndependentLetters(isolate, c, one_byte, chars, 4);
+ if (length < 1) {
+ // This can't match. Must be an one-byte subject and a non-one-byte
+ // character. We do not need to do anything since the one-byte pass
+ // already handled this.
+ return false; // Bounds not checked.
+ }
+ bool checked = false;
+ // We handle the length > 1 case in a later pass.
+ if (length == 1) {
+ if (one_byte && c > String::kMaxOneByteCharCodeU) {
+ // Can't match - see above.
+ return false; // Bounds not checked.
+ }
+ if (!preloaded) {
+ macro_assembler->LoadCurrentCharacter(cp_offset, on_failure, check);
+ checked = check;
+ }
+ macro_assembler->CheckNotCharacter(c, on_failure);
+ }
+ return checked;
+}
+
+static bool ShortCutEmitCharacterPair(RegExpMacroAssembler* macro_assembler,
+ bool one_byte, uc16 c1, uc16 c2,
+ Label* on_failure) {
+ uc16 char_mask;
+ if (one_byte) {
+ char_mask = String::kMaxOneByteCharCode;
+ } else {
+ char_mask = String::kMaxUtf16CodeUnit;
+ }
+ uc16 exor = c1 ^ c2;
+ // Check whether exor has only one bit set.
+ if (((exor - 1) & exor) == 0) {
+ // If c1 and c2 differ only by one bit.
+ // Ecma262UnCanonicalize always gives the highest number last.
+ DCHECK(c2 > c1);
+ uc16 mask = char_mask ^ exor;
+ macro_assembler->CheckNotCharacterAfterAnd(c1, mask, on_failure);
+ return true;
+ }
+ DCHECK(c2 > c1);
+ uc16 diff = c2 - c1;
+ if (((diff - 1) & diff) == 0 && c1 >= diff) {
+ // If the characters differ by 2^n but don't differ by one bit then
+ // subtract the difference from the found character, then do the or
+ // trick. We avoid the theoretical case where negative numbers are
+ // involved in order to simplify code generation.
+ uc16 mask = char_mask ^ diff;
+ macro_assembler->CheckNotCharacterAfterMinusAnd(c1 - diff, diff, mask,
+ on_failure);
+ return true;
+ }
+ return false;
+}
+
+using EmitCharacterFunction = bool(Isolate* isolate, RegExpCompiler* compiler,
+ uc16 c, Label* on_failure, int cp_offset,
+ bool check, bool preloaded);
+
+// Only emits letters (things that have case). Only used for case independent
+// matches.
+static inline bool EmitAtomLetter(Isolate* isolate, RegExpCompiler* compiler,
+ uc16 c, Label* on_failure, int cp_offset,
+ bool check, bool preloaded) {
+ RegExpMacroAssembler* macro_assembler = compiler->macro_assembler();
+ bool one_byte = compiler->one_byte();
+ unibrow::uchar chars[4];
+ int length = GetCaseIndependentLetters(isolate, c, one_byte, chars, 4);
+ if (length <= 1) return false;
+ // We may not need to check against the end of the input string
+ // if this character lies before a character that matched.
+ if (!preloaded) {
+ macro_assembler->LoadCurrentCharacter(cp_offset, on_failure, check);
+ }
+ Label ok;
+ switch (length) {
+ case 2: {
+ if (ShortCutEmitCharacterPair(macro_assembler, one_byte, chars[0],
+ chars[1], on_failure)) {
+ } else {
+ macro_assembler->CheckCharacter(chars[0], &ok);
+ macro_assembler->CheckNotCharacter(chars[1], on_failure);
+ macro_assembler->Bind(&ok);
+ }
+ break;
+ }
+ case 4:
+ macro_assembler->CheckCharacter(chars[3], &ok);
+ V8_FALLTHROUGH;
+ case 3:
+ macro_assembler->CheckCharacter(chars[0], &ok);
+ macro_assembler->CheckCharacter(chars[1], &ok);
+ macro_assembler->CheckNotCharacter(chars[2], on_failure);
+ macro_assembler->Bind(&ok);
+ break;
+ default:
+ UNREACHABLE();
+ }
+ return true;
+}
+
+static void EmitBoundaryTest(RegExpMacroAssembler* masm, int border,
+ Label* fall_through, Label* above_or_equal,
+ Label* below) {
+ if (below != fall_through) {
+ masm->CheckCharacterLT(border, below);
+ if (above_or_equal != fall_through) masm->GoTo(above_or_equal);
+ } else {
+ masm->CheckCharacterGT(border - 1, above_or_equal);
+ }
+}
+
+static void EmitDoubleBoundaryTest(RegExpMacroAssembler* masm, int first,
+ int last, Label* fall_through,
+ Label* in_range, Label* out_of_range) {
+ if (in_range == fall_through) {
+ if (first == last) {
+ masm->CheckNotCharacter(first, out_of_range);
+ } else {
+ masm->CheckCharacterNotInRange(first, last, out_of_range);
+ }
+ } else {
+ if (first == last) {
+ masm->CheckCharacter(first, in_range);
+ } else {
+ masm->CheckCharacterInRange(first, last, in_range);
+ }
+ if (out_of_range != fall_through) masm->GoTo(out_of_range);
+ }
+}
+
+// even_label is for ranges[i] to ranges[i + 1] where i - start_index is even.
+// odd_label is for ranges[i] to ranges[i + 1] where i - start_index is odd.
+static void EmitUseLookupTable(RegExpMacroAssembler* masm,
+ ZoneList<int>* ranges, int start_index,
+ int end_index, int min_char, Label* fall_through,
+ Label* even_label, Label* odd_label) {
+ static const int kSize = RegExpMacroAssembler::kTableSize;
+ static const int kMask = RegExpMacroAssembler::kTableMask;
+
+ int base = (min_char & ~kMask);
+ USE(base);
+
+ // Assert that everything is on one kTableSize page.
+ for (int i = start_index; i <= end_index; i++) {
+ DCHECK_EQ(ranges->at(i) & ~kMask, base);
+ }
+ DCHECK(start_index == 0 || (ranges->at(start_index - 1) & ~kMask) <= base);
+
+ char templ[kSize];
+ Label* on_bit_set;
+ Label* on_bit_clear;
+ int bit;
+ if (even_label == fall_through) {
+ on_bit_set = odd_label;
+ on_bit_clear = even_label;
+ bit = 1;
+ } else {
+ on_bit_set = even_label;
+ on_bit_clear = odd_label;
+ bit = 0;
+ }
+ for (int i = 0; i < (ranges->at(start_index) & kMask) && i < kSize; i++) {
+ templ[i] = bit;
+ }
+ int j = 0;
+ bit ^= 1;
+ for (int i = start_index; i < end_index; i++) {
+ for (j = (ranges->at(i) & kMask); j < (ranges->at(i + 1) & kMask); j++) {
+ templ[j] = bit;
+ }
+ bit ^= 1;
+ }
+ for (int i = j; i < kSize; i++) {
+ templ[i] = bit;
+ }
+ Factory* factory = masm->isolate()->factory();
+ // TODO(erikcorry): Cache these.
+ Handle<ByteArray> ba = factory->NewByteArray(kSize, AllocationType::kOld);
+ for (int i = 0; i < kSize; i++) {
+ ba->set(i, templ[i]);
+ }
+ masm->CheckBitInTable(ba, on_bit_set);
+ if (on_bit_clear != fall_through) masm->GoTo(on_bit_clear);
+}
+
+static void CutOutRange(RegExpMacroAssembler* masm, ZoneList<int>* ranges,
+ int start_index, int end_index, int cut_index,
+ Label* even_label, Label* odd_label) {
+ bool odd = (((cut_index - start_index) & 1) == 1);
+ Label* in_range_label = odd ? odd_label : even_label;
+ Label dummy;
+ EmitDoubleBoundaryTest(masm, ranges->at(cut_index),
+ ranges->at(cut_index + 1) - 1, &dummy, in_range_label,
+ &dummy);
+ DCHECK(!dummy.is_linked());
+ // Cut out the single range by rewriting the array. This creates a new
+ // range that is a merger of the two ranges on either side of the one we
+ // are cutting out. The oddity of the labels is preserved.
+ for (int j = cut_index; j > start_index; j--) {
+ ranges->at(j) = ranges->at(j - 1);
+ }
+ for (int j = cut_index + 1; j < end_index; j++) {
+ ranges->at(j) = ranges->at(j + 1);
+ }
+}
+
+// Unicode case. Split the search space into kSize spaces that are handled
+// with recursion.
+static void SplitSearchSpace(ZoneList<int>* ranges, int start_index,
+ int end_index, int* new_start_index,
+ int* new_end_index, int* border) {
+ static const int kSize = RegExpMacroAssembler::kTableSize;
+ static const int kMask = RegExpMacroAssembler::kTableMask;
+
+ int first = ranges->at(start_index);
+ int last = ranges->at(end_index) - 1;
+
+ *new_start_index = start_index;
+ *border = (ranges->at(start_index) & ~kMask) + kSize;
+ while (*new_start_index < end_index) {
+ if (ranges->at(*new_start_index) > *border) break;
+ (*new_start_index)++;
+ }
+ // new_start_index is the index of the first edge that is beyond the
+ // current kSize space.
+
+ // For very large search spaces we do a binary chop search of the non-Latin1
+ // space instead of just going to the end of the current kSize space. The
+ // heuristics are complicated a little by the fact that any 128-character
+ // encoding space can be quickly tested with a table lookup, so we don't
+ // wish to do binary chop search at a smaller granularity than that. A
+ // 128-character space can take up a lot of space in the ranges array if,
+ // for example, we only want to match every second character (eg. the lower
+ // case characters on some Unicode pages).
+ int binary_chop_index = (end_index + start_index) / 2;
+ // The first test ensures that we get to the code that handles the Latin1
+ // range with a single not-taken branch, speeding up this important
+ // character range (even non-Latin1 charset-based text has spaces and
+ // punctuation).
+ if (*border - 1 > String::kMaxOneByteCharCode && // Latin1 case.
+ end_index - start_index > (*new_start_index - start_index) * 2 &&
+ last - first > kSize * 2 && binary_chop_index > *new_start_index &&
+ ranges->at(binary_chop_index) >= first + 2 * kSize) {
+ int scan_forward_for_section_border = binary_chop_index;
+ int new_border = (ranges->at(binary_chop_index) | kMask) + 1;
+
+ while (scan_forward_for_section_border < end_index) {
+ if (ranges->at(scan_forward_for_section_border) > new_border) {
+ *new_start_index = scan_forward_for_section_border;
+ *border = new_border;
+ break;
+ }
+ scan_forward_for_section_border++;
+ }
+ }
+
+ DCHECK(*new_start_index > start_index);
+ *new_end_index = *new_start_index - 1;
+ if (ranges->at(*new_end_index) == *border) {
+ (*new_end_index)--;
+ }
+ if (*border >= ranges->at(end_index)) {
+ *border = ranges->at(end_index);
+ *new_start_index = end_index; // Won't be used.
+ *new_end_index = end_index - 1;
+ }
+}
+
+// Gets a series of segment boundaries representing a character class. If the
+// character is in the range between an even and an odd boundary (counting from
+// start_index) then go to even_label, otherwise go to odd_label. We already
+// know that the character is in the range of min_char to max_char inclusive.
+// Either label can be nullptr 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, 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;
+
+ DCHECK_LT(min_char, first);
+
+ // Just need to test if the character is before or on-or-after
+ // a particular character.
+ if (start_index == end_index) {
+ EmitBoundaryTest(masm, first, fall_through, even_label, odd_label);
+ return;
+ }
+
+ // Another almost trivial case: There is one interval in the middle that is
+ // different from the end intervals.
+ if (start_index + 1 == end_index) {
+ EmitDoubleBoundaryTest(masm, first, last, fall_through, even_label,
+ odd_label);
+ return;
+ }
+
+ // It's not worth using table lookup if there are very few intervals in the
+ // character class.
+ if (end_index - start_index <= 6) {
+ // It is faster to test for individual characters, so we look for those
+ // first, then try arbitrary ranges in the second round.
+ static int kNoCutIndex = -1;
+ int cut = kNoCutIndex;
+ for (int i = start_index; i < end_index; i++) {
+ if (ranges->at(i) == ranges->at(i + 1) - 1) {
+ cut = i;
+ break;
+ }
+ }
+ if (cut == kNoCutIndex) cut = start_index;
+ CutOutRange(masm, ranges, start_index, end_index, cut, even_label,
+ odd_label);
+ DCHECK_GE(end_index - start_index, 2);
+ GenerateBranches(masm, ranges, start_index + 1, end_index - 1, min_char,
+ max_char, fall_through, even_label, odd_label);
+ return;
+ }
+
+ // If there are a lot of intervals in the regexp, then we will use tables to
+ // determine whether the character is inside or outside the character class.
+ static const int kBits = RegExpMacroAssembler::kTableSizeBits;
+
+ if ((max_char >> kBits) == (min_char >> kBits)) {
+ EmitUseLookupTable(masm, ranges, start_index, end_index, min_char,
+ fall_through, even_label, odd_label);
+ return;
+ }
+
+ if ((min_char >> kBits) != (first >> kBits)) {
+ masm->CheckCharacterLT(first, odd_label);
+ GenerateBranches(masm, ranges, start_index + 1, end_index, first, max_char,
+ fall_through, odd_label, even_label);
+ return;
+ }
+
+ int new_start_index = 0;
+ int new_end_index = 0;
+ int border = 0;
+
+ SplitSearchSpace(ranges, start_index, end_index, &new_start_index,
+ &new_end_index, &border);
+
+ Label handle_rest;
+ Label* above = &handle_rest;
+ if (border == last + 1) {
+ // We didn't find any section that started after the limit, so everything
+ // above the border is one of the terminal labels.
+ above = (end_index & 1) != (start_index & 1) ? odd_label : even_label;
+ DCHECK(new_end_index == end_index - 1);
+ }
+
+ DCHECK_LE(start_index, new_end_index);
+ DCHECK_LE(new_start_index, end_index);
+ DCHECK_LT(start_index, new_start_index);
+ DCHECK_LT(new_end_index, end_index);
+ DCHECK(new_end_index + 1 == new_start_index ||
+ (new_end_index + 2 == new_start_index &&
+ border == ranges->at(new_end_index + 1)));
+ DCHECK_LT(min_char, border - 1);
+ DCHECK_LT(border, max_char);
+ DCHECK_LT(ranges->at(new_end_index), border);
+ DCHECK(border < ranges->at(new_start_index) ||
+ (border == ranges->at(new_start_index) &&
+ new_start_index == end_index && new_end_index == end_index - 1 &&
+ border == last + 1));
+ DCHECK(new_start_index == 0 || border >= ranges->at(new_start_index - 1));
+
+ masm->CheckCharacterGT(border - 1, above);
+ Label dummy;
+ GenerateBranches(masm, ranges, start_index, new_end_index, min_char,
+ border - 1, &dummy, even_label, odd_label);
+ if (handle_rest.is_linked()) {
+ masm->Bind(&handle_rest);
+ bool flip = (new_start_index & 1) != (start_index & 1);
+ GenerateBranches(masm, ranges, new_start_index, end_index, border, max_char,
+ &dummy, flip ? odd_label : even_label,
+ flip ? even_label : odd_label);
+ }
+}
+
+static void EmitCharClass(RegExpMacroAssembler* macro_assembler,
+ RegExpCharacterClass* cc, bool one_byte,
+ Label* on_failure, int cp_offset, bool check_offset,
+ bool preloaded, Zone* zone) {
+ ZoneList<CharacterRange>* ranges = cc->ranges(zone);
+ CharacterRange::Canonicalize(ranges);
+
+ int max_char;
+ if (one_byte) {
+ max_char = String::kMaxOneByteCharCode;
+ } else {
+ max_char = String::kMaxUtf16CodeUnit;
+ }
+
+ int range_count = ranges->length();
+
+ int last_valid_range = range_count - 1;
+ while (last_valid_range >= 0) {
+ CharacterRange& range = ranges->at(last_valid_range);
+ if (range.from() <= max_char) {
+ break;
+ }
+ last_valid_range--;
+ }
+
+ if (last_valid_range < 0) {
+ if (!cc->is_negated()) {
+ macro_assembler->GoTo(on_failure);
+ }
+ if (check_offset) {
+ macro_assembler->CheckPosition(cp_offset, on_failure);
+ }
+ return;
+ }
+
+ if (last_valid_range == 0 && ranges->at(0).IsEverything(max_char)) {
+ if (cc->is_negated()) {
+ macro_assembler->GoTo(on_failure);
+ } else {
+ // 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)) {
+ return;
+ }
+
+ // A new list with ascending entries. Each entry is a code unit
+ // where there is a boundary between code units that are part of
+ // the class and code units that are not. Normally we insert an
+ // entry at zero which goes to the failure label, but if there
+ // was already one there we fall through for success on that entry.
+ // Subsequent entries have alternating meaning (success/failure).
+ ZoneList<int>* range_boundaries =
+ new (zone) ZoneList<int>(last_valid_range, zone);
+
+ bool zeroth_entry_is_failure = !cc->is_negated();
+
+ for (int i = 0; i <= last_valid_range; i++) {
+ CharacterRange& range = ranges->at(i);
+ if (range.from() == 0) {
+ DCHECK_EQ(i, 0);
+ zeroth_entry_is_failure = !zeroth_entry_is_failure;
+ } else {
+ range_boundaries->Add(range.from(), zone);
+ }
+ range_boundaries->Add(range.to() + 1, zone);
+ }
+ int end_index = range_boundaries->length() - 1;
+ if (range_boundaries->at(end_index) > max_char) {
+ end_index--;
+ }
+
+ Label fall_through;
+ GenerateBranches(macro_assembler, range_boundaries,
+ 0, // start_index.
+ end_index,
+ 0, // min_char.
+ max_char, &fall_through,
+ zeroth_entry_is_failure ? &fall_through : on_failure,
+ zeroth_entry_is_failure ? on_failure : &fall_through);
+ macro_assembler->Bind(&fall_through);
+}
+
+RegExpNode::~RegExpNode() = default;
+
+RegExpNode::LimitResult RegExpNode::LimitVersions(RegExpCompiler* compiler,
+ Trace* trace) {
+ // If we are generating a greedy loop then don't stop and don't reuse code.
+ if (trace->stop_node() != nullptr) {
+ return CONTINUE;
+ }
+
+ RegExpMacroAssembler* macro_assembler = compiler->macro_assembler();
+ if (trace->is_trivial()) {
+ if (label_.is_bound() || on_work_list() || !KeepRecursing(compiler)) {
+ // If a generic version is already scheduled to be generated or we have
+ // recursed too deeply then just generate a jump to that code.
+ macro_assembler->GoTo(&label_);
+ // This will queue it up for generation of a generic version if it hasn't
+ // already been queued.
+ compiler->AddWork(this);
+ return DONE;
+ }
+ // Generate generic version of the node and bind the label for later use.
+ macro_assembler->Bind(&label_);
+ return CONTINUE;
+ }
+
+ // We are being asked to make a non-generic version. Keep track of how many
+ // non-generic versions we generate so as not to overdo it.
+ trace_count_++;
+ if (KeepRecursing(compiler) && compiler->optimize() &&
+ trace_count_ < kMaxCopiesCodeGenerated) {
+ return CONTINUE;
+ }
+
+ // If we get here code has been generated for this node too many times or
+ // recursion is too deep. Time to switch to a generic version. The code for
+ // generic versions above can handle deep recursion properly.
+ bool was_limiting = compiler->limiting_recursion();
+ compiler->set_limiting_recursion(true);
+ trace->Flush(compiler, this);
+ compiler->set_limiting_recursion(was_limiting);
+ return DONE;
+}
+
+bool RegExpNode::KeepRecursing(RegExpCompiler* compiler) {
+ return !compiler->limiting_recursion() &&
+ compiler->recursion_depth() <= RegExpCompiler::kMaxRecursion;
+}
+
+int ActionNode::EatsAtLeast(int still_to_find, int budget, bool not_at_start) {
+ if (budget <= 0) return 0;
+ if (action_type_ == POSITIVE_SUBMATCH_SUCCESS) return 0; // Rewinds input!
+ return on_success()->EatsAtLeast(still_to_find, budget - 1, not_at_start);
+}
+
+void ActionNode::FillInBMInfo(Isolate* isolate, int offset, int budget,
+ BoyerMooreLookahead* bm, bool not_at_start) {
+ if (action_type_ == POSITIVE_SUBMATCH_SUCCESS) {
+ // Anything may follow a positive submatch success, thus we need to accept
+ // all characters from this position onwards.
+ bm->SetRest(offset);
+ } else {
+ on_success()->FillInBMInfo(isolate, offset, budget - 1, bm, not_at_start);
+ }
+ SaveBMInfo(bm, not_at_start, offset);
+}
+
+int AssertionNode::EatsAtLeast(int still_to_find, int budget,
+ bool not_at_start) {
+ if (budget <= 0) return 0;
+ // If we know we are not at the start and we are asked "how many characters
+ // will you match if you succeed?" then we can answer anything since false
+ // implies false. So lets just return the max answer (still_to_find) since
+ // that won't prevent us from preloading a lot of characters for the other
+ // branches in the node graph.
+ if (assertion_type() == AT_START && not_at_start) return still_to_find;
+ return on_success()->EatsAtLeast(still_to_find, budget - 1, not_at_start);
+}
+
+void AssertionNode::FillInBMInfo(Isolate* isolate, int offset, int budget,
+ BoyerMooreLookahead* bm, bool not_at_start) {
+ // Match the behaviour of EatsAtLeast on this node.
+ if (assertion_type() == AT_START && not_at_start) return;
+ on_success()->FillInBMInfo(isolate, offset, budget - 1, bm, not_at_start);
+ SaveBMInfo(bm, not_at_start, offset);
+}
+
+int BackReferenceNode::EatsAtLeast(int still_to_find, int budget,
+ bool not_at_start) {
+ if (read_backward()) return 0;
+ if (budget <= 0) return 0;
+ return on_success()->EatsAtLeast(still_to_find, budget - 1, not_at_start);
+}
+
+int TextNode::EatsAtLeast(int still_to_find, int budget, bool not_at_start) {
+ if (read_backward()) return 0;
+ int answer = Length();
+ if (answer >= still_to_find) return answer;
+ if (budget <= 0) return answer;
+ // We are not at start after this node so we set the last argument to 'true'.
+ return answer +
+ on_success()->EatsAtLeast(still_to_find - answer, budget - 1, true);
+}
+
+int NegativeLookaroundChoiceNode::EatsAtLeast(int still_to_find, int budget,
+ bool not_at_start) {
+ if (budget <= 0) return 0;
+ // Alternative 0 is the negative lookahead, alternative 1 is what comes
+ // afterwards.
+ RegExpNode* node = alternatives_->at(1).node();
+ return node->EatsAtLeast(still_to_find, budget - 1, not_at_start);
+}
+
+void NegativeLookaroundChoiceNode::GetQuickCheckDetails(
+ QuickCheckDetails* details, RegExpCompiler* compiler, int filled_in,
+ bool not_at_start) {
+ // Alternative 0 is the negative lookahead, alternative 1 is what comes
+ // afterwards.
+ RegExpNode* node = alternatives_->at(1).node();
+ return node->GetQuickCheckDetails(details, compiler, filled_in, not_at_start);
+}
+
+int ChoiceNode::EatsAtLeastHelper(int still_to_find, int budget,
+ RegExpNode* ignore_this_node,
+ bool not_at_start) {
+ if (budget <= 0) return 0;
+ int min = 100;
+ int choice_count = alternatives_->length();
+ budget = (budget - 1) / choice_count;
+ for (int i = 0; i < choice_count; i++) {
+ RegExpNode* node = alternatives_->at(i).node();
+ if (node == ignore_this_node) continue;
+ int node_eats_at_least =
+ node->EatsAtLeast(still_to_find, budget, not_at_start);
+ if (node_eats_at_least < min) min = node_eats_at_least;
+ if (min == 0) return 0;
+ }
+ return min;
+}
+
+int LoopChoiceNode::EatsAtLeast(int still_to_find, int budget,
+ bool not_at_start) {
+ return EatsAtLeastHelper(still_to_find, budget - 1, loop_node_, not_at_start);
+}
+
+int ChoiceNode::EatsAtLeast(int still_to_find, int budget, bool not_at_start) {
+ return EatsAtLeastHelper(still_to_find, budget, nullptr, not_at_start);
+}
+
+// Takes the left-most 1-bit and smears it out, setting all bits to its right.
+static inline uint32_t SmearBitsRight(uint32_t v) {
+ v |= v >> 1;
+ v |= v >> 2;
+ v |= v >> 4;
+ v |= v >> 8;
+ v |= v >> 16;
+ return v;
+}
+
+bool QuickCheckDetails::Rationalize(bool asc) {
+ bool found_useful_op = false;
+ uint32_t char_mask;
+ if (asc) {
+ char_mask = String::kMaxOneByteCharCode;
+ } else {
+ char_mask = String::kMaxUtf16CodeUnit;
+ }
+ mask_ = 0;
+ value_ = 0;
+ int char_shift = 0;
+ for (int i = 0; i < characters_; i++) {
+ Position* pos = &positions_[i];
+ if ((pos->mask & String::kMaxOneByteCharCode) != 0) {
+ found_useful_op = true;
+ }
+ mask_ |= (pos->mask & char_mask) << char_shift;
+ value_ |= (pos->value & char_mask) << char_shift;
+ char_shift += asc ? 8 : 16;
+ }
+ return found_useful_op;
+}
+
+bool RegExpNode::EmitQuickCheck(RegExpCompiler* compiler,
+ Trace* bounds_check_trace, Trace* trace,
+ bool preload_has_checked_bounds,
+ Label* on_possible_success,
+ QuickCheckDetails* details,
+ bool fall_through_on_failure) {
+ if (details->characters() == 0) return false;
+ GetQuickCheckDetails(details, compiler, 0,
+ trace->at_start() == Trace::FALSE_VALUE);
+ if (details->cannot_match()) return false;
+ if (!details->Rationalize(compiler->one_byte())) return false;
+ DCHECK(details->characters() == 1 ||
+ compiler->macro_assembler()->CanReadUnaligned());
+ uint32_t mask = details->mask();
+ uint32_t value = details->value();
+
+ RegExpMacroAssembler* assembler = compiler->macro_assembler();
+
+ if (trace->characters_preloaded() != details->characters()) {
+ DCHECK(trace->cp_offset() == bounds_check_trace->cp_offset());
+ // We are attempting to preload the minimum number of characters
+ // any choice would eat, so if the bounds check fails, then none of the
+ // choices can succeed, so we can just immediately backtrack, rather
+ // than go to the next choice.
+ assembler->LoadCurrentCharacter(
+ trace->cp_offset(), bounds_check_trace->backtrack(),
+ !preload_has_checked_bounds, details->characters());
+ }
+
+ bool need_mask = true;
+
+ if (details->characters() == 1) {
+ // If number of characters preloaded is 1 then we used a byte or 16 bit
+ // load so the value is already masked down.
+ uint32_t char_mask;
+ if (compiler->one_byte()) {
+ char_mask = String::kMaxOneByteCharCode;
+ } else {
+ char_mask = String::kMaxUtf16CodeUnit;
+ }
+ if ((mask & char_mask) == char_mask) need_mask = false;
+ mask &= char_mask;
+ } 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 & kTwoByteMask) == kTwoByteMask) need_mask = false;
+ } else if (details->characters() == 1 && !compiler->one_byte()) {
+ if ((mask & kTwoByteMask) == kTwoByteMask) need_mask = false;
+ } else {
+ if (mask == kFourByteMask) need_mask = false;
+ }
+ }
+
+ if (fall_through_on_failure) {
+ if (need_mask) {
+ assembler->CheckCharacterAfterAnd(value, mask, on_possible_success);
+ } else {
+ assembler->CheckCharacter(value, on_possible_success);
+ }
+ } else {
+ if (need_mask) {
+ assembler->CheckNotCharacterAfterAnd(value, mask, trace->backtrack());
+ } else {
+ assembler->CheckNotCharacter(value, trace->backtrack());
+ }
+ }
+ return true;
+}
+
+// Here is the meat of GetQuickCheckDetails (see also the comment on the
+// super-class in the .h file).
+//
+// We iterate along the text object, building up for each character a
+// mask and value that can be used to test for a quick failure to match.
+// The masks and values for the positions will be combined into a single
+// machine word for the current character width in order to be used in
+// generating a quick check.
+void TextNode::GetQuickCheckDetails(QuickCheckDetails* details,
+ RegExpCompiler* compiler,
+ int characters_filled_in,
+ bool not_at_start) {
+ // Do not collect any quick check details if the text node reads backward,
+ // since it reads in the opposite direction than we use for quick checks.
+ if (read_backward()) return;
+ Isolate* isolate = compiler->macro_assembler()->isolate();
+ DCHECK(characters_filled_in < details->characters());
+ int characters = details->characters();
+ int char_mask;
+ if (compiler->one_byte()) {
+ char_mask = String::kMaxOneByteCharCode;
+ } else {
+ char_mask = String::kMaxUtf16CodeUnit;
+ }
+ for (int k = 0; k < elements()->length(); k++) {
+ TextElement elm = elements()->at(k);
+ if (elm.text_type() == TextElement::ATOM) {
+ Vector<const uc16> quarks = elm.atom()->data();
+ for (int i = 0; i < characters && i < quarks.length(); i++) {
+ QuickCheckDetails::Position* pos =
+ details->positions(characters_filled_in);
+ uc16 c = quarks[i];
+ if (elm.atom()->ignore_case()) {
+ unibrow::uchar chars[4];
+ int length = GetCaseIndependentLetters(
+ isolate, c, compiler->one_byte(), chars, 4);
+ if (length == 0) {
+ // This can happen because all case variants are non-Latin1, but we
+ // know the input is Latin1.
+ details->set_cannot_match();
+ pos->determines_perfectly = false;
+ return;
+ }
+ if (length == 1) {
+ // This letter has no case equivalents, so it's nice and simple
+ // and the mask-compare will determine definitely whether we have
+ // a match at this character position.
+ pos->mask = char_mask;
+ pos->value = c;
+ pos->determines_perfectly = true;
+ } else {
+ uint32_t common_bits = char_mask;
+ uint32_t bits = chars[0];
+ for (int j = 1; j < length; j++) {
+ uint32_t differing_bits = ((chars[j] & common_bits) ^ bits);
+ common_bits ^= differing_bits;
+ bits &= common_bits;
+ }
+ // If length is 2 and common bits has only one zero in it then
+ // our mask and compare instruction will determine definitely
+ // whether we have a match at this character position. Otherwise
+ // it can only be an approximate check.
+ uint32_t one_zero = (common_bits | ~char_mask);
+ if (length == 2 && ((~one_zero) & ((~one_zero) - 1)) == 0) {
+ pos->determines_perfectly = true;
+ }
+ pos->mask = common_bits;
+ pos->value = bits;
+ }
+ } else {
+ // Don't ignore case. Nice simple case where the mask-compare will
+ // determine definitely whether we have a match at this character
+ // position.
+ if (c > char_mask) {
+ details->set_cannot_match();
+ pos->determines_perfectly = false;
+ return;
+ }
+ pos->mask = char_mask;
+ pos->value = c;
+ pos->determines_perfectly = true;
+ }
+ characters_filled_in++;
+ DCHECK(characters_filled_in <= details->characters());
+ if (characters_filled_in == details->characters()) {
+ return;
+ }
+ }
+ } else {
+ QuickCheckDetails::Position* pos =
+ details->positions(characters_filled_in);
+ RegExpCharacterClass* tree = elm.char_class();
+ ZoneList<CharacterRange>* ranges = tree->ranges(zone());
+ DCHECK(!ranges->is_empty());
+ if (tree->is_negated()) {
+ // A quick check uses multi-character mask and compare. There is no
+ // useful way to incorporate a negative char class into this scheme
+ // so we just conservatively create a mask and value that will always
+ // succeed.
