The driver would be used in async Python environments, and then need to allow async calls on connections.

The rust driver has native asynchronous & pipelining support, we could find a way to benefit from them on Python side.

The performances of the current binding are bad due to multiple calls to the underlying driver, and multiple object instanciations on Python side.

e.g., for a list(Connection.from_env().query('SELECT a, b, c FROM foo WHERE d = $1 LIMIT 2', 'bar')) we have the following calls:

• connect
• new_query
• query_param
• query_exec_result
• 2 times:
  • next_row
  • row_len
  • row_item (3 times)
  • row_close
• result_close
• close

Each call resulting in 2 new Python objects.

We could improve the binding to reduce the number of calls, by handling arrays instead of using only scalar types.
And we could avoid many Python objects as their interface is not exposed to the end-user.

Hi:
what would be the license for this fine code?
Thanks!

Currently, the driver handles int/float/str as Python input types, and int2/int4/int8/float8/text/bpchar/varchar/json/jsonb/uuid as PG output types.

It should handle all PG types, and conversions from standard Python types (as much as possible).

The binding code uses a lot of unsafe sections to keep pointers alive hen returning them to Python.

There may be some places where these allocated resources are not correctly freed.
And other places where we may execute invalid operations on pointers that could result in segmentation faults.