You're encountering a hang because Python’s Global Interpreter Lock (GIL) is not correctly managed in a multithreaded context. When Py_Initialize() is called, the main thread holds the GIL by default. However, if you later call PyGILState_Ensure() from other threads (like Taskflow tasks) without releasing the GIL in the main thread first, you cause a deadlock — the other threads block forever waiting for the GIL.

After Py_Initialize(), immediately release the GIL using PyEval_SaveThread() before spawning threads:

Py_Initialize();

PyEval_InitThreads(); // deprecated in Python 3.9+, but harmless

PyThreadState* tstate = PyEval_SaveThread(); // releases GIL

Then in each task/thread that wants to run Python code:
PyGILState_STATE gil = PyGILState_Ensure();

// ... run Python code

PyGILState_Release(gil);

Finally, before Py_FinalizeEx(), restore the GIL to the main thread:
PyEval_RestoreThread(tstate);

Py_FinalizeEx();

Notes:

• Never call Python APIs from threads without ensuring the GIL.

• Python 3.9+ automatically initializes threading, so PyEval_InitThreads() is legacy-safe.

• Always balance PyGILState_Ensure() with PyGILState_Release().

Regards