HISTORY.txt

Changes from 5.1 to 5.1.1
 1) Simplify management of threads for callbacks performed by database change
    notification and eliminate a crash that occurred under high load in
    certain situations. Thanks to Calvin S. for noting the issue and suggesting
    a solution and testing the patch.
 2) Force server detach on close so that the connection is completely closed
    and not just the session as before.
 3) Force use of OCI_UTF16ID for NCLOBs as using the default character set
    would result in ORA-03127 with Oracle 11.2.0.2 and UTF8 character set.
 4) Avoid attempting to clear temporary LOBs a second time when destroying the
    variable as in certain situations this results in spurious errors.
 5) Added additional parameter service_name to makedsn() which can be used to
    use the service_name rather than the SID in the DSN string that is
    generated.
 6) Fix cursor description in test suite to take into account the number of
    bytes per character.
 7) Added tests for NCLOBS to the test suite.
 8) Removed redundant code in setup.py for calculating the library path.

Changes from 5.0.4 to 5.1
 1) Remove support for UNICODE mode and permit Unicode to be passed through in
    everywhere a string may be passed in. This means that strings will be
    passed through to Oracle using the value of the NLS_LANG environment
    variable in Python 3.x as well. Doing this eliminated a bunch of problems
    that were discovered by using UNICODE mode and also removed an unnecessary
    restriction in Python 2.x that Unicode could not be used in connect strings
    or SQL statements, for example.
 2) Added support for creating an empty object variable via a named type, the
    first step to adding full object support.
 3) Added support for Python 3.2.
 4) Account for lib64 used on x86_64 systems. Thanks to Alex Wood for supplying
    the patch.
 5) Clear up potential problems when calling cursor.close() ahead of the
    cursor being freed by going out of scope.
 6) Avoid compilation difficulties on AIX5 as OCIPing does not appear to be
    available on that platform under Oracle 10g Release 2. Thanks to
    Pierre-Yves Fontaniere for the patch.
 7) Free temporary LOBs prior to each fetch in order to avoid leaking them.
    Thanks to Uwe Hoffmann for the initial patch.

Changes from 5.0.3 to 5.0.4
 1) Added support for Python 2.7.
 2) Added support for new parameter (port) for subscription() call which allows
    the client to specify the listening port for callback notifications from
    the database server. Thanks to Geoffrey Weber for the initial patch.
 3) Fixed compilation under Oracle 9i.
 4) Fixed a few error messages.

Changes from 5.0.2 to 5.0.3
 1) Added support for 64-bit Windows.
 2) Added support for Python 3.1 and dropped support for Python 3.0.
 3) Added support for keyword arguments in cursor.callproc() and
    cursor.callfunc().
 4) Added documentation for the UNICODE and FIXED_UNICODE variable types.
 5) Added extra link arguments required for Mac OS X as suggested by Jason
    Woodward.
 6) Added additional error codes to the list of error codes that raise
    OperationalError rather than DatabaseError.
 7) Fixed calculation of display size for strings with national database
    character sets that are not the default AL16UTF16.
 8) Moved the resetting of the setinputsizes flag before the binding takes
    place so that if an error takes place and a new statement is prepared
    subsequently, spurious errors will not occur.
 9) Fixed compilation with Oracle 10g Release 1.
10) Tweaked documentation based on feedback from a number of people.
11) Added support for running the test suite using "python setup.py test"
12) Added support for setting the CLIENT_IDENTIFIER value in the v$session
    table for connections.
13) Added exception when attempting to call executemany() with arrays which is
    not supported by the OCI.
14) Fixed bug when converting from decimal would result in OCI-22062 because
    the locale decimal point was not a period. Thanks to Amaury Forgeot d'Arc
    for the solution to this problem.

Changes from 5.0.1 to 5.0.2
 1) Fix creation of temporary NCLOB values and the writing of NCLOB values in
    non Unicode mode.
 2) Re-enabled parsing of non select statements as requested by Roy Terrill.
 3) Implemented a parse error offset as requested by Catherine Devlin.
 4) Removed lib subdirectory when forcing RPATH now that the library directory
    is being calculated exactly in setup.py.
 5) Added an additional cast in order to support compiling by Microsoft
    Visual C++ 2008 as requested by Marco de Paoli.
 6) Added additional include directory to setup.py in order to support
    compiling by Microsoft Visual Studio was requested by Jason Coombs.
 7) Fixed a few documentation issues.

