Patch to fix bugs 15556 and 22742.

Double byte handeling of SSTDeserializer now works.


git-svn-id: https://svn.apache.org/repos/asf/jakarta/poi/branches/REL_2_BRANCH@353396 13f79535-47bb-0310-9956-ffa450edef68
This commit is contained in:
Jason Height 2003-10-14 07:18:17 +00:00
parent 926262df8d
commit aec407bceb
2 changed files with 49 additions and 34 deletions

View File

@ -62,13 +62,14 @@ import org.apache.poi.util.LittleEndianConsts;
* Handles the task of deserializing a SST string. The two main entry points are
*
* @author Glen Stampoultzis (glens at apache.org)
* @author Jason Height (jheight at apache.org)
*/
class SSTDeserializer
{
private BinaryTree strings;
/** this is the number of characters we expect in the first sub-record in a subsequent continuation record */
private int continuationExpectedChars;
/** this is the number of characters that have been read prior to the continuation */
private int continuationReadChars;
/** this is the string we were working on before hitting the end of the current record. This string is NOT finished. */
private String unfinishedString;
/** this is true if the string uses wide characters */
@ -82,6 +83,7 @@ class SSTDeserializer
/** Number of characters in current string */
private int charCount;
private int extensionLength;
private int continueSkipBytes = 0;
public SSTDeserializer( BinaryTree strings )
@ -93,13 +95,14 @@ class SSTDeserializer
private void initVars()
{
runCount = 0;
continuationExpectedChars = 0;
continuationReadChars = 0;
unfinishedString = "";
// bytesInCurrentSegment = 0;
// stringDataOffset = 0;
wideChar = false;
richText = false;
extendedText = false;
continueSkipBytes = 0;
}
/**
@ -107,14 +110,15 @@ class SSTDeserializer
* strings may span across multiple continuations. Read the SST record
* carefully before beginning to hack.
*/
public void manufactureStrings( final byte[] data, final int initialOffset, short dataSize )
public void manufactureStrings( final byte[] data, final int initialOffset)
{
initVars();
int offset = initialOffset;
while ( ( offset - initialOffset ) < dataSize )
final int dataSize = data.length;
while ( offset < dataSize )
{
int remaining = dataSize - offset + initialOffset;
int remaining = dataSize - offset;
if ( ( remaining > 0 ) && ( remaining < LittleEndianConsts.SHORT_SIZE ) )
{
@ -122,26 +126,31 @@ class SSTDeserializer
}
if ( remaining == LittleEndianConsts.SHORT_SIZE )
{
setContinuationExpectedChars( LittleEndian.getUShort( data, offset ) );
//JMH Dont know about this
setContinuationCharsRead( 0 );//LittleEndian.getUShort( data, offset ) );
unfinishedString = "";
break;
}
charCount = LittleEndian.getUShort( data, offset );
int charsRead = charCount;
readStringHeader( data, offset );
boolean stringContinuesOverContinuation = remaining < totalStringSize();
if ( stringContinuesOverContinuation )
{
int remainingBytes = ( initialOffset + dataSize ) - offset - stringHeaderOverhead();
setContinuationExpectedChars( charCount - calculateCharCount( remainingBytes ) );
charCount -= getContinuationExpectedChars();
int remainingBytes = dataSize - offset - stringHeaderOverhead();
//Only read the size of the string or whatever is left before the
//continuation
charsRead = Math.min(charsRead, calculateCharCount( remainingBytes ));
setContinuationCharsRead( charsRead );
if (charsRead == charCount) {
//Since all of the characters will have been read, but the entire string (including formatting runs etc)
//hasnt, Compute the number of bytes to skip when the continue record starts
continueSkipBytes = offsetForContinuedRecord(0) - (remainingBytes - calculateByteCount(charsRead));
}
else
{
setContinuationExpectedChars( 0 );
}
processString( data, offset, charCount );
processString( data, offset, charsRead );
offset += totalStringSize();
if ( getContinuationExpectedChars() != 0 )
if ( stringContinuesOverContinuation )
{
break;
}
@ -222,6 +231,7 @@ class SSTDeserializer
UnicodeString string = new UnicodeString( UnicodeString.sid,
(short) unicodeStringBuffer.length,
unicodeStringBuffer );
setContinuationCharsRead( calculateCharCount(bytesRead));
if ( isStringFinished() )
{
@ -238,7 +248,7 @@ class SSTDeserializer
private boolean isStringFinished()
{
return getContinuationExpectedChars() == 0;
return getContinuationCharsRead() == charCount;
