All Downloads are FREE. Search and download functionalities are using the official Maven repository.

org.apache.openjpa.jdbc.sql.DBDictionary Maven / Gradle / Ivy

There is a newer version: 4.0.1
Show newest version
/*
 * Licensed to the Apache Software Foundation (ASF) under one
 * or more contributor license agreements.  See the NOTICE file
 * distributed with this work for additional information
 * regarding copyright ownership.  The ASF licenses this file
 * to you 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.
 */
package org.apache.openjpa.jdbc.sql;

import java.io.BufferedReader;
import java.io.ByteArrayInputStream;
import java.io.CharArrayReader;
import java.io.IOException;
import java.io.InputStream;
import java.io.InputStreamReader;
import java.io.OutputStream;
import java.io.Reader;
import java.io.StringReader;
import java.io.Writer;
import java.math.BigDecimal;
import java.math.BigInteger;
import java.sql.Array;
import java.sql.Blob;
import java.sql.Clob;
import java.sql.Connection;
import java.sql.DatabaseMetaData;
import java.sql.PreparedStatement;
import java.sql.Ref;
import java.sql.ResultSet;
import java.sql.SQLException;
import java.sql.SQLWarning;
import java.sql.Statement;
import java.sql.Time;
import java.sql.Timestamp;
import java.sql.Types;
import java.text.MessageFormat;
import java.time.LocalDate;
import java.time.LocalDateTime;
import java.time.LocalTime;
import java.time.OffsetDateTime;
import java.time.OffsetTime;
import java.time.ZoneId;
import java.util.ArrayList;
import java.util.Arrays;
import java.util.Calendar;
import java.util.Collection;
import java.util.Collections;
import java.util.Date;
import java.util.GregorianCalendar;
import java.util.HashMap;
import java.util.HashSet;
import java.util.Iterator;
import java.util.LinkedHashSet;
import java.util.List;
import java.util.Locale;
import java.util.Map;
import java.util.Set;
import java.util.TreeMap;

import javax.sql.DataSource;

import org.apache.openjpa.jdbc.conf.JDBCConfiguration;
import org.apache.openjpa.jdbc.identifier.ColumnDefIdentifierRule;
import org.apache.openjpa.jdbc.identifier.DBIdentifier;
import org.apache.openjpa.jdbc.identifier.DBIdentifier.DBIdentifierType;
import org.apache.openjpa.jdbc.identifier.DBIdentifierRule;
import org.apache.openjpa.jdbc.identifier.DBIdentifierUtil;
import org.apache.openjpa.jdbc.identifier.Normalizer;
import org.apache.openjpa.jdbc.kernel.JDBCFetchConfiguration;
import org.apache.openjpa.jdbc.kernel.JDBCStore;
import org.apache.openjpa.jdbc.kernel.exps.ExpContext;
import org.apache.openjpa.jdbc.kernel.exps.ExpState;
import org.apache.openjpa.jdbc.kernel.exps.FilterValue;
import org.apache.openjpa.jdbc.kernel.exps.Null;
import org.apache.openjpa.jdbc.kernel.exps.Val;
import org.apache.openjpa.jdbc.meta.ClassMapping;
import org.apache.openjpa.jdbc.meta.FieldMapping;
import org.apache.openjpa.jdbc.meta.JavaSQLTypes;
import org.apache.openjpa.jdbc.schema.Column;
import org.apache.openjpa.jdbc.schema.DataSourceFactory;
import org.apache.openjpa.jdbc.schema.ForeignKey;
import org.apache.openjpa.jdbc.schema.ForeignKey.FKMapKey;
import org.apache.openjpa.jdbc.schema.Index;
import org.apache.openjpa.jdbc.schema.NameSet;
import org.apache.openjpa.jdbc.schema.PrimaryKey;
import org.apache.openjpa.jdbc.schema.Schema;
import org.apache.openjpa.jdbc.schema.SchemaGroup;
import org.apache.openjpa.jdbc.schema.Sequence;
import org.apache.openjpa.jdbc.schema.Table;
import org.apache.openjpa.jdbc.schema.Unique;
import org.apache.openjpa.kernel.Filters;
import org.apache.openjpa.kernel.OpenJPAStateManager;
import org.apache.openjpa.kernel.Seq;
import org.apache.openjpa.kernel.StateManagerImpl;
import org.apache.openjpa.kernel.exps.Path;
import org.apache.openjpa.lib.conf.Configurable;
import org.apache.openjpa.lib.conf.Configuration;
import org.apache.openjpa.lib.identifier.IdentifierConfiguration;
import org.apache.openjpa.lib.identifier.IdentifierRule;
import org.apache.openjpa.lib.identifier.IdentifierUtil;
import org.apache.openjpa.lib.jdbc.ConnectionDecorator;
import org.apache.openjpa.lib.jdbc.LoggingConnectionDecorator;
import org.apache.openjpa.lib.log.Log;
import org.apache.openjpa.lib.util.Localizer;
import org.apache.openjpa.lib.util.Localizer.Message;
import org.apache.openjpa.lib.util.StringUtil;
import org.apache.openjpa.meta.FieldMetaData;
import org.apache.openjpa.meta.JavaTypes;
import org.apache.openjpa.meta.ValueStrategies;
import org.apache.openjpa.util.ExceptionInfo;
import org.apache.openjpa.util.GeneralException;
import org.apache.openjpa.util.InternalException;
import org.apache.openjpa.util.InvalidStateException;
import org.apache.openjpa.util.LockException;
import org.apache.openjpa.util.ObjectExistsException;
import org.apache.openjpa.util.ObjectNotFoundException;
import org.apache.openjpa.util.OpenJPAException;
import org.apache.openjpa.util.OptimisticException;
import org.apache.openjpa.util.ProxyManager;
import org.apache.openjpa.util.QueryException;
import org.apache.openjpa.util.ReferentialIntegrityException;
import org.apache.openjpa.util.Serialization;
import org.apache.openjpa.util.StoreException;
import org.apache.openjpa.util.UnsupportedException;
import org.apache.openjpa.util.UserException;


/**
 * Class which allows the creation of SQL dynamically, in a
 * database agnostic fashion. Subclass for the nuances of different data stores.
 */
public class DBDictionary
    implements Configurable, ConnectionDecorator, JoinSyntaxes,
    LoggingConnectionDecorator.SQLWarningHandler, IdentifierConfiguration {

    public static final String VENDOR_OTHER      = "other";
    public static final String VENDOR_DATADIRECT = "datadirect";

    public static final String SCHEMA_CASE_UPPER     = IdentifierUtil.CASE_UPPER;
    public static final String SCHEMA_CASE_LOWER     = IdentifierUtil.CASE_LOWER;
    public static final String SCHEMA_CASE_PRESERVE  = IdentifierUtil.CASE_PRESERVE;

    public static final String CONS_NAME_BEFORE = "before";
    public static final String CONS_NAME_MID    = "mid";
    public static final String CONS_NAME_AFTER  = "after";

    public int blobBufferSize = 50000;
    public int clobBufferSize = 50000;

    protected static final int RANGE_POST_SELECT   = 0;
    protected static final int RANGE_PRE_DISTINCT  = 1;
    protected static final int RANGE_POST_DISTINCT = 2;
    protected static final int RANGE_POST_LOCK     = 3;

    protected static final int NANO  = 1;
    protected static final int MICRO = NANO * 1000;
    protected static final int MILLI = MICRO * 1000;
    protected static final int CENTI = MILLI * 10;
    protected static final int DECI  = MILLI * 100;
    protected static final int SEC   = MILLI * 1000;

    protected static final int NAME_ANY      = DBIdentifierUtil.ANY;
    protected static final int NAME_TABLE    = DBIdentifierUtil.TABLE;
    protected static final int NAME_SEQUENCE = DBIdentifierUtil.SEQUENCE;

    protected static final int UNLIMITED = -1;
    protected static final int NO_BATCH = 0;

    private static final String ZERO_DATE_STR      = "'" + new java.sql.Date(0) + "'";
    private static final String ZERO_TIME_STR      = "'" + new Time(0) + "'";
    private static final String ZERO_TIMESTAMP_STR = "'" + new Timestamp(0) + "'";

    private static final Localizer _loc = Localizer.forPackage(DBDictionary.class);

    // Database version info preferably set from Connection metadata
    private int major;
    private int minor;

    // schema data
    public String platform = "Generic";
    public String databaseProductName = "";
    public String databaseProductVersion = "";
    public String driverVendor = null;
    public boolean createPrimaryKeys = true;
    public String constraintNameMode = CONS_NAME_BEFORE;
    public int maxTableNameLength = 128;
    public int maxColumnNameLength = 128;
    public int maxConstraintNameLength = 128;
    public int maxIndexNameLength = 128;
    public int maxIndexesPerTable = Integer.MAX_VALUE;
    public boolean supportsForeignKeys = true;
    public boolean supportsParameterInSelect = true;
    public boolean supportsForeignKeysComposite = true;
    public boolean supportsUniqueConstraints = true;
    public boolean supportsDeferredConstraints = true;
    public boolean supportsRestrictDeleteAction = true;
    public boolean supportsCascadeDeleteAction = true;
    public boolean supportsNullDeleteAction = true;
    public boolean supportsNullUniqueColumn = true;
    public boolean supportsDefaultDeleteAction = true;
    public boolean supportsRestrictUpdateAction = true;
    public boolean supportsCascadeUpdateAction = true;
    public boolean supportsNullUpdateAction = true;
    public boolean supportsDefaultUpdateAction = true;
    public boolean supportsAlterTableWithAddColumn = true;
    public boolean supportsAlterTableWithDropColumn = true;
    public boolean supportsComments = false;
    public Boolean supportsGetGeneratedKeys = null;
    public String reservedWords = null;
    public String systemSchemas = null;
    public String systemTables = null;
    public String selectWords = null;
    public String fixedSizeTypeNames = null;
    public String schemaCase = SCHEMA_CASE_UPPER;
    public boolean setStringRightTruncationOn = true;
    public boolean fullResultCollectionInOrderByRelation = false;
    public boolean disableSchemaFactoryColumnTypeErrors = false; //OPENJPA-2627

    // sql
    public boolean disableAlterSeqenceIncrementBy=false;
    public String validationSQL = null;
    public String closePoolSQL = null;
    public String initializationSQL = null;
    public int joinSyntax = SYNTAX_SQL92;
    public String outerJoinClause = "LEFT OUTER JOIN";
    public String innerJoinClause = "INNER JOIN";
    public String crossJoinClause = "CROSS JOIN";
    public boolean requiresConditionForCrossJoin = false;
    public String forUpdateClause = "FOR UPDATE";
    public String tableForUpdateClause = null;
    public String distinctCountColumnSeparator = null;
    public boolean supportsSelectForUpdate = true;
    public boolean supportsLockingWithDistinctClause = true;
    public boolean supportsLockingWithMultipleTables = true;
    public boolean supportsLockingWithOrderClause = true;
    public boolean supportsLockingWithOuterJoin = true;
    public boolean supportsLockingWithInnerJoin = true;
    public boolean supportsLockingWithSelectRange = true;
    public boolean supportsQueryTimeout = true;
    public boolean allowQueryTimeoutOnFindUpdate = false; //OPENJPA-2517
    public boolean simulateLocking = false;
    public boolean supportsSubselect = true;
    public boolean supportsCorrelatedSubselect = true;
    public boolean supportsHaving = true;
    public boolean supportsSelectStartIndex = false;
    public boolean supportsSelectEndIndex = false;
    public int rangePosition = RANGE_POST_SELECT;
    public boolean requiresAliasForSubselect = false;
    public boolean requiresTargetForDelete = false;
    public boolean allowsAliasInBulkClause = true;
    public boolean supportsMultipleNontransactionalResultSets = true;
    public boolean requiresSearchStringEscapeForLike = false;
    public String searchStringEscape = "\\";
    public boolean requiresCastForMathFunctions = false;
    public boolean requiresCastForComparisons = false;
    public boolean supportsModOperator = false;
    public boolean supportsXMLColumn = false;
    public boolean supportsCaseConversionForLob = false;
    public boolean reportsSuccessNoInfoOnBatchUpdates = false;
    public boolean supportsSelectFromFinalTable = false;
    public boolean supportsSimpleCaseExpression = true;
    public boolean supportsGeneralCaseExpression = true;
    public boolean useWildCardForCount = false;

    /**
     * Some Databases append whitespace after the schema name
     */
    public boolean trimSchemaName = false;

    // functions
    public String castFunction = "CAST({0} AS {1})";
    public String toLowerCaseFunction = "LOWER({0})";
    public String toUpperCaseFunction = "UPPER({0})";
    public String stringLengthFunction = "CHAR_LENGTH({0})";
    public String bitLengthFunction = "(OCTET_LENGTH({0}) * 8)";
    public String trimLeadingFunction = "TRIM(LEADING {1} FROM {0})";
    public String trimTrailingFunction = "TRIM(TRAILING {1} FROM {0})";
    public String trimBothFunction = "TRIM(BOTH {1} FROM {0})";
    public String concatenateFunction = "({0}||{1})";
    public String concatenateDelimiter = "'OPENJPATOKEN'";
    public String substringFunctionName = "SUBSTRING";
    public String currentDateFunction = "CURRENT_DATE";
    public String currentTimeFunction = "CURRENT_TIME";
    public String currentTimestampFunction = "CURRENT_TIMESTAMP";
    public String dropTableSQL = "DROP TABLE {0}";

    // types
    public boolean storageLimitationsFatal = false;
    public boolean storeLargeNumbersAsStrings = false;
    public boolean storeCharsAsNumbers = true;
    public boolean trimStringColumns = false;
    public boolean useGetBytesForBlobs = false;
    public boolean useSetBytesForBlobs = false;
    public boolean useGetObjectForBlobs = false;
    public boolean useGetStringForClobs = false;
    public boolean useSetStringForClobs = false;
    public boolean useJDBC4SetBinaryStream = true;//OPENJPA-2067
    public int maxEmbeddedBlobSize = -1;
    public int maxEmbeddedClobSize = -1;
    public int inClauseLimit = -1;
    public int datePrecision = MILLI;

    /**
     * @deprecated Use 'dateMillisecondBehavior' instead.
     */
    @Deprecated
    public boolean roundTimeToMillisec = true;

    /*
     * This defines how the milliseconds of a Date field are handled
     * when the Date is retrieved from the database, as follows:
     *
     * ROUND: This is the default.  The
     * Date will be rounded to the nearest millisecond.
     * DROP: The milliseconds will be dropped, thus rounding is not
     * performed.  As an example, a date of '2010-01-01 12:00:00.687701'
     * stored in the database will become '2010-01-01 12:00:00.000' in
     * the Date field of the entity.
     * RETAIN: The milliseconds will not be rounded and retained.  As an
     * example, a date of '2010-01-01 12:00:00.687701' stored in the
     * database will become '2010-01-01 12:00:00.687' in the Date field
     * of the entity.
     */
    public enum DateMillisecondBehaviors { DROP, ROUND, RETAIN };
    private DateMillisecondBehaviors dateMillisecondBehavior;

    /**
     * Defines how {@code Boolean} and {@code boolean} values get represented
     * in OpenJPA. Default to {@code INT_10}.
     * for backward compatibility.
     */
    protected BooleanRepresentation booleanRepresentation = BooleanRepresentationFactory.INT_10;

    public int characterColumnSize = 255;
    public String arrayTypeName = "ARRAY";
    public String bigintTypeName = "BIGINT";
    public String binaryTypeName = "BINARY";
    public String bitTypeName = "BIT";
    public String blobTypeName = "BLOB";
    public String booleanTypeName = "BOOLEAN";
    public String charTypeName = "CHAR";
    public String clobTypeName = "CLOB";
    public String dateTypeName = "DATE";
    public String decimalTypeName = "DECIMAL";
    public String distinctTypeName = "DISTINCT";
    public String doubleTypeName = "DOUBLE";
    public String floatTypeName = "FLOAT";
    public String integerTypeName = "INTEGER";
    public String javaObjectTypeName = "JAVA_OBJECT";
    public String longVarbinaryTypeName = "LONGVARBINARY";
    public String longVarcharTypeName = "LONGVARCHAR";
    public String nullTypeName = "NULL";
    public String numericTypeName = "NUMERIC";
    public String otherTypeName = "OTHER";
    public String realTypeName = "REAL";
    public String refTypeName = "REF";
    public String smallintTypeName = "SMALLINT";
    public String structTypeName = "STRUCT";
    public String timeTypeName = "TIME";
    public String timestampTypeName = "TIMESTAMP";
    public String timeWithZoneTypeName = "TIME WITH TIME ZONE";
    public String timestampWithZoneTypeName = "TIMESTAMP WITH TIME ZONE";
    public String tinyintTypeName = "TINYINT";
    public String varbinaryTypeName = "VARBINARY";
    public String varcharTypeName = "VARCHAR";
    public String xmlTypeName = "XML";
    public String xmlTypeEncoding = "UTF-8";
    public String getStringVal = "";

    // schema metadata
    public boolean useSchemaName = true;
    public String tableTypes = "TABLE";
    public boolean supportsSchemaForGetTables = true;
    public boolean supportsSchemaForGetColumns = true;
    public boolean supportsNullTableForGetColumns = true;
    public boolean supportsNullTableForGetPrimaryKeys = false;
    public boolean supportsNullTableForGetIndexInfo = false;
    public boolean supportsNullTableForGetImportedKeys = false;
    public boolean useGetBestRowIdentifierForPrimaryKeys = false;
    public boolean requiresAutoCommitForMetaData = false;
    public boolean tableLengthIncludesSchema = false;

    // auto-increment
    public int maxAutoAssignNameLength = 31;
    public String autoAssignClause = null;
    public String autoAssignTypeName = null;
    public boolean supportsAutoAssign = false;
    public String lastGeneratedKeyQuery = null;
    public String nextSequenceQuery = null;
    public String sequenceSQL = null;
    public String sequenceSchemaSQL = null;
    public String sequenceNameSQL = null;
    // most native sequences can be run inside the business transaction
    public int nativeSequenceType= Seq.TYPE_CONTIGUOUS;

    /**
     * This variable was used in 2.1.x and prior releases to indicate that
     * OpenJPA should not use the CACHE clause when getting a native
     * sequence; instead the INCREMENT BY clause gets its value equal to the
     * allocationSize property.  Post 2.1.x, code was added to allow
     * said functionality by default (see OPENJPA-1376).  For forward
     * compatibility, this variable should not be removed.
     */
    @Deprecated
    public boolean useNativeSequenceCache = true;

    /**
     * If a user sets the previous variable (useNativeSequenceCache) to false, we should log a
     * warning indicating that the variable no longer has an effect due to the code changes
     * of OPENJPA-1376.  We only want to log the warning once per instance, thus this
     * variable will be used to indicate if the warning should be printed or not.
     */
    @Deprecated
    private boolean logNativeSequenceCacheWarning = true;

    protected JDBCConfiguration conf = null;
    protected Log log = null;
    protected boolean connected = false;
    protected boolean isJDBC3 = false;
    protected boolean isJDBC4 = false;
    protected final Set reservedWordSet = new HashSet<>();
    // reservedWordSet subset that CANNOT be used as valid column names
    // (i.e., without surrounding them with double-quotes)
    protected Set invalidColumnWordSet = new HashSet<>();
    protected final Set systemSchemaSet = new HashSet<>();
    protected final Set systemTableSet = new HashSet<>();
    protected final Set fixedSizeTypeNameSet = new HashSet<>();

    /**
     * set of types that might have an optional fraction (sub-second) part.
     * @{code @Column(scale=-1)} sets the fraction to zero.
     * @{code @Column(scale=0)} will use the default behaviour (no fraction information generated)
     */
    public final Set fractionalTypeNameSet = new HashSet<>();

    /**
     * Default amount of digits for fractional Types.
     * This is not supported/required by every database.
     *
     * This value is only being used if no explicit {@code @Column(scale=n)} is set or n is zero.
     * Use {@code @Column(scale=-1)} to disable the explicit fraction part in the SQL generator.
     * @see #fractionalTypeNameSet
     * @see #getDateFractionDigits(Column, String)
     */
    public int dateFractionDigits = 6;


    protected final Set typeModifierSet = new HashSet<>();

    // NamingConfiguration properties
    private boolean delimitIdentifiers = false;
    public Boolean supportsDelimitedIdentifiers = null;
    public String leadingDelimiter = "\"";
    public String trailingDelimiter = "\"";
    public String nameConcatenator = "_";
    public String delimitedCase = SCHEMA_CASE_PRESERVE;
    public String catalogSeparator = ".";
    protected String defaultSchemaName = null;
    private String conversionKey = null;

    // Naming utility and naming rules
    private DBIdentifierUtil namingUtil = null;
    private Map namingRules = new HashMap<>();
    private IdentifierRule defaultNamingRule = null;  // cached for performance

    /**
     * If a native query begins with any of the values found here then it will
     * be treated as a select statement.
     */
    protected final Set selectWordSet = new HashSet<>();

    // when we store values that lose precision, track the types so that the
    // first time it happens we can warn the user
    private Set> _precisionWarnedTypes = null;

    // batchLimit value:
    // -1 = unlimited
    // 0  = no batch
    // any positive number = batch limit
    public int batchLimit = NO_BATCH;

    public final Map> sqlStateCodes =
        new HashMap<>();

    protected ProxyManager _proxyManager;

    public DBDictionary() {
        fixedSizeTypeNameSet.addAll(Arrays.asList(new String[]{
            "BIGINT", "BIT", "BLOB", "CLOB", "DATE", "DECIMAL", "DISTINCT",
            "DOUBLE", "FLOAT", "INTEGER", "JAVA_OBJECT", "NULL", "NUMERIC",
            "OTHER", "REAL", "REF", "SMALLINT", "STRUCT", "TIME", "TIMESTAMP",
            "TINYINT",
        }));

        selectWordSet.add("SELECT");
    }

    /**
     * This method is called when the dictionary first sees any connection.
     * It is used to initialize dictionary metadata if needed. If you
     * override this method, be sure to call
     * super.connectedConfiguration.
     */
    public void connectedConfiguration(Connection conn)
        throws SQLException {
        if (!connected) {
            DatabaseMetaData metaData = null;
            try {
                metaData = conn.getMetaData();

                databaseProductName    = nullSafe(metaData.getDatabaseProductName());
                databaseProductVersion = nullSafe(metaData.getDatabaseProductVersion());
                setMajorVersion(metaData.getDatabaseMajorVersion());
                setMinorVersion(metaData.getDatabaseMinorVersion());
                try {
                    // JDBC3-only method, so it might throw an
                    // AbstractMethodError
                    int JDBCMajorVersion = metaData.getJDBCMajorVersion();
                    isJDBC3 = JDBCMajorVersion >= 3;
                    isJDBC4 = JDBCMajorVersion >= 4;
                } catch (Throwable t) {
                    // ignore if not JDBC3
                }
            } catch (Exception e) {
                if (log.isTraceEnabled())
                    log.trace(e.toString(), e);
            }

            if (log.isTraceEnabled()) {
                log.trace(DBDictionaryFactory.toString(metaData));

                if (isJDBC3) {
                    try {
                        log.trace(_loc.get("connection-defaults", new Object[]{
                            conn.getAutoCommit(), conn.getHoldability(),
                            conn.getTransactionIsolation()}));
                    } catch (Throwable t) {
                        log.trace("Unable to trace connection settings", t);
                    }
                }
            }