Changes from 5.0 to 5.0.1
 1) Added support for database change notification available in Oracle 10g
    Release 2 and higher.
 2) Fix bug where NCLOB data would be corrupted upon retrieval (non Unicode
    mode) or would generate exception ORA-24806 (LOB form mismatch). Oracle
    insists upon differentiating between CLOB and NCLOB no matter which
    character set is being used for retrieval.
 3) Add new attributes size, bufferSize and numElements to variable objects,
    deprecating allocelems (replaced by numElements) and maxlength (replaced
    by bufferSize)
 4) Avoid increasing memory allocation for strings when using variable width
    character sets and increasing the number of elements in a variable during
    executemany().
 5) Tweaked code in order to ensure that cx_Oracle can compile with Python
    3.0.1.

Changes from 4.4.1 to 5.0
 1) Added support for Python 3.0 with much help from Amaury Forgeot d'Arc.
 2) Removed support for Python 2.3 and Oracle 8i.
 3) Added support for full unicode mode in Python 2.x where all strings are
    passed in and returned as unicode (module must be built in this mode)
    rather than encoded strings
 4) nchar and nvarchar columns now return unicode instead of encoded strings
 5) Added support for an output type handler and/or an input type handler to be
    specified at the connection and cursor levels.
 6) Added support for specifying both input and output converters for variables
 7) Added support for specifying the array size of variables that are created
    using the cursor.var() method
 8) Added support for events mode and database resident connection pooling
    (DRCP) in Oracle 11g.
 9) Added support for changing the password during construction of a new
    connection object as well as after the connection object has been created
10) Added support for the interval day to second data type in Oracle,
    represented as datetime.timedelta objects in Python.
11) Added support for getting and setting the current_schema attribute for a
    session
12) Added support for proxy authentication in session pools as requested by
    Michael Wegrzynek (and thanks for the initial patch as well).
13) Modified connection.prepare() to return a boolean indicating if a
    transaction was actually prepared in order to avoid the error ORA-24756
    (transaction does not exist).
14) Raise a cx_Oracle.Error instance rather than a string for column
    truncation errors as requested by Helge Tesdal.
15) Fixed handling of environment handles in session pools in order to allow
    session pools to fetch objects without exceptions taking place.

Changes from 4.4 to 4.4.1
 1) Make the bind variables and fetch variables accessible although they need
    to be treated carefully since they are used internally; support added for
    forward compatibility with version 5.x.
 2) Include the "cannot insert null value" in the list of errors that are
    treated as integrity errors as requested by Matt Boersma.
 3) Use a cx_Oracle.Error instance rather than a string to hold the error when
    truncation (ORA-1406) takes place as requested by Helge Tesdal.
 4) Added support for fixed char, old style varchar and timestamp attribute
    values in objects.
 5) Tweaked setup.py to check for the Oracle version up front rather than
    during the build in order to produce more meaningful errors and simplify
    the code.
 6) In setup.py added proper detection for the instant client on Mac OS X as
    recommended by Martijn Pieters.
 7) In setup.py, avoided resetting the extraLinkArgs on Mac OS X as doing so
    prevents simple modification where desired as expressed by Christian
    Zagrodnick.
 8) Added documentation on exception handling as requested by Andreas Mock, who
    also graciously provided an initial patch.
 9) Modified documentation indicating that the password attribute on connection
    objects can be written.
10) Added documentation warning that parameters not passed in during subsequent
    executions of a statement will retain their original values as requested by
    Harald Armin Massa.
11) Added comments indicating that an Oracle client is required since so many
    people find this surprising.
12) Removed all references to Oracle 8i from the documentation and version 5.x
    will eliminate all vestiges of support for this version of the Oracle
    client.
13) Added additional link arguments for Cygwin as requested by Rob Gillen.

