HDF5 mounting files (with H5Fmount)

Question

I try to mount different files that all have the same structure to a single new one (in C++). Therefore I create a new File "new.h5" and in there a new Group "/G". The existing two files ("file1.h5" and "file2.h5") have some Attributes in the root-Group "/" and then a path "dataset1/data1/data", where "dataset1/data1" contains some Attributes and the "data"-Path with numerical Data in it.

 /dataset1
 /attribute1
 /attribute2
  /data1
   /data
   /attribute1
   /attribute2

In the new file I want a structure:

/G
 /dataset1
 /attribute1
 /attribute2
  /data1
   /data
   /attribute1
   /attribute2
 /dataset2
 /attribute1
 /attribute2
  /data1
   /data
   /attribute1
   /attribute2

I tried it this way:

#include "hdf5.h"
#ifdef OLD_HEADER_FILENAME
#include <iostream.h>
#else
#include <iostream>
#endif
#include <string>
#ifndef H5_NO_NAMESPACE
#ifndef H5_NO_STD
using std::cout;
using std::endl;
#endif // H5_NO_STD
#endif
#include "H5Cpp.h"
#ifndef H5_NO_NAMESPACE
using namespace H5;
#endif

#define FILE_NAME_1 "file1.h5"
#define FILE_NAME_2 "file2.h5"
#define FILE_NAME_ALL "new.h5"

int main(void)
{

   hid_t fid1, fid2, fid3, gid;  /* Files and group identifiers */

   fid1 = H5Fopen(FILE_NAME_1, H5F_ACC_RDWR, H5P_DEFAULT);
   fid2 = H5Fopen(FILE_NAME_2, H5F_ACC_RDWR, H5P_DEFAULT);
   fid3 = H5Fcreate(FILE_NAME_ALL, H5F_ACC_TRUNC, H5P_DEFAULT, H5P_DEFAULT);

   gid = H5Gcreate2(fid3, "/G", H5P_DEFAULT, H5P_DEFAULT, H5P_DEFAULT);
   H5Gclose(gid);

   /*
    * Mount second file under G in the first file.
    */
   H5Fmount(fid3, "/G", fid1, H5P_DEFAULT);
   H5Fmount(fid3, "/G", fid2, H5P_DEFAULT);

   H5Fclose(fid1);
   H5Fclose(fid2);
   H5Fclose(fid3);

   return 0;
}

When I take a look to new.h5 with h5dump I get:

h5dump new.h5
HDF5 "new.h5" {
GROUP "/" {
   GROUP "G" {
   }
}
}

so I created the Group G but there are no files in it?! What went wrong?

And is there a possibility to rename the "dataset1" from the second file to "dataset2" while mounting the files or do I have to do it before?

Thank you!

EDIT:

I might have found a problem: I want to mount files from the root-group. What I mean: I want to have a common file that contains everything from the other two files. That makes it difficult to find a mounting point. It can't be the root-directory but there are no groups called "/G" in the existing files. How can I solve this?

score 2 · Accepted Answer · answered Apr 02 '14 at 15:51

This is the correct behavior. Using H5Fmount only mounts the file at that location in memory. So when you close the files and exit your program the newfile (new.h5) does not (and will not) show the contents of the mounted files.

If you want this kind of behavior, use H5Lcreate_external to create external links.

I've changed your code to be C (not C++ - sorry, but the idea is the same):

#include <stdlib.h>
#include <stdio.h>
#include <hdf5.h>

#define FILE_NAME_1 "file1.h5"
#define FILE_NAME_2 "file2.h5"
#define FILE_NAME_ALL "new.h5"

int main(void)
{

   hid_t fid1;  /* Files identifier */

   fid1 = H5Fcreate(FILE_NAME_ALL, H5F_ACC_TRUNC, H5P_DEFAULT, H5P_DEFAULT);

   /*
    * Link "file1.h5" to group F1
    */
   H5Lcreate_external(FILE_NAME_1, "/", fid1, "/F1", H5P_DEFAULT, H5P_DEFAULT);

   /*
    * Link "file2.h5" to group F2
    */
   H5Lcreate_external(FILE_NAME_2, "/", fid1, "/F2", H5P_DEFAULT, H5P_DEFAULT);

   H5Fclose(fid1);

   return 0;
}

Where the contents of file1.h5 and file2.h5 are the same for the test case:

h5dump file1.h5
HDF5 "file1.h5" {
GROUP "/" {
   DATASET "D" {
      DATATYPE  H5T_STD_I32LE
      DATASPACE  SIMPLE { ( 4, 5 ) / ( 4, 5 ) }
      DATA {
      (0,0): 0, 1, 2, 3, 4,
      (1,0): 1, 2, 3, 4, 5,
      (2,0): 2, 3, 4, 5, 6,
      (3,0): 3, 4, 5, 6, 7
      }
   }
}
}

Then if we compile and run the test program:

h5cc -o test test.c && ./test

Finally dump the new file:

h5dump new.h5
HDF5 "new.h5" {
GROUP "/" {
   EXTERNAL_LINK "F1" {
      TARGETFILE "file1.h5"
      TARGETPATH "/"
         GROUP "/" {
            DATASET "D" {
               DATATYPE  H5T_STD_I32LE
               DATASPACE  SIMPLE { ( 4, 5 ) / ( 4, 5 ) }
               DATA {
               (0,0): 0, 1, 2, 3, 4,
               (1,0): 1, 2, 3, 4, 5,
               (2,0): 2, 3, 4, 5, 6,
               (3,0): 3, 4, 5, 6, 7
               }
            }
         }
   }
   EXTERNAL_LINK "F2" {
      TARGETFILE "file2.h5"
      TARGETPATH "/"
         GROUP "/" {
            DATASET "D" {
               DATATYPE  H5T_STD_I32LE
               DATASPACE  SIMPLE { ( 4, 5 ) / ( 4, 5 ) }
               DATA {
               (0,0): 0, 1, 2, 3, 4,
               (1,0): 1, 2, 3, 4, 5,
               (2,0): 2, 3, 4, 5, 6,
               (3,0): 3, 4, 5, 6, 7
               }
            }
         }
   }
}
}

That worked, thank you! You said you used C not C++, this is a stupid question, sorry for this, but how do I know this? I tried to figure out the difference between HDF5 API (http://ftp.hdfgroup.org/HDF5/doc/cpplus_RM/classH5_1_1AbstractDs.html) and the HDF5 C Library (Tutorial here: http://www.hdfgroup.org/HDF5/doc/RM/RM_H5Front.html) but I don't get it. When I use a .cpp file, can I mix those types? — smaica, Apr 03 '14 at 08:40
@smaica Yes, you can mix the C++ and C api. The C++ objects are just thin wrappers around the C API and will provide a method to get their ID (e.g. H5::Group::getLocId) that can be passed to the C API. In many cases, there is no C++ class that provides the same functionality as the C API. I don't believe that the H5Lcreate_external C function has a C++ counterpart. Also, there is not nearly as much documentation for the C++ API as the C API. — Nathan, Apr 06 '15 at 21:08

HDF5 mounting files (with H5Fmount)

1 Answers1