DatabaseMetaData.java
上传用户:blenddy
上传日期:2007-01-07
资源大小:6495k
文件大小:85k
- package postgresql.jdbc1;
- // IMPORTANT NOTE: This file implements the JDBC 1 version of the driver.
- // If you make any modifications to this file, you must make sure that the
- // changes are also made (if relevent) to the related JDBC 2 class in the
- // postgresql.jdbc2 package.
- import java.sql.*;
- import java.util.*;
- import postgresql.Field;
- /**
- * This class provides information about the database as a whole.
- *
- * <p>Many of the methods here return lists of information in ResultSets. You
- * can use the normal ResultSet methods such as getString and getInt to
- * retrieve the data from these ResultSets. If a given form of metadata is
- * not available, these methods should throw a SQLException.
- *
- * <p>Some of these methods take arguments that are String patterns. These
- * arguments all have names such as fooPattern. Within a pattern String,
- * "%" means match any substring of 0 or more characters, and "_" means
- * match any one character. Only metadata entries matching the search
- * pattern are returned. if a search pattern argument is set to a null
- * ref, it means that argument's criteria should be dropped from the
- * search.
- *
- * <p>A SQLException will be throws if a driver does not support a meta
- * data method. In the case of methods that return a ResultSet, either
- * a ResultSet (which may be empty) is returned or a SQLException is
- * thrown.
- *
- * @see java.sql.DatabaseMetaData
- */
- public class DatabaseMetaData implements java.sql.DatabaseMetaData
- {
- Connection connection; // The connection association
-
- // These define various OID's. Hopefully they will stay constant.
- static final int iVarcharOid = 1043; // OID for varchar
- static final int iBoolOid = 16; // OID for bool
- static final int iInt2Oid = 21; // OID for int2
- static final int iInt4Oid = 23; // OID for int4
- static final int VARHDRSZ = 4; // length for int4
-
- // This is a default value for remarks
- private static final byte defaultRemarks[]="no remarks".getBytes();
-
- public DatabaseMetaData(Connection conn)
- {
- this.connection = conn;
- }
-
- /**
- * Can all the procedures returned by getProcedures be called
- * by the current user?
- *
- * @return true if so
- * @exception SQLException if a database access error occurs
- */
- public boolean allProceduresAreCallable() throws SQLException
- {
- return true; // For now...
- }
-
- /**
- * Can all the tables returned by getTable be SELECTed by
- * the current user?
- *
- * @return true if so
- * @exception SQLException if a database access error occurs
- */
- public boolean allTablesAreSelectable() throws SQLException
- {
- return true; // For now...
- }
-
- /**
- * What is the URL for this database?
- *
- * @return the url or null if it cannott be generated
- * @exception SQLException if a database access error occurs
- */
- public String getURL() throws SQLException
- {
- return connection.getURL();
- }
-
- /**
- * What is our user name as known to the database?
- *
- * @return our database user name
- * @exception SQLException if a database access error occurs
- */
- public String getUserName() throws SQLException
- {
- return connection.getUserName();
- }
-
- /**
- * Is the database in read-only mode?
- *
- * @return true if so
- * @exception SQLException if a database access error occurs
- */
- public boolean isReadOnly() throws SQLException
- {
- return connection.isReadOnly();
- }
-
- /**
- * Are NULL values sorted high?
- *
- * @return true if so
- * @exception SQLException if a database access error occurs
- */
- public boolean nullsAreSortedHigh() throws SQLException
- {
- return false;
- }
-
- /**
- * Are NULL values sorted low?
- *
- * @return true if so
- * @exception SQLException if a database access error occurs
- */
- public boolean nullsAreSortedLow() throws SQLException
- {
- return false;
- }
-
- /**
- * Are NULL values sorted at the start regardless of sort order?
- *
- * @return true if so
- * @exception SQLException if a database access error occurs
- */
- public boolean nullsAreSortedAtStart() throws SQLException
- {
- return false;
- }
-
- /**
- * Are NULL values sorted at the end regardless of sort order?
- *
- * @return true if so
- * @exception SQLException if a database access error occurs
- */
- public boolean nullsAreSortedAtEnd() throws SQLException
- {
- return true;
- }
-
- /**
- * What is the name of this database product - we hope that it is
- * PostgreSQL, so we return that explicitly.
- *
- * @return the database product name
- * @exception SQLException if a database access error occurs
- */
- public String getDatabaseProductName() throws SQLException
- {
- return new String("PostgreSQL");
- }
-
- /**
- * What is the version of this database product.
- *
- * <p>Note that PostgreSQL 6.3 has a system catalog called pg_version -
- * however, select * from pg_version on any database retrieves
- * no rows.
- *
- * <p>For now, we will return the version 6.3 (in the hope that we change
- * this driver as often as we change the database)
- *
- * @return the database version
- * @exception SQLException if a database access error occurs
- */
- public String getDatabaseProductVersion() throws SQLException
- {
- return ("6.4");
- }
-
- /**
- * What is the name of this JDBC driver? If we don't know this
- * we are doing something wrong!
- *
- * @return the JDBC driver name
- * @exception SQLException why?
- */
- public String getDriverName() throws SQLException
- {
- return new String("PostgreSQL Native Driver");
- }
-
- /**
- * What is the version string of this JDBC driver? Again, this is
- * static.
- *
- * @return the JDBC driver name.
- * @exception SQLException why?
- */
- public String getDriverVersion() throws SQLException
- {
- return new String(Integer.toString(connection.this_driver.getMajorVersion())+"."+Integer.toString(connection.this_driver.getMinorVersion()));
- }
-
- /**
- * What is this JDBC driver's major version number?
- *
- * @return the JDBC driver major version
- */
- public int getDriverMajorVersion()
- {
- return connection.this_driver.getMajorVersion();
- }
-
- /**
- * What is this JDBC driver's minor version number?
- *
- * @return the JDBC driver minor version
- */
- public int getDriverMinorVersion()
- {
- return connection.this_driver.getMinorVersion();
- }
-
- /**
- * Does the database store tables in a local file? No - it
- * stores them in a file on the server.
- *
- * @return true if so
- * @exception SQLException if a database access error occurs
- */
- public boolean usesLocalFiles() throws SQLException
- {
- return false;
- }
-
- /**
- * Does the database use a file for each table? Well, not really,
- * since it doesnt use local files.
- *
- * @return true if so
- * @exception SQLException if a database access error occurs
- */
- public boolean usesLocalFilePerTable() throws SQLException
- {
- return false;
- }
-
- /**
- * Does the database treat mixed case unquoted SQL identifiers
- * as case sensitive and as a result store them in mixed case?
- * A JDBC-Compliant driver will always return false.
- *
- * <p>Predicament - what do they mean by "SQL identifiers" - if it
- * means the names of the tables and columns, then the answers
- * given below are correct - otherwise I don't know.
- *
- * @return true if so
- * @exception SQLException if a database access error occurs
- */
- public boolean supportsMixedCaseIdentifiers() throws SQLException
- {
- return false;
- }
-
- /**
- * Does the database treat mixed case unquoted SQL identifiers as
- * case insensitive and store them in upper case?
- *
- * @return true if so
- */
- public boolean storesUpperCaseIdentifiers() throws SQLException
- {
- return false;
- }
-
- /**
- * Does the database treat mixed case unquoted SQL identifiers as
- * case insensitive and store them in lower case?
- *
- * @return true if so
- */
- public boolean storesLowerCaseIdentifiers() throws SQLException
- {
- return true;
- }
-
- /**
- * Does the database treat mixed case unquoted SQL identifiers as
- * case insensitive and store them in mixed case?
- *
- * @return true if so
- */
- public boolean storesMixedCaseIdentifiers() throws SQLException
- {
- return false;
- }
-
- /**
- * Does the database treat mixed case quoted SQL identifiers as
- * case sensitive and as a result store them in mixed case? A
- * JDBC compliant driver will always return true.
- *
- * <p>Predicament - what do they mean by "SQL identifiers" - if it
- * means the names of the tables and columns, then the answers
- * given below are correct - otherwise I don't know.
- *
- * @return true if so
- * @exception SQLException if a database access error occurs
- */
- public boolean supportsMixedCaseQuotedIdentifiers() throws SQLException
- {
- return true;
- }
-
- /**
- * Does the database treat mixed case quoted SQL identifiers as
- * case insensitive and store them in upper case?
- *
- * @return true if so
- */
- public boolean storesUpperCaseQuotedIdentifiers() throws SQLException
- {
- return false;
- }
-
- /**
- * Does the database treat mixed case quoted SQL identifiers as case
- * insensitive and store them in lower case?
- *
- * @return true if so
- */
- public boolean storesLowerCaseQuotedIdentifiers() throws SQLException
- {
- return false;
- }
-
- /**
- * Does the database treat mixed case quoted SQL identifiers as case
- * insensitive and store them in mixed case?
- *
- * @return true if so
- */
- public boolean storesMixedCaseQuotedIdentifiers() throws SQLException
- {
- return false;
- }
-
- /**
- * What is the string used to quote SQL identifiers? This returns
- * a space if identifier quoting isn't supported. A JDBC Compliant
- * driver will always use a double quote character.
- *
- * <p>If an SQL identifier is a table name, column name, etc. then
- * we do not support it.
- *
- * @return the quoting string
- * @exception SQLException if a database access error occurs
- */
- public String getIdentifierQuoteString() throws SQLException
- {
- return null;
- }
-
- /**
- * Get a comma separated list of all a database's SQL keywords that
- * are NOT also SQL92 keywords.
- *
- * <p>Within PostgreSQL, the keywords are found in
- * src/backend/parser/keywords.c
- *
- * <p>For SQL Keywords, I took the list provided at
- * <a href="http://web.dementia.org/~shadow/sql/sql3bnf.sep93.txt">
- * http://web.dementia.org/~shadow/sql/sql3bnf.sep93.txt</a>
- * which is for SQL3, not SQL-92, but it is close enough for
- * this purpose.
- *
- * @return a comma separated list of keywords we use
- * @exception SQLException if a database access error occurs
- */
- public String getSQLKeywords() throws SQLException
- {
- return new String("abort,acl,add,aggregate,append,archive,arch_store,backward,binary,change,cluster,copy,database,delimiters,do,extend,explain,forward,heavy,index,inherits,isnull,light,listen,load,merge,nothing,notify,notnull,oids,purge,rename,replace,retrieve,returns,rule,recipe,setof,stdin,stdout,store,vacuum,verbose,version");
- }
-
- public String getNumericFunctions() throws SQLException
- {
- // XXX-Not Implemented
- return "";
- }
-
- public String getStringFunctions() throws SQLException
- {
- // XXX-Not Implemented
- return "";
- }
-
- public String getSystemFunctions() throws SQLException
- {
- // XXX-Not Implemented
- return "";
- }
-
- public String getTimeDateFunctions() throws SQLException
- {
- // XXX-Not Implemented
- return "";
- }
-
- /**
- * This is the string that can be used to escape '_' and '%' in
- * a search string pattern style catalog search parameters
- *
- * @return the string used to escape wildcard characters
- * @exception SQLException if a database access error occurs
- */
- public String getSearchStringEscape() throws SQLException
- {
- return new String("\");
- }
-
- /**
- * Get all the "extra" characters that can bew used in unquoted
- * identifier names (those beyond a-zA-Z0-9 and _)
- *
- * <p>From the file src/backend/parser/scan.l, an identifier is
- * {letter}{letter_or_digit} which makes it just those listed
- * above.
