[Top][All Lists]

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

Re: [h5md-user] The first H5MD modules: "particles" and "observables"

From: Felix Höfling
Subject: Re: [h5md-user] The first H5MD modules: "particles" and "observables"
Date: Wed, 02 Oct 2013 16:48:36 +0200
User-agent: Opera Mail/12.15 (Linux)

Am 02.10.2013, 16:34 Uhr, schrieb Pierre de Buyl


On Mon, Sep 30, 2013 at 11:38:40AM +0200, Felix Höfling wrote:
Now that we have modules defined in the spec, I suggest to take this
concept seriously and to promote the sections on  "particles" and
"observables" as definitions of modules.

These modules will already be part of the H5MD spec v1.0, others may
be added later (updating the version of the spec but not of the H5MD
format itself). The structure of the file would not be changed, just
the way it is presented.

The core specification will then mainly specify the metadata and the
storage of time-independent and time-dependent data and introduce
the concept of a "data group". Further it contains the box
specification (but does not say where to put it, this is left to
each module).

For the fate of the parameters group, I have no strong opinion. I
could become a module as well or stay next to h5md as it is.

What do you think?

I would prefer very much that the current spec does not rely on additional
modules, however standard those modules are.

The spec is clear that a lot is optional. I feel that making /particles and
/observables part of a module would void H5MD of much of its purpose.


The two modules would be part of the H5MD spec, of course. The two root
groups are optional anyway. We would not loose anything by calling them
"modules" IMHO.

My concern is the following: "particles" and "observables" are not
distinguished in the set of possible modules, but take a distinguished
place in the spec now. Think about lattice-Boltzmann simulations: why are
lattice data qualified as "module", but particle positions in the
continuum belong to the core specification? Or why are time-series data
(in observables) "better" than a pair distribution function or a
mean-square displacement?


reply via email to

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