            // Configure the naming utility
            if (supportsDelimitedIdentifiers == null) // not explicitly set
                configureNamingUtil(metaData);

            // Auto-detect generated keys retrieval support unless user specified it.
            if (supportsGetGeneratedKeys == null) {
                supportsGetGeneratedKeys =  (isJDBC3) ? metaData.supportsGetGeneratedKeys() : false;
            }
            if (log.isInfoEnabled()) {
                log.info(_loc.get("dict-info", new Object[] {
                        metaData.getDatabaseProductName(), getMajorVersion(), getMinorVersion(),
                        metaData.getDriverName(), metaData.getDriverVersion()}));
            }
        }
        connected = true;
    }

    private void configureNamingUtil(DatabaseMetaData metaData) {
        // Get the naming utility from the configuration
        setSupportsDelimitedIdentifiers(metaData);
        setDelimitedCase(metaData);
    }

    /**
     * Configures the naming rules for this dictionary.  Subclasses should
     * override this method, providing their own naming rules.
     */
    protected void configureNamingRules() {
        // Add the default naming rule
        DBIdentifierRule defRule = new DBIdentifierRule(DBIdentifierType.DEFAULT, reservedWordSet);
        namingRules.put(defRule.getName(), defRule);
        // Disable delimiting of column definition.  DB platforms are very
        // picky about delimiters in column definitions. Base column types
        // do not require delimiters and will cause failures if delimited.
        DBIdentifierRule cdRule = new ColumnDefIdentifierRule();
        cdRule.setCanDelimit(false);
        namingRules.put(cdRule.getName(), cdRule);
    }

    //////////////////////
    // ResultSet wrappers
    //////////////////////

    /**
     * Convert the specified column of the SQL ResultSet to the proper
     * java type.
     */
    public Array getArray(ResultSet rs, int column)
        throws SQLException {
        return rs.getArray(column);
    }

    /**
     * Convert the specified column of the SQL ResultSet to the proper
     * java type.
     */
    public InputStream getAsciiStream(ResultSet rs, int column)
        throws SQLException {
        return rs.getAsciiStream(column);
    }

    /**
     * Convert the specified column of the SQL ResultSet to the proper
     * java type.
     */
    public BigDecimal getBigDecimal(ResultSet rs, int column)
        throws SQLException {
        if (storeLargeNumbersAsStrings) {
            String str = getString(rs, column);
            return (str == null) ? null : new BigDecimal(str);
        }
        return rs.getBigDecimal(column);
    }

    /**
     * Returns the specified column value as an unknown numeric type;
     * we try from the most generic to the least generic.
     */
    public Number getNumber(ResultSet rs, int column)
        throws SQLException {
        // try from the most generic, and if errors occur, try
        // less generic types; this enables us to handle values
        // like Double.NaN without having to introspect on the
        // ResultSetMetaData (bug #1053). Note that we handle
        // generic exceptions, since some drivers may throw
        // NumberFormatExceptions, whereas others may throw SQLExceptions
        try {
            return getBigDecimal(rs, column);
        } catch (Exception e1) {
            try {
                return Double.valueOf(getDouble(rs, column));
            } catch (Exception e2) {
                try {
                    return Float.valueOf(getFloat(rs, column));
                } catch (Exception e3) {
                    try {
                        return getLong(rs, column);
                    } catch (Exception e4) {
                        try {
                            return getInt(rs, column);
                        } catch (Exception e5) {
                        }
                    }
                }
            }

            if (e1 instanceof RuntimeException)
                throw(RuntimeException) e1;
            if (e1 instanceof SQLException)
                throw(SQLException) e1;
        }

        return null;
    }

    /**
     * Convert the specified column of the SQL ResultSet to the proper
     * java type.
     */
    public BigInteger getBigInteger(ResultSet rs, int column)
        throws SQLException {
        if (storeLargeNumbersAsStrings) {
            String str = getString(rs, column);
            return (str == null) ? null : new BigDecimal(str).toBigInteger();
        }
        BigDecimal bd = getBigDecimal(rs, column);
        return (bd == null) ? null : bd.toBigInteger();
    }

    /**
     * Convert the specified column of the SQL ResultSet to the proper
     * java type.
     */
    public InputStream getBinaryStream(ResultSet rs, int column)
        throws SQLException {
        return rs.getBinaryStream(column);
    }

    public InputStream getLOBStream(JDBCStore store, ResultSet rs,
        int column) throws SQLException {
        return rs.getBinaryStream(column);
    }

    /**
     * Convert the specified column of the SQL ResultSet to the proper
     * java type.
     */
    public Blob getBlob(ResultSet rs, int column)
        throws SQLException {
        return rs.getBlob(column);
    }

    /**
     * Convert the specified column of the SQL ResultSet to the proper
     * java type.
     */
    public Object getBlobObject(ResultSet rs, int column, JDBCStore store)
        throws SQLException {
        InputStream in = null;
        if (useGetBytesForBlobs || useGetObjectForBlobs) {
            byte[] bytes = getBytes(rs, column);
            if (bytes != null && bytes.length > 0)
                in = new ByteArrayInputStream(bytes);
        } else {
            Blob blob = getBlob(rs, column);
            if (blob != null && blob.length() > 0)
                in = blob.getBinaryStream();
        }
        if (in == null)
            return null;

        try {
            if (store == null)
                return Serialization.deserialize(in, null);
            return Serialization.deserialize(in, store.getContext());
        } finally {
            try {
                in.close();
            } catch (IOException ioe) {
            }
        }
    }

    /**
     * Convert the specified column of the SQL ResultSet to the proper
     * java type.
     */
    public boolean getBoolean(ResultSet rs, int column)
        throws SQLException {
        return booleanRepresentation.getBoolean(rs, column);
    }

    /**
     * Convert the specified column of the SQL ResultSet to the proper
     * java type.
     */
    public byte getByte(ResultSet rs, int column)
        throws SQLException {
        return rs.getByte(column);
    }

    /**
     * Convert the specified column of the SQL ResultSet to the proper
     * java type.
     */
    public byte[] getBytes(ResultSet rs, int column)
        throws SQLException {
        if (useGetBytesForBlobs)
            return rs.getBytes(column);
        if (useGetObjectForBlobs)
            return (byte[]) rs.getObject(column);

        Blob blob = getBlob(rs, column);
        if (blob == null)
            return null;
        int length = (int) blob.length();
        if (length == 0)
            return null;
        return blob.getBytes(1, length);
    }

    /**
     * Convert the specified column of the SQL ResultSet to the proper
     * java type. Converts the date from a {@link Timestamp} by default.
     */
    public Calendar getCalendar(ResultSet rs, int column) throws SQLException {
        Date d = getDate(rs, column);
        if (d == null)
            return null;
        Calendar cal = (Calendar) getProxyManager().newCalendarProxy(GregorianCalendar.class, null);
        cal.setTime(d);
        return cal;
    }

    /**
     * Retrieve the specified column of the SQL ResultSet to the proper
     * {@link LocalDate} java type.
     */
    public LocalDate getLocalDate(ResultSet rs, int column) throws SQLException {
        java.sql.Date date = rs.getDate(column);
        return date != null ? date.toLocalDate() : null;
    }

    /**
     * Retrieve the specified column of the SQL ResultSet to the proper
     * {@link LocalTime} java type.
     */
    public LocalTime getLocalTime(ResultSet rs, int column) throws SQLException {
        java.sql.Time time = rs.getTime(column);
        return time != null ? time.toLocalTime() : null;
    }

    /**
     * Retrieve the specified column of the SQL ResultSet to the proper
     * {@link LocalDateTime} java type.
     */
    public LocalDateTime getLocalDateTime(ResultSet rs, int column) throws SQLException {
        Timestamp tst = rs.getTimestamp(column);
        return tst != null ? tst.toLocalDateTime() : null;
    }

    /**
     * Retrieve the specified column of the SQL ResultSet to the proper
     * {@link OffsetTime} java type.
     */
    public OffsetTime getOffsetTime(ResultSet rs, int column) throws SQLException {
        java.sql.Time time = rs.getTime(column);
        return time != null ? time.toLocalTime().atOffset(OffsetDateTime.now().getOffset()) : null;
    }

    /**
     * Retrieve the specified column of the SQL ResultSet to the proper
     * {@link OffsetDateTime} java type.
     */
    public OffsetDateTime getOffsetDateTime(ResultSet rs, int column) throws SQLException {
        Timestamp tst = rs.getTimestamp(column);
        return tst != null ? tst.toLocalDateTime().atZone(ZoneId.systemDefault()).toOffsetDateTime() : null;
    }

    private ProxyManager getProxyManager() {
        if (_proxyManager == null) {
            _proxyManager = conf.getProxyManagerInstance();
        }
        return _proxyManager;
    }

    /**
     * Convert the specified column of the SQL ResultSet to the proper
     * java type.
     */
    public char getChar(ResultSet rs, int column)
        throws SQLException {
        if (storeCharsAsNumbers)
            return (char) getInt(rs, column);

        String str = getString(rs, column);
        return (StringUtil.isEmpty(str)) ? 0 : str.charAt(0);
    }

    /**
     * Convert the specified column of the SQL ResultSet to the proper
     * java type.
     */
    public Reader getCharacterStream(ResultSet rs, int column)
        throws SQLException {
        return rs.getCharacterStream(column);
    }

    /**
     * Convert the specified column of the SQL ResultSet to the proper
     * java type.
     */
    public Clob getClob(ResultSet rs, int column)
        throws SQLException {
        return rs.getClob(column);
    }

    /**
     * Convert the specified column of the SQL ResultSet to the proper
     * java type.
     */
    public String getClobString(ResultSet rs, int column)
        throws SQLException {
        if (useGetStringForClobs)
            return rs.getString(column);

        Clob clob = getClob(rs, column);
        if (clob == null)
            return null;
        if (clob.length() == 0)
            return "";

        // unlikely that we'll have strings over Integer.MAX_VALUE chars
        return clob.getSubString(1, (int) clob.length());
    }

    /**
     * Convert the specified column of the SQL ResultSet to the proper
     * java type. Converts the date from a {@link Timestamp} by default.
     */
    public Date getDate(ResultSet rs, int column)
        throws SQLException {
        Timestamp tstamp = getTimestamp(rs, column, null);
        if (tstamp == null)
            return null;

        // get the fractional seconds component, rounding away anything beyond
        // milliseconds
        int fractional = 0;
        switch (getMillisecondBehavior()) {
            case DROP :
                fractional = 0;
                break;
            case RETAIN :
                fractional = (int) (tstamp.getNanos() / (double) MILLI);
                break;
            case ROUND :
                fractional = (int) Math.round(tstamp.getNanos() / (double) MILLI);
                break;
        }

        // get the millis component; some JDBC drivers round this to the
        // nearest second, while others do not
        long millis = (tstamp.getTime() / 1000L) * 1000L;
        return new Date(millis + fractional);
    }

    /**
     * Convert the specified column of the SQL ResultSet to the proper
     * java type.
     */
    public java.sql.Date getDate(ResultSet rs, int column, Calendar cal)
        throws SQLException {
        if (cal == null)
            return rs.getDate(column);
        return rs.getDate(column, cal);
    }

    /**
     * Convert the specified column of the SQL ResultSet to the proper
     * java type.
     */
    public double getDouble(ResultSet rs, int column)
        throws SQLException {
        return rs.getDouble(column);
    }

    /**
     * Convert the specified column of the SQL ResultSet to the proper
     * java type.
     */
    public float getFloat(ResultSet rs, int column)
        throws SQLException {
        return rs.getFloat(column);
    }

    /**
     * Convert the specified column of the SQL ResultSet to the proper
     * java type.
     */
    public int getInt(ResultSet rs, int column)
        throws SQLException {
        return rs.getInt(column);
    }

    /**
     * Convert the specified column of the SQL ResultSet to the proper
     * java type.
     */
    public Locale getLocale(ResultSet rs, int column)
        throws SQLException {
        String str = getString(rs, column);
        if (StringUtil.isEmpty(str))
            return null;

        String[] params = StringUtil.split(str, "_", 3);
        if (params.length < 3)
            return null;
        return new Locale(params[0], params[1], params[2]);
    }

    /**
     * Convert the specified column of the SQL ResultSet to the proper
     * java type.
     */
    public long getLong(ResultSet rs, int column)
        throws SQLException {
        return rs.getLong(column);
    }

    /**
     * Convert the specified column of the SQL ResultSet to the proper
     * java type.
     */
    public Object getObject(ResultSet rs, int column, Map map)
        throws SQLException {
        if (map == null)
            return rs.getObject(column);
        return rs.getObject(column, map);
    }

    /**
     * Convert the specified column of the SQL ResultSet to the proper
     * java type.
     */
    public Ref getRef(ResultSet rs, int column, Map map)
        throws SQLException {
        return rs.getRef(column);
    }

    /**
     * Convert the specified column of the SQL ResultSet to the proper
     * java type.
     */
    public short getShort(ResultSet rs, int column)
        throws SQLException {
        return rs.getShort(column);
    }

    /**
     * Convert the specified column of the SQL ResultSet to the proper
     * java type.
     */
    public String getString(ResultSet rs, int column)
        throws SQLException {
        String res = rs.getString(column);
        if ((res != null) && trimStringColumns) {
            res = res.trim();
        }
        return res;
    }

    /**
     * Convert the specified column of the SQL ResultSet to the proper
     * java type.
     */
    public Time getTime(ResultSet rs, int column, Calendar cal)
        throws SQLException {
        if (cal == null)
            return rs.getTime(column);
        return rs.getTime(column, cal);
    }

    /**
     * Convert the specified column of the SQL ResultSet to the proper
     * java type.
     */
    public Timestamp getTimestamp(ResultSet rs, int column, Calendar cal)
        throws SQLException {
        if (cal == null)
            return rs.getTimestamp(column);
        return rs.getTimestamp(column, cal);
    }

    //////////////////////////////
    // PreparedStatement wrappers
    //////////////////////////////

    /**
     * Set the given value as a parameter to the statement.
     */
    public void setArray(PreparedStatement stmnt, int idx, Array val,
        Column col)
        throws SQLException {
        stmnt.setArray(idx, val);
    }

    /**
     * Set the given value as a parameter to the statement.
     */
    public void setAsciiStream(PreparedStatement stmnt, int idx, InputStream val, int length, Column col)
        throws SQLException {
        stmnt.setAsciiStream(idx, val, length);
    }

    /**
     * Set the given value as a parameter to the statement.
     */
    public void setBigDecimal(PreparedStatement stmnt, int idx, BigDecimal val, Column col)
        throws SQLException {
        if ((col != null && col.isCompatible(Types.VARCHAR, null, 0, 0))
            || (col == null && storeLargeNumbersAsStrings)) {
            setString(stmnt, idx, val.toString(), col);
        }
        else {
            stmnt.setBigDecimal(idx, val);
        }
    }

    /**
     * Set the given value as a parameter to the statement.
     */
    public void setBigInteger(PreparedStatement stmnt, int idx, BigInteger val, Column col)
        throws SQLException {
        if ((col != null && col.isCompatible(Types.VARCHAR, null, 0, 0))
            || (col == null && storeLargeNumbersAsStrings)) {
            setString(stmnt, idx, val.toString(), col);
        }
        else {
            setBigDecimal(stmnt, idx, new BigDecimal(val), col);
        }
    }

    /**
     * Set the given value as a parameter to the statement.
     */
    public void setBinaryStream(PreparedStatement stmnt, int idx, InputStream val, int length, Column col)
        throws SQLException {

        //OPENJPA-2067: If the user has set the 'useJDBC4SetBinaryStream' property
        //then lets use the JDBC 4.0 version of the setBinaryStream method.
        if (useJDBC4SetBinaryStream) {
            if (isJDBC4){
                stmnt.setBinaryStream(idx, val);
                return;
            }
            else {
                log.trace(_loc.get("jdbc4-setbinarystream-unsupported"));
            }
        }

        stmnt.setBinaryStream(idx, val, length);
    }

    /**
     * Set the given value as a parameter to the statement.
     */
    public void setBlob(PreparedStatement stmnt, int idx, Blob val, Column col)
        throws SQLException {
        stmnt.setBlob(idx, val);
    }

    /**
     * Set the given value as a parameter to the statement. Uses the
     * {@link #serialize} method to serialize the value.
     */
    public void setBlobObject(PreparedStatement stmnt, int idx, Object val, Column col, JDBCStore store)
        throws SQLException {
        setBytes(stmnt, idx, serialize(val, store), col);
    }

    /**
     * Set the given value as a parameter to the statement.
     */
    public void setBoolean(PreparedStatement stmnt, int idx, boolean val, Column col)
        throws SQLException {
        booleanRepresentation.setBoolean(stmnt, idx, val);
    }

    /**
     * Set the given value as a parameter to the statement.
     */
    public void setByte(PreparedStatement stmnt, int idx, byte val, Column col)
        throws SQLException {
        stmnt.setByte(idx, val);
    }

    /**
     * Set the given value as a parameter to the statement.
     */
    public void setBytes(PreparedStatement stmnt, int idx, byte[] val,
        Column col)
        throws SQLException {
        if (useSetBytesForBlobs)
            stmnt.setBytes(idx, val);
        else
            setBinaryStream(stmnt, idx, new ByteArrayInputStream(val),
                val.length, col);
    }

    /**
     * Set the given value as a parameter to the statement.
     */
    public void setChar(PreparedStatement stmnt, int idx, char val, Column col)
        throws SQLException {
        if ((col != null && col.isCompatible(Types.INTEGER, null, 0, 0))
            || (col == null && storeCharsAsNumbers))
            setInt(stmnt, idx, (int) val, col);
        else
            setString(stmnt, idx, String.valueOf(val), col);
    }

    /**
     * Set the given value as a parameter to the statement.
     */
    public void setCharacterStream(PreparedStatement stmnt, int idx,
        Reader val, int length, Column col)
        throws SQLException {
        stmnt.setCharacterStream(idx, val, length);
    }

    /**
     * Set the given value as a parameter to the statement.
     */
    public void setClob(PreparedStatement stmnt, int idx, Clob val, Column col)
        throws SQLException {
        stmnt.setClob(idx, val);
    }

    /**
     * Set the given value as a parameter to the statement.
     */
    public void setClobString(PreparedStatement stmnt, int idx, String val,
        Column col)
        throws SQLException {
        if (useSetStringForClobs)
            stmnt.setString(idx, val);
        else {
            // set reader from string
            StringReader in = new StringReader(val);
            setCharacterStream(stmnt, idx, in, val.length(), col);
        }
    }

    /**
     * Set the given value as a parameter to the statement.
     */
    public void setDate(PreparedStatement stmnt, int idx, Date val, Column col)
        throws SQLException {
        if (col != null && col.getType() == Types.DATE)
            setDate(stmnt, idx, new java.sql.Date(val.getTime()), null, col);
        else if (col != null && col.getType() == Types.TIME)
            setTime(stmnt, idx, new Time(val.getTime()), null, col);
        else if (val instanceof Timestamp)
            setTimestamp(stmnt, idx,(Timestamp) val, null, col);
        else
            setTimestamp(stmnt, idx, new Timestamp(val.getTime()), null, col);
    }

    /**
     * Set the given value as a parameter to the statement.
     */
    public void setDate(PreparedStatement stmnt, int idx, java.sql.Date val, Calendar cal, Column col)
        throws SQLException {
        if (cal == null) {
            stmnt.setDate(idx, val);
        }
        else {
            stmnt.setDate(idx, val, cal);
        }
    }

    /**
     * Set the given value as a parameter to the statement.
     */
    public void setCalendar(PreparedStatement stmnt, int idx, Calendar val, Column col)
        throws SQLException {
        // by default we merely delegate to the Date parameter
        setDate(stmnt, idx, val.getTime(), col);
    }

    /**
     * Set the given LocalDate value as a parameter to the statement.
     *
     */
    public void setLocalDate(PreparedStatement stmnt, int idx, LocalDate val, Column col)
        throws SQLException {
        setDate(stmnt, idx, java.sql.Date.valueOf(val), null, col);
    }

    /**
     * Set the given LocalTime value as a parameter to the statement.
     *
     */
    public void setLocalTime(PreparedStatement stmnt, int idx, LocalTime val, Column col)
        throws SQLException {
        setTime(stmnt, idx, java.sql.Time.valueOf(val), null, col);
    }

    /**
     * Set the given LocalTime value as a parameter to the statement.
     *
     */
    public void setLocalDateTime(PreparedStatement stmnt, int idx, LocalDateTime val, Column col)
        throws SQLException {
        setTimestamp(stmnt, idx, java.sql.Timestamp.valueOf(val), null, col);
    }


    /**
     * Set the given LocalTime value as a parameter to the statement.
     *
     */
    public void setOffsetTime(PreparedStatement stmnt, int idx, OffsetTime val, Column col)
            throws SQLException {
        // adjust to the default timezone right now.
        // This is an ugly hack and cries for troubles in case the daylight saving changes...
        // Which is also the reason why we cannot cache the offset.
        // According to the Oracle docs the JDBC driver always assumes 'local time' ...
        LocalTime localTime = val.withOffsetSameInstant(OffsetDateTime.now().getOffset()).toLocalTime();
        setLocalTime(stmnt, idx, localTime, col);
    }

    /**
     * Set the given LocalTime value as a parameter to the statement.
     *
     */
    public void setOffsetDateTime(PreparedStatement stmnt, int idx, OffsetDateTime val, Column col)
            throws SQLException {
        LocalDateTime localdt = val.atZoneSameInstant(ZoneId.systemDefault()).toLocalDateTime();
        setLocalDateTime(stmnt, idx, localdt, col);
    }

    /**
     * Set the given value as a parameter to the statement.
     */
    public void setDouble(PreparedStatement stmnt, int idx, double val, Column col)
        throws SQLException {
        stmnt.setDouble(idx, val);
    }

    /**
     * Set the given value as a parameter to the statement.
     */
    public void setFloat(PreparedStatement stmnt, int idx, float val, Column col)
        throws SQLException {
        stmnt.setFloat(idx, val);
    }

    /**
     * Set the given value as a parameter to the statement.
     */
    public void setInt(PreparedStatement stmnt, int idx, int val, Column col)
        throws SQLException {
        stmnt.setInt(idx, val);
    }

    /**
     * Set the given value as a parameter to the statement.
     */
    public void setLong(PreparedStatement stmnt, int idx, long val, Column col)
        throws SQLException {
        stmnt.setLong(idx, val);
    }

    /**
     * Set the given value as a parameter to the statement.
     */
    public void setLocale(PreparedStatement stmnt, int idx, Locale val, Column col)
        throws SQLException {
        setString(stmnt, idx, val.getLanguage() + "_" + val.getCountry()
            + "_" + val.getVariant(), col);
    }

    /**
     * Set null as a parameter to the statement. The column
     * type will come from {@link Types}.
     */
    public void setNull(PreparedStatement stmnt, int idx, int colType, Column col)
        throws SQLException {
        stmnt.setNull(idx, colType);
    }

