changelog.html

<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Strict//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-strict.dtd">
<!--
Copyright 2004-2009 H2 Group. Multiple-Licensed under the H2 License, Version 1.0,
and under the Eclipse Public License, Version 1.0
(http://h2database.com/html/license.html).
Initial Developer: H2 Group
-->
<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en">
<head><meta http-equiv="Content-Type" content="text/html;charset=utf-8" /><title>
Change Log
</title><link rel="stylesheet" type="text/css" href="stylesheet.css" />
<!-- [search] { -->
<script type="text/javascript" src="navigation.js"></script>
</head><body onload="frameMe();">
<table class="content"><tr class="content"><td class="content"><div class="contentDiv">
<!-- } -->

<h1>Change Log</h1>

<h2>Next Version (unreleased)</h2>
<ul><li>Page store: a rollback of a relatively large transaction could fail with an ArrayIndexOutOfBoundsException
    or a 'row not found' exception in the PageBtreeIndex in some cases.
</li><li>JaQu: the decompiler has been improved, and a few test cases already work. It is still incomplete however.
</li><li>LIKE: any letter is now allowed after the escape character (which is still '\' by default).
    Previously, an exception was thrown (unlike other databases) if it was not the escape character, '_' or '%'.
    If the escape character appears at the end of the pattern, the result is  it is ignored (like PostgreSQL and MS SQL Server).
</li><li>The reserve heap memory is no longer used.
</li><li>Fixed NullPointerException in Database.checkpoint().
</li></ul>

<h2>Version 1.2.125 (2009-12-06)</h2>
<ul><li>Lucene fulltext search: the Lucene field names now match the table column names,
    except if the column names start with _ (in which case another _ is prepended).
    Unfortunately this change means existing fulltext indexes need to be re-built.
</li><li>The shell tool now has a very simple statement history.
</li><li>The zip file system implementation now supports the '~' home directory prefix.
    Example database URL: jdbc:h2:zip:~/test.zip!/test
</li><li>Right outer joins on tables that were already 'inner joined' was processed incorrectly.
</li><li>Temporary files from LOB objects were not deleted early enough when using the server mode.
</li><li>Trying to alter a temporary table threw a strange exception.
    It is still not possible to do that, but the exception message is better now.
</li><li>When the system property h2.maxMemoryRowsDistinct was set, and
    using SELECT DISTINCT, the temporary table was not correctly dropped.
    This could cause problems in recovery when the process was killed.
</li><li>Trigger that are called before a select statement are now supported.
    This allows to create tables that work like materialized views.
</li><li>Non-row based triggers were called even if the action didn't match the
    declared action (INSERT triggers were also called when deleting rows).
    This has been changed. The MERGE statement calls both INSERT and DELETE triggers.
</li><li>Statements with IN(..) conditions could produce the wrong result or a data conversion error (since version 1.2.120).
    Examples: index on id, name, condition: id=1 and name in('Hello', 'x'); index on id, query:
    select * from (select * from test) where id=1 and name in('Hello', 'World').
</li><li>The CompressTool was not multithreading safe. Because of this, the following database
    operations where also not multithreading safe (even when using different databases): the SCRIPT command
    (only when using compression), the COMPRESS function, and storing CLOB or BLOB data (only when compression is enabled).
</li><li>The compression algorithm "LZF" is now about 33% faster than before when compressing small block
    (around 2 KB). It is much faster than Deflate, but the compression ratio is lower.
    Some of the optimizations are from Sam Van Oort, thanks a lot!
</li><li>Compressing large blocks of data didn't work when using the "Deflate" compression algorithm.
    Compressing a lot of data could run out of heap memory.
</li><li>The test cases don't access the file system directly, this simplifies GAE for Java testing.
    Thanks to Vince Bonfanti.
</li><li>More bugs in the server-less multi-connection mode have been fixed.
</li><li>When running against an old database, the SCRIPT statement could generate a
    SQL script that contained duplicate indexes (PRIMARY_KEY_E).
</li><li>JdbcConnectionPool.getConnection() could throw a NullPointerException.
</li><li>User defined functions: the source code is now available using
    SELECT SOURCE FROM INFORMATION_SCHEMA.FUNCTION_ALIASES.
</li><li>User defined functions with source code didn't work after re-opening the database.
</li><li>The newsfeeds are now Atom 1.0 standard compliant.
</li><li>The database is now closed after an out of memory exception, because
    the database could get corrupt otherwise.
</li><li>Better error message if both AUTO_SERVER and SERIALIZED parameters are set to TRUE.
</li><li>Drop table did not delete lob files in old file store (not PAGE_STORE).
</li></ul>