- *
- * @return a string containing the extra characters
- * @exception SQLException if a database access error occurs
- */
- public String getExtraNameCharacters() throws SQLException
- {
- return new String("");
- }
-
- /**
- * Is "ALTER TABLE" with an add column supported?
- * Yes for PostgreSQL 6.1
- *
- * @return true if so
- * @exception SQLException if a database access error occurs
- */
- public boolean supportsAlterTableWithAddColumn() throws SQLException
- {
- return true;
- }
-
- /**
- * Is "ALTER TABLE" with a drop column supported?
- * Yes for PostgreSQL 6.1
- *
- * @return true if so
- * @exception SQLException if a database access error occurs
- */
- public boolean supportsAlterTableWithDropColumn() throws SQLException
- {
- return true;
- }
-
- /**
- * Is column aliasing supported?
- *
- * <p>If so, the SQL AS clause can be used to provide names for
- * computed columns or to provide alias names for columns as
- * required. A JDBC Compliant driver always returns true.
- *
- * <p>e.g.
- *
- * <br><pre>
- * select count(C) as C_COUNT from T group by C;
- *
- * </pre><br>
- * should return a column named as C_COUNT instead of count(C)
- *
- * @return true if so
- * @exception SQLException if a database access error occurs
- */
- public boolean supportsColumnAliasing() throws SQLException
- {
- return true;
- }
-
- /**
- * Are concatenations between NULL and non-NULL values NULL? A
- * JDBC Compliant driver always returns true
- *
- * @return true if so
- * @exception SQLException if a database access error occurs
- */
- public boolean nullPlusNonNullIsNull() throws SQLException
- {
- return true;
- }
-
- public boolean supportsConvert() throws SQLException
- {
- // XXX-Not Implemented
- return false;
- }
-
- public boolean supportsConvert(int fromType, int toType) throws SQLException
- {
- // XXX-Not Implemented
- return false;
- }
-
- public boolean supportsTableCorrelationNames() throws SQLException
- {
- // XXX-Not Implemented
- return false;
- }
-
- public boolean supportsDifferentTableCorrelationNames() throws SQLException
- {
- // XXX-Not Implemented
- return false;
- }
-
- /**
- * Are expressions in "ORCER BY" lists supported?
- *
- * <br>e.g. select * from t order by a + b;
- *
- * @return true if so
- * @exception SQLException if a database access error occurs
- */
- public boolean supportsExpressionsInOrderBy() throws SQLException
- {
- return true;
- }
-
- /**
- * Can an "ORDER BY" clause use columns not in the SELECT?
- * I checked it, and you can't.
- *
- * @return true if so
- * @exception SQLException if a database access error occurs
- */
- public boolean supportsOrderByUnrelated() throws SQLException
- {
- return false;
- }
-
- /**
- * Is some form of "GROUP BY" clause supported?
- * I checked it, and yes it is.
- *
- * @return true if so
- * @exception SQLException if a database access error occurs
- */
- public boolean supportsGroupBy() throws SQLException
- {
- return true;
- }
-
- /**
- * Can a "GROUP BY" clause use columns not in the SELECT?
- * I checked it - it seems to allow it
- *
- * @return true if so
- * @exception SQLException if a database access error occurs
- */
- public boolean supportsGroupByUnrelated() throws SQLException
- {
- return true;
- }
-
- /**
- * Can a "GROUP BY" clause add columns not in the SELECT provided
- * it specifies all the columns in the SELECT? Does anyone actually
- * understand what they mean here?
- *
- * @return true if so
- * @exception SQLException if a database access error occurs
- */
- public boolean supportsGroupByBeyondSelect() throws SQLException
- {
- return true; // For now...
- }
-
- /**
- * Is the escape character in "LIKE" clauses supported? A
- * JDBC compliant driver always returns true.
- *
- * @return true if so
- * @exception SQLException if a database access error occurs
- */
- public boolean supportsLikeEscapeClause() throws SQLException
- {
- return true;
- }
-
- /**
- * Are multiple ResultSets from a single execute supported?
- * Well, I implemented it, but I dont think this is possible from
- * the back ends point of view.
- *
- * @return true if so
- * @exception SQLException if a database access error occurs
- */
- public boolean supportsMultipleResultSets() throws SQLException
- {
- return false;
- }
-
- /**
- * Can we have multiple transactions open at once (on different
- * connections?)
- * I guess we can have, since Im relying on it.
- *
- * @return true if so
- * @exception SQLException if a database access error occurs
- */
- public boolean supportsMultipleTransactions() throws SQLException
- {
- return true;
- }
-
- /**
- * Can columns be defined as non-nullable. A JDBC Compliant driver
- * always returns true.
- *
- * <p>This changed from false to true in v6.2 of the driver, as this
- * support was added to the backend.
- *
- * @return true if so
- * @exception SQLException if a database access error occurs
- */
- public boolean supportsNonNullableColumns() throws SQLException
- {
- return true;
- }
-
- /**
- * Does this driver support the minimum ODBC SQL grammar. This
- * grammar is defined at:
- *
- * <p><a href="http://www.microsoft.com/msdn/sdk/platforms/doc/odbc/src/intropr.htm">http://www.microsoft.com/msdn/sdk/platforms/doc/odbc/src/intropr.htm</a>
- *
- * <p>In Appendix C. From this description, we seem to support the
- * ODBC minimal (Level 0) grammar.
- *
- * @return true if so
- * @exception SQLException if a database access error occurs
- */
- public boolean supportsMinimumSQLGrammar() throws SQLException
- {
- return true;
- }
-
- /**
- * Does this driver support the Core ODBC SQL grammar. We need
- * SQL-92 conformance for this.
- *
- * @return true if so
- * @exception SQLException if a database access error occurs
- */
- public boolean supportsCoreSQLGrammar() throws SQLException
- {
- return false;
- }
-
- /**
- * Does this driver support the Extended (Level 2) ODBC SQL
- * grammar. We don't conform to the Core (Level 1), so we can't
- * conform to the Extended SQL Grammar.
- *
- * @return true if so
- * @exception SQLException if a database access error occurs
- */
- public boolean supportsExtendedSQLGrammar() throws SQLException
- {
- return false;
- }
-
- /**
- * Does this driver support the ANSI-92 entry level SQL grammar?
- * All JDBC Compliant drivers must return true. I think we have
- * to support outer joins for this to be true.
- *
- * @return true if so
- * @exception SQLException if a database access error occurs
- */
- public boolean supportsANSI92EntryLevelSQL() throws SQLException
- {
- return false;
- }
-
- /**
- * Does this driver support the ANSI-92 intermediate level SQL
- * grammar? Anyone who does not support Entry level cannot support
- * Intermediate level.
- *
- * @return true if so
- * @exception SQLException if a database access error occurs
- */
- public boolean supportsANSI92IntermediateSQL() throws SQLException
- {
- return false;
- }
-
- /**
- * Does this driver support the ANSI-92 full SQL grammar?
- *
- * @return true if so
- * @exception SQLException if a database access error occurs
- */
- public boolean supportsANSI92FullSQL() throws SQLException
- {
- return false;
- }
-
- /**
- * Is the SQL Integrity Enhancement Facility supported?
- * I haven't seen this mentioned anywhere, so I guess not
- *
- * @return true if so
- * @exception SQLException if a database access error occurs
- */
- public boolean supportsIntegrityEnhancementFacility() throws SQLException
- {
- return false;
- }
-
- /**
- * Is some form of outer join supported? From my knowledge, nope.
- *
- * @return true if so
- * @exception SQLException if a database access error occurs
- */
- public boolean supportsOuterJoins() throws SQLException
- {
- return false;
- }
-
- /**
- * Are full nexted outer joins supported? Well, we dont support any
- * form of outer join, so this is no as well
- *
- * @return true if so
- * @exception SQLException if a database access error occurs
- */
- public boolean supportsFullOuterJoins() throws SQLException
- {
- return false;
- }
-
- /**
- * Is there limited support for outer joins? (This will be true if
- * supportFullOuterJoins is true)
- *
- * @return true if so
- * @exception SQLException if a database access error occurs
- */
- public boolean supportsLimitedOuterJoins() throws SQLException
- {
- return false;
- }
-
- /**
- * What is the database vendor's preferred term for "schema" - well,
- * we do not provide support for schemas, so lets just use that
- * term.
- *
- * @return the vendor term
- * @exception SQLException if a database access error occurs
- */
- public String getSchemaTerm() throws SQLException
- {
- return new String("Schema");
- }
-
- /**
- * What is the database vendor's preferred term for "procedure" -
- * I kind of like "Procedure" myself.
- *
- * @return the vendor term
- * @exception SQLException if a database access error occurs
- */
- public String getProcedureTerm() throws SQLException
- {
- return new String("Procedure");
- }
-
- /**
- * What is the database vendor's preferred term for "catalog"? -
- * we dont have a preferred term, so just use Catalog
- *
- * @return the vendor term
- * @exception SQLException if a database access error occurs
- */
- public String getCatalogTerm() throws SQLException
- {
- return new String("Catalog");
- }
-
- /**
- * Does a catalog appear at the start of a qualified table name?
- * (Otherwise it appears at the end).
- *
- * @return true if so
- * @exception SQLException if a database access error occurs
- */
- public boolean isCatalogAtStart() throws SQLException
- {
- return false;
- }
-
- /**
- * What is the Catalog separator. Hmmm....well, I kind of like
- * a period (so we get catalog.table definitions). - I don't think
- * PostgreSQL supports catalogs anyhow, so it makes no difference.
- *
- * @return the catalog separator string
- * @exception SQLException if a database access error occurs
- */
- public String getCatalogSeparator() throws SQLException
- {
- // PM Sep 29 97 - changed from "." as we don't support catalogs.
- return new String("");
- }
-
- /**
- * Can a schema name be used in a data manipulation statement? Nope.
- *
- * @return true if so
- * @exception SQLException if a database access error occurs
- */
- public boolean supportsSchemasInDataManipulation() throws SQLException
- {
- return false;
- }
-
- /**
- * Can a schema name be used in a procedure call statement? Nope.
- *
- * @return true if so
- * @exception SQLException if a database access error occurs
- */
- public boolean supportsSchemasInProcedureCalls() throws SQLException
- {
- return false;
- }
-
- /**
- * Can a schema be used in a table definition statement? Nope.
- *
- * @return true if so
- * @exception SQLException if a database access error occurs
- */
- public boolean supportsSchemasInTableDefinitions() throws SQLException
- {
- return false;
- }
-
- /**
- * Can a schema name be used in an index definition statement?
- *
- * @return true if so
- * @exception SQLException if a database access error occurs
- */
- public boolean supportsSchemasInIndexDefinitions() throws SQLException
- {
- return false;
- }
-
- /**
- * Can a schema name be used in a privilege definition statement?
- *
- * @return true if so
- * @exception SQLException if a database access error occurs
- */
- public boolean supportsSchemasInPrivilegeDefinitions() throws SQLException
- {
- return false;
- }
-
- /**
- * Can a catalog name be used in a data manipulation statement?
- *
- * @return true if so
- * @exception SQLException if a database access error occurs
- */
- public boolean supportsCatalogsInDataManipulation() throws SQLException
- {
- return false;
- }
-
- /**
- * Can a catalog name be used in a procedure call statement?
- *
- * @return true if so
- * @exception SQLException if a database access error occurs
- */
- public boolean supportsCatalogsInProcedureCalls() throws SQLException
- {
- return false;
- }
-
- /**
- * Can a catalog name be used in a table definition statement?
- *
- * @return true if so
- * @exception SQLException if a database access error occurs
- */
- public boolean supportsCatalogsInTableDefinitions() throws SQLException
- {
- return false;
- }
-
- /**
- * Can a catalog name be used in an index definition?
- *
- * @return true if so
- * @exception SQLException if a database access error occurs
- */
- public boolean supportsCatalogsInIndexDefinitions() throws SQLException
- {
- return false;
- }
-
- /**
- * Can a catalog name be used in a privilege definition statement?
- *
- * @return true if so
- * @exception SQLException if a database access error occurs
- */
- public boolean supportsCatalogsInPrivilegeDefinitions() throws SQLException
- {
- return false;
- }
-
- /**
- * We support cursors for gets only it seems. I dont see a method
- * to get a positioned delete.
- *
- * @return true if so
- * @exception SQLException if a database access error occurs
- */
- public boolean supportsPositionedDelete() throws SQLException
- {
- return false; // For now...
- }
-
- /**
- * Is positioned UPDATE supported?
- *
- * @return true if so
- * @exception SQLException if a database access error occurs
- */
- public boolean supportsPositionedUpdate() throws SQLException
- {
- return false; // For now...
- }
-
- public boolean supportsSelectForUpdate() throws SQLException
- {
- // XXX-Not Implemented
- return false;
- }
-
- public boolean supportsStoredProcedures() throws SQLException
- {
- // XXX-Not Implemented
- return false;
- }
-
- public boolean supportsSubqueriesInComparisons() throws SQLException
- {
- // XXX-Not Implemented
- return false;
- }
-
- public boolean supportsSubqueriesInExists() throws SQLException
- {
- // XXX-Not Implemented
- return false;
- }
-
- public boolean supportsSubqueriesInIns() throws SQLException
- {
- // XXX-Not Implemented
- return false;
- }
-
- public boolean supportsSubqueriesInQuantifieds() throws SQLException
- {
- // XXX-Not Implemented
- return false;
- }
-
- public boolean supportsCorrelatedSubqueries() throws SQLException
- {
- // XXX-Not Implemented
- return false;
- }
-
- /**
- * Is SQL UNION supported? Nope.
- *
- * @return true if so
- * @exception SQLException if a database access error occurs
- */
- public boolean supportsUnion() throws SQLException
- {
- return false;
- }
-
- /**
- * Is SQL UNION ALL supported? Nope.
- *
- * @return true if so
- * @exception SQLException if a database access error occurs
- */
- public boolean supportsUnionAll() throws SQLException
- {
- return false;
- }
-
- /**
- * In PostgreSQL, Cursors are only open within transactions.
- *
- * @return true if so
- * @exception SQLException if a database access error occurs
- */
- public boolean supportsOpenCursorsAcrossCommit() throws SQLException
- {
- return false;
- }
-
- /**
- * Do we support open cursors across multiple transactions?
- *
- * @return true if so
- * @exception SQLException if a database access error occurs
- */
- public boolean supportsOpenCursorsAcrossRollback() throws SQLException
- {
- return false;
- }
-
- /**
- * Can statements remain open across commits? They may, but
- * this driver cannot guarentee that. In further reflection.
- * we are talking a Statement object jere, so the answer is
- * yes, since the Statement is only a vehicle to ExecSQL()
- *
- * @return true if they always remain open; false otherwise
- * @exception SQLException if a database access error occurs
- */
- public boolean supportsOpenStatementsAcrossCommit() throws SQLException
- {
- return true;
- }
-
- /**
- * Can statements remain open across rollbacks? They may, but
- * this driver cannot guarentee that. In further contemplation,
- * we are talking a Statement object here, so the answer is yes,
- * since the Statement is only a vehicle to ExecSQL() in Connection
- *
- * @return true if they always remain open; false otherwise
- * @exception SQLException if a database access error occurs
- */
- public boolean supportsOpenStatementsAcrossRollback() throws SQLException
- {
- return true;
- }
-
- /**
- * How many hex characters can you have in an inline binary literal
- *
- * @return the max literal length
- * @exception SQLException if a database access error occurs
- */
- public int getMaxBinaryLiteralLength() throws SQLException
- {
- return 0; // For now...
- }
-
- /**
- * What is the maximum length for a character literal
- * I suppose it is 8190 (8192 - 2 for the quotes)
- *
- * @return the max literal length
- * @exception SQLException if a database access error occurs
- */
- public int getMaxCharLiteralLength() throws SQLException
- {
- return 8190;
- }
-
- /**
- * Whats the limit on column name length. The description of
- * pg_class would say '32' (length of pg_class.relname) - we
- * should probably do a query for this....but....
- *
- * @return the maximum column name length
- * @exception SQLException if a database access error occurs
- */
- public int getMaxColumnNameLength() throws SQLException
- {
- return 32;
- }
-
- /**
- * What is the maximum number of columns in a "GROUP BY" clause?
- *
- * @return the max number of columns
- * @exception SQLException if a database access error occurs
- */
- public int getMaxColumnsInGroupBy() throws SQLException
- {
- return getMaxColumnsInTable();
- }
-
- /**
- * What's the maximum number of columns allowed in an index?
- * 6.0 only allowed one column, but 6.1 introduced multi-column
- * indices, so, theoretically, its all of them.
- *
- * @return max number of columns
- * @exception SQLException if a database access error occurs
- */
- public int getMaxColumnsInIndex() throws SQLException
- {
- return getMaxColumnsInTable();
- }
-
- /**
- * What's the maximum number of columns in an "ORDER BY clause?
- * Theoretically, all of them!
- *
- * @return the max columns
- * @exception SQLException if a database access error occurs
- */
- public int getMaxColumnsInOrderBy() throws SQLException
- {
- return getMaxColumnsInTable();
- }
-
- /**
- * What is the maximum number of columns in a "SELECT" list?
- * Theoretically, all of them!
- *
- * @return the max columns
- * @exception SQLException if a database access error occurs
- */
- public int getMaxColumnsInSelect() throws SQLException
- {
- return getMaxColumnsInTable();
- }
-
- /**
- * What is the maximum number of columns in a table? From the
- * create_table(l) manual page...
- *
- * <p>"The new class is created as a heap with no initial data. A
- * class can have no more than 1600 attributes (realistically,
- * this is limited by the fact that tuple sizes must be less than
- * 8192 bytes)..."
- *
- * @return the max columns
- * @exception SQLException if a database access error occurs
- */
- public int getMaxColumnsInTable() throws SQLException
- {
- return 1600;
- }
-
- /**
- * How many active connection can we have at a time to this
- * database? Well, since it depends on postmaster, which just
- * does a listen() followed by an accept() and fork(), its
- * basically very high. Unless the system runs out of processes,
- * it can be 65535 (the number of aux. ports on a TCP/IP system).
- * I will return 8192 since that is what even the largest system
- * can realistically handle,
- *
- * @return the maximum number of connections
- * @exception SQLException if a database access error occurs
- */
- public int getMaxConnections() throws SQLException
- {
- return 8192;
- }
-
- /**
- * What is the maximum cursor name length (the same as all
- * the other F***** identifiers!)
- *
- * @return max cursor name length in bytes
- * @exception SQLException if a database access error occurs
- */
- public int getMaxCursorNameLength() throws SQLException
- {
- return 32;
- }
-
- /**
- * What is the maximum length of an index (in bytes)? Now, does
- * the spec. mean name of an index (in which case its 32, the
- * same as a table) or does it mean length of an index element
- * (in which case its 8192, the size of a row) or does it mean
- * the number of rows it can access (in which case it 2^32 -
- * a 4 byte OID number)? I think its the length of an index
- * element, personally, so Im setting it to 8192.
- *
- * @return max index length in bytes
- * @exception SQLException if a database access error occurs
- */
- public int getMaxIndexLength() throws SQLException
- {
- return 8192;
- }
-
- public int getMaxSchemaNameLength() throws SQLException
- {
- // XXX-Not Implemented
- return 0;
- }
-
- /**
- * What is the maximum length of a procedure name?
- * (length of pg_proc.proname used) - again, I really
- * should do a query here to get it.
- *
- * @return the max name length in bytes
- * @exception SQLException if a database access error occurs
- */
- public int getMaxProcedureNameLength() throws SQLException
- {
- return 32;
- }
-
- public int getMaxCatalogNameLength() throws SQLException
- {
- // XXX-Not Implemented
- return 0;
- }
-
- /**
- * What is the maximum length of a single row? (not including
- * blobs). 8192 is defined in PostgreSQL.
- *
- * @return max row size in bytes
- * @exception SQLException if a database access error occurs
- */
- public int getMaxRowSize() throws SQLException
- {
- return 8192;
- }
-
- /**
- * Did getMaxRowSize() include LONGVARCHAR and LONGVARBINARY
- * blobs? We don't handle blobs yet
- *
- * @return true if so
- * @exception SQLException if a database access error occurs
- */
- public boolean doesMaxRowSizeIncludeBlobs() throws SQLException
- {
- return false;
- }
-
- /**
- * What is the maximum length of a SQL statement?
- *
- * @return max length in bytes
- * @exception SQLException if a database access error occurs
- */
- public int getMaxStatementLength() throws SQLException
- {
- return 8192;
- }
-
- /**
- * How many active statements can we have open at one time to
- * this database? Basically, since each Statement downloads
- * the results as the query is executed, we can have many. However,
- * we can only really have one statement per connection going
- * at once (since they are executed serially) - so we return
- * one.
- *
- * @return the maximum
- * @exception SQLException if a database access error occurs
- */
- public int getMaxStatements() throws SQLException
- {
- return 1;
- }
-
- /**
- * What is the maximum length of a table name? This was found
- * from pg_class.relname length
- *
- * @return max name length in bytes
- * @exception SQLException if a database access error occurs
- */
- public int getMaxTableNameLength() throws SQLException
- {
- return 32;
- }
-
- /**
- * What is the maximum number of tables that can be specified
- * in a SELECT? Theoretically, this is the same number as the
- * number of tables allowable. In practice tho, it is much smaller
- * since the number of tables is limited by the statement, we
- * return 1024 here - this is just a number I came up with (being
- * the number of tables roughly of three characters each that you
- * can fit inside a 8192 character buffer with comma separators).
- *
- * @return the maximum
- * @exception SQLException if a database access error occurs
- */
- public int getMaxTablesInSelect() throws SQLException
- {
- return 1024;
- }
-
- /**
- * What is the maximum length of a user name? Well, we generally
- * use UNIX like user names in PostgreSQL, so I think this would
- * be 8. However, showing the schema for pg_user shows a length
- * for username of 32.
- *
- * @return the max name length in bytes
- * @exception SQLException if a database access error occurs
- */
- public int getMaxUserNameLength() throws SQLException
- {
- return 32;
- }
-
-
- /**
- * What is the database's default transaction isolation level? We
- * do not support this, so all transactions are SERIALIZABLE.
- *
- * @return the default isolation level
- * @exception SQLException if a database access error occurs
- * @see Connection
- */
- public int getDefaultTransactionIsolation() throws SQLException
- {
- return Connection.TRANSACTION_SERIALIZABLE;
- }
-
- /**
- * Are transactions supported? If not, commit and rollback are noops
- * and the isolation level is TRANSACTION_NONE. We do support
- * transactions.
- *
- * @return true if transactions are supported
- * @exception SQLException if a database access error occurs
- */
- public boolean supportsTransactions() throws SQLException
- {
- return true;
- }
-
- /**
- * Does the database support the given transaction isolation level?
- * We only support TRANSACTION_SERIALIZABLE
- *
- * @param level the values are defined in java.sql.Connection
- * @return true if so
- * @exception SQLException if a database access error occurs
- * @see Connection
- */
- public boolean supportsTransactionIsolationLevel(int level) throws SQLException
- {
- if (level == Connection.TRANSACTION_SERIALIZABLE)
- return true;
- else
- return false;
- }
-
- /**
- * Are both data definition and data manipulation transactions
- * supported? I checked it, and could not do a CREATE TABLE
- * within a transaction, so I am assuming that we don't
- *
- * @return true if so
- * @exception SQLException if a database access error occurs
- */
- public boolean supportsDataDefinitionAndDataManipulationTransactions() throws SQLException
- {
- return false;
- }
-
- /**
- * Are only data manipulation statements withing a transaction
- * supported?
- *
- * @return true if so
- * @exception SQLException if a database access error occurs
- */
- public boolean supportsDataManipulationTransactionsOnly() throws SQLException
- {
- return true;
- }
-
- /**
- * Does a data definition statement within a transaction force
- * the transaction to commit? I think this means something like:
- *
- * <p><pre>
- * CREATE TABLE T (A INT);
- * INSERT INTO T (A) VALUES (2);
- * BEGIN;
- * UPDATE T SET A = A + 1;
- * CREATE TABLE X (A INT);
- * SELECT A FROM T INTO X;
- * COMMIT;
- * </pre><p>
- *
- * does the CREATE TABLE call cause a commit? The answer is no.
- *
- * @return true if so
- * @exception SQLException if a database access error occurs
- */
- public boolean dataDefinitionCausesTransactionCommit() throws SQLException
- {
- return false;
- }
-
- /**
- * Is a data definition statement within a transaction ignored?
- * It seems to be (from experiment in previous method)
- *
- * @return true if so
- * @exception SQLException if a database access error occurs
- */
- public boolean dataDefinitionIgnoredInTransactions() throws SQLException
- {
- return true;
- }
-
- /**
- * Get a description of stored procedures available in a catalog
- *
- * <p>Only procedure descriptions matching the schema and procedure
- * name criteria are returned. They are ordered by PROCEDURE_SCHEM
- * and PROCEDURE_NAME
- *
- * <p>Each procedure description has the following columns:
- * <ol>
- * <li><b>PROCEDURE_CAT</b> String => procedure catalog (may be null)
- * <li><b>PROCEDURE_SCHEM</b> String => procedure schema (may be null)
- * <li><b>PROCEDURE_NAME</b> String => procedure name
- * <li><b>Field 4</b> reserved (make it null)
- * <li><b>Field 5</b> reserved (make it null)
- * <li><b>Field 6</b> reserved (make it null)
- * <li><b>REMARKS</b> String => explanatory comment on the procedure
- * <li><b>PROCEDURE_TYPE</b> short => kind of procedure
- * <ul>
- * <li> procedureResultUnknown - May return a result
- * <li> procedureNoResult - Does not return a result
- * <li> procedureReturnsResult - Returns a result
- * </ul>
- * </ol>
- *
- * @param catalog - a catalog name; "" retrieves those without a
- * catalog; null means drop catalog name from criteria
- * @param schemaParrern - a schema name pattern; "" retrieves those
- * without a schema - we ignore this parameter
- * @param procedureNamePattern - a procedure name pattern
- * @return ResultSet - each row is a procedure description
- * @exception SQLException if a database access error occurs
- */
- public java.sql.ResultSet getProcedures(String catalog, String schemaPattern, String procedureNamePattern) throws SQLException
- {
- // the field descriptors for the new ResultSet
- Field f[] = new Field[8];
- java.sql.ResultSet r; // ResultSet for the SQL query that we need to do
- Vector v = new Vector(); // The new ResultSet tuple stuff
-
- byte remarks[] = defaultRemarks;
-
- f[0] = new Field(connection, "PROCEDURE_CAT", iVarcharOid, 32);
- f[1] = new Field(connection, "PROCEDURE_SCHEM", iVarcharOid, 32);
- f[2] = new Field(connection, "PROCEDURE_NAME", iVarcharOid, 32);
- f[3] = f[4] = f[5] = null; // reserved, must be null for now
- f[6] = new Field(connection, "REMARKS", iVarcharOid, 8192);
- f[7] = new Field(connection, "PROCEDURE_TYPE", iInt2Oid, 2);
-
- // If the pattern is null, then set it to the default
- if(procedureNamePattern==null)
- procedureNamePattern="%";
-
- r = connection.ExecSQL("select proname, proretset from pg_proc where proname like '"+procedureNamePattern.toLowerCase()+"' order by proname");
-
- while (r.next())
- {
- byte[][] tuple = new byte[8][0];
-
- tuple[0] = null; // Catalog name
- tuple[1] = null; // Schema name
- tuple[2] = r.getBytes(1); // Procedure name
- tuple[3] = tuple[4] = tuple[5] = null; // Reserved
- tuple[6] = remarks; // Remarks
-
- if (r.getBoolean(2))
- tuple[7] = Integer.toString(java.sql.DatabaseMetaData.procedureReturnsResult).getBytes();
- else
- tuple[7] = Integer.toString(java.sql.DatabaseMetaData.procedureNoResult).getBytes();
-
- v.addElement(tuple);
- }
- return new ResultSet(connection, f, v, "OK", 1);
- }
-
- /**
- * Get a description of a catalog's stored procedure parameters
- * and result columns.
- *
- * <p>Only descriptions matching the schema, procedure and parameter
- * name criteria are returned. They are ordered by PROCEDURE_SCHEM
- * and PROCEDURE_NAME. Within this, the return value, if any, is
- * first. Next are the parameter descriptions in call order. The
- * column descriptions follow in column number order.
- *
- * <p>Each row in the ResultSet is a parameter description or column
- * description with the following fields:
- * <ol>
- * <li><b>PROCEDURE_CAT</b> String => procedure catalog (may be null)
- * <li><b>PROCEDURE_SCHE</b>M String => procedure schema (may be null)
- * <li><b>PROCEDURE_NAME</b> String => procedure name
- * <li><b>COLUMN_NAME</b> String => column/parameter name
- * <li><b>COLUMN_TYPE</b> Short => kind of column/parameter:
- * <ul><li>procedureColumnUnknown - nobody knows
- * <li>procedureColumnIn - IN parameter
- * <li>procedureColumnInOut - INOUT parameter
- * <li>procedureColumnOut - OUT parameter
- * <li>procedureColumnReturn - procedure return value
- * <li>procedureColumnResult - result column in ResultSet
- * </ul>
- * <li><b>DATA_TYPE</b> short => SQL type from java.sql.Types
- * <li><b>TYPE_NAME</b> String => SQL type name
- * <li><b>PRECISION</b> int => precision
- * <li><b>LENGTH</b> int => length in bytes of data
- * <li><b>SCALE</b> short => scale
- * <li><b>RADIX</b> short => radix
- * <li><b>NULLABLE</b> short => can it contain NULL?
- * <ul><li>procedureNoNulls - does not allow NULL values
- * <li>procedureNullable - allows NULL values
- * <li>procedureNullableUnknown - nullability unknown
- * <li><b>REMARKS</b> String => comment describing parameter/column
- * </ol>
- * @param catalog This is ignored in postgresql, advise this is set to null
- * @param schemaPattern This is ignored in postgresql, advise this is set to null
- * @param procedureNamePattern a procedure name pattern
- * @param columnNamePattern a column name pattern
- * @return each row is a stored procedure parameter or column description
- * @exception SQLException if a database-access error occurs
- * @see #getSearchStringEscape
- */
- // Implementation note: This is required for Borland's JBuilder to work
- public java.sql.ResultSet getProcedureColumns(String catalog, String schemaPattern, String procedureNamePattern, String columnNamePattern) throws SQLException
- {
- if(procedureNamePattern==null)
- procedureNamePattern="%";
-
- if(columnNamePattern==null)
- columnNamePattern="%";
-
- // for now, this returns an empty result set.
- Field f[] = new Field[13];
- ResultSet r; // ResultSet for the SQL query that we need to do
- Vector v = new Vector(); // The new ResultSet tuple stuff
-
- f[0] = new Field(connection, new String("PROCEDURE_CAT"), iVarcharOid, 32);
- f[1] = new Field(connection, new String("PROCEDURE_SCHEM"), iVarcharOid, 32);
- f[2] = new Field(connection, new String("PROCEDURE_NAME"), iVarcharOid, 32);
- f[3] = new Field(connection, new String("COLUMN_NAME"), iVarcharOid, 32);
- f[4] = new Field(connection, new String("COLUMN_TYPE"), iInt2Oid, 2);
- f[5] = new Field(connection, new String("DATA_TYPE"), iInt2Oid, 2);
- f[6] = new Field(connection, new String("TYPE_NAME"), iVarcharOid, 32);
- f[7] = new Field(connection, new String("PRECISION"), iInt4Oid, 4);
- f[8] = new Field(connection, new String("LENGTH"), iInt4Oid, 4);
- f[9] = new Field(connection, new String("SCALE"), iInt2Oid, 2);
- f[10] = new Field(connection, new String("RADIX"), iInt2Oid, 2);
- f[11] = new Field(connection, new String("NULLABLE"), iInt2Oid, 2);
- f[12] = new Field(connection, new String("REMARKS"), iVarcharOid, 32);
-
- // add query loop here
-
- return new ResultSet(connection, f, v, "OK", 1);
- }
-
- /**
- * Get a description of tables available in a catalog.
- *
- * <p>Only table descriptions matching the catalog, schema, table
- * name and type criteria are returned. They are ordered by
- * TABLE_TYPE, TABLE_SCHEM and TABLE_NAME.
- *
- * <p>Each table description has the following columns:
- *
- * <ol>
- * <li><b>TABLE_CAT</b> String => table catalog (may be null)
- * <li><b>TABLE_SCHEM</b> String => table schema (may be null)
- * <li><b>TABLE_NAME</b> String => table name
- * <li><b>TABLE_TYPE</b> String => table type. Typical types are "TABLE",
- * "VIEW", "SYSTEM TABLE", "GLOBAL TEMPORARY", "LOCAL
- * TEMPORARY", "ALIAS", "SYNONYM".
- * <li><b>REMARKS</b> String => explanatory comment on the table
- * </ol>
- *
- * <p>The valid values for the types parameter are:
- * "TABLE", "INDEX", "LARGE OBJECT", "SEQUENCE", "SYSTEM TABLE" and
- * "SYSTEM INDEX"
- *
- * @param catalog a catalog name; For postgresql, this is ignored, and
- * should be set to null
- * @param schemaPattern a schema name pattern; For postgresql, this is ignored, and
- * should be set to null
- * @param tableNamePattern a table name pattern. For all tables this should be "%"
- * @param types a list of table types to include; null returns
- * all types
- * @return each row is a table description
- * @exception SQLException if a database-access error occurs.
- */
- public java.sql.ResultSet getTables(String catalog, String schemaPattern, String tableNamePattern, String types[]) throws SQLException
- {
- // Handle default value for types
- if(types==null)
- types = defaultTableTypes;
-
- if(tableNamePattern==null)
- tableNamePattern="%";
-
- // the field descriptors for the new ResultSet
- Field f[] = new Field[5];
- java.sql.ResultSet r; // ResultSet for the SQL query that we need to do
- Vector v = new Vector(); // The new ResultSet tuple stuff
-
- f[0] = new Field(connection, new String("TABLE_CAT"), iVarcharOid, 32);
- f[1] = new Field(connection, new String("TABLE_SCHEM"), iVarcharOid, 32);
- f[2] = new Field(connection, new String("TABLE_NAME"), iVarcharOid, 32);
- f[3] = new Field(connection, new String("TABLE_TYPE"), iVarcharOid, 32);
- f[4] = new Field(connection, new String("REMARKS"), iVarcharOid, 32);
-
- // Now form the query
- StringBuffer sql = new StringBuffer("select relname,oid from pg_class where (");
- boolean notFirst=false;
- for(int i=0;i<types.length;i++) {
- if(notFirst)
- sql.append(" or ");
- for(int j=0;j<getTableTypes.length;j++)
- if(getTableTypes[j][0].equals(types[i])) {
- sql.append(getTableTypes[j][1]);
- notFirst=true;
- }
- }
-
- // Added by Stefan Andreasen <stefan@linux.kapow.dk>
- // Now take the pattern into account
- sql.append(") and relname like '");
- sql.append(tableNamePattern.toLowerCase());
- sql.append("'");
-
- // Now run the query
- r = connection.ExecSQL(sql.toString());
-
- byte remarks[];
-
- while (r.next())
- {
- byte[][] tuple = new byte[5][0];
-
- // Fetch the description for the table (if any)
- java.sql.ResultSet dr = connection.ExecSQL("select description from pg_description where objoid="+r.getInt(2));
- if(((postgresql.ResultSet)dr).getTupleCount()==1) {
- dr.next();
- remarks = dr.getBytes(1);
- } else
- remarks = defaultRemarks;
- dr.close();
-
- tuple[0] = null; // Catalog name
- tuple[1] = null; // Schema name
- tuple[2] = r.getBytes(1); // Table name
- tuple[3] = null; // Table type
- tuple[4] = remarks; // Remarks
- v.addElement(tuple);
- }
- r.close();
- return new ResultSet(connection, f, v, "OK", 1);
- }
-
- // This array contains the valid values for the types argument
- // in getTables().
- //
- // Each supported type consists of it's name, and the sql where
- // clause to retrieve that value.
- //
- // IMPORTANT: the query must be enclosed in ( )
- private static final String getTableTypes[][] = {
- {"TABLE", "(relkind='r' and relname !~ '^pg_' and relname !~ '^xinv')"},
- {"INDEX", "(relkind='i' and relname !~ '^pg_' and relname !~ '^xinx')"},
- {"LARGE OBJECT", "(relkind='r' and relname ~ '^xinv')"},
- {"SEQUENCE", "(relkind='S' and relname !~ '^pg_')"},
- {"SYSTEM TABLE", "(relkind='r' and relname ~ '^pg_')"},
- {"SYSTEM INDEX", "(relkind='i' and relname ~ '^pg_')"}
- };
-
- // These are the default tables, used when NULL is passed to getTables
- // The choice of these provide the same behaviour as psql's d
- private static final String defaultTableTypes[] = {
- "TABLE","INDEX","SEQUENCE"
- };
-
- /**
- * Get the schema names available in this database. The results
- * are ordered by schema name.
- *
- * <P>The schema column is:
- * <OL>
- * <LI><B>TABLE_SCHEM</B> String => schema name
- * </OL>
- *
- * @return ResultSet each row has a single String column that is a
- * schema name
- */
- public java.sql.ResultSet getSchemas() throws SQLException
- {
- // We don't use schemas, so we simply return a single schema name "".
- //
- Field f[] = new Field[1];
- Vector v = new Vector();
- byte[][] tuple = new byte[1][0];
- f[0] = new Field(connection,new String("TABLE_SCHEM"),iVarcharOid,32);
- tuple[0] = "".getBytes();
- v.addElement(tuple);
- return new ResultSet(connection,f,v,"OK",1);
- }
-
- /**
- * Get the catalog names available in this database. The results
- * are ordered by catalog name.
- *
- * <P>The catalog column is:
- * <OL>
- * <LI><B>TABLE_CAT</B> String => catalog name
- * </OL>
- *
- * @return ResultSet each row has a single String column that is a
- * catalog name
- */
- public java.sql.ResultSet getCatalogs() throws SQLException
- {
- // We don't use catalogs, so we simply return a single catalog name "".
- Field f[] = new Field[1];
- Vector v = new Vector();
- byte[][] tuple = new byte[1][0];
- f[0] = new Field(connection,new String("TABLE_CAT"),iVarcharOid,32);
- tuple[0] = "".getBytes();
- v.addElement(tuple);
- return new ResultSet(connection,f,v,"OK",1);
- }
-
- /**
- * Get the table types available in this database. The results
- * are ordered by table type.
- *
- * <P>The table type is:
- * <OL>
- * <LI><B>TABLE_TYPE</B> String => table type. Typical types are "TABLE",
- * "VIEW", "SYSTEM TABLE", "GLOBAL TEMPORARY",
- * "LOCAL TEMPORARY", "ALIAS", "SYNONYM".
- * </OL>
- *
- * @return ResultSet each row has a single String column that is a
- * table type
- */
- public java.sql.ResultSet getTableTypes() throws SQLException
- {
- Field f[] = new Field[1];
- Vector v = new Vector();
- byte[][] tuple = new byte[1][0];
- f[0] = new Field(connection,new String("TABLE_TYPE"),iVarcharOid,32);
- for(int i=0;i<getTableTypes.length;i++) {
- tuple[0] = getTableTypes[i][0].getBytes();
- v.addElement(tuple);
- }
- return new ResultSet(connection,f,v,"OK",1);
- }
-
- /**
- * Get a description of table columns available in a catalog.
- *
- * <P>Only column descriptions matching the catalog, schema, table
- * and column name criteria are returned. They are ordered by
- * TABLE_SCHEM, TABLE_NAME and ORDINAL_POSITION.
- *
- * <P>Each column description has the following columns:
- * <OL>
- * <LI><B>TABLE_CAT</B> String => table catalog (may be null)
- * <LI><B>TABLE_SCHEM</B> String => table schema (may be null)
- * <LI><B>TABLE_NAME</B> String => table name
- * <LI><B>COLUMN_NAME</B> String => column name
- * <LI><B>DATA_TYPE</B> short => SQL type from java.sql.Types
- * <LI><B>TYPE_NAME</B> String => Data source dependent type name
- * <LI><B>COLUMN_SIZE</B> int => column size. For char or date
- * types this is the maximum number of characters, for numeric or
- * decimal types this is precision.
- * <LI><B>BUFFER_LENGTH</B> is not used.
- * <LI><B>DECIMAL_DIGITS</B> int => the number of fractional digits
- * <LI><B>NUM_PREC_RADIX</B> int => Radix (typically either 10 or 2)
- * <LI><B>NULLABLE</B> int => is NULL allowed?
- * <UL>
- * <LI> columnNoNulls - might not allow NULL values
- * <LI> columnNullable - definitely allows NULL values
- * <LI> columnNullableUnknown - nullability unknown
- * </UL>
- * <LI><B>REMARKS</B> String => comment describing column (may be null)
- * <LI><B>COLUMN_DEF</B> String => default value (may be null)
- * <LI><B>SQL_DATA_TYPE</B> int => unused
- * <LI><B>SQL_DATETIME_SUB</B> int => unused
- * <LI><B>CHAR_OCTET_LENGTH</B> int => for char types the
- * maximum number of bytes in the column
- * <LI><B>ORDINAL_POSITION</B> int => index of column in table
- * (starting at 1)
- * <LI><B>IS_NULLABLE</B> String => "NO" means column definitely
- * does not allow NULL values; "YES" means the column might
- * allow NULL values. An empty string means nobody knows.
