How to change the behavior of getting a bound method to avoid GIL?

[huweiATgithub]

For a C++ class A that is exported to Python:

from bindings import A, g
a = A()
a.f. # This results in a bound method in python
g(a.f) # pybind11 will wrap a.f into a std::function and acquire GIL when it is called

For multithreaded programs that run a.f in multiple threads, one must release GIL in function g. Otherwise, a deadlock is presented as other threads running a.f will never successfully acquire GIL.
Even if one releases GIL in g, the multithreaded program is running serially.

Would it be possible to change the behavior of a.f so that instead of creating a bound method in Python, it creates a bound in C++, and return that function? so that operations on the class A will achieve nearly the same performance that native C++ has.