<h2>Version 1.2.124 (2009-11-20)</h2>
<ul><li>Clustering: there is now a way to detect which cluster instances are running.
</li><li>ConvertTraceFile: the SQL statement statistics are better formatted (newline are removed).
</li><li>The file lock thread is now stopped when the database is closed.
</li><li>Issue 140: the Script tool now supports writing to a stream.
</li><li>Issue 138: the trace output of Statement.execute(String, int) and executeUpdate was incorrect.
</li><li>Page store: new databases can not be opened with older versions.
</li><li>Page store: multi-column indexes didn't work if the cache was very small.
</li><li>Page store: opening a database could theoretically result in an endless loop.
</li><li>Page store: adding large indexed columns could get very slow.
</li><li>Page store: after a duplicate key exception, an ArrayIndexOutOfBoundsException could be thrown (only for very large rows).
</li><li>Page store: the recover tool sometimes generated a script file that contained duplicate data.
</li><li>Page store: sometimes opening a read-only database failed.
</li><li>Page store: opening a database sometimes failed if large rows where updated, or if a table was truncated before.
</li><li>Page store: when using a very small page size (128 bytes or smaller), writing a large row could result in an endless recursion. This is only a theoretical problem, as the page size is 2 KB.
</li><li>Page store: getting the min value from a descending index with NULL entries could return the wrong result.
</li><li>Page store: improved auto-recovery after power failure.
</li><li>The JDBC client did not detect that it was not talking to an H2 server. This could result in strange exceptions when trying to connect to another kind of server.
</li><li>User defined functions can be created with source code. Example: CREATE ALIAS HI AS 'String hi() { return "Hello"; }'
</li><li>Database file lock: the exception "lock file modified in the future" is no longer thrown; instead, opening the file will be delayed by 2 seconds.
</li><li>Inserting LOBs got slower each time the process was restarted. It could loop endlessly after about 1000 process restarts.
</li><li>Issue 117: Multi-version concurrency: concurrent MERGE statements now work.
</li><li>Improved read-only database detection.
</li></ul>

<h2>Version 1.2.123 (2009-11-08)</h2>
<ul><li>Page store: new databases can not be opened with older versions.
</li><li>Page store: updating large rows (rows with more than 2000 bytes of data)
    could corrupt the database.
</li><li>Page store: inserting very large rows failed with ArrayIndexOutOfBoundsException.
</li><li>When using multi-threaded kernel mode, setting disabling locking (LOCK_MODE=0)
    will now throw an exception. At this time this combination will result in corruption when
    multiple threads try to update the same table.
</li><li>The fulltext search methods and fields are now protected instead of private,
    to make the classes more easily extensible.
</li><li>The Lucene fulltext search now also returns the score.
</li><li> New function SCOPE_IDENTITY() to avoid problems when inserting
    rows in a trigger.
</li><li>Statement.getGeneratedKeys() returned the wrong value if a trigger
    changed the identity value after inserting the row.
</li><li>Improved error messages: identifiers and values are now quoted.
</li><li>Improved error message when casting a value failed:
    the column name and type is included in the message.
</li><li>Improved support for GAE for Java thanks to Vince Bonfanti.
</li></ul>

<h2>Version 1.2.122 (2009-10-28)</h2>
<ul><li>The native fulltext search now supports streaming CLOB data.
</li><li>If the database URL ends with ;PAGE_STORE=TRUE and a database in
    the old format exists, it is automatically converted to the new page store format
    if possible. A backup of the database is created first. Automatic conversion is not supported
    if the database was not closed normally (if it contains uncommitted transactions).
</li><li>Dropping the current user is now allowed if another admin user exists.
</li><li>Values of type BINARY or BLOB could not be converted to the data type OTHER.
</li><li>SHUTDOWN COMPACT now fully compacts the database.
</li><li>New system properties h2.maxCompactCount and h2.maxCompactTime
    to allow changing the default behavior (at most 2 seconds compacting when closing the database).
</li><li>New sorted insert optimization (see Performance / Database Performance Tuning).
</li><li>Issue 116: The files h2*-sources.jar and h2*-javadoc.jar are
    now in the Maven repository.
</li><li>Page store: opening a large database was slow if it was not closed before.
</li><li>Page store: new write and read counters in the meta data table. Use
    SELECT * FROM INFORMATION_SCHEMA.SETTINGS WHERE NAME IN(
    'info.FILE_WRITE_TOTAL', 'info.FILE_WRITE', 'info.FILE_READ',
    'info.CACHE_MAX_SIZE', 'info.CACHE_SIZE')
</li><li>The SQL syntax is documented using (railroad) diagrams.
    The diagrams are HTML.
</li><li>The documentation is no longer available in Japanese because the
    translation was too much out of sync. Please use the Google translation instead.
</li><li>Certain queries were not sorted if subselect queries were involved
</li><li>More bugs in the server-less multi-connection mode have been fixed:
    90097 The database is read only, caches must be cleared on reconnect, etc.
</li></ul>

<h2>Version 1.2.121 (2009-10-11)</h2>
<ul><li>Better support GaeVFS (Google App Engine Virtual File System) thanks to Thanks to Vince Bonfanti.
</li><li>CSVREAD didn't close the file. Thanks to Vince Bonfanti for the patch!
</li><li>If a database in the old format exists, it is now used.
    The system property is used for new databases, or if databases exist
    in both formats. In any case, the flag in the URL overrides this logic.
</li><li>Page store bugs were fixed. Large values in indexed columns could corrupt the index.
</li><li>The page store did not work when using Retrotranslator (because the Retrotranslator doesn't
    support Integer.reverse and Long.reverse).
</li><li>New system property h2.pageStoreTrim to disable shrinking the database when closing
    (disabled by default, meaning by default the database is trimmed).
</li></ul>

<h2>Version 1.2.120 (2009-10-04)</h2>
<ul><li>This is a beta version.
</li><li>Large updates could throw an ArrayIndexOutOfBoundsException in RowList.writeRow.
</li><li>In version 1.2, the following system properties are now enabled by default:
    h2.pageStore, h2.nullConcatIsNull, h2.optimizeInList. The default value for
    h2.defaultMaxLengthInplaceLob is now 4096 (it was 1024 with version 1.1).
</li><li>New databases are now stored in the new 'page store' file format.
    Existing databases are kept in the old file format. To use the old file format,
    append ;PAGE_STORE=FALSE to the database URL or set the system property h2.pageStore to false.
</li><li>Issue 125: Renaming primary keys was not persistent. Fixed.
    Unfortunately, databases created by this version can not be opened with older versions because of this change.
</li><li>Issue 124: Hibernate schema validation failed for decimal/numeric columns.
    This problem is fixed in the Hibernate dialect that is included with H2
    (src/tools/org/hibernate/dialect/H2Dialect.java.txt), but not in Hibernate yet.
</li><li>PostgreSQL compatibility: function LASTVAL() as an alias for IDENTITY().
</li><li>Linked tables now support default values when inserting, updating or merging.
</li><li>Bugfixes in the page store.
</li><li>Possibility to set a vendor id in Constants.java, so that unofficial builds are distinguishable
    from official releases.
</li><li>Allow writing to linked tables in readonly databases.
</li></ul>

<h2>Version 1.1.119 (2009-09-26)</h2>
<ul><li>SQL statements in the exception message are no longer included if they contain '--hide--'.
</li><li>Temporary local tables did not always work after reconnect if AUTO_SERVER=TRUE
</li><li>New system property h2.defaultMaxLengthInplaceLob to change the default maximum size
    of an in-place LOB object.
</li><li>New system property h2.nullConcatIsNull to change the default null concatenation behavior.
    The default will be enabled in version 1.2.
</li><li>The cache algorithm TQ is disabled in this version, because it is unstable, and
    because the current implementation does not have any measurable advantages over the default.
</li><li>New committer: Christian Peter. He works for <a href="http://www.docware.com">Docware</a> and
    helped a lot finding and fixing bugs, and generally improving the database. He is now a committer.
</li><li>ChangeFileEncryption did not work with Lob subdirectories. Fixed.
</li><li>Issue 121: JaQu: new simple update and merge methods.
</li><li>Issue 120: JaQu didn't close result sets.
</li><li>Issue 119: JaQu creates wrong WHERE conditions on some inputs.
</li><li>The new page store mechanism is now alpha-level quality. The next release
    will most likely be "1.2.120 beta" where this mode is enabled by default. To use
    it right now, append ;PAGE_STORE=TRUE to the database URL. The file format
    of this mode will probably not change any more.
</li><li>SELECT COUNT(*) FROM SYSTEM_RANGE(...) returned the wrong result. Fixed.
</li><li>The Recover tool now also processes the log files, however applying those changes
    is still a manual process.
</li><li>New sample application that shows how to pass data to a trigger.
</li><li>More bugs in the server-less multi-connection mode have been fixed:
    On Windows, two processes could write to the same database at the same time.
</li><li>When loading triggers or other client classes
    (static functions, database event listener, user aggregate functions, other JDBC drivers),
    the database now uses the context class loader if the class could not be found using Class.forName().
</li><li>Updating many rows with the same CLOB or BLOB values could result in FileNotFoundException.
</li><li>Statement.getConnection() threw an exception if the connection was already closed.
</li><li>The native fulltext index kept a reference to a database after the database was closed.
</li><li>Non-unique in-memory hash indexes are now supported. Thanks a lot to Sergi Vladykin for the patch!
</li><li>The optimizer does a better job for joins if indexes are missing.
</li></ul>