Changes from 4.3.3 to 4.4
 1) Fix setup.py to handle the Oracle instant client and Oracle XE on both
    Linux and Windows as pointed out by many. Thanks also to the many people
    who also provided patches.
 2) Set the default array size to 50 instead of 1 as the DB API suggests
    because the performance difference is so drastic and many people have
    recommended that the default be changed.
 3) Added Py_BEGIN_ALLOW_THREADS and Py_END_ALLOW_THREADS around each blocking
    call for LOBs as requested by Jason Conroy who also provided an initial
    patch and performed a number of tests that demonstrate the new code is much
    more responsive.
 4) Add support for acquiring cursor.description after a parse.
 5) Defer type assignment when performing executemany() until the last possible
    moment if the value being bound in is null as suggested by Dragos Dociu.
 6) When dropping a connection from the pool, ignore any errors that occur
    during the rollback; unfortunately, Oracle decides to commit data even when
    dropping a connection from the pool instead of rolling it back so the
    attempt still has to be made.
 7) Added support for setting CLIENT_DRIVER in V$SESSION_CONNECT_INFO in Oracle
    11g and higher.
 8) Use cx_Oracle.InterfaceError rather than the builtin RuntimeError when
    unable to create the Oracle environment object as requested by Luke Mewburn
    since the error is specific to Oracle and someone attempting to catch any
    exception cannot simply use cx_Oracle.Error.
 9) Translated some error codes to OperationalError as requested by Matthew
    Harriger; translated if/elseif/else logic to switch statement to make it
    more readable and to allow for additional translation if desired.
10) Transformed documentation to new format using restructured text. Thanks to
    Waldemar Osuch for contributing the initial draft of the new documentation.
11) Allow the password to be overwritten by a new value as requested by Alex
    VanderWoude; this value is retained as a convenience to the user and not
    used by anything in the module; if changed externally it may be convenient
    to keep this copy up to date.
12) Cygwin is on Windows so should be treated in the same way as noted by
    Matthew Cahn.
13) Add support for using setuptools if so desired as requested by Shreya
    Bhatt.
14) Specify that the version of Oracle 10 that is now primarily used is 10.2,
    not 10.1.

Changes from 4.3.2 to 4.3.3
 1) Added method ping() on connections which can be used to test whether or not
    a connection is still active (available in Oracle 10g R2).
 2) Added method cx_Oracle.clientversion() which returns a 5-tuple giving the
    version of the client that is in use (available in Oracle 10g R2).
 3) Added methods startup() and shutdown() on connections which can be used to
    startup and shutdown databases (available in Oracle 10g R2).
 4) Added support for Oracle 11g.
 5) Added samples directory which contains a handful of scripts containing
    sample code for more advanced techniques. More will follow in future
    releases.
 6) Prevent error "ORA-24333: zero iteration count" when calling executemany()
    with zero rows as requested by Andreas Mock.
 7) Added methods __enter__() and __exit__() on connections to support using
    connections as context managers in Python 2.5 and higher. The context
    managed is the transaction state. Upon exit the transaction is either
    rolled back or committed depending on whether an exception took place or
    not.
 8) Make the search for the lib32 and lib64 directories automatic for all
    platforms.
 9) Tweak the setup configuration script to include all of the metadata and
    allow for building the module within another setup configuration script
10) Include the Oracle version in addition to the Python version in the build
    directories that are created and in the names of the binary packages that
    are created.
11) Remove unnecessary dependency on win32api to build module on Windows.

Changes from 4.3.1 to 4.3.2
 1) Added methods open(), close(), isopen() and getchunksize() in order to
    improve performance of reading/writing LOB values in chunks.
 2) Fixed support for native doubles and floats in Oracle 10g; added new type
    NATIVE_FLOAT to allow specification of a variable of that specific type
    where desired. Thanks to D.R. Boxhoorn for pointing out the fact that this
    was not working properly when the arraysize was anything other than 1.
 3) When calling connection.begin(), only create a new tranasction handle if
    one is not already associated with the connection. Thanks to Andreas Mock
    for discovering this and for Amaury Forgeot d'Arc for diagnosing the
    problem and pointing the way to a solution.
 4) Added attribute cursor.rowfactory which allows a method to be called for
    each row that is returned; this is about 20% faster than calling the method
    in Python using the idiom [method(*r) for r in cursor].
 5) Attempt to locate an Oracle installation by looking at the PATH if the
    environment variable ORACLE_HOME is not set; this is of primary use on
    Windows where this variable should not normally be set.
 6) Added support for autocommit mode as requested by Ian Kelly.
 7) Added support for connection.stmtcachesize which allows for both reading
    and writing the size of the statement cache size. This parameter can make a
    huge difference with the length of time taken to prepare statements. Added
    support for setting the statement tag when preparing a statement. Both of
    these were requested by Bjorn Sandberg who also provided an initial patch.
 8) When copying the value of a variable, copy the return code as well.

