#####################
KInterbasDB Changelog
#####################

Version 3.3
===========

New Features
------------

    - It is now possible to use multiple transactions simultaneously on a
      single kinterbasdb.Connection.
      See discussion:
      http://sourceforge.net/forum/forum.php?thread_id=1597658&forum_id=30917

    - If a unicode object is passed as the SQL statement to any of

        + kinterbasdb.create_database
        + Connection.execute_immediate
        + Cursor.prep
        + Cursor.execute

      KInterbasDB will attempt to encode it to the character set of file system
      (in the case of kinterbasdb.create_database), or to the character set of
      the connection, in the other cases.  Previously, only unicode objects
      that could be encoded to ASCII were accepted.

    - Documentation was extended and completely redone using reStructured text
      and Sphinx (http://sphinx.pocoo.org)

Backward-incompatibilities
--------------------------

    - Default type conversion setting was changed to `type_conv=200`. Applications
      that doesn't call :func:`kinterbasdb.init()` and rely on Python older than 2.4
      and/or mx.DateTime and/or explicit unicode conversion must call 
      :func:`kinterbasdb.init()` with `type_conv=1` as first thing after `kinterbasdb`
      import. Applications that explicitly call :func:`kinterbasdb.init()` doesn't need
      to be changed at all.

      Details about new default setting are described in `Parameter Conversion` section
      `Deferred Loading of Dynamic Type Translators` of KInterbasDB documentation.


Version 3.2.2
=============

Bug Fixes
---------

    - Fixed bug with wrong Transaction Parameter Block structure. It surfaces with
      Firebird 2.1 that's more strict about TPB correctness.

    - Fixed bug with Services under Firebird 2.1.

Version 3.2.1
=============

Bug Fixes
---------

    - Inadequate compatibility with Interbase 7's boolean data type.
      All official Win32 binaries are compiled and linked against Firebird, not
      Interbase.  Typically, Interbase users either compile their own
      KInterbasDB binaries or use the official Firebird 1.0-oriented binaries.
      Previously, the latter were not fully compatible with Interbase 7's
      boolean type.

      `Thanks to rmacdurmon for reporting this problem.`

    - Conditionalized C reference to some constants not included in the
      Interbase C API.

      SF bug 1631461:
      http://sourceforge.net/tracker/index.php?func=detail&aid=1631461&group_id=9913&atid=109913

    - The "Signature Specifications for Input and Output Translators" section
      of the Usage Guide specified the Dynamic Type Translator signature of the
      TIMESTAMP type incorrectly.

      SF bug 1554643:
      http://sourceforge.net/tracker/index.php?func=detail&aid=1554643&group_id=9913&atid=109913

    - Removed unnecessary reference to <semaphore.h> that prevented KInterbasDB
      from compiling on some primitive Unixy operating systems.

    - An assertion was triggered when BLOB DTT {'mode': 'materialized'} was set
      explicitly.  This did not actually affect 3.2.0-final binaries, since
      they were delivered with assertions off, but it is now fixed.

      SF bug 1652413:
      http://sourceforge.net/tracker/index.php?func=detail&aid=1652413&group_id=9913&atid=109913

    - KInterbasDB had various problems when any of the following were installed
      at a path that contained non-ASCII characters:

        - Python
        - KInterbasDB
        - Firebird

      It now works properly.
      See discussion:
      http://sourceforge.net/forum/forum.php?thread_id=1695175&forum_id=30917

      - As a side effect of that change, SF bug 1676482 was fixed:
        http://sourceforge.net/tracker/index.php?func=detail&aid=1676482&group_id=9913&atid=109913


Version 3.2
===========

Bug Fixes
---------

    - At concurrency_level 1, it was possible for a deadlock to occur if
      KInterbasDB simultaneously raised an exception in one thread while
      executing a SQL statement in another.  This problem did not affect
      concurrency_level 2.

      `Thanks to Atsuo Ishimoto for reporting this bug.`

    - The official implementation of the automagic TEXT_UNICODE type translator
      (in the kinterbasdb.typeconv_text_unicode module) was missing support for
      the new character sets introduced in Firebird 2.0 (namely, the corrected
      version of UTF8, plus KOI8-R, KOI8-U, and WIN1258).

      `Thanks to Oleg Deribas for bringing this to my attention.`


Version 3.2rc1
==============

Bug Fixes
---------

    - KInterbasDB's "Implicit Conversion of Input Parameters from Strings" now
      accepts not only str objects, but also unicode objects, as long as
      they're convertible to ASCII.

      `Thanks to Ronald Lew for reporting this bug.`

      For general info about the "Implicit Conversion of Input Parameters
      from Strings" feature, see:
      http://kinterbasdb.sf.net/dist_docs/usage.html#adv_param_conv_implicit_from_string

New Features
------------

    - KInterbasDB now stores and retrieves the undocumented sub-second
      component of TIME and TIMESTAMP fields.

      `Thanks to Petr Jakes and Helen Borrie for bringing the availability of
      the sub-second data to my attention.`

    - Passing None to Cursor.execute (instead of a SQL string or a
      PreparedStatement) now executes the most recently prepared/executed
      statement, if any.  This can enhance convenience because it frees the
      client programmer from the responsibility of separately tracking the most
      recent statement in order to execute it again.

      `Thanks to Igor Youdytsky for suggesting this feature.`

    - PreparedStatements now have a read-only 'description' property that
      contains a Python DB API 2.0 description sequence of the same format as
      Cursor.description.

      `Thanks to Alexandr Zamaraev for suggesting this feature.`
      For more info, see:
      http://kinterbasdb.sf.net/dist_docs/usage.html#PreparedStatement_description

    - The following query and resolution methods for limbo transactions have
      been added to the kinterbasdb.services.Connection class:
      getLimboTransactionIDs, commitLimboTransaction, rollbackLimboTransaction.

Backward-incompatibilities
--------------------------

    - Dynamic Type Translators for TIME and TIMESTAMP fields must now
      accomodate an additional tuple element:  an integer which represents
      microseconds.  The official TIME and TIMESTAMP type translators in
      typeconv_datetime_naked.py, typeconv_datetime_stdlib.py, and
      typeconv_datetime_mx.py have been updated, and can be used as a guide.


Version 3.2b1
=============

Bug Fixes
---------

    - Previously, if KInterbasDB detected that the field to which a NULL value
      was bound was defined as NOT NULL, KInterbasDB itself immediately raised
      an exception.  This caused problems for fields defined as NOT NULL but
      populated by BEFORE triggers, so KInterbasDB now submits the illegal NULL
      and allows the database engine to make the decision.

