[regexp] Reduce public API surface

This further reduces the number of things declared in the public
regexp API file, currently still named jsregexp.h.

* Move JSRegExp::Flags convenience functions to regexp-compiler.h.
* Set RegExpImpl methods private if possible (these will later be
  moved to a new hidden impl class).
* Merge RegExpEngine::CompilationResult into RegExpCompileData.
* Move remaining RegExpEngine methods to RegExpImpl and delete
  RegExpEngine.
* Extract RegExpGlobalCache.
* Document a few data structures.

Upcoming CLs will rename RegExpImpl to RegExp and jsregexp.h to
regexp.h. This should then be the only header included from other
directories.

Bug: v8:9359
Change-Id: I78c8f4cca495a2b95735a48b6181583bc3310bdf
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/1662294Reviewed-by: Peter Marshall <petermarshall@chromium.org>
Commit-Queue: Jakob Gruber <jgruber@chromium.org>
Cr-Commit-Position: refs/heads/master@{#62218}

BUILD.gn

@@ -2677,7 +2677,6 @@ v8_source_set("v8_base_without_compiler") {
     "src/profiler/tick-sample.h",
     "src/profiler/tracing-cpu-profiler.cc",
     "src/profiler/tracing-cpu-profiler.h",
-    "src/regexp/jsregexp-inl.h",
     "src/regexp/jsregexp.cc",
     "src/regexp/jsregexp.h",
     "src/regexp/property-sequences.cc",

src/regexp/jsregexp-inl.h

-// 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];
-}
-
-}  // namespace internal
-}  // namespace v8
-
-#endif  // V8_REGEXP_JSREGEXP_INL_H_

src/regexp/jsregexp.cc

@@ -6,7 +6,7 @@
 
 #include "src/codegen/compilation-cache.h"
 #include "src/heap/heap-inl.h"
-#include "src/regexp/jsregexp-inl.h"
+#include "src/objects/js-regexp-inl.h"
 #include "src/regexp/regexp-compiler.h"
 #include "src/regexp/regexp-dotprinter.h"
 #include "src/regexp/regexp-interpreter.h"
@@ -290,28 +290,22 @@ bool RegExpImpl::CompileIrregexp(Isolate* isolate, Handle<JSRegExp> re,
     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);
+  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), result.code);
+  data->set(JSRegExp::code_index(is_one_byte), compile_data.code);
   SetIrregexpCaptureNameMap(*data, compile_data.capture_name_map);
   int register_max = IrregexpMaxRegisterCount(*data);
-  if (result.num_registers > register_max) {
-    SetIrregexpMaxRegisterCount(*data, result.num_registers);
+  if (compile_data.register_count > register_max) {
+    SetIrregexpMaxRegisterCount(*data, compile_data.register_count);
   }
 
   return true;
@@ -552,88 +546,59 @@ Handle<RegExpMatchInfo> RegExpImpl::SetLastMatchInfo(
   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;
+void RegExpImpl::DotPrintForTesting(const char* label, RegExpNode* node,
+                                    bool ignore_case) {
+  DotPrinter::DotPrint(label, node, ignore_case);
+}
 
-  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;
-    }
-  }
+namespace {
 
-  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;
-  }
+// 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;
 
-  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;
+  Heap* heap = isolate->heap();
+  if (pattern->length() > RegExpImpl::kRegExpTooLargeToOptimize) return true;
+  return (isolate->total_regexp_code_generated() > kRegExpCompiledLimit &&
+          heap->CommittedMemoryExecutable() > kRegExpExecutableMemoryLimit);
 }
 
-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;
-}
+}  // namespace
 
-void RegExpEngine::DotPrint(const char* label, RegExpNode* node,
-                            bool ignore_case) {
-  DotPrinter::DotPrint(label, node, ignore_case);
+bool RegExpImpl::CompileForTesting(Isolate* isolate, Zone* zone,
+                                   RegExpCompileData* data,
+                                   JSRegExp::Flags flags,
+                                   Handle<String> pattern,
+                                   Handle<String> sample_subject,
+                                   bool is_one_byte) {
+  return Compile(isolate, zone, data, flags, pattern, sample_subject,
+                 is_one_byte);
 }
 
