Re: [h5md-user] box and observables

[Peter Colberg]

On Tue, Sep 24, 2013 at 05:14:10PM +0200, Felix Höfling wrote:
> Nothing is said about the sampling intervals in observables. The
> data groups within one subgroup may have very different time grids.
> Of course, this might become a mess, but an analysis routine may
> simply refuse to process such a file (unless it implements some
> search algorithm for equal IDs).
> 
> For the box, essentially nothing changes. The group has merely been
> moved from "observables" to "observables/global".

I must say that I am beginning to side with Pierre on this one, who
sees “premature complexification” of the observables group.

Let’s consider the implementation for a second. Whether box is
replicated to multiple locations or not, a writer would probably
implement a single function that appends the box information at the
current time to the box group, which is invoked by particle and
observables modules that rely on the box information.

Now, from the implementer’s point of view, if there is a global box,
this function also possesses the box group. Otherwise, if there are
multiple box locations, the modules that rely on the box have to
posses a box group each. However, this means that two observables
modules sampled on the same time grid, in the same subsystem group,
now need to share that box group among themselves (if they desire
not to unnecessarily duplicate the box data).

Do you see what I mean with shifting the problem down one level?
Instead of sharing a global box group, now you have the problem of
sharing local box groups among modules of the program.

So, with regard to what Konrad wrote earlier, that readers favour a
global box group, and writers favour local box groups, I would object;
upon closer inspection, both readers and writers favour a single box
group.

Peter