WeeklyTelcon_20220419

Open MPI Weekly Telecon ---

4.1.4 - Looking to do a quick turnaround to get coll_ucc component released.
A bug, William added a check to compare fabric names.
- Thinks the check is correct, but fabric names might not be?
- Hopefully today or tomorrow he'll have it fixed.
Want to make an 4.1.4rc1 soon.
Mellanox is testing this internally, but not in MTT.
- If anyone else wants to test coll_ucc, need to up the priority (it's default=0)

v5.0.0rc6 went out last week.
srun issue opened up and giles already PRed a fix.
Nothing new yet for an rc this week.
Remaining major issues
- orte -> prrte docs
- Ralph has a laundry list for prrte v2.1 that sound like we need fixed for OMPI v5.0.0
  - And sound like a lot of work
  - Two that stand out:
    - mpirun (no DVM) - issue in how resources are being allocated for anything more complex than -np 2
      - Markalle opened a prrte issue on Power9
      - Need to understand this disconnect
    - Bouncing data back between PMIx and PRRTE repeatedly.
  - opal_show help not being aggregated correctly. Possibly above?
  - Java binding issue isn't a blocker, but represents a bigger disconnect between us an PRRTE.
Brian and Jeff are working on an mpirun requirements doc.
- Will need to beef up the mpirun before the exec of prterun.
- No hook in prrte for setting up path in back end daemons
- Brian opened an Issue #10252
- Brian wrote up a first cut at it, and Jeff hasn't yet read it.
- He will paste it into the Issue later today or tomorrow.
  - v5.0.0 blocker items.
Jeff gave the packagers a heads up that we're now including the html docs and they might want to package that.

Reminder: Delete master branch in your fork. It will save you much pain in the future.
- Do it NOW!
No longer master nightly tarballs, so if MTT isn't updated, you're not running recent code.
Howard wants to talk about Fortran.
- He's been trying to use recessitate and what Giles started to do a while ago.
  - OMPI doesn't doe f08 correctly at ALL.
  - Something is wrong in the Fortran->C bindings for years.
  - Howard rebased a fix that Giles did
  - XL Fortran compiler is very inadequate in F08.
  - configury that Giles added is based on the assumption (correct one).
    - If you have cscoping, and #include <ISO_Fortran_binding_support.h> something header.
    - Compiler will pull this in because it's part of the compiler.
    - GNU <8 doesn't use this ISO
- Jeff, Howard and Mark will meet next Thursday to talk about this.
  - Howard things the bindings generations are somewhat seperate.
  - Jeff swears that Rolf told him that the descriptor way, and the older way of just taking a pointer... that the descriptor way is compatible, but since the first element of the struct is a pointer, then it's compatible.
    - Howard agrees with this argument, but it can be a scalar, but scalar will be passed as a pointer as well.
    - Don't think it'll misbehave, but if the compiler supports descriptors we're supposed to support it.
- Reviving Giles work is good and neccesary,
- This is post v5.0.0 topic.

IBM asked nVidia if they would take over PGI compiler build/testing.
- nVidia is still looking into it.

Wiki for face to face: https://github.com/open-mpi/ompi/wiki/Meeting-2022
- Should think about schedule, location, and topics.
- Some new topics added this week. Please consider adding more topics.
Domestic travel might be opening up next few months.