`DataLayouts` F Dimension Constructor Checks

[jb-mackay]

Adds F dimension checks on backing data arrays for DataLayouts. See #664

[jb-mackay]

• [x] VF
• [x] IF
• [x] IFH
• [x] IJF
• [x] IJFH
• [x] VIFH
• [x] VIJFH
• [x] IJKFVH

[jb-mackay]

@simonbyrne is this all sensible?

[jb-mackay]

https://github.com/CliMA/ClimaCore.jl/blob/5540fc43fc00819d0c21fdbb102d9986191dd2a4/test/DataLayouts/data1d.jl#L114-L117

This does not make sense - data1 and data2 are arrays, not VF structs which we are testing. The code without @test_prop errors. They should be replaced with dl1, dl2(?), which then fail the JET tests and also error on the mapping functions.

[jb-mackay]

@sriharshakandala would you be able to help me understand/debug the errors now in dss_2d!?

[simonbyrne]

Can you figure out why the test is failing?

[sriharshakandala]

@sriharshakandala would you be able to help me understand/debug the errors now in dss_2d!?

DSS buffer sizing can be different from the number of fields. The sizing is conservative and typically determined based on maximum number of fields being dss'd in a given simulation. When building the spectral element space, we are dss'g the jacobian, which is a length 1 field. But, the buffer is sized for maximum number of fields that will be dss'd at a time in the simulation.

[simonbyrne]

Maybe change the assertions to inequalities?

[jb-mackay]

Maybe change the assertions to inequalities?

It seems like the info necessary to compute a conservative upper bound aren't really accessible at the DataLayout level. I'm inclined to close this PR and leave things as is.