[regexp] Support unicode capture names in non-unicode patterns

src/regexp/regexp-parser.cc

Index: src/regexp/regexp-parser.cc
diff --git a/src/regexp/regexp-parser.cc b/src/regexp/regexp-parser.cc
index 68c5a6d2633cd31ee8ee17c0487ec87ea44d7b7d..b6250b98b51319cbcfa3d721420ed575f9ca0acf 100644
--- a/src/regexp/regexp-parser.cc
+++ b/src/regexp/regexp-parser.cc
@@ -47,12 +47,13 @@ RegExpParser::RegExpParser(FlatStringReader* in, Handle<String>* error,
 }
 
 template <bool update_position>

[Yang, 2017/04/05 12:52:38]

I wonder whether it makes sense to make the scan mode a template param. Or
alternatively make this bool a normal parameter. I sort of doubt that this has
any performance implications.

[jgruber, 2017/04/05 12:57:30]

Agreed, will do that.

[jgruber, 2017/04/07 07:10:56]

Done.

-inline uc32 RegExpParser::ReadNext() {
+inline uc32 RegExpParser::ReadNext(ScanMode mode) {
   int position = next_pos_;
   uc32 c0 = in()->Get(position);
   position++;
-  // Read the whole surrogate pair in case of unicode flag, if possible.
-  if (unicode() && position < in()->length() &&
+  const bool try_combine_surrogate_pairs =
+      (unicode() || mode == ScanMode::FORCE_COMBINE_SURROGATE_PAIRS);
+  if (try_combine_surrogate_pairs && position < in()->length() &&
       unibrow::Utf16::IsLeadSurrogate(static_cast<uc16>(c0))) {
     uc16 c1 = in()->Get(position);
     if (unibrow::Utf16::IsTrailSurrogate(c1)) {
@@ -67,14 +68,13 @@ inline uc32 RegExpParser::ReadNext() {
 
 uc32 RegExpParser::Next() {
   if (has_next()) {
-    return ReadNext<false>();
+    return ReadNext<false>(ScanMode::DEFAULT);
   } else {
     return kEndMarker;
   }
 }
 
-
-void RegExpParser::Advance() {
+void RegExpParser::Advance(ScanMode mode) {
   if (has_next()) {
     StackLimitCheck check(isolate());
     if (check.HasOverflowed()) {
@@ -84,7 +84,7 @@ void RegExpParser::Advance() {
     } else if (zone()->excess_allocation()) {
       ReportError(CStrVector("Regular expression too large"));
     } else {
-      current_ = ReadNext<true>();
+      current_ = ReadNext<true>(mode);
     }
   } else {
     current_ = kEndMarker;
@@ -102,10 +102,9 @@ void RegExpParser::Reset(int pos) {
   Advance();
 }
 
-
-void RegExpParser::Advance(int dist) {
+void RegExpParser::Advance(int dist, ScanMode mode) {
   next_pos_ += dist - 1;
-  Advance();
+  Advance(mode);
 }
 
 
@@ -329,7 +328,6 @@ RegExpTree* RegExpParser::ParseDisjunction() {
               if (FLAG_harmony_regexp_named_captures) {
                 has_named_captures_ = true;
                 is_named_capture = true;
-                Advance();

[Yang, 2017/04/05 12:52:38]

How come we can remove this Advance here and below?

[jgruber, 2017/04/05 12:57:30]

The advance is now done in ParseCaptureGroupName. That was necessary since
advancing from the '<' already needs to use the FORCE_COMBINE_SURROGATE_PAIRS
mode.

                 break;
               }
             // Fall through.
@@ -769,20 +767,26 @@ static void push_code_unit(ZoneVector<uc16>* v, uint32_t code_unit) {
 
 const ZoneVector<uc16>* RegExpParser::ParseCaptureGroupName() {
   DCHECK(FLAG_harmony_regexp_named_captures);
+  DCHECK_EQ(current(), '<');
 
   ZoneVector<uc16>* name =
       new (zone()->New(sizeof(ZoneVector<uc16>))) ZoneVector<uc16>(zone());
 
+  // Capture names can always contain surrogate pairs, and we need to scan
+  // accordingly.
+  const ScanMode scan_mode = ScanMode::FORCE_COMBINE_SURROGATE_PAIRS;
+  Advance(scan_mode);
+
   bool at_start = true;
   while (true) {
     uc32 c = current();
-    Advance();
+    Advance(scan_mode);
 
     // Convert unicode escapes.
     if (c == '\\' && current() == 'u') {
       // TODO(jgruber): Reconsider this once the spec has settled.
       // https://github.com/tc39/proposal-regexp-named-groups/issues/23
-      Advance();
+      Advance(scan_mode);
       if (!ParseUnicodeEscape(&c)) {
         ReportError(CStrVector("Invalid Unicode escape sequence"));
         return nullptr;
@@ -847,7 +851,6 @@ bool RegExpParser::ParseNamedBackReference(RegExpBuilder* builder,
     return false;
   }
 
-  Advance();
   const ZoneVector<uc16>* name = ParseCaptureGroupName();
   if (name == nullptr) {
     return false;