[Mpi3-ft] RTS Proposal - 3rd Draft

[Josh Hursey]

I updated the draft in SVN and on the Wiki page from today's teleconf.
Change log at bottom.

We changed the name of the global file handle validation functions
from MPI_File_validate_sync to MPI_File_validate_all to match the
naming convention for groups/communicators/windows. However, the
'Sync' is a reminder that the operation not only agrees upon a global
list of failed processes, but also synchronizes the file to disk. How
do folks feel about this change? Should we change it back?

-- Josh

Files can be found here:
  https://svn.mpi-forum.org/trac/mpi-forum-web/wiki/ft/run_through_stabilization
  https://svn.mpi-forum.org/trac/mpi-forum-web/attachment/wiki/ft/run_through_stabilization/FTWG-Process-FT-Draft-3.pdf


Change Log:
 * 17.5: Added italics to 'local' and 'global' list upon first introduction.
 * 17.5: Renamed the 'type' parameter to a more descriptive 'mask'
 * 17.5.1: Changed wording for MPI_Group_validate to that which was
posted by Darius on the list
 * 17.5.1: Add a short description of the 'mask' in
MPI_Group_validate_get_num_state
 * 17.5.2: Update second sentence of MPI_Comm_validate_all paragraph
to be a bit more clear.
 * 17.5.2: Update fourth sentence of MPI_Comm_validate_all paragraph
to be a bit more clear.
 * 17.5.2: Added a couple sentences about outstanding collective
across a MPI_Comm_validate_all
 * 17.5.4: Changed "MPI_File_validate_sync" to "MPI_File_validate_all"
to be consistent with the naming convention
 * 17.5.4: Changed 'Advice to implementors' to 'Advice to users' for
the 'sync-barrier-sync' advice.

-- 
Joshua Hursey
Postdoctoral Research Associate
Oak Ridge National Laboratory
http://users.nccs.gov/~jjhursey