Python and fortran with ctypes: lists with objects

I need to write a library in fortran that will take an array of objects from Python, process it and return it back.

I've tried a lot of options, the last one looks like this:

Code Python:

import ctypes as ct
import os
import numpy as np

here = os.path.dirname(__file__)
lib_file = os.path.abspath(os.path.join(here, '..', 'fortran', 'svector_test_mod.dll'))
fortlib = ct.CDLL(lib_file, winmode=0)


class StateVector(ct.Structure):
    _fields_ = [
        ('x', ct.c_longdouble),
        ('y', ct.c_longdouble),
        ('z', ct.c_longdouble),
        ('vx', ct.c_longdouble),
        ('vy', ct.c_longdouble),
        ('vz', ct.c_longdouble),
        ('t', ct.c_longdouble),
    ]


size = 5
array = (StateVector * size)()

for i in range(size):
    array[i].x = ct.c_longdouble(2.5)
    array[i].y = ct.c_longdouble(2.5)
    array[i].z = ct.c_longdouble(2.5)
    array[i].vx = ct.c_longdouble(2.5)
    array[i].vy = ct.c_longdouble(2.5)
    array[i].vz = ct.c_longdouble(2.5)
    array[i].t = ct.c_longdouble(2.5)


np_objects = np.ctypeslib.as_array(array)

arr_ptr = np_objects.ctypes.data_as(ct.POINTER(ct.c_void_p))

fortlib.lists_change(ct.byref(arr_ptr), size)

Code Fortran:

module state_vector_mod
    use, intrinsic :: iso_c_binding, only: C_LONG_DOUBLE, c_int, C_PTR, C_F_POINTER
    implicit none

    type, bind(C) :: state_vector
        real(C_LONG_DOUBLE) :: x
        real(C_LONG_DOUBLE) :: y
        real(C_LONG_DOUBLE) :: z
        real(C_LONG_DOUBLE) :: vx
        real(C_LONG_DOUBLE) :: vy
        real(C_LONG_DOUBLE) :: vz
        real(C_LONG_DOUBLE) :: t
    end type state_vector

    contains

    subroutine make_sv(x, y, z, vx, vy, vz, t, sv) bind(c, name='make_sv')
        real(C_LONG_DOUBLE), intent(in) :: x
        real(C_LONG_DOUBLE), intent(in) :: y
        real(C_LONG_DOUBLE), intent(in) :: z
        real(C_LONG_DOUBLE), intent(in) :: vx
        real(C_LONG_DOUBLE), intent(in) :: vy
        real(C_LONG_DOUBLE), intent(in) :: vz
        real(C_LONG_DOUBLE), intent(in) :: t
        type(state_vector), intent(out), target :: sv

        sv = state_vector(x, y, z, vx, vy, vz, t)
    end subroutine make_sv

    subroutine lists_change(in_list, length) bind(C, name='lists_change')
        type(C_PTR), intent(inout) :: in_list
        !type(state_vector), dimension(length), intent(out) :: out_list
        type(state_vector), dimension(:), pointer :: my_array
        integer(c_int), intent(in) :: length
        integer(c_int) :: i

        call C_F_POINTER(in_list, my_array, [length])

        DO i = 1, length
            my_array(i)%x = my_array(i)%x * 2
            my_array(i)%y = my_array(i)%y * 2
            my_array(i)%z = my_array(i)%z * 2
            my_array(i)%vx = my_array(i)%vx * 2
            my_array(i)%vy = my_array(i)%vy * 2
            my_array(i)%vz = my_array(i)%vz * 2
            my_array(i)%t = my_array(i)%t * 2
        END DO

        !out_list = my_array
        deallocate(my_array)
    end subroutine lists_change
end module state_vector_mod

But when I call the lists_change function, an error appears "OSError: exception: access violation reading 0x0000000000000005"

Please tell me how you can fix it and whether the code is written correctly in general?

The address in the error message is peculiar. It points to an integer value of 5 being used as an address value. The variable size has this value. We can conclude that the argument length

    integer(c_int), intent(in) :: length

should actually have the value attribute

    integer(c_int), value :: length

or that it should be called from Python by reference (ct.byref).

Regarding general points, francescalus is right that the in_list array pointer is modified and should likely be passed as value so you can get rid of the ct.byref or with intent(in). Or you can also just pass the array itself by reference and avoid casting de-referencing the pointer.

    type(state_vector), intent(inout) :: my_array(length)

or with some obvious modifications if you introduce a separate out_list argument that is now commented out.