Stop doing unnecessary UTF-8 to UTF-16 conversions in JSONWriter.

base/json/string_escape.cc

 // Copyright (c) 2006-2008 The Chromium Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
 #include "base/json/string_escape.h"
 
 #include <string>
 
 #include "base/strings/string_util.h"
 #include "base/strings/stringprintf.h"
+#include "base/strings/utf_string_conversion_utils.h"
+#include "base/strings/utf_string_conversions.h"
+#include "base/third_party/icu/icu_utf.h"
 
 namespace base {
 
 namespace {
 
-// Try to escape |c| as a "SingleEscapeCharacter" (\n, etc).  If successful,
-// returns true and appends the escape sequence to |dst|.  This isn't required
-// by the spec, but it's more readable by humans than the \uXXXX alternatives.
-template<typename CHAR>
-static bool JsonSingleEscapeChar(const CHAR c, std::string* dst) {
+// Format string for printing a \uXXXX escape sequence.
+const char kU16EscapeFormat[] = "\\u%04X";
+
+// The code point to output for an invalid input code unit.
+const uint32 kReplacementCodePoint = 0xFFFD;
+
+// Try to escape the |code_point| if it is a known special character. If
+// successful, returns true and appends the escape sequence to |dest|. This
+// isn't required by the spec, but it's more readable by humans.
+bool EscapeCodePoint(uint32 code_point, std::string* dest) {

[Mark Mentovai, 2013/12/06 15:35:12]

This name makes it sound like it can escape any code point, but it really only
operates on a few special characters. EscapeSpecialCodePoint?

[jungshik at Google, 2013/12/07 13:16:59]

Agree. I had to come back up here to remind me what this function does while
reading EscapeJSONString() when this function is used. 

[Robert Sesek, 2013/12/09 19:52:09]

Done.

   // WARNING: if you add a new case here, you need to update the reader as well.
   // Note: \v is in the reader, but not here since the JSON spec doesn't
   // allow it.
-  switch (c) {
+  switch (code_point) {
     case '\b':
-      dst->append("\\b");
+      dest->append("\\b");
       break;
     case '\f':
-      dst->append("\\f");
+      dest->append("\\f");
       break;
     case '\n':
-      dst->append("\\n");
+      dest->append("\\n");
       break;
     case '\r':
-      dst->append("\\r");
+      dest->append("\\r");
       break;
     case '\t':
-      dst->append("\\t");
+      dest->append("\\t");
       break;
     case '\\':
-      dst->append("\\\\");
+      dest->append("\\\\");
       break;
     case '"':
-      dst->append("\\\"");
+      dest->append("\\\"");
+      break;
+    // Escape <, > to prevent script execution.

[Mark Mentovai, 2013/12/06 15:35:12]

You really only need < for this, you can save a couple of bytes in the output by
leaving > alone.

[Robert Sesek, 2013/12/09 19:52:09]

Done.

+    case '<':
+      dest->append("\\u003C");
+      break;
+    case '>':
+      dest->append("\\u003E");
       break;
     default:
       return false;
   }
   return true;
 }
 
-template <class STR>
-void JsonDoubleQuoteT(const STR& str,
+template <typename S>
+int32 StringLength(const S& str) {
+  // Casting is necessary because ICU uses int32. Try and do so safely.
+  CHECK_LE(str.length(), static_cast<size_t>(kint32max));

[Mark Mentovai, 2013/12/06 15:35:12]

std::numeric_limits<int32>::max() instead of kint32max?

[Robert Sesek, 2013/12/09 19:52:09]

There seems to be a preference for the constant when using the Chromium
basictypes.h integer types.

+  return static_cast<int32>(str.length());
+}
+
+}  // namespace
+
+void EscapeJSONString(const StringPiece& str,
                       bool put_in_quotes,
-                      std::string* dst) {
+                      std::string* dest) {
   if (put_in_quotes)
-    dst->push_back('"');
+    dest->push_back('"');
 
