By setting options(covr.branch_coverage = TRUE) before measuring
coverage, covr also records which branches of each if and switch() in R
code were taken. Line and expression coverage are the same as without it.
Branch coverage is experimental: what counts as a branch, and how branches
are stored and reported, may change.
An if has two branches, "TRUE" and "FALSE". The "FALSE" branch
of an if without an else is taken when the condition is FALSE.
Nothing is recorded if the condition is NA or otherwise signals an
error.
A switch() has a branch for each of its arms, labelled with the name of
the arm. Arms that fall through to the next arm (switch(x, a = , b = 1))
share its branch, labelled "a, b". An unnamed arm is labelled
"<default>", or with its position if all arms are unnamed. A switch()
without a default arm has another branch, "<no match>", taken when no
arm matches, including a number that selects no arm. A switch() with
arms in ... has no branches.
&&, || and loops have no branches, nor does compiled code.
If the line where an if or switch() starts is excluded, e.g. with
# nocov (see exclusions), all its branches are excluded too.
The branches are in the branches attribute of the coverage object, a list
with an element for each if and switch(), named like the elements of
the coverage object (see coverage). Each is a list with the srcref of
the if or switch(), its functions, the labels of its branches, and
value, the number of times each branch was taken. Use
tally_coverage(by = "branch") for a data frame with one row for each
branch, and percent_coverage(by = "branch") for the percentage of
branches taken.
report(), file_report(), lcov(), cobertura(), sonarqube(),
codecov() and coveralls() include the branches. The reports show the
number of branches taken on each line and the percentage of branches taken
in each file, and show a line with branches that weren't all taken as partly
covered. Codecov shows such lines as partly covered too, and counts them as
not covered in the coverage percentage, so the percentage on Codecov is
lower with branch coverage.
The option is read when the code is instrumented, so it has no cost when it
is not set. When it is set, every if and switch() that runs calls a
function of covr, which makes the code slower.
withr::with_options(list(covr.branch_coverage = TRUE), {
cov <- code_coverage(
"f <- function(x) {\n if (x > 0) 'positive'\n 'done'\n}",
"f(1)"
)
})
tally_coverage(cov, by = "branch")
percent_coverage(cov, by = "branch")
Run the code above in your browser using DataLab