    - Fixed an obscure memory leak in Connection.drop_database.

    - Fixed a few more compatibility problems with Interbase 7.x.

    - kinterbasdb.Cursor can again be used as a base class for user-defined
      subclasses.  This capability had been removed in KInterbasDB 3.2a1.

New Features
------------

    - Connection timeouts:  KInterbasDB connections can now be asked to time
      out after a specified period of inactivity.  This feature is not
      supported by the Firebird C API, so it is implemented entirely at the
      KInterbasDB level.
      For more info, see:
      http://kinterbasdb.sf.net/dist_docs/usage.html#adv_ct

    - Added a TPB class to assist with the construction of complex Transaction
      Parameter Buffers.

      This feature has not yet been documented.  In the meantime, you can
      find example code in the test_transactions.py module of the KInterbasDB
      test suite:
      http://kinterbasdb.sf.net/test-suite/releases/

    - Added methods Connection.transaction_info and Connection.trans_info.
      transaction_info is a thin wrapper around the C function
      isc_transaction_info, while trans_info is a Pythonic wrapper around
      transaction_info.

      This feature has not yet been documented.  In the meantime, you can
      find example code in the test_services.py module of the KInterbasDB test
      suite:
      http://kinterbasdb.sf.net/test-suite/releases/

    - Exposed the Firebird header constant FB_API_VER as
      kinterbasdb.FB_API_VER.  This integer represents the version of Firebird
      against which KInterbasDB was compiled, as follows:

        - Any version of Interbase, or Firebird 1.0.x:      10
        - Firebird 1.5.x:                                   15
        - Firebird 2.0.x:                                   20

    - KInterbasDB now raises a kinterbasdb.TransactionConflict exception
      (instead of the rather generic ProgrammingError) when it receives a
      server-side transaction conflict notification.  This makes it easier for
      the client programmer to detect and resolve deadlocks.

      TransactionConflict is a subclass of kinterbasdb.OperationalError.

Backward-incompatibilities
--------------------------

    - Client programs that encounter transaction conflicts in routine
      operation, *and* which contain logic to deal with this type of exception
      specifically (on the basis of the payload of the ProgrammingError) should
      be updated to use::

        try:
          ...
        except kinterbasdb.TransactionConflict:
          ...

      instead.

      For more info, see the last item under "New Features" above.


Version 3.2a1
=============

New Features
------------

    - Better concurrency support.  This requires a capable database client
      library, and must be explicitly activated by specifying keyword argument
      concurrency_level to function kinterbasdb.init.
      For more info, see:
      http://kinterbasdb.sf.net/dist_docs/usage.html#special_issue_concurrency

    - Streaming blob support (blobs can now be handled without fully
      materializing them in memory).
      For more info, see:
      http://kinterbasdb.sf.net/dist_docs/usage.html#adv_param_conv_blobs

    - Support for manual creation and manipulation of prepared statements.
      KInterbasDB has always used prepared statements under the hood, but now
      they're directly manipulable by the client programmer.  The implicit
      prepared statement cache has also become more sophisticated.
      For more info, see:
      http://kinterbasdb.sf.net/dist_docs/usage.html#adv_prepared_statements

    - Database event system entirely rewritten.  Client programs can now create
      any number of EventConduits per process, and a given conduit can listen
      for more than 15 events (the limit is in the hundreds, but the specific
      limit depends on the database server version and operating system rather
      than on KInterbasDB).
      For more info, see:
      http://kinterbasdb.sf.net/dist_docs/usage.html#adv_event

    - Added method Connection.db_info, a Pythonic wrapper around
      Connection.database_info.  Thanks to Pavel Cisar for implementing this.
      For more info, see:
      http://kinterbasdb.sf.net/dist_docs/usage.html#adv_prog_maint_db_info

    - Positional Dynamic Type Translation:  It's now possible to specify DTT
      settings at the Cursor-column level in addition to the previously
      available levels.
      For more info, see:
      http://kinterbasdb.sf.net/dist_docs/usage.html#adv_param_conv_dynamic_type_translation_positional

    - Added official support for the Python 2.4+ standard library decimal
      module, including a new kinterbasdb.init(type_conv=200) convenience code.
      For more info, see:
      http://kinterbasdb.sf.net/dist_docs/usage.html#adv_param_conv_dynamic_type_translation_deferred_loading

    - KInterbasDB now detects the "PSQL Stack Trace" generated by FB 2.0+, and
      includes a nicely formatted rendition of the stack trace in the exception
      message.
      For more info, see:
      Firebird 2.0 Release Notes, section "PSQL Stack Trace"


Bug Fixes
---------

    - KInterbasDB should now compile and run out of the box with Interbase
      7.x.  DSR doesn't have that version of Interbase, however, so KInterbasDB
      is not actually tested with it.


Backward-incompatibilities
--------------------------

    - KInterbasDB 3.2 has dropped support for Python versions earlier than 2.3,
      and (officially) supports only Firebird 1.0 and later.  However,
      Interbase 7.x support has been considerably enhanced, so it could be said
      that Interbase is "unofficially" supported.

    - Most Python classes in KInterbasDB have become new-style, for symmetry
      with the new-style C classes added in 3.2.  Notably, kinterbasdb.Cursor
      is now a new-style class written in pure C.

        Impact rating:  Low (There is practically no reason for a client
        program to access the affected KInterbasDB classes in such a way that
        this change would matter.)

    - Previously, the "infinite timeout value" for EventConduit.wait was 0.0.
      The choice of that value was a terrible mistake, because attempting to
      specify an extremely short timeout with a value such as 0.000000000000001
      in fact created an infinite timeout.  The new "infinite timeout value" is
      -1.0.

        Impact rating:  Low-Medium (The Usage Guide for KInterbasDB 3.1
        specified that "The default timeout is infinite.", but it did not
        guarantee a particular value.  Client programs that use both events and
        event timeouts should be checked, however.)


Version 3.1.3
=============

Bug Fixes
---------

    - Fixed leak of weak reference objects used by a Connection to track its
      Cursors.  These objects were collected when the Connection closed, but
      never earlier.

      `Thanks to Robby Dermody for reporting this bug.`
      Refs:
      http://sf.net/forum/forum.php?thread_id=1380653&forum_id=30917

    - The database engine's Services API makes no provision for Unicode
      handling.  kinterbasdb should have gracefully rejected Python unicode
      objects submitted to the Services API, but instead, it choked.

      `Thanks to Garrett Smith for reporting this bug.`
      Refs:
      http://sf.net/forum/forum.php?thread_id=1366918&forum_id=30917


Version 3.1.2
=============

Bug Fixes
---------

    - Attempting to apply dynamic type translation settings dictionaries that
      had non-string keys caused a segfault under some some circumstances.

