[petsc-users] PCSetup() changes between 3.0 and 3.1

[Gaetan Kenway]

Hello 

I've run across a somewhat curious problem. I have included a snipped of
code below, that is from a Newton-Krylov flow solver.  The code is from
the "FormJacobian" function for a snes.  The code works fine with PETSc
3.1 but when I tried it with Petsc 3.0, I get a PETSc Error code 73:

"object in argument is in wrong state, e.g. unassembled mat "

on the PCSetup(pc,ierr) call. The subroutine setupNK_KSP_PC(dRdwPre)
assembled the preconditioner matrix (dRdwPre) and performs the
MatAssemblyBegin/End functions. I do realize, it is possible to just use
PETSc 3.1, but I would like to know if its just a fluke that it works,
or I've done something incorrect. I also realize, it is not typical to
set options like this directly in code, but is is necessary for our
code. I apologize I don't have a minimum representative example, but
this section of code is quite buried and an example would have to be
coded from scratch.

Any suggestions are greatly appreciated

Gaetan Kenway



*********** BEGIN CODE ****************

  ! Dummy assembly begin/end calls for the matrix-free Matrx
  call MatAssemblyBegin(dRdw,MAT_FINAL_ASSEMBLY,ierr)
  call EChk(ierr,__FILE__,__LINE__)
  call MatAssemblyEnd(dRdw,MAT_FINAL_ASSEMBLY,ierr)
  call EChk(ierr,__FILE__,__LINE__)

  ! Assemble the approximate PC
  call setupNK_KSP_PC(dRdwPre)
 
  ! Setup the required options for the KSP solver
  call SNESGetKSP(snes,ksp,ierr); 
  call KSPSetType(ksp,ksp_solver_type,ierr);
  call KSPGMRESSetRestart(ksp, ksp_subspace,ierr); 
  call KSPSetPreconditionerSide(ksp,PC_RIGHT,ierr);

  ! Setup the required options for the Global PC
  call KSPGetPC(ksp,pc,ierr);    
  call PCSetType(pc,global_pc_type,ierr);

  if (trim(global_pc_type) == 'asm') then
     call PCASMSetOverlap(pc,asm_overlap,ierr);
     call PCSetup(pc,ierr);              
     call PCASMGetSubKSP( pc, nlocal,  first, subksp, ierr );          
  end if

************** END CODE ****************