- * </OL>
- *
- * @param catalog a catalog name; "" retrieves those without a catalog
- * @param schemaPattern a schema name pattern; "" retrieves those
- * without a schema
- * @param tableNamePattern a table name pattern
- * @param columnNamePattern a column name pattern
- * @return ResultSet each row is a column description
- * @see #getSearchStringEscape
- */
- public java.sql.ResultSet getColumns(String catalog, String schemaPattern, String tableNamePattern, String columnNamePattern) throws SQLException
- {
- // the field descriptors for the new ResultSet
- Field f[] = new Field[18];
- java.sql.ResultSet r; // ResultSet for the SQL query that we need to do
- Vector v = new Vector(); // The new ResultSet tuple stuff
-
- f[0] = new Field(connection, new String("TABLE_CAT"), iVarcharOid, 32);
- f[1] = new Field(connection, new String("TABLE_SCHEM"), iVarcharOid, 32);
- f[2] = new Field(connection, new String("TABLE_NAME"), iVarcharOid, 32);
- f[3] = new Field(connection, new String("COLUMN_NAME"), iVarcharOid, 32);
- f[4] = new Field(connection, new String("DATA_TYPE"), iInt2Oid, 2);
- f[5] = new Field(connection, new String("TYPE_NAME"), iVarcharOid, 32);
- f[6] = new Field(connection, new String("COLUMN_SIZE"), iInt4Oid, 4);
- f[7] = new Field(connection, new String("BUFFER_LENGTH"), iVarcharOid, 32);
- f[8] = new Field(connection, new String("DECIMAL_DIGITS"), iInt4Oid, 4);
- f[9] = new Field(connection, new String("NUM_PREC_RADIX"), iInt4Oid, 4);
- f[10] = new Field(connection, new String("NULLABLE"), iInt4Oid, 4);
- f[11] = new Field(connection, new String("REMARKS"), iVarcharOid, 32);
- f[12] = new Field(connection, new String("COLUMN_DEF"), iVarcharOid, 32);
- f[13] = new Field(connection, new String("SQL_DATA_TYPE"), iInt4Oid, 4);
- f[14] = new Field(connection, new String("SQL_DATETIME_SUB"), iInt4Oid, 4);
- f[15] = new Field(connection, new String("CHAR_OCTET_LENGTH"), iVarcharOid, 32);
- f[16] = new Field(connection, new String("ORDINAL_POSITION"), iInt4Oid,4);
- f[17] = new Field(connection, new String("IS_NULLABLE"), iVarcharOid, 32);
-
- // Added by Stefan Andreasen <stefan@linux.kapow.dk>
- // If the pattern are null then set them to %
- if (tableNamePattern == null) tableNamePattern="%";
- if (columnNamePattern == null) columnNamePattern="%";
-
- // Now form the query
- // Modified by Stefan Andreasen <stefan@linux.kapow.dk>
- r = connection.ExecSQL("select a.oid,c.relname,a.attname,a.atttypid,a.attnum,a.attnotnull,a.attlen,a.atttypmod from pg_class c, pg_attribute a where a.attrelid=c.oid and c.relname like '"+tableNamePattern.toLowerCase()+"' and a.attname like '"+columnNamePattern.toLowerCase()+"' and a.attnum>0 order by c.relname,a.attnum");
-
- byte remarks[];
-
- while(r.next()) {
- byte[][] tuple = new byte[18][0];
-
- // Fetch the description for the table (if any)
- java.sql.ResultSet dr = connection.ExecSQL("select description from pg_description where objoid="+r.getInt(1));
- if(((postgresql.ResultSet)dr).getTupleCount()==1) {
- dr.next();
- tuple[11] = dr.getBytes(1);
- } else
- tuple[11] = defaultRemarks;
-
- dr.close();
-
- tuple[0] = "".getBytes(); // Catalog name
- tuple[1] = "".getBytes(); // Schema name
- tuple[2] = r.getBytes(2); // Table name
- tuple[3] = r.getBytes(3); // Column name
-
- dr = connection.ExecSQL("select typname from pg_type where oid = "+r.getString(4));
- dr.next();
- String typname=dr.getString(1);
- dr.close();
- tuple[4] = Integer.toString(Field.getSQLType(typname)).getBytes(); // Data type
- tuple[5] = typname.getBytes(); // Type name
-
- // Column size
- // Looking at the psql source,
- // I think the length of a varchar as specified when the table was created
- // should be extracted from atttypmod which contains this length + sizeof(int32)
- if (typname.equals("bpchar") || typname.equals("varchar")) {
- int atttypmod = r.getInt(8);
- tuple[6] = Integer.toString(atttypmod != -1 ? atttypmod - VARHDRSZ : 0).getBytes();
- } else
- tuple[6] = r.getBytes(7);
-
- tuple[7] = null; // Buffer length
-
- tuple[8] = "0".getBytes(); // Decimal Digits - how to get this?
- tuple[9] = "10".getBytes(); // Num Prec Radix - assume decimal
-
- // tuple[10] is below
- // tuple[11] is above
-
- tuple[12] = null; // column default
-
- tuple[13] = null; // sql data type (unused)
- tuple[14] = null; // sql datetime sub (unused)
-
- tuple[15] = tuple[6]; // char octet length
-
- tuple[16] = r.getBytes(5); // ordinal position
-
- String nullFlag = r.getString(6);
- tuple[10] = Integer.toString(nullFlag.equals("f")?java.sql.DatabaseMetaData.columnNullable:java.sql.DatabaseMetaData.columnNoNulls).getBytes(); // Nullable
- tuple[17] = (nullFlag.equals("f")?"YES":"NO").getBytes(); // is nullable
-
- v.addElement(tuple);
- }
- r.close();
- return new ResultSet(connection, f, v, "OK", 1);
- }
-
- /**
- * Get a description of the access rights for a table's columns.
- *
- * <P>Only privileges matching the column name criteria are
- * returned. They are ordered by COLUMN_NAME and PRIVILEGE.
- *
- * <P>Each privilige description has the following columns:
- * <OL>
- * <LI><B>TABLE_CAT</B> String => table catalog (may be null)
- * <LI><B>TABLE_SCHEM</B> String => table schema (may be null)
- * <LI><B>TABLE_NAME</B> String => table name
- * <LI><B>COLUMN_NAME</B> String => column name
- * <LI><B>GRANTOR</B> => grantor of access (may be null)
- * <LI><B>GRANTEE</B> String => grantee of access
- * <LI><B>PRIVILEGE</B> String => name of access (SELECT,
- * INSERT, UPDATE, REFRENCES, ...)
- * <LI><B>IS_GRANTABLE</B> String => "YES" if grantee is permitted
- * to grant to others; "NO" if not; null if unknown
- * </OL>
- *
- * @param catalog a catalog name; "" retrieves those without a catalog
- * @param schema a schema name; "" retrieves those without a schema
- * @param table a table name
- * @param columnNamePattern a column name pattern
- * @return ResultSet each row is a column privilege description
- * @see #getSearchStringEscape
- */
- public java.sql.ResultSet getColumnPrivileges(String catalog, String schema, String table, String columnNamePattern) throws SQLException
- {
- Field f[] = new Field[8];
- Vector v = new Vector();
-
- if(table==null)
- table="%";
-
- if(columnNamePattern==null)
- columnNamePattern="%";
- else
- columnNamePattern=columnNamePattern.toLowerCase();
-
- f[0] = new Field(connection,new String("TABLE_CAT"),iVarcharOid,32);
- f[1] = new Field(connection,new String("TABLE_SCHEM"),iVarcharOid,32);
- f[2] = new Field(connection,new String("TABLE_NAME"),iVarcharOid,32);
- f[3] = new Field(connection,new String("COLUMN_NAME"),iVarcharOid,32);
- f[4] = new Field(connection,new String("GRANTOR"),iVarcharOid,32);
- f[5] = new Field(connection,new String("GRANTEE"),iVarcharOid,32);
- f[6] = new Field(connection,new String("PRIVILEGE"),iVarcharOid,32);
- f[7] = new Field(connection,new String("IS_GRANTABLE"),iVarcharOid,32);
-
- // This is taken direct from the psql source
- java.sql.ResultSet r = connection.ExecSQL("SELECT relname, relacl FROM pg_class, pg_user WHERE ( relkind = 'r' OR relkind = 'i') and relname !~ '^pg_' and relname !~ '^xin[vx][0-9]+' and usesysid = relowner and relname like '"+table.toLowerCase()+"' ORDER BY relname");
- while(r.next()) {
- byte[][] tuple = new byte[8][0];
- tuple[0] = tuple[1]= "".getBytes();
- DriverManager.println("relname=""+r.getString(1)+"" relacl=""+r.getString(2)+""");
-
- // For now, don't add to the result as relacl needs to be processed.
- //v.addElement(tuple);
- }
-
- return new ResultSet(connection,f,v,"OK",1);
- }
-
- /**
- * Get a description of the access rights for each table available
- * in a catalog.
- *
- * <P>Only privileges matching the schema and table name
- * criteria are returned. They are ordered by TABLE_SCHEM,
- * TABLE_NAME, and PRIVILEGE.
- *
- * <P>Each privilige description has the following columns:
- * <OL>
- * <LI><B>TABLE_CAT</B> String => table catalog (may be null)
- * <LI><B>TABLE_SCHEM</B> String => table schema (may be null)
- * <LI><B>TABLE_NAME</B> String => table name
- * <LI><B>COLUMN_NAME</B> String => column name
- * <LI><B>GRANTOR</B> => grantor of access (may be null)
- * <LI><B>GRANTEE</B> String => grantee of access
- * <LI><B>PRIVILEGE</B> String => name of access (SELECT,
- * INSERT, UPDATE, REFRENCES, ...)
- * <LI><B>IS_GRANTABLE</B> String => "YES" if grantee is permitted
- * to grant to others; "NO" if not; null if unknown
- * </OL>
- *
- * @param catalog a catalog name; "" retrieves those without a catalog
- * @param schemaPattern a schema name pattern; "" retrieves those
- * without a schema
- * @param tableNamePattern a table name pattern
- * @return ResultSet each row is a table privilege description
- * @see #getSearchStringEscape
- */
- public java.sql.ResultSet getTablePrivileges(String catalog, String schemaPattern, String tableNamePattern) throws SQLException
- {
- // XXX-Not Implemented
- return null;
- }
-
- /**
- * Get a description of a table's optimal set of columns that
- * uniquely identifies a row. They are ordered by SCOPE.
- *
- * <P>Each column description has the following columns:
- * <OL>
- * <LI><B>SCOPE</B> short => actual scope of result
- * <UL>
- * <LI> bestRowTemporary - very temporary, while using row
- * <LI> bestRowTransaction - valid for remainder of current transaction
- * <LI> bestRowSession - valid for remainder of current session
- * </UL>
- * <LI><B>COLUMN_NAME</B> String => column name
- * <LI><B>DATA_TYPE</B> short => SQL data type from java.sql.Types
- * <LI><B>TYPE_NAME</B> String => Data source dependent type name
- * <LI><B>COLUMN_SIZE</B> int => precision
- * <LI><B>BUFFER_LENGTH</B> int => not used
- * <LI><B>DECIMAL_DIGITS</B> short => scale
- * <LI><B>PSEUDO_COLUMN</B> short => is this a pseudo column
- * like an Oracle ROWID
- * <UL>
- * <LI> bestRowUnknown - may or may not be pseudo column
- * <LI> bestRowNotPseudo - is NOT a pseudo column
- * <LI> bestRowPseudo - is a pseudo column
- * </UL>
- * </OL>
- *
- * @param catalog a catalog name; "" retrieves those without a catalog
- * @param schema a schema name; "" retrieves those without a schema
- * @param table a table name
- * @param scope the scope of interest; use same values as SCOPE
- * @param nullable include columns that are nullable?
- * @return ResultSet each row is a column description
- */
- // Implementation note: This is required for Borland's JBuilder to work
- public java.sql.ResultSet getBestRowIdentifier(String catalog, String schema, String table, int scope, boolean nullable) throws SQLException
- {
- // for now, this returns an empty result set.