    - kinterbasdb's Services API infrastructure parsed integer results from
      the engine as recommended by the IB 6 API Guide, but this was
      inappropriate on non-Windows x86-64, and could cause invalid memory
      access.

    - Input handling of INTEGER ARRAY fields did not work correctly on
      non-Windows x86-64.

    - Overridding a connection's dynamic type translation settings for a
      particular slot with the "naked" translator by passing None as the
      translator to Cursor.set_type_trans_[in|out] did not work.

    - The FIXED dynamic type translation slot was not handled properly on
      dialect 1 connections (dialect 1 databases store NUMERIC/DECIMAL values
      with precisions 10-18 internally as floating point).

    - Documentation bug:  The "Using KInterbasDB with Embedded Firebird"
      section of the Usage Guide stated that the Services API did not work with
      the embedded server architecture.  That was written when Firebird 1.5 was
      in alpha; the Services API *does* work with embedded Firebird 1.5.2.


Version 3.1.1
=============

Bug Fixes
---------

    - kinterbasdb.init(type_conv=100|200) didn't work under Python 2.4.
      Late in the Python 2.4 development cycle, additional constraints were
      introduced in Python's funcobject.c that defeated kinterbasdb's attempts
      to manipulate the 'func_code' attribute of some functions during the
      execution of kinterbasdb.init.

    - C type ConnectionType's destructor was called directly (rather than as a
      result of DECREF) if an error arose in kinterbasdb.connect or
      kinterbasdb.create_database.  This triggered a refcount assertion in
      debug builds of Python.

    - Fixed a reference count leak in the C layer's central exception-raising
      function.

    - Fixed some potential memory handling problems in exceptional situations
      in the event handling code.

    - A trivial problem prevented kinterbasdb 3.1 from compiling with the
      prereleases of Firebird 2.0.


New Features
------------

    - In typeconv_text_unicode.py, enabled auto-translation of some Asian
      Unicode codecs that didn't enter the Python standard library until Python
      2.4.


Version 3.1
===========

Bug Fixes
---------

    - Fixed minor problems with the Connection.database_info method.


Version 3.1_pre9
================

Version 3.1_pre9 is being released instead of 3.1 final primarily to test
Python 2.4 compatibility.  Since the first beta of Python 2.4 has now been
released, it is expected that these binaries will continue to work throughout
2.4's lifespan (including maintenance releases - 2.4.x).

New Features
------------

    - Python 2.4 support (that is, a few build script changes and the
      availability of official Windows binaries for Python 2.4).

Bug Fixes
---------

    - kinterbasdb sometimes caused an exception to be raised during the Python
      interpreter's shutdown process.
      Refs:
      http://sourceforge.net/tracker/index.php?func=detail&aid=1011513&group_id=9913&atid=109913

    - Fixed a potential concurrency problem regarding memory allocation in
      kinterbasdb's event handling code.


Version 3.1_pre8
================

Version 3.1_pre8 is the recommended stable version of kinterbasdb.

New Features
------------

    - kinterbasdb._RowMapping has a richer dict-like interface (now implements
      __len__, __getitem__, get, __contains__, keys, values, items, __iter__,
      iterkeys, itervalues, iteritems).

Bug Fixes
---------

    - The kinterbasdb.typeconv_fixed_fixedpoint.fixed_conv_out_precise dynamic
      type translator was unable to convert some NUMERIC/DECIMAL database
      values.
      Refs:
      http://sourceforge.net/tracker/index.php?func=detail&aid=949669&group_id=9913&atid=109913

    - The kinterbasdb.typeconv_text_unicode.unicode_conv_[in|out] dynamic type
      translators did not work with non-default collations.
      Refs:
      http://sourceforge.net/tracker/index.php?func=detail&aid=876564&group_id=9913&atid=109913

    - The kinterbasdb.services.Connection.getLog method should not have
      accepted a database parameter; it no longer does.

    - The kinterbasdb.services.Connection.backup method now returns a
      gbak-style log string (as the kinterbasdb.services.Connection.restore
      method has done all along).

    - Applied Mac OS X compatibility patch to setup.py.
      Refs:
      http://sourceforge.net/tracker/index.php?func=detail&aid=909886&group_id=9913&atid=309913

    - Ported to the AMD64 architecture.  Tested with a prerelease version of
      Firebird 1.5.1/AMD64 on Fedora Core 1/AMD64.
      Refs:
      http://firebird.sourceforge.net/download/prerelease/1.5.1
      http://firebird.sourceforge.net/download/prerelease/1.5.1/FirebirdSS-1.5.1.4424-public3.amd64.rpm

Backward-incompatibilities
--------------------------

    - The kinterbasdb.services.Connection.getEnvironmentMessage method has been
      renamed to getMessageFileDir.

    - The kinterbasdb.services.Connection.getLog method should not have
      accepted a database parameter; it no longer does.

DOCUMENTATION CHANGES
---------------------

    - Documented about 66% of the Services API (kinterbasdb.services module) in
      the KInterbasDB Usage Guide.

KNOWN ISSUES
------------

    - The third-party fixedpoint.py module contains an incompatibility with
      Python 2.1 that is exposed by a bugfix applied to the
      kinterbasdb.typeconv_fixed_fixedpoint module in 3.1_pre8.

      No attempt will be made to fix this problem (which is a fixedpoint bug,
      not a kinterbasdb bug); users should either upgrade to a newer version of
      Python or refrain from using fixedpoint.


Version 3.1_pre7
================

Version 3.1_pre7 should be considered a release candidate.  It is thought
to be ready for production use.

New Features
------------

    - Introduced dynamic type translation slot TEXT_UNICODE, which applies to
      all CHAR/VARCHAR fields except those with character sets NONE, OCTETS,
      or ASCII.  Used in combination with the official translators in the
      kinterbasdb.typeconv_text_unicode module, TEXT_UNICODE enables automatic
      encoding/decoding of Unicode values.

      This translator is not active by default except when kinterbasdb is
      initialized with kinterbasdb.init(type_conv=100); the backward
      compatibility implications are discussed in detail in the
      Backward-incompatibilities section below.

      Refs:
      docs/usage.html#faq_fep_unicode

    - Added read-only .charset attribute to Connection.

    - On Windows, kinterbasdb now conforms to the client library loading scheme
      introduced in FB 1.5 RC7, so fbclient.dll need not be explicitly placed
      in a directory on the PATH if the registry settings are present.

