Add support for surrogates when serializing and deserializing for native ports

runtime/vm/unicode.h

Index: runtime/vm/unicode.h
diff --git a/runtime/vm/unicode.h b/runtime/vm/unicode.h
index 03a4b29d879898c2c83f06aac4a175d4729db209..68747fa969a32eb82d87a6a65e2736cb98a861d3 100644
--- a/runtime/vm/unicode.h
+++ b/runtime/vm/unicode.h
@@ -73,6 +73,36 @@ class Utf16 : AllStatic {
 
   static const int32_t kSurrogateOffset = (0x10000 - (0xD800 << 10) - 0xDC00);
 
+  class CodePointIterator {

[cshapiro, 2012/11/30 02:49:08]

Consider Utf16::Next, which is a lighter-weight, and requires no more code at
the use site.

[Søren Gjesse, 2012/11/30 12:23:07]

Are you suggestion a static function like this:
bool Utf16::Next(const uint16_t* utf16_array, intptr_t array_len, int* index,
uint32_t* ch);

And used this way:

uint16_t* data = ...
intptr_t len = ...

intptr_t index = 0;
uint32_t ch = -1;
while (Utf16::Next(data, len, &index, &ch) {
  // ... use ch.
}

Seems simpler to keep the state in a stack allocated object.

[Søren Gjesse, 2012/11/30 13:06:03]

I didn't realize that Utf16::Next was already in there as I did not rebase since
starting this CL. Used that and removed Utf16::CodePointIterator.

+   public:
+    CodePointIterator(const uint16_t* utf16_array, intptr_t array_len)
+        : utf16_array_(utf16_array),
+          array_len_(array_len),
+          index_(-1),
+          ch_(-1) {
+    }
+
+    int32_t Current() const {
+      ASSERT(index_ >= 0);
+      ASSERT(index_ < array_len_);
+      return ch_;
+    }
+
+    bool Next();
+
+    void Reset() {
+      index_ = -1;
+      ch_ = -1;
+    }
+
+   private:
+    const uint16_t* utf16_array_;
+    intptr_t array_len_;
+    intptr_t index_;
+    int32_t ch_;
+    DISALLOW_IMPLICIT_CONSTRUCTORS(CodePointIterator);
+  };
+
   // Returns the length of the code point in UTF-16 code units.
   static intptr_t Length(int32_t ch) {
     return (ch <= kMaxBmpCodepoint) ? 1 : 2;