    /**
     * Set the given value as a parameter to the statement.
     */
    public void setNumber(PreparedStatement stmnt, int idx, Number num, Column col)
        throws SQLException {
        // check for known floating point types to give driver a chance to
        // handle special numbers like NaN and infinity; bug #1053
        if (num instanceof Double)
            setDouble(stmnt, idx, ((Double) num).doubleValue(), col);
        else if (num instanceof Float)
            setFloat(stmnt, idx, ((Float) num).floatValue(), col);
        else
            setBigDecimal(stmnt, idx, new BigDecimal(num.toString()), col);
    }

    /**
     * Set the given value as a parameter to the statement. The column
     * type will come from {@link Types}.
     */
    public void setObject(PreparedStatement stmnt, int idx, Object val, int colType, Column col)
        throws SQLException {
        if (colType == -1 || colType == Types.OTHER)
            stmnt.setObject(idx, val);
        else
            stmnt.setObject(idx, val, colType);
    }

    /**
     * Set the given value as a parameter to the statement.
     */
    public void setRef(PreparedStatement stmnt, int idx, Ref val, Column col)
        throws SQLException {
        stmnt.setRef(idx, val);
    }

    /**
     * Set the given value as a parameter to the statement.
     */
    public void setShort(PreparedStatement stmnt, int idx, short val, Column col)
        throws SQLException {
        stmnt.setShort(idx, val);
    }

    /**
     * Set the given value as a parameter to the statement.
     */
    public void setString(PreparedStatement stmnt, int idx, String val, Column col)
        throws SQLException {
        stmnt.setString(idx, val);
    }

    /**
     * Set the given value as a parameter to the statement.
     */
    public void setTime(PreparedStatement stmnt, int idx, Time val, Calendar cal, Column col)
        throws SQLException {
        if (cal == null)
            stmnt.setTime(idx, val);
        else
            stmnt.setTime(idx, val, cal);
    }

    /**
     * Set the given value as a parameter to the statement.
     */
    public void setTimestamp(PreparedStatement stmnt, int idx, Timestamp val, Calendar cal, Column col)
        throws SQLException {

        val = StateManagerImpl.roundTimestamp(val, datePrecision);

        if (cal == null)
            stmnt.setTimestamp(idx, val);
        else
            stmnt.setTimestamp(idx, val, cal);
    }

    /**
     * Set a column value into a prepared statement.
     *
     * @param stmnt the prepared statement to parameterize
     * @param idx the index of the parameter in the prepared statement
     * @param val the value of the column
     * @param col the column being set
     * @param type the field mapping type code for the value
     * @param store the store manager for the current context
     */
    public void setTyped(PreparedStatement stmnt, int idx, Object val, Column col, int type, JDBCStore store)
        throws SQLException {
        if (val == null) {
            setNull(stmnt, idx, (col == null) ? Types.OTHER : col.getType(),
                col);
            return;
        }

        Sized s;
        Calendard c;
        switch (type) {
            case JavaTypes.BOOLEAN:
            case JavaTypes.BOOLEAN_OBJ:
                setBoolean(stmnt, idx, ((Boolean) val).booleanValue(), col);
                break;
            case JavaTypes.BYTE:
            case JavaTypes.BYTE_OBJ:
                setByte(stmnt, idx, ((Number) val).byteValue(), col);
                break;
            case JavaTypes.CHAR:
            case JavaTypes.CHAR_OBJ:
                setChar(stmnt, idx, ((Character) val).charValue(), col);
                break;
            case JavaTypes.DOUBLE:
            case JavaTypes.DOUBLE_OBJ:
                setDouble(stmnt, idx, ((Number) val).doubleValue(), col);
                break;
            case JavaTypes.FLOAT:
            case JavaTypes.FLOAT_OBJ:
                setFloat(stmnt, idx, ((Number) val).floatValue(), col);
                break;
            case JavaTypes.INT:
            case JavaTypes.INT_OBJ:
                setInt(stmnt, idx, ((Number) val).intValue(), col);
                break;
            case JavaTypes.LONG:
            case JavaTypes.LONG_OBJ:
                setLong(stmnt, idx, ((Number) val).longValue(), col);
                break;
            case JavaTypes.SHORT:
            case JavaTypes.SHORT_OBJ:
                setShort(stmnt, idx, ((Number) val).shortValue(), col);
                break;
            case JavaTypes.STRING:
                if (col != null && (col.getType() == Types.CLOB
                    || col.getType() == Types.LONGVARCHAR))
                    setClobString(stmnt, idx, (String) val, col);
                else {
                    if (val instanceof String)
                        setString(stmnt, idx, (String) val, col);
                    else
                        setString(stmnt, idx, val.toString(), col);
                }
                break;
            case JavaTypes.OBJECT:
                setBlobObject(stmnt, idx, val, col, store);
                break;
            case JavaTypes.DATE:
                setDate(stmnt, idx, (Date) val, col);
                break;
            case JavaTypes.CALENDAR:
                setCalendar(stmnt, idx, (Calendar) val, col);
                break;
            case JavaTypes.LOCAL_DATE:
                setLocalDate(stmnt, idx, (LocalDate) val, col);
                break;
            case JavaTypes.LOCAL_TIME:
                setLocalTime(stmnt, idx, (LocalTime) val, col);
                break;
            case JavaTypes.LOCAL_DATETIME:
                setLocalDateTime(stmnt, idx, (LocalDateTime) val, col);
                break;
            case JavaTypes.OFFSET_TIME:
                setOffsetTime(stmnt, idx, (OffsetTime) val, col);
                break;
            case JavaTypes.OFFSET_DATETIME:
                setOffsetDateTime(stmnt, idx, (OffsetDateTime) val, col);
                break;
            case JavaTypes.BIGDECIMAL:
                setBigDecimal(stmnt, idx, (BigDecimal) val, col);
                break;
            case JavaTypes.BIGINTEGER:
                setBigInteger(stmnt, idx, (BigInteger) val, col);
                break;
            case JavaTypes.NUMBER:
                setNumber(stmnt, idx, (Number) val, col);
                break;
            case JavaTypes.LOCALE:
                setLocale(stmnt, idx, (Locale) val, col);
                break;
            case JavaSQLTypes.SQL_ARRAY:
                setArray(stmnt, idx, (Array) val, col);
                break;
            case JavaSQLTypes.ASCII_STREAM:
                s = (Sized) val;
                setAsciiStream(stmnt, idx, (InputStream) s.value, s.size, col);
                break;
            case JavaSQLTypes.BINARY_STREAM:
                s = (Sized) val;
                setBinaryStream(stmnt, idx, (InputStream) s.value, s.size, col);
                break;
            case JavaSQLTypes.BLOB:
                setBlob(stmnt, idx, (Blob) val, col);
                break;
            case JavaSQLTypes.BYTES:
                setBytes(stmnt, idx, (byte[]) val, col);
                break;
            case JavaSQLTypes.CHAR_STREAM:
                s = (Sized) val;
                setCharacterStream(stmnt, idx, (Reader) s.value, s.size, col);
                break;
            case JavaSQLTypes.CLOB:
                setClob(stmnt, idx, (Clob) val, col);
                break;
            case JavaSQLTypes.SQL_DATE:
                if (val instanceof Calendard) {
                    c = (Calendard) val;
                    setDate(stmnt, idx, (java.sql.Date) c.value, c.calendar,
                        col);
                } else
                    setDate(stmnt, idx, (java.sql.Date) val, null, col);
                break;
            case JavaSQLTypes.REF:
                setRef(stmnt, idx, (Ref) val, col);
                break;
            case JavaSQLTypes.TIME:
                if (val instanceof Calendard) {
                    c = (Calendard) val;
                    setTime(stmnt, idx, (Time) c.value, c.calendar, col);
                } else
                    setTime(stmnt, idx, (Time) val, null, col);
                break;
            case JavaSQLTypes.TIMESTAMP:
                if (val instanceof Calendard) {
                    c = (Calendard) val;
                    setTimestamp(stmnt, idx, (Timestamp) c.value, c.calendar,
                        col);
                } else
                    setTimestamp(stmnt, idx, (Timestamp) val, null, col);
                break;
            default:
                if (col != null && (col.getType() == Types.BLOB
                    || col.getType() == Types.VARBINARY))
                    setBlobObject(stmnt, idx, val, col, store);
                else
                    setObject(stmnt, idx, val, col.getType(), col);
        }
    }

    /**
     * Set a completely unknown parameter into a prepared statement.
     */
    public void setUnknown(PreparedStatement stmt, int idx, Object val, Column col) throws SQLException {
        if (val instanceof Object[]) {
            Object[] valArray = (Object[])val;
            for (Object object : valArray) {
                setUnknown(stmt, idx, col, object);
            }
        } else {
            setUnknown(stmt, idx, col, val);
        }
    }

    /**
     * Set a completely unknown parameter into a prepared statement.
     */
    public void setUnknown(PreparedStatement stmnt, int idx, Column col, Object val) throws SQLException {
        Sized sized = null;
        Calendard cald = null;
        if (val instanceof Sized) {
            sized = (Sized) val;
            val = sized.value;
        } else if (val instanceof Calendard) {
            cald = (Calendard) val;
            val = cald.value;
        }

        if (val == null)
            setNull(stmnt, idx, (col == null) ? Types.OTHER : col.getType(),
                col);
        else if (val instanceof String)
            setString(stmnt, idx, val.toString(), col);
        else if (val instanceof Integer)
            setInt(stmnt, idx, ((Integer) val).intValue(), col);
        else if (val instanceof Boolean)
            setBoolean(stmnt, idx, ((Boolean) val).booleanValue(), col);
        else if (val instanceof Long)
            setLong(stmnt, idx, ((Long) val).longValue(), col);
        else if (val instanceof Float)
            setFloat(stmnt, idx, ((Float) val).floatValue(), col);
        else if (val instanceof Double)
            setDouble(stmnt, idx, ((Double) val).doubleValue(), col);
        else if (val instanceof Byte)
            setByte(stmnt, idx, ((Byte) val).byteValue(), col);
        else if (val instanceof Character)
            setChar(stmnt, idx, ((Character) val).charValue(), col);
        else if (val instanceof Short)
            setShort(stmnt, idx, ((Short) val).shortValue(), col);
        else if (val instanceof Locale)
            setLocale(stmnt, idx, (Locale) val, col);
        else if (val instanceof BigDecimal)
            setBigDecimal(stmnt, idx, (BigDecimal) val, col);
        else if (val instanceof BigInteger)
            setBigInteger(stmnt, idx, (BigInteger) val, col);
        else if (val instanceof Array)
            setArray(stmnt, idx, (Array) val, col);
        else if (val instanceof Blob)
            setBlob(stmnt, idx, (Blob) val, col);
        else if (val instanceof byte[])
            setBytes(stmnt, idx, (byte[]) val, col);
        else if (val instanceof Clob)
            setClob(stmnt, idx, (Clob) val, col);
        else if (val instanceof Ref)
            setRef(stmnt, idx, (Ref) val, col);
        else if (val instanceof java.sql.Date)
            setDate(stmnt, idx, (java.sql.Date) val,
                (cald == null) ? null : cald.calendar, col);
        else if (val instanceof Timestamp)
            setTimestamp(stmnt, idx, (Timestamp) val,
                (cald == null) ? null : cald.calendar, col);
        else if (val instanceof Time)
            setTime(stmnt, idx, (Time) val,
                (cald == null) ? null : cald.calendar, col);
        else if (val instanceof Date)
            setDate(stmnt, idx, (Date) val, col);
        else if (val instanceof Calendar)
            setDate(stmnt, idx, ((Calendar) val).getTime(), col);
        else if (val instanceof LocalDate) {
            setLocalDate(stmnt, idx, (LocalDate) val, col);
        }
        else if (val instanceof LocalTime) {
            setLocalTime(stmnt, idx, (LocalTime) val, col);
        }
        else if (val instanceof LocalDateTime) {
            setLocalDateTime(stmnt, idx, (LocalDateTime) val, col);
        }
        else if (val instanceof Reader)
            setCharacterStream(stmnt, idx, (Reader) val,
                (sized == null) ? 0 : sized.size, col);
        else
            throw new UserException(_loc.get("bad-param", val.getClass()));
    }

    /**
     * Return the serialized bytes for the given object.
     */
    public byte[] serialize(Object val, JDBCStore store)
        throws SQLException {
        if (val == null)
            return null;
        if (val instanceof SerializedData)
            return ((SerializedData) val).bytes;
        return Serialization.serialize(val, store.getContext());
    }

    /**
     * Invoke the JDK 1.4 setBytes method on the given BLOB object.
     */
    public void putBytes(Blob blob, byte[] data)
        throws SQLException {
        blob.setBytes(1L, data);
    }

    /**
     * Invoke the JDK 1.4 setString method on the given CLOB
     * object.
     */
    public void putString(Clob clob, String data)
        throws SQLException {
        clob.setString(1L, data);
    }

    /**
     * Invoke the JDK 1.4 setCharacterStream method on the given
     * CLOB object.
     */
    public void putChars(Clob clob, char[] data)
        throws SQLException {
        Writer writer = clob.setCharacterStream(1L);
        try {
            writer.write(data);
            writer.flush();
        } catch (IOException ioe) {
            throw new SQLException(ioe.toString());
        }
    }

    /**
     * Warn that a particular value could not be stored precisely.
     * After the first warning for a particular type, messages
     * will be turned into trace messages.
     */
    protected void storageWarning(Object orig, Object converted) {
        boolean warn;
        synchronized (this) {
            if (_precisionWarnedTypes == null)
                _precisionWarnedTypes = new HashSet<>();
            warn = _precisionWarnedTypes.add(orig.getClass());
        }

        if (storageLimitationsFatal || (warn && log.isWarnEnabled())
            || (!warn && log.isTraceEnabled())) {
            Message msg = _loc.get("storage-restriction", new Object[]{
                platform,
                orig,
                orig.getClass().getName(),
                converted,
            });

            if (storageLimitationsFatal)
                throw new StoreException(msg);

            if (warn)
                log.warn(msg);
            else
                log.trace(msg);
        }
    }

    /////////
    // Types
    /////////

    /**
     * Return the preferred {@link Types} constant for the given
     * {@link JavaTypes} or {@link JavaSQLTypes} constant.
     */
    public int getJDBCType(int metaTypeCode, boolean lob) {
        return getJDBCType(metaTypeCode, lob, 0, 0);
    }

    /**
     * Return the preferred {@link Types} constant for the given
     * {@link JavaTypes} or {@link JavaSQLTypes} constant.
     */
    public int getJDBCType(int metaTypeCode, boolean lob, int precis,
        int scale, boolean xml) {
        return getJDBCType(metaTypeCode, lob, precis, scale);
    }

    /**
     * Return the preferred {@link Types} constant for the given
     * {@link JavaTypes} or {@link JavaSQLTypes} constant.
     */
    public int getJDBCType(int metaTypeCode, boolean lob, int precis,
        int scale) {
        if (lob) {
            switch (metaTypeCode) {
                case JavaTypes.STRING:
                case JavaSQLTypes.ASCII_STREAM:
                case JavaSQLTypes.CHAR_STREAM:
                    return getPreferredType(Types.CLOB);
                default:
                    return getPreferredType(Types.BLOB);
            }
        }

        switch (metaTypeCode) {
            case JavaTypes.BOOLEAN:
            case JavaTypes.BOOLEAN_OBJ:
                return getPreferredType(Types.BIT);
            case JavaTypes.BYTE:
            case JavaTypes.BYTE_OBJ:
                return getPreferredType(Types.TINYINT);
            case JavaTypes.CHAR:
            case JavaTypes.CHAR_OBJ:
                if (storeCharsAsNumbers)
                    return getPreferredType(Types.INTEGER);
                return getPreferredType(Types.CHAR);
            case JavaTypes.DOUBLE:
            case JavaTypes.DOUBLE_OBJ:
                if(precis > 0 || scale > 0) {
                    return getPreferredType(Types.NUMERIC);
                }
                else {
                    return getPreferredType(Types.DOUBLE);
                }
            case JavaTypes.FLOAT:
            case JavaTypes.FLOAT_OBJ:
                if(precis > 0 || scale > 0) {
                    return getPreferredType(Types.NUMERIC);
                }
                else {
                    return getPreferredType(Types.REAL);
                }
            case JavaTypes.INT:
            case JavaTypes.INT_OBJ:
                return getPreferredType(Types.INTEGER);
            case JavaTypes.LONG:
            case JavaTypes.LONG_OBJ:
                return getPreferredType(Types.BIGINT);
            case JavaTypes.SHORT:
            case JavaTypes.SHORT_OBJ:
                return getPreferredType(Types.SMALLINT);
            case JavaTypes.STRING:
            case JavaTypes.LOCALE:
            case JavaSQLTypes.ASCII_STREAM:
            case JavaSQLTypes.CHAR_STREAM:
                return getPreferredType(Types.VARCHAR);
            case JavaTypes.BIGINTEGER:
                if (storeLargeNumbersAsStrings)
                    return getPreferredType(Types.VARCHAR);
                return getPreferredType(Types.BIGINT);
            case JavaTypes.BIGDECIMAL:
                if (storeLargeNumbersAsStrings)
                    return getPreferredType(Types.VARCHAR);
                return getPreferredType(Types.NUMERIC);
            case JavaTypes.NUMBER:
                if (storeLargeNumbersAsStrings)
                    return getPreferredType(Types.VARCHAR);
                return getPreferredType(Types.NUMERIC);
            case JavaTypes.CALENDAR:
            case JavaTypes.DATE:
                return getPreferredType(Types.TIMESTAMP);
            case JavaTypes.LOCAL_DATE:
                return getPreferredType(Types.DATE);
            case JavaTypes.LOCAL_TIME:
                return getPreferredType(Types.TIME);
            case JavaTypes.LOCAL_DATETIME:
                return getPreferredType(Types.TIMESTAMP);
            case JavaTypes.OFFSET_TIME:
                return getPreferredType(Types.TIME_WITH_TIMEZONE);
            case JavaTypes.OFFSET_DATETIME:
                return getPreferredType(Types.TIMESTAMP_WITH_TIMEZONE);
            case JavaSQLTypes.SQL_ARRAY:
                return getPreferredType(Types.ARRAY);
            case JavaSQLTypes.BINARY_STREAM:
            case JavaSQLTypes.BLOB:
            case JavaSQLTypes.BYTES:
                return getPreferredType(Types.BLOB);
            case JavaSQLTypes.CLOB:
                return getPreferredType(Types.CLOB);
            case JavaSQLTypes.SQL_DATE:
                return getPreferredType(Types.DATE);
            case JavaSQLTypes.TIME:
                return getPreferredType(Types.TIME);
            case JavaSQLTypes.TIMESTAMP:
                return getPreferredType(Types.TIMESTAMP);
            default:
                return getPreferredType(Types.BLOB);
        }
    }

    /**
     * Return the preferred {@link Types} type for the given one. Returns
     * the given type by default.
     */
    public int getPreferredType(int type) {
        return type;
    }

    /**
     * Return the preferred database type name for the given column's type
     * from {@link Types}.
     */
    public String getTypeName(Column col) {
        if (!DBIdentifier.isEmpty(col.getTypeIdentifier()))
            return appendSize(col, toDBName(col.getTypeIdentifier()));

        if (col.isAutoAssigned() && autoAssignTypeName != null)
            return appendSize(col, autoAssignTypeName);

        return appendSize(col, getTypeName(col.getType()));
    }

    /**
     * Returns the type name for the specific constant as defined
     * by {@link java.sql.Types}.
     *
     * @param type the type
     * @return the name for the type
     */
    public String getTypeName(int type) {
        switch (type) {
            case Types.ARRAY:
                return arrayTypeName;
            case Types.BIGINT:
                return bigintTypeName;
            case Types.BINARY:
                return binaryTypeName;
            case Types.BIT:
                return bitTypeName;
            case Types.BLOB:
                return blobTypeName;
            case Types.BOOLEAN:
                return booleanTypeName;
            case Types.CHAR:
                return charTypeName;
            case Types.CLOB:
                return clobTypeName;
            case Types.DATE:
                return dateTypeName;
            case Types.DECIMAL:
                return decimalTypeName;
            case Types.DISTINCT:
                return distinctTypeName;
            case Types.DOUBLE:
                return doubleTypeName;
            case Types.FLOAT:
                return floatTypeName;
            case Types.INTEGER:
                return integerTypeName;
            case Types.JAVA_OBJECT:
                return javaObjectTypeName;
            case Types.LONGVARBINARY:
                return longVarbinaryTypeName;
            case Types.LONGVARCHAR:
                return longVarcharTypeName;
            case Types.NULL:
                return nullTypeName;
            case Types.NUMERIC:
                return numericTypeName;
            case Types.OTHER:
                return otherTypeName;
            case Types.REAL:
                return realTypeName;
            case Types.REF:
                return refTypeName;
            case Types.SMALLINT:
                return smallintTypeName;
            case Types.STRUCT:
                return structTypeName;
            case Types.TIME:
                return timeTypeName;
            case Types.TIMESTAMP:
                return timestampTypeName;
            case Types.TIME_WITH_TIMEZONE:
                return timeWithZoneTypeName;
            case Types.TIMESTAMP_WITH_TIMEZONE:
                return timestampWithZoneTypeName;
            case Types.TINYINT:
                return tinyintTypeName;
            case Types.VARBINARY:
                return varbinaryTypeName;
            case Types.VARCHAR:
                return varcharTypeName;
            default:
                return otherTypeName;
        }
    }

    /**
     * Helper method to add size properties to the specified type.
     * If present, the string "{0}" will be replaced with the size definition;
     * otherwise the size definition will be appended to the type name.
     * If your database has column types that don't allow size definitions,
     * override this method to return the unaltered type name for columns of
     * those types (or add the type names to the
     * fixedSizeTypeNameSet).
     *
     * 