Bug Fixes
---------

    - The type slot in cursor.description is now updated dynamically to reflect
      dynamic type translation settings.
      Refs:
      http://sourceforge.net/tracker/index.php?func=detail&aid=814276&group_id=9913&atid=109913

    - Added logic to prevent inappropriate calls to isc_dsql_free_statement,
      which were observed to cause a segfault in a heavily multithreaded
      environment.

    - Added special case to field precision determination code to accommodate
      the database client library's irregular handling of RDB$DATABASE.RDB$DB_KEY
      Refs:
      http://sourceforge.net/tracker/index.php?func=detail&aid=818609&group_id=9913&atid=109913

Backward-incompatibilities
--------------------------

    - Programs that use BOTH of the following:

        - the TEXT dynamic type translation slot
        - unicode database fields

      will need to be updated to take the new TEXT_UNICODE slot into account.
      Since the TEXT slot is not particularly useful, this incompatibility is
      expected to affect very few existing programs.

      Refs:
      docs/usage.html#faq_fep_unicode

    - Convenience code 100 for the kinterbasdb.init function now activates the
      new TEXT_UNICODE translation slot, so unicode values are automatically
      encoded and decoded.

      Convenience code 1 remains the default, however, and it does not
      activate the TEXT_UNICODE slot.  Programs that do BOTH of the following:

        - invoke kinterbasdb.init(type_conv=100)
        - use unicode database fields

      will need to be updated to take the new TEXT_UNICODE slot into account.

      Refs:
      docs/usage.html#adv_param_conv_dynamic_type_translation_tbl_convenience_codes


Version 3.1_pre6
================

Version 3.1_pre6 should be considered a release candidate.  It is thought
to be stable.

New Features
------------

    - Added support for manual control over the phases of two-phase commit.
      The client programmer now has the option of triggering the first phase
      manually via Connection.prepare() or ConnectionGroup.prepare().

      This is useful when integrating with third-party transaction managers.

    - KInterbasDB can now be compiled "out of the box" with MinGW when building
      against Firebird 1.5 (but not Firebird 1.0 or Interbase).

      See docs/installation-source.html for instructions.

Bug Fixes
---------

    - Connection.drop_database() now rolls back the connection's active
      transaction (if any) before dropping the database.

      Previously, the database could be dropped with the transaction still
      active; when the connection was subsequently garbage collected, a rollback
      request was issued for the transaction (in a nonexistent database),
      resulting in memory corruption.

    - String values returned by input dynamic type translators were sometimes
      prematurely garbage collected before the database engine had read their
      contents.

    - SQL fields with dynamic definitions (such as expressions in a SELECT list)
      that involved fixed point data types (NUMERIC or DECIMAL) didn't get
      passed through the FIXED dynamic type translator because the database
      engine does not flag dynamically defined fields properly.

      Though this is a bug in the database engine rather than KInterbasDB, a
      workaround was added to KInterbasDB.

      `Thanks to Bert Hughes for reporting this bug.`

    - The installation action of the setup script ('setup.py install') did not
      place the supporting files (documentation) in the proper directory on
      Linux.

      `Thanks to Treeve Jelbert for reporting this bug.`


Version 3.1_pre5
================

Version 3.1_pre5 should be considered a release candidate.  It is thought
to be stable.

New Features
------------

    - Deferred loading of dynamic type translators:

      KInterbasDB's choice of initial dynamic type translators for date/time
      and fixed point types is now deferred as late as possible, and the
      programmer has the *option* of controlling the choice via the type_conv
      parameter of the new kinterbasdb.init function.

      This feature is documented in the Usage Guide at:
      usage.html#adv_param_conv_dynamic_type_translation_deferred_loading

    - KInterbasDB's setup script is now capable of compiling the source
      distribution "out of the box" with MinGW on Windows, but only with
      Firebird 1.5 or later (Borland C++ can be used with Firebird 1.0).

      This feature is documented in the installation guide for the source
      distribution at:

      installation-source.html#compiler_specific_compilation_notes

Bug Fixes
---------

    - During blob insertion, not enough memory was allocated to hold the
      blob ID returned by the database engine, resulting in an overflow.

    - Implicit conversion of DATE/TIME/TIMESTAMP input parameters from strings
      to the appropriate internal types was accidentally disallowed in
      3.1_pre4.  This feature has been enabled again.

    - The Services API method kinterbasdb.services.Connection.restore was
      incapable of restoring a backup into a multi-file database because it
      sent the wrong cluster identifier for destination file page counts.

Backward-incompatibilities
--------------------------

    - Because of the new "Deferred loading of dynamic type translators" feature,
      the DB API type comparison singleton kinterbasdb.DATETIME will not compare
      equal to *any* type until the kinterbasdb.init function has been called
      (whether explicitly or implicitly).

      This issue--which is expected to affect little or no existing code--is
      documented in the Usage Guide at:

      usage.html#adv_param_conv_dynamic_type_translation_deferred_loading_backcompat

    - The dynamic type translation module typeconv_preferred has been renamed
      to typeconv_23plus.


Version 3.1_pre4
================

Version 3.1_pre4 should be considered a late beta release.  It is thought
to be stable, and there are no plans to add new features before 3.1 final
(only to fix bugs and finish updating the documentation).

Note that the KInterbasDB Usage Guide has been considerably updated, though
it is not quite complete.  When complete, it will document all of the
numerous new features in kinterbasdb 3.1; it's a "must read" even now.

The Usage Guide is distributed with KInterbasDB
(kinterbasdb-installation-dir/docs/usage.html), and is available online at:
http://cvs.sourceforge.net/cgi-bin/viewcvs.cgi/*checkout*/kinterbasdb/Kinterbasdb-3.0/docs/usage.html


New Features
------------

    - DATABASE EVENT HANDLING has been reinstated, ported to POSIX, and timeout
      support has been added.

      This feature is thoroughly documented in the updated Usage Guide.

      Refs:
      http://sourceforge.net/tracker/index.php?func=detail&aid=637796&group_id=9913&atid=109913

    - DISTRIBUTED TRANSACTIONS are now supported via the
      kinterbasdb.ConnectionGroup class.

      Although the Usage Guide does not yet fully document this feature, it
      does contain an example program and a few hints:
      http://cvs.sourceforge.net/cgi-bin/viewcvs.cgi/*checkout*/kinterbasdb/Kinterbasdb-3.0/docs/usage.html#adv_trans_control_distributed

    - DYNAMIC TYPE TRANSLATION

      KInterbasDB 3.1_pre4 implements two-way "dynamic type translation".
      This feature allows the client programmer to change the type conversion
      methods for specific SQL data types and achieve complete "type
      transparency".  For example, KInterbasDB 3.1_pre4 includes reference
      implementations of converters for both input and output of
      'mx.DateTime' and Python 2.3 stdlib 'datetime' for TIME/DATE/TIMESTAMP
      fields.

