void IndexWriter::_writeBatch( indri::file::SequentialWriteBuffer* buffer, int document, int length, indri::utility::Buffer& data ) {
assert( length < 100*1000*1000 );
_writeSkip( buffer, document, length );
if( data.position() != 0 ) {
buffer->write( data.front(), data.position() );
data.clear();
}
}
static void zlib_read_document( z_stream_s& stream, indri::file::File& infile, UINT64 offset, indri::utility::Buffer& outputBuffer ) {
// read in data from the file until the stream ends
// split up the data as necessary
// decompress positional info
// read some data
char inputBuffer[INPUT_BUFFER_SIZE];
outputBuffer.grow( INPUT_BUFFER_SIZE );
outputBuffer.write( sizeof(indri::api::ParsedDocument) );
stream.avail_in = 0;
stream.avail_out = 0;
while(true) {
if( !stream.avail_in ) {
UINT64 readSize = infile.read( inputBuffer, offset, sizeof inputBuffer );
offset += readSize;
stream.avail_in = readSize;
stream.next_in = (Bytef*) inputBuffer;
}
stream.avail_out = outputBuffer.size() - outputBuffer.position();
stream.next_out = (Bytef*) outputBuffer.write( outputBuffer.size() - outputBuffer.position() );
int result = inflate( &stream, Z_NO_FLUSH );
outputBuffer.unwrite( stream.avail_out );
if( result == Z_STREAM_END ) {
result = inflate( &stream, Z_FINISH );
if( result < 0 )
LEMUR_THROW( result, "Something bad happened while trying to finish decompressing a document." );
inflateEnd( &stream );
break;
}
if( result < 0 ) {
LEMUR_THROW( result, "Something bad happened while trying to decompress a document." );
}
if( stream.avail_out == 0 ) {
outputBuffer.grow();
}
}
}
bool lemur::file::SortMergeTextFiles::_readLine(FILE *_in, char*& beginLine, size_t& lineLength, indri::utility::Buffer &_buffer ) {
lineLength = 0;
size_t actual;
// make a buffer of a reasonable size so we're not always allocating
if( _buffer.size() < 1024*1024 )
_buffer.grow( 1024*1024 );
// if we're running out of room, add 25MB
if( (_buffer.size() - _buffer.position()) < 512*1024 ) {
_buffer.grow( _buffer.size() + 1024*1024*25 );
}
if( _buffer.position() ) {
// get rid of null terminator from previous call
_buffer.unwrite(1);
}
size_t readAmount = _buffer.size() - _buffer.position() - 2;
// fetch next document line
char* buffer = _buffer.write( readAmount );
char* result = fgets( buffer, (int)readAmount, _in );
if(!result) {
return false;
}
actual = strlen(buffer);
lineLength += actual;
_buffer.unwrite( readAmount - actual );
// all finished reading
*_buffer.write(1) = 0;
beginLine = _buffer.front() + _buffer.position() - lineLength - 1;
// strip the \n off the end
if (beginLine[lineLength-1]=='\n') {
beginLine[lineLength-1]=0;
}
return true;
}
void IndexWriter::_addInvertedListData( indri::utility::greedy_vector<WriterIndexContext*>& lists, indri::index::TermData* termData, indri::utility::Buffer& listBuffer, UINT64& endOffset ) {
indri::utility::greedy_vector<WriterIndexContext*>::iterator iter;
const int minimumSkip = 1<<12; // 4k
int documentsWritten = 0;
const float topdocsFraction = 0.01f;
bool hasTopdocs = termData->corpus.documentCount > TOPDOCS_DOCUMENT_COUNT;
bool isFrequent = termData->corpus.totalCount > FREQUENT_TERM_COUNT;
int topdocsCount = hasTopdocs ? int(termData->corpus.documentCount * 0.01) : 0;
int topdocsSpace = hasTopdocs ? ((topdocsCount*3*sizeof(UINT32)) + sizeof(int)) : 0;
// write a control byte
char control = (hasTopdocs ? 0x01 : 0) | (isFrequent ? 0x02 : 0);
_invertedOutput->write( &control, 1 );
UINT64 initialPosition = _invertedOutput->tell();
// leave some room for the topdocs list
if( hasTopdocs ) {
_invertedOutput->seek( topdocsSpace + initialPosition );
}
// maintain a list of top documents
std::priority_queue<DocListIterator::TopDocument,
std::vector<DocListIterator::TopDocument>,
DocListIterator::TopDocument::greater> topdocs;
double threshold = 0;
int lastDocument = 0;
int positions = 0;
int docs = 0;
// for each matching list:
for( iter = lists.begin(); iter != lists.end(); ++iter ) {
indri::index::DocListFileIterator::DocListData* listData = (*iter)->iterator->currentEntry();
DocListIterator* iterator = listData->iterator;
Index* index = (*iter)->index;
indri::utility::RVLCompressStream stream( listBuffer );
int listDocs = 0;
int listPositions = 0;
while( !iterator->finished() ) {
// get the latest entry from the list
DocListIterator::DocumentData* documentData = iterator->currentEntry();
// add to document counter
docs++; listDocs++;
// update the topdocs list
if( hasTopdocs ) {
int length = index->documentLength( documentData->document );
int count = documentData->positions.size();
// compute DocListIterator::TopDocument::greater (current, top())
// if false, no reason to insert this entry.
// note that the test is inverted.
// int(length * threshold) <= count is equivalent to
// count/length > topdocs.top().count/topdocs.top().length
// but we use < to force breaking a tie in favor of keeping
// the first seen document.
if( int(length * threshold) < count || topdocs.size() < topdocsCount ) {
// form a topdocs entry for this document
DocListIterator::TopDocument topDocument( documentData->document,
count,
length );
topdocs.push( topDocument );
while( topdocs.size() > topdocsCount )
topdocs.pop();
threshold = topdocs.top().count / double(topdocs.top().length);
}
}
if( listBuffer.position() > minimumSkip ) {
// time to write in a skip
_writeBatch( _invertedOutput, documentData->document, listBuffer.position(), listBuffer );
// delta encode documents by batch
lastDocument = 0;
}
assert( documentData->document > lastDocument );
// write this entry out to the list
stream << documentData->document - lastDocument;
stream << (int) documentData->positions.size();
lastDocument = documentData->document;
int lastPosition = 0;
for( int i=0; i<documentData->positions.size(); i++ ) {
stream << (documentData->positions[i] - lastPosition);
lastPosition = documentData->positions[i];
positions++; listPositions++;
}
iterator->nextEntry();
}
indri::index::TermData* td = iterator->termData();
assert( listPositions == td->corpus.totalCount );
assert( listDocs == td->corpus.documentCount );
}
assert( docs == termData->corpus.documentCount );
assert( positions == termData->corpus.totalCount );
// write in the final skip info
_writeBatch( _invertedOutput, -1, listBuffer.position(), listBuffer );
UINT64 finalPosition = _invertedOutput->tell();
if( hasTopdocs ) {
_invertedOutput->seek( initialPosition );
_invertedOutput->write( &topdocsCount, sizeof(int) );
assert( topdocs.size() == topdocsCount );
// write these into the topdocs list in order from smallest fraction to largest fraction,
// where fraction = c(w;D)/|D|
while( topdocs.size() ) {
DocListIterator::TopDocument topDocument = topdocs.top();
_invertedOutput->write( &topDocument.document, sizeof(int) );
_invertedOutput->write( &topDocument.count, sizeof(int) );
_invertedOutput->write( &topDocument.length, sizeof(int) );
topdocs.pop();
}
assert( (_invertedOutput->tell() - initialPosition) == topdocsSpace );
_invertedOutput->seek( finalPosition );
}
endOffset = finalPosition;
}
void _fetchText( indri::utility::greedy_vector<TagExtent *>& tags, indri::utility::greedy_vector<char*>& terms ) {
// now, fetch the additional terms
char line[65536];
_buffer.clear();
for( int i=0; i<_count; i++ ) {
// LINK
_in.getline( line, sizeof line-1 );
// LINKDOCNO
_in.getline( line, sizeof line-1 );
// TEXT=
_in.getline( line, sizeof line-1 );
int textLen = strlen(line+6);
strcpy( _buffer.write(textLen+1), line+6 );
_buffer.unwrite(1);
assert( *(_buffer.front()+_buffer.position()-1) == '\"' && "Last character should be a quote" );
}
*(_buffer.write(1)) = 0;
// now there's a bunch of text in _buffer, space separated, with each
// link separated by a " symbol
char* beginWord = 0;
int beginIndex = 0;
char* buffer = _buffer.front();
for( unsigned int i=0; i<_buffer.position(); i++ ) {
if( isalnum(buffer[i]) && !beginWord ) {
beginWord = buffer+i;
if(!beginIndex)
beginIndex = terms.size();
} else if( isspace(buffer[i]) ) {
buffer[i] = 0;
if( beginWord )
terms.push_back( beginWord );
beginWord = 0;
} else if( buffer[i] == '\"' ) {
buffer[i] = 0;
if( beginWord )
terms.push_back( beginWord );
beginWord = 0;
TagExtent * extent = new TagExtent;
extent->name = "inlink";
extent->begin = beginIndex;
extent->end = terms.size();
extent->number = 0;
extent->parent = 0;
assert( extent->begin <= extent->end );
if( beginIndex ) {
tags.push_back(extent);
if( terms.size() > 125000 )
break;
}
beginIndex = 0;
}
}
}