- Field f[] = new Field[8];
- ResultSet r; // ResultSet for the SQL query that we need to do
- Vector v = new Vector(); // The new ResultSet tuple stuff
-
- f[0] = new Field(connection, new String("SCOPE"), iInt2Oid, 2);
- f[1] = new Field(connection, new String("COLUMN_NAME"), iVarcharOid, 32);
- f[2] = new Field(connection, new String("DATA_TYPE"), iInt2Oid, 2);
- f[3] = new Field(connection, new String("TYPE_NAME"), iVarcharOid, 32);
- f[4] = new Field(connection, new String("COLUMN_SIZE"), iInt4Oid, 4);
- f[5] = new Field(connection, new String("BUFFER_LENGTH"), iInt4Oid, 4);
- f[6] = new Field(connection, new String("DECIMAL_DIGITS"), iInt2Oid, 2);
- f[7] = new Field(connection, new String("PSEUDO_COLUMN"), iInt2Oid, 2);
-
- return new ResultSet(connection, f, v, "OK", 1);
- }
-
- /**
- * Get a description of a table's columns that are automatically
- * updated when any value in a row is updated. They are
- * unordered.
- *
- * <P>Each column description has the following columns:
- * <OL>
- * <LI><B>SCOPE</B> short => is not used
- * <LI><B>COLUMN_NAME</B> String => column name
- * <LI><B>DATA_TYPE</B> short => SQL data type from java.sql.Types
- * <LI><B>TYPE_NAME</B> String => Data source dependent type name
- * <LI><B>COLUMN_SIZE</B> int => precision
- * <LI><B>BUFFER_LENGTH</B> int => length of column value in bytes
- * <LI><B>DECIMAL_DIGITS</B> short => scale
- * <LI><B>PSEUDO_COLUMN</B> short => is this a pseudo column
- * like an Oracle ROWID
- * <UL>
- * <LI> versionColumnUnknown - may or may not be pseudo column
- * <LI> versionColumnNotPseudo - is NOT a pseudo column
- * <LI> versionColumnPseudo - is a pseudo column
- * </UL>
- * </OL>
- *
- * @param catalog a catalog name; "" retrieves those without a catalog
- * @param schema a schema name; "" retrieves those without a schema
- * @param table a table name
- * @return ResultSet each row is a column description
- */
- public java.sql.ResultSet getVersionColumns(String catalog, String schema, String table) throws SQLException
- {
- // XXX-Not Implemented
- return null;
- }
-
- /**
- * Get a description of a table's primary key columns. They
- * are ordered by COLUMN_NAME.
- *
- * <P>Each column description has the following columns:
- * <OL>
- * <LI><B>TABLE_CAT</B> String => table catalog (may be null)
- * <LI><B>TABLE_SCHEM</B> String => table schema (may be null)
- * <LI><B>TABLE_NAME</B> String => table name
- * <LI><B>COLUMN_NAME</B> String => column name
- * <LI><B>KEY_SEQ</B> short => sequence number within primary key
- * <LI><B>PK_NAME</B> String => primary key name (may be null)
- * </OL>
- *
- * @param catalog a catalog name; "" retrieves those without a catalog
- * @param schema a schema name pattern; "" retrieves those
- * without a schema
- * @param table a table name
- * @return ResultSet each row is a primary key column description
- */
- public java.sql.ResultSet getPrimaryKeys(String catalog, String schema, String table) throws SQLException
- {
- return connection.createStatement().executeQuery("SELECT " +
- "'' as TABLE_CAT," +
- "'' AS TABLE_SCHEM," +
- "bc.relname AS TABLE_NAME," +
- "ic.relname AS COLUMN_NAME," +
- "'1' as KEY_SEQ,"+ // -- fake it as a String for now
- "t.typname as PK_NAME " +
- " FROM pg_class bc, pg_class ic, pg_index i, pg_attribute a, pg_type t " +
- " WHERE bc.relkind = 'r' " + // -- not indices
- " and bc.relname ~ '"+table+"'" +
- " and i.indrelid = bc.oid" +
- " and i.indexrelid = ic.oid" +
- " and i.indkey[0] = a.attnum" +
- " and i.indproc = '0'::oid" +
- " and a.attrelid = bc.oid" +
- " ORDER BY TABLE_NAME, COLUMN_NAME;"
- );
- }
-
- /**
- * Get a description of the primary key columns that are
- * referenced by a table's foreign key columns (the primary keys
- * imported by a table). They are ordered by PKTABLE_CAT,
- * PKTABLE_SCHEM, PKTABLE_NAME, and KEY_SEQ.
- *
- * <P>Each primary key column description has the following columns:
- * <OL>
- * <LI><B>PKTABLE_CAT</B> String => primary key table catalog
- * being imported (may be null)
- * <LI><B>PKTABLE_SCHEM</B> String => primary key table schema
- * being imported (may be null)
- * <LI><B>PKTABLE_NAME</B> String => primary key table name
- * being imported
- * <LI><B>PKCOLUMN_NAME</B> String => primary key column name
- * being imported
- * <LI><B>FKTABLE_CAT</B> String => foreign key table catalog (may be null)
- * <LI><B>FKTABLE_SCHEM</B> String => foreign key table schema (may be null)
- * <LI><B>FKTABLE_NAME</B> String => foreign key table name
- * <LI><B>FKCOLUMN_NAME</B> String => foreign key column name
- * <LI><B>KEY_SEQ</B> short => sequence number within foreign key
- * <LI><B>UPDATE_RULE</B> short => What happens to
- * foreign key when primary is updated:
- * <UL>
- * <LI> importedKeyCascade - change imported key to agree
- * with primary key update
- * <LI> importedKeyRestrict - do not allow update of primary
- * key if it has been imported
- * <LI> importedKeySetNull - change imported key to NULL if
- * its primary key has been updated
- * </UL>
- * <LI><B>DELETE_RULE</B> short => What happens to
- * the foreign key when primary is deleted.
- * <UL>
- * <LI> importedKeyCascade - delete rows that import a deleted key
- * <LI> importedKeyRestrict - do not allow delete of primary
- * key if it has been imported
- * <LI> importedKeySetNull - change imported key to NULL if
- * its primary key has been deleted
- * </UL>
- * <LI><B>FK_NAME</B> String => foreign key name (may be null)
- * <LI><B>PK_NAME</B> String => primary key name (may be null)
- * </OL>
- *
- * @param catalog a catalog name; "" retrieves those without a catalog
- * @param schema a schema name pattern; "" retrieves those
- * without a schema
- * @param table a table name
- * @return ResultSet each row is a primary key column description
- * @see #getExportedKeys
- */
- public java.sql.ResultSet getImportedKeys(String catalog, String schema, String table) throws SQLException
- {
- // XXX-Not Implemented
- return null;
- }
-
- /**
- * Get a description of a foreign key columns that reference a
- * table's primary key columns (the foreign keys exported by a
- * table). They are ordered by FKTABLE_CAT, FKTABLE_SCHEM,
- * FKTABLE_NAME, and KEY_SEQ.
- *
- * <P>Each foreign key column description has the following columns:
- * <OL>
- * <LI><B>PKTABLE_CAT</B> String => primary key table catalog (may be null)
- * <LI><B>PKTABLE_SCHEM</B> String => primary key table schema (may be null)
- * <LI><B>PKTABLE_NAME</B> String => primary key table name
- * <LI><B>PKCOLUMN_NAME</B> String => primary key column name
- * <LI><B>FKTABLE_CAT</B> String => foreign key table catalog (may be null)
- * being exported (may be null)
- * <LI><B>FKTABLE_SCHEM</B> String => foreign key table schema (may be null)
- * being exported (may be null)
- * <LI><B>FKTABLE_NAME</B> String => foreign key table name
- * being exported
- * <LI><B>FKCOLUMN_NAME</B> String => foreign key column name
- * being exported
- * <LI><B>KEY_SEQ</B> short => sequence number within foreign key
- * <LI><B>UPDATE_RULE</B> short => What happens to
- * foreign key when primary is updated:
- * <UL>
- * <LI> importedKeyCascade - change imported key to agree
- * with primary key update
- * <LI> importedKeyRestrict - do not allow update of primary
- * key if it has been imported
- * <LI> importedKeySetNull - change imported key to NULL if
- * its primary key has been updated
- * </UL>
- * <LI><B>DELETE_RULE</B> short => What happens to
- * the foreign key when primary is deleted.
- * <UL>
- * <LI> importedKeyCascade - delete rows that import a deleted key
- * <LI> importedKeyRestrict - do not allow delete of primary
- * key if it has been imported
- * <LI> importedKeySetNull - change imported key to NULL if
- * its primary key has been deleted
- * </UL>
- * <LI><B>FK_NAME</B> String => foreign key identifier (may be null)
- * <LI><B>PK_NAME</B> String => primary key identifier (may be null)
- * </OL>
- *
- * @param catalog a catalog name; "" retrieves those without a catalog
- * @param schema a schema name pattern; "" retrieves those
- * without a schema
- * @param table a table name
- * @return ResultSet each row is a foreign key column description
- * @see #getImportedKeys
- */
- public java.sql.ResultSet getExportedKeys(String catalog, String schema, String table) throws SQLException
- {
- // XXX-Not Implemented
- return null;
- }
-
- /**
- * Get a description of the foreign key columns in the foreign key
- * table that reference the primary key columns of the primary key
- * table (describe how one table imports another's key.) This
- * should normally return a single foreign key/primary key pair
- * (most tables only import a foreign key from a table once.) They
- * are ordered by FKTABLE_CAT, FKTABLE_SCHEM, FKTABLE_NAME, and
- * KEY_SEQ.
- *
- * <P>Each foreign key column description has the following columns:
- * <OL>
- * <LI><B>PKTABLE_CAT</B> String => primary key table catalog (may be null)
- * <LI><B>PKTABLE_SCHEM</B> String => primary key table schema (may be null)
- * <LI><B>PKTABLE_NAME</B> String => primary key table name
- * <LI><B>PKCOLUMN_NAME</B> String => primary key column name
- * <LI><B>FKTABLE_CAT</B> String => foreign key table catalog (may be null)
- * being exported (may be null)
- * <LI><B>FKTABLE_SCHEM</B> String => foreign key table schema (may be null)
- * being exported (may be null)
- * <LI><B>FKTABLE_NAME</B> String => foreign key table name
- * being exported
- * <LI><B>FKCOLUMN_NAME</B> String => foreign key column name
- * being exported
- * <LI><B>KEY_SEQ</B> short => sequence number within foreign key
- * <LI><B>UPDATE_RULE</B> short => What happens to
- * foreign key when primary is updated:
- * <UL>
- * <LI> importedKeyCascade - change imported key to agree
- * with primary key update
- * <LI> importedKeyRestrict - do not allow update of primary
- * key if it has been imported
- * <LI> importedKeySetNull - change imported key to NULL if
- * its primary key has been updated
- * </UL>
- * <LI><B>DELETE_RULE</B> short => What happens to
- * the foreign key when primary is deleted.
