Advanced F2PY use cases#

Adding user-defined functions to F2PY generated modules#

User-defined Python C/API functions can be defined inside signature files using usercode and pymethoddef statements (they must be used inside the python module block). For example, the following signature file spam.pyf

!    -*- f90 -*-
python module spam
    usercode '''
  static char doc_spam_system[] = "Execute a shell command.";
  static PyObject *spam_system(PyObject *self, PyObject *args)
  {
    char *command;
    int sts;

    if (!PyArg_ParseTuple(args, "s", &command))
        return NULL;
    sts = system(command);
    return Py_BuildValue("i", sts);
  }
    '''
    pymethoddef '''
    {"system",  spam_system, METH_VARARGS, doc_spam_system},
    '''
end python module spam

wraps the C library function system():

f2py -c spam.pyf

In Python this can then be used as:

>>> import spam
>>> status = spam.system('whoami')
pearu
>>> status = spam.system('blah')
sh: line 1: blah: command not found

Adding user-defined variables#

The following example illustrates how to add user-defined variables to a F2PY generated extension module by modifying the dictionary of a F2PY generated module. Consider the following signature file (compiled with f2py -c var.pyf):

!    -*- f90 -*-
python module var
  usercode '''
    int BAR = 5;
  '''
  interface
    usercode '''
      PyDict_SetItemString(d,"BAR",PyInt_FromLong(BAR));
    '''
  end interface
end python module

Notice that the second usercode statement must be defined inside an interface block and the module dictionary is available through the variable d (see varmodule.c generated by f2py var.pyf for additional details).

Usage in Python:

>>> import var
>>> var.BAR
5

Dealing with KIND specifiers#

Currently, F2PY can handle only <type spec>(kind=<kindselector>) declarations where <kindselector> is a numeric integer (e.g. 1, 2, 4,…), but not a function call KIND(..) or any other expression. F2PY needs to know what would be the corresponding C type and a general solution for that would be too complicated to implement.

However, F2PY provides a hook to overcome this difficulty, namely, users can define their own <Fortran type> to <C type> maps. For example, if Fortran 90 code contains:

REAL(kind=KIND(0.0D0)) ...

then create a mapping file containing a Python dictionary:

{'real': {'KIND(0.0D0)': 'double'}}

for instance.

Use the --f2cmap command-line option to pass the file name to F2PY. By default, F2PY assumes file name is .f2py_f2cmap in the current working directory.

More generally, the f2cmap file must contain a dictionary with items:

<Fortran typespec> : {<selector_expr>:<C type>}

that defines mapping between Fortran type:

<Fortran typespec>([kind=]<selector_expr>)

and the corresponding <C type>. The <C type> can be one of the following:

double
float
long_double
char
signed_char
unsigned_char
short
unsigned_short
int
long
long_long
unsigned
complex_float
complex_double
complex_long_double
string

For example, for a Fortran file func1.f containing:

      subroutine func1(n, x, res)
        use, intrinsic :: iso_fortran_env, only: int64, real64
        implicit none
        integer(int64), intent(in) :: n
        real(real64), intent(in) :: x(n)
        real(real64), intent(out) :: res
Cf2py   intent(hide) :: n
        res = sum(x)
      end

In order to convert int64 and real64 to valid C data types, a .f2py_f2cmap file with the following content can be created in the current directory:

dict(real=dict(real64='double'), integer=dict(int64='long long'))

and create the module as usual. F2PY checks if a .f2py_f2cmap file is present in the current directory and will use it to map KIND specifiers to C data types.

f2py -c func1.f -m func1

Alternatively, the mapping file can be saved with any other name, for example mapfile.txt, and this information can be passed to F2PY by using the --f2cmap option.

f2py -c func1.f -m func1 --f2cmap mapfile.txt

For more information, see F2Py source code numpy/f2py/capi_maps.py.

Character strings#

Assumed length character strings#

In Fortran, assumed length character string arguments are declared as character*(*) or character(len=*), that is, the length of such arguments are determined by the actual string arguments at runtime. For intent(in) arguments, this lack of length information poses no problems for f2py to construct functional wrapper functions. However, for intent(out) arguments, the lack of length information is problematic for f2py generated wrappers because there is no size information available for creating memory buffers for such arguments and F2PY assumes the length is 0. Depending on how the length of assumed length character strings are specified, there exist ways to workaround this problem, as exemplified below.

If the length of the character*(*) output argument is determined by the state of other input arguments, the required connection can be established in a signature file or within a f2py-comment by adding an extra declaration for the corresponding argument that specifies the length in character selector part. For example, consider a Fortran file asterisk1.f90:

subroutine foo1(s)
  character*(*), intent(out) :: s
  !f2py character(f2py_len=12) s
  s = "123456789A12"
end subroutine foo1

Compile it with f2py -c asterisk1.f90 -m asterisk1 and then in Python:

>>> import asterisk1
>>> asterisk1.foo1()
b'123456789A12'

Notice that the extra declaration character(f2py_len=12) s is interpreted only by f2py and in the f2py_len= specification one can use C-expressions as a length value.

In the following example:

subroutine foo2(s, n)
  character(len=*), intent(out) :: s
  integer, intent(in) :: n
  !f2py character(f2py_len=n), depend(n) :: s
  s = "123456789A123456789B"(1:n)
end subroutine foo2

the length of the output assumed length string depends on an input argument n, after wrapping with F2PY, in Python:

>>> import asterisk
>>> asterisk.foo2(2)
b'12'
>>> asterisk.foo2(12)
b'123456789A12'
>>>