python-oracledb Release Notes

For deprecations, see Deprecations.

oracledb 2.2.0 (TBD)

Thin Mode Changes

Thick Mode Changes

Common Changes

  1. Added support for database type oracledb.DB_TYPE_INTERVAL_YM which is represented in Python by instances of the new oracledb.IntervalYM class (issue 310).

  2. Added support for getting a list of the network service names found in a tnsnames.ora file by adding the method ConnectParams.get_network_service_names() (issue 313).

  3. Added support for iterating over DbObject instances that are collections (issue 314).

  4. Error ORA-24545: invalid value of POOL_BOUNDARY specified in connect string is now raised consistently for both Thick and Thin modes. Previously, Thin mode was raising the error DPY-4030: invalid DRCP pool boundary {boundary}.

  5. Tightened definition of easy connect string regular expression to avoid parsing errors with connect descriptors containing the / character.

oracledb 2.1.1 (March 2024)

Thin Mode Changes

  1. Fixed bug when calling gettype() with an object type name containing %ROWTYPE (issue 304).

  2. Error DPY-2048: the bind variable placeholder ":{name}" cannot be used both before and after the RETURNING clause in a DML RETURNING statement is now raised when the same bind variable placeholder name is used both before and after the RETURNING clause in a DML RETURNING statement. Previously, various internal errors were raised.

  3. Restored the error message raised when attempting to connect to Oracle Database 11g.

  4. Internal change: tightened up code looking for the end of a database request.

  5. Internal change: packet output is now immediately flushed in order to avoid losing output due to buffering when multiple threads are running.

Thick Mode Changes

Common Changes

oracledb 2.1.0 (March 2024)

Thin Mode Changes

  1. Oracle Database 23c feature support:

  2. asyncio changes:

    • Support for asyncio is no longer considered a pre-release.

    • Internal change to improve handling of packets.

    • Fixed bug when using DRCP.

    • Fixed bug in processing metadata that spans multiple packets.

    • Fixed bug when connecting to a database using listener redirects (issue 285).

  3. Added support for Easy Connect strings found in tnsnames.ora files.

  4. Added support for writing UTF-8 encoded bytes to CLOB and NCLOB values and writing strings to BLOB values in order to be consistent with what is done for string variables.

  5. User-defined errors raised by the database no longer display an error help portal URL.

  6. Fixed potential cursor issues when using Database Resident Connection Pooling (DRCP).

  7. Fixed regression when using IAM token authentication (issue 288).

  8. Fixed bug connecting to databases that are only mounted and not opened (issue 294).

  9. Fixed bug in identifying bind variables in SQL statements containing a single line comment at the end of the statement.

  10. Fixed bug in determining the list of attributes for PL/SQL collections.

  11. Fixed bug in calculating the Connection.thin attribute.

  12. Fixed type declaration for the connectiontype parameter to oracledb.create_pool_async() and the return value of AsyncConnectionPool.acquire().

Thick Mode Changes

  1. Added support for internal use of JSON in SODA with Oracle Client 23. This allows for seamless transfer of extended data types.

  2. Fixed bug when calling SodaDoc.getContent() for SODA documents that do not contain JSON.

  3. Corrected support for sharding.

  4. Errors DPY-4011: the database or network closed the connection and DPY-4024: call timeout of {timeout} ms exceeded now retain the original error message raised by the Oracle Client library.

Common Changes

  1. Added a boolean property FetchInfo.is_oson which is set when a column has the check constraint IS JSON FORMAT OSON enabled.

  2. Added methods Connection.decode_oson() and Connection.encode_oson() to support fetching and inserting into columns which have the check constraint IS JSON FORMAT OSON enabled.

  3. Added class oracledb.JsonId to represent JSON ID values returned by SODA in Oracle Database 23.4 and later in the _id attribute of documents stored in native collections.

  4. Added support for columns of type VECTOR (currently requires access to a limited availability release of the database).

  5. Errors raised when calling Cursor.executemany() with PL/SQL now have the oracledb._Error.offset attribute populated with the last iteration that succeeded (issue 283).

  6. A number of performance improvements were made.

  7. Error DPY-2045: arraysize must be an integer greater than zero is now raised when an invalid value is used for the attribute Cursor.arraysize. Previously, a variety of errors (TypeError, OverflowError or ORA-03147: missing mandatory TTC field) were raised.

  8. Error DPY-2016: variable array size of %d is too small (should be at least %d) is now raised when Cursor.executemany() is called with an integer number of iterations that is too large for the existing bind variables. Previously, the python-oracledb Thin mode raised IndexError and python-oracledb Thick mode raised DPI-1018: array size of %d is too small.

  9. Error DPY-1001: not connected to database is now raised when an attempt is made to perform an operation on a LOB using a closed connection. Previously, the python-oracledb Thin mode raised an AttributeError exception and python-oracledb Thick mode raised DPI-1040: LOB was already closed.

  10. Fixed bug in ConnectParams.get_connect_string() when a value for the connection parameter purity has been specified.

  11. Fixed bug in ConnectParams.set() that would clear the ssl_context, appcontext, shardingkey and supershardingkey parameters if they were not included in the parameters. This also affected calls to oracledb.connect() and oracledb.create_pool() that made use of the DSN with credentials format.

  12. The error DPY-2047: LOB amount must be greater than zero is now raised when the amount parameter in LOB.read() is set to zero or negative.

  13. Fixed bug in the calculation of Cursor.rowcount under some circumstances.

  14. Connection parameters that are strings now treat an empty string in the same way as the value None.

oracledb 2.0.1 (January 2024)

Thin Mode Changes

  1. Added support for using alternative event loop implementations such as uvloop with asyncio (issue 276).

  2. Added support for the asynchronous context manager protocol on the AsyncCursor class as a convenience.

  3. Fixed regression when connecting to a database using listener redirects with either a connection pool or using asyncio (issue 275).

  4. Fixed bug when an intermittent hang occurs on some versions of Oracle Database while using asyncio and the database raises an error and output variables are present (issue 278).

  5. Fixed bug when fetch variables contain output converters and a query is re-executed (issue 271).

  6. Corrected typing declaration for oracledb.connect_async().

  7. Internal change to ensure that connection pools are closed gracefully when the main thread terminates.

  8. Internal change to slightly improve performance of LOB reads and writes.

Common Changes

  1. Fixed regression which prevented a null value from being set on DbObject attributes or used as elements of collections (issue 273).

  2. Fixed regression from cx_Oracle which ignored the value of the encoding_errors parameter when creating variables by calling the method Cursor.var() (issue 279).

  3. Bumped minimum requirement of Cython to 3.0.

oracledb 2.0.0 (December 2023)

Thin Mode Changes

  1. Added support for concurrent programming with asyncio (issue 6).

  2. Added parameter ConnectParams.sdu for configuring the Session Data Unit (SDU) size for sizing internal buffers used for tuning communication with the database. The connection property Connection.sdu was also added.

  3. Added parameter ConnectParams.ssl_context to modify the SSL context used when connecting via TLS (issue 259).

  4. Added support for an Oracle Database 23c JSON feature allowing field names with more than 255 UTF-8 encoded bytes.

  5. Added support for the FAILOVER clause in full connect descriptors.

  6. Fixed bug in detecting the current time zone (issue 257).

  7. Fixed bug in handling database response in certain unusual circumstances.

  8. Fixed bug in handling exceptions raised during connection establishment.

  9. Fixed bug in identifying bind variables in SQL statements containing multiple line comments with multiple asterisks before the closing slash.

  10. A more meaningful error is raised when the wrong type of data is passed to LOB.write().

  11. Internal change to support an Oracle Database 23c JSON feature improving JSON storage usage.

  12. Internal change to ensure that all connections in a pool have been closed gracefully before the pool is closed.

  13. Internal changes to improve handling of the network protocol between python-oracledb and Oracle Database.

  14. Internal changes to improve handling of multiple address and description lists in full connect descriptors.

Thick Mode Changes

  1. Fixed bug in return value of SodaOperation.replaceOne().

Common Changes

  1. Dropped support for Python 3.6.

  2. Desupported a number of parameters and attributes that were previously deprecated. See desupport notices for details.

  3. Added property Cursor.warning for database warnings (such as PL/SQL compilation warnings) generated by calls to Cursor.execute() or Cursor.executemany().

  4. Added property Connection.warning for warnings (such as the password being in the grace period) generated during connection.

  5. Added properties that provide information about the database: Connection.db_domain, Connection.db_name, Connection.max_open_cursors, Connection.service_name and Connection.transaction_in_progress.

  6. Added property Connection.proxy_user to show the name of the user which was used as a proxy when connecting (issue 250).

  7. Added properties FetchInfo.domain_schema, FetchInfo.domain_name and FetchInfo.annotations for the SQL domain and annotations associated with columns that are being fetched. SQL domains and annotations require Oracle Database 23c. If using python-oracledb Thick mode, Oracle Client 23c is also required.

  8. Added parameter data to Connection.createlob() to allow data to be written at LOB creation time.

  9. Added type DB_TYPE_XMLTYPE to represent data of type SYS.XMLTYPE in the database. Previously the value of FetchInfo.type_code for data of this type was DB_TYPE_LONG in Thick mode and DB_TYPE_OBJECT in Thin mode.

  10. Attribute and element values of Oracle Object instances that contain strings or bytes now have their maximum size constraints checked. Errors DPY-2043 (attributes) and DPY-2044 (element values) are now raised when constraints are violated.

  11. Attribute and element values of Oracle Object instances that are numbers are now returned as integers if the precision and scale allow for it. This is the same way that numbers are fetched from the database (issue 99).

  12. Errors that have entries in the troubleshooting documentation now have links to that documentation shown in the message text.

  13. Fixed bug with binding boolean values with Oracle Database 23c (issue 263).

  14. Fixed bug with getting unknown attributes from Oracle Object instances.

  15. Error DPY-4029: errors in array DML exceed 65535 is now raised when the number of batch errors exceeds 65535 when calling Cursor.executemany() with the parameter batcherrors set to the value True. Note that in thick mode this error is not raised unless the number of batch errors is a multiple of 65536; instead, the number of batch errors returned is modulo 65536 (issue 262).