-RegExpEngine::CompilationResult RegExpEngine::Compile(
-    Isolate* isolate, Zone* zone, RegExpCompileData* data,
-    JSRegExp::Flags flags, Handle<String> pattern,
-    Handle<String> sample_subject, bool 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) {
-    return CompilationResult::RegExpTooBig();
+    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())
+  if (compiler.optimize()) {
     compiler.set_optimize(!TooMuchRegExpCode(isolate, pattern));
+  }
 
   // Sample some characters from the middle of the string.
   static const int kSampleSize = 128;
@@ -693,7 +658,9 @@ RegExpEngine::CompilationResult RegExpEngine::Compile(
   Analysis analysis(isolate, is_one_byte);
   analysis.EnsureAnalyzed(node);
   if (analysis.has_failed()) {
-    return CompilationResult(analysis.error_message());
+    data->error =
+        isolate->factory()->NewStringFromAsciiChecked(analysis.error_message());
+    return false;
   }
 
   // Create the correct assembler for the architecture.
@@ -763,20 +730,140 @@ RegExpEngine::CompilationResult RegExpEngine::Compile(
   RegExpCompiler::CompilationResult result = compiler.Assemble(
       isolate, macro_assembler.get(), node, data->capture_count, pattern);
 
-  return RegExpEngine::CompilationResult(result.error_message, result.code,
-                                         result.num_registers);
+  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();
 }
 
-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;
+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_ =
+        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;
+}
+
+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,

src/regexp/jsregexp.h

@@ -13,46 +13,46 @@ namespace internal {
 class RegExpNode;
 class RegExpTree;
 
-inline bool IgnoreCase(JSRegExp::Flags flags) {
-  return (flags & JSRegExp::kIgnoreCase) != 0;
-}
+// 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;
 
-inline bool IsUnicode(JSRegExp::Flags flags) {
-  return (flags & JSRegExp::kUnicode) != 0;
-}
+  // 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;
 
-inline bool IsSticky(JSRegExp::Flags flags) {
-  return (flags & JSRegExp::kSticky) != 0;
-}
+  // True, iff the pattern is anchored at the start of the string with '^'.
+  bool contains_anchor = false;
 
-inline bool IsGlobal(JSRegExp::Flags flags) {
-  return (flags & JSRegExp::kGlobal) != 0;
-}
+  // 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;
 
-inline bool DotAll(JSRegExp::Flags flags) {
-  return (flags & JSRegExp::kDotAll) != 0;
-}
+  // The error message. Only used if an error occurred during parsing or
+  // compilation.
+  Handle<String> error;
 
-inline bool Multiline(JSRegExp::Flags flags) {
-  return (flags & JSRegExp::kMultiline) != 0;
-}
+  // The number of capture groups, without the global capture \0.
+  int capture_count = 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);
-}
+  // The number of registers used by the generated code.
+  int register_count = 0;
+};
 
-class RegExpImpl final {
+class RegExpImpl final : public AllStatic {
  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.
@@ -67,6 +67,43 @@ class RegExpImpl final {
       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);
+
+  // 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,
+                                                   bool ignore_case);
+
+  static const int kRegExpTooLargeToOptimize = 20 * KB;
+
+ private:
+  // 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,
@@ -84,18 +121,6 @@ class RegExpImpl final {
                                  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.
@@ -114,44 +139,17 @@ class RegExpImpl final {
       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);
+  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);
 
-    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_;
-  };
+  // 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);
@@ -163,68 +161,47 @@ class RegExpImpl final {
   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);
-};
-
-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;
+  friend class RegExpGlobalCache;
 };
 
-class RegExpEngine final : public AllStatic {
+// 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:
-  struct CompilationResult {
-    explicit CompilationResult(const char* error_message)
-        : error_message(error_message) {}
-    CompilationResult(const char* error_message, Object code, int registers)
-        : error_message(error_message), code(code), num_registers(registers) {}
-
-    static CompilationResult RegExpTooBig() {
-      return CompilationResult("RegExp too big");
-    }
-
-    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);
+  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 };
@@ -239,14 +216,15 @@ class RegExpResultsCache final : public AllStatic {
                     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;
+
+  static constexpr 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;
+  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