+ pos->mask = 0;
+ pos->value = 0;
+ } else {
+ int first_range = 0;
+ while (ranges->at(first_range).from() > char_mask) {
+ first_range++;
+ if (first_range == ranges->length()) {
+ details->set_cannot_match();
+ pos->determines_perfectly = false;
+ return;
+ }
+ }
+ CharacterRange range = ranges->at(first_range);
+ uc16 from = range.from();
+ uc16 to = range.to();
+ if (to > char_mask) {
+ to = char_mask;
+ }
+ uint32_t differing_bits = (from ^ to);
+ // A mask and compare is only perfect if the differing bits form a
+ // number like 00011111 with one single block of trailing 1s.
+ if ((differing_bits & (differing_bits + 1)) == 0 &&
+ from + differing_bits == to) {
+ pos->determines_perfectly = true;
+ }
+ uint32_t common_bits = ~SmearBitsRight(differing_bits);
+ uint32_t bits = (from & common_bits);
+ for (int i = first_range + 1; i < ranges->length(); i++) {
+ CharacterRange range = ranges->at(i);
+ uc16 from = range.from();
+ uc16 to = range.to();
+ if (from > char_mask) continue;
+ if (to > char_mask) to = char_mask;
+ // Here we are combining more ranges into the mask and compare
+ // value. With each new range the mask becomes more sparse and
+ // so the chances of a false positive rise. A character class
+ // with multiple ranges is assumed never to be equivalent to a
+ // mask and compare operation.
+ pos->determines_perfectly = false;
+ uint32_t new_common_bits = (from ^ to);
+ new_common_bits = ~SmearBitsRight(new_common_bits);
+ common_bits &= new_common_bits;
+ bits &= new_common_bits;
+ uint32_t differing_bits = (from & common_bits) ^ bits;
+ common_bits ^= differing_bits;
+ bits &= common_bits;
+ }
+ pos->mask = common_bits;
+ pos->value = bits;
+ }
+ characters_filled_in++;
+ DCHECK(characters_filled_in <= details->characters());
+ if (characters_filled_in == details->characters()) {
+ return;
+ }
+ }
+ }
+ DCHECK(characters_filled_in != details->characters());
+ if (!details->cannot_match()) {
+ on_success()->GetQuickCheckDetails(details, compiler, characters_filled_in,
+ true);
+ }
+}
+
+void QuickCheckDetails::Clear() {
+ for (int i = 0; i < characters_; i++) {
+ positions_[i].mask = 0;
+ positions_[i].value = 0;
+ positions_[i].determines_perfectly = false;
+ }
+ characters_ = 0;
+}
+
+void QuickCheckDetails::Advance(int by, bool one_byte) {
+ if (by >= characters_ || by < 0) {
+ DCHECK_IMPLIES(by < 0, characters_ == 0);
+ Clear();
+ return;
+ }
+ DCHECK_LE(characters_ - by, 4);
+ DCHECK_LE(characters_, 4);
+ for (int i = 0; i < characters_ - by; i++) {
+ positions_[i] = positions_[by + i];
+ }
+ for (int i = characters_ - by; i < characters_; i++) {
+ positions_[i].mask = 0;
+ positions_[i].value = 0;
+ positions_[i].determines_perfectly = false;
+ }
+ characters_ -= by;
+ // We could change mask_ and value_ here but we would never advance unless
+ // they had already been used in a check and they won't be used again because
+ // it would gain us nothing. So there's no point.
+}
+
+void QuickCheckDetails::Merge(QuickCheckDetails* other, int from_index) {
+ DCHECK(characters_ == other->characters_);
+ if (other->cannot_match_) {
+ return;
+ }
+ if (cannot_match_) {
+ *this = *other;
+ return;
+ }
+ for (int i = from_index; i < characters_; i++) {
+ QuickCheckDetails::Position* pos = positions(i);
+ QuickCheckDetails::Position* other_pos = other->positions(i);
+ if (pos->mask != other_pos->mask || pos->value != other_pos->value ||
+ !other_pos->determines_perfectly) {
+ // Our mask-compare operation will be approximate unless we have the
+ // exact same operation on both sides of the alternation.
+ pos->determines_perfectly = false;
+ }
+ pos->mask &= other_pos->mask;
+ pos->value &= pos->mask;
+ other_pos->value &= pos->mask;
+ uc16 differing_bits = (pos->value ^ other_pos->value);
+ pos->mask &= ~differing_bits;
+ pos->value &= pos->mask;
+ }
+}
+
+class VisitMarker {
+ public:
+ explicit VisitMarker(NodeInfo* info) : info_(info) {
+ DCHECK(!info->visited);
+ info->visited = true;
+ }
+ ~VisitMarker() { info_->visited = false; }
+
+ private:
+ NodeInfo* info_;
+};
+
+RegExpNode* SeqRegExpNode::FilterOneByte(int depth) {
+ if (info()->replacement_calculated) return replacement();
+ if (depth < 0) return this;
+ DCHECK(!info()->visited);
+ VisitMarker marker(info());
+ return FilterSuccessor(depth - 1);
+}
+
+RegExpNode* SeqRegExpNode::FilterSuccessor(int depth) {
+ RegExpNode* next = on_success_->FilterOneByte(depth - 1);
+ if (next == nullptr) return set_replacement(nullptr);
+ on_success_ = next;
+ return set_replacement(this);
+}
+
+// We need to check for the following characters: 0x39C 0x3BC 0x178.
+bool RangeContainsLatin1Equivalents(CharacterRange range) {
+ // TODO(dcarney): this could be a lot more efficient.
+ return range.Contains(0x039C) || range.Contains(0x03BC) ||
+ range.Contains(0x0178);
+}
+
+static bool RangesContainLatin1Equivalents(ZoneList<CharacterRange>* ranges) {
+ for (int i = 0; i < ranges->length(); i++) {
+ // TODO(dcarney): this could be a lot more efficient.
+ if (RangeContainsLatin1Equivalents(ranges->at(i))) return true;
+ }
+ return false;
+}
+
+RegExpNode* TextNode::FilterOneByte(int depth) {
+ if (info()->replacement_calculated) return replacement();
+ if (depth < 0) return this;
+ DCHECK(!info()->visited);
+ VisitMarker marker(info());
+ int element_count = elements()->length();
+ for (int i = 0; i < element_count; i++) {
+ TextElement elm = elements()->at(i);
+ if (elm.text_type() == TextElement::ATOM) {
+ Vector<const uc16> quarks = elm.atom()->data();
+ for (int j = 0; j < quarks.length(); j++) {
+ uint16_t c = quarks[j];
+ if (elm.atom()->ignore_case()) {
+ c = unibrow::Latin1::TryConvertToLatin1(c);
+ }
+ if (c > unibrow::Latin1::kMaxChar) return set_replacement(nullptr);
+ // Replace quark in case we converted to Latin-1.
+ uint16_t* writable_quarks = const_cast<uint16_t*>(quarks.begin());
+ writable_quarks[j] = c;
+ }
+ } else {
+ DCHECK(elm.text_type() == TextElement::CHAR_CLASS);
+ RegExpCharacterClass* cc = elm.char_class();
+ ZoneList<CharacterRange>* ranges = cc->ranges(zone());
+ 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()) {
+ if (range_count != 0 && ranges->at(0).from() == 0 &&
+ ranges->at(0).to() >= String::kMaxOneByteCharCode) {
+ // This will be handled in a later filter.
+ if (IgnoreCase(cc->flags()) && RangesContainLatin1Equivalents(ranges))
+ continue;
+ return set_replacement(nullptr);
+ }
+ } else {
+ if (range_count == 0 ||
+ ranges->at(0).from() > String::kMaxOneByteCharCode) {
+ // This will be handled in a later filter.
+ if (IgnoreCase(cc->flags()) && RangesContainLatin1Equivalents(ranges))
+ continue;
+ return set_replacement(nullptr);
+ }
+ }
+ }
+ }
+ return FilterSuccessor(depth - 1);
+}
+
+RegExpNode* LoopChoiceNode::FilterOneByte(int depth) {
+ if (info()->replacement_calculated) return replacement();
+ if (depth < 0) return this;
+ if (info()->visited) return this;
+ {
+ VisitMarker marker(info());
+
+ RegExpNode* continue_replacement = continue_node_->FilterOneByte(depth - 1);
+ // If we can't continue after the loop then there is no sense in doing the
+ // loop.
+ if (continue_replacement == nullptr) return set_replacement(nullptr);
+ }
+
+ return ChoiceNode::FilterOneByte(depth - 1);
+}
+
+RegExpNode* ChoiceNode::FilterOneByte(int depth) {
+ if (info()->replacement_calculated) return replacement();
+ if (depth < 0) return this;
+ if (info()->visited) return this;
+ VisitMarker marker(info());
+ int choice_count = alternatives_->length();
+
+ for (int i = 0; i < choice_count; i++) {
+ GuardedAlternative alternative = alternatives_->at(i);
+ if (alternative.guards() != nullptr &&
+ alternative.guards()->length() != 0) {
+ set_replacement(this);
+ return this;
+ }
+ }
+
+ int surviving = 0;
+ RegExpNode* survivor = nullptr;
+ for (int i = 0; i < choice_count; i++) {
+ GuardedAlternative alternative = alternatives_->at(i);
+ RegExpNode* replacement = alternative.node()->FilterOneByte(depth - 1);
+ DCHECK(replacement != this); // No missing EMPTY_MATCH_CHECK.
+ if (replacement != nullptr) {
+ alternatives_->at(i).set_node(replacement);
+ surviving++;
+ survivor = replacement;
+ }
+ }
+ if (surviving < 2) return set_replacement(survivor);
+
+ set_replacement(this);
+ if (surviving == choice_count) {
+ return this;
+ }
+ // Only some of the nodes survived the filtering. We need to rebuild the
+ // alternatives list.
+ ZoneList<GuardedAlternative>* new_alternatives =
+ new (zone()) ZoneList<GuardedAlternative>(surviving, zone());
+ for (int i = 0; i < choice_count; i++) {
+ RegExpNode* replacement =
+ alternatives_->at(i).node()->FilterOneByte(depth - 1);
+ if (replacement != nullptr) {
+ alternatives_->at(i).set_node(replacement);
+ new_alternatives->Add(alternatives_->at(i), zone());
+ }
+ }
+ alternatives_ = new_alternatives;
+ return this;
+}
+
+RegExpNode* NegativeLookaroundChoiceNode::FilterOneByte(int depth) {
+ if (info()->replacement_calculated) return replacement();
+ if (depth < 0) return this;
+ if (info()->visited) return this;
+ VisitMarker marker(info());
+ // Alternative 0 is the negative lookahead, alternative 1 is what comes
+ // afterwards.
+ RegExpNode* node = alternatives_->at(1).node();
+ RegExpNode* replacement = node->FilterOneByte(depth - 1);
+ if (replacement == nullptr) return set_replacement(nullptr);
+ alternatives_->at(1).set_node(replacement);
+
+ RegExpNode* neg_node = alternatives_->at(0).node();
+ RegExpNode* neg_replacement = neg_node->FilterOneByte(depth - 1);
+ // If the negative lookahead is always going to fail then
+ // we don't need to check it.
+ if (neg_replacement == nullptr) return set_replacement(replacement);
+ alternatives_->at(0).set_node(neg_replacement);
+ return set_replacement(this);
+}
+
+void LoopChoiceNode::GetQuickCheckDetails(QuickCheckDetails* details,
+ RegExpCompiler* compiler,
+ int characters_filled_in,
+ bool not_at_start) {
+ if (body_can_be_zero_length_ || info()->visited) return;
+ VisitMarker marker(info());
+ return ChoiceNode::GetQuickCheckDetails(details, compiler,
+ characters_filled_in, not_at_start);
+}
+
+void LoopChoiceNode::FillInBMInfo(Isolate* isolate, int offset, int budget,
+ BoyerMooreLookahead* bm, bool not_at_start) {
+ if (body_can_be_zero_length_ || budget <= 0) {
+ bm->SetRest(offset);
+ SaveBMInfo(bm, not_at_start, offset);
+ return;
+ }
+ ChoiceNode::FillInBMInfo(isolate, offset, budget - 1, bm, not_at_start);
+ SaveBMInfo(bm, not_at_start, offset);
+}
+
+void ChoiceNode::GetQuickCheckDetails(QuickCheckDetails* details,
+ RegExpCompiler* compiler,
+ int characters_filled_in,
+ bool not_at_start) {
+ not_at_start = (not_at_start || not_at_start_);
+ int choice_count = alternatives_->length();
+ DCHECK_LT(0, choice_count);
+ alternatives_->at(0).node()->GetQuickCheckDetails(
+ details, compiler, characters_filled_in, not_at_start);
+ for (int i = 1; i < choice_count; i++) {
+ QuickCheckDetails new_details(details->characters());
+ RegExpNode* node = alternatives_->at(i).node();
+ node->GetQuickCheckDetails(&new_details, compiler, characters_filled_in,
+ not_at_start);
+ // Here we merge the quick match details of the two branches.
+ details->Merge(&new_details, characters_filled_in);
+ }
+}
+
+// Check for [0-9A-Z_a-z].
+static void EmitWordCheck(RegExpMacroAssembler* assembler, Label* word,
+ Label* non_word, bool fall_through_on_word) {
+ if (assembler->CheckSpecialCharacterClass(
+ fall_through_on_word ? 'w' : 'W',
+ fall_through_on_word ? non_word : word)) {
+ // Optimized implementation available.
+ return;
+ }
+ assembler->CheckCharacterGT('z', non_word);
+ assembler->CheckCharacterLT('0', non_word);
+ assembler->CheckCharacterGT('a' - 1, word);
+ assembler->CheckCharacterLT('9' + 1, word);
+ assembler->CheckCharacterLT('A', non_word);
+ assembler->CheckCharacterLT('Z' + 1, word);
+ if (fall_through_on_word) {
+ assembler->CheckNotCharacter('_', non_word);
+ } else {
+ assembler->CheckCharacter('_', word);
+ }
+}
+
+// Emit the code to check for a ^ in multiline mode (1-character lookbehind
+// that matches newline or the start of input).
+static void EmitHat(RegExpCompiler* compiler, RegExpNode* on_success,
+ Trace* trace) {
+ RegExpMacroAssembler* assembler = compiler->macro_assembler();
+ // We will be loading the previous character into the current character
+ // register.
+ Trace new_trace(*trace);
+ new_trace.InvalidateCurrentCharacter();
+
+ Label ok;
+ if (new_trace.cp_offset() == 0) {
+ // The start of input counts as a newline in this context, so skip to
+ // ok if we are at the start.
+ assembler->CheckAtStart(&ok);
+ }
+ // We already checked that we are not at the start of input so it must be
+ // OK to load the previous character.
+ assembler->LoadCurrentCharacter(new_trace.cp_offset() - 1,
+ new_trace.backtrack(), false);
+ if (!assembler->CheckSpecialCharacterClass('n', new_trace.backtrack())) {
+ // Newline means \n, \r, 0x2028 or 0x2029.
+ if (!compiler->one_byte()) {
+ assembler->CheckCharacterAfterAnd(0x2028, 0xFFFE, &ok);
+ }
+ assembler->CheckCharacter('\n', &ok);
+ assembler->CheckNotCharacter('\r', new_trace.backtrack());
+ }
+ assembler->Bind(&ok);
+ on_success->Emit(compiler, &new_trace);
+}
+
+// Emit the code to handle \b and \B (word-boundary or non-word-boundary).
+void AssertionNode::EmitBoundaryCheck(RegExpCompiler* compiler, Trace* trace) {
+ RegExpMacroAssembler* assembler = compiler->macro_assembler();
+ Isolate* isolate = assembler->isolate();
+ Trace::TriBool next_is_word_character = Trace::UNKNOWN;
+ bool not_at_start = (trace->at_start() == Trace::FALSE_VALUE);
+ BoyerMooreLookahead* lookahead = bm_info(not_at_start);
+ if (lookahead == nullptr) {
+ int eats_at_least = Min(kMaxLookaheadForBoyerMoore,
+ EatsAtLeast(kMaxLookaheadForBoyerMoore,
+ kRecursionBudget, not_at_start));
+ if (eats_at_least >= 1) {
+ BoyerMooreLookahead* bm =
+ new (zone()) BoyerMooreLookahead(eats_at_least, compiler, zone());
+ FillInBMInfo(isolate, 0, kRecursionBudget, bm, not_at_start);
+ if (bm->at(0)->is_non_word()) next_is_word_character = Trace::FALSE_VALUE;
+ if (bm->at(0)->is_word()) next_is_word_character = Trace::TRUE_VALUE;
+ }
+ } else {
+ if (lookahead->at(0)->is_non_word())
+ next_is_word_character = Trace::FALSE_VALUE;
+ if (lookahead->at(0)->is_word()) next_is_word_character = Trace::TRUE_VALUE;
+ }
+ bool at_boundary = (assertion_type_ == AssertionNode::AT_BOUNDARY);
+ if (next_is_word_character == Trace::UNKNOWN) {
+ Label before_non_word;
+ Label before_word;
+ if (trace->characters_preloaded() != 1) {
+ assembler->LoadCurrentCharacter(trace->cp_offset(), &before_non_word);
+ }
+ // Fall through on non-word.
+ EmitWordCheck(assembler, &before_word, &before_non_word, false);
+ // Next character is not a word character.
+ assembler->Bind(&before_non_word);
+ Label ok;
+ BacktrackIfPrevious(compiler, trace, at_boundary ? kIsNonWord : kIsWord);
+ assembler->GoTo(&ok);
+
+ assembler->Bind(&before_word);
+ BacktrackIfPrevious(compiler, trace, at_boundary ? kIsWord : kIsNonWord);
+ assembler->Bind(&ok);
+ } else if (next_is_word_character == Trace::TRUE_VALUE) {
+ BacktrackIfPrevious(compiler, trace, at_boundary ? kIsWord : kIsNonWord);
+ } else {
+ DCHECK(next_is_word_character == Trace::FALSE_VALUE);
+ BacktrackIfPrevious(compiler, trace, at_boundary ? kIsNonWord : kIsWord);
+ }
+}
+
+void AssertionNode::BacktrackIfPrevious(
+ RegExpCompiler* compiler, Trace* trace,
+ AssertionNode::IfPrevious backtrack_if_previous) {
+ RegExpMacroAssembler* assembler = compiler->macro_assembler();
+ Trace new_trace(*trace);
+ new_trace.InvalidateCurrentCharacter();
+
+ Label fall_through, dummy;
+
+ Label* non_word = backtrack_if_previous == kIsNonWord ? new_trace.backtrack()
+ : &fall_through;
+ Label* word = backtrack_if_previous == kIsNonWord ? &fall_through
+ : new_trace.backtrack();
+
+ if (new_trace.cp_offset() == 0) {
+ // The start of input counts as a non-word character, so the question is
+ // decided if we are at the start.
+ assembler->CheckAtStart(non_word);
+ }
+ // We already checked that we are not at the start of input so it must be
+ // OK to load the previous character.
+ assembler->LoadCurrentCharacter(new_trace.cp_offset() - 1, &dummy, false);
+ EmitWordCheck(assembler, word, non_word, backtrack_if_previous == kIsNonWord);
+
+ assembler->Bind(&fall_through);
+ on_success()->Emit(compiler, &new_trace);
+}
+
+void AssertionNode::GetQuickCheckDetails(QuickCheckDetails* details,
+ RegExpCompiler* compiler,
+ int filled_in, bool not_at_start) {
+ if (assertion_type_ == AT_START && not_at_start) {
+ details->set_cannot_match();
+ return;
+ }
+ return on_success()->GetQuickCheckDetails(details, compiler, filled_in,
+ not_at_start);
+}
+
+void AssertionNode::Emit(RegExpCompiler* compiler, Trace* trace) {
+ RegExpMacroAssembler* assembler = compiler->macro_assembler();
+ switch (assertion_type_) {
+ case AT_END: {
+ Label ok;
+ assembler->CheckPosition(trace->cp_offset(), &ok);
+ assembler->GoTo(trace->backtrack());
+ assembler->Bind(&ok);
+ break;
+ }
+ case AT_START: {
+ if (trace->at_start() == Trace::FALSE_VALUE) {
+ assembler->GoTo(trace->backtrack());
+ return;
+ }
+ if (trace->at_start() == Trace::UNKNOWN) {
+ assembler->CheckNotAtStart(trace->cp_offset(), trace->backtrack());
+ Trace at_start_trace = *trace;
+ at_start_trace.set_at_start(Trace::TRUE_VALUE);
+ on_success()->Emit(compiler, &at_start_trace);
+ return;
+ }
+ } break;
+ case AFTER_NEWLINE:
+ EmitHat(compiler, on_success(), trace);
+ return;
+ case AT_BOUNDARY:
+ case AT_NON_BOUNDARY: {
+ EmitBoundaryCheck(compiler, trace);
+ return;
+ }
+ }
+ on_success()->Emit(compiler, trace);
+}
+
+static bool DeterminedAlready(QuickCheckDetails* quick_check, int offset) {
+ if (quick_check == nullptr) return false;
+ if (offset >= quick_check->characters()) return false;
+ return quick_check->positions(offset)->determines_perfectly;
+}
+
+static void UpdateBoundsCheck(int index, int* checked_up_to) {
+ if (index > *checked_up_to) {
+ *checked_up_to = index;
+ }
+}
+
+// We call this repeatedly to generate code for each pass over the text node.
+// The passes are in increasing order of difficulty because we hope one
+// of the first passes will fail in which case we are saved the work of the
+// later passes. for example for the case independent regexp /%[asdfghjkl]a/
+// we will check the '%' in the first pass, the case independent 'a' in the
+// second pass and the character class in the last pass.
+//
+// The passes are done from right to left, so for example to test for /bar/
+// we will first test for an 'r' with offset 2, then an 'a' with offset 1
+// and then a 'b' with offset 0. This means we can avoid the end-of-input
+// bounds check most of the time. In the example we only need to check for
+// end-of-input when loading the putative 'r'.
+//
+// A slight complication involves the fact that the first character may already
+// be fetched into a register by the previous node. In this case we want to
+// do the test for that character first. We do this in separate passes. The
+// 'preloaded' argument indicates that we are doing such a 'pass'. If such a
+// pass has been performed then subsequent passes will have true in
+// first_element_checked to indicate that that character does not need to be
+// checked again.
+//
+// In addition to all this we are passed a Trace, which can
+// contain an AlternativeGeneration object. In this AlternativeGeneration
+// object we can see details of any quick check that was already passed in
+// order to get to the code we are now generating. The quick check can involve
+// loading characters, which means we do not need to recheck the bounds
+// up to the limit the quick check already checked. In addition the quick
+// check can have involved a mask and compare operation which may simplify
+// or obviate the need for further checks at some character positions.
+void TextNode::TextEmitPass(RegExpCompiler* compiler, TextEmitPassType pass,
+ bool preloaded, Trace* trace,
+ bool first_element_checked, int* checked_up_to) {
+ RegExpMacroAssembler* assembler = compiler->macro_assembler();
+ Isolate* isolate = assembler->isolate();
+ bool one_byte = compiler->one_byte();
+ Label* backtrack = trace->backtrack();
+ QuickCheckDetails* quick_check = trace->quick_check_performed();
+ int element_count = elements()->length();
+ int backward_offset = read_backward() ? -Length() : 0;
+ for (int i = preloaded ? 0 : element_count - 1; i >= 0; i--) {
+ TextElement elm = elements()->at(i);
+ int cp_offset = trace->cp_offset() + elm.cp_offset() + backward_offset;
+ if (elm.text_type() == TextElement::ATOM) {
+ if (SkipPass(pass, elm.atom()->ignore_case())) continue;
+ Vector<const uc16> quarks = elm.atom()->data();
+ for (int j = preloaded ? 0 : quarks.length() - 1; j >= 0; j--) {
+ if (first_element_checked && i == 0 && j == 0) continue;
+ if (DeterminedAlready(quick_check, elm.cp_offset() + j)) continue;
+ EmitCharacterFunction* emit_function = nullptr;
+ uc16 quark = quarks[j];
+ if (elm.atom()->ignore_case()) {
+ // Everywhere else we assume that a non-Latin-1 character cannot match
+ // a Latin-1 character. Avoid the cases where this is assumption is
+ // invalid by using the Latin1 equivalent instead.
+ quark = unibrow::Latin1::TryConvertToLatin1(quark);
+ }
+ switch (pass) {
+ case NON_LATIN1_MATCH:
+ DCHECK(one_byte);
+ if (quark > String::kMaxOneByteCharCode) {
+ assembler->GoTo(backtrack);
+ return;
+ }
+ break;
+ case NON_LETTER_CHARACTER_MATCH:
+ emit_function = &EmitAtomNonLetter;
+ break;
+ case SIMPLE_CHARACTER_MATCH:
+ emit_function = &EmitSimpleCharacter;
+ break;
+ case CASE_CHARACTER_MATCH:
+ emit_function = &EmitAtomLetter;
+ break;
+ default:
+ break;
+ }
+ if (emit_function != nullptr) {
+ bool bounds_check = *checked_up_to < cp_offset + j || read_backward();
+ bool bound_checked =
+ emit_function(isolate, compiler, quark, backtrack, cp_offset + j,
+ bounds_check, preloaded);
+ if (bound_checked) UpdateBoundsCheck(cp_offset + j, checked_up_to);
+ }
+ }
+ } else {
+ DCHECK_EQ(TextElement::CHAR_CLASS, elm.text_type());
+ if (pass == CHARACTER_CLASS_MATCH) {
+ if (first_element_checked && i == 0) continue;
+ if (DeterminedAlready(quick_check, elm.cp_offset())) continue;
+ RegExpCharacterClass* cc = elm.char_class();
+ bool bounds_check = *checked_up_to < cp_offset || read_backward();
+ EmitCharClass(assembler, cc, one_byte, backtrack, cp_offset,
+ bounds_check, preloaded, zone());
+ UpdateBoundsCheck(cp_offset, checked_up_to);
+ }
+ }
+ }
+}
+
+int TextNode::Length() {
+ TextElement elm = elements()->last();
+ DCHECK_LE(0, elm.cp_offset());
+ return elm.cp_offset() + elm.length();
+}
+
+bool TextNode::SkipPass(TextEmitPassType pass, bool ignore_case) {
+ if (ignore_case) {
+ return pass == SIMPLE_CHARACTER_MATCH;
+ } else {
+ return pass == NON_LETTER_CHARACTER_MATCH || pass == CASE_CHARACTER_MATCH;
+ }
+}
+
+TextNode* TextNode::CreateForCharacterRanges(Zone* zone,
+ ZoneList<CharacterRange>* ranges,
+ bool read_backward,
+ RegExpNode* on_success,
+ JSRegExp::Flags flags) {
+ DCHECK_NOT_NULL(ranges);
+ ZoneList<TextElement>* elms = new (zone) ZoneList<TextElement>(1, zone);
+ elms->Add(TextElement::CharClass(
+ new (zone) RegExpCharacterClass(zone, ranges, flags)),
+ 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,
+ JSRegExp::Flags flags) {
+ 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(zone, lead_ranges, flags)),
+ zone);
+ elms->Add(TextElement::CharClass(
+ new (zone) RegExpCharacterClass(zone, trail_ranges, flags)),
+ 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
+// pass from left to right. Instead we pass over the text node several times,
+// emitting code for some character positions every time. See the comment on
+// TextEmitPass for details.
+void TextNode::Emit(RegExpCompiler* compiler, Trace* trace) {
+ LimitResult limit_result = LimitVersions(compiler, trace);
+ if (limit_result == DONE) return;
+ DCHECK(limit_result == CONTINUE);
+
+ if (trace->cp_offset() + Length() > RegExpMacroAssembler::kMaxCPOffset) {
+ compiler->SetRegExpTooBig();
+ return;
+ }
+
+ if (compiler->one_byte()) {
+ int dummy = 0;
+ TextEmitPass(compiler, NON_LATIN1_MATCH, false, trace, false, &dummy);
+ }
+
+ bool first_elt_done = false;
+ int bound_checked_to = trace->cp_offset() - 1;
+ bound_checked_to += trace->bound_checked_up_to();
+
+ // If a character is preloaded into the current character register then
+ // check that now.
+ if (trace->characters_preloaded() == 1) {
+ for (int pass = kFirstRealPass; pass <= kLastPass; pass++) {
+ TextEmitPass(compiler, static_cast<TextEmitPassType>(pass), true, trace,
+ false, &bound_checked_to);
+ }
+ first_elt_done = true;
+ }
+
+ for (int pass = kFirstRealPass; pass <= kLastPass; pass++) {
+ TextEmitPass(compiler, static_cast<TextEmitPassType>(pass), false, trace,
+ first_elt_done, &bound_checked_to);
+ }
+
+ Trace successor_trace(*trace);
+ // If we advance backward, we may end up at the start.
+ successor_trace.AdvanceCurrentPositionInTrace(
+ read_backward() ? -Length() : Length(), compiler);
+ successor_trace.set_at_start(read_backward() ? Trace::UNKNOWN
+ : Trace::FALSE_VALUE);
+ RecursionCheck rc(compiler);
+ on_success()->Emit(compiler, &successor_trace);
+}
+
+void Trace::InvalidateCurrentCharacter() { characters_preloaded_ = 0; }
+
+void Trace::AdvanceCurrentPositionInTrace(int by, RegExpCompiler* compiler) {
+ // We don't have an instruction for shifting the current character register
+ // down or for using a shifted value for anything so lets just forget that
+ // we preloaded any characters into it.
+ characters_preloaded_ = 0;
+ // Adjust the offsets of the quick check performed information. This
+ // information is used to find out what we already determined about the
+ // characters by means of mask and compare.
+ quick_check_performed_.Advance(by, compiler->one_byte());
+ cp_offset_ += by;
+ if (cp_offset_ > RegExpMacroAssembler::kMaxCPOffset) {
+ compiler->SetRegExpTooBig();
+ cp_offset_ = 0;
+ }
+ bound_checked_up_to_ = Max(0, bound_checked_up_to_ - by);
+}
+
+void TextNode::MakeCaseIndependent(Isolate* isolate, bool is_one_byte) {
+ int element_count = elements()->length();
+ for (int i = 0; i < element_count; i++) {
+ TextElement elm = elements()->at(i);
+ if (elm.text_type() == TextElement::CHAR_CLASS) {
+ RegExpCharacterClass* cc = elm.char_class();
+#ifdef V8_INTL_SUPPORT
+ bool case_equivalents_already_added =
+ NeedsUnicodeCaseEquivalents(cc->flags());
+#else
+ bool case_equivalents_already_added = false;
+#endif
+ if (IgnoreCase(cc->flags()) && !case_equivalents_already_added) {
+ // None of the standard character classes is different in the case
+ // 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());
+ CharacterRange::AddCaseEquivalents(isolate, zone(), ranges,
+ is_one_byte);
+ }
+ }
+ }
+}
+
+int TextNode::GreedyLoopTextLength() { return Length(); }
+
+RegExpNode* TextNode::GetSuccessorOfOmnivorousTextNode(
+ RegExpCompiler* compiler) {
+ if (read_backward()) return nullptr;
+ if (elements()->length() != 1) return nullptr;
+ TextElement elm = elements()->at(0);
+ if (elm.text_type() != TextElement::CHAR_CLASS) return nullptr;
+ RegExpCharacterClass* node = elm.char_class();
+ ZoneList<CharacterRange>* ranges = node->ranges(zone());
+ CharacterRange::Canonicalize(ranges);
+ if (node->is_negated()) {
+ return ranges->length() == 0 ? on_success() : nullptr;
+ }
+ if (ranges->length() != 1) return nullptr;
+ uint32_t max_char;
+ if (compiler->one_byte()) {
+ max_char = String::kMaxOneByteCharCode;
+ } else {
+ max_char = String::kMaxUtf16CodeUnit;
+ }
+ return ranges->at(0).IsEverything(max_char) ? on_success() : nullptr;
+}
+
+// Finds the fixed match length of a sequence of nodes that goes from
+// this alternative and back to this choice node. If there are variable
+// length nodes or other complications in the way then return a sentinel
+// value indicating that a greedy loop cannot be constructed.
