Re: [h5md-user] proposal 101 - update

[Felix Höfling]

Am 27.05.2015, 14:35 Uhr, schrieb Pierre de Buyl  
<address@hidden>:

> Hi all,
>
> Felix suggests in
> http://nongnu.org/h5md/proposals/0101_particles_and_tuples_lists.html
>
> """Alternatively, let `particles_group` point to the respective group in
> `/particles` using a hard/soft link, which ensures that this group  
> exists."""
>
> It cannot be a hard link because if the list is time-independent you  
> cannot have
> this hard link anywhere (hard links go into groups).
>
> However, if we have
>
> <list_name>: Integer[N]
>   +-- particles_group: HDF5 Reference
>
> (see http://www.hdfgroup.org/HDF5/doc/RM/RM_H5R.html)
>
> References can be stored as attributes. I vaguely remember that we  
> discussed the
> use of references in the past but cannot find this topic on the list.
>
> Has anyone a comment on the use of HDF5 references?
>
> Cheers,
>
> Pierre

Hi Pierre,

I see the problem with hard links. "Object references" seem to be a good  
alternative. They do not guarantee that the referenced object exists, but  
at least it must exist upon creation of the reference. I think this small  
change is a clear improvement over the original proposal, where the  
reference to the group in /particles occurred by means of a simple string.

Note that we should restrict to "Object references", in contrast to  
references to a Dataset region, see
https://www.hdfgroup.org/HDF5/doc/RM/RM_H5R.html#Reference-Create.

Best wishes,

Felix