      One consequence of two-way dynamic type translation support is that
      KInterbasDB 3.1_pre4 can be used with Python 2.3's datetime module
      occupying the former role mx.DateTime.  For backward compatibility,
      mx.DateTime is still the default, and it will remain so.

      This feature is documented in the updated Usage Guide.

    - Cursor.rowcount support has been added (insofar as the database engine
      supports it).

      This feature is documented in the updated Usage Guide.

      Refs:
      http://sourceforge.net/forum/forum.php?thread_id=866629&forum_id=30917

    - SAVEPOINTs (a Firebird 1.5 feature) are exposed at the Python level via
      the Connection.savepoint(savepoint='name') method and the optional
      $savepoint argument to the Cursor.rollback method.

      This feature is documented in the updated Usage Guide.

    - New attributes suggested by the "Optional DB API Extensions" section
      of PEP 249:

        - Access to a cursor's connection via the Cursor.connection attribute.
        - Access to kinterbasdb's exception classes via connection attributes.

      Refs:
      http://www.python.org/peps/pep-0249.html

    - A cursor can now be reused after it has caused an exception.

Bug Fixes
---------

    - Passing the wrong number of parameters for a parameterized SQL statement
      sometimes caused a crash instead of an exception with kinterbasdb 3.1_pre3.
      This would not have affected client programs that were written correctly,
      but it was still a bug.

    - The kinterbasdb.create_database function leaked memory if it encountered
      an error.

    - Additional Windows binaries are being released to avoid dynamic linking
      problems with Interbase 5.5 and Firebird 1.5-embedded.

      Refs:
      http://sourceforge.net/tracker/index.php?func=detail&aid=707644&group_id=9913&atid=109913
      http://sourceforge.net/forum/forum.php?thread_id=855348&forum_id=30917

    - kinterbasdb now builds with less hassle on FreeBSD.

      Refs:
      http://sourceforge.net/tracker/index.php?func=detail&aid=720021&group_id=9913&atid=109913

    - Whenever a transactional context is needed, a transaction is now started
      implicitly if the Python programmer has not started one explicitly
      with Connection.begin.  This implicit behavior is implicitly required
      by the Python DB API specification.

      Refs:
      http://mail.python.org/pipermail/db-sig/2003-February/003158.html

    - The mapping objects returned from the Cursor.fetch*map() method now
      accept "double-quoted" field names (lookup keys).  If the field name is
      double-quoted, its original case will be preserved--instead of being
      normalized to upper case--when the result set is searched for a field
      with that name.

      For example, if a table were defined this way::

          create table tbl ("sTRanGelyCasEDfieldnAme" integer)

      and the statement::

          cur.execute("select * from tbl")

      were executed against it, the mapping objects returned by::

          cur.fetchonemap()

      would have rejected the lookup key 'sTRanGelyCasEDfieldnAme', converting
      it instead to 'STRANGELYCASEDFIELDNAME' and then failing to find the
      upper-cased field name.

      The solution available in 3.1_pre4 is to perform the lookup this way::

          cur.execute("select * from tbl")
          mapping = cur.fetchonemap()
          mapping['"sTRanGelyCasEDfieldnAme"']
                   ^-----double-quoted-----^

      which will force the preservation of the field name's case.

      An easy way to avoid problems such as this is to refrain from using
      quoted identifiers; in that case, the database engine will treat
      identifiers in a case-insensitive manner.

      Refs:
      http://sourceforge.net/tracker/index.php?func=detail&aid=720130&group_id=9913&atid=109913

INTERNAL CHANGES
----------------

    - kinterbasdb now implements its standard date/time and fixed point
      handling via the new, general-purpose dynamic type translation feature.

      This eliminates the C-compile-time dependency on mx.DateTime.  Although
      mx.DateTime (for date/time types) and Connection.precision_mode (for
      precise fixed point types) still function as before, dynamic type
      translation allows other types to be transparently substituted (such as
      those in Python 2.3's standard library datetime module for date/time
      types, or those in the fixedpoint module for precise fixed point types).

      For more information, see the Usage Guide.

Backward-incompatibilities
--------------------------

    There are no outright incompatibilities, but there is one deprecation:

    - Although Connection.precision_mode continues to function as in earlier
      versions, it is deprecated in favor of dynamic type translation.  The
      functionality that Connection.precision_mode delivers (precise I/O of
      fixed point values) is now implemented at the Python level via dynamic
      type translation, rather than at the C level.

      If you explicitly use both Connection.precision_mode *and* dynamic
      type translation, beware that changing the value of
      `Connection.precision_mode` will cause changes to the registered dynamic
      type converters under the hood.

      For more information, see the INTERNAL CHANGES section above, and the
      Usage Guide.


Version 3.1_pre3
================

Version 3.1_pre3 should be considered a beta release.

New Features
------------

    - database array support

      Database arrays are mapped from Python sequences (except strings) on
      input; to Python lists on output.  On output, the lists will be nested
      if the database array has multiple dimensions.

      I'm not impressed by the Interbase/Firebird implementation of database
      arrays.  The database engine claims to support up to 16 dimensions, but
      actually malfunctions catastrophically* above 10.

      The arrays are of fixed size, with a predeclared number of dimensions
      and number of elements per dimension.  Individual array elements cannot
      be set to NULL/None**, so the mapping between Python lists (which have
      dynamic length and are therefore not normally null-padded) and
      non-trivial database arrays is clumsy.

      Arrays cannot be passed as parameters to, or returned from, stored
      procedures.

      Finally, many interface libraries, GUIs, and even the isql command line
      utility do not support arrays.
      Refs:

         * http://sourceforge.net/tracker/?func=detail&aid=659610&group_id=9028&atid=109028
        ** Interbase 6 API Guide page 153.

    - retaining commit/retaining rollback

      The commit() and rollback() methods of kinterbasdb.Connection now
      accept an optional boolean parameter 'retaining' (default False).  If
      retaining is True, the infrastructural support for the transaction active
      at the time of the method call will be "retained" (efficiently and
      transparently recycled) after the database server has committed or rolled
      back the conceptual transaction.

      In code that commits or rolls back frequently, 'retaining' the
      transaction yields considerably better performance.  'retaining' will
      become the default at some point in the future if the switch can be made
      without serious backward compatibility issues.

      Refs:
      http://sourceforge.net/forum/forum.php?thread_id=799246&forum_id=30917
      Interbase 6 API Guide page 74.