+int ChoiceNode::GreedyLoopTextLengthForAlternative(
+ GuardedAlternative* alternative) {
+ int length = 0;
+ RegExpNode* node = alternative->node();
+ // Later we will generate code for all these text nodes using recursion
+ // so we have to limit the max number.
+ int recursion_depth = 0;
+ while (node != this) {
+ if (recursion_depth++ > RegExpCompiler::kMaxRecursion) {
+ return kNodeIsTooComplexForGreedyLoops;
+ }
+ int node_length = node->GreedyLoopTextLength();
+ if (node_length == kNodeIsTooComplexForGreedyLoops) {
+ return kNodeIsTooComplexForGreedyLoops;
+ }
+ length += node_length;
+ SeqRegExpNode* seq_node = static_cast<SeqRegExpNode*>(node);
+ node = seq_node->on_success();
+ }
+ return read_backward() ? -length : length;
+}
+
+void LoopChoiceNode::AddLoopAlternative(GuardedAlternative alt) {
+ DCHECK_NULL(loop_node_);
+ AddAlternative(alt);
+ loop_node_ = alt.node();
+}
+
+void LoopChoiceNode::AddContinueAlternative(GuardedAlternative alt) {
+ DCHECK_NULL(continue_node_);
+ AddAlternative(alt);
+ continue_node_ = alt.node();
+}
+
+void LoopChoiceNode::Emit(RegExpCompiler* compiler, Trace* trace) {
+ RegExpMacroAssembler* macro_assembler = compiler->macro_assembler();
+ if (trace->stop_node() == this) {
+ // Back edge of greedy optimized loop node graph.
+ int text_length =
+ GreedyLoopTextLengthForAlternative(&(alternatives_->at(0)));
+ DCHECK_NE(kNodeIsTooComplexForGreedyLoops, text_length);
+ // Update the counter-based backtracking info on the stack. This is an
+ // optimization for greedy loops (see below).
+ DCHECK(trace->cp_offset() == text_length);
+ macro_assembler->AdvanceCurrentPosition(text_length);
+ macro_assembler->GoTo(trace->loop_label());
+ return;
+ }
+ DCHECK_NULL(trace->stop_node());
+ if (!trace->is_trivial()) {
+ trace->Flush(compiler, this);
+ return;
+ }
+ ChoiceNode::Emit(compiler, trace);
+}
+
+int ChoiceNode::CalculatePreloadCharacters(RegExpCompiler* compiler,
+ int eats_at_least) {
+ int preload_characters = Min(4, eats_at_least);
+ DCHECK_LE(preload_characters, 4);
+ if (compiler->macro_assembler()->CanReadUnaligned()) {
+ bool one_byte = compiler->one_byte();
+ if (one_byte) {
+ // We can't preload 3 characters because there is no machine instruction
+ // to do that. We can't just load 4 because we could be reading
+ // beyond the end of the string, which could cause a memory fault.
+ if (preload_characters == 3) preload_characters = 2;
+ } else {
+ if (preload_characters > 2) preload_characters = 2;
+ }
+ } else {
+ if (preload_characters > 1) preload_characters = 1;
+ }
+ return preload_characters;
+}
+
+// This class is used when generating the alternatives in a choice node. It
+// records the way the alternative is being code generated.
+class AlternativeGeneration : public Malloced {
+ public:
+ AlternativeGeneration()
+ : possible_success(),
+ expects_preload(false),
+ after(),
+ quick_check_details() {}
+ Label possible_success;
+ bool expects_preload;
+ Label after;
+ QuickCheckDetails quick_check_details;
+};
+
+// Creates a list of AlternativeGenerations. If the list has a reasonable
+// size then it is on the stack, otherwise the excess is on the heap.
+class AlternativeGenerationList {
+ public:
+ AlternativeGenerationList(int count, Zone* zone) : alt_gens_(count, zone) {
+ for (int i = 0; i < count && i < kAFew; i++) {
+ alt_gens_.Add(a_few_alt_gens_ + i, zone);
+ }
+ for (int i = kAFew; i < count; i++) {
+ alt_gens_.Add(new AlternativeGeneration(), zone);
+ }
+ }
+ ~AlternativeGenerationList() {
+ for (int i = kAFew; i < alt_gens_.length(); i++) {
+ delete alt_gens_[i];
+ alt_gens_[i] = nullptr;
+ }
+ }
+
+ AlternativeGeneration* at(int i) { return alt_gens_[i]; }
+
+ private:
+ static const int kAFew = 10;
+ ZoneList<AlternativeGeneration*> alt_gens_;
+ AlternativeGeneration a_few_alt_gens_[kAFew];
+};
+
+void BoyerMoorePositionInfo::Set(int character) {
+ SetInterval(Interval(character, character));
+}
+
+namespace {
+
+ContainedInLattice AddRange(ContainedInLattice containment, const int* ranges,
+ int ranges_length, Interval new_range) {
+ DCHECK_EQ(1, ranges_length & 1);
+ DCHECK_EQ(String::kMaxCodePoint + 1, ranges[ranges_length - 1]);
+ if (containment == kLatticeUnknown) return containment;
+ bool inside = false;
+ int last = 0;
+ for (int i = 0; i < ranges_length; inside = !inside, last = ranges[i], i++) {
+ // Consider the range from last to ranges[i].
+ // We haven't got to the new range yet.
+ if (ranges[i] <= new_range.from()) continue;
+ // New range is wholly inside last-ranges[i]. Note that new_range.to() is
+ // inclusive, but the values in ranges are not.
+ if (last <= new_range.from() && new_range.to() < ranges[i]) {
+ return Combine(containment, inside ? kLatticeIn : kLatticeOut);
+ }
+ return kLatticeUnknown;
+ }
+ return containment;
+}
+
+int BitsetFirstSetBit(BoyerMoorePositionInfo::Bitset bitset) {
+ STATIC_ASSERT(BoyerMoorePositionInfo::kMapSize ==
+ 2 * kInt64Size * kBitsPerByte);
+
+ // Slight fiddling is needed here, since the bitset is of length 128 while
+ // CountTrailingZeros requires an integral type and std::bitset can only
+ // convert to unsigned long long. So we handle the most- and least-significant
+ // bits separately.
+
+ {
+ static constexpr BoyerMoorePositionInfo::Bitset mask(~uint64_t{0});
+ BoyerMoorePositionInfo::Bitset masked_bitset = bitset & mask;
+ STATIC_ASSERT(kInt64Size >= sizeof(decltype(masked_bitset.to_ullong())));
+ uint64_t lsb = masked_bitset.to_ullong();
+ if (lsb != 0) return base::bits::CountTrailingZeros(lsb);
+ }
+
+ {
+ BoyerMoorePositionInfo::Bitset masked_bitset = bitset >> 64;
+ uint64_t msb = masked_bitset.to_ullong();
+ if (msb != 0) return 64 + base::bits::CountTrailingZeros(msb);
+ }
+
+ return -1;
+}
+
+} // namespace
+
+void BoyerMoorePositionInfo::SetInterval(const Interval& interval) {
+ w_ = AddRange(w_, kWordRanges, kWordRangeCount, interval);
+
+ if (interval.size() >= kMapSize) {
+ map_count_ = kMapSize;
+ map_.set();
+ return;
+ }
+
+ for (int i = interval.from(); i <= interval.to(); i++) {
+ int mod_character = (i & kMask);
+ if (!map_[mod_character]) {
+ map_count_++;
+ map_.set(mod_character);
+ }
+ if (map_count_ == kMapSize) return;
+ }
+}
+
+void BoyerMoorePositionInfo::SetAll() {
+ w_ = kLatticeUnknown;
+ if (map_count_ != kMapSize) {
+ map_count_ = kMapSize;
+ map_.set();
+ }
+}
+
+BoyerMooreLookahead::BoyerMooreLookahead(int length, RegExpCompiler* compiler,
+ Zone* zone)
+ : length_(length), compiler_(compiler) {
+ if (compiler->one_byte()) {
+ max_char_ = String::kMaxOneByteCharCode;
+ } else {
+ max_char_ = String::kMaxUtf16CodeUnit;
+ }
+ bitmaps_ = new (zone) ZoneList<BoyerMoorePositionInfo*>(length, zone);
+ for (int i = 0; i < length; i++) {
+ bitmaps_->Add(new (zone) BoyerMoorePositionInfo(), zone);
+ }
+}
+
+// Find the longest range of lookahead that has the fewest number of different
+// characters that can occur at a given position. Since we are optimizing two
+// different parameters at once this is a tradeoff.
+bool BoyerMooreLookahead::FindWorthwhileInterval(int* from, int* to) {
+ int biggest_points = 0;
+ // If more than 32 characters out of 128 can occur it is unlikely that we can
+ // be lucky enough to step forwards much of the time.
+ const int kMaxMax = 32;
+ for (int max_number_of_chars = 4; max_number_of_chars < kMaxMax;
+ max_number_of_chars *= 2) {
+ biggest_points =
+ FindBestInterval(max_number_of_chars, biggest_points, from, to);
+ }
+ if (biggest_points == 0) return false;
+ return true;
+}
+
+// Find the highest-points range between 0 and length_ where the character
+// information is not too vague. 'Too vague' means that there are more than
+// max_number_of_chars that can occur at this position. Calculates the number
+// of points as the product of width-of-the-range and
+// probability-of-finding-one-of-the-characters, where the probability is
+// calculated using the frequency distribution of the sample subject string.
+int BoyerMooreLookahead::FindBestInterval(int max_number_of_chars,
+ int old_biggest_points, int* from,
+ int* to) {
+ int biggest_points = old_biggest_points;
+ static const int kSize = RegExpMacroAssembler::kTableSize;
+ for (int i = 0; i < length_;) {
+ while (i < length_ && Count(i) > max_number_of_chars) i++;
+ if (i == length_) break;
+ int remembered_from = i;
+
+ BoyerMoorePositionInfo::Bitset union_bitset;
+ for (; i < length_ && Count(i) <= max_number_of_chars; i++) {
+ union_bitset |= bitmaps_->at(i)->raw_bitset();
+ }
+
+ int frequency = 0;
+
+ // Iterate only over set bits.
+ int j;
+ while ((j = BitsetFirstSetBit(union_bitset)) != -1) {
+ DCHECK(union_bitset[j]); // Sanity check.
+ // Add 1 to the frequency to give a small per-character boost for
+ // the cases where our sampling is not good enough and many
+ // characters have a frequency of zero. This means the frequency
+ // can theoretically be up to 2*kSize though we treat it mostly as
+ // a fraction of kSize.
+ frequency += compiler_->frequency_collator()->Frequency(j) + 1;
+ union_bitset.reset(j);
+ }
+
+ // We use the probability of skipping times the distance we are skipping to
+ // judge the effectiveness of this. Actually we have a cut-off: By
+ // dividing by 2 we switch off the skipping if the probability of skipping
+ // is less than 50%. This is because the multibyte mask-and-compare
+ // skipping in quickcheck is more likely to do well on this case.
+ bool in_quickcheck_range =
+ ((i - remembered_from < 4) ||
+ (compiler_->one_byte() ? remembered_from <= 4 : remembered_from <= 2));
+ // Called 'probability' but it is only a rough estimate and can actually
+ // be outside the 0-kSize range.
+ int probability = (in_quickcheck_range ? kSize / 2 : kSize) - frequency;
+ int points = (i - remembered_from) * probability;
+ if (points > biggest_points) {
+ *from = remembered_from;
+ *to = i - 1;
+ biggest_points = points;
+ }
+ }
+ return biggest_points;
+}
+
+// Take all the characters that will not prevent a successful match if they
+// occur in the subject string in the range between min_lookahead and
+// max_lookahead (inclusive) measured from the current position. If the
+// character at max_lookahead offset is not one of these characters, then we
+// can safely skip forwards by the number of characters in the range.
+int BoyerMooreLookahead::GetSkipTable(int min_lookahead, int max_lookahead,
+ Handle<ByteArray> boolean_skip_table) {
+ const int kSkipArrayEntry = 0;
+ const int kDontSkipArrayEntry = 1;
+
+ std::memset(boolean_skip_table->GetDataStartAddress(), kSkipArrayEntry,
+ boolean_skip_table->length());
+
+ for (int i = max_lookahead; i >= min_lookahead; i--) {
+ BoyerMoorePositionInfo::Bitset bitset = bitmaps_->at(i)->raw_bitset();
+
+ // Iterate only over set bits.
+ int j;
+ while ((j = BitsetFirstSetBit(bitset)) != -1) {
+ DCHECK(bitset[j]); // Sanity check.
+ boolean_skip_table->set(j, kDontSkipArrayEntry);
+ bitset.reset(j);
+ }
+ }
+
+ const int skip = max_lookahead + 1 - min_lookahead;
+ return skip;
+}
+
+// See comment above on the implementation of GetSkipTable.
+void BoyerMooreLookahead::EmitSkipInstructions(RegExpMacroAssembler* masm) {
+ const int kSize = RegExpMacroAssembler::kTableSize;
+
+ int min_lookahead = 0;
+ int max_lookahead = 0;
+
+ if (!FindWorthwhileInterval(&min_lookahead, &max_lookahead)) return;
+
+ // Check if we only have a single non-empty position info, and that info
+ // contains precisely one character.
+ bool found_single_character = false;
+ int single_character = 0;
+ for (int i = max_lookahead; i >= min_lookahead; i--) {
+ BoyerMoorePositionInfo* map = bitmaps_->at(i);
+ if (map->map_count() == 0) continue;
+
+ if (found_single_character || map->map_count() > 1) {
+ found_single_character = false;
+ break;
+ }
+
+ DCHECK(!found_single_character);
+ DCHECK_EQ(map->map_count(), 1);
+
+ found_single_character = true;
+ single_character = BitsetFirstSetBit(map->raw_bitset());
+
+ DCHECK_NE(single_character, -1);
+ }
+
+ int lookahead_width = max_lookahead + 1 - min_lookahead;
+
+ if (found_single_character && lookahead_width == 1 && max_lookahead < 3) {
+ // The mask-compare can probably handle this better.
+ return;
+ }
+
+ if (found_single_character) {
+ Label cont, again;
+ masm->Bind(&again);
+ masm->LoadCurrentCharacter(max_lookahead, &cont, true);
+ if (max_char_ > kSize) {
+ masm->CheckCharacterAfterAnd(single_character,
+ RegExpMacroAssembler::kTableMask, &cont);
+ } else {
+ masm->CheckCharacter(single_character, &cont);
+ }
+ masm->AdvanceCurrentPosition(lookahead_width);
+ masm->GoTo(&again);
+ masm->Bind(&cont);
+ return;
+ }
+
+ Factory* factory = masm->isolate()->factory();
+ Handle<ByteArray> boolean_skip_table =
+ factory->NewByteArray(kSize, AllocationType::kOld);
+ int skip_distance =
+ GetSkipTable(min_lookahead, max_lookahead, boolean_skip_table);
+ DCHECK_NE(0, skip_distance);
+
+ Label cont, again;
+ masm->Bind(&again);
+ masm->LoadCurrentCharacter(max_lookahead, &cont, true);
+ masm->CheckBitInTable(boolean_skip_table, &cont);
+ masm->AdvanceCurrentPosition(skip_distance);
+ masm->GoTo(&again);
+ masm->Bind(&cont);
+}
+
+/* Code generation for choice nodes.
+ *
+ * We generate quick checks that do a mask and compare to eliminate a
+ * choice. If the quick check succeeds then it jumps to the continuation to
+ * do slow checks and check subsequent nodes. If it fails (the common case)
+ * it falls through to the next choice.
+ *
+ * Here is the desired flow graph. Nodes directly below each other imply
+ * fallthrough. Alternatives 1 and 2 have quick checks. Alternative
+ * 3 doesn't have a quick check so we have to call the slow check.
+ * Nodes are marked Qn for quick checks and Sn for slow checks. The entire
+ * regexp continuation is generated directly after the Sn node, up to the
+ * next GoTo if we decide to reuse some already generated code. Some
+ * nodes expect preload_characters to be preloaded into the current
+ * character register. R nodes do this preloading. Vertices are marked
+ * F for failures and S for success (possible success in the case of quick
+ * nodes). L, V, < and > are used as arrow heads.
+ *
+ * ----------> R
+ * |
+ * V
+ * Q1 -----> S1
+ * | S /
+ * F| /
+ * | F/
+ * | /
+ * | R
+ * | /
+ * V L
+ * Q2 -----> S2
+ * | S /
+ * F| /
+ * | F/
+ * | /
+ * | R
+ * | /
+ * V L
+ * S3
+ * |
+ * F|
+ * |
+ * R
+ * |
+ * backtrack V
+ * <----------Q4
+ * \ F |
+ * \ |S
+ * \ F V
+ * \-----S4
+ *
+ * For greedy loops we push the current position, then generate the code that
+ * eats the input specially in EmitGreedyLoop. The other choice (the
+ * continuation) is generated by the normal code in EmitChoices, and steps back
+ * in the input to the starting position when it fails to match. The loop code
+ * looks like this (U is the unwind code that steps back in the greedy loop).
+ *
+ * _____
+ * / \
+ * V |
+ * ----------> S1 |
+ * /| |
+ * / |S |
+ * F/ \_____/
+ * /
+ * |<-----
+ * | \
+ * V |S
+ * Q2 ---> U----->backtrack
+ * | F /
+ * S| /
+ * V F /
+ * S2--/
+ */
+
+GreedyLoopState::GreedyLoopState(bool not_at_start) {
+ counter_backtrack_trace_.set_backtrack(&label_);
+ if (not_at_start) counter_backtrack_trace_.set_at_start(Trace::FALSE_VALUE);
+}
+
+void ChoiceNode::AssertGuardsMentionRegisters(Trace* trace) {
+#ifdef DEBUG
+ int choice_count = alternatives_->length();
+ for (int i = 0; i < choice_count - 1; i++) {
+ GuardedAlternative alternative = alternatives_->at(i);
+ ZoneList<Guard*>* guards = alternative.guards();
+ int guard_count = (guards == nullptr) ? 0 : guards->length();
+ for (int j = 0; j < guard_count; j++) {
+ DCHECK(!trace->mentions_reg(guards->at(j)->reg()));
+ }
+ }
+#endif
+}
+
+void ChoiceNode::SetUpPreLoad(RegExpCompiler* compiler, Trace* current_trace,
+ PreloadState* state) {
+ if (state->eats_at_least_ == PreloadState::kEatsAtLeastNotYetInitialized) {
+ // Save some time by looking at most one machine word ahead.
+ state->eats_at_least_ =
+ EatsAtLeast(compiler->one_byte() ? 4 : 2, kRecursionBudget,
+ current_trace->at_start() == Trace::FALSE_VALUE);
+ }
+ state->preload_characters_ =
+ CalculatePreloadCharacters(compiler, state->eats_at_least_);
+
+ state->preload_is_current_ =
+ (current_trace->characters_preloaded() == state->preload_characters_);
+ state->preload_has_checked_bounds_ = state->preload_is_current_;
+}
+
+void ChoiceNode::Emit(RegExpCompiler* compiler, Trace* trace) {
+ int choice_count = alternatives_->length();
+
+ if (choice_count == 1 && alternatives_->at(0).guards() == nullptr) {
+ alternatives_->at(0).node()->Emit(compiler, trace);
+ return;
+ }
+
+ AssertGuardsMentionRegisters(trace);
+
+ LimitResult limit_result = LimitVersions(compiler, trace);
+ if (limit_result == DONE) return;
+ DCHECK(limit_result == CONTINUE);
+
+ // For loop nodes we already flushed (see LoopChoiceNode::Emit), but for
+ // other choice nodes we only flush if we are out of code size budget.
+ if (trace->flush_budget() == 0 && trace->actions() != nullptr) {
+ trace->Flush(compiler, this);
+ return;
+ }
+
+ RecursionCheck rc(compiler);
+
+ PreloadState preload;
+ preload.init();
+ GreedyLoopState greedy_loop_state(not_at_start());
+
+ int text_length = GreedyLoopTextLengthForAlternative(&alternatives_->at(0));
+ AlternativeGenerationList alt_gens(choice_count, zone());
+
+ if (choice_count > 1 && text_length != kNodeIsTooComplexForGreedyLoops) {
+ trace = EmitGreedyLoop(compiler, trace, &alt_gens, &preload,
+ &greedy_loop_state, text_length);
+ } else {
+ // TODO(erikcorry): Delete this. We don't need this label, but it makes us
+ // match the traces produced pre-cleanup.
+ Label second_choice;
+ compiler->macro_assembler()->Bind(&second_choice);
+
+ preload.eats_at_least_ = EmitOptimizedUnanchoredSearch(compiler, trace);
+
+ EmitChoices(compiler, &alt_gens, 0, trace, &preload);
+ }
+
+ // At this point we need to generate slow checks for the alternatives where
+ // the quick check was inlined. We can recognize these because the associated
+ // label was bound.
+ int new_flush_budget = trace->flush_budget() / choice_count;
+ for (int i = 0; i < choice_count; i++) {
+ AlternativeGeneration* alt_gen = alt_gens.at(i);
+ Trace new_trace(*trace);
+ // If there are actions to be flushed we have to limit how many times
+ // they are flushed. Take the budget of the parent trace and distribute
+ // it fairly amongst the children.
+ if (new_trace.actions() != nullptr) {
+ new_trace.set_flush_budget(new_flush_budget);
+ }
+ bool next_expects_preload =
+ i == choice_count - 1 ? false : alt_gens.at(i + 1)->expects_preload;
+ EmitOutOfLineContinuation(compiler, &new_trace, alternatives_->at(i),
+ alt_gen, preload.preload_characters_,
+ next_expects_preload);
+ }
+}
+
+Trace* ChoiceNode::EmitGreedyLoop(RegExpCompiler* compiler, Trace* trace,
+ AlternativeGenerationList* alt_gens,
+ PreloadState* preload,
+ GreedyLoopState* greedy_loop_state,
+ int text_length) {
+ RegExpMacroAssembler* macro_assembler = compiler->macro_assembler();
+ // Here we have special handling for greedy loops containing only text nodes
+ // and other simple nodes. These are handled by pushing the current
+ // position on the stack and then incrementing the current position each
+ // time around the switch. On backtrack we decrement the current position
+ // and check it against the pushed value. This avoids pushing backtrack
+ // information for each iteration of the loop, which could take up a lot of
+ // space.
+ DCHECK(trace->stop_node() == nullptr);
+ macro_assembler->PushCurrentPosition();
+ Label greedy_match_failed;
+ Trace greedy_match_trace;
+ if (not_at_start()) greedy_match_trace.set_at_start(Trace::FALSE_VALUE);
+ greedy_match_trace.set_backtrack(&greedy_match_failed);
+ Label loop_label;
+ macro_assembler->Bind(&loop_label);
+ greedy_match_trace.set_stop_node(this);
+ greedy_match_trace.set_loop_label(&loop_label);
+ alternatives_->at(0).node()->Emit(compiler, &greedy_match_trace);
+ macro_assembler->Bind(&greedy_match_failed);
+
+ Label second_choice; // For use in greedy matches.
+ macro_assembler->Bind(&second_choice);
+
+ Trace* new_trace = greedy_loop_state->counter_backtrack_trace();
+
+ EmitChoices(compiler, alt_gens, 1, new_trace, preload);
+
+ macro_assembler->Bind(greedy_loop_state->label());
+ // If we have unwound to the bottom then backtrack.
+ macro_assembler->CheckGreedyLoop(trace->backtrack());
+ // Otherwise try the second priority at an earlier position.
+ macro_assembler->AdvanceCurrentPosition(-text_length);
+ macro_assembler->GoTo(&second_choice);
+ return new_trace;
+}
+
+int ChoiceNode::EmitOptimizedUnanchoredSearch(RegExpCompiler* compiler,
+ Trace* trace) {
+ int eats_at_least = PreloadState::kEatsAtLeastNotYetInitialized;
+ if (alternatives_->length() != 2) return eats_at_least;
+
+ GuardedAlternative alt1 = alternatives_->at(1);
+ if (alt1.guards() != nullptr && alt1.guards()->length() != 0) {
+ return eats_at_least;
+ }
+ RegExpNode* eats_anything_node = alt1.node();
+ if (eats_anything_node->GetSuccessorOfOmnivorousTextNode(compiler) != this) {
+ return eats_at_least;
+ }
+
+ // Really we should be creating a new trace when we execute this function,
+ // but there is no need, because the code it generates cannot backtrack, and
+ // we always arrive here with a trivial trace (since it's the entry to a
+ // loop. That also implies that there are no preloaded characters, which is
+ // good, because it means we won't be violating any assumptions by
+ // overwriting those characters with new load instructions.
+ DCHECK(trace->is_trivial());
+
+ RegExpMacroAssembler* macro_assembler = compiler->macro_assembler();
+ Isolate* isolate = macro_assembler->isolate();
+ // At this point we know that we are at a non-greedy loop that will eat
+ // any character one at a time. Any non-anchored regexp has such a
+ // loop prepended to it in order to find where it starts. We look for
+ // a pattern of the form ...abc... where we can look 6 characters ahead
+ // and step forwards 3 if the character is not one of abc. Abc need
+ // not be atoms, they can be any reasonably limited character class or
+ // small alternation.
+ BoyerMooreLookahead* bm = bm_info(false);
+ if (bm == nullptr) {
+ eats_at_least =
+ Min(kMaxLookaheadForBoyerMoore,
+ EatsAtLeast(kMaxLookaheadForBoyerMoore, kRecursionBudget, false));
+ if (eats_at_least >= 1) {
+ bm = new (zone()) BoyerMooreLookahead(eats_at_least, compiler, zone());
+ GuardedAlternative alt0 = alternatives_->at(0);
+ alt0.node()->FillInBMInfo(isolate, 0, kRecursionBudget, bm, false);
+ }
+ }
+ if (bm != nullptr) {
+ bm->EmitSkipInstructions(macro_assembler);
+ }
+ return eats_at_least;
+}
+
+void ChoiceNode::EmitChoices(RegExpCompiler* compiler,
+ AlternativeGenerationList* alt_gens,
+ int first_choice, Trace* trace,
+ PreloadState* preload) {
+ RegExpMacroAssembler* macro_assembler = compiler->macro_assembler();
+ SetUpPreLoad(compiler, trace, preload);
+
+ // For now we just call all choices one after the other. The idea ultimately
+ // is to use the Dispatch table to try only the relevant ones.
+ int choice_count = alternatives_->length();
+
+ int new_flush_budget = trace->flush_budget() / choice_count;
+
+ for (int i = first_choice; i < choice_count; i++) {
+ bool is_last = i == choice_count - 1;
+ bool fall_through_on_failure = !is_last;
+ GuardedAlternative alternative = alternatives_->at(i);
+ AlternativeGeneration* alt_gen = alt_gens->at(i);
+ alt_gen->quick_check_details.set_characters(preload->preload_characters_);
+ ZoneList<Guard*>* guards = alternative.guards();
+ int guard_count = (guards == nullptr) ? 0 : guards->length();
+ Trace new_trace(*trace);
+ new_trace.set_characters_preloaded(
+ preload->preload_is_current_ ? preload->preload_characters_ : 0);
+ if (preload->preload_has_checked_bounds_) {
+ new_trace.set_bound_checked_up_to(preload->preload_characters_);
+ }
+ new_trace.quick_check_performed()->Clear();
+ if (not_at_start_) new_trace.set_at_start(Trace::FALSE_VALUE);
+ if (!is_last) {
+ new_trace.set_backtrack(&alt_gen->after);
+ }
+ alt_gen->expects_preload = preload->preload_is_current_;
+ bool generate_full_check_inline = false;
+ if (compiler->optimize() &&
+ try_to_emit_quick_check_for_alternative(i == 0) &&
+ alternative.node()->EmitQuickCheck(
+ compiler, trace, &new_trace, preload->preload_has_checked_bounds_,
+ &alt_gen->possible_success, &alt_gen->quick_check_details,
+ fall_through_on_failure)) {
+ // Quick check was generated for this choice.
+ preload->preload_is_current_ = true;
+ preload->preload_has_checked_bounds_ = true;
+ // If we generated the quick check to fall through on possible success,
+ // we now need to generate the full check inline.
+ if (!fall_through_on_failure) {
+ macro_assembler->Bind(&alt_gen->possible_success);
+ new_trace.set_quick_check_performed(&alt_gen->quick_check_details);
+ new_trace.set_characters_preloaded(preload->preload_characters_);
+ new_trace.set_bound_checked_up_to(preload->preload_characters_);
+ generate_full_check_inline = true;
+ }
+ } else if (alt_gen->quick_check_details.cannot_match()) {
+ if (!fall_through_on_failure) {
+ macro_assembler->GoTo(trace->backtrack());
+ }
+ continue;
+ } else {
+ // No quick check was generated. Put the full code here.
+ // If this is not the first choice then there could be slow checks from
+ // previous cases that go here when they fail. There's no reason to
+ // insist that they preload characters since the slow check we are about
+ // to generate probably can't use it.
+ if (i != first_choice) {
+ alt_gen->expects_preload = false;
+ new_trace.InvalidateCurrentCharacter();
+ }
+ generate_full_check_inline = true;
+ }
+ if (generate_full_check_inline) {
+ if (new_trace.actions() != nullptr) {
+ new_trace.set_flush_budget(new_flush_budget);
+ }
+ for (int j = 0; j < guard_count; j++) {
+ GenerateGuard(macro_assembler, guards->at(j), &new_trace);
+ }
+ alternative.node()->Emit(compiler, &new_trace);
+ preload->preload_is_current_ = false;
+ }
+ macro_assembler->Bind(&alt_gen->after);
+ }
+}
+
+void ChoiceNode::EmitOutOfLineContinuation(RegExpCompiler* compiler,
+ Trace* trace,
+ GuardedAlternative alternative,
+ AlternativeGeneration* alt_gen,
+ int preload_characters,
+ bool next_expects_preload) {
+ if (!alt_gen->possible_success.is_linked()) return;
+
+ RegExpMacroAssembler* macro_assembler = compiler->macro_assembler();
+ macro_assembler->Bind(&alt_gen->possible_success);
+ Trace out_of_line_trace(*trace);
+ out_of_line_trace.set_characters_preloaded(preload_characters);
+ out_of_line_trace.set_quick_check_performed(&alt_gen->quick_check_details);
+ if (not_at_start_) out_of_line_trace.set_at_start(Trace::FALSE_VALUE);
+ ZoneList<Guard*>* guards = alternative.guards();
+ int guard_count = (guards == nullptr) ? 0 : guards->length();
+ if (next_expects_preload) {
+ Label reload_current_char;
+ out_of_line_trace.set_backtrack(&reload_current_char);
+ for (int j = 0; j < guard_count; j++) {
+ GenerateGuard(macro_assembler, guards->at(j), &out_of_line_trace);
+ }
+ alternative.node()->Emit(compiler, &out_of_line_trace);
+ macro_assembler->Bind(&reload_current_char);
+ // Reload the current character, since the next quick check expects that.
+ // We don't need to check bounds here because we only get into this
+ // code through a quick check which already did the checked load.
