[Sichao25]

By default, the field API gives owned data access by default. The ghost data needs to be accessed from the backend. So far the only use case of the ghost data is in OmegaHLagrangeEvaluator.

[jacobmerson]

We may want to discuss offline. Do we see a situation where we want to expose non-owned dofs through the PCMS API? I'd suspect that information should only be used internally to perform evaluations, etc.

[jacobmerson]

We should be able to rewrite this using tag data fairly easily.

[Sichao25]

We should be able to rewrite this using tag data fairly easily.

@jacobmerson Which part are you referring to? Do you mean rewriting ghost synchronization to store the values as mesh tags?

[jacobmerson]

I mean anything inside the evaluator already knows that it is a field stored as Omega_h tags, so we can directly use the tag info rather than going through the higher level PCMS API.

[Sichao25]

There is one point of mismatch here. I don't think we previously made the assumption that omega h Lagrange backend data must come from a mesh tag, at least it is not explicitly guaranteed by code.

In the current synchronization implementation, I use sync_array, which synchronizes an array according to the distribution pattern of the mesh if I understand correctly. This is different from sync_tag, which synchronizes a specific tag stored on the mesh.

Therefore, the evaluator should get ghost data from FieldData rather than directly from mesh tags, unless we explicitly decide that field data must always be from mesh tags.

[jacobmerson]

Based on offline discussion @Sichao25 will move the Omega_h data out of simple data into it's own FieldData type. SimpleData is intended to just be a really rough data type with pure arrays.

[Sichao25]

The update of OmegaHFieldData is done.