A more informative error when the user accidentally spreads levels in col_factor(...)

[yjunechoe]

Issue

In col_factor(), levels of the factor must be specified in the first argument levels, supplied as a vector:

x <- c("M", "F", NA)

collector <- col_factor(c("M", "F"))

parse_vector(x, collector)
#> [1] M    F    <NA>
#> Levels: M F

In practice, users may accidentally spread out the levels across arguments of col_factor(). In current implementation, the collector is still successfully constructed, and the error is only caught at the point of parsing, with a rather cryptic message.

collector <- col_factor("M", "F")

parse_vector(x, collector)
#> Error: Expected single logical value

In so far as the second argument of col_factor() (ordered) expects a scalar logical, I think it'd be helpful for col_factor() to check for this early at its creation. Additionally, it would be helpful for col_factor() to inform the likely source of the error if ordered receives a scalar character instead (the user mistakenly treated levels as ...).

Proposed solution

A behavior like this would be helpful for debugging:

col_factor("M", "F")
#> Error in `col_factor()`:
#> ! `ordered` must be a logical, not a string.
#> ℹ Did you intend to pass a vector of levels to `levels`?

And here's the toy implementation for the above (inserts an if block to check ordered):

col_factor <- function(levels = NULL, ordered = FALSE, include_na = FALSE) {
  if (!(is.null(levels) || is.character(levels))) {
    stop(sprintf("`levels` must be `NULL` or a character vector:\n- `levels` is a '%s'", class(levels)), call. = FALSE)
  }
  
  if (!rlang::is_scalar_logical(ordered)) {
    cli::cli_abort(c(
      "{.arg ordered} must be a logical, not {.obj_type_friendly {ordered}}.",
      if (rlang::is_scalar_character(ordered)) {
        c("i" = "Did you intend to pass a vector of levels to {.arg levels}?")
      }
    ))
  }
  
  collector("factor", levels = levels, ordered = ordered, include_na = include_na)
}

Happy to PR this if this is within scope!