  16. Black is now used to format Python code and ruff to lint Python code.

oracledb 1.4.2 (October 2023)

Thick Changes

  1. Fixed bug resulting in a segfault on some platforms when using two-phase commit.

Common Changes

  1. Pre-built binaries are now being created for Python 3.12 (issue 237).

oracledb 1.4.1 (September 2023)

Thin Mode Changes

  1. Improved statement bind variable placeholder parser performance, handle statements which use the Alternative Quoting Mechanism (‘Q’ strings), and fix some issues identifying bind variable placeholders in embedded quotes and in JSON syntax.

Thick Changes

  1. Fixed error checking when getting or setting the connection pool parameters ping_interval and soda_metadata_cache.

Common Changes

  1. Fixed bug when calling Cursor.execute() or Cursor.executemany() with missing bind data after calling Cursor.setinputsizes() with at least one of the values supplied as None (issue 217).

  2. SQL statement parsing now raises DPY-2041: missing ending quote (') in string or DPY-2042: missing ending quote (") in identifier for statements with the noted invalid syntax. Previously, thick mode gave ORA-1756 or ORA-1740, respectively, while thin mode did not throw an error.

  3. Added missing “>” to repr() of SodaDatabase Objects.

oracledb 1.4.0 (August 2023)

Thin Mode Changes

  1. Added support for an Oracle Database 23c feature that can improve the performance of connection creation by reducing the number of round trips required to create the second and subsequent connections to the same database.

  2. Added support for shrinking the connection pool back to the specified minimum size when the pool is idle for ConnectionPool.timeout seconds.

  3. Added support for growing the connection pool back to the minimum number of connections after connections are killed or otherwise made unusable.

  4. A default connection class is now generated when DRCP is used with a connection pool and no connection class was specified when the pool was created. The default connection class will be of the form DPY: followed by a 16-byte unique identifier converted to base64 encoding.

  5. Changed internal connection feature negotiation for more accurate Oracle Database 23c support.

  6. Added support for sending a generated connection identifier to the database used for tracing. An application specific prefix is prepended to this value if specified via a new connection_id_prefix parameter when creating standalone connections or connection pools.

  7. Added URL to the Oracle Database Error Help Portal in Oracle Database error messages similar to when Thick mode uses Oracle Client 23c.

  8. Added support for the ORA_SDTZ environment variable used to set the session time zone used by the database.

  9. Fixed bug when a dynamically sized connection pool is created with an increment of zero and the pool needs to grow.

  10. Fixed bug affecting connection reuse when connections were acquired from the connection pool with a cclass different to the one used to create the pool.

  11. Fixed bug when a connection is discarded from the connection pool during ConnectionPool.acquire() and the ping check fails due to the connection being dead.

  12. Fixed bug when an output type handler is used and the value of Cursor.prefetchrows exceeds Cursor.arraysize (issue 173).

  13. Fixed bug when an Application Continuity replay context is returned during connection to the database (issue 176).

  14. Fixed bug when socket is not closed immediately upon failure to establish a connection to the database (issue 211).

Thick Mode Changes

  1. Added function SodaCollection.listIndexes() for getting the indexes on a SODA collection.

  2. Added support for specifying if documents should be locked when fetched from SODA collections. A new non-terminal method lock() was added which requires Oracle Client 21.3 or higher (or Oracle Client 19 from 19.11).

  3. Relaxed restriction for end-to-end tracing string connection attributes. These values can now be set to the value None which will be treated the same as an empty string.

  4. Fixed bug when using external authentication with an Easy Connect connection string.

  5. Fixed memory leak when accessing objects embedded within other objects.

Common Changes

  1. Use of Python 3.6 and 3.7 is deprecated and support for them will be removed in a future release. A warning is issued when these versions are used but otherwise they will continue to function as usual. The warning can be suppressed by importing warnings and adding a call like warnings.filterwarnings(action='ignore', module="oracledb") before importing oracledb.

  2. Added support for the outconverter being called when a null value is fetched from the database and the new parameter convert_nulls to the method Cursor.var() is passed the value True (issue 107).

  3. Replaced fixed 7-tuple for the cursor metadata found in Cursor.description with a class which provides additional information such as the database object type and whether the column contains JSON data.

  4. Changed the signature for output type handlers to handler(cursor, metadata) where the metadata parameter is a FetchInfo object containing the same information found in Cursor.description. The original signature for output type handlers is deprecated and will be removed in a future version.

  5. Added support for fetching VARCHAR2 and LOB columns which contain JSON (and have the “IS JSON” check constraint enabled) in the same way as columns of type JSON (which requires Oracle Database 21c or higher) are fetched. In thick mode this requires Oracle Client 19c or higher. The attribute oracledb.__future__.old_json_col_as_obj must be set to the value True for this behavior to occur. In version 2.0 this will become the normal behavior and setting this attribute will no longer be needed.

  6. Added new property Connection.instance_name which provides the Oracle Database instance name associated with the connection. This is the same value as the SQL expression sys_context('userenv', 'instance_name').

  7. Added support for relational queries on the underlying tables of SODA collections created in Oracle Database 23c if they contain JSON documents with embedded OIDs.

  8. Automatically retry a query if the error ORA-00932: inconsistent data types is raised (which can occur if a table or view is recreated with a data type that is incompatible with the column’s previous data type).

  9. The repr() value of the DbObject class now shows the string “DbObject” instead of the string “Object” for consistency with the name of the class and the other repr() values for DbObjectType and DbObjectAttr.

  10. Fixed bug when binding sequences other than lists and tuples (issue 205).

  11. Added support for using the Cython 3.0 release (issue 204).

  12. Improved test suite and documentation.

oracledb 1.3.2 (June 2023)

Thin Mode Changes

  1. Fixed bug using Cursor.arraysize for tuning data fetches from REF CURSORS.

  2. Fixed bug connecting to databases with older 11g password verifiers (issue 189).

  3. Fixed bugs in the implementation of the statement cache.

  4. Fixed bug which caused a cursor leak if an error was thrown while processing the execution of a query.

  5. Eliminated unneeded round trip when using token authentication to connect to the database.

  6. Fixed bug which could cause a redirect loop with improperly configured listener redirects.

  7. Fixed bug when executing PL/SQL with a large number of binds.

  8. Fixed bug when using DRCP with Oracle Database 23c.

Thick Mode Changes

  1. Fixed bug when using external authentication with a Net Service Name connection string (issue 178).

  2. Fixed bug when using external authentication with an Easy Connect connection string.

Common Changes

  1. When fetching rows from REF CURSORS, the cursor’s prefetchrows attribute is now ignored. Use Cursor.arraysize for tuning these fetches. This change allows consistency between Thin and Thick modes.

oracledb 1.3.1 (April 2023)

Thin Mode Changes

  1. Improved performance of regular expressions used for parsing SQL (issue 172).

  2. Fixed bug with Oracle Database 23c when SQL is executed after first being parsed.

  3. Fixed bug when ConnectionPool.timeout is not None when creating a connection pool (issue 166).

  4. Fixed bug when a query is re-executed after an underlying table is dropped and recreated, and the query select list contains LOBs or JSON data.

  5. Fixed bug when warning message such as for impending password expiry is encountered during connect (issue 171).

Common Changes

  1. Improved test suite and samples.

oracledb 1.3.0 (March 2023)

Thin Mode Changes

  1. Added direct support for the Oracle Database 21c JSON data type, removing the need to use an output type handler.

  2. Added implementation for ConnectionPool.timeout to allow pools to shrink to min connections.

  3. Added check to prevent adding too many elements to bounded database collections.

  4. Removed internally set fixed size for database collections. Collections of any size supported by the database can now be created.

  5. Added support for connecting to databases that accept passwords longer than 30 UTF-8 encoded bytes.

  6. Detect the time zone on the OS and set the session timezone using this value to be consistent with thick mode (issue 144).

  7. Improved BOOLEAN handling.

  8. Error DPY-6005: cannot connect to database is now raised for all failures to connect to the database and the phrase cannot connect to database is removed from all other error messages (since this can be confusing when these errors are raised from ConnectParams.parse_connect_string()).

  9. Fixed bug when calling Cursor.executemany() with PL/SQL when the size of the bound data increases on subsequent calls (issue 132).

  10. Fixed bug when binding data of type TIMESTAMP WITH TIME ZONE but with zero fractional seconds.

  11. Fixed bug with incorrect values of Cursor.rowcount when fetching data (issue 147).

  12. Fixed bug with SQL containing multibyte characters with certain database character sets (issue 133).

  13. Fixed bug with ordering of binds in SQL when the database version is 12.1 (issue 135).

  14. Fixed bug with ordering of binds in PL/SQL when the bind variable may potentially exceed the 32767 byte limit but the actual value bound does not (issue 146).

  15. Fixed bug connecting to an IPv6 address with IAM tokens.

  16. Fixed bug determining RETURNING binds in a SQL statement when RETURNING and INTO keywords are not separated by whitespace, but are separated by parentheses.

  17. The exception DPY-3022: named time zones are not supported in thin mode is now raised when attempting to fetch data of type TIMESTAMP WITH TIME ZONE when the time zone associated with the data is a named time zone. Previously invalid data was returned (disc 131).

  18. Internal implementation changes:

    • Added internal support for prefetching the LOB size and chunk size, thereby eliminating a round-trip when calling LOB.size() and LOB.getchunksize().

    • Made the pool implementation LIFO to improve locality, reduce the number of times any session callback must be invoked, and allow connections to be timed out.

    • Removed packet for negotiating network services which are not supported in thin mode.

    • Removed unneeded packet for changing the password of the connected user.

Thick Mode Changes

  1. Raise a more meaningful error when an unsupported type in a JSON value is detected.

  2. Added support for the “signed int”, “signed long” and “decimal128” scalar types in JSON (generally only seen when converting from MongoDB).

  3. Defer raising an exception when calling Connection.gettype() for a type containing an attribute or element with an unsupported data type until the first attempt to reference the attribute or element with the unsupported data type.

  4. Fixed bug when attempting to create bequeath connections when the DSN contains credentials.

Common Changes

  1. Improved type annotations.

  2. Added method ConnectParams.parse_dsn_with_credentials() for parsing a DSN that contains credentials.

  3. Error DPY-2038: element at index {index} does not exist is now raised whenever an element in a database collection is missing. Previously, thick mode raised DPI-1024: element at index {index} does not exist and thin mode raised KeyError or IndexError.