+ macro_assembler->LoadCurrentCharacter(trace->cp_offset(), nullptr, false,
+ preload_characters);
+ macro_assembler->GoTo(&(alt_gen->after));
+ } else {
+ out_of_line_trace.set_backtrack(&(alt_gen->after));
+ for (int j = 0; j < guard_count; j++) {
+ GenerateGuard(macro_assembler, guards->at(j), &out_of_line_trace);
+ }
+ alternative.node()->Emit(compiler, &out_of_line_trace);
+ }
+}
+
+void ActionNode::Emit(RegExpCompiler* compiler, Trace* trace) {
+ RegExpMacroAssembler* assembler = compiler->macro_assembler();
+ LimitResult limit_result = LimitVersions(compiler, trace);
+ if (limit_result == DONE) return;
+ DCHECK(limit_result == CONTINUE);
+
+ RecursionCheck rc(compiler);
+
+ switch (action_type_) {
+ case STORE_POSITION: {
+ Trace::DeferredCapture new_capture(data_.u_position_register.reg,
+ data_.u_position_register.is_capture,
+ trace);
+ Trace new_trace = *trace;
+ new_trace.add_action(&new_capture);
+ on_success()->Emit(compiler, &new_trace);
+ break;
+ }
+ case INCREMENT_REGISTER: {
+ Trace::DeferredIncrementRegister new_increment(
+ data_.u_increment_register.reg);
+ Trace new_trace = *trace;
+ new_trace.add_action(&new_increment);
+ on_success()->Emit(compiler, &new_trace);
+ break;
+ }
+ case SET_REGISTER: {
+ Trace::DeferredSetRegister new_set(data_.u_store_register.reg,
+ data_.u_store_register.value);
+ Trace new_trace = *trace;
+ new_trace.add_action(&new_set);
+ on_success()->Emit(compiler, &new_trace);
+ break;
+ }
+ case CLEAR_CAPTURES: {
+ Trace::DeferredClearCaptures new_capture(Interval(
+ data_.u_clear_captures.range_from, data_.u_clear_captures.range_to));
+ Trace new_trace = *trace;
+ new_trace.add_action(&new_capture);
+ on_success()->Emit(compiler, &new_trace);
+ break;
+ }
+ case BEGIN_SUBMATCH:
+ if (!trace->is_trivial()) {
+ trace->Flush(compiler, this);
+ } else {
+ assembler->WriteCurrentPositionToRegister(
+ data_.u_submatch.current_position_register, 0);
+ assembler->WriteStackPointerToRegister(
+ data_.u_submatch.stack_pointer_register);
+ on_success()->Emit(compiler, trace);
+ }
+ break;
+ case EMPTY_MATCH_CHECK: {
+ int start_pos_reg = data_.u_empty_match_check.start_register;
+ int stored_pos = 0;
+ int rep_reg = data_.u_empty_match_check.repetition_register;
+ bool has_minimum = (rep_reg != RegExpCompiler::kNoRegister);
+ bool know_dist = trace->GetStoredPosition(start_pos_reg, &stored_pos);
+ if (know_dist && !has_minimum && stored_pos == trace->cp_offset()) {
+ // If we know we haven't advanced and there is no minimum we
+ // can just backtrack immediately.
+ assembler->GoTo(trace->backtrack());
+ } else if (know_dist && stored_pos < trace->cp_offset()) {
+ // If we know we've advanced we can generate the continuation
+ // immediately.
+ on_success()->Emit(compiler, trace);
+ } else if (!trace->is_trivial()) {
+ trace->Flush(compiler, this);
+ } else {
+ Label skip_empty_check;
+ // If we have a minimum number of repetitions we check the current
+ // number first and skip the empty check if it's not enough.
+ if (has_minimum) {
+ int limit = data_.u_empty_match_check.repetition_limit;
+ assembler->IfRegisterLT(rep_reg, limit, &skip_empty_check);
+ }
+ // If the match is empty we bail out, otherwise we fall through
+ // to the on-success continuation.
+ assembler->IfRegisterEqPos(data_.u_empty_match_check.start_register,
+ trace->backtrack());
+ assembler->Bind(&skip_empty_check);
+ on_success()->Emit(compiler, trace);
+ }
+ break;
+ }
+ case POSITIVE_SUBMATCH_SUCCESS: {
+ if (!trace->is_trivial()) {
+ trace->Flush(compiler, this);
+ return;
+ }
+ assembler->ReadCurrentPositionFromRegister(
+ data_.u_submatch.current_position_register);
+ assembler->ReadStackPointerFromRegister(
+ data_.u_submatch.stack_pointer_register);
+ int clear_register_count = data_.u_submatch.clear_register_count;
+ if (clear_register_count == 0) {
+ on_success()->Emit(compiler, trace);
+ return;
+ }
+ int clear_registers_from = data_.u_submatch.clear_register_from;
+ Label clear_registers_backtrack;
+ Trace new_trace = *trace;
+ new_trace.set_backtrack(&clear_registers_backtrack);
+ on_success()->Emit(compiler, &new_trace);
+
+ assembler->Bind(&clear_registers_backtrack);
+ int clear_registers_to = clear_registers_from + clear_register_count - 1;
+ assembler->ClearRegisters(clear_registers_from, clear_registers_to);
+
+ DCHECK(trace->backtrack() == nullptr);
+ assembler->Backtrack();
+ return;
+ }
+ default:
+ UNREACHABLE();
+ }
+}
+
+void BackReferenceNode::Emit(RegExpCompiler* compiler, Trace* trace) {
+ RegExpMacroAssembler* assembler = compiler->macro_assembler();
+ if (!trace->is_trivial()) {
+ trace->Flush(compiler, this);
+ return;
+ }
+
+ LimitResult limit_result = LimitVersions(compiler, trace);
+ if (limit_result == DONE) return;
+ DCHECK(limit_result == CONTINUE);
+
+ RecursionCheck rc(compiler);
+
+ DCHECK_EQ(start_reg_ + 1, end_reg_);
+ if (IgnoreCase(flags_)) {
+ assembler->CheckNotBackReferenceIgnoreCase(
+ start_reg_, read_backward(), IsUnicode(flags_), 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 (IsUnicode(flags_) && !compiler->one_byte()) {
+ assembler->CheckNotInSurrogatePair(trace->cp_offset(), trace->backtrack());
+ }
+ on_success()->Emit(compiler, trace);
+}
+
+// -------------------------------------------------------------------
+// Analysis
+
+void Analysis::EnsureAnalyzed(RegExpNode* that) {
+ StackLimitCheck check(isolate());
+ if (check.HasOverflowed()) {
+ fail("Stack overflow");
+ return;
+ }
+ if (that->info()->been_analyzed || that->info()->being_analyzed) return;
+ that->info()->being_analyzed = true;
+ that->Accept(this);
+ that->info()->being_analyzed = false;
+ that->info()->been_analyzed = true;
+}
+
+void Analysis::VisitEnd(EndNode* that) {
+ // nothing to do
+}
+
+void TextNode::CalculateOffsets() {
+ int element_count = elements()->length();
+ // Set up the offsets of the elements relative to the start. This is a fixed
+ // quantity since a TextNode can only contain fixed-width things.
+ int cp_offset = 0;
+ for (int i = 0; i < element_count; i++) {
+ TextElement& elm = elements()->at(i);
+ elm.set_cp_offset(cp_offset);
+ cp_offset += elm.length();
+ }
+}
+
+void Analysis::VisitText(TextNode* that) {
+ that->MakeCaseIndependent(isolate(), is_one_byte_);
+ EnsureAnalyzed(that->on_success());
+ if (!has_failed()) {
+ that->CalculateOffsets();
+ }
+}
+
+void Analysis::VisitAction(ActionNode* that) {
+ RegExpNode* target = that->on_success();
+ EnsureAnalyzed(target);
+ if (!has_failed()) {
+ // If the next node is interested in what it follows then this node
+ // has to be interested too so it can pass the information on.
+ that->info()->AddFromFollowing(target->info());
+ }
+}
+
+void Analysis::VisitChoice(ChoiceNode* that) {
+ NodeInfo* info = that->info();
+ for (int i = 0; i < that->alternatives()->length(); i++) {
+ RegExpNode* node = that->alternatives()->at(i).node();
+ EnsureAnalyzed(node);
+ if (has_failed()) return;
+ // Anything the following nodes need to know has to be known by
+ // this node also, so it can pass it on.
+ info->AddFromFollowing(node->info());
+ }
+}
+
+void Analysis::VisitLoopChoice(LoopChoiceNode* that) {
+ NodeInfo* info = that->info();
+ for (int i = 0; i < that->alternatives()->length(); i++) {
+ RegExpNode* node = that->alternatives()->at(i).node();
+ if (node != that->loop_node()) {
+ EnsureAnalyzed(node);
+ if (has_failed()) return;
+ info->AddFromFollowing(node->info());
+ }
+ }
+ // Check the loop last since it may need the value of this node
+ // to get a correct result.
+ EnsureAnalyzed(that->loop_node());
+ if (!has_failed()) {
+ info->AddFromFollowing(that->loop_node()->info());
+ }
+}
+
+void Analysis::VisitBackReference(BackReferenceNode* that) {
+ EnsureAnalyzed(that->on_success());
+}
+
+void Analysis::VisitAssertion(AssertionNode* that) {
+ EnsureAnalyzed(that->on_success());
+}
+
+void BackReferenceNode::FillInBMInfo(Isolate* isolate, int offset, int budget,
+ BoyerMooreLookahead* bm,
+ bool not_at_start) {
+ // Working out the set of characters that a backreference can match is too
+ // hard, so we just say that any character can match.
+ bm->SetRest(offset);
+ SaveBMInfo(bm, not_at_start, offset);
+}
+
+STATIC_ASSERT(BoyerMoorePositionInfo::kMapSize ==
+ RegExpMacroAssembler::kTableSize);
+
+void ChoiceNode::FillInBMInfo(Isolate* isolate, int offset, int budget,
+ BoyerMooreLookahead* bm, bool not_at_start) {
+ ZoneList<GuardedAlternative>* alts = alternatives();
+ budget = (budget - 1) / alts->length();
+ for (int i = 0; i < alts->length(); i++) {
+ GuardedAlternative& alt = alts->at(i);
+ if (alt.guards() != nullptr && alt.guards()->length() != 0) {
+ bm->SetRest(offset); // Give up trying to fill in info.
+ SaveBMInfo(bm, not_at_start, offset);
+ return;
+ }
+ alt.node()->FillInBMInfo(isolate, offset, budget, bm, not_at_start);
+ }
+ SaveBMInfo(bm, not_at_start, offset);
+}
+
+void TextNode::FillInBMInfo(Isolate* isolate, int initial_offset, int budget,
+ BoyerMooreLookahead* bm, bool not_at_start) {
+ if (initial_offset >= bm->length()) return;
+ int offset = initial_offset;
+ int max_char = bm->max_char();
+ for (int i = 0; i < elements()->length(); i++) {
+ if (offset >= bm->length()) {
+ if (initial_offset == 0) set_bm_info(not_at_start, bm);
+ return;
+ }
+ TextElement text = elements()->at(i);
+ if (text.text_type() == TextElement::ATOM) {
+ RegExpAtom* atom = text.atom();
+ for (int j = 0; j < atom->length(); j++, offset++) {
+ if (offset >= bm->length()) {
+ if (initial_offset == 0) set_bm_info(not_at_start, bm);
+ return;
+ }
+ uc16 character = atom->data()[j];
+ if (IgnoreCase(atom->flags())) {
+ unibrow::uchar chars[4];
+ int length = GetCaseIndependentLetters(
+ isolate, character, bm->max_char() == String::kMaxOneByteCharCode,
+ chars, 4);
+ for (int j = 0; j < length; j++) {
+ bm->Set(offset, chars[j]);
+ }
+ } else {
+ if (character <= max_char) bm->Set(offset, character);
+ }
+ }
+ } else {
+ DCHECK_EQ(TextElement::CHAR_CLASS, text.text_type());
+ RegExpCharacterClass* char_class = text.char_class();
+ ZoneList<CharacterRange>* ranges = char_class->ranges(zone());
+ if (char_class->is_negated()) {
+ bm->SetAll(offset);
+ } else {
+ for (int k = 0; k < ranges->length(); k++) {
+ CharacterRange& range = ranges->at(k);
+ if (range.from() > max_char) continue;
+ int to = Min(max_char, static_cast<int>(range.to()));
+ bm->SetInterval(offset, Interval(range.from(), to));
+ }
+ }
+ offset++;
+ }
+ }
+ if (offset >= bm->length()) {
+ if (initial_offset == 0) set_bm_info(not_at_start, bm);
+ return;
+ }
+ on_success()->FillInBMInfo(isolate, offset, budget - 1, bm,
+ true); // Not at start after a text node.
+ if (initial_offset == 0) set_bm_info(not_at_start, bm);
+}
+
+// static
+RegExpNode* RegExpCompiler::OptionallyStepBackToLeadSurrogate(
+ RegExpCompiler* compiler, RegExpNode* on_success, JSRegExp::Flags flags) {
+ 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, flags);
+ RegExpLookaround::Builder builder(true, step_back, stack_register,
+ position_register);
+ RegExpNode* match_trail = TextNode::CreateForCharacterRanges(
+ zone, trail_surrogates, false, builder.on_match_success(), flags);
+
+ optional_step_back->AddAlternative(
+ GuardedAlternative(builder.ForMatch(match_trail)));
+ optional_step_back->AddAlternative(GuardedAlternative(on_success));
+
+ return optional_step_back;
+}
+
+} // namespace internal
+} // namespace v8
diff --git a/deps/v8/src/regexp/regexp-compiler.h b/deps/v8/src/regexp/regexp-compiler.h
new file mode 100644
index 0000000000..1b70abfd98
--- /dev/null
+++ b/deps/v8/src/regexp/regexp-compiler.h
@@ -0,0 +1,657 @@
+// Copyright 2019 the V8 project authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+#ifndef V8_REGEXP_REGEXP_COMPILER_H_
+#define V8_REGEXP_REGEXP_COMPILER_H_
+
+#include <bitset>
+
+#include "src/base/small-vector.h"
+#include "src/regexp/regexp-nodes.h"
+
+namespace v8 {
+namespace internal {
+
+class DynamicBitSet;
+class Isolate;
+
+namespace regexp_compiler_constants {
+
+// 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.
+constexpr uc32 kRangeEndMarker = 0x110000;
+constexpr int kSpaceRanges[] = {
+ '\t', '\r' + 1, ' ', ' ' + 1, 0x00A0, 0x00A1, 0x1680,
+ 0x1681, 0x2000, 0x200B, 0x2028, 0x202A, 0x202F, 0x2030,
+ 0x205F, 0x2060, 0x3000, 0x3001, 0xFEFF, 0xFF00, kRangeEndMarker};
+constexpr int kSpaceRangeCount = arraysize(kSpaceRanges);
+
+constexpr int kWordRanges[] = {'0', '9' + 1, 'A', 'Z' + 1, '_',
+ '_' + 1, 'a', 'z' + 1, kRangeEndMarker};
+constexpr int kWordRangeCount = arraysize(kWordRanges);
+constexpr int kDigitRanges[] = {'0', '9' + 1, kRangeEndMarker};
+constexpr int kDigitRangeCount = arraysize(kDigitRanges);
+constexpr int kSurrogateRanges[] = {kLeadSurrogateStart,
+ kLeadSurrogateStart + 1, kRangeEndMarker};
+constexpr int kSurrogateRangeCount = arraysize(kSurrogateRanges);
+constexpr int kLineTerminatorRanges[] = {0x000A, 0x000B, 0x000D, 0x000E,
+ 0x2028, 0x202A, kRangeEndMarker};
+constexpr int kLineTerminatorRangeCount = arraysize(kLineTerminatorRanges);
+
+// More makes code generation slower, less makes V8 benchmark score lower.
+constexpr int kMaxLookaheadForBoyerMoore = 8;
+// In a 3-character pattern you can maximally step forwards 3 characters
+// at a time, which is not always enough to pay for the extra logic.
+constexpr int kPatternTooShortForBoyerMoore = 2;
+
+} // namespace regexp_compiler_constants
+
+inline bool IgnoreCase(JSRegExp::Flags flags) {
+ return (flags & JSRegExp::kIgnoreCase) != 0;
+}
+
+inline bool IsUnicode(JSRegExp::Flags flags) {
+ return (flags & JSRegExp::kUnicode) != 0;
+}
+
+inline bool IsSticky(JSRegExp::Flags flags) {
+ return (flags & JSRegExp::kSticky) != 0;
+}
+
+inline bool IsGlobal(JSRegExp::Flags flags) {
+ return (flags & JSRegExp::kGlobal) != 0;
+}
+
+inline bool DotAll(JSRegExp::Flags flags) {
+ return (flags & JSRegExp::kDotAll) != 0;
+}
+
+inline bool Multiline(JSRegExp::Flags flags) {
+ return (flags & JSRegExp::kMultiline) != 0;
+}
+
+inline bool NeedsUnicodeCaseEquivalents(JSRegExp::Flags flags) {
+ // Both unicode and ignore_case flags are set. We need to use ICU to find
+ // the closure over case equivalents.
+ return IsUnicode(flags) && IgnoreCase(flags);
+}
+
+// Details of a quick mask-compare check that can look ahead in the
+// input stream.
+class QuickCheckDetails {
+ public:
+ QuickCheckDetails()
+ : characters_(0), mask_(0), value_(0), cannot_match_(false) {}
+ explicit QuickCheckDetails(int characters)
+ : characters_(characters), mask_(0), value_(0), cannot_match_(false) {}
+ bool Rationalize(bool one_byte);
+ // Merge in the information from another branch of an alternation.
+ void Merge(QuickCheckDetails* other, int from_index);
+ // Advance the current position by some amount.
+ void Advance(int by, bool one_byte);
+ void Clear();
+ bool cannot_match() { return cannot_match_; }
+ void set_cannot_match() { cannot_match_ = true; }
+ struct Position {
+ Position() : mask(0), value(0), determines_perfectly(false) {}
+ uc16 mask;
+ uc16 value;
+ bool determines_perfectly;
+ };
+ int characters() { return characters_; }
+ void set_characters(int characters) { characters_ = characters; }
+ Position* positions(int index) {
+ DCHECK_LE(0, index);
+ DCHECK_GT(characters_, index);
+ return positions_ + index;
+ }
+ uint32_t mask() { return mask_; }
+ uint32_t value() { return value_; }
+
+ private:
+ // How many characters do we have quick check information from. This is
+ // the same for all branches of a choice node.
+ int characters_;
+ Position positions_[4];
+ // These values are the condensate of the above array after Rationalize().
+ uint32_t mask_;
+ uint32_t value_;
+ // If set to true, there is no way this quick check can match at all.
+ // E.g., if it requires to be at the start of the input, and isn't.
+ bool cannot_match_;
+};
+
+// Improve the speed that we scan for an initial point where a non-anchored
+// regexp can match by using a Boyer-Moore-like table. This is done by
+// identifying non-greedy non-capturing loops in the nodes that eat any
+// character one at a time. For example in the middle of the regexp
+// /foo[\s\S]*?bar/ we find such a loop. There is also such a loop implicitly
+// inserted at the start of any non-anchored regexp.
+//
+// When we have found such a loop we look ahead in the nodes to find the set of
+// characters that can come at given distances. For example for the regexp
+// /.?foo/ we know that there are at least 3 characters ahead of us, and the
+// sets of characters that can occur are [any, [f, o], [o]]. We find a range in
+// the lookahead info where the set of characters is reasonably constrained. In
+// our example this is from index 1 to 2 (0 is not constrained). We can now
+// look 3 characters ahead and if we don't find one of [f, o] (the union of
+// [f, o] and [o]) then we can skip forwards by the range size (in this case 2).
+//
+// For Unicode input strings we do the same, but modulo 128.
+//
+// We also look at the first string fed to the regexp and use that to get a hint
+// of the character frequencies in the inputs. This affects the assessment of
+// whether the set of characters is 'reasonably constrained'.
+//
+// We also have another lookahead mechanism (called quick check in the code),
+// which uses a wide load of multiple characters followed by a mask and compare
+// to determine whether a match is possible at this point.
+enum ContainedInLattice {
+ kNotYet = 0,
+ kLatticeIn = 1,
+ kLatticeOut = 2,
+ kLatticeUnknown = 3 // Can also mean both in and out.
+};
+
+inline ContainedInLattice Combine(ContainedInLattice a, ContainedInLattice b) {
+ return static_cast<ContainedInLattice>(a | b);
+}
+
+class BoyerMoorePositionInfo : public ZoneObject {
+ public:
+ bool at(int i) const { return map_[i]; }
+
+ static constexpr int kMapSize = 128;
+ static constexpr int kMask = kMapSize - 1;
+
+ int map_count() const { return map_count_; }
+
+ void Set(int character);
+ void SetInterval(const Interval& interval);
+ void SetAll();
+
+ bool is_non_word() { return w_ == kLatticeOut; }
+ bool is_word() { return w_ == kLatticeIn; }
+
+ using Bitset = std::bitset<kMapSize>;
+ Bitset raw_bitset() const { return map_; }
+
+ private:
+ Bitset map_;
+ int map_count_ = 0; // Number of set bits in the map.
+ ContainedInLattice w_ = kNotYet; // The \w character class.
+};
+
+class BoyerMooreLookahead : public ZoneObject {
+ public:
+ BoyerMooreLookahead(int length, RegExpCompiler* compiler, Zone* zone);
+
+ int length() { return length_; }
+ int max_char() { return max_char_; }
+ RegExpCompiler* compiler() { return compiler_; }
+
+ int Count(int map_number) { return bitmaps_->at(map_number)->map_count(); }
+
+ BoyerMoorePositionInfo* at(int i) { return bitmaps_->at(i); }
+
+ void Set(int map_number, int character) {
+ if (character > max_char_) return;
+ BoyerMoorePositionInfo* info = bitmaps_->at(map_number);
+ info->Set(character);
+ }
+
+ void SetInterval(int map_number, const Interval& interval) {
+ if (interval.from() > max_char_) return;
+ BoyerMoorePositionInfo* info = bitmaps_->at(map_number);
+ if (interval.to() > max_char_) {
+ info->SetInterval(Interval(interval.from(), max_char_));
+ } else {
+ info->SetInterval(interval);
+ }
+ }
+
+ void SetAll(int map_number) { bitmaps_->at(map_number)->SetAll(); }
+
+ void SetRest(int from_map) {
+ for (int i = from_map; i < length_; i++) SetAll(i);
+ }
+ void EmitSkipInstructions(RegExpMacroAssembler* masm);
+
+ private:
+ // This is the value obtained by EatsAtLeast. If we do not have at least this
+ // many characters left in the sample string then the match is bound to fail.
+ // Therefore it is OK to read a character this far ahead of the current match
+ // point.
+ int length_;
+ RegExpCompiler* compiler_;
+ // 0xff for Latin1, 0xffff for UTF-16.
+ int max_char_;
+ ZoneList<BoyerMoorePositionInfo*>* bitmaps_;
+
+ int GetSkipTable(int min_lookahead, int max_lookahead,
+ Handle<ByteArray> boolean_skip_table);
+ bool FindWorthwhileInterval(int* from, int* to);
+ int FindBestInterval(int max_number_of_chars, int old_biggest_points,
+ int* from, int* to);
+};
+
+// There are many ways to generate code for a node. This class encapsulates
+// the current way we should be generating. In other words it encapsulates
+// the current state of the code generator. The effect of this is that we
+// generate code for paths that the matcher can take through the regular
+// expression. A given node in the regexp can be code-generated several times
+// as it can be part of several traces. For example for the regexp:
+// /foo(bar|ip)baz/ the code to match baz will be generated twice, once as part
+// of the foo-bar-baz trace and once as part of the foo-ip-baz trace. The code
+// to match foo is generated only once (the traces have a common prefix). The
+// code to store the capture is deferred and generated (twice) after the places
+// where baz has been matched.
+class Trace {
+ public:
+ // A value for a property that is either known to be true, know to be false,
+ // or not known.
+ enum TriBool { UNKNOWN = -1, FALSE_VALUE = 0, TRUE_VALUE = 1 };
+
+ class DeferredAction {
+ public:
+ DeferredAction(ActionNode::ActionType action_type, int reg)
+ : action_type_(action_type), reg_(reg), next_(nullptr) {}
+ DeferredAction* next() { return next_; }
+ bool Mentions(int reg);
+ int reg() { return reg_; }
+ ActionNode::ActionType action_type() { return action_type_; }
+
+ private:
+ ActionNode::ActionType action_type_;
+ int reg_;
+ DeferredAction* next_;
+ friend class Trace;
+ };
+
+ class DeferredCapture : public DeferredAction {
+ public:
+ DeferredCapture(int reg, bool is_capture, Trace* trace)
+ : DeferredAction(ActionNode::STORE_POSITION, reg),
+ cp_offset_(trace->cp_offset()),
+ is_capture_(is_capture) {}
+ int cp_offset() { return cp_offset_; }
+ bool is_capture() { return is_capture_; }
+
+ private:
+ int cp_offset_;
+ bool is_capture_;
+ void set_cp_offset(int cp_offset) { cp_offset_ = cp_offset; }
+ };
+
+ class DeferredSetRegister : public DeferredAction {
+ public:
+ DeferredSetRegister(int reg, int value)
+ : DeferredAction(ActionNode::SET_REGISTER, reg), value_(value) {}
+ int value() { return value_; }
+
+ private:
+ int value_;
+ };
+
+ class DeferredClearCaptures : public DeferredAction {
+ public:
+ explicit DeferredClearCaptures(Interval range)
+ : DeferredAction(ActionNode::CLEAR_CAPTURES, -1), range_(range) {}
+ Interval range() { return range_; }
+
+ private:
+ Interval range_;
+ };
+
+ class DeferredIncrementRegister : public DeferredAction {
+ public:
+ explicit DeferredIncrementRegister(int reg)
+ : DeferredAction(ActionNode::INCREMENT_REGISTER, reg) {}
+ };
+
+ Trace()
+ : cp_offset_(0),
+ actions_(nullptr),
+ backtrack_(nullptr),
+ stop_node_(nullptr),
+ loop_label_(nullptr),
+ characters_preloaded_(0),
+ bound_checked_up_to_(0),
+ flush_budget_(100),
+ at_start_(UNKNOWN) {}
+
+ // End the trace. This involves flushing the deferred actions in the trace
+ // and pushing a backtrack location onto the backtrack stack. Once this is
+ // done we can start a new trace or go to one that has already been
+ // generated.
+ void Flush(RegExpCompiler* compiler, RegExpNode* successor);
+ int cp_offset() { return cp_offset_; }
+ DeferredAction* actions() { return actions_; }
+ // A trivial trace is one that has no deferred actions or other state that
+ // affects the assumptions used when generating code. There is no recorded
+ // backtrack location in a trivial trace, so with a trivial trace we will
+ // generate code that, on a failure to match, gets the backtrack location
+ // from the backtrack stack rather than using a direct jump instruction. We
+ // always start code generation with a trivial trace and non-trivial traces
+ // are created as we emit code for nodes or add to the list of deferred
+ // actions in the trace. The location of the code generated for a node using
+ // a trivial trace is recorded in a label in the node so that gotos can be
+ // generated to that code.
+ bool is_trivial() {
+ return backtrack_ == nullptr && actions_ == nullptr && cp_offset_ == 0 &&
+ characters_preloaded_ == 0 && bound_checked_up_to_ == 0 &&
+ quick_check_performed_.characters() == 0 && at_start_ == UNKNOWN;
+ }
+ TriBool at_start() { return at_start_; }
+ void set_at_start(TriBool at_start) { at_start_ = at_start; }
+ Label* backtrack() { return backtrack_; }
+ Label* loop_label() { return loop_label_; }
+ RegExpNode* stop_node() { return stop_node_; }
+ int characters_preloaded() { return characters_preloaded_; }
+ int bound_checked_up_to() { return bound_checked_up_to_; }
+ int flush_budget() { return flush_budget_; }
+ QuickCheckDetails* quick_check_performed() { return &quick_check_performed_; }
+ bool mentions_reg(int reg);
+ // Returns true if a deferred position store exists to the specified
+ // register and stores the offset in the out-parameter. Otherwise
+ // returns false.
+ bool GetStoredPosition(int reg, int* cp_offset);
+ // These set methods and AdvanceCurrentPositionInTrace should be used only on
+ // new traces - the intention is that traces are immutable after creation.
+ void add_action(DeferredAction* new_action) {
+ DCHECK(new_action->next_ == nullptr);
+ new_action->next_ = actions_;
+ actions_ = new_action;
+ }
+ void set_backtrack(Label* backtrack) { backtrack_ = backtrack; }
+ void set_stop_node(RegExpNode* node) { stop_node_ = node; }
+ void set_loop_label(Label* label) { loop_label_ = label; }
+ void set_characters_preloaded(int count) { characters_preloaded_ = count; }
+ void set_bound_checked_up_to(int to) { bound_checked_up_to_ = to; }
+ void set_flush_budget(int to) { flush_budget_ = to; }
+ void set_quick_check_performed(QuickCheckDetails* d) {
+ quick_check_performed_ = *d;
+ }
+ void InvalidateCurrentCharacter();
+ void AdvanceCurrentPositionInTrace(int by, RegExpCompiler* compiler);
+
+ private:
+ int FindAffectedRegisters(DynamicBitSet* affected_registers, Zone* zone);
+ void PerformDeferredActions(RegExpMacroAssembler* macro, int max_register,
+ const DynamicBitSet& affected_registers,
+ DynamicBitSet* registers_to_pop,
+ DynamicBitSet* registers_to_clear, Zone* zone);
+ void RestoreAffectedRegisters(RegExpMacroAssembler* macro, int max_register,
+ const DynamicBitSet& registers_to_pop,
+ const DynamicBitSet& registers_to_clear);
+ int cp_offset_;
+ DeferredAction* actions_;
+ Label* backtrack_;
+ RegExpNode* stop_node_;
+ Label* loop_label_;
+ int characters_preloaded_;
+ int bound_checked_up_to_;
+ QuickCheckDetails quick_check_performed_;
+ int flush_budget_;
+ TriBool at_start_;
+};
+
+class GreedyLoopState {
+ public:
+ explicit GreedyLoopState(bool not_at_start);
+
+ Label* label() { return &label_; }
+ Trace* counter_backtrack_trace() { return &counter_backtrack_trace_; }
+
+ private:
+ Label label_;
+ Trace counter_backtrack_trace_;
+};
+
+struct PreloadState {
+ static const int kEatsAtLeastNotYetInitialized = -1;
+ bool preload_is_current_;
+ bool preload_has_checked_bounds_;
+ int preload_characters_;
+ int eats_at_least_;
+ void init() { eats_at_least_ = kEatsAtLeastNotYetInitialized; }
+};
+
+// Assertion propagation moves information about assertions such as
+// \b to the affected nodes. For instance, in /.\b./ information must
+// be propagated to the first '.' that whatever follows needs to know
+// if it matched a word or a non-word, and to the second '.' that it
+// has to check if it succeeds a word or non-word. In this case the
+// result will be something like:
+//
+// +-------+ +------------+
+// | . | | . |
+// +-------+ ---> +------------+
+// | word? | | check word |
+// +-------+ +------------+
+class Analysis : public NodeVisitor {
+ public:
+ Analysis(Isolate* isolate, bool is_one_byte)
+ : isolate_(isolate), is_one_byte_(is_one_byte), error_message_(nullptr) {}
+ void EnsureAnalyzed(RegExpNode* node);
+
+#define DECLARE_VISIT(Type) void Visit##Type(Type##Node* that) override;
+ FOR_EACH_NODE_TYPE(DECLARE_VISIT)
+#undef DECLARE_VISIT
+ void VisitLoopChoice(LoopChoiceNode* that) override;
+
+ bool has_failed() { return error_message_ != nullptr; }
+ const char* error_message() {
+ DCHECK(error_message_ != nullptr);
+ return error_message_;
+ }
+ void fail(const char* error_message) { error_message_ = error_message; }
+
+ Isolate* isolate() const { return isolate_; }
+
+ private:
+ Isolate* isolate_;
+ bool is_one_byte_;
+ const char* error_message_;
+
+ DISALLOW_IMPLICIT_CONSTRUCTORS(Analysis);
+};
+
+class FrequencyCollator {
+ public:
+ FrequencyCollator() : total_samples_(0) {
+ for (int i = 0; i < RegExpMacroAssembler::kTableSize; i++) {
+ frequencies_[i] = CharacterFrequency(i);
+ }
+ }
+
+ void CountCharacter(int character) {
+ int index = (character & RegExpMacroAssembler::kTableMask);
+ frequencies_[index].Increment();
+ total_samples_++;
+ }
+
+ // Does not measure in percent, but rather per-128 (the table size from the
+ // regexp macro assembler).
