String TextCodecICU::decode(const char* bytes, size_t length, FlushBehavior flush, bool stopOnError, bool& sawError)
{
// Get a converter for the passed-in encoding.
if (!m_converterICU) {
createICUConverter();
ASSERT(m_converterICU);
if (!m_converterICU) {
WTF_LOG_ERROR("error creating ICU encoder even though encoding was in table");
return String();
}
}
ErrorCallbackSetter callbackSetter(m_converterICU, stopOnError);
StringBuilder result;
UChar buffer[ConversionBufferSize];
UChar* bufferLimit = buffer + ConversionBufferSize;
const char* source = reinterpret_cast<const char*>(bytes);
const char* sourceLimit = source + length;
int32_t* offsets = nullptr;
UErrorCode err = U_ZERO_ERROR;
do {
int ucharsDecoded = decodeToBuffer(buffer, bufferLimit, source, sourceLimit, offsets, flush != DoNotFlush, err);
result.append(buffer, ucharsDecoded);
} while (err == U_BUFFER_OVERFLOW_ERROR);
if (U_FAILURE(err)) {
// flush the converter so it can be reused, and not be bothered by this error.
do {
decodeToBuffer(buffer, bufferLimit, source, sourceLimit, offsets, true, err);
} while (source < sourceLimit);
sawError = true;
}
#if !defined(USING_SYSTEM_ICU)
// Chrome's copy of ICU does not have the issue described below.
return result.toString();
#else
String resultString = result.toString();
// <http://bugs.webkit.org/show_bug.cgi?id=17014>
// Simplified Chinese pages use the code A3A0 to mean "full-width space", but ICU decodes it as U+E5E5.
if (!strcmp(m_encoding.name(), "GBK")) {
if (!strcasecmp(m_encoding.name(), "gb18030"))
resultString.replace(0xE5E5, ideographicSpaceCharacter);
// Make GBK compliant to the encoding spec and align with GB18030
resultString.replace(0x01F9, 0xE7C8);
// FIXME: Once https://www.w3.org/Bugs/Public/show_bug.cgi?id=28740#c3
// is resolved, add U+1E3F => 0xE7C7.
}
return resultString;
#endif
}
String TextCodecICU::decode(const char* bytes, size_t length, bool flush, bool stopOnError, bool& sawError)
{
// Get a converter for the passed-in encoding.
if (!m_converterICU) {
createICUConverter();
ASSERT(m_converterICU);
if (!m_converterICU) {
LOG_ERROR("error creating ICU encoder even though encoding was in table");
return String();
}
}
ErrorCallbackSetter callbackSetter(m_converterICU, stopOnError);
StringBuilder result;
UChar buffer[ConversionBufferSize];
UChar* bufferLimit = buffer + ConversionBufferSize;
const char* source = reinterpret_cast<const char*>(bytes);
const char* sourceLimit = source + length;
int32_t* offsets = NULL;
UErrorCode err = U_ZERO_ERROR;
do {
int ucharsDecoded = decodeToBuffer(buffer, bufferLimit, source, sourceLimit, offsets, flush, err);
result.append(buffer, ucharsDecoded);
} while (err == U_BUFFER_OVERFLOW_ERROR);
if (U_FAILURE(err)) {
// flush the converter so it can be reused, and not be bothered by this error.
do {
decodeToBuffer(buffer, bufferLimit, source, sourceLimit, offsets, true, err);
} while (source < sourceLimit);
sawError = true;
}
String resultString = result.toString();
// <http://bugs.webkit.org/show_bug.cgi?id=17014>
// Simplified Chinese pages use the code A3A0 to mean "full-width space", but ICU decodes it as U+E5E5.
// FIXME: strcasecmp is locale sensitive, we should not be using it.
if (strcmp(m_encodingName, "GBK") == 0 || strcasecmp(m_encodingName, "gb18030") == 0)
resultString.replace(0xE5E5, ideographicSpace);
return resultString;
}
