Make PETScLUSolver subclass of PETScKrylovSolver

[Prof Garth Wells]

The would mirror the PETSc design for solvers, and substantially reduce redundant code.

In PETSc, an LU solver is a KSP with an LU preconditioner and pre-only.

[Jan Blechta]

Do you want to follow this logic in parent (generics, wrappers, factories) objects and sister (other backends) objects?

[Prof Garth Wells]

Not for now.

In PETSc, a LU solver is just a KSP object. But DOLFIN treats them differently, which leads to code duplication and interface inconsistencies.

[Prof Garth Wells]

For reference, see http://www.mcs.anl.gov/petsc/petsc-current/docs/manualpages/Mat/MatSetOption.html#MatSetOption for setting matrix type, from which an appropriate solver could be chosen.

[Prof Garth Wells]

Merge branch 'garth/fix-issue-815'. Fixes Issue #815.

→ <<cset a0f5c7c3d475>>