Problem with LUSolver

[Praveen C]

LUSolver complains that matrix is not set even though it is. See attached code. It works with version 1.4 but not with 1.5

[Praveen C]

Also see this discussion http://fenicsproject.org/qa/6790/getting-petsc-error-code-is-73-with-version-1-5

[Miro Kuchta]

Hi, the code works if you first assemble the matrix and then construct the LUSolver solver.

A = PETScMatrix()

assemble(a, tensor=A)
b = assemble(L)
bc.apply(A,b)

solver = LUSolver(A)

[Praveen C]

Yes, but assembling after creating solver used to work before. If I am doing a time dependent problem, I want to create a LUSolver once and reuse it in each time step, making use of some optimizations like same_nonzero_pattern. My matrix changes in each time step but same sparsity pattern, so I wanted to take advantage of that. Has the behaviour been deliberately changed in the new versions, we have to assemble matrix before creating LUSolver ?

[Prof Garth Wells]

This should be an easy fix.

[Prof Garth Wells]

Fixes Issue #492. Print error message if PETScLUSolver is initialised with empty matrix.

→ <<cset 14b79381bbde>>

[Martin Sandve Alnæs]

Removing milestone: 1.6 (automated comment)