There is a bug in PETSc 3.21.1, that is affecting the Latest workflow and the nightly integration tests:
File "/mdao/u/swryan/.conda/envs/dymos_20240501-003059/lib/python3.12/site-packages/openmdao/solvers/linear/petsc_ksp.py", line 489, in _get_ksp_solver
jac_mat = PETSc.Mat().createPython([(lsize, size), (lsize, size)],
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "petsc4py/PETSc/Mat.pyx", line 1627, in petsc4py.PETSc.Mat.createPython
petsc4py.PETSc.Error: error code 83
[0] MatSetUp() at /home/conda/feedstock_root/build_artifacts/petsc_1714498151856/work/src/mat/interface/matrix.c:935
[0] MatSetUp_Python() at src/petsc4py/PETSc.c:381406
[0] Python context not set, call one of
* MatPythonSetType(mat, "[package.]module.class")
* MatSetFromOptions(mat) and pass option -mat_python_type [package.]module.class
Task Description
There is a bug in PETSc 3.21.1, that is affecting the
Latest
workflow and the nightly integration tests: