Hi Simon,
> we have quite big output files from a seismic wavefield simulation and
> want to fourier transform them for further use.
>
> The idea would be that the output file is opened, the wavefields are
> read and transformed and copied into a new file.
> However, there are a lot of smaller variables in the file, which should
> be copied to the new file as well. I do not want to declare temporary
> arrays for all of them and so I gave NF_COPY_VAR a try.
>
>
> The questions:
>
> 1. NetCDF-C has the routine nc_copy_var, which seems to be perfectly
> suited. Can you call it from a Fortran90 code?
The only portable way (that would work with different Fortran compilers) is if
you
use the Fortran-2003 C interoperability feature that provides a standard way to
call C functions from Fortran. Otherwise you have to either customize your
solution
to the Fortran compiler you have, adapting to it's non-standard way of
referring to
a C function and character string arguments, or learn to use a more general
solution
such as the cfortran.h macros that come with the netCDF Fortran software.
We are providing a beta-release of netcdf-fortran software that implements
use of the Fortran-2003 solution for calling C functions from Fortran, including
nf_copy_var, as well as a more recent daily snapshot of that software that has
additional bug fixes:
ftp://ftp.unidata.ucar.edu/pub/netcdf/netcdf-fortran-4.4-beta1.tar.gz
ftp://ftp.unidata.ucar.edu/pub/netcdf/snapshot/netcdf-fortran-4.4-daily.tar.gz
> 2. There is the undocumented NF_COPY_VAR. However, using this like
>
> status = nf_copy_var(ncin_grpid, varid, ncout_id)
> print *, nf_strerror(status), status
>
> Results in
>
> NetCDF: Invalid dimension ID or name -46
>
> even though nf90_get_var can read perfectly from this variable with the
> same group and variable id. Also, the error does not seem to make much
> sense, since neither dimensions or names are even involved.
Note that nc_copy_var (and nf_copy_var that calls it) both depend on the
dimensions that the variable uses having already been defined in the
output file. If you try to copy a variable with a dimension not yet
defined, you will get that error.
That assumption seems to have been omitted from the documentation. We'll
fix that.
> 3. Is there any reliable documentation for NF_COPY_VAR or a Fortran90
> version of it?
No, the Fortran documentation doesn't appear to include that function. We'll
fix that too.
> Since the code is to be distributed, we would not like it to rely on NCO
> or similar external routines and do everything with the normal Fortran
> interface.
>
> Any ideas?
Keep in mind that using nc_copy_var or nf_copy_var can be very inefficient if
the target is a large netCDF-3 (classic or 64-bit offset format) file, as
described by this paragraph from the documentation:
For large netCDF-3 files, this can be a very inefficient way to copy
data from one file to another, because adding a new variable to the
target file may require more space in the header and thus result in
moving data for other variables in the target file. This is not a problem
for netCDF-4 files, which support efficient addition of variables
without moving data for other variables.
--Russ
Russ Rew UCAR Unidata Program
russ@xxxxxxxxxxxxxxxx http://www.unidata.ucar.edu
Ticket Details
===================
Ticket ID: AYG-518147
Department: Support netCDF
Priority: Normal
Status: Closed
