[regexp] implement /ui to mirror the implementation for /i.

src/regexp/jsregexp.cc

 // 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 "src/ast/ast.h"
 #include "src/base/platform/platform.h"
 #include "src/compilation-cache.h"
 #include "src/compiler.h"
@@ skipping 1570 matching lines @@
       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 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;
+static int GetCaseIndependentLetters(RegExpCompiler* compiler, uc16 character,
+                                     bool one_byte_subject, uc32* letters) {

[erikcorry, 2016/02/01 15:21:40]

The compiler knows whether we have a one-byte subject, so we don't need that
argument any more.

[Yang, 2016/02/02 06:06:48]

Done.

+  int length;
+#ifdef V8_I18N_SUPPORT
+  if (compiler->unicode()) {
+    USet* set = uset_open(character, character);
+    uset_closeOver(set, USET_CASE_INSENSITIVE);
+    uset_removeAllStrings(set);
+    length = uset_size(set);
+    for (int i = 0; i < length; i++) {
+      letters[i] = uset_charAt(set, i);
+    }
+    uset_close(set);
+  } else  // NOLINT
+  // Fallback in case ICU is not included.
+#endif  // V8_I18N_SUPPORT
+  {
+    length = compiler->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;
 }
 
-
-static inline bool EmitSimpleCharacter(Isolate* isolate,
-                                       RegExpCompiler* compiler,
-                                       uc16 c,
-                                       Label* on_failure,
-                                       int cp_offset,
-                                       bool check,
-                                       bool preloaded) {
+static inline bool EmitSimpleCharacter(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) {
+static inline bool EmitAtomNonLetter(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[unibrow::Ecma262UnCanonicalize::kMaxWidth];
-  int length = GetCaseIndependentLetters(isolate, c, one_byte, chars);
+  int length = GetCaseIndependentLetters(compiler, c, one_byte, chars);
   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) {

[erikcorry, 2016/02/01 15:21:40]

I think this is still worth having.  For these characters we will emit an
unconditional backtrack elsewhere (they can never match on a one-byte subject
string) so it is just bloat to emit code here to load and check them.

We could do more to avoid outputting code in that case, but this looks like an
OK thing to have here.

[Yang, 2016/02/02 06:06:48]

I'm confused here. Wouldn't the FilterOneByte pass remove this case before we
reach here, unless FilterOneByte gives up due to too deep recursion?

I put this line back just in case you disagree.

-      // 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;
 }
 
 
@@ skipping 26 matching lines @@
     uc16 mask = char_mask ^ diff;
     macro_assembler->CheckNotCharacterAfterMinusAnd(c1 - diff,
                                                     diff,
                                                     mask,
                                                     on_failure);
     return true;
   }
   return false;
 }
 
-
-typedef bool EmitCharacterFunction(Isolate* isolate,
-                                   RegExpCompiler* compiler,
-                                   uc16 c,
-                                   Label* on_failure,
-                                   int cp_offset,
-                                   bool check,
+typedef bool EmitCharacterFunction(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,
+static inline bool EmitAtomLetter(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[unibrow::Ecma262UnCanonicalize::kMaxWidth];
-  int length = GetCaseIndependentLetters(isolate, c, one_byte, chars);
+  int length = GetCaseIndependentLetters(compiler, c, one_byte, chars);
   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;
   DCHECK(unibrow::Ecma262UnCanonicalize::kMaxWidth == 4);
   switch (length) {
     case 2: {
@@ skipping 528 matching lines @@
 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,
+void ActionNode::FillInBMInfo(RegExpCompiler* compiler, int offset, int budget,
                               BoyerMooreLookahead* bm, bool not_at_start) {
   if (action_type_ == BEGIN_SUBMATCH) {
     bm->SetRest(offset);
   } else if (action_type_ != POSITIVE_SUBMATCH_SUCCESS) {
-    on_success()->FillInBMInfo(isolate, offset, budget - 1, bm, not_at_start);
+    on_success()->FillInBMInfo(compiler, 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) {
+void AssertionNode::FillInBMInfo(RegExpCompiler* compiler, 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);
+  on_success()->FillInBMInfo(compiler, 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,
@@ skipping 193 matching lines @@
 // 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 (compiler->ignore_case()) {
           unibrow::uchar chars[unibrow::Ecma262UnCanonicalize::kMaxWidth];
-          int length = GetCaseIndependentLetters(isolate, c,
+          int length = GetCaseIndependentLetters(compiler, c,
                                                  compiler->one_byte(), chars);
           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
@@ skipping 185 matching lines @@
     DCHECK(!info->visited);
     info->visited = true;
   }
   ~VisitMarker() {
     info_->visited = false;
   }
  private:
   NodeInfo* info_;
 };
 