Some databases support "type modifiers", for example the unsigned * "modifier" in MySQL. In these cases the size should go between the type * and the "modifier", instead of after the modifier. For example * CREATE table FOO ( myint INT (10) UNSIGNED . . .) instead of * CREATE table FOO ( myint INT UNSIGNED (10) . . .). * Type modifiers should be added to typeModifierSet in * subclasses. */ protected String appendSize(Column col, String typeName) { String upperCaseTypeName = typeName.toUpperCase(Locale.ENGLISH); if (fixedSizeTypeNameSet.contains(upperCaseTypeName)) { return typeName; } if (typeName.indexOf('(') != -1) { return typeName; } String size = null; int colSize = col.getSize(); if (colSize<=0 && fractionalTypeNameSet.contains(upperCaseTypeName)){ // special handling for types with fractions // Attention! We abuse @Column(scale=n) // One can disable all fractions with @Column(scale=-1) if (col.getDecimalDigits() != 0) { // the default colSize = col.getDecimalDigits() == -1 ? 0 : col.getDecimalDigits(); } else { colSize = getDateFractionDigits(col, typeName); } size = "(" + colSize + ")"; } else if (colSize > 0) { StringBuilder buf = new StringBuilder(10); buf.append("(").append(colSize); if (col.getDecimalDigits() > 0) buf.append(", ").append(col.getDecimalDigits()); buf.append(")"); size = buf.toString(); } return insertSize(typeName, size); } /** * Only get's used if no explicit scale is used (@{code @Column(scale=0)} which is the default). * Attention! @{code @Column(scale=-1)} disables the scale of a column * @return the fraction length of types which have a fraction * @see #fractionalTypeNameSet */ protected int getDateFractionDigits(Column col, String typeName) { return dateFractionDigits; } /** * Helper method that inserts a size clause for a given SQL type. * * @param typeName The SQL type e.g. INT * @param size The size clause e.g. (10) * @return The typeName + size clause. Usually the size clause will * be appended to typeName. If the typeName contains a * marker : {0} or if typeName contains a modifier the * size clause will be inserted appropriately. */ protected String insertSize(String typeName, String size) { if (StringUtil.isEmpty(size)) { int idx = typeName.indexOf("{0}"); if (idx != -1) { return typeName.substring(0, idx); } return typeName; } int idx = typeName.indexOf("{0}"); if (idx != -1) { // replace '{0}' with size String ret = typeName.substring(0, idx); if (size != null) ret = ret + size; if (typeName.length() > idx + 3) ret = ret + typeName.substring(idx + 3); return ret; } if (!typeModifierSet.isEmpty()) { String s; idx = typeName.length(); int curIdx = -1; for (Iterator i = typeModifierSet.iterator(); i.hasNext();) { s = i.next(); if (typeName.toUpperCase(Locale.ENGLISH).indexOf(s) != -1) { curIdx = typeName.toUpperCase(Locale.ENGLISH).indexOf(s); if (curIdx != -1 && curIdx < idx) { idx = curIdx; } } } if(idx != typeName.length()) { String ret = typeName.substring(0, idx); ret = ret + size; ret = ret + ' ' + typeName.substring(idx); return ret; } } return typeName + size; } /////////// // Selects /////////// /** * Set the name of the join syntax to use: sql92, traditional, database. */ public void setJoinSyntax(String syntax) { if ("sql92".equals(syntax)) joinSyntax = SYNTAX_SQL92; else if ("traditional".equals(syntax)) joinSyntax = SYNTAX_TRADITIONAL; else if ("database".equals(syntax)) joinSyntax = SYNTAX_DATABASE; else if (!StringUtil.isEmpty(syntax)) throw new IllegalArgumentException(syntax); } public boolean isImplicitJoin() { return false; } /** * Return a SQL string to act as a placeholder for the given column. */ public String getPlaceholderValueString(Column col) { switch (col.getType()) { case Types.BIGINT: case Types.BIT: case Types.INTEGER: case Types.NUMERIC: case Types.SMALLINT: case Types.TINYINT: return "0"; case Types.CHAR: return (storeCharsAsNumbers) ? "0" : "' '"; case Types.CLOB: case Types.LONGVARCHAR: case Types.VARCHAR: return "''"; case Types.DATE: return ZERO_DATE_STR; case Types.DECIMAL: case Types.DOUBLE: case Types.FLOAT: case Types.REAL: return "0.0"; case Types.TIME: return ZERO_TIME_STR; case Types.TIMESTAMP: return ZERO_TIMESTAMP_STR; default: return "NULL"; } } /** * Create a SELECT COUNT statement in the proper join syntax for the * given instance. */ public SQLBuffer toSelectCount(Select sel) { SQLBuffer selectSQL = new SQLBuffer(this); SQLBuffer from; sel.addJoinClassConditions(); if (sel.getFromSelect() != null) from = getFromSelect(sel, false); else from = getFrom(sel, false); SQLBuffer where = getWhere(sel, false); // if no grouping and no range, we might be able to get by without // a subselect if (sel.getGrouping() == null && sel.getStartIndex() == 0 && sel.getEndIndex() == Long.MAX_VALUE) { // if the select has no identifier cols, use COUNT(*) List aliases = (!sel.isDistinct()) ? Collections.EMPTY_LIST : sel.getIdentifierAliases(); if (useWildCardForCount || aliases.isEmpty()) { selectSQL.append("COUNT(*)"); return toSelect(selectSQL, null, from, where, null, null, null, false, false, 0, Long.MAX_VALUE); } // if there is a single distinct col, use COUNT(DISTINCT col) if (aliases.size() == 1) { selectSQL.append("COUNT(DISTINCT "). append(aliases.get(0).toString()).append(")"); return toSelect(selectSQL, null, from, where, null, null, null, false, false, 0, Long.MAX_VALUE); } // can we combine distinct cols? if (distinctCountColumnSeparator != null) { selectSQL.append("COUNT(DISTINCT "); for (int i = 0; i < aliases.size(); i++) { if (i > 0) { selectSQL.append(" "); selectSQL.append(distinctCountColumnSeparator); selectSQL.append(" "); } selectSQL.append(aliases.get(i).toString()); } selectSQL.append(")"); return toSelect(selectSQL, null, from, where, null, null, null, false, false, 0, Long.MAX_VALUE); } } // since we can't combine distinct cols, we have to perform an outer // COUNT(*) select using the original select as a subselect in the // FROM clause assertSupport(supportsSubselect, "SupportsSubselect"); SQLBuffer subSelect = getSelects(sel, true, false); SQLBuffer subFrom = from; from = new SQLBuffer(this); from.append("("); from.append(toSelect(subSelect, null, subFrom, where, sel.getGrouping(), sel.getHaving(), null, sel.isDistinct(), false, sel.getStartIndex(), sel.getEndIndex(), true, sel)); from.append(")"); if (requiresAliasForSubselect) from.append(" ").append(Select.FROM_SELECT_ALIAS); selectSQL.append("COUNT(*)"); return toSelect(selectSQL, null, from, null, null, null, null, false, false, 0, Long.MAX_VALUE); } /** * Create a DELETE statement for the specified Select. If the * database does not support the bulk delete statement (such as * cases where a subselect is required and the database doesn't support * subselects), this method should return null. */ public SQLBuffer toDelete(ClassMapping mapping, Select sel, Object[] params) { return toBulkOperation(mapping, sel, null, params, null); } public SQLBuffer toUpdate(ClassMapping mapping, Select sel, JDBCStore store, Object[] params, Map updates) { return toBulkOperation(mapping, sel, store, params, updates); } /** * Returns the SQL for a bulk operation, either a DELETE or an UPDATE. * * @param mapping the mapping against which we are operating * @param sel the Select that will constitute the WHERE clause * @param store the current store * @param updateParams the Map that holds the update parameters; a null * value indicates that this is a delete operation * @return the SQLBuffer for the update, or null if it is not * possible to perform the bulk update */ protected SQLBuffer toBulkOperation(ClassMapping mapping, Select sel, JDBCStore store, Object[] params, Map updateParams) { SQLBuffer sql = new SQLBuffer(this); if (updateParams == null) { if (requiresTargetForDelete) { sql.append("DELETE "); SQLBuffer deleteTargets = getDeleteTargets(sel); sql.append(deleteTargets); sql.append(" FROM "); } else { sql.append("DELETE FROM "); } } else sql.append("UPDATE "); sel.addJoinClassConditions(); // if there is only a single table in the select, then we can // just issue a single DELETE FROM TABLE WHERE // statement; otherwise, since SQL doesn't allow deleting // from one of a multi-table select, we need to issue a subselect // like DELETE FROM TABLE WHERE EXISTS // (SELECT 1 FROM TABLE t0 WHERE t0.ID = TABLE.ID); also, some // databases do not allow aliases in delete statements, which // also causes us to use a subselect Collection selectedTables = getSelectTableAliases(sel); if (selectedTables.size() == 1 && supportsSubselect && allowsAliasInBulkClause) { SQLBuffer from; if (sel.getFromSelect() != null) from = getFromSelect(sel, false); else from = getFrom(sel, false); sql.append(from); appendUpdates(sel, store, sql, params, updateParams, allowsAliasInBulkClause); SQLBuffer where = sel.getWhere(); if (where != null && !where.isEmpty()) { sql.append(" WHERE "); sql.append(where); } return sql; } Table table = mapping.getTable(); String tableName = getFullName(table, false); // only use a subselect if the where is not empty; otherwise // an unqualified delete or update will work if (sel.getWhere() == null || sel.getWhere().isEmpty()) { sql.append(tableName); appendUpdates(sel, store, sql, params, updateParams, false); return sql; } // we need to use a subselect if we are to bulk delete where // the select includes multiple tables; if the database // doesn't support it, then we need to signal this by returning null if (!supportsSubselect || !supportsCorrelatedSubselect) return null; Column[] pks = mapping.getPrimaryKeyColumns(); sel.clearSelects(); sel.setDistinct(true); // if we have only a single PK, we can use a non-correlated // subquery (using an IN statement), which is much faster than // a correlated subquery (since a correlated subquery needs // to be executed once for each row in the table) if (pks.length == 1) { sel.select(pks[0]); sql.append(tableName); appendUpdates(sel, store, sql, params, updateParams, false); sql.append(" WHERE "). append(pks[0]).append(" IN ("). append(sel.toSelect(false, null)).append(")"); } else { sel.clearSelects(); sel.setDistinct(false); // since the select is using a correlated subquery, we // only need to select a bogus virtual column sel.select("1", null); // add in the joins to the table Column[] cols = table.getPrimaryKey().getColumns(); SQLBuffer buf = new SQLBuffer(this); buf.append("("); for (int i = 0; i < cols.length; i++) { if (i > 0) buf.append(" AND "); // add in "t0.PK = MYTABLE.PK" buf.append(sel.getColumnAlias(cols[i])).append(" = "). append(table).append(catalogSeparator).append(cols[i]); } buf.append(")"); sel.where(buf, null); sql.append(tableName); appendUpdates(sel, store, sql, params, updateParams, false); sql.append(" WHERE EXISTS ("). append(sel.toSelect(false, null)).append(")"); } return sql; } protected Collection getSelectTableAliases(Select sel) { return sel.getTableAliases(); } protected SQLBuffer getDeleteTargets(Select sel) { SQLBuffer deleteTargets = new SQLBuffer(this); Collection aliases = sel.getTableAliases(); // Assumes aliases are of the form "TABLENAME t0" // or "\"TABLE NAME\" t0" for (Iterator itr = aliases.iterator(); itr.hasNext();) { String tableAlias = itr.next(); String[] names = Normalizer.splitName(tableAlias, IdentifierUtil.SPACE); if (names.length > 1) { if (allowsAliasInBulkClause) { deleteTargets.append(names[1]); } else { deleteTargets.append(toDBName(DBIdentifier.newTable(names[0]))); } } else { deleteTargets.append(toDBName(DBIdentifier.newTable(tableAlias))); } if (itr.hasNext()) deleteTargets.append(", "); } return deleteTargets; } protected void appendUpdates(Select sel, JDBCStore store, SQLBuffer sql, Object[] params, Map updateParams, boolean allowAlias) { if (updateParams == null || updateParams.size() == 0) return; // manually build up the SET clause for the UPDATE statement sql.append(" SET "); ExpContext ctx = new ExpContext(store, params, store.getFetchConfiguration()); // If the updates map contains any version fields, assume that the // optimistic lock version data is being handled properly by the // caller. Otherwise, give the version indicator an opportunity to // add more update clauses as needed. boolean augmentUpdates = true; for (Iterator i = updateParams.entrySet().iterator(); i.hasNext();) { Map.Entry next = (Map.Entry) i.next(); Path path = (Path) next.getKey(); FieldMapping fmd = (FieldMapping) path.last(); if (fmd.isVersion()) augmentUpdates = false; Val val = (Val) next.getValue(); if (val == null) val = new Null(); Column col = fmd.getColumns()[0]; if (allowAlias) { sql.append(sel.getColumnAlias(col)); } else { sql.append(toDBName(col.getIdentifier())); } sql.append(" = "); ExpState state = val.initialize(sel, ctx, 0); // JDBC Paths are always PCPaths; PCPath implements Val ExpState pathState = ((Val) path).initialize(sel, ctx, 0); calculateValue(val, sel, ctx, state, path, pathState); // append the value with a null for the Select; i // indicates that the int length = val.length(sel, ctx, state); for (int j = 0; j < length; j++) val.appendTo((allowAlias) ? sel : null, ctx, state, sql, j); if (i.hasNext()) sql.append(", "); } if (augmentUpdates) { Path path = (Path) updateParams.keySet().iterator().next(); FieldMapping fm = (FieldMapping) path.last(); ClassMapping meta = fm.getDefiningMapping(); Map updates = meta.getVersion().getBulkUpdateValues(); for (Map.Entry e : updates.entrySet()) { Column col = (Column) e.getKey(); Object val = e.getValue(); sql.append(", ").append(toDBName(col.getIdentifier())).append(" = "); // Version update value for Numeric version is encoded in a String // to make SQL such as version = version+1 while Time stamp version is parameterized if (val instanceof String) { sql.append((String)val); } else { sql.appendValue(val); } } } } /** * Create SQL to delete the contents of the specified tables. * The default implementation drops all non-deferred RESTRICT foreign key * constraints involving the specified tables, issues DELETE statements * against the tables, and then adds the dropped constraints back in. * Databases with more optimal ways of deleting the contents of several * tables should override this method. */ public String[] getDeleteTableContentsSQL(Table[] tables,Connection conn) { Collection sql = new ArrayList<>(); // collect and drop non-deferred physical restrict constraints, and // collect the DELETE FROM statements Collection deleteSQL = new ArrayList<>(tables.length); Collection restrictConstraints = new LinkedHashSet<>(); for (int i = 0; i < tables.length; i++) { ForeignKey[] fks = tables[i].getForeignKeys(); for (int j = 0; j < fks.length; j++) { if (!fks[j].isLogical() && !fks[j].isDeferred() && fks[j].getDeleteAction() == ForeignKey.ACTION_RESTRICT) restrictConstraints.add(fks[j]); } deleteSQL.add("DELETE FROM " + toDBName(tables[i].getFullIdentifier())); } for(ForeignKey fk : restrictConstraints) { String[] constraintSQL = getDropForeignKeySQL(fk,conn); sql.addAll(Arrays.asList(constraintSQL)); } // add the delete statements after all the constraint mutations sql.addAll(deleteSQL); // add the deleted constraints back to the schema for (ForeignKey fk : restrictConstraints) { String[] constraintSQL = getAddForeignKeySQL(fk); sql.addAll(Arrays.asList(constraintSQL)); } return (String[]) sql.toArray(new String[sql.size()]); } /** * Create a SELECT statement in the proper join syntax for the given * instance. */ public SQLBuffer toSelect(Select sel, boolean forUpdate, JDBCFetchConfiguration fetch) { sel.addJoinClassConditions(); boolean update = forUpdate && sel.getFromSelect() == null; SQLBuffer select = getSelects(sel, false, update); SQLBuffer ordering = null; if (!sel.isAggregate() || sel.getGrouping() != null) ordering = sel.getOrdering(); SQLBuffer from; if (sel.getFromSelect() != null) from = getFromSelect(sel, forUpdate); else from = getFrom(sel, update); SQLBuffer where = getWhere(sel, update); return toSelect(select, fetch, from, where, sel.getGrouping(), sel.getHaving(), ordering, sel.isDistinct(), forUpdate, sel.getStartIndex(), sel.getEndIndex(), sel); } /** * Return the portion of the select statement between the FROM keyword * and the WHERE keyword. */ protected SQLBuffer getFrom(Select sel, boolean forUpdate) { SQLBuffer fromSQL = new SQLBuffer(this); Collection aliases = sel.getTableAliases(); if (aliases.size() < 2 || sel.getJoinSyntax() != SYNTAX_SQL92) { for (Iterator itr = aliases.iterator(); itr.hasNext();) { fromSQL.append(itr.next().toString()); if (forUpdate && tableForUpdateClause != null) fromSQL.append(" ").append(tableForUpdateClause); if (itr.hasNext()) fromSQL.append(", "); } if (aliases.size() < 2 && sel.getParent() != null) { // subquery may contain correlated joins Iterator itr = sel.getJoinIterator(); while (itr.hasNext()) { Join join = (Join) itr.next(); // append where clause if (join.isCorrelated() && join.getForeignKey() != null) { SQLBuffer where = new SQLBuffer(this); where.append("(").append(toTraditionalJoin(join)).append(")"); sel.where(where.getSQL()); } } } } else { Iterator itr = sel.getJoinIterator(); boolean first = true; while (itr.hasNext()) { Join join = (Join) itr.next(); if (correlatedJoinCondition(join, sel)) continue; if (join.isCorrelated()) toCorrelatedJoin(sel, join, forUpdate, first); else fromSQL.append(toSQL92Join(sel, join, forUpdate, first)); first = false; if (itr.hasNext() && join.isCorrelated()) { if (fromSQL.getSQL().length() > 0) fromSQL.append(", "); first = true; } } for (Iterator itr2 = aliases.iterator(); itr2.hasNext();) { String tableAlias = itr2.next().toString(); if (fromSQL.getSQL().indexOf(tableAlias) == -1) { if (!first && fromSQL.getSQL().length() > 0) fromSQL.append(", "); fromSQL.append(tableAlias); if (forUpdate && tableForUpdateClause != null) fromSQL.append(" ").append(tableForUpdateClause); first = false; } } } return fromSQL; } private boolean correlatedJoinCondition(Join join, Select sel) { if (!join.isCorrelated()) return false; Iterator itr = sel.getJoinIterator(); boolean skip = false; //if table1 in join is in the main query, table2 is in //subquery, and table2 participates in other joins //in subquery, the join condition can only be placed in //the where clause in the subquery while (itr.hasNext()) { Join join1 = (Join) itr.next(); if (join == join1 && !join.isForeignKeyInversed()) { continue; } if (join.getIndex2() == join1.getIndex1() || join.getIndex2() == join1.getIndex2()) { skip = true; if (join.getForeignKey() != null){ SQLBuffer where = new SQLBuffer(this); where.append("(").append(toTraditionalJoin(join)).append(")"); sel.where(where.getSQL()); } break; } } return skip; } /** * Return the FROM clause for a select that selects from a tmp table * created by an inner select. */ protected SQLBuffer getFromSelect(Select sel, boolean forUpdate) { SQLBuffer fromSQL = new SQLBuffer(this); fromSQL.append("("); fromSQL.append(toSelect(sel.getFromSelect(), forUpdate, null)); fromSQL.append(")"); if (requiresAliasForSubselect) fromSQL.append(" ").append(Select.FROM_SELECT_ALIAS); return fromSQL; } /** * Return the WHERE portion of the select statement, or null if no where * conditions. */ protected SQLBuffer getWhere(Select sel, boolean forUpdate) { Joins joins = sel.getJoins(); if (sel.getJoinSyntax() == SYNTAX_SQL92 || joins == null || joins.isEmpty()) return sel.getWhere(); SQLBuffer where = new SQLBuffer(this); if (sel.getWhere() != null) where.append(sel.getWhere()); if (joins != null) sel.append(where, joins); return where; } /** * Use the given join instance to create SQL joining its tables in * the traditional style. */ public SQLBuffer toTraditionalJoin(Join join) { ForeignKey fk = join.getForeignKey(); if (fk == null) return null; boolean inverse = join.isForeignKeyInversed(); Column[] from = (inverse) ? fk.getPrimaryKeyColumns() : fk.getColumns(); Column[] to = (inverse) ? fk.getColumns() : fk.getPrimaryKeyColumns(); // do column joins SQLBuffer buf = new SQLBuffer(this); int count = 0; for (int i = 0; i < from.length; i++, count++) { if (count > 0) buf.append(" AND "); buf.append(join.getAlias1()).append(".").append(from[i]); buf.append(" = "); buf.append(join.getAlias2()).append(".").append(to[i]); } // do constant joins Column[] constCols = fk.getConstantColumns(); for (int i = 0; i < constCols.length; i++, count++) { if (count > 0) buf.append(" AND "); if (inverse) buf.appendValue(fk.getConstant(constCols[i]), constCols[i]); else buf.append(join.getAlias1()).append("."). append(constCols[i]); buf.append(" = "); if (inverse) buf.append(join.getAlias2()).append("."). append(constCols[i]); else buf.appendValue(fk.getConstant(constCols[i]), constCols[i]); } Column[] constColsPK = fk.getConstantPrimaryKeyColumns(); for (int i = 0; i < constColsPK.length; i++, count++) { if (count > 0) buf.append(" AND "); if (inverse) buf.append(join.getAlias1()).append("."). append(constColsPK[i]); else buf.appendValue(fk.getPrimaryKeyConstant(constColsPK[i]), constColsPK[i]); buf.append(" = "); if (inverse) buf.appendValue(fk.getPrimaryKeyConstant(constColsPK[i]), constColsPK[i]); else buf.append(join.getAlias2()).append("."). append(constColsPK[i]); } return buf; } /** * Use the given join instance to create SQL joining its tables in * the SQL92 style. */ public SQLBuffer toSQL92Join(Select sel, Join join, boolean forUpdate, boolean first) { SQLBuffer buf = new SQLBuffer(this); if (first) { buf.append(join.getTable1()).append(" "). append(join.getAlias1()); if (forUpdate && tableForUpdateClause != null) buf.append(" ").append(tableForUpdateClause); } buf.append(" "); if (join.getType() == Join.TYPE_OUTER) buf.append(outerJoinClause); else if (join.getType() == Join.TYPE_INNER) buf.append(innerJoinClause); else // cross buf.append(crossJoinClause); buf.append(" "); buf.append(join.getTable2()).append(" ").append(join.getAlias2()); if (forUpdate && tableForUpdateClause != null) buf.append(" ").append(tableForUpdateClause); if (join.getForeignKey() != null) buf.append(" ON ").append(toTraditionalJoin(join)); else if (requiresConditionForCrossJoin && join.getType() == Join.TYPE_CROSS) buf.append(" ON (1 = 1)"); return buf; } private SQLBuffer toCorrelatedJoin(Select sel, Join join, boolean forUpdate, boolean first) { if (join.getForeignKey() != null){ SQLBuffer where = new SQLBuffer(this); where.append("(").append(toTraditionalJoin(join)).append(")"); sel.where(where.getSQL()); } return null; } /** * Use the given join instance to create SQL joining its tables in * the database's native syntax. Throws an exception by default. */ public SQLBuffer toNativeJoin(Join join) { throw new UnsupportedException(); } /** * Returns if the given foreign key can be eagerly loaded using other joins. */ public boolean canOuterJoin(int syntax, ForeignKey fk) { return syntax != SYNTAX_TRADITIONAL; } /** * Combine