Can we make it easier to test UTF-8 handling?

[lionel-]

Porting the UTF-8 tests of rlang to testthat3 was a bit tricky. The waldo_compare() function called by all testthat expectations calls reporter$local_user_output() which automatically skips the test when the locale is set to non-utf-8.

I found this workaround:

local_utf8_test <- function(frame = caller_env()) {
  reporter <- get_reporter()

  old <- reporter$unicode
  defer(reporter$unicode <- old, envir = frame)

  reporter$unicode <- FALSE
}

This doesn't seem like a proper long-term solution though.

[gaborcsardi]

Yeah, that should not happen. local_reproducible_output() should probably not skip anything when not called for its primary purpose.

[gaborcsardi]

@lionel- FWIW this is how cli tests Unicode and colors now: https://github.com/r-lib/cli/pull/231/files#diff-6481dd5af605948a35f35754a60e76df1c572fed58f647c29822fa5861eac1b9R6

[krlmlr]

This affects tests in {utf8}.

Reprex:

library(testthat)

test_that("WAT", {
  withr::local_locale(LC_CTYPE = "C")
  expect_true(TRUE)
})
#> ── Skip (<text>:5:3): WAT ──────────────────────────────────────────────────────
#> Reason: non utf8 locale

^{Created on 2021-07-24 by the reprex package (v2.0.0.9000)}

I have confirmed that this comes from local_reproducible_output(unicode = TRUE), I'm not sure which component is calling this.

[hadley]

I think the behaviour of local_reproducible_output(unicode = TRUE) is ok; if you're deliberately requesting unicode output and you're in a non-UTF-8 locale, there's nothing you can do except for skipping. The problem arises because it's called automatically by the reporters, which have to switch back and forth between "reproducible output" (inside tests) and "user output" (outside testthat).