Jürgen,
Chuck Alexander and I provided Netcdf-3 compatible methods because you
asked for a "backward compatible" method in your original message.
However, it seems that you really want to use the variable length
array feature of Netcdf-4. I think you are on the right track.
I do not have the experience with VLEN types to identify your error.
However, there is a simple working example program that writes a
Netcdf-4 file containing a VLEN type, i.e. a true variable length
array. It is in the test suite in the Netcdf-4.2 source distribution,
probably in 4.2.1 as well. The program is src/ncdump/tst_vlen_data.c.
It may be helpful to compare this example with your program.
Since it is in the test suite, it uses some special include files.
The following compile command command gave me a working executable
with no problems and no source code tweaking:
gcc tst_vlen_data.c -I$HOME/netcdf/4.2/core/src/include \
-I$HOME/netcdf/4.2/core/src -lnetcdf
Replace $HOME/netcdf/4.2/core with the path to your own distribution
source, of course. I would expect that Netcdf's configure script must
have been run at least once, to get valid include files. HTH.
--Dave
On Tue, Jul 31, 2012 at 7:34 AM, Jürgen Lorenz Simon
<juergen_simon@xxxxxxx> wrote:
> Hi,
>
> I have revised the code a little and found the point of failure:
>
> NcDim *dim = file->add_dim("dim", (int)
> featureSpace->coordinate_system->size() );
> NcDim *N = file->add_dim( "N", clusters.size() );
>
> int var_id, dim_id;
> nc_inq_dimid( ncid, "dim", &dim_id );
> int dims[1] = {dim_id};
> nc_def_var( ncid, ss.str().c_str(), type_id, 1, dims, &var_id );
>
> ...
>
> size_t start=0, len=clusters[ci].points.size();
>
> status = nc_put_vara_double( nc_file_id, var_id, &start, &len,
> &data[0][0] );
>
> check_nc_error( "nc_put_vara_double", status );
>
>
> nc_put_vara_double returns status code -57, which according to the man page
> translates to:
>
> NC_EEDGE The specified edge lengths added to the specified corner would have
> referenced data out of range for the rank of the specified variable. For
> example, an edge length that is larger than the corresponding dimension
> length minus the corner index will cause an error.
>
> I revised the code again, making the dimension variable N 'unlimited' and
> the variables X1 ... X_n depending on N and d:
>
> NcDim *dim = file->add_dim("dim", (int)
> featureSpace->coordinate_system->size() );
> NcDim *N = file->add_dim( "N", NC_UNLIMITED );
>
> ....
>
> int var_id, dim_id, N_id;
> status = nc_inq_dimid( nc_file_id, "dim", &dim_id );
> check_nc_error( "nc_inq_dimid", status );
> status = nc_inq_dimid( nc_file_id, "N", &N_id );
> check_nc_error( "nc_inq_dimid", status );
> int dims[2] = {N_id,dim_id};
> status = nc_def_var( nc_file_id, ss.str().c_str(), type_id, 2,
> dims, &var_id );
>
> and now I get the following error:
>
> NC_EINVALCOORDS The specified corner indices were out of range for the rank
> of the specified variable. For example, a negative index, or an index that
> is larger than the corresponding dimension length will cause an error.
>
> Something I'm overlooking here! Anyone know what it is?
>
> Kind Regards,
> Jürgen
>
>