<h2>Version 1.1.118 (2009-09-04)</h2>
<ul><li>SHOW COLUMNS only listed indexed columns.
</li><li>When calling SHUTDOWN IMMEDIATELY in the server mode, the .trace.db file was not closed.
</li><li>DatabaseMetaData.getPrimaryKeys: the wrong constraint name was reported
    if there was another constraint on the same table and columns.
</li><li>AUTO_INCREMENT now works in the same way in ALTER TABLE ALTER COLUMN
    as in CREATE TABLE (it does not create a primary key).
</li><li>Native fulltext search: before searching, FT_INIT() had to be called.
    This is no longer required.
</li><li>Better support GaeVFS (Google App Engine Virtual File System).
</li><li>JaQu: the plan is to support natural (pure Java / Scala) conditions such as
    (id == 1 &amp;&amp; name.equals("Test")). A proof of concept decompiler is now included (it doesn't work yet).
</li><li>Various bugfixes and improvements in the page store mechanism (still experimental).
</li><li>PreparedStatement.setObject now converts a java.lang.Character to a string.
</li><li>H2 Console: PierPaolo Ucchino has completed the Italian translation. Thanks a lot!
</li><li>Various tools now use Java 5 var-args, such as main the methods and SimpleResultSet.addRow.
</li><li>H2 Console: indexes of tables of non-default schemas are now also listed.
</li><li>Issue 111: Multi-version concurrency / duplicate primary key after rollback.
</li><li>Issue 110: Multi-version concurrency / wrong exception is thrown.
</li><li>Parser: sequenceName.NEXTVAL and CURRVAL did not respect the schema search path.
</li><li>Issue 101: The following sequence could throw the exception "Row not found when trying to delete":
    start a transaction, insert many rows, delete many rows, rollback. The number of rows depends
    on the cache size.
</li><li>The stack trace of very common exceptions is no longer written to the .trace.db file by default.
</li><li>An optimization for OR is implemented, but disabled by default.
    Expressions of the type X=1 OR X=2 are converted to X IN(1, 2).
    To enable, set the system property h2.optimizeInList to true before loading the H2 JDBC driver.
</li><li>An optimization for IN(..) and IN(SELECT...) is implemented, but disabled by default.
    To enable, set the system property h2.optimizeInList to true before loading the H2 JDBC driver.
    If enabled, this overrides h2.optimizeIn and h2.optimizeInJoin. Unlike now, this optimization
    will also speed up updates and deletes.
</li></ul>

<h2>Version 1.1.117 (2009-08-09)</h2>
<ul><li>New committer: Sam Van Oort has been contributing to H2 since quite some time
    in many ways (on the mailing list, documentation, and in the form of patches).
    He is now a committer.
</li><li>JaQu: the order of the fields in the database no longer needs to match the order in the database.
</li><li>Issue 103: MVCC: the setting MAX_MEMORY_UNDO can currently not be supported when using
    multi-version concurrency, that means the complete undo log must fit in memory.
</li><li>LIKE: the escape mechanism can now be disable using ESCAPE ''.
    The default escape character can be changed using the system property h2.defaultEscape.
    The default is still '\' (as in MySQL and PostgreSQL).
</li><li>Views using functions were not re-evaluated when necessary.
</li><li>Improved MySQL compatibility for SHOW COLUMNS.
</li><li>Improved PostgreSQL compatibility for timestamp literals with timezone.
</li><li>Sergi Vladykin translated the error messages to Russian. Thanks a lot!
</li><li>Support for Java 6 DatabaseMetaData.getTables, getColumns, getProcedures, and getProcedureColumns.
</li><li>Issue 101: Rollback of a large transaction (more than 100000 rows) could fail.
</li><li>Various bugfixes and improvements in the page store mechanism (still experimental).
</li><li>The functions LENGTH, OCTET_LENGTH, and BIT_LENGTH now return BIGINT.
</li><li>Data types CLOB and BLOB: the maximum precision was Integer.MAX_VALUE, it is now Long.MAX_VALUE.
</li><li>Multi-threaded kernel: creating and dropping temporary database objects
    and the potentially free pages list was not correctly synchronized. Thanks a lot
    to Eric Faulhaber for the test case and patch!
</li><li>Parsing SQL script files is now faster.
</li><li>CSV reading is now faster.
</li><li>SimpleResultSet.newInstance(SimpleRowSource rs) did not work.
</li></ul>

<h2>Version 1.1.116 (2009-07-18)</h2>
<ul><li>Server-less multi-connection mode: more bugs are fixed.
</li><li>The built-in help (INFORMATION_SCHEMA.HELP) is smaller, shrinking the jar file size a bit.
</li><li>H2 Console: column of tables of non-default schemas are now also listed,
    except for schemas starting with 'INFO'.
</li><li>ALTER TABLE: removing an auto-increment or identity column didn't remove the sequence.
</li><li>Creating indexes is now a bit faster.
</li><li>PG Server: new system property h2.pgClientEncoding to explicitly set the encoding
    for clients that don't send the encoding (the default encoding is UTF-8).
    Thanks a lot to Sergi Vladykin for the patch!
</li><li>PG Server: improved compatibility by using the type ids of the PostgreSQL driver.
    Thanks a lot to Sergi Vladykin for the patch!
</li><li>H2 Console: Oracle system tables are no longer listed, improving performance.
</li><li>Result sets are now read-only except if the statement or prepared statement was created
    with the concurrency ResultSet.CONCUR_UPDATABLE. This change is required because the old behavior
    (all result set are updatable) violated the JDBC spec. For backward compatibility, use the
    system property h2.defaultResultSetConcurrency.
</li><li>New system property h2.defaultResultSetConcurrency to change the default result set concurrency.
</li><li>JDBC: using an invalid result set type or concurrency now throws an exception.
</li><li>If a pooled connection was not closed but garbage collected, a NullPointerException could occur.
</li><li>Fulltext search: a NullPointerException was thrown when updating a value that
    was NULL previously.
</li><li>The Recover tool did not work with .data.db files of the wrong size.
</li><li>Triggers: if there was an exception when initializing a trigger, this exception could be hidden,
    and in some cases (specially when using the Lucene fulltext index mechanism) a NullPointerException was
    thrown later on. Now the exception that occurred on init is thrown when changing data.
</li><li>The soft-references cache (CACHE_TYPE=SOFT_LRU) could throw a NullPointerException.
</li><li>To enable the new page store mechanism, append ;PAGE_STORE=TRUE to the database URL.
    or set the system property h2.pageStore to true.
    This mechanism is still experimental, and the file format will change, but it is quite stable now.
</li></ul>

