edu.harvard.iq.dataverse.ingest.tabulardata.impl.plugins.sav.SAVFileReader.java Source code

Java tutorial

Introduction

Here is the source code for edu.harvard.iq.dataverse.ingest.tabulardata.impl.plugins.sav.SAVFileReader.java

Source

/*
   Copyright (C) 2005-2012, by the President and Fellows of Harvard College.
    
   Licensed under the Apache License, Version 2.0 (the "License");
   you may not use this file except in compliance with the License.
   You may obtain a copy of the License at
    
     http://www.apache.org/licenses/LICENSE-2.0
    
   Unless required by applicable law or agreed to in writing, software
   distributed under the License is distributed on an "AS IS" BASIS,
   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
   See the License for the specific language governing permissions and
   limitations under the License.
    
   Dataverse Network - A web application to share, preserve and analyze research data.
   Developed at the Institute for Quantitative Social Science, Harvard University.
   Version 3.0.
*/
package edu.harvard.iq.dataverse.ingest.tabulardata.impl.plugins.sav;

import java.io.*;
import java.nio.*;
import java.util.logging.*;

import java.util.*;
import java.util.regex.*;
import java.text.*;

import org.apache.commons.lang.*;
import org.apache.commons.codec.binary.Hex;

import edu.harvard.iq.dataverse.DataTable;
import edu.harvard.iq.dataverse.datavariable.DataVariable;
import edu.harvard.iq.dataverse.datavariable.SummaryStatistic;
import edu.harvard.iq.dataverse.datavariable.VariableCategory;
import edu.harvard.iq.dataverse.datavariable.VariableRange;

import edu.harvard.iq.dataverse.ingest.tabulardata.TabularDataFileReader;
import edu.harvard.iq.dataverse.ingest.tabulardata.spi.TabularDataFileReaderSpi;
import edu.harvard.iq.dataverse.ingest.tabulardata.TabularDataIngest;
import edu.harvard.iq.dataverse.ingest.tabulardata.InvalidData;

/**
 * ingest plugin for SPSS SAV file format.
 *
 * This reader plugin has been fully re-implemented for the DVN 4.0;
 * It is still borrows heavily from, and builds on the basis of the 
 * old implementation by Akio Sone, that was in use in the versions 
 * 2-3 of the DVN.
 * 
 * @author Akio Sone at UNC-Odum
 * @author Leonid Andreev
 */

public class SAVFileReader extends TabularDataFileReader {

    // static fields ---------------------------------------------------------//
    private static String[] FORMAT_NAMES = { "sav", "SAV" };
    private static String[] EXTENSIONS = { "sav" };
    private static String[] MIME_TYPE = { "application/x-spss-sav" };

    private static final int LENGTH_SAV_INT_BLOCK = 4;
    // note: OBS block is either double or String, not Integer
    private static final int LENGTH_SAV_OBS_BLOCK = 8;

    private static final int SAV_MAGIC_NUMBER_LENGTH = LENGTH_SAV_INT_BLOCK;

    private static String SAV_FILE_SIGNATURE = "$FL2";

    // Record Type 1 fields
    private static final int LENGTH_RECORDTYPE1 = 172;

    private static final int LENGTH_SPSS_PRODUCT_INFO = 60;

    private static final int FILE_LAYOUT_CONSTANT = 2;

    private static final int LENGTH_FILE_LAYOUT_CODE = LENGTH_SAV_INT_BLOCK;

    private static final int LENGTH_NUMBER_OF_OBS_UNITS_PER_CASE = LENGTH_SAV_INT_BLOCK;

    private static final int LENGTH_COMPRESSION_SWITCH = LENGTH_SAV_INT_BLOCK;

    private static final int LENGTH_CASE_WEIGHT_VARIABLE_INDEX = LENGTH_SAV_INT_BLOCK;

    private static final int LENGTH_NUMBER_OF_CASES = LENGTH_SAV_INT_BLOCK;

    private static final int LENGTH_COMPRESSION_BIAS = LENGTH_SAV_OBS_BLOCK;

    private static final int LENGTH_FILE_CREATION_INFO = 84;

    private static final int length_file_creation_date = 9;
    private static final int length_file_creation_time = 8;
    private static final int length_file_creation_label = 64;
    private static final int length_file_creation_padding = 3;

    // Recorde Type 2

    private static final int LENGTH_RECORDTYPE2_FIXED = 32;
    private static final int LENGTH_RECORD_TYPE2_CODE = 4;
    private static final int LENGTH_TYPE_CODE = 4;
    private static final int LENGTH_LABEL_FOLLOWS = 4;
    private static final int LENGTH_MISS_VALUE_FORMAT_CODE = 4;
    private static final int LENGTH_PRINT_FORMAT_CODE = 4;;
    private static final int LENGTH_WRITE_FORMAT_CODE = 4;
    private static final int LENGTH_VARIABLE_NAME = 8;
    private static final int LENGTH_VARIABLE_LABEL = 4;

    private static final int LENGTH_MISS_VAL_OBS_CODE = LENGTH_SAV_OBS_BLOCK;

    // Record Type 3/4
    private static final int LENGTH_RECORDTYPE3_HEADER_CODE = 4;
    private static final int LENGTH_RECORD_TYPE3_CODE = 4;
    private static final int LENGTH_RT3_HOW_MANY_LABELS = 4;
    private static final int LENGTH_RT3_VALUE = LENGTH_SAV_OBS_BLOCK;
    private static final int LENGTH_RT3_LABEL_LENGTH = 1;

    private static final int LENGTH_RECORD_TYPE4_CODE = 4;
    private static final int LENGTH_RT4_HOW_MANY_VARIABLES = 4;
    private static final int LENGTH_RT4_VARIABLE_INDEX = 4;

    // Record Type 6
    private static final int LENGTH_RECORD_TYPE6_CODE = 4;
    private static final int LENGTH_RT6_HOW_MANY_LINES = 4;
    private static final int LENGTH_RT6_DOCUMENT_LINE = 80;

    // Record Type 7
    private static final int LENGTH_RECORD_TYPE7_CODE = 4;
    private static final int LENGTH_RT7_SUB_TYPE_CODE = 4;

    // Record Type 999
    private static final int LENGTH_RECORD_TYPE999_CODE = 4;
    private static final int LENGTH_RT999_FILLER = 4;

    private static final List<String> RecordType7SubType4Fields = new ArrayList<String>();
    private static final Set<Integer> validMissingValueCodeSet = new HashSet<Integer>();
    private static final Map<Integer, Integer> missingValueCodeUnits = new HashMap<Integer, Integer>();

    private static double SYSMIS_LITTLE = 0xFFFFFFFFFFFFEFFFL;
    private static double SYSMIS_BIG = 0xFFEFFFFFFFFFFFFFL;

    private static Calendar GCO = new GregorianCalendar();

    private String[] dateFormatList;

    static {

        // initialize validMissingValueCodeSet
        validMissingValueCodeSet.add(3);
        validMissingValueCodeSet.add(2);
        validMissingValueCodeSet.add(1);
        validMissingValueCodeSet.add(0);
        validMissingValueCodeSet.add(-2);
        validMissingValueCodeSet.add(-3);

        // initialize missingValueCodeUnits

        missingValueCodeUnits.put(1, 1);
        missingValueCodeUnits.put(2, 2);
        missingValueCodeUnits.put(3, 3);
        missingValueCodeUnits.put(-2, 2);
        missingValueCodeUnits.put(-3, 3);
        missingValueCodeUnits.put(0, 0);

        RecordType7SubType4Fields.add("SYSMIS");
        RecordType7SubType4Fields.add("HIGHEST");
        RecordType7SubType4Fields.add("LOWEST");

        // set the origin of GCO to 1582-10-15
        GCO.set(1, 1582);// year
        GCO.set(2, 9); // month
        GCO.set(5, 15);// day of month
        GCO.set(9, 0);// AM(0) or PM(1)
        GCO.set(10, 0);// hh
        GCO.set(12, 0);// mm
        GCO.set(13, 0);// ss
        GCO.set(14, 0); // SS millisecond
        GCO.set(15, 0);// z
    }

    private static final long SPSS_DATE_BIAS = 60 * 60 * 24 * 1000;

    private static final long SPSS_DATE_OFFSET = SPSS_DATE_BIAS + Math.abs(GCO.getTimeInMillis());

    // instance fields -------------------------------------------------------//
    private static String unfVersionNumber = "6";

    // instance fields -------------------------------------------------------//

    private static Logger dbgLog = Logger.getLogger(SAVFileReader.class.getPackage().getName());

    TabularDataIngest ingesteddata = new TabularDataIngest();
    private DataTable dataTable = new DataTable();

    Map<String, String> shortToLongVariableNameTable = new LinkedHashMap<String, String>();
    Map<String, String> formatCategoryTable = new LinkedHashMap<String, String>();

    private boolean isLittleEndian = false;
    private boolean isDataSectionCompressed = true;

    private Map<Integer, String> OBSIndexToVariableName = new LinkedHashMap<Integer, String>();

    private int OBSUnitsPerCase;

    private List<Integer> variableTypelList = new ArrayList<Integer>();
    private List<Integer> OBSwiseTypelList = new ArrayList<Integer>();

    Map<String, String> printFormatTable = new LinkedHashMap<String, String>();

    Set<Integer> obsNonVariableBlockSet = new LinkedHashSet<Integer>();

    Map<String, String> valueVariableMappingTable = new LinkedHashMap<String, String>();

    Map<String, Integer> extendedVariablesSizeTable = new LinkedHashMap<String, Integer>();

    List<String> variableNameList = new ArrayList<String>();

    Map<String, InvalidData> invalidDataTable = new LinkedHashMap<String, InvalidData>(); // this variable used in 2 methods; only one uses it to set the smd value -- ??

    NumberFormat doubleNumberFormatter = new DecimalFormat();

    Set<Integer> decimalVariableSet = new HashSet<Integer>();

    String[] variableFormatTypeList = null;

    List<Integer> formatDecimalPointPositionList = new ArrayList<Integer>();

    int caseWeightVariableOBSIndex = 0;

    // date/time data formats

    private SimpleDateFormat sdf_ymd = new SimpleDateFormat("yyyy-MM-dd");
    private SimpleDateFormat sdf_ymdhms = new SimpleDateFormat("yyyy-MM-dd HH:mm:ss");
    private SimpleDateFormat sdf_dhms = new SimpleDateFormat("DDD HH:mm:ss");
    private SimpleDateFormat sdf_hms = new SimpleDateFormat("HH:mm:ss");

    Map<String, String> OBSTypeHexValue = new LinkedHashMap<String, String>();

    /*
     * TODO: add a comment explaining the whole thing about this default
     * character set. -- L.A. 4.0 beta
    */
    private String defaultCharSet = "ISO-8859-1";
    //private String defaultCharSet = "US-ASCII"; // -- temporary! -- 4.0 beta 6
    private int spssVersionNumber = 0;

    /**
     * The <code>String</code> that represents the numeric missing value 
     * in the final tab-delimited data file.
     */
    private String MissingValueForTextDataFileNumeric = "";

    public String getMissingValueForTextDataFileNumeric() {
        return MissingValueForTextDataFileNumeric;
    }

    public void setMissingValueForTextDataFileNumeric(String MissingValueToken) {
        this.MissingValueForTextDataFileNumeric = MissingValueToken;
    }

    String MissingValueForTextDataFileString = "";

    public String getMissingValueForTextDataFileString() {
        return MissingValueForTextDataFileString;
    }

    public void setMissingValueForTextDataFileString(String MissingValueToken) {
        this.MissingValueForTextDataFileString = MissingValueToken;
    }

    public SAVFileReader(TabularDataFileReaderSpi originator) {
        super(originator);
    }

    // Methods ---------------------------------------------------------------//

    private void init() throws IOException {

        sdf_ymd.setTimeZone(TimeZone.getTimeZone("GMT"));
        sdf_ymdhms.setTimeZone(TimeZone.getTimeZone("GMT"));
        sdf_dhms.setTimeZone(TimeZone.getTimeZone("GMT"));
        sdf_hms.setTimeZone(TimeZone.getTimeZone("GMT"));

        doubleNumberFormatter.setGroupingUsed(false);
        doubleNumberFormatter.setMaximumFractionDigits(340);

        if (getDataLanguageEncoding() != null) {
            defaultCharSet = getDataLanguageEncoding();
        }
    }

    public TabularDataIngest read(BufferedInputStream stream, File dataFile) throws IOException {
        dbgLog.info("SAVFileReader: read() start");

        if (dataFile != null) {
            throw new IOException("this plugin does not support external raw data files");
        }

        /* 
         * this "try" block is for catching unknown/unexpected exceptions 
         * thrown anywhere in the ingest code:
         */
        try {
            /* ingest happens here ... */

            // the following methods are now executed, in this order:

            // decodeHeader -- this method doesn't read any [meta]data and 
            //    doesn't initialize any values; its only purpose is to 
            //    make sure that the file is indeed an SPSS/SAV file. 
            // 
            // decodeRecordType1 -- there's always one RT1 record; it is 
            //    always 176 byte long. it contains the very basic metadata
            //    about the data file. most notably, the number of observations
            //    and the number of OBS (8 byte values) per observation.
            //
            // decodeRecordType2 -- there are multiple RT2 records. there's 
            //    one RT2 for every OBS (8 byte value); i.e. one per variable,
            //    or more per every String variable split into multiple OBS
            //    segments. this one is a 400 line method, that may benefit 
            //    from being split into smaller methods.
            //
            // decodeRecordType3and4 -- these sections come in pairs, each
            //    pair dedicated to one set of variable labels. 
            // decodeRecordType6,
            //
            // decodeRecordType7 -- this RT contains some extended 
            //    metadata for the data file. (including the information 
            //    about the extended variables, i.e. variables longer than
            //    255 bytes split into 255 byte fragments that are stored 
            //    in the data file as independent variables). 
            //
            // decodeRecordType999 -- this RT does not contain any data; 
            //    its sole function is to indicate that the metadata portion 
            //    of the data file is over and the data section follows. 
            // 
            // decodeRecordTypeData -- this method decodes the data section 
            //    of the file. Inside this method, 2 distinct methods are 
            //    called to process compressed or uncompressed data, depending
            //    on which method is used in this data file. 

            String methodCurrentlyExecuted = null;

            try {
                methodCurrentlyExecuted = "decodeHeader";
                dbgLog.fine("***** SAVFileReader: executing method decodeHeader");
                decodeHeader(stream);

                methodCurrentlyExecuted = "decodeRecordType1";
                dbgLog.fine("***** SAVFileReader: executing method decodeRecordType1");
                decodeRecordType1(stream);

                methodCurrentlyExecuted = "decodeRecordType2";
                dbgLog.fine("***** SAVFileReader: executing method decodeRecordType1");
                decodeRecordType2(stream);

                methodCurrentlyExecuted = "decodeRecordType3and4";
                dbgLog.fine("***** SAVFileReader: executing method decodeRecordType3and4");
                decodeRecordType3and4(stream);

                methodCurrentlyExecuted = "decodeRecordType6";
                dbgLog.fine("***** SAVFileReader: executing method decodeRecordType6");
                decodeRecordType6(stream);

                methodCurrentlyExecuted = "decodeRecordType7";
                dbgLog.fine("***** SAVFileReader: executing method decodeRecordType7");
                decodeRecordType7(stream);

                methodCurrentlyExecuted = "decodeRecordType999";
                dbgLog.fine("***** SAVFileReader: executing method decodeRecordType999");
                decodeRecordType999(stream);

                methodCurrentlyExecuted = "decodeRecordTypeData";
                dbgLog.fine("***** SAVFileReader: executing method decodeRecordTypeData");
                decodeRecordTypeData(stream);

            } catch (IllegalArgumentException e) {
                //Throwable cause = e.getCause();
                dbgLog.fine("***** SAVFileReader: ATTENTION: IllegalArgumentException thrown while executing "
                        + methodCurrentlyExecuted);
                e.printStackTrace();
                throw new IOException(
                        "IllegalArgumentException in method " + methodCurrentlyExecuted + ": " + e.getMessage());
            } catch (IOException e) {
                dbgLog.fine("***** SAVFileReader: ATTENTION: IOException thrown while executing "
                        + methodCurrentlyExecuted);
                e.printStackTrace();
                throw new IOException("IO Exception in method " + methodCurrentlyExecuted + ": " + e.getMessage());
            }

            /* 
             * Final variable type assignments;
             * TODO: (maybe?) 
             * Instead of doing it here, perhaps all the type assignments need to 
             * be done on DataVariable objects directly;  without relying on 
             * maps and lists here... -- L.A. 4.0 beta (?)
             */

            for (int indx = 0; indx < variableTypelList.size(); indx++) {
                String varName = dataTable.getDataVariables().get(indx).getName();
                int simpleType = 0;
                if (variableTypelList.get(indx) != null) {
                    simpleType = variableTypelList.get(indx).intValue();
                }

                if (simpleType <= 0) {
                    // We need to make one last type adjustment:
                    // Dates and Times will be stored as character values in the 
                    // dataverse tab files; even though they are not typed as 
                    // strings at this point:
                    // TODO: 
                    // Make sure the date/time format is properly preserved!
                    // (see the setFormatCategory below... but double-check!)
                    // -- L.A. 4.0 alpha
                    String variableFormatType = variableFormatTypeList[indx];
                    if (variableFormatType != null) {
                        if (variableFormatType.equals("time") || variableFormatType.equals("date")) {
                            simpleType = 1;

                            String formatCategory = formatCategoryTable.get(varName);

                            if (formatCategory != null) {
                                if (dateFormatList[indx] != null) {
                                    dbgLog.fine("setting format category to " + formatCategory);
                                    dataTable.getDataVariables().get(indx).setFormatCategory(formatCategory);
                                    dbgLog.fine("setting formatschemaname to " + dateFormatList[indx]);
                                    dataTable.getDataVariables().get(indx).setFormat(dateFormatList[indx]);
                                }
                            }
                        } else if (variableFormatType.equals("other")) {
                            dbgLog.fine("Variable of format type \"other\"; type adjustment may be needed");
                            dbgLog.fine("SPSS print format: "
                                    + printFormatTable.get(dataTable.getDataVariables().get(indx).getName()));

                            if (printFormatTable.get(dataTable.getDataVariables().get(indx).getName())
                                    .equals("WKDAY")
                                    || printFormatTable.get(dataTable.getDataVariables().get(indx).getName())
                                            .equals("MONTH")) {
                                // week day or month; 
                                // These are not treated as time/date values (meaning, we 
                                // don't define time/date formats for them; there's likely 
                                // no valid ISO time/date format for just a month or a day 
                                // of week). However, the
                                // values will be stored in the TAB files as strings, 
                                // and not as numerics - as they were stored in the 
                                // SAV file. So we need to adjust the type here.
                                // -- L.A. 

