Common field registry and virtual fields

[pgrete]

I was wondering if there's need to have some machinery to create virtual fields (or if this is too much of an overhead upstream and should be handled downstream) -- similar to the discussion about a common name registry we talked about during the sync yesterday.

With virtual fields, I imagine some Metadata::Derived field that is also Metadata::Virtual/Unallocated that translates to the field being specified by a callback function (e.g., to get a temperature from the conserved variables).
The main advantages over existing machinery that I see is that it saves memory.
IIRC this is also a feature that is currently being implemented (or maybe already is) in ADIOS2 so that new fields are available for direct access in post processing tools without requiring additional disk space.

[Yurlungur]

👍 I think this would be great to have. We pondered adding this to riot but never found a great non-intrusive mechanism for doing so.