Using Ctrl + Alt + T gives failures because of localized error messages

[graue70]

I am developing a package and I have written tests using the expect_error function from the testthat package. I am using its 'regexp' argument to check whether the correct error is thrown. My test looks like this:

test_that('weird localization issue', {
  skip_if_translated()
  expect_error(if (logical(0)) {print("yes")} else {print("no")}, "argument is of length zero")
})

If I run devtools::test() in the RStudio console, all tests pass. If I use Ctrl + Alt + T, the tests fail. The actual message is the German translation of the expected message:

test_cmsaf.add.R:482: failure: weird localization issue
`if (...) NULL` threw an error with unexpected message.
Expected match: "argument is of length zero"
Actual message: "Argument hat L�nge 0"

My questions are the following:

1. Why do the two ways of running tests behave differently?
2. How can I prevent RStudio from automatically translating error messages?
3. Why does the function skip_if_translated() not work as I expect it to?
4. Who is responsible for � being shown instead of the German Umlaut in the actual message?

This is the first part of the output of sessionInfo():

> sessionInfo()
R version 3.5.3 (2019-03-11)
Platform: x86_64-w64-mingw32/x64 (64-bit)
Running under: Windows >= 8 x64 (build 9200)

Matrix products: default

locale:
[1] LC_COLLATE=German_Germany.1252  LC_CTYPE=German_Germany.1252    LC_MONETARY=German_Germany.1252 LC_NUMERIC=C                   
[5] LC_TIME=German_Germany.1252

I use version 2.0.1 of testthat and RStudio version 1.1.463.

[kevinushey]

This is actually an RStudio bug:

Issue: cannot change R message language at runtime

Normally, messages should be translated according to the active locale (which appears to be German in your case) but that's not happening within RStudio, even though it happens in (child) R processes.

[graue70]

I don't immediately see the direct link between what I experienced and what you posted there, but thank you anyway.

This issue has some very annoying implications: As long as I test error messages, I cannot ...

1. use covr::package_coverage() or covr::report().
2. use Ctrl + Alt + E to check my package, even though running devtools::check() works fine.

I would have said that using commands in the RStudio Console works fine all the time and using the RStudio shortcuts does not work at all, but since covr::package_coverage() fails, this is not true.