Remove NewtonSolver from hyperelasticity demo #3929
Conversation
|
On a related note, where are we with providing the things needed to interface with SNES from C++? |
| VecGhostUpdateBegin(x, INSERT_VALUES, SCATTER_FORWARD); | ||
| VecGhostUpdateEnd(x, INSERT_VALUES, SCATTER_FORWARD); | ||
| PetscReal residual = 0; | ||
| VecNorm(r, NORM_2, &residual); |
There was a problem hiding this comment.
Should we use NORM_INFINITY for residual norms instead (as a naive fix)? I always struggle that NORM_2 grows with numbers of elements so for a finer mesh you'll violate atol very soon.
Btw. are C++ demos run in CI? Could not find the job.
There was a problem hiding this comment.
Btw. are C++ demos run in CI? Could not find the job.
https://github.com/FEniCS/dolfinx/blob/main/.github/workflows/ccpp.yml#L230 - CI step has a confusing name.
|
Is the long term plan to remove the DOLFINx C++ Newton solver, in favor of another solver (like SNES)? If we plan to remove it, I think we need to give some deprecation notice to the C++ users + an alternative that is not copying solver logic across multiple user files. |
|
Yes, the plan is to remove NewtonSolver and provide some tooling/examples to help people interface SNES from C++. However, I've never done it myself, so I have no idea where to even begin. |
|
I think that the example here is sufficiently simple that reproducing the logic of NewtonSolver (and simplifying) is OK (only adds 70 lines of code). |
|
Could you add that deprecated message so we can get this merged for 0.10? |
5 participants
No description provided.