BUG: Use of constant in tests/slate/test_hdg_poisson.py::test_hdg_convergence causes errors

[JDBetteridge]

Describe the bug
There is something strange happening in the tests/slate/test_hdg_poisson.py::test_hdg_convergence test if run in parallel, but not in the "default pytest order". This is seen in the test suite cleanup PR #3385 when using pytest-split:

https://github.com/firedrakeproject/firedrake/actions/runs/11355091421/job/31583739399?pr=3385

This error goes away if a function in the real space is used

Steps to Reproduce
Steps to reproduce the behavior:
This is quite difficult you have to run

pytest \
    --splitting-algorithm least_duration \
    --splits 4 \
    --group ??? \
    --timeout=1800 \
    --timeout-method=thread \
    -o faulthandler_timeout=1860 \
    -m "parallel[3] and not broken" \
    -v tests

With ??? replaced with 1 to 4 and one of the runs should fail. Reproducing the error locally has proven tricky though.

Expected behavior
This test should pass with a Constant it seems something in some part of the test suite is throwing off the numbering 😕

Error message
https://github.com/firedrakeproject/firedrake/actions/runs/11355091421/job/31583739399?pr=3385

Environment:

• CI

Additional Info
Add any other context about the problem here.

[connorjward]

Can you share the current "fix" for this so it's clear where I need to look?

[wence-]

This line

firedrake/tests/slate/test_hdg_poisson.py

Line 53 in 6cc4ff8

tau = Constant(1)

wants to be a function in Real, I suspect.

My guess would be Constants are not renumbered correctly by UFL and one process gets out of sync?

[JDBetteridge]

The fix was in this commit: 911b0ad

[connorjward]

This line

firedrake/tests/slate/test_hdg_poisson.py

Line 53 in 6cc4ff8

tau = Constant(1)

wants to be a function in Real, I suspect.

My guess would be Constants are not renumbered correctly by UFL and one process gets out of sync?

I think it's very likely to be an issue with out-of-sync ranks. We also do some renumbering tricks in PyOP2 so the problem isn't necessarily from UFL. I will investigate.

[connorjward]

Fixed in #3808. We were hardcoding constant names in SLATE (which I wrote some time ago).