BUG: Firedrake tests fail in firedrake container

[inducer]

Describe the bug
It looks like Firedrake tests assumes a different MPI than is installed in the container.

Steps to Reproduce

docker pull docker.io/firedrakeproject/firedrake
docker run -t -i docker.io/firedrakeproject/firedrake
cd firedrake
pytest -k test_dg_advection_icosahedral_sphere tests/firedrake/regression/

Expected behavior
Passes as before.

Error message

(Lengthy traceback snipped)
/usr/lib/python3.12/subprocess.py:571: CalledProcessError
--------------------------------------- Captured stderr call ---------------------------------------
mpiexec: Error: unknown option "-genv"
Type 'mpiexec --help' for usage.

Environment:

• OS: Linux
• Python version: 3.12
• Output of firedrake-status

firedrake-status says

Unable to retrieve virtualenv name from the environment.
 Please ensure the virtualenv is active before running firedrake-update.

when run in the container. That might be a separate bug.

[connorjward]

I think this can be fixed by running make check instead of invoking pytest directly. The container now ships with OpenMPI instead of MPICH so the previous invocation is no longer valid.

[inducer]

Thanks! That almost works.

https://github.com/inducer/loopy/actions/runs/13702656334/job/38320135568?pr=924

I think it might be good to add an --oversubscribe in mpiexec. (But 🤔 that'll make it OMPI-specific again 😢.)

[connorjward]

Thanks! That almost works.

https://github.com/inducer/loopy/actions/runs/13702656334/job/38320135568?pr=924

I think it might be good to add an --oversubscribe in mpiexec. (But 🤔 that'll make it OMPI-specific again 😢.)

I think that the right thing to do is to allow different arguments to be passed to make check. This has cropped up already with a case where mpiexec doesn't exist (somehow use srun instead). @JHopeCollins has been quite active in having ideas for this.

Perhaps this should be another discussion but I am also tempted to drop make check now that we have another entrypoint for the same code (firedrake-check).

[inducer]

LMK if you do decide to go that route. I've just adapted the loopy downstream tests (inducer/loopy#924), so I'm not super eager to do it again, but... I'll do it if I have to.

[connorjward]

LMK if you do decide to go that route. I've just adapted the loopy downstream tests (inducer/loopy#924), so I'm not super eager to do it again, but... I'll do it if I have to.

Will do. I will try and submit a PR to loopy ASAP if it does happen.

[JHopeCollins]

I wouldn't be averse to getting rid of make check and just consolidating it in firedrake-check. It's really good to be able to run the smoke tests with a single call rather than copying the whole line from the website each time, but trying to stay consistent with PETSc's make check if it's causing issues is unnecessary.
Just having firedrake-check would make handling additional arguments easier too, they wouldn't need to be passed through multiple scripts before hitting pytest.