Hi!
I wanted to start using binary I/O for my model code, being written in Fortran
90 (95, 2003 in parts).
Now, having used NetCDF before (with C++, though writing my own set of wrapper
classes over the C api), I naturally wanted to use it here, too.
I first started with the include file that contains the f77 interface, then
switched to the netcdf module as that's what is actually described in the
NetCDF Fortran 90 guide (the nf90_ symbols).
It offers some nice convenience with overloading functions for the different
types, and, apart from my trouble realizing that writing text array attributes
won't fly (I faintly remember that I circumvented that one before, in C), it
works fine.
I had the test code runnning on a GNU/Linux system with the gfortran compiler
corresponding to the GCC that compiled the whole system (well, at least NetCDF
for sure), so it worked fine to use /usr/include/netcdf.mod .
But now I am on a different system where we work with intel fortran, or even
with gfortran, which seems to be a different gfortran than that that compiled
netcdf a long time ago.
Problem: The .mod file with the Fortran 90 interface is not portable across
compilers, architectures, or even different releases of the same compiler.
In a related way, this came up in fedora packaging, and subsequently on the
gfortran mailing list:
http://gcc.gnu.org/ml/fortran/2007-10/msg00304.html
So... my question is: What should I do to use the Fortran 90 interface of
NetCDF?
Seems like the only safe way is to include a copy of the source of that module
from NetCDF sources and compile that together with my other code. Is that The
Right Way(tm)?
Or, should one rather have the OS distribution install the fortran module
source in /usr/include (or some other place where you can pull in the source
from)?
What is the NetCDF folks' angle on this?
Alrighty then,
Thomas.
--
Dipl. Phys. Thomas Orgis
Atmospheric Modelling
Alfred-Wegener-Institute for Polar and Marine Research
