made GSD windows support optional, now raises Error

[richardjgowers]

Fixes #1923

Changes made in this Pull Request:

• now optionally import GSD modules

PR Checklist

• Tests?
• Docs?
• CHANGELOG updated?
• Issue raised/referenced?

[richardjgowers]

@tylerjereddy I don't have a windows box, but I think this should work.. some tests will need to be marked as xfail for windows (if they're not already)

[tylerjereddy]

Looks like the os checks are reversed -- interestingly, I only had to fix one of them, in topology path, to get tests to pass on win locally, but probably best to adjust both.

[tylerjereddy]

You probably want != here

[tylerjereddy]

Probably want != here

[codecov]

Codecov Report

Merging #2090 into develop will decrease coverage by <.01%.
The diff coverage is 77.77%.

@@             Coverage Diff             @@
##           develop    #2090      +/-   ##
===========================================
- Coverage    89.41%   89.41%   -0.01%     
===========================================
  Files          159      159              
  Lines        18731    18737       +6     
  Branches      2696     2698       +2     
===========================================
+ Hits         16749    16753       +4     
  Misses        1381     1381              
- Partials       601      603       +2

Impacted Files	Coverage Δ
package/MDAnalysis/topology/GSDParser.py	84% <75%> (-1.11%)	⬇️
package/MDAnalysis/coordinates/GSD.py	85.1% <80%> (-1.26%)	⬇️

---

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 edb1643...be2d47e. Read the comment docs.

[richardjgowers]

Ok my last fix wasn't any good as it changed the namespace (ie MDAnalysis.coordinates.GSD.GSDReader wouldn't exist) so it might break scripts rather than fail gracefully as intended. This fix is even more fixier than before.

[tylerjereddy]

LGTM now -- I don't think we need to worry too much about the coverage failures; I suppose we could check for the exception raising on windows if we really wanted.

Once this is merged I think we can remove the gsd pip install from appveyor; this may also mean that 2.7 becomes tractable on Windows too since I think issues there were related primarily to gsd.

[tylerjereddy]

Ok, in it goes. I think we can safely ignore codecov patch diff coverage being a little off target here.

This should mean we can safely remove gsd installs to Windows builds, which were only acting as "mock" imports really.