Re: [h5md-user] attribute modification_time

[Felix Höfling]

Am 29.05.2013, 16:51 Uhr, schrieb Peter Colberg  
<address@hidden>:

> On Wed, May 29, 2013 at 01:07:36PM +0200, Felix Höfling wrote:
> > the /h5md group contains the attribute "creation_time". I suggest to
> > provide "modification_time" (or "last_modified") as well to document
> > manipulations of the H5MD file after the initial simulation. The
> > description could be:
> >
> > modification_time: The time when then file was modified last. It
> > uses the same format as "creation_time". H5MD writers are advised to
> > update the field just before closing the file.
> >
> > What do you think?
>
> When a simulation has finished successfully, I `chmod 444` the input
> and output files and never write to them again. That practice should
> be standard for molecular simulations.
>
> Of course users are free to modify their files after the fact, but
> should this behaviour be encouraged by the specification?
>
> Regards,
> Peter

Just some examples:

A user may decide to continue a simulation and to append the results to  
the same file.

One may want to delete some unneeded data groups after a while (data that  
should not have been written in the first place).

A H5MD file may also facilitate the setup of more complex systems und  
therefore undergo further processing: first equilibrate a homogeneous  
system, then cut out a region and replace it by something else. Or  
bringing two separate systems into contact.

Of course, these things could be split across several files which is at  
some variance to the HDF5 idea of having everything in one file. And for  
such cases, it makes sense to record what happened. As this easily opens  
Pandora's box we may forget about it for the upcoming first release.

Felix