Correct way to shut down Python interpreter avoiding false leakage abort?

[oscargus]

cocotb shuts down the Python interpreter as:

https://github.com/cocotb/cocotb/blob/279e6bc055bc01f394d519b10ac00880521edc9a/src/cocotb/share/lib/embed/gpi_embed.cpp#L191-L204

However, this triggers the leakage code as not everything seems to be cleaned up at this stage, so the question is if there is a more correct way to do this? (It is sort of a cocotb problem, but I am pretty sure that more is known about the shutdown procedure and the leakage test here...)

See apytypes/apytypes#533 for a bit more context. However, we created a minimal nanobind module based on #69 (comment) (but without the dummy part, so no reference cycle(?)) and just importing Color as part of the the cocotb testbench triggers the leakage leading to abort which is not well handled by the rest of the code...

A stack trace is found in apytypes/apytypes#530 (comment) and it is clearly the call to Py_Finalize that triggers the leakage check. Is there something else that should be done before calling Py_Finalize to avoid this?

(We can of course disable the leakage check in our module, but it would be nice to solve it properly as cocotb will have this problem for any module using nanobind...)

[oscargus]

The abort went away when using the nanobind-wrapper from WrapDB (we had our own that seems to have been bad). Still leakage though, but it may be related to using static variables on the C++-side.