Parallel tests use different language

[flying-sheep]

I have Config/testthat/parallel: true in my DESCRIPTION

data_uris <- function(..., mime = '', encoding = 'base64', files) {
	stopifnot(length(list(...)) == 0L)
	vapply(
		files,
		function(f) base64enc::dataURI(mime = mime, encoding = encoding, file = f),
		character(1L))
}

test_that('Extra arguments to data_uris error', {
	expect_error(
		data_uris('one', mime = 'text/html'),
		'length(list(...)) == 0L is not TRUE',
		fixed = TRUE
	)
})

When I run it (edit: with german locale set), I get

Error (test_utils.r:3:2): Extra arguments to data_uris error
Error: length(list(...)) == 0L ist nicht TRUE
Backtrace:
 1. testthat::expect_error(...)
 7. repr:::f()
 8. repr::data_uris("one", mime = "text/html")
 9. base::stopifnot(length(list(...)) == 0L) R/utils.r:137:8

[gaborcsardi]

I cannot reproduce this with testthat 3.0.0 from CRAN.

[gaborcsardi]

I have put this code in testthat, in this branch: https://github.com/r-lib/testthat/tree/fix/issue-1213 and the tests run fine in parallel on macOS.

@flying-sheep Does this work for you?

[flying-sheep]

Aha! The problem seems to be that other than in sequential mode, in parallel mode tests don’t get run in a normalized locale. Therefore I get ist nicht TRUE.

[gaborcsardi]

Oh, that is a testthat bug indeed then!

[hadley]

@gaborcsardi Should we be setting LANGUAGE / LANG in local_reproducible_output()?

[hadley]

No that doesn't seem right, because it would be a rather large change. Why aren't the child processes inheriting LANG from the parent?

[gaborcsardi]

Yeah, they do inherit all env vars. And testthat does not set the language afaik. So I am not sure why it is different in the main process. Let me try to reproduce this.