Outstanding tasks for History3G that will need generic3G and geom manger work

[bena-nasa]

No action required. Just a placeholder to keep track development tasks needed for History3G that is not local to History3G but I would say "above", i.e. requires changes most likely in generic3g or geom manager that are needed to get stuff into History3G for the most part. I just want to enumerate these in one place so we keep track of the "big" picture and what needs done. We can update this as things get done. These are in no particular order.

1. Time averaging in couplers, beyond implementation, how do we specify this as part of the import spec?
2. Where does wait/done for server go? (see issue Using PFIO server in new History3G design (stage, done, wait) #2792)
3. Handling of vector items coupled to History3G
4. vertical regridding
5. Subsetting such as subsetting vertical levels
6. Since History3G is not doing the regridding now. How do we pass requested regrid options in general to coupler?
7. Bit shaving, in History3G or somewhere else, coupler, server? If in History3G will require a copy to be made potentially.
8. If time averaging may require checkpointing of coupler state (monthly for example)
9. Writing bundles contents
10. Outputting things directly from tile components onto a grid (presumable coupler will handle)
11. Extending geom manager for all grids, especially cube and other structures (locstream/xgrid tile components)
12. Sampler stuff with time dependent geometries
13. Split field capability that was in old History to preserve file spec?

[tclune]

Won't attempt to answer all of the above right away. But ...

• For time averaging, my plan is that the import spec will have an "average" setting. Generic layer will use that and the time step ratio to implement a suitable coupler.
• I think bit shaving should be done inside History layer. It could be implemented as a coupler though if that proves to be advantageous. But the other compression is all on the History side, so it seems to me this should too.
• We need to consider a few approaches to passing regrid options. I think I know more-or-less how regrid options can be one more aspect of the VarSpec, but we may also want to have a more intelligent default method (default bilinear vs conservative) that is obtained from the Field dictionary. (Need to get ESMF to add to their dictionary spec.)