Skip to content

Firedrake meeting 2024 11 20

Connor Ward edited this page Nov 20, 2024 · 13 revisions

Date and time 2024-11-20 1600 UTC

Action Items

  1. Pick Chair and Minuter (PB to pick)
  2. ALL: (ongoing) triage the open issues and confirm if they are indeed still open (and perhaps provide labels)
  3. CW (formerly JB): Move PyOP2 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. JB: Enable merge queues (minutes)
  9. LC: Try to merge RNH' PR: Movable VOM
  10. ANY: Add Python 3.13 to PyOP2, TSFC, FIAT, FInAT CI (and others?)
  11. UZ: ngsPETSc releases (minutes) (more or less done)
  12. PB: Speed up some tests (minutes)
  13. RK: sort out Firedrake USA details and website before Christmas time
  14. CW: tackle Loopy warnings

Minutes

Present: CW (minuter), UZ, FA, JHC, PB, DH, KS, DD, RK, LC, IM

Apologies:

PB: speed up some tests

  • DH: multi-minute compile times seems like a bug. We are possibly having a recursive tree walk or exploding a DAG into a tree.
  • Action point: PB profile codegen to see what is slow.

Allow set_working_tape and stop_annotating to be used as function decorators, and implement no_annotations in terms of stop_annotating.

  • Interface bikeshedding. Small TODOs.

CW: Merging JB's big merge

https://github.com/firedrakeproject/firedrake/pull/3817 is very nearly ready I believe. I am keen to get it merged ASAP as it is holding up any changes to PyOP2.

  • DH: --doi to firedrake-install never really worked. Is prepared to lose this.
  • Action point: fix firedrake-zenodo.
  • Action point: check install and zenodo pages for inaccuracies.
  • Once these and checklist items addressed happy-ish to merge.
  • DH: Script entrypoints may not work. Good candidate for fdtools.

CW: (related) What is the long-term vision for Firedrake installation?

  • I believe that we want to be moving towards pip install firedrake (even without a wheel) as the sole installation method.
  • firedrake-install should be made much more minimal, effectively building PETSc and then called pip install firedrake. Importantly this means:
    • The script should not manage homebrew or apt
    • Extra packages like pytorch, gusto etc should not be installed via special flags. They can just be pip-installed.

Thoughts? If people agree then I can work to rip out a lot of code and incrementally improve the installation process.

  • DH: firedrake-install can go away! Likes idea of having a firedrake-build-petsc script and pip install firedrake. The former could spit out the environment variables that need setting.
  • DH: Not concerned about supporting firedrake-update.

PB: RestrictedFunctionSpace

Ongoing discussion on what is the right interface to restrict any FunctionSpace given the bcs of a problem.

My proposal: RestrictedFunctionSpace(MixedFunctionSpace, bcs)

PR needs review

  • More discussion (and clarity) needed

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.

  1. UZ+PB: ngsPETSc was not installed via pip PR#3873 (merged)

  2. DD: Disk checkpointing for adjoints with schedules: Firedrake PR 3812 (tiny change requested) + Pyadjoint PR 173 (basically ready)

  3. KS: dtype (merged)

Date of next meeting

1600 UTC 2024-11-20

Home

Building locally
Tips

Install Frequently Asked Questions

Running on HPC

Users

Developers Notes

Minutes and agenda of Firedrake meetings


Policies and procedures

Gravity wave scaling

Merge Complex Sprint

Reading Group

Firedrake 2021 Planning Meetings
Clone this wiki locally