Reference implementations

[Markus Demleitner]

Dear DM,

On Tue, May 10, 2016 at 04:52:23PM +0200, Mireille Louys wrote:
> A data model is usually designed to meet the requirements expressed in
> formalized use-cases .
> To be valid , it should meet the requirements.
> 
> VO-DML will help to express a data model in a machine readable document.
> This does not warrant the classes and attributes defined in the document
> mean something in the use-case context.
> 
> There should be a service or an application that consumes the serialisations
> instances of the model.
> The serialisation documents should be validated with respect to the
> constraints , data types , vocabulary terms expressed in the DM description.
> 
> Reference implementation should work on real data.

I couldn't agree more.  Especially with data models it's terribly
easy to get something subtly wrong that will make it difficult to use
(to put it mildly).

So, yes, before approving a DM standard I'd like to see at least one
thing producing instances and another thing consuming it, and if both
come from the same source, there should be yet another
implementation.

As to how much of the data model actually is exercised in these
implementations... ah well, I think realistically we cannot really
say more than "as much as possible".

Finally, on the point that there are DMs that are meaningless
stand-alone and therefore cannot have reference implementations--
well, these would simply wait in PR until they're used in by DMs that
actually are meaningful by themselves.  It's already happened for
char, no?

Cheers,

       Markus