[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [h5md-user] Multiple H5MD root groups per file

From: Konrad Hinsen
Subject: Re: [h5md-user] Multiple H5MD root groups per file
Date: Fri, 20 Sep 2013 13:14:59 +0200

Pierre de Buyl writes:

 > I propose one of these updates:
 > 1. Mandate a list of H5MD paths somewhere at the root of the file (for
 > autodiscovery of H5MD containers).

If we want to be able to use H5MD inside a collection of data managed
by some other specification, then we cannot mandate anything to be at
the root level. Any prescription about the root level more or less
implies "we own the whole file".

It's actually not so difficult to let programs search for H5MD data
inside a file. The real difficulty is what to do when there are
multiple H5MD trajectories but the operation of the program requires a
single one.

 > 2. Allow the extension but mandate that H5MD data be communicated as
 > file.h5/path/to/some/h5md_root_group.

No problem with that but it's a bit outside of the specification of the
file format.

BTW, I'd use some other separator between filename and path, otherwise
it's difficult for Unix software to identify the filename. Of course anything
else could conflict with some other platform's path separator.

 > The hierarchical nature of HDF5 make the idea very natural to allow for H5MD
 > "file within a file", indeed. I am not 100% that we should do it if we have 
 > not
 > a very (very) straightforward way to open a file and find the H5MD root 
 > groups.
 > We need to consider also the impact on the implementations.

Indeed. For Python this is pretty simple: just use file.root.visit and grab
those groups that contain a group called "h5md". It's probably a lot less
straightforward in C or Fortran.

 > > plans. In fact, I would probably have done this sooner or later even if
 > > the H5MD specification did not allow it :-)
 > This indeed a good motivation! But it is a bit more obvious in your 
 > situation as
 > one would have a link to the trajectories from the administrative part.

Yes, but a program that knows about H5MD but not about ActivePapers
would still have to locate the trajectories on its own.

Konrad Hinsen
Centre de Biophysique Moléculaire, CNRS Orléans
Synchrotron Soleil - Division Expériences
Saint Aubin - BP 48
91192 Gif sur Yvette Cedex, France
Tel. +33-1 69 35 97 15
E-Mail: research AT khinsen DOT fastmail DOT net
ORCID: http://orcid.org/0000-0003-0330-9428
Twitter: @khinsen

reply via email to

[Prev in Thread] Current Thread [Next in Thread]