Just use DIFFERENT_NONZERO_PATTERN in PetscMatrix::operator=

Using MatGetInfo, particularly in the way it was being used, to
compare the nonzero pattern is brittle. There are several issues
- Likely the NONZERO_PATTERN is a function of zeroes used, not alloc'd
- Global nz comparison is obviously fraught
- Even comparing nz on process and doing a parallel min is fraught since
  you could have mismatches for different rows that balance out

Author: lindsayad

src/numerics/petsc_matrix.C

@@ -1225,27 +1225,17 @@ PetscMatrix<T> & PetscMatrix<T>::operator= (const PetscMatrix<T> & v)
     {
       PetscBool assembled;
       LibmeshPetscCall(MatAssembled(this->_mat, &assembled));
-      bool same_nonzero_pattern = false;
-
-      if (assembled)
-        {
-          MatInfo our_info, v_info;
+#ifndef NDEBUG
+      const bool cxx_assembled = (assembled == PETSC_TRUE) ? true : false;
+      libmesh_assert(this->_communicator.verify(cxx_assembled));
+#endif
 
-          LibmeshPetscCall(MatGetInfo(this->_mat, MAT_GLOBAL_SUM, &our_info));
-          LibmeshPetscCall(MatGetInfo(v._mat, MAT_GLOBAL_SUM, &v_info));
-          if (our_info.nz_allocated == v_info.nz_allocated)
-            same_nonzero_pattern = true;
-        }
-      else
+      if (!assembled)
         // MatCopy does not work with an unassembled matrix. We could use MatDuplicate but then we
         // would have to destroy the matrix we manage and others might be relying on that data. So
         // we just assemble here regardless of the preceding level of matrix fill
         this->close();
-
-      if (same_nonzero_pattern)
-        LibmeshPetscCall(MatCopy(v._mat, this->_mat, SAME_NONZERO_PATTERN));
-      else
-        LibmeshPetscCall(MatCopy(v._mat, this->_mat, DIFFERENT_NONZERO_PATTERN));
+      LibmeshPetscCall(MatCopy(v._mat, this->_mat, DIFFERENT_NONZERO_PATTERN));
     }
   else
       LibmeshPetscCall(MatDuplicate(v._mat, MAT_COPY_VALUES, &this->_mat));