Framework: Recent STK update breaks ROL's downward dependency testing
Created by: dridzal
@trilinos/framework @trilinos/stk @trilinos/panzer @trilinos/rol
Expectations
Restore ROL's downward dependency testing.
Current Behavior
https://testing-vm.sandia.gov/cdash/viewBuildError.php?buildid=3913799
Build error:
/ascldap/users/dridzal/development/TEST/rol-trilinos/Trilinos/packages/stk/stk_util/stk_util/parallel/CommNeighbors.cpp: In member function ‘virtual ompi_communicator_t* stk::CommNeighbors::setup_neighbor_comm(stk::ParallelMachine, const std::vector<int>&, const std::vector<int>&)’: /ascldap/users/dridzal/development/TEST/rol-trilinos/Trilinos/packages/stk/stk_util/stk_util/parallel/CommNeighbors.cpp:81:30: error: ‘MPI_UNWEIGHTED’ was not declared in this scope const int* weights = (int*)MPI_UNWEIGHTED; ^ /ascldap/users/dridzal/development/TEST/rol-trilinos/Trilinos/packages/stk/stk_util/stk_util/parallel/CommNeighbors.cpp:86:43: error: ‘MPI_Dist_graph_create_adjacent’ was not declared in this scope info, reorder, &neighborComm); ^ /ascldap/users/dridzal/development/TEST/rol-trilinos/Trilinos/packages/stk/stk_util/stk_util/parallel/CommNeighbors.cpp: In member function ‘virtual void stk::CommNeighbors::perform_neighbor_communication(MPI_Comm, const std::vector<unsigned char>&, const std::vector<int>&, const std::vector<int>&, std::vector<unsigned char>&, std::vector<int>&, std::vector<int>&)’: /ascldap/users/dridzal/development/TEST/rol-trilinos/Trilinos/packages/stk/stk_util/stk_util/parallel/CommNeighbors.cpp:221:71: error: ‘MPI_Neighbor_alltoall’ was not declared in this scope (void*)recvCountsPtr, 1, MPI_INT, neighborComm); ^ /ascldap/users/dridzal/development/TEST/rol-trilinos/Trilinos/packages/stk/stk_util/stk_util/parallel/CommNeighbors.cpp:236:78: error: ‘MPI_Neighbor_alltoallv’ was not declared in this scope (void*)recvBufPtr, recvCountsPtr, recvDisplsPtr, MPI_BYTE, neighborComm); ^
Possible Solution
I believe that MPI that is loaded for ROL's nightly testing (1.6.5??) no longer works for STK. This needs to be updated ASAP. I simply call
$TRILINOS_DIR/cmake/load_sems_dev_env.sh "default"
so the issue must be in load_sems_dev_env or the definition of "default". I wonder why others are not seeing this issue. Is the "default" abandoned or renamed? On the other hand, pre-push test scripts seem to work just fine ... how are they different?