    - unicode strings can now be executed via:

        - kinterbasdb.Cursor.execute[many]()
        - kinterbasdb.Cursor.callproc()
        - kinterbasdb.Connection.execute_immediate()

      However, the encoding of the incoming unicode string is rather
      simplistic--via PyUnicode_AsASCIIString.


Bug Fixes
---------

    - Addressed buffer overflow potential in:

      - kinterbasdb.create_database()
      - kinterbasdb.connect()
      - kinterbasdb.Connection.begin()
      - kinterbasdb.Connection.execute_immediate()
      - kinterbasdb.Cursor.execute() (and thence, executemany() and callproc())

    - Fixed reference count leaks in:

      - exception handling (_exception_functions.c)
      - field precision determination (_kiconversion_field_precision.c)

    - Fixed kinterbasdb.Connection.close() bug:  The physical connection to
      the database server was not actually closed until the
      kinterbasdb.Connection instance was garbage collected.

    - Fixed a bug in the kinterbasdb.services.Connection.userExists() method.
      Usernames are now normalized to upper case.

    - Database version compatibility:

      - kinterbasdb compiles properly against Firebird 1.5.
      - kinterbasdb compiles against and ought to work with (but has not been
        tested with) Interbase 5.5, albeit with some lost functionality,
        namely:

          - field precision determination (the precision entry in cursor.description)
          - Services API support
          - retaining rollback
          - various data storage options, such as precise 64-bit integer storage
            of NUMERIC and DECIMAL values (IB 5.5 uses doubles instead, which
            is not really adequate) and more diverse date/time types.

      Refs:
      http://sourceforge.net/tracker/index.php?func=detail&aid=627816&group_id=9913&atid=109913
      IB 6 Data Definition Guide page 65.

    - Improved DB API compliance:

      - Now, there need not be an active transaction before any execute(),
        commit(), or rollback() call; transaction establishment is implicit in
        these cases.
      - Cursors no longer need to be discarded after an exception; the same
        cursor can be reused.  Of course if the cursor was in the process of
        fetching a result set, the remainder of the set will not be available
        after the exception.

INTERNAL CHANGES
----------------

    - Numerous modest optimizations, especially with regard to memory handling.
      Among these is a move to take advantage of Python 2.3's specialized,
      Python-oriented memory manager.
    - MAJOR code refactoring and tidying.


Backward-incompatibilities
--------------------------

    - Invalid argument combinations to the connect() function now raise a
      ProgrammingError rather than an InterfaceError.  Note that this refers to
      invalid *combinations* of arguments, not necessarily to invalid *values*
      for those arguments.

    - Non-keyword-argument forms of connect() are now deprecated; passing
      non-keyword arguments to connect() results in a DeprecationWarning being
      issued via the standard Python warning framework.  This is a warning, not
      an incompatibility in the strict sense.

      Refs:
      http://www.python.org/doc/current/lib/module-warnings.html

    - Official support for database event handling has been deferred until 3.2.
      A Win32-only prototype will still be included with the kinterbasdb 3.1
      source distribution (but not compiled by default).

      Refs:
      docs/usage.html#database_events_unsupported


Version 3.1_pre2
================

New Features
------------

    - Global Interpreter Lock management

      Previously, kinterbasdb operated in a serial manner, with the sole
      exception of the event handling code, whose parallelism is "under the
      hood".  Aside from event handling, all kinterbasdb operations, including
      potentially long-running Firebird API calls, were serialized by the
      Python GIL.

      With the advent of kinterbasdb 3.1_pre2, kinterbasdb releases the GIL
      where appropriate--that is, when it is about to make a potentially long-
      running Firebird API call, and can do so without invoking the Python API,
      or otherwise operating on Python structures.

      However, the Firebird client library itself is not threadsafe, so
      Firebird API calls must also be serialized.  To that end, kinterbasdb
      maintains a process-wide thread lock around which all Firebird API calls
      are serialized.

      When kinterbasdb is about to make a potentially long-running Firebird
      API call, it follows these steps:

        1.  Extract necessary parameter data from Python structures
        2.  Release the Python GIL
        3.  Acquire the kinterbasdb process-wide Firebird client thread lock
        4.  Execute the Firebird API call
        5.  Release the kinterbasdb process-wide Firebird client thread lock
        6.  Acquire the Python GIL
        7.  Modify Python structures to reflect the results of the Firebird API
              call

      The addition of GIL management should improve kinterbasdb's maximum
      possible throughput for multi-threaded Python programs on multi-processor
      systems (one processor can run the Python interpreter while another
      executes a Firebird client library operation).  GIL management may also
      yield greater "responsiveness" for multi-threaded Python programs running
      on single-processor systems.

      The addition of GIL management required fairly extensive internal
      changes, and therefore warranted a whole prerelease version virtually
      unto itself.

    - Cursor name support

      The read/write property Cursor.name allows the Python programmer to
      perform scrolling UPDATEs or DELETions via the "SELECT ... FOR UPDATE"
      syntax.  If you don't know what this means, refer to the database SQL
      syntax documentation of the FOR UPDATE clause of the SELECT statement.
      The Cursor.name property can be ignored entirely if you don't need to
      use it.

      Here's an example code fragment:

      .. sourcecode:: python

          con = ... # establish a kinterbasdb.Connection
          curScroll = con.cursor()
          curUpdate = con.cursor()

          curScroll.execute('select city from customer for update')
          curScroll.name = 'city_scroller'
          update = 'update customer set city=? where current of ' + curScroll.name

          for (city,) in curScroll:
              city = ... # make some changes to city
              curUpdate.execute( update, (city,) )


Version 3.1_pre1
================

Version 3.1_pre1 should be considered an early alpha release.

New Features
------------

This list of new features represents the state of kinterbasdb 3.1_pre1,
which does not include some features slated for inclusion in the final
release of kinterbasdb 3.1.  For a discussion of the ultimate goals of
version 3.1, see:
http://sourceforge.net/forum/forum.php?thread_id=696302&forum_id=30917

Also, the documentation has not yet been updated to cover these new
features, nor will it be for at least another month.  In the meantime,
those who need to use the new features must refer to the source code.


    - Cursor Iteration Support
        When used with Python 2.2 or later, kinterbasdb's Cursors now support
      "natural" iteration.  For example:

      .. sourcecode:: python

         # Index-based field lookup (based on Cursor.fetchone):
         cur = con.cursor()
         cur.execute("select col1, col2 from the_table")
         for row in cur:
             col1 = row[0]

         # Key-based field lookup (based on Cursor.fetchonemap):
         cur = con.cursor()
         cur.execute("select col1, col2 from the_table")
         for rowMap in cur.itermap():
             col1 = rowMap['col1']

      The iterator-based pattern supercedes the ugly fetch pattern of old
      (though of course the old pattern will still work):

      .. sourcecode:: python

         # Index-based field lookup (based on Cursor.fetchone):
         cur = con.cursor()
         cur.execute("select col1, col2 from the_table")
         while 1:
             row = cur.fetchone()
             if not row:
                 break

             col1 = row[0]


    - Implicit Parameter Conversion

      Implicit parameter conversion allows any SQL datatype supported by
      kinterbasdb to be passed to the database engine as a Python string.

