Question

I am using HDF5 API and I am trying to create a dataset with variable-length string.
The struct is

struct dataX
{
  std::string data;
};

I was using char[256] with a static hard coded size.
But I want it to be dynamic so after reading the HDF5 Doc, I found H5T_VARIABLE and used it as follows but it still fails.

H5Dcreate returns a negative value (means error).

hid_t mem_type;
mem_type = H5Tcopy( H5T_C_S1 );
H5Tset_size(mem_type,H5T_VARIABLE);

/* Create the memory data type. */
if ((mem_type_id = H5Tcreate (H5T_COMPOUND, mem_type )) < 0 ) {
  return -1;
}
/* Insert fields. */
if ( H5Tinsert(mem_type_id, "field", 0, mem_type_id ) < 0 ) {
  return -1;
}

/* Create a simple data space with unlimited size */
//   hsize_t dims[1]={0};
//   hsize_t maxdimsk[1]={ H5S_UNLIMITED };
if ( (sid = H5Screate_simple( 1, dims, maxdims )) < 0 ){
  return -1;
}

/* Modify dataset creation properties, i.e. enable chunking  */
plist_id = H5Pcreate (H5P_DATASET_CREATE);
//chunk==1
if ( H5Pset_chunk ( plist_id, 1, chunk ) < 0 ){
  return -1;
}

H5Pset_alloc_time( plist_id, H5D_ALLOC_TIME_EARLY )
/* Set the fill value using a struct as the data type. */
// fill_data=0
if ( fill_data )
{
  if ( H5Pset_fill_value( plist_id, mem_type_id, fill_data ) < 0 ){
    LOG_ERROR << "cannot fill value " << LOG_ENDL;
    return -1;
  }
}
else {
  if ( H5Pset_fill_time( plist_id, H5D_FILL_TIME_NEVER ) < 0 ) {
    LOG_ERROR << "error" << LOG_ENDL;
  }
}

/* Create the dataset. */
did = H5Dcreate( loc_id, dset_name, mem_type_id, sid, plist_id )

I tried H5D_ALLOC_TIME_LATE, thinking that maybe if it allocated the memory just before writing it would work but … it didn’t.

Now I’m stuck and I don’t know what to do.

Did I miss something ?

Answer 1

Your mem_type_id doubly invalid:

• the second argument of H5Tcreate should be the size of the compound datatype
• in H5Tinsert, the last argument should be the datatype of the inserted field. Here I guess you meant mem_type instead of mem_type_id.

I don’t know anything about what you are doing in particular, but to write variable length string, you do not need to create a compound type nor to set any special property lists. Basically your 3 first line are enough to create a valid variable-length string datatype (mem_type). Then you create the simple dataspace, then the dataset.

Have a look at this example, you will see it’s pretty simple.