[SciPy-Dev] f2py, the fortran integer type, and npy_intp

[Charles R Harris]

On Sat, Jul 10, 2010 at 12:30 PM, Kurt Smith <kwmsmith at gmail.com> wrote:

> On Sat, Jul 10, 2010 at 12:31 PM, Charles R Harris
> <charlesr.harris at gmail.com> wrote:
> > I note that some c programs try to call f2py generated fortran interfaces
> > using the npy_intp type for the array dimension. There are two problems
> > here, first the prototype for the generated wrapper uses int causing
> > compiler warnings on a 64 bit os, and second, the fortran subroutines
> > themselves use integer types. So some questions. What precision are
> fortran
> > integers? Should we be using npy_intp for array dimensions at all? Is
> there
> > any transparent way to have 64 bit support?
>
> I can comment on some of the above (not the f2py-specific bits, though).
>
> 1) "What precision are fortran integers?"
>
> According to a somewhat authoritative interlanguage programming site
> [0], a fortran 'integer' corresponds to a C 'int'.  This can be relied
> upon, but other fortran type declarations cannot (e.g. 'integer*8' !=
> a C 'long' on all platforms, with all compilers).  If sizeof(npy_intp)
> == sizeof(int), then everything will be fine, but it's a near
> certainty there are platforms where this doesn't hold, and may lead to
> problems if array sizes are large.
>
>
Looks like most FORTRANs allow the size of integer to be set by a
compilerflag<http://www.nccs.gov/computing-resources/ewok/compiling/fortran/>.
But hooking it up to work with f2py and C looks like a significant problem
involving the build infrastructure among other things. I think for the time
being we should just avoid using npy_intp as a type passed to FORTRAN
subroutines. This will limit the size of one dimensional arrays that can be
passed to the subroutines to 2GB but I don't see any easy way around it.

Chuck
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.python.org/pipermail/scipy-dev/attachments/20100710/f032beb9/attachment.html>