[regexp] implement case-insensitive unicode regexps.

src/regexp/jsregexp.cc

Index: src/regexp/jsregexp.cc
diff --git a/src/regexp/jsregexp.cc b/src/regexp/jsregexp.cc
index 3559bcd111a4c19eb5755dd264d6a4a8cf39a88a..2229dd8e175cc9483741ad75bff773a0c0a7797f 100644
--- a/src/regexp/jsregexp.cc
+++ b/src/regexp/jsregexp.cc
@@ -25,6 +25,11 @@
 #include "src/string-search.h"
 #include "src/unicode-decoder.h"
 
+#ifdef V8_I18N_SUPPORT
+#include "unicode/uset.h"
+#include "unicode/utypes.h"
+#endif  // V8_I18N_SUPPORT
+
 #ifndef V8_INTERPRETED_REGEXP
 #if V8_TARGET_ARCH_IA32
 #include "src/regexp/ia32/regexp-macro-assembler-ia32.h"
@@ -4390,8 +4395,8 @@ void BackReferenceNode::Emit(RegExpCompiler* compiler, Trace* trace) {
 
   DCHECK_EQ(start_reg_ + 1, end_reg_);
   if (compiler->ignore_case()) {
-    assembler->CheckNotBackReferenceIgnoreCase(start_reg_, read_backward(),
-                                               trace->backtrack());
+    assembler->CheckNotBackReferenceIgnoreCase(
+        start_reg_, read_backward(), compiler->unicode(), trace->backtrack());
   } else {
     assembler->CheckNotBackReference(start_reg_, read_backward(),
                                      trace->backtrack());
@@ -4861,21 +4866,6 @@ bool RegExpCharacterClass::is_standard(Zone* zone) {
 }
 
 
-bool RegExpCharacterClass::NeedsDesugaringForUnicode(Zone* zone) {
-  ZoneList<CharacterRange>* ranges = this->ranges(zone);
-  CharacterRange::Canonicalize(ranges);
-  for (int i = ranges->length() - 1; i >= 0; i--) {
-    uc32 from = ranges->at(i).from();
-    uc32 to = ranges->at(i).to();
-    // Check for non-BMP characters.
-    if (to >= kNonBmpStart) return true;
-    // Check for lone surrogates.
-    if (from <= kTrailSurrogateEnd && to >= kLeadSurrogateStart) return true;
-  }
-  return false;
-}
-
-
 UnicodeRangeSplitter::UnicodeRangeSplitter(Zone* zone,
                                            ZoneList<CharacterRange>* base)
     : zone_(zone),
@@ -5086,11 +5076,49 @@ void AddLoneTrailSurrogates(RegExpCompiler* compiler, ChoiceNode* result,
 }
 
 
+void AddUnicodeCaseEquivalents(RegExpCompiler* compiler,
+                               ZoneList<CharacterRange>* ranges) {
+#ifdef V8_I18N_SUPPORT
+  // Use ICU to compute the case fold closure over the ranges.
+  DCHECK(compiler->unicode());
+  DCHECK(compiler->ignore_case());
+  USet* set = uset_openEmpty();
+  for (int i = 0; i < ranges->length(); i++) {
+    uset_addRange(set, ranges->at(i).from(), ranges->at(i).to());
+  }
+  ranges->Clear();
+  uset_closeOver(set, USET_CASE_INSENSITIVE);
+  // Full case mapping map single characters to multiple characters.
+  // Those are represented as strings in the set. Remove them so that
+  // we end up with only simple and common case mappings.
+  uset_removeAllStrings(set);
+  int item_count = uset_getItemCount(set);
+  int item_result = 0;
+  UErrorCode ec = U_ZERO_ERROR;
+  Zone* zone = compiler->zone();
+  for (int i = 0; i < item_count; i++) {
+    uc32 start = 0;
+    uc32 end = 0;
+    item_result += uset_getItem(set, i, &start, &end, nullptr, 0, &ec);
+    ranges->Add(CharacterRange::Range(start, end), zone);
+  }
+  // No errors and everything we collected have been ranges.
+  DCHECK_EQ(U_ZERO_ERROR, ec);
+  DCHECK_EQ(0, item_result);
+  uset_close(set);
+  CharacterRange::Canonicalize(ranges);
+#endif  // V8_I18N_SUPPORT
+}
+
+
 RegExpNode* RegExpCharacterClass::ToNode(RegExpCompiler* compiler,
                                          RegExpNode* on_success) {
   set_.Canonicalize();
   Zone* zone = compiler->zone();
   ZoneList<CharacterRange>* ranges = this->ranges(zone);
+  if (compiler->unicode() && compiler->ignore_case()) {
+    AddUnicodeCaseEquivalents(compiler, ranges);
+  }
   if (compiler->unicode() && !compiler->one_byte()) {
     if (is_negated()) {
       ZoneList<CharacterRange>* negated =