encoding on windows with expect_known_output

[tklebel]

I'm having difficulties with expect_known_output on windows due to encoding issues. The reference file was created on unix, and tests pass there. On windows several tests fail. I checked the ouput of my functions, it is UTF-8, so there is no reason why a mismatch between output and reference file should occur.

There seems to be a problem in testthat, as the following reprex suggests:

library(testthat)

text <- c("Acohido “ Snowden")
file <- tempfile()

expect_known_output(text, file, print = T, update = FALSE)
#> Warning: Creating reference output
expect_known_output(text, file, print = T, update = FALSE)
#> Error: `text` has changed from known value recorded in 'C:\\Users\\...'.
#> 1/1 mismatches
#> x[1]: "[1] \"Acohido “ Snowden\""
#> y[1]: "[1] \"Acohido “ Snowden\""

[gaborcsardi]

What's your testthat version?

[tklebel]

2.0.0. Current dev version does not work for me either.

[gaborcsardi]

Looks like this is a bug indeed. The strings are actually equal, but they have different marked encodings, one is "unknown", the other one "latin1". I don't have an immediate workaround unfortunately, except for using expect_output, or capture.output.

[patperry]

This might be related to a bug in R that is fixed in R-devel: wch/r-source@64f889d . (enc2utf8/translateCharUTF8 is broken for Windows native strings for characters in CP1252 that aren't Latin-1, like curly quotes.) You can probably use utf8::as_utf8 to work around this if you need something that works in older versions of R. Alternatively, compare expected output by translating to native encoding, not by translating to UTF-8.

[gaborcsardi]

@patperry they don't seem related to me.

[patperry]

you know better than I do. any time I see a curly quote and "Windows encoding" I assume it's this CP1252/Latin-1 issue

[jonthegeek]

I do not get the bug output with testthat 2.0.1 and R 3.5.1 on Windows.

[hadley]

Closing since it seems like this has fixed itself in the mean time.