Need for more flexibility in naming conventions

[tsjackson-noaa]

As we shift focus from CMIP data to model development runs, we lose the benefit of having the metadata post-processed by CMOR et. al. In particular, the design assumption that "each modeling center has one variable naming convention" is becoming too restrictive in practice: at GFDL, different conventions are frequently used in the output of a single model run (issue #257).

The purpose of this discussion thread is to solicit advice on what the "one convention per center" assumption should be replaced with, in a way that's both practical to implement and future-proof for all the use cases we'll want to handle.

One option, proposed in #257, is allowing multiple conventions per data source plug-in. It may be worth going beyond this and allowing multiple name possibilities at the level of individual variables (e.g. "'eastward_wind' might be called 'u', or 'ua', or 'ucomp', or...")