There seems a bug inside ICUTranscoder::transcodeFrom and another in IconvGNUTranscoder::transcodeFrom

[Kirby Zhou <ki...@gmail.com>]

ICUTranscoder::transcodeFrom 

 495     UErrorCode  err = U_ZERO_ERROR;
 496     ucnv_toUnicode
 497     (
 498         fConverter
 499         , &startTarget
 500         , startTarget + maxChars
 501         , (const char**)&startSrc
 502         , (const char*)endSrc
 503         , (fFixed ? 0 : (int32_t*)fSrcOffsets)
 504         , false
 505         , &err
 506     );

There seems need a mute to protect fConverter.
ICULCPTranscoder::calcRequiredSize called ' XMLMutexLock
lockConverter(&fMutex); ' to do it.


IconvGNUTranscoder::transcodeFrom

1027     for (size_t cnt = 0; cnt < maxChars && srcLen; cnt++) {
1028         size_t    rc = iconvFrom(startSrc, &srcLen, &orgTarget,
uChSize());
1029         if (rc == (size_t)-1) {
1030             if (errno != E2BIG || prevSrcLen == srcLen) {
1031                 ThrowXMLwithMemMgr(TranscodingException,
XMLExcepts::Trans_BadSrcSeq, getMemoryManager());
1032             }
1033         }
1034         charSizes[cnt] = prevSrcLen - srcLen;
1035         prevSrcLen = srcLen;

It do not deal with errno==EINVAL, which will happen when it is called by
XMLReader::refreshCharBuffer with an incomplete MBCS character.
I mean, one half is in buffer, and the other half is left on disk.


  Regards
  Kirby Zhou




---------------------------------------------------------------------
To unsubscribe, e-mail: c-dev-unsubscribe@xerces.apache.org
For additional commands, e-mail: c-dev-help@xerces.apache.org