Dear Forum Members:
We would like to schedule a meeting about "new language bindings" as an action item from our virtual last meeting (just ended).
This covers this topic with sub-topics raised in the Virtual Meeting (plus more):
"MPI object that contains an entire language binding" ... which is due to Jeff Squyres :-)
With these questions (among others to be answered)
--- How to descrtibe this, such as with a normative header, from C, C++, Fortran
--- How to guarantee that a given library (e.g., MPI-4) can also allow internal and layered components (e.g., MPI3) to work.
[Imagine an MPI-3 based library receiving a communicator from an MPI-4 main program in this way]
--- How this approach can alleviate the need for the following
a) _X Big MPI APIs, because the MPI.fn() [maybe MPI->fn() in C] provides the old and new APIs either through multiplicity
of APIs, or through compile-time polymorphism
b) Explanation of how this is strongly typed at compile time
c) Explanation of how this works with PMPI and tools (or what changes it may imply)
d) DIscussion of how this could interact with
--- eliminating MPI_COMM_WORLD and new MPI_Init() variants
--- Sessions and Topological groups
--- Subset APIs
--- Fault Tolerance
[others?]
e) Timing of such a proposal once we agree to formalize it. Probably December. TBD.
Right now, I am proposing 9am US Eastern on October 31, 2018 (Halloween in the US and elsewhere), since some of you may find this to be a scary conversation. This is 2 hours before our next virtual meeting (11am Eastern on 10/31). Alt time:
10am Same day. Third alternative, November 2 @ 8am eastern.
AGAIN:
9am Eastern, 10/31
10am Eastern, 10/31 (right before Virtual Meeting)
8am Eastern, 11/1 ,
Of those of you who would actually attend, do these possible times work :-) . [Let me know -- if you can make and can't make, and I'll try to intersect the data.]
If all time fails, I will do a formal doodle poll.
Thanks,
Tony