String:WriteUtf8: Add REPLACE_INVALID_UTF8 option

src/unicode.h

Index: src/unicode.h
diff --git a/src/unicode.h b/src/unicode.h
index 6ba61d0e17b2a0b6ff7702c422c38ddf9276318a..f1bc2c77c1e9d9ea720c4076bfc5405e61b66e65 100644
--- a/src/unicode.h
+++ b/src/unicode.h
@@ -46,6 +46,12 @@ typedef unsigned char byte;
  */
 const int kMaxMappingSize = 4;
 
+/**
+ * The unicode replacement character, used to signal invalid unicode sequences
+ * (e.g. an orphan surrogate) when converting to a UTF encoding.
+ */
+const int kReplacementCharacter = 0xFFFD;

[dcarney, 2014/01/04 15:56:45]

should probably be in the Utf8 class

[haimuiba, 2014/01/06 05:40:18]

I don't feel strongly about it, but technically it's not directly related to
UTF-8, but rather provided for all unicode encoders/decoders that might need it.

+
 template <class T, int size = 256>
 class Predicate {
  public:
@@ -102,6 +108,9 @@ class UnicodeData {
 
 class Utf16 {
  public:
+  static inline bool IsSurrogatePair(int lead, int trail) {
+    return IsLeadSurrogate(lead) && IsTrailSurrogate(trail);
+  }
   static inline bool IsLeadSurrogate(int code) {
     if (code == kNoPreviousCharacter) return false;
     return (code & 0xfc00) == 0xd800;
@@ -146,8 +155,7 @@ class Utf8 {
  public:
   static inline uchar Length(uchar chr, int previous);
   static inline unsigned EncodeOneByte(char* out, uint8_t c);
-  static inline unsigned Encode(
-      char* out, uchar c, int previous);
+  static inline unsigned Encode(char* str, uchar c, bool replace_surrogates);

[dcarney, 2014/01/04 15:56:45]

this is not backwards compatible.  All existing call sites which you did not cut
over are casting int to bool for the third argument

[haimuiba, 2014/01/06 05:40:18]

Is there a good reason to keep compatibility, rather than changing the call
sites?

   static uchar CalculateValue(const byte* str,
                               unsigned length,
                               unsigned* cursor);