the given components into a SELECT statement. */ public SQLBuffer toSelect(SQLBuffer selects, JDBCFetchConfiguration fetch, SQLBuffer from, SQLBuffer where, SQLBuffer group, SQLBuffer having, SQLBuffer order, boolean distinct, boolean forUpdate, long start, long end) { return toOperation(getSelectOperation(fetch), selects, from, where, group, having, order, distinct, start, end, getForUpdateClause(fetch, forUpdate, null)); } /** * Combine the given components into a SELECT statement. */ protected SQLBuffer toSelect(SQLBuffer selects, JDBCFetchConfiguration fetch, SQLBuffer from, SQLBuffer where, SQLBuffer group, SQLBuffer having, SQLBuffer order, boolean distinct, boolean forUpdate, long start, long end, boolean subselect, Select sel) { return toOperation(getSelectOperation(fetch), selects, from, where, group, having, order, distinct, start, end, getForUpdateClause(fetch, forUpdate, null), subselect); } public SQLBuffer toSelect(SQLBuffer selects, JDBCFetchConfiguration fetch, SQLBuffer from, SQLBuffer where, SQLBuffer group, SQLBuffer having, SQLBuffer order, boolean distinct, boolean forUpdate, long start, long end, boolean subselect, boolean checkTableForUpdate) { return toOperation(getSelectOperation(fetch), selects, from, where, group, having, order, distinct, start, end, getForUpdateClause(fetch, forUpdate, null), subselect, checkTableForUpdate); } /** * Combine the given components into a SELECT statement. */ protected SQLBuffer toSelect(SQLBuffer selects, JDBCFetchConfiguration fetch, SQLBuffer from, SQLBuffer where, SQLBuffer group, SQLBuffer having, SQLBuffer order, boolean distinct, boolean forUpdate, long start, long end, Select sel) { return toOperation(getSelectOperation(fetch), selects, from, where, group, having, order, distinct, start, end, getForUpdateClause(fetch, forUpdate, sel)); } /** * Get the update clause for the query based on the * updateClause and isolationLevel hints */ protected String getForUpdateClause(JDBCFetchConfiguration fetch, boolean isForUpdate, Select sel) { if (fetch != null && fetch.getIsolation() != -1) { throw new InvalidStateException(_loc.get( "isolation-level-config-not-supported", getClass().getName())); } else if (isForUpdate && !simulateLocking) { assertSupport(supportsSelectForUpdate, "SupportsSelectForUpdate"); return forUpdateClause; } else { return null; } } /** * Return true if the dictionary uses isolation level to compute the * returned getForUpdateClause() SQL clause. */ public boolean supportsIsolationForUpdate() { return false; } /** * Return the "SELECT" operation clause, adding any available hints, etc. */ public String getSelectOperation(JDBCFetchConfiguration fetch) { return "SELECT"; } /** * Return the SQL for the given selecting operation. */ public SQLBuffer toOperation(String op, SQLBuffer selects, SQLBuffer from, SQLBuffer where, SQLBuffer group, SQLBuffer having, SQLBuffer order, boolean distinct, long start, long end, String forUpdateClause) { return toOperation(op, selects, from, where, group, having, order, distinct, start, end, forUpdateClause, false); } /** * Return the SQL for the given selecting operation. */ public SQLBuffer toOperation(String op, SQLBuffer selects, SQLBuffer from, SQLBuffer where, SQLBuffer group, SQLBuffer having, SQLBuffer order, boolean distinct, long start, long end, String forUpdateClause, boolean subselect) { return toOperation(op, selects, from, where, group, having, order, distinct, start, end, forUpdateClause, subselect, false); } /** * Return the SQL for the given selecting operation. */ private SQLBuffer toOperation(String op, SQLBuffer selects, SQLBuffer from, SQLBuffer where, SQLBuffer group, SQLBuffer having, SQLBuffer order, boolean distinct, long start, long end, String forUpdateClause, boolean subselect, boolean checkTableForUpdate) { SQLBuffer buf = new SQLBuffer(this); buf.append(op); boolean range = start != 0 || end != Long.MAX_VALUE; if (range && rangePosition == RANGE_PRE_DISTINCT) appendSelectRange(buf, start, end, subselect); if (distinct) buf.append(" DISTINCT"); if (range && rangePosition == RANGE_POST_DISTINCT) appendSelectRange(buf, start, end, subselect); buf.append(" ").append(selects).append(" FROM ").append(from); if (checkTableForUpdate && (StringUtil.isEmpty(forUpdateClause) && !StringUtil.isEmpty(tableForUpdateClause))) { buf.append(" ").append(tableForUpdateClause); } if (where != null && !where.isEmpty()) buf.append(" WHERE ").append(where); if (group != null && !group.isEmpty()) buf.append(" GROUP BY ").append(group); if (having != null && !having.isEmpty()) { assertSupport(supportsHaving, "SupportsHaving"); buf.append(" HAVING ").append(having); } if (order != null && !order.isEmpty()) buf.append(" ORDER BY ").append(order); if (range && rangePosition == RANGE_POST_SELECT) appendSelectRange(buf, start, end, subselect); if (forUpdateClause != null) buf.append(" ").append(forUpdateClause); if (range && rangePosition == RANGE_POST_LOCK) appendSelectRange(buf, start, end, subselect); return buf; } /** * If this dictionary can select ranges, * use this method to append the range SQL. */ protected void appendSelectRange(SQLBuffer buf, long start, long end, boolean subselect) { } /** * Return the portion of the select statement between the SELECT keyword * and the FROM keyword. */ protected SQLBuffer getSelects(Select sel, boolean distinctIdentifiers, boolean forUpdate) { // append the aliases for all the columns SQLBuffer selectSQL = new SQLBuffer(this); List aliases; if (distinctIdentifiers) aliases = sel.getIdentifierAliases(); else aliases = sel.getSelectAliases(); Object alias; for (int i = 0; i < aliases.size(); i++) { alias = aliases.get(i); if (alias instanceof String) { alias = getNamingUtil().convertAlias((String)alias); } appendSelect(selectSQL, alias, sel, i); if (i < aliases.size() - 1) selectSQL.append(", "); } return selectSQL; } /** * Append elem to selectSQL. * @param selectSQL The SQLBuffer to append to. * @param elem A {@link SQLBuffer} or a {@link String} to append. * * @since 1.1.0 */ protected void appendSelect(SQLBuffer selectSQL, Object elem, Select sel, int idx) { if (elem instanceof SQLBuffer) selectSQL.append((SQLBuffer) elem); else selectSQL.append(elem.toString()); } /** * Returns true if a "FOR UPDATE" clause can be used for the specified * Select object. */ public boolean supportsLocking(Select sel) { if (sel.isAggregate()) return false; if (!supportsSelectForUpdate) return false; if (!supportsLockingWithSelectRange && (sel.getStartIndex() != 0 || sel.getEndIndex() != Long.MAX_VALUE)) return false; // only inner select is locked if (sel.getFromSelect() != null) sel = sel.getFromSelect(); if (!supportsLockingWithDistinctClause && sel.isDistinct()) return false; if (!supportsLockingWithMultipleTables && sel.getTableAliases().size() > 1) return false; if (!supportsLockingWithOrderClause && sel.getOrdering() != null) return false; if (!supportsLockingWithOuterJoin || !supportsLockingWithInnerJoin) { for (Iterator itr = sel.getJoinIterator(); itr.hasNext();) { Join join = (Join) itr.next(); if (!supportsLockingWithOuterJoin && join.getType() == Join.TYPE_OUTER) return false; if (!supportsLockingWithInnerJoin && join.getType() == Join.TYPE_INNER) return false; } } return true; } /** * Return false if the given select requires a forward-only result set. */ public boolean supportsRandomAccessResultSet(Select sel, boolean forUpdate) { return !sel.isAggregate(); } /** * Assert that the given dictionary flag is true. If it is not true, * throw an error saying that the given setting needs to return true for * the current operation to work. */ public void assertSupport(boolean feature, String property) { if (!feature) throw new UnsupportedException(_loc.get("feature-not-supported", getClass(), property)); } //////////////////// // Query functions //////////////////// /** * Invoke this database's substring function. * Numeric parameters are inlined if possible. This is to handle grouping by SUBSTRING - * most databases do not allow parameter binding in this case. * * @param buf the SQL buffer to write the substring invocation to * @param str a query value representing the target string * @param start a query value representing the start index * @param length a query value representing the length of substring, or null for none */ public void substring(SQLBuffer buf, FilterValue str, FilterValue start, FilterValue length) { buf.append(substringFunctionName).append("("); str.appendTo(buf); buf.append(", "); if (start.getValue() instanceof Number) { buf.append(Long.toString(toLong(start))); } else { start.appendTo(buf); } if (length != null) { buf.append(", "); if (length.getValue() instanceof Number) { buf.append(Long.toString(toLong(length))); } else { length.appendTo(buf); } } buf.append(")"); } long toLong(FilterValue litValue) { return ((Number) litValue.getValue()).longValue(); } /** * Invoke this database's indexOf function. * * @param buf the SQL buffer to write the indexOf invocation to * @param str a query value representing the target string * @param find a query value representing the search string * @param start a query value representing the start index, or null * to start at the beginning */ public void indexOf(SQLBuffer buf, FilterValue str, FilterValue find, FilterValue start) { buf.append("(INSTR(("); if (start != null) substring(buf, str, start, null); else str.appendTo(buf); buf.append("), ("); find.appendTo(buf); buf.append("))"); if (start != null) { buf.append(" - 1 + "); start.appendTo(buf); } buf.append(")"); } /** * Append the numeric parts of a mathematical function. * * @param buf the SQL buffer to write the math function * @param op the mathematical operation to perform * @param lhs the left hand side of the math function * @param rhs the right hand side of the math function */ public void mathFunction(SQLBuffer buf, String op, FilterValue lhs, FilterValue rhs) { boolean castlhs = false; boolean castrhs = false; Class lc = Filters.wrap(lhs.getType()); Class rc = Filters.wrap(rhs.getType()); int type = 0; if (requiresCastForMathFunctions && (lc != rc || (lhs.isConstant() || rhs.isConstant()))) { Class c = Filters.promote(lc, rc); type = getJDBCType(JavaTypes.getTypeCode(c), false); if (type != Types.VARBINARY && type != Types.BLOB) { castlhs = (lhs.isConstant() && rhs.isConstant()) || lc != c; castrhs = (lhs.isConstant() && rhs.isConstant()) || rc != c; } } boolean mod = "MOD".equals(op); if (mod) { if (supportsModOperator) op = "%"; else buf.append(op); } buf.append("("); if (castlhs) appendCast(buf, lhs, type); else lhs.appendTo(buf); if (mod && !supportsModOperator) buf.append(", "); else buf.append(" ").append(op).append(" "); if (castrhs) appendCast(buf, rhs, type); else rhs.appendTo(buf); buf.append(")"); } /** * Append a comparison. * * @param buf the SQL buffer to write the comparison * @param op the comparison operation to perform * @param lhs the left hand side of the comparison * @param rhs the right hand side of the comparison */ public void comparison(SQLBuffer buf, String op, FilterValue lhs, FilterValue rhs) { boolean lhsxml = lhs.getXPath() != null; boolean rhsxml = rhs.getXPath() != null; if (lhsxml || rhsxml) { appendXmlComparison(buf, op, lhs, rhs, lhsxml, rhsxml); return; } boolean castlhs = false; boolean castrhs = false; Class lc = Filters.wrap(lhs.getType()); Class rc = Filters.wrap(rhs.getType()); // special case of comparison of two boolean constants // because some databases do not like false = false or false = true // but all databases understand 1 = 0 or 0 <> 1 etc. if (lc == rc && lc == Boolean.class && lhs.isConstant() && rhs.isConstant()) { String lvalue = Boolean.TRUE.equals(lhs.getValue()) ? "1" : "0"; String rvalue = Boolean.TRUE.equals(rhs.getValue()) ? "1" : "0"; buf.append(lvalue).append(op).append(rvalue); return; } int type = 0; if (requiresCastForComparisons && (lc != rc || (lhs.isConstant() && rhs.isConstant()))) { Class c = Filters.promote(lc, rc); type = getJDBCType(JavaTypes.getTypeCode(c), false); if (type != Types.VARBINARY && type != Types.BLOB) { castlhs = (lhs.isConstant() && rhs.isConstant()) || lc != c; castrhs = (lhs.isConstant() && rhs.isConstant()) || rc != c; castlhs = castlhs && lhs.requiresCast(); castrhs = castrhs && rhs.requiresCast(); } } if (castlhs) appendCast(buf, lhs, type); else lhs.appendTo(buf); buf.append(" ").append(op).append(" "); if (castrhs) appendCast(buf, rhs, type); else rhs.appendTo(buf); } /** * If this dictionary supports XML type, * use this method to append xml predicate. */ public void appendXmlComparison(SQLBuffer buf, String op, FilterValue lhs, FilterValue rhs, boolean lhsxml, boolean rhsxml) { assertSupport(supportsXMLColumn, "SupportsXMLColumn"); } /** * Append SQL for the given numeric value to the buffer, casting as needed. */ protected void appendNumericCast(SQLBuffer buf, FilterValue val) { if (val.isConstant()) appendCast(buf, val, Types.NUMERIC); else val.appendTo(buf); } /** * Cast the specified value to the specified type. * * @param buf the buffer to append the cast to * @param val the value to cast * @param type the type of the case, e.g. {@link Types#NUMERIC} */ public void appendCast(SQLBuffer buf, Object val, int type) { // Convert the cast function: "CAST({0} AS {1})" int firstParam = castFunction.indexOf("{0}"); String pre = castFunction.substring(0, firstParam); // "CAST(" String mid = castFunction.substring(firstParam + 3); int secondParam = mid.indexOf("{1}"); String post; if (secondParam > -1) { post = mid.substring(secondParam + 3); // ")" mid = mid.substring(0, secondParam); // " AS " } else post = ""; buf.append(pre); if (val instanceof FilterValue) ((FilterValue) val).appendTo(buf); else if (val instanceof SQLBuffer) buf.append(((SQLBuffer) val)); else buf.append(val.toString()); buf.append(mid); buf.append(getTypeName(type)); appendLength(buf, type); buf.append(post); } protected void appendLength(SQLBuffer buf, int type) { } /** * add CAST for a function operator where operand is a param * @param func function name * @param val * @return updated func */ public String addCastAsType(String func, Val val) { return null; } /////////// // DDL SQL /////////// /** * Increment the reference count of any table components that this * dictionary adds that are not used by mappings. Does nothing by default. */ public void refSchemaComponents(Table table) { } /** * Returns the name of the column using database specific delimiters. */ public DBIdentifier getColumnIdentifier(Column column) { if (column == null) { return DBIdentifier.NULL; } return column.getIdentifier(); } public String getColumnDBName(Column column) { return toDBName(getColumnIdentifier(column)); } /** * Returns the full name of the table, including the schema (delimited * by {@link #catalogSeparator}). */ public DBIdentifier getFullIdentifier(Table table, boolean logical) { if (!useSchemaName || DBIdentifier.isNull(table.getSchemaIdentifier())) return table.getIdentifier(); return table.getFullIdentifier(); } public String getFullName(Table table, boolean logical) { if (!useSchemaName || DBIdentifier.isNull(table.getSchemaIdentifier())) return toDBName(table.getIdentifier()); return toDBName(table.getFullIdentifier()); } /** * Returns the full name of the index, including the schema (delimited * by the result of {@link #catalogSeparator}). */ public String getFullName(Index index) { if (!useSchemaName || DBIdentifier.isNull(index.getSchemaIdentifier())) return toDBName(index.getIdentifier()); return toDBName(index.getFullIdentifier()); } /** * Returns the full name of the sequence, including the schema (delimited * by the result of {@link #catalogSeparator}). */ public String getFullName(Sequence seq) { if (!useSchemaName || DBIdentifier.isNull(seq.getSchemaIdentifier())) return toDBName(seq.getIdentifier()); return toDBName(seq.getFullIdentifier()); } /** * Return the subset of the words in reservedWordSet that cannot be used as * valid column names for the current DB. If the column name is invalid the * getValidColumnName method of the DB dictionary should be invoked to make * it valid. * * @see #getValidColumnName(org.apache.openjpa.jdbc.identifier.DBIdentifier, org.apache.openjpa.jdbc.schema.Table) */ public final Set getInvalidColumnWordSet() { return invalidColumnWordSet; } /** * Make any necessary changes to the given table name to make it valid for * the current DB. * @deprecated */ @Deprecated public String getValidTableName(String name, Schema schema) { return getValidTableName(DBIdentifier.newTable(name), schema).getName(); } /** * Make any necessary changes to the given table name to make it valid for * the current DB. */ public DBIdentifier getValidTableName(DBIdentifier name, Schema schema) { return namingUtil.getValidTableIdentifier(name, schema, maxTableNameLength); } /** * Make any necessary changes to the given sequence name to make it valid * for the current DB. * @deprecated */ @Deprecated public String getValidSequenceName(String name, Schema schema) { return getValidSequenceName(DBIdentifier.newSequence(name), schema).getName(); } /** * Make any necessary changes to the given sequence name to make it valid * for the current DB. */ public DBIdentifier getValidSequenceName(DBIdentifier name, Schema schema) { return namingUtil.getValidSequenceIdentifier(name, schema, maxTableNameLength); } /** * Make any necessary changes to the given column name to make it valid * for the current DB. The column name will be made unique for the * specified table. * @deprecated */ @Deprecated public String getValidColumnName(String name, Table table) { return getValidColumnName(DBIdentifier.newColumn(name), table, true).getName(); } /** * Make any necessary changes to the given column name to make it valid * for the current DB. The column name will be made unique for the * specified table. */ public DBIdentifier getValidColumnName(DBIdentifier name, Table table) { return getValidColumnName(name, table, true); } /** * Make any necessary changes to the given column name to make it valid * for the current DB. If checkForUniqueness is true, the column name will * be made unique for the specified table. * @deprecated */ @Deprecated public String getValidColumnName(String name, Table table, boolean checkForUniqueness) { return getValidColumnName(DBIdentifier.newColumn(name), table, checkForUniqueness).toString(); } /** * Make any necessary changes to the given column name to make it valid * for the current DB. If checkForUniqueness is true, the column name will * be made unique for the specified table. */ public DBIdentifier getValidColumnName(DBIdentifier name, Table table, boolean checkForUniqueness) { return getNamingUtil().getValidColumnIdentifier(name, table, maxColumnNameLength, checkForUniqueness); } /** * Make any necessary changes to the given primary key name to make it * valid for the current DB. */ public String getValidPrimaryKeyName(String name, Table table) { while (name.startsWith("_")) name = name.substring(1); return makeNameValid("P_" + name, table.getSchema().getSchemaGroup(), maxConstraintNameLength, NAME_ANY); } /** * Make any necessary changes to the given foreign key name to make it * valid for the current DB. * @deprecated */ @Deprecated public String getValidForeignKeyName(String name, Table table, Table toTable) { return getValidForeignKeyName(DBIdentifier.newForeignKey(name), table, toTable).getName(); } /** * Make any necessary changes to the given foreign key name to make it * valid for the current DB. */ public DBIdentifier getValidForeignKeyName(DBIdentifier name, Table table, Table toTable) { return namingUtil.getValidForeignKeyIdentifier(name, table, toTable, maxConstraintNameLength); } /** * Make any necessary changes to the given index name to make it valid * for the current DB. * @deprecated */ @Deprecated public String getValidIndexName(String name, Table table) { return getValidIndexName(DBIdentifier.newIndex(name), table).getName(); } /** * Make any necessary changes to the given index name to make it valid * for the current DB. */ public DBIdentifier getValidIndexName(DBIdentifier name, Table table) { return getNamingUtil().getValidIndexIdentifier(name, table, maxIndexNameLength); } /** * Make any necessary changes to the given unique constraint name to make * it valid for the current DB. * @deprecated */ @Deprecated public String getValidUniqueName(String name, Table table) { return getValidUniqueName(DBIdentifier.newConstraint(name), table).getName(); } /** * Make any necessary changes to the given unique constraint name to make * it valid for the current DB. */ public DBIdentifier getValidUniqueName(DBIdentifier name, Table table) { return namingUtil.getValidUniqueIdentifier(name, table, maxConstraintNameLength); } /** * Shorten the specified name to the specified target name. This will * be done by first stripping out the vowels, and then removing * characters from the middle of the word until it reaches the target * length. */ public static String shorten(String name, int targetLength) { if (name == null || name.length() <= targetLength) return name; StringBuilder nm = new StringBuilder(name); while (nm.length() > targetLength) { if (!stripVowel(nm)) { // cut out the middle char nm.replace(nm.length() / 2, (nm.length() / 2) + 1, ""); } } return nm.toString(); } /** * Remove vowels from the specified StringBuilder. * * @return true if any vowels have been removed */ private static boolean stripVowel(StringBuilder name) { if (name == null || name.length() == 0) return false; char[] vowels = { 'A', 'E', 'I', 'O', 'U', }; for (int i = 0; i < vowels.length; i++) { int index = name.toString().toUpperCase(Locale.ENGLISH).indexOf(vowels[i]); if (index != -1) { name.replace(index, index + 1, ""); return true; } } return false; } /** * Shortens the given name to the given maximum length, then checks that * it is not a reserved word. If it is reserved, appends a "0". If * the name conflicts with an existing schema component, the last * character is replace with '0', then '1', etc. * Note that the given max len may be 0 if the database metadata is * incomplete. * @deprecated */ @Deprecated protected String makeNameValid(String name, NameSet set, int maxLen, int nameType) { return makeNameValid(name, set, maxLen, nameType, true); } /** * Shortens the given name to the given maximum length, then checks that * it is not a reserved word. If it is reserved, appends a "0". If * the name conflicts with an existing schema component, the last * character is replace with '0', then '1', etc. * Note that the given max len may be 0 if the database metadata is * incomplete. */ protected DBIdentifier makeNameValid(DBIdentifier name, NameSet set, int maxLen, int nameType) { return