Version 4.1.0

• Release wheel also for MacOS ARM 64 architectures to PyPi. Thanks to FlyCI for their free tier!

Version 4.0.0

• Removed parameter driver_returns_memory_garbage_for_indicators from read_arrow_batches_from_odbc as it was intended as a workaround for IBM/DB2 drivers. Turns out IBM offers drivers which work correctly with 64Bit driver managers. Look for file names ending in 'o'.
• Add support for mapping inferred schemas via the map_schema parameter. It accepts a callable taking and returning an Arrow Schema. This allows you to avoid data types which are not supported in downstream operations and map them e.g. to string. It also enables you to work around quirks of your ODBC driver and map float 32 to float 64 if precisions inferred by the driver are too small.

Version 3.0.1

• Updated Rust dependencies. This includes an update to odbc-api which fixes a bug in decimal parsing. Decimal values with a magnitude smaller 1 and negative sign would have been interpreted as positive without this fix.

Version 3.0.0

• read_arrow_batches_from_odbc now requires the paramteter driver_returns_memory_garbage_for_indicators explicitly set to True in order to compensate for weaknesses in the IBM/DB2 driver and no longer tries to autodetect this. This change in interface has been made in order to not accidentially apply the workaround to drivers which in actuallity report indices just fine.

Version 2.1.6

• BatchReader.more_results now shares the same default value for fetch buffer size as read_arrow_batches_from_odbc.

Version 2.1.5

• Updated Rust dependencies. This includes an update to odbc-api which activates db2 specific workaround for any platform

Version 2.1.4

v2.1.4

decimal now robust against missing trailing zeroes

Version 2.1.3

v2.1.3

Fix decimal parsing - update arrow odbc

Version 2.1.2

v2.1.2

more robust decimal parsing

Version 2.1.1

v2.1.1

change default max batch size in bytes to 256 MiB