Adjust API to be more consistent

[chrisrichardson]

• use squared_norm() for both Vector and MatrixCSR
• Use InsertMode not ScatterMode
• Use finalize for both MatrixCSR and SparsityPattern
• index_map() not map() for Vector

If any others seem obvious, we can add to this branch.

[chrisrichardson]

Now also contains a bugfix for MatrixCSR::finalize which was leading to random results

[garth-wells]

The order of this line and the one below should be switched.

[garth-wells]

I don't think this is clearer. What's being controlled is what happens during a scatter. Moreover, the previous naming had a consistency with PETSc, see https://petsc.org/release/manualpages/Vec/VecGhostUpdateBegin/.

The name change could be confused with switching between insert and set during assembly.

[chrisrichardson]

I think it was inconsistent with PETSc before, and is more inline now: PETSc has a scattermode (forward or reverse).

[garth-wells]

I think it was inconsistent with PETSc before, and is more inline now: PETSc has a scattermode (forward or reverse).

But the change in this PR is PyScatterMode -> PyInsertMode?

[chrisrichardson]

Yes because PyScatterMode was add or insert - it basically was the "Insert Mode" just wrongly named.

[IgorBaratta]

assemble not finalize for MatrixCSR (similar to PETSc and we also use assemble for SparsityPattern)

I still prefer finalize over assemble. Or maybe a more descriptive name like scatter_values()?

[chrisrichardson]

assemble not finalize for MatrixCSR (similar to PETSc and we also use assemble for SparsityPattern)

I still prefer finalize over assemble. Or maybe a more descriptive name like scatter_values()?

Yes, I think so too. I can change it back, unless anyone has a better idea - I'd like to change SparsityPattern to also use finalize() in which case...

[garth-wells]

assemble not finalize for MatrixCSR (similar to PETSc and we also use assemble for SparsityPattern)

I still prefer finalize over assemble. Or maybe a more descriptive name like scatter_values()?

Yes, I think so too. I can change it back, unless anyone has a better idea - I'd like to change SparsityPattern to also use finalize() in which case...

Not crazy about finalize for the matrix :). Sounds as if it can't be re-used. Happy with it for SparsityPattern.

[chrisrichardson]

In which case, the other alternative is some variant of scatter_foo*. Basically this operation is the equivalent operation to scatter_reverse that we do for Vector, except for matrix rows.
Could also be gather_rows() ...