Code coverage of R itself

[HenrikBengtsson]

Doing code coverage analysis on the R source code would:

• help identifying core code that is poorly tested or not tested at all
• be a great guide for writing additional tests
• these new tests will result in new bugs being identified
• new tests will lower the risk for new bugs being introduced
• ...

The covr package should be useful for this. It is not obvious to me how to run covr on the core R packages since they're not standalone packages per se, e.g.

> c <- covr::package_coverage(quiet = FALSE)
Error: '[...]/r-source/src/library/stats' does not contain a package!

[jimhester]

I agree this could be very useful, e.g. rho tracks C/C++ coverage for instance.

Tracking the R code with covr is a little tricky though, many of the tests in base R are not in the individual packages themselves, but in the https://github.com/wch/r-source/tree/trunk/tests directory, the packages (particuarly the base package) uses different methods to load than a user package and covr obviously uses base functions when it runs, which may muddle the counts.