Updates for upstream PETSc assembly changes

[francesco-ballarin]

I am finally working on bringing multiphenicsx up to date with changes on dolfinx main branch related to PETSc assembly.

The main upstream changes are (will tick the box once I've worked on that):

• Unify PETSc matrix and vector factory functions FEniCS/dolfinx#3667
• Unify assembly functions for PETSc data structures (Python) FEniCS/dolfinx#3668
• dolfinx.fem.petsc.LinearProblem for nest and block systems FEniCS/dolfinx#3684 (and follow ups)
• SNES solver interface for standard, block and nest systems FEniCS/dolfinx#3648 (and follow ups)

For now I've worked on the first two items, and made sure that the tests pass.
I'll update the tutorials after working on the last two items, therefore CI is still failing for now.

cc @ordinary-slim in case you want to have a look, test it on your codes or review it.

[ordinary-slim]

My tests do work after the API changes. Note that the old API used to call create_vector(form, restriction); this still runs, but with restriction=Noneand kind=restriction, so it produces an unrestricted vector. I don't think this needs to be addressed.

[francesco-ballarin]

My tests do work after the API changes.

Great, thanks for checking. I'll merge this as soon as CI is green.

Note that the old API used to call create_vector(form, restriction); this still runs, but with restriction=Noneand kind=restriction, so it produces an unrestricted vector. I don't think this needs to be addressed.

I do understand your point, but there is nothing I can make about this. The rationale is that we should copy the signature from dolfinx, and add the restriction arguments at the end. A fix for this specific API change would be to swap kind and restriction, but doing so would make that function inconsistent with the rationale used in the others. Still, I don't think it's worth doing this, considering that there are much major API changes and the users will need to review again their entire code...

[francesco-ballarin]

Note that the old API used to call create_vector(form, restriction); this still runs, but with restriction=Noneand kind=restriction, so it produces an unrestricted vector. I don't think this needs to be addressed.

I do understand your point, but there is nothing I can make about this. The rationale is that we should copy the signature from dolfinx, and add the restriction arguments at the end. A fix for this specific API change would be to swap kind and restriction, but doing so would make that function inconsistent with the rationale used in the others. Still, I don't think it's worth doing this, considering that there are much major API changes and the users will need to review again their entire code...

I've pushed a commit which adds an assert for this case, which hopefully should be enough for the user to realize that there has been an API change.

[ordinary-slim]

Code looks good to me. assign function in petsc submodule is a very welcome addition.

[francesco-ballarin]

Will delay merging until FEniCS/dolfinx#3785 is merged upstream, because that fixes on failed assertion in the tutorials.