line coverage value mismatch between print()/tally_coverage() and report()

[kforner]

With my package I have an example where the coverage given by print() is different from the one give by report().

I did not manage to create a reproducible example, but I think the problem lies in a different implementation between the two methods.

The code in question is in that function:

Using report() or package_coverage(), line 177 appears covered.

But using print() or tally_coverage(), I get:

tally_coverage(x)

90 R/pkg_test.R       summary.pkg_test  183    16
842        R/pkg_test.R                       summary.pkg_test  176    16
843        R/pkg_test.R                       summary.pkg_test  177     0
844        R/pkg_test.R                       summary.pkg_test  178    16
845        R/pkg_test.R                       summary.pkg_test  179    16
846        R/pkg_test.R                       summary.pkg_test  180     2
847        R/pkg_test.R                       summary.pkg_test  181    16
848        R/pkg_test.R                       summary.pkg_test  183    16

print(x)
> x
srcpkgs Coverage: 99.88%
R/pkg_test.R: 98.89%
[...]

as.data.frame(x)
>
60 R/pkg_test.R       summary.pkg_test        176          3       178         3            3           3         1052        1054    16
61 R/pkg_test.R       summary.pkg_test        177          9       177        21            9          21         1053        1053   196
62 R/pkg_test.R       summary.pkg_test        177         24       177        29           24          29         1053        1053     0
63 R/pkg_test.R       summary.pkg_test        177         36       177        41           36          41         1053        1053   196

> covr:::per_line(x)[["R/pkg_test.R"]]$coverage[177]
[1] 196

As you can see, for tally_coverage(x), line 177 has 0 coverage, while for report() and per_line(), it has 196 coverage.

I guess this is caused by line if (is.logical(x)) any(x) else sum(x) , for which its 2 clauses are in the same line, and one is actually tested but not the other.

Still it seems that there are two different ways of computing the coverage, and probably the correct way is the most conservative, as computed by tally_coverage(x)