expect_comparison could use waldo to determine digits in the failure case?

[MichaelChirico]

testthat/R/expect-comparison.R

Line 39 in 8c2bd6e

sprintf("%s is %s %s. Difference: %.3g", act$lab, msg, exp$lab, act$val - exp$val),

Motivation: imbs-hl/ranger#747

I know {waldo} has recently improved "what's the right number of digits to show when compare() fails", so presumably we could re-use that logic here, e.g. compare(act - exp, 0). The specifics may need to be more careful, but I think it's clearly not helpful to have expect_gt() fail with message "Difference: 0".

[hadley]

Or just inline min_digits()?

min_digits <- function(x, y, tolerance = default_tol()) {
  if (is.integer(x) && is.integer(y)) {
    return(0L)
  }

  attributes(x) <- NULL
  attributes(y) <- NULL

  n <- digits(abs(x - y))
  if (!is.null(tolerance)) {
    n <- min(n, digits(tolerance))
  }

  as.integer(n) + 1L
}