Case independent back references.

Reviewed at http://codereview.chromium.org/12406


git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@833 ce2b1a6d-e550-0410-aec6-3dcde31c8c00

src/assembler-irregexp.cc

@@ -239,6 +239,14 @@ void IrregexpAssembler::CheckNotBackReference(int capture_index,
 }
 
 
+void IrregexpAssembler::CheckNotBackReferenceNoCase(int capture_index,
+                                                    Label* on_mismatch) {
+  Emit(BC_CHECK_NOT_BACK_REF_NO_CASE);
+  Emit(capture_index);
+  EmitOrLink(on_mismatch);
+}
+
+
 void IrregexpAssembler::CheckRegister(int byte_code,
                                       int reg_index,
                                       uint16_t vs,

src/assembler-irregexp.h

@@ -75,6 +75,7 @@ class IrregexpAssembler {
   // the register after.  If a register contains -1 then the other register
   // must always contain -1 and the on_mismatch label will never be called.
   void CheckNotBackReference(int capture_index, Label* on_mismatch);
+  void CheckNotBackReferenceNoCase(int capture_index, Label* on_mismatch);
 
   // Checks a register for strictly-less-than or greater-than-or-equal.
   void CheckRegisterLT(int reg_index, uint16_t vs, Label* on_less_than);

src/bytecodes-irregexp.h

@@ -57,12 +57,13 @@ V(MINUS_OR_CHECK_NOT_CHAR, 21, 9) /* minus_or_check_not_char uc16 uc16 ad...*/ \
 V(CHECK_LT,          22, 7) /* check_lt uc16 addr32                         */ \
 V(CHECK_GT,          23, 7) /* check_gr uc16 addr32                         */ \
 V(CHECK_NOT_BACK_REF, 24, 6) /* check_not_back_ref capture_idx addr32       */ \
-V(LOOKUP_MAP1,       25, 11) /* l_map1 start16 bit_map_addr32 addr32        */ \
-V(LOOKUP_MAP2,       26, 99) /* l_map2 start16 half_nibble_map_addr32*      */ \
-V(LOOKUP_MAP8,       27, 99) /* l_map8 start16 byte_map addr32*             */ \
-V(LOOKUP_HI_MAP8,    28, 99) /* l_himap8 start8 byte_map_addr32 addr32*     */ \
-V(CHECK_REGISTER_LT, 29, 8) /* check_reg_lt register_index value16 addr32   */ \
-V(CHECK_REGISTER_GE, 30, 8) /* check_reg_ge register_index value16 addr32   */ \
+V(CHECK_NOT_BACK_REF_NO_CASE, 25, 6) /* check_not_back_ref_no_case captu... */ \
+V(LOOKUP_MAP1,       26, 11) /* l_map1 start16 bit_map_addr32 addr32        */ \
+V(LOOKUP_MAP2,       27, 99) /* l_map2 start16 half_nibble_map_addr32*      */ \
+V(LOOKUP_MAP8,       28, 99) /* l_map8 start16 byte_map addr32*             */ \
+V(LOOKUP_HI_MAP8,    29, 99) /* l_himap8 start8 byte_map_addr32 addr32*     */ \
+V(CHECK_REGISTER_LT, 30, 8) /* check_reg_lt register_index value16 addr32   */ \
+V(CHECK_REGISTER_GE, 31, 8) /* check_reg_ge register_index value16 addr32   */ \
 
 #define DECLARE_BYTECODES(name, code, length) \
   static const int BC_##name = code;

src/interpreter-irregexp.cc

@@ -29,6 +29,7 @@
 
 
 #include "v8.h"
+#include "unicode.h"
 #include "utils.h"
 #include "ast.h"
 #include "bytecodes-irregexp.h"
@@ -38,6 +39,27 @@
 namespace v8 { namespace internal {
 
 
+static unibrow::Mapping<unibrow::Ecma262Canonicalize> canonicalize;
+
+
+static bool BackRefMatchesNoCase(int from,
+                                 int current,
+                                 int len,
+                                 Vector<const uc16> subject) {
+  for (int i = 0; i < len; i++) {
+    unibrow::uchar old_char = subject[from++];
+    unibrow::uchar new_char = subject[current++];
+    if (old_char == new_char) continue;
+    canonicalize.get(old_char, '\0', &old_char);
+    canonicalize.get(new_char, '\0', &new_char);
+    if (old_char != new_char) {
+      return false;
+    }
+  }
+  return true;
+}
+
+
 #ifdef DEBUG
 static void TraceInterpreter(const byte* code_base,
                              const byte* pc,
@@ -319,6 +341,21 @@ static bool RawMatch(const byte* code_base,
         pc += BC_CHECK_NOT_BACK_REF_LENGTH;
         break;
       }
+      BYTECODE(CHECK_NOT_BACK_REF_NO_CASE) {
+        int from = registers[pc[1]];
+        int len = registers[pc[1] + 1] - from;
+        if (current + len > subject.length()) {
+          pc = code_base + Load32(pc + 2);
+          break;
+        } else {
+          if (BackRefMatchesNoCase(from, current, len, subject)) {
+            pc += BC_CHECK_NOT_BACK_REF_NO_CASE_LENGTH;
+          } else {
+            pc = code_base + Load32(pc + 2);
+          }
+        }
+        break;
+      }
       default:
         UNREACHABLE();
         break;

