[pera-noaa]

Under Open MPI 4.1.4, programs did not launch, and after a checkpoint they hung.

1. Open MPI's predefined handles are addresses, often with bit 31 set, so the sign-extended lookup missed MPI_COMM_WORLD. Try the exact value first (no change for MPICH).
2. Open MPI's MPI_REQUEST_NULL differs between the halves. A send the drain completed reached the lower half's MPI_Test as the upper half's null, and never completed. Pass the lower half's null.
3. libmpistub.so lacks Open MPI's ompi_mpi_* objects, so libmana.so failed to load. Keep the real library under Open MPI. @xuyao00, if you'd rather the stub define them, I'll drop this commit.

Hercules, 2 nodes (with #527): 18/23 in both modes. The rest are the Fortran tests (the wrappers assume MPICH's handles), attributes (#531), and two tests not built for Open MPI. Intel MPI: unchanged, 22/23 (#524).

🤖 Generated with Claude Code

[coderabbitai]

Warning

Review limit reached

You've used all free OSS reviews for now. Wait for the free limit to reset to keep reviewing this public repository.

Next included review available in 59 minutes.

Check out review usage here.

View limit details

Limit details: You’ve used the included review currently available.

Learn how review limits work.

Review configuration:

⚙️ Run configuration

• Configuration used: defaults
• Review profile: CHILL
• Plan: Advanced
• Run ID: c0066739-8f4d-4b65-9bf4-855332645434

📥 Commits

Reviewing files that changed from the base of the PR and between d602397 and 58eeda4.

📒 Files selected for processing: 3

• mpi-proxy-split/lower-half/lower-half.cpp
• mpi-proxy-split/mpi-wrappers/mpi_request_wrappers.cpp
• mpi-proxy-split/virtual_id.cpp

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[xuyao00]

Can you check if PR#529 solves you problem? I think it's covered in PR#529.

[pera-noaa]

Sorry, I hadn't seen #529 when I opened this. Yes, #529 covers all three: the lookup key by handle size, the lower half's MPI_REQUEST_NULL for requests MANA completed, and the stub's Open MPI objects. On Hercules, with #529 merged onto main (2 nodes, 3 checkpoints per test):

MPI	polling	blocking
Open MPI 4.1.4 (ob1, TCP)	41 of 41	41 of 41
Intel MPI 2021.7.1	40 of 41	41 of 41

This PR had 18 of 23 with Open MPI 4.1.4. The Intel failure is p2p_probe ("told to stop at iteration N"), which fails on main too (5 of 10 runs, against 4 of 10 with #529).

The merge has two conflicts with #522, both small. In p2p_drain_send_recv.cpp, commName() now names MPI_COMM_WORLD 0 and MPI_COMM_SELF 1, and commByName() maps them back. In test/Makefile, fortran_handles is added to the new lists.

#529 also fixes #531: on Open MPI, the first user keyval is now 7. #532 is still open with #529: MPI_UNIVERSE_SIZE, MPI_APPNUM and MPI_LASTUSEDCODE return MPI_ERR_KEYVAL under both MPIs. I'll do it once #529 is in.

I'm closing this in favor of #529.