Test threaded_legacy fails with T8code.jl v0.4.0

[jmark]

Test threaded_legacy fails with T8code.jl v0.4.0 with the following error while loading required packages:

 T8code [d0cc0030] log:
 ├─possible versions are: 0.2.0-0.4.0 or uninstalled
 ├─restricted to versions 0.4 by Trixi [a7f1ee26], leaving only versions 0.4.0
 │ └─Trixi [a7f1ee26] log:
 │   ├─possible versions are: 0.5.36 or uninstalled
 │   └─Trixi [a7f1ee26] is fixed to version 0.5.36-pre
 └─restricted by julia compatibility requirements to versions: 0.2.0 or uninstalled — no versions left

This is due to the fact that T8code.jl requires Julia v1.9.0 or newer. But this test runs with Julia v1.8.5.
T8code.jl does not run stably with Julia versions older than 1.9.

@sloede @ranocha @jlchan

[ranocha]

So it looks like we would need to drop support of Julia v1.8 in Trixi.jl to depend on T8code.jl. Any thoughts, @trixi-framework/principal-developers ?

[sloede]

I am not a fan. We haven't really had a chance to test Julia for HPC with v1.9 yet. From PASC23 and some discussions on the Julia Slack I know that there are some potential pitfalls with using v1.9 on HPC clusters, mostly related to the new caching stuff. While that can be disabled, it would be nice if we can support v1.8 a while longer.

However, if v1.9 is really a hard requirement for T8code.jl, I wouldn't be opposed to bumping the minimum for Trixi.jl. What does it mean it runs not stable with v1.8? Is this a a general statement or only for certain workloads?

[ranocha]

That's in line with my current thoughts. Julia v1.9 has some negative side effects (in spite of the really good stuff!), so it's hard to argue to drop support in Trixi.jl. The other option would be to allow installing T8code.jl on Julia v1.8 but to strictly tell people not to use it there - e.g., by some very drastic messages during precompilation and/or initialization on Julia v1.8.

[jmark]

I downgrade compat to Julia v1.6 with which the CI of T8code.jl is actually run. New T8code.jl v0.4.1 is already online.

I added comments in the examples and the README and also opened an issue DLR-AMR/T8code.jl#26
to inform resp. warn users about this problem. Hopefully, in the near future this issue will resolve once Julia v1.9 or newer will become more mainstream.

[jmark]

The legacy test passes now: https://github.com/trixi-framework/Trixi.jl/actions/runs/5620674377

[ranocha]

So we can close this issue?

[jmark]

@ranocha From my side, yes! :)