-
-RegExpNode* SeqRegExpNode::FilterOneByte(int depth, bool ignore_case) {
+RegExpNode* SeqRegExpNode::FilterOneByte(int depth, RegExpCompiler* compiler) {
   if (info()->replacement_calculated) return replacement();
   if (depth < 0) return this;
   DCHECK(!info()->visited);
   VisitMarker marker(info());
-  return FilterSuccessor(depth - 1, ignore_case);
+  return FilterSuccessor(depth - 1, compiler);
 }
 
-
-RegExpNode* SeqRegExpNode::FilterSuccessor(int depth, bool ignore_case) {
-  RegExpNode* next = on_success_->FilterOneByte(depth - 1, ignore_case);
+RegExpNode* SeqRegExpNode::FilterSuccessor(int depth,
+                                           RegExpCompiler* compiler) {
+  RegExpNode* next = on_success_->FilterOneByte(depth - 1, compiler);
   if (next == NULL) return set_replacement(NULL);
   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(0x39c) ||
       range.Contains(0x3bc) || range.Contains(0x178);
 }
 
 
 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;
 }
 
+static uc16 ConvertNonLatin1ToEquivalentLatin1(bool unicode, uc16 c) {
+#ifdef V8_I18N_SUPPORT
+  if (unicode) {
+    USet* set = uset_open(c, c);
+    uset_closeOver(set, USET_CASE_INSENSITIVE);
+    uset_removeAllStrings(set);
+    int length = uset_size(set);
+    uc16 result = 0;
+    for (int i = 0; i < length; i++) {
+      uc32 c = uset_charAt(set, i);
+      if (c <= String::kMaxOneByteCharCode) {
+        result = static_cast<uc16>(c);
+        break;
+      }
+    }
+    uset_close(set);
+    return result;
+  }
+  // Fallback to unibrow if ICU is not included.
+#endif  // V8_I18N_SUPPORT
+  return unibrow::Latin1::ConvertNonLatin1ToLatin1(c);
+}
 
-RegExpNode* TextNode::FilterOneByte(int depth, bool ignore_case) {
+RegExpNode* TextNode::FilterOneByte(int depth, RegExpCompiler* compiler) {
   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];
+        uc16 c = quarks[j];
         if (c <= String::kMaxOneByteCharCode) continue;
-        if (!ignore_case) return set_replacement(NULL);
+        if (!compiler->ignore_case()) return set_replacement(NULL);
         // Here, we need to check for characters whose upper and lower cases
         // are outside the Latin-1 range.
-        uint16_t converted = unibrow::Latin1::ConvertNonLatin1ToLatin1(c);
+        uc16 converted =
+            ConvertNonLatin1ToEquivalentLatin1(compiler->unicode(), c);
         // Character is outside Latin-1 completely
         if (converted == 0) return set_replacement(NULL);
         // Convert quark to Latin-1 in place.
-        uint16_t* copy = const_cast<uint16_t*>(quarks.start());
+        uc16* copy = const_cast<uc16*>(quarks.start());
         copy[j] = converted;
       }
     } 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 (ignore_case && RangesContainLatin1Equivalents(ranges)) continue;
+          if (compiler->ignore_case() && RangesContainLatin1Equivalents(ranges))
+            continue;
           return set_replacement(NULL);
         }
       } else {
         if (range_count == 0 ||
             ranges->at(0).from() > String::kMaxOneByteCharCode) {
           // This will be handled in a later filter.
-          if (ignore_case && RangesContainLatin1Equivalents(ranges)) continue;
+          if (compiler->ignore_case() && RangesContainLatin1Equivalents(ranges))
+            continue;
           return set_replacement(NULL);
         }
       }
     }
   }
-  return FilterSuccessor(depth - 1, ignore_case);
+  return FilterSuccessor(depth - 1, compiler);
 }
 
