Due to the current gap in continued funding from the U.S. National Science Foundation (NSF), the NSF Unidata Program Center has temporarily paused most operations. See NSF Unidata Pause in Most Operations for details.
I'm having problems in the creation of files for parallel IO: using the standard file creation: nf90_create(FILE_NAME, NF90_NETCDF4, ncid, comm = MPI_COMM_WORLD,info = MPI_INFO_NULL) I receive the following error: "Can't add HDF5 file metadata" However, an empty file gets created and when I run the program (the parallel test from the netcdf tutorials) a second time, everything works out. I've found the following information: http://www.unidata.ucar.edu/support/help/MailArchives/netcdf/msg08863.html where it is suggested, that instead of using NF90_NETCDF4 one should use NC_MPIIO or NC_MPIPOSIX, however, those are part of the c-interface and not of the nf90 interface and corresponding constants for the f90 interface don't exist?! Now, if I declare them myself locally as in: integer, parameter :: NF90_MPIIO=8192,NF90_MPIPOSIX=16384 correspnding to netcdf.h: #define NC_MPIIO 0x2000 #define NC_MPIPOSIX 0x4000 and use them instead of NF90_NETCDF4 the small test from the tutorial works neatly. So now I'm wondering: -why don't the constants corresponding to NC_MPIIO and NC_MPIPOSIX exist in the F90 interface? -why do the parallel examples in tutoial and distribution examples as of netcdf-4.1.1 use the NF90_NETCDF4 mode, if it doesn't work? I'm sure I'm missing something here or getting something wrong, so anybody that could shed some light on this, please answer... Thanks, Momme
netcdfgroup
archives: