simplify temporary language settings

[MichaelChirico]

Originally wrote this because I noticed another place where we relied on English locale:

https://github.com/jimhester/lintr/blob/9104a004d50c66d3d77cc4e0d4a2c83067f8b43a/R/get_source_expressions.R#L89

But that's covered by #663 already.

Anyway, I think this approach looks nicer. Considered an approach like with_lang("en", {...}) but the expressions within {...} would be huge, don't think it would actually be better.

Also note that before, we weren't actually restoring LANGUAGE in the event that it was currently unset, e.g.

ORIG = Sys.getenv("LANGUAGE", unset=NA)

Sys.unsetenv("LANGUAGE")

old = Sys.getenv("LANGUAGE")
# do stuff
Sys.setenv(LANGUAGE = old)

# initially, LANGUAGE was unset, now it's set equal to nothing
Sys.getenv("LANGUAGE", unset=NA)
# [1] ""

# true reset. Sys.setenv(LANGUAGE = NA) actually sets "NA"
if (is.na(ORIG)) Sys.unsetenv("LANGUAGE") else Sys.setenv(LANGUAGE = ORIG)

[AshesITR]

Are you sure the current behaviour isn't resetting env?

"LANGUAGE" %in% names(Sys.getenv())
#> [1] FALSE
Sys.setenv(LANGUAGE = "en")
"LANGUAGE" %in% names(Sys.getenv())
#> [1] TRUE
Sys.setenv(LANGUAGE = "")
"LANGUAGE" %in% names(Sys.getenv())
#> [1] FALSE

Tested on R 4.0.3 @ Windows 10.

Also, if we do make a helper, we should use it in all places (I remember R/expect_lint.R and tests/testthat/test-error.R)

[MichaelChirico]

This might be a platform-dependent thing (I'm on Linux Mint ~ Ubuntu):

Sys.unsetenv("LANGUAGE")
"LANGUAGE" %in% names(Sys.getenv())
# [1] FALSE
Sys.setenv(LANGUAGE="")
"LANGUAGE" %in% names(Sys.getenv())
# [1] TRUE

[AshesITR]

Okay then we should fix it in all places. Thanks for pointing it out!

[AshesITR]

#670 causes spurious lints here. I'll whip up a quick fix PR.

[MichaelChirico]

For tests/testthat/test-error.R we'd have to either extract with ::: or re-define it, is it worth it?

[AshesITR]

Using internal functions in tests seems okay to me.

[MichaelChirico]

OK done

[AshesITR]

LGTM