<h2>Version 1.1.115 (2009-06-21)</h2>
<ul><li>The new storage mechanism is now alpha quality. To try it out, set the system property
    "h2.pageStore" to "true" (java -Dh2.pageStore=true). There are still bugs to be found and fixed,
    for example inserting many rows references a lot of main memory. Performance is currently
    about the same as with the regular storage mechanism, but the database file size is smaller.
    The file format is not stable yet.
</li><li>ALTER TABLE could throw an exception "object already exists" in some cases.
</li><li>Views: in some situations, an ArrayIndexOutOfBoundsException
    was thrown when using the same view concurrently.
</li><li>java.util.UUID is now supported in PreparedStatement.setObject and user defined Java functions.
    ResultSet.getObject() returns a java.util.UUID when using the UUID data type.
</li><li>H2 Console: the language was reset to the browser language when disconnecting.
</li><li>H2 Console: improved Polish translation.
</li><li>Server-less multi-connection mode: more bugs are fixed.
</li><li>The download page now included the SHA1 checksums.
</li><li>Shell tool: the file encoding workaround is now documented
    if you run java org.h2.tools.Shell -?.
</li><li>The RunScript tool and SQL statement did not work with the compression method LZF.
</li><li>Fulltext search: searching for NULL or an empty string threw an exception.
</li><li>Lucene fulltext search: FTL_DROP_ALL did not drop the triggers.
</li><li>Backup: if the database contained CLOB or BLOB data, the backup
    included a file entry for the LOB directory. This caused the
    restore to fail.
</li><li>Data types: LONG is now an alias for BIGINT.
</li></ul>

<h2>Version 1.1.114 (2009-06-01)</h2>
<ul><li>ResultSetMetaData.getColumnClassName returned the wrong
    class for CLOB and BLOB columns.
</li><li>Fulltext search: data is no longer deleted and
    re-inserted if the indexed columns didn't change.
</li><li>In some situations, an ArrayIndexOutOfBoundsException was thrown when adding rows.
    This was caused by a bug in the b-tree code.
</li><li>Microsoft Windows Vista: when using the the installer, Vista wrote
    "This program may not have installed correctly." This message should no longer appear
    (in the h2.nsi file, the line 'RequestExecutionLevel highest' was added).
</li><li>The Recover tool did not always work when the database contains
    referential integrity constraints.
</li><li>Java 1.5 is now required to run H2. If required, Retrotranslator can be used
    to create a Java 1.4 version (http://retrotranslator.sourceforge.net/).
</li></ul>

<h2>Version 1.1.113 (2009-05-21)</h2>
<ul><li>Shell tool: the built-in commands EXIT, HELP, ?, LIST, and so on didn't
    work with a semicolon at the end.
</li><li>JDK 1.5 is now required to build the jar file. However it is still possible to create
    a jar file for Java 1.4. For details, see buildRelease.sh and buildRelease.bat.
    As an alternative, compile using JDK 1.5 or 1.6 and use Retrotranslator to create a Java 1.4
    version (http://retrotranslator.sourceforge.net/).
</li><li>When deleting or updating many rows in a table, the space in the index
    file was not re-used in the default mode (persistent database, b-tree index,
    LOG=1). This caused the index file to grow over time. Workarounds were to
    delete and re-created the index file, alter the table (add a remove a column),
    or append ;LOG=2 to the database URL. To disable the change, set the system
    property h2.reuseSpaceBtreeIndex to false.
</li><li>Identifiers with a digit and then a dollar sign didn't work. Example: A1$B.
</li><li>MS SQL Server compatibility: support for linked tables with
    NVARCHAR, NCHAR, NCLOB, and LONGNVARCHAR.
</li><li>Android: workaround for a problem when using read-only databases in zip files
    (skip seems to be implemented incorrectly on the Android system).
</li><li>Calling execute() or prepareStatement() with null as the SQL statement
    now throws an exception.
</li><li>Benchmark: the number of executed statements was incorrect. The H2 database
    was loaded at the beginning of the test to collect results, now it is loaded at the very end.
    Thanks to Fred Toussi from HSQLDB for reporting those problems. However the changed
    do not affect the relative performance.
</li><li>H2 Console: command line settings are no longer stored in the properties file.
    They are now only used for the current process, except if they are explicitly saved.
</li><li>Cache: support for a second level soft-references cache.
    To enable it, append ;CACHE_TYPE=SOFT_LRU (or SOFT_TQ) to the database URL, or
    set the system property h2.cacheTypeDefault to "SOFT_LRU" / "SOFT_TQ".
    Enabling the second level cache reduces performance for
    small databases, but speeds up large databases. It makes sense to use it
    if the available memory size is unknown. Thanks a lot to Jan Kotek!
</li></ul>

