Tighten code, increase test coverage

[dmbates]

No description provided.

[palday]

I was unsure if the removal of some of the multiplication methods would cause problems -- I probably don't have time to check the CIs in detail tonight, but some of them seem to be failing.

What inspired these changes? Got tired of waiting for tests? :)

[codecov]

Codecov Report

Merging #301 into master will increase coverage by 1.43%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master     #301      +/-   ##
==========================================
+ Coverage   93.39%   94.82%   +1.43%     
==========================================
  Files          20       20              
  Lines        1513     1508       -5     
==========================================
+ Hits         1413     1430      +17     
+ Misses        100       78      -22     

Impacted Files	Coverage Δ
src/MixedModels.jl	100.00% <ø> (ø)
src/bootstrap.jl	97.77% <ø> (+2.17%)	⬆️
src/likelihoodratiotest.jl	96.61% <ø> (+3.16%)	⬆️
src/linalg.jl	100.00% <ø> (+4.08%)	⬆️
src/linalg/rankUpdate.jl	98.43% <ø> (+1.51%)	⬆️
src/arraytypes.jl	100.00% <100.00%> (ø)
src/blockdescription.jl	100.00% <100.00%> (+17.85%)	⬆️
src/linearmixedmodel.jl	100.00% <100.00%> (+2.80%)	⬆️
... and 1 more

---

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 6f9ac4f...92798c6. Read the comment docs.

[dmbates]

Grrr. It seems that the issingular method is producing different answers under Windows. Probably need to tweak the tolerances. @palday do you have easy access to a Windows machine for testing? I can reboot a computer into Windows but I only have 64-bit Windows.

[palday]

Since I can no longer go into the office, the same holds for me -- I have a laptop I can boot into 64-bit Windows.

Actually .... 64-bit Windows gets its 32-bit compatibility by basically running 32-bit programs a 32-Windows running on top of 64-bit Windows (hence the WoW acronym in some of the filenames). If you have 64-bit Windows, you may be able to replicate the 32-bit failure just by using the 32-bit julia build.

Otherwise for such things, I tend to modify the test so that gives a bit more debugging output. Once it's fixed, I remove the debugging output and push again -- the commits adding and removing the output cancel out in the squash commit anyway.

There's also the bigger question of whether we want to support x86 Windows .... I haven't seen an x86 Windows box in years, just lots of never-ported 32-bit software running on 64-bit Windows

[dmbates]

@palday I think this is ready to go. Can you take a look at it please? If it seems okay then please squash and merge.

[palday]

Mostly editorial comments, but two real things:

1. testing zerocorr! instead of just zerocorr.
2. big change in the objective of one model

[palday]

The last remaining test failure looks semi-legitimate.: it can't find the correlation value in the kb07 output.

[dmbates]

I think the test failure was spurious. If you agree @palday please squash and merge.

[dmbates]

But all of the other architectures did find that value and previous CI runs on this architecture found it. Is there a way of restarting a single test?

[dmbates]

Also, I don't understand why there is a Tier 2 test of the same combination of Julia version, architecture and operating system. @Nosferican have I made a mistake in changing Tier1.yml and Tier2.yml for the minimum Julia version of 1.4? We do want a minimum Julia version of 1.4 - what I am asking is did I create an accidental duplication of test runs?