During the last online optimade meeting, the point was raised that we should also define how the metadata fields should be defined in the property definitions.

I expect that some metadata subfields occur in several metadata properties with the same meaning. For example, many numerical properties may have a confidence interval. For now, I have specified that these should have the same $id field. (I hope that this was indeed the intention of the original $id field) but should otherwise be declared repeatedly.
Another option would be to allow custom data types that could be defined elsewhere, but I think that would be too complicated to include in this PR.

I have also added a metadata_for field to the metadata dictionary, as asked by @sauliusg. I do not think we reached a consensus on whether this field should be added, so I can easily remove it if that is what you prefer.

Do we want to return only the field name or the complete path including the field name in the metadata_for field?

As far as I can see, we still don't have a solution to the concerns of some people in the last meeting that it is an issue to limit the field name namespace; i.e., we really should use a different meta name delimiter?

(@sauliusg alternatively argued to reduce the _meta suffix naming to SHOULD or MAY level, and rely completely on metadata_for in the property definition for the connection was the solution. But, I'm still reluctant to that, since because of code design convenience I want to be able to write code that can detect and parse ranged properties in the output without having to know about the property definitions. That will only be possible if the metadata field name is derivable from the field name.)

argued to reduce the _meta suffix naming to SHOULD or MAY level, and rely completely on metadata_for in the property definition for the connection was the solution. But, I'm still reluctant to that, since because of code design convenience I want to be able to write code that can detect and parse ranged properties in the output without having to know about the property definitions. That will only be possible if the metadata field name is derivable from the field name.

I would once again advise against assigning special meanings or special requirements to names. IMHO this is a technological debt, and a pretty nasty one.

Special names pollute the name space, and if we go that way and try to apply it universally, we might very soon arrive at contradictory requirements. We were bitten several times by a decision to use "special" names, do I am wary about such choices.

If a data element "is_metadata_for" is embedded into the data response, what is a problem to test it and get the necessary information during processing?

Is anyone working on resolving the conflict? If not, I can give it a stab.

@merkys Sorry, didn't see your comment. Already did the merge - it was a bit nasty, but not too bad.

I'm happy with this PR as it now is. There is apparently some disagreement over what the key inside "meta" should be to communicate metadata for properties, property_metadata or attributes_metadata. Perhaps @giovannipizzi and @sauliusg who were part of the referenced discussion can take a side on that? Or, for that matter, anyone else.

@merkys Sorry, didn't see your comment. Already did the merge - it was a bit nasty, but not too bad.

No problem, I wanted to get an ACK before embarking on it anyway.