Hello, I'm working under the able mentoring of Dag Sverre Seljebotn to implement a GSoC project informally known as 'f2cy'. From the 10,000 meter view, f2cy will (1) wrap fortran 77/90/95 code into a python module (reproducing f2py in this regard) with full support for assumed-shape fortran arrays (beyond f2py, if I'm not mistaken); and (2) will wrap the same fortran code in *cython* cdef functions that can be manipulated at the cython level with no interaction with the python API -- yielding close to the metal speed, even for small functions.
We're making use of the new ISO_C_BINDING intrinsic module implemented in most every extant Fortran 95 compiler (including gfortran) to make the resulting code very portable. This introduces a small wrinkle to the compilation of the python/cython module, and that's the content of my question. Basically, f2cy generates a 'genconfig.f95' fortran source file that, when compiled and run, creates another fortran source file 'config.f95' and a C header file 'config.h'. config.f95 defines a module that contains the necessary mappings between the ISO_C_BINDING kind-type-parameters (ktps) and the ktps used in the rest of the generated source code. The genconfig.f95 file needs to be compiled and run before the python module can be compiled, since the python module depends on the 'config.f95' file. Once config.f95 exists, it is easy to make the python module. The trick is in generating the 'config.f95' file in a portable way, using the right fortran compiler with all the right flags. We plan on leveraging the impressive body of work in the numpy.distutils.fcompiler package for this purpose; I don't think it would be too hard to compile the 'genconfig.f95' into an object file, but we then need to compile it to an executable. Is there a way for numpy.distutils to compile a fortran source file into an executable? I'm not well-versed in the black arts of distutils or numpy.distutils. I know the basics and am quickly reading up on it -- any pointers would be very appreciated. Thanks, Kurt Smith _______________________________________________ Numpy-discussion mailing list Numpy-discussion@scipy.org http://mail.scipy.org/mailman/listinfo/numpy-discussion