- * <UL>
- * <LI> importedKeyCascade - delete rows that import a deleted key
- * <LI> importedKeyRestrict - do not allow delete of primary
- * key if it has been imported
- * <LI> importedKeySetNull - change imported key to NULL if
- * its primary key has been deleted
- * </UL>
- * <LI><B>FK_NAME</B> String => foreign key identifier (may be null)
- * <LI><B>PK_NAME</B> String => primary key identifier (may be null)
- * </OL>
- *
- * @param catalog a catalog name; "" retrieves those without a catalog
- * @param schema a schema name pattern; "" retrieves those
- * without a schema
- * @param table a table name
- * @return ResultSet each row is a foreign key column description
- * @see #getImportedKeys
- */
- public java.sql.ResultSet getCrossReference(String primaryCatalog, String primarySchema, String primaryTable, String foreignCatalog, String foreignSchema, String foreignTable) throws SQLException
- {
- // XXX-Not Implemented
- return null;
- }
-
- /**
- * Get a description of all the standard SQL types supported by
- * this database. They are ordered by DATA_TYPE and then by how
- * closely the data type maps to the corresponding JDBC SQL type.
- *
- * <P>Each type description has the following columns:
- * <OL>
- * <LI><B>TYPE_NAME</B> String => Type name
- * <LI><B>DATA_TYPE</B> short => SQL data type from java.sql.Types
- * <LI><B>PRECISION</B> int => maximum precision
- * <LI><B>LITERAL_PREFIX</B> String => prefix used to quote a literal
- * (may be null)
- * <LI><B>LITERAL_SUFFIX</B> String => suffix used to quote a literal
- (may be null)
- * <LI><B>CREATE_PARAMS</B> String => parameters used in creating
- * the type (may be null)
- * <LI><B>NULLABLE</B> short => can you use NULL for this type?
- * <UL>
- * <LI> typeNoNulls - does not allow NULL values
- * <LI> typeNullable - allows NULL values
- * <LI> typeNullableUnknown - nullability unknown
- * </UL>
- * <LI><B>CASE_SENSITIVE</B> boolean=> is it case sensitive?
- * <LI><B>SEARCHABLE</B> short => can you use "WHERE" based on this type:
- * <UL>
- * <LI> typePredNone - No support
- * <LI> typePredChar - Only supported with WHERE .. LIKE
- * <LI> typePredBasic - Supported except for WHERE .. LIKE
- * <LI> typeSearchable - Supported for all WHERE ..
- * </UL>
- * <LI><B>UNSIGNED_ATTRIBUTE</B> boolean => is it unsigned?
- * <LI><B>FIXED_PREC_SCALE</B> boolean => can it be a money value?
- * <LI><B>AUTO_INCREMENT</B> boolean => can it be used for an
- * auto-increment value?
- * <LI><B>LOCAL_TYPE_NAME</B> String => localized version of type name
- * (may be null)
- * <LI><B>MINIMUM_SCALE</B> short => minimum scale supported
- * <LI><B>MAXIMUM_SCALE</B> short => maximum scale supported
- * <LI><B>SQL_DATA_TYPE</B> int => unused
- * <LI><B>SQL_DATETIME_SUB</B> int => unused
- * <LI><B>NUM_PREC_RADIX</B> int => usually 2 or 10
- * </OL>
- *
- * @return ResultSet each row is a SQL type description
- */
- public java.sql.ResultSet getTypeInfo() throws SQLException
- {
- java.sql.ResultSet rs = connection.ExecSQL("select typname from pg_type");
- if(rs!=null) {
- Field f[] = new Field[18];
- ResultSet r; // ResultSet for the SQL query that we need to do
- Vector v = new Vector(); // The new ResultSet tuple stuff
-
- f[0] = new Field(connection, new String("TYPE_NAME"), iVarcharOid, 32);
- f[1] = new Field(connection, new String("DATA_TYPE"), iInt2Oid, 2);
- f[2] = new Field(connection, new String("PRECISION"), iInt4Oid, 4);
- f[3] = new Field(connection, new String("LITERAL_PREFIX"), iVarcharOid, 32);
- f[4] = new Field(connection, new String("LITERAL_SUFFIX"), iVarcharOid, 32);
- f[5] = new Field(connection, new String("CREATE_PARAMS"), iVarcharOid, 32);
- f[6] = new Field(connection, new String("NULLABLE"), iInt2Oid, 2);
- f[7] = new Field(connection, new String("CASE_SENSITIVE"), iBoolOid, 1);
- f[8] = new Field(connection, new String("SEARCHABLE"), iInt2Oid, 2);
- f[9] = new Field(connection, new String("UNSIGNED_ATTRIBUTE"), iBoolOid, 1);
- f[10] = new Field(connection, new String("FIXED_PREC_SCALE"), iBoolOid, 1);
- f[11] = new Field(connection, new String("AUTO_INCREMENT"), iBoolOid, 1);
- f[12] = new Field(connection, new String("LOCAL_TYPE_NAME"), iVarcharOid, 32);
- f[13] = new Field(connection, new String("MINIMUM_SCALE"), iInt2Oid, 2);
- f[14] = new Field(connection, new String("MAXIMUM_SCALE"), iInt2Oid, 2);
- f[15] = new Field(connection, new String("SQL_DATA_TYPE"), iInt4Oid, 4);
- f[16] = new Field(connection, new String("SQL_DATETIME_SUB"), iInt4Oid, 4);
- f[17] = new Field(connection, new String("NUM_PREC_RADIX"), iInt4Oid, 4);
-
- // cache some results, this will keep memory useage down, and speed
- // things up a little.
- byte b9[] = "9".getBytes();
- byte b10[] = "10".getBytes();
- byte bf[] = "f".getBytes();
- byte bnn[] = Integer.toString(typeNoNulls).getBytes();
- byte bts[] = Integer.toString(typeSearchable).getBytes();
-
- while(rs.next()) {
- byte[][] tuple = new byte[18][];
- String typname=rs.getString(1);
- tuple[0] = typname.getBytes();
- tuple[1] = Integer.toString(Field.getSQLType(typname)).getBytes();
- tuple[2] = b9; // for now
- tuple[6] = bnn; // for now
- tuple[7] = bf; // false for now - not case sensitive
- tuple[8] = bts;
- tuple[9] = bf; // false for now - it's signed
- tuple[10] = bf; // false for now - must handle money
- tuple[11] = bf; // false for now - handle autoincrement
- // 12 - LOCAL_TYPE_NAME is null
- // 13 & 14 ?
- // 15 & 16 are unused so we return null
- tuple[17] = b10; // everything is base 10
- v.addElement(tuple);
- }
- rs.close();
- return new ResultSet(connection, f, v, "OK", 1);
- }
-
- return null;
- }
-
- /**
- * Get a description of a table's indices and statistics. They are
- * ordered by NON_UNIQUE, TYPE, INDEX_NAME, and ORDINAL_POSITION.
- *
- * <P>Each index column description has the following columns:
- * <OL>
- * <LI><B>TABLE_CAT</B> String => table catalog (may be null)
- * <LI><B>TABLE_SCHEM</B> String => table schema (may be null)
- * <LI><B>TABLE_NAME</B> String => table name
- * <LI><B>NON_UNIQUE</B> boolean => Can index values be non-unique?
- * false when TYPE is tableIndexStatistic
- * <LI><B>INDEX_QUALIFIER</B> String => index catalog (may be null);
- * null when TYPE is tableIndexStatistic
- * <LI><B>INDEX_NAME</B> String => index name; null when TYPE is
- * tableIndexStatistic
- * <LI><B>TYPE</B> short => index type:
- * <UL>
- * <LI> tableIndexStatistic - this identifies table statistics that are
- * returned in conjuction with a table's index descriptions
- * <LI> tableIndexClustered - this is a clustered index
- * <LI> tableIndexHashed - this is a hashed index
- * <LI> tableIndexOther - this is some other style of index
- * </UL>
- * <LI><B>ORDINAL_POSITION</B> short => column sequence number
- * within index; zero when TYPE is tableIndexStatistic
- * <LI><B>COLUMN_NAME</B> String => column name; null when TYPE is
- * tableIndexStatistic
- * <LI><B>ASC_OR_DESC</B> String => column sort sequence, "A" => ascending
- * "D" => descending, may be null if sort sequence is not supported;
- * null when TYPE is tableIndexStatistic
- * <LI><B>CARDINALITY</B> int => When TYPE is tableIndexStatisic then
- * this is the number of rows in the table; otherwise it is the
- * number of unique values in the index.
- * <LI><B>PAGES</B> int => When TYPE is tableIndexStatisic then
- * this is the number of pages used for the table, otherwise it
- * is the number of pages used for the current index.
- * <LI><B>FILTER_CONDITION</B> String => Filter condition, if any.
- * (may be null)
- * </OL>
- *
- * @param catalog a catalog name; "" retrieves those without a catalog
- * @param schema a schema name pattern; "" retrieves those without a schema
- * @param table a table name
- * @param unique when true, return only indices for unique values;
- * when false, return indices regardless of whether unique or not
- * @param approximate when true, result is allowed to reflect approximate
- * or out of data values; when false, results are requested to be
- * accurate
- * @return ResultSet each row is an index column description
- */
- // Implementation note: This is required for Borland's JBuilder to work
- public java.sql.ResultSet getIndexInfo(String catalog, String schema, String table, boolean unique, boolean approximate) throws SQLException
- {
- // for now, this returns an empty result set.
- Field f[] = new Field[13];
- ResultSet r; // ResultSet for the SQL query that we need to do
- Vector v = new Vector(); // The new ResultSet tuple stuff
-
- f[0] = new Field(connection, new String("TABLE_CAT"), iVarcharOid, 32);
- f[1] = new Field(connection, new String("TABLE_SCHEM"), iVarcharOid, 32);
- f[2] = new Field(connection, new String("TABLE_NAME"), iVarcharOid, 32);
- f[3] = new Field(connection, new String("NON_UNIQUE"), iBoolOid, 1);
- f[4] = new Field(connection, new String("INDEX_QUALIFIER"), iVarcharOid, 32);
- f[5] = new Field(connection, new String("INDEX_NAME"), iVarcharOid, 32);
- f[6] = new Field(connection, new String("TYPE"), iInt2Oid, 2);
- f[7] = new Field(connection, new String("ORDINAL_POSITION"), iInt2Oid, 2);
- f[8] = new Field(connection, new String("COLUMN_NAME"), iVarcharOid, 32);
- f[9] = new Field(connection, new String("ASC_OR_DESC"), iVarcharOid, 32);
- f[10] = new Field(connection, new String("CARDINALITY"), iInt4Oid, 4);
- f[11] = new Field(connection, new String("PAGES"), iInt4Oid, 4);
- f[12] = new Field(connection, new String("FILTER_CONDITION"), iVarcharOid, 32);
-
- return new ResultSet(connection, f, v, "OK", 1);
- }
- }