Fortran optional arguments in Python using ctypes

Question

How should I properly deal with optional Fortran arguments in Python using CTypes? For example let's assume the Fortran subroutine looks something like:

MODULE test_module
INCLUDES
   SUBROUTINE fstr_test(iunit, istat, file1, file2, file3, test, file4, file5)
      IMPLICIT NONE
      INTEGER, INTENT(inout) :: iunit
      INTEGER, INTENT(out) :: istat
      CHARACTER(LEN=*), INTENT(in) :: file1, file2, file3
      CHARACTER(LEN=*), INTENT(in), OPTIONAL :: file4, file5
      INTEGER, OPTIONAL :: test
      WRITE(6,*) file1, file2, file3
      RETURN
   END SUBROUTINE fstr_test
END MODULE test_module

Now the python should look something like

import ctypes as ct
libtest = ct.cdll.LoadLibrary("libtest.so")
fstr_h = getattr(libtest,'__test_module_MOD_fstr_test')
fstr_h.argparse =     [ ct.c_int, \ #iunit
                        ct.c_int, \ #istat
                        ct.c_char_p, \ #file1
                        ct.c_char_p, \ #file2
                        ct.c_char_p, \ #file3
                        ct.c_int, \ #test
                        ct.c_char_p,\ #file4
                        ct.c_char_p, \ #file5
    ct.c_bool, ct.c_bool, ct.c_bool,\ # Three optional arguments
    ct.c_long, ct.c_long, ct.c_long, ct.c_long, ct.c_long] # lengths
fstr_h.restype=None
iunit_temp = ct.c_int(0)
istat_temp = ct.c_int(0)
record_in_temp = ct.c_int(0)
opt1 = ct.c_bool(True)
opt2 = ct.c_bool(True)
opt3 = ct.c_bool(True)
file1 = "one"
file2 = "two"
file3 = "three"
file4 = "four"
file5 = "five"
fstr_h(iunit_temp,ct.byref(istat_temp), \
    file1.encode('UTF-8'),file2.encode('UTF-8'),file3.encode('UTF-8'),\
    record_in_temp,file4.encode('UTF-8'),file5.encode('UTF-8'), \
    ct.byref(opt1), ct.byref(opt2), ct.byref(opt3),\
    len(file1),len(file2),len(file3),len(file4),len(file5))

From gcc's documentation on its argument passing conventions I believe that I'm passing the variables correctly but there may be a subtlety I'm missing? When I try to run the code it crashes.

Can you match your (Python-side) arguments to what you understand gcc (through the argument passing convention) is expecting? That would help me (who doesn't do python/ctypes) assess your belief. — francescalus
– francescalus, Commented Feb 24, 2018 at 9:46
What is argparse? Try setting the argtypes attribute instead. — 101
– 101, Commented Feb 24, 2018 at 10:57
If passing by reference, you want the argtype to be ct.POINTER(ct.c_int) to pass ct.byref(iunit_temp), for example. You can also use file1 = b'one' and skip encoding the files explicitly. — Mark Tolonen
– Mark Tolonen, Commented Feb 25, 2018 at 4:16
I switched to argtypes this doesn't solve my problem but possibly fixes some other code I had which was using argparse (see solution to this question stackoverflow.com/questions/42328866/…) — Lazer
– Lazer, Commented Feb 25, 2018 at 9:38

Lazer · Accepted Answer · 2018-02-25 10:24:17Z

2

OK figured a few things out.

I needed to be using argtypes not argparse
OPTIONAL arguments only need the hidden logical variable when they're defined as passed by VALUE in Fortran. If they're not passed then a NULL pointer would still need to be passed.
The integers all need to be passed in by reference.

So the Fortran code remains as above but the Python code should read like

import ctypes as ct
libtest = ct.cdll.LoadLibrary("libtest.so")
fstr_h = getattr(libtest,'__test_module_MOD_fstr_test')
fstr_h.argtypes =     [ ct.POINTER(ct.c_int), \ #iunit
                        ct.POINTER(ct.c_int), \ #istat
                        ct.c_char_p, \ #file1
                        ct.c_char_p, \ #file2
                        ct.c_char_p, \ #file3
                        ct.POINTER(ct.c_int), \ #test
                        ct.c_char_p,\ #file4
                        ct.c_char_p, \ #file5
                        ct.c_long, ct.c_long, ct.c_long, ct.c_long, ct.c_long] # lengths
fstr_h.restype=None
iunit_temp = ct.c_int(0)
istat_temp = ct.c_int(0)
record_in_temp = ct.c_int(0)
file1 = "one"
file2 = "two"
file3 = "three"
file4 = "four"
file5 = "five"
fstr_h(ct.byref(iunit_temp),ct.byref(istat_temp), \
       file1.encode('UTF-8'),file2.encode('UTF-8'),file3.encode('UTF-8'),\
       ct.byref(record_in_temp),file4.encode('UTF-8'),file5.encode('UTF-8'), \
       len(file1),len(file2),len(file3),len(file4),len(file5))

Now this is conjecture but I suspect any variable typed as INTNET(in) in Fortran can be passed as a value and not a pointer. However, since INTENT(inout) and INTENT(out) variables have their value changed, I believe the need to be passed by pointer. The same can be said of just regularly typed arguments in Fortran.

answered Feb 25, 2018 at 10:24

Lazer

3894 silver badges16 bronze badges

Sign up to request clarification or add additional context in comments.

1 Comment

Vladimir F Героям слава Over a year ago

It is much MUCH better the official Fortran to C interoperability. See fortran-iso-c-binding. Only if you use bind(C) you can count with the fact that optional arguments use NULL pointers. Without bind(C) the compiler can do whatever it wants. It can alse change the weird mangeled name __test_module_MOD_fstr_test to whatever it wants, it may differ between compiler versions and it will differ between different compilers. You can read about passing-by-reference and similar there as well.

Collectives™ on Stack Overflow

Fortran optional arguments in Python using ctypes

1 Answer 1

1 Comment

Your Answer

Linked

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

1 Comment

Your Answer

Sign up or log in

Post as a guest

Linked

Related