Changes from 4.3 to 4.3.1
 1) Ensure that if the client buffer size exceeds 4000 bytes that the server
    buffer size does not as strings may only contain 4000 bytes; this allows
    handling of multibyte character sets on the server as well as the client.
 2) Added support for using buffer objects to populate binary data and made the
    Binary() constructor the buffer type as requested by Ken Mason.
 3) Fix potential crash when using full optimization with some compilers.
    Thanks to Aris Motas for noticing this and providing the initial patch and
    to Amaury Forgeot d'Arc for providing an even simpler solution.
 4) Pass the correct charset form in to the write call in order to support
    writing to national character set LOB values properly. Thanks to Ian Kelly
    for noticing this discrepancy.

Changes from 4.2.1 to 4.3
 1) Added preliminary support for fetching Oracle objects (SQL types) as
    requested by Kristof Beyls (who kindly provided an initial patch).
    Additional work needs to be done to support binding and updating objects
    but the basic structure is now in place.
 2) Added connection.maxBytesPerCharacter which indicates the maximum number of
    bytes each character can use; use this value to also determine the size of
    local buffers in order to handle discrepancies between the client character
    set and the server character set. Thanks to Andreas Mock for providing the
    initial patch and working with me to resolve this issue.
 3) Added support for querying native floats in Oracle 10g as requested by
    Danny Boxhoorn.
 4) Add support for temporary LOB variables created via PL/SQL instead of only
    directly by cx_Oracle; thanks to Henning von Bargen for discovering this
    problem.
 5) Added support for specifying variable types using the builtin types int,
    float, str and datetime.date which allows for finer control of what type of
    Python object is returned from cursor.callfunc() for example.
 6) Added support for passing booleans to callproc() and callfunc() as
    requested by Anana Aiyer.
 7) Fixed support for 64-bit environments in Python 2.5.
 8) Thanks to Filip Ballegeer and a number of his co-workers, an intermittent
    crash was tracked down; specifically, if a connection is closed, then the
    call to OCIStmtRelease() will free memory twice. Preventing the call when
    the connection is closed solves the problem.

Changes from 4.2 to 4.2.1
 1) Added additional type (NCLOB) to handle CLOBs that use the national
    character set as requested by Chris Dunscombe.
 2) Added support for returning cursors from functions as requested by Daniel
    Steinmann.
 3) Added support for getting/setting the "get" mode on session pools as
    requested by Anand Aiyer.
 4) Added support for binding subclassed cursors.
 5) Fixed binding of decimal objects with absolute values less than 0.1.

Changes from 4.1.2 to 4.2
 1) Added support for parsing an Oracle statement as requested by Patrick
    Blackwill.
 2) Added support for BFILEs at the request of Matthew Cahn.
 3) Added support for binding decimal.Decimal objects to cursors.
 4) Added support for reading from NCLOBs as requested by Chris Dunscombe.
 5) Added connection attributes encoding and nencoding which return the IANA
    character set name for the character set and national character set in use
    by the client.
 6) Rework module initialization to use the techniques recommended by the
    Python documentation as one user was experiencing random segfaults due
    to the use of the module dictionary after the initialization was complete.
 7) Removed support for the OPT_Threading attribute. Use the threaded keyword
    when creating connections and session pools instead.
 8) Removed support for the OPT_NumbersAsStrings attribute. Use the
    numbersAsStrings attribute on cursors instead.
 9) Use type long rather than type int in order to support long integers on
    64-bit machines as reported by Uwe Hoffmann.
10) Add cursor attribute "bindarraysize" which is defaulted to 1 and is used
    to determine the size of the arrays created for bind variables.
11) Added repr() methods to provide something a little more useful than the
    standard type name and memory address.
12) Added keyword argument support to the functions that imply such in the
    documentation as requested by Harald Armin Massa.
13) Treat an empty dictionary passed through to cursor.execute() as keyword
    arguments the same as if no keyword arguments were specified at all, as
    requested by Fabien Grumelard.
14) Fixed memory leak when a LOB read would fail.
15) Set the LDFLAGS value in the environment rather than directly in the
    setup.py file in order to satisfy those who wish to enable the use of
    debugging symbols.
16) Use __DATE__ and __TIME__ to determine the date and time of the build
    rather than passing it through directly.
17) Use Oracle types and add casts to reduce warnings as requested by Amaury
    Forgeot d'Arc.
18) Fixed typo in error message.

Changes from 4.1.1 to 4.1.2
 1) Restore support of Oracle 9i features when using the Oracle 10g client.

