Is there a way to identify group names in an H5 file programmatically without using h5info?

I am working with large (0.5-2 GB) and complex h5 data files and trying to identify the high level group names in the files. The names of the groups change for each file, so I need to be able to programmatically identify them. Below the high level of groups, the file structure is consistent, so I can efficiently use h5read once I have these 5-10 group names. Using hdf5info works, but is very slow because it is scanning the entire file and giving me much more metadata than I really care about (each high level group has thousands of nested lower level groups/datasets/attributes). The MATLAB recommended h5info is much slower for some reason. In fact, I have never actually let it run to completion, usually giving up after half an hour.

I have also tried setting the "ReadAttributes" bool to FALSE which for some reason made hdf5info take even longer to run. Is there a more efficient way to identify only the top level of group names in the h5 file?

Thanks,

Answer 1

[Jacob]

I finally figured it out by using the low level H5 functions built into MATLAB (H5G.get_objname_by_idx). This is exponentially faster than running the full hdf5info.

fid = H5F.open('test.H5');
idx = 0;
while true
    this_name = H5G.get_objname_by_idx(fid,idx);
    if isempty(this_name)
       break 
    end
    group_names{idx+1,1} = this_name;
    idx = idx+1;
end
num_grps = length(group_names);
H5F.close(fid);

[John Wolter]

Note that Jacob's routine does not test the object names found to see if they are actually groups. I wrote a version of this to traverse the full groups tree and discovered that H5G.get_objname_by_idx found groups, datasets, and datatypes in my file. There were no links in my file, so I don't know if it would find those as well.

I used the routine below to distingush between datasets and groups, but I wouldn't be surprised if there is a more elegant solution.

try
    did = H5D.open(fid, full_name);
    % dataset open succeeded; do action appropriate for a dataset
    ...
catch
    try
        gid = H5G.open(fid, full_name);
        % group open succeeded; do action appropriate for a group
        ...
    catch
        ...
    end
end