[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[netCDF #VCD-725374]: Trouble inheriting variable ID into group

Hi Sourish,

> I found a problem with netcdf 4.1.3 while creating a group variable
> where some of the dimensions should come from the parent group.
> Basically, while trying to create a file with the structure
> 
> netcdf wrong_file {
> dimensions:
> dim1 = 2 ;
> dim2 = 1 ;
> 
> group: grp1 {
> dimensions:
> dim3 = 45 ;
> dim4 = 60 ;
> variables:
> double var1(dim1, dim2, dim3, dim4) ;
> double var2(dim1, dim2, dim3, dim4) ;
> } // group grp1
> }
> 
> the variables var1 and var2 get wrong dimensions IDs. Please find
> attached a fortran 90 program to reproduce the problem. On my system,
> with the IBM xlf90 compiler, I do the following:
> 
> 1. xlf90_r -o netcdf_4.1.3_error -I$NETCDF_DIR/include
> netcdf_4.1.3_error.F90 -L$NETCDF_DIR/lib -lnetcdff -lnetcdf
> 
> 2. ./netcdf_4.1.3_error
> 
> 3. ncdump -h wrong_file.nc4
> 
> I see the following structure, which is absolutely not correct:
> 
> netcdf wrong_file {
> dimensions:
> categories = 2 ;
> months = 1 ;
> 
> group: glb600x400 {
> dimensions:
> latitude = 45 ;
> longitude = 60 ;
> variables:
> double poste_emission(longitude, latitude, longitude, latitude) ;
> double prior_emission(longitude, latitude, longitude, latitude) ;
> } // group glb600x400
> }
> 
> Could you please look into this?

Yes, it appears it is currently necessary to reorder the calls or call 

  nf90_enddef(nc_id) 

after the definitions of dimensions and variables, but before
writing data to variables.  This is similar to a documented
requirement for the netCDF-3 API, in which definitions must be invoked
in "define mode" before data is accessed in "data mode" with calls to
nf90_endef and nf90_redef separating transitions from define mode to
data mode.

In your example, if you replace

  ! define variable
  io_status = nf90_def_var(grp_id, 'poste_emission', NF90_DOUBLE, dim_id, 
var_id(1))
  ! fill variable
  io_status = nf90_put_var(grp_id, var_id(1), x)

  ! define variable
  io_status = nf90_def_var(grp_id, 'prior_emission', NF90_DOUBLE, dim_id, 
var_id(2))
  ! fill variable
  io_status = nf90_put_var(grp_id, var_id(2), x)

with

  ! define variable
  nf90_def_var(grp_id, 'poste_emission', NF90_DOUBLE, dim_id, var_id(1))

  ! define variable
  nf90_def_var(grp_id, 'prior_emission', NF90_DOUBLE, dim_id, var_id(2))

  ! leave define mode
  nf90_enddef(nc_id)

  ! fill variable
  nf90_put_var(grp_id, var_id(1), x)

  ! fill variable
  nf90_put_var(grp_id, var_id(2), x)

then I believe the resulting file will be OK, which you can check with
ncdump.  The file also appears OK if you leave out the call to
nf90_enddef but reorder the calls so that all the nf90_def calls
appear before all the nf90_put_var calls.  This is consistent with the
documentation, that says the call to nf90_enddef is not necessary for
netCDF-4 calls, as it is made automatically by the library.

But you have identified a bug, that no error status is returned if
you make the calls out of the order required by the library.  Another
bug is that the library doesn't say the dimension defines have to be
made prior to the put_var calls.

I have created a Jira issue for this bug, so that we will fix it in a
future release.  It's really a bug in the C library, but it shows in
other APIs built on top of the C library, like the f90 and f77 APIs.
If you want to track progress on the bug or refer to it in the future,
you can follow it here:

  https://www.unidata.ucar.edu/jira/browse/NCF-134

Thanks for reporting the problem and creating the example that
demonstrates it!

--Russ

Russ Rew                                         UCAR Unidata Program
address@hidden                      http://www.unidata.ucar.edu



Ticket Details
===================
Ticket ID: VCD-725374
Department: Support netCDF
Priority: Normal
Status: Closed