[MAINT] Trying to re-enable testing for non-x86 archs

[IAlibay]

Partially fixes MDAnalysis/mdaencore#38 and #3127

Changes made in this Pull Request:

• Make encore compile args a separate list so we can selectively reduce optimisation (it's the only part of the code that seems more sensitive to floating point errors).
• Adds Travis cron jobs for ppc64le and aarch64

PR Checklist

• Tests?
• Docs? - N/A?
• CHANGELOG updated?
• Issue raised/referenced?

[codecov]

Codecov Report

Merging #3149 (266bf61) into develop (a35608f) will decrease coverage by 0.00%.
The diff coverage is n/a.

@@             Coverage Diff             @@
##           develop    MDAnalysis/mdanalysis#3149      +/-   ##
===========================================
- Coverage    92.68%   92.68%   -0.01%     
===========================================
  Files          171      168       -3     
  Lines        22737    22677      -60     
  Branches      3212     3212              
===========================================
- Hits         21074    21018      -56     
+ Misses        1615     1611       -4     
  Partials        48       48              

Impacted Files	Coverage Δ
package/MDAnalysis/coordinates/__init__.py	100.00% <0.00%> (ø)
package/MDAnalysis/coordinates/chemfiles.py	31.81% <0.00%> (ø)
package/MDAnalysis/analysis/encore/cutils.pyx
...e/dimensionality_reduction/stochasticproxembed.pyx
...alysis/analysis/encore/clustering/affinityprop.pyx

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update a35608f...266bf61. Read the comment docs.

[IAlibay]

closing & re-opening due to some GH issues that have stalled updating my latest commits.

[IAlibay]

That's weird, looks like conda is picking up an old version of scipy even thought numpy is picked up as 1.20. I can pin CI to >1.5 I guess. Any thoughts @tylerjereddy ?

https://travis-ci.com/github/MDAnalysis/mdanalysis/jobs/490647828#L2460

[tylerjereddy]

Hmm, one difference is that SciPy is listed explicitly while NumPy is not (presumably implicit), for the conda install line?

I wouldn't think that would be the cause though; the ppc64le arch is certainly not as well supported, but there seems to be an appropriate conda binary for modern SciPy on that platform here: https://repo.anaconda.com/pkgs/main/linux-ppc64le/

[IAlibay]

Hmm, one difference is that SciPy is listed explicitly while NumPy is not (presumably implicit), for the conda install line?

I wouldn't think that would be the cause though; the ppc64le arch is certainly not as well supported, but there seems to be an appropriate conda binary for modern SciPy on that platform here: https://repo.anaconda.com/pkgs/main/linux-ppc64le/

Strange it does pick it up if told to (or afterwards if doing it interactively). I won't question it too much, we can always pin the versions when we release on conda-forge.

[IAlibay]

The provenance of 98 new user warnings is baffling to me: https://github.com/MDAnalysis/mdanalysis/pull/3149/checks?check_run_id=2099327091#step:10:317

[IAlibay]

I'll be honest, I'm not 100% sure this will work. I have enabled the cron job on develop to be daily for now (eventually we can change it to weekly). So maybe it'll get triggered tomorrow, but it's not clear to me that the cron job will run if this doesn't get merged first.

[IAlibay]

We can probably add s390x here for the sake of #1389, but is IBM Z really that prevalent? I guess the z15 was released in 2019, but I can't think of a single cluster that uses it? (definitely nothing in the top500 list)

[IAlibay]

Cheers @richardjgowers I'll merge so we can test that today's cron job is working as intended. We can discuss s390x further in MDAnalysis/mdaencore#38