-  for (typename STR::const_iterator it = str.begin(); it != str.end(); ++it) {
-    typename ToUnsigned<typename STR::value_type>::Unsigned c = *it;
-    if (!JsonSingleEscapeChar(c, dst)) {
-      if (c < 32 || c > 126 || c == '<' || c == '>') {
-        // 1. Escaping <, > to prevent script execution.
-        // 2. Technically, we could also pass through c > 126 as UTF8, but this
-        //    is also optional.  It would also be a pain to implement here.
-        unsigned int as_uint = static_cast<unsigned int>(c);
-        base::StringAppendF(dst, "\\u%04X", as_uint);
+  const int32 length = StringLength(str);
+  for (int32 i = 0; i < length; ++i) {
+    uint32 code_point;
+    if (!ReadUnicodeCharacter(str.data(), length, &i, &code_point))
+      code_point = kReplacementCodePoint;
+
+    if (EscapeCodePoint(code_point, dest))
+      continue;
+
+    // Escape non-printing characters.
+    if (code_point < 32)
+      base::StringAppendF(dest, kU16EscapeFormat, code_point);
+    else
+      WriteUnicodeCharacter(code_point, dest);
+  }
+
+  if (put_in_quotes)
+    dest->push_back('"');
+}
+
+void EscapeJSONString(const StringPiece16& str,
+                      bool put_in_quotes,
+                      std::string* dest) {
+  if (put_in_quotes)
+    dest->push_back('"');
+
+  const int32 length = StringLength(str);
+  for (int32 i = 0; i < length; ++i) {
+    uint32 code_point;
+    if (!ReadUnicodeCharacter(str.data(), length, &i, &code_point))
+      code_point = kReplacementCodePoint;

[jungshik at Google, 2013/12/07 13:16:59]

ReadUnicodeCharacter passes througn non-character code points. I think it's all
right (see http://www.unicode.org/faq/private_use.html#noncharacters ). 

[Robert Sesek, 2013/12/09 19:52:09]

base::StringValue performs a DCHECK(IsStrintUTF8()) when being created, which
ensures that all characters are IsValidCharacter(). Also, considering:

"""
Q: Are noncharacters intended for interchange?
A: No. They are intended explicity for internal use.
"""

and that JSON is an interchange format, I don't think it makes sense to do this.

+
+    if (EscapeCodePoint(code_point, dest))
+      continue;
+
+    if (code_point < 32 || code_point > 126) {
+      if (CBU16_LENGTH(code_point) == 1) {
+        base::StringAppendF(dest, kU16EscapeFormat, code_point);
       } else {
-        unsigned char ascii = static_cast<unsigned char>(*it);
-        dst->push_back(ascii);
+        base::StringAppendF(dest, kU16EscapeFormat, CBU16_LEAD(code_point));
+        base::StringAppendF(dest, kU16EscapeFormat, CBU16_TRAIL(code_point));
       }
+    } else {
+      dest->push_back(static_cast<char>(code_point));
     }
   }
 
   if (put_in_quotes)
-    dst->push_back('"');
+    dest->push_back('"');
 }
 
-}  // namespace
-
-void JsonDoubleQuote(const StringPiece& str,
-                     bool put_in_quotes,
-                     std::string* dst) {
-  JsonDoubleQuoteT(str, put_in_quotes, dst);
+std::string GetQuotedJSONString(const StringPiece& str) {
+  std::string dest;
+  EscapeJSONString(str, true, &dest);
+  return dest;
 }
 
-std::string GetDoubleQuotedJson(const StringPiece& str) {
-  std::string dst;
-  JsonDoubleQuote(str, true, &dst);
-  return dst;
+std::string GetQuotedJSONString(const StringPiece16& str) {
+  std::string dest;
+  EscapeJSONString(str, true, &dest);
+  return dest;
 }
 
-void JsonDoubleQuote(const StringPiece16& str,
-                     bool put_in_quotes,
-                     std::string* dst) {
-  JsonDoubleQuoteT(str, put_in_quotes, dst);
-}
+std::string EscapeBytesAsInvalidJSONString(const StringPiece& str,
+                                           bool put_in_quotes) {
+  std::string dest;
 
-std::string GetDoubleQuotedJson(const StringPiece16& str) {
-  std::string dst;
-  JsonDoubleQuote(str, true, &dst);
-  return dst;
+  if (put_in_quotes)
+    dest.push_back('"');
+
+  for (StringPiece::const_iterator it = str.begin(); it != str.end(); ++it) {
+    ToUnsigned<StringPiece::value_type>::Unsigned c = *it;
+    if (EscapeCodePoint(c, &dest))
+      continue;
+
+    if (c < 32 || c > 126)
+      base::StringAppendF(&dest, kU16EscapeFormat, c);
+    else
+      dest.push_back(*it);
+  }
+
+  if (put_in_quotes)
+    dest.push_back('"');
+
+  return dest;
 }
 
 }  // namespace base