Reading/writing HDF5 file in the same order it was created

Question

I'm using HDF5 file to do a sort of "serialization" of C++ class data. Is there a way to transverse an HDF5 file visiting each dataset in the same order they were created?

I chose HDF5 because I need to access the data from outside my program as well as to be able to resume my program state.

Thanks!

Answer 1

For me it is enough to be able to traverse the datasets in a group (and possibly its subgroups) in the same order they were created. To achieve this, (like embert said above) the groups within the file must be created using the H5P_CRT_ORDER_TRACKED | H5P_CRT_ORDER_INDEXED flags set. That is, call:

hid_t group_creation_plist;
group_creation_plist = H5Pcreate(H5P_GROUP_CREATE);
status = H5Pset_link_creation_order(group_creation_plist, 
                                 H5P_CRT_ORDER_TRACKED | H5P_CRT_ORDER_INDEXED);

And then create groups using:

hid_t group_id;
group_id = H5Gcreate(id_file, 
                     "/mydata", 
                     H5P_DEFAULT, 
                     group_creation_plist, 
                     H5P_DEFAULT);

Add all your data to your new group. To traverse the group in sorted creation order you must use the H5Literate function, making sure to set the index_type argument to H5_INDEX_CRT_ORDER.

    status = H5Literate (group_id, 
                         H5_INDEX_CRT_ORDER, // Note this argument
                         H5_ITER_INC, 
                         NULL, 
                         op_func,
                     (void *) &od);

It is important to start the iteration from the group, and not from root. I have not been able to make the root group maintain the order of creation of links. The solution to this is just to make a group at the root with the appropriate flags.

Using H5Literate is a bit tricky, specially if one wants to do recursive traversal of subgroups. A good example is found here and in the documentation.