+ int Frequency(int in_character) {
+ DCHECK((in_character & RegExpMacroAssembler::kTableMask) == in_character);
+ if (total_samples_ < 1) return 1; // Division by zero.
+ int freq_in_per128 =
+ (frequencies_[in_character].counter() * 128) / total_samples_;
+ return freq_in_per128;
+ }
+
+ private:
+ class CharacterFrequency {
+ public:
+ CharacterFrequency() : counter_(0), character_(-1) {}
+ explicit CharacterFrequency(int character)
+ : counter_(0), character_(character) {}
+
+ void Increment() { counter_++; }
+ int counter() { return counter_; }
+ int character() { return character_; }
+
+ private:
+ int counter_;
+ int character_;
+ };
+
+ private:
+ CharacterFrequency frequencies_[RegExpMacroAssembler::kTableSize];
+ int total_samples_;
+};
+
+class RegExpCompiler {
+ public:
+ RegExpCompiler(Isolate* isolate, Zone* zone, int capture_count,
+ bool is_one_byte);
+
+ int AllocateRegister() {
+ if (next_register_ >= RegExpMacroAssembler::kMaxRegister) {
+ reg_exp_too_big_ = true;
+ return next_register_;
+ }
+ 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_;
+ }
+
+ struct CompilationResult final {
+ explicit CompilationResult(const char* error_message)
+ : error_message(error_message) {}
+ CompilationResult(Object code, int registers)
+ : code(code), num_registers(registers) {}
+
+ static CompilationResult RegExpTooBig() {
+ return CompilationResult("RegExp too big");
+ }
+
+ bool Succeeded() const { return error_message == nullptr; }
+
+ const char* const error_message = nullptr;
+ Object code;
+ int num_registers = 0;
+ };
+
+ CompilationResult Assemble(Isolate* isolate, RegExpMacroAssembler* assembler,
+ RegExpNode* start, int capture_count,
+ Handle<String> pattern);
+
+ // If the regexp matching starts within a surrogate pair, step back to the
+ // lead surrogate and start matching from there.
+ static RegExpNode* OptionallyStepBackToLeadSurrogate(RegExpCompiler* compiler,
+ RegExpNode* on_success,
+ JSRegExp::Flags flags);
+
+ inline void AddWork(RegExpNode* node) {
+ if (!node->on_work_list() && !node->label()->is_bound()) {
+ node->set_on_work_list(true);
+ work_list_->push_back(node);
+ }
+ }
+
+ static const int kImplementationOffset = 0;
+ static const int kNumberOfRegistersOffset = 0;
+ static const int kCodeOffset = 1;
+
+ RegExpMacroAssembler* macro_assembler() { return macro_assembler_; }
+ EndNode* accept() { return accept_; }
+
+ static const int kMaxRecursion = 100;
+ inline int recursion_depth() { return recursion_depth_; }
+ inline void IncrementRecursionDepth() { recursion_depth_++; }
+ inline void DecrementRecursionDepth() { recursion_depth_--; }
+
+ void SetRegExpTooBig() { reg_exp_too_big_ = true; }
+
+ inline bool one_byte() { return one_byte_; }
+ inline bool optimize() { return optimize_; }
+ inline void set_optimize(bool value) { optimize_ = value; }
+ inline bool limiting_recursion() { return limiting_recursion_; }
+ inline void set_limiting_recursion(bool value) {
+ limiting_recursion_ = value;
+ }
+ bool read_backward() { return read_backward_; }
+ void set_read_backward(bool value) { read_backward_ = value; }
+ FrequencyCollator* frequency_collator() { return &frequency_collator_; }
+
+ int current_expansion_factor() { return current_expansion_factor_; }
+ void set_current_expansion_factor(int value) {
+ current_expansion_factor_ = value;
+ }
+
+ Isolate* isolate() const { return isolate_; }
+ Zone* zone() const { return zone_; }
+
+ static const int kNoRegister = -1;
+
+ private:
+ EndNode* accept_;
+ int next_register_;
+ int unicode_lookaround_stack_register_;
+ int unicode_lookaround_position_register_;
+ std::vector<RegExpNode*>* work_list_;
+ int recursion_depth_;
+ RegExpMacroAssembler* macro_assembler_;
+ bool one_byte_;
+ bool reg_exp_too_big_;
+ bool limiting_recursion_;
+ bool optimize_;
+ bool read_backward_;
+ int current_expansion_factor_;
+ FrequencyCollator frequency_collator_;
+ Isolate* isolate_;
+ Zone* zone_;
+};
+
+// Categorizes character ranges into BMP, non-BMP, lead, and trail surrogates.
+class UnicodeRangeSplitter {
+ public:
+ V8_EXPORT_PRIVATE UnicodeRangeSplitter(ZoneList<CharacterRange>* base);
+
+ static constexpr int kInitialSize = 8;
+ using CharacterRangeVector = base::SmallVector<CharacterRange, kInitialSize>;
+
+ const CharacterRangeVector* bmp() const { return &bmp_; }
+ const CharacterRangeVector* lead_surrogates() const {
+ return &lead_surrogates_;
+ }
+ const CharacterRangeVector* trail_surrogates() const {
+ return &trail_surrogates_;
+ }
+ const CharacterRangeVector* non_bmp() const { return &non_bmp_; }
+
+ private:
+ void AddRange(CharacterRange range);
+
+ CharacterRangeVector bmp_;
+ CharacterRangeVector lead_surrogates_;
+ CharacterRangeVector trail_surrogates_;
+ CharacterRangeVector non_bmp_;
+};
+
+// We need to check for the following characters: 0x39C 0x3BC 0x178.
+// TODO(jgruber): Move to CharacterRange.
+bool RangeContainsLatin1Equivalents(CharacterRange range);
+
+} // namespace internal
+} // namespace v8
+
+#endif // V8_REGEXP_REGEXP_COMPILER_H_
diff --git a/deps/v8/src/regexp/regexp-dotprinter.cc b/deps/v8/src/regexp/regexp-dotprinter.cc
new file mode 100644
index 0000000000..a6d72aaf5b
--- /dev/null
+++ b/deps/v8/src/regexp/regexp-dotprinter.cc
@@ -0,0 +1,244 @@
+// Copyright 2019 the V8 project authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+#include "src/regexp/regexp-dotprinter.h"
+
+#include "src/regexp/regexp-compiler.h"
+#include "src/utils/ostreams.h"
+
+namespace v8 {
+namespace internal {
+
+// -------------------------------------------------------------------
+// Dot/dotty output
+
+#ifdef DEBUG
+
+class DotPrinterImpl : public NodeVisitor {
+ public:
+ explicit DotPrinterImpl(std::ostream& os) : os_(os) {}
+ void PrintNode(const char* label, RegExpNode* node);
+ void Visit(RegExpNode* node);
+ void PrintAttributes(RegExpNode* from);
+ void PrintOnFailure(RegExpNode* from, RegExpNode* to);
+#define DECLARE_VISIT(Type) virtual void Visit##Type(Type##Node* that);
+ FOR_EACH_NODE_TYPE(DECLARE_VISIT)
+#undef DECLARE_VISIT
+ private:
+ std::ostream& os_;
+};
+
+void DotPrinterImpl::PrintNode(const char* label, RegExpNode* node) {
+ os_ << "digraph G {\n graph [label=\"";
+ for (int i = 0; label[i]; i++) {
+ switch (label[i]) {
+ case '\\':
+ os_ << "\\\\";
+ break;
+ case '"':
+ os_ << "\"";
+ break;
+ default:
+ os_ << label[i];
+ break;
+ }
+ }
+ os_ << "\"];\n";
+ Visit(node);
+ os_ << "}" << std::endl;
+}
+
+void DotPrinterImpl::Visit(RegExpNode* node) {
+ if (node->info()->visited) return;
+ node->info()->visited = true;
+ node->Accept(this);
+}
+
+void DotPrinterImpl::PrintOnFailure(RegExpNode* from, RegExpNode* on_failure) {
+ os_ << " n" << from << " -> n" << on_failure << " [style=dotted];\n";
+ Visit(on_failure);
+}
+
+class AttributePrinter {
+ public:
+ explicit AttributePrinter(std::ostream& os) // NOLINT
+ : os_(os), first_(true) {}
+ void PrintSeparator() {
+ if (first_) {
+ first_ = false;
+ } else {
+ os_ << "|";
+ }
+ }
+ void PrintBit(const char* name, bool value) {
+ if (!value) return;
+ PrintSeparator();
+ os_ << "{" << name << "}";
+ }
+ void PrintPositive(const char* name, int value) {
+ if (value < 0) return;
+ PrintSeparator();
+ os_ << "{" << name << "|" << value << "}";
+ }
+
+ private:
+ std::ostream& os_;
+ bool first_;
+};
+
+void DotPrinterImpl::PrintAttributes(RegExpNode* that) {
+ os_ << " a" << that << " [shape=Mrecord, color=grey, fontcolor=grey, "
+ << "margin=0.1, fontsize=10, label=\"{";
+ AttributePrinter printer(os_);
+ NodeInfo* info = that->info();
+ printer.PrintBit("NI", info->follows_newline_interest);
+ printer.PrintBit("WI", info->follows_word_interest);
+ printer.PrintBit("SI", info->follows_start_interest);
+ Label* label = that->label();
+ if (label->is_bound()) printer.PrintPositive("@", label->pos());
+ os_ << "}\"];\n"
+ << " a" << that << " -> n" << that
+ << " [style=dashed, color=grey, arrowhead=none];\n";
+}
+
+void DotPrinterImpl::VisitChoice(ChoiceNode* that) {
+ os_ << " n" << that << " [shape=Mrecord, label=\"?\"];\n";
+ for (int i = 0; i < that->alternatives()->length(); i++) {
+ GuardedAlternative alt = that->alternatives()->at(i);
+ os_ << " n" << that << " -> n" << alt.node();
+ }
+ for (int i = 0; i < that->alternatives()->length(); i++) {
+ GuardedAlternative alt = that->alternatives()->at(i);
+ alt.node()->Accept(this);
+ }
+}
+
+void DotPrinterImpl::VisitText(TextNode* that) {
+ Zone* zone = that->zone();
+ os_ << " n" << that << " [label=\"";
+ for (int i = 0; i < that->elements()->length(); i++) {
+ if (i > 0) os_ << " ";
+ TextElement elm = that->elements()->at(i);
+ switch (elm.text_type()) {
+ case TextElement::ATOM: {
+ Vector<const uc16> data = elm.atom()->data();
+ for (int i = 0; i < data.length(); i++) {
+ os_ << static_cast<char>(data[i]);
+ }
+ break;
+ }
+ case TextElement::CHAR_CLASS: {
+ RegExpCharacterClass* node = elm.char_class();
+ os_ << "[";
+ if (node->is_negated()) os_ << "^";
+ for (int j = 0; j < node->ranges(zone)->length(); j++) {
+ CharacterRange range = node->ranges(zone)->at(j);
+ os_ << AsUC16(range.from()) << "-" << AsUC16(range.to());
+ }
+ os_ << "]";
+ break;
+ }
+ default:
+ UNREACHABLE();
+ }
+ }
+ os_ << "\", shape=box, peripheries=2];\n";
+ PrintAttributes(that);
+ os_ << " n" << that << " -> n" << that->on_success() << ";\n";
+ Visit(that->on_success());
+}
+
+void DotPrinterImpl::VisitBackReference(BackReferenceNode* that) {
+ os_ << " n" << that << " [label=\"$" << that->start_register() << "..$"
+ << that->end_register() << "\", shape=doubleoctagon];\n";
+ PrintAttributes(that);
+ os_ << " n" << that << " -> n" << that->on_success() << ";\n";
+ Visit(that->on_success());
+}
+
+void DotPrinterImpl::VisitEnd(EndNode* that) {
+ os_ << " n" << that << " [style=bold, shape=point];\n";
+ PrintAttributes(that);
+}
+
+void DotPrinterImpl::VisitAssertion(AssertionNode* that) {
+ os_ << " n" << that << " [";
+ switch (that->assertion_type()) {
+ case AssertionNode::AT_END:
+ os_ << "label=\"$\", shape=septagon";
+ break;
+ case AssertionNode::AT_START:
+ os_ << "label=\"^\", shape=septagon";
+ break;
+ case AssertionNode::AT_BOUNDARY:
+ os_ << "label=\"\\b\", shape=septagon";
+ break;
+ case AssertionNode::AT_NON_BOUNDARY:
+ os_ << "label=\"\\B\", shape=septagon";
+ break;
+ case AssertionNode::AFTER_NEWLINE:
+ os_ << "label=\"(?<=\\n)\", shape=septagon";
+ break;
+ }
+ os_ << "];\n";
+ PrintAttributes(that);
+ RegExpNode* successor = that->on_success();
+ os_ << " n" << that << " -> n" << successor << ";\n";
+ Visit(successor);
+}
+
+void DotPrinterImpl::VisitAction(ActionNode* that) {
+ os_ << " n" << that << " [";
+ switch (that->action_type_) {
+ case ActionNode::SET_REGISTER:
+ os_ << "label=\"$" << that->data_.u_store_register.reg
+ << ":=" << that->data_.u_store_register.value << "\", shape=octagon";
+ break;
+ case ActionNode::INCREMENT_REGISTER:
+ os_ << "label=\"$" << that->data_.u_increment_register.reg
+ << "++\", shape=octagon";
+ break;
+ case ActionNode::STORE_POSITION:
+ os_ << "label=\"$" << that->data_.u_position_register.reg
+ << ":=$pos\", shape=octagon";
+ break;
+ case ActionNode::BEGIN_SUBMATCH:
+ os_ << "label=\"$" << that->data_.u_submatch.current_position_register
+ << ":=$pos,begin\", shape=septagon";
+ break;
+ case ActionNode::POSITIVE_SUBMATCH_SUCCESS:
+ os_ << "label=\"escape\", shape=septagon";
+ break;
+ case ActionNode::EMPTY_MATCH_CHECK:
+ os_ << "label=\"$" << that->data_.u_empty_match_check.start_register
+ << "=$pos?,$" << that->data_.u_empty_match_check.repetition_register
+ << "<" << that->data_.u_empty_match_check.repetition_limit
+ << "?\", shape=septagon";
+ break;
+ case ActionNode::CLEAR_CAPTURES: {
+ os_ << "label=\"clear $" << that->data_.u_clear_captures.range_from
+ << " to $" << that->data_.u_clear_captures.range_to
+ << "\", shape=septagon";
+ break;
+ }
+ }
+ os_ << "];\n";
+ PrintAttributes(that);
+ RegExpNode* successor = that->on_success();
+ os_ << " n" << that << " -> n" << successor << ";\n";
+ Visit(successor);
+}
+
+#endif // DEBUG
+
+void DotPrinter::DotPrint(const char* label, RegExpNode* node) {
+#ifdef DEBUG
+ StdoutStream os;
+ DotPrinterImpl printer(os);
+ printer.PrintNode(label, node);
+#endif // DEBUG
+}
+
+} // namespace internal
+} // namespace v8
diff --git a/deps/v8/src/regexp/regexp-dotprinter.h b/deps/v8/src/regexp/regexp-dotprinter.h
new file mode 100644
index 0000000000..d9c75fc1f2
--- /dev/null
+++ b/deps/v8/src/regexp/regexp-dotprinter.h
@@ -0,0 +1,23 @@
+// Copyright 2019 the V8 project authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+#ifndef V8_REGEXP_REGEXP_DOTPRINTER_H_
+#define V8_REGEXP_REGEXP_DOTPRINTER_H_
+
+#include "src/common/globals.h"
+
+namespace v8 {
+namespace internal {
+
+class RegExpNode;
+
+class DotPrinter final : public AllStatic {
+ public:
+ static void DotPrint(const char* label, RegExpNode* node);
+};
+
+} // namespace internal
+} // namespace v8
+
+#endif // V8_REGEXP_REGEXP_DOTPRINTER_H_
diff --git a/deps/v8/src/regexp/interpreter-irregexp.cc b/deps/v8/src/regexp/regexp-interpreter.cc
index 04bb63ee7a..881758861c 100644
--- a/deps/v8/src/regexp/interpreter-irregexp.cc
+++ b/deps/v8/src/regexp/regexp-interpreter.cc
@@ -4,13 +4,14 @@
// A simple interpreter for the Irregexp byte code.
-#include "src/regexp/interpreter-irregexp.h"
+#include "src/regexp/regexp-interpreter.h"
#include "src/ast/ast.h"
+#include "src/base/small-vector.h"
#include "src/objects/objects-inl.h"
-#include "src/regexp/bytecodes-irregexp.h"
-#include "src/regexp/jsregexp.h"
+#include "src/regexp/regexp-bytecodes.h"
#include "src/regexp/regexp-macro-assembler.h"
+#include "src/regexp/regexp.h"
#include "src/strings/unicode.h"
#include "src/utils/utils.h"
@@ -33,7 +34,6 @@ static bool BackRefMatchesNoCase(Isolate* isolate, int from, int current,
offset_a, offset_b, length, unicode ? nullptr : isolate) == 1;
}
-
static bool BackRefMatchesNoCase(Isolate* isolate, int from, int current,
int len, Vector<const uint8_t> subject,
bool unicode) {
@@ -55,28 +55,19 @@ static bool BackRefMatchesNoCase(Isolate* isolate, int from, int current,
return true;
}
-
#ifdef DEBUG
-static void TraceInterpreter(const byte* code_base,
- const byte* pc,
- int stack_depth,
- int current_position,
- uint32_t current_char,
- int bytecode_length,
+static void TraceInterpreter(const byte* code_base, const byte* pc,
+ int stack_depth, int current_position,
+ uint32_t current_char, int bytecode_length,
const char* bytecode_name) {
if (FLAG_trace_regexp_bytecodes) {
bool printable = (current_char < 127 && current_char >= 32);
const char* format =
- printable ?
- "pc = %02x, sp = %d, curpos = %d, curchar = %08x (%c), bc = %s" :
- "pc = %02x, sp = %d, curpos = %d, curchar = %08x .%c., bc = %s";
- PrintF(format,
- pc - code_base,
- stack_depth,
- current_position,
- current_char,
- printable ? current_char : '.',
- bytecode_name);
+ printable
+ ? "pc = %02x, sp = %d, curpos = %d, curchar = %08x (%c), bc = %s"
+ : "pc = %02x, sp = %d, curpos = %d, curchar = %08x .%c., bc = %s";
+ PrintF(format, pc - code_base, stack_depth, current_position, current_char,
+ printable ? current_char : '.', bytecode_name);
for (int i = 0; i < bytecode_length; i++) {
printf(", %02x", pc[i]);
}
@@ -93,54 +84,57 @@ static void TraceInterpreter(const byte* code_base,
}
}
-
-#define BYTECODE(name) \
- case BC_##name: \
- TraceInterpreter(code_base, \
- pc, \
- static_cast<int>(backtrack_sp - backtrack_stack_base), \
- current, \
- current_char, \
- BC_##name##_LENGTH, \
- #name);
+#define BYTECODE(name) \
+ case BC_##name: \
+ TraceInterpreter(code_base, pc, backtrack_stack.sp(), current, \
+ current_char, BC_##name##_LENGTH, #name);
#else
-#define BYTECODE(name) \
- case BC_##name:
+#define BYTECODE(name) case BC_##name:
#endif
-
static int32_t Load32Aligned(const byte* pc) {
DCHECK_EQ(0, reinterpret_cast<intptr_t>(pc) & 3);
- return *reinterpret_cast<const int32_t *>(pc);
+ return *reinterpret_cast<const int32_t*>(pc);
}
-
static int32_t Load16Aligned(const byte* pc) {
DCHECK_EQ(0, reinterpret_cast<intptr_t>(pc) & 1);
- return *reinterpret_cast<const uint16_t *>(pc);
+ return *reinterpret_cast<const uint16_t*>(pc);
}
-
// A simple abstraction over the backtracking stack used by the interpreter.
-// This backtracking stack does not grow automatically, but it ensures that the
-// the memory held by the stack is released or remembered in a cache if the
-// matching terminates.
+//
+// Despite the name 'backtracking' stack, it's actually used as a generic stack
+// that stores both program counters (= offsets into the bytecode) and generic
+// integer values.
class BacktrackStack {
public:
- BacktrackStack() { data_ = NewArray<int>(kBacktrackStackSize); }
+ BacktrackStack() = default;
- ~BacktrackStack() {
- DeleteArray(data_);
+ void push(int v) { data_.emplace_back(v); }
+ int peek() const {
+ DCHECK(!data_.empty());
+ return data_.back();
+ }
+ int pop() {
+ int v = peek();
+ data_.pop_back();
+ return v;
}
- int* data() const { return data_; }
-
- int max_size() const { return kBacktrackStackSize; }
+ // The 'sp' is the index of the first empty element in the stack.
+ int sp() const { return static_cast<int>(data_.size()); }
+ void set_sp(int new_sp) {
+ DCHECK_LE(new_sp, sp());
+ data_.resize_no_init(new_sp);
+ }
private:
- static const int kBacktrackStackSize = 10000;
+ // Semi-arbitrary. Should be large enough for common cases to remain in the
+ // static stack-allocated backing store, but small enough not to waste space.
+ static constexpr int kStaticCapacity = 64;
- int* data_;
+ base::SmallVector<int, kStaticCapacity> data_;
DISALLOW_COPY_AND_ASSIGN(BacktrackStack);
};
@@ -163,28 +157,30 @@ IrregexpInterpreter::Result HandleInterrupts(Isolate* isolate,
StackLimitCheck check(isolate);
if (check.JsHasOverflowed()) {
- // A real stack overflow.
- return StackOverflow(isolate);
+ return StackOverflow(isolate); // A real stack overflow.
}
- const bool was_one_byte =
- String::IsOneByteRepresentationUnderneath(*subject_string);
+ // Handle interrupts if any exist.
+ if (check.InterruptRequested()) {
+ const bool was_one_byte =
+ String::IsOneByteRepresentationUnderneath(*subject_string);
- Object result;
- {
- AllowHeapAllocation yes_gc;
- result = isolate->stack_guard()->HandleInterrupts();
- }
+ Object result;
+ {
+ AllowHeapAllocation yes_gc;
+ result = isolate->stack_guard()->HandleInterrupts();
+ }
- if (result.IsException(isolate)) {
- return IrregexpInterpreter::EXCEPTION;
- }
+ if (result.IsException(isolate)) {
+ return IrregexpInterpreter::EXCEPTION;
+ }
- // If we changed between a LATIN1 and a UC16 string, we need to restart
- // regexp matching with the appropriate template instantiation of RawMatch.
- if (String::IsOneByteRepresentationUnderneath(*subject_string) !=
- was_one_byte) {
- return IrregexpInterpreter::RETRY;
+ // If we changed between a LATIN1 and a UC16 string, we need to restart
+ // regexp matching with the appropriate template instantiation of RawMatch.
+ if (String::IsOneByteRepresentationUnderneath(*subject_string) !=
+ was_one_byte) {
+ return IrregexpInterpreter::RETRY;
+ }
}
return IrregexpInterpreter::SUCCESS;
@@ -221,121 +217,108 @@ IrregexpInterpreter::Result RawMatch(Isolate* isolate,
const byte* pc = code_array->GetDataStartAddress();
const byte* code_base = pc;
- // BacktrackStack ensures that the memory allocated for the backtracking stack
- // is returned to the system or cached if there is no stack being cached at
- // the moment.
BacktrackStack backtrack_stack;
- int* backtrack_stack_base = backtrack_stack.data();
- int* backtrack_sp = backtrack_stack_base;
- int backtrack_stack_space = backtrack_stack.max_size();
+
#ifdef DEBUG
if (FLAG_trace_regexp_bytecodes) {
PrintF("\n\nStart bytecode interpreter\n\n");
}
#endif
while (true) {
- int32_t insn = Load32Aligned(pc);
+ const int32_t insn = Load32Aligned(pc);
switch (insn & BYTECODE_MASK) {
- BYTECODE(BREAK)
- UNREACHABLE();
- BYTECODE(PUSH_CP)
- if (--backtrack_stack_space < 0) {
- return StackOverflow(isolate);
- }
- *backtrack_sp++ = current;
+ BYTECODE(BREAK) { UNREACHABLE(); }
+ BYTECODE(PUSH_CP) {
+ backtrack_stack.push(current);
pc += BC_PUSH_CP_LENGTH;
break;
- BYTECODE(PUSH_BT)
- if (--backtrack_stack_space < 0) {
- return StackOverflow(isolate);
- }
- *backtrack_sp++ = Load32Aligned(pc + 4);
+ }
+ BYTECODE(PUSH_BT) {
+ backtrack_stack.push(Load32Aligned(pc + 4));
pc += BC_PUSH_BT_LENGTH;
break;
- BYTECODE(PUSH_REGISTER)
- if (--backtrack_stack_space < 0) {
- return StackOverflow(isolate);
- }
- *backtrack_sp++ = registers[insn >> BYTECODE_SHIFT];
+ }
+ BYTECODE(PUSH_REGISTER) {
+ backtrack_stack.push(registers[insn >> BYTECODE_SHIFT]);
pc += BC_PUSH_REGISTER_LENGTH;
break;
- BYTECODE(SET_REGISTER)
+ }
+ BYTECODE(SET_REGISTER) {
registers[insn >> BYTECODE_SHIFT] = Load32Aligned(pc + 4);
pc += BC_SET_REGISTER_LENGTH;
break;
- BYTECODE(ADVANCE_REGISTER)
+ }
+ BYTECODE(ADVANCE_REGISTER) {
registers[insn >> BYTECODE_SHIFT] += Load32Aligned(pc + 4);
pc += BC_ADVANCE_REGISTER_LENGTH;
break;
- BYTECODE(SET_REGISTER_TO_CP)
+ }
+ BYTECODE(SET_REGISTER_TO_CP) {
registers[insn >> BYTECODE_SHIFT] = current + Load32Aligned(pc + 4);
pc += BC_SET_REGISTER_TO_CP_LENGTH;
break;
- BYTECODE(SET_CP_TO_REGISTER)
+ }
+ BYTECODE(SET_CP_TO_REGISTER) {
current = registers[insn >> BYTECODE_SHIFT];
pc += BC_SET_CP_TO_REGISTER_LENGTH;
break;
- BYTECODE(SET_REGISTER_TO_SP)
- registers[insn >> BYTECODE_SHIFT] =
- static_cast<int>(backtrack_sp - backtrack_stack_base);
+ }
+ BYTECODE(SET_REGISTER_TO_SP) {
+ registers[insn >> BYTECODE_SHIFT] = backtrack_stack.sp();
pc += BC_SET_REGISTER_TO_SP_LENGTH;
break;
- BYTECODE(SET_SP_TO_REGISTER)
- backtrack_sp = backtrack_stack_base + registers[insn >> BYTECODE_SHIFT];
- backtrack_stack_space = backtrack_stack.max_size() -
- static_cast<int>(backtrack_sp - backtrack_stack_base);
+ }
+ BYTECODE(SET_SP_TO_REGISTER) {
+ backtrack_stack.set_sp(registers[insn >> BYTECODE_SHIFT]);
pc += BC_SET_SP_TO_REGISTER_LENGTH;
break;
- BYTECODE(POP_CP)
- backtrack_stack_space++;
- --backtrack_sp;
- current = *backtrack_sp;
+ }
+ BYTECODE(POP_CP) {
+ current = backtrack_stack.pop();
pc += BC_POP_CP_LENGTH;
break;
- // clang-format off
+ }
BYTECODE(POP_BT) {
- IrregexpInterpreter::Result return_code = HandleInterrupts(
- isolate, subject_string);
+ IrregexpInterpreter::Result return_code =
+ HandleInterrupts(isolate, subject_string);
if (return_code != IrregexpInterpreter::SUCCESS) return return_code;
UpdateCodeAndSubjectReferences(isolate, code_array, subject_string,
- &code_base, &pc, &subject);
+ &code_base, &pc, &subject);
- backtrack_stack_space++;
- --backtrack_sp;
- pc = code_base + *backtrack_sp;
+ pc = code_base + backtrack_stack.pop();
break;
}
- BYTECODE(POP_REGISTER) // clang-format on
- backtrack_stack_space++;
- --backtrack_sp;
- registers[insn >> BYTECODE_SHIFT] = *backtrack_sp;
+ BYTECODE(POP_REGISTER) {
+ registers[insn >> BYTECODE_SHIFT] = backtrack_stack.pop();
pc += BC_POP_REGISTER_LENGTH;
break;
- BYTECODE(FAIL)
- return IrregexpInterpreter::FAILURE;
- BYTECODE(SUCCEED)
- return IrregexpInterpreter::SUCCESS;
- BYTECODE(ADVANCE_CP)
+ }
+ BYTECODE(FAIL) { return IrregexpInterpreter::FAILURE; }
+ BYTECODE(SUCCEED) { return IrregexpInterpreter::SUCCESS; }
+ BYTECODE(ADVANCE_CP) {
current += insn >> BYTECODE_SHIFT;
pc += BC_ADVANCE_CP_LENGTH;
break;
- BYTECODE(GOTO)
+ }
+ BYTECODE(GOTO) {
pc = code_base + Load32Aligned(pc + 4);
break;
- BYTECODE(ADVANCE_CP_AND_GOTO)
+ }
+ BYTECODE(ADVANCE_CP_AND_GOTO) {
current += insn >> BYTECODE_SHIFT;
pc = code_base + Load32Aligned(pc + 4);
break;
- BYTECODE(CHECK_GREEDY)
- if (current == backtrack_sp[-1]) {
- backtrack_sp--;
- backtrack_stack_space++;
+ }
+ BYTECODE(CHECK_GREEDY) {
+ if (current == backtrack_stack.peek()) {
+ backtrack_stack.pop();
pc = code_base + Load32Aligned(pc + 4);
} else {
pc += BC_CHECK_GREEDY_LENGTH;
}
break;
+ }
BYTECODE(LOAD_CURRENT_CHAR) {
int pos = current + (insn >> BYTECODE_SHIFT);
if (pos >= subject.length() || pos < 0) {
@@ -380,10 +363,8 @@ IrregexpInterpreter::Result RawMatch(Isolate* isolate,
Char next1 = subject[pos + 1];
Char next2 = subject[pos + 2];
Char next3 = subject[pos + 3];
- current_char = (subject[pos] |
- (next1 << 8) |
- (next2 << 16) |
- (next3 << 24));
+ current_char =
+ (subject[pos] | (next1 << 8) | (next2 << 16) | (next3 << 24));
pc += BC_LOAD_4_CURRENT_CHARS_LENGTH;
}
break;
@@ -394,10 +375,8 @@ IrregexpInterpreter::Result RawMatch(Isolate* isolate,
Char next1 = subject[pos + 1];
Char next2 = subject[pos + 2];
Char next3 = subject[pos + 3];
- current_char = (subject[pos] |
- (next1 << 8) |
- (next2 << 16) |
- (next3 << 24));
+ current_char =
+ (subject[pos] | (next1 << 8) | (next2 << 16) | (next3 << 24));
pc += BC_LOAD_4_CURRENT_CHARS_UNCHECKED_LENGTH;
break;
}
@@ -533,28 +512,31 @@ IrregexpInterpreter::Result RawMatch(Isolate* isolate,
}
break;
}
- BYTECODE(CHECK_REGISTER_LT)
+ BYTECODE(CHECK_REGISTER_LT) {
if (registers[insn >> BYTECODE_SHIFT] < Load32Aligned(pc + 4)) {
pc = code_base + Load32Aligned(pc + 8);
} else {
pc += BC_CHECK_REGISTER_LT_LENGTH;
}
break;
- BYTECODE(CHECK_REGISTER_GE)
+ }
+ BYTECODE(CHECK_REGISTER_GE) {
if (registers[insn >> BYTECODE_SHIFT] >= Load32Aligned(pc + 4)) {
pc = code_base + Load32Aligned(pc + 8);
} else {
pc += BC_CHECK_REGISTER_GE_LENGTH;
}
break;
- BYTECODE(CHECK_REGISTER_EQ_POS)
+ }
+ BYTECODE(CHECK_REGISTER_EQ_POS) {
if (registers[insn >> BYTECODE_SHIFT] == current) {
pc = code_base + Load32Aligned(pc + 4);
} else {
pc += BC_CHECK_REGISTER_EQ_POS_LENGTH;
}
break;
- BYTECODE(CHECK_NOT_REGS_EQUAL)
+ }
+ BYTECODE(CHECK_NOT_REGS_EQUAL) {
if (registers[insn >> BYTECODE_SHIFT] ==
registers[Load32Aligned(pc + 4)]) {
pc += BC_CHECK_NOT_REGS_EQUAL_LENGTH;
@@ -562,6 +544,7 @@ IrregexpInterpreter::Result RawMatch(Isolate* isolate,
pc = code_base + Load32Aligned(pc + 8);
}
break;
+ }
BYTECODE(CHECK_NOT_BACK_REF) {
int from = registers[insn >> BYTECODE_SHIFT];
int len = registers[(insn >> BYTECODE_SHIFT) + 1] - from;
@@ -628,20 +611,22 @@ IrregexpInterpreter::Result RawMatch(Isolate* isolate,
pc += BC_CHECK_NOT_BACK_REF_NO_CASE_BACKWARD_LENGTH;
break;
}
- BYTECODE(CHECK_AT_START)
+ BYTECODE(CHECK_AT_START) {
if (current == 0) {
pc = code_base + Load32Aligned(pc + 4);
} else {
pc += BC_CHECK_AT_START_LENGTH;
}
break;
- BYTECODE(CHECK_NOT_AT_START)
+ }
+ BYTECODE(CHECK_NOT_AT_START) {
if (current + (insn >> BYTECODE_SHIFT) == 0) {
pc += BC_CHECK_NOT_AT_START_LENGTH;
} else {
pc = code_base + Load32Aligned(pc + 4);
}
break;
+ }
BYTECODE(SET_CURRENT_POSITION_FROM_END) {
int by = static_cast<uint32_t>(insn) >> BYTECODE_SHIFT;
if (subject.length() - current > by) {
@@ -658,6 +643,8 @@ IrregexpInterpreter::Result RawMatch(Isolate* isolate,
}
}
+#undef BYTECODE
+
} // namespace
// static
diff --git a/deps/v8/src/regexp/interpreter-irregexp.h b/deps/v8/src/regexp/regexp-interpreter.h
index a57d40854e..ad27dcd296 100644
--- a/deps/v8/src/regexp/interpreter-irregexp.h
+++ b/deps/v8/src/regexp/regexp-interpreter.h
@@ -4,20 +4,22 @@
// A simple interpreter for the Irregexp byte code.
-#ifndef V8_REGEXP_INTERPRETER_IRREGEXP_H_
-#define V8_REGEXP_INTERPRETER_IRREGEXP_H_
+#ifndef V8_REGEXP_REGEXP_INTERPRETER_H_
+#define V8_REGEXP_REGEXP_INTERPRETER_H_
-#include "src/regexp/jsregexp.h"
+#include "src/regexp/regexp.h"
namespace v8 {
namespace internal {
class V8_EXPORT_PRIVATE IrregexpInterpreter {
public:
- enum Result { RETRY = -2, EXCEPTION = -1, FAILURE = 0, SUCCESS = 1 };
- STATIC_ASSERT(EXCEPTION == static_cast<int>(RegExpImpl::RE_EXCEPTION));
- STATIC_ASSERT(FAILURE == static_cast<int>(RegExpImpl::RE_FAILURE));
- STATIC_ASSERT(SUCCESS == static_cast<int>(RegExpImpl::RE_SUCCESS));
+ enum Result {
+ FAILURE = RegExp::kInternalRegExpFailure,
+ SUCCESS = RegExp::kInternalRegExpSuccess,
+ EXCEPTION = RegExp::kInternalRegExpException,
+ RETRY = RegExp::kInternalRegExpRetry,
+ };
// The caller is responsible for initializing registers before each call.
static Result Match(Isolate* isolate, Handle<ByteArray> code_array,
@@ -28,4 +30,4 @@ class V8_EXPORT_PRIVATE IrregexpInterpreter {
} // namespace internal
} // namespace v8
-#endif // V8_REGEXP_INTERPRETER_IRREGEXP_H_
+#endif // V8_REGEXP_REGEXP_INTERPRETER_H_
diff --git a/deps/v8/src/regexp/regexp-macro-assembler-arch.h b/deps/v8/src/regexp/regexp-macro-assembler-arch.h
new file mode 100644
index 0000000000..2dc6739e42
--- /dev/null
+++ b/deps/v8/src/regexp/regexp-macro-assembler-arch.h
@@ -0,0 +1,30 @@
+// Copyright 2019 the V8 project authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+#ifndef V8_REGEXP_REGEXP_MACRO_ASSEMBLER_ARCH_H_
+#define V8_REGEXP_REGEXP_MACRO_ASSEMBLER_ARCH_H_
+
+#include "src/regexp/regexp-macro-assembler.h"
+
+#if V8_TARGET_ARCH_IA32
+#include "src/regexp/ia32/regexp-macro-assembler-ia32.h"
+#elif V8_TARGET_ARCH_X64
+#include "src/regexp/x64/regexp-macro-assembler-x64.h"
+#elif V8_TARGET_ARCH_ARM64
+#include "src/regexp/arm64/regexp-macro-assembler-arm64.h"
+#elif V8_TARGET_ARCH_ARM
+#include "src/regexp/arm/regexp-macro-assembler-arm.h"
+#elif V8_TARGET_ARCH_PPC
+#include "src/regexp/ppc/regexp-macro-assembler-ppc.h"
+#elif V8_TARGET_ARCH_MIPS
+#include "src/regexp/mips/regexp-macro-assembler-mips.h"
+#elif V8_TARGET_ARCH_MIPS64
+#include "src/regexp/mips64/regexp-macro-assembler-mips64.h"
+#elif V8_TARGET_ARCH_S390
+#include "src/regexp/s390/regexp-macro-assembler-s390.h"
+#else
+#error Unsupported target architecture.
+#endif
+
+#endif // V8_REGEXP_REGEXP_MACRO_ASSEMBLER_ARCH_H_
diff --git a/deps/v8/src/regexp/regexp-macro-assembler.cc b/deps/v8/src/regexp/regexp-macro-assembler.cc
index cfe827ef4e..68fa16db61 100644
--- a/deps/v8/src/regexp/regexp-macro-assembler.cc
+++ b/deps/v8/src/regexp/regexp-macro-assembler.cc
@@ -133,7 +133,8 @@ int NativeRegExpMacroAssembler::CheckStackGuardState(
Isolate* isolate, int start_index, bool is_direct_call,
Address* return_address, Code re_code, Address* subject,
const byte** input_start, const byte** input_end) {
- AllowHeapAllocation allow_allocation;
+ DisallowHeapAllocation no_gc;
+
DCHECK(re_code.raw_instruction_start() <= *return_address);
DCHECK(*return_address <= re_code.raw_instruction_end());
int return_value = 0;
@@ -154,15 +155,15 @@ int NativeRegExpMacroAssembler::CheckStackGuardState(
// forcing the call through the runtime system.
return_value = js_has_overflowed ? EXCEPTION : RETRY;
} else if (js_has_overflowed) {
+ AllowHeapAllocation yes_gc;
isolate->StackOverflow();
return_value = EXCEPTION;
- } else {
+ } else if (check.InterruptRequested()) {
+ AllowHeapAllocation yes_gc;
Object result = isolate->stack_guard()->HandleInterrupts();
if (result.IsException(isolate)) return_value = EXCEPTION;
}
- DisallowHeapAllocation no_gc;
-
if (*code_handle != re_code) { // Return address no longer valid
intptr_t delta = code_handle->address() - re_code.address();
// Overwrite the return address on the stack.
diff --git a/deps/v8/src/regexp/regexp-macro-assembler.h b/deps/v8/src/regexp/regexp-macro-assembler.h
index 8626d1a19e..b55ac13590 100644
--- a/deps/v8/src/regexp/regexp-macro-assembler.h
+++ b/deps/v8/src/regexp/regexp-macro-assembler.h
@@ -7,6 +7,7 @@
#include "src/codegen/label.h"
#include "src/regexp/regexp-ast.h"
+#include "src/regexp/regexp.h"
namespace v8 {
namespace internal {
@@ -206,7 +207,12 @@ class NativeRegExpMacroAssembler: public RegExpMacroAssembler {
// FAILURE: Matching failed.
// SUCCESS: Matching succeeded, and the output array has been filled with
// capture positions.
- enum Result { RETRY = -2, EXCEPTION = -1, FAILURE = 0, SUCCESS = 1 };
+ enum Result {
+ FAILURE = RegExp::kInternalRegExpFailure,
+ SUCCESS = RegExp::kInternalRegExpSuccess,
+ EXCEPTION = RegExp::kInternalRegExpException,
+ RETRY = RegExp::kInternalRegExpRetry,
+ };
NativeRegExpMacroAssembler(Isolate* isolate, Zone* zone);
~NativeRegExpMacroAssembler() override;
diff --git a/deps/v8/src/regexp/regexp-nodes.h b/deps/v8/src/regexp/regexp-nodes.h
new file mode 100644
index 0000000000..4c13b74926
--- /dev/null
+++ b/deps/v8/src/regexp/regexp-nodes.h
@@ -0,0 +1,675 @@
+// Copyright 2019 the V8 project authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+#ifndef V8_REGEXP_REGEXP_NODES_H_
+#define V8_REGEXP_REGEXP_NODES_H_
+
+#include "src/regexp/regexp-macro-assembler.h"
+#include "src/zone/zone.h"
+
+namespace v8 {
+namespace internal {
+
+class AlternativeGenerationList;
+class BoyerMooreLookahead;
+class GreedyLoopState;
+class Label;
+class NodeVisitor;
+class QuickCheckDetails;
+class RegExpCompiler;
+class Trace;
+struct PreloadState;
+
+#define FOR_EACH_NODE_TYPE(VISIT) \
+ VISIT(End) \
+ VISIT(Action) \
+ VISIT(Choice) \
+ VISIT(BackReference) \
+ VISIT(Assertion) \
+ VISIT(Text)
+
+struct NodeInfo final {
+ NodeInfo()
+ : being_analyzed(false),
+ been_analyzed(false),
+ follows_word_interest(false),
+ follows_newline_interest(false),
+ follows_start_interest(false),
+ at_end(false),
+ visited(false),
+ replacement_calculated(false) {}
+
+ // Returns true if the interests and assumptions of this node
+ // matches the given one.
+ bool Matches(NodeInfo* that) {
+ return (at_end == that->at_end) &&
+ (follows_word_interest == that->follows_word_interest) &&
+ (follows_newline_interest == that->follows_newline_interest) &&
+ (follows_start_interest == that->follows_start_interest);
+ }
+
+ // Updates the interests of this node given the interests of the
+ // node preceding it.
+ void AddFromPreceding(NodeInfo* that) {
+ at_end |= that->at_end;
+ follows_word_interest |= that->follows_word_interest;
+ follows_newline_interest |= that->follows_newline_interest;
+ follows_start_interest |= that->follows_start_interest;
+ }
+
+ bool HasLookbehind() {
+ return follows_word_interest || follows_newline_interest ||
+ follows_start_interest;
+ }
+
+ // Sets the interests of this node to include the interests of the
+ // following node.
+ void AddFromFollowing(NodeInfo* that) {
+ follows_word_interest |= that->follows_word_interest;
+ follows_newline_interest |= that->follows_newline_interest;
+ follows_start_interest |= that->follows_start_interest;
+ }
+
+ void ResetCompilationState() {
+ being_analyzed = false;
+ been_analyzed = false;
+ }
+
+ bool being_analyzed : 1;
+ bool been_analyzed : 1;
+
+ // These bits are set of this node has to know what the preceding
+ // character was.
+ bool follows_word_interest : 1;
+ bool follows_newline_interest : 1;
+ bool follows_start_interest : 1;
+
+ bool at_end : 1;
+ bool visited : 1;
+ bool replacement_calculated : 1;
+};
+
+class RegExpNode : public ZoneObject {
+ public:
+ explicit RegExpNode(Zone* zone)
+ : replacement_(nullptr),
+ on_work_list_(false),
+ trace_count_(0),
+ zone_(zone) {
+ bm_info_[0] = bm_info_[1] = nullptr;
+ }
+ virtual ~RegExpNode();
+ virtual void Accept(NodeVisitor* visitor) = 0;
+ // Generates a goto to this node or actually generates the code at this point.
+ virtual void Emit(RegExpCompiler* compiler, Trace* trace) = 0;
+ // How many characters must this node consume at a minimum in order to
+ // succeed. If we have found at least 'still_to_find' characters that
+ // must be consumed there is no need to ask any following nodes whether
+ // they are sure to eat any more characters. The not_at_start argument is
+ // used to indicate that we know we are not at the start of the input. In
+ // this case anchored branches will always fail and can be ignored when
+ // determining how many characters are consumed on success.
+ virtual int EatsAtLeast(int still_to_find, int budget, bool not_at_start) = 0;
+ // Emits some quick code that checks whether the preloaded characters match.
+ // Falls through on certain failure, jumps to the label on possible success.
+ // If the node cannot make a quick check it does nothing and returns false.
+ bool EmitQuickCheck(RegExpCompiler* compiler, Trace* bounds_check_trace,
+ Trace* trace, bool preload_has_checked_bounds,
+ Label* on_possible_success,
+ QuickCheckDetails* details_return,
+ bool fall_through_on_failure);
+ // For a given number of characters this returns a mask and a value. The
+ // next n characters are anded with the mask and compared with the value.
+ // A comparison failure indicates the node cannot match the next n characters.
+ // A comparison success indicates the node may match.
+ virtual void GetQuickCheckDetails(QuickCheckDetails* details,
+ RegExpCompiler* compiler,
+ int characters_filled_in,
+ bool not_at_start) = 0;
+ static const int kNodeIsTooComplexForGreedyLoops = kMinInt;
+ virtual int GreedyLoopTextLength() { return kNodeIsTooComplexForGreedyLoops; }
+ // Only returns the successor for a text node of length 1 that matches any
+ // character and that has no guards on it.
+ virtual RegExpNode* GetSuccessorOfOmnivorousTextNode(
+ RegExpCompiler* compiler) {
+ return nullptr;
+ }
+
+ // Collects information on the possible code units (mod 128) that can match if
+ // we look forward. This is used for a Boyer-Moore-like string searching
+ // implementation. TODO(erikcorry): This should share more code with
+ // EatsAtLeast, GetQuickCheckDetails. The budget argument is used to limit
+ // the number of nodes we are willing to look at in order to create this data.
+ static const int kRecursionBudget = 200;
+ bool KeepRecursing(RegExpCompiler* compiler);
+ virtual void FillInBMInfo(Isolate* isolate, int offset, int budget,
+ BoyerMooreLookahead* bm, bool not_at_start) {
+ UNREACHABLE();
+ }
+
+ // If we know that the input is one-byte then there are some nodes that can
+ // never match. This method returns a node that can be substituted for
+ // itself, or nullptr if the node can never match.
+ virtual RegExpNode* FilterOneByte(int depth) { return this; }
+ // Helper for FilterOneByte.
+ RegExpNode* replacement() {
+ DCHECK(info()->replacement_calculated);
+ return replacement_;
+ }
+ RegExpNode* set_replacement(RegExpNode* replacement) {
+ info()->replacement_calculated = true;
+ replacement_ = replacement;
+ return replacement; // For convenience.
+ }
+
+ // We want to avoid recalculating the lookahead info, so we store it on the
+ // node. Only info that is for this node is stored. We can tell that the
+ // info is for this node when offset == 0, so the information is calculated
+ // relative to this node.
+ void SaveBMInfo(BoyerMooreLookahead* bm, bool not_at_start, int offset) {
+ if (offset == 0) set_bm_info(not_at_start, bm);
+ }
+
+ Label* label() { return &label_; }
+ // If non-generic code is generated for a node (i.e. the node is not at the
+ // start of the trace) then it cannot be reused. This variable sets a limit
+ // on how often we allow that to happen before we insist on starting a new
+ // trace and generating generic code for a node that can be reused by flushing
+ // the deferred actions in the current trace and generating a goto.
+ static const int kMaxCopiesCodeGenerated = 10;
+
+ bool on_work_list() { return on_work_list_; }
+ void set_on_work_list(bool value) { on_work_list_ = value; }
+
+ NodeInfo* info() { return &info_; }
+
+ BoyerMooreLookahead* bm_info(bool not_at_start) {
+ return bm_info_[not_at_start ? 1 : 0];
+ }
+
+ Zone* zone() const { return zone_; }
+
+ protected:
+ enum LimitResult { DONE, CONTINUE };
+ RegExpNode* replacement_;
+
+ LimitResult LimitVersions(RegExpCompiler* compiler, Trace* trace);
+
+ void set_bm_info(bool not_at_start, BoyerMooreLookahead* bm) {
+ bm_info_[not_at_start ? 1 : 0] = bm;
+ }
+
+ private:
+ static const int kFirstCharBudget = 10;
+ Label label_;
+ bool on_work_list_;
+ NodeInfo info_;
+ // This variable keeps track of how many times code has been generated for
+ // this node (in different traces). We don't keep track of where the
+ // generated code is located unless the code is generated at the start of
+ // a trace, in which case it is generic and can be reused by flushing the
+ // deferred operations in the current trace and generating a goto.
+ int trace_count_;
+ BoyerMooreLookahead* bm_info_[2];
+
+ Zone* zone_;
+};
+
+class SeqRegExpNode : public RegExpNode {
+ public:
+ explicit SeqRegExpNode(RegExpNode* on_success)
+ : RegExpNode(on_success->zone()), on_success_(on_success) {}
+ RegExpNode* on_success() { return on_success_; }
+ void set_on_success(RegExpNode* node) { on_success_ = node; }
+ RegExpNode* FilterOneByte(int depth) override;
+ void FillInBMInfo(Isolate* isolate, int offset, int budget,
+ BoyerMooreLookahead* bm, bool not_at_start) override {
+ on_success_->FillInBMInfo(isolate, offset, budget - 1, bm, not_at_start);
+ if (offset == 0) set_bm_info(not_at_start, bm);
+ }
+
+ protected:
+ RegExpNode* FilterSuccessor(int depth);
+
+ private:
+ RegExpNode* on_success_;
+};
+
+class ActionNode : public SeqRegExpNode {
+ public:
+ enum ActionType {
+ SET_REGISTER,
+ INCREMENT_REGISTER,
+ STORE_POSITION,
+ BEGIN_SUBMATCH,
+ POSITIVE_SUBMATCH_SUCCESS,
+ EMPTY_MATCH_CHECK,
+ CLEAR_CAPTURES
+ };
+ static ActionNode* SetRegister(int reg, int val, RegExpNode* on_success);
+ static ActionNode* IncrementRegister(int reg, RegExpNode* on_success);
+ static ActionNode* StorePosition(int reg, bool is_capture,
+ RegExpNode* on_success);
+ static ActionNode* ClearCaptures(Interval range, RegExpNode* on_success);
+ static ActionNode* BeginSubmatch(int stack_pointer_reg, int position_reg,
+ RegExpNode* on_success);
+ static ActionNode* PositiveSubmatchSuccess(int stack_pointer_reg,
+ int restore_reg,
+ int clear_capture_count,
+ int clear_capture_from,
+ RegExpNode* on_success);
+ static ActionNode* EmptyMatchCheck(int start_register,
+ int repetition_register,
+ int repetition_limit,
+ RegExpNode* on_success);
+ void Accept(NodeVisitor* visitor) override;
+ void Emit(RegExpCompiler* compiler, Trace* trace) override;
+ int EatsAtLeast(int still_to_find, int budget, bool not_at_start) override;
+ void GetQuickCheckDetails(QuickCheckDetails* details,
+ RegExpCompiler* compiler, int filled_in,
+ bool not_at_start) override {
+ return on_success()->GetQuickCheckDetails(details, compiler, filled_in,
+ not_at_start);
+ }
+ void FillInBMInfo(Isolate* isolate, int offset, int budget,
+ BoyerMooreLookahead* bm, bool not_at_start) override;
+ ActionType action_type() { return action_type_; }
+ // TODO(erikcorry): We should allow some action nodes in greedy loops.
+ int GreedyLoopTextLength() override {
+ return kNodeIsTooComplexForGreedyLoops;
+ }
+
+ private:
+ union {
+ struct {
+ int reg;
+ int value;
+ } u_store_register;
+ struct {
+ int reg;
+ } u_increment_register;
+ struct {
+ int reg;
+ bool is_capture;
+ } u_position_register;
+ struct {
+ int stack_pointer_register;
+ int current_position_register;
+ int clear_register_count;
+ int clear_register_from;
+ } u_submatch;
+ struct {
+ int start_register;
+ int repetition_register;
+ int repetition_limit;
+ } u_empty_match_check;
+ struct {
+ int range_from;
+ int range_to;
+ } u_clear_captures;
+ } data_;
+ ActionNode(ActionType action_type, RegExpNode* on_success)
+ : SeqRegExpNode(on_success), action_type_(action_type) {}
+ ActionType action_type_;
+ friend class DotPrinterImpl;
+};
+
+class TextNode : public SeqRegExpNode {
+ public:
+ TextNode(ZoneList<TextElement>* elms, bool read_backward,
+ RegExpNode* on_success)
+ : SeqRegExpNode(on_success), elms_(elms), read_backward_(read_backward) {}
+ TextNode(RegExpCharacterClass* that, bool read_backward,
+ RegExpNode* on_success)
+ : SeqRegExpNode(on_success),
+ elms_(new (zone()) ZoneList<TextElement>(1, zone())),
+ read_backward_(read_backward) {
+ elms_->Add(TextElement::CharClass(that), zone());
+ }
+ // Create TextNode for a single character class for the given ranges.
+ static TextNode* CreateForCharacterRanges(Zone* zone,
+ ZoneList<CharacterRange>* ranges,
+ bool read_backward,
+ RegExpNode* on_success,
+ JSRegExp::Flags flags);
+ // Create TextNode for a surrogate pair with a range given for the
+ // lead and the trail surrogate each.
+ static TextNode* CreateForSurrogatePair(Zone* zone, CharacterRange lead,
+ CharacterRange trail,
+ bool read_backward,
+ RegExpNode* on_success,
+ JSRegExp::Flags flags);
+ void Accept(NodeVisitor* visitor) override;
+ void Emit(RegExpCompiler* compiler, Trace* trace) override;
+ int EatsAtLeast(int still_to_find, int budget, bool not_at_start) override;
+ void GetQuickCheckDetails(QuickCheckDetails* details,
+ RegExpCompiler* compiler, int characters_filled_in,
+ bool not_at_start) override;
+ ZoneList<TextElement>* elements() { return elms_; }
+ bool read_backward() { return read_backward_; }
+ void MakeCaseIndependent(Isolate* isolate, bool is_one_byte);
+ int GreedyLoopTextLength() override;
+ RegExpNode* GetSuccessorOfOmnivorousTextNode(
+ RegExpCompiler* compiler) override;
+ void FillInBMInfo(Isolate* isolate, int offset, int budget,
+ BoyerMooreLookahead* bm, bool not_at_start) override;
+ void CalculateOffsets();
+ RegExpNode* FilterOneByte(int depth) override;
+
+ private:
+ enum TextEmitPassType {
+ NON_LATIN1_MATCH, // Check for characters that can't match.
+ SIMPLE_CHARACTER_MATCH, // Case-dependent single character check.
+ NON_LETTER_CHARACTER_MATCH, // Check characters that have no case equivs.
+ CASE_CHARACTER_MATCH, // Case-independent single character check.
+ CHARACTER_CLASS_MATCH // Character class.
+ };
+ static bool SkipPass(TextEmitPassType pass, bool ignore_case);
+ static const int kFirstRealPass = SIMPLE_CHARACTER_MATCH;
+ static const int kLastPass = CHARACTER_CLASS_MATCH;
+ void TextEmitPass(RegExpCompiler* compiler, TextEmitPassType pass,
+ bool preloaded, Trace* trace, bool first_element_checked,
+ int* checked_up_to);
+ int Length();
+ ZoneList<TextElement>* elms_;
+ bool read_backward_;
+};
+
+class AssertionNode : public SeqRegExpNode {
+ public:
+ enum AssertionType {
+ AT_END,
+ AT_START,
+ AT_BOUNDARY,
+ AT_NON_BOUNDARY,
+ AFTER_NEWLINE
+ };
+ static AssertionNode* AtEnd(RegExpNode* on_success) {
+ return new (on_success->zone()) AssertionNode(AT_END, on_success);
+ }
+ static AssertionNode* AtStart(RegExpNode* on_success) {
+ return new (on_success->zone()) AssertionNode(AT_START, on_success);
+ }
+ static AssertionNode* AtBoundary(RegExpNode* on_success) {
+ return new (on_success->zone()) AssertionNode(AT_BOUNDARY, on_success);
+ }
+ static AssertionNode* AtNonBoundary(RegExpNode* on_success) {
+ return new (on_success->zone()) AssertionNode(AT_NON_BOUNDARY, on_success);
+ }
+ static AssertionNode* AfterNewline(RegExpNode* on_success) {
+ return new (on_success->zone()) AssertionNode(AFTER_NEWLINE, on_success);
+ }
+ void Accept(NodeVisitor* visitor) override;
+ void Emit(RegExpCompiler* compiler, Trace* trace) override;
+ int EatsAtLeast(int still_to_find, int budget, bool not_at_start) override;
+ void GetQuickCheckDetails(QuickCheckDetails* details,
+ RegExpCompiler* compiler, int filled_in,
+ bool not_at_start) override;
+ void FillInBMInfo(Isolate* isolate, int offset, int budget,
+ BoyerMooreLookahead* bm, bool not_at_start) override;
+ AssertionType assertion_type() { return assertion_type_; }
+
+ private:
+ void EmitBoundaryCheck(RegExpCompiler* compiler, Trace* trace);
+ enum IfPrevious { kIsNonWord, kIsWord };
+ void BacktrackIfPrevious(RegExpCompiler* compiler, Trace* trace,
+ IfPrevious backtrack_if_previous);
+ AssertionNode(AssertionType t, RegExpNode* on_success)
+ : SeqRegExpNode(on_success), assertion_type_(t) {}
+ AssertionType assertion_type_;
+};
+
+class BackReferenceNode : public SeqRegExpNode {
+ public:
+ BackReferenceNode(int start_reg, int end_reg, JSRegExp::Flags flags,
+ bool read_backward, RegExpNode* on_success)
+ : SeqRegExpNode(on_success),
+ start_reg_(start_reg),
+ end_reg_(end_reg),
+ flags_(flags),
+ read_backward_(read_backward) {}
+ void Accept(NodeVisitor* visitor) override;
+ int start_register() { return start_reg_; }
+ int end_register() { return end_reg_; }
+ bool read_backward() { return read_backward_; }
+ void Emit(RegExpCompiler* compiler, Trace* trace) override;
+ int EatsAtLeast(int still_to_find, int recursion_depth,
+ bool not_at_start) override;
+ void GetQuickCheckDetails(QuickCheckDetails* details,
+ RegExpCompiler* compiler, int characters_filled_in,
+ bool not_at_start) override {
+ return;
+ }
+ void FillInBMInfo(Isolate* isolate, int offset, int budget,
+ BoyerMooreLookahead* bm, bool not_at_start) override;
+
+ private:
+ int start_reg_;
+ int end_reg_;
+ JSRegExp::Flags flags_;
+ bool read_backward_;
+};
+
+class EndNode : public RegExpNode {
+ public:
+ enum Action { ACCEPT, BACKTRACK, NEGATIVE_SUBMATCH_SUCCESS };
+ EndNode(Action action, Zone* zone) : RegExpNode(zone), action_(action) {}
+ void Accept(NodeVisitor* visitor) override;
+ void Emit(RegExpCompiler* compiler, Trace* trace) override;
+ int EatsAtLeast(int still_to_find, int recursion_depth,
+ bool not_at_start) override {
+ return 0;
+ }
+ void GetQuickCheckDetails(QuickCheckDetails* details,
+ RegExpCompiler* compiler, int characters_filled_in,
+ bool not_at_start) override {
+ // Returning 0 from EatsAtLeast should ensure we never get here.
+ UNREACHABLE();
+ }
+ void FillInBMInfo(Isolate* isolate, int offset, int budget,
+ BoyerMooreLookahead* bm, bool not_at_start) override {
+ // Returning 0 from EatsAtLeast should ensure we never get here.
+ UNREACHABLE();
+ }
+
+ private:
+ Action action_;
+};
+
+class NegativeSubmatchSuccess : public EndNode {
+ public:
+ NegativeSubmatchSuccess(int stack_pointer_reg, int position_reg,
+ int clear_capture_count, int clear_capture_start,
+ Zone* zone)
+ : EndNode(NEGATIVE_SUBMATCH_SUCCESS, zone),
+ stack_pointer_register_(stack_pointer_reg),
+ current_position_register_(position_reg),
+ clear_capture_count_(clear_capture_count),
+ clear_capture_start_(clear_capture_start) {}
+ void Emit(RegExpCompiler* compiler, Trace* trace) override;
+
+ private:
+ int stack_pointer_register_;
+ int current_position_register_;
+ int clear_capture_count_;
+ int clear_capture_start_;
+};
+
+class Guard : public ZoneObject {
+ public:
+ enum Relation { LT, GEQ };
+ Guard(int reg, Relation op, int value) : reg_(reg), op_(op), value_(value) {}
+ int reg() { return reg_; }
+ Relation op() { return op_; }
+ int value() { return value_; }
+
+ private:
+ int reg_;
+ Relation op_;
+ int value_;
+};
+
+class GuardedAlternative {
+ public:
+ explicit GuardedAlternative(RegExpNode* node)
+ : node_(node), guards_(nullptr) {}
+ void AddGuard(Guard* guard, Zone* zone);
+ RegExpNode* node() { return node_; }
+ void set_node(RegExpNode* node) { node_ = node; }
+ ZoneList<Guard*>* guards() { return guards_; }
+
+ private:
+ RegExpNode* node_;
+ ZoneList<Guard*>* guards_;
+};
+
+class AlternativeGeneration;
+
+class ChoiceNode : public RegExpNode {
+ public:
+ explicit ChoiceNode(int expected_size, Zone* zone)
+ : RegExpNode(zone),
+ alternatives_(new (zone)
+ ZoneList<GuardedAlternative>(expected_size, zone)),
+ not_at_start_(false),
+ being_calculated_(false) {}
+ void Accept(NodeVisitor* visitor) override;
+ void AddAlternative(GuardedAlternative node) {
+ alternatives()->Add(node, zone());
+ }
+ ZoneList<GuardedAlternative>* alternatives() { return alternatives_; }
+ void Emit(RegExpCompiler* compiler, Trace* trace) override;
+ int EatsAtLeast(int still_to_find, int budget, bool not_at_start) override;
+ int EatsAtLeastHelper(int still_to_find, int budget,
+ RegExpNode* ignore_this_node, bool not_at_start);
+ void GetQuickCheckDetails(QuickCheckDetails* details,
+ RegExpCompiler* compiler, int characters_filled_in,
+ bool not_at_start) override;
+ void FillInBMInfo(Isolate* isolate, int offset, int budget,
+ BoyerMooreLookahead* bm, bool not_at_start) override;
+
+ bool being_calculated() { return being_calculated_; }
+ bool not_at_start() { return not_at_start_; }
+ void set_not_at_start() { not_at_start_ = true; }
+ void set_being_calculated(bool b) { being_calculated_ = b; }
+ virtual bool try_to_emit_quick_check_for_alternative(bool is_first) {
+ return true;
+ }
+ RegExpNode* FilterOneByte(int depth) override;
+ virtual bool read_backward() { return false; }
+
+ protected:
+ int GreedyLoopTextLengthForAlternative(GuardedAlternative* alternative);
+ ZoneList<GuardedAlternative>* alternatives_;
+
+ private:
+ friend class Analysis;
+
+ void GenerateGuard(RegExpMacroAssembler* macro_assembler, Guard* guard,
+ Trace* trace);
+ int CalculatePreloadCharacters(RegExpCompiler* compiler, int eats_at_least);
+ void EmitOutOfLineContinuation(RegExpCompiler* compiler, Trace* trace,
+ GuardedAlternative alternative,
+ AlternativeGeneration* alt_gen,
+ int preload_characters,
+ bool next_expects_preload);
+ void SetUpPreLoad(RegExpCompiler* compiler, Trace* current_trace,
+ PreloadState* preloads);
+ void AssertGuardsMentionRegisters(Trace* trace);
+ int EmitOptimizedUnanchoredSearch(RegExpCompiler* compiler, Trace* trace);
+ Trace* EmitGreedyLoop(RegExpCompiler* compiler, Trace* trace,
+ AlternativeGenerationList* alt_gens,
+ PreloadState* preloads,
+ GreedyLoopState* greedy_loop_state, int text_length);
+ void EmitChoices(RegExpCompiler* compiler,
+ AlternativeGenerationList* alt_gens, int first_choice,
+ Trace* trace, PreloadState* preloads);
+
+ // If true, this node is never checked at the start of the input.
+ // Allows a new trace to start with at_start() set to false.
+ bool not_at_start_;
+ bool being_calculated_;
+};
+
+class NegativeLookaroundChoiceNode : public ChoiceNode {
+ public:
+ explicit NegativeLookaroundChoiceNode(GuardedAlternative this_must_fail,
+ GuardedAlternative then_do_this,
+ Zone* zone)
+ : ChoiceNode(2, zone) {
+ AddAlternative(this_must_fail);
+ AddAlternative(then_do_this);
+ }
+ int EatsAtLeast(int still_to_find, int budget, bool not_at_start) override;
+ void GetQuickCheckDetails(QuickCheckDetails* details,
+ RegExpCompiler* compiler, int characters_filled_in,
+ bool not_at_start) override;
+ void FillInBMInfo(Isolate* isolate, int offset, int budget,
+ BoyerMooreLookahead* bm, bool not_at_start) override {
+ alternatives_->at(1).node()->FillInBMInfo(isolate, offset, budget - 1, bm,
+ not_at_start);
+ if (offset == 0) set_bm_info(not_at_start, bm);
+ }
+ // For a negative lookahead we don't emit the quick check for the
+ // alternative that is expected to fail. This is because quick check code
+ // starts by loading enough characters for the alternative that takes fewest
+ // characters, but on a negative lookahead the negative branch did not take
+ // part in that calculation (EatsAtLeast) so the assumptions don't hold.
+ bool try_to_emit_quick_check_for_alternative(bool is_first) override {
+ return !is_first;
+ }
+ RegExpNode* FilterOneByte(int depth) override;
+};
+
+class LoopChoiceNode : public ChoiceNode {
+ public:
+ LoopChoiceNode(bool body_can_be_zero_length, bool read_backward, Zone* zone)
+ : ChoiceNode(2, zone),
+ loop_node_(nullptr),
+ continue_node_(nullptr),
+ body_can_be_zero_length_(body_can_be_zero_length),
+ read_backward_(read_backward) {}
+ void AddLoopAlternative(GuardedAlternative alt);
+ void AddContinueAlternative(GuardedAlternative alt);
+ void Emit(RegExpCompiler* compiler, Trace* trace) override;
+ int EatsAtLeast(int still_to_find, int budget, bool not_at_start) override;
+ void GetQuickCheckDetails(QuickCheckDetails* details,
+ RegExpCompiler* compiler, int characters_filled_in,
+ bool not_at_start) override;
+ void FillInBMInfo(Isolate* isolate, int offset, int budget,
+ BoyerMooreLookahead* bm, bool not_at_start) override;
+ RegExpNode* loop_node() { return loop_node_; }
+ RegExpNode* continue_node() { return continue_node_; }
+ bool body_can_be_zero_length() { return body_can_be_zero_length_; }
+ bool read_backward() override { return read_backward_; }
+ void Accept(NodeVisitor* visitor) override;
+ RegExpNode* FilterOneByte(int depth) override;
+
+ private:
+ // AddAlternative is made private for loop nodes because alternatives
+ // should not be added freely, we need to keep track of which node
+ // goes back to the node itself.
+ void AddAlternative(GuardedAlternative node) {
+ ChoiceNode::AddAlternative(node);
+ }
+
+ RegExpNode* loop_node_;
+ RegExpNode* continue_node_;
+ bool body_can_be_zero_length_;
+ bool read_backward_;
+};
+
+class NodeVisitor {
+ public:
+ virtual ~NodeVisitor() = default;
+#define DECLARE_VISIT(Type) virtual void Visit##Type(Type##Node* that) = 0;
+ FOR_EACH_NODE_TYPE(DECLARE_VISIT)
+#undef DECLARE_VISIT
+ virtual void VisitLoopChoice(LoopChoiceNode* that) { VisitChoice(that); }
+};
+
+} // namespace internal
+} // namespace v8
+
+#endif // V8_REGEXP_REGEXP_NODES_H_
diff --git a/deps/v8/src/regexp/regexp-parser.cc b/deps/v8/src/regexp/regexp-parser.cc
index 7cae456f56..3647680969 100644
--- a/deps/v8/src/regexp/regexp-parser.cc
+++ b/deps/v8/src/regexp/regexp-parser.cc
@@ -9,8 +9,9 @@
#include "src/execution/isolate.h"
#include "src/heap/factory.h"
#include "src/objects/objects-inl.h"
-#include "src/regexp/jsregexp.h"
#include "src/regexp/property-sequences.h"
+#include "src/regexp/regexp-macro-assembler.h"
+#include "src/regexp/regexp.h"
#include "src/strings/char-predicates-inl.h"
#include "src/utils/ostreams.h"
#include "src/utils/utils.h"
@@ -879,24 +880,25 @@ bool RegExpParser::CreateNamedCaptureAtIndex(const ZoneVector<uc16>* name,
DCHECK(0 < index && index <= captures_started_);
DCHECK_NOT_NULL(name);
+ RegExpCapture* capture = GetCapture(index);
+ DCHECK_NULL(capture->name());
+
+ capture->set_name(name);
+
if (named_captures_ == nullptr) {
- named_captures_ = new (zone()) ZoneList<RegExpCapture*>(1, zone());
+ named_captures_ = new (zone_->New(sizeof(*named_captures_)))
+ ZoneSet<RegExpCapture*, RegExpCaptureNameLess>(zone());
} else {
// Check for duplicates and bail if we find any.
- // TODO(jgruber): O(n^2).
- for (const auto& named_capture : *named_captures_) {
- if (*named_capture->name() == *name) {
- ReportError(CStrVector("Duplicate capture group name"));
- return false;
- }
+
+ const auto& named_capture_it = named_captures_->find(capture);
+ if (named_capture_it != named_captures_->end()) {
+ ReportError(CStrVector("Duplicate capture group name"));
+ return false;
}
}
- RegExpCapture* capture = GetCapture(index);
- DCHECK_NULL(capture->name());
-
- capture->set_name(name);
- named_captures_->Add(capture, zone());
+ named_captures_->emplace(capture);
return true;
}
@@ -943,20 +945,22 @@ void RegExpParser::PatchNamedBackReferences() {
}
// Look up and patch the actual capture for each named back reference.
- // TODO(jgruber): O(n^2), optimize if necessary.
for (int i = 0; i < named_back_references_->length(); i++) {
RegExpBackReference* ref = named_back_references_->at(i);
- int index = -1;
- for (const auto& capture : *named_captures_) {
- if (*capture->name() == *ref->name()) {
- index = capture->index();
- break;
- }
- }
+ // Capture used to search the named_captures_ by name, index of the
+ // capture is never used.
+ static const int kInvalidIndex = 0;
+ RegExpCapture* search_capture = new (zone()) RegExpCapture(kInvalidIndex);
+ DCHECK_NULL(search_capture->name());
+ search_capture->set_name(ref->name());
- if (index == -1) {
+ int index = -1;
+ const auto& capture_it = named_captures_->find(search_capture);
+ if (capture_it != named_captures_->end()) {
+ index = (*capture_it)->index();
+ } else {
ReportError(CStrVector("Invalid named capture referenced"));
return;
}
@@ -981,16 +985,17 @@ RegExpCapture* RegExpParser::GetCapture(int index) {
}
Handle<FixedArray> RegExpParser::CreateCaptureNameMap() {
- if (named_captures_ == nullptr || named_captures_->is_empty())
+ if (named_captures_ == nullptr || named_captures_->empty()) {
return Handle<FixedArray>();
+ }
Factory* factory = isolate()->factory();
- int len = named_captures_->length() * 2;
+ int len = static_cast<int>(named_captures_->size()) * 2;
Handle<FixedArray> array = factory->NewFixedArray(len);
- for (int i = 0; i < named_captures_->length(); i++) {
- RegExpCapture* capture = named_captures_->at(i);
+ int i = 0;
+ for (const auto& capture : *named_captures_) {
Vector<const uc16> capture_name(capture->name()->data(),
capture->name()->size());
// CSA code in ConstructNewResultFromMatchInfo requires these strings to be
@@ -998,7 +1003,10 @@ Handle<FixedArray> RegExpParser::CreateCaptureNameMap() {
Handle<String> name = factory->InternalizeString(capture_name);
array->set(i * 2, *name);
array->set(i * 2 + 1, Smi::FromInt(capture->index()));
+
+ i++;
}
+ DCHECK_EQ(i * 2, len);
return array;
}
@@ -1963,12 +1971,6 @@ void RegExpBuilder::AddTerm(RegExpTree* term) {
void RegExpBuilder::AddAssertion(RegExpTree* assert) {
FlushText();
- if (terms_.length() > 0 && terms_.last()->IsAssertion()) {
- // Omit repeated assertions of the same type.
- RegExpAssertion* last = terms_.last()->AsAssertion();
- RegExpAssertion* next = assert->AsAssertion();
- if (last->assertion_type() == next->assertion_type()) return;
- }
terms_.Add(assert, zone());
LAST(ADD_ASSERT);
}
diff --git a/deps/v8/src/regexp/regexp-parser.h b/deps/v8/src/regexp/regexp-parser.h
index bf9e62ed71..36cec7e984 100644
--- a/deps/v8/src/regexp/regexp-parser.h
+++ b/deps/v8/src/regexp/regexp-parser.h
@@ -326,11 +326,19 @@ class V8_EXPORT_PRIVATE RegExpParser {
FlatStringReader* in() { return in_; }
void ScanForCaptures();
+ struct RegExpCaptureNameLess {
+ bool operator()(const RegExpCapture* lhs, const RegExpCapture* rhs) const {
+ DCHECK_NOT_NULL(lhs);
+ DCHECK_NOT_NULL(rhs);
+ return *lhs->name() < *rhs->name();
+ }
+ };
+
Isolate* isolate_;
Zone* zone_;
Handle<String>* error_;
ZoneList<RegExpCapture*>* captures_;
- ZoneList<RegExpCapture*>* named_captures_;
+ ZoneSet<RegExpCapture*, RegExpCaptureNameLess>* named_captures_;
ZoneList<RegExpBackReference*>* named_back_references_;
FlatStringReader* in_;
uc32 current_;
diff --git a/deps/v8/src/regexp/regexp-utils.cc b/deps/v8/src/regexp/regexp-utils.cc
index 49f9d4476b..ad50270fdc 100644
--- a/deps/v8/src/regexp/regexp-utils.cc
+++ b/deps/v8/src/regexp/regexp-utils.cc
@@ -8,7 +8,7 @@
#include "src/heap/factory.h"
#include "src/objects/js-regexp-inl.h"
#include "src/objects/objects-inl.h"
-#include "src/regexp/jsregexp.h"
+#include "src/regexp/regexp.h"
namespace v8 {
namespace internal {
@@ -179,7 +179,9 @@ bool RegExpUtils::IsUnmodifiedRegExp(Isolate* isolate, Handle<Object> obj) {
return false;
}
- if (!isolate->IsRegExpSpeciesLookupChainIntact()) return false;
+ if (!isolate->IsRegExpSpeciesLookupChainIntact(isolate->native_context())) {
+ return false;
+ }
// The smi check is required to omit ToLength(lastIndex) calls with possible
// user-code execution on the fast path.
diff --git a/deps/v8/src/regexp/regexp.cc b/deps/v8/src/regexp/regexp.cc
new file mode 100644
index 0000000000..15b0321c46
--- /dev/null
+++ b/deps/v8/src/regexp/regexp.cc
@@ -0,0 +1,1018 @@
+// Copyright 2012 the V8 project authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+#include "src/regexp/regexp.h"
+
+#include "src/codegen/compilation-cache.h"
+#include "src/heap/heap-inl.h"
+#include "src/objects/js-regexp-inl.h"
+#include "src/regexp/regexp-bytecode-generator.h"
+#include "src/regexp/regexp-compiler.h"
+#include "src/regexp/regexp-dotprinter.h"
+#include "src/regexp/regexp-interpreter.h"
+#include "src/regexp/regexp-macro-assembler-arch.h"
+#include "src/regexp/regexp-parser.h"
+#include "src/strings/string-search.h"
+
+namespace v8 {
+namespace internal {
+
+using namespace regexp_compiler_constants; // NOLINT(build/namespaces)
+
+class RegExpImpl final : public AllStatic {
+ public:
+ // Returns a string representation of a regular expression.
+ // Implements RegExp.prototype.toString, see ECMA-262 section 15.10.6.4.
+ // This function calls the garbage collector if necessary.
+ static Handle<String> ToString(Handle<Object> value);
+
+ // Prepares a JSRegExp object with Irregexp-specific data.
+ static void IrregexpInitialize(Isolate* isolate, Handle<JSRegExp> re,
+ Handle<String> pattern, JSRegExp::Flags flags,
+ int capture_register_count);
+
+ static void AtomCompile(Isolate* isolate, Handle<JSRegExp> re,
+ Handle<String> pattern, JSRegExp::Flags flags,
+ Handle<String> match_pattern);
+
+ static int AtomExecRaw(Isolate* isolate, Handle<JSRegExp> regexp,
+ Handle<String> subject, int index, int32_t* output,
+ int output_size);
+
+ static Handle<Object> AtomExec(Isolate* isolate, Handle<JSRegExp> regexp,
+ Handle<String> subject, int index,
+ Handle<RegExpMatchInfo> last_match_info);
+
+ // Execute a regular expression on the subject, starting from index.
+ // If matching succeeds, return the number of matches. This can be larger
+ // than one in the case of global regular expressions.
+ // The captures and subcaptures are stored into the registers vector.
+ // If matching fails, returns RE_FAILURE.
+ // If execution fails, sets a pending exception and returns RE_EXCEPTION.
+ static int IrregexpExecRaw(Isolate* isolate, Handle<JSRegExp> regexp,
+ Handle<String> subject, int index, int32_t* output,
+ int output_size);
+
+ // Execute an Irregexp bytecode pattern.
+ // On a successful match, the result is a JSArray containing
+ // captured positions. On a failure, the result is the null value.
+ // Returns an empty handle in case of an exception.
+ V8_WARN_UNUSED_RESULT static MaybeHandle<Object> IrregexpExec(
+ Isolate* isolate, Handle<JSRegExp> regexp, Handle<String> subject,
+ int index, Handle<RegExpMatchInfo> last_match_info);
+
+ static bool CompileIrregexp(Isolate* isolate, Handle<JSRegExp> re,
+ Handle<String> sample_subject, bool is_one_byte);
+ static inline bool EnsureCompiledIrregexp(Isolate* isolate,
+ Handle<JSRegExp> re,
+ Handle<String> sample_subject,
+ bool is_one_byte);
+
+ // Returns true on success, false on failure.
+ static bool Compile(Isolate* isolate, Zone* zone, RegExpCompileData* input,
+ JSRegExp::Flags flags, Handle<String> pattern,
+ Handle<String> sample_subject, bool is_one_byte);
+
+ // For acting on the JSRegExp data FixedArray.
+ static int IrregexpMaxRegisterCount(FixedArray re);
+ static void SetIrregexpMaxRegisterCount(FixedArray re, int value);
+ static void SetIrregexpCaptureNameMap(FixedArray re,
+ Handle<FixedArray> value);
+ static int IrregexpNumberOfCaptures(FixedArray re);
+ static int IrregexpNumberOfRegisters(FixedArray re);
+ static ByteArray IrregexpByteCode(FixedArray re, bool is_one_byte);
+ static Code IrregexpNativeCode(FixedArray re, bool is_one_byte);
+};
+
+V8_WARN_UNUSED_RESULT
+static inline MaybeHandle<Object> ThrowRegExpException(
+ Isolate* isolate, Handle<JSRegExp> re, Handle<String> pattern,
+ Handle<String> error_text) {
+ THROW_NEW_ERROR(
+ isolate,
+ NewSyntaxError(MessageTemplate::kMalformedRegExp, pattern, error_text),
+ Object);
+}
+
+inline void ThrowRegExpException(Isolate* isolate, Handle<JSRegExp> re,
+ Handle<String> error_text) {
+ USE(ThrowRegExpException(isolate, re, Handle<String>(re->Pattern(), isolate),
+ error_text));
+}
+
+// Identifies the sort of regexps where the regexp engine is faster
+// than the code used for atom matches.
+static bool HasFewDifferentCharacters(Handle<String> pattern) {
+ int length = Min(kMaxLookaheadForBoyerMoore, pattern->length());
+ if (length <= kPatternTooShortForBoyerMoore) return false;
+ const int kMod = 128;
+ bool character_found[kMod];
+ int different = 0;
+ memset(&character_found[0], 0, sizeof(character_found));
+ for (int i = 0; i < length; i++) {
+ int ch = (pattern->Get(i) & (kMod - 1));
+ if (!character_found[ch]) {
+ character_found[ch] = true;
+ different++;
+ // We declare a regexp low-alphabet if it has at least 3 times as many
+ // characters as it has different characters.
+ if (different * 3 > length) return false;
+ }
+ }
+ return true;
+}
+
+// Generic RegExp methods. Dispatches to implementation specific methods.
+
+// static
+MaybeHandle<Object> RegExp::Compile(Isolate* isolate, Handle<JSRegExp> re,
+ Handle<String> pattern,
+ JSRegExp::Flags flags) {
+ DCHECK(pattern->IsFlat());
+
+ Zone zone(isolate->allocator(), ZONE_NAME);
+ CompilationCache* compilation_cache = isolate->compilation_cache();
+ MaybeHandle<FixedArray> maybe_cached =
+ compilation_cache->LookupRegExp(pattern, flags);
+ Handle<FixedArray> cached;
+ if (maybe_cached.ToHandle(&cached)) {
+ re->set_data(*cached);
+ return re;
+ }
+
+ PostponeInterruptsScope postpone(isolate);
+ RegExpCompileData parse_result;
+ FlatStringReader reader(isolate, pattern);
+ DCHECK(!isolate->has_pending_exception());
+ if (!RegExpParser::ParseRegExp(isolate, &zone, &reader, flags,
+ &parse_result)) {
+ // Throw an exception if we fail to parse the pattern.
+ return ThrowRegExpException(isolate, re, pattern, parse_result.error);
+ }
+
+ bool has_been_compiled = false;
+
+ if (parse_result.simple && !IgnoreCase(flags) && !IsSticky(flags) &&
+ !HasFewDifferentCharacters(pattern)) {
+ // Parse-tree is a single atom that is equal to the pattern.
+ RegExpImpl::AtomCompile(isolate, re, pattern, flags, pattern);
+ has_been_compiled = true;
+ } else if (parse_result.tree->IsAtom() && !IsSticky(flags) &&
+ parse_result.capture_count == 0) {
+ RegExpAtom* atom = parse_result.tree->AsAtom();
+ Vector<const uc16> atom_pattern = atom->data();
+ Handle<String> atom_string;
+ ASSIGN_RETURN_ON_EXCEPTION(
+ isolate, atom_string,
+ isolate->factory()->NewStringFromTwoByte(atom_pattern), Object);
+ if (!IgnoreCase(atom->flags()) && !HasFewDifferentCharacters(atom_string)) {
+ RegExpImpl::AtomCompile(isolate, re, pattern, flags, atom_string);
+ has_been_compiled = true;
+ }
+ }
+ if (!has_been_compiled) {
+ RegExpImpl::IrregexpInitialize(isolate, re, pattern, flags,
+ parse_result.capture_count);
+ }
+ DCHECK(re->data().IsFixedArray());
+ // Compilation succeeded so the data is set on the regexp
+ // and we can store it in the cache.
+ Handle<FixedArray> data(FixedArray::cast(re->data()), isolate);
+ compilation_cache->PutRegExp(pattern, flags, data);
+
+ return re;
+}
+
+// static
+MaybeHandle<Object> RegExp::Exec(Isolate* isolate, Handle<JSRegExp> regexp,
+ Handle<String> subject, int index,
+ Handle<RegExpMatchInfo> last_match_info) {
+ switch (regexp->TypeTag()) {
+ case JSRegExp::ATOM:
+ return RegExpImpl::AtomExec(isolate, regexp, subject, index,
+ last_match_info);
+ case JSRegExp::IRREGEXP: {
+ return RegExpImpl::IrregexpExec(isolate, regexp, subject, index,
+ last_match_info);
+ }
+ default:
+ UNREACHABLE();
+ }
+}
+
+// RegExp Atom implementation: Simple string search using indexOf.
+
+void RegExpImpl::AtomCompile(Isolate* isolate, Handle<JSRegExp> re,
+ Handle<String> pattern, JSRegExp::Flags flags,
+ Handle<String> match_pattern) {
+ isolate->factory()->SetRegExpAtomData(re, JSRegExp::ATOM, pattern, flags,
+ match_pattern);
+}
+
+static void SetAtomLastCapture(Isolate* isolate,
+ Handle<RegExpMatchInfo> last_match_info,
+ String subject, int from, int to) {
+ SealHandleScope shs(isolate);
+ last_match_info->SetNumberOfCaptureRegisters(2);
+ last_match_info->SetLastSubject(subject);
+ last_match_info->SetLastInput(subject);
+ last_match_info->SetCapture(0, from);
+ last_match_info->SetCapture(1, to);
+}
+
+int RegExpImpl::AtomExecRaw(Isolate* isolate, Handle<JSRegExp> regexp,
+ Handle<String> subject, int index, int32_t* output,
+ int output_size) {
+ DCHECK_LE(0, index);
+ DCHECK_LE(index, subject->length());
+
+ subject = String::Flatten(isolate, subject);
+ DisallowHeapAllocation no_gc; // ensure vectors stay valid
+
+ String needle = String::cast(regexp->DataAt(JSRegExp::kAtomPatternIndex));
+ int needle_len = needle.length();
+ DCHECK(needle.IsFlat());
+ DCHECK_LT(0, needle_len);
+
+ if (index + needle_len > subject->length()) {
+ return RegExp::RE_FAILURE;
+ }
+
+ for (int i = 0; i < output_size; i += 2) {
+ String::FlatContent needle_content = needle.GetFlatContent(no_gc);
+ String::FlatContent subject_content = subject->GetFlatContent(no_gc);
+ DCHECK(needle_content.IsFlat());
+ DCHECK(subject_content.IsFlat());
+ // dispatch on type of strings
+ index =
+ (needle_content.IsOneByte()
+ ? (subject_content.IsOneByte()
+ ? SearchString(isolate, subject_content.ToOneByteVector(),
+ needle_content.ToOneByteVector(), index)
+ : SearchString(isolate, subject_content.ToUC16Vector(),
+ needle_content.ToOneByteVector(), index))
+ : (subject_content.IsOneByte()
+ ? SearchString(isolate, subject_content.ToOneByteVector(),
+ needle_content.ToUC16Vector(), index)
+ : SearchString(isolate, subject_content.ToUC16Vector(),
+ needle_content.ToUC16Vector(), index)));
+ if (index == -1) {
+ return i / 2; // Return number of matches.
+ } else {
+ output[i] = index;
+ output[i + 1] = index + needle_len;
+ index += needle_len;
+ }
+ }
+ return output_size / 2;
+}
+
+Handle<Object> RegExpImpl::AtomExec(Isolate* isolate, Handle<JSRegExp> re,
+ Handle<String> subject, int index,
+ Handle<RegExpMatchInfo> last_match_info) {
+ static const int kNumRegisters = 2;
+ STATIC_ASSERT(kNumRegisters <= Isolate::kJSRegexpStaticOffsetsVectorSize);
+ int32_t* output_registers = isolate->jsregexp_static_offsets_vector();
+
+ int res =
+ AtomExecRaw(isolate, re, subject, index, output_registers, kNumRegisters);
+
+ if (res == RegExp::RE_FAILURE) return isolate->factory()->null_value();
+
+ DCHECK_EQ(res, RegExp::RE_SUCCESS);
+ SealHandleScope shs(isolate);
+ SetAtomLastCapture(isolate, last_match_info, *subject, output_registers[0],
+ output_registers[1]);
+ return last_match_info;
+}
+
+// Irregexp implementation.
+
+// Ensures that the regexp object contains a compiled version of the
+// source for either one-byte or two-byte subject strings.
+// If the compiled version doesn't already exist, it is compiled
+// from the source pattern.
+// If compilation fails, an exception is thrown and this function
+// returns false.
+bool RegExpImpl::EnsureCompiledIrregexp(Isolate* isolate, Handle<JSRegExp> re,
+ Handle<String> sample_subject,
+ bool is_one_byte) {
+ Object compiled_code = re->DataAt(JSRegExp::code_index(is_one_byte));
+ if (compiled_code != Smi::FromInt(JSRegExp::kUninitializedValue)) {
+ DCHECK(FLAG_regexp_interpret_all ? compiled_code.IsByteArray()
+ : compiled_code.IsCode());
+ return true;
+ }
+ return CompileIrregexp(isolate, re, sample_subject, is_one_byte);
+}
+
+bool RegExpImpl::CompileIrregexp(Isolate* isolate, Handle<JSRegExp> re,
+ Handle<String> sample_subject,
+ bool is_one_byte) {
+ // Compile the RegExp.
+ Zone zone(isolate->allocator(), ZONE_NAME);
+ PostponeInterruptsScope postpone(isolate);
+#ifdef DEBUG
+ Object entry = re->DataAt(JSRegExp::code_index(is_one_byte));
+ // When arriving here entry can only be a smi representing an uncompiled
+ // regexp.
+ DCHECK(entry.IsSmi());
+ int entry_value = Smi::ToInt(entry);
+ DCHECK_EQ(JSRegExp::kUninitializedValue, entry_value);
+#endif
+
+ JSRegExp::Flags flags = re->GetFlags();
+
+ Handle<String> pattern(re->Pattern(), isolate);
+ pattern = String::Flatten(isolate, pattern);
+ RegExpCompileData compile_data;
+ FlatStringReader reader(isolate, pattern);
+ 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(isolate, re, pattern, compile_data.error));
+ return false;
+ }
+ const bool compilation_succeeded =
+ Compile(isolate, &zone, &compile_data, flags, pattern, sample_subject,
+ is_one_byte);
+ if (!compilation_succeeded) {
+ DCHECK(!compile_data.error.is_null());
+ ThrowRegExpException(isolate, re, compile_data.error);
+ return false;
+ }
+
+ Handle<FixedArray> data =
+ Handle<FixedArray>(FixedArray::cast(re->data()), isolate);
+ data->set(JSRegExp::code_index(is_one_byte), compile_data.code);
+ SetIrregexpCaptureNameMap(*data, compile_data.capture_name_map);
+ int register_max = IrregexpMaxRegisterCount(*data);
+ if (compile_data.register_count > register_max) {
+ SetIrregexpMaxRegisterCount(*data, compile_data.register_count);
+ }
+
+ return true;
+}
+
+int RegExpImpl::IrregexpMaxRegisterCount(FixedArray re) {
+ return Smi::cast(re.get(JSRegExp::kIrregexpMaxRegisterCountIndex)).value();
+}
+
+void RegExpImpl::SetIrregexpMaxRegisterCount(FixedArray re, int value) {
+ re.set(JSRegExp::kIrregexpMaxRegisterCountIndex, Smi::FromInt(value));
+}
+
+void RegExpImpl::SetIrregexpCaptureNameMap(FixedArray re,
+ Handle<FixedArray> value) {
+ if (value.is_null()) {
+ re.set(JSRegExp::kIrregexpCaptureNameMapIndex, Smi::kZero);
+ } else {
+ re.set(JSRegExp::kIrregexpCaptureNameMapIndex, *value);
+ }
+}
+
+int RegExpImpl::IrregexpNumberOfCaptures(FixedArray re) {
+ return Smi::ToInt(re.get(JSRegExp::kIrregexpCaptureCountIndex));
+}
+
+int RegExpImpl::IrregexpNumberOfRegisters(FixedArray re) {
+ return Smi::ToInt(re.get(JSRegExp::kIrregexpMaxRegisterCountIndex));
+}
+
+ByteArray RegExpImpl::IrregexpByteCode(FixedArray re, bool is_one_byte) {
+ return ByteArray::cast(re.get(JSRegExp::code_index(is_one_byte)));
+}
+
+Code RegExpImpl::IrregexpNativeCode(FixedArray re, bool is_one_byte) {
+ return Code::cast(re.get(JSRegExp::code_index(is_one_byte)));
+}
+
+void RegExpImpl::IrregexpInitialize(Isolate* isolate, Handle<JSRegExp> re,
+ Handle<String> pattern,
+ JSRegExp::Flags flags, int capture_count) {
+ // Initialize compiled code entries to null.
+ isolate->factory()->SetRegExpIrregexpData(re, JSRegExp::IRREGEXP, pattern,
+ flags, capture_count);
+}
+
+// static
+int RegExp::IrregexpPrepare(Isolate* isolate, Handle<JSRegExp> regexp,
+ Handle<String> subject) {
+ DCHECK(subject->IsFlat());
+
+ // Check representation of the underlying storage.
+ bool is_one_byte = String::IsOneByteRepresentationUnderneath(*subject);
+ if (!RegExpImpl::EnsureCompiledIrregexp(isolate, regexp, subject,
+ is_one_byte)) {
+ return -1;
+ }
+
+ DisallowHeapAllocation no_gc;
+ FixedArray data = FixedArray::cast(regexp->data());
+ if (FLAG_regexp_interpret_all) {
+ // Byte-code regexp needs space allocated for all its registers.
+ // The result captures are copied to the start of the registers array
+ // if the match succeeds. This way those registers are not clobbered
+ // when we set the last match info from last successful match.
+ return RegExpImpl::IrregexpNumberOfRegisters(data) +
+ (RegExpImpl::IrregexpNumberOfCaptures(data) + 1) * 2;
+ } else {
+ // Native regexp only needs room to output captures. Registers are handled
+ // internally.
