Re: [f2py] f2py bug?

Pearu Peterson <[email protected]> Fri, 30 Jul 2010 06:46:52 +0300
Newsgroups gmane.comp.python.f2py.user
Message-ID <[email protected]>
On Thu, Jul 29, 2010 at 11:52 PM, Al Niessner <[email protected]> wrote:
> I do a 'f2py -c a.f90 b.f90 c.f90 x.f -m x' and it plows away for a
> while and then has the following error:
>
> gcc: /tmp/tmp4DtGEi/src.linux-x86_64-2.6/xmodule.c
> /tmp/tmp4DtGEi/src.linux-x86_64-2.6/xmodule.c: In function
> ‘long_double_from_pyobj’:
> /tmp/tmp4DtGEi/src.linux-x86_64-2.6/xmodule.c:335: error: expected ‘;’
> before ‘return’
> /tmp/tmp4DtGEi/src.linux-x86_64-2.6/xmodule.c: In function
> ‘long_double_from_pyobj’:
> /tmp/tmp4DtGEi/src.linux-x86_64-2.6/xmodule.c:335: error: expected ‘;’
> before ‘return’
>
> Here is the offending code:
>
> static int long_double_from_pyobj(long_double* v,PyObject *obj,const
> char *errmess) {
>  double d=0;
>  if (PyArray_CheckScalar(obj)){
>    if PyArray_IsScalar(obj, LongDouble) {
>      PyArray_ScalarAsCtype(obj, v);
>      return 1;
>    }
>    else if (PyArray_Check(obj) &&
> PyArray_TYPE(obj)==PyArray_LONGDOUBLE) {
>      (*v) = *((npy_longdouble *)PyArray_DATA(obj))
>      return 1;
>    }
>  }
>  if (double_from_pyobj(&d,obj,errmess)) {
>    *v = (long_double)d;
>    return 1;
>  }
>  return 0;
> }
>
> where line 335 is:
> (*v) = *((npy_longdouble *)PyArray_DATA(obj))
>
> I attach a ; at the end of that line and compile it by hand and it
> works. So, it seems like whatever generated this missed a semicolon. Is
> this known and is there is a fix? I looked through some of the archive
> but did not see anything nor did I find anything useful from the
> obligatory google search.
>
> While the problem is easy to understand, it is very unclear to me how to
> fix it or work around it. I can compile it by hand but I cannot get f2py
> to pick up where it left off. I have not looked at the code generator to
> fix it, but that does not sound easy either. Any other ideas?
>
> Thanks for any and all help in advance.

What numpy version (numpy.__version__) are you using? numpy in svn
does not have this problem, that is, the bug is fixed there.
If you cannot upgrade numpy at the moment, then edit the
numpy/f2py/cfuncs.py file by adding the semicolon to the corresponding
line, search for the definition of long_double_from_pyobj.

Pearu