-
-RegExpNode* LoopChoiceNode::FilterOneByte(int depth, bool ignore_case) {
+RegExpNode* LoopChoiceNode::FilterOneByte(int depth, RegExpCompiler* compiler) {
   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, ignore_case);
+        continue_node_->FilterOneByte(depth - 1, compiler);
     // If we can't continue after the loop then there is no sense in doing the
     // loop.
     if (continue_replacement == NULL) return set_replacement(NULL);
   }
 
-  return ChoiceNode::FilterOneByte(depth - 1, ignore_case);
+  return ChoiceNode::FilterOneByte(depth - 1, compiler);
 }
 
-
-RegExpNode* ChoiceNode::FilterOneByte(int depth, bool ignore_case) {
+RegExpNode* ChoiceNode::FilterOneByte(int depth, RegExpCompiler* compiler) {
   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() != NULL && alternative.guards()->length() != 0) {
       set_replacement(this);
       return this;
     }
   }
 
   int surviving = 0;
   RegExpNode* survivor = NULL;
   for (int i = 0; i < choice_count; i++) {
     GuardedAlternative alternative = alternatives_->at(i);
     RegExpNode* replacement =
-        alternative.node()->FilterOneByte(depth - 1, ignore_case);
+        alternative.node()->FilterOneByte(depth - 1, compiler);
     DCHECK(replacement != this);  // No missing EMPTY_MATCH_CHECK.
     if (replacement != NULL) {
       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, ignore_case);
+        alternatives_->at(i).node()->FilterOneByte(depth - 1, compiler);
     if (replacement != NULL) {
       alternatives_->at(i).set_node(replacement);
       new_alternatives->Add(alternatives_->at(i), zone());
     }
   }
   alternatives_ = new_alternatives;
   return this;
 }
 
-
-RegExpNode* NegativeLookaroundChoiceNode::FilterOneByte(int depth,
-                                                        bool ignore_case) {
+RegExpNode* NegativeLookaroundChoiceNode::FilterOneByte(
+    int depth, RegExpCompiler* compiler) {
   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, ignore_case);
+  RegExpNode* replacement = node->FilterOneByte(depth - 1, compiler);
   if (replacement == NULL) return set_replacement(NULL);
   alternatives_->at(1).set_node(replacement);
 
   RegExpNode* neg_node = alternatives_->at(0).node();
-  RegExpNode* neg_replacement = neg_node->FilterOneByte(depth - 1, ignore_case);
+  RegExpNode* neg_replacement = neg_node->FilterOneByte(depth - 1, compiler);
   // If the negative lookahead is always going to fail then
   // we don't need to check it.
   if (neg_replacement == NULL) 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) {
+void LoopChoiceNode::FillInBMInfo(RegExpCompiler* compiler, 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);
+  ChoiceNode::FillInBMInfo(compiler, 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();
@@ skipping 71 matching lines @@
     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 == NULL) {
     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);
+      FillInBMInfo(compiler, 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;
   }
@@ skipping 151 matching lines @@
 // 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) {
       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 = NULL;
         switch (pass) {
           case NON_LATIN1_MATCH:
             DCHECK(one_byte);
-            if (quarks[j] > String::kMaxOneByteCharCode) {
+            if (quarks[j] > String::kMaxOneByteCharCode &&
+                !(compiler->unicode() && compiler->ignore_case())) {
+              // In the non-unicode case, if the pattern is non-Latin1, it

[erikcorry, 2016/02/01 15:21:40]

I think you mean in the one-byte case (non-unicode sounds like you mean in the
absence of the /u flag, but that doesn't fit with the rest of the sentence).

[Yang, 2016/02/02 06:06:48]

I did mean this. With /i, if a character is outside of the latin1 range, it
cannot match a character inside the latin1 range, so we had this optimization
here.
With /ui, we have to be careful not to miss the case where \u212b matches
\u00c5.

