Question

so I am getting a segfault from ctypes on a 32 bit linux machine that I can’t reproduce on 64 bit darwin or linux.

Here is the C:

typedef struct {
  void *ptr;
  } doodle;

doodle C_intpointerfunction(int *a)
{
  *a = 41;
  doodle foo;
  foo.ptr = a;
  return foo;
}

which is compiled with:

gcc -c intpointerlibrary.c
gcc -shared intpointerlibrary.o -o libintpointerlib.so

and here is the Python:

import numpy as N
from ctypes import *

_libintpointer = N.ctypeslib.load_library('libintpointerlib.so','.')

_libintpointer.C_intpointerfunction.restype = c_void_p
_libintpointer.C_intpointerfunction.argtypes = [POINTER(c_int)]
def P_intpointerfunction():
  lrc = c_int(0)
  print "lrc before (should be 0) = "+str(lrc.value)
  return_val = _libintpointer.C_intpointerfunction(byref(lrc))
  print "lrc after (should be 41) = "+str(lrc.value)
  return return_val

so now, when I call this function:

  rc = P_intpointerfunction()

I get a segfault. I’ve tried creating a Python class to wrap the returned struct that is created on the C side with the same results. If passing back a native ctypes type (like c_int) everything works fine.. Is this truly just a problem with 32 bit Linux, or is there something I have not considered? Thanks!

Answer 1

Not sure why you’re setting restype = c_void_p;

class DOODLE(Structure):
    _fields_ = [('ptr', c_void_p)]
_libintpointer.C_intpointerfunction.restype = DOODLE

works as expected for me, on Linux x86-64.