  4. Error DPY-2039: given index {index} must be in the range of {min_index} to {max_index} is now raised whenever an element in a database collection is set outside the bounds of the collection. Previously, thick mode raised OCI-22165: given index [{index}] must be in the range of [{min_index}] to [{max_index}] and thin mode raised IndexError.

  5. Error DPY-2040: parameters "batcherrors" and "arraydmlrowcounts" may only be true when used with insert, update, delete and merge statements is now raised when either of the parameters batcherrors and arraydmlrowcounts is set to the value True when calling Cursor.executemany(). Previously, thick mode raised DPI-1063: modes DPI_MODE_EXEC_BATCH_ERRORS and DPI_MODE_EXEC_ARRAY_DML_ROWCOUNTS can only be used with insert, update, delete and merge statements and thin mode raised ORA-03137: malformed TTC packet from client rejected (issue 128).

  6. Internal changes to ensure that errors taking place while raising exceptions are handled more gracefully.

oracledb 1.2.2 (January 2023)

Thin Mode Changes

  1. Any exception raised while finding the operating system user for database logging is now ignored (issue 112).

  2. Fixed bug when binding OUT a NULL boolean value. (issue 119).

  3. Fixed bug when getting a record type based on a table (%ROWTYPE) (issue 123).

  4. Fixed bug when using a select * from table query and columns are added to the table (issue 125).

Thick Mode Changes

  1. Fixed bug when attempting to create bequeath connections to a local database (issue 114).

Common Changes

  1. Fixed bug when attempting to populate an array variable with too many elements.

oracledb 1.2.1 (December 2022)

Thin Mode Changes

  1. Fixed bug determining RETURNING binds in a SQL statement when RETURNING and INTO keywords are not separated by spaces, but are separated by other whitespace characters (issue 104).

  2. Fixed bug determining bind variables when found between two comment blocks (issue 105).

Thick Mode Changes

  1. Fixed bug creating a homogeneous connection pool with a proxy user (issue 101).

  2. Fixed bug closing a SODA document cursor explicitly (instead of simply allowing it to be closed automatically when it goes out of scope).

  3. Fixed bug when calling Subscription.registerquery() with bind values.

  4. Fixed bug that caused Message.dbname to always be the value None.

Common Changes

  1. Corrected __repr__() of connections to include the actual class name instead of a hard-coded oracledb.

oracledb 1.2.0 (November 2022)

Thin Mode Changes

  1. Added support for binding and fetching data of type DB_TYPE_OBJECT. Note that some of the error codes and messages have changed as a result: DPY errors are raised instead of ones specific to ODPI-C and OCI (issue 43).

  2. Added support for fetching SYS.XMLTYPE data as strings. Note that unlike in Thick mode, fetching longer values does not require using XMLTYPE.GETCLOBVAL().

  3. Added support for using a wallet for one-way TLS connections, rather than requiring OS recognition of certificates (issue 65).

  4. Added support for connecting to CMAN using (SOURCE_ROUTE=YES) in the connect string (issue 81).

  5. Fixed bug when fetching nested cursors with more columns than the parent cursor.

  6. Fixed bug preventing a cursor from being reused after it was bound as a REF CURSOR to a PL/SQL block that closes it.

  7. Fixed bug preventing binding OUT data of type DB_TYPE_UROWID that exceeds 3950 bytes in length.

  8. Fixed bug preventing correct parsing of connect descriptors with both ADDRESS and ADDRESS_LIST components at the same level.

  9. The complete connect string is now sent to the server instead of just the actual components being used. This is important for some configurations.

  10. Fixed bug resulting in an internal protocol error when handling database responses.

  11. Fixed bug when calling Cursor.executemany() with the batcherrors parameter set to True multiple times with each call resulting in at least one batch error.

Thick Mode Changes

  1. Connections acquired from a homogeneous pool now show the username and dsn to which they are connected in their repr().

Common Changes

  1. Added support for Python 3.11.

  2. Added attribute DbObjectType.package_name which contains the name of the package if the type is a PL/SQL type (otherwise, it will be None).

  3. Added sample for loading data from a CSV file.

  4. Improved test suite and documentation.

oracledb 1.1.1 (September 2022)

Thin Mode Changes

  1. Fixed bug that prevented binding data of types DB_TYPE_ROWID and DB_TYPE_UROWID.

  2. Fixed bug that caused Connection.is_healthy() to return True after a connection has been killed.

  3. Internally, before a connection is returned from a pool, perform additional checks in order to avoid returning a dead connection from the pool.

Thick Mode Changes

  1. Fixed bug returning metadata of SODA documents inserted into a collection using SodaCollection.saveAndGet().

Common Changes

  1. Fixed type checking errors (issue 52).

  2. Enhanced type checking (issue 54), (issue 60).

  3. The mode of python-oracledb is now fixed only after a call to oracledb.init_oracle_client(), oracledb.connect() or oracledb.create_pool() has completed successfully (issue 44).

  4. Improved test suite and documentation.

oracledb 1.1.0 (September 2022)

Thin Mode Changes

  1. Added support for getting the LOB chunk size (issue 14).

  2. The error DPY-2030: LOB offset must be greater than zero is now raised when the offset parameter to LOB.read() is zero or negative (issue 13).

  3. Internally, before a connection is returned from a pool, check for control packets from the server (which may inform the client that the connection needs to be closed and a new one established).

  4. Internally make use of the TCP_NODELAY socket option to remove delays in socket reads.

  5. Fixed bug when calling Cursor.parse() multiple times with the same SQL statement.

  6. Fixed bug that prevented connecting to Oracle Database 12.1.0.1.

  7. Fixed bug that prevented the database error message from being returned when connecting to a database that the listener configuration file states exists but actually doesn’t (issue 51).

  8. The error DPY-3016: python-oracledb thin mode cannot be used because the cryptography package is not installed is now raised when the cryptography package is not installed, instead of an ImportError. This allows platforms that are not capable of building the cryptography package to still use Thick mode.

  9. Fixed bug that prevented the full_code attribute from being populated on the errors returned by Cursor.getbatcherrors().

Thick Mode Changes

  1. Added support for getting the message id of the AQ message which generated a notification.

  2. Added support for enqueuing and dequeing AQ messages as JSON.

  3. Added the ability to use externalauth as a connection parameter for standalone connections in addition to creating pools. For standalone connections, this parameter is optional.

Common Changes

  1. Added support for Azure Active Directory OAuth 2.0 and Oracle Cloud Infrastructure Identity and Access Management (IAM) token authentication via the new parameter access_token to oracledb.connect() and oracledb.create_pool().

  2. Added method oracledb.is_thin_mode() to support determining whether the driver is using Thin mode or not (issue 16).

  3. Improved samples and documentation.

oracledb 1.0.3 (August 2022)

Thin Mode Changes

  1. The error DPY-3015: password verifier type is not supported by python-oracledb in thin mode is now raised when the database sends a password challenge with a verifier type that is not recognized, instead of ORA-01017: invalid username/password (issue 26).

  2. Fixed bug with handling of redirect data returned by some SCAN listeners (issue 39).

  3. Fixed bug with re-execution of SQL that requires a define, such as occurs when setting oracledb.defaults.fetch_lobs to the value False (issue 41).

  4. Fixed bug that prevented cursors from implicit results sets from being closed.

Common Changes

  1. Fixed bug with the deferral of type assignment when creating variables for Cursor.executemany() (issue 35).

oracledb 1.0.2 (July 2022)

Thin Mode Changes

  1. Connecting to a database with national character set UTF8 is now supported; an error is now raised only when the first attempt to use NCHAR, NVARCHAR2 or NCLOB data is made (issue 16).

  2. Fixed a bug when calling cursor.executemany() with a PL/SQL statement and a single row of data (issue 30).

  3. When using the connection parameter https_proxy while using protocol tcp, a more meaningful exception is now raised: DPY-2029: https_proxy requires use of the tcps protocol.

  4. Fixed a bug that caused TLS renegotiation to be skipped in some configurations, thereby causing the connection to fail to be established (https://github.com/oracle/python-oracledb/discussions/34).

Thick Mode Changes

  1. Fixed the ability to use external authentication with connection pools.

Common Changes

  1. The compiler flag -arch x86_64 no longer needs to be explicitly specified when building from source code on macOS (Intel x86) without Universal Python binaries.

  2. Binary packages have been added for the Linux ARM 64-bit platform.

  3. Improved samples and documentation.

oracledb 1.0.1 (June 2022)

Thin Mode Changes

  1. Added support for multiple aliases in one entry in tnsnames.ora (issue 3).

  2. Fixed connection retry count handling to work in cases where the database listener is running but the service is down (issue 3).

  3. Return the same value for TIMESTAMP WITH TIME ZONE columns as thick mode (issue 7).

  4. Fixed order in which bind data is sent to the server when LONG and non-LONG column data is interspersed (issue 12).

  5. If an error occurs during the creation of a connection to the database, the error is wrapped by DPY-6005 (so that it can be caught with an exception handler on class oracledb.DatabaseError).

  6. Ensured that errors occurring during fetch are detected consistently.

  7. Fixed issue when fetching null values in implicit results.

  8. Small performance optimization when sending column metadata.

Thick Mode Changes

  1. Fixed the ability to create bequeath connections to a local database.

  2. Fixed issue fetching NCLOB columns with oracledb.defaults.fetch_lobs = False.

Common Changes

  1. Fixed issue where unconstrained numbers containing integer values would be fetched as floats when oracledb.defaults.fetch_lobs = False. (issue 15).

  2. Ensured connection error messages contain the function name instead of wrapped().

  3. Improved samples, including adding a Dockerfile that starts a container with a running database and the samples.

  4. A binary package has been added for Python 3.7 on macOS (Intel x86).

  5. Improved documentation.

oracledb 1.0.0 (May 2022)

  1. Renamed cx_Oracle to python-oracledb. See Appendix C: The python-oracledb and cx_Oracle Drivers.

  2. Python-oracledb is a ‘Thin’ driver by default that connects directly to Oracle Database. Optional use of Oracle Client libraries enables a ‘Thick’ mode with some additional functionality. Both modes support the Python Database API v2.0 Specification.

  3. Added a Connection.thin attribute which shows whether the connection was established in the python-oracledb Thin mode or Thick mode.

