Am Tue, 02 Feb 2010 08:25:31 -0600
schrieb Ted Mansell <Ted.Mansell@xxxxxxxx>:
> According to my config.log, I used this config:
>
> ./configure --prefix=/storage/users/local/netcdf4sun
> --enable-netcdf-4 --with-hdf5=/storage/users/local/hdf5sun F77=f77
> FC=f90 --disable-cxx CPPFLAGS=-DpgiFortran
>
> This was with 4.1-beta2 on a Linux x86_64 Opteron system and Sun
> Studio 12.1. With the most recent versions of netcdf4 I have had to
> turn off opendap (--disable-dap) on other systems because I was
> getting some errors.
OK, I finally got it: I encountered a separate issue... because I had not the
path of sun's rtlibs in LD_LIBRARY_PATH, the conftest program for determining
NF_INT_IS_C_INT / NF_INT_IS_C_LONG failed to run. In the end, the following
error message pointed into the right direction, but it might be a good idea to
make the configure script fail in this case ... the purpose of that test is to
find out which of these two symbols shall be definded. If none of them applies,
we got an ill-configured build.
I managed to get NetCDF compiled now, using -DpgiFortran (also a second time
with -DsunFortran, which seems more appropriate), make check also is positive.
Oh, and I should note that it is indeed mandatory to omit the C++ interface
because of conflicts of system headers with Sun Studio 12.1 ... see
http://forums.sun.com/thread.jspa?threadID=5393796 . Good that I diverted from
the C++ NetCDF API in the old days and made my own C++ wrapper around the C
interface for my C++ apps:-/
Thanks for poking me to a successful build... now I can try to debug why Sun
Studio is so horrific with my OpenMP code...
Alrighty then,
Thomas.
--
Dipl. Phys. Thomas Orgis
Atmospheric Modelling
Alfred-Wegener-Institute for Polar and Marine Research
