remove -ffast-math compiler flag

[orbeckst]

Fixes #4211 and #3821

Changes made in this Pull Request:

• removed -ffast-math from the compiler flags in setup.py
• Without -ffast-math, the AtomGroup.center_of_charge(..., unwrap=True) pass again on Intel x86_64 macOS.

PR Checklist

• Tests? (existing, checked locally on Intel mac 15.2)
• n/a Docs?
• CHANGELOG updated?
• Issue raised/referenced?

Developers certificate of origin

• I certify that this contribution is covered by the LGPLv2.1+ license as defined in our LICENSE and adheres to the Developer Certificate of Origin.

---

📚 Documentation preview 📚: https://mdanalysis--4220.org.readthedocs.build/en/4220/

[github-actions]

Linter Bot Results:

Hi @orbeckst! Thanks for making this PR. We linted your code and found the following:

There are currently no issues detected! 🎉

[codecov]

Codecov Report

Patch and project coverage have no change.

Comparison is base (66a2b21) 93.62% compared to head (0b9b2d7) 93.62%.

Additional details and impacted files

@@           Coverage Diff            @@
##           develop    #4220   +/-   ##
========================================
  Coverage    93.62%   93.62%           
========================================
  Files          193      193           
  Lines        25295    25295           
  Branches      4063     4063           
========================================
  Hits         23683    23683           
  Misses        1096     1096           
  Partials       516      516           

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

[IAlibay]

Makes sense, it's probably going to have a performance impact but we've been meaning to get rid of that flag anyways.

Do we want to backport this into a 2.5.1? If so could you the move the changelog entry under a 2.5.1 and I'll make a release.

[RMeli]

Not sure how we deal with this stuff, but is it worth adding an entry under "Changed" too?

[orbeckst]

Yeah, I wasn't sure, especially now considering that @hmacdope said in #4211 (comment) that the failing test is not scientifically wrong (under PBC), just technically wrong.

[orbeckst]

I updated CHANGELOG and also included @hmacdope suggestion to include the URL of the original blog post. That's a bit unusual for a CHANGELOG but in this case it's so esoteric that it seems justified and easier than having to comb through the linked issue.

[hmacdope]

Could also note that we are not infectious any more https://moyix.blogspot.com/2022/09/someones-been-messing-with-my-subnormals.html

[orbeckst]

Please let me know with "changes requested" what you need me to do. Sorry, I have a ton of other things to attend to and it's a lot easier just to be told what to fix than having to think hard what ought to be done. Thanks!

[orbeckst]

Re backport #4220 (review) : given @hmacdope 's assessment that the result of the test is still scientifically valid, I don't think we need a backport as I would consider it a small reproducibility issue but not a correctness issue and we (=you @IAlibay ) can spend time on other things. But if anyone else thinks we should backport, I am of course happy to change the CHANGELOG.

[IAlibay]

I don't think we need a backport as I would consider it a small reproducibility issue but not a correctness issue

Sounds good, I think we're likely close to needing another release soon (we're at 2.2 months roughly and we're meant to have a 3 monthly cadence) so we could just say we'll release at some point in August.

[orbeckst]

@RMeli can I please ask you to shepherd the PR to completion?

[RMeli]

Thanks @orbeckst, LGTM