246 flexible delayed choices

[chlebowa]

Closes #246

Added functions first_choices and last_choices to allow for specifying subset as "first n" and "last n" of available choices.
Both functions take argument n which must be integer-line and greated than 1. This way specifying first_choices(2) automatically sets multiple = TRUE where appropriate.
Class all_choices is changed to class multiple_choices.

[gogonzo]

@donyunardi please wait with the release until this is merged 👍

[gogonzo]

first_choice should be now:

first_choice <- function() {
  out <- first_choices(1)
  class(out) <- c("delayed_choices", "delayed_data")
  out
}

[chlebowa]

first_choice should be now:

first_choice <- function() {
  out <- first_choices(1)
  class(out) <- c("delayed_choices", "delayed_data")
  out
}

No. There must be a distinction because the multiple argument defaults to FALSE for first_choice() and to TRUE for first_choices(n), so the class multiple_choices is necessary.

Allowing n = 1 in first_choices is a valid point, though, because one might want to allow multiple but select only one. Do you want to me to loosen the assertion?

[chlebowa]

first_choice should be now:

first_choice <- function() {
  out <- first_choices(1)
  class(out) <- c("delayed_choices", "delayed_data")
  out
}

No. There must be a distinction because the multiple argument defaults to FALSE for first_choice() and to TRUE for first_choices(n), so the class multiple_choices is necessary.

Allowing n = 1 in first_choices is a valid point, though, because one might want to allow multiple but select only one. Do you want to me to loosen the assertion?

Sorry, I misread your comment.

I think it's better to have independent implementations but if you prefer this way, I'll change it.

[gogonzo]

first_choice should be now:

first_choice <- function() {
  out <- first_choices(1)
  class(out) <- c("delayed_choices", "delayed_data")
  out
}

No. There must be a distinction because the multiple argument defaults to FALSE for first_choice() and to TRUE for first_choices(n), so the class multiple_choices is necessary.
Allowing n = 1 in first_choices is a valid point, though, because one might want to allow multiple but select only one. Do you want to me to loosen the assertion?

Sorry, I misread your comment.

I think it's better to have independent implementations but if you prefer this way, I'll change it.

I think there is even more room for improvement. All these functions repeat the same condition. Please create a private function which can evaluate custom functions:

• first_choice: function(x) head(x, 1)
• last_choice: function(x) tail(x, 1)
• first_choices: function(x) head(x, n)
• last_choices: function(x) tail(x, n)

.delayed_choices <- function(fun) {
   structure(
    function(x) {
      if (inherits(x, "delayed_choices")) {
        x
      } else if (length(x) == 0L) {
        x
      } else if (is.atomic(x)) {
        fun(x)
      } else if (inherits(x, "delayed_data")) {
        if (is.null(x$subset)) return(x)
        original_fun <- x$subset
        x$subset <- function(data) {
          fun_x(original_fun(data))
        }
        x
      }
    },
    class = c("delayed_choices", "delayed_data")
  )
}

first_choices <- function(n) {
  out <- .delayed_choices(fun = function(x) tail(x, n))
  class(out) <- c("multiple_choices", class(out))
  out
}

[chlebowa]

Something like that was an idea I had but I didn't want to overcomplicate things prematurely 😉

[gogonzo]

Great! 👍