summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
-rw-r--r--tools/source/misc/json_writer.cxx135
1 files changed, 79 insertions, 56 deletions
diff --git a/tools/source/misc/json_writer.cxx b/tools/source/misc/json_writer.cxx
index 09f34c25c3c7..7024b580c7fd 100644
--- a/tools/source/misc/json_writer.cxx
+++ b/tools/source/misc/json_writer.cxx
@@ -119,6 +119,56 @@ void JsonWriter::endStruct()
mbFirstFieldInNode = false;
}
+static char getEscapementChar(char ch)
+{
+ switch (ch)
+ {
+ case '\b':
+ return 'b';
+ case '\t':
+ return 't';
+ case '\n':
+ return 'n';
+ case '\f':
+ return 'f';
+ case '\r':
+ return 'r';
+ default:
+ return ch;
+ }
+}
+
+static bool writeEscapedSequence(sal_uInt32 ch, char*& pos)
+{
+ switch (ch)
+ {
+ case '\b':
+ case '\t':
+ case '\n':
+ case '\f':
+ case '\r':
+ case '"':
+ case '/':
+ case '\\':
+ *pos++ = '\\';
+ *pos++ = getEscapementChar(ch);
+ return true;
+ // Special processing of U+2028 and U+2029, which are valid JSON, but invalid JavaScript
+ // Write them in escaped '\u2028' or '\u2029' form
+ case 0x2028:
+ case 0x2029:
+ *pos++ = '\\';
+ *pos++ = 'u';
+ *pos++ = '2';
+ *pos++ = '0';
+ *pos++ = '2';
+ *pos++ = ch == 0x2028 ? '8' : '9';
+ return true;
+ default:
+ return false;
+ }
+}
+
void JsonWriter::writeEscapedOUString(const OUString& rPropVal)
{
// Convert from UTF-16 to UTF-8 and perform escaping
@@ -126,42 +176,9 @@ void JsonWriter::writeEscapedOUString(const OUString& rPropVal)
while (i < rPropVal.getLength())
{
sal_uInt32 ch = rPropVal.iterateCodePoints(&i);
- if (ch == '\\')
- {
- *mPos = static_cast<char>(ch);
- ++mPos;
- *mPos = static_cast<char>(ch);
- ++mPos;
- }
- else if (ch == '"')
- {
- *mPos = '\\';
- ++mPos;
- *mPos = static_cast<char>(ch);
- ++mPos;
- }
- else if (ch == '\n')
- {
- *mPos = '\\';
- ++mPos;
- *mPos = 'n';
- ++mPos;
- }
- else if (ch == '\r')
- {
- *mPos = '\\';
- ++mPos;
- *mPos = 'r';
- ++mPos;
- }
- else if (ch == '\f')
- {
- *mPos = '\\';
- ++mPos;
- *mPos = 'f';
- ++mPos;
- }
- else if (ch <= 0x7F)
+ if (writeEscapedSequence(ch, mPos))
+ continue;
+ if (ch <= 0x7F)
{
*mPos = static_cast<char>(ch);
++mPos;
@@ -200,9 +217,8 @@ void JsonWriter::put(const char* pPropName, const OUString& rPropVal)
{
auto nPropNameLength = strlen(pPropName);
// But values can be any UTF-8,
- // see rtl_ImplGetFastUTF8ByteLen in sal/rtl/string.cxx for why a factor 3
- // is the worst case
- auto nWorstCasePropValLength = rPropVal.getLength() * 3;
+ // if the string only contains of 0x2028, it will be expanded 6 times (see writeEscapedSequence)
+ auto nWorstCasePropValLength = rPropVal.getLength() * 6;
ensureSpace(nPropNameLength + nWorstCasePropValLength + 8);
addCommaBeforeField();
@@ -241,24 +257,31 @@ void JsonWriter::put(const char* pPropName, std::string_view rPropVal)
for (size_t i = 0; i < rPropVal.size(); ++i)
{
char ch = rPropVal[i];
- if (ch == '\\')
- {
- *mPos = ch;
- ++mPos;
- *mPos = ch;
- ++mPos;
- }
- else if (ch == '"')
+ switch (ch)
{
- *mPos = '\\';
- ++mPos;
- *mPos = ch;
- ++mPos;
- }
- else
- {
- *mPos = ch;
- ++mPos;
+ case '\b':
+ case '\t':
+ case '\n':
+ case '\f':
+ case '\r':
+ case '"':
+ case '/':
+ case '\\':
+ writeEscapedSequence(ch, mPos);
+ break;
+ case '\xE2': // Special processing of U+2028 and U+2029
+ if (i + 2 < rPropVal.size() && rPropVal[i + 1] == '\x80'
+ && (rPropVal[i + 2] == '\xA8' || rPropVal[i + 2] == '\xA9'))
+ {
+ writeEscapedSequence(rPropVal[i + 2] == '\xA8' ? 0x2028 : 0x2029, mPos);
+ i += 2;
+ break;
+ }
+ [[fallthrough]];
+ default:
+ *mPos = ch;
+ ++mPos;
+ break;
}
}