Notes 2019 12 19

Meeting 2019-12-19

Participants

• Kathryn Mohror
• Josh Cunningham
• Bengisu Elis
• Chris Chambreau
• Marc-Andre Hermanns

Topics

MPI_INFO_ENV before init

• Ticket passed a no no-vote in ABQ
• One piece of the puzzle to allow MPI_Info objects before MPI initialization
• Will be out for first vote in Portland

Sessions

• Ticket passed the reading in ABQ
• The other piece of the puzzle to allow MPI_Info objects before MPI initiation
• Will be out for first vote in Portland

MPI_T Events

• Passed first vote in ABQ
• Should receive update if MPI_Info is allowed before initialization
• Need to prepare PR with suitable changes
  • Should be part of a no no-vote
    • Need to discuss process with Wesley and Martin

MPI_T binding objects

• OSU reported an ambiguity with which objects can be given as an argument to a binding object argument
  • Providing MPI_COMM_WORLD as a binding object before MPI_Init caused a crash
  • It's currently erroneous to omit a binding object
    • Impossible to register MPI_T entities with comm binding object before MPI_Init
• 2-fold solution
  1. Provide an Erratum to fix the ambiguity
     • State that it is erroneous to pass in invalid objects
     • Add Advice to users to specifically note that MPI_COMM_WORLD and MPI_COMM_SELF are invalid objects before MPI_Init
  2. Provide an addition to the standard (don't know if this can be an Erratum)
     • Allow passing NULL handles to mean "unbound".
     • This may enable tracking communication within MPI_Init

Fix to Backwards Compatibility Chapter

• Marc-Andre proposed some fixes to the \label of the BC Chapter
• Marc-Andre proposed some boilerplate text fixes for entries in the ChangeLog and BC.