A function that produces both output and error

[otoomet]

Consider the function:

root <- function(x) {
   if(x < 0) {
      cat("You cannot compute root of such numbers\n")
      print(x)
      stop("Root of negative number")
   }
   sqrt(x)
}

What is the best way to test the error behavior in such a function?

• If I do tinytest::expect_error(root(-2)) the test passes but I also get output on screen. I'd like the tests to run with no clutter.
• If I test output tinytest::expect_output(root(-2)) I get an error
• If I wrap error in output tinytest::expect_stdout(tinytest::expect_error(root(-2))) I am not quite sure what I am testing, the test still passes if I remove stop.

While I can easily put the value inside the error message in this example, it may be more tricky with complex object with multi-line printout. Is it considered a good habit that complete output of error message should be spit out by stop itself, and not complemented with ordinary printing?

Thoughts/suggestions?

[markvanderloo]

Hi Ott,

It is good practice to give a message only once. So here, I would try to give one single clean message in stop and avoid the use of cat(). Of course you are allowed to do

capture.output({
  expect_error(root(-2))
})

to avoid output to stdout. The test failure will still be captured. This is better than tinytest swallowing everything sent to stdout because now, output capturing is explicit in the test file.

[markvanderloo]

Fixed it in fda29e5:

> library(tinytest)
> x <- 0
> attr(x,"foo") <- "bar"
> x
[1] 0
attr(,"foo")
[1] "bar"
> tinytest::expect_equal(x,0)
----- FAILED[attr]: <-->
 call| tinytest::expect_equal(x, 0)
 diff| Attributes differ 

Not quite as good as I want it to be, but I'm working on better reporting (it's a challenge).