- Improve compatibility with ODBC drivers (e.g. FreeTDS) that do not
support ODBC's
SQLDescribeParam()
function by using a default parameter type. - Improve compatibility with some ODBC drivers (e.g. Microsoft's official MSSQL ODBC driver) for setting timestamps with fractional seconds.
- Add support for chaining operations to
Cursor.execute()
andCursor.executemany()
. This allows one-liners such ascursor.execute("SELECT 42").fetchallnumpy()
. - Right before a database connection is closed, any open transactions are explicitly rolled back. This improves compatibility with ODBC drivers that do not perform automatic rollbacks such as Microsoft's official ODBC driver.
- Improved stability of turbodbc when facing errors while closing connections, statements, and environments. In earlier versions, connection timeouts etc. could have lead to the Python process's termination.
- Source distribution contains license, readme, and changelog.
- Add support for OSX
- Added support for Python 3. Python 2 is still supported as well. Tested with Python 2.7, 3.4, 3.5, and 3.6.
- Added
six
package as dependency - Turbodbc uses pybind11 instead of Boost.Python to generate its Python bindings. pybind11 is available as a Python package and automatically installed when you install turbodbc. Other boost libraries are still required for other aspects of the code.
- A more modern compiler is required due to the pybind11 dependency. GCC 4.8 will suffice.
- Internal: Move remaining stuff depending on python to turbodbc_python
- Internal: Now requires cmake 2.8.12+ (get it with
pip install cmake
)
- Fixed build issue with older numpy versions, e.g., 1.8 (thanks @xhochy)
- Improved performance of parameter-based operations.
- Internal: Major modifications to the way parameters are handled.
- The size of the input buffers for retrieving result sets can now be set
to a certain amount of memory instead of using a fixed number of rows.
Use the optional
read_buffer_size
parameter ofturbodbc.connect()
and set it to instances of the new top-level classesMegabytes
andRows
(thanks @LukasDistel). - The read buffer size's default value has changed from 1,000 rows to 20 MB.
- The parameter
rows_to_buffer
ofturbodbc.connect()
is deprecated. You can set theread_buffer_size
toturbodbc.Rows(1000)
for the same effect, though it is recommended to specify the buffer size in MB. - Internal: Libraries no longer link
libpython.so
for local development (linking is already done by the Python interpreter). This was always the case for the libraries in the packages uploaded to PyPI, so no change was necessary here. - Internal: Some modifications to the structure of the underlying C++ code.
- NumPy support is introduced to turbodbc for retrieving result sets.
Use
cursor.fetchallnumpy
to retrieve a result set as anOrderedDict
ofcolumn_name: column_data
pairs, wherecolumn_data
is a NumPyMaskedArray
of appropriate type. - Internal: Single
turbodbc_intern
library was split up into three libraries to keep NumPy support optional. A few files were moved because of this.
-
turbodbc now supports asynchronous I/O operations for retrieving result sets. This means that while the main thread is busy converting an already retrieved batch of results to Python objects, another thread fetches an additional batch in the background. This may yield substantial performance improvements in the right circumstances (results are retrieved in roughly the same speed as they are converted to Python objects).
Ansynchronous I/O support is experimental. Enable it with
turbodbc.connect('My data source name', use_async_io=True)
- C++ backend:
turbodbc::column
no longer automatically binds on construction. Callbind()
instead.
- Result set rows are returned as native Python lists instead of a not easily printable custom type.
- Improve performance of Python object conversion while reading result sets. In tests with an Exasol database, performance got about 15% better.
- C++ backend:
turbodbc::cursor
no longer allows direct access to the C++field
type. Instead, please use thecursor
'sget_query()
method, and construct aturbodbc::result_sets::field_result_set
using theget_results()
method.
- Fix issue that only lists were allowed for specifying parameters for queries
- Improve parameter memory consumption when the database reports very large string parameter sizes
- C++ backend: Provides more low-level ways to access the result set
- Fix issue that
dsn
parameter was always present in the connection string even if it was not set by the user's call toconnect()
- Internal: First version to run on Travis.
- Internal: Use pytest instead of unittest for testing
- Internal: Allow for integration tests to run in custom environment
- Internal: Simplify integration test configuration
- Internal: Change C++ test framework to Google Test
- New parameter types supported:
bool
,datetime.date
,datetime.datetime
cursor.rowcount
returns number of affected rows for manipulating queriesConnection
supportsrollback()
- Improved handling of string parameters
Initial release