Changes from 4.1 to 4.1.1
 1) Add support for dropping a connection from a session pool.
 2) Add support for write only attributes "module", "action" and "clientinfo"
    which work only in Oracle 10g as requested by Egor Starostin.
 3) Add support for pickling database errors.
 4) Use the previously created bind variable as a template if available when
    creating a new variable of a larger size. Thanks to Ted Skolnick for the
    initial patch.
 5) Fixed tests to work properly in the Python 2.4 environment where dates and
    timestamps are different Python types. Thanks to Henning von Bargen for
    pointing this out.
 6) Added additional directories to search for include files and libraries in
    order to better support the Oracle 10g instant client.
 7) Set the internal fetch number to 0 in order to satisfy very picky source
    analysis tools as requested by Amaury Fogeot d'Arc.
 8) Improve the documentation for building and installing the module from
    source as some people are unaware of the standard methods for building
    Python modules using distutils.
 9) Added note in the documentation indicating that the arraysize attribute
    can drastically affect performance of queries since this seems to be a
    common misunderstanding of first time users of cx_Oracle.
10) Add a comment indicating that on HP-UX Itanium with Oracle 10g the library
    ttsh10 must alos be linked against. Thanks to Bernard Delmee for the
    information.

Changes from 4.1 beta 1 to 4.1
 1) Fixed bug where subclasses of Cursor do not pass the connection in the
    constructor causing a segfault.
 2) DDL statements must be reparsed before execution as noted by Mihai
    Ibanescu.
 3) Add support for setting input sizes by position.
 4) Fixed problem with catching an exception during execute and then still
    attempting to perform a fetch afterwards as noted by Leith Parkin.
 5) Rename the types so that they can be pickled and unpickled. Thanks to Harri
    Pasanen for pointing out the problem.
 6) Handle invalid NLS_LANG setting properly (Oracle seems to like to provide a
    handle back even though it is invalid) and determine the number of bytes
    per character in order to allow for proper support in the future of
    multibyte and variable width character sets.
 7) Remove date checking from the native case since Python already checks that
    dates are valid; enhance error message when invalid dates are encountered
    so that additional processing can be done.
 8) Fix bug executing SQL using numeric parameter names with predefined
    variables (such as what takes place when calling stored procedures with out
    parameters).
 9) Add support for reading CLOB values using multibyte or variable length
    character sets.

Changes from 4.0.1 to 4.1 beta 1
 1) Added support for Python 2.4. In Python 2.4, the datetime module is used
    for both binding and fetching of date and timestamp data. In Python 2.3,
    objects from the datetime module can be bound but the internal datetime
    objects will be returned from queries.
 2) Added pickling support for LOB and datetime data.
 3) Fully qualified the table name that was missing in an alter table
    statement in the setup test script as noted by Marc Gehling.
 4) Added a section allowing for the setting of the RPATH linker directive in
    setup.py as requested by Iustin Pop.
 5) Added code to raise a programming error exception when an attempt is made
    to access a LOB locator variable in a subsequent fetch.
 6) The username, password and dsn (tnsentry) are stored on the connection
    object when specified, regardless of whether or not a standard connection
    takes place.
 7) Added additional module level constant called "LOB" as requested by Joseph
    Canedo.
 8) Changed exception type to IntegrityError for constraint violations as
    requested by Joseph Canedo.
 9) If scale and precision are not specified, an attempt is made to return a
    long integer as requested by Joseph Canedo.
10) Added workaround for Oracle bug which returns an invalid handle when the
    prepare call fails. Thanks to alantam@hsbc.com for providing the code that
    demonstrated the problem.
11) The cusor method arravar() will now accept the actual list so that it is
    not necessary to call cursor.arrayvar() followed immediately by
    var.setvalue().
12) Fixed bug where attempts to execute the statement "None" with bind
    variables would cause a segmentation fault.
13) Added support for binding by position (paramstyle = "numeric").
14) Removed memory leak created by calls to OCIParamGet() which were not
    mirrored by calls to OCIDescriptorFree(). Thanks to Mihai Ibanescu for
    pointing this out and providing a patch.
15) Added support for calling cursor.executemany() with statement None
    implying that the previously prepared statement ought to be executed.
    Thanks to Mihai Ibanescu for providing a patch.
16) Added support for rebinding variables when a subsequent call to
    cursor.executemany() uses a different number of rows. Thanks to Mihai
    Ibanescu for supplying a patch.
17) The microseconds are now displayed in datetime variables when nonzero
    similar to method used in the datetime module.
18) Added support for binary_float and binary_double columns in Oracle 10g.

