Optimize non-ASCII string splitting with single-character search pattern

src/objects.h

Index: src/objects.h
diff --git a/src/objects.h b/src/objects.h
index 7d46c068589bfb707a43e93d84bf5cc74e328275..5f04fe1e61d929ba47e86c45e3e804ac5644da0d 100644
--- a/src/objects.h
+++ b/src/objects.h
@@ -7189,6 +7189,13 @@ class String: public HeapObject {
       ASSERT_EQ(TWO_BYTE, state_);
       return Vector<const uc16>::cast(buffer_);
     }
+    // Return the character length of the string.
+    unsigned Length() const {
+      if (state_ == ASCII) return buffer_.length();
+      if (state_ == TWO_BYTE) return buffer_.length() / 2;

[Yang, 2012/11/26 08:45:30]

you could do an
if (state_ == ASCII) {
  ...
} else {
  ASSERT(state == TWO_BYTE) {
  ...
}

to be cleaner.

Even cooler would be without any branching, like,

ASSERT(ASCII == 0 && TWO_BYTE == 1);
ASSERT(state_ == ASCII || state_ == TWO_BYTE);
return buffer_.length() >> state_;

(Make sure the enum elements have values explicitly assigned.)

+      UNREACHABLE();
+      return 0;  // Placate compiler.
+    }
 
    private:
     enum State { NON_FLAT, ASCII, TWO_BYTE };