Double-check if covr catches all functions that we defined (user-level and anything (cpp, c, etc...)). Does it appear in the coverage report even with 0%?

[matteodelucchi]

No description provided.

[j-i-l]

covr reports the %-lines covered in tests in all files:

> pcov = package_coverage()
> pcov
abn Coverage: 40.43%
R/calc_node_inla_glm.R: 0.00%
R/calc_node_inla_glmm.R: 0.00%
R/fitabn_bayes.R: 0.00%
R/search_heuristic.R: 0.00%
src/abn_toolbox.cpp: 0.00%
src/irls_binomial_br.cpp: 0.00%
src/irls_binomial_fast.cpp: 0.00%
src/mi.cpp: 0.00%
src/node_binomial_marginals_rv.c: 0.00%
src/node_binomial_rv_inner.c: 0.00%
src/node_binomial_rv.c: 0.00%
src/node_gaussian_marginals_rv.c: 0.00%
src/node_gaussian_rv_inner.c: 0.00%
src/node_gaussian_rv.c: 0.00%
src/node_poisson_marginals_rv.c: 0.00%
src/node_poisson_rv_inner.c: 0.00%
src/node_poisson_rv.c: 0.00%
src/node_poisson.c: 0.00%
R/abn-methods.R: 4.93%
R/link-strength.R: 30.86%
R/build_score_cache_bayes.R: 35.87%
R/abn-infotheo.R: 40.37%
R/build_score_cache_mle.R: 52.59%
R/plot-abn.R: 52.76%
R/fitabn_mle.R: 53.41%
src/utility.c: 57.56%
src/node_binomial.c: 58.52%
R/fitabn.R: 63.55%
R/search_hillclimber.R: 72.31%
R/abn-internal.R: 72.56%
R/build_score_cache.R: 74.43%
src/fit_single_node.c: 76.47%
R/mostprobable.R: 78.79%
R/tographviz.R: 80.36%
R/validation.R: 80.43%
src/fitabn_marginals.c: 85.19%
R/getmarginals.R: 88.50%
R/abn-toolbox.R: 89.42%
R/simulateabn.R: 90.83%
src/mobius.c: 93.62%
R/markov-blanket.R: 93.94%
src/irls_poisson_fast.cpp: 95.24%
src/node_gaussian.c: 96.10%
src/mostprobable.c: 98.69%
src/buildcachematrix.c: 100.00%
src/cycles.c: 100.00%
src/irls_binomial_fast_br.cpp: 100.00%
src/irls_binomial.cpp: 100.00%
src/irls_gaussian_fast.cpp: 100.00%
src/irls_gaussian.cpp: 100.00%
src/irls_poisson.cpp: 100.00%
src/rank.cpp: 100.00%
src/searchhill.c: 100.00%

Here are the available functions listed for covr

Setting covr.record_tests=TRUE (so options(covr.record_tests=TRUE) before running package_coverage) we should be able to see what test(s) covers what line.

Considering #109: Having above report (or even better the change in coverage introduced by some pull request) in a nicely visible form on GH would indeed be helpful.

[matteodelucchi]

Does it also cover tests, that are embraced with skip_on_cran()?

[j-i-l]

With setting the env variable NOT_CRAN we can ignore the skip_on_cran():

> Sys.setenv(NOT_CRAN = "true")
> pcov = package_coverage()
> pcov
abn Coverage: 66.43%
src/abn_toolbox.cpp: 0.00%
src/irls_binomial_br.cpp: 0.00%
src/irls_binomial_fast.cpp: 0.00%
src/mi.cpp: 0.00%
src/node_binomial_marginals_rv.c: 0.00%
src/node_gaussian_marginals_rv.c: 0.00%
src/node_poisson_marginals_rv.c: 0.00%
src/node_poisson_rv_inner.c: 0.00%
src/node_poisson_rv.c: 0.00%
R/link-strength.R: 30.86%
R/abn-infotheo.R: 40.37%
R/search_heuristic.R: 48.67%
R/fitabn_mle.R: 56.26%
R/build_score_cache_mle.R: 56.40%
src/utility.c: 58.14%
R/build_score_cache_bayes.R: 68.09%
R/abn-methods.R: 68.31%
src/node_binomial_rv.c: 69.54%
R/fitabn_bayes.R: 72.24%
src/node_gaussian_rv.c: 72.51%
R/abn-internal.R: 72.56%
R/plot-abn.R: 74.80%
R/build_score_cache.R: 74.89%
R/search_hillclimber.R: 78.46%
R/mostprobable.R: 78.79%
R/tographviz.R: 80.36%
R/validation.R: 80.43%
R/calc_node_inla_glm.R: 81.52%
R/fitabn.R: 85.71%
R/calc_node_inla_glmm.R: 88.68%
R/getmarginals.R: 89.42%
R/abn-toolbox.R: 89.42%
R/simulateabn.R: 90.83%
src/fitabn_marginals.c: 92.59%
src/mobius.c: 93.62%
src/node_gaussian_rv_inner.c: 93.85%
R/markov-blanket.R: 93.94%
src/fit_single_node.c: 94.12%
src/node_poisson.c: 94.29%
src/irls_poisson_fast.cpp: 95.24%
src/node_gaussian.c: 96.10%
src/node_binomial.c: 96.11%
src/node_binomial_rv_inner.c: 97.05%
src/mostprobable.c: 98.69%
src/buildcachematrix.c: 100.00%
src/cycles.c: 100.00%
src/irls_binomial_fast_br.cpp: 100.00%
src/irls_binomial.cpp: 100.00%
src/irls_gaussian_fast.cpp: 100.00%
src/irls_gaussian.cpp: 100.00%
src/irls_poisson.cpp: 100.00%
src/rank.cpp: 100.00%
src/searchhill.c: 100.00%

For example fitabn_bayes which as tests that are skipped on CRAN:

R/fitabn_bayes.R: 72.24% as compared to R/fitabn_bayes.R: 0.00% before.

Basically, we want to do Sys.setenv(NOT_CRAN = "true") before assessing the code coverage.

Going to implement this in the existing workflow.