                                simpleType = 1;
                            }
                        }
                    }
                }

                // OK, we can now assign the types: 

                if (simpleType > 0) {
                    // String: 
                    dataTable.getDataVariables().get(indx).setTypeCharacter();
                    dataTable.getDataVariables().get(indx).setIntervalDiscrete();
                } else {
                    // Numeric: 
                    dataTable.getDataVariables().get(indx).setTypeNumeric();
                    // discrete or continuous?
                    // "decimal variables" become dataverse data variables of interval type "continuous":

                    if (decimalVariableSet.contains(indx)) {
                        dataTable.getDataVariables().get(indx).setIntervalContinuous();
                    } else {
                        dataTable.getDataVariables().get(indx).setIntervalDiscrete();
                    }

                }

                // TODO: take care of the SPSS "shortToLongVariableNameTable"
                // mapping before returning the ingested data object. -- 4.0 alpha
                // (done, below - but verify!)

                if (shortToLongVariableNameTable.containsKey(varName)) {
                    String longName = shortToLongVariableNameTable.get(varName);
                    if (longName != null && !longName.equals("")) {
                        dataTable.getDataVariables().get(indx).setName(longName);
                    }
                }

            }

            ingesteddata.setDataTable(dataTable);
        } catch (Exception ex) {
            dbgLog.fine("***** SAVFileReader: ATTENTION: unknown exception thrown.");
            ex.printStackTrace();
            String failureMessage = "Unknown exception in SPSS/SAV reader";
            if (ex.getMessage() != null) {
                failureMessage = failureMessage.concat(": " + ex.getMessage());
            } else {
                failureMessage = failureMessage.concat("; no further information is available.");
            }
            throw new IOException(failureMessage);
        }
        dbgLog.info("SAVFileReader: read() end");
        return ingesteddata;
    }

    void decodeHeader(BufferedInputStream stream) throws IOException {
        dbgLog.fine("decodeHeader(): start");

        if (stream == null) {
            throw new IllegalArgumentException("stream == null!");
        }
        // the length of the magic number is 4 (1-byte character * 4)
        // its value is expected to be $FL2

        byte[] b = new byte[SAV_MAGIC_NUMBER_LENGTH];

        try {
            if (stream.markSupported()) {
                stream.mark(100);
            }
            int nbytes = stream.read(b, 0, SAV_MAGIC_NUMBER_LENGTH);

            if (nbytes == 0) {
                throw new IOException();
            }

        } catch (IOException ex) {
            //ex.printStackTrace();
            throw ex;
        }

        //printHexDump(b, "hex dump of the byte-array");

        String hdr4sav = new String(b);
        dbgLog.fine("from string=" + hdr4sav);

        if (hdr4sav.equals(SAV_FILE_SIGNATURE)) {
            dbgLog.fine("this file is spss-sav type");
            // initialize version-specific parameter
            init();

            dataTable.setOriginalFileFormat(MIME_TYPE[0]);

            dataTable.setUnf("UNF:6:");

        } else {
            dbgLog.fine("this file is NOT spss-sav type");

            throw new IllegalArgumentException("given file is not spss-sav type");
        }

        dbgLog.fine("***** decodeHeader(): end *****");

    }

    void decodeRecordType1(BufferedInputStream stream) throws IOException {
        dbgLog.fine("***** decodeRecordType1(): start *****");

        if (stream == null) {
            throw new IllegalArgumentException("stream == null!");
        }
        // how to read each recordType
        // 1. set-up the following objects before reading bytes
        // a. the working byte array
        // b. the storage object
        // the length of this field: 172bytes = 60 + 4 + 12 + 4 + 8 + 84
        // this field consists of 6 distinct blocks

        byte[] recordType1 = new byte[LENGTH_RECORDTYPE1];
        // int caseWeightVariableOBSIndex = 0; 

        try {
            int nbytes = stream.read(recordType1, 0, LENGTH_RECORDTYPE1);

            //printHexDump(recordType1, "recordType1");

            if (nbytes == 0) {
                throw new IOException("reading recordType1: no byte was read");
            }

            // 1.1 60 byte-String that tells the platform/version of SPSS that
            // wrote this file

            int offset_start = 0;
            int offset_end = LENGTH_SPSS_PRODUCT_INFO; // 60 bytes

            String productInfo = new String(Arrays.copyOfRange(recordType1, offset_start, offset_end), "US-ASCII");

            dbgLog.fine("productInfo:\n" + productInfo + "\n");
            dataTable.setOriginalFormatVersion(productInfo);

            // try to parse out the SPSS version that created this data
            // file: 

            String spssVersionTag = null;

            String regexpVersionNumber = ".*Release ([0-9]*)";
            Pattern versionTagPattern = Pattern.compile(regexpVersionNumber);
            Matcher matcher = versionTagPattern.matcher(productInfo);
            if (matcher.find()) {
                spssVersionTag = matcher.group(1);
                dbgLog.fine("SPSS Version Number: " + spssVersionTag);
            }

            // TODO: 
            // try a more elaborate regex (like the one for the "new-style" 
            // productInfo line, below), to select the version number, the 
            // minor version number and the platform (windows vs. mac) separately. 
            // would be cleaner to save just that, rather than the entire 
            // productInfo tag. 
            // -- L.A. 4.0 beta

            if (spssVersionTag == null || spssVersionTag.equals("")) {
                // Later versions of SPSS have different formatting of the
                // productInfo line:
                regexpVersionNumber = ".* IBM SPSS STATISTICS.* ([^ ]*) ([0-9][0-9]*)([^ ]*)";
                versionTagPattern = Pattern.compile(regexpVersionNumber);
                matcher = versionTagPattern.matcher(productInfo);
                if (matcher.find()) {
                    String spssPlatformTag = matcher.group(1);
                    spssVersionTag = matcher.group(2);
                    String spssVersionTagMinor = matcher.group(3);

                    dbgLog.fine("SPSS Version Number (new style): " + spssVersionTag);
                    dbgLog.fine("SPSS Version/Platform Identification (new style:) " + spssPlatformTag + " "
                            + spssVersionTag + spssVersionTagMinor);
                    dataTable
                            .setOriginalFormatVersion(spssVersionTag + spssVersionTagMinor + " " + spssPlatformTag);

                }
            }

            if (spssVersionTag != null && !spssVersionTag.equals("")) {
                spssVersionNumber = Integer.valueOf(spssVersionTag).intValue();

                /*
                 *  Starting with SPSS version 16, the default encoding is 
                 *  UTF-8. 
                 *  But we are only going to use it if the user did not explicitly
                 *  specify the encoding on the addfiles page. Then we'd want 
                 *  to stick with whatever they entered. 
                 *  (also, it appears that (starting with the same version 16?)
                 *  it is actually possible to define the locale/character set
                 *  in the file - section 7, sub-type 20; TODO: decide which 
                 *  one takes precedence, if we have the encoding defined both
                 *  in the file and through the UI. -- L.A. 4.0 beta)
                 */
                if (spssVersionNumber > 15) {
                    if (getDataLanguageEncoding() == null) {
                        //defaultCharSet = "windows-1252"; // temporary! -- L.A. "UTF-8";
                        defaultCharSet = "UTF-8";
                    }
                }
            }

            // TODO: 
            // decide if we want to save the [determined/guessed] character set
            // somewhere in the dataset object. 
            // this may be relevant in cases when accented/non-latin characters
            // get ingested incorrectly; 
            // -- L.A. 4.0 beta

            // 1.2) 4-byte file-layout-code (byte-order)

            offset_start = offset_end;
            offset_end += LENGTH_FILE_LAYOUT_CODE; // 4 byte

            ByteBuffer bb_fileLayout_code = ByteBuffer.wrap(recordType1, offset_start, LENGTH_FILE_LAYOUT_CODE);

            ByteBuffer byteOderTest = bb_fileLayout_code.duplicate();
            // interprete the 4 byte as int

            int int2test = byteOderTest.getInt();

            if (int2test == 2 || int2test == 3) {
                dbgLog.fine("integer == " + int2test + ": the byte-oder of the writer is the same "
                        + "as the counterpart of Java: Big Endian");
            } else {
                // Because Java's byte-order is always big endian, 
                // this(!=2) means this sav file was  written on a little-endian machine
                // non-string, multi-bytes blocks must be byte-reversed

                bb_fileLayout_code.order(ByteOrder.LITTLE_ENDIAN);

                int2test = bb_fileLayout_code.getInt();

                if (int2test == 2 || int2test == 3) {
                    dbgLog.fine("The sav file was saved on a little endian machine");
                    dbgLog.fine("Reveral of the bytes is necessary to decode " + "multi-byte, non-string blocks");

                    isLittleEndian = true;

                } else {
                    throw new IOException("reading recordType1:unknown file layout code=" + int2test);
                }
            }

            dbgLog.fine("Endian of this platform:" + ByteOrder.nativeOrder().toString());

            // 1.3 4-byte Number_Of_OBS_Units_Per_Case 
            // (= how many RT2 records => how many varilables)

            offset_start = offset_end;
            offset_end += LENGTH_NUMBER_OF_OBS_UNITS_PER_CASE; // 4 byte

            ByteBuffer bb_OBS_units_per_case = ByteBuffer.wrap(recordType1, offset_start,
                    LENGTH_NUMBER_OF_OBS_UNITS_PER_CASE);

            if (isLittleEndian) {
                bb_OBS_units_per_case.order(ByteOrder.LITTLE_ENDIAN);
            }

            OBSUnitsPerCase = bb_OBS_units_per_case.getInt();

            dbgLog.fine("RT1: OBSUnitsPerCase=" + OBSUnitsPerCase);

            // 1.4 4-byte Compression_Switch

            offset_start = offset_end;
            offset_end += LENGTH_COMPRESSION_SWITCH; // 4 byte

            ByteBuffer bb_compression_switch = ByteBuffer.wrap(recordType1, offset_start,
                    LENGTH_COMPRESSION_SWITCH);

            if (isLittleEndian) {
                bb_compression_switch.order(ByteOrder.LITTLE_ENDIAN);
            }

            int compression_switch = bb_compression_switch.getInt();
            if (compression_switch == 0) {
                // data section is not compressed
                isDataSectionCompressed = false;
                dbgLog.fine("data section is not compressed");
            } else {
                dbgLog.fine("data section is compressed:" + compression_switch);
            }

            // 1.5 4-byte Case-Weight Variable Index
            // warning: this variable index starts from 1, not 0

            offset_start = offset_end;
            offset_end += LENGTH_CASE_WEIGHT_VARIABLE_INDEX; // 4 byte

            ByteBuffer bb_Case_Weight_Variable_Index = ByteBuffer.wrap(recordType1, offset_start,
                    LENGTH_CASE_WEIGHT_VARIABLE_INDEX);

            if (isLittleEndian) {
                bb_Case_Weight_Variable_Index.order(ByteOrder.LITTLE_ENDIAN);
            }

            caseWeightVariableOBSIndex = bb_Case_Weight_Variable_Index.getInt();

            /// caseWeightVariableOBSIndex will be used later on to locate 
            /// the weight variable; so we'll be able to mark the corresponding
            /// variables properly. 
            // TODO: make sure case weight variables are properly handled! 
            // -- L.A. 4.0 beta
            ///smd.getFileInformation().put("caseWeightVariableOBSIndex", caseWeightVariableOBSIndex);

            // 1.6 4-byte Number of Cases

            offset_start = offset_end;
            offset_end += LENGTH_NUMBER_OF_CASES; // 4 byte

            ByteBuffer bb_Number_Of_Cases = ByteBuffer.wrap(recordType1, offset_start, LENGTH_NUMBER_OF_CASES);

            if (isLittleEndian) {
                bb_Number_Of_Cases.order(ByteOrder.LITTLE_ENDIAN);
            }

            int numberOfCases = bb_Number_Of_Cases.getInt();

            if (numberOfCases < 0) {
                // -1 if numberOfCases is unknown
                throw new RuntimeException("number of cases is not recorded in the header");
            } else {
                dbgLog.fine("RT1: number of cases is recorded= " + numberOfCases);
                dataTable.setCaseQuantity(new Long(numberOfCases));
                ///caseQnty = numberOfCases;
                ///smd.getFileInformation().put("caseQnty", numberOfCases);
            }

            // 1.7 8-byte compression-bias [not long but double]

            offset_start = offset_end;
            offset_end += LENGTH_COMPRESSION_BIAS; // 8 byte

            ByteBuffer bb_compression_bias = ByteBuffer
                    .wrap(Arrays.copyOfRange(recordType1, offset_start, offset_end));

            if (isLittleEndian) {
                bb_compression_bias.order(ByteOrder.LITTLE_ENDIAN);
            }

            Double compressionBias = bb_compression_bias.getDouble();

            // TODO: 
            // check if this "compression bias" is being used anywhere? 
            // doesn't seem to be!
            // -- 4.0 alpha
            if (compressionBias == 100d) {
                // 100 is expected
                dbgLog.fine("compressionBias is 100 as expected");
                ///smd.getFileInformation().put("compressionBias", 100);
            } else {
                dbgLog.fine("compression bias is not 100: " + compressionBias);
                ///smd.getFileInformation().put("compressionBias", compressionBias);
            }

            // 1.8 84-byte File Creation Information (date/time: dd MM yyhh:mm:ss +
            // 64-bytelabel)

            offset_start = offset_end;
            offset_end += LENGTH_FILE_CREATION_INFO; // 84 bytes

            String fileCreationInfo = getNullStrippedString(
                    new String(Arrays.copyOfRange(recordType1, offset_start, offset_end), "US-ASCII"));

            dbgLog.fine("fileCreationInfo:\n" + fileCreationInfo + "\n");

            String fileCreationDate = fileCreationInfo.substring(0, length_file_creation_date);
            int dateEnd = length_file_creation_date + length_file_creation_time;
            String fileCreationTime = fileCreationInfo.substring(length_file_creation_date, (dateEnd));
            String fileCreationNote = fileCreationInfo.substring(dateEnd, length_file_creation_label);

            dbgLog.fine("fileDate=" + fileCreationDate);
            dbgLog.fine("fileTime=" + fileCreationTime);
            dbgLog.fine("fileNote" + fileCreationNote);

        } catch (IOException ex) {
            throw ex;
        }

        dbgLog.fine("decodeRecordType1(): end");
    }

    void decodeRecordType2(BufferedInputStream stream) throws IOException {
        dbgLog.fine("decodeRecordType2(): start");
        if (stream == null) {
            throw new IllegalArgumentException("stream == null!");
        }

        Map<String, String> printFormatNameTable = new LinkedHashMap<String, String>();
        Map<String, String> variableLabelMap = new LinkedHashMap<String, String>();
        Map<String, List<String>> missingValueTable = new LinkedHashMap<String, List<String>>();
        List<Integer> printFormatList = new ArrayList<Integer>();

        String caseWeightVariableName = null;
        int caseWeightVariableIndex = 0;

        boolean lastVariableIsExtendable = false;
        boolean extendedVariableMode = false;
        boolean obs255 = false;

        String lastVariableName = null;
        String lastExtendedVariable = null;

        // this field repeats as many as the number of variables in
        // this sav file

        // (note that the above statement is not technically correct, this
        //  record repeats not just for every variable in the file, but for
        //  every OBS (8 byte unit); i.e., if a string is split into multiple
        //  OBS units, each one will have its own RT2 record -- L.A.).

