How to write a package test in R to see if warning is correctly thrown?

Question

I'm writing some tests for an R package and would like to have R CMD check verify that functions display the correct warnings for certain inputs. But I can't figure out how to capture the warning output so that I can test it.

So if I have a function like:

throwsWarning<-function(x){
  if(x>0){
    warning('Argument "x" is greater than zero, results may be incorrect')
  }
  # do something useful ...
}

I'd want a something in my test file like:

warningOutput <-try( throwsWarning(1))
if (warningOutput!='Argument "x" is greater than zero, results may be incorrect'){
  stop('function "throwsWarning" did not produce correct warning when x>0')
}

So far I've found possible partial solutions by changing options so that warnings are treated as errors and the surrounding with a trycatch block. Also considered testing value of last.warning, but that seems dangerous if the warning is not thrown (would test previous value). Seems like there must be an easy way to do this that I'm missing?

Answer 1

The testthat package has an expect_warning and gives_warning function that you can use.

From the examples, you would do something like this:

R> library(testthat)
R> expect_that(warning("this is a warning"), gives_warning("is a"))
## This does not raise an error, but:
R> expect_that(warning("this is a warning"), gives_warning("nope"))
Error: warning("this is a warning") does not match 'nope'. Actual value: 
this is a warning

So, gives_warning is regular expression that is matched against the warning that is supposed to be emitted. If the regex does not match (or no warning is thrown), then a red flag is raised.

Equally, using the shorter expect_warning:

R> expect_warning(warning("this is a warning"), "is a")

Answer 2

If you're writing your own package, it might make sense to make use of R's condition system by throwing (and catching) particular types of errors or warnings. So

myFun <- function(x) {
    if (any(is.na(x))) {
        w <- simpleWarning("'x' contains NA values")
        class(w) <- c("HasNA", class(w))
        warning(w)
    }
    if (any(x < 0))
        warning("'x' contains values less than 0")
    x
}

and then in your test, e.g., with library(RUnit), use tryCatch and pick off just the conditions that you're interested in testing, i.e., warnings with class HasNA:

test_myFun_NAwarning <- function() {
    warnOccurred <- FALSE
    tryCatch(myFun(1:5), HasNA = function(w) warnOcccurred <<- TRUE)
    checkTrue(!warnOccurred)
    tryCatch(myFun(-(1:5)), HasNA = function(w) warnOcccurred <<- TRUE)
    checkTrue(!warnOccurred)
    tryCatch(myFun(c(1:5, NA)), HasNA = function(w) warnOccurred <<- TRUE)
    checkTrue(warnOccurred)
}

leading to

> test_myFun_NAwarning()
[1] TRUE
Warning message:
In myFun(-(1:5)) : 'x' contains values less than 0

which shows that tryCatch is catching just the particular warning you're interested in, and doing so in a way that does not rely on matching the text of the string. Maybe you'd have a helper function .warn to make all warnings for your package. See ?withCallingHandlers for additional detail; withCallingHandlers and muffleRestart are how one might cope with continuing evaluation after a warning occurred, rather than stopping the way tryCatch does.