Re: [h5md-user] H5MD developer's guide

[Konrad Hinsen]

Pierre de Buyl writes:

 > > Fine with me, but as far as I can tell, no implementation currently
 > > supports all H5MD features from the specification. Ideally the
 > > implementations should then have sufficient comments that explain why
 > > certain approaches were chosen.
 > 
 > This is totally true. I aim to have fully compliant libraries, but
 > no firm deadline on that :-/

And that's why I suggested a Developer's Guide. It would already be of
value as a means of communication between the people on this list
implementing different H5MD libraries for different languages.

I don't believe it is possible to design a solid data model or file
format without implementing it completely. Every single line in a
specification needs to have some demonstrable impact on a reference
implementation. If it doesn't, then it's useless, unclear, impossible
to do, or all of that. Designing a file format without implementing
it is like writing a program without testing it: bugs are guaranteed.

Going through the messages on this list (not *all* of them of course),
I see too much theoretical discussion and not enough feedback from
practice. People argue based on expected problems rather than on real
ones, and discuss unimportant details rather than fundamental
semantics. Could it be that we are all theoreticians?  ;-)

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
---------------------------------------------------------------------