makeNameValid(name, set, maxLen, nameType, true); } /** * Shortens the given name to the given maximum length, then checks that * it is not a reserved word. If it is reserved, appends a "0". If * the name conflicts with an existing schema component and uniqueness * checking is enabled, the last character is replace with '0', then * '1', etc. * Note that the given max len may be 0 if the database metadata is * incomplete. * * Note: If the name is delimited, make sure the ending delimiter is * not stripped off. */ protected String makeNameValid(String name, NameSet set, int maxLen, int nameType, boolean checkForUniqueness) { return namingUtil.makeNameValid(name, set, maxLen, nameType, checkForUniqueness).toString(); } /** * Shortens the given name to the given maximum length, then checks that * it is not a reserved word. If it is reserved, appends a "0". If * the name conflicts with an existing schema component and uniqueness * checking is enabled, the last character is replace with '0', then * '1', etc. * Note that the given max len may be 0 if the database metadata is * incomplete. * * Note: If the name is delimited, make sure the ending delimiter is * not stripped off. */ protected DBIdentifier makeNameValid(DBIdentifier name, NameSet set, int maxLen, int nameType, boolean checkForUniqueness) { return namingUtil.makeIdentifierValid(name, set, maxLen, checkForUniqueness); } /** * Return a series of SQL statements to create the given table, complete * with columns. Indexes and constraints will be created separately. */ public String[] getCreateTableSQL(Table table, SchemaGroup group) { return getCreateTableSQL(table); } /** * Return a series of SQL statements to create the given table, complete * with columns. Indexes and constraints will be created separately. */ public String[] getCreateTableSQL(Table table) { StringBuilder buf = new StringBuilder(); String tableName = checkNameLength(getFullIdentifier(table, false), maxTableNameLength, "long-table-name", tableLengthIncludesSchema); buf.append("CREATE TABLE ").append(tableName); if (supportsComments && table.hasComment()) { buf.append(" "); comment(buf, table.getComment()); buf.append("\n ("); } else { buf.append(" ("); } // do this before getting the columns so we know how to handle // the last comma StringBuilder endBuf = new StringBuilder(); PrimaryKey pk = table.getPrimaryKey(); String pkStr; if (pk != null) { pkStr = getPrimaryKeyConstraintSQL(pk); if (pkStr != null) endBuf.append(pkStr); } Unique[] unqs = table.getUniques(); String unqStr; for (int i = 0; i < unqs.length; i++) { unqStr = getUniqueConstraintSQL(unqs[i]); if (unqStr != null) { if (endBuf.length() > 0) endBuf.append(", "); endBuf.append(unqStr); } } Column[] cols = table.getColumns(); for (int i = 0; i < cols.length; i++) { buf.append(getDeclareColumnSQL(cols[i], false)); if (i < cols.length - 1 || endBuf.length() > 0) buf.append(", "); if (supportsComments && cols[i].hasComment()) { comment(buf, cols[i].getComment()); buf.append("\n "); } } buf.append(endBuf.toString()); buf.append(")"); return new String[]{ buf.toString() }; } public int getBatchFetchSize(int batchFetchSize) { return batchFetchSize; } protected StringBuilder comment(StringBuilder buf, String comment) { return buf.append("-- ").append(comment); } /** * Return a series of SQL statements to drop the given table. Indexes * will be dropped separately. Returns * DROP TABLE <table name> by default. */ public String[] getDropTableSQL(Table table) { String drop = MessageFormat.format(dropTableSQL, new Object[]{ getFullName(table, false) }); return new String[]{ drop }; } /** * Return a series of SQL statements to create the given sequence. Returns * CREATE SEQUENCE <sequence name>[ START WITH <start>] * [ INCREMENT BY <increment>] by default. */ public String[] getCreateSequenceSQL(Sequence seq) { return commonCreateAlterSequenceSQL(seq, true); } public String getAlterSequenceSQL(Sequence seq) { return commonCreateAlterSequenceSQL(seq, false)[0]; } private String[] commonCreateAlterSequenceSQL(Sequence seq, boolean create) { if (nextSequenceQuery == null) return null; //We need a place to detect if the user is setting the 'useNativeSequenceCache' property. //While in previous releases this property had meaning, it is no longer useful //given the code added via OPENJPA-1327. As such, we need to warn user's the //property no longer has meaning. While it would be nice to have a better way //to detect if the useNativeSequenceCache property has been set, the best we can do //is detect the variable in this code path as this is the path a user's code //would go down if they are still executing code which actually made use of //the support provided via setting useNativeSequenceCache. if (!useNativeSequenceCache && logNativeSequenceCacheWarning){ log.warn(_loc.get("sequence-cache-warning")); logNativeSequenceCacheWarning=false; } StringBuilder buf = new StringBuilder(); buf.append(create ? "CREATE" : "ALTER").append(" SEQUENCE "); //Strip off the schema and verify the sequence name is within the legal length, NOT //the schema name + sequence name. checkNameLength(toDBName(seq.getFullIdentifier().getUnqualifiedName()), maxTableNameLength, "long-seq-name"); //Now use the full sequence name (schema + sequence name). String seqName = getFullName(seq); buf.append(seqName); if (create && seq.getInitialValue() != 0) buf.append(" START WITH ").append(seq.getInitialValue()); if ((seq.getIncrement() >= 1) || (seq.getAllocate() >= 1)) buf.append(" INCREMENT BY ").append(seq.getIncrement() * seq.getAllocate()); return new String[]{ buf.toString() }; } /** * Return a series of SQL statements to drop the given sequence. Returns * DROP SEQUENCE <sequence name> by default. */ public String[] getDropSequenceSQL(Sequence seq) { return new String[]{ "DROP SEQUENCE " + getFullName(seq) }; } /** * Return a series of SQL statements to create the given index. Returns * CREATE [UNIQUE] INDEX <index name> ON <table name> * (<col list>) by default. */ public String[] getCreateIndexSQL(Index index) { StringBuilder buf = new StringBuilder(); buf.append("CREATE "); if (index.isUnique()) buf.append("UNIQUE "); DBIdentifier fullIdxName = index.getIdentifier(); DBIdentifier unQualifiedName = fullIdxName.getUnqualifiedName(); checkNameLength(toDBName(unQualifiedName), maxIndexNameLength, "long-index-name"); String indexName = toDBName(fullIdxName); buf.append("INDEX ").append(indexName); buf.append(" ON ").append(getFullName(index.getTable(), false)); buf.append(" (").append(namingUtil.appendColumns(index.getColumns())). append(")"); return new String[]{ buf.toString() }; } /** * Return a series of SQL statements to drop the given index. Returns * DROP INDEX <index name> by default. */ public String[] getDropIndexSQL(Index index) { return new String[]{ "DROP INDEX " + getFullName(index) }; } /** * Return a series of SQL statements to add the given column to * its table. Return an empty array if operation not supported. Returns * ALTER TABLE <table name> ADD (<col dec>) * by default. */ public String[] getAddColumnSQL(Column column) { if (!supportsAlterTableWithAddColumn) return new String[0]; String dec = getDeclareColumnSQL(column, true); if (dec == null) return new String[0]; return new String[]{ "ALTER TABLE " + getFullName(column.getTable(), false) + " ADD " + dec }; } /** * Return a series of SQL statements to drop the given column from * its table. Return an empty array if operation not supported. Returns * ALTER TABLE <table name> DROP COLUMN <col name> * by default. */ public String[] getDropColumnSQL(Column column) { if (!supportsAlterTableWithDropColumn) return new String[0]; return new String[]{ "ALTER TABLE " + getFullName(column.getTable(), false) + " DROP COLUMN " + column }; } /** * Return a series of SQL statements to add the given primary key to * its table. Return an empty array if operation not supported. * Returns ALTER TABLE <table name> ADD * <pk cons sql > by default. */ public String[] getAddPrimaryKeySQL(PrimaryKey pk) { String pksql = getPrimaryKeyConstraintSQL(pk); if (pksql == null) return new String[0]; return new String[]{ "ALTER TABLE " + getFullName(pk.getTable(), false) + " ADD " + pksql }; } /** * Return a series of SQL statements to drop the given primary key from * its table. Return an empty array if operation not supported. * Returns ALTER TABLE <table name> DROP CONSTRAINT * <pk name> by default. */ public String[] getDropPrimaryKeySQL(PrimaryKey pk) { if (DBIdentifier.isNull(pk.getIdentifier())) return new String[0]; return new String[]{ "ALTER TABLE " + getFullName(pk.getTable(), false) + " DROP CONSTRAINT " + toDBName(pk.getIdentifier()) }; } /** * Return a series of SQL statements to add the given foreign key to * its table. Return an empty array if operation not supported. * Returns ALTER TABLE <table name> ADD * <fk cons sql > by default. */ public String[] getAddForeignKeySQL(ForeignKey fk) { String fkSQL = getForeignKeyConstraintSQL(fk); if (fkSQL == null) return new String[0]; return new String[]{ "ALTER TABLE " + getFullName(fk.getTable(), false) + " ADD " + fkSQL }; } /** * Return a series of SQL statements to drop the given foreign key from * its table. Return an empty array if operation not supported. * Returns ALTER TABLE <table name> DROP CONSTRAINT * <fk name> by default. */ public String[] getDropForeignKeySQL(ForeignKey fk, Connection conn) { if (DBIdentifier.isNull(fk.getIdentifier())) { String[] retVal; DBIdentifier fkName = fk.loadIdentifierFromDB(this,conn); retVal = (fkName == null || fkName.getName() == null) ? new String[0] : new String[]{ "ALTER TABLE " + getFullName(fk.getTable(), false) + " DROP CONSTRAINT " + toDBName(fkName) }; return retVal; } return new String[]{ "ALTER TABLE " + getFullName(fk.getTable(), false) + " DROP CONSTRAINT " + toDBName(fk.getIdentifier()) }; } /** * Return the declaration SQL for the given column. This method is used * for each column from within {@link #getCreateTableSQL} and * {@link #getAddColumnSQL}. */ protected String getDeclareColumnSQL(Column col, boolean alter) { StringBuilder buf = new StringBuilder(); String columnName = checkNameLength(toDBName(col.getIdentifier()), maxColumnNameLength, "long-column-name"); buf.append(columnName).append(" "); buf.append(getTypeName(col)); // can't add constraints to a column we're adding after table // creation, cause some data might already be inserted if (!alter) { if (col.getDefaultString() != null && !col.isAutoAssigned()) buf.append(" DEFAULT ").append(col.getDefaultString()); if (col.isNotNull() || (!supportsNullUniqueColumn && col.hasConstraint(Unique.class))) buf.append(" NOT NULL"); } if (col.isAutoAssigned()) { if (!supportsAutoAssign) log.warn(_loc.get("invalid-autoassign", platform, col)); else if (autoAssignClause != null) buf.append(" ").append(autoAssignClause); } return buf.toString(); } /** * Return the declaration SQL for the given primary key. This method is * used from within {@link #getCreateTableSQL} and * {@link #getAddPrimaryKeySQL}. Returns * CONSTRAINT <pk name> PRIMARY KEY (<col list>) * by default. */ protected String getPrimaryKeyConstraintSQL(PrimaryKey pk) { // if we have disabled the creation of primary keys, abort here if (!createPrimaryKeys) return null; String name = toDBName(pk.getIdentifier()); if (name != null && reservedWordSet.contains(name.toUpperCase(Locale.ENGLISH))) name = null; StringBuilder buf = new StringBuilder(); if (name != null && CONS_NAME_BEFORE.equals(constraintNameMode)) buf.append("CONSTRAINT ").append(name).append(" "); buf.append("PRIMARY KEY "); if (name != null && CONS_NAME_MID.equals(constraintNameMode)) buf.append(name).append(" "); buf.append("(").append(namingUtil.appendColumns(pk.getColumns())). append(")"); if (name != null && CONS_NAME_AFTER.equals(constraintNameMode)) buf.append(" CONSTRAINT ").append(name); return buf.toString(); } /** * Return the declaration SQL for the given foreign key, or null if it is * not supported. This method is used from within * {@link #getCreateTableSQL} and {@link #getAddForeignKeySQL}. Returns * CONSTRAINT <cons name> FOREIGN KEY (<col list>) * REFERENCES <foreign table> (<col list>) * [ON DELETE <action>] [ON UPDATE <action>] by default. */ protected String getForeignKeyConstraintSQL(ForeignKey fk) { if (!supportsForeignKeys) return null; if (fk.getColumns().length > 0 && !supportsForeignKeysComposite) return null; if (fk.getDeleteAction() == ForeignKey.ACTION_NONE) return null; if (fk.isDeferred() && !supportsDeferredForeignKeyConstraints()) return null; if (!supportsDeleteAction(fk.getDeleteAction()) || !supportsUpdateAction(fk.getUpdateAction())) return null; Column[] locals = fk.getColumns(); Column[] foreigns = fk.getPrimaryKeyColumns(); int delActionId = fk.getDeleteAction(); if (delActionId == ForeignKey.ACTION_NULL) { for (int i = 0; i < locals.length; i++) { if (locals[i].isNotNull()) delActionId = ForeignKey.ACTION_NONE; } } String delAction = getActionName(delActionId); String upAction = getActionName(fk.getUpdateAction()); StringBuilder buf = new StringBuilder(); if (!DBIdentifier.isNull(fk.getIdentifier()) && CONS_NAME_BEFORE.equals(constraintNameMode)) buf.append("CONSTRAINT ").append(toDBName(fk.getIdentifier())).append(" "); buf.append("FOREIGN KEY "); if (!DBIdentifier.isNull(fk.getIdentifier()) && CONS_NAME_MID.equals(constraintNameMode)) buf.append(toDBName(fk.getIdentifier())).append(" "); buf.append("(").append(namingUtil.appendColumns(locals)).append(")"); buf.append(" REFERENCES "); buf.append(getFullName(foreigns[0].getTable(), false)); buf.append(" (").append(namingUtil.appendColumns(foreigns)).append(")"); if (delAction != null) buf.append(" ON DELETE ").append(delAction); if (upAction != null) buf.append(" ON UPDATE ").append(upAction); if (fk.isDeferred()) buf.append(" INITIALLY DEFERRED"); if (supportsDeferredForeignKeyConstraints()) buf.append(" DEFERRABLE"); if (!DBIdentifier.isNull(fk.getIdentifier()) && CONS_NAME_AFTER.equals(constraintNameMode)) buf.append(" CONSTRAINT ").append(toDBName(fk.getIdentifier())); return buf.toString(); } /** * Whether or not this dictionary supports deferred foreign key constraints. * This implementation returns {@link #supportsUniqueConstraints}. * * @since 1.1.0 */ protected boolean supportsDeferredForeignKeyConstraints() { return supportsDeferredConstraints; } /** * Return the name of the given foreign key action. */ private String getActionName(int action) { switch (action) { case ForeignKey.ACTION_CASCADE: return "CASCADE"; case ForeignKey.ACTION_NULL: return "SET NULL"; case ForeignKey.ACTION_DEFAULT: return "SET DEFAULT"; default: return null; } } /** * Whether this database supports the given foreign key delete action. */ public boolean supportsDeleteAction(int action) { if (action == ForeignKey.ACTION_NONE) return true; if (!supportsForeignKeys) return false; switch (action) { case ForeignKey.ACTION_RESTRICT: return supportsRestrictDeleteAction; case ForeignKey.ACTION_CASCADE: return supportsCascadeDeleteAction; case ForeignKey.ACTION_NULL: return supportsNullDeleteAction; case ForeignKey.ACTION_DEFAULT: return supportsDefaultDeleteAction; default: return false; } } /** * Whether this database supports the given foreign key update action. */ public boolean supportsUpdateAction(int action) { if (action == ForeignKey.ACTION_NONE) return true; if (!supportsForeignKeys) return false; switch (action) { case ForeignKey.ACTION_RESTRICT: return supportsRestrictUpdateAction; case ForeignKey.ACTION_CASCADE: return supportsCascadeUpdateAction; case ForeignKey.ACTION_NULL: return supportsNullUpdateAction; case ForeignKey.ACTION_DEFAULT: return supportsDefaultUpdateAction; default: return false; } } /** * Return the declaration SQL for the given unique constraint. This * method is used from within {@link #getCreateTableSQL}. * Returns CONSTRAINT <name> UNIQUE (<col list>) * by default. */ protected String getUniqueConstraintSQL(Unique unq) { if (!supportsUniqueConstraints || (unq.isDeferred() && !supportsDeferredUniqueConstraints())) return null; StringBuilder buf = new StringBuilder(); if (!DBIdentifier.isNull(unq.getIdentifier()) && CONS_NAME_BEFORE.equals(constraintNameMode)) buf.append("CONSTRAINT ").append(checkNameLength(toDBName(unq.getIdentifier()), maxConstraintNameLength, "long-constraint-name")).append(" "); buf.append("UNIQUE "); if (!DBIdentifier.isNull(unq.getIdentifier()) && CONS_NAME_MID.equals(constraintNameMode)) buf.append(toDBName(unq.getIdentifier())).append(" "); buf.append("(").append(namingUtil.appendColumns(unq.getColumns())). append(")"); if (unq.isDeferred()) buf.append(" INITIALLY DEFERRED"); if (supportsDeferredUniqueConstraints()) buf.append(" DEFERRABLE"); if (!DBIdentifier.isNull(unq.getIdentifier()) && CONS_NAME_AFTER.equals(constraintNameMode)) buf.append(" CONSTRAINT ").append(toDBName(unq.getIdentifier())); return buf.toString(); } /** * Whether or not this dictionary supports deferred unique constraints. * This implementation returns {@link #supportsUniqueConstraints}. * * @since 1.1.0 */ protected boolean supportsDeferredUniqueConstraints() { return supportsDeferredConstraints; } ///////////////////// // Database metadata ///////////////////// /** * This method is used to filter system tables from database metadata. * Return true if the given table name represents a system table that * should not appear in the schema definition. By default, returns * true only if the given table is in the internal list of system tables, * or if the given schema is in the list of system schemas and is not * the target schema. * * @param name the table name * @param schema the table schema; may be null * @param targetSchema if true, then the given schema was listed by * the user as one of his schemas * @deprecated */ @Deprecated public boolean isSystemTable(String name, String schema, boolean targetSchema) { return isSystemTable(DBIdentifier.newTable(name), DBIdentifier.newSchema(schema), targetSchema); } /** * This method is used to filter system tables from database metadata. * Return true if the given table name represents a system table that * should not appear in the schema definition. By default, returns * true only if the given table is in the internal list of system tables, * or if the given schema is in the list of system schemas and is not * the target schema. * * @param name the table name * @param schema the table schema; may be null * @param targetSchema if true, then the given schema was listed by * the user as one of his schemas */ public boolean isSystemTable(DBIdentifier name, DBIdentifier schema, boolean targetSchema) { DBIdentifier sName = DBIdentifier.toUpper(name); if (systemTableSet.contains(sName.getName())) return true; DBIdentifier schName = DBIdentifier.toUpper(schema); return !targetSchema && schema != null && systemSchemaSet.contains(schName.getName()); } /** * This method is used to filter system indexes from database metadata. * Return true if the given index name represents a system index that * should not appear in the schema definition. Returns false by default. * * @param name the index name * @param table the index table * @deprecated */ @Deprecated public boolean isSystemIndex(String name, Table table) { return false; } /** * This method is used to filter system indexes from database metadata. * Return true if the given index name represents a system index that * should not appear in the schema definition. Returns false by default. * * @param name the index name * @param table the index table */ public boolean isSystemIndex(DBIdentifier name, Table table) { return false; } /** * This method is used to filter system sequences from database metadata. * Return true if the given sequence represents a system sequence that * should not appear in the schema definition. Returns true if system * schema by default. * * @param name the table name * @param schema the table schema; may be null * @param targetSchema if true, then the given schema was listed by * the user as one of his schemas * @deprecated */ @Deprecated public boolean isSystemSequence(String name, String schema, boolean targetSchema) { return isSystemSequence(DBIdentifier.newSequence(name), DBIdentifier.newSchema(schema), targetSchema); } /** * This method is used to filter system sequences from database metadata. * Return true if the given sequence represents a system sequence that * should not appear in the schema definition. Returns true if system * schema by default. * * @param name the table name * @param schema the table schema; may be null * @param targetSchema if true, then the given schema was listed by * the user as one of his schemas */ public boolean isSystemSequence(DBIdentifier name, DBIdentifier schema, boolean targetSchema) { return !targetSchema && !DBIdentifier.isNull(schema) && systemSchemaSet.contains(DBIdentifier.toUpper(schema).getName()); } /** * This method is used to filter system sequences from database metadata. * Return true if the given sequence represents a system sequence that * should not appear in the schema definition. Returns true if system * schema by default. * * @param name the table name * @param schema the table schema; may be null * @param targetSchema if true, then the given schema was listed by * the user as one of his schemas * @param conn connection to the database */ public boolean isSystemSequence(DBIdentifier name, DBIdentifier schema, boolean targetSchema, Connection conn) { return isSystemSequence(name, schema, targetSchema); } /** * Reflect on the schema to find tables matching the given name pattern. * @deprecated not used by openjpa internally anymore * @see #getTables(DatabaseMetaData, DBIdentifier, DBIdentifier, DBIdentifier, Connection) */ @Deprecated public Table[] getTables(DatabaseMetaData meta, String catalog, String schemaName, String tableName, Connection conn) throws SQLException { return getTables(meta, DBIdentifier.newCatalog(catalog), DBIdentifier.newSchema(schemaName), DBIdentifier.newTable(tableName), conn); } /** * Reflect on the schema to find tables matching the given name pattern. */ public Table[] getTables(DatabaseMetaData meta, DBIdentifier sqlCatalog, DBIdentifier sqlSchemaName, DBIdentifier sqlTableName, Connection conn) throws SQLException { String schemaName = DBIdentifier.isNull(sqlSchemaName) ? null : sqlSchemaName.getName(); if (!supportsSchemaForGetTables) schemaName = null; else { schemaName = getSchemaNameForMetadata(sqlSchemaName); } String[] types = StringUtil.split(tableTypes, ",", 0); for (int i = 0; i < types.length; i++) types[i] = types[i].trim(); beforeMetadataOperation(conn); ResultSet tables = null; try { tables = meta.getTables(getCatalogNameForMetadata(sqlCatalog), schemaName, getTableNameForMetadata(sqlTableName), types); List

