Rewind ASCII files after getting their size and skip EOF character transmission

   limitations under the License.

 */

/*! \file cluster.cpp

/*! \file cluster.cp

 *

 * \brief Implementation of the calculation for a cluster of spheres.

 */

void cluster(const string& config_file, const string& data_file, const string& output_path, const mixMPI *mpidata) {

  chrono::time_point<chrono::high_resolution_clock> t_start = chrono::high_resolution_clock::now();

  chrono::duration<double> elapsed;

  string message;

  string timing_name = output_path + "/c_timing_mpi"+ to_string(mpidata->rank) +".log";

  FILE *timing_file = fopen(timing_name.c_str(), "w");

  Logger *time_logger = new Logger(LOG_DEBG, timing_file);

#pragma omp barrier

	{

	  // thread 0 already wrote on global files, skip it and take care of appending the others

	  for (int ri = 0; ri < ompnumthreads; ri++) {

	    // still, we need to remove all c_OCLU_RANK_0 files

	  for (int ri = 1; ri < ompnumthreads; ri++) {

	    string partial_file_name = output_path + "/c_OCLU_" + to_string(mpidata->rank) + "_" + to_string(ri);

	    if (ri == 0) {

	      remove(partial_file_name.c_str());

	    } else {

	    string message = "Copying ASCII output in MPI process " + to_string(mpidata->rank) + " of thread " + to_string(ri) + " of " + to_string(ompnumthreads - 1) + "... ";

	    logger->log(message, LOG_DEBG);

	    FILE *partial_output = fopen(partial_file_name.c_str(), "r");

	    logger->log("done.\n", LOG_DEBG);

	  }

	}

	}

#endif

	// here go the code to append the files written in MPI processes > 0 to the ones on MPI process 0

#ifdef MPI_VERSION

	    int remotethreads;

	    MPI_Recv(&remotethreads, 1, MPI_INT, rr, 1, MPI_COMM_WORLD, MPI_STATUS_IGNORE);

	    for (int ri=0; ri<remotethreads; ri++) {

	      // first get the ASCII local file

	      char *chunk_buffer;

	      int chunk_buffer_size = -1;

	      MPI_Recv(&chunk_buffer_size, 1, MPI_INT, rr, 1, MPI_COMM_WORLD, MPI_STATUS_IGNORE);

		delete[] chunk_buffer;

		MPI_Recv(&chunk_buffer_size, 1, MPI_INT, rr, 1, MPI_COMM_WORLD, MPI_STATUS_IGNORE);

	      }

	      fputc(10, output);

	      // now get the binary local file

	      long buffer_size = 0;

	partial_output.open(partial_file_name.c_str(), ios::in | ios::binary);

	partial_output.seekg(0, ios::end);

	const long partial_output_size = partial_output.tellg();

	partial_output.close();

	partial_output.open(partial_file_name.c_str(), ios::in | ios::binary);

	int chunk_buffer_size = 25165824; // Length of char array  with 24Mb size

	char *chunk_buffer = new char[chunk_buffer_size]();

	int full_chunks = (int)(partial_output_size / chunk_buffer_size);

	for (int fi = 0; fi < full_chunks; fi++) {

	  partial_output.read(chunk_buffer, chunk_buffer_size);

	  // If EOF is reached, do not send EOF character.

	  long ptr_position = partial_output.tellg();

	  if (ptr_position == partial_output_size) {

	    chunk_buffer_size--;

	  }

	  // Send the size of the buffer that is being transmitted (Node-0 does not know whether it is full or not)

	  MPI_Send(&chunk_buffer_size, 1, MPI_INT, 0, 1, MPI_COMM_WORLD);

	  // Actually send the file contents to Node-0

	long ptr_position = partial_output.tellg();

	if (ptr_position < partial_output_size) {

	  // Send the last partial buffer

	  chunk_buffer_size = partial_output_size - ptr_position;

	  chunk_buffer_size = partial_output_size - ptr_position - 1;

	  partial_output.read(chunk_buffer, chunk_buffer_size);

	  // Send the size of the buffer that is being transmitted (Node-0 does not know whether it is full or not)

	  MPI_Send(&chunk_buffer_size, 1, MPI_INT, 0, 1, MPI_COMM_WORLD);

	partial_output.close();

	delete[] chunk_buffer;

	remove(partial_file_name.c_str());

	logger->log("done.\n", LOG_DEBG);

	partial_file_name = output_path + "/c_TPPOAN_" + to_string(mpidata->rank) + "_" + to_string(ri);

	message = "Copying binary output in MPI process " + to_string(mpidata->rank) + " of thread " + to_string(ri) + " of " + to_string(ompnumthreads - 1) + "... ";