feat: test thread-safety

[jkrimmer]

This PR adds a basic test for the thread-safety of FINUFFT in julia. According to #63 , this test fails on MacOS but not on Windows or Linux.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 58.83%. Comparing base (2a54a96) to head (e87be1d).

Additional details and impacted files

@@           Coverage Diff           @@
##           master      #65   +/-   ##
=======================================
  Coverage   58.83%   58.83%           
=======================================
  Files           8        8           
  Lines         566      566           
=======================================
  Hits          333      333           
  Misses        233      233           

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

[ludvigak]

Thanks for the PR!
I don't have a Mac to test this, and I am not convinced that we should add the test to the standard test suite unless we can actually make it fail in the CI. (Although I realize that this isn't the case for the cuFINUFFT tests either.)
Maybe we can put this and other tests that catch thread issues under test/extras for now?

[jkrimmer]

I have now modified the CI to run the tests also on an arm64-Mac runner on julia 1.10. Surprisingly, the CI can neither reproduce the issues #63 nor #64 ... Probably, the issue is related to my M3 chip then.
Still, I can move the new test to test/extras if you prefer! :)

[jkrimmer]

I forgot to make sure the tests actually run with multiple threads. Now, the segmentation fault on the Mac is back - besides additional errors which I have to investigate further!

[jkrimmer]

Actually, this one seems to be ready - I have rerun the failing test on Ubuntu and it went smoothly. Now, only the multithreading tests on MacOS fail, probably due to some OpenMP issue that I have not been able to identify yet.

[ludvigak]

Do the tests pass if you sync this PR with master, so that it includes your fix for FFTW thread safety?

[jkrimmer]

I have played around with some more CI tweaks; however, the FFTW thread-safety adaptations do not help with the issues on MacOS...

[ludvigak]

Ok. I don't think we should merge this as long as it's failing in the CI. We could hold off with it until a fix is in place, and then merge it in so that we can detect any future regression on the thread safety. Or just disable the test on MacOS for now, but then we would need to remember to turn it back on.

[jkrimmer]

Sounds good! Then, let's keep this PR open until the MacOS issue has been resolved.