src/jsregexp.cc

@@ -1269,17 +1269,17 @@ static void EmitCharClass(RegExpMacroAssembler* macro_assembler,
     }
   } else {
     if (from != 0) {
-      if (!cc->is_negated()) {
-        macro_assembler->CheckCharacterLT(from, on_failure);
-      } else {
+      if (cc->is_negated()) {
         macro_assembler->CheckCharacterLT(from, &success);
+      } else {
+        macro_assembler->CheckCharacterLT(from, on_failure);
       }
     }
     if (to != 0xffff) {
-      if (!cc->is_negated()) {
-        macro_assembler->CheckCharacterGT(to, on_failure);
-      } else {
+      if (cc->is_negated()) {
         macro_assembler->CheckCharacterLT(to + 1, on_failure);
+      } else {
+        macro_assembler->CheckCharacterGT(to, on_failure);
       }
     } else {
       if (cc->is_negated()) {
@@ -1302,16 +1302,16 @@ bool TextNode::Emit(RegExpCompiler* compiler) {
     TextElement elm = elms_->at(i);
     if (elm.type == TextElement::ATOM) {
       Vector<const uc16> quarks = elm.data.u_atom->data();
-      if (!compiler->is_case_independent()) {
-        macro_assembler->CheckCharacters(quarks,
-                                         cp_offset,
-                                         on_failure_->label());
-      } else {
+      if (compiler->is_case_independent()) {
         EmitAtomNonLetters(macro_assembler,
                            elm,
                            quarks,
                            on_failure_->label(),
                            cp_offset);
+      } else {
+        macro_assembler->CheckCharacters(quarks,
+                                         cp_offset,
+                                         on_failure_->label());
       }
       cp_offset += quarks.length();
     } else {
@@ -1474,7 +1474,12 @@ bool BackReferenceNode::Emit(RegExpCompiler* compiler) {
   macro->IfRegisterLT(start_reg_, 0, on_success()->label());
   macro->IfRegisterLT(end_reg_, 0, on_success()->label());
   ASSERT_EQ(start_reg_ + 1, end_reg_);
-  macro->CheckNotBackReference(start_reg_, on_failure_->label());
+  if (compiler->is_case_independent()) {
+    macro->CheckNotBackReferenceCaseIndependent(start_reg_,
+                                                on_failure_->label());
+  } else {
+    macro->CheckNotBackReference(start_reg_, on_failure_->label());
+  }
   return on_success()->GoTo(compiler);
 }
 
@@ -2140,8 +2145,9 @@ void CharacterRange::AddCaseEquivalents(ZoneList<CharacterRange>* ranges) {
         uc32 c = range[i];
         uc16 range_from = c + (pos - start);
         uc16 range_to = c + (end - start);
-        if (!(from() <= range_from && range_to <= to()))
+        if (!(from() <= range_from && range_to <= to())) {
           ranges->Add(CharacterRange(range_from, range_to));
+        }
       }
       start = pos = block_end + 1;
     }

src/regexp-macro-assembler-ia32.cc

@@ -213,11 +213,14 @@ void RegExpMacroAssemblerIA32::CheckCurrentPosition(int register_index,
 }
 
 
+void RegExpMacroAssemblerIA32::CheckNotBackReferenceCaseIndependent(
+    int start_reg, Label* on_no_match) {
+  UNIMPLEMENTED();
+}
+
+
 void RegExpMacroAssemblerIA32::CheckNotBackReference(
     int start_reg, Label* on_no_match) {
-  if (ignore_case_) {
-    UNIMPLEMENTED();
-  }
   Label fallthrough;
   __ mov(eax, register_location(start_reg));
   __ mov(ecx, register_location(start_reg + 1));

src/regexp-macro-assembler-ia32.h

@@ -53,6 +53,8 @@ class RegExpMacroAssemblerIA32: public RegExpMacroAssembler {
                                Label* on_failure);
   virtual void CheckCurrentPosition(int register_index, Label* on_equal);
   virtual void CheckNotBackReference(int start_reg, Label* on_no_match);
+  virtual void CheckNotBackReferenceCaseIndependent(int start_reg,
+                                                    Label* on_no_match);
   virtual void CheckNotCharacter(uc16 c, Label* on_not_equal);
   virtual void CheckNotCharacterAfterOr(uc16 c, uc16 mask, Label* on_not_equal);
   virtual void CheckNotCharacterAfterMinusOr(uc16 c,

src/regexp-macro-assembler-irregexp.cc

@@ -199,6 +199,13 @@ void RegExpMacroAssemblerIrregexp::CheckNotBackReference(int start_reg,
 }
 
 
+void RegExpMacroAssemblerIrregexp::CheckNotBackReferenceCaseIndependent(
+    int start_reg,
+    Label* on_not_equal) {
+  assembler_->CheckNotBackReferenceNoCase(start_reg, on_not_equal);
+}
+
+
 void RegExpMacroAssemblerIrregexp::CheckBitmap(uc16 start,
                                            Label* bitmap,
                                            Label* on_zero) {

src/regexp-macro-assembler-irregexp.h

@@ -65,6 +65,8 @@ class RegExpMacroAssemblerIrregexp: public RegExpMacroAssembler {
                                              uc16 mask,
                                              Label* on_not_equal);
   virtual void CheckNotBackReference(int start_reg, Label* on_no_match);
+  virtual void CheckNotBackReferenceCaseIndependent(int start_reg,
+                                                    Label* on_no_match);
   virtual void CheckCharacters(Vector<const uc16> str,
                                int cp_offset,
                                Label* on_failure);

src/regexp-macro-assembler.h

@@ -76,6 +76,8 @@ class RegExpMacroAssembler {
       int register_index,
       Label* on_equal) = 0;
   virtual void CheckNotBackReference(int start_reg, Label* on_no_match) = 0;
+  virtual void CheckNotBackReferenceCaseIndependent(int start_reg,
+                                                    Label* on_no_match) = 0;
   // Check the current character for a match with a literal character.  If we
   // fail to match then goto the on_failure label.  End of input always
   // matches.  If the label is NULL then we should pop a backtrack address off