How does `geom_abline()` not affect the x and y scales?

Question

Does anyone know which argument within geom_abline() is responsible for not affecting the x and y scales?

The function draw_panel() w/i GeomAbline gets the underlying "ranges", but the line should than be typically lie outside the original scales:

GeomAbline <- ggproto("GeomAbline", Geom,
  draw_panel = function(data, panel_params, coord) {
    ranges <- coord$backtransform_range(panel_params)

    data$x    <- ranges$x[1]
    data$xend <- ranges$x[2]
    data$y    <- ranges$x[1] * data$slope + data$intercept
    data$yend <- ranges$x[2] * data$slope + data$intercept

    GeomSegment$draw_panel(unique(data), panel_params, coord)
  },

  default_aes = aes(colour = "black", size = 0.5, linetype = 1, alpha = NA),
  required_aes = c("slope", "intercept"),

  draw_key = draw_key_abline
)

My guess are the arguments/functions ggplot2::StatIdentity and ggplot2::PositionIdentity set in ggplot2:layer() of geom_abline(). But I don't understand how this works? My motivation is to write a new geom_* that also does not affect the x and y scales.

The constructor of `geom_abline()` reparametrizes the data to not have any aesthetics related to the x- or y-scales [here](https://github.com/tidyverse/ggplot2/blob/759c63c2fd9e00ba3322c1b74b227f63c98d2e06/R/geom-abline.r#L99-L103). Note that `slope` and `intercept` don't appear in `scale_x_continuous()$aesthetics` (same for y). — teunbrand, Oct 20 '21 at 14:54
Thanks for the correct answer. If you want to write it as an answer I certainly will accept it. I was not aware that just certain aesthetics can affect the x/y scaling. Do you know, if it's possible to add additional aesthetics to be taken into account or if there is a way to tell a certain `geom_*()` to not affect the scaling? — moremo, Oct 21 '21 at 14:27
I'm not sure it is possible to add additional position aesthetics without writing a custom scale constructor. I don't know how one would exclude a layer from participating in scale training. I think scales control this and not layers, so I wouldn't know of an opt-out method for layers. — teunbrand, Oct 21 '21 at 14:58
Ok, thanks! Still a very helpful comment. I have found a very awkward workaround. I have both a customized `stat_*()` and `geom_*()`, whereby the `stat_*()` uses `required_aes = c("x", "y")` but returns `c("x_noaes", "y_noaes")`. The latter is equivalent to the `required_aes` used in the `geom_*()`. Hence, I don't need a new mapping but the scales are not effected by my `geom_*()`. — moremo, Oct 21 '21 at 15:21

moremo · Accepted Answer · 2021-10-21T14:29:10.490

According to @teunbrand's comment, the solution is the following:

Inside ggplot2::scale_x_continuous() only the following aesthetics can affect the x-scaling: c("x", "xmin", "xmax", "xend", "xintercept", "xmin_final", "xmax_final", "xlower", "xmiddle", "xupper", "x0"). However, since geom_abline() does not contain any of these aesthetics, the scales are not affected.

The same applies to the y scaling. By the way, the variables correspond to the characters in ggplot2:::ggplot_global$x_aes and ggplot2:::ggplot_global$y_aes, but these are not used for reasons unknown to me.

For illustration I have rewritten geom_point(), but with aesthetics c("x_new", "y").

geom_my_point <- function(mapping = NULL, data = NULL, stat = "identity",
                            position = "identity", na.rm = FALSE,
                            show.legend = NA, inherit.aes = TRUE, ...) {
  ggplot2::layer(
    geom = GeomMyPoint, mapping = mapping,
    data = data, stat = stat, position = position,
    show.legend = show.legend, inherit.aes = inherit.aes,
    params = list(na.rm = na.rm, ...)
  )
}


GeomMyPoint <- ggplot2::ggproto("GeomMyPoint", ggplot2::GeomPoint,
  required_aes = c("x_new", "y"),
  
  draw_panel = function(data, panel_scales, coord) {
    if (is.character(data$shape)) {
      data$shape <- translate_shape_string(data$shape)
    }

    ## Transform the data first
    coords <- coord$transform(data, panel_scales)

    ## Construct a grid grob
    grid::pointsGrob(
      x = coords$x_new,
      y = coords$y,
      pch = coords$shape,
      gp = grid::gpar(
        col = alpha(coords$colour, coords$alpha),
        fill = alpha(coords$fill, coords$alpha),
        fontsize = coords$size * ggplot2::.pt + coords$stroke * ggplot2::.stroke / 2,
        lwd = coords$stroke * ggplot2::.stroke / 2
      )
    )
  }
)

The output of geom_my_point() with no x aesthetic corresponds to figure "gg2"; hence the red points do not effect the x scales:

d <- data.frame(x = runif(200))
d$y <- 1 * d$x + rnorm(200, 0, 0.2)
d$x2 <- d$x * 2

require("ggplot2")
gg1 <- ggplot(d) + geom_point(aes(x, y)) + geom_point(aes(x = x2, y = y), col = 2) + ggtitle("gg1")
gg2 <- ggplot(d) + geom_point(aes(x, y)) + geom_my_point(aes(x_new = x2, y = y), col = 2) + ggtitle("gg2")

These unknown reasons about why the global settings aren't used are discussed here: https://github.com/tidyverse/ggplot2/issues/4135. TL;DR it's an implementation detail. — teunbrand, Oct 21 '21 at 14:34
Thanks for the link. Ok, so it's basically planned but not yet implemented (and not on the top of their to-dos ;) — moremo, Oct 21 '21 at 14:43

How does `geom_abline()` not affect the x and y scales?

1 Answers1