<h2>Version 1.1.112 (2009-05-01)</h2>
<ul><li>JdbcPreparedStatement.toString() could throw a NullPointerException.
</li><li>EclipseLink: added H2Platform.supportsIdentity().
</li><li>Connection pool: the default login timeout is now 5 minutes.
</li><li>After truncating tables, opening large databases could become slow
    because indexes were always re-built unnecessarily when opening.
</li><li>More bugs in the server-less multi-connection mode have been fixed:
    Sometimes parameters of prepared statements were lost when a reconnecting.
    Concurrent read operations were slow.
    To improve performance, executeQuery(..) must be used for queries
    (execute(..) switches to the write mode, which is slow).
</li><li>GROUP BY queries with a self-join (join to the same table) that were grouped by
    columns with indexes returned the wrong result in some cases.
</li><li>Improved error message when the .lock.db file modification time is in the future.
</li><li>The MERGE statement now returns 0 as the generated key if the row was updated.
</li><li>Running code coverage is now automated.
</li><li>A file system implementation can now be registered using FileSystem.register.
</li><li>The database file system is no longer included in the jar file, it moved to the test section.
</li></ul>

<h2>Version 1.1.111 (2009-04-10)</h2>
<ul><li>In-memory databases can now run inside the Google App Engine.
</li><li>Queries that are ordered by an indexed column returned no rows in certain cases
    (if all rows were deleted from the table previously, and there is a low number of rows
    in the table, and when not using other conditions, and when using the default b tree index).
</li><li>The wrong exception was thrown when using unquoted text for
    the SQL statements COMMENT, ALTER USER, and SET PASSWORD.
</li><li>The built-in connection pool did not roll back transactions and
    enable autocommit enabled after closing a connection.
</li><li>Sometimes a StackOverflow occurred when checking for deadlock. See also
    http://code.google.com/p/h2database/issues/detail?id=61
</li><li>The Shell tool no longer truncates results with only one column, and displays
    a message if data was truncated.
</li></ul>

<h2>Version 1.1.110 (2009-04-03)</h2>
<ul><li>Support for not persistent in-memory tables in regular (persistent) databases
    using CREATE MEMORY TABLE(..) NOT PERSISTENT. Thanks a lot to Sergi Vladykin for the patch!
</li><li>The H2 Console trimmed the password (removed leading and trailing spaces).
    This is no longer the case, to support encrypted H2 database with an empty user password.
</li><li>The data type of a SUBSTRING method was wrong.
</li><li>ResultSet.findColumn and get methods with column label parameters
    now also check for matching column names (like most databases except MySQL).
</li><li>H2 Console: the browser system property now supports a list of arguments.
    Example: java -Dh2.browser="open,-a,Safari,%url" ...
</li><li>Improved Javadoc navigation (similar to Scaladoc).
</li><li>H2 Console: auto-complete of identifiers did not work correctly
    for H2 databases in MySQL mode.
</li><li>DISTINCT and GROUP BY on a CLOB column was broken.
</li><li>The FTP server moved to the tools section and is no longer included in the h2*.jar file.
</li><li>Improved error message for unsupported features:
    now the message says what exactly is not supported.
</li><li>Improved OSGi support.
</li><li>Some internal caches did not use the LRU mechanism. Fixed
    (LOB file list, optimizer cost cache, trace system, view indexes, collection keys,
    compressed in-memory file system).
</li><li>The API of the tools changed a bit (each tool now returns an exit code).
</li><li>Command line help of the tools now match the javadocs.
    The build converts the javadocs to a resource that is read by the tool at runtime.
    This should not have an effect on using the database, but it reduces duplicate
    and out-of-sync documentation.
</li><li>CREATE TABLE: improved compatibility (support for UNIQUE NOT NULL).
</li><li>DatabaseMetaData.getSQLKeywords now returns the correct list.
</li><li>Deterministic user defined functions did not work when the parameter was a column. Fixed.
</li><li>JdbcConnectionPool.setLoginTimeout with 0 now uses the default timeout.
</li><li>Creating a JdbcConnectionPool has been simplified a bit.
</li><li>The built-in connection pool did not re-use connections.
    Getting a connection using the built-in JdbcConnectionPool is now about 70 times faster
    than opening connections using DriverManager.getConnection.
</li><li>More bugs in the server-less multi-connection mode have been fixed:
    If a process terminated while writing, other open connections were blocked.
    If two processes were writing to the database, sometimes the database was corrupt after closing.
</li><li>Linked tables to SQLite database can now be created.
</li><li>Nested IN(IN(...)) didn't work.
</li><li>NIO storage: the nio: prefix was using memory mapped files instead of FileChannel.
</li></ul>

