- currently one can enter any path for the solver executables, a check is missing
- moreover for Z88 the executable name is error-prone (just happened to me and cost me half an hour to find out what is wrong), therefore check also the correct filename. For Elmer this cannot be done since there are different executables one can choose of (e.g. for multi-threading etc.)
- we must not hardcode the number of coincidence vector places
The user must have a chance to change this setting for the iterative solvers according to the Z88 docs.
We use as default the number Z88 uses in its distributed example.
- we must not hardcode the number of stiffness matrix places, especially not for the Cholesky solver
The user must have a chance to change this setting in case of a solver error. This is also according to the Z88 docs that says it is a two step process:
- the solver first computed the necessary places
- the user must set the value accordingly in case it is too low
- also some code style fixes
- especially don't hardcode the choly solver that fails quickly for finer meshes
- setting sorcg as default since it shows the best performance with the cantilever tests