      This is especially useful for parameterized statements that involve
      date/time datatypes, because they can now accept server-computed "magic"
      values such as 'now' and 'current_date' more naturally.  Implicit
      parameter conversion is also likely to yield a speedup for programs that
      load external data from flat files into the database, since the incoming
      values do not need to be converted from their original string
      representation into an acceptable Python type before being forwarded to
      the database.

      For a more thorough discussion of this new feature, see:
      http://sourceforge.net/tracker/index.php?func=detail&aid=531828&group_id=9913&atid=309913

    - Services API Support (see IB 6 API Guide Chapter 12)

      The database engine provides an API (the Services API) to facilitate
      programmatic invocation of the maintenance tasks available through the
      command-line tools gbak, gfix, etc.

      I've wrapped nearly the entire Services API in a thick Python API of
      my own design.  My API design is only provisional; I seek feedback as
      to how it could be made more elegant.  The Services API support is
      accessible via the kinterbasdb.services module.

    - Database Event Support (see IB 6 API Guide Chapter 11)

      The database engine allows client programs to register to be informed
      of the occurrence of database events, which can be raised with the
      POST_EVENT statement in stored procedures or triggers.  kinterbasdb 3.1
      supports a subset of this functionality (synchronous waiting only) via
      the Connection.wait(eventNames) method.

      The current implementation is only a rough prototype; though usable,
      it is not recommended for production environments.

      The current implementation suffers from a major limitation:  only one
      thread per process is allowed to listen for event notification.  This is
      so because the current implementation resorts to some roundabout trickery
      to circumvent the lack of database API support for synchronous event
      notification on Windows.  Because the database API only starts one
      asynchronous event handler thread per process, I doubt that support for
      multiple event-listening threads in a single process will materialize.

Bug Fixes
---------

    - In the past, the opaque mapping object returned by the Cursor.fetch*map
      methods returned None when asked for a field not in its select list,
      rather than raising a KeyError.  It now raises a KeyError in such a case.
      For example:

      .. sourcecode:: python

         cur = con.cursor()
         cur.execute("select col1, col2 from the_table")
         for rowMap in cur.itermap():
             x = rowMap['col3'] # Used to return None.  Now raises KeyError,
                                # because col3 was not SELECTed.


Backward-incompatibilities
--------------------------

    - Although kinterbasdb 3.1 is significantly different internally, there is
      only one known API incompatibility with version 3.0.2.  It would only arise
      in code that relies on the erroneous behavior of the mapping-fetch bug
      mentioned above.

    - Python versions prior to 2.1 are no longer officially supported.
      Although kinterbasdb might still compile against Python 2.0 and earlier,
      I will not go out of my way to ensure that it does.


Version 3.0.2
=============

Bug Fixes
---------

    - Fixed a CHAR-handling bug that caused CHAR values inserted into the
      database to lack their trailing spaces.  Instead, the values were null-
      terminated.  This left CHAR values inserted by kinterbasdb incompatible
      with standard tools, which expect trailing spaces.

      For more information, see
      http://sourceforge.net/tracker/index.php?func=detail&aid=594908&group_id=9913&atid=109913


Version 3.0.1
=============

Bug Fixes
---------

    - Adjusted input handling of NULL values.  The new scheme raises an
      exception immediately when it detects that a Python None value has
      arrived for storage in a database field or parameter that disallows
      NULL values.

      The old scheme simply accepted the Python None value and then tried
      to execute the query, relying on the database API to detect the error.
      With certain data types, the database API would silently insert a bogus
      value rather than detecting the error.

    - Scrutinized the datetime input/output facilities, found some
      incompatibilities with the DB API, and corrected them.  These changes
      are backward-incompatible, but are warranted because the previous
      behavior was in defiance of the specification.  See further notes about
      the nature of these changes in the backward-incompatibilities section.

    - Fixed a memory leak that affected the storage of Python string input
      parameters in BLOB fields.

    - Fixed a rollback-related bug that arose if a connection to a database
      was established, but a transaction was never started on that connection.
      In such a case, a spurious exception was raised when the connection was
      garbage collected.

      Normal code would not have invoked this bug, but it was still a bug.

Backward-incompatibilities
--------------------------

    - Datetime input/output has changed to better comply with the DB API (see
      datetime bugfix discussion above).

      Code that uses the mx.DateTime module directly (rather than the
      kinterbasdb DB API datetime constructors) should not be affected.

      For details, see the comments in the code block in __init__.py tagged
      with "DSR:2002.07.19".

Version 3.0.1_pre3
==================

Bug Fixes
---------

    - Bug #572326 (which was not present in kinterbasdb 3.0 and never affected
      Python 2.2+) caused several numeric types to not be transferred from
      Python to the database engine when they were passed as query parameters.

      This was a serious bug; it caused even such fundamental operations as:
      cursor.execute("insert into the_table values (?)", (1,))
      to not work correctly.


Version 3.0.1_pre2
==================

Bug Fixes
---------

    - CHAR output now doesn't have such problems with multibyte character sets
      and values shorter than the maximum declared length of the field.

      CHARs are no longer returned with their trailing blanks intact.  The
      trailing blanks have been abandoned because they were in fact NULL
      characters, not spaces.  kinterbasdb would fill in the spaces manually,
      except for the problems that approach causes with multibyte character
      sets.

    - Fixed a potential buffer overflow, but the fix only applies when compiled
      against Python 2.2 or later.

Backward-incompatibilities
--------------------------

    - See coverage of CHAR output changes in the 'Bug Fixes' section.  In a
      nutshell:  CHAR output values no longer have trailing NULL bytes.


Version 3.0.1_pre1
==================

New Features
------------

    - It is now possible to connect to a database under a specific role by using
      the 'role' keyword argument of the kinterbasdb.connect function.

    - The following methods now accept any sequence except a string for their
      'parameter' argument, rather than demanding a tuple:  Cursor.execute,
      Cursor.executemany and Cursor.callproc.

Bug Fixes
---------

    - kinterbasdb supports IB 5.x again.

