[h5md-user] Modules

[Konrad Hinsen]

Peter Colberg writes:

 > On Wed, Sep 18, 2013 at 09:45:13AM +0200, Konrad Hinsen wrote:
 > > Leaving units to modules is a good idea. But for now, modules are just
 > > a proposal, the specification doesn't mention them.
 > 
 > Please write a paragraph or two to add them to the specification ;-).

OK, here's a proposition. I have added a version number to Olaf's
original specification. Everything is likely to change with time, so
version numbers have their use.

The storage of module information is of course debatable. I like
tables using compound data types, but others hate them. Personally I
don't really care, so if someone sees a good reason to store the
information otherwise, that's fine with me.

------------------------------------------------------------
Modules

The H5MD specification can be complemented by domain-specific modules,
which can add conditions that the data in the trajectory must satisfy,
and rules for the semantic interpretation of the data. A single
trajectory can use multiple modules, as long as their prescriptions
are not contradictory. Each module is identified by a name and a version
number.

The list of modules that apply to a specific trajectory file is stored
in the dataset h5md/modules, which is a one-dimensional array. Each
element of the array is a compound data element containing the module
name as a variable-length string, and the version number stored as an
integer.
------------------------------------------------------------

 > It sounds nice at first; but don't be fooled, figshare will
 > inevitably see its demise just like most other web service,
 > leaving the permanent DOIs to the data files inaccessible.

Eternity is of course valid only on the time scale of the lifetime of
any Web service. That said, people with good contacts to the
publishing sector tell me that Figshare has serious backing and will
likely be around for years even if it makes no benefit. A few
years is good enough for me for the present purpose. It's certainly
longer than anything my own research structure, the CNRS, would ever
be willing to promise.

 > So if figshare provides URLs that we can refer to in the git-annex
 > repository, that will do the job as a mirror. Persistence would be

Figshare provides a URL for a page describing a dataset, and another
one for direct download. Plus a DOI, through which the two URLs can be
obtained.

Konrad.
-- 
---------------------------------------------------------------------
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
http://dirac.cnrs-orleans.fr/~hinsen/
ORCID: http://orcid.org/0000-0003-0330-9428
Twitter: @khinsen
---------------------------------------------------------------------