[Mpi3-ft] More notes from jul19 mpi forum discussion

[Darius Buntinas]

Here are some notes from today's discussion.

-d


In group_validate_get_state, process states are defined in 17.4 not
17.4.1.  Also flags are mentioned in 17.4.1, so both sections should
be referenced.

Add text that the user cannot use the group routines to change the
rank state to NULL

SET_STATE_NULL:  add a mask parameter so one does not need to allocate
a rank_infos array to set all to NULL

MPI_COMM_KILL: require comm to be an intracomm?

Order of reduction operations:  There's something in the standard
(advice to user possibly) that says that the order of evaluation is
not defined, but it will not change from one call to the other.
However, it's possible that the order may change after a validate_all
because the tree has changed.  We should possibly include an advice to
users on this in the FT chapter.