[netCDF #NJG-578265]: Netcdf4 and multi-unlimited dimension

Ludovic,

I've simplified your bug demonstration a little bit, and the program
behaves differently on two different platforms.  On a Linux system
using GNU compilers it runs to completion, but calling nc_inq_dimlen on
beam dimension gives a much larger dimension length than it should just
before closing the file.

On a Sun/i386 platform with vendor C compiler, it gets an error in the
HDF5 layer on a call to H5Dwrite.  Ed will look at the problem next week.

In the meantime, I think I can suggest a workaround or maybe a better
approach.  Your demonstration code is writing a "ragged array" that has
an unlimited number of rows and different numbers of elements in each row.
Another way to do this is to use the variable-length array feature of
netCDF-4 for each row, so you would have an array of variable-length
arrays.  There is a test/demonstration program in ncdump/tst_vlen_data.c
that shows how to do this.  It writes a file that has this structure:

  netcdf tst_vlen_data {
  types:
    float(*) row_of_floats ;
  dimensions:
        m = 5 ;
  variables:
          row_of_floats ragged_array(m) ;
                row_of_floats ragged_array:_FillValue = {-999} ;
  data:

   ragged_array = {10, 11, 12, 13, 14}, {20, 21, 22, 23}, {30, 31, 32},
      {40, 41}, _ ;
  }

If a ragged array is what you really need for your application, I think
this sort of structure may offer a better way to implement it.

--Russ

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



Ticket Details
===================
Ticket ID: NJG-578265
Department: Support netCDF
Priority: Normal
Status: Closed