void open( const std::string& anchorFile ) {
_in.close();
_in.clear();
_in.open( anchorFile.c_str() );
_buffer.clear();
_readDocumentHeader();
}
void _mergeData() {
// fill in tags from the document vector fields
// unnecessary to filter out any fields to remove, as they will
// be ignored when indexing.
indri::utility::greedy_vector<indri::index::FieldExtent>& fields = _docIter->currentEntry()->fields() ;
tl.clear();
for (size_t i = 0; i < fields.size(); i++) {
const indri::index::FieldExtent& field = fields[i];
std::string fieldName = _index->field(fields[i].id);;
tl.addTag(fieldName.c_str(), fieldName.c_str(), field.begin);
tl.endTag(fieldName.c_str(), fieldName.c_str(), field.end);
}
//stuff it into the parsed doc
tl.writeTagList(parsed->tags);
// fill in terms from the document text so that they will stop/stem
// correctly when added to the new repository.
// Potentially issues with url injection here...
// probably best not to do this with trecweb/html docs....
// TODO: test this
termBuffer.clear();
termBuffer.grow( parsed->textLength * 2 );
for (size_t i = 0; i < parsed->positions.size(); i++ ) {
int start = parsed->positions[i].begin;
int end = parsed->positions[i].end;
int token_len = end - start;
const char *token = parsed->text + start;
char* write_loc = termBuffer.write( token_len + 1 );
strncpy( write_loc, token, token_len );
write_loc[token_len] = '\0';
parsed->terms.push_back( write_loc );
}
}
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();
}
}
indri::api::ParsedDocument* transform( indri::api::ParsedDocument* document ) {
// find the url metadata pair
indri::utility::greedy_vector<indri::parse::MetadataPair>::iterator iter;
iter = std::find_if( document->metadata.begin(),
document->metadata.end(),
indri::parse::MetadataPair::key_equal( "url" ) );
// no need to stick around if there is no url
if( iter == document->metadata.end() )
return document;
// need to copy this into the buffer and parse it:
_buffer.clear();
_buffer.grow( iter->valueLength + 1 );
char* urlText = _buffer.write( iter->valueLength );
// pushes the buffer pointer, trash in urlText
// memcpy( _buffer.write( iter->valueLength ), iter->value, iter->valueLength );
memcpy( urlText, iter->value, iter->valueLength );
*_buffer.write(1) = '\0';
// now we're pointing to the copied urlText, so we can start parsing
int urlStart = (int)document->terms.size();
char* c = urlText;
bool lastSkipped = true;
bool foundSlash = false;
int remainingStart = -1;
// skip the beginning stuff (http://)
for( c = urlText; *c; c++ ) {
if( *c == '/' && c[1] && c[1] == '/' ) {
urlText = c + 2;
}
}
int cnt = 0;
// now, try to find the
for( c = urlText; *c; c++ ) {
if( (*c >= 'A' && *c <= 'Z') ||
(*c >= 'a' && *c <= 'z') ||
(*c >= '0' && *c <= '9') )
{
if( lastSkipped ) {
lastSkipped = false;
document->terms.push_back( c );
cnt++;
}
} else if( *c == '/' && remainingStart < 0 ) {
*c = 0;
lastSkipped = true;
remainingStart = document->terms.size();
} else {
lastSkipped = true;
*c = 0;
}
}
// put in phony positions entries
int tokEnd = document->positions.size() ? document->positions[document->positions.size()-1].end : 0;
for (size_t n = document->terms.size()-cnt; n < document->terms.size(); n++) {
TermExtent extent;
extent.begin = tokEnd++; // hope this doesn't run off the end
extent.end = tokEnd;
document->positions.push_back( extent );
}
// the URL text is now parsed and stored in the document
// all we need to do now is put some tags around the text.
TagExtent *url = new TagExtent;
url->begin = urlStart;
url->end = document->terms.size();
url->name = "url";
url->number = 0;
document->tags.push_back(url);
TagExtent *domain = new TagExtent;
domain->begin = urlStart;
domain->end = (remainingStart >= 0) ? remainingStart : document->terms.size();
domain->name = "urldomain";
domain->number = 0;
document->tags.push_back(domain);
if( remainingStart > 0 ) {
indri::parse::TagExtent *urlpath = new TagExtent;
urlpath->begin = remainingStart;
urlpath->end = document->terms.size();
urlpath->name = "urlpath";
urlpath->number = 0;
document->tags.push_back(urlpath);
}
return document;
}
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;
}
}
}
