Enhance Supporting Nonstandard Classes in train_continuous

[billdenney]

Could train_continuous also include an if block for the case when existing is NULL to better work with nonstandard classes??

train_continuous <- function(new, existing = NULL) {
  if (is.null(new)) return(existing)
  
  if (is.factor(new) || !typeof(new) %in% c("integer", "double")) {
    stop("Discrete value supplied to continuous scale", call. = FALSE)
  }
  if (is.null(existing)) {
    suppressWarnings(range(new, na.rm = TRUE, finite = TRUE))
  } else {
    suppressWarnings(range(existing, new, na.rm = TRUE, finite = TRUE))
  }
}

This issue relates to r-quantities/units#164.

When trying to use train_continuous with a nonstandard class, it ends up calling range(existing, new) (

scales/R/scale-continuous.r

Line 41 in 3c17ebf

suppressWarnings(range(existing, new, na.rm = TRUE, finite = TRUE))

). With classes beyond base classes (in the case listed above, the 'units' class), this causes the class to be lost and yields issues down the line when using classes that cannot be directly compared.

[hadley]

Looks good. Want to do a PR?

[billdenney]

PR is coming shortly. FYI, I looked at train_discrete, and it wasn't obvious if it should be updated or not, so I didn't change its behavior.

[hadley]

I'm reverting your PR because it breaks ggforce, and I see from the other thread in units it doesn't fix the motivating problem.