        // Each field constists of a fixed (32-byte) segment and
        // then a few variable segments:
        // if the variable has a label (3rd INT4 set to 1), then there's 4 more
        // bytes specifying the length of the label, and then that many bytes
        // holding the label itself (no more than 256).
        // Then if there are optional missing value units (4th INT4 set to 1)
        // there will be 3 more OBS units attached = 24 extra bytes.

        int variableCounter = 0;
        int obsSeqNumber = 0;

        int j;

        dbgLog.fine("RT2: Reading " + OBSUnitsPerCase + " OBS units.");

        for (j = 0; j < OBSUnitsPerCase; j++) {

            dbgLog.fine("RT2: " + j + "-th RT2 unit is being decoded.");
            // 2.0: read the fixed[=non-optional] 32-byte segment
            byte[] recordType2Fixed = new byte[LENGTH_RECORDTYPE2_FIXED];

            try {
                int nbytes = stream.read(recordType2Fixed, 0, LENGTH_RECORDTYPE2_FIXED);

                //printHexDump(recordType2Fixed, "recordType2 part 1");

                if (nbytes == 0) {
                    throw new IOException("reading recordType2: no bytes read!");
                }

                int offset = 0;

                // 2.1: create int-view of the bytebuffer for the first 16-byte segment
                int rt2_1st_4_units = 4;
                ByteBuffer[] bb_record_type2_fixed_part1 = new ByteBuffer[rt2_1st_4_units];
                int[] recordType2FixedPart1 = new int[rt2_1st_4_units];
                for (int i = 0; i < rt2_1st_4_units; i++) {

                    bb_record_type2_fixed_part1[i] = ByteBuffer.wrap(recordType2Fixed, offset,
                            LENGTH_SAV_INT_BLOCK);

                    offset += LENGTH_SAV_INT_BLOCK;
                    if (isLittleEndian) {
                        bb_record_type2_fixed_part1[i].order(ByteOrder.LITTLE_ENDIAN);
                    }
                    recordType2FixedPart1[i] = bb_record_type2_fixed_part1[i].getInt();
                }

                ///dbgLog.fine("recordType2FixedPart="+
                ///        ReflectionToStringBuilder.toString(recordType2FixedPart1, ToStringStyle.MULTI_LINE_STYLE));

                // 1st ([0]) element must be 2 otherwise no longer Record Type 2
                if (recordType2FixedPart1[0] != 2) {
                    dbgLog.warning(j + "-th RT header value is no longet RT2! " + recordType2FixedPart1[0]);
                    break;
                }
                dbgLog.fine("variable type[must be 2]=" + recordType2FixedPart1[0]);

                // 2.3 variable name: 8 byte(space[x20]-padded)
                // This field is located at the very end of the 32 byte
                // fixed-size RT2 header (bytes 24-31).
                // We are processing it now, so that
                // we can make the decision on whether this variable is part
                // of a compound variable:

                String RawVariableName = getNullStrippedString(new String(
                        Arrays.copyOfRange(recordType2Fixed, 24, (24 + LENGTH_VARIABLE_NAME)), defaultCharSet));
                //offset +=LENGTH_VARIABLE_NAME;
                String variableName = null;
                if (RawVariableName.indexOf(' ') >= 0) {
                    variableName = RawVariableName.substring(0, RawVariableName.indexOf(' '));
                } else {
                    variableName = RawVariableName;
                }

                // 2nd ([1]) element: numeric variable = 0 :for string variable
                // this block indicates its datum-length, i.e, >0 ;
                // if -1, this RT2 unit is a non-1st RT2 unit for a string variable
                // whose value is longer than 8 character.

                boolean isNumericVariable = false;

                dbgLog.fine("variable type(0: numeric; > 0: String;-1 continue )=" + recordType2FixedPart1[1]);

                //OBSwiseTypelList.add(recordType2FixedPart1[1]);

                int HowManyRt2Units = 1;

                if (recordType2FixedPart1[1] == -1) {
                    dbgLog.fine("this RT2 is an 8 bit continuation chunk of an earlier string variable");
                    if (obs255) {
                        if (obsSeqNumber < 30) {
                            OBSwiseTypelList.add(recordType2FixedPart1[1]);
                            obsSeqNumber++;
                        } else {
                            OBSwiseTypelList.add(-2);
                            obs255 = false;
                            obsSeqNumber = 0;
                        }
                    } else {
                        OBSwiseTypelList.add(recordType2FixedPart1[1]);
                    }

                    obsNonVariableBlockSet.add(j);
                    continue;
                } else if (recordType2FixedPart1[1] == 0) {
                    // This is a numeric variable
                    extendedVariableMode = false;
                    // And as such, it cannot be an extension of a
                    // previous, long string variable.
                    OBSwiseTypelList.add(recordType2FixedPart1[1]);
                    variableCounter++;
                    isNumericVariable = true;
                    variableTypelList.add(recordType2FixedPart1[1]);
                } else if (recordType2FixedPart1[1] > 0) {

                    // This looks like a regular string variable. However,
                    // it may still be a part of a compound variable
                    // (a String > 255 bytes that was split into 255 byte
                    // chunks, stored as individual String variables).

                    if (recordType2FixedPart1[1] == 255) {
                        obs255 = true;
                    }

                    if (lastVariableIsExtendable) {
                        String varNameBase = null;
                        if (lastVariableName.length() > 5) {
                            varNameBase = lastVariableName.substring(0, 5);
                        } else {
                            varNameBase = lastVariableName;
                        }

                        if (extendedVariableMode) {
                            if (variableNameIsAnIncrement(varNameBase, lastExtendedVariable, variableName)) {
                                OBSwiseTypelList.add(-1);
                                lastExtendedVariable = variableName;
                                // OK, we stay in the "extended variable" mode;
                                // but we can't move on to the next OBS (hence the commented out
                                // "continue" below:
                                //continue;
                                // see the next comment below for the explanation.
                                //
                                // Should we also set "extendable" flag to false at this point
                                // if it's shorter than 255 bytes, i.e. the last extended chunk?
                            } else {
                                extendedVariableMode = false;
                            }
                        } else {
                            if (variableNameIsAnIncrement(varNameBase, variableName)) {
                                OBSwiseTypelList.add(-1);
                                extendedVariableMode = true;
                                dbgLog.fine("RT2: in extended variable mode; variable " + variableName);
                                lastExtendedVariable = variableName;
                                // Before we move on to the next OBS unit, we need to check
                                // if this current extended variable has its own label specified;
                                // If so, we need to determine its length, then read and skip
                                // that many bytes.
                                // Hence the commented out "continue" below:
                                //continue;
                            }
                        }
                    }

                    if (!extendedVariableMode) {
                        // OK, this is a "real"
                        // string variable, and not a continuation chunk of a compound
                        // string.

                        OBSwiseTypelList.add(recordType2FixedPart1[1]);
                        variableCounter++;

                        if (recordType2FixedPart1[1] == 255) {
                            // This variable is 255 bytes long, i.e. this is
                            // either the single "atomic" variable of the
                            // max allowed size, or it's a 255 byte segment
                            // of a compound variable. So we will check
                            // the next variable and see if it is the continuation
                            // of this one.

                            lastVariableIsExtendable = true;
                        } else {
                            lastVariableIsExtendable = false;
                        }

                        if (recordType2FixedPart1[1] % LENGTH_SAV_OBS_BLOCK == 0) {
                            HowManyRt2Units = recordType2FixedPart1[1] / LENGTH_SAV_OBS_BLOCK;
                        } else {
                            HowManyRt2Units = recordType2FixedPart1[1] / LENGTH_SAV_OBS_BLOCK + 1;
                        }
                        variableTypelList.add(recordType2FixedPart1[1]);
                    }
                }

                if (!extendedVariableMode) {
                    // Again, we only want to do the following steps for the "real"
                    // variables, not the chunks of split mega-variables:

                    dbgLog.fine("RT2: HowManyRt2Units for this variable=" + HowManyRt2Units);

                    lastVariableName = variableName;

                    // caseWeightVariableOBSIndex starts from 1: 0 is used for does-not-exist cases
                    if (j == (caseWeightVariableOBSIndex - 1)) {
                        caseWeightVariableName = variableName;
                        // TODO: do we need this "index"? -- 4.0 alpha
                        caseWeightVariableIndex = variableCounter;

                        ///smd.setCaseWeightVariableName(caseWeightVariableName);
                        ///smd.getFileInformation().put("caseWeightVariableIndex", caseWeightVariableIndex);
                    }

                    OBSIndexToVariableName.put(j, variableName);

                    //dbgLog.fine("\nvariable name="+variableName+"<-");
                    dbgLog.fine("RT2: " + j + "-th variable name=" + variableName + "<-");
                    dbgLog.fine("RT2: raw variable: " + RawVariableName);

                    variableNameList.add(variableName);
                }

                // 3rd ([2]) element: = 1 variable-label block follows; 0 = no label
                //
                dbgLog.fine("RT: variable label follows?(1:yes; 0: no)=" + recordType2FixedPart1[2]);
                boolean hasVariableLabel = recordType2FixedPart1[2] == 1 ? true : false;
                if ((recordType2FixedPart1[2] != 0) && (recordType2FixedPart1[2] != 1)) {
                    throw new IOException("RT2: reading error: value is neither 0 or 1" + recordType2FixedPart1[2]);
                }

                // 2.4 [optional]The length of a variable label followed: 4-byte int
                // 3rd element of 2.1 indicates whether this field exists
                // *** warning: The label block is padded to a multiple of the 4-byte
                // NOT the raw integer value of this 4-byte block

                if (hasVariableLabel) {
                    byte[] length_variable_label = new byte[4];
                    int nbytes_2_4 = stream.read(length_variable_label);
                    if (nbytes_2_4 == 0) {
                        throw new IOException("RT 2: error reading recordType2.4: no bytes read!");
                    } else {
                        dbgLog.fine("nbytes_2_4=" + nbytes_2_4);
                    }
                    ByteBuffer bb_length_variable_label = ByteBuffer.wrap(length_variable_label, 0,
                            LENGTH_VARIABLE_LABEL);
                    if (isLittleEndian) {
                        bb_length_variable_label.order(ByteOrder.LITTLE_ENDIAN);
                    }
                    int rawVariableLabelLength = bb_length_variable_label.getInt();

                    dbgLog.fine("rawVariableLabelLength=" + rawVariableLabelLength);
                    int variableLabelLength = getSAVintAdjustedBlockLength(rawVariableLabelLength);
                    dbgLog.fine("RT2: variableLabelLength=" + variableLabelLength);

                    // 2.5 [optional]variable label whose length is found at 2.4

                    String variableLabel = "";

                    if (rawVariableLabelLength > 0) {
                        byte[] variable_label = new byte[variableLabelLength];
                        int nbytes_2_5 = stream.read(variable_label);
                        if (nbytes_2_5 == 0) {
                            throw new IOException("RT 2: error reading recordType2.5: " + variableLabelLength
                                    + " bytes requested, no bytes read!");
                        } else {
                            dbgLog.fine("nbytes_2_5=" + nbytes_2_5);
                        }
                        variableLabel = getNullStrippedString(new String(
                                Arrays.copyOfRange(variable_label, 0, rawVariableLabelLength), defaultCharSet));
                        dbgLog.fine("RT2: variableLabel=" + variableLabel + "<-");

                        dbgLog.fine(variableName + " => " + variableLabel);
                    } else {
                        dbgLog.fine("RT2: defaulting to empty variable label.");
                    }

                    if (!extendedVariableMode) {
                        // We only have any use for this label if it's a "real" variable.
                        // Thinking about it, it doesn't make much sense for the "fake"
                        // variables that are actually chunks of large strings to store
                        // their own labels. But in some files they do. Then failing to read
                        // the bytes would result in getting out of sync with the RT record
                        // borders. So we always read the bytes, but only use them for
                        // the real variable entries.
                        /*String variableLabel = new String(Arrays.copyOfRange(variable_label,
                            0, rawVariableLabelLength),"US-ASCII");*/

                        variableLabelMap.put(variableName, variableLabel);
                    }
                }

                if (extendedVariableMode) {
                    // there's nothing else left for us to do in this iteration of the loop.
                    // Once again, this was not a real variable, but a dummy variable entry
                    // created for a chunk of a string variable longer than 255 bytes --
                    // that's how SPSS stores them.
                    continue;
                }

                // 4th ([3]) element: Missing value type code
                // 0[none], 1, 2, 3 [point-type],-2[range], -3 [range type+ point]

                dbgLog.fine("RT: missing value unit follows?(if 0, none)=" + recordType2FixedPart1[3]);
                boolean hasMissingValues = (validMissingValueCodeSet.contains(recordType2FixedPart1[3])
                        && (recordType2FixedPart1[3] != 0)) ? true : false;

                InvalidData invalidDataInfo = null;

                if (recordType2FixedPart1[3] != 0) {
                    invalidDataInfo = new InvalidData(recordType2FixedPart1[3]);
                    dbgLog.fine("RT: missing value type=" + invalidDataInfo.getType());
                }

                // 2.2: print/write formats: 4-byte each = 8 bytes

                byte[] printFormt = Arrays.copyOfRange(recordType2Fixed, offset, offset + LENGTH_PRINT_FORMAT_CODE);
                dbgLog.fine("printFrmt=" + new String(Hex.encodeHex(printFormt)));

                offset += LENGTH_PRINT_FORMAT_CODE;
                int formatCode = isLittleEndian ? printFormt[2] : printFormt[1];
                int formatWidth = isLittleEndian ? printFormt[1] : printFormt[2];

                // TODO: 
                // What should we be doing with these "format decimal positions" 
                // in 4.0? 
                // -- L.A. 4.0 alpha

                int formatDecimalPointPosition = isLittleEndian ? printFormt[0] : printFormt[3];
                dbgLog.fine("RT2: format code{5=F, 1=A[String]}=" + formatCode);

                formatDecimalPointPositionList.add(formatDecimalPointPosition);

                if (!SPSSConstants.FORMAT_CODE_TABLE_SAV.containsKey(formatCode)) {
                    throw new IOException("Unknown format code was found = " + formatCode);
                } else {
                    printFormatList.add(formatCode);
                }

                byte[] writeFormt = Arrays.copyOfRange(recordType2Fixed, offset, offset + LENGTH_WRITE_FORMAT_CODE);

                dbgLog.fine("RT2: writeFrmt=" + new String(Hex.encodeHex(writeFormt)));
                if (writeFormt[3] != 0x00) {
                    dbgLog.fine("byte-order(write format): reversal required");
                }

                offset += LENGTH_WRITE_FORMAT_CODE;

                if (!SPSSConstants.ORDINARY_FORMAT_CODE_SET.contains(formatCode)) {
                    StringBuilder sb = new StringBuilder(
                            SPSSConstants.FORMAT_CODE_TABLE_SAV.get(formatCode) + formatWidth);
                    if (formatDecimalPointPosition > 0) {
                        sb.append("." + formatDecimalPointPosition);
                    }
                    dbgLog.fine("formattable[i] = " + variableName + " -> " + sb.toString());
                    printFormatNameTable.put(variableName, sb.toString());

                }

                printFormatTable.put(variableName, SPSSConstants.FORMAT_CODE_TABLE_SAV.get(formatCode));

                // 2.6 [optional] missing values:4-byte each if exists
                // 4th element of 2.1 indicates the structure of this sub-field

                // Should we perhaps check for this for the "fake" variables too?
                //

                if (hasMissingValues) {
                    dbgLog.fine("RT2: decoding missing value: type=" + recordType2FixedPart1[3]);
                    int howManyMissingValueUnits = missingValueCodeUnits.get(recordType2FixedPart1[3]);
                    //int howManyMissingValueUnits = recordType2FixedPart1[3] > 0 ? recordType2FixedPart1[3] :  0;

                    dbgLog.fine("RT2: howManyMissingValueUnits=" + howManyMissingValueUnits);

                    byte[] missing_value_code_units = new byte[LENGTH_SAV_OBS_BLOCK * howManyMissingValueUnits];
                    int nbytes_2_6 = stream.read(missing_value_code_units);

                    if (nbytes_2_6 == 0) {
                        throw new IOException("RT 2: reading recordType2.6: no byte was read");
                    } else {
                        dbgLog.fine("nbytes_2_6=" + nbytes_2_6);
                    }

                    //printHexDump(missing_value_code_units, "missing value");

                    if (isNumericVariable) {

                        double[] missingValues = new double[howManyMissingValueUnits];
                        //List<String> mvp = new ArrayList<String>();
                        List<String> mv = new ArrayList<String>();

                        ByteBuffer[] bb_missig_value_code = new ByteBuffer[howManyMissingValueUnits];

                        int offset_start = 0;

                        for (int i = 0; i < howManyMissingValueUnits; i++) {

                            bb_missig_value_code[i] = ByteBuffer.wrap(missing_value_code_units, offset_start,
                                    LENGTH_SAV_OBS_BLOCK);

                            offset_start += LENGTH_SAV_OBS_BLOCK;
                            if (isLittleEndian) {
                                bb_missig_value_code[i].order(ByteOrder.LITTLE_ENDIAN);
                            }

                            ByteBuffer temp = bb_missig_value_code[i].duplicate();

                            missingValues[i] = bb_missig_value_code[i].getDouble();
                            if (Double.toHexString(missingValues[i]).equals("-0x1.ffffffffffffep1023")) {
                                dbgLog.fine("1st value is LOWEST");
                                mv.add(Double.toHexString(missingValues[i]));
                            } else if (Double.valueOf(missingValues[i]).equals(Double.MAX_VALUE)) {
                                dbgLog.fine("2nd value is HIGHEST");
                                mv.add(Double.toHexString(missingValues[i]));
                            } else {
                                mv.add(doubleNumberFormatter.format(missingValues[i]));
                            }
                            dbgLog.fine(i + "-th missing value=" + Double.toHexString(missingValues[i]));
                        }

                        dbgLog.fine("variableName=" + variableName);
                        if (recordType2FixedPart1[3] > 0) {
                            // point cases only
                            dbgLog.fine("mv(>0)=" + mv);
                            missingValueTable.put(variableName, mv);
                            invalidDataInfo.setInvalidValues(mv);
                        } else if (recordType2FixedPart1[3] == -2) {
                            dbgLog.fine("mv(-2)=" + mv);
                            // range
                            invalidDataInfo.setInvalidRange(mv);
                        } else if (recordType2FixedPart1[3] == -3) {
                            // mixed case
                            dbgLog.fine("mv(-3)=" + mv);
                            invalidDataInfo.setInvalidRange(mv.subList(0, 2));
                            invalidDataInfo.setInvalidValues(mv.subList(2, 3));
                            missingValueTable.put(variableName, mv.subList(2, 3));
                        }

                        dbgLog.fine("missing value=" + StringUtils.join(missingValueTable.get(variableName), "|"));
                        dbgLog.fine("invalidDataInfo(Numeric):\n" + invalidDataInfo);
                        invalidDataTable.put(variableName, invalidDataInfo);
                    } else {
                        // string variable case
                        String[] missingValues = new String[howManyMissingValueUnits];
                        List<String> mv = new ArrayList<String>();
                        int offset_start = 0;
                        int offset_end = LENGTH_SAV_OBS_BLOCK;
                        for (int i = 0; i < howManyMissingValueUnits; i++) {

                            missingValues[i] = StringUtils.stripEnd(new String(
                                    Arrays.copyOfRange(missing_value_code_units, offset_start, offset_end),
                                    defaultCharSet), " ");
                            dbgLog.fine("missing value=" + missingValues[i] + "<-");

                            offset_start = offset_end;
                            offset_end += LENGTH_SAV_OBS_BLOCK;

                            mv.add(missingValues[i]);
                        }
                        invalidDataInfo.setInvalidValues(mv);
                        missingValueTable.put(variableName, mv);
                        invalidDataTable.put(variableName, invalidDataInfo);
                        dbgLog.fine(
                                "missing value(str)=" + StringUtils.join(missingValueTable.get(variableName), "|"));
                        dbgLog.fine("invalidDataInfo(String):\n" + invalidDataInfo);

                    } // string case
                    dbgLog.fine("invalidDataTable:\n" + invalidDataTable);
                } // if msv

            } catch (IOException ex) {
                //ex.printStackTrace();
                throw ex;
            } catch (Exception ex) {
                ex.printStackTrace();
                // should we be throwing some exception here?
            }
        } // j-loop

        if (j != OBSUnitsPerCase) {
            dbgLog.fine("RT2: attention! didn't reach the end of the OBS list!");
            throw new IOException("RT2: didn't reach the end of the OBS list!");
        }

        dbgLog.fine("RT2 metadata-related exit-chores");
        ///smd.getFileInformation().put("varQnty", variableCounter);
        dataTable.setVarQuantity(new Long(variableCounter));
        dbgLog.fine("RT2: varQnty=" + variableCounter);

        // 4.0 Initialize variables: 
        List<DataVariable> variableList = new ArrayList<DataVariable>();

        for (int i = 0; i < variableCounter; i++) {
            DataVariable dv = new DataVariable();
            String varName = variableNameList.get(i);
            dbgLog.fine("name: " + varName);
            dv.setName(varName);
            String varLabel = variableLabelMap.get(varName);
            if (varLabel != null && varLabel.length() > 255) {
                // TODO: 
                // variable labels will be changed into type 'TEXT' in the 
                // database - this will eliminate the 255 char. limit. 
                // -- L.A. 4.0 beta11
                dbgLog.fine("Have to truncate label: " + varLabel);
                varLabel = varLabel.substring(0, 255);
            }
            dbgLog.fine("label: " + varLabel);
            dv.setLabel(varLabel);

            dv.setInvalidRanges(new ArrayList<VariableRange>());
            dv.setSummaryStatistics(new ArrayList<SummaryStatistic>());
            dv.setUnf("UNF:6:");
            dv.setCategories(new ArrayList<VariableCategory>());
            variableList.add(dv);

            dv.setFileOrder(i);

            dv.setDataTable(dataTable);
        }

        dataTable.setDataVariables(variableList);

        ///smd.setVariableName(variableNameList.toArray(new String[variableNameList.size()]));
        ///smd.setVariableLabel(variableLabelMap);
        // TODO: 
        // figure out what to do with the missing value table!
        // -- 4.0 alpha
        // well, they were used to generate merged summary statistics for 
        // the variable. So need to verify what the DDI import was doing 
        // with them and replicate the same in 4.0.
        // (add appropriate value labels?)
        ///TODO: 4.0 smd.setMissingValueTable(missingValueTable);
        ///smd.getFileInformation().put("caseWeightVariableName", caseWeightVariableName);

        dbgLog.fine("sumstat:long case=" + Arrays.deepToString(variableTypelList.toArray()));

        dbgLog.fine("RT2: OBSwiseTypelList=" + OBSwiseTypelList);

        dbgLog.fine("decodeRecordType2(): end");
    }

    void decodeRecordType3and4(BufferedInputStream stream) throws IOException {
        dbgLog.fine("decodeRecordType3and4(): start");
        Map<String, Map<String, String>> valueLabelTable = new LinkedHashMap<String, Map<String, String>>();

        int safteyCounter = 0;
        while (true) {
            try {
                if (stream == null) {
                    throw new IllegalArgumentException("stream == null!");
                }
                // this secton may not exit so first check the 4-byte header value
                //if (stream.markSupported()){
                stream.mark(1000);
                //}
                // 3.0 check the first 4 bytes
                byte[] headerCode = new byte[LENGTH_RECORD_TYPE3_CODE];

                int nbytes_rt3 = stream.read(headerCode, 0, LENGTH_RECORD_TYPE3_CODE);
                // to-do check against nbytes
                //printHexDump(headerCode, "RT3 header test");
                ByteBuffer bb_header_code = ByteBuffer.wrap(headerCode, 0, LENGTH_RECORD_TYPE3_CODE);
                if (isLittleEndian) {
                    bb_header_code.order(ByteOrder.LITTLE_ENDIAN);
                }

                int intRT3test = bb_header_code.getInt();
                dbgLog.fine("header test value: RT3=" + intRT3test);
                if (intRT3test != 3) {
                    //if (stream.markSupported()){
                    dbgLog.fine("iteration=" + safteyCounter);

                    // We have encountered a record that's not type 3. This means we've
                    // processed all the type 3/4 record pairs. So we want to rewind
                    // the stream and return -- so that the appropriate record type
                    // reader can be called on it.
                    // But before we return, we need to save all the value labels
                    // we have found:
                    //smd.setValueLabelTable(valueLabelTable);
                    assignValueLabels(valueLabelTable);

                    stream.reset();
                    return;
                    //}
                }
                // 3.1 how many value-label pairs follow
                byte[] number_of_labels = new byte[LENGTH_RT3_HOW_MANY_LABELS];

                int nbytes_3_1 = stream.read(number_of_labels);
                if (nbytes_3_1 == 0) {
                    throw new IOException("RT 3: reading recordType3.1: no byte was read");
                }
                ByteBuffer bb_number_of_labels = ByteBuffer.wrap(number_of_labels, 0, LENGTH_RT3_HOW_MANY_LABELS);
                if (isLittleEndian) {
                    bb_number_of_labels.order(ByteOrder.LITTLE_ENDIAN);
                }

                int numberOfValueLabels = bb_number_of_labels.getInt();
                dbgLog.fine("number of value-label pairs=" + numberOfValueLabels);

                ByteBuffer[] tempBB = new ByteBuffer[numberOfValueLabels];

                String valueLabel[] = new String[numberOfValueLabels];

                for (int i = 0; i < numberOfValueLabels; i++) {

                    // read 8-byte as value          
                    byte[] value = new byte[LENGTH_RT3_VALUE];
                    int nbytes_3_value = stream.read(value);

                    if (nbytes_3_value == 0) {
                        throw new IOException("RT 3: reading recordType3 value: no byte was read");
                    }
                    // note these 8 bytes are interpreted later
                    // currently no information about which variable's (=> type unknown)
                    ByteBuffer bb_value = ByteBuffer.wrap(value, 0, LENGTH_RT3_VALUE);
                    if (isLittleEndian) {
                        bb_value.order(ByteOrder.LITTLE_ENDIAN);
                    }
                    tempBB[i] = bb_value;
                    dbgLog.fine("bb_value=" + Hex.encodeHex(bb_value.array()));
                    /*
                     double valueD = bb_value.getDouble();                
                     dbgLog.fine("value="+valueD);
                     */
                    // read 1st byte as unsigned integer = label_length

                    // read label_length byte as label
                    byte[] labelLengthByte = new byte[LENGTH_RT3_LABEL_LENGTH];

                    int nbytes_3_label_length = stream.read(labelLengthByte);

                    // add check-routine here
                    dbgLog.fine("labelLengthByte" + Hex.encodeHex(labelLengthByte));
                    dbgLog.fine("label length = " + labelLengthByte[0]);
                    // the net-length of a value label is saved as
                    // unsigned byte; however, the length is less than 127
                    // byte should be ok
                    int rawLabelLength = labelLengthByte[0] & 0xFF;
                    dbgLog.fine("rawLabelLength=" + rawLabelLength);
                    // -1 =>1-byte already read
                    int labelLength = getSAVobsAdjustedBlockLength(rawLabelLength + 1) - 1;
                    byte[] valueLabelBytes = new byte[labelLength];
                    int nbytes_3_value_label = stream.read(valueLabelBytes);

                    // ByteBuffer bb_label = ByteBuffer.wrap(valueLabel,0,labelLength);
                    valueLabel[i] = StringUtils.stripEnd(
                            new String(Arrays.copyOfRange(valueLabelBytes, 0, rawLabelLength), defaultCharSet),
                            " ");
                    dbgLog.fine(i + "-th valueLabel=" + valueLabel[i] + "<-");

                } // iter rt3

                dbgLog.fine("end of RT3 block");
                dbgLog.fine("start of RT4 block");

                // 4.0 check the first 4 bytes
                byte[] headerCode4 = new byte[LENGTH_RECORD_TYPE4_CODE];

                int nbytes_rt4 = stream.read(headerCode4, 0, LENGTH_RECORD_TYPE4_CODE);

                if (nbytes_rt4 == 0) {
                    throw new IOException("RT4: reading recordType4 value: no byte was read");
                }

                //printHexDump(headerCode4, "RT4 header test");
                ByteBuffer bb_header_code_4 = ByteBuffer.wrap(headerCode4, 0, LENGTH_RECORD_TYPE4_CODE);
                if (isLittleEndian) {
                    bb_header_code_4.order(ByteOrder.LITTLE_ENDIAN);
                }

                int intRT4test = bb_header_code_4.getInt();
                dbgLog.fine("header test value: RT4=" + intRT4test);

                if (intRT4test != 4) {
                    throw new IOException("RT 4: reading recordType4 header: no byte was read");
                }

                // 4.1 read the how-many-variables bytes
                byte[] howManyVariablesfollow = new byte[LENGTH_RT4_HOW_MANY_VARIABLES];

                int nbytes_rt4_1 = stream.read(howManyVariablesfollow, 0, LENGTH_RT4_HOW_MANY_VARIABLES);

                ByteBuffer bb_howManyVariablesfollow = ByteBuffer.wrap(howManyVariablesfollow, 0,
                        LENGTH_RT4_HOW_MANY_VARIABLES);
                if (isLittleEndian) {
                    bb_howManyVariablesfollow.order(ByteOrder.LITTLE_ENDIAN);
                }

                int howManyVariablesRT4 = bb_howManyVariablesfollow.getInt();
                dbgLog.fine("how many variables follow: RT4=" + howManyVariablesRT4);

                int length_indicies = LENGTH_RT4_VARIABLE_INDEX * howManyVariablesRT4;
                byte[] variableIdicesBytes = new byte[length_indicies];

                int nbytes_rt4_2 = stream.read(variableIdicesBytes, 0, length_indicies);

                // !!!!! Caution: variableIndex in RT4 starts from 1 NOT ** 0 **
                int[] variableIndex = new int[howManyVariablesRT4];
                int offset = 0;
                for (int i = 0; i < howManyVariablesRT4; i++) {

                    ByteBuffer bb_variable_index = ByteBuffer.wrap(variableIdicesBytes, offset,
                            LENGTH_RT4_VARIABLE_INDEX);
                    offset += LENGTH_RT4_VARIABLE_INDEX;

                    if (isLittleEndian) {
                        bb_variable_index.order(ByteOrder.LITTLE_ENDIAN);
                    }

                    variableIndex[i] = bb_variable_index.getInt();
                    dbgLog.fine(i + "-th variable index number=" + variableIndex[i]);
                }

                dbgLog.fine("variable index set=" + ArrayUtils.toString(variableIndex));
                dbgLog.fine("subtract 1 from variableIndex for getting a variable info");

                boolean isNumeric = OBSwiseTypelList.get(variableIndex[0] - 1) == 0 ? true : false;

                Map<String, String> valueLabelPair = new LinkedHashMap<String, String>();
                if (isNumeric) {
                    // numeric variable
                    dbgLog.fine("processing of a numeric value-label table");
                    for (int j = 0; j < numberOfValueLabels; j++) {
                        valueLabelPair.put(doubleNumberFormatter.format(tempBB[j].getDouble()), valueLabel[j]);
                    }
                } else {
                    // String variable
                    dbgLog.fine("processing of a string value-label table");
                    for (int j = 0; j < numberOfValueLabels; j++) {
                        valueLabelPair.put(
                                StringUtils.stripEnd(new String((tempBB[j].array()), defaultCharSet), " "),
                                valueLabel[j]);
                    }
                }

                dbgLog.fine("valueLabePair=" + valueLabelPair);
                dbgLog.fine("key variable's (raw) index =" + variableIndex[0]);

                valueLabelTable.put(OBSIndexToVariableName.get(variableIndex[0] - 1), valueLabelPair);

                dbgLog.fine("valueLabelTable=" + valueLabelTable);

                // create a mapping table that finds the key variable for this mapping table
                String keyVariableName = OBSIndexToVariableName.get(variableIndex[0] - 1);
                for (int vn : variableIndex) {
                    valueVariableMappingTable.put(OBSIndexToVariableName.get(vn - 1), keyVariableName);
                }

                dbgLog.fine("valueVariableMappingTable:\n" + valueVariableMappingTable);
            } catch (IOException ex) {
                //ex.printStackTrace();
                throw ex;
            }

            safteyCounter++;
            if (safteyCounter >= 1000000) {
                break;
            }
        } //while

        ///smd.setValueLabelTable(valueLabelTable);
        assignValueLabels(valueLabelTable);

        dbgLog.fine("***** decodeRecordType3and4(): end *****");
    }

    void assignValueLabels(Map<String, Map<String, String>> valueLabelTable) {
        // Let's go through all the categorical value label mappings and 
        // assign them to the correct variables: 

        for (int i = 0; i < dataTable.getVarQuantity().intValue(); i++) {

            String varName = dataTable.getDataVariables().get(i).getName();

            Map<String, String> valueLabelPairs = valueLabelTable.get(valueVariableMappingTable.get(varName));
            if (valueLabelPairs != null && !valueLabelPairs.isEmpty()) {
                for (String value : valueLabelPairs.keySet()) {

                    VariableCategory cat = new VariableCategory();
                    cat.setValue(value);
                    cat.setLabel(valueLabelPairs.get(value));

                    /* cross-link the variable and category to each other: */
                    cat.setDataVariable(dataTable.getDataVariables().get(i));
                    dataTable.getDataVariables().get(i).getCategories().add(cat);
                }
            }
        }
    }

    void decodeRecordType6(BufferedInputStream stream) throws IOException {
        dbgLog.fine("***** decodeRecordType6(): start *****");
        try {
            if (stream == null) {
                throw new IllegalArgumentException("stream == null!");
            }
            // this section is optional; so let's first check the 4-byte header 
            // value and see what type it is. 
            //if (stream.markSupported()){ // -- ? L.A. 4.0 alpha
            stream.mark(1000);
            //}
            // 6.0 check the first 4 bytes
            byte[] headerCodeRt6 = new byte[LENGTH_RECORD_TYPE6_CODE];

            int nbytes_rt6 = stream.read(headerCodeRt6, 0, LENGTH_RECORD_TYPE6_CODE);
            // to-do check against nbytes
            //printHexDump(headerCodeRt6, "RT6 header test");
            ByteBuffer bb_header_code_rt6 = ByteBuffer.wrap(headerCodeRt6, 0, LENGTH_RECORD_TYPE6_CODE);
            if (isLittleEndian) {
                bb_header_code_rt6.order(ByteOrder.LITTLE_ENDIAN);
            }

            int intRT6test = bb_header_code_rt6.getInt();
            dbgLog.fine("RT6: header test value=" + intRT6test);
            if (intRT6test != 6) {
                //if (stream.markSupported()){
                //out.print("iteration="+safteyCounter);
                //dbgLog.fine("iteration="+safteyCounter);
                dbgLog.fine("intRT6test failed=" + intRT6test);

                stream.reset();
                return;
                //}
            }
            // 6.1 check 4-byte integer that tells how many lines follow

            byte[] length_how_many_line_bytes = new byte[LENGTH_RT6_HOW_MANY_LINES];

            int nbytes_rt6_1 = stream.read(length_how_many_line_bytes, 0, LENGTH_RT6_HOW_MANY_LINES);
            // to-do check against nbytes

            //printHexDump(length_how_many_line_bytes, "RT6 how_many_line_bytes");
            ByteBuffer bb_how_many_lines = ByteBuffer.wrap(length_how_many_line_bytes, 0,
                    LENGTH_RT6_HOW_MANY_LINES);
            if (isLittleEndian) {
                bb_how_many_lines.order(ByteOrder.LITTLE_ENDIAN);
            }

            int howManyLinesRt6 = bb_how_many_lines.getInt();
            dbgLog.fine("how Many lines follow=" + howManyLinesRt6);

            // 6.2 read 80-char-long lines 
            String[] documentRecord = new String[howManyLinesRt6];

            for (int i = 0; i < howManyLinesRt6; i++) {

                byte[] line = new byte[80];
                int nbytes_rt6_line = stream.read(line);

                documentRecord[i] = StringUtils.stripEnd(
                        new String(Arrays.copyOfRange(line, 0, LENGTH_RT6_DOCUMENT_LINE), defaultCharSet), " ");

                dbgLog.fine(i + "-th line =" + documentRecord[i] + "<-");
            }
            dbgLog.fine("documentRecord:\n" + StringUtils.join(documentRecord, "\n"));

        } catch (IOException ex) {
            //ex.printStackTrace();
            throw ex;
        }

        dbgLog.fine("decodeRecordType6(): end");
    }

    /*
     * TODO: 
     * Add an explanation note here documenting what "record type 7" is 
     * and what information it stores. This is not obvious from the code
     * below. -- L.A. 4.0 alpha
    */
    void decodeRecordType7(BufferedInputStream stream) throws IOException {
        dbgLog.fine("decodeRecordType7(): start");
        int counter = 0;
        int[] headerSection = new int[2];

        // the variables below may no longer needed; 
        // but they may be useful for debugging/logging purposes.