Changed the comment to be more clear about this.

+              // cannot possibly match a Latin1 character, unless we have the
+              // /ui flags, e.g. /\u212b/ui matches "\u00c5".
               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 != NULL) {
           bool bounds_check = *checked_up_to < cp_offset + j || read_backward();
           bool bound_checked =
-              emit_function(isolate, compiler, quarks[j], backtrack,
-                            cp_offset + j, bounds_check, preloaded);
+              emit_function(compiler, quarks[j], 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();
@@ skipping 62 matching lines @@
 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()) {

[erikcorry, 2016/02/01 15:21:40]

You should check for the /ui flags here instead of inside TextEmitPass, since
there's no point stepping through the text, only to decide every time that
there's nothing to do.

[Yang, 2016/02/02 06:06:48]

Done.

     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.
@@ skipping 730 matching lines @@
 
   // 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 == NULL) {
     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);
+      alt0.node()->FillInBMInfo(compiler, 0, kRecursionBudget, bm, false);
     }
   }
   if (bm != NULL) {
     bm->EmitSkipInstructions(macro_assembler);
   }
   return eats_at_least;
 }
 
 
 void ChoiceNode::EmitChoices(RegExpCompiler* compiler,
@@ skipping 2236 matching lines @@
 
 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,
+void BackReferenceNode::FillInBMInfo(RegExpCompiler* compiler, 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,
+void ChoiceNode::FillInBMInfo(RegExpCompiler* compiler, 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() != NULL && 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);
+    alt.node()->FillInBMInfo(compiler, 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) {
+void TextNode::FillInBMInfo(RegExpCompiler* compiler, 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 (bm->compiler()->ignore_case()) {
           unibrow::uchar chars[unibrow::Ecma262UnCanonicalize::kMaxWidth];
           int length = GetCaseIndependentLetters(
-              isolate, character, bm->max_char() == String::kMaxOneByteCharCode,
-              chars);
+              compiler, character,
+              bm->max_char() == String::kMaxOneByteCharCode, chars);
           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,
+  on_success()->FillInBMInfo(compiler, 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) {
@@ skipping 137 matching lines @@
 }
 
 
 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 ignore_case = flags & JSRegExp::kIgnoreCase;
   bool is_sticky = flags & JSRegExp::kSticky;
   bool is_global = flags & JSRegExp::kGlobal;
   RegExpCompiler compiler(isolate, zone, data->capture_count, flags,
                           is_one_byte);
 
   if (compiler.optimize()) compiler.set_optimize(!TooMuchRegExpCode(pattern));
 
   // Sample some characters from the middle of the string.
   static const int kSampleSize = 128;
 
@@ skipping 28 matching lines @@
       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('*'), false, loop_node)));
       node = first_step_node;
     } else {
       node = loop_node;
     }
   }
   if (is_one_byte) {
-    node = node->FilterOneByte(RegExpCompiler::kMaxRecursion, ignore_case);
+    node = node->FilterOneByte(RegExpCompiler::kMaxRecursion, &compiler);
     // Do it again to propagate the new nodes to places where they were not
     // put because they had not been calculated yet.
     if (node != NULL) {
-      node = node->FilterOneByte(RegExpCompiler::kMaxRecursion, ignore_case);
+      node = node->FilterOneByte(RegExpCompiler::kMaxRecursion, &compiler);
     }
   } else if (compiler.unicode() && (is_global || is_sticky)) {
     node = OptionallyStepBackToLeadSurrogate(&compiler, node);
   }
 
   if (node == NULL) node = new(zone) EndNode(EndNode::BACKTRACK, zone);
   data->node = node;
   Analysis analysis(isolate, flags, is_one_byte);
   analysis.EnsureAnalyzed(node);
   if (analysis.has_failed()) {
@@ skipping 174 matching lines @@
 
 
 void RegExpResultsCache::Clear(FixedArray* cache) {
   for (int i = 0; i < kRegExpResultsCacheSize; i++) {
     cache->set(i, Smi::FromInt(0));
   }
 }
 
 }  // namespace internal
 }  // namespace v8