  4. Creating connections or connection pools now requires keyword parameters be passed. This brings python-oracledb into compliance with the Python Database API specification PEP 249.

  5. Threaded mode is now always enabled for standalone connections (Thick mode).

  6. The function oracledb.init_oracle_client() must now always be called to load Oracle Client libraries, which enables Thick mode.

  7. Allow oracledb.init_oracle_client() to be called multiple times in each process as long as the same parameters are used each time.

  8. Improved some connection and binding error messages (Thin mode only).

  9. Added oracledb.defaults containing attributes that can be used to adjust the default behavior of the python-oracledb driver. In particular oracledb.defaults.fetch_lobs obsoletes the need for a LOB type handler .

  10. Added a ConnectParams Class which provides the ability to define connection parameters in one place.

  11. Added a PoolParams Class which provides the ability to define pool parameters in one place.

  12. Added a ConnectionPool Class which is equivalent to the SessionPool class previously used in cx_Oracle. The new oracledb.create_pool() function is now the preferred method for creating connection pools.

  13. Changed the default oracledb.create_pool() getmode parameter value to POOL_GETMODE_WAIT to remove potential transient errors when calling ConnectionPool.acquire() during pool growth.

  14. Connection pools in python-oracledb Thin mode support all connection mode privileges.

  15. Added new Two-phase commit functionality.

  16. Added Connection.is_healthy() to do a local check of a connection’s health.

  17. Added a boolean parameter cache_statement to Cursor.prepare(), giving applications control over statement caching.

  18. Made improvements to statement cache invalidation (Thin mode only)

  19. Added a recipient attribute to support recipient lists in Oracle Advanced Queuing.

  20. Added a full_code attribute to the Error object giving the top-level error prefix and the error number.

  21. Added a DB_TYPE_LONG_NVARCHAR constant.

cx_Oracle 8.3 (November 2021)

  1. Updated embedded ODPI-C to version 4.3.0.

  2. Added official support for Python 3.10.

  3. Support for dequeuing messages from Oracle Transactional Event Queue (TEQ) queues was restored.

  4. Corrected calculation of attribute MessageProperties.msgid. Note that the attribute is now also read only.

  5. Binary integer variables now explicitly convert values to integers (since implicit conversion to integer has become an error in Python 3.10) and values that are not int, float or decimal.Decimal are explicitly rejected.

  6. Improved samples and test suite.

cx_Oracle 8.2.1 (June 2021)

  1. Updated embedded ODPI-C to version 4.2.1.

  2. Added support for caching the database version in pooled connections with Oracle Client 19 and earlier (later Oracle Clients handle this caching internally). This optimization eliminates a round-trip previously often required when reusing a pooled connection.

  3. Fixed a regression with error messages when creating a connection fails.

  4. Fixed crash when using the deprecated parameter name keywordParameters with Cursor.callproc().

  5. Improved documentation and the test suite.

cx_Oracle 8.2 (May 2021)

  1. Updated embedded ODPI-C to version 4.2.0.

  2. Threaded mode is now always enabled when creating connection pools with cx_Oracle.SessionPool(). Any threaded parameter value is ignored.

  3. Added SessionPool.reconfigure() to support pool reconfiguration. This method provides the ability to change properties such as the size of existing pools instead of having to restart the application or create a new pool.

  4. Added parameter max_sessions_per_shard to cx_Oracle.SessionPool() to allow configuration of the maximum number of sessions per shard in the pool. In addition, the attribute SessionPool.max_sessions_per_shard was added in order to permit making adjustments after the pool has been created. They are usable when using Oracle Client version 18.3 and higher.

  5. Added parameter stmtcachesize to cx_Oracle.connect() and cx_Oracle.SessionPool() in order to permit specifying the size of the statement cache during the creation of pools and standalone connections.

  6. Added parameter ping_interval to cx_Oracle.SessionPool() to specify the ping interval when acquiring pooled connections. In addition, the attribute SessionPool.ping_interval was added in order to permit making adjustments after the pool has been created. In previous cx_Oracle releases a fixed ping interval of 60 seconds was used.

  7. Added parameter soda_metadata_cache to cx_Oracle.SessionPool() for SODA metadata cache support. In addition, the attribute SessionPool.soda_metadata_cache was added in order to permit making adjustments after the pool has been created. This feature significantly improves the performance of methods SodaDatabase.createCollection() (when not specifying a value for the metadata parameter) and SodaDatabase.openCollection(). Caching is available when using Oracle Client version 21.3 and higher (or Oracle Client 19 from 19.11).

  8. Added support for supplying hints to SODA operations. A new non-terminal method hint() was added and a hint parameter was added to the methods SodaCollection.insertOneAndGet(), SodaCollection.insertManyAndGet() and SodaCollection.saveAndGet(). All of these require Oracle Client 21.3 or higher (or Oracle Client 19 from 19.11).

  9. Added parameter bypass_decode to Cursor.var() in order to allow the decode step to be bypassed when converting data from Oracle Database into Python strings (issue 385). Initial work was done in PR 549.

  10. Enhanced dead connection detection. If an Oracle Database error indicates that a connection is no longer usable, the error DPI-1080: connection was closed by ORA-%d is now returned. The %d will be the Oracle error causing the connection to be closed. Using the connection after this will give DPI-1010: not connected. This behavior also applies for Connection.call_timeout errors that result in an unusable connection.

  11. Eliminated a memory leak when calling SodaOperation.filter() with a dictionary.

  12. The distributed transaction handle assosciated with the connection is now cleared on commit or rollback (issue 530).

  13. Added a check to ensure that when setting variables or object attributes, the type of the temporary LOB must match the expected type.

  14. A small number of parameter, method, and attribute names were updated to follow the PEP 8 style guide. This brings better consistency to the cx_Oracle API. The old names are still usable but may be removed in a future release of cx_Oracle. See Deprecated in cx_Oracle 8.2 for details.

  15. Improved the test suite.

cx_Oracle 8.1 (December 2020)

  1. Updated embedded ODPI-C to version 4.1.0.

  2. Added support for new JSON data type available in Oracle Client and Database 21 and higher.

  3. Dropped support for Python 3.5. Added support for Python 3.9.

  4. Added internal methods for getting/setting OCI attributes that are otherwise not supported by cx_Oracle. These methods should only be used as directed by Oracle.

  5. Minor code improvement supplied by Alex Henrie (PR 472).

  6. Builds are now done with setuptools and most metadata has moved from setup.py to setup.cfg in order to take advantage of Python packaging improvements.

  7. The ability to pickle/unpickle Database and API types has been restored.

  8. Tests can now be run with tox in order to automate testing of the different environments that are supported.

  9. The value of prefetchrows for REF CURSOR variables is now honored.

  10. Improved documentation, samples and test suite.

cx_Oracle 8.0.1 (August 2020)

  1. Updated embedded ODPI-C to version 4.0.2. This includes the fix for binding and fetching numbers with 39 or 40 decimal digits (issue 459).

  2. Added build metadata specifying that Python 3.5 and higher is required in order to avoid downloading and failing to install with Python 2. The exception message when running setup.py directly was updated to inform those using Python 2 to use version 7.3 instead.

  3. Documentation improvements.

cx_Oracle 8.0 (June 2020)

  1. Dropped support for Python 2.

  2. Updated embedded ODPI-C to version 4.0.1.

  3. Reworked type management to clarify and simplify code

    • Added constants for all database types. The database types cx_Oracle.DB_TYPE_BINARY_FLOAT, cx_Oracle.DB_TYPE_INTERVAL_YM, cx_Oracle.DB_TYPE_TIMESTAMP_LTZ and cx_Oracle.DB_TYPE_TIMESTAMP_TZ are completely new. The other types were found in earlier releases under a different name. These types will be found in Cursor.description and passed as the defaultType parameter to the Connection.outputtypehandler and Cursor.outputtypehandler functions.

    • Added synonyms from the old type names to the new type names for backwards compatibility. They are deprecated and will be removed in a future version of cx_Oracle.

    • The DB API constants are now a specialized constant that matches to the corresponding database types, as recommended by the DB API.

    • The variable attribute type now refers to one of the new database type constants if the variable does not contain objects (previously it was None in that case).

    • The attribute type was added to LOB values.

    • The attribute type was added to attributes of object types.

    • The attribute element_type was added to object types.

    • Object types now compare equal if they were created by the same connection or session pool and their schemas and names match.

    • All variables are now instances of the same class (previously each type was an instance of a separate variable type). The attribute type can be examined to determine the database type it is associated with.

    • The string representation of variables has changed to include the type in addition to the value.

  4. Added function cx_Oracle.init_oracle_client() in order to enable programmatic control of the initialization of the Oracle Client library.

  5. The default encoding for all character data is now UTF-8 and any character set specified in the environment variable NLS_LANG is ignored.

  6. Added functions SodaCollection.save(), SodaCollection.saveAndGet() and SodaCollection.truncate() available in Oracle Client 20 and higher.

  7. Added function SodaOperation.fetchArraySize() available in Oracle Client 19.5 and higher.

  8. Added attribute Cursor.prefetchrows to control the number of rows that the Oracle Client library fetches into internal buffers when a query is executed.

  9. Internally make use of new mode available in Oracle Client 20 and higher in order to avoid a round-trip when accessing Connection.version for the first time.

  10. Added support for starting up a database using a parameter file (PFILE), as requested (issue 295).

  11. Fixed overflow issue when calling Cursor.getbatcherrors() with row offsets exceeding 65536.

  12. Eliminated spurious error when accessing Cursor.lastrowid after executing an INSERT ALL statement.

  13. Miscellaneous improvements supplied by Alex Henrie (pull requests 419, 420, 421, 422, 423, 437 and 438).

  14. Python objects bound to boolean variables are now converted to True or False based on whether they would be considered True or False in a Python if statement. Previously, only True was treated as True and all other Python values (including 1, 1.0, and “foo”) were treated as False (pull request 435).

  15. Documentation, samples and test suite improvements.

cx_Oracle 7.3 (December 2019)

  1. Added support for Python 3.8.

  2. Updated embedded ODPI-C to version 3.3.

  3. Added support for CQN and other subscription client initiated connections to the database (as opposed to the default server initiated connections) created by calling Connection.subscribe().