      Various identifiers specific to IB 6.x/Firebird had crept into unguarded
      areas of __init__.py and _kinterbasdb.c, but this has been changed so
      that kinterbasdb compiles gracefully with IB 5.x.
      See:
      http://sourceforge.net/tracker/index.php?func=detail&aid=553184&group_id=9913&atid=209913

    - The distutils setup script no longer raises a ValueError on Windows 2000
      or XP.

    - The precision slot in Cursor.description was always zero.  It now contains
      the correct value if that value can reasonably be determined.

      Note that the database engine records the precision of some fields as
      zero (e.g., FLOAT), and the slot will also be zero in cases where the
      database engine does not expose the precision of the field (e.g., dynamic
      fields such as "SELECT 33.5 FROM RDB$DATABASE").

      Since the database API does not provide the field's precision figure in
      the XSQLVAR structure, it is necessary to query the system tables.  In
      order to minimize the performance hit, precision figures are cached per
      Connection; the determination of a given field's precision figure in the
      context of a given Connection will require only dictionary lookups after
      it is determined the first time with a system table query.

      An unfortunate side effect of this caching is that if a field's
      precision is altered after the figure has been cached in by a Connection,
      cursors based on that Connection will still show the old precision figure.
      In practice, this situation will almost never arise.
      See:
      http://sourceforge.net/tracker/index.php?func=detail&aid=549982&group_id=9913&atid=109913

    - On Linux, attempting to fetch immediately after having executed a
      non-query statement resulted in a segfault.  An exception is now raised
      instead.  The problem did not afflict Windows, which always raised the
      exception.
      See:
      http://sourceforge.net/tracker/index.php?func=detail&aid=551098&group_id=9913&atid=109913

      - The message carried by this exception grew without bound in on both
        Windows and Linux.  It no longer does.

    - Under some circumstances, the fetched values of CHAR fields were
      incorrect.  CHAR values now appear as expected (they are left-padded with
      spaces and always of length equal to their field's designated maximum
      length).

    - Cursor.fetchmany raised an error if there were no remaining values to
      fetch.  It now returns an empty sequence instead, as required by the DB
      API Specification.

    - Field domains are checked more strictly.  It is now impossible to (for
      example) issue a statement that attempts to insert a 12-character string
      into a 10-character CHAR field without encountering an exception.

      This checking is not perfect, since it validates against the field's
      internal storage type rather than the field's declared type.  For example,
      a NUMERIC(1,1), which is stored internally as a short, will erroneously
      accept the value 12.5 because 125 fits in a short.

    - When operating in imprecise mode (connection.precision_mode == 0),
      kinterbasdb 3.0 sometimes interpreted integer values as though it were
      operating in precise mode.


Version 3.0 versus 2.0-0.3.1
============================

New Features
------------

    The new features are thoroughly documented in the KInterbasDB Usage Guide
    (usage.html); they need not be reiterated here.
    However, backward-incompatible changes *have* been documented in this
    changelog (see the Backward-incompatibilities section).

Bug Fixes
---------

    Many bugs have been fixed, including (but not limited to) the following,
    which were registered with the KInterbasDB bug tracker at SourceForge
    ( http://sourceforge.net/tracker/index.php?group_id=9913&atid=109913 ):

      - 433090  cannot connect to firebird server
      - 438130  cursor.callproc not adding param code
      - 468304  fetchmany return all record
      - 498086  ignores column aliases in select
      - 498403  fetching after a callproc hangs program
      - 498414  execute procedure message length error
      - 505950  inconsistent fetch* return types
      - 515974  Wrong decoding of FB isc_version
      - 517093  broken fixed-point handling in 3.0
      - 517840  C function normalize_double inf. loop
      - 517842  fetch bug - program hangs
      - 520793  poor DB API compliance
        ^ a *BIG* fix that entailed many changes
      - 522230  error with blobs larger than (2^16) - 1
      - 522774  inconsistent fixed-point conv in 3.0-rc2
      - 523348  memory leak in Blob2PyObject

      - immediate execution facilities unreliable in 2.x

Backward-incompatibilities
--------------------------

    As a result of the changes required for some of the bugfixes (especially
    #520793 - "poor DB API compliance") and general reengineering, several
    areas of backward-incompatibility have arisen:

      - fetch* return types

        The standard fetch(one|many|all) methods now return just a sequence,
        not a combined sequence/mapping.  If you want a mapping, use one of
        the fetch(one|many|all)map methods.

        Note the "'absolutely no guarantees' except..." caveats in the
        KInterbasDB Usage Guide regarding the return types of the
        Cursor.fetch* methods and the contents of the Cursor.description
        attribute.

        This is a significant backward-incompatibility, and was not
        undertaken without serious consideration (for evidence see
        http://sourceforge.net/forum/forum.php?thread_id=622782&forum_id=30919
        ).

      - Fixed point number handling

        Fixed point number handling has been remodelled.  By default, fixed
        point numbers (NUMERIC/DECIMAL field values) are now represented
        (with a potential loss of precision) as Python floats.

        A Connection.precision_mode attribute has been added so that precise
        representation of fixed point values as scaled Python integers (as in
        KInterbasDB 2.x) can be used at will.

        For more information, see the KInterbasDB Usage Guide.

      - Connection.dialect

        In KInterbasDB 2.x, the default connection dialect was 1 (the
        backward-compatibility dialect for use with Interbase 5.5 and earlier).

        KInterbasDB 3.0 is being released into quite a different climate.
        Interbase 6.0 was released nearly two years ago, and Firebird 1.0 has
        recently been released.  Because it is expected that KInterbasDB 3.0
        will be used most frequently with Interbase 6.0+ and Firebird, the
        default connection dialect is 3.

        Using KInterbasDB 3.0 with Interbase 5.5 and earlier is still
        possible, though untested by the developers of KInterbasDB 3.0.  See
        the Connection.dialect documentation in the KInterbasDB Usage Guide
        for an explanation of how to initialize a connection with a dialect
        other than 3.

      - Connection.server_version

        The Connection.server_version attribute is now a string rather than
        an integer.  An integer simply was not expressive enough to represent
        the numerous Interbase variants that exist today (including Firebird,
        which does not fit neatly into the Interbase version progression).

        For more information, see the KInterbasDB Usage Guide.

      - kinterbasdb.execute_immediate

        The kinterbasdb.execute_immediate function has been removed.  A
        similar function named kinterbasdb.create_database has been added.
        The primary differences between kinterbasdb.execute_immediate and
        kinterbasdb.create_database are:

          - kinterbasdb.create_database is not as general
          - kinterbasdb.create_database actually works

        The execute_immediate method of the Connection class has been
        retained.

        For more information, see the KInterbasDB Usage Guide.