<h2>Version 1.1.109 (2009-03-14)</h2>
<ul><li>The optimization for IN(...) is now only used if comparing a column with an index.
</li><li>User defined functions can now be deterministic (see CREATE ALIAS documentation).
</li><li>Multiple nested queries in the FROM clause with parameters did not always work.
</li><li>When converting CLOB to BINARY, each character resulted in one byte.
    Now, the text is parsed as a hex as when converting VARCHAR.
</li><li>New experimental NIO storage mechanism with both FileChannel and
    memory mapped files. To use it, use the file name prefix nio: or nioMapped:
    as in jdbc:h2:nio:~/test. So far it looks like NIO storage is faster on Mac OS
    but slower on some Windows systems. Thanks a lot to Jan Kotek for the patch!
</li><li>The functions BITOR, BITAND, BITXOR, and MOD now accept
    and return BIGINT instead of INT.
</li><li>Could not use the same linked table multiple times in the same query.
</li><li>Bugs in the server-less multi-connection mode have been fixed.
</li><li>Column names could not be named "UNIQUE" (with the quotes).
</li><li>New system function TRANSACTION_ID() to get the current transaction
    identifier for a session.
</li></ul>

<h2>Version 1.1.108 (2009-02-28)</h2>
<ul><li>When the shutdown hook closed the database, the last log file
    was deleted too early. This could cause uncommitted changes to be persisted.
    In some cases, this could cause data corruption.
</li><li>JdbcConnectionPool: it was possible to set a negative connection pool size.
</li><li>Fulltext search did not support table names with a backslash.
</li><li>The internal IntArray class did not work correctly when initialized with a zero length array.
</li><li>The H2 Console web application (war file) did only support ASCII characters.
    Now UTF-8 is supported.
</li><li>DATEADD does no longer require that the argument is a timestamp.
</li><li>The database file locking mechanism didn't work correctly on Mac OS.
</li><li>Some built-in functions reported the wrong precision, scale, and display size.
</li><li>MySQL compatibility for CREATE TABLE is improved (UNSIGNED, KEY).
</li><li>Recovery did not work if there were more than 255 lobs stored as files.
</li><li>New experimental mode to support multiple read-write connections without starting
    a server. To enable this mode, append ;FILE_LOCK=SERIALIZED;OPEN_NEW=TRUE to the database URL.
    Don't expect high performance when multiple concurrent writers.
</li><li>In a web application, the database classes are not unloaded if a connection is open.
    This may cause out of memory when re-deploying a web application.
    The DbStarter is changed to close all connections to the configured database
    (by executing SHUTDOWN).
</li><li>The WebServlet did not close the database when un-deploying the web application.
</li><li>The exception message of failed INSERT or MERGE statements now includes all values and the row number.
</li><li>If opening a database failed with an out of memory exception, some files were not closed.
</li><li>Optimizer: the expected runtime calculation was incorrect. The fixed calculation
    should give slightly better query plans when using many joins.
</li><li>Improved exception message when connecting to a just started server fails.
</li><li>Connection.isValid is a bit faster.
</li><li>H2 Console: the autocomplete feature has been improved a bit. It can now better
    parse conditions.
</li><li>When restarting a web application in Tomcat, an exception was thrown sometimes.
    In most cases this was a NullPointerException. A workaround in H2 has been implemented.
    The root cause of the problem is now documented in the FAQ:
    Tomcat sets all static fields (final or non-final) to null when unloading a web application.
    A workaround is to put the h2.jar in the lib directory, or set the system property
    org.apache.catalina.loader.WebappClassLoader.ENABLE_CLEAR_REFERENCES
    to false.
</li></ul>

<h2>Version 1.1.107 (2009-01-24)</h2>
<ul><li>Some DatabaseMetaData operations did not work for non-admin users for versions 1.1.x.
</li><li>The MySQL compatibility extension fromUnixTime now used the English locale.
</li><li>When using LOG=2 and repeatedly updating the last row rows of a table, the index file grew quickly.
</li><li>In versions 1.1.105 and 1.1.106, encrypted script files of earlier versions could not be processed.
    This is now again possible. The problem was that such script files were stored in a special format
    (STORAGE=TEXT) but support for this format was removed in version 1.1.105.
</li><li>Enabling the trace mechanism by creating a specially named file is no longer supported.
</li></ul>