        /// // RecordType 7 
        /// // Subtype 3
        /// List<Integer> releaseMachineSpecificInfo = new ArrayList<Integer>();
        /// List<String> releaseMachineSpecificInfoHex = new ArrayList<String>();

        /// // Subytpe 4
        /// Map<String, Double> OBSTypeValue = new LinkedHashMap<String, Double>();
        /// Map<String, String> OBSTypeHexValue = new LinkedHashMap<String, String>();    
        //Subtype 11
        /// List<Integer> measurementLevel = new ArrayList<Integer>();
        /// List<Integer> columnWidth = new ArrayList<Integer>();
        /// List<Integer> alignment = new ArrayList<Integer>();

        while (true) {
            try {
                if (stream == null) {
                    throw new IllegalArgumentException("RT7: stream == null!");
                }
                // first check the 4-byte header value
                //if (stream.markSupported()){
                stream.mark(1000);
                //}
                // 7.0 check the first 4 bytes
                byte[] headerCodeRt7 = new byte[LENGTH_RECORD_TYPE7_CODE];

                int nbytes_rt7 = stream.read(headerCodeRt7, 0, LENGTH_RECORD_TYPE7_CODE);
                // to-do check against nbytes
                //printHexDump(headerCodeRt7, "RT7 header test");
                ByteBuffer bb_header_code_rt7 = ByteBuffer.wrap(headerCodeRt7, 0, LENGTH_RECORD_TYPE7_CODE);
                if (isLittleEndian) {
                    bb_header_code_rt7.order(ByteOrder.LITTLE_ENDIAN);
                }

                int intRT7test = bb_header_code_rt7.getInt();
                dbgLog.fine("RT7: header test value=" + intRT7test);
                if (intRT7test != 7) {
                    //if (stream.markSupported()){
                    //out.print("iteration="+safteyCounter);
                    //dbgLog.fine("iteration="+safteyCounter);
                    dbgLog.fine("intRT7test failed=" + intRT7test);
                    dbgLog.fine("counter=" + counter);
                    stream.reset();
                    return;
                    //}
                }

                // 7.1 check 4-byte integer Sub-Type Code

                byte[] length_sub_type_code = new byte[LENGTH_RT7_SUB_TYPE_CODE];

                int nbytes_rt7_1 = stream.read(length_sub_type_code, 0, LENGTH_RT7_SUB_TYPE_CODE);
                // to-do check against nbytes

                //printHexDump(length_how_many_line_bytes, "RT7 how_many_line_bytes");
                ByteBuffer bb_sub_type_code = ByteBuffer.wrap(length_sub_type_code, 0, LENGTH_RT7_SUB_TYPE_CODE);
                if (isLittleEndian) {
                    bb_sub_type_code.order(ByteOrder.LITTLE_ENDIAN);
                }

                int subTypeCode = bb_sub_type_code.getInt();
                dbgLog.fine("RT7: subTypeCode=" + subTypeCode);

                switch (subTypeCode) {
                case 3:
                    // 3: Release andMachine-Specific Integer Information

                    //parseRT7SubTypefield(stream);

                    headerSection = parseRT7SubTypefieldHeader(stream);
                    if (headerSection != null) {
                        int unitLength = headerSection[0];
                        int numberOfUnits = headerSection[1];

                        for (int i = 0; i < numberOfUnits; i++) {
                            dbgLog.finer(i + "-th fieldData");
                            byte[] work = new byte[unitLength];

                            int nb = stream.read(work);
                            dbgLog.finer("raw bytes in Hex:" + new String(Hex.encodeHex(work)));
                            ByteBuffer bb_field = ByteBuffer.wrap(work);
                            if (isLittleEndian) {
                                bb_field.order(ByteOrder.LITTLE_ENDIAN);
                            }
                            String dataInHex = new String(Hex.encodeHex(bb_field.array()));
                            /// releaseMachineSpecificInfoHex.add(dataInHex);

                            dbgLog.finer("raw bytes in Hex:" + dataInHex);
                            if (unitLength == 4) {
                                int fieldData = bb_field.getInt();
                                dbgLog.finer("fieldData(int)=" + fieldData);
                                dbgLog.finer("fieldData in Hex=0x" + Integer.toHexString(fieldData));
                                /// releaseMachineSpecificInfo.add(fieldData);
                            }

                        }

                        /// dbgLog.fine("releaseMachineSpecificInfo="+releaseMachineSpecificInfo);
                        /// dbgLog.fine("releaseMachineSpecificInfoHex="+releaseMachineSpecificInfoHex);

                    } else {
                        // throw new IOException
                    }

                    dbgLog.fine("***** end of subType 3 ***** \n");

                    break;
                case 4:
                    // Release andMachine-SpecificOBS-Type Information
                    headerSection = parseRT7SubTypefieldHeader(stream);
                    if (headerSection != null) {
                        int unitLength = headerSection[0];
                        int numberOfUnits = headerSection[1];

                        for (int i = 0; i < numberOfUnits; i++) {
                            dbgLog.finer(i + "-th fieldData:" + RecordType7SubType4Fields.get(i));
                            byte[] work = new byte[unitLength];

                            int nb = stream.read(work);

                            dbgLog.finer("raw bytes in Hex:" + new String(Hex.encodeHex(work)));
                            ByteBuffer bb_field = ByteBuffer.wrap(work);
                            dbgLog.finer("byte order=" + bb_field.order().toString());
                            if (isLittleEndian) {
                                bb_field.order(ByteOrder.LITTLE_ENDIAN);
                            }
                            ByteBuffer bb_field_dup = bb_field.duplicate();
                            OBSTypeHexValue.put(RecordType7SubType4Fields.get(i),
                                    new String(Hex.encodeHex(bb_field.array())));
                            //                            dbgLog.finer("raw bytes in Hex:"+
                            //                                OBSTypeHexValue.get(RecordType7SubType4Fields.get(i)));
                            if (unitLength == 8) {
                                double fieldData = bb_field.getDouble();
                                /// OBSTypeValue.put(RecordType7SubType4Fields.get(i), fieldData);
                                dbgLog.finer("fieldData(double)=" + fieldData);
                                OBSTypeHexValue.put(RecordType7SubType4Fields.get(i),
                                        Double.toHexString(fieldData));
                                dbgLog.fine("fieldData in Hex=" + Double.toHexString(fieldData));
                            }
                        }
                        /// dbgLog.fine("OBSTypeValue="+OBSTypeValue);
                        /// dbgLog.fine("OBSTypeHexValue="+OBSTypeHexValue);

                    } else {
                        // throw new IOException
                    }

                    dbgLog.fine("***** end of subType 4 ***** \n");
                    break;
                case 5:
                    // Variable Sets Information
                    parseRT7SubTypefield(stream);
                    break;
                case 6:
                    // Trends date information
                    parseRT7SubTypefield(stream);
                    break;
                case 7:
                    // Multiple response groups
                    parseRT7SubTypefield(stream);
                    break;
                case 8:
                    // Windows Data Entry data
                    parseRT7SubTypefield(stream);
                    break;
                case 9:
                    //
                    parseRT7SubTypefield(stream);
                    break;
                case 10:
                    // TextSmart data
                    parseRT7SubTypefield(stream);
                    break;
                case 11:
                    // Msmt level, col width, & alignment
                    //parseRT7SubTypefield(stream);

                    headerSection = parseRT7SubTypefieldHeader(stream);
                    if (headerSection != null) {
                        int unitLength = headerSection[0];
                        int numberOfUnits = headerSection[1];

                        for (int i = 0; i < numberOfUnits; i++) {
                            dbgLog.finer(i + "-th fieldData");
                            byte[] work = new byte[unitLength];

                            int nb = stream.read(work);
                            dbgLog.finer("raw bytes in Hex:" + new String(Hex.encodeHex(work)));
                            ByteBuffer bb_field = ByteBuffer.wrap(work);
                            if (isLittleEndian) {
                                bb_field.order(ByteOrder.LITTLE_ENDIAN);
                            }
                            dbgLog.finer("raw bytes in Hex:" + new String(Hex.encodeHex(bb_field.array())));

                            if (unitLength == 4) {
                                int fieldData = bb_field.getInt();
                                dbgLog.finer("fieldData(int)=" + fieldData);
                                dbgLog.finer("fieldData in Hex=0x" + Integer.toHexString(fieldData));

                                int remainder = i % 3;
                                dbgLog.finer("remainder=" + remainder);
                                if (remainder == 0) {
                                    /// measurementLevel.add(fieldData);
                                } else if (remainder == 1) {
                                    /// columnWidth.add(fieldData);
                                } else if (remainder == 2) {
                                    /// alignment.add(fieldData);
                                }
                            }

                        }

                    } else {
                        // throw new IOException
                    }
                    /// dbgLog.fine("measurementLevel="+measurementLevel);
                    /// dbgLog.fine("columnWidth="+columnWidth);
                    /// dbgLog.fine("alignment="+alignment);
                    dbgLog.fine("end of subType 11\n");

                    break;
                case 12:
                    // Windows Data Entry GUID
                    parseRT7SubTypefield(stream);
                    break;
                case 13:
                    // Extended variable names
                    // parseRT7SubTypefield(stream);
                    headerSection = parseRT7SubTypefieldHeader(stream);

                    if (headerSection != null) {
                        int unitLength = headerSection[0];
                        dbgLog.fine("RT7: unitLength=" + unitLength);
                        int numberOfUnits = headerSection[1];
                        dbgLog.fine("RT7: numberOfUnits=" + numberOfUnits);
                        byte[] work = new byte[unitLength * numberOfUnits];
                        int nbtyes13 = stream.read(work);

                        String[] variableShortLongNamePairs = new String(work, "US-ASCII").split("\t");

                        for (int i = 0; i < variableShortLongNamePairs.length; i++) {
                            dbgLog.fine("RT7: " + i + "-th pair" + variableShortLongNamePairs[i]);
                            String[] pair = variableShortLongNamePairs[i].split("=");
                            shortToLongVariableNameTable.put(pair[0], pair[1]);
                        }

                        dbgLog.fine("RT7: shortToLongVarialbeNameTable" + shortToLongVariableNameTable);
                        // We are saving the short-to-long name map; at the
                        // end of ingest, we'll go through the data variables and
                        // change the names accordingly. 

                        // smd.setShortToLongVarialbeNameTable(shortToLongVarialbeNameTable);
                    } else {
                        // throw new IOException
                    }

                    break;
                case 14:
                    // Extended strings
                    //parseRT7SubTypefield(stream);
                    headerSection = parseRT7SubTypefieldHeader(stream);

                    if (headerSection != null) {
                        int unitLength = headerSection[0];
                        dbgLog.fine("RT7.14: unitLength=" + unitLength);
                        int numberOfUnits = headerSection[1];
                        dbgLog.fine("RT7.14: numberOfUnits=" + numberOfUnits);
                        byte[] work = new byte[unitLength * numberOfUnits];
                        int nbtyes13 = stream.read(work);

                        String[] extendedVariablesSizePairs = new String(work, defaultCharSet).split("\000\t");

                        for (int i = 0; i < extendedVariablesSizePairs.length; i++) {
                            dbgLog.fine("RT7.14: " + i + "-th pair" + extendedVariablesSizePairs[i]);
                            if (extendedVariablesSizePairs[i].indexOf("=") > 0) {
                                String[] pair = extendedVariablesSizePairs[i].split("=");
                                extendedVariablesSizeTable.put(pair[0], Integer.valueOf(pair[1]));
                            }
                        }

                        dbgLog.fine("RT7.14: extendedVariablesSizeTable" + extendedVariablesSizeTable);
                    } else {
                        // throw new IOException
                    }

                    break;
                case 15:
                    // Clementine Metadata
                    parseRT7SubTypefield(stream);
                    break;
                case 16:
                    // 64 bit N of cases
                    parseRT7SubTypefield(stream);
                    break;
                case 17:
                    // File level attributes
                    parseRT7SubTypefield(stream);
                    break;
                case 18:
                    // Variable attributes
                    parseRT7SubTypefield(stream);
                    break;
                case 19:
                    // Extended multiple response groups
                    parseRT7SubTypefield(stream);
                    break;
                case 20:
                    // Character encoding, aka code page.
                    // Must be a version 16+ feature (?).
                    // Starting v.16, the default character encoding for SAV
                    // files is UTF-8; but then it is possible to specify an 
                    // alternative encoding here. 
                    // A typical use case would be people setting it to "ISO-Latin" 
                    // or "windows-1252", or a similar 8-bit encoding to store 
                    // text with standard Western European accents.
                    // -- L.A.

                    headerSection = parseRT7SubTypefieldHeader(stream);

                    if (headerSection != null) {
                        int unitLength = headerSection[0];
                        dbgLog.fine("RT7-20: unitLength=" + unitLength);
                        int numberOfUnits = headerSection[1];
                        dbgLog.fine("RT7-20: numberOfUnits=" + numberOfUnits);
                        byte[] rt7st20bytes = new byte[unitLength * numberOfUnits];
                        int nbytes20 = stream.read(rt7st20bytes);

                        String dataCharSet = new String(rt7st20bytes, "US-ASCII");

                        if (dataCharSet != null && !(dataCharSet.equals(""))) {
                            dbgLog.fine("RT7-20: data charset: " + dataCharSet);
                            defaultCharSet = dataCharSet;
                        }
                    } /*else {
                      // TODO: 
                      // decide if the exception should actually be thrown here!
                      // -- L.A. 4.0 beta
                      // throw new IOException
                      }*/

                    break;
                case 21:
                    // Value labels for long strings
                    parseRT7SubTypefield(stream);
                    break;
                case 22:
                    // Missing values for long strings
                    parseRT7SubTypefield(stream);
                    break;
                default:
                    parseRT7SubTypefield(stream);
                }

            } catch (IOException ex) {
                //ex.printStackTrace();
                throw ex;
            }

            counter++;

            if (counter > 20) {
                break;
            }
        }

        dbgLog.fine("RT7: counter=" + counter);
        dbgLog.fine("RT7: decodeRecordType7(): end");
    }

    void decodeRecordType999(BufferedInputStream stream) throws IOException {
        dbgLog.fine("decodeRecordType999(): start");
        try {
            if (stream == null) {
                throw new IllegalArgumentException("RT999: stream == null!");
            }
            // first check the 4-byte header value
            //if (stream.markSupported()){
            stream.mark(1000);
            //}
            // 999.0 check the first 4 bytes
            byte[] headerCodeRt999 = new byte[LENGTH_RECORD_TYPE999_CODE];

            //dbgLog.fine("RT999: stream position="+stream.pos);

            int nbytes_rt999 = stream.read(headerCodeRt999, 0, LENGTH_RECORD_TYPE999_CODE);
            // to-do check against nbytes
            //printHexDump(headerCodeRt999, "RT999 header test");
            ByteBuffer bb_header_code_rt999 = ByteBuffer.wrap(headerCodeRt999, 0, LENGTH_RECORD_TYPE999_CODE);
            if (isLittleEndian) {
                bb_header_code_rt999.order(ByteOrder.LITTLE_ENDIAN);
            }

            int intRT999test = bb_header_code_rt999.getInt();
            dbgLog.fine("header test value: RT999=" + intRT999test);
            if (intRT999test != 999) {
                //if (stream.markSupported()){
                dbgLog.fine("intRT999test failed=" + intRT999test);
                stream.reset();
                throw new IOException("RT999:Header value(999) was not correctly detected:" + intRT999test);
                //}
            }

            // 999.1 check 4-byte integer Filler block

            byte[] length_filler = new byte[LENGTH_RT999_FILLER];

            int nbytes_rt999_1 = stream.read(length_filler, 0, LENGTH_RT999_FILLER);
            // to-do check against nbytes

            //printHexDump(length_how_many_line_bytes, "RT999 how_many_line_bytes");
            ByteBuffer bb_filler = ByteBuffer.wrap(length_filler, 0, LENGTH_RT999_FILLER);
            if (isLittleEndian) {
                bb_filler.order(ByteOrder.LITTLE_ENDIAN);
            }

            int rt999filler = bb_filler.getInt();
            dbgLog.fine("rt999filler=" + rt999filler);

            if (rt999filler == 0) {
                dbgLog.fine("the end of the dictionary section");
            } else {
                throw new IOException("RT999: failed to detect the end mark(0): value=" + rt999filler);
            }

            // missing value processing concerning HIGHEST/LOWEST values

            Set<Map.Entry<String, InvalidData>> msvlc = invalidDataTable.entrySet();
            for (Iterator<Map.Entry<String, InvalidData>> itc = msvlc.iterator(); itc.hasNext();) {
                Map.Entry<String, InvalidData> et = itc.next();
                String variable = et.getKey();
                dbgLog.fine("variable=" + variable);
                InvalidData invalidDataInfo = et.getValue();

                if (invalidDataInfo.getInvalidRange() != null && !invalidDataInfo.getInvalidRange().isEmpty()) {
                    if (invalidDataInfo.getInvalidRange().get(0).equals(OBSTypeHexValue.get("LOWEST"))) {
                        dbgLog.fine("1st value is LOWEST");
                        invalidDataInfo.getInvalidRange().set(0, "LOWEST");
                    } else if (invalidDataInfo.getInvalidRange().get(1).equals(OBSTypeHexValue.get("HIGHEST"))) {
                        dbgLog.fine("2nd value is HIGHEST");
                        invalidDataInfo.getInvalidRange().set(1, "HIGHEST");
                    }
                }
            }
            dbgLog.fine("invalidDataTable:\n" + invalidDataTable);
            // TODO: take care of the invalid data! - add the appropriate 
            // value labels (?) 
            // should it be done here, or at the end of ingest?
            // -- L.A. 4.0 alpha
            ///smd.setInvalidDataTable(invalidDataTable);
        } catch (IOException ex) {
            //ex.printStackTrace();
            //exit(1);
            throw ex;
        }

        dbgLog.fine("decodeRecordType999(): end");
    }

    void decodeRecordTypeData(BufferedInputStream stream) throws IOException {
        dbgLog.fine("decodeRecordTypeData(): start");

        ///String fileUnfValue = null;
        ///String[] unfValues = null;

        if (stream == null) {
            throw new IllegalArgumentException("stream == null!");
        }
        if (isDataSectionCompressed) {
            decodeRecordTypeDataCompressed(stream);
        } else {
            decodeRecordTypeDataUnCompressed(stream);
        }

