[REVIEW] Adding Cython to Code Coverage

[mdemoret-nv]

Closes #3110

Quick update to add cython to the code coverage.

Since adding --linetrace=1 to the build has a performance impact, the ./build.sh scripts have been updated to only perform this during GPU CI. I used an environment variable to support adding any argument to the python build via BUILD_PYTHON_ARGS.

[GPUtester]

Please update the changelog in order to start CI tests.

View the gpuCI docs here.

[codecov-io]

Codecov Report

Merging #3111 into branch-0.17 will increase coverage by 10.18%.
The diff coverage is n/a.

@@               Coverage Diff                @@
##           branch-0.17    #3111       +/-   ##
================================================
+ Coverage        59.20%   69.38%   +10.18%     
================================================
  Files              142      193       +51     
  Lines             8966    14704     +5738     
================================================
+ Hits              5308    10203     +4895     
- Misses            3658     4501      +843     

Impacted Files	Coverage Δ
python/cuml/metrics/regression.pyx	95.40% <0.00%> (ø)
python/cuml/linear_model/mbsgd_regressor.pyx	94.73% <0.00%> (ø)
python/cuml/metrics/cluster/homogeneity_score.pyx	100.00% <0.00%> (ø)
python/cuml/internals/internals.pyx	100.00% <0.00%> (ø)
python/cuml/ensemble/randomforestclassifier.pyx	72.72% <0.00%> (ø)
python/cuml/metrics/cluster/mutual_info_score.pyx	100.00% <0.00%> (ø)
python/cuml/metrics/accuracy.pyx	100.00% <0.00%> (ø)
python/cuml/common/pointer_utils.pyx	50.00% <0.00%> (ø)
python/cuml/tsa/seasonality.pyx	32.14% <0.00%> (ø)
python/cuml/neighbors/kneighbors_classifier.pyx	96.26% <0.00%> (ø)
... and 41 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 1f563ee...d95d80e. Read the comment docs.

[JohnZed]

Looks great to me. Seems safe for the non-CI case too. The codecov.io displays don't show lines, though - I just see file-level summaries.

[JohnZed]

Maybe this should be a command-line flag instead, like "--python-args"? Seems like that would be more consistent with the other inputs

[mdemoret-nv]

I can try and convert it to a command line argument, but it will be a little riskier since it will be a little different than all of our current command line args. This would be the only one that accepts an argument value. All of the current options (-g, -n, --singlegpu, etc.) are boolean flags which is much easier to parse. I'm just a little concerned that we may miss some edge cases when parsing arbitrary arguments since this runs on several platforms.

I'll give it a shot and see if there are any other build.sh scripts in rapids that take arbitrary values

[mdemoret-nv]

I ended up sticking with boolean flags and added a new --codecov option to build.sh. This doesn't add any risk from parsing argument values and is generic enough to possibly include C++ code coverage one day (if we can figure out how to do it).

I'm open to different argument name. Let me know if you have a better selection. Link to the help section: https://github.com/rapidsai/cuml/pull/3111/files#diff-4d2a8eefdf2a9783512a35da4dc7676a66404b6f3826a8af9aad038722da6823R46-R47

[mdemoret-nv]

Looks great to me. Seems safe for the non-CI case too. The codecov.io displays don't show lines, though - I just see file-level summaries.

FYI, I think it doesnt show up in codecov.io since some of the cython names get mangled. I would look into it but I cant get codecov.io to even load this PR for me.

[JohnZed]

Looks great to me. Seems safe for the non-CI case too. The codecov.io displays don't show lines, though - I just see file-level summaries.

FYI, I think it doesnt show up in codecov.io since some of the cython names get mangled. I would look into it but I cant get codecov.io to even load this PR for me.

Ok it's showing up in codecov for me.. no line level info but that seems to be because there is no baseline from which to diff?

[JohnZed]

I like the --codecov option... matches the existing approach well.