Hi,
On Thu, Dec 09, 2010 at 05:06:13PM -0500, Collignon, Barbara C. wrote:
> I am having problems when doing serial HDF5 to read a 2D array of characters.
>
> The writing of the array in hdf5_file works fine using basic HDF5 functions :
> (...)
> dimsfx[0] = FILE_LEN ;
> dimsfx[1] = LINE_LEN ;
> sprintf (buf_nameX,"%s,FileName);
> dataset =
> H5LTmake_dataset(hdf5_file,buf_nameX,2,dimsfx,H5T_NATIVE_CHAR,buf_lig);
> (...)
>
> However I am having problems when reading this 2D-array from hdf5_file::
> (...)
> char **data;
> data = (char**)malloc((FILE_LEN)*sizeof(char*));
> for (int i = 0; i < FILE_LEN; i++) {
> data[i] = (char*)malloc(LINE_LEN*sizeof(char));
> }
This isn't a 2D array in the normal C/C++ sense - what you have
here is an array of FILE_LEN pointers to char *, and then each
element pointing to an array of FILE_LEN chars. That's quite a
different thing from a
char data[ FILE_LEN ][ FILE_LEN ];
While in the data structure you're using the 1D arrays of chars
can be allocated at more or less random places in memory, in the
second case (for the classic C 2D array) they are all are follo-
wing each other. And that's what the read function expects.
> if(status = H5LTread_dataset_char(hdf5_file,DATASETNAME,*data) == NULL) {
> for (i = 0; i < FILE_LEN; i++) {
> printf("%s\n",data[i]);
Probably the simplest fix is to change the way you allocate the
memory. If you change it to
char **data = malloc( FILE_LEN * sizeof *data );
char *tmp = malloc( FILE_LEN * FILE_LEN );
for ( int i = 0; i < FILE_LEN; i++ ) {
data[ i ] = tmp + i * FILE_LEN;
}
Calling malloc() only once makes sure that all the memory is
continuous. But, of course, you then have to deallocate only
data[0] (and, of course, data) once when you're done with the
array since all memory was allocated also in a single call of
malloc() instead of FILE_LEN calls.
Regards, Jens
--
\ Jens Thoms Toerring ________ [email protected]
\_______________________________ http://toerring.de
_______________________________________________
Hdf-forum is for HDF software users discussion.
[email protected]
http://mail.hdfgroup.org/mailman/listinfo/hdf-forum_hdfgroup.org