+ return (RegExpImpl::IrregexpNumberOfCaptures(data) + 1) * 2;
+ }
+}
+
+int RegExpImpl::IrregexpExecRaw(Isolate* isolate, Handle<JSRegExp> regexp,
+ Handle<String> subject, int index,
+ int32_t* output, int output_size) {
+ Handle<FixedArray> irregexp(FixedArray::cast(regexp->data()), isolate);
+
+ DCHECK_LE(0, index);
+ DCHECK_LE(index, subject->length());
+ DCHECK(subject->IsFlat());
+
+ bool is_one_byte = String::IsOneByteRepresentationUnderneath(*subject);
+
+ if (!FLAG_regexp_interpret_all) {
+ DCHECK(output_size >= (IrregexpNumberOfCaptures(*irregexp) + 1) * 2);
+ do {
+ EnsureCompiledIrregexp(isolate, regexp, subject, is_one_byte);
+ Handle<Code> code(IrregexpNativeCode(*irregexp, is_one_byte), isolate);
+ // The stack is used to allocate registers for the compiled regexp code.
+ // This means that in case of failure, the output registers array is left
+ // untouched and contains the capture results from the previous successful
+ // match. We can use that to set the last match info lazily.
+ int res = NativeRegExpMacroAssembler::Match(code, subject, output,
+ output_size, index, isolate);
+ if (res != NativeRegExpMacroAssembler::RETRY) {
+ DCHECK(res != NativeRegExpMacroAssembler::EXCEPTION ||
+ isolate->has_pending_exception());
+ STATIC_ASSERT(static_cast<int>(NativeRegExpMacroAssembler::SUCCESS) ==
+ RegExp::RE_SUCCESS);
+ STATIC_ASSERT(static_cast<int>(NativeRegExpMacroAssembler::FAILURE) ==
+ RegExp::RE_FAILURE);
+ STATIC_ASSERT(static_cast<int>(NativeRegExpMacroAssembler::EXCEPTION) ==
+ RegExp::RE_EXCEPTION);
+ return res;
+ }
+ // If result is RETRY, the string has changed representation, and we
+ // must restart from scratch.
+ // In this case, it means we must make sure we are prepared to handle
+ // the, potentially, different subject (the string can switch between
+ // being internal and external, and even between being Latin1 and UC16,
+ // but the characters are always the same).
+ RegExp::IrregexpPrepare(isolate, regexp, subject);
+ is_one_byte = String::IsOneByteRepresentationUnderneath(*subject);
+ } while (true);
+ UNREACHABLE();
+ } else {
+ DCHECK(FLAG_regexp_interpret_all);
+ DCHECK(output_size >= IrregexpNumberOfRegisters(*irregexp));
+ // We must have done EnsureCompiledIrregexp, so we can get the number of
+ // registers.
+ int number_of_capture_registers =
+ (IrregexpNumberOfCaptures(*irregexp) + 1) * 2;
+ int32_t* raw_output = &output[number_of_capture_registers];
+
+ do {
+ // We do not touch the actual capture result registers until we know there
+ // has been a match so that we can use those capture results to set the
+ // last match info.
+ for (int i = number_of_capture_registers - 1; i >= 0; i--) {
+ raw_output[i] = -1;
+ }
+ Handle<ByteArray> byte_codes(IrregexpByteCode(*irregexp, is_one_byte),
+ isolate);
+
+ IrregexpInterpreter::Result result = IrregexpInterpreter::Match(
+ isolate, byte_codes, subject, raw_output, index);
+ DCHECK_IMPLIES(result == IrregexpInterpreter::EXCEPTION,
+ isolate->has_pending_exception());
+
+ switch (result) {
+ case IrregexpInterpreter::SUCCESS:
+ // Copy capture results to the start of the registers array.
+ MemCopy(output, raw_output,
+ number_of_capture_registers * sizeof(int32_t));
+ return result;
+ case IrregexpInterpreter::EXCEPTION:
+ case IrregexpInterpreter::FAILURE:
+ return result;
+ case IrregexpInterpreter::RETRY:
+ // The string has changed representation, and we must restart the
+ // match.
+ is_one_byte = String::IsOneByteRepresentationUnderneath(*subject);
+ EnsureCompiledIrregexp(isolate, regexp, subject, is_one_byte);
+ break;
+ }
+ } while (true);
+ UNREACHABLE();
+ }
+}
+
+MaybeHandle<Object> RegExpImpl::IrregexpExec(
+ Isolate* isolate, Handle<JSRegExp> regexp, Handle<String> subject,
+ int previous_index, Handle<RegExpMatchInfo> last_match_info) {
+ DCHECK_EQ(regexp->TypeTag(), JSRegExp::IRREGEXP);
+
+ subject = String::Flatten(isolate, subject);
+
+ // Prepare space for the return values.
+#ifdef DEBUG
+ if (FLAG_regexp_interpret_all && FLAG_trace_regexp_bytecodes) {
+ String pattern = regexp->Pattern();
+ PrintF("\n\nRegexp match: /%s/\n\n", pattern.ToCString().get());
+ PrintF("\n\nSubject string: '%s'\n\n", subject->ToCString().get());
+ }
+#endif
+ int required_registers = RegExp::IrregexpPrepare(isolate, regexp, subject);
+ if (required_registers < 0) {
+ // Compiling failed with an exception.
+ DCHECK(isolate->has_pending_exception());
+ return MaybeHandle<Object>();
+ }
+
+ int32_t* output_registers = nullptr;
+ if (required_registers > Isolate::kJSRegexpStaticOffsetsVectorSize) {
+ output_registers = NewArray<int32_t>(required_registers);
+ }
+ std::unique_ptr<int32_t[]> auto_release(output_registers);
+ if (output_registers == nullptr) {
+ output_registers = isolate->jsregexp_static_offsets_vector();
+ }
+
+ int res =
+ RegExpImpl::IrregexpExecRaw(isolate, regexp, subject, previous_index,
+ output_registers, required_registers);
+ if (res == RegExp::RE_SUCCESS) {
+ int capture_count =
+ IrregexpNumberOfCaptures(FixedArray::cast(regexp->data()));
+ return RegExp::SetLastMatchInfo(isolate, last_match_info, subject,
+ capture_count, output_registers);
+ }
+ if (res == RegExp::RE_EXCEPTION) {
+ DCHECK(isolate->has_pending_exception());
+ return MaybeHandle<Object>();
+ }
+ DCHECK(res == RegExp::RE_FAILURE);
+ return isolate->factory()->null_value();
+}
+
+// static
+Handle<RegExpMatchInfo> RegExp::SetLastMatchInfo(
+ Isolate* isolate, Handle<RegExpMatchInfo> last_match_info,
+ Handle<String> subject, int capture_count, int32_t* match) {
+ // This is the only place where match infos can grow. If, after executing the
+ // regexp, RegExpExecStub finds that the match info is too small, it restarts
+ // execution in RegExpImpl::Exec, which finally grows the match info right
+ // here.
+
+ int capture_register_count = (capture_count + 1) * 2;
+ Handle<RegExpMatchInfo> result = RegExpMatchInfo::ReserveCaptures(
+ isolate, last_match_info, capture_register_count);
+ result->SetNumberOfCaptureRegisters(capture_register_count);
+
+ if (*result != *last_match_info) {
+ if (*last_match_info == *isolate->regexp_last_match_info()) {
+ // This inner condition is only needed for special situations like the
+ // regexp fuzzer, where we pass our own custom RegExpMatchInfo to
+ // RegExpImpl::Exec; there actually want to bypass the Isolate's match
+ // info and execute the regexp without side effects.
+ isolate->native_context()->set_regexp_last_match_info(*result);
+ }
+ }
+
+ DisallowHeapAllocation no_allocation;
+ if (match != nullptr) {
+ for (int i = 0; i < capture_register_count; i += 2) {
+ result->SetCapture(i, match[i]);
+ result->SetCapture(i + 1, match[i + 1]);
+ }
+ }
+ result->SetLastSubject(*subject);
+ result->SetLastInput(*subject);
+ return result;
+}
+
+// static
+void RegExp::DotPrintForTesting(const char* label, RegExpNode* node) {
+ DotPrinter::DotPrint(label, node);
+}
+
+namespace {
+
+// Returns true if we've either generated too much irregex code within this
+// isolate, or the pattern string is too long.
+bool TooMuchRegExpCode(Isolate* isolate, Handle<String> pattern) {
+ // Limit the space regexps take up on the heap. In order to limit this we
+ // would like to keep track of the amount of regexp code on the heap. This
+ // is not tracked, however. As a conservative approximation we track the
+ // total regexp code compiled including code that has subsequently been freed
+ // and the total executable memory at any point.
+ static constexpr size_t kRegExpExecutableMemoryLimit = 16 * MB;
+ static constexpr size_t kRegExpCompiledLimit = 1 * MB;
+
+ Heap* heap = isolate->heap();
+ if (pattern->length() > RegExp::kRegExpTooLargeToOptimize) return true;
+ return (isolate->total_regexp_code_generated() > kRegExpCompiledLimit &&
+ heap->CommittedMemoryExecutable() > kRegExpExecutableMemoryLimit);
+}
+
+} // namespace
+
+// static
+bool RegExp::CompileForTesting(Isolate* isolate, Zone* zone,
+ RegExpCompileData* data, JSRegExp::Flags flags,
+ Handle<String> pattern,
+ Handle<String> sample_subject,
+ bool is_one_byte) {
+ return RegExpImpl::Compile(isolate, zone, data, flags, pattern,
+ sample_subject, is_one_byte);
+}
+
+bool RegExpImpl::Compile(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) {
+ data->error =
+ isolate->factory()->NewStringFromAsciiChecked("RegExp too big");
+ return false;
+ }
+
+ bool is_sticky = IsSticky(flags);
+ bool is_global = IsGlobal(flags);
+ bool is_unicode = IsUnicode(flags);
+ RegExpCompiler compiler(isolate, zone, data->capture_count, is_one_byte);
+
+ if (compiler.optimize()) {
+ compiler.set_optimize(!TooMuchRegExpCode(isolate, pattern));
+ }
+
+ // Sample some characters from the middle of the string.
+ static const int kSampleSize = 128;
+
+ sample_subject = String::Flatten(isolate, sample_subject);
+ int chars_sampled = 0;
+ int half_way = (sample_subject->length() - kSampleSize) / 2;
+ for (int i = Max(0, half_way);
+ i < sample_subject->length() && chars_sampled < kSampleSize;
+ i++, chars_sampled++) {
+ compiler.frequency_collator()->CountCharacter(sample_subject->Get(i));
+ }
+
+ // Wrap the body of the regexp in capture #0.
+ RegExpNode* captured_body =
+ RegExpCapture::ToNode(data->tree, 0, &compiler, compiler.accept());
+ RegExpNode* node = captured_body;
+ bool is_end_anchored = data->tree->IsAnchoredAtEnd();
+ bool is_start_anchored = data->tree->IsAnchoredAtStart();
+ int max_length = data->tree->max_match();
+ if (!is_start_anchored && !is_sticky) {
+ // Add a .*? at the beginning, outside the body capture, unless
+ // this expression is anchored at the beginning or sticky.
+ JSRegExp::Flags default_flags = JSRegExp::Flags();
+ RegExpNode* loop_node = RegExpQuantifier::ToNode(
+ 0, RegExpTree::kInfinity, false,
+ new (zone) RegExpCharacterClass('*', default_flags), &compiler,
+ captured_body, data->contains_anchor);
+
+ if (data->contains_anchor) {
+ // Unroll loop once, to take care of the case that might start
+ // at the start of input.
+ ChoiceNode* first_step_node = new (zone) ChoiceNode(2, zone);
+ first_step_node->AddAlternative(GuardedAlternative(captured_body));
+ first_step_node->AddAlternative(GuardedAlternative(new (zone) TextNode(
+ new (zone) RegExpCharacterClass('*', default_flags), false,
+ loop_node)));
+ node = first_step_node;
+ } else {
+ node = loop_node;
+ }
+ }
+ if (is_one_byte) {
+ node = node->FilterOneByte(RegExpCompiler::kMaxRecursion);
+ // Do it again to propagate the new nodes to places where they were not
+ // put because they had not been calculated yet.
+ if (node != nullptr) {
+ node = node->FilterOneByte(RegExpCompiler::kMaxRecursion);
+ }
+ } else if (is_unicode && (is_global || is_sticky)) {
+ node = RegExpCompiler::OptionallyStepBackToLeadSurrogate(&compiler, node,
+ flags);
+ }
+
+ if (node == nullptr) node = new (zone) EndNode(EndNode::BACKTRACK, zone);
+ data->node = node;
+ Analysis analysis(isolate, is_one_byte);
+ analysis.EnsureAnalyzed(node);
+ if (analysis.has_failed()) {
+ data->error =
+ isolate->factory()->NewStringFromAsciiChecked(analysis.error_message());
+ return false;
+ }
+
+ // Create the correct assembler for the architecture.
+ std::unique_ptr<RegExpMacroAssembler> macro_assembler;
+ if (!FLAG_regexp_interpret_all) {
+ // Native regexp implementation.
+ DCHECK(!FLAG_jitless);
+
+ NativeRegExpMacroAssembler::Mode mode =
+ is_one_byte ? NativeRegExpMacroAssembler::LATIN1
+ : NativeRegExpMacroAssembler::UC16;
+
+#if V8_TARGET_ARCH_IA32
+ macro_assembler.reset(new RegExpMacroAssemblerIA32(
+ isolate, zone, mode, (data->capture_count + 1) * 2));
+#elif V8_TARGET_ARCH_X64
+ macro_assembler.reset(new RegExpMacroAssemblerX64(
+ isolate, zone, mode, (data->capture_count + 1) * 2));
+#elif V8_TARGET_ARCH_ARM
+ macro_assembler.reset(new RegExpMacroAssemblerARM(
+ isolate, zone, mode, (data->capture_count + 1) * 2));
+#elif V8_TARGET_ARCH_ARM64
+ macro_assembler.reset(new RegExpMacroAssemblerARM64(
+ isolate, zone, mode, (data->capture_count + 1) * 2));
+#elif V8_TARGET_ARCH_S390
+ macro_assembler.reset(new RegExpMacroAssemblerS390(
+ isolate, zone, mode, (data->capture_count + 1) * 2));
+#elif V8_TARGET_ARCH_PPC
+ macro_assembler.reset(new RegExpMacroAssemblerPPC(
+ isolate, zone, mode, (data->capture_count + 1) * 2));
+#elif V8_TARGET_ARCH_MIPS
+ macro_assembler.reset(new RegExpMacroAssemblerMIPS(
+ isolate, zone, mode, (data->capture_count + 1) * 2));
+#elif V8_TARGET_ARCH_MIPS64
+ macro_assembler.reset(new RegExpMacroAssemblerMIPS(
+ isolate, zone, mode, (data->capture_count + 1) * 2));
+#else
+#error "Unsupported architecture"
+#endif
+ } else {
+ DCHECK(FLAG_regexp_interpret_all);
+
+ // Interpreted regexp implementation.
+ macro_assembler.reset(new RegExpBytecodeGenerator(isolate, zone));
+ }
+
+ macro_assembler->set_slow_safe(TooMuchRegExpCode(isolate, pattern));
+
+ // Inserted here, instead of in Assembler, because it depends on information
+ // in the AST that isn't replicated in the Node structure.
+ static const int kMaxBacksearchLimit = 1024;
+ if (is_end_anchored && !is_start_anchored && !is_sticky &&
+ max_length < kMaxBacksearchLimit) {
+ macro_assembler->SetCurrentPositionFromEnd(max_length);
+ }
+
+ if (is_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);
+ }
+
+ RegExpCompiler::CompilationResult result = compiler.Assemble(
+ isolate, macro_assembler.get(), node, data->capture_count, pattern);
+
+ if (FLAG_correctness_fuzzer_suppressions &&
+ strncmp(result.error_message, "Stack overflow", 15) == 0) {
+ FATAL("Aborting on stack overflow");
+ }
+
+ if (result.error_message != nullptr) {
+ data->error =
+ isolate->factory()->NewStringFromAsciiChecked(result.error_message);
+ }
+ data->code = result.code;
+ data->register_count = result.num_registers;
+
+ return result.Succeeded();
+}
+
+RegExpGlobalCache::RegExpGlobalCache(Handle<JSRegExp> regexp,
+ Handle<String> subject, Isolate* isolate)
+ : register_array_(nullptr),
+ register_array_size_(0),
+ regexp_(regexp),
+ subject_(subject),
+ isolate_(isolate) {
+ bool interpreted = FLAG_regexp_interpret_all;
+
+ if (regexp_->TypeTag() == JSRegExp::ATOM) {
+ static const int kAtomRegistersPerMatch = 2;
+ registers_per_match_ = kAtomRegistersPerMatch;
+ // There is no distinction between interpreted and native for atom regexps.
+ interpreted = false;
+ } else {
+ registers_per_match_ = RegExp::IrregexpPrepare(isolate_, regexp_, subject_);
+ if (registers_per_match_ < 0) {
+ num_matches_ = -1; // Signal exception.
+ return;
+ }
+ }
+
+ DCHECK(IsGlobal(regexp->GetFlags()));
+ if (!interpreted) {
+ register_array_size_ =
+ Max(registers_per_match_, Isolate::kJSRegexpStaticOffsetsVectorSize);
+ max_matches_ = register_array_size_ / registers_per_match_;
+ } else {
+ // Global loop in interpreted regexp is not implemented. We choose
+ // the size of the offsets vector so that it can only store one match.
+ register_array_size_ = registers_per_match_;
+ max_matches_ = 1;
+ }
+
+ if (register_array_size_ > Isolate::kJSRegexpStaticOffsetsVectorSize) {
+ register_array_ = NewArray<int32_t>(register_array_size_);
+ } else {
+ register_array_ = isolate->jsregexp_static_offsets_vector();
+ }
+
+ // Set state so that fetching the results the first time triggers a call
+ // to the compiled regexp.
+ current_match_index_ = max_matches_ - 1;
+ num_matches_ = max_matches_;
+ DCHECK_LE(2, registers_per_match_); // Each match has at least one capture.
+ DCHECK_GE(register_array_size_, registers_per_match_);
+ int32_t* last_match =
+ &register_array_[current_match_index_ * registers_per_match_];
+ last_match[0] = -1;
+ last_match[1] = 0;
+}
+
+RegExpGlobalCache::~RegExpGlobalCache() {
+ // Deallocate the register array if we allocated it in the constructor
+ // (as opposed to using the existing jsregexp_static_offsets_vector).
+ if (register_array_size_ > Isolate::kJSRegexpStaticOffsetsVectorSize) {
+ DeleteArray(register_array_);
+ }
+}
+
+int RegExpGlobalCache::AdvanceZeroLength(int last_index) {
+ if (IsUnicode(regexp_->GetFlags()) && 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;
+}
+
+int32_t* RegExpGlobalCache::FetchNext() {
+ current_match_index_++;
+ if (current_match_index_ >= num_matches_) {
+ // Current batch of results exhausted.
+ // Fail if last batch was not even fully filled.
+ if (num_matches_ < max_matches_) {
+ num_matches_ = 0; // Signal failed match.
+ return nullptr;
+ }
+
+ int32_t* last_match =
+ &register_array_[(current_match_index_ - 1) * registers_per_match_];
+ int last_end_index = last_match[1];
+
+ if (regexp_->TypeTag() == JSRegExp::ATOM) {
+ num_matches_ =
+ RegExpImpl::AtomExecRaw(isolate_, regexp_, subject_, last_end_index,
+ register_array_, register_array_size_);
+ } else {
+ int last_start_index = last_match[0];
+ if (last_start_index == last_end_index) {
+ // Zero-length match. Advance by one code point.
+ last_end_index = AdvanceZeroLength(last_end_index);
+ }
+ if (last_end_index > subject_->length()) {
+ num_matches_ = 0; // Signal failed match.
+ return nullptr;
+ }
+ num_matches_ = RegExpImpl::IrregexpExecRaw(
+ isolate_, regexp_, subject_, last_end_index, register_array_,
+ register_array_size_);
+ }
+
+ if (num_matches_ <= 0) return nullptr;
+ current_match_index_ = 0;
+ return register_array_;
+ } else {
+ return &register_array_[current_match_index_ * registers_per_match_];
+ }
+}
+
+int32_t* RegExpGlobalCache::LastSuccessfulMatch() {
+ int index = current_match_index_ * registers_per_match_;
+ if (num_matches_ == 0) {
+ // After a failed match we shift back by one result.
+ index -= registers_per_match_;
+ }
+ return &register_array_[index];
+}
+
+Object RegExpResultsCache::Lookup(Heap* heap, String key_string,
+ Object key_pattern,
+ FixedArray* last_match_cache,
+ ResultsCacheType type) {
+ FixedArray cache;
+ if (!key_string.IsInternalizedString()) return Smi::kZero;
+ if (type == STRING_SPLIT_SUBSTRINGS) {
+ DCHECK(key_pattern.IsString());
+ if (!key_pattern.IsInternalizedString()) return Smi::kZero;
+ cache = heap->string_split_cache();
+ } else {
+ DCHECK(type == REGEXP_MULTIPLE_INDICES);
+ DCHECK(key_pattern.IsFixedArray());
+ cache = heap->regexp_multiple_cache();
+ }
+
+ uint32_t hash = key_string.Hash();
+ uint32_t index = ((hash & (kRegExpResultsCacheSize - 1)) &
+ ~(kArrayEntriesPerCacheEntry - 1));
+ if (cache.get(index + kStringOffset) != key_string ||
+ cache.get(index + kPatternOffset) != key_pattern) {
+ index =
+ ((index + kArrayEntriesPerCacheEntry) & (kRegExpResultsCacheSize - 1));
+ if (cache.get(index + kStringOffset) != key_string ||
+ cache.get(index + kPatternOffset) != key_pattern) {
+ return Smi::kZero;
+ }
+ }
+
+ *last_match_cache = FixedArray::cast(cache.get(index + kLastMatchOffset));
+ return cache.get(index + kArrayOffset);
+}
+
+void RegExpResultsCache::Enter(Isolate* isolate, Handle<String> key_string,
+ Handle<Object> key_pattern,
+ Handle<FixedArray> value_array,
+ Handle<FixedArray> last_match_cache,
+ ResultsCacheType type) {
+ Factory* factory = isolate->factory();
+ Handle<FixedArray> cache;
+ if (!key_string->IsInternalizedString()) return;
+ if (type == STRING_SPLIT_SUBSTRINGS) {
+ DCHECK(key_pattern->IsString());
+ if (!key_pattern->IsInternalizedString()) return;
+ cache = factory->string_split_cache();
+ } else {
+ DCHECK(type == REGEXP_MULTIPLE_INDICES);
+ DCHECK(key_pattern->IsFixedArray());
+ cache = factory->regexp_multiple_cache();
+ }
+
+ uint32_t hash = key_string->Hash();
+ uint32_t index = ((hash & (kRegExpResultsCacheSize - 1)) &
+ ~(kArrayEntriesPerCacheEntry - 1));
+ if (cache->get(index + kStringOffset) == Smi::kZero) {
+ cache->set(index + kStringOffset, *key_string);
+ cache->set(index + kPatternOffset, *key_pattern);
+ cache->set(index + kArrayOffset, *value_array);
+ cache->set(index + kLastMatchOffset, *last_match_cache);
+ } else {
+ uint32_t index2 =
+ ((index + kArrayEntriesPerCacheEntry) & (kRegExpResultsCacheSize - 1));
+ if (cache->get(index2 + kStringOffset) == Smi::kZero) {
+ cache->set(index2 + kStringOffset, *key_string);
+ cache->set(index2 + kPatternOffset, *key_pattern);
+ cache->set(index2 + kArrayOffset, *value_array);
+ cache->set(index2 + kLastMatchOffset, *last_match_cache);
+ } else {
+ cache->set(index2 + kStringOffset, Smi::kZero);
+ cache->set(index2 + kPatternOffset, Smi::kZero);
+ cache->set(index2 + kArrayOffset, Smi::kZero);
+ cache->set(index2 + kLastMatchOffset, Smi::kZero);
+ cache->set(index + kStringOffset, *key_string);
+ cache->set(index + kPatternOffset, *key_pattern);
+ cache->set(index + kArrayOffset, *value_array);
+ cache->set(index + kLastMatchOffset, *last_match_cache);
+ }
+ }
+ // If the array is a reasonably short list of substrings, convert it into a
+ // list of internalized strings.
+ if (type == STRING_SPLIT_SUBSTRINGS && value_array->length() < 100) {
+ for (int i = 0; i < value_array->length(); i++) {
+ Handle<String> str(String::cast(value_array->get(i)), isolate);
+ Handle<String> internalized_str = factory->InternalizeString(str);
+ value_array->set(i, *internalized_str);
+ }
+ }
+ // Convert backing store to a copy-on-write array.
+ value_array->set_map_no_write_barrier(
+ ReadOnlyRoots(isolate).fixed_cow_array_map());
+}
+
+void RegExpResultsCache::Clear(FixedArray cache) {
+ for (int i = 0; i < kRegExpResultsCacheSize; i++) {
+ cache.set(i, Smi::kZero);
+ }
+}
+
+} // namespace internal
+} // namespace v8
diff --git a/deps/v8/src/regexp/regexp.h b/deps/v8/src/regexp/regexp.h
new file mode 100644
index 0000000000..0f3ed463da
--- /dev/null
+++ b/deps/v8/src/regexp/regexp.h
@@ -0,0 +1,177 @@
+// Copyright 2012 the V8 project authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+#ifndef V8_REGEXP_REGEXP_H_
+#define V8_REGEXP_REGEXP_H_
+
+#include "src/objects/js-regexp.h"
+
+namespace v8 {
+namespace internal {
+
+class RegExpNode;
+class RegExpTree;
+
+// TODO(jgruber): Consider splitting between ParseData and CompileData.
+struct RegExpCompileData {
+ // The parsed AST as produced by the RegExpParser.
+ RegExpTree* tree = nullptr;
+
+ // The compiled Node graph as produced by RegExpTree::ToNode methods.
+ RegExpNode* node = nullptr;
+
+ // The generated code as produced by the compiler. Either a Code object (for
+ // irregexp native code) or a ByteArray (for irregexp bytecode).
+ Object code;
+
+ // True, iff the pattern is a 'simple' atom with zero captures. In other
+ // words, the pattern consists of a string with no metacharacters and special
+ // regexp features, and can be implemented as a standard string search.
+ bool simple = true;
+
+ // True, iff the pattern is anchored at the start of the string with '^'.
+ bool contains_anchor = false;
+
+ // Only use if the pattern contains named captures. If so, this contains a
+ // mapping of capture names to capture indices.
+ Handle<FixedArray> capture_name_map;
+
+ // The error message. Only used if an error occurred during parsing or
+ // compilation.
+ Handle<String> error;
+
+ // The number of capture groups, without the global capture \0.
+ int capture_count = 0;
+
+ // The number of registers used by the generated code.
+ int register_count = 0;
+};
+
+class RegExp final : public AllStatic {
+ public:
+ // Whether the irregexp engine generates native code or interpreter bytecode.
+ static bool GeneratesNativeCode() { return !FLAG_regexp_interpret_all; }
+
+ // Parses the RegExp pattern and prepares the JSRegExp object with
+ // generic data and choice of implementation - as well as what
+ // the implementation wants to store in the data field.
+ // Returns false if compilation fails.
+ V8_WARN_UNUSED_RESULT static MaybeHandle<Object> Compile(
+ Isolate* isolate, Handle<JSRegExp> re, Handle<String> pattern,
+ JSRegExp::Flags flags);
+
+ // See ECMA-262 section 15.10.6.2.
+ // This function calls the garbage collector if necessary.
+ V8_EXPORT_PRIVATE V8_WARN_UNUSED_RESULT static MaybeHandle<Object> Exec(
+ Isolate* isolate, Handle<JSRegExp> regexp, Handle<String> subject,
+ int index, Handle<RegExpMatchInfo> last_match_info);
+
+ // Integral return values used throughout regexp code layers.
+ static constexpr int kInternalRegExpFailure = 0;
+ static constexpr int kInternalRegExpSuccess = 1;
+ static constexpr int kInternalRegExpException = -1;
+ static constexpr int kInternalRegExpRetry = -2;
+
+ enum IrregexpResult {
+ RE_FAILURE = kInternalRegExpFailure,
+ RE_SUCCESS = kInternalRegExpSuccess,
+ RE_EXCEPTION = kInternalRegExpException,
+ };
+
+ // Prepare a RegExp for being executed one or more times (using
+ // IrregexpExecOnce) on the subject.
+ // This ensures that the regexp is compiled for the subject, and that
+ // the subject is flat.
+ // Returns the number of integer spaces required by IrregexpExecOnce
+ // as its "registers" argument. If the regexp cannot be compiled,
+ // an exception is set as pending, and this function returns negative.
+ static int IrregexpPrepare(Isolate* isolate, Handle<JSRegExp> regexp,
+ Handle<String> subject);
+
+ // Set last match info. If match is nullptr, then setting captures is
+ // omitted.
+ static Handle<RegExpMatchInfo> SetLastMatchInfo(
+ Isolate* isolate, Handle<RegExpMatchInfo> last_match_info,
+ Handle<String> subject, int capture_count, int32_t* match);
+
+ V8_EXPORT_PRIVATE static bool CompileForTesting(Isolate* isolate, Zone* zone,
+ RegExpCompileData* input,
+ JSRegExp::Flags flags,
+ Handle<String> pattern,
+ Handle<String> sample_subject,
+ bool is_one_byte);
+
+ V8_EXPORT_PRIVATE static void DotPrintForTesting(const char* label,
+ RegExpNode* node);
+
+ static const int kRegExpTooLargeToOptimize = 20 * KB;
+};
+
+// Uses a special global mode of irregexp-generated code to perform a global
+// search and return multiple results at once. As such, this is essentially an
+// iterator over multiple results (retrieved batch-wise in advance).
+class RegExpGlobalCache final {
+ public:
+ RegExpGlobalCache(Handle<JSRegExp> regexp, Handle<String> subject,
+ Isolate* isolate);
+
+ ~RegExpGlobalCache();
+
+ // Fetch the next entry in the cache for global regexp match results.
+ // This does not set the last match info. Upon failure, nullptr is
+ // returned. The cause can be checked with Result(). The previous result is
+ // still in available in memory when a failure happens.
+ int32_t* FetchNext();
+
+ int32_t* LastSuccessfulMatch();
+
+ bool HasException() { return num_matches_ < 0; }
+
+ private:
+ int AdvanceZeroLength(int last_index);
+
+ int num_matches_;
+ int max_matches_;
+ int current_match_index_;
+ int registers_per_match_;
+ // Pointer to the last set of captures.
+ int32_t* register_array_;
+ int register_array_size_;
+ Handle<JSRegExp> regexp_;
+ Handle<String> subject_;
+ Isolate* isolate_;
+};
+
+// Caches results for specific regexp queries on the isolate. At the time of
+// writing, this is used during global calls to RegExp.prototype.exec and
+// @@split.
+class RegExpResultsCache final : public AllStatic {
+ public:
+ enum ResultsCacheType { REGEXP_MULTIPLE_INDICES, STRING_SPLIT_SUBSTRINGS };
+
+ // Attempt to retrieve a cached result. On failure, 0 is returned as a Smi.
+ // On success, the returned result is guaranteed to be a COW-array.
+ static Object Lookup(Heap* heap, String key_string, Object key_pattern,
+ FixedArray* last_match_out, ResultsCacheType type);
+ // Attempt to add value_array to the cache specified by type. On success,
+ // value_array is turned into a COW-array.
+ static void Enter(Isolate* isolate, Handle<String> key_string,
+ Handle<Object> key_pattern, Handle<FixedArray> value_array,
+ Handle<FixedArray> last_match_cache, ResultsCacheType type);
+ static void Clear(FixedArray cache);
+
+ static constexpr int kRegExpResultsCacheSize = 0x100;
+
+ private:
+ static constexpr int kStringOffset = 0;
+ static constexpr int kPatternOffset = 1;
+ static constexpr int kArrayOffset = 2;
+ static constexpr int kLastMatchOffset = 3;
+ static constexpr int kArrayEntriesPerCacheEntry = 4;
+};
+
+} // namespace internal
+} // namespace v8
+
+#endif // V8_REGEXP_REGEXP_H_