        /* UNF calculation was here... */

        dbgLog.fine("***** decodeRecordTypeData(): end *****");
    }

    PrintWriter createOutputWriter(BufferedInputStream stream) throws IOException {
        PrintWriter pwout = null;
        FileOutputStream fileOutTab = null;

        try {

            // create a File object to save the tab-delimited data file
            File tabDelimitedDataFile = File.createTempFile("tempTabfile.", ".tab");

            String tabDelimitedDataFileName = tabDelimitedDataFile.getAbsolutePath();

            // save the temp file name in the metadata object
            ///smd.getFileInformation().put("tabDelimitedDataFileLocation", tabDelimitedDataFileName);
            ingesteddata.setTabDelimitedFile(tabDelimitedDataFile);

            fileOutTab = new FileOutputStream(tabDelimitedDataFile);

            pwout = new PrintWriter(new OutputStreamWriter(fileOutTab, "utf8"), true);

        } catch (FileNotFoundException ex) {
            ex.printStackTrace();
        } catch (UnsupportedEncodingException ex) {
            ex.printStackTrace();
        } catch (IOException ex) {
            //ex.printStackTrace();
            throw ex;
        }

        return pwout;

    }

    void decodeRecordTypeDataCompressed(BufferedInputStream stream) throws IOException {

        dbgLog.fine("***** decodeRecordTypeDataCompressed(): start *****");

        if (stream == null) {
            throw new IllegalArgumentException("decodeRecordTypeDataCompressed: stream == null!");
        }

        PrintWriter pwout = createOutputWriter(stream);

        int varQnty = dataTable.getVarQuantity().intValue();
        int caseQnty = dataTable.getCaseQuantity().intValue();

        dbgLog.fine("varQnty: " + varQnty);

        dateFormatList = new String[varQnty];

        boolean hasStringVarContinuousBlock = obsNonVariableBlockSet.size() > 0 ? true : false;
        dbgLog.fine("hasStringVarContinuousBlock=" + hasStringVarContinuousBlock);

        int ii = 0;

        int OBS = LENGTH_SAV_OBS_BLOCK;
        int nOBS = OBSUnitsPerCase;

        dbgLog.fine("OBSUnitsPerCase=" + OBSUnitsPerCase);

        int caseIndex = 0;

        dbgLog.fine("printFormatTable:\n" + printFormatTable);
        variableFormatTypeList = new String[varQnty];

        for (int i = 0; i < varQnty; i++) {
            variableFormatTypeList[i] = SPSSConstants.FORMAT_CATEGORY_TABLE
                    .get(printFormatTable.get(variableNameList.get(i)));
            dbgLog.fine("i=" + i + "th variableFormatTypeList=" + variableFormatTypeList[i]);
            formatCategoryTable.put(variableNameList.get(i), variableFormatTypeList[i]);
        }
        dbgLog.fine("variableFormatType:\n" + Arrays.deepToString(variableFormatTypeList));
        dbgLog.fine("formatCategoryTable:\n" + formatCategoryTable);

        // TODO: 
        // Make sure the date formats are actually preserved! 
        // (this is something that was collected in the code below and passed
        // to the UNF calculator). 
        // -- L.A. 4.0 alpha
        List<String> casewiseRecordForTabFile = new ArrayList<String>();

        try {
            // this compression is applied only to non-float data, i.e. integer;
            // 8-byte float datum is kept in tact
            boolean hasReachedEOF = false;

            OBSERVATION: while (true) {

                dbgLog.fine("SAV Reader: compressed: ii=" + ii + "-th iteration");

                byte[] octate = new byte[LENGTH_SAV_OBS_BLOCK];

                int nbytes = stream.read(octate);

                // processCompressedOBSblock ()

                // (this means process a block of 8 compressed OBS
                // values -- should result in 64 bytes of data total)

                for (int i = 0; i < LENGTH_SAV_OBS_BLOCK; i++) {

                    dbgLog.finer("i=" + i + "-th iteration");
                    int octate_i = octate[i];
                    //dbgLog.fine("octate="+octate_i);
                    if (octate_i < 0) {
                        octate_i += 256;
                    }
                    int byteCode = octate_i;//octate_i & 0xF;
                    //out.println("byeCode="+byteCode);

                    // processCompressedOBS

                    switch (byteCode) {
                    case 252:
                        // end of the file
                        dbgLog.fine("SAV Reader: compressed: end of file mark [FC] was found");
                        hasReachedEOF = true;
                        break;
                    case 253:
                        // FD: uncompressed data follows after this octate
                        // long string datum or float datum
                        // read the following octate
                        byte[] uncompressedByte = new byte[LENGTH_SAV_OBS_BLOCK];
                        int ucbytes = stream.read(uncompressedByte);
                        int typeIndex = (ii * OBS + i) % nOBS;

                        if ((OBSwiseTypelList.get(typeIndex) > 0) || (OBSwiseTypelList.get(typeIndex) == -1)) {
                            // code= >0 |-1: string or its conitiguous block
                            // decode as a string object
                            String strdatum = new String(
                                    Arrays.copyOfRange(uncompressedByte, 0, LENGTH_SAV_OBS_BLOCK), defaultCharSet);
                            //out.println("str_datum="+strdatum+"<-");
                            // add this non-missing-value string datum
                            casewiseRecordForTabFile.add(strdatum);
                            //out.println("casewiseRecordForTabFile(String)="+casewiseRecordForTabFile);
                        } else if (OBSwiseTypelList.get(typeIndex) == -2) {
                            String strdatum = new String(
                                    Arrays.copyOfRange(uncompressedByte, 0, LENGTH_SAV_OBS_BLOCK - 1),
                                    defaultCharSet);
                            casewiseRecordForTabFile.add(strdatum);
                            //out.println("casewiseRecordForTabFile(String)="+casewiseRecordForTabFile);
                        } else if (OBSwiseTypelList.get(typeIndex) == 0) {
                            // code= 0: numeric

                            ByteBuffer bb_double = ByteBuffer.wrap(uncompressedByte, 0, LENGTH_SAV_OBS_BLOCK);
                            if (isLittleEndian) {
                                bb_double.order(ByteOrder.LITTLE_ENDIAN);
                            }

                            Double ddatum = bb_double.getDouble();
                            // out.println("ddatum="+ddatum);
                            // add this non-missing-value numeric datum
                            casewiseRecordForTabFile.add(doubleNumberFormatter.format(ddatum));
                            dbgLog.fine("SAV Reader: compressed: added value to dataLine: " + ddatum);

                        } else {
                            dbgLog.fine("SAV Reader: out-of-range exception");
                            throw new IOException("out-of-range value was found");
                        }

                        /*
                        // EOF-check after reading this octate
                        if (stream.available() == 0){
                        hasReachedEOF = true;
                        dbgLog.fine(
                        "SAV Reader: *** After reading an uncompressed octate," +
                        " reached the end of the file at "+ii
                        +"th iteration and i="+i+"th octate position [0-start] *****");
                        }
                         */

                        break;
                    case 254:
                        // FE: used as the missing value for string variables
                        // an empty case in a string variable also takes this value
                        // string variable does not accept space-only data
                        // cf: uncompressed case
                        // 20 20 20 20 20 20 20 20
                        // add the string missing value
                        // out.println("254: String missing data");

                        casewiseRecordForTabFile.add(" "); // add "." here?

                        // Note that technically this byte flag (254/xFE) means
                        // that *eight* white space characters should be
                        // written to the output stream. This caused me
                        // a great amount of confusion, because it appeared
                        // to me that there was a mismatch between the number
                        // of bytes advertised in the variable metadata and
                        // the number of bytes actually found in the data
                        // section of a compressed SAV file; this is because
                        // these 8 bytes "come out of nowhere"; they are not
                        // written in the data section, but this flag specifies
                        // that they should be added to the output.
                        // Also, as I pointed out above, we are only writing
                        // out one whitespace character, not 8 as instructed.
                        // This appears to be legit; these blocks of 8 spaces
                        // seem to be only used for padding, and all such
                        // multiple padding spaces are stripped anyway during
                        // the post-processing.

                        break;
                    case 255:
                        // FF: system missing value for numeric variables
                        // cf: uncompressed case (sysmis)
                        // FF FF FF FF FF FF eF FF(little endian)
                        // add the numeric missing value
                        dbgLog.fine("SAV Reader: compressed: Missing Value, numeric");
                        casewiseRecordForTabFile.add(MissingValueForTextDataFileNumeric);

                        break;
                    case 0:
                        // 00: do nothing
                        dbgLog.fine("SAV Reader: compressed: doing nothing (zero); ");

                        break;
                    default:
                        //out.println("byte code(default)="+ byteCode);
                        if ((byteCode > 0) && (byteCode < 252)) {
                            // datum is compressed
                            //Integer unCompressed = Integer.valueOf(byteCode -100);
                            // add this uncompressed numeric datum
                            Double unCompressed = Double.valueOf(byteCode - 100);
                            dbgLog.fine("SAV Reader: compressed: default case: " + unCompressed);

                            casewiseRecordForTabFile.add(doubleNumberFormatter.format(unCompressed));
                            // out.println("uncompressed="+unCompressed);
                            // out.println("dataline="+casewiseRecordForTabFile);
                        }
                    }// end of switch

                    // out.println("end of switch");

                    // The-end-of-a-case(row)-processing

                    // this line that follows, and the code around it
                    // is really confusing:
                    int varCounter = (ii * OBS + i + 1) % nOBS;
                    // while both OBS and LENGTH_SAV_OBS_BLOCK = 8
                    // (OBS was initialized as OBS=LENGTH_SAV_OBS_BLOCK),
                    // the 2 values mean different things:
                    // LENGTH_SAV_OBS_BLOCK is the number of bytes in one OBS;
                    // and OBS is the number of OBS blocks that we process
                    // at a time. I.e., we process 8 chunks of 8 bytes at a time.
                    // This is how data is organized inside an SAV file:
                    // 8 bytes of compression flags, followd by 8x8 or fewer
                    // (depending on the flags) bytes of compressed data.
                    // I should rename this OBS variable something more
                    // meaningful.
                    //
                    // Also, the "varCounter" variable name is entirely
                    // misleading -- it counts not variables, but OBS blocks.

                    dbgLog.fine("SAV Reader: compressed: OBS counter=" + varCounter + "(ii=" + ii + ")");

                    if ((ii * OBS + i + 1) % nOBS == 0) {

                        //out.println("casewiseRecordForTabFile(before)="+casewiseRecordForTabFile);

                        // out.println("all variables in a case are parsed == nOBS");
                        // out.println("hasStringVarContinuousBlock="+hasStringVarContinuousBlock);

                        // check whether a string-variable's continuous block exits
                        // if so, they must be joined

                        if (hasStringVarContinuousBlock) {

                            // string-variable's continuous-block-concatenating-processing

                            //out.println("concatenating process starts");
                            //out.println("casewiseRecordForTabFile(before)="+casewiseRecordForTabFile);
                            //out.println("casewiseRecordForTabFile(before:size)="+casewiseRecordForTabFile.size());

                            StringBuilder sb = new StringBuilder("");
                            int firstPosition = 0;

                            Set<Integer> removeJset = new HashSet<Integer>();
                            for (int j = 0; j < nOBS; j++) {
                                dbgLog.fine("RTD: j=" + j + "-th type =" + OBSwiseTypelList.get(j));
                                if ((OBSwiseTypelList.get(j) == -1) || (OBSwiseTypelList.get(j) == -2)) {
                                    // Continued String variable found at j-th
                                    // position. look back the j-1
                                    firstPosition = j - 1;
                                    int lastJ = j;
                                    String concatenated = null;

                                    removeJset.add(j);
                                    sb.append(casewiseRecordForTabFile.get(j - 1));
                                    sb.append(casewiseRecordForTabFile.get(j));

                                    for (int jc = 1;; jc++) {
                                        if ((j + jc == nOBS) || ((OBSwiseTypelList.get(j + jc) != -1)
                                                && (OBSwiseTypelList.get(j + jc) != -2))) {

                                            // j is the end unit of this string variable
                                            concatenated = sb.toString();
                                            sb.setLength(0);
                                            lastJ = j + jc;
                                            break;
                                        } else {
                                            sb.append(casewiseRecordForTabFile.get(j + jc));
                                            removeJset.add(j + jc);
                                        }
                                    }
                                    casewiseRecordForTabFile.set(j - 1, concatenated);

                                    //out.println(j-1+"th concatenated="+concatenated);
                                    j = lastJ - 1;

                                } // end-of-if: continuous-OBS only

                            } // end of loop-j

                            //out.println("removeJset="+removeJset);

                            // a new list that stores a new case with concatanated string data
                            List<String> newDataLine = new ArrayList<String>();

                            for (int jl = 0; jl < casewiseRecordForTabFile.size(); jl++) {
                                //out.println("jl="+jl+"-th datum =["+casewiseRecordForTabFile.get(jl)+"]");

                                if (!removeJset.contains(jl)) {

                                    //                                if (casewiseRecordForTabFile.get(jl).equals(MissingValueForTextDataFileString)){
                                    //                                    out.println("NA-S jl= "+jl+"=["+casewiseRecordForTabFile.get(jl)+"]");
                                    //                                } else if (casewiseRecordForTabFile.get(jl).equals(MissingValueForTextDataFileNumeric)){
                                    //                                    out.println("NA-N jl= "+jl+"=["+casewiseRecordForTabFile.get(jl)+"]");
                                    //                                } else if (casewiseRecordForTabFile.get(jl)==null){
                                    //                                    out.println("null case jl="+jl+"=["+casewiseRecordForTabFile.get(jl)+"]");
                                    //                                } else if (casewiseRecordForTabFile.get(jl).equals("NaN")){
                                    //                                    out.println("NaN jl= "+jl+"=["+casewiseRecordForTabFile.get(jl)+"]");
                                    //                                } else if (casewiseRecordForTabFile.get(jl).equals("")){
                                    //                                    out.println("blank jl= "+jl+"=["+casewiseRecordForTabFile.get(jl)+"]");
                                    //                                } else if (casewiseRecordForTabFile.get(jl).equals(" ")){
                                    //                                    out.println("space jl= "+jl+"=["+casewiseRecordForTabFile.get(jl)+"]");
                                    //                                }

                                    newDataLine.add(casewiseRecordForTabFile.get(jl));
                                } else {
                                    //                                out.println("Excluded: jl="+jl+"-th datum=["+casewiseRecordForTabFile.get(jl)+"]");
                                }
                            } // end of loop-jl

                            //out.println("new casewiseRecordForTabFile="+newDataLine);
                            //out.println("new casewiseRecordForTabFile(size)="+newDataLine.size());

                            casewiseRecordForTabFile = newDataLine;

                        } // end-if: stringContinuousVar-exist case

                        // caseIndex starts from 1 not 0
                        caseIndex = (ii * OBS + i + 1) / nOBS;

                        for (int k = 0; k < casewiseRecordForTabFile.size(); k++) {

                            dbgLog.fine("k=" + k + "-th variableTypelList=" + variableTypelList.get(k));

                            if (variableTypelList.get(k) > 0) {

                                // Strip the String variables off the
                                // whitespace padding:

                                // [ snipped ]

                                // I've removed the block of code above where
                                // String values were substring()-ed to the
                                // length specified in the variable metadata;
                                // Doing that was not enough, since a string
                                // can still be space-padded inside its
                                // advertised capacity. (note that extended
                                // variables can have many kylobytes of such
                                // padding in them!) Plus it was completely
                                // redundant, since we are stripping all the
                                // trailing white spaces with
                                // StringUtils.stripEnd() below:

                                String paddRemoved = StringUtils
                                        .stripEnd(casewiseRecordForTabFile.get(k).toString(), null);
                                // TODO: clean this up.  For now, just make sure that strings contain at least one blank space.
                                if (paddRemoved.equals("")) {
                                    paddRemoved = " ";
                                }
                                //casewiseRecordForTabFile.set(k, "\"" + paddRemoved.replaceAll("\"", Matcher.quoteReplacement("\\\"")) + "\"");
                                casewiseRecordForTabFile.set(k, escapeCharacterString(paddRemoved));

                                // end of String var case

                            } // end of variable-type check

                            if (casewiseRecordForTabFile.get(k) != null && !casewiseRecordForTabFile.get(k)
                                    .equals(MissingValueForTextDataFileNumeric)) {

                                String variableFormatType = variableFormatTypeList[k];
                                dbgLog.finer("k=" + k + "th printFormatTable format="
                                        + printFormatTable.get(variableNameList.get(k)));

                                int formatDecimalPointPosition = formatDecimalPointPositionList.get(k);

                                if (variableFormatType.equals("date")) {
                                    dbgLog.finer("date case");

                                    long dateDatum = Long.parseLong(casewiseRecordForTabFile.get(k).toString())
                                            * 1000L - SPSS_DATE_OFFSET;

                                    String newDatum = sdf_ymd.format(new Date(dateDatum));
                                    dbgLog.finer("k=" + k + ":" + newDatum);
                                    /* saving date format */
                                    dbgLog.finer("saving dateFormat[k] = " + sdf_ymd.toPattern());
                                    casewiseRecordForTabFile.set(k, newDatum);
                                    dateFormatList[k] = sdf_ymd.toPattern();
                                    //formatCategoryTable.put(variableNameList.get(k), "date");
                                } else if (variableFormatType.equals("time")) {
                                    dbgLog.finer("time case:DTIME or DATETIME or TIME");
                                    //formatCategoryTable.put(variableNameList.get(k), "time");

                                    if (printFormatTable.get(variableNameList.get(k)).equals("DTIME")) {
                                        // We're not even going to try to handle "DTIME"
                                        // values as time/dates in dataverse; this is a weird
                                        // format that nobody uses outside of SPSS.
                                        // (but we do need to remember to treat the resulting values 
                                        // as character strings, not numerics!)