Changes from 4.0 to 4.0.1
 1) Fixed bugs on 64-bit platforms that caused segmentation faults and bus
    errors in session pooling and determining the bind variables associated
    with a statement.
 2) Modified test suite so that 64-bit platforms are tested properly.
 3) Added missing commit statements in the test setup scripts. Thanks to Keith
    Lyon for pointing this out.
 4) Fix setup.py for Cygwin environments. Thanks to Doug Henderson for
    providing the necessary fix.
 5) Added support for compiling cx_Oracle without thread support. Thanks to
    Andre Reitz for pointing this out.
 6) Added support for a new keyword parameter called threaded on connections
    and session pools. This parameter defaults to False and indicates whether
    threaded mode ought to be used. It replaces the module level attribute
    OPT_Threading although examining the attribute will be retained until the
    next release at least.
 7) Added support for a new keyword parameter called twophase on connections.
    This parameter defaults to False and indicates whether support for two
    phase (distributed or global) transactions ought to be present. Note that
    support for distributed transactions is buggy when crossing major version
    boundaries (Oracle 8i to Oracle 9i for example).
 8) Ensure that the rowcount attribute is set properly when an exception is
    raised during execution. Thanks to Gary Aviv for pointing out this problem
    and its solution.

Changes from 3.1 to 4.0
 1) Added support for subclassing connections, cursors and session pools. The
    changes involved made it necessary to drop support for Python 2.1 and
    earlier although a branch exists in CVS to allow for support of Python 2.1
    and earlier if needed.
 2) Connections and session pools can now be created with keyword parameters,
    not just sequential parameters.
 3) Queries now return integers whenever possible and long integers if the
    number will overflow a simple integer. Floats are only returned when it is
    known that the number is a floating point number or the integer conversion
    fails.
 4) Added initial support for user callbacks on OCI functions. See the
    documentation for more details.
 5) Add support for retrieving the bind variable names associated with a
    cursor with a new method bindnames().
 6) Add support for temporary LOB variables. This means that setinputsizes()
    can be used with the values CLOB and BLOB to create these temporary LOB
    variables and allow for the equivalent of empty_clob() and empty_blob()
    since otherwise Oracle will treat empty strings as NULL values.
 7) Automatically switch to long strings when the data size exceeds the
    maximum string size that Oracle allows (4000 characters) and raise an
    error if an attempt is made to set a string variable to a size that it
    does not support. This avoids truncation errors as reported by Jon Franz.
 8) Add support for global (distributed) transactions and two phase commit.
 9) Force the NLS settings for the session so that test tables are populated
    correctly in all circumstances; problems were noted by Ralf Braun and
    Allan Poulsen.
10) Display error messages using the environment handle when the error handle
    has not yet been created; this provides better error messages during this
    rather rare situation.
11) Removed memory leak in callproc() that was reported by Todd Whiteman.
12) Make consistent the calls to manipulate memory; otherwise segfaults can
    occur when the pymalloc option is used, as reported by Matt Hoskins.
13) Force a rollback when a session is released back to the session pool.
    Apparently the connections are not as stateless as Oracle's documentation
    suggests and this makes the logic consistent with normal connections.
14) Removed module method attach(). This can be replaced with a call to
    Connection(handle=) if needed.

Changes from 3.0a to 3.1
 1) Added support for connecting with SYSDBA and SYSOPER access which is
    needed for connecting as sys in Oracle 9i.
 2) Only check the dictionary size if the variable is not NULL; otherwise, an
    error takes place which is not caught or cleared; this eliminates a
    spurious "Objects/dictobject.c:1258: bad argument to internal function" in
    Python 2.3.
 3) Add support for session pooling. This is only support for Oracle 9i but
    is amazingly fast -- about 100 times faster than connecting.
 4) Add support for statement caching when pooling sessions, this reduces the
    parse time considerably. Unfortunately, the Oracle OCI does not allow this
    to be easily turned on for normal sessions.
 5) Add method trim() on CLOB and BLOB variables for trimming the size.
 6) Add support for externally identified users; to use this feature leave the
    username and password fields empty when connecting.
 7) Add method cancel() on connection objects to cancel long running queries.
    Note that this only works on non-Windows platforms.
 8) Add method callfunc() on cursor objects to allow calling a function
    without using an anonymous PL/SQL block.
 9) Added documentation on objects that were not documented. At this point all
    objects, methods and constants in cx_Oracle have been documented.
10) Added support for timestamp columns in Oracle 9i.
11) Added module level method makedsn() which creates a data source name given
    the host, port and SID.
12) Added constant "buildtime" which is the time when the module was built as
    an additional means of identifying the build that is in use.
13) Binding a value that is incompatible to the previous value that was bound
    (data types do not match or array size is larger) will now result in a
    new bind taking place. This is more consistent with the DB API although
    it does imply a performance penalty when used.