<h2>Version 1.1.106 (2009-01-04)</h2>
<ul><li>Statement.setQueryTimeout did not work correctly for some statements.
</li><li>CREATE DOMAIN: built-in data types can now only be changed if no tables exist.
</li><li>Linked tables: a workaround for Oracle DATE columns has been implemented.
</li><li>DatabaseMetaData.getPrimaryKeys: the column PK_NAME now contains the
    constraint name instead of the index name (compatibility for PostgreSQL and Derby).
</li><li>Using IN(..) inside a IN(SELECT..) did not always work.
</li><li>Views with IN(..) that used a view itself did not work.
</li><li>Union queries with LIMIT or ORDER BY that are used in a view or subquery did not work.
</li><li>The license change a bit: so far the license was modified to say
    'Swiss law'. This is now changed back to the original 'US law'.
    This was requested by a user, and I don't see a problem.
</li><li>Constraints for local temporary tables now session scoped. So far they were global.
    Thanks a lot to Eric Faulhaber for finding and fixing this problem!
</li><li>When using the auto-server mode, and if the lock file was modified in the future,
    the wrong exception was thrown ('Connection is broken' instead of 'Error opening database: lock file modified in the future').
</li></ul>

<h2>Version 1.1.105 (2008-12-19)</h2>
<ul><li>The setting STORAGE=TEXT is no longer supported.
</li><li>Deleting a database using the tool DeleteDbFiles deleted LOB files
    of other databases in the same directory.
</li><li>When used in a subquery, LIKE and IN(..) did not work correctly sometimes.
</li><li>The fulltext search documentation has been improved.
</li><li>ARRAY_GET returned the wrong data type (ARRAY). Now it returns VARCHAR.
</li><li>Natural join: the joined columns are not repeated any more when using SELECT *.
</li><li>User defined aggregate functions: the method getType expected internal data types
    instead of SQL types.
</li><li>User defined aggregate functions did not work if there was no group by expression.
</li><li>MySQL compatibility: support for := assignment as in @sum:=@sum+x
</li><li>INSERT INTO TEST(SELECT * FROM TEST) is now supported.
</li><li>Each session threw an invisible exception when garbage collected.
</li><li>Foreign key constraints that refer to a quoted column did not work.
</li><li>New meta data column INFORMATION_SCHEMA.TABLES.LAST_MODIFICATION to get
    the table modification counter.
</li><li>Shell: line comments didn't work correctly.
</li><li>H2 Console: columns are now listed for up to 500 tables instead of 100.
</li><li>H2 Console: Cmd+Enter executes the current statement, Alt+Space for autocomplete.
</li><li>JaQu: the maximum length of a column can now be defined using maxLength.
    For an example, see Product.java (maxLength(category, 255)).
</li><li>R&#305;dvan A&#287;ar has completed the Turkish translation of the H2 Console. Thanks a lot!
</li></ul>

<h2>Version 1.1.104 (2008-11-28)</h2>
<ul><li>If a query that was used like a table contained group by and was ordered by an expression that
    is not in the column list, an exception was thrown.
</li><li>JaQu: tables are now auto-created when running a query.
</li><li>The optimizer had problems with function tables (for example CSVREAD and FTL_SEARCH).
    A new system property h2.estimatedFunctionTableRows (default 1000) defines how many rows
    can be expected in the table.
</li><li>The function SUM could overflow when using large values. It returns now a data type that is safe.
</li><li>The function AVG could overflow when using large values. Fixed.
</li><li>The emergency reserve file has been removed. It didn't provide an appropriate
    solution for the problem. It is still possible for an application to detect and deal with
    the low disk space problem (deleting temporary files for example)
    using DatabaseEventListener.diskSpaceIsLow, but this method is now always called
    with stillAvailable=0.
</li><li>Build: JAVA_HOME is now automatically detected on Mac OS X.
</li><li>Testing for local connections was very slow on some systems.
</li><li>The cache memory usage calculation is more conservative.
</li><li>Allocating space got slower and slower the larger the database.
</li><li>ALTER TABLE ALTER COLUMN could throw the wrong exception in the last version
    (Table not found).
</li><li>Updatable result sets: the key columns can now be updated.
</li><li>The H2DatabaseProvider for ActiveObjects is now included in the tools section.
</li><li>The H2Platform for Oracle Toplink Essential has been improved a bit.
</li><li>The Windows service to start H2 didn't work in version 1.1.
</li><li>File systems with a maximum file size (for example FAT) are now supported using
    the file prefix 'split:'. In this case the files are split in parts of 1 GB.
    Example URL: jdbc:h2:split:~/db/test. If you want to split into parts of 1 MB, use
    jdbc:h2:split:20:~/db/test (the part size is 1 &lt;&lt; x, the default is 30 meaning 1 GB).
</li><li>The database now tries to detect if the classloader or virtual machine has
    almost shut down by checking if static final variables are set to null.
    This should help reduce exceptions when stopping the web application.
</li><li>Compatibility for MS SQL Server DATEDIFF(YYYY, .., ..)
</li><li>ResultSet.getObject for CLOB or BLOB will return a java.sql.Clob / java.sql.Blob object instead of
    a java.io.Reader / java.io.InputStream as in version 1.0. This behavior can be changed using the system
    property h2.returnLobObjects (true by default for version 1.1).
</li><li>The interface CloseListener has a new method 'remove' that is called when the trigger is dropped.
</li><li>Fulltext search: there was a memory leak when creating and dropping fulltext indexes in a loop.
</li></ul>

<!-- [close] { --></div></td></tr></table><!-- } --><!-- analytics --></body></html>