                                        if (casewiseRecordForTabFile.get(k).toString().indexOf(".") < 0) {
                                            long dateDatum = Long
                                                    .parseLong(casewiseRecordForTabFile.get(k).toString()) * 1000L
                                                    - SPSS_DATE_BIAS;
                                            String newDatum = sdf_dhms.format(new Date(dateDatum));
                                            dbgLog.finer("k=" + k + ":" + newDatum);
                                            casewiseRecordForTabFile.set(k, newDatum);
                                        } else {
                                            // decimal point included
                                            String[] timeData = casewiseRecordForTabFile.get(k).toString()
                                                    .split("\\.");

                                            dbgLog.finer(StringUtils.join(timeData, "|"));
                                            long dateDatum = Long.parseLong(timeData[0]) * 1000L - SPSS_DATE_BIAS;
                                            StringBuilder sb_time = new StringBuilder(
                                                    sdf_dhms.format(new Date(dateDatum)));
                                            dbgLog.finer(sb_time.toString());

                                            if (formatDecimalPointPosition > 0) {
                                                sb_time.append(
                                                        "." + timeData[1].substring(0, formatDecimalPointPosition));
                                            }

                                            dbgLog.finer("k=" + k + ":" + sb_time.toString());
                                            casewiseRecordForTabFile.set(k, sb_time.toString());
                                        }
                                    } else if (printFormatTable.get(variableNameList.get(k)).equals("DATETIME")) {
                                        // TODO: 
                                        // (for both datetime and "dateless" time)
                                        // keep the longest of the matching formats - i.e., if there are *some*
                                        // values in the vector that have thousands of a second, that should be 
                                        // part of the saved format!
                                        //  -- L.A. Aug. 12 2014 
                                        if (casewiseRecordForTabFile.get(k).toString().indexOf(".") < 0) {
                                            long dateDatum = Long
                                                    .parseLong(casewiseRecordForTabFile.get(k).toString()) * 1000L
                                                    - SPSS_DATE_OFFSET;
                                            String newDatum = sdf_ymdhms.format(new Date(dateDatum));
                                            dbgLog.finer("k=" + k + ":" + newDatum);
                                            casewiseRecordForTabFile.set(k, newDatum);
                                            dateFormatList[k] = sdf_ymdhms.toPattern();
                                        } else {
                                            // decimal point included
                                            String[] timeData = casewiseRecordForTabFile.get(k).toString()
                                                    .split("\\.");

                                            //dbgLog.finer(StringUtils.join(timeData, "|"));
                                            long dateDatum = Long.parseLong(timeData[0]) * 1000L - SPSS_DATE_OFFSET;
                                            StringBuilder sb_time = new StringBuilder(
                                                    sdf_ymdhms.format(new Date(dateDatum)));
                                            //dbgLog.finer(sb_time.toString());

                                            if (formatDecimalPointPosition > 0) {
                                                sb_time.append(
                                                        "." + timeData[1].substring(0, formatDecimalPointPosition));
                                            }
                                            dbgLog.finer("k=" + k + ":" + sb_time.toString());
                                            casewiseRecordForTabFile.set(k, sb_time.toString());
                                            dateFormatList[k] = sdf_ymdhms.toPattern()
                                                    + (formatDecimalPointPosition > 0 ? ".S" : "");
                                        }
                                    } else if (printFormatTable.get(variableNameList.get(k)).equals("TIME")) {
                                        // TODO: 
                                        // double-check that we are handling "dateless" time correctly... -- L.A. Aug. 2014
                                        if (casewiseRecordForTabFile.get(k).toString().indexOf(".") < 0) {
                                            long dateDatum = Long
                                                    .parseLong(casewiseRecordForTabFile.get(k).toString()) * 1000L;
                                            String newDatum = sdf_hms.format(new Date(dateDatum));
                                            dbgLog.finer("k=" + k + ":" + newDatum);
                                            casewiseRecordForTabFile.set(k, newDatum);
                                            if (dateFormatList[k] == null) {
                                                dateFormatList[k] = sdf_hms.toPattern();
                                            }
                                        } else {
                                            // decimal point included
                                            String[] timeData = casewiseRecordForTabFile.get(k).toString()
                                                    .split("\\.");

                                            //dbgLog.finer(StringUtils.join(timeData, "|"));
                                            long dateDatum = Long.parseLong(timeData[0]) * 1000L;
                                            StringBuilder sb_time = new StringBuilder(
                                                    sdf_hms.format(new Date(dateDatum)));
                                            //dbgLog.finer(sb_time.toString());

                                            if (formatDecimalPointPosition > 0) {
                                                sb_time.append(
                                                        "." + timeData[1].substring(0, formatDecimalPointPosition));
                                            }
                                            dbgLog.finer("k=" + k + ":" + sb_time.toString());
                                            casewiseRecordForTabFile.set(k, sb_time.toString());

                                            String format_hmsS = sdf_hms.toPattern()
                                                    + (formatDecimalPointPosition > 0 ? ".S" : "");
                                            if (dateFormatList[k] == null
                                                    || (format_hmsS.length() > dateFormatList[k].length())) {
                                                dateFormatList[k] = format_hmsS;
                                            }
                                        }
                                    }

                                } else if (variableFormatType.equals("other")) {
                                    dbgLog.finer("other non-date/time case:=" + i);

                                    if (printFormatTable.get(variableNameList.get(k)).equals("WKDAY")) {
                                        // day of week
                                        dbgLog.finer("data k=" + k + ":" + casewiseRecordForTabFile.get(k));
                                        dbgLog.finer("data k=" + k + ":" + SPSSConstants.WEEKDAY_LIST.get(
                                                Integer.valueOf(casewiseRecordForTabFile.get(k).toString()) - 1));
                                        String newDatum = SPSSConstants.WEEKDAY_LIST.get(
                                                Integer.valueOf(casewiseRecordForTabFile.get(k).toString()) - 1);
                                        casewiseRecordForTabFile.set(k, newDatum);
                                        dbgLog.finer("wkday:k=" + k + ":" + casewiseRecordForTabFile.get(k));
                                    } else if (printFormatTable.get(variableNameList.get(k)).equals("MONTH")) {
                                        // month
                                        dbgLog.finer("data k=" + k + ":" + casewiseRecordForTabFile.get(k));
                                        dbgLog.finer("data k=" + k + ":" + SPSSConstants.MONTH_LIST.get(
                                                Integer.valueOf(casewiseRecordForTabFile.get(k).toString()) - 1));
                                        String newDatum = SPSSConstants.MONTH_LIST.get(
                                                Integer.valueOf(casewiseRecordForTabFile.get(k).toString()) - 1);
                                        casewiseRecordForTabFile.set(k, newDatum);
                                        dbgLog.finer("month:k=" + k + ":" + casewiseRecordForTabFile.get(k));
                                    }
                                }

                            } // end: date-time-datum check

                        } // end: loop-k(2nd: variable-wise-check)

                        // write to tab file
                        if (casewiseRecordForTabFile.size() > 0) {
                            pwout.println(StringUtils.join(casewiseRecordForTabFile, "\t"));
                        }

                        // numeric contents-check
                        for (int l = 0; l < casewiseRecordForTabFile.size(); l++) {
                            if (variableFormatTypeList[l].equals("date") || variableFormatTypeList[l].equals("time")
                                    || printFormatTable.get(variableNameList.get(l)).equals("WKDAY")
                                    || printFormatTable.get(variableNameList.get(l)).equals("MONTH")) {
                                // TODO: 
                                // figure out if any special handling is still needed here in 4.0. 
                                // -- L.A. - Aug. 2014

                            } else {
                                if (variableTypelList.get(l) <= 0) {
                                    if (casewiseRecordForTabFile.get(l).toString().indexOf(".") >= 0) {
                                        decimalVariableSet.add(l);
                                    }
                                }
                            }
                        }

                        // reset the case-wise working objects
                        casewiseRecordForTabFile.clear();

                        if (caseQnty > 0) {
                            if (caseIndex == caseQnty) {
                                hasReachedEOF = true;
                            }
                        }

                        if (hasReachedEOF) {
                            break;
                        }

                    } // if(The-end-of-a-case(row)-processing)

                } // loop-i (OBS unit)

                if ((hasReachedEOF) || (stream.available() == 0)) {
                    // reached the end of this file
                    // do exit-processing

                    dbgLog.fine("***** reached the end of the file at " + ii + "th iteration *****");

                    break OBSERVATION;
                }

                ii++;

            } // while loop

            pwout.close();
        } catch (IOException ex) {
            throw ex;
        }

        dbgLog.fine("<<<<<<");
        dbgLog.fine("formatCategoryTable = " + formatCategoryTable);
        dbgLog.fine(">>>>>>");

        dbgLog.fine("decimalVariableSet=" + decimalVariableSet);

        dbgLog.fine("decodeRecordTypeDataCompressed(): end");
    }

    void decodeRecordTypeDataUnCompressed(BufferedInputStream stream) throws IOException {
        dbgLog.fine("***** decodeRecordTypeDataUnCompressed(): start *****");

        if (stream == null) {
            throw new IllegalArgumentException("decodeRecordTypeDataUnCompressed: stream == null!");
        }

        int varQnty = dataTable.getVarQuantity().intValue();

        // 
        // set-up tab file

        PrintWriter pwout = createOutputWriter(stream);

        boolean hasStringVarContinuousBlock = obsNonVariableBlockSet.size() > 0 ? true : false;
        dbgLog.fine("hasStringVarContinuousBlock=" + hasStringVarContinuousBlock);

        int ii = 0;

        int OBS = LENGTH_SAV_OBS_BLOCK;
        int nOBS = OBSUnitsPerCase;

        dbgLog.fine("OBSUnitsPerCase=" + OBSUnitsPerCase);

        int caseIndex = 0;

        dbgLog.fine("printFormatTable:\n" + printFormatTable);

        variableFormatTypeList = new String[varQnty];
        dateFormatList = new String[varQnty];

        for (int i = 0; i < varQnty; i++) {
            variableFormatTypeList[i] = SPSSConstants.FORMAT_CATEGORY_TABLE
                    .get(printFormatTable.get(variableNameList.get(i)));
            dbgLog.fine("i=" + i + "th variableFormatTypeList=" + variableFormatTypeList[i]);
            formatCategoryTable.put(variableNameList.get(i), variableFormatTypeList[i]);
        }
        dbgLog.fine("variableFormatType:\n" + Arrays.deepToString(variableFormatTypeList));
        dbgLog.fine("formatCategoryTable:\n" + formatCategoryTable);

        int numberOfDecimalVariables = 0;

        // TODO: 
        // Make sure the date formats are actually preserved! 
        // (this is something that was collected in the code below and passed
        // to the UNF calculator). 
        // -- L.A. 4.0 alpha

        List<String> casewiseRecordForTabFile = new ArrayList<String>();

        // missing values are written to the tab-delimited file by
        // using the default or user-specified missing-value  strings;
        // however, to calculate UNF/summary statistics,
        // classes for these calculations require their specific 
        // missing values that differ from the above missing-value
        // strings; therefore, after row data for the tab-delimited 
        // file are written, missing values in a row are changed to
        // UNF/summary-statistics-OK ones.

        // data-storage object for sumStat
        ///dataTable2 = new Object[varQnty][caseQnty];
        // storage of date formats to pass to UNF   
        ///dateFormats = new String[varQnty][caseQnty];

        try {
            for (int i = 0;; i++) { // case-wise loop

                byte[] buffer = new byte[OBS * nOBS];

                int nbytesuc = stream.read(buffer);

                StringBuilder sb_stringStorage = new StringBuilder("");

                for (int k = 0; k < nOBS; k++) {
                    int offset = OBS * k;

                    // uncompressed case
                    // numeric missing value == sysmis
                    // FF FF FF FF FF FF eF FF(little endian)
                    // string missing value
                    // 20 20 20 20 20 20 20 20
                    // cf: compressed case 
                    // numeric type:sysmis == 0xFF
                    // string type: missing value == 0xFE
                    // 

                    boolean isNumeric = OBSwiseTypelList.get(k) == 0 ? true : false;

                    if (isNumeric) {
                        dbgLog.finer(k + "-th variable is numeric");
                        // interprete as double
                        ByteBuffer bb_double = ByteBuffer.wrap(buffer, offset, LENGTH_SAV_OBS_BLOCK);
                        if (isLittleEndian) {
                            bb_double.order(ByteOrder.LITTLE_ENDIAN);
                        }
                        //char[] hexpattern =
                        String dphex = new String(Hex.encodeHex(
                                Arrays.copyOfRange(bb_double.array(), offset, offset + LENGTH_SAV_OBS_BLOCK)));
                        dbgLog.finer("dphex=" + dphex);

                        if ((dphex.equals("ffffffffffffefff")) || (dphex.equals("ffefffffffffffff"))) {
                            //casewiseRecordForTabFile.add(systemMissingValue);
                            // add the numeric missing value
                            dbgLog.fine("SAV Reader: adding: Missing Value (numeric)");
                            casewiseRecordForTabFile.add(MissingValueForTextDataFileNumeric);
                        } else {
                            Double ddatum = bb_double.getDouble();
                            dbgLog.fine("SAV Reader: adding: ddatum=" + ddatum);

                            // add this non-missing-value numeric datum
                            casewiseRecordForTabFile.add(doubleNumberFormatter.format(ddatum));
                        }

                    } else {
                        dbgLog.finer(k + "-th variable is string");
                        // string case
                        // strip space-padding
                        // do not trim: string might have spaces within it
                        // the missing value (hex) for a string variable is:
                        // "20 20 20 20 20 20 20 20"

                        String strdatum = new String(
                                Arrays.copyOfRange(buffer, offset, (offset + LENGTH_SAV_OBS_BLOCK)),
                                defaultCharSet);
                        dbgLog.finer("str_datum=" + strdatum);
                        // add this non-missing-value string datum 
                        casewiseRecordForTabFile.add(strdatum);

                    } // if isNumeric

                } // k-loop

                // String-variable's continuous block exits:
                if (hasStringVarContinuousBlock) {
                    // continuous blocks: string case
                    // concatenating process
                    //dbgLog.fine("concatenating process starts");

                    //dbgLog.fine("casewiseRecordForTabFile(before)="+casewiseRecordForTabFile);
                    //dbgLog.fine("casewiseRecordForTabFile(before:size)="+casewiseRecordForTabFile.size());

                    StringBuilder sb = new StringBuilder("");
                    int firstPosition = 0;

                    Set<Integer> removeJset = new HashSet<Integer>();
                    for (int j = 0; j < nOBS; j++) {
                        dbgLog.finer("j=" + j + "-th type =" + OBSwiseTypelList.get(j));
                        if (OBSwiseTypelList.get(j) == -1) {
                            // String continued fount at j-th 
                            // look back the j-1 
                            firstPosition = j - 1;
                            int lastJ = j;
                            String concatanated = null;

                            removeJset.add(j);
                            sb.append(casewiseRecordForTabFile.get(j - 1));
                            sb.append(casewiseRecordForTabFile.get(j));
                            for (int jc = 1;; jc++) {
                                if (OBSwiseTypelList.get(j + jc) != -1) {
                                    // j is the end unit of this string variable
                                    concatanated = sb.toString();
                                    sb.setLength(0);
                                    lastJ = j + jc;
                                    break;
                                } else {
                                    sb.append(casewiseRecordForTabFile.get(j + jc));
                                    removeJset.add(j + jc);
                                }
                            }
                            casewiseRecordForTabFile.set(j - 1, concatanated);

                            //out.println(j-1+"th concatanated="+concatanated);
                            j = lastJ - 1;

                        } // end-of-if: continuous-OBS only
                    } // end of loop-j

                    List<String> newDataLine = new ArrayList<String>();

                    for (int jl = 0; jl < casewiseRecordForTabFile.size(); jl++) {
                        //out.println("jl="+jl+"-th datum =["+casewiseRecordForTabFile.get(jl)+"]");

                        if (!removeJset.contains(jl)) {
                            newDataLine.add(casewiseRecordForTabFile.get(jl));
                        }
                    }

                    dbgLog.fine("new casewiseRecordForTabFile=" + newDataLine);
                    dbgLog.fine("new casewiseRecordForTabFile(size)=" + newDataLine.size());

                    casewiseRecordForTabFile = newDataLine;

                } // end-if: stringContinuousVar-exist case

                caseIndex++;
                dbgLog.finer("caseIndex=" + caseIndex);
                for (int k = 0; k < casewiseRecordForTabFile.size(); k++) {

                    if (variableTypelList.get(k) > 0) {

                        // See my comments for this padding removal logic
                        // in the "compressed" method -- L.A.

                        String paddRemoved = StringUtils.stripEnd(casewiseRecordForTabFile.get(k).toString(), null);
                        // TODO: clean this up.  For now, just make sure that strings contain at least one blank space.
                        if (paddRemoved.equals("")) {
                            paddRemoved = " ";
                        }

                        //casewiseRecordForTabFile.set(k, "\"" + paddRemoved.replaceAll("\"", Matcher.quoteReplacement("\\\"")) + "\"");
                        casewiseRecordForTabFile.set(k, escapeCharacterString(paddRemoved));

                        // end of String var case

                    } // end of variable-type check

                    if (casewiseRecordForTabFile.get(k) != null
                            && !casewiseRecordForTabFile.get(k).equals(MissingValueForTextDataFileNumeric)) {

                        // to do date conversion
                        String variableFormatType = variableFormatTypeList[k];
                        dbgLog.finer("k=" + k + "th variable format=" + variableFormatType);

                        int formatDecimalPointPosition = formatDecimalPointPositionList.get(k);

                        if (variableFormatType.equals("date")) {
                            dbgLog.finer("date case");

                            long dateDatum = Long.parseLong(casewiseRecordForTabFile.get(k).toString()) * 1000L
                                    - SPSS_DATE_OFFSET;

                            String newDatum = sdf_ymd.format(new Date(dateDatum));
                            dbgLog.finer("k=" + k + ":" + newDatum);

                            casewiseRecordForTabFile.set(k, newDatum);
                            dateFormatList[k] = sdf_ymd.toPattern();
                        } else if (variableFormatType.equals("time")) {
                            dbgLog.finer("time case:DTIME or DATETIME or TIME");
                            //formatCategoryTable.put(variableNameList.get(k), "time");
                            // not treating DTIME as date/time; see comment elsewhere in 
                            // the code; 
                            // (but we do need to remember to treat the resulting values 
                            // as character strings, not numerics!)