  4. Added support for returning the rowid of the last row modified by an operation on a cursor (or None if no row was modified).

  5. Added support for setting the maxSessionsPerShard attribute when creating session pools.

  6. Added check to ensure sharding key is specified when a super sharding key is specified.

  7. Improved error message when the Oracle Client library is loaded successfully but the attempt to detect the version of that library fails, either due to the fact that the library is too old or the method could not be called for some reason (node-oracledb issue 1168).

  8. Adjusted support for creating a connection using an existing OCI service context handle. In order to avoid potential memory corruption and unsupported behaviors, the connection will now use the same encoding as the existing OCI service context handle when it was created.

  9. Added ORA-3156: OCI call timed out to the list of error messages that result in error DPI-1067.

  10. Adjusted samples and the test suite so that they can be run against Oracle Cloud databases.

  11. Fixed bug when attempting to create a scrollable cursor on big endian platforms like AIX on PPC.

  12. Eliminated reference leak and ensure that memory is properly initialized in case of error when using sharding keys.

  13. Eliminated reference leak when splitting the password and DSN components out of a full connect string.

  14. Corrected processing of DATE sharding keys (sharding requires a slightly different format to be passed to the server).

  15. Eliminated reference leak when creating message property objects.

  16. Attempting to use proxy authentication with a homogeneous pool will now raise a DatabaseError exception with the message DPI-1012: proxy authentication is not possible with homogeneous pools instead of a ProgrammingError exception with the message pool is homogeneous. Proxy authentication is not possible. since this check is done by ODPI-C. An empty string (or None) for the user name will no longer generate an exception.

  17. Exception InterfaceError: not connected is now always raised when an operation is attempted with a closed connection. Previously, a number of different exceptions were raised depending on the operation.

  18. Added ORA-40479: internal JSON serializer error to the list of exceptions that result in cx_Oracle.IntegrityError.

  19. Improved documentation.

cx_Oracle 7.2.3 (October 2019)

  1. Updated embedded ODPI-C to version 3.2.2.

  2. Restored support for setting numeric bind variables with boolean values.

  3. Ensured that sharding keys are dedicated to the connection that is acquired using them in order to avoid possible hangs, crashes or unusual errors.

  4. Corrected support for PLS_INTEGER and BINARY_INTEGER types when used in PL/SQL records (ODPI-C issue 112).

  5. Improved documentation.

cx_Oracle 7.2.2 (August 2019)

  1. Updated embedded ODPI-C to version 3.2.1.

  2. A more meaningful error is now returned when calling SodaCollection.insertMany() with an empty list.

  3. A more meaningful error is now returned when calling Subscription.registerquery() with SQL that is not a SELECT statement.

  4. Eliminated segfault when a connection is closed after being created by a call to cx_Oracle.connect() with the parameter cclass set to a non-empty string.

  5. Added user guide documentation.

  6. Updated default connect strings to use 19c and XE 18c defaults.

cx_Oracle 7.2.1 (July 2019)

  1. Resolved MemoryError exception on Windows when using an output type handler (issue 330).

  2. Improved test suite and samples.

  3. Improved documentation.

cx_Oracle 7.2 (July 2019)

  1. Updated embedded ODPI-C to version 3.2.

  2. Improved AQ support

    • added support for enqueue and dequeue of RAW payloads

    • added support for bulk enqueue and dequeue of messages

    • added new method Connection.queue() which creates a new queue object in order to simplify AQ usage

    • enhanced method Connection.msgproperties() to allow the writable properties of the newly created object to be initialized.

    • the original methods for enqueuing and dequeuing (Connection.deq(), Connection.deqoptions(), Connection.enq() and Connection.enqoptions()) are now deprecated and will be removed in a future version.

  3. Removed preview status from existing SODA functionality. See this tracking issue for known issues with SODA.

  4. Added support for a preview of SODA bulk insert, available in Oracle Client 18.5 and higher.

  5. Added support for setting LOB object attributes, as requested (issue 299).

  6. Added mode cx_Oracle.DEFAULT_AUTH as requested (issue 293).

  7. Added support for using the LOB prefetch length indicator in order to reduce the number of round trips when fetching LOBs and then subsequently calling LOB.size(), LOB.getchunksize() or LOB.read(). This is always enabled.

  8. Added support for types BINARY_INTEGER, PLS_INTEGER, ROWID, LONG and LONG RAW when used in PL/SQL.

  9. Eliminated deprecation of attribute Subscription.id. It is now populated with the value of REGID found in the database view USER_CHANGE_NOTIFICATION_REGS or the value of REG_ID found in the database view USER_SUBSCR_REGISTRATIONS. For AQ subscriptions, the value is 0.

  10. Enabled PY_SSIZE_T_CLEAN, as required by Python 3.8 (issue 317).

  11. Eliminated memory leak when fetching objects that are atomically null (issue 298).

  12. Eliminated bug when processing the string representation of numbers like 1e-08 and 1e-09 (issue 300).

  13. Improved error message when the parent cursor is closed before a fetch is attempted from an implicit result cursor.

  14. Improved test suite and samples.

  15. Improved documentation.

cx_Oracle 7.1.3 (April 2019)

  1. Updated to ODPI-C 3.1.4.

  2. Added support for getting the row count for PL/SQL statements (issue 285).

  3. Corrected parsing of connect string so that the last @ symbol is searched for instead of the first @ symbol; otherwise, passwords containing an @ symbol will result in the incorrect DSN being extracted (issue 290).

  4. Adjusted return value of cursor.callproc() to follow documentation (only positional arguments are returned since the order of keyword parameters cannot be guaranteed in any case) (PR 287).

  5. Corrected code getting sample and test parameters by user input when using Python 2.7.

cx_Oracle 7.1.2 (March 2019)

  1. Updated to ODPI-C 3.1.3.

  2. Ensured that the strings “-0” and “-0.0” are correctly handled as zero values (issue 274).

  3. Eliminated error when startup and shutdown events are generated (ODPI-C issue 102).

  4. Enabled the types specified in Cursor.setinputsizes() and Cursor.callfunc() to be an object type in addition to a Python type, just like in Cursor.var().

  5. Reverted changes to return decimal numbers when the numeric precision was too great to be returned accurately as a floating point number. This change had too great an impact on existing functionality and an output type handler can be used to return decimal numbers where that is desirable (issue 279).

  6. Eliminated discrepancies in character sets between an external connection handle and the newly created connection handle that references the external connection handle (issue 273).

  7. Eliminated memory leak when receiving messages received from subscriptions.

  8. Improved test suite and documentation.

cx_Oracle 7.1.1 (February 2019)

  1. Updated to ODPI-C 3.1.2.

  2. Corrected code for freeing CQN message objects when multiple queries are registered (ODPI-C issue 96).

  3. Improved error messages and installation documentation.

cx_Oracle 7.1 (February 2019)

  1. Updated to ODPI-C 3.1.

  2. Improved support for session tagging in session pools by allowing a session callback to be specified when creating a pool via cx_Oracle.SessionPool(). Callbacks can be written in Python or in PL/SQL and can be used to improve performance by decreasing round trips to the database needed to set session state. Callbacks written in Python will be invoked for brand new connections (that have never been acquired from the pool before) or when the tag assigned to the connection doesn’t match the one that was requested. Callbacks written in PL/SQL will only be invoked when the tag assigned to the connection doesn’t match the one that was requested.

  3. Added attribute Connection.tag to provide access to the actual tag assigned to the connection. Setting this attribute will cause the connection to be retagged when it is released back to the pool.

  4. Added support for fetching SYS.XMLTYPE values as strings, as requested (issue 14). Note that this support is limited to the size of VARCHAR2 columns in the database (either 4000 or 32767 bytes).

  5. Added support for allowing the typename parameter in method Cursor.var() to be None or a valid object type created by the method Connection.gettype(), as requested (issue 231).

  6. Added support for getting and setting attributes of type RAW on Oracle objects, as requested (ODPI-C issue 72).

  7. Added support for performing external authentication with proxy for standalone connections.

  8. Added support for mixing integers, floating point and decimal values in data passed to Cursor.executemany() (issue 241). The error message raised when a value cannot be converted to an Oracle number was also improved.

  9. Adjusted fetching of numeric values so that no precision is lost. If an Oracle number cannot be represented by a Python floating point number a decimal value is automatically returned instead.

  10. Corrected handling of multiple calls to method Cursor.executemany() where all of the values in one of the columns passed to the first call are all None and a subsequent call has a value other than None in the same column (issue 236).

  11. Added additional check for calling Cursor.setinputsizes() with an empty dictionary in order to avoid the error “cx_Oracle.ProgrammingError: positional and named binds cannot be intermixed” (issue 199).

  12. Corrected handling of values that exceed the maximum value of a plain integer object on Python 2 on Windows (issue 257).

  13. Added error message when attempting external authentication with proxy without placing the user name in [] (proxy authentication was previously silently ignored).

  14. Exempted additional error messages from forcing a statement to be dropped from the cache (ODPI-C issue 76).

  15. Improved dead session detection when using session pools for Oracle Client 12.2 and higher.

  16. Ensured that the connection returned from a pool after a failed ping (such as due to a killed session) is not itself marked as needing to be dropped from the pool.

  17. Eliminated memory leak under certain circumstances when pooled connections are released back to the pool.

  18. Eliminated memory leak when connections are dropped from the pool.

  19. Eliminated memory leak when calling Connection.close() after fetching collections from the database.

  20. Adjusted order in which memory is freed when the last references to SODA collections, documents, document cursors and collection cursors are released, in order to prevent a segfault under certain circumstances.

  21. Improved code preventing a statement from binding itself, in order to avoid a potential segfault under certain circumstances.

  22. Worked around OCI bug when attempting to free objects that are PL/SQL records, in order to avoid a potential segfault.

  23. Improved test suite and samples. Note that default passwords are no longer supplied. New environment variables can be set to specify passwords if desired, or the tests and samples will prompt for the passwords when needed. In addition, a Python script is now available to create and drop the schemas used for the tests and samples.

  24. Improved documentation.

cx_Oracle 7.0 (September 2018)

  1. Update to ODPI-C 3.0.

  2. Added support for Oracle Client 18 libraries.

  3. Added support for SODA (as preview). See SODA Database, SODA Collection and SODA Document for more information.

  4. Added support for call timeouts available in Oracle Client 18.1 and higher. See Connection.call_timeout.

  5. Added support for getting the contents of a SQL collection object as a dictionary, where the keys are the indices of the collection and the values are the elements of the collection. See function Object.asdict().

  6. Added support for closing a session pool via the function SessionPool.close(). Once closed, further attempts to use any connection that was acquired from the pool will result in the error “DPI-1010: not connected”.

  7. Added support for setting a LOB attribute of an object with a string or bytes (instead of requiring a temporary LOB to be created).

  8. Added support for the packed decimal type used by object attributes with historical types DECIMAL and NUMERIC (issue 212).

  9. On Windows, first attempt to load oci.dll from the same directory as the cx_Oracle module.

  10. SQL objects that are created or fetched from the database are now tracked and marked unusable when a connection is closed. This was done in order to avoid a segfault under certain circumstances.

  11. Re-enabled dead session detection functionality when using pools for Oracle Client 12.2 and higher in order to handle classes of connection errors such as resource profile limits.

  12. Improved error messages when the Oracle Client or Oracle Database need to be at a minimum version in order to support a particular feature.

  13. When a connection is used as a context manager, the connection is now closed when the block ends. Attempts to set cx_Oracle.__future__.ctx_mgr_close are now ignored.

  14. When a DML returning statement is executed, variables bound to it will return an array when calling Variable.getvalue(). Attempts to set cx_Oracle.__future__.dml_ret_array_val are now ignored.

  15. Support for Python 3.4 has been dropped.

  16. Added additional test cases.

  17. Improved documentation.

cx_Oracle 6.4.1 (July 2018)

  1. Update to ODPI-C 2.4.2.

    • Avoid buffer overrun due to improper calculation of length byte when converting some negative 39 digit numbers from string to the internal Oracle number format (ODPI-C issue 67).

  2. Prevent error “cx_Oracle.ProgrammingError: positional and named binds cannot be intermixed” when calling cursor.setinputsizes() without any parameters and then calling cursor.execute() with named bind parameters (issue 199).

cx_Oracle 6.4 (July 2018)

  1. Update to ODPI-C 2.4.1.

    • Added support for grouping subscriptions. See parameters groupingClass, groupingValue and groupingType to function Connection.subscribe().

    • Added support for specifying the IP address a subscription should use instead of having the Oracle Client library determine the IP address on its own. See parameter ipAddress to function Connection.subscribe().

    • Added support for subscribing to notifications when messages are available to dequeue in an AQ queue. The new constant cx_Oracle.SUBSCR_NAMESPACE_AQ should be passed to the namespace parameter of function Connection.subscribe() in order to get this functionality. Attributes Message.queueName and Message.consumerName will be populated in notification messages that are received when this namespace is used.

    • Added attribute Message.registered to let the notification callback know when the subscription that generated the notification is no longer registered with the database.

    • Added support for timed waits when acquiring a session from a session pool. Use the new constant cx_Oracle.SPOOL_ATTRVAL_TIMEDWAIT in the parameter getmode to function cx_Oracle.SessionPool() along with the new parameter waitTimeout.

    • Added support for specifying the timeout and maximum lifetime session for session pools when they are created using function cx_Oracle.SessionPool(). Previously the pool had to be created before these values could be changed.

    • Avoid memory leak when dequeuing from an empty queue.

    • Ensure that the row count for queries is reset to zero when the statement is executed (issue 193).

    • If the statement should be deleted from the statement cache, first check to see that there is a statement cache currently being used; otherwise, the error “ORA-24300: bad value for mode” will be raised under certain conditions.

  2. Added support for using the cursor as a context manager (issue 190).

  3. Added parameter “encodingErrors” to function Cursor.var() in order to add support for specifying the “errors” parameter to the decode() that takes place internally when fetching strings from the database (issue 162).

  4. Added support for specifying an integer for the parameters argument to Cursor.executemany(). This allows for batch execution when no parameters are required or when parameters have previously been bound. This replaces Cursor.executemanyprepared() (which is now deprecated and will be removed in cx_Oracle 7).

  5. Adjusted the binding of booleans so that outside of PL/SQL they are bound as integers (issue 181).

  6. Added support for binding decimal.Decimal values to cx_Oracle.NATIVE_FLOAT as requested (issue 184).

  7. Added checks on passing invalid type parameters to methods Cursor.arrayvar(), Cursor.callfunc() and Cursor.setinputsizes().

  8. Corrected handling of cursors and rowids in DML Returning statements.

  9. Added sample from David Lapp demonstrating the use of GeoPandas with SDO_GEOMETRY and a sample for demonstrating the use of REF cursors.

  10. Adjusted samples and documentation for clarity.

  11. Added additional test cases.

cx_Oracle 6.3.1 (May 2018)

  1. Update to ODPI-C 2.3.2.

    • Ensure that a call to unregister a subscription only occurs if the subscription is still registered.

    • Ensure that before a statement is executed any buffers used for DML returning statements are reset.

  2. Ensure that behavior with cx_Oracle.__future__.dml_ret_array_val not set or False is the same as the behavior in cx_Oracle 6.2 (issue 176).

cx_Oracle 6.3 (April 2018)

  1. Update to ODPI-C 2.3.1.

    • Fixed binding of LONG data (values exceeding 32KB) when using the function Cursor.executemany().

    • Added code to verify that a CQN subscription is open before permitting it to be used. Error “DPI-1060: subscription was already closed” will now be raised if an attempt is made to use a subscription that was closed earlier.

    • Stopped attempting to unregister a CQN subscription before it was completely registered. This prevents errors encountered during registration from being masked by an error stating that the subscription has not been registered!

    • Added error “DPI-1061: edition is not supported when a new password is specified” to clarify the fact that specifying an edition and a new password at the same time is not supported when creating a connection. Previously the edition value was simply ignored.

    • Improved error message when older OCI client libraries are being used that don’t have the method OCIClientVersion().

    • Fixed the handling of ANSI types REAL and DOUBLE PRECISION as implemented by Oracle. These types are just subtypes of NUMBER and are different from BINARY_FLOAT and BINARY_DOUBLE (issue 163).

    • Fixed support for true heterogeneous session pools that use different user/password combinations for each session acquired from the pool.

    • Added error message indicating that setting either of the parameters arraydmlrowcounts and batcherrors to True in Cursor.executemany() is only supported with insert, update, delete and merge statements.

  2. Fixed support for getting the OUT values of bind variables bound to a DML Returning statement when calling the function Cursor.executemany(). Note that the attribute dml_ret_array_val in cx_Oracle.__future__ must be set to True first.

  3. Added support for binding integers and floats as cx_Oracle.NATIVE_FLOAT.

  4. A cx_Oracle._Error object is now the value of all cx_Oracle exceptions raised by cx_Oracle. (issue 51).

  5. Added support for building cx_Oracle with a pre-compiled version of ODPI-C, as requested (issue 103).

  6. Default values are now provided for all parameters to cx_Oracle.SessionPool().

  7. Improved error message when an unsupported Oracle type is encountered.

  8. The Python GIL is now prevented from being held while performing a round trip for the call to get the attribute Connection.version (issue 158).

  9. Added check for the validity of the year for Python 2.x since it doesn’t do that itself like Python 3.x does (issue 166).

  10. Adjusted documentation to provide additional information on the use of Cursor.executemany() as requested (issue 153).

  11. Adjusted documentation to state that batch errors and array DML row counts can only be used with insert, update, delete and merge statements (issue 31).

  12. Updated tutorial to import common connection information from files in order to make setup a bit more generic.

cx_Oracle 6.2.1 (March 2018)

  1. Make sure cxoModule.h is included in the source archive (issue 155).

cx_Oracle 6.2 (March 2018)

  1. Update to ODPI-C 2.2.1.

    • eliminate error “DPI-1054: connection cannot be closed when open statements or LOBs exist” (issue 138).

    • avoid a round trip to the database when a connection is released back to the pool by preventing a rollback from being called when no transaction is in progress.

    • improve error message when the use of bind variables is attempted with DLL statements, which is not supported by Oracle.

    • if an Oracle object is retrieved from an attribute of another Oracle object or a collection, prevent the “owner” from being destroyed until the object that was retrieved has itself been destroyed.

    • correct handling of boundary numbers 1e126 and -1e126

    • eliminate memory leak when calling Connection.enq() and Connection.deq()

    • eliminate memory leak when setting NCHAR and NVARCHAR attributes of objects.

    • eliminate memory leak when fetching collection objects from the database.

  2. Added support for creating a temporary CLOB, BLOB or NCLOB via the method Connection.createlob().

  3. Added support for binding a LOB value directly to a cursor.

  4. Added support for closing the connection when reaching the end of a with code block controlled by the connection as a context manager, but in a backwards compatible way (issue 113). See cx_Oracle.__future__ for more information.

  5. Reorganized code to simplify continued maintenance and consolidate transformations to/from Python objects.

  6. Ensure that the number of elements in the array is not lost when the buffer size is increased to accommodate larger strings.

  7. Corrected support in Python 3.x for cursor.parse() by permitting a string to be passed, instead of incorrectly requiring a bytes object.

  8. Eliminate reference leak with LOBs acquired from attributes of objects or elements of collections.

  9. Eliminate reference leak when extending an Oracle collection.

  10. Documentation improvements.

  11. Added test cases to the test suite.

cx_Oracle 6.1 (December 2017)

  1. Update to ODPI-C 2.1.

    • Support was added for accessing sharded databases via sharding keys (new in Oracle 12.2). NOTE: the underlying OCI library has a bug when using standalone connections. There is a small memory leak proportional to the number of connections created/dropped. There is no memory leak when using session pools, which is recommended.

    • Added options for authentication with SYSBACKUP, SYSDG, SYSKM and SYSRAC, as requested (issue 101).

    • Attempts to release statements or free LOBs after the connection has been closed (by, for example, killing the session) are now prevented.

    • An error message was added when specifying an edition and a connection class since this combination is not supported.

    • Attempts to close the session for connections created with an external handle are now prevented.

    • Attempting to ping a database earlier than 10g results in ORA-1010: invalid OCI operation, but that implies a response from the database and therefore a successful ping, so treat it that way! (see https://github.com/rana/ora/issues/224 for more information).

