find_edition() / find_description() may have too strong a preference for an installed package

[jennybc]

This is a discovery made during the package development masterclass as posit conf.

We asked participants to check the test coverage for hadley/stringb and there would be no reason for anyone to have installed that package. And devtools::test_coverage() fails in this case.

> devtools:::test_coverage()
ℹ Computing test coverage for stringb
Error in idesc_create_package(self, private, package) : 
  Cannot find DESCRIPTION for installed package stringb

It goes like this: devtools::test_coverage() ➡️ testthat::local_test_directory(pkg$path, pkg$package) ➡️ find_edition(path, package) ➡️ find_description(path, package):

https://github.com/r-lib/testthat/blob/29018e067f87b07805e55178f387d2a04ff8311f/R/edition.R#L14C1-L23

I think find_description() is too eager to consult an installed package. It seems to assume that's safe and preferred if you happen to know the package's name.

I was able to "fix" the devtools problem by not passing the package name, but I don't think that's a great solution (r-lib/devtools#2536). That has other downsides, e.g., the package name won't be available for setting the TESTTHAT_PKG environment variable.

testthat/R/local.R

Line 188 in 29018e0

TESTTHAT_PKG = package,

[hadley]

Consulting that path isn't going to work either (and I can't see that it ever did) because the path is to the test directory, not the package directory.

[hadley]

Maybe test_coverage() just shouldn't call local_test_directory() at all? package_coverage() installs the package to a temporary location and then calls testInstalledPackage() which presumably calls tests/testthat and there is already existing infrastructure to correctly detect the edition?

[hadley]

Also need to consider test_coverage_active_file().

[hadley]

I've made https://github.com/hadley/testcoverage as a test — it contains two tests that will fail if either the edition or the TESTTHAT_PKG are set incorrectly.