                            if (printFormatTable.get(variableNameList.get(k)).equals("DTIME")) {

                                if (casewiseRecordForTabFile.get(k).toString().indexOf(".") < 0) {
                                    long dateDatum = Long.parseLong(casewiseRecordForTabFile.get(k).toString())
                                            * 1000L - SPSS_DATE_BIAS;
                                    String newDatum = sdf_dhms.format(new Date(dateDatum));
                                    // Note: DTIME is not a complete date, so we don't save a date format with it
                                    dbgLog.finer("k=" + k + ":" + newDatum);
                                    casewiseRecordForTabFile.set(k, newDatum);
                                } else {
                                    // decimal point included
                                    String[] timeData = casewiseRecordForTabFile.get(k).toString().split("\\.");

                                    dbgLog.finer(StringUtils.join(timeData, "|"));
                                    long dateDatum = Long.parseLong(timeData[0]) * 1000L - SPSS_DATE_BIAS;
                                    StringBuilder sb_time = new StringBuilder(sdf_dhms.format(new Date(dateDatum)));

                                    if (formatDecimalPointPosition > 0) {
                                        sb_time.append("." + timeData[1].substring(0, formatDecimalPointPosition));
                                    }

                                    dbgLog.finer("k=" + k + ":" + sb_time.toString());
                                    casewiseRecordForTabFile.set(k, sb_time.toString());
                                }
                            } else if (printFormatTable.get(variableNameList.get(k)).equals("DATETIME")) {
                                // TODO: 
                                // (for both datetime and "dateless" time)
                                // keep the longest of the matching formats - i.e., if there are *some*
                                // values in the vector that have thousands of a second, that should be 
                                // part of the saved format!
                                //  -- L.A. Aug. 12 2014 

                                if (casewiseRecordForTabFile.get(k).toString().indexOf(".") < 0) {
                                    long dateDatum = Long.parseLong(casewiseRecordForTabFile.get(k).toString())
                                            * 1000L - SPSS_DATE_OFFSET;
                                    String newDatum = sdf_ymdhms.format(new Date(dateDatum));
                                    dbgLog.finer("k=" + k + ":" + newDatum);
                                    casewiseRecordForTabFile.set(k, newDatum);
                                    dateFormatList[k] = sdf_ymdhms.toPattern();
                                } else {
                                    // decimal point included
                                    String[] timeData = casewiseRecordForTabFile.get(k).toString().split("\\.");

                                    //dbgLog.finer(StringUtils.join(timeData, "|"));
                                    long dateDatum = Long.parseLong(timeData[0]) * 1000L - SPSS_DATE_OFFSET;
                                    StringBuilder sb_time = new StringBuilder(
                                            sdf_ymdhms.format(new Date(dateDatum)));
                                    //dbgLog.finer(sb_time.toString());

                                    if (formatDecimalPointPosition > 0) {
                                        sb_time.append("." + timeData[1].substring(0, formatDecimalPointPosition));
                                    }
                                    dbgLog.finer("k=" + k + ":" + sb_time.toString());
                                    casewiseRecordForTabFile.set(k, sb_time.toString());
                                    // datetime with milliseconds:
                                    dateFormatList[k] = sdf_ymdhms.toPattern()
                                            + (formatDecimalPointPosition > 0 ? ".S" : "");
                                }
                            } else if (printFormatTable.get(variableNameList.get(k)).equals("TIME")) {
                                if (casewiseRecordForTabFile.get(k).toString().indexOf(".") < 0) {
                                    long dateDatum = Long.parseLong(casewiseRecordForTabFile.get(k).toString())
                                            * 1000L;
                                    String newDatum = sdf_hms.format(new Date(dateDatum));
                                    dbgLog.finer("k=" + k + ":" + newDatum);
                                    casewiseRecordForTabFile.set(k, newDatum);
                                    if (dateFormatList[k] == null) {
                                        dateFormatList[k] = sdf_hms.toPattern();
                                    }
                                } else {
                                    // decimal point included
                                    String[] timeData = casewiseRecordForTabFile.get(k).toString().split("\\.");

                                    //dbgLog.finer(StringUtils.join(timeData, "|"));
                                    long dateDatum = Long.parseLong(timeData[0]) * 1000L;
                                    StringBuilder sb_time = new StringBuilder(sdf_hms.format(new Date(dateDatum)));
                                    //dbgLog.finer(sb_time.toString());

                                    if (formatDecimalPointPosition > 0) {
                                        sb_time.append("." + timeData[1].substring(0, formatDecimalPointPosition));
                                    }
                                    dbgLog.finer("k=" + k + ":" + sb_time.toString());
                                    casewiseRecordForTabFile.set(k, sb_time.toString());
                                    // time, possibly with milliseconds:
                                    String format_hmsS = sdf_hms.toPattern()
                                            + (formatDecimalPointPosition > 0 ? ".S" : "");
                                    if (dateFormatList[k] == null
                                            || (format_hmsS.length() > dateFormatList[k].length())) {
                                        dateFormatList[k] = format_hmsS;
                                    }
                                }
                            }
                        } else if (variableFormatType.equals("other")) {
                            dbgLog.finer("other non-date/time case");

                            if (printFormatTable.get(variableNameList.get(k)).equals("WKDAY")) {
                                // day of week
                                dbgLog.finer("data k=" + k + ":" + casewiseRecordForTabFile.get(k));
                                dbgLog.finer("data k=" + k + ":" + SPSSConstants.WEEKDAY_LIST
                                        .get(Integer.valueOf(casewiseRecordForTabFile.get(k).toString()) - 1));
                                String newDatum = SPSSConstants.WEEKDAY_LIST
                                        .get(Integer.valueOf(casewiseRecordForTabFile.get(k).toString()) - 1);
                                casewiseRecordForTabFile.set(k, newDatum);
                                dbgLog.finer("wkday:k=" + k + ":" + casewiseRecordForTabFile.get(k));
                            } else if (printFormatTable.get(variableNameList.get(k)).equals("MONTH")) {
                                // month
                                dbgLog.finer("data k=" + k + ":" + casewiseRecordForTabFile.get(k));
                                dbgLog.finer("data k=" + k + ":" + SPSSConstants.MONTH_LIST
                                        .get(Integer.valueOf(casewiseRecordForTabFile.get(k).toString()) - 1));
                                String newDatum = SPSSConstants.MONTH_LIST
                                        .get(Integer.valueOf(casewiseRecordForTabFile.get(k).toString()) - 1);
                                casewiseRecordForTabFile.set(k, newDatum);
                                dbgLog.finer("month:k=" + k + ":" + casewiseRecordForTabFile.get(k));

                            }
                        }
                        // end of date/time block
                    } // end: date-time-datum check

                } // end: loop-k(2nd: variablte-wise-check)

                // write to tab file
                if (casewiseRecordForTabFile.size() > 0) {
                    pwout.println(StringUtils.join(casewiseRecordForTabFile, "\t"));
                }

                // numeric contents-check
                for (int l = 0; l < casewiseRecordForTabFile.size(); l++) {
                    if (variableFormatTypeList[l].equals("date") || variableFormatTypeList[l].equals("time")
                            || printFormatTable.get(variableNameList.get(l)).equals("WKDAY")
                            || printFormatTable.get(variableNameList.get(l)).equals("MONTH")) {

                    } else {
                        if (variableTypelList.get(l) <= 0) {
                            if (casewiseRecordForTabFile.get(l).toString().indexOf(".") >= 0) {
                                decimalVariableSet.add(l);
                            }
                        }
                    }
                }

                // reset the case-wise working objects
                casewiseRecordForTabFile.clear();

                if (stream.available() == 0) {
                    // reached the end of this file
                    // do exit-processing

                    dbgLog.fine("reached the end of the file at " + ii + "th iteration");

                    break;
                } // if eof processing
            } //i-loop: case(row) iteration

            // close the writer
            pwout.close();

        } catch (IOException ex) {
            throw ex;
        }

        // contents check
        dbgLog.fine("numberOfDecimalVariables=" + numberOfDecimalVariables);
        dbgLog.fine("decimalVariableSet=" + decimalVariableSet);

        dbgLog.fine("***** decodeRecordTypeDataUnCompressed(): end *****");
    }

    // Utility Methods  -----------------------------------------------------//

    private boolean variableNameIsAnIncrement(String varNameBase, String variableName) {
        if (varNameBase == null) {
            return false;
        }

        if (varNameBase.concat("0").equals(variableName)) {
            return true;
        }

        return false;
    }

    private boolean variableNameIsAnIncrement(String varNameBase, String lastExtendedVariable,
            String currentVariable) {

        if (varNameBase == null || lastExtendedVariable == null || currentVariable == null) {
            return false;
        }

        if (varNameBase.length() >= lastExtendedVariable.length()) {
            return false;
        }

        if (varNameBase.length() >= currentVariable.length()) {
            return false;
        }

        if (!(varNameBase.equals(currentVariable.substring(0, varNameBase.length())))) {
            return false;
        }

        String lastSuffix = lastExtendedVariable.substring(varNameBase.length());
        String currentSuffix = currentVariable.substring(varNameBase.length());

        if (currentSuffix.length() > 2) {
            return false;
        }

        //if ( !currentSuffix.matches("^[0-9A-Z]*$") ) {
        //    return false; 
        //}

        return suffixIsAnIncrement(lastSuffix, currentSuffix);
    }

    private boolean suffixIsAnIncrement(String lastSuffix, String currentSuffix) {
        // Extended variable suffixes are base-36 number strings in the 
        // [0-9A-Z] alphabet. I.e. the incremental suffixes go from 
        // 0 to 9 to A to Z to 10 to 1Z ... etc. 

        int lastSuffixValue = intBase36(lastSuffix);
        int currentSuffixValue = intBase36(currentSuffix);

        if (currentSuffixValue - lastSuffixValue > 0) {
            return true;
        }

        return false;
    }

    private int intBase36(String stringBase36) {

        // integer value of a base-36 string in [0-9A-Z] alphabet;
        // i.e. "0" = 0, "9" = 9, "A" = 10, 
        // "Z"  = 35, "10" = 36, "1Z" = 71 ...

        byte[] stringBytes = stringBase36.getBytes();

        int ret = 0;

        for (int i = 0; i < stringBytes.length; i++) {
            int value = 0;
            if (stringBytes[i] >= 48 && stringBytes[i] <= 57) {
                // [0-9]
                value = (int) stringBytes[i] - 48;
            } else if (stringBytes[i] >= 65 && stringBytes[i] <= 90) {
                // [A-Z] 
                value = (int) stringBytes[i] - 55;
            }

            ret = (ret * 36) + value;
        }

        return ret;
    }

    private int getSAVintAdjustedBlockLength(int rawLength) {
        int adjustedLength = rawLength;
        if ((rawLength % LENGTH_SAV_INT_BLOCK) != 0) {
            adjustedLength = LENGTH_SAV_INT_BLOCK * (rawLength / LENGTH_SAV_INT_BLOCK + 1);
        }
        return adjustedLength;
    }

    private int getSAVobsAdjustedBlockLength(int rawLength) {
        int adjustedLength = rawLength;
        if ((rawLength % LENGTH_SAV_OBS_BLOCK) != 0) {
            adjustedLength = LENGTH_SAV_OBS_BLOCK * (rawLength / LENGTH_SAV_OBS_BLOCK + 1);
        }
        return adjustedLength;
    }

    private int[] parseRT7SubTypefieldHeader(BufferedInputStream stream) throws IOException {
        int length_unit_length = 4;
        int length_number_of_units = 4;
        int storage_size = length_unit_length + length_number_of_units;

        int[] headerSection = new int[2];

        byte[] byteStorage = new byte[storage_size];

        try {
            int nbytes = stream.read(byteStorage);
            // to-do check against nbytes

            //printHexDump(byteStorage, "RT7:storage");

            ByteBuffer bb_data_type = ByteBuffer.wrap(byteStorage, 0, length_unit_length);
            if (isLittleEndian) {
                bb_data_type.order(ByteOrder.LITTLE_ENDIAN);
            }

            int unitLength = bb_data_type.getInt();
            dbgLog.fine("parseRT7 SubTypefield: unitLength=" + unitLength);

            ByteBuffer bb_number_of_units = ByteBuffer.wrap(byteStorage, length_unit_length,
                    length_number_of_units);
            if (isLittleEndian) {
                bb_number_of_units.order(ByteOrder.LITTLE_ENDIAN);
            }

            int numberOfUnits = bb_number_of_units.getInt();
            dbgLog.fine("parseRT7 SubTypefield: numberOfUnits=" + numberOfUnits);

            headerSection[0] = unitLength;
            headerSection[1] = numberOfUnits;
            return headerSection;
        } catch (IOException ex) {
            throw ex;
        }
    }

    // TODO: 
    // rename this method "skipRT7SubTypefield or parseAndSkip... 
    // -- because that's what it really does. We only call it 
    // on RT7 sub-fields that we don't know what to do with.
    // -- L.A. 4.0 beta
    private void parseRT7SubTypefield(BufferedInputStream stream) throws IOException {
        int length_unit_length = 4;
        int length_number_of_units = 4;
        int storage_size = length_unit_length + length_number_of_units;

        int[] headerSection = new int[2];

        byte[] byteStorage = new byte[storage_size];

        try {
            int nbytes = stream.read(byteStorage);
            // to-do check against nbytes

            //printHexDump(byteStorage, "RT7:storage");

            ByteBuffer bb_data_type = ByteBuffer.wrap(byteStorage, 0, length_unit_length);
            if (isLittleEndian) {
                bb_data_type.order(ByteOrder.LITTLE_ENDIAN);
            }

            int unitLength = bb_data_type.getInt();
            dbgLog.fine("parseRT7 SubTypefield: unitLength=" + unitLength);

            ByteBuffer bb_number_of_units = ByteBuffer.wrap(byteStorage, length_unit_length,
                    length_number_of_units);
            if (isLittleEndian) {
                bb_number_of_units.order(ByteOrder.LITTLE_ENDIAN);
            }

            int numberOfUnits = bb_number_of_units.getInt();
            dbgLog.fine("parseRT7 SubTypefield: numberOfUnits=" + numberOfUnits);

            headerSection[0] = unitLength;
            headerSection[1] = numberOfUnits;

            for (int i = 0; i < numberOfUnits; i++) {
                byte[] work = new byte[unitLength];

                int nb = stream.read(work);
                dbgLog.finer("raw bytes in Hex:" + new String(Hex.encodeHex(work)));
                ByteBuffer bb_field = ByteBuffer.wrap(work);
                if (isLittleEndian) {
                    bb_field.order(ByteOrder.LITTLE_ENDIAN);
                }
                dbgLog.fine("RT7ST: raw bytes in Hex:" + new String(Hex.encodeHex(bb_field.array())));
                if (unitLength == 4) {
                    int fieldData = bb_field.getInt();
                    dbgLog.fine("RT7ST: " + i + "-th fieldData=" + fieldData);
                    dbgLog.fine("RT7ST: fieldData in Hex=" + Integer.toHexString(fieldData));
                } else if (unitLength == 8) {
                    double fieldData = bb_field.getDouble();
                    dbgLog.finer("RT7ST: " + i + "-th fieldData=" + fieldData);
                    dbgLog.finer("RT7ST: fieldData in Hex=" + Double.toHexString(fieldData));

                }
                dbgLog.finer("");
            }

        } catch (IOException ex) {
            //ex.printStackTrace();
            throw ex;
        }

    }

    private List<byte[]> getRT7SubTypefieldData(BufferedInputStream stream) throws IOException {
        int length_unit_length = 4;
        int length_number_of_units = 4;
        int storage_size = length_unit_length + length_number_of_units;
        List<byte[]> dataList = new ArrayList<byte[]>();
        int[] headerSection = new int[2];

        byte[] byteStorage = new byte[storage_size];

        try {
            int nbytes = stream.read(byteStorage);
            // to-do check against nbytes

            //printHexDump(byteStorage, "RT7:storage");

            ByteBuffer bb_data_type = ByteBuffer.wrap(byteStorage, 0, length_unit_length);
            if (isLittleEndian) {
                bb_data_type.order(ByteOrder.LITTLE_ENDIAN);
            }

            int unitLength = bb_data_type.getInt();
            dbgLog.fine("parseRT7SubTypefield: unitLength=" + unitLength);

            ByteBuffer bb_number_of_units = ByteBuffer.wrap(byteStorage, length_unit_length,
                    length_number_of_units);
            if (isLittleEndian) {
                bb_number_of_units.order(ByteOrder.LITTLE_ENDIAN);
            }

            int numberOfUnits = bb_number_of_units.getInt();
            dbgLog.fine("parseRT7SubTypefield: numberOfUnits=" + numberOfUnits);

            headerSection[0] = unitLength;
            headerSection[1] = numberOfUnits;

            for (int i = 0; i < numberOfUnits; i++) {

                byte[] work = new byte[unitLength];
                int nb = stream.read(work);
                dbgLog.finer(new String(Hex.encodeHex(work)));
                dataList.add(work);
            }

        } catch (IOException ex) {
            //ex.printStackTrace();
            throw ex;
        }
        return dataList;
    }

    void print2Darray(Object[][] datatable, String title) {
        dbgLog.fine(title);
        for (int i = 0; i < datatable.length; i++) {
            dbgLog.fine(StringUtils.join(datatable[i], "|"));
        }
    }

}