Firedrake meeting 2024 09 04

Date and time 2024-09-04 1600 BST (1500 UTC)

Action Items

1. Pick Chair and Minuter (KS to pick)
2. ALL: (ongoing) triage the open issues and confirm if they are indeed still open (and perhaps provide labels)
3. JB: Move pyop3 and TSFC to firedrake and move FInAT to FIAT
4. ALL: do things with SV's branches
5. DH: Email to Andreas to have 2 (+ others!!!) loopy PRs merged TODO: FIND OUT WHICH PRS THESE ARE
6. DH: Get Firedrake a docker open source account (link here)
7. DH: Talk to GregVernon about PR#2116.
8. DH: Order more Firedrake stickers DONE
9. JB: Enable merge queues (minutes)

Notices

1. Firedrake User Meeting 16-18 September 2024 Firedrake (Registration 25th August/Abstracts 18th August)

Minutes

Present: CW (minuter), DD, KS, DH, RK, IM, NB, PB

Apologies: JB

JB: Stuff

JB is away this week and probably won't make the meeting, but there are several things up in the air:

• I have a bunch of small PRs open that should probably just be merges (see Merge PRs).
• I think the FInAT hash stuff is ready, I have written a test, I don't think coverage of more elements is necessary.
• The caching changes that are causing deadlock need another look. This is much closer to being merged, this needs:
  • Another look over the PyOP2 changes/review.
    • DH: thinks a meeting with JB to review would be more productive.
  • Comments/feedback on PYOP2_SPMD_STRICT mode (not switched on by default, but I've already found it exceedingly useful).
    • DH: thinks this is sensible.
  • Feedback/review on the corresponding Firedrake changes, specifically I'd like some group thought/wisdom about the hashing functions that are used (ensuring that they are unique, I don't necessarily know all the details of the arguments).
    • See feedback from DH.
• I have had this PR open for a log time as a bit of a pet project, but find myself needing it more and more. The idea is to allow running the full test suite on systems that do not allow a recursive MPI init call (Side note: there is an officially blessed MPI way of doing this, see: https://github.com/JDBetteridge/mpispawn for a proof of concept solution). This Firedrake branch now contains a bunch of test fixes that ought to be moved back to master. What I want to know is:
  • Can we switch to "MPI on the outside" test execution? Or will I have to keep maintaining this branch to test on HPC and other pedantic systems?
  • Additionally, does someone want to fix, or at least tell me how to fix tests/output/test_adjoint_disk_checkpointing.py::test_disk_checkpointing? I have had to mark the test broken in this PR. The test pollutes the tape and only passes on master because each parallel test is run in isolation! Also, I don't think pytest-mpi propagates markers like xfail. (I (Daiane) can see this.)
  • DH: Considers this too complex a PR to review without JB present.

Can whoever minutes also make it very clear what the outcomes of the discussions are so that I can pick this up next week when I'm back.

Merge PRs

Note that PRs put in this section should either be trivial or already have been reviewed. Discussion-worthy PRs should be separate agenda items.

JB: FIAT #81 Use some relevant versions of Python for CI and fix missing pkg_resources. - merged

JB: FInAT #136 Requires the above to pass tests, this can be re-run "live" the tests take <1min. - approved

JB: FInAT #134 Update FInAT UFL element hashes so they are persistent across invocations (so expressions can be retrieved from disk cache). There is proof that Firedrake works with these changes. - merged

PB: FD #3719 PatchPC: Log flops - merged

PB: FD #3734 Fix TransferManager reuse for MixedFunctionSpace - merged

PB: pyop2 #727 Fix Function.sub(i).zero() - merged

PB: FD #3749 Test Function.sub(i).zero() - approved, set to auto-merge

PB: FIAT #82 Macro: bin points to nearest cell - not ready

PB: FInAT #137 sympy2gem: if-then-else - merged

DD: Fix MUMPS macOS #3752 - small comment, otherwise ready to merge

• DH: Suspects something is wrong inside MUMPS causing the pytest issue. Thinks it is acceptable to treat as a 'wontfix'.

DD: Fix link in FWI demo #3756 - merged

Date of next meeting

1600 BST (1500 UTC) 2024-09-11