Running sub-interpreters in parallel with Python 3.12

parallel_python_subinterpreters.cpp

#include <Windows.h>
#include <Python.h>

void worker(PyThreadState * tstate) {
    PyThreadState_Swap(tstate);
    PyRun_SimpleString("fib = lambda x: x if x < 2 else fib(x - 1) + fib(x - 2)");
    PyRun_SimpleString("print(f'{fib(38) = }')");
}

PyThreadState * new_interpreter() {
    PyInterpreterConfig config = {};
    config.use_main_obmalloc = 0;
    config.allow_fork = 0;
    config.allow_exec = 0;
    config.allow_threads = 0;
    config.allow_daemon_threads = 0;
    config.check_multi_interp_extensions = 1;
    config.gil = PyInterpreterConfig_OWN_GIL;
    PyThreadState * tstate = NULL;
    PyThreadState * save_tstate = PyThreadState_Get();
    PyStatus status = Py_NewInterpreterFromConfig(&tstate, &config);
    PyThreadState_Swap(save_tstate);
    return tstate;
}

int main() {
    PyConfig config = {};
    // PyConfig_InitIsolatedConfig(&config);
    PyConfig_InitPythonConfig(&config);
    Py_InitializeFromConfig(&config);

    const int num_threads = 4;
    PyThreadState * interpreter[num_threads] = {};
    for (int i = 0; i < num_threads; ++i) {
        interpreter[i] = new_interpreter();
    }
    // PyThreadState_Swap(NULL);

    HANDLE threads[num_threads] = {};
    for (int i = 0; i < num_threads; ++i) {
        threads[i] = CreateThread(NULL, 0, (LPTHREAD_START_ROUTINE)worker, interpreter[i], 0, NULL);
    }
    WaitForMultipleObjects(num_threads, threads, true, INFINITE);
}