Adds option for JSON reader to allow invalid utf characters

base/json/json_parser.cc

Index: base/json/json_parser.cc
diff --git a/base/json/json_parser.cc b/base/json/json_parser.cc
index 9ce6a2c1e73fa7105402ee84580eb564bcddd700..ca0015ceb66efdffde7bae2de042628a8144867a 100644
--- a/base/json/json_parser.cc
+++ b/base/json/json_parser.cc
@@ -626,11 +626,19 @@ bool JSONParser::ConsumeStringRaw(StringBuilder* out) {
   int32_t next_char = 0;
 
   while (CanConsume(1)) {
+    const int start_index = index_;

[sky, 2016/11/03 19:14:14]

It may be possible to backtrack for the first valid character, but I'm keeping
it simple.

     pos_ = start_pos_ + index_;  // CBU8_NEXT is postcrement.
     CBU8_NEXT(start_pos_, index_, length, next_char);
     if (next_char < 0 || !IsValidCharacter(next_char)) {
-      ReportError(JSONReader::JSON_UNSUPPORTED_ENCODING, 1);
-      return false;
+      if ((options_ & JSON_ALLOW_NON_UTF_STRINGS) == 0) {
+        ReportError(JSONReader::JSON_UNSUPPORTED_ENCODING, 1);
+        return false;
+      }
+      string.Convert();
+      std::string invalid_string(index_ - start_index, 0);
+      memcpy(&invalid_string.front(), pos_, index_ - start_index);

[brettw, 2016/11/04 23:17:13]

Personally, I would prefer substituting it with the "Unicode replacement
character" than keeping it as-is. Various Chrome code assumes strings are
encoded properly, and I think this will be safer.

The replacement character = U+FFFD.

[sky, 2016/11/08 01:02:20]

Done.

+      string.AppendString(invalid_string);
+      continue;
     }
 
     if (next_char == '"') {