One of the most pernicious challenges to debug is when a test runs fine in your test suite, but fails when you run it interactively (or similarly, it fails randomly when running your tests in parallel). One of the most common causes of this problem is accidentally changing global state in a previous test (e.g. changing an option, an environment variable, or the working directory). This is hard to debug, because it's very hard to figure out which test made the change.
Luckily testthat provides a tool to figure out if tests are changing global
state. You can register a state inspector with set_state_inspector() and
testthat will run it before and after each test, store the results, then
report if there are any differences. For example, if you wanted to see if
any of your tests were changing options or environment variables, you could
put this code in tests/testthat/helper-state.R:
set_state_inspector(function() {
list(
options = options(),
envvars = Sys.getenv()
)
})
(You might discover other packages outside your control are changing the global state, in which case you might want to modify this function to ignore those values.)
Other problems that can be troublesome to resolve are CRAN check notes that report things like connections being left open. You can easily debug that problem with:
set_state_inspector(function() {
getAllConnections()
})
set_state_inspector(callback)Either a zero-argument function that returns an object
capturing global state that you're interested in, or NULL.