Refactor Geom, Stat & Position

[hadley]

Need to make as homogeneous as possible.

• setup_data(): called once per layer. Returns modified data. (aka geom$parameterise(),
• setup_params(): called one per layer. Returns modified params.
• compute_layer() calls compute_panel() calls compute_group(). Only ever override one of the three. The further up the stack, the more work, but also the more options for efficiency savings.

Biggest difference is scales/panel ranges, because geom called after facets are trained, so get panel ranges rather than scales. I think the main difference here is due to the interaction between coords and scales for computing the expansion. It's too hard to change this now, so it just needs to be clearly documented.

Other stuff to change:

• layer() needs compute_geom(), compute_stat(), compute_position()
• build_grob() in ggplot_gtable() should call compute_geom() with by_layer()
• put ggplot_build() and ggplot_gtable() in the same file
• need to rename panel_info argument to scales in Stat$compute_panel() and scales argument to panel_ranges in Geom$compute_panel()
• Need to call remove_missing() by default in compute_data(). Need to document that if you override this method you need to call it yourself.

Some code ideas

adjust_position = function(self, data) {
  if (empty(data)) return(data.frame())
  params <- self$position$compute_defaults(data)

  plyr::ddply(data, "PANEL", function(data) {
    if (empty(data)) return(data.frame())

    self$position$adjust(data, params)
  })
}

compute_position = function(self, data, scales) {
  if (empty(data)) return(data.frame())

  params <- self$position$setup_params(data)
  data <- self$position$setup_data(data, params)
  self$position$compute_layer(data, scales)
}

# --------------------------------------------

calc_statistic = function(self, data, panel) {
  if (empty(data))
    return(data.frame())

  params <- self$stat$compute_defaults(data, self$stat_params)
  data <- self$stat$compute_data(data, self$stat_params)

  check_required_aesthetics(
    self$stat$required_aes,
    c(names(data), names(params)),
    snake_class(self$stat)
  )

  args <- c(list(data = quote(panel_data), panel_info = quote(panel_info)), params)
  plyr::ddply(data, "PANEL", function(panel_data) {
    panel_info <- panel_scales(panel, panel_data$PANEL[1])
    tryCatch(do.call(self$stat$compute_panel, args), error = function(e) {
      warning("Computation failed in `", snake_class(self$stat), "()`:\n",
        e$message, call. = FALSE)
      data.frame()
    })
  })
},

compute_statistic = function(self, data, panel) {
  if (empty(data)) return(data.frame())

  params <- self$stat$setup_params(data, self$stat_param)
  data <- self$stat$setup_data(data, params)
  self$stat$compute_layer(data, scales)
}