Adding per property meta_data field.

[JPBergsma]

In issue #410 the idea was raised to create a per property field for metadata.
There was still some discussion on how to name such a field.
But for the moment, appending "_meta" to the end of the field name seems the most popular way to name this field.
I would like to use this metadata field for the ranged properties (See PR #452)
Before I can write a proposal for the metadata field: I however, still have a few questions/ ideas to discuss.

• Nested metadata fields ? Can a field in metadata have a metadata field of its own ?

• Do we want to return the metadata field when the main field is requested via response_fields ?

• Do we want to be able to “override” fields set in the property definition ?
  i.e. the value in the property definition would be the default for all entries, but individual entries can deviate from this.

[merkys]

@JPBergsma internal GitHub links in your post do not seem to work, could you please fix them? This way it would be easier to see the connected issues/PR, and possibly more people from related issues/PRs would find their way here.

To answer your questions:

• Nested metadata fields ? Can a field in metadata have a metadata field of its own ?

Intuitively I would say no, but maybe there is a use case for this?

• Do we want to return the metadata field when the main field is requested via response_fields ?

I would say yes.

• Do we want to be able to “override” fields set in the property definition ?

I would say no. This might overcomplicate OPTIMADE.

[JPBergsma]

Intuitively I would say no, but maybe there is a use case for this?

Would you consider an uncertainty margin/error bar for a value metadata ?
If so, the number of values could be just as large, as the number of values in the main property.
In that case, it may be necessary to treat it as a ranged property (see PR#452) for which we also plan to use the metadata field.

[merkys]

Would you consider an uncertainty margin/error bar for a value metadata ?

Good point. Currently OPTIMADE has no mechanism to communicate uncertainties and metadata could be a place to do so. Alternatively we may define properties with arbitrary suffices (for example, _SU), but I like metadata better.