ScanRLEArray::ScanRLEArray(ArrayDesc const& arr, std::string path) :
RLEArray(arr), _dirPath(path), _maxChunkNo(0), logger(log4cxx::Logger::getLogger("scidb.query.ops.ScanRQArray"))
{
filesystem::path full_path = filesystem::system_complete(filesystem::path(_dirPath));
if (!filesystem::exists(full_path))
{
throw SYSTEM_EXCEPTION(SCIDB_SE_INTERNAL, SCIDB_LE_CANT_OPEN_PATH) << _dirPath;
}
if (!filesystem::is_directory(full_path))
{
throw SYSTEM_EXCEPTION(SCIDB_SE_INTERNAL, SCIDB_LE_DIRECTORY_EXPECTED) << _dirPath;
}
_maxChunkNo = arr.getNumberOfChunks() / arr.getAttributes().size();
}
shared_ptr< Array > execute(vector< shared_ptr< Array> >& inputArrays, shared_ptr<Query> query)
{
// I maintain the log of the operator in a local file named after Correlation_N.log, N is the instance ID.
stringstream logFileName;
logFileName << "/home/scidb/preselect_" << query->getInstanceID() << ".log";
FILE *logFile;
logFile = fopen(logFileName.str().c_str(), "w");
shared_ptr<Array> originalArray = inputArrays[0];
shared_ptr<Array> correlationArray = inputArrays[1];
ArrayDesc originalSchema = originalArray->getArrayDesc();
ArrayDesc corrSchema = correlationArray->getArrayDesc();
Dimensions originalDims = originalSchema.getDimensions();
Dimensions corrDims = corrSchema.getDimensions();
DimensionDesc originalDimsP = originalDims[1];
DimensionDesc corrDimsP = corrDims[0];
// Note the correlation array doesn't have Y column.
Coordinate p = corrDimsP.getCurrLength();
fprintf(logFile, "p = %ld\n # of chunk = %ld\n", p, corrSchema.getNumberOfChunks());
fflush(logFile);
shared_ptr<ConstArrayIterator> corrArrayIter = correlationArray->getIterator(0);
if(! corrArrayIter->end() )
{
correlation *corr = new correlation[p];
// The correlation array will always have only 1 chunk (we designed correlation array like this), so no loops here.
shared_ptr<ConstChunkIterator> corrChunkIter = corrArrayIter->getChunk().getConstIterator();
for(Coordinate i=0; i<p; ++i)
{
corr[i].id = i+1;
corr[i].corr = corrChunkIter->getItem().getDouble();
//fprintf(logFile, "%d, %f\n", corr[i].id, corr[i].corr);
++(*corrChunkIter);
}
//fflush(logFile);
qsort(corr, p, sizeof(correlation), &comp);
for(Coordinate i=0; i<p; ++i)
{
fprintf(logFile, "%d, %f\n", corr[i].id, corr[i].corr);
}
fflush(logFile);
Coordinate d = ((boost::shared_ptr<OperatorParamPhysicalExpression>&)_parameters[0])->getExpression()->evaluate().getInt64();
fprintf(logFile, "d=%ld\n", d);
stringstream ss;
vector<string> names;
names.push_back("j");
vector<TypeId> types;
types.push_back(TID_INT64);
for(Coordinate i=0; i<d; ++i)
{
ss << "j=" << corr[i].id << " or ";
}
ss << "j=" << p+1;
fprintf(logFile, "%s\n", ss.str().c_str());
fflush(logFile);
Expression e;
e.compile(ss.str(), names, types);
fclose(logFile);
boost::shared_ptr<scidb::Query> emptyQuery;
return boost::shared_ptr<Array>(new FilterArray(_schema, inputArrays[0], boost::make_shared<Expression>(e), emptyQuery, _tileMode));
}
else
{
shared_ptr<Array> outputArray(new MemArray(_schema, query));
fclose(logFile);
return outputArray;
}
}