CI: Multiple CI and test suite changes

[DilumAluthge]

• Run CI on more Julia versions
• Remove Julia 1.9 from the CI matrix
• Run CI on both Intel macOS and Apple Silicon macOS, and set arch correctly on each
• Disable the interactive threadpool

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 87.94%. Comparing base (d8f44f5) to head (83c2eea).

Additional details and impacted files

@@            Coverage Diff             @@
##           master      #50      +/-   ##
==========================================
- Coverage   88.00%   87.94%   -0.06%     
==========================================
  Files          11       11              
  Lines        2117     2115       -2     
==========================================
- Hits         1863     1860       -3     
- Misses        254      255       +1     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[DilumAluthge]

Alright, so good news and bad news.

Good news: This fixes CI on Julia 1.12

Bad news: This breaks CI on Julia 1.9 (which is green on master).

So let's try gating the new code on a Julia 1.12+ version check.

[JamesWrigley]

Nah let's just make 1.10 the new requirement 🤷 It's fine since it's the LTS.

[JamesWrigley]

This is also in #25 BTW though I didn't get around to fixing that.

[DilumAluthge]

Sure, I can remove Julia 1.9 from the CI matrix.

But Julia 1.9 through 1.11 are all broken the same way by this PR: (example CI log on 1.11)

[58](https://github.com/JuliaParallel/DistributedNext.jl/actions/runs/22278423994/job/64444325942?pr=50#step:6:161)
addprocs()/rmprocs(): Test Failed at /home/runner/work/DistributedNext.jl/DistributedNext.jl/test/distributed_exec.jl:1182
  Expression: ex.captured.ex.msg == "Only process 1 can add and remove workers"
   Evaluated: "syntax: expected assignment after \"const\"" == "Only process 1 can add and remove workers"

So I think we still need to gate some of this logic on Julia 1.12+ vs Julia <1.12.

[DilumAluthge]

This is also in #25 BTW though I didn't get around to fixing that.

Is #25 ready to go? If so, wanna merge #25 instead?

[JamesWrigley]

Honestly yeah might as well. There was a race condition I was trying to debug at the time but that I can look at later.

[JamesWrigley]

Done.

[DilumAluthge]

Alright, this PR now consists only of the CI changes.

[JamesWrigley]

Thanks ❤️ Note that there are some test failures on master, most likely because LibSSH.jl isn't threadsafe and 1.12 has an interactive thread by default. You can ignore those failures, I'll fix it at some point.

[DilumAluthge]

1.12 has an interactive thread by default

To get CI green, should we just disable the interactive thread?

E.g. something like -t1 should do the trick - IIRC, -t1 expands to -t1,0 (unlike -t4, which IIRC would expand to -t4,1).

[JamesWrigley]

Yeah I think this environment variable might need to be changed to 1,0:

DistributedNext.jl/test/runtests.jl

Line 17 in d8f44f5

run(addenv(`$cmd $sshtestfile`, "JULIA_NUM_THREADS" => "1"))

[DilumAluthge]

Alright, macOS jobs are failing on this PR, but when I look on master, macOS is failing there as well.

• Tip of master (d8f44f5): 2 macOS jobs, both failed
• One commit before tip (fbca46f): 3 macOS jobs, all three failed
• Two commits before tip (3c591ec): 3 macOS jobs, all three failed

[DilumAluthge]

Okay, it looks like killing workers doesn't work on Julia 1.10 on Windows. (IIUC, it didn't work on Julia 1.9 on Windows either.)

Unhandled Task ERROR: IOError: kill: function not implemented (ENOSYS)
Stacktrace:
 [1] kill(p::Base.Process, signum::Int64)
   @ Base .\process.jl:595
 [2] (::DistributedNext.var"#289#290"{Int64, Int64, Int64, Int64, WorkerConfig})()
   @ DistributedNext D:\a\DistributedNext.jl\DistributedNext.jl\src\managers.jl:788

Windows seems fine on Julia 1.11, 1.12, and nightly. So I'll exclude Julia 1.10 on Windows.

[DilumAluthge]

Alright, so the only CI failures here are on macOS, which as I mentioned above, are also seen on master.

I think this is good to go from my point-of-view.