    • Support was added for converting numeric values in an object type attribute to integer and text, as requested (ODPI-C issue 35).

    • Setting attributes DeqOptions.msgId and MessageProperties.msgId now works as expected.

    • The overflow check when using double values (Python floats) as input to float attributes of objects or elements of collections was removed as it didn’t work anyway and is a well-known issue that cannot be prevented without removing desired functionality. The developer should ensure that the source value falls within the limits of floats, understand the consequent precision loss or use a different data type.

    • Variables of string/raw types are restricted to 2 bytes less than 1 GB (1,073,741,822 bytes), since OCI cannot handle more than that currently.

    • Support was added for identifying the id of the transaction which spawned a CQN subscription message, as requested (ODPI-C issue 32).

    • Corrected use of subscription port number (issue 115).

    • Problems reported with the usage of FormatMessage() on Windows were addressed (ODPI-C issue 47).

    • On Windows, if oci.dll cannot be loaded because it is the wrong architecture (32-bit vs 64-bit), attempt to find the offending DLL and include the full path of the DLL in the message, as suggested. (ODPI-C issue 49).

    • Force OCI prefetch to always use the value 2; the OCI default is 1 but setting the ODPI-C default to 2 ensures that single row fetches don’t require an extra round trip to determine if there are more rows to fetch; this change also reduces the potential memory consumption when fetchArraySize was set to a large value and also avoids performance issues discovered with larger values of prefetch.

  2. Fix build with PyPy 5.9.0-alpha0 in libpython mode (PR 54).

  3. Ensure that the edition is passed through to the database when a session pool is created.

  4. Corrected handling of Python object references when an invalid keyword parameter is passed to cx_Oracle.SessionPool().

  5. Corrected handling of Connection.handle and the handle parameter to cx_Oracle.connect() on Windows.

  6. Documentation improvements.

  7. Added test cases to the test suite.

cx_Oracle 6.0.3 (November 2017)

  1. Update to ODPI-C 2.0.3.

    • Prevent use of uninitialized data in certain cases (issue 77).

    • Attempting to ping a database earlier than 10g results in error “ORA-1010: invalid OCI operation”, but that implies a response from the database and therefore a successful ping, so treat it that way!

    • Correct handling of conversion of some numbers to NATIVE_FLOAT.

    • Prevent use of NaN with Oracle numbers since it produces corrupt data (issue 91).

    • Verify that Oracle objects bound to cursors, fetched from cursors, set in object attributes or appended to collection objects are of the correct type.

    • Correct handling of NVARCHAR2 when used as attributes of Oracle objects or as elements of collections.

  2. Ensure that a call to setinputsizes() with an invalid type prior to a call to executemany() does not result in a type error, but instead gracefully ignores the call to setinputsizes() as required by the DB API (issue 75).

  3. Check variable array size when setting variable values and raise IndexError, as is already done for getting variable values.

cx_Oracle 6.0.2 (August 2017)

  1. Update to ODPI-C 2.0.2.

    • Don’t prevent connection from being explicitly closed when a fatal error has taken place (issue 67).

    • Correct handling of objects when dynamic binding is performed.

    • Process deregistration events without an error.

    • Eliminate memory leak when creating objects.

  2. Added missing type check to prevent coercion of decimal to float (issue 68).

  3. On Windows, sizeof(long) = 4, not 8, which meant that integers between 10 and 18 digits were not converted to Python correctly (issue 70).

  4. Eliminate memory leak when repeatedly executing the same query.

  5. Eliminate segfault when attempting to reuse a REF cursor that has been closed.

  6. Updated documentation.

cx_Oracle 6.0.1 (August 2017)

  1. Update to ODPI-C 2.0.1.

    • Ensure that queries registered via Subscription.registerquery() do not prevent the associated connection from being closed (ODPI-C issue 27).

    • Deprecated attribute Subscription.id as it was never intended to be exposed (ODPI-C issue 28). It will be dropped in version 6.1.

    • Add support for DML Returning statements that require dynamically allocated variable data (such as CLOBs being returned as strings).

  2. Correct packaging of Python 2.7 UCS4 wheels on Linux (issue 64).

  3. Updated documentation.

cx_Oracle 6.0 (August 2017)

  1. Update to ODPI-C 2.0.

    • Prevent closing the connection when there are any open statements or LOBs and add new error “DPI-1054: connection cannot be closed when open statements or LOBs exist” when this situation is detected; this is needed to prevent crashes under certain conditions when statements or LOBs are being acted upon while at the same time (in another thread) a connection is being closed; it also prevents leaks of statements and LOBs when a connection is returned to a session pool.

    • On platforms other than Windows, if the regular method for loading the Oracle Client libraries fails, try using $ORACLE_HOME/lib/libclntsh.so (ODPI-C issue 20).

    • Use the environment variable DPI_DEBUG_LEVEL at runtime, not compile time.

    • Added support for DPI_DEBUG_LEVEL_ERRORS (reports errors and has the value 8) and DPI_DEBUG_LEVEL_SQL (reports prepared SQL statement text and has the value 16) in order to further improve the ability to debug issues.

    • Correct processing of Cursor.scroll() in some circumstances.

  2. Delay initialization of the ODPI-C library until the first standalone connection or session pool is created so that manipulation of the environment variable NLS_LANG can be performed after the module has been imported; this also has the added benefit of reducing the number of errors that can take place when the module is imported.

  3. Prevent binding of null values from generating the exception “ORA-24816: Expanded non LONG bind data supplied after actual LONG or LOB column” in certain circumstances (issue 50).

  4. Added information on how to run the test suite (issue 33).

  5. Documentation improvements.

cx_Oracle 6.0 rc 2 (July 2017)

  1. Update to ODPI-C rc 2.

    • Provide improved error message when OCI environment cannot be created, such as when the oraaccess.xml file cannot be processed properly.

    • On Windows, convert system message to Unicode first, then to UTF-8; otherwise, the error message returned could be in a mix of encodings (issue 40).

    • Corrected support for binding decimal values in object attribute values and collection element values.

    • Corrected support for binding PL/SQL boolean values to PL/SQL procedures with Oracle client 11.2.

  2. Define exception classes on the connection object in addition to at module scope in order to simplify error handling in multi-connection environments, as specified in the Python DB API.

  3. Ensure the correct encoding is used for setting variable values.

  4. Corrected handling of CLOB/NCLOB when using different encodings.

  5. Corrected handling of TIMESTAMP WITH TIME ZONE attributes on objects.

  6. Ensure that the array position passed to var.getvalue() does not exceed the number of elements allocated in the array.

  7. Reworked test suite and samples so that they are independent of each other and so that the SQL scripts used to create/drop schemas are easily adjusted to use different schema names, if desired.

  8. Updated DB API test suite stub to support Python 3.

  9. Added additional test cases and samples.

  10. Documentation improvements.

cx_Oracle 6.0 rc 1 (June 2017)

  1. Update to ODPI-C rc 1.

  2. The method Cursor.setoutputsize() no longer needs to do anything, since ODPI-C automatically manages buffer sizes of LONG and LONG RAW columns.

  3. Handle case when both precision and scale are zero, as occurs when retrieving numeric expressions (issue 34).

  4. OCI requires that both encoding and nencoding have values or that both encoding and encoding do not have values. These parameters are used in functions cx_Oracle.connect() and cx_Oracle.SessionPool(). The missing value is set to its default value if one of the values is set and the other is not (issue 36).

  5. Permit use of both string and unicode for Python 2.7 for creating session pools and for changing passwords (issue 23).

  6. Corrected handling of BFILE LOBs.

  7. Add script for dropping test schemas.

  8. Documentation improvements.

cx_Oracle 6.0 beta 2 (May 2017)

  1. Added support for getting/setting attributes of objects or element values in collections that contain LOBs, BINARY_FLOAT values, BINARY_DOUBLE values and NCHAR and NVARCHAR2 values. The error message for any types that are not supported has been improved as well.

  2. Enable temporary LOB caching in order to avoid disk I/O as suggested.

  3. Added support for setting the debug level in ODPI-C, if desirable, by setting environment variable DPI_DEBUG_LEVEL prior to building cx_Oracle.

  4. Correct processing of strings in Cursor.executemany() when a larger string is found after a shorter string in the list of data bound to the statement.

  5. Correct handling of long Python integers that cannot fit inside a 64-bit C integer (issue 18).

  6. Correct creation of pool using external authentication.

  7. Handle edge case when an odd number of zeroes trail the decimal point in a value that is effectively zero (issue 22).

  8. Prevent segfault under load when the attempt to create an error fails.

  9. Eliminate resource leak when a standalone connection or pool is freed.

  10. Correct typo.

  11. Correct handling of REF cursors when the array size is manipulated.

  12. Prevent attempts from binding the cursor being executed to itself.

  13. Correct reference count handling of parameters when creating a cursor.

  14. Correct determination of the names of the bind variables in prepared SQL statements (which behaves a little differently from PL/SQL statements).

cx_Oracle 6.0 beta 1 (April 2017)

  1. Simplify building cx_Oracle considerably by use of ODPI-C. This means that cx_Oracle can now be built without Oracle Client header files or libraries and that at runtime cx_Oracle can adapt to Oracle Client 11.2, 12.1 or 12.2 libraries without needing to be rebuilt. This also means that wheels can now be produced and installed via pip.

  2. Added attribute SessionPool.stmtcachesize to support getting and setting the default statement cache size for connections in the pool.

  3. Added attribute Connection.dbop to support setting the database operation that is to be monitored.

  4. Added attribute Connection.handle to facilitate testing the creation of a connection using a OCI service context handle.

  5. Added parameters tag and matchanytag to the cx_Oracle.connect() and SessionPool.acquire() methods and added parameters tag and retag to the SessionPool.release() method in order to support session tagging.

  6. Added parameter edition to the cx_Oracle.SessionPool() method.

  7. Added support for universal rowids.

  8. Added support for DML Returning of multiple rows.

  9. Added attributes Variable.actualElements and Variable.values to variables.

  10. Added parameters region, sharding_key and super_sharding_key to the cx_Oracle.makedsn() method to support connecting to a sharded database (new in Oracle Database 12.2).

  11. Added support for smallint and float data types in Oracle objects, as requested.

  12. An exception is no longer raised when a collection is empty for methods Object.first() and Object.last(). Instead, the value None is returned to be consistent with the methods Object.next() and Object.prev().

  13. If the environment variables NLS_LANG and NLS_NCHAR are being used, they must be set before the module is imported. Using the encoding and nencoding parameters to the cx_Oracle.connect() and cx_Oracle.SessionPool() methods is a simpler alternative to setting these environment variables.

  14. Removed restriction on fetching LOBs across round trips to the database (eliminates error “LOB variable no longer valid after subsequent fetch”).

  15. Removed requirement for specifying a maximum size when fetching LONG or LONG raw columns. This also allows CLOB, NCLOB, BLOB and BFILE columns to be fetched as strings or bytes without needing to specify a maximum size.

  16. Dropped deprecated parameter twophase from the cx_Oracle.connect() method. Applications should set the Connection.internal_name and Connection.external_name attributes instead to a value appropriate to the application.

  17. Dropped deprecated parameters action, module and clientinfo from the cx_Oracle.connect() method. The appcontext parameter should be used instead as shown in this sample.

  18. Dropped deprecated attribute numbersAsString from cursor objects. Use an output type handler instead as shown in this sample.

  19. Dropped deprecated attributes cqqos and rowids from subscription objects. Use the qos attribute instead as shown in this sample.

  20. Dropped deprecated parameters cqqos and rowids from the Connection.subscribe() method. Use the qos parameter instead as shown in this sample.

cx_Oracle 5.3 (March 2017)

  1. Added support for Python 3.6.

  2. Dropped support for Python versions earlier than 2.6.

  3. Dropped support for Oracle clients earlier than 11.2.

  4. Added support for fetching implicit results (available in Oracle 12.1)

  5. Added support for Transaction Guard (available in Oracle 12.1).

  6. Added support for setting the maximum lifetime of pool connections (available in Oracle 12.1).

  7. Added support for large row counts (larger than 2 ** 32, available in Oracle 12.1)

  8. Added support for advanced queuing.

  9. Added support for scrollable cursors.

  10. Added support for edition based redefinition.

  11. Added support for creating, modifying and binding user defined types and collections.

  12. Added support for creating, modifying and binding PL/SQL records and collections (available in Oracle 12.1).

  13. Added support for binding native integers.

  14. Enabled statement caching.

  15. Removed deprecated variable attributes maxlength and allocelems.

  16. Corrected support for setting the encoding and nencoding parameters when creating a connection and added support for setting these when creating a session pool. These can now be used instead of setting the environment variables NLS_LANG and NLS_NCHAR.

  17. Use None instead of 0 for items in the Cursor.description attribute that do not have any validity.

  18. Changed driver name to match informal driver name standard used by Oracle for other drivers.

  19. Add check for maximum of 10,000 parameters when calling a stored procedure or function in order to prevent a possible improper memory access from taking place.

  20. Removed -mno-cygwin compile flag since it is no longer used in newer versions of the gcc compiler for Cygwin.

  21. Simplified test suite by combining Python 2 and 3 scripts into one script and separated out 12.1 features into a single script.

  22. Updated samples to use code that works on both Python 2 and 3

  23. Added support for pickling/unpickling error objects (Issue #23)

  24. Dropped support for callbacks on OCI functions.

  25. Removed deprecated types UNICODE, FIXED_UNICODE and LONG_UNICODE (use NCHAR, FIXED_NCHAR and LONG_NCHAR instead).

  26. Increased default array size to 100 (from 50) to match other drivers.

  27. Added support for setting the internal_name and external_name on the connection directly. The use of the twophase parameter is now deprecated. Applications should set the internal_name and external_name attributes directly to a value appropriate to the application.

  28. Added support for using application context when creating a connection. This should be used in preference to the module, action and clientinfo parameters which are now deprecated.

  29. Reworked database change notification and continuous query notification to more closely align with the PL/SQL implementation and prepare for sending notifications for AQ messages. The following changes were made:

    • added constant SUBSCR_QOS_BEST_EFFORT to replace deprecated constant SUBSCR_CQ_QOS_BEST_EFFORT

    • added constant SUBSCR_QOS_QUERY to replace deprecated constant SUBSCR_CQ_QOS_QUERY

    • added constant SUBSCR_QOS_DEREG_NFY to replace deprecated constant SUBSCR_QOS_PURGE_ON_NTFN

    • added constant SUBSCR_QOS_ROWIDS to replace parameter rowids for method Connection.subscribe()

    • deprecated parameter cqqos for method Connection.subscribe(). The qos parameter should be used instead.

    • dropped constants SUBSCR_CQ_QOS_CLQRYCACHE, SUBSCR_QOS_HAREG, SUBSCR_QOS_MULTICBK, SUBSCR_QOS_PAYLOAD, SUBSCR_QOS_REPLICATE, and SUBSCR_QOS_SECURE since they were never actually used

  30. Deprecated use of the numbersAsStrings attribute on cursors. An output type handler should be used instead.

cx_Oracle 5.2.1 (January 2016)

  1. Added support for Python 3.5.

  2. Removed password attribute from connection and session pool objects in order to promote best security practices (if stored in RAM in cleartext it can be read in process dumps, for example). For those who would like to retain this feature, a subclass of Connection could be used to store the password.

  3. Added optional parameter externalauth to SessionPool() which enables wallet based or other external authentication mechanisms to be used.

  4. Use the national character set encoding when required (when char set form is SQLCS_NCHAR); otherwise, the wrong encoding would be used if the environment variable NLS_NCHAR is set.

  5. Added support for binding boolean values to PL/SQL blocks and stored procedures (available in Oracle 12.1).

cx_Oracle 5.2 (June 2015)

  1. Added support for strings up to 32k characters (new in Oracle 12c).

  2. Added support for getting array DML row counts (new in Oracle 12c).

  3. Added support for fetching batch errors.

  4. Added support for LOB values larger than 4 GB.

  5. Added support for connections as SYSASM.

  6. Added support for building without any configuration changes to the machine when using instant client RPMs on Linux.

  7. Added types NCHAR, FIXED_NCHAR and LONG_NCHAR to replace the types UNICODE, FIXED_UNICODE and LONG_UNICODE (which are now deprecated). These types are available in Python 3 as well so they can be used to specify the use of NCHAR type fields when binding or using setinputsizes().

  8. Fixed binding of booleans in Python 3.x.

  9. Test suite now sets NLS_LANG if not already set.

  10. Enhanced documentation for connection.action attribute and added note on cursor.parse() method to make clear that DDL statements are executed when parsed.

  11. Removed remaining remnants of support Oracle 9i.

  12. Added __version__ attribute to conform with PEP 396.

  13. Ensure that sessions are released to the pool when calling connection.close() (Issue #2)

  14. Fixed handling of datetime intervals (Issue #7)

cx_Oracle 5.1.3 (May 2014)

  1. Added support for Oracle 12c.

  2. Added support for Python 3.4.

  3. Added support for query result set change notification. Thanks to Glen Walker for the patch.

  4. Ensure that in Python 3.x that NCHAR and NVARCHAR2 and NCLOB columns are retrieved properly without conversion issues. Thanks to Joakim Andersson for pointing out the issue and the possible solution.

  5. Fix bug when an exception is caught and then another exception is raised while handling that exception in Python 3.x. Thanks to Boris Dzuba for pointing out the issue and providing a test case.

  6. Enhance performance returning integers between 10 and 18 digits on 64-bit platforms that support it. Thanks for Shai Berger for the initial patch.

  7. Fixed two memory leaks.

  8. Fix to stop current_schema from throwing a MemoryError on 64-bit platforms on occasion. Thanks to Andrew Horton for the fix.

  9. Class name of cursors changed to real name cx_Oracle.Cursor.

cx_Oracle 5.1.2 (July 2012)

  1. Added support for LONG_UNICODE which is a type used to handle long unicode strings. These are not explicitly supported in Oracle but can be used to bind to NCLOB, for example, without getting the error “unimplemented or unreasonable conversion requested”.

  2. Set the row number in a cursor when executing PL/SQL blocks as requested by Robert Ritchie.

  3. Added support for setting the module, action and client_info attributes during connection so that logon triggers will see the supplied values, as requested by Rodney Barnett.

cx_Oracle 5.1.1 (October 2011)

  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.

cx_Oracle 5.1 (March 2011)

  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.

cx_Oracle 5.0.4 (July 2010)

  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.

cx_Oracle 5.0.3 (February 2010)

  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 parameters 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.

cx_Oracle 5.0.2 (May 2009)

  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.

cx_Oracle 5.0.1 (February 2009)

  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.

cx_Oracle 5.0 (December 2008)

  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.

cx_Oracle 4.4.1 (October 2008)

  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.

cx_Oracle 4.4 (June 2008)

  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.

cx_Oracle 4.3.3 (October 2007)

  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.

cx_Oracle 4.3.2 (August 2007)

  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 transaction 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.

cx_Oracle 4.3.1 (April 2007)

  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.

cx_Oracle 4.3 (March 2007)

  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.

cx_Oracle 4.2.1 (September 2006)

  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.

cx_Oracle 4.2 (July 2006)

  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 parameter 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 parameters the same as if no keyword parameters 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.

cx_Oracle 4.1.2 (December 2005)

  1. Restore support of Oracle 9i features when using the Oracle 10g client.

cx_Oracle 4.1.1 (December 2005)

  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 also be linked against. Thanks to Bernard Delmee for the information.

cx_Oracle 4.1 (January 2005)

  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.

cx_Oracle 4.1 beta 1 (September 2004)

  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 cursor method arrayvar() 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.

cx_Oracle 4.0.1 (February 2004)

  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.

cx_Oracle 4.0 (December 2003)

  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.

cx_Oracle 3.1 (August 2003)

  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.

cx_Oracle 3.0a (June 2003)

  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

cx_Oracle 3.0 (March 2003)

  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 parameters 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 parameters 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

cx_Oracle 2.5a (August 2002)

  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

cx_Oracle 2.5 (July 2002)

  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 parameters

  8. Implemented cursor method callproc()

  9. Added cursor 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 parameter 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

cx_Oracle 2.4 (January 2002)

  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 parameter 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 parameter 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.

cx_Oracle 2.3 (October 2001)

  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.

cx_Oracle 2.2 (July 2001)

  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.