Changes from 3.0 to 3.0a
 1) Fixed bug where zero length PL/SQL arrays were being mishandled
 2) Fixed support for the data type "float" in Oracle; added one to the
    display size to allow for the sign of the number, if necessary; changed
    the display size of unconstrained numbers to 127, which is the largest
    number that Oracle can handle
 3) Added support for retrieving the description of a bound cursor before
    fetching it
 4) Fixed a couple of build issues on Mac OS X, AIX and Solaris (64-bit)
 5) Modified documentation slightly based on comments from several people
 6) Included files in MANIFEST that are needed to generate the binaries
 7) Modified test suite to work within the test environment at Computronix
    as well as within the packages that are distributed

Changes from 2.5a to 3.0
 1) Removed support for connection to Oracle7 databases; it is entirely
    possible that it will still work but I no longer have any way of testing
    and Oracle has dropped any meaningful support for Oracle7 anyway
 2) Fetching of strings is now done with predefined memory areas rather than
    dynamic memory areas; dynamic fetching of strings was causing problems
    with Oracle 9i in some instances and databases using a different character
    set other than US ASCII
 3) Fixed bug where segfault would occur if the '/' character preceded the '@'
    character in a connect string
 4) Added two new cursor methods var() and arrayvar() in order to eliminate
    the need for setinputsizes() when defining PL/SQL arrays and as a generic
    method of acquiring bind variables directly when needed
 5) Fixed support for binding cursors and added support for fetching cursors
    (these are known as ref cursors in PL/SQL).
 6) Eliminated discrepancy between the array size used internally and the
    array size specified by the interface user; this was done earlier to avoid
    bus errors on 64-bit platforms but another way has been found to get
    around that issue and a number of people were getting confused because of
    the discrepancy
 7) Added support for the attribute "connection" on cursors, an optional
    DB API extension
 8) Added support for passing a dictionary as the second parameter for the
    cursor.execute() method in order to comply with the DB API more closely;
    the method of passing parameters with keyword arguments is still supported
    and is in fact preferred
 9) Added support for the attribute "statement" on cursors which is a
    reference to the last SQL statement prepared or executed
10) Added support for passing any sequence to callproc() rather than just
    lists as before
11) Fixed bug where segfault would occur if the array size was changed after
    the cursor was executed but before it was fetched
12) Ignore array size when performing executemany() and use the length of the
    list of arguments instead
13) Rollback when connection is closed or destroyed to follow DB API rather
    than use the Oracle default (which is commit)
14) Added check for array size too large causing an integer overflow
15) Added support for iterators for Python 2.2 and above
16) Added test suite based on PyUnitTest
17) Added documentation in HTML format similar to the documentation for the
    core Python library

Changes from 2.5 to 2.5a
 1) Fix problem with Oracle 9i and retrieving strings; it seems that Oracle 9i
    uses the correct method for dynamic callback but Oracle 8i will not work
    with that method so an #ifdef was added to check for the existence of an
    Oracle 9i feature; thanks to Paul Denize for discovering this problem