tableList = new ArrayList<>(); while (tables != null && tables.next()) { tableList.add(newTable(tables)); } return tableList.toArray(new Table[tableList.size()]); } finally { if (tables != null) { try { tables.close(); } catch (Exception e) { } } } } /** * Create a new table from the information in the schema metadata. */ protected Table newTable(ResultSet tableMeta) throws SQLException { Table t = new Table(); t.setIdentifier(fromDBName(tableMeta.getString("TABLE_NAME"), DBIdentifierType.TABLE)); return t; } /** * Reflect on the schema to find sequences matching the given name pattern. * Returns an empty array by default, as there is no standard way to * retrieve a list of sequences. * @deprecated */ @Deprecated public Sequence[] getSequences(DatabaseMetaData meta, String catalog, String schemaName, String sequenceName, Connection conn) throws SQLException { return getSequences(meta, DBIdentifier.newCatalog(catalog), DBIdentifier.newSchema(schemaName), DBIdentifier.newSequence(sequenceName), conn); } public Sequence[] getSequences(DatabaseMetaData meta, DBIdentifier catalog, DBIdentifier schemaName, DBIdentifier sequenceName, Connection conn) throws SQLException { String str = getSequencesSQL(schemaName, sequenceName); if (str == null) return new Sequence[0]; PreparedStatement stmnt = prepareStatement(conn, str); ResultSet rs = null; try { int idx = 1; if (!DBIdentifier.isNull(schemaName)) stmnt.setString(idx++, DBIdentifier.toUpper(schemaName).getName()); if (!DBIdentifier.isNull(sequenceName)) stmnt.setString(idx++, sequenceName.getName()); setQueryTimeout(stmnt, conf.getQueryTimeout()); rs = executeQuery(conn, stmnt, str); return getSequence(rs); } finally { if (rs != null) try { rs.close(); } catch (SQLException se) { } if (stmnt != null) try { stmnt.close(); } catch (SQLException se) { } } } /** * Create a new sequence from the information in the schema metadata. */ protected Sequence newSequence(ResultSet sequenceMeta) throws SQLException { Sequence seq = new Sequence(); seq.setSchemaIdentifier(fromDBName(StringUtil.stripEnd(sequenceMeta.getString("SEQUENCE_SCHEMA"), null), DBIdentifierType.SCHEMA)); seq.setIdentifier(fromDBName(StringUtil.stripEnd(sequenceMeta.getString("SEQUENCE_NAME"), null), DBIdentifierType.SEQUENCE)); return seq; } /** * Return the SQL needed to select the list of sequences. * @deprecated */ @Deprecated protected String getSequencesSQL(String schemaName, String sequenceName) { return null; } protected String getSequencesSQL(DBIdentifier schemaName, DBIdentifier sequenceName) { return null; } /** * Reflect on the schema to find columns matching the given table and * column patterns. * @deprecated */ @Deprecated public Column[] getColumns(DatabaseMetaData meta, String catalog, String schemaName, String tableName, String columnName, Connection conn) throws SQLException { return getColumns(meta, DBIdentifier.newCatalog(catalog), DBIdentifier.newSchema(schemaName), DBIdentifier.newTable(tableName), DBIdentifier.newColumn(columnName), conn); } /** * Reflect on the schema to find columns matching the given table and * column patterns. */ public Column[] getColumns(DatabaseMetaData meta, DBIdentifier catalog, DBIdentifier schemaName, DBIdentifier tableName, DBIdentifier columnName, Connection conn) throws SQLException { if (DBIdentifier.isNull(tableName) && !supportsNullTableForGetColumns) return null; String sqlSchemaName = null; if (!DBIdentifier.isNull(schemaName)) { sqlSchemaName = schemaName.getName(); } if (!supportsSchemaForGetColumns) sqlSchemaName = null; else sqlSchemaName = getSchemaNameForMetadata(schemaName); beforeMetadataOperation(conn); ResultSet cols = null; try { cols = meta.getColumns(getCatalogNameForMetadata(catalog), sqlSchemaName, getTableNameForMetadata(tableName), getColumnNameForMetadata(columnName)); List columnList = new ArrayList(); while (cols != null && cols.next()) columnList.add(newColumn(cols)); return (Column[]) columnList.toArray (new Column[columnList.size()]); } finally { if (cols != null) try { cols.close(); } catch (Exception e) { } } } /** * Create a new column from the information in the schema metadata. */ protected Column newColumn(ResultSet colMeta) throws SQLException { Column c = new Column(); c.setSchemaIdentifier(fromDBName(colMeta.getString("TABLE_SCHEM"), DBIdentifierType.SCHEMA)); c.setTableIdentifier(fromDBName(colMeta.getString("TABLE_NAME"), DBIdentifierType.TABLE)); c.setIdentifier(fromDBName(colMeta.getString("COLUMN_NAME"), DBIdentifierType.COLUMN)); c.setType(colMeta.getInt("DATA_TYPE")); c.setTypeIdentifier(fromDBName(colMeta.getString("TYPE_NAME"), DBIdentifierType.COLUMN_DEFINITION)); c.setSize(colMeta.getInt("COLUMN_SIZE")); c.setDecimalDigits(colMeta.getInt("DECIMAL_DIGITS")); c.setNotNull(colMeta.getInt("NULLABLE") == DatabaseMetaData.columnNoNulls); String def = colMeta.getString("COLUMN_DEF"); if (!StringUtil.isEmpty(def) && !"null".equalsIgnoreCase(def)) c.setDefaultString(def); return c; } /** * Reflect on the schema to find primary keys for the given table pattern. * @deprecated */ @Deprecated public PrimaryKey[] getPrimaryKeys(DatabaseMetaData meta, String catalog, String schemaName, String tableName, Connection conn) throws SQLException { return getPrimaryKeys(meta, DBIdentifier.newCatalog(catalog), DBIdentifier.newSchema(schemaName), DBIdentifier.newTable(tableName), conn); } /** * Reflect on the schema to find primary keys for the given table pattern. */ public PrimaryKey[] getPrimaryKeys(DatabaseMetaData meta, DBIdentifier catalog, DBIdentifier schemaName, DBIdentifier tableName, Connection conn) throws SQLException { if (useGetBestRowIdentifierForPrimaryKeys) return getPrimaryKeysFromBestRowIdentifier(meta, catalog, schemaName, tableName, conn); return getPrimaryKeysFromGetPrimaryKeys(meta, catalog, schemaName, tableName, conn); } /** * Reflect on the schema to find primary keys for the given table pattern. * @deprecated */ @Deprecated protected PrimaryKey[] getPrimaryKeysFromGetPrimaryKeys (DatabaseMetaData meta, String catalog, String schemaName, String tableName, Connection conn) throws SQLException { return getPrimaryKeysFromGetPrimaryKeys(meta, DBIdentifier.newCatalog(catalog), DBIdentifier.newSchema(schemaName), DBIdentifier.newTable(tableName), conn); } /** * Reflect on the schema to find primary keys for the given table pattern. */ protected PrimaryKey[] getPrimaryKeysFromGetPrimaryKeys (DatabaseMetaData meta, DBIdentifier catalog, DBIdentifier schemaName, DBIdentifier tableName, Connection conn) throws SQLException { if (tableName == null && !supportsNullTableForGetPrimaryKeys) return null; beforeMetadataOperation(conn); ResultSet pks = null; try { pks = meta.getPrimaryKeys(getCatalogNameForMetadata(catalog), getSchemaNameForMetadata(schemaName), getTableNameForMetadata(tableName)); List pkList = new ArrayList(); while (pks != null && pks.next()) pkList.add(newPrimaryKey(pks)); return (PrimaryKey[]) pkList.toArray (new PrimaryKey[pkList.size()]); } finally { if (pks != null) try { pks.close(); } catch (Exception e) { } } } /** * Create a new primary key from the information in the schema metadata. */ protected PrimaryKey newPrimaryKey(ResultSet pkMeta) throws SQLException { PrimaryKey pk = new PrimaryKey(); pk.setSchemaIdentifier(fromDBName(pkMeta.getString("TABLE_SCHEM"), DBIdentifierType.SCHEMA)); pk.setTableIdentifier(fromDBName(pkMeta.getString("TABLE_NAME"), DBIdentifierType.TABLE)); pk.setColumnIdentifier(fromDBName(pkMeta.getString("COLUMN_NAME"), DBIdentifierType.COLUMN)); pk.setIdentifier(fromDBName(pkMeta.getString("PK_NAME"), DBIdentifierType.CONSTRAINT)); return pk; } /** * Reflect on the schema to find primary keys for the given table pattern. * @deprecated */ @Deprecated protected PrimaryKey[] getPrimaryKeysFromBestRowIdentifier (DatabaseMetaData meta, String catalog, String schemaName, String tableName, Connection conn) throws SQLException { return getPrimaryKeysFromBestRowIdentifier(meta, DBIdentifier.newCatalog(catalog), DBIdentifier.newSchema(schemaName), DBIdentifier.newTable(tableName), conn); } /** * Reflect on the schema to find primary keys for the given table pattern. */ protected PrimaryKey[] getPrimaryKeysFromBestRowIdentifier (DatabaseMetaData meta, DBIdentifier catalog, DBIdentifier schemaName, DBIdentifier tableName, Connection conn) throws SQLException { if (tableName == null) return null; beforeMetadataOperation(conn); ResultSet pks = null; try { pks = meta.getBestRowIdentifier(toDBName(catalog), toDBName(schemaName), toDBName(tableName), 0, false); List pkList = new ArrayList(); while (pks != null && pks.next()) { PrimaryKey pk = new PrimaryKey(); pk.setSchemaIdentifier(schemaName); pk.setTableIdentifier(tableName); pk.setColumnIdentifier(fromDBName(pks.getString("COLUMN_NAME"), DBIdentifierType.COLUMN)); pkList.add(pk); } return (PrimaryKey[]) pkList.toArray (new PrimaryKey[pkList.size()]); } finally { if (pks != null) try { pks.close(); } catch (Exception e) { } } } /** * Reflect on the schema to find indexes matching the given table pattern. * @deprecated */ @Deprecated public Index[] getIndexInfo(DatabaseMetaData meta, String catalog, String schemaName, String tableName, boolean unique, boolean approx, Connection conn) throws SQLException { return getIndexInfo(meta, DBIdentifier.newCatalog(catalog), DBIdentifier.newSchema(schemaName), DBIdentifier.newTable(tableName), unique, approx, conn); } /** * Reflect on the schema to find indexes matching the given table pattern. */ public Index[] getIndexInfo(DatabaseMetaData meta, DBIdentifier catalog, DBIdentifier schemaName, DBIdentifier tableName, boolean unique, boolean approx, Connection conn) throws SQLException { if (tableName == null && !supportsNullTableForGetIndexInfo) return null; beforeMetadataOperation(conn); ResultSet indexes = null; try { indexes = meta.getIndexInfo(getCatalogNameForMetadata(catalog), getSchemaNameForMetadata(schemaName), getTableNameForMetadata(tableName), unique, approx); List indexList = new ArrayList(); while (indexes != null && indexes.next()) indexList.add(newIndex(indexes)); return (Index[]) indexList.toArray(new Index[indexList.size()]); } finally { if (indexes != null) try { indexes.close(); } catch (Exception e) { } } } /** * Create a new index from the information in the schema metadata. */ protected Index newIndex(ResultSet idxMeta) throws SQLException { Index idx = new Index(); idx.setSchemaIdentifier(fromDBName(idxMeta.getString("TABLE_SCHEM"), DBIdentifierType.SCHEMA)); idx.setTableIdentifier(fromDBName(idxMeta.getString("TABLE_NAME"), DBIdentifierType.TABLE)); idx.setColumnIdentifier(fromDBName(idxMeta.getString("COLUMN_NAME"), DBIdentifierType.COLUMN)); idx.setIdentifier(fromDBName(idxMeta.getString("INDEX_NAME"), DBIdentifierType.INDEX)); idx.setUnique(!idxMeta.getBoolean("NON_UNIQUE")); return idx; } /** * Reflect on the schema to return foreign keys imported by the given * table pattern. * @deprecated */ @Deprecated public ForeignKey[] getImportedKeys(DatabaseMetaData meta, String catalog, String schemaName, String tableName, Connection conn) throws SQLException { return getImportedKeys(meta, catalog, schemaName, tableName, conn, true); } /** * Reflect on the schema to return foreign keys imported by the given * table pattern. */ public ForeignKey[] getImportedKeys(DatabaseMetaData meta, DBIdentifier catalog, DBIdentifier schemaName, DBIdentifier tableName, Connection conn) throws SQLException { return getImportedKeys(meta, catalog, schemaName, tableName, conn, true); } /** * Reflect on the schema to return full foreign keys imported by the given * table pattern. * @deprecated */ @Deprecated public ForeignKey[] getImportedKeys(DatabaseMetaData meta, String catalog, String schemaName, String tableName, Connection conn, boolean partialKeys) throws SQLException { return getImportedKeys(meta, DBIdentifier.newCatalog(catalog), DBIdentifier.newSchema(schemaName), DBIdentifier.newTable(tableName), conn, partialKeys); } /** * Reflect on the schema to return full foreign keys imported by the given * table pattern. */ public ForeignKey[] getImportedKeys(DatabaseMetaData meta, DBIdentifier catalog, DBIdentifier schemaName, DBIdentifier tableName, Connection conn, boolean partialKeys) throws SQLException { if (!supportsForeignKeys) return null; if (tableName == null && !supportsNullTableForGetImportedKeys) return null; beforeMetadataOperation(conn); ResultSet keys = null; try { keys = meta.getImportedKeys(getCatalogNameForMetadata(catalog), getSchemaNameForMetadata(schemaName), getTableNameForMetadata(tableName)); List importedKeyList = new ArrayList<>(); Map fkMap = new HashMap<>(); while (keys != null && keys.next()) { ForeignKey nfk = newForeignKey(keys); if (!partialKeys) { ForeignKey fk = combineForeignKey(fkMap, nfk); // If the key returned != new key, fk col was combined // with existing fk. if (fk != nfk) { continue; } } importedKeyList.add(nfk); } return (ForeignKey[]) importedKeyList.toArray (new ForeignKey[importedKeyList.size()]); } finally { if (keys != null) { try { keys.close(); } catch (Exception e) { } } } } /** * Combines partial foreign keys into singular key */ protected ForeignKey combineForeignKey(Map fkMap, ForeignKey fk) { FKMapKey fkmk = new FKMapKey(fk); ForeignKey baseKey = fkMap.get(fkmk); // Found the FK, add the additional column if (baseKey != null) { baseKey.addColumn(fk); return baseKey; } // fkey is new fkMap.put(fkmk, fk); return fk; } /** * Create a new foreign key from the information in the schema metadata. */ protected ForeignKey newForeignKey(ResultSet fkMeta) throws SQLException { ForeignKey fk = new ForeignKey(); fk.setSchemaIdentifier(fromDBName(fkMeta.getString("FKTABLE_SCHEM"), DBIdentifierType.SCHEMA)); fk.setTableIdentifier(fromDBName(fkMeta.getString("FKTABLE_NAME"), DBIdentifierType.TABLE)); fk.setColumnIdentifier(fromDBName(fkMeta.getString("FKCOLUMN_NAME"), DBIdentifierType.COLUMN)); fk.setIdentifier(fromDBName(fkMeta.getString("FK_NAME"), DBIdentifierType.FOREIGN_KEY)); fk.setPrimaryKeySchemaIdentifier(fromDBName(fkMeta.getString("PKTABLE_SCHEM"), DBIdentifierType.SCHEMA)); fk.setPrimaryKeyTableIdentifier(fromDBName(fkMeta.getString("PKTABLE_NAME"), DBIdentifierType.TABLE)); fk.setPrimaryKeyColumnIdentifier(fromDBName(fkMeta.getString("PKCOLUMN_NAME"), DBIdentifierType.COLUMN)); fk.setKeySequence(fkMeta.getShort("KEY_SEQ")); fk.setDeferred(fkMeta.getShort("DEFERRABILITY") == DatabaseMetaData.importedKeyInitiallyDeferred); int del = fkMeta.getShort("DELETE_RULE"); switch (del) { case DatabaseMetaData.importedKeySetNull: fk.setDeleteAction(ForeignKey.ACTION_NULL); break; case DatabaseMetaData.importedKeySetDefault: fk.setDeleteAction(ForeignKey.ACTION_DEFAULT); break; case DatabaseMetaData.importedKeyCascade: fk.setDeleteAction(ForeignKey.ACTION_CASCADE); break; default: fk.setDeleteAction(ForeignKey.ACTION_RESTRICT); break; } return fk; } /** * Returns the table name that will be used for obtaining information * from {@link DatabaseMetaData}. */ protected String getTableNameForMetadata(String tableName) { return convertSchemaCase(DBIdentifier.newTable(tableName)); } /** * Returns the table name that will be used for obtaining information * from {@link DatabaseMetaData}. */ protected String getTableNameForMetadata(DBIdentifier tableName) { if (tableName == null) { return null; } return convertSchemaCase(tableName.getUnqualifiedName()); } /** * Returns the schema name that will be used for obtaining information * from {@link DatabaseMetaData}. */ protected String getSchemaNameForMetadata(String schemaName) { if (schemaName == null) schemaName = conf.getSchema(); return convertSchemaCase(DBIdentifier.newSchema(schemaName)); } /** * Returns the schema name that will be used for obtaining information * from {@link DatabaseMetaData}. */ protected String getSchemaNameForMetadata(DBIdentifier schemaName) { if (DBIdentifier.isNull(schemaName)) schemaName = DBIdentifier.newSchema(conf.getSchema()); return convertSchemaCase(schemaName); } /** * Returns the catalog name that will be used for obtaining information * from {@link DatabaseMetaData}. */ protected String getCatalogNameForMetadata(String catalogName) { return convertSchemaCase(DBIdentifier.newCatalog(catalogName)); } /** * Returns the catalog name that will be used for obtaining information * from {@link DatabaseMetaData}. */ protected String getCatalogNameForMetadata(DBIdentifier catalogName) { return convertSchemaCase(catalogName); } /** * Returns the column name that will be used for obtaining information * from {@link DatabaseMetaData}. */ protected String getColumnNameForMetadata(String columnName) { return convertSchemaCase(DBIdentifier.newColumn(columnName)); } /** * Returns the column name that will be used for obtaining information * from {@link DatabaseMetaData}. */ protected String getColumnNameForMetadata(DBIdentifier columnName) { return convertSchemaCase(columnName); } /** * Convert the specified schema name to a name that the database will * be able to understand. */ public String convertSchemaCase(String objectName) { return convertSchemaCase(DBIdentifier.newIdentifier(objectName, DBIdentifierType.DEFAULT, false)); } /** * Convert the specified schema name to a name that the database will * be able to understand. */ public String convertSchemaCase(DBIdentifier objectName) { return toDBName(namingUtil.convertSchemaCase(objectName), false); } /** * Return DB specific schemaCase */ @Override public String getSchemaCase(){ return schemaCase; } /** * Prepared the connection for metadata operations. */ private void beforeMetadataOperation(Connection c) { if (requiresAutoCommitForMetaData) { try { c.rollback(); } catch (SQLException sqle) { } try { if (!c.getAutoCommit()) c.setAutoCommit(true); } catch (SQLException sqle) { } } } ///////////////////////////// // Sequences and Auto-Assign ///////////////////////////// /** * Return the last generated value for the given column. * Throws an exception by default if {@link #lastGeneratedKeyQuery} is null. */ public Object getGeneratedKey(Column col, Connection conn) throws SQLException { if (lastGeneratedKeyQuery == null) throw new StoreException(_loc.get("no-auto-assign")); // replace things like "SELECT MAX({0}) FROM {1}" String query = lastGeneratedKeyQuery; if (query.indexOf('{') != -1) // only if the token is in the string { query = getGenKeySeqName(query, col); } PreparedStatement stmnt = prepareStatement(conn, query); ResultSet rs = null; try { setQueryTimeout(stmnt, conf.getQueryTimeout()); rs = executeQuery(conn, stmnt, query); return getKey(rs, col); } finally { if (rs != null) try { rs.close(); } catch (SQLException se) {} if (stmnt != null) try { stmnt.close(); } catch (SQLException se) {} } } protected String getGenKeySeqName(String query, Column col) { return MessageFormat.format(query, new Object[]{ toDBName(col.getIdentifier()), getFullName(col.getTable(), false), getGeneratedKeySequenceName(col), }); } /** * Return the sequence name used by databases for the given autoassigned * column. This is only used by databases that require an explicit name * to be used for auto-assign support. */ protected String getGeneratedKeySequenceName(Column col) { return toDBName(namingUtil.getGeneratedKeySequenceName(col, maxAutoAssignNameLength)); } protected Map _procs = new TreeMap<>(); /** * Gets the metadata of the stored procedure by the given name either from the cached version or * by enquiring the database. * @param meta the database meta data * @param catalog the catalog name or null * @param schema the schema name or null * @param procedure the procedure name * @return metadata about the named procedure or null * @throws SQLException when metadata query goes wrong */ public StoredProcedure getStoredProcedure(DatabaseMetaData meta, DBIdentifier catalog, DBIdentifier schema, String procedure) throws SQLException { if (_procs.containsKey(procedure)) { return _procs.get(procedure); } ResultSet rs = meta.getProcedureColumns( getCatalogNameForMetadata(catalog), getSchemaNameForMetadata(schema), procedure, null); StoredProcedure sp = null; if (rs.next()) { sp = new StoredProcedure(rs); } else { ResultSet rsExist = meta.getProcedures( getCatalogNameForMetadata(catalog), getSchemaNameForMetadata(schema), procedure); if (rsExist.next()) { sp = new StoredProcedure((String) null); sp.setCatalog(catalog); sp.setSchema(schema); sp.setName(procedure); } } _procs.put(procedure, sp); return sp; } /////////////////////////////// // Configurable implementation /////////////////////////////// @Override public void setConfiguration(Configuration conf) { this.conf = (JDBCConfiguration) conf; this.log = this.conf.getLog(JDBCConfiguration.LOG_JDBC); // Create the naming utility namingUtil = this.conf.getIdentifierUtilInstance(); namingUtil.setIdentifierConfiguration(this); configureNamingRules(); // warn about unsupported dicts if (log.isWarnEnabled() && !isSupported()) log.warn(_loc.get("dict-not-supported", getClass())); } private boolean isSupported() { // if this is a custom dict, traverse to whatever openjpa dict it // extends Class c = getClass(); while (!c.getName().startsWith("org.apache.openjpa.")) c = c.getSuperclass(); // the generic dbdictionary is not considered a supported dict; all // other concrete dictionaries are if (c == DBDictionary.class) return false; return true; } @Override public void startConfiguration() { } @Override public void endConfiguration() { // initialize the set of reserved SQL92 words from resource InputStream in = DBDictionary.class.getResourceAsStream ("sql-keywords.rsrc"); try { String keywords = new BufferedReader(new InputStreamReader(in)). readLine(); reservedWordSet.addAll(Arrays.asList(StringUtil.split (keywords, ",", 0))); } catch (IOException ioe) { throw new GeneralException(ioe); } finally { try { in.close(); } catch (IOException e) {} } // add additional reserved words set by user if (reservedWords != null) reservedWordSet.addAll(Arrays.asList(StringUtil.split(reservedWords.toUpperCase(Locale.ENGLISH), ",", 0))); // add system schemas set by user if (systemSchemas != null) systemSchemaSet.addAll(Arrays.asList(StringUtil.split(systemSchemas.toUpperCase(Locale.ENGLISH), ",", 0))); // add system tables set by user if (systemTables != null) systemTableSet.addAll(Arrays.asList(StringUtil.split(systemTables.toUpperCase(Locale.ENGLISH), ",", 0))); // add fixed size type names set by the user if (fixedSizeTypeNames != null) fixedSizeTypeNameSet.addAll(Arrays.asList(StringUtil.split(fixedSizeTypeNames.toUpperCase(Locale.ENGLISH), ",", 0))); // if user has unset sequence sql, null it out so we know sequences // aren't supported nextSequenceQuery = StringUtil.trimToNull(nextSequenceQuery); if (selectWords != null) selectWordSet.addAll(Arrays.asList(StringUtil.split(selectWords.toUpperCase(Locale.ENGLISH), ",", 0))); // initialize the error codes SQLErrorCodeReader codeReader = new SQLErrorCodeReader(); String rsrc = "sql-error-state-codes.xml"; // We'll allow sub-classes to override the stream for custom err codes // @edu.umd.cs.findbugs.annotations.SuppressWarnings(value="UI_INHERITANCE_UNSAFE_GETRESOURCE") InputStream stream = getClass().getResourceAsStream(rsrc); String dictionaryClassName = getClass().getName(); if (stream == null) { // User supplied dictionary but no error codes xml // use default stream = DBDictionary.class.getResourceAsStream(rsrc); dictionaryClassName = getClass().getSuperclass().getName(); } codeReader.parse(stream, dictionaryClassName, this); } public void addErrorCode(int errorType, String errorCode) { if (errorCode == null || errorCode.trim().length() == 0) return; Set codes = sqlStateCodes.get(errorType); if (codes == null) { codes = new HashSet<>(); codes.add(errorCode.trim()); sqlStateCodes.put(errorType, codes); } else { codes.add(errorCode.trim()); } } /** * FIXME - OPENJPA-957 - lockTimeout is a server-side function and * shouldn't be using client-side setQueryTimeout for lock timeouts. * * This method is to provide override for non-JDBC or JDBC-like * implementation of setting query and lock timeouts. * * @param stmnt * @param fetch - optional lock and query timeouts in milliseconds * @param forUpdate - true if we should also try setting a lock timeout * @throws SQLException */ public void setTimeouts(PreparedStatement stmnt, JDBCFetchConfiguration fetch, boolean forUpdate) throws SQLException { if (supportsQueryTimeout) { int timeout = fetch.getQueryTimeout(); if (forUpdate) { // if this is a locking select and the lock timeout is greater // than the configured query timeout, use the lock timeout timeout = Math.max(fetch.getQueryTimeout(), fetch.getLockTimeout()); } setQueryTimeout(stmnt, timeout); } } /** * FIXME - OPENJPA-957 - lockTimeout is a server-side function and * shouldn't be using client-side setQueryTimeout for