}
/**
@ -301,8 +311,9 @@ class SSTDeserializer
{
if ( isStringFinished() )
{
final int offset = continueSkipBytes;
initVars();
manufactureStrings( record, 0, (short) record.length );
manufactureStrings( record, offset);
}
else
{
@ -330,13 +341,12 @@ class SSTDeserializer
*/
private void readStringRemainder( final byte[] record )
{
int stringRemainderSizeInBytes = calculateByteCount( getContinuationExpectedChars() );
// stringDataOffset = LittleEndianConsts.BYTE_SIZE;
int stringRemainderSizeInBytes = calculateByteCount( charCount-getContinuationCharsRead() );
byte[] unicodeStringData = new byte[SSTRecord.STRING_MINIMAL_OVERHEAD
+ calculateByteCount( getContinuationExpectedChars() )];
+ stringRemainderSizeInBytes];
// write the string length
LittleEndian.putShort( unicodeStringData, 0, (short) getContinuationExpectedChars() );
LittleEndian.putShort( unicodeStringData, 0, (short) (charCount-getContinuationCharsRead()) );
// write the options flag
unicodeStringData[LittleEndianConsts.SHORT_SIZE] = createOptionByte( wideChar, richText, extendedText );
@ -345,7 +355,7 @@ class SSTDeserializer
// past all the overhead of the str_data array
arraycopy( record, LittleEndianConsts.BYTE_SIZE, unicodeStringData,
SSTRecord.STRING_MINIMAL_OVERHEAD,
unicodeStringData.length - SSTRecord.STRING_MINIMAL_OVERHEAD );
stringRemainderSizeInBytes );
// use special constructor to create the final string
UnicodeString string = new UnicodeString( UnicodeString.sid,
@ -356,7 +366,7 @@ class SSTDeserializer
addToStringTable( strings, integer, string );
int newOffset = offsetForContinuedRecord( stringRemainderSizeInBytes );
manufactureStrings( record, newOffset, (short) ( record.length - newOffset ) );
manufactureStrings( record, newOffset);
}
/**
@ -388,8 +398,12 @@ class SSTDeserializer
private int offsetForContinuedRecord( int stringRemainderSizeInBytes )
{
return stringRemainderSizeInBytes + LittleEndianConsts.BYTE_SIZE
+ runCount * LittleEndianConsts.INT_SIZE + extensionLength;
int offset = stringRemainderSizeInBytes + runCount * LittleEndianConsts.INT_SIZE + extensionLength;
if (stringRemainderSizeInBytes != 0)
//If a portion of the string remains then the wideChar options byte is repeated,
//so need to skip this.
offset += + LittleEndianConsts.BYTE_SIZE;
return offset;
}
private byte createOptionByte( boolean wideChar, boolean richText, boolean farEast )
@ -409,17 +423,18 @@ class SSTDeserializer
int dataLengthInBytes = record.length - LittleEndianConsts.BYTE_SIZE;
byte[] unicodeStringData = new byte[record.length + LittleEndianConsts.SHORT_SIZE];
LittleEndian.putShort( unicodeStringData, (byte) 0, (short) calculateCharCount( dataLengthInBytes ) );
int charsRead = calculateCharCount( dataLengthInBytes );
LittleEndian.putShort( unicodeStringData, (byte) 0, (short) charsRead );
arraycopy( record, 0, unicodeStringData, LittleEndianConsts.SHORT_SIZE, record.length );
UnicodeString ucs = new UnicodeString( UnicodeString.sid, (short) unicodeStringData.length, unicodeStringData );
unfinishedString = unfinishedString + ucs.getString();
setContinuationExpectedChars( getContinuationExpectedChars() - calculateCharCount( dataLengthInBytes ) );
setContinuationCharsRead( charsRead );
}
private boolean stringSpansContinuation( int continuationSizeInBytes )
{
return calculateByteCount( getContinuationExpectedChars() ) > continuationSizeInBytes;
return calculateByteCount( charCount - getContinuationCharsRead() ) > continuationSizeInBytes;
}
/**
@ -427,14 +442,14 @@ class SSTDeserializer
* sub-record in a subsequent continuation record
*/
int getContinuationExpectedChars()
int getContinuationCharsRead()
{
return continuationExpectedChars;
return continuationReadChars;
}
private void setContinuationExpectedChars( final int count )
private void setContinuationCharsRead( final int count )
{
continuationExpectedChars = count;
continuationReadChars = count;
}
private int calculateByteCount( final int character_count )

View File

@ -482,7 +482,7 @@ public class SSTRecord
field_2_num_unique_strings = LittleEndian.getInt( data, 4 + offset );
field_3_strings = new BinaryTree();
deserializer = new SSTDeserializer(field_3_strings);
deserializer.manufactureStrings( data, 8 + offset, (short)(size - 8) );
deserializer.manufactureStrings( data, 8 + offset);
}