Changes from 2.4 to 2.5
 1) Added flag OPT_NoOracle7 which, if set, assumes that connections are being
    made to Oracle8 or higher databases; this allows for eliminating the
    overhead in performing this check at connect time
 2) Added flag OPT_NumbersAsStrings which, if set, returns all numbers as
    strings rather than integers or floats; this flag is used when defined
    variables are created (during select statements only)
 3) Added flag OPT_Threading which, if set, uses OCI threading mode; there is a
    significant performance degradation in this mode (about 15-20%) but it does
    allow threads to share connections (threadsafety level 2 according to the
    Python Database API 2.0); note that in order to support this, Oracle 8i or
    higher is now required
 4) Added Py_BEGIN_ALLOW_THREADS and Py_END_ALLOW_THREADS pairs where
    applicable to support threading during blocking OCI calls
 5) Added global method attach() to cx_Oracle to support attaching to an
    existing database handle (as provided by PowerBuilder, for example)
 6) Eliminated the cursor method fetchbinds() which was used for returning the
    list of bind variables after execution to get the values of out variables;
    the cursor method setinputsizes() was modified to return the list of bind
    variables and the cursor method execute() was modified to return the list
    of defined variables in the case of a select statement being executed;
    these variables have three methods available to them: getvalue([<pos>]) to
    get the value of a variable, setvalue(<pos>, <value>) to set its value and
    copy(<var>, <src_pos>, <targ_pos>) to copy the value from a variable in a
    more efficient manner than setvalue(getvalue())
 7) Implemented cursor method executemany() which expects a list of
    dictionaries for the arguments
 8) Implemented cursor method callproc()
 9) Added cursr method prepare() which parses (prepares) the statement for
    execution; subsequent execute() or executemany() calls can pass None as the
    statement which will imply use of the previously prepared statement; used
    for high performance only
10) Added cursor method fetchraw() which will perform a raw fetch of the cursor
    returning the number of rows thus fetched; this is used to avoid the
    overhead of generating result sets; used for high performance only
11) Added cursor method executemanyprepared() which is identical to the method
    executemany() except that it takes a single argument which is the number of
    times to execute a previously prepared statement and it assumes that the
    bind variables already have their values set; used for high performance
    only
12) Added support for rowid being returned in a select statement
13) Added support for comparing dates returned by cx_Oracle
14) Integrated patch from Andre Reitz to set the null ok flag in the
    description attribute of the cursor
15) Integrated patch from Andre Reitz to setup.py to support compilation with
    Python 1.5
16) Integrated patch from Benjamin Kearns to setup.py to support compilation
    on Cygwin

Changes from 2.3 to 2.4

1) String variables can now be made any length (previously restricted to the
   64K limit imposed by Oracle for default binding); use the type
   cx_Oracle.LONG_STRING as the argument to setinputsizes() for binding in
   string values larger than 4000 bytes.
2) Raw and long raw columns are now supported; use the types cx_Oracle.BINARY
   and cx_Oracle.LONG_BINARY as the argument to setinputsizes() for binding in
   values of these types.
3) Functions DateFromTicks(), TimeFromTicks() and TimestampFromTicks()
   are now implemented.
4) Function cursor.setoutputsize() implemented
5) Added the ability to bind arrays as out parameters to procedures; use the
   format [cx_Oracle.<DataType>, <NumElems>] as the input to the function
   setinputsizes() for binding arrays
6) Discovered from the Oracle 8.1.6 version of the documentation of the OCI
   libraries, that the size of the memory location required for the precision
   variable is larger than the printed documentation says; this was causing a
   problem with the code on the Sun platform.
7) Now support building RPMs for Linux.

Changes from 2.2 to 2.3

 1) Incremental performance enhancements (dealing with reusing cursors and
    bind handles)
 2) Ensured that arrays of integers with a single float in them are all
    treated as floats, as suggested by Martin Koch.
 3) Fixed code dealing with scale and precision for both defining a numeric
    variable and for providing the cursor description; this eliminates the
    problem of an underflow error (OCI-22054) when retrieving data with
    non-zero scale.

Changes from 2.1 to 2.2

 1) Upgraded thread safety to level 1 (according to the Python DB API 2.0) as
    an internal project required the ability to share the module between
    threads.
 2) Added ability to bind ref cursors to PL/SQL blocks as requested by
    Brad Powell.
 3) Added function write(Value, [Offset]) to LOB variables as requested by
    Matthias Kirst.
 4) Procedure execute() on Cursor objects now permits a value None for the
    statement which means that the previously prepared statement will be
    executed and any input sizes set earlier will be retained. This was done to
    improve the performance of scripts that execute one statement many times.
 5) Modified module global constants BINARY and DATETIME to point to the
    external representations of those types so that the expression
    type(var) == cx_Oracle.DATETIME will work as expected.
 6) Added global constant version to provide means of determining the current
    version of the module.
 7) Modified error checking routine to distinguish between an Oracle error and
    invalid handles.
 8) Added error checking to avoid setting the value of a bind variable to a
    value that it cannot support and raised an exception to indicate this fact.
 9) Added extra compile arguments for the AIX platform as suggested by Jehwan
    Ryu.
10) Added section to the README to indicate the method for a binary
    installation as suggested by Steve Holden.
11) Added simple usage example as requested by many people.
12) Added HISTORY file to the distribution.