src/regexp/regexp-compiler.cc

@@ -7,7 +7,6 @@
 #include "src/diagnostics/code-tracer.h"
 #include "src/execution/isolate.h"
 #include "src/objects/objects-inl.h"
-#include "src/regexp/jsregexp.h"  // TODO(jgruber): Only needed for IgnoreCase.
 #include "src/regexp/regexp-macro-assembler-arch.h"
 #include "src/regexp/regexp-macro-assembler-tracer.h"
 #include "src/strings/unicode-inl.h"

src/regexp/regexp-compiler.h

@@ -45,6 +45,36 @@ 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);
+}
+
 // A set of unsigned integers that behaves especially well on small
 // integers (< 32).  May do zone-allocation.
 class OutSet : public ZoneObject {
@@ -645,6 +675,8 @@ class RegExpCompiler {
       return CompilationResult("RegExp too big");
     }
 
+    bool Succeeded() const { return error_message == nullptr; }
+
     const char* const error_message = nullptr;
     Object code;
     int num_registers = 0;

src/runtime/runtime-regexp.cc

@@ -11,7 +11,8 @@
 #include "src/logging/counters.h"
 #include "src/numbers/conversions-inl.h"
 #include "src/objects/js-array-inl.h"
-#include "src/regexp/jsregexp-inl.h"
+#include "src/objects/js-regexp-inl.h"
+#include "src/regexp/jsregexp.h"
 #include "src/regexp/regexp-utils.h"
 #include "src/runtime/runtime-utils.h"
 #include "src/strings/string-builder-inl.h"
@@ -638,7 +639,7 @@ V8_WARN_UNUSED_RESULT static Object StringReplaceGlobalRegExpWithString(
     }
   }
 
-  RegExpImpl::GlobalCache global_cache(regexp, subject, isolate);
+  RegExpGlobalCache global_cache(regexp, subject, isolate);
   if (global_cache.HasException()) return ReadOnlyRoots(isolate).exception();
 
   int32_t* current_match = global_cache.FetchNext();
@@ -703,7 +704,7 @@ V8_WARN_UNUSED_RESULT static Object StringReplaceGlobalRegExpWithEmptyString(
     }
   }
 
-  RegExpImpl::GlobalCache global_cache(regexp, subject, isolate);
+  RegExpGlobalCache global_cache(regexp, subject, isolate);
   if (global_cache.HasException()) return ReadOnlyRoots(isolate).exception();
 
   int32_t* current_match = global_cache.FetchNext();
@@ -1115,7 +1116,7 @@ static Object SearchRegExpMultiple(Isolate* isolate, Handle<String> subject,
     }
   }
 
-  RegExpImpl::GlobalCache global_cache(regexp, subject, isolate);
+  RegExpGlobalCache global_cache(regexp, subject, isolate);
   if (global_cache.HasException()) return ReadOnlyRoots(isolate).exception();
 
   // Ensured in Runtime_RegExpExecMultiple.

src/runtime/runtime-strings.cc

@@ -10,7 +10,6 @@
 #include "src/objects/objects-inl.h"
 #include "src/objects/slots.h"
 #include "src/objects/smi.h"
-#include "src/regexp/jsregexp-inl.h"
 #include "src/regexp/regexp-utils.h"
 #include "src/runtime/runtime-utils.h"
 #include "src/strings/string-builder-inl.h"

test/cctest/test-regexp.cc

@@ -547,8 +547,8 @@ static RegExpNode* Compile(const char* input, bool multiline, bool unicode,
                                .ToHandleChecked();
   Handle<String> sample_subject =
       isolate->factory()->NewStringFromUtf8(CStrVector("")).ToHandleChecked();
-  RegExpEngine::Compile(isolate, zone, &compile_data, flags, pattern,
-                        sample_subject, is_one_byte);
+  RegExpImpl::CompileForTesting(isolate, zone, &compile_data, flags, pattern,
+                                sample_subject, is_one_byte);
   return compile_data.node;
 }
 
@@ -561,7 +561,7 @@ static void Execute(const char* input, bool multiline, bool unicode,
   USE(node);
 #ifdef DEBUG
   if (dot_output) {
-    RegExpEngine::DotPrint(input, node, false);
+    RegExpImpl::DotPrintForTesting(input, node, false);
   }
 #endif  // DEBUG
 }