lock timeouts. * * This method is to provide override for non-JDBC or JDBC-like * implementation of setting query and lock timeouts. * * @param stmnt * @param conf - optional lock and query timeouts in milliseconds * @param forUpdate - true if we should also try setting a lock timeout * @throws SQLException */ public void setTimeouts(PreparedStatement stmnt, JDBCConfiguration conf, boolean forUpdate) throws SQLException { if (supportsQueryTimeout) { int timeout = conf.getQueryTimeout(); if (forUpdate) { // if this is a locking select and the lock timeout is greater // than the configured query timeout, use the lock timeout timeout = Math.max(conf.getQueryTimeout(), conf.getLockTimeout()); } setQueryTimeout(stmnt, timeout); } } /** * Provides the default validation handling of setting a query timeout. * @param stmnt * @param timeout in milliseconds * @throws SQLException */ public void setQueryTimeout(PreparedStatement stmnt, int timeout) throws SQLException { if (supportsQueryTimeout) { if (timeout == -1) { // special OpenJPA allowed case denoting no timeout timeout = 0; } else if (timeout < 0) { if (log.isWarnEnabled()) log.warn(_loc.get("invalid-timeout", Integer.valueOf(timeout))); return; } else if (timeout > 0 && timeout < 1000) { // round up to 1 sec timeout = 1000; if (log.isWarnEnabled()) log.warn(_loc.get("millis-query-timeout")); } setStatementQueryTimeout(stmnt, timeout); } } /** * Allow subclasses to provide DB unique override implementations of * setting query timeouts, while preserving the default timeout logic * in the public setQueryTimeout method. * @param stmnt * @param timeout in milliseconds * @throws SQLException */ protected void setStatementQueryTimeout(PreparedStatement stmnt, int timeout) throws SQLException { // JDBC uses seconds, so we'll do a simple round-down conversion here stmnt.setQueryTimeout(timeout / 1000); } ////////////////////////////////////// // ConnectionDecorator implementation ////////////////////////////////////// /** * Decorate the given connection if needed. Some databases require special * handling for JDBC bugs. This implementation issues any * {@link #initializationSQL} that has been set for the dictionary but * does not decorate the connection. */ @Override public Connection decorate(Connection conn) throws SQLException { if (!connected) connectedConfiguration(conn); if (!StringUtil.isEmpty(initializationSQL)) { PreparedStatement stmnt = null; try { stmnt = conn.prepareStatement(initializationSQL); stmnt.execute(); } catch (Exception e) { if (log.isTraceEnabled()) log.trace(e.toString(), e); } finally { if (stmnt != null) try { stmnt.close(); } catch (SQLException se) { } } } return conn; } /** * Implementation of the * {@link LoggingConnectionDecorator.SQLWarningHandler} interface * that allows customization of the actions to perform when a * {@link SQLWarning} occurs at any point on a {@link Connection}, * {@link Statement}, or {@link ResultSet}. This method may * be used determine those warnings the application wants to * consider critical failures, and throw the warning in those * cases. By default, this method does nothing. * * @see LoggingConnectionDecorator#setWarningAction * @see LoggingConnectionDecorator#setWarningHandler */ @Override public void handleWarning(SQLWarning warning) throws SQLException { } /** * Return a new exception that wraps causes. * However, the details of exactly what type of exception is returned can * be determined by the implementation. This may take into account * DB-specific exception information in causes. */ public OpenJPAException newStoreException(String msg, SQLException[] causes, Object failed) { if (causes != null && causes.length > 0) { OpenJPAException ret = narrow(msg, causes[0], failed); ret.setFailedObject(failed).setNestedThrowables(causes); return ret; } return new StoreException(msg).setFailedObject(failed). setNestedThrowables(causes); } /** * Gets the category of StoreException by matching the given SQLException's * error state code to the list of error codes supplied by the dictionary. * * @return a StoreException of {@link ExceptionInfo#GENERAL general} category * if the given SQL Exception can not be further categorized. * * @see #matchErrorState(Map, SQLException) */ OpenJPAException narrow(String msg, SQLException ex, Object failed) { int errorType = matchErrorState(sqlStateCodes, ex); StoreException storeEx; switch (errorType) { case StoreException.LOCK: storeEx = new LockException(failed); break; case StoreException.OBJECT_EXISTS: storeEx = new ObjectExistsException(msg); break; case StoreException.OBJECT_NOT_FOUND: storeEx = new ObjectNotFoundException(failed); break; case StoreException.OPTIMISTIC: storeEx = new OptimisticException(failed); break; case StoreException.REFERENTIAL_INTEGRITY: storeEx = new ReferentialIntegrityException(msg); break; case StoreException.QUERY: storeEx = new QueryException(msg); break; default: storeEx = new StoreException(msg); } storeEx.setFatal(isFatalException(errorType, ex)); return storeEx; } /** * Determine the more appropriate type of store exception by matching the SQL Error State of the * the given SQLException to the given Error States categorized by error types. * Dictionary subclass can override this method and extract * SQLException data to figure out if the exception is recoverable. * * @param errorStates classification of SQL error states by their specific nature. The keys of the * map represent one of the constants defined in {@link StoreException}. The value corresponding to * a key represent the set of SQL Error States representing specific category of database error. * This supplied map is sourced from sql-error-state-codes.xml and filtered the * error states for the current database. * * @param ex original SQL Exception as raised by the database driver. * * @return A constant indicating the category of error as defined in {@link StoreException}. */ protected int matchErrorState(Map> errorStates, SQLException ex) { String errorState = ex.getSQLState(); for (Map.Entry> states : errorStates.entrySet()) { if (states.getValue().contains(errorState)) return states.getKey(); } return ExceptionInfo.GENERAL; } /** * Determine if the given SQL Exception is fatal or recoverable (such as a timeout). * This implementation always returns true (i.e. all exceptions are fatal). * The current dictionary implementation can overwrite this method to mark certain * exception conditions as recoverable error. * @param subtype A constant indicating the category of error as defined in {@link StoreException}. * @param ex original SQL Exception as raised by the database driver. * * @return false if the error is fatal. */ public boolean isFatalException(int subtype, SQLException ex) { return true; } /** * Closes the specified {@link DataSource} and releases any * resources associated with it. * * @param dataSource the DataSource to close */ public void closeDataSource(DataSource dataSource) { DataSourceFactory.closeDataSource(dataSource); } /** * Used by some mappings to represent data that has already been * serialized so that we don't have to serialize multiple times. */ public static class SerializedData { public final byte[] bytes; public SerializedData(byte[] bytes) { this.bytes = bytes; } } /** * Return version column name * @param column * @param tableAlias : this is needed for platform specific version column */ public String getVersionColumn(Column column, String tableAlias) { return getVersionColumn(column, DBIdentifier.newTable(tableAlias)).toString(); } public DBIdentifier getVersionColumn(Column column, DBIdentifier tableAlias) { return column.getIdentifier(); } public void insertBlobForStreamingLoad(Row row, Column col, JDBCStore store, Object ob, Select sel) throws SQLException { if (ob != null) { row.setBinaryStream(col, new ByteArrayInputStream(new byte[0]), 0); } else { row.setNull(col); } } public void insertClobForStreamingLoad(Row row, Column col, Object ob) throws SQLException { if (ob != null) { row.setCharacterStream(col, new CharArrayReader(new char[0]), 0); } else { row.setNull(col); } } public void updateBlob(Select sel, JDBCStore store, InputStream is) throws SQLException { SQLBuffer sql = sel.toSelect(true, store.getFetchConfiguration()); ResultSet res = null; Connection conn = store.getConnection(); PreparedStatement stmnt = null; try { stmnt = sql.prepareStatement(conn, store.getFetchConfiguration(), ResultSet.TYPE_SCROLL_SENSITIVE, ResultSet.CONCUR_UPDATABLE); setTimeouts(stmnt, store.getFetchConfiguration(), true); res = stmnt.executeQuery(); if (!res.next()) { throw new InternalException(_loc.get("stream-exception")); } Blob blob = res.getBlob(1); OutputStream os = blob.setBinaryStream(1); copy(is, os); os.close(); res.updateBlob(1, blob); res.updateRow(); } catch (IOException ioe) { throw new StoreException(ioe); } finally { if (res != null) try { res.close (); } catch (SQLException e) {} if (stmnt != null) try { stmnt.close (); } catch (SQLException e) {} if (conn != null) try { conn.close (); } catch (SQLException e) {} } } public void updateClob(Select sel, JDBCStore store, Reader reader) throws SQLException { SQLBuffer sql = sel.toSelect(true, store.getFetchConfiguration()); ResultSet res = null; Connection conn = store.getConnection(); PreparedStatement stmnt = null; try { stmnt = sql.prepareStatement(conn, store.getFetchConfiguration(), ResultSet.TYPE_SCROLL_SENSITIVE, ResultSet.CONCUR_UPDATABLE); setTimeouts(stmnt, store.getFetchConfiguration(), true); res = stmnt.executeQuery(); if (!res.next()) { throw new InternalException(_loc.get("stream-exception")); } Clob clob = res.getClob(1); if (clob != null) { Writer writer = clob.setCharacterStream(1); copy(reader, writer); writer.close(); res.updateClob(1, clob); res.updateRow(); } } catch (IOException ioe) { throw new StoreException(ioe); } finally { if (res != null) try { res.close (); } catch (SQLException e) {} if (stmnt != null) try { stmnt.close (); } catch (SQLException e) {} if (conn != null) try { conn.close (); } catch (SQLException e) {} } } protected long copy(InputStream in, OutputStream out) throws IOException { byte[] copyBuffer = new byte[blobBufferSize]; long bytesCopied = 0; int read = -1; while ((read = in.read(copyBuffer, 0, copyBuffer.length)) != -1) { out.write(copyBuffer, 0, read); bytesCopied += read; } return bytesCopied; } protected long copy(Reader reader, Writer writer) throws IOException { char[] copyBuffer = new char[clobBufferSize]; long bytesCopied = 0; int read = -1; while ((read = reader.read(copyBuffer, 0, copyBuffer.length)) != -1) { writer.write(copyBuffer, 0, read); bytesCopied += read; } return bytesCopied; } /** * Attach CAST to the current function if necessary * * @param val operand value * @parma func the sql function statement * @return a String with the correct CAST function syntax */ public String getCastFunction(Val val, String func) { return func; } /** * Return the correct CAST function syntax. This should be overriden by subclasses * that need access to the Column information. * * @param val operand of cast * @param func original string * @param col database column * @return a String with the correct CAST function syntax */ public String getCastFunction(Val val, String func, Column col) { return getCastFunction (val, func); } /** * Create an index if necessary for some database tables * @deprecated */ @Deprecated public void createIndexIfNecessary(Schema schema, String table, Column pkColumn) { } public void createIndexIfNecessary(Schema schema, DBIdentifier table, Column pkColumn) { } /** * Return the batchLimit */ public int getBatchLimit(){ return batchLimit; } /** * Set the batchLimit value */ public void setBatchLimit(int limit){ batchLimit = limit; } /** * Validate the batch process. In some cases, we can't batch the statements * due to some restrictions. For example, if the GeneratedType=IDENTITY, * we have to disable the batch process because we need to get the ID value * right away for the in-memory entity to use. */ public boolean validateBatchProcess(RowImpl row, Column[] autoAssign, OpenJPAStateManager sm, ClassMapping cmd ) { boolean disableBatch = false; if (getBatchLimit()== 0) return false; if (autoAssign != null && sm != null) { FieldMetaData[] fmd = cmd.getPrimaryKeyFields(); int i = 0; while (!disableBatch && i < fmd.length) { if (fmd[i].getValueStrategy() == ValueStrategies.AUTOASSIGN) disableBatch = true; i++; } } // go to each Dictionary to validate the batch capability if (!disableBatch) disableBatch = validateDBSpecificBatchProcess(disableBatch, row, autoAssign, sm, cmd); return disableBatch; } /** * Allow each Dictionary to validate its own batch process. */ public boolean validateDBSpecificBatchProcess (boolean disableBatch, RowImpl row, Column[] autoAssign, OpenJPAStateManager sm, ClassMapping cmd ) { return disableBatch; } /** * This method is to provide override for non-JDBC or JDBC-like * implementation of executing query. */ protected ResultSet executeQuery(Connection conn, PreparedStatement stmnt, String sql ) throws SQLException { return stmnt.executeQuery(); } /** * This method is to provide override for non-JDBC or JDBC-like * implementation of preparing statement. */ protected PreparedStatement prepareStatement(Connection conn, String sql) throws SQLException { return conn.prepareStatement(sql); } /** * This method is to provide override for non-JDBC or JDBC-like * implementation of getting sequence from the result set. */ protected Sequence[] getSequence(ResultSet rs) throws SQLException { List seqList = new ArrayList(); while (rs != null && rs.next()) seqList.add(newSequence(rs)); return (Sequence[]) seqList.toArray(new Sequence[seqList.size()]); } /** * This method is to provide override for non-JDBC or JDBC-like * implementation of getting key from the result set. */ protected Object getKey (ResultSet rs, Column col) throws SQLException { if (!rs.next()) throw new StoreException(_loc.get("no-genkey")); Object key = rs.getObject(1); if (key == null) log.warn(_loc.get("invalid-genkey", col)); return key; } /** * This method is to provide override for non-JDBC or JDBC-like * implementation of calculating value. */ protected void calculateValue(Val val, Select sel, ExpContext ctx, ExpState state, Path path, ExpState pathState) { val.calculateValue(sel, ctx, state, (Val) path, pathState); } /** * Determine whether the provided sql may be treated as a * select statement on this database. * * @param sql A sql statement. * * @return true if sql represents a select statement. */ public boolean isSelect(String sql) { Iterator i = selectWordSet.iterator(); String cur; while (i.hasNext()) { cur = (String) i.next(); if (sql.length() >= cur.length() && sql.substring(0, cur.length()).equalsIgnoreCase(cur)) { return true; } } return false; } public boolean needsToCreateIndex(Index idx, Table table, Unique[] uniques) { return needsToCreateIndex(idx, table); } public boolean needsToCreateIndex(Index idx, Table table) { return true; } /** * Return batched statements update success count * @param ps A PreparedStatement * @return return update count */ public int getBatchUpdateCount(PreparedStatement ps) throws SQLException { return 0; } public boolean getTrimSchemaName() { return trimSchemaName; } public void setTrimSchemaName(boolean trimSchemaName) { this.trimSchemaName = trimSchemaName; } public void deleteStream(JDBCStore store, Select sel) throws SQLException { // Do nothing } /** * Validate that the given name is not longer than given maximum length. * If the given name is indeed longer then raises a UserException with the * given message key otherwise returns the same name. */ final String checkNameLength(String name, int length, String msgKey) { if (name.length() > length) { throw new UserException(_loc.get(msgKey, name, name.length(), length)); } return name; } /** * Validate that the given name is not longer than given maximum length. Uses the unqualified name * from the supplied {@link DBIdentifier} by default.. * * @param identifier The database identifier to check. * @param length Max length for this type of identifier * @param msgKey message identifier for the exception. * * @throws {@link UserException} with the given message key if the given name is indeed longer. * @return the same name. */ final String checkNameLength(DBIdentifier identifier, int length, String msgKey) { return checkNameLength(identifier, length, msgKey, false); } /** * Validate that the given name is not longer than given maximum length. Conditionally uses the unqualified name * from the supplied {@link DBIdentifier}. * * @param identifier The database identifier to check. * @param length Max length for this type of identifier * @param msgKey message identifier for the exception. * @param qualified If true the qualified name of the DBIdentifier will be used. * * @throws {@link UserException} with the given message key if the given name is indeed longer. * @return the same name. */ final String checkNameLength(DBIdentifier identifier, int length, String msgKey, boolean qualified) { // always return the input name, String name = toDBName(identifier); String compareName = qualified ? name : toDBName(identifier.getUnqualifiedName()); if (compareName.length() > length) { throw new UserException(_loc.get(msgKey, name, name.length(), length)); } return name; } protected void setDelimitedCase(DatabaseMetaData metaData) { try { if (metaData.storesMixedCaseQuotedIdentifiers()) { delimitedCase = SCHEMA_CASE_PRESERVE; } else if (metaData.storesUpperCaseQuotedIdentifiers()) { delimitedCase = SCHEMA_CASE_UPPER; } else if (metaData.storesLowerCaseQuotedIdentifiers()) { delimitedCase = SCHEMA_CASE_LOWER; } } catch (SQLException e) { getLog().warn("cannot-determine-identifier-case"); if (getLog().isTraceEnabled()) { getLog().trace(e.toString(), e); } } } /** * @return the supportsDelimitedIds */ @Override public boolean getSupportsDelimitedIdentifiers() { return (supportsDelimitedIdentifiers == null ? false : supportsDelimitedIdentifiers); } /** * @param supportsDelimitedIds the supportsDelimitedIds to set */ public void setSupportsDelimitedIdentifiers(boolean supportsDelimitedIds) { supportsDelimitedIdentifiers = Boolean.valueOf(supportsDelimitedIds); } /** * @param metaData the DatabaseMetaData to use to determine whether delimiters can be supported */ private void setSupportsDelimitedIdentifiers(DatabaseMetaData metaData) { try { supportsDelimitedIdentifiers = Boolean.valueOf( metaData.supportsMixedCaseQuotedIdentifiers() || metaData.storesLowerCaseQuotedIdentifiers() || metaData.storesUpperCaseQuotedIdentifiers()); } catch (SQLException e) { supportsDelimitedIdentifiers = Boolean.valueOf(false); getLog().warn(_loc.get("unknown-delim-support", e)); } } /** * @return the delimitIds */ public boolean getDelimitIdentifiers() { return delimitIdentifiers; } /** * @param delimitIds the delimitIds to set */ public void setDelimitIdentifiers(boolean delimitIds) { delimitIdentifiers = delimitIds; } /** * @return supportsXMLColumn */ public boolean getSupportsXMLColumn() { return supportsXMLColumn; } /** * @param b boolean representing if XML columns are supported */ public void setSupportsXMLColumn(boolean b) { supportsXMLColumn = b; } /** * @return xmlTypeEncoding */ public String getXMLTypeEncoding() { return xmlTypeEncoding; } /** * @param encoding database required JAXB encoding for the XML value */ public void setXMLTypeEncoding(String encoding) { xmlTypeEncoding = encoding; } public Log getLog() { return log; } @Override public boolean delimitAll() { return delimitIdentifiers; } @Override public String getLeadingDelimiter() { return leadingDelimiter; } public void setLeadingDelimiter(String delim) { leadingDelimiter = delim; } @Override public String getIdentifierDelimiter() { return catalogSeparator; } @Override public String getIdentifierConcatenator() { return nameConcatenator; } @Override public String getTrailingDelimiter() { return trailingDelimiter; } public void setTrailingDelimiter(String delim) { trailingDelimiter = delim; } @Override public IdentifierRule getDefaultIdentifierRule() { if (defaultNamingRule == null) { defaultNamingRule = namingRules.get(DBIdentifierType.DEFAULT.name()); } return defaultNamingRule; } @Override public IdentifierRule getIdentifierRule(T t) { if (t.equals(DBIdentifierType.DEFAULT.name())) { return getDefaultIdentifierRule(); } IdentifierRule nrule = namingRules.get(t); if (nrule == null) { return getDefaultIdentifierRule(); } return nrule; } @Override @SuppressWarnings("unchecked") public Map getIdentifierRules() { return namingRules; } /** * Returns the naming utility used by this dictionary instance */ public DBIdentifierUtil getNamingUtil() { return namingUtil; } @Override public String getDelimitedCase() { return delimitedCase; } public String toDBName(DBIdentifier name) { if (!getSupportsDelimitedIdentifiers()) return name.getName(); else return getNamingUtil().toDBName(name); } public String toDBName(DBIdentifier name, boolean delimit) { if (!getSupportsDelimitedIdentifiers()) return name.getName(); else return getNamingUtil().toDBName(name, delimit); } public DBIdentifier fromDBName(String name, DBIdentifierType id) { return getNamingUtil().fromDBName(name, id); } public void setDefaultSchemaName(String defaultSchemaName) { this.defaultSchemaName = defaultSchemaName; } public String getDefaultSchemaName() { return defaultSchemaName; } @Override public String getConversionKey() { if (conversionKey == null) { conversionKey = getLeadingDelimiter() + getIdentifierDelimiter() + getTrailingDelimiter(); } return conversionKey; } /** * Return parameter marker for INSERT and UPDATE statements. * Usually it is ? but some database-specific types might require customization. * * @param col column definition * @param val value to be inserted/updated * @return parameter marker */ public String getMarkerForInsertUpdate(Column col, Object val) { return "?"; } public String getIsNullSQL(String colAlias, int colType) { return String.format("%s IS NULL", colAlias); } public String getIsNotNullSQL(String colAlias, int colType) { return String.format("%s IS NOT NULL", colAlias); } public String getIdentityColumnName() { return null; } /** * Default behavior is ROUND */ public DateMillisecondBehaviors getMillisecondBehavior() { // If the user hasn't configured this property, fall back // to looking at the old property in the mean time till // we can get rid of it. if (dateMillisecondBehavior == null) { if (roundTimeToMillisec) { dateMillisecondBehavior = DateMillisecondBehaviors.ROUND; } else { dateMillisecondBehavior = DateMillisecondBehaviors.DROP; } } return dateMillisecondBehavior; } public void setDateMillisecondBehavior(String str) { if (str != null) { // Tolerate different case str = str.toUpperCase(Locale.ENGLISH); dateMillisecondBehavior = DateMillisecondBehaviors.valueOf(str); } else { dateMillisecondBehavior = null; } } public BooleanRepresentation getBooleanRepresentation() { return booleanRepresentation; } public void setBooleanRepresentation(String booleanRepresentationKey) { BooleanRepresentation evaluatedBooleanRepresentation = null; if (booleanRepresentationKey != null && booleanRepresentationKey.length() > 0) { ClassLoader cl = conf.getUserClassLoader(); evaluatedBooleanRepresentation = BooleanRepresentationFactory.valueOf(booleanRepresentationKey, cl); } booleanRepresentation = evaluatedBooleanRepresentation != null ? evaluatedBooleanRepresentation : BooleanRepresentationFactory.INT_10; if (log.isInfoEnabled()) { log.info(_loc.get("using-booleanRepresentation", booleanRepresentation)); } } protected boolean isUsingRange(long start, long end) { return isUsingOffset(start) || isUsingLimit(end); } protected boolean isUsingOffset(long start) { return start != 0; } protected boolean isUsingLimit(long end) { return end != Long.MAX_VALUE; } protected boolean isUsingOrderBy(SQLBuffer sql) { return sql != null && !sql.isEmpty(); } protected boolean versionEqualOrLaterThan(int maj, int min) { return (major > maj) || (major == maj && minor >= min); } protected boolean versionEqualOrEarlierThan(int maj, int min) { return (major < maj) || (major == maj && minor <= min); } protected boolean versionLaterThan(int maj) { return (major > maj); } /** * Gets major version of the database server. */ public final int getMajorVersion() { return major; } /** * Sets major version of the database server. */ public void setMajorVersion(int maj) { major = maj; } /** * Gets minor version of the database server. */ public final int getMinorVersion() { return major; } /** * Sets minor version of the database server. */ public void setMinorVersion(int min) { minor = min; } String nullSafe(String s